Transcriptional Characterization of Glioma Neural Stem Cells Diva ...

Transcriptional Characterization
of
Glioma Neural Stem Cells
Diva Tommei
Clare Hall College
A dissertation submitted to the University of Cambridge
for the degree of Doctor of Philosophy
European Molecular Biology Laboratory
European Bioinformatics Institute
Wellcome Trust Genome Campus
dt322@cam.ac.uk

A Claudia e Daniela.

This dissertation is my own work and includes nothing which is the outcome
of work done in collaboration, except when specified in the text. It is not sub-
stantially the same as any other that I have submitted for a degree, diploma
or other qualification and no part has already been, or is currently being sub-
mitted for any degree, diploma or other qualification. This dissertation does
not exceed the specified length limit of 60,000 words as defined by The Biology
Degree Committee. This dissertation has been typeset in 12pt Palatino with
one-and-a-half spacing using L ATEX 2ε according to the specifications defined
by the Board of Graduate Studies and the Biology Degree Committee.
Diva Tommei
March 25, 2013

Abstract
Tumours affecting the glial portion of brain parenchyma are termed gliomas and consti-
tute the most frequent and lethal cancers affecting the central nervous system. Glioblastoma
multiforme is the most aggressive glioma in adults and a World Health Organisation clas-
sified grade IV astrocytoma, characterised by widespread intra-tumoural heterogeneity. A
recent advance in the study of gliomas has been the establishment of glioma-derived neural
stem (GNS) cell lines that may represent the glioma cell of origin. While these cell lines
show many similarities to normal neural stem (NS) cells, an important difference is their
capacity to give rise to authentic glioma-like tumours when xenografted into subventricular
strata of immunocompromised mice.
Here I describe an in-depth characterisation of the transcriptome of GNS cells, to identify
differences in the gene expression between normal and glioma-derived cell lines that may
underlie tumorigenesis. Analyses were carried out at the levels of gene expression, molecular
signature profiling, transcript isoform detection and the quantitation of small non-coding
RNAs, taking genetic alterations into account at both the karyotype and mutational level.
Importantly, the cell lines studied were established from tumours with differing histology,
allowing us to sample the breadth of the disease rather than focus on the differences between
unhealthy versus healthy counterparts.
We identified a large cohort of significantly differentially-expressed genes and a smaller
subset of strictly up- and down-regulated ones, including several known glioma oncogenes
as well as novel candidates. An extensive glioblastoma pathway was manually curated
to show the expression of our dataset on the known and unknown glioblastoma-affected
pathways. Interestingly, gene set enrichment analysis revealed a consistent up-regulation of
inflammatory genes in the GNS lines belonging to the MHC class II family, suggesting an
immune-evasion phenotype that has been noted in a number of early glioma studies.
Gliomas have been classified into a small number of subtypes on the basis of patient
survival and response to therapy. We found that the expression signatures of GNS cell lines
closely resembled the mesenchymal and proneural subtypes, as well as reflecting their known
histopathological features. To characterise genes correlating with patient survival time, we
tested for the association between survival time and gene expression in publicly available
glioma and glioblastoma data sets and found four genes to be strongly positively correlated
with patient survival time and patient age. Together these studies provide an in-depth
analysis of a model of glioma pathology driven by an aberrant population of NS cells.
Finally, a package for the performance evaluation of eight leading microRNA target pre-
diction algorithms was built using exon array and microRNA array data from the same GNS
cell lines. This data was used to validate experimentally the target prediction algorithms
that were assessed in their performance as single and combinations of them. The combi-
natorial weight analysis allowed us to conclude that (i) tissue specificity bears a non-trivial
weight in predicting what set of genes a certain microRNA regulates and, therefore, should
be included in future versions of these algorithms, and that (ii) the ElMMO prediction
algorithm fares better than any other combination of prediction algorithms.

Acknowledgements
I would like to thank on a professional note all the members of my lab, starting
from my supervisor Paul Bertone, who has given me the opportunity of con-
ducting research at the University of Cambridge and has taught me over the
years invaluable life lessons that I will always remember. An incredibly special
thank you goes to Pär Engstrom, who supervised my work throughout my time
as a doctoral student and has always been there for me in any matter scientific.
On a more personal note, I would like to thank my parents. Mamma, grazie
dell’appoggio psicologico, culinario, telefonico, automobilistico e soprattutto
affettivo che mi hai dato negli anni passati lontano da casa. Senza le tue cure
ed il tuo affetto non sarei mai arrivata viva a questo giorno. Babbo, grazie
del DNA che condividiamo, grazie del costante supporto mentale, pecuniario
e filosofico che mi dai sempre incondizionatamente, a mo’ di martello pneu-
matico. Grazie degli stimoli incessanti sui quali posso sempre contare. I also
want to thank Sean Cheng for being there for me since March of 2009. Thanks
for these past four years of exciting PhD life that you lived with me. Who
knows what’s next for us. Thanks for being a great person, thanks for under-
standing what goes on in my mind, thanks for the love and affection that you
give me every day and thanks for ruining movies by always anticipating what
will happen next. I secretly love that.

Contents
Page
Introduction 1
1 Glioblastoma 2
1.1 Glial Cells in the Central Nervous System . . . . . . . . . . . . 2
1.2 Glioblastoma Multiforme . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Primary and Secondary Glioblastomas . . . . . . . . . . . . . . 10
1.4 Pathways Involved in Glioblastoma . . . . . . . . . . . . . . . . 21
1.5 Pathway Crosstalk . . . . . . . . . . . . . . . . . . . . . . . . . 30
2 Neurogenesis 33
2.1 Radial Glia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Neural Stem Cells . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3 Brain Cancer Stem Cells 57
3.1 The Cancer Stem Cell Hypothesis . . . . . . . . . . . . . . . . . 57
3.2 Brain Cancer Stem Cells . . . . . . . . . . . . . . . . . . . . . . 63
3.3 Glioma Culture Systems . . . . . . . . . . . . . . . . . . . . . . 72
4 The Non-Coding RNA World 85
4.1 MicroRNA regulation . . . . . . . . . . . . . . . . . . . . . . . . 86
4.2 Target Prediction and Validation . . . . . . . . . . . . . . . . . 90
Methods 93
5 Methods 94
5.1 Tag-sequencing Data Processing . . . . . . . . . . . . . . . . . . 94
5.2 Array Comparative Genomic Hybridization . . . . . . . . . . . . 98
5.3 Differential Gene Expression . . . . . . . . . . . . . . . . . . . . 100
5.4 Quantitative Real Time-PCR Validation . . . . . . . . . . . . . 101
5.5 Literature Mining . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.6 Differential Isoform Expression . . . . . . . . . . . . . . . . . . . 110
5.7 Differential Long ncRNA Expression . . . . . . . . . . . . . . . 112
5.8 Glioma Expression Signatures . . . . . . . . . . . . . . . . . . . 112
i

5.9 External Dataset Expression Correlation . . . . . . . . . . . . . 112
5.10 Glioblastoma Pathway Construction . . . . . . . . . . . . . . . . 114
5.11 MicroRNA Target Prediction Analysis . . . . . . . . . . . . . . 117
Results 119
6 Digital Transcriptome Profiling 120
6.1 Clinical Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.2 Tag mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.3 Copy Number Aberrations . . . . . . . . . . . . . . . . . . . . . 125
6.4 Core Differentially Expressed Genes . . . . . . . . . . . . . . . . 129
6.5 Large-scale qRT-PCR Validation . . . . . . . . . . . . . . . . . 136
6.6 Literature Mining for Differentially Expressed Genes . . . . . . 142
6.7 Isoform Differential Expression . . . . . . . . . . . . . . . . . . . 144
6.8 Long ncRNA Differential Expression . . . . . . . . . . . . . . . 157
7 Dataset Correlation Analyses 161
7.1 Enrichment Analysis . . . . . . . . . . . . . . . . . . . . . . . . 161
7.2 Glioblastoma Expression Signatures . . . . . . . . . . . . . . . . 168
7.3 Tumour Expression Correlation . . . . . . . . . . . . . . . . . . 170
7.4 Survival Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 182
7.5 Glioblastoma Pathway Analysis . . . . . . . . . . . . . . . . . . 187
8 MicroRNA Target Prediction Ensemble Software 202
8.1 Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
8.2 Workflows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
8.3 Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8.4 Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.5 Target Prediction Ensemble Analysis . . . . . . . . . . . . . . . 211
Conclusions 219
9 Discussion 220
9.1 Digital Profiling of GNS Cell Lines . . . . . . . . . . . . . . . . 220
9.2 MicroRNA Target Prediction Analysis . . . . . . . . . . . . . . 231
9.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . 232
Appendix A Differentially Expressed Genes 234
A.1 Differential Expression . . . . . . . . . . . . . . . . . . . . . . . 234
A.2 Classified Differential Expression . . . . . . . . . . . . . . . . . 250
A.3 Quantitative RT-PCR . . . . . . . . . . . . . . . . . . . . . . . 256
A.4 Tag-seq vs qRT-PCR Correlation . . . . . . . . . . . . . . . . . 265
Appendix B Literature Mining Script 266
ii

Appendix C Long ncRNAs 273
Appendix D Glioblastoma Pathway 276
D.1 Pathway Interactions . . . . . . . . . . . . . . . . . . . . . . . . 276
D.2 Pathway Images . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
Appendix E Exon Array Data 286
Appendix F MicroRNA Array Data 300
List of Abbreviations 306
List of Figures 311
List of Tables 314
Bibliography 315
iii

Introduction
1

Chapter 1
Glioblastoma
Contents
1.1 Glial Cells in the Central Nervous System . . . . . . . . . . 2
1.2 Glioblastoma Multiforme . . . . . . . . . . . . . . . . . . . 5
1.3 Primary and Secondary Glioblastomas . . . . . . . . . . . . 10
1.4 Pathways Involved in Glioblastoma . . . . . . . . . . . . . 21
1.5 Pathway Crosstalk . . . . . . . . . . . . . . . . . . . . . . . 30
1.1 Glial Cells in the Central Nervous System
The Central Nervous System (CNS) is built up of neurons and special kinds
of supporting cells called glial cells. Neurons are responsible for the functions
that are unique to the nervous system, whereas the glial cells primarily serve
the needs of the neurons. The functions of the glial cells are still not completely
known. It has been established, however, that they are responsible for isolat-
ing neuronal processes and controlling the environment of neurons as well as
taking part in repair processes. Importantly, glial cells are fundamental during
the development of the nervous system by providing surfaces and scaffoldings
for migrating neurons and outgrowing axons [10,73]. The accepted hypothesis
that glial cells outnumber the 100 billion neurons present in the human brain
with a ratio of up to 50 glial cells per neuron [529] has recently been challenged
and scaled down to a ratio that is closer to one glial cell per neuron [33,73].
Independently of the neuronal and glial counts, the importance of glial cells
remains unquestioned and their functional roles continue to expand away from
the conservative notion of a support structure for neurons [73,173,516]. In fact,
although glial cells do not send precise signals over long distances, they can
produce brief electric currents by opening the membrane channels for Ca 2+
2

1.1 Glial Cells in the Central Nervous System Introduction
and producing "calcium signals”. These signals can spread rapidly and thus
influence many neurons almost simultaneously. Also, since neurotransmitter
release depends on extracellular Ca 2+ concentrations, glial cells can contribute
to coordination of synaptic activity [10,17,73,516]. Finally, glial activation
and inflammation has been implicated in neurodegenerative diseases such as
Alzheimer’s disease, Parkinson’s disease and multiple sclerosis [269]. Glial cells
are usually divided into three categories: astrocytes, oligodendrocytes and mi-
croglial cells, each carrying functional as well as structural differences. Whilst
astrocytes have many processes of various shapes and appear to serve a home-
ostatic function, oligodendrocytes tend to have fewer and shorter processes,
their main function being production of myelin sheaths for axon insulation.
Microglial cells are much smaller than other glial cells and serve to the brain
similar functions that macrophages 1 serve to the rest of the body [73].
In addition to the three main kinds described above, there are other spe-
cialised forms of glial cells. Schwann cells are present only in the peripheral
nervous system and form myelin sheaths that influence axonal thickness, ax-
onal transport and neurofilament content [73]. Ependyma cells are epithelial
cylindrical cells that line the surface of the ventricles 2 of the CNS and are
responsible for the production, transport and absorption of cerebrospinal fluid
(CSF). These cells have recently become candidates for the location of the
neural stem cell niche [81,328]. Müller cells are glial cells that reside in the
vertebrate retina and have recently been observed to undergo dedifferentiation
in vitro into multi-potent progenitor cells in fish, chicken and mouse, to then
differentiate into a number of retinal cell types, although in vivo evidence is
not conclusive [144]. They have also been shown to act as a light collector in
the mammalian eye [53,298]. Bergman cells are astrocytes that reside in the
cerebellum and are responsible for the migration, dendritogenesis, synaptoge-
nesis and maturation of Purkinje neurons 3 [534]. Finally, pituicytes are glial
cells that reside in the posterior of the pituitary gland, where they participate
in the control of secretory events [434].
1 Macrophages are white blood cells that protect the body from harmful bacteria, particles
and dying cells by ingesting them in a process called phagocytosis.
2 The ventricular system of the brain is composed of four communicating ventricles filled
with cerebrospinal fluid that bathes and cushions the brain and communicates with the
central canal of the spinal chord.
3 Purkinje cells are very large neurons in the cerebellum that release the inhibitory neurotransmitter
gamma-aminobutyric acid (GABA) and are responsible for transmission of all
motor coordination signals.
3

1.1 Glial Cells in the Central Nervous System Introduction
The numerous short and long processes of astrocytes amount to a very large
exposed surface that makes these glial cells well suited for efficient exchange
of molecules and ions. This fact, put together with the intimate contacts that
astrocytes establish with neurons, capillaries and the CSF, puts them in the
unique position to control the environment in which neurons function [394].
Neuronal homeostasis is crucial due to the extreme sensitivity of neurons to
changes in concentrations of ions and neurotransmitters and the fact that these
changes can become significant at very low additional amounts of substance
given the very limited extracellular space 4 . Extracellular neurotransmitter
concentration must be kept very low at all times in order for the synaptic re-
lease to generate a significant change in neurotransmitter concentration [73].
Astrocytes also contribute to extracellular pH by removing CO2 [136] and
help control extracellular osmotic pressure by maintaining the water balance
of the brain [20]. Mechanisms for the control of extracellular osmolarity may
involve exchange of small neutral molecules like the amino acid taurin [384]
and special channels for transport of water called aquaporins that are present
on the membrane of astrocytic processes in contact with capillaries [540,545].
By surrounding capillaries with their processes and forming very extensive
tight junctions 5 between the endothelial cells, astrocytes decrease the perme-
ability of brain capillaries and help establish a fully functional blood-brain
barrier 6 [172,279,438].
Since their discovery, oligodendrocytes have been classified in many different
ways due to their inherent morphological heterogeneity including variable num-
ber of processes, thickness of myelin sheath, density of cytoplasm and clumping
of nuclear chromatin [44]. The function of most oligodendrocytes is to pro-
duce myelin sheaths to wrap around axons in order to isolate and quicken
transportation of the nerve impulse [73]. This contributes to the much higher
conduction speeds observed in myelinated axons versus unmyelinated ones,
with the thickest myelinated axons conducting at about 120 meters/sec and
the slowest unmyelinated ones at 1 meter/sec [44,73]. A myelin sheath consists
almost exclusively of numerous layers of oligodendrocyte or Schwann cell mem-
brane (lamellae) that in the process of wrapping around the axon squeeze away
4 Less than 20% of the total brain volume.
5 Type of junctional complexes present only in vertebrates that seal together the membranes
of two cells, effectively preventing passage of water and ions.
6 Unlike most organs in our body, the composition of the extracellular fluid of the brain
differs from that in the blood plasma.
4

1.2 Glioblastoma Multiforme Introduction
the cytoplasm, thus allowing the membrane layers to lie closely apposed [73].
Although the structure and function of the myelin sheath produced by oligo-
dendrocytes and Schwann cells is identical, one oligodendrocyte can extend its
processes to 50 axons, whilst each Schwann cell can only wrap its processes
around one axon [44,413]. Recent observations have suggested that oligoden-
drocytes and Schwann cells are also responsible for the long-term functional
integrity of axons [356]. Furthermore, there exists a pool of oligodendrocytes
named "satellite" that are found next to neuronal cell bodies where they can-
not serve an isolating role, but rather it is suggested they regulate neuronal
homeostasis in much a similar way as do astrocytes [44].
1.2 Glioblastoma multiforme 7
Primary brain tumours comprise a wide range of histological and pathologi-
cal entities, each with a distinct natural history [77,128]. For simplicity, CNS
tumours are classified as gliomas or non-gliomas. Gliomas affect the glial por-
tion of brain parenchyma and they are the most frequent and lethal cancers
affecting the CNS [383]. Diagnosis of gliomas is heavily based on the predom-
inantly affected cell type, which in turn is indicative of prognosis [128] (Table
1.1). Gliomas of astrocytic, oligodendroglial, oligoastrocytic 8 and ependymal
origin account for more than 70% of all brain tumours [383], with glioblas-
toma multiforme (GBM) being the most frequent (65%) and malignant of all
adult gliomas [154,300,365]. Grade I astrocytomas are well circumscribed and
curative surgical resections may be attempted. However, most gliomas are
characterised by diffuse infiltration of brain parenchyma, making surgical ex-
tirpation impossible [77]. Some lower-grade astrocytic tumours change their
identity over time and turn into more aggressive forms. Depending on the his-
tology of the tumour, patients undergo surgery, chemotherapy, radiotherapy
or a combination of these treatments, although the chances of cure are very
low [128].
In 1979 the World Health Organisation (WHO) published the first edition
of Histological Typing of Tumours of the Central Nervous System with the
7 The term glioblastoma is used synonymously with "glioblastoma multiforme".
8 Oligoastrocytic tumours are composed of roughly equivalent amounts of astrocyte-like
and oligodendrocyte-like cells.
5

1.2 Glioblastoma Multiforme Introduction
Table 1.1: Histological Types and Prognosis of Gliomas (y, years). Taken from
Doyle et al 2005 [128].
Tumour Cell Type Tumour WHO
grade
Incidence
Relative Survival
Rates (%)
1y 2y 3y
Astrocytic tumours Pilocytic astrocytoma l 10% of cerebral astrocytomas
most common
in children
95.7 94.3 91.3
Astrocytic tumours Diffuse astrocytoma II 10-15% of astrocytic
tumours
73.9 61.8 46.9
Astrocytic tumours Anaplastic astrocytoma III - 60.3 44.0 29.4
Astrocytic tumours Glioblastoma multiforme IV 50-60% of astrocy- 29.3 8.7 3.3
tomas; 12-15% of all
Oligodendroglial
tumours
Oligodendroglioma II
intracranial neoplasms
- 89.7 83.4 70.5
Ependymal
tumours
Ependymoma or
Anaplastic ependymoma
II-III - 87.6 81.4 70.6
aim of establishing a comprehensive four-tiered malignancy 9 grading guideline
for the evaluation of brain tumour progression. The WHO system is based on
the appearance of specific characteristics, such as atypia 10 , mitosis, endothelial
proliferation and necrosis. By reflecting the malignant potential of the tumour,
these features help assess the choice of therapies [301]. The four grades are
detailed as follows:
· Grade I lesions are considered low grade and include lesions with low
proliferative potential that are often cured through surgical resection
alone;
· Grade II lesions are also considered low grade, but they are infiltrative
in nature and often recur despite surgical resection;
· Grade III lesions carry histological evidence of malignancy, such as
atypia and fast mitotic activity, and are considered intermediate to high
grade lesions;
· Grade IV lesions are considered high grade and are cytologically ma-
lignant, mitotically active, necrosis-prone neoplasms often associated
with rapid pre- and post-operative disease evolution and a fatal out-
come [154,300,301].
The WHO classifies glioblastoma as a grade IV astrocytoma, accounting for
approximately 12-15% of all intracranial neoplasms and 60-75% of astrocytic
9 Medical term used to describe the state of tumours that are resistant to therapy, spread
rapidly and have a destructive clinical course.
10 Term that indicates a general cellular abnormality.
6

1.2 Glioblastoma Multiforme Introduction
tumours [154,301,383]. As the moniker "multiforme" implies, GBM is charac-
terised by a widespread intratumoural heterogeneity that makes it extremely
hard to understand and treat [154,383]. The histopathological features of GBM
include nuclear atypia, cellular pleomorphism, mitotic activity, vascular throm-
bosis, microvascular proliferation and necrosis [154,301]. This complexity, com-
bined with a putative cancer stem cell subpopulation and an incomplete atlas
of epigenetic and genetic lesions, has contributed to make this cancer one of
the most difficult to understand and to treat [154].
The average life expectancy of a patient with glioblastoma lies between several
weeks and several months after postoperative radiotherapy, with a protracted
course when treated with temozolomide in addition to radiotherapy alone (Fig
1.1). Median survival is generally less than one year from the time of diag-
nosis and most patients die within two years, with most long-term survivors
being given the wrong histological diagnosis at first [128,472]. Unless the
neoplasm has developed from a lower grade astrocytoma, in more than 50%
of cases the clinical history is less than 3 months. In most European and
North American countries, the incidence lies in the range of 3-4 new cases per
100,000 population per year, preferentially affecting adults and with a slightly
higher incidence in men than women [128,301]. Although infiltrative spread
Figure 1.1: Estimates of survival amongst GBM patients treated with radiotherapy
alone or radiotherapy with the alkylating agent temozolomide. Taken from Stupp et
al 2005 [472].
is a common feature of all diffuse astrocytic tumours, glioblastoma is particu-
larly notorious for its rapid invasion of the neighboring brain structures [154].
7

1.2 Glioblastoma Multiforme Introduction
Invading cells reside outside the contrast-enhancing rim of the tumour thereby
escaping surgical resection and evading radiotherapy. However, the generation
of metastases outside of the CNS remains very rare because the subarach-
noidal space 11 and CSF tend to remain unaffected [301]. The transforming
growth factor β (TGFβ) and Akt signaling pathways have been reported to
act as molecular mediators for glioblastoma invasion [245,527], as well as the
possibility of activation through hypoxia with the hypoxia-inducible factors
HIF1 12 [219]. An important aspect of glioblastoma invasion is the production
of a thick extracellular matrix to support migration and that of proteolytic
enzymes to enhance invasion across this matrix [301]. The current standard
for glioma patients involves resection of the tumour followed by extensive radi-
ation therapy and chemotherapy with the temozolomide alkylating agent - or
carmustine, a nitrosourea drug, in the United States - that grants the patient
a median survival of 15 months [472].
Methods and Technologies in Cancer Genomics The field of cancer
genomics has been evolving over the past decade at a very fast pace, bring-
ing in gargantuan amounts of fresh data to analyse from increasingly more
efficient technology platforms. A brief overview of the methods and technolo-
gies referred to in the rest of this chapter is given below. It should be noted
that this paragraph is by no means a comprehensive representation of all the
technologies currently deployed in the field of cancer genomics.
· Expression profiling determines the expression level of transcripts within
a cell using platforms that can be distinguished between array and non-
array. In the case of microarrays, the expression level is measured through
probe-transcript interaction for those transcripts that are represented on
the array. An a priori knowledge of what sequences to measure is nec-
essary in terms of the type of transcript (i.e. microRNAs, mRNAs), the
specific transcripts of interest for the correct probe design, and the re-
gions within that transcript (i.e. exons, introns, UTRs, single nucleotide
polymorphisms (SNPs), etc), making this platform inadequate for novel
gene discovery. Unlike microarrays, RNA sequencing measures expres-
sion levels by averaging the number of sequenced "reads", fragments of
11Interval between two membranes that protect the brain, the arachnoid membrane and
the pia mater.
12Transcription factors that respond to decreased availability of oxygen in the cellular
environment and are highly conserved transcriptional complexes of heterodimers constituted
by an α and a β subunit.
8

1.2 Glioblastoma Multiforme Introduction
up to several hundred base pairs, depending on the technology used, col-
lected along the entire length of the transcript. Potentially any RNA
population can be selected for sequencing but the mRNA and small non-
coding RNA population are the most commonly measured. Due to its
probe-free nature, sequencing is an appropriate platform for novel gene
discovery.
· Epigenetic profiling measures the amount of DNA methylation that oc-
curs at CpG islands in promoters using different types of assays on array
and non-array platforms. Many high-throughput profiling assays have
genomic DNA treated with a bisulfite conversion kit that converts un-
methylated cytosine residues to uracil and leaves methylated cytosine
residues unaffected. This treatment yields single nucleotide resolution
information about the methylation status of a segment of DNA and var-
ious analyses can be performed that depend on the platform used, to
retrieve this information.
· Exome sequencing is a selective method of DNA and RNA sequencing in
which samples are enriched for the coding regions of the genome, or the
"exome", and are successively sequenced. Targeted enrichment of the
exome may be performed by capture, using hybridization to microarrays
with probe sequences defined, for example, by the National Centre for
Biotechnology Information (NCBI) Consensus Coding Sequence (CCDS)
database [358], or by amplicon-based PCR amplification employing se-
quencing adaptors to amplify specific loci. As coding regions constitute
approximately 1% of the human genome [98], exome sequencing is a
potentially efficient strategy for the identification of rare functional mu-
tations, thereby detaining clinical relevance in the assessment of the role
of sequence variation in genetic disorders [56,98,358]. Although meth-
ods of enriching targeted genomic segments by hybridization have been
historically limited by the large amounts of genomic DNA required and
the modest throughput of the coupled sequence platforms, nowadays ad-
vances in hybridization specificity and sequencing technology have suc-
cessfully reduced the costs and increased the coverage of the exome se-
quencing process. However, cost effectiveness and completeness of the
information obtained are still key considerations [98].
· Array comparative genomic hybridization is a method for the detection
of copy number aberrations that uses an array platform containing thou-
9

1.3 Primary and Secondary Glioblastomas Introduction
sands of defined DNA probes that are hybridised to a sample mixture
containing DNA from tumour cells and DNA from a normal control, each
labeled with a different fluorescent dye. Abnormal regions in the genome
are then detected by calculating the ratio of the fluorescence intensity of
the hybridised sample to that of the reference DNA.
1.3 Primary and Secondary Glioblastomas
High-grade malignant gliomas are uniformly fatal despite the therapeutic ag-
gressiveness with which they are treated. As already mentioned, a classical
feature of these tumours is their widespread morphological and lineage het-
erogeneity. This plasticity is especially appreciable in gliomas presenting both
astrocytic and oligodendroglial histopathological features, the basis of which,
however, remains unknown [549].
Glioblastomas have been classified into two subtypes, primary or de novo
and secondary glioblastomas [154,301,383,549]. Primary glioblastomas rep-
resent the majority (more than 90%) of diagnosed cases and they develop
very rapidly without clinical or histopathological evidence of a pre-existing,
less malignant precursor lesion [154,301,334,450]. Secondary glioblastomas, on
the other hand, have a history of malignant progression from a lower-grade tu-
mour such as diffuse astrocytoma (WHO grade II) or anaplastic 13 astrocytoma
(WHO grade III), with more than 70% of WHO grade II gliomas transform-
ing into WHO grade III/IV diseases within five to 10 years of diagnosis [154].
Only 5% of glioblastomas are classified as secondary and they tend to pertain
to a younger cohort of patients (average age 45 years) compared to primary
glioblastoma patients (average age 60 years) [154,301,383]. Interestingly, de-
spite their distinct clinical histories, primary and secondary glioblastomas are
morphologically and clinically indistinguishable and their prognoses are equally
poor after having performed age-adjusted analysis [154,383,390].
Presently, the tools used to study glioblastoma are: primary tissue, genetically
modified mouse models (GEMMs), and glioma cell lines [308].
Primary Tissue The genetic events involved in the initiation and progres-
sion of glioblastoma are still unknown because of the limited availability of
early stage neoplastic tissue [308]. A number of studies today have focused
on large-scale sequencing of glioblastoma samples from different patients to
13 A cancer that is very poorly differentiated is called anaplastic.
10

1.3 Primary and Secondary Glioblastomas Introduction
attempt focusing on this problem. In the Parsons et al study from 2008 [383],
mutational data obtained from the exome sequencing of 22 human glioblas-
tomas was analysed for copy number aberrations (CNAs), and integrated to
identify glioblastoma candidate driver genes, i.e. genes that carry mutations
providing a selective advantage to the tumour cell. Interestingly, in 12% of the
glioblastoma patients mutations were found in the active site of the Isocitrate
dehydrogenase 1 (IDH1) gene on the long arm of chromosome 22, a gene never
previously associated with glioblastoma. IDH1 encodes an isocitrate dehydro-
genase, which catalyses the carboxylation of isocitrate to α-ketoglutarate and
nicotinamide adenine dinucleotide phosphate (NADPH), a coenzyme used as a
reducing agent in anabolic 14 biosynthetic reactions [225]. Five isocitrate dehy-
drogenase genes exist in humans and three are localised in the mitochondria,
while IDH1 is localised within the cytoplasm and peroxisomes. The func-
tion of IDH1 is to help release cellular stress from oxidative damage through
the generation of NADPH. Mutations in IDH1 were observed preferentially
in younger glioblastoma patients, on average 33 years of age, as opposed to
wild type carriers, on average 53 years of age, and most of them were found
in patients with secondary glioblastomas. These patients had a longer median
survival time of 3.8 years as compared to 1.1 years for patients with wild-type
IDH1. A similar pattern was also observed in the subgroup of young patients
with Tumour protein 53 (TP53) mutations. All mutations of IDH1 resulted
in an amino acid activating substitution at an evolutionary conserved residue
located within the enzyme’s binding site, reminiscent of known activating al-
terations in oncogenes such as BRAF, KRAS and PIK3CA [383].
The study by Watanabe et al [522] carried these results further by finding a
total of 130 IDH1 mutations involving amino acid 132 in 321 gliomas that
specifically affected 88% of the low-grade diffuse astrocytomas, 82% of the
secondary glioblastomas that developed through progression from low-grade
diffuse or anaplastic astrocytoma, 79% of the oligodendrogliomas and 94% of
the oligoastrocytomas. Interestingly, analyses of multiple biopsies from the
same patient showed that an IDH1 mutation never occurred after the acquisi-
tion of a TP53 mutation, suggesting that IDH1 mutation is a very early event
in gliomagenesis that may affect a common glial precursor cell population.
IDH1 mutations were co-present with TP53 mutations in 63% of low-grade
14 An anabolic reaction, as opposed to a catabolic one, is a metabolic pathway that constructs
molecules from smaller units and requires energy.
11

1.3 Primary and Secondary Glioblastomas Introduction
diffuse astrocytomas, but only 10% of pilocytic astrocytomas, 5% of primary
glioblastomas and none of the ependymomas. The frequent presence of IDH1
mutations in secondary glioblastomas and their near complete absence in pri-
mary glioblastomas reinforces the concept that, despite their histological sim-
ilarities, these subtypes are genetically and clinically distinct entities [522].
In an even larger study by Yan et al [535], the sequences of the IDH1 and
closely related IDH2 genes were determined in 445 tumours of the CNS and
494 tumours that did not affect the CNS. In corroboration of the Parsons et
al [383] and Watanabe et al [522] studies, mutations in the IDH1 gene were
found in more than 70% of WHO grade II and III astrocytomas and oligoden-
drogliomas, as well as in the glioblastomas that developed from these lower-
grade lesions. Each of these mutations affected amino acid 132 and reduced
the enzymatic activity of the encoded protein [535]. Interestingly, tumours
that did not carry mutations in the IDH1 gene often had mutations affect-
ing the analogous amino acid (R172) on the closely related IDH2 gene that
also reduced the enzymatic activity of the encoded protein, suggesting a form
of functional redundancy between the two genes. Similarly to the results by
Parsons et al [383], the tumours carrying IDH1 or IDH2 mutations showed dis-
tinctive genetic and clinical characteristics that resulted in better outcomes for
those patients with respect to the patients carrying wild-type IDH genes [535].
In a study by Zhao et al [548] the functional impact of the IDH1 mutation
was assessed in cultured glioma cells. By using the human cytosolic IDH1
crystal structure reported by Xu et al in 2004 [532], this study showed that
the tumour-derived IDH1 mutation impaired the affinity of the enzyme for its
substrate by forming catalytically inactive heterodimers. Interestingly, when
the expression of a mutant IDH1 was forced in cultured glioma cells, the for-
mation of α-ketoglutarate was greatly reduced but levels of the subunit α of
hypoxia-inducible factor 1 (HIF1A) were greatly increased. In fact, the tran-
scription factor HIF1A is regulated by the product of the reaction catalysed by
IDH1, α-ketoglutarate, and as a result, the IDH1 mutated human glioma cul-
tures displayed higher levels of HIF1A unlike wild-type IDH1 glioma cultures.
This was indicative that IDH1 may function as a tumour suppressor and, when
inactivated by mutation, may contribute to tumourigenesis through induction
of the HIF1 pathway [548]. In summary, the possibility of detecting an IDH1
mutation in a patient has the clinical potential, for a subpopulation of mostly
12

1.3 Primary and Secondary Glioblastomas Introduction
secondary and few primary glioblastoma patients, to hypothesise a protracted
clinical course [383]. New treatments could be designed to take advantage
of IDH1 alterations in these patients, especially since the inhibition of the
IDH2 enzyme has recently been shown to increase sensitivity of tumour cells
to chemotherapeutic agents [225].
The glioblastoma cancer genome was the first to be characterised in the con-
certed efforts of the Cancer Genome Atlas project (TCGA) [326]. The aim
was to "catalogue and discover major cancer-causing genome alterations in
large cohorts of human tumours through integrated multi-dimensional analy-
ses". The pilot project screened a total of 587 samples down to 206, which
were used to conduct genome-wide analysis of DNA copy number and gene
expression, and DNA methylation screening on a total of 2,305 assayed genes.
Of the 206 chosen biospecimens, 21 were post-treatment glioblastoma cases
and the remaining 185 represented predominantly primary glioblastomas. A
total of 91 samples and 601 genes, inclusive of 7932 exons, were chosen from
the 206 sample pool for re-sequencing towards mutational analysis. Although
the method of biospecimen selection ensured high-quality data, the stringency
of selection criteria may have introduced a degree of bias since small samples
and samples with high levels of necrosis were excluded [326].
In the TCGA study, upon a statistical analysis of mutation significance in the
91 matched glioblastoma-normal pairs selected for detection of somatic muta-
tions in 601 selected genes, eight genes were found to be significantly mutated:
TP53, PTEN, NF1, EGFR, ERBB2, RB1, PIK3R1 and PIK3CA. All the mu-
tations involving TP53 were clustered in its DNA-binding domain, a known
hotspot for TP53 mutations in human cancers. TP53 is an important tran-
scription factor and tumour suppressor involved in most cell survival pathways
and for this reason often referred to as "the guardian of the genome" [133].
Given that 27 of the 72 untreated samples and 11 of the 19 treated samples
harboured TP53 mutations and given that most of the 91 samples were pri-
mary glioblastomas, one can conclude that TP53 mutation is a common event
in primary glioblastoma [326]. Neurofibromin 1 (NF1) is a tumour suppressor
that when mutated in humans is responsible for the development of neurofi-
bromatosis type I and is associated with increased risk of optic gliomas, as-
trocytomas and glioblastomas [35]. The protein encoded by NF1 is a negative
regulator of the rat sarcoma (RAS) signaling pathway of small guanosine-5’-
13

1.3 Primary and Secondary Glioblastomas Introduction
triphosphate hydrolases (GTPase) that activate the mitogen activated protein
kinase (MAPK) signaling cascade amongst other pathways [496]. Overall, at
least 47 of the 206 samples harboured somatic NF1 inactivating mutations or
deletions, confirming the relevance of this gene in sporadic human glioblas-
toma [326].
Furthermore, within the TCGA dataset it was observed that mutation rates
between untreated and treated glioblastomas were markedly different, aver-
aging at 1.4 and 5.8 somatic silent mutations per sample, respectively. The
higher average in the treated samples was mostly due to the contributions
from a cohort of seven hypermutated tumours treated with temozolomide or
lomustine. The hypermutator phenotypes previously described in glioblas-
toma [79,204] were known to carry mutations in MutS homolog 6 (MSH6),
a component of the post-replicative DNA mismatch repair system (MMR).
MSH6 heterodimerizes with MSH2 to form MutSα, which binds to DNA mis-
matches thereby initiating DNA repair [6]. An analysis of the genes involved
in mismatch repair within the TCGA dataset uncovered that six out of the
seven hypermutated samples harbored mutations in at least one of the mis-
match repair genes MLH1, MSH2, MSH6 or PMS2 [326].
Recurrent focal alterations found in the TCGA samples that have already
been described and are common in glioblastoma are the amplification of Epi-
dermal growth factor receptor (EGFR), cyclin-dependent-kinase (CDK) CDK4
and CDK6, PDGFRA, MDM2, MDM4, MET, MYCN, CCND2 and PIK2CA
[104,255,262,292,374,423,429]. Interestingly, uncommon focal alterations were
also found, such as the amplification of the serine/threonine protein kinase
AKT3 and the homozygous deletions of NF1 and PARK2. V-akt murine thy-
moma viral oncogene homolog 3 (AKT3) belongs to the AKT family of ser-
ine/threonine kinases together with AKT1 and AKT2. The three AKT kinases
are now known to represent central nodes in a variety of signaling cascades that
regulate normal cellular process such as cell size and growth, proliferation,
survival, glucose metabolism, genome stability, and neo-vascularization. It is
currently less clear, however, whether AKT1, AKT2, and AKT3 are function-
ally redundant or whether each carries out a specific functional role [45,46].
Parkinson protein 2 (PARK2) encodes for a component of an E3 ubiquitin
ligase complex that mediates the targeting of substrate proteins for protea-
somal degradation. The functions carried by PARK2 are currently unknown,
14

1.3 Primary and Secondary Glioblastomas Introduction
but mutations in this gene are known to cause a familial form of Parkinson’s
disease known as autosomal recessive juvenile Parkinson disease [229]. The
most significant loss-of-heterozygosity (LOH 15 ) event identified in the TCGA
dataset was observed on the long arm of chromosome 17 where the TP53 gene
resides [326].
Cytosine-phosphate-Guanine (CpG 16 ) islands are regions of DNA in which a
cytosine is linked to a guanine via a phopsphodiester bond to form the dinu-
cleotide CpG that is repeated many times along a linear sequence. In formal
definitions, a CpG island must occupy more than 50% of a 200bp sequence
and have an expected/observed CpG ratio of 0.6 [102,159]. Methylation of the
5’ carbon of cytosine is a form of epigenetic modification that affects regula-
tion of gene expression via non-sequence based interactions. Such methylated
cytosines are present in the coding regions of mammalian genes and, over evo-
lutionary time, spontaneously deaminate to become thymines [102,287]. In
mammals, 70% to 80% of CpG cytosines are methylated [208]. Oppositely,
CpG islands in promoters tend to be unmethylated when genes are expressed,
suggesting that methylation is an inhibiting event for gene expression [142,214].
Methylation of CpG sites within promoters has been observed as a mechanism
of tumour suppressor gene silencing in a number of human cancers. In contrast,
the hypomethylation of CpG sites has been associated with the over-expression
of oncogenes within cancer cells [214].
Evaluation of promoter methylation was conducted in the TCGA project across
91 glioblastoma samples. A pattern emerged between the methylation of the
O-6-methylguanine-DNA methyltransferase (MGMT) promoter and the sub-
stitution spectrum of treated samples. MGMT is a DNA repair enzyme that
repairs damaged alkylated guanine residues and its promoter methylation sta-
tus has already been associated to sensitivity to the temozolomide alkylating
agent, which is the current standard of care for glioblastoma patients [79,204].
Amongst the 13 samples treated with alkylating agent that did not show
MGMT methylation, the validated somatic mutations from GC to AT, caused
by the spontaneous deamination of methylated cytosines to thymines, occurred
in comparable amounts between CpG and non-CpG dinucleotides. However,
in the six treated samples with MGMT methylation, the GC to AT transitions
15 The loss of normal function of one allele of a gene in which the other allele was already
inactivated. In the context of oncogenesis it refers to when the remaining functional allele
in a somatic cell of the offspring becomes inactivated by mutation.
16 The "CpG" notation is used to distinguish CG base-pairing of cytosine and guanine.
15

1.3 Primary and Secondary Glioblastomas Introduction
were found mostly in all non-CpG dinucleotides. This pattern is consistent
with the failure to repair alkylated guanine residues that is caused by treat-
ment if MGMT methylation is also shifting the mutation spectrum of treated
samples to a preponderance of GC to AT transitions at non-CpG sites [326].
The molecular mechanisms lying behind such pattern could find an expla-
nation in the interesting observation that the mutation spectra of mismatch
repair genes reflected MGMT promoter methylation as well. In fact, in treated
hypermutated samples with methylated MGMT, mismatch repair genes accu-
mulated GC to AT transitions in non-CpG islands, which was not observed
in any of the hypermutated tumours with unmethylated MGMT. Thus, mis-
match repair deficiency and MGMT methylation status together could have
powerful clinical implications in the context of treatment, raising the possibil-
ity that patients who initially respond to the frontline therapy in use today may
evolve not only treatment resistance, but also an MMR-defective hypermuta-
tor phenotype. The fact that newly diagnosed glioblastomas with methylated
MGMT respond well to treatment with alkylating agents, is in part due to
the initiation of many mismatch repair cycles that attempt to repair the alky-
lated guanines and in doing so lead to cell death, which is consistent with
the observation that the mismatch repair genes themselves are mutated with
CG to AT transitions at non-CpG sites [326]. Therefore, initial methylation
of MGMT in this scenario would have an effect on two fronts: shifting the
mutation spectrum that will affect mutations at mismatch repair genes, and
increasing the selective pressure to lose mismatch repair function, resulting
in aggressive recurrent tumours with a hypermutator phenotype [363]. These
findings highlight the importance of designing selective strategies that target
mismatch-repair-deficient cells in combination with alkylating agents, in order
to prevent or minimise the emergence of treatment resistance [326].
Genetically Engineered Mouse Models Murine gliomas that appear to
develop in the absence of lower grade precursors are very important disease
models because they reproduce de novo conditions for the onset of glioblas-
toma. GEMMs are useful research tools towards that end because they can
accurately reproduce the initiation and progression stages in the human pathol-
ogy upon introduction of few mutations, although it is still debated whether
they can accurately recreate the genomic and expression heterogeneity of the
original human disease [308]. In some cases these mouse models helped pre-
dict the importance in human gliomas of events such as TP53 and NF1 in-
16

1.3 Primary and Secondary Glioblastomas Introduction
activation [426,549]. Thus, GEMMs have been instrumental so far in the
molecular understanding of the causes of human gliomagenesis. Mutations in
Phosphatase and tensin homolog (PTEN), TP53 and Retinoblastoma 1 (RB1)
have all been tested in mouse models expressing Cre specifically in the brain
and their phenotypical effects have all been cell-type as well as developmental-
stage specific [100]. The use of the Cre-lox system allows specific recombination
events to occur in genomic DNA. The Cre protein is a site-specific DNA re-
combinase that catalyses the recombination of DNA between specific loxP sites
that have a directional core sequence between them. When cells express Cre,
the DNA is cut at both loxP sites and the result of the recombination depends
on whether the lox sites are located on the same chromosome and whether in
an inverted or direct repeat fashion. Same chromosome inverted lox sites pro-
duce an insertion, while direct repeats cause a deletion. Different chromosome
lox sites may cause translocation events [371].
Glioma Cell Lines Cancer cell lines have been the historical standard
both for exploring the biology of human tumours and as preclinical models
for screening of potential therapeutic agents [261]. Due to the inherent diffi-
culty in establishing and maintaining primary tumour cell cultures, established
cell lines have been traditionally used to characterize the genomic aberrations
identified in primary tumours [275]. However, it is possible that the genetic
aberrations accumulated by repeatedly passaging the cells in vitro may cause
their phenotypic characteristics to bear little resemblance with the primary
human tumour [66,261].
The laboratory of Howard Fine has attempted a systematic genomic survey
of five of the most commonly used glioma cell lines - A172, Hs683, T98G,
U251, and U87 - for the purpose of evaluating their similarity to primary
gliomas [275]. Their research, conducted with high-density Single Nucleotide
Polymorphism (SNP) arrays, showed that established glioma cell lines and pri-
mary tumour have significant differences in both genomic alterations and gene
expression, indicating that glioma cell lines may not be an accurate representa-
tion or model system for primary gliomas [275]. The differences observed in the
biological phenotype of glioma cell lines compared with primary tumours are
further confounded by the lack of molecular hallmarks in serially passaged cell
lines, such as the over-expression of EGFR, the silencing of cyclin-dependent
kinase inhibitor (CDKN) CDKN2A and the loss of PTEN [207,445]. This ex-
plains why in vitro and in vivo cancer cell line-based preclinical therapeutic
17

1.3 Primary and Secondary Glioblastomas Introduction
screening models have been poorly predictive of useful therapeutic agents and
may have led to important misinterpretations on the relevance of aberrant sig-
naling pathways within cell lines compared to primary tumours [261].
Nonetheless, cell lines don’t contain the typical mixture of genetically distinct
cells of primary tumours and can therefore be more easily characterised. In
fact, the problematic presence of non-tumour cells in primary tumours, makes
it harder to pinpoint the rare mutations that are not spread uniformly through-
out the tumour [275]. On the basis of this claim, cancer cell line sequencing
efforts are recently flourishing, including works on the commonly studied grade
IV glioma cell line U87MG [101].
With these pros and cons in mind, in the study by Lee et al [261] they went on
to search for a more biologically relevant model system for exploring glioma bi-
ology and for the screening of new therapeutic targets, which they found in neu-
ral stem (NS) cells. These cells have characteristics of continuous self-renewal,
extensive migration and infiltration of brain parenchyma and the potential for
full or partial differentiation, which are lost in glioma cell lines [261,286,347]
and will be discussed in greater detail in chapter three (see Section 3.3).
Classification Systems
Several decades of experimentation on glioblastoma have highlighted that
specific genetic lesions are more commonly observed in certain subclasses of
glioblastoma. Primary glioblastoma typically harbours mutations in the EGFR
receptor tyrosine kinase gene, tumour suppressor PTEN and cyclin inhibitor
CDKN2A, while secondary glioblastoma harbours mutations in Platelet-derived
growth factor (PDGF) and tumour suppressor TP53. However, the latter as-
sociation is now starting to be considered a historical one, since an increasing
number of studies are showing that TP53 mutations occur in a significant
amount of primary glioblastomas [383,549]. These alterations can become pre-
dictive of glioma subclasses. Glioblastomas with intact expression of the PTEN
and EGFR vIII 17 proteins, for example, correlate with increased EGFR kinase
inhibitor response as compared to tumours expressing EGFR vIII but lacking
PTEN [329].
Immunohistochemical markers have been important tools so far in the classifi-
cation and diagnosis of malignant gliomas, with Glial fibrillary acidic protein
(GFAP) and Oligodendrocyte lineage transcription factor 2 (OLIG2) being two
17 The vIII mutant of EGFR is the most common in glioblastoma and results from a
non-random 801bp in-frame deletion of exons 2 to 7 of the EGFR gene [371].
18

1.3 Primary and Secondary Glioblastomas Introduction
of the most specific ones [154]. GFAP is universally expressed in astrocytic
and ependymal tumours and OLIG2 is an oligodendroglial as well as stem cell
marker expressed at high levels only in diffuse gliomas [281,338,435]. Recently
investigated novel markers are stem and progenitor cell markers. Intensive
research efforts are attempting to uncover agents that may target subpopula-
tions of these cells with high tumourigenic potential and increased resistance
to current therapies [154]. The cell surface marker CD133 and other mark-
ers of stem cells, such as Nestin, Musashi and Sex determining region y-box
2 (SOX2), have been shown to negatively correlate with outcome parame-
ters [304].
In an attempt to optimise the association of different prognoses with differ-
ent therapies, several studies have focused their efforts in building an accurate
classification system [154,383,390]. Elucidating patterns between prognosis
and specific genetic lesions would allow therapies to tailor to the group of
patients who will most likely respond to them, an approach also known as
"stratification of treatment" [383]. Genome-wide profiling studies such as the
ones conducted by Freije et al in 2004 [148] Phillips et al in 2006 [390] and
Verhaak et al in 2010 [511], have tried to categorise glioblastoma in molecular
subclasses that could be predictive of survival outcomes. Thus, microarray
gene expression data for hundreds of high-grade glioma samples was analysed
and has shown that most tumours can be classified into a small number of
subtypes correlated with survival and response to therapy.
The largest such study to date [511] built a dataset from 200 GBM and two
normal brain samples that was used to identify four glioblastoma subtypes
named Proneural, Neural, Classical and Mesenchymal, each characterised by
a distinct gene expression signature encompassing a set of 210 up-regulated
genes. An independent set of 260 GBM expression profiles was compiled from
the public domain, including TCGA and Phillips et al [390], that successfully
assessed subtype reproducibility. The Proneural subtype was associated with
younger age, Platelet-derived growth factor receptor α (PDGFRA) abnormal-
ities, and IDH1 and TP53 mutations, all of which have previously been as-
sociated with secondary GBM and correlate with longer survival times [326].
In confirmation of this pattern, the Proneural subtype previously identified
in the study by Phillips et al also included most grade III gliomas and 75%
of lower grade gliomas [390]. The Classical subtype was strongly associated
with the astrocytic signature and contained all common genomic aberrations
19

1.3 Primary and Secondary Glioblastomas Introduction
observed in GBM, such as chromosome 7 amplifications, chromosome 10 dele-
tions, EGFR amplification, and deletion of the TP53-stabilising isoform of the
cyclin-dependent inhibitor CDKN2A:ARF [326]. As already observed in the
study by Phillips et al, the Mesenchymal subtype was characterised by high
expression of Chitinase 3-like 1 (CHI3L1) and Met proto-oncogene (MET) and
also a lack of association with a specific signature, but rather an equal cor-
relation with the neural, astrocytic, and oligodendrocytic gene signatures. A
striking characteristic of this class was the strong association with NF1 dele-
tion, already known to induce GBM in Nf1;p53 double knockout mice [426] and
shown to occur in a variety of tumours such as neurofibromas [550], but only
recently observed in human GBMs [69,320]. Since the Proneural subtype was
associated with a trend toward longer survival and the samples did not show a
survival advantage from aggressive treatment protocols, but a clear treatment
effect was observed in the Classical and Mesenchymal samples, the results of
this profiling-based classification study may find important roles in suggesting
different therapeutic strategies of high clinical impact. For example, extend-
ing the current biomarker assays for GBM to include subtyping tests for key
genetic events, including NF1 and PTEN loss, IDH1 and Phosphoinositide-3-
kinase (PI3K) mutation, PDGFRA and EGFR amplification [511].
Other studies attempting to define distinct subgroups of glioma identified CpG
island methylator phenotypes [363] and microRNA profiles [231] as part of
the same goal towards a new therapeutic approach. In the former study by
Noushmehr et al, promoter DNA methylation was assessed in 272 glioblas-
tomas from the TCGA dataset and validated in a different set of non-TCGA
glioblastomas and low-grade gliomas. Three DNA methylation clusters were
identified on array-based methylation assay platforms and one of these formed
a particularly tight cluster with a highly characteristic DNA methylation pro-
file designated as the "glioma CpG island methylator phenotype" or G-CIMP.
The G-CIMP sample cluster was highly enriched for the Proneural expression
profile defined by Verhaak et al [511] and the 24 G-CIMP-positive patients
were all significantly associated with IDH1 somatic mutations and a longer
survival time, making G-CIMP status a potential predictor of improved pa-
tient survival. The authors claim that if a transacting factor were involved in
the protection from methylation of the CpG island promoters in the G-CIMP
cluster, then the loss of its function could provide a favourable context for the
acquisition of specific genetic lesions such as IDH1 mutation. The two GBM
20

1.4 Pathways Involved in Glioblastoma Introduction
subgroups identified through the G-CIMP status would therefore have impor-
tant implications in the assessment of therapeutic strategies for different GBM
patients [363].
The microRNA profiling study by Kim et al [231] analysed 261 microRNA
expression profiles from TCGA, identifying five clinically and genetically dis-
tinct subclasses of glioblastoma that each related to a different neural precursor
cell type: radial glia, oligoneuronal precursors, neuronal precursors, neuroep-
ithelial/neural crest precursors and astrocyte precursors, suggesting a rela-
tionship between each subclass and a distinct stage of neural differentiation.
Interestingly, when compared to the glioblastoma subclasses identified by Ver-
haak et al [511], the microRNA-based oligoneural, radial glial, and astrocytic
subclasses were enriched in tumours from the Proneural, Classical, and Mes-
enchymal subtypes, respectively. MicroRNA-based consensus clustering also
yielded robust survival differences, with oligoneural glioblastoma patients liv-
ing significantly longer, and a distinct pattern of somatic mutations, with the
oligoneural subclass enriched for IDH1 and Phosphoinositide-3-kinase receptor
1 (PIK3R1) mutations but lacking NF1 mutations. A very high connectivity,
i.e. the number of directly connected mRNAs, was displayed by miR-9 and
miR-222 to the oligoneural and astrocytic precursor subtypes, respectively,
suggesting that these microRNAs might serve as core regulators of subclass-
specific gene expression in glioblastoma. Overall, this study provided strong
evidence that glioblastomas can arise from the transformation of neural precur-
sors at each of the stages represented by a microRNA-identified subclass and
that microRNAs are therefore useful for sub-classifying glioblastomas to gener-
ate accurate prognoses and for the development of molecular-based treatment
decisions [231].
1.4 Pathways Involved in Glioblastoma
Glioblastoma often involves the concurrent deregulation of three core path-
ways: RTK/PI3K/PTEN signaling, p53 signaling and Rb-mediated control of
cell cycle progression (Fig 1.2) [2,286,326,383]. Therefore, important genetic
events in human glioblastoma are the deregulation of growth factor signaling
pathways via amplification or mutational activation of receptor tyrosine ki-
nases (RTKs), the activation of the PI3K pathway and inactivation of the p53
and Rb tumour suppressor pathways [286,326]. The Rb and p53 pathways,
21

1.4 Pathways Involved in Glioblastoma Introduction
which regulate cell cycle primarily by governing the G1/S 18 phase transition,
are major targets of inactivating mutations in glioblastoma and their absence
renders tumours particularly susceptible to inappropriate cell division driven
by constitutively active mitogenic signaling effectors, such as PI3K and MAP
kinases [154]. The same three pathways were singled out in the TCGA project
when mapping the somatic nucleotide substitutions, homozygous deletions and
focal amplifications, onto the major pathways implicated in glioblastoma. A
statistical tendency was observed in which components within each pathway
were altered in a mutually exclusive fashion, hinting at deregulation of one com-
ponent relieving the selective pressure for additional ones in the same pathway.
Also, 74% of the samples harbored aberrations in all three pathways, suggest-
ing a functional non-redundancy between the three as a key requirement for
glioblastoma pathogenesis [326].
The study by Parsons et al in 2008 [383] detected critical genes within im-
portant cell signaling pathways in glioma: TP53, Mdm2 p53 binding protein
homolog (MDM2) and Mdm4 p53 binding protein homolog (MDM4) in the
p53 pathway; RB1, CDK4 and CDKN2A in the Rb pathway and PIK3CA,
PIK3R1, PTEN and IRS1 in the RTK/PI3K/PTEN pathway. All but one of
the cancers bearing mutations in members of one of these three pathways did
not show alterations in any of the other two, suggesting functional redundancy
for these mutations in tumourigenesis [383].
RTK/PI3K/PTEN Pathway
Tumour cells acquire genomic alterations that greatly reduce their dependence
on exogenous growth stimulation via transmembrane receptor contact with dif-
fusible growth factors, cell to cell adhesion, or extracellular matrix. Most often
these cells enable inappropriate cell division, survival, and motility through the
constitutive activation of the PI3K and MAPK pathways. The predominant
mechanism of mitogenic signaling activation for gliomas occurs through RTKs,
high-affinity cell surface receptors that bind and transduce the signal from cy-
tokines, growth factors and hormones, and integrins, membrane-bound recep-
tors that mediate the interaction between the extracellular matrix and the cy-
toskeleton [154]. Receptors that belong to the RTK family are EGFR, PDGFR,
MET, FGFR and VEGFR [431]. The epidermal growth factor EGF and the
platelet-derived growth factor PDGF pathways play important roles in both
18 Major checkpoint in the regulation of cell cycle beyond which the cell is committed to
dividing.
22

1.4 Pathways Involved in Glioblastoma Introduction
23
Figure 1.2: KEGG Glioma Pathway [2]. Oncogenes and tumour suppressors are highlighted in red.

1.4 Pathways Involved in Glioblastoma Introduction
CNS development and gliomagenesis [154]. The EGFR family is composed
of four structurally related members: EGFR, ERBB2, ERBB3, and ERBB4.
The importance of RTKs is made obvious by the fact that 58 out of the 90
unique tyrosine kinase genes in the human genome, encode for receptor tyro-
sine kinase proteins [431]. EGFR gene amplification occurs in roughly 40% of
all glioblastomas, and the amplified genes are frequently rearranged [154,262].
Alterations in this family were found in 41 of the 91 TCGA samples, includ-
ing the vIII mutant, extracellular domain point mutations and cytoplasmic
domain deletions [326]. The vIII EGFR mutant contains deletions of exons
two to seven and occurs in 20-30% of all human glioblastoma, making it the
most common EGFR mutant [154]. Although ERBB2 mutation was previously
reported in only one glioblastoma, seven samples of the 91 TCGA harbored
11 somatic ERBB2 mutations, including mostly missense 19 and one splice-site
mutation. Unlike in breast cancer, however, no amplifications were observed
for ERBB2 [326]. The PDGFR family of receptors contains two members,
PDGFRα and PDGFRβ, that homodimerize or heterodimerize depending on
which growth factor is bound to them [191]. PDGFRα and its ligands, PDGFA
and PDGFB, are expressed in gliomas, particularly in high-grade tumours,
while strong expression of PDGFRβ occurs in proliferating endothelial cells
in glioblastoma [154]. In contrast to EGFR, amplification or rearrangement
of PDGFR is much less common, and a relatively rare oncogenic deletion of
exons eight and nine, similarly to EGFR vIII, is constitutively active and en-
hances tumourigenicity [154]. Given the tumoural co-expression of PDGF and
PDGFR, autocrine and paracrine loops may be the primary means by which
this growth factor axis exerts its effects [154]. Although RTKs are known to
signal through both the MAPK and PI3K pathways, GEMMs showed con-
sistent high activation of just the PI3K pathway in high-grade astrocytomas,
hinting at the fact that downstream consequences of RTK activation may vary
greatly [100].
The PI3K complex belongs to class I of the lipid kinase family of phospho-
inositide 3-kinases and is composed of a catalytic subunit, PIK3CA, and a
regulatory subunit, PIK3R1. Activating missense mutations in the adaptor
binding and kinase domains of the catalytic subunit of class I PI3K complexes
are known to occur in different types of tumours, including glioblastoma, while
19 A missense mutation is a point mutation that results in a codon coding for a different
amino acid.
24

1.4 Pathways Involved in Glioblastoma Introduction
mutations in the regulatory subunits are less common [35,36,286]. Of the 91
samples from the TCGA study, nine carried somatic mutations in the regu-
latory subunit that clustered around the three amino acids acting as contact
points for the catalytic subunit, suggesting the possibility that these mutations
prevent inhibitory contacts between the two subunits and cause constitutive
PI3K activity [326]. The PI3K family of enzymes phosphorylates the inositol
ring of phosphatidylinositol 20 on the three, four and five hydroxyl groups in
many different combinations [157,440]. Class I PI3K complexes can be acti-
vated by small GTPases like RAS or RTKs [154].
Upon activation from RTKs, PI3K catalyses the phosphorylation of phos-
phatidylinositol (3,4)-bisphosphate to phosphatidylinositol (3,4,5)-trisphosphate
(PIP3) [154,267]. The generation of PIP3 in the cytosolic side of the cell mem-
brane acts as a docking site for the serine/threonine protein kinases of the
Akt family and the 3-phosphoinositide dependent protein kinase-1 (PDPK1).
Upon relocation, PDPK1 and mammalian target of rapamycin mTOR activate
AKT1 via phosphorylation of two key residues, starting the AKT-mediated sig-
naling cascade that promotes cell survival and proliferation [154]. AKT activa-
tion may be compromised via two other mechanisms in glioblastoma: dephos-
phorylation from the PH domain and leucine rich repeat protein phosphatase 1
(PHLPP) [74] or inhibition of phosphorylation through the C-terminal modu-
lator protein (CTMP) inhibitor [307]. One of the targets of AKT1 is phospho-
rylation of the Forkhead box O (FOXO) family of transcription factors, which
promotes their exclusion from the nucleus and reduces the expression of im-
portant target genes such as the cyclin-dependent kinase inhibitors CDKN1A
and CDKN1B, both also directly targeted by AKT1, and the Rb family mem-
ber p130 [154]. The action of class I PI3Ks is directly antagonized by PTEN
through dephosphorylation of PIP3 and inhibition of AKT1 relocation, which
strongly reduces AKT1-mediated cell cycle promotion [84,99,111].
Furthermore, in 86% of the TCGA samples, at least one genetic event was har-
boured in the RTK/PI3K/PTEN pathway as well as frequent deletions of the
PTEN lipid phosphatase gene. Within the RTK/PI3K/PTEN pathway, fre-
quent aberrations were shown in EGFR, ERBB2, PDGFRA and MET [326].
Patients with PTEN deletions or activating mutations in the catalytic and
regulatory subunits of class I PI3K complexes, might benefit from PI3K or
PDPK1 inhibitors, whereas patients in which the PI3K pathway is altered by
20 Negatively charged phospholipid and a minor component in the cytosolic side of eukary-
otic cell membranes.
25

1.4 Pathways Involved in Glioblastoma Introduction
AKT amplification might be refractory. Also, the co-amplification exhibited
by multiple RTKs in the same glioblastoma sample may be tailored with anti-
RTK therapies to specific patterns of RTK mutation [326].
PTEN is a major tumour suppressor that is inactivated in 50% of high-grade
gliomas by mutations or epigenetic mechanisms, each resulting in uncontrolled
PI3K signaling [154]. PTEN expression is completely extinguished in tumour
cells of hGFAP-Cre + ;p53 lox/lox ;Pten lox/+ GEMMs 21 with developed anaplastic
astrocytomas and glioblastomas, but it is present in the surrounding normal
tissue and vessels supplying tumour tissue. Such LOH effect is very frequent in
human high-grade gliomas [549]. PTEN is located in the long arm of chromo-
some 10 and acts as the central negative regulator of the PI3K/AKT pathway
due to its lipid phosphatase activity that affects RTK signaling [84,99,111].
As a consequence, the RTK/PI3K pathway is commonly affected by the bial-
lelic inactivation of PTEN or LOH of the long arm of chromosome 10. Loss
of PTEN most often results in constitutive activation of AKT1 but is not, in
mature astrocytes, sufficient to drive proliferation and initiate gliomagenesis
in the absence of other mutations. This suggests that the PI3K/AKT pathway
is not sufficiently stimulated by the absence of its main negative regulator to
elevate pathway activity in astrocytes [100]. Furthermore, PTEN may act to
suppress transformation and tumour progression beyond regulation of PI3K
signaling. In a study by Shen et al [453], quiescent cells from mouse model
cell systems 22 harboured high levels of nuclear PTEN, which appeared to fulfill
important roles in the maintenance of genomic integrity, through centromere
stabilisation and promotion of DNA repair [453]. In other studies, a number
of PTEN point mutations found in familial cancer predisposition syndromes
had no effect on enzyme activity and lied within important sequences for the
localisation of PTEN. Analysis of such mutants has confirmed that aberrant
sequestration of PTEN into either the nucleus or the cytoplasm compromises
its tumour suppressor function [117,154,495].
21 In the study by Zheng et al [549], the hGFAP-Cre transgene was used to delete p53 alone
or in combination with Pten in all CNS lineages using conditional p53 and Pten alleles, with
modelling efforts directed towards the Pten lox/+ genotype since broad CNS deletion of Pten
results in lethal hydrocephalus in early mouse postnatal life.
22 This mouse system included mouse embryonic fibroblasts and mouse embryonic stem
cells.
26

1.4 Pathways Involved in Glioblastoma Introduction
p53 Pathway
The tumour suppressor and transcription factor TP53 is the most commonly
mutated gene in human cancers and a master regulator of cell survival path-
ways, as the number of solely its protein interactors suggests (Fig 1.3). After
activation by cellular stresses, TP53 functions to trans-activate genes that
mediate cell cycle arrest, apoptosis, DNA repair, inhibition of angiogenesis
and metastasis, and other p53-dependent activities [185]. In humans, TP53
Figure 1.3: Visualisation generated from list of 345 interactors (orange) of TP53
(yellow) from the BioGRID 3.1 [62] repository for interaction datasets.
is located on the short arm of chromosome 17 and may be inactivated di-
rectly by gene mutations or indirectly by alterations in genes that promote
degradation of the TP53 protein [100,491,515]. TP53 signaling is commonly
affected by biallelic inactivation of TP53 and sometimes via amplification of
MDM2 or loss or mutation of the cyclin inhibitor CDKN2A:ARF [100]. While
MDM2 is a direct inhibitor of TP53 through its ubiquitin ligase activity,
CDKN2A has been identified in at least three alternatively spliced variants,
two of which encode isoforms inhibitors of the CDK4 kinase. The remain-
ing transcript, however, includes an alternate first exon that contains an al-
ternate open reading frame (ARF) specifying a protein that is structurally
unrelated to the products of the other variants and stabilises TP53 by se-
questering MDM2 [233,491,515]. CDKN2A is predominantly inactivated by
biallelic loss or hypermethylation in 50% to 70% of high-grade gliomas and
roughly 90% of cultured glioma cell lines. Concordantly, the chromosomal re-
gion containing MDM2 is amplified in roughly 10% of primary glioblastomas,
the majority of which contain intact TP53 [154]. Furthermore, the discovery
27

1.4 Pathways Involved in Glioblastoma Introduction
of the MDM2-related gene MDM4, which inhibits TP53 and enhances the ac-
tivity of MDM2, prompted the finding of 4% of glioblastomas with MDM4
amplification and no TP53 mutation nor MDM2 amplification [284]. Through
CDKN2A:ARF mediated stabilisation and activation, TP53 is able to acti-
vate a potent cyclin-dependent kinase inhibitor, CDKN1A. By binding and
inhibiting CDK4 and CDK6 complexes, CDKN1A acts as a regulator of the
G1 progression of cell cycle [160,233]. Another cyclin-dependent inhibitor of
the same family as CDKN2A is CDKN2C, which has recently been suggested
to drive glioblastoma pathogenesis. CDKN2C inhibits the formation of the
CDK4/CDK6 complex with cyclin dependent kinases, needed to keep the cell
cycle from stalling at the G1 phase. Homozygous deletions of CDKN2C were
reported in glioblastoma multiforme as well as missense mutations that dis-
turb its binding with CDK6. Suggestions based on these GBM studies are that
CDKN2C is a tumour suppressor that compensates for CDKN2A homozygous
deletion by being up-regulated through the action of the E2F1 transcription
factor [467].
In the TCGA project, inactivation of the p53 pathway occurred mostly in the
form of TP53 mutations, but also of CDKN2A:ARF deletions and MDM2 and
MDM4 amplifications. While genetic lesions in TP53 were mutually exclusive
of those in MDM2 or MDM4, CDKN2A:ARF deletions were concurrent to
TP53 mutations in 30% of the samples [326]. The best-characterised effector
of TP53 is the transcriptional target CDKN1A. Although this gene has not
been found to be altered in gliomas, its expression is frequently abrogated
by TP53 functional inactivity as well as by mitogenic signaling through the
PI3K and MAPK pathways [154]. Although TP53 mutation was historically
associated with low-grade gliomas and secondary human glioblastomas, works
done with GEMMs prompted the re-sequencing of both TP53 and PTEN to
re-evaluate their combinatorial role in the disease [549]. PTEN had already
been associated with primary glioblastoma [423] and the results of these works
showed that 60% of the clinically annotated human primary glioblastomas with
TP53 mutations also harboured a PTEN mutation or homozygous deletion,
indicating that TP53, together with PTEN, is also a key player in human
primary glioblastoma, as the TCGA data also reports [326,549].
28

1.4 Pathways Involved in Glioblastoma Introduction
Rb Pathway
The importance of the inactivation of the Rb pathway in glioma progression is
evidenced by the near-universal and mutually exclusive alteration of Rb path-
way effectors and inhibitors in both primary and secondary glioblastoma [154].
RB1 is located on the long arm of chromosome 13 and similarly to the p53
pathway, the Rb pathway is also affected by homozygous deletions of the
CDKN2A locus. The CDKN2A gene is, with the exception of its alternate
reading frame isoform CDKN2A:ARF, a negative regulator of p53 signaling
as well as a regulator of the G1 checkpoint in the Rb-mediated progression of
cell cycle [154,398]. The RB1 gene is mutated in roughly 25% of high-grade
astrocytomas and the loss of the long arm of chromosome 13 characterises
the transition from low to intermediate grade gliomas. Amplification of the
CDK4 gene accounts for the functional inactivation of RB1 in roughly 15% of
high-grade gliomas, and CDK6 is also amplified but at a lower frequency [154].
Deregulation of Rb signaling leading to G1/S progression appears to be a crit-
ical event in gliomagenesis whether or not inactivation of RB1 is an initiating
event [100]. Cell cycle progression is regulated by the activities of complexes of
cyclins and CDKs, which phosphorylate RB1 and block its growth-inhibitory
functions [318]. G1 progression is controlled by the D-type cyclins, which
form active complexes with CDK4 or CDK6, and E-type cyclins in associa-
tion with CDK2 (Fig 1.4) [398]. Within the TCGA dataset, 77% of samples
showed genetic alterations in the Rb pathway. Among these, the deletion of the
CDKN2A/CDKN2B locus on the short arm of chromosome nine was the most
common event, followed by amplification of the CDK4 locus [326]. CDKN2B
and CDKN2A lie adjacent in the short arm of chromosome nine and together
define a region that is frequently mutated and deleted in a wide variety of
tumours [233]. CDKN2A and CDKN2B both form complexes with CDK4,
CDK6 and cyclin D to block their activation and progression of cell cycle into
the G1/S phase [382,454]. Interestingly, all samples with RB1 nucleotide sub-
stitutions lacked CDKN2A/CDKN2B locus deletion, suggesting that this type
of RB1 inactivation obviates the genetic pressure for activation of upstream
cyclin and cyclin-dependent kinase complexes. Thus, it would be reasonable
to speculate that patients with deletions in CDKN2A or CDKN2B or with
amplifications in CDK4 or CDK6 could benefit from a treatment with CDK
inhibitors, unlikely to affect patients with RB1 mutation [326]. However, nu-
merous in vitro and in vivo assays have demonstrated that the neutralisation
29

1.5 Pathway Crosstalk Introduction
Figure 1.4: The Biocarta pathway for Rb signaling [61]. The cell cycle checkpoints
at the G1/S and G2/M transitions prevent progression when DNA is damaged. The
cyclin-dependent kinase CDK2 targets and phosphorylates RB1 to allow progression
to the G1/S transition. When the cell is in a quiescent state, hypophosphorylated
RB1 blocks proliferation by binding and sequestering the E2F family of transcription
factors, which prevents the transactivation of genes essential for progression through
the cell cycle [154]. Upon stimulation and activation of the MAPK cascade, cyclin D
forms complexes with cyclin-dependent kinases CDK4 and CDK6 and CDK2 is released
from the inhibitory interaction with CDKN1B and binds to cyclin E. Together
these activated complexes phosphorylate RB1, which stops inhibiting E2F transcription
factors so that the cell cycle can proceed through the G1/S checkpoint [68,151].
of this pathway alone is insufficient to abrogate cell cycle control to the ex-
tent needed for cellular transformation, suggesting that other important cell
cycle regulation pathways complement its activities in preventing gliomagene-
sis [154].
1.5 Pathway Crosstalk
While the RTK/PI3K/PTEN, p53, and Rb pathways are often considered as
distinct entities, there is significant crosstalk among them, reinforcing the in-
appropriate regulation of single pathway perturbations [154].
While 70% of secondary glioblastomas share the common event of IDH1 mu-
tation, which initiates them into the development of the higher-grade pathol-
ogy, primary glioblastomas arise de novo and lack a similar common initi-
ating event. In trying to assess this, the study by Chow et al looked into
the cooperativity of the three most important pathways in glioblastoma using
GEMMs [100]. Mutations in the three tumour suppressors Pten, p53 and Rb1
were introduced in various combinations in astrocytes and neural precursors
30

1.5 Pathway Crosstalk Introduction
and eventually developed into astrocytomas ranging from grade III to grade
IV [100,308]. Interestingly, none of the GEMMs carrying a deletion in only one
of the three tumour suppressors developed high-grade astrocytomas. Only the
deletion of p53 caused a late onset and low frequency of astrocytomas. The
earliest tumour onset was observed in triple knockout mice and the highest
frequency of astrocytomas in double knockout mice that carried a p53 dele-
tion. This observation supports a role for p53 inactivation in astrocytoma
initiation, alongside other factors such as the low frequency of RB1 and PTEN
mutations and high frequency of TP53 mutations in grade II human astro-
cytomas [300,526]. Furthermore, the earlier onset of high-grade gliomas in
Pten:p53 knockout mice shows that Pten cooperates more efficiently with p53
mutation than Rb1 mutation in double knockout mice. Interestingly, Pten and
Rb1 mutations together fail to cause gliomagenesis in GEMMs, indicating that
in the absence of other mutations these two pathways fail to cooperate [100].
Moreover, these pathways can negatively regulate each other by having TP53
inhibiting the activation of the FOXO transcription factors via the activation
of the Serum/glucocorticoid regulated kinase 1 (SGK1). Such kinase, in fact,
could induce through phosphorylation the translocation of FOXO transcription
factors out of the nucleus. This is turn would cause the inhibition of TP53 tran-
scriptional activity via a FOXO-mediated increase in the association of TP53
with the nuclear export receptors translocating it to the cytoplasm [154,541].
In the study by Chow et al, the array comparative genomics hybridization
(CGH) showed a subset of focal and large-scale genomic aberrations typical of
human glioblastoma subclasses. For example, p53:Pten tumours had acquired
the secondary amplifications of RTKs such as Pdgfra, Egfr and Met that are
considered hallmarks of the human pathology. Also, p53:Pten:Rb1 tumours
seemed to lose the selective pressure for RTK amplification, since only Pdgfra
was found to be amplified. The understanding of how tumour suppressor losses
induce secondary genomic alterations is a key to the direction of glioma pro-
gression [100,308]. Thus, Rb1 inactivation seems to further cooperate with
Pten and p53 deletions to generate high-grade gliomas with similar histologi-
cal and biochemical signatures as in Pten:p53 double knockout mice, but with
different selective pressure for RTK amplifications. Other genes affected by
CNAs aside from the RTKs mentioned above are Cdk4, Cdk6, Ccnd1, Ccnd2
and Ccnd3, which all directly regulate the G1/S cell cycle checkpoint and thus
Rb1 activity [68,100,382,454].
31

1.5 Pathway Crosstalk Introduction
The complicated interplay among these critical molecules highlights the need
for detailed dissection of the pathways that are aberrant in each tumour to
accurately guide the choice of combination therapies that can simultaneously
target multiple pathways [154].
32

Chapter 2
Neurogenesis
Contents
2.1 Radial Glia . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Neural Stem Cells . . . . . . . . . . . . . . . . . . . . . . . 37
2.1 Radial Glia
Early ultrastructural studies with electron microscopy revealed that, during
the development of the mammalian brain, newly born neurons used the ex-
tended bipolar processes of radial glia as a structural support and guide to
migrate to a new location [417]. However, the role of radial glia has recently
been extended to that of a progenitor population that can divide in the devel-
oping cortex and possibly the entire CNS, producing daughter cells including
neurons, astrocytes and glia [400]. Historically, radial glia were believed to co-
exist with neural progenitors in the ventricular zone of the brain, but the past
decade has seen an increase in the amount of experimental data challenging
this theory, with radial glia isolated in vitro displaying neuronal differentiation
capacity [310], and dividing precursors in the developing cortex displaying a
radial glia phenotype, as revealed by morphology and markers such as Brain
lipid binding protein (BLBP) and Vimentin [187,361]. After cortical neuroge-
nesis is complete, radial glia retract their processes and convert to multi-polar
astrocytes, leaving specialized forms of radial glia persisting in the adult CNS
in locations such as the cerebellum, retina and adult hippocampus [343]. In
mammalian species, for example, radial glia persist into adulthood in the form
of Bergmann glia in the cerebellum and Müller glia in the retina [400]. In
non-mammalian vertebrates the number of radial glia that persist through-
33

2.1 Radial Glia Introduction
out adulthood is much higher and more widely spread across the CNS, which
might account for the considerable regenerative capacity seen in such species
with respect to that in mammalian species [552].
During the development of the mammalian CNS, the steps that precede the
appearance of the radial glia progenitor population are the following:
1. Commitment of an early population of cells to the neural lineage [400];
2. Induction of neuroectoderm, promoted by the absence of Bone morpho-
genetic protein (BMP) and Fibroblast growth factor 2 (FGF2), and com-
posed of a progenitor population termed "neuroepithelial progenitors"
(NEP) cells. Although continuous self-renewal has not been demon-
strated for NEP cells, all cells of the CNS directly or indirectly derive
from them [400];
3. NEP cells undergo a process of interkinetic nuclear migration, in which
the nucleus oscillates between the apical and basal cellular membrane
in synchrony with cell cycle progression, leading to the formation of a
pseudo-stratified epithelium of which Sox1 is the earliest marker [389];
4. The neural plate defined by the action of NEP cells undergoes a mor-
phogenetic movement that results in the formation of the neural tube;
5. Signaling molecules (i.e. retinoic acid (RA), BMP, notochord-derived
Sonic hedgehog (SHH)) are secreted from the nearby tissues to establish a
positional gradient that defines sub-regions of the CNS, in which distinct
neuronal and glial subtypes are specified [72];
6. At approximately embryonic day 9.5-10.5 in mouse, a second morpho-
logically distinct cell type appears, termed "radial glia" [400].
Structural features of radial glia that distinguish them from their earlier NEP
progenitors are the bipolar morphology, with one extension at the luminal sur-
face 23 of the ventricular zone (VZ) and a longer process extending in the op-
posite direction through to the basement membrane adjacent to the pia mater
24 [193] (Fig 2.1); the ovoid cell body, where the nucleus is positioned in the
23This term indicates the surface that looks into the space defined by the interior of a
tubular structure.
24The CNS is enclosed in three connective tissue membranes called meninges, one of them
being the pia mater that follows the surface of the brain and spinal cord closely, extending
into all sulci and depressions of the surface [73].
34

2.1 Radial Glia Introduction
VZ adjacent to the lumen; electron lucent processes; abundant intermediate
filaments 25 ; numerous glycogen granules condensed at the end of the process
closest to the luminal surface. A second relevant NEP-distinguishing feature of
radial glia is the expression of markers characteristic of the astrocytic lineage,
such as the Glutamate aspartate transporter (GLAST), BLBP and GFAP, as
well as the consistent immunoreactivity shown towards the Nestin and Vi-
mentin antibodies [122,310,400]. In figure 2.1, dividing NEP cells appear in
blue and populate the VZ and the sub-ventricular zone (SVZ), while mature
migrated neurons are highlighted in yellow and populate the stratum closest to
the pia mater. Radial glia are shown in green and their process extends with
a bipolar morphology across from the VZ to the pia mater to support mature
neuronal migration in the developing cortex. The stratification of the NEP
cell population in the VZ is concordant with cell cycle phase (G1, G1/G2, M),
while the more superficial SVZ displays continued mitotic activity but does not
host a similar cellular stratification. Therefore, the mature cortex eventually
displays an inside-out pattern of layering, with the early-born neurons residing
in the deeper layers and the late-born ones residing more superficially next to
the pia mater [193].
Figure 2.1: Cross-section through the neural tube with morphological zones indicated
on the left. NEP cells appear in blue, mature migrated neurons are highlighted
in yellow and radial glia are shown in green. Adapted from Herrup et al 2007 [193].
25 Components of the cytoskeletal system that are distinguishable from microfilaments by
the size of their diameter, 8-12 nanometers. They function as a tension-bearing element to
help maintain cell shape and rigidity as well as anchor in place several organelles.
35

2.1 Radial Glia Introduction
Radial glia are by no means a uniform cell population. In fact, they are
found within the cortex and also throughout the developing brain and spinal
cord. This spatial and temporal heterogeneity likely generates the diversity of
cellular phenotypes within the nervous system. For example, region-specific
expression of transcription factors in radial glia is likely to determine the fate
of progeny towards one of the lineages of the CNS [249]. Furthermore, there
are circumstances in which the radial glia phenotype is reacquired, such as af-
ter injury, during reprogramming and dedifferentiation in vitro, and following
epigenetic disruptions in tumorigenesis [400].
Anatomically, a ventricular system is present within the cerebrum 26 that is
composed of four communicating compartments, or ventricles, filled by CSF
[387]. The SVZ is a paired brain structure situated in the lining of the two
lateral ventricles and is one of the major germinal layers during embryogene-
sis [18,41] together with the sub-granular zone (SGZ 27 ), as well as the largest
district in which NS cells with the characteristics of astrocytes persist after
birth in the mammalian adult brain [18,332]. Several studies have demon-
strated that radial glia not only give rise to multiple classes of brain cells, but
also generate adult SVZ stem cells that maintain the neurogenic lineage in the
adult brain [293,332,413], with a similar relationship proposed also between
radial glia and hippocampal progenitors [223,449]. Specifically, these adult
SVZ stem cells have been shown to arise from a subpopulation of radial glia
present within the developing striatum and display characteristics intermedi-
ate between normal astrocytes and radial glia, hinting that NEP cells, radial
glia and adult SVZ stem cells are the components of a continuous lineage with
multipotent neural differentiation potential [107,332]. Although these in vitro
studies have demonstrated the shared molecular and morphological charac-
teristics between radial glia and adult SVZ stem cells (Fig 2.2) [107], several
aspects of in vivo biology cannot be accounted for. For example, the fact
that radial glia exist only transiently during fetal development makes it harder
experimentally to validate whether they function as self-renewing stem cells.
Also, the artificial environment of cultures may result in a unique synthetic cell
state, and the combination of transcription factors expressed in cultured SVZ
stem cells is not found in vivo. Therefore, it is fairest to term these precursor
26Largest structure in the mammalian and human brain composed of the white and grey
matter in the cranial cavity.
27Adult mammalian neural stem cells have also been isolated from the sub-granular zone
(SGZ) of the dentate gyrus in the hippocampus, and the subcortical white matter [442].
36

2.2 Neural Stem Cells Introduction
cells "radial glia-like NS cells", although in the rest of this thesis they will
be referred to simply as NS cells [400]. Radial glia can be obtained from the
Figure 2.2: Surface markers of radial glia are expressed by NS cell lines, indicating
that these cells may provide the biological context to work with progenitors of
the CNS. For example, the GFAP is a type III Intermediate filament; GLAST is
an astrocyte-specific glutamate transporter; Prominin, also known as CD133, is a
glycoprotein that is a neural and hematopoietic stem cell marker. Adapted from
Conti et al 2005 [107].
dissociation of fetal CNS tissues and the subsequent establishment of primary
cultures. The heterogeneity linked to these primary cultures was overcome by
the development of cell type specific monoclonal antibodies like RC1, which
reduces the presence of multiple immature and differentiated cell types and
distinct radial glia subtypes. Fluorescent activated cell sorting (FACS) can
also be used with cell surface markers like CD15 and CD133, which, however,
enable only the enrichment and not the isolation of radial glia subpopula-
tions. Another technique uses reporter mice in which an endogenous radial
glia promoter drives the expression of fluorescent reporters, and cells with ac-
tivated reporter expression are then isolated using FACS. Although primary
cell cultures are a useful tool to isolate and characterise radial glia cells iso-
lated directly from neural tissues, the mitogen-driven expansion of cells in vitro
leads to the formation of NS cell lines that are an invaluable tool for molec-
ular and biochemical studies on nervous system pathological models amongst
others [400].
2.2 Neural Stem Cells
NS cells are defined as clonogenic cells capable of self-renewal and multipotent
differentiation into the three main cell types of the CNS: neurons, astrocytes
37

2.2 Neural Stem Cells Introduction
and oligodendrocytes. NS cells don’t express the pluripotent specific transcrip-
tion factors Oct-4 and Nanog, but rather show expression of neural genes and
lack the expression of mesoderm and endoderm specific genes. As shown in
figure 2.3, pure NS cells can be derived from ES cells taken from the inner cell
mass (ICM) of the blastocyst or from the SVZ and germinative area of adult
and fetal brain tissue, respectively [400,403]. NS cells can be isolated from the
Figure 2.3: Sources of NS cells: (a) cultured indirectly starting from ES cells
derived from the ICM of the blastocyst; (b) cultured directly from the dissociation
of germinative areas of the fetal brain or SVZ of the adult brain. Adapted from
Pollard et al 2007 [400].
embryonic or adult mammalian brain of mice [297,427,525], primates [171] and
humans [135,145,149,436,443,512], although their precise purification remains
elusive since they cannot yet be unambiguously identified with markers. Until
recently, NS cells were mainly defined by the expression of Nestin, a cyto-
plasmic intermediate filament protein discovered by Hockfield and McKay in
1985 [195], although it is now clear that Nestin identifies neural progenitors as
well as stem cells. Nestin expression is lost in vitro with differentiation of NS
cells and, in vivo, is retained postnatally only in proliferative zones. Direct iso-
lation of NS cells from human fetal brain using flow cytometry for the cell sur-
face marker Prominin-1 (CD133) was reported by Uchida et al. in 2000 [502].
CD133 was originally shown to be a hematopoietic stem cell marker, but it is
also expressed in the SVZ of developing mice and humans on the apical mem-
brane of cells lining the lateral ventricles, with more restricted expression than
Nestin. In vitro NS cell-like cells with a marked stem cell activity have been
isolated from human fetal brain cells expressing CD133 [122]. In vivo, NS cells
have been identified as radial glial cells, expressing markers BLBP, GLAST
38

2.2 Neural Stem Cells Introduction
and RC2, an antibody recognising the presence of Nestin [310]. Until the late
1990s, the only cell line that could consistently generate human neuronal cells
in vitro was NTERA-2, a teratocarcinoma 28 derived cell line that required the
performance of complex manipulations to induce differentiation [22,397]. In
1997, Sah et al. established the first immortalized adherent human fetal neu-
ral precursor cell line using retrovirally expressed avian v-myc [441] that led
to subsequent independent reports using similar strategies [114,145,513], until
the possibility of expanding human fetal neural precursors in suspension cul-
tures was explored [83,428,483]). The floating aggregates of cells were termed
"neurospheres" [427] and only recently adherent monoculture protocols have
been developed as an alternative.
Neurospheres With the exception of ES cells, it has always proven difficult
to obtain homogeneous propagation of stem cell cultures ex vivo since they tend
to be accompanied by differentiation. In 1992 Weiss and Reynolds discovered
that cells from fetal mouse CNS could be propagated in suspension culture
with EGF, as a cluster of floating cells that they termed "neurospheres" (Fig
2.4). A neurosphere represents a clonal single cell-derived floating cluster of
Figure 2.4: (a,b) Contrast microscopy images of early phase neurosphere formation,
in which individual cells form small clusters. (c,d) Immunofluorescence microscopy
images in which (c) EGFR (green) and the Nestin protein (red) are detected on an
intact neurosphere, and (d) nuclei (DAPI staining, blue) and cell mitosis (5-bromo-
2’-deoxyuridine (BrdU) incorporation, green) are detected on a frozen neurosphere
section. Image adapted from [375].
proliferating cells [427] that contains thousands of cells and is a mixture of
stem and progenitor cells, with only up to 5% stem cells [122]. The number of
stem cells in a neurosphere is evaluated in a clonogenic assay by determining
28 A germ cell tumor.
39

2.2 Neural Stem Cells Introduction
the number of singly dissociated cells from primary spheres that can give rise
to secondary spheres that, in turn, can differentiate down all three lineages of
the CNS (Fig 2.5) [494]. Therefore, the re-plating efficiency of neurospheres
is a measure of the number of stem cells, and the size of the sphere reflects
progenitor proliferative efficiency. This assay also allows for a quantitation of
self-renewal, which is distinct from proliferation in that self-renewal involves a
cell division with a cell fate decision, so that at least one daughter cell retains
the full stem cell potential of the parent cell (see Fig 3.2). A multipotent
secondary sphere can only form from a stem cell. Undifferentiated neurospheres
can be extensively passaged in suspension, but when plated onto an adherent
substrate in serum, they differentiate into the three main neural lineages of
the CNS: neurons, astrocytes and oligodendrocytes [122].
In their 1992 neurosphere assay Weiss and Reynolds [427] employed a serum-
free culture system, whereby the majority of primary differentiated CNS cells
did not survive but a small population of EGF-responsive cells were maintained
in an undifferentiated state and proliferated to form clusters, that could then
be dissociated to form numerous secondary spheres or induced to differentiate
into the three major cell types of the CNS. After approximately seven days
in growth medium containing EGF, the neurospheres isolated measured 100-
200µm in diameter, were composed of 3,000-5,000 cells, and differentiated into
the three primary CNS phenotypes when, as intact clusters or dissociated cells,
they were plated without growth factors on an adhesive substrate (Figure 2.5).
Over the past decade the use of the neurosphere assay has demonstrated that
a population of cells existed in the fetal through to the adult mammalian CNS
that could be isolated in culture, and exhibited the critical stem cell attributes
of proliferation, self-renewal, and the ability to give rise to a number of dif-
ferentiated, functional progeny [116]. The neurosphere assay has proven to be
an excellent technique to isolate NS cells and progenitor cells to investigate
the differentiation and potential of cell lineages. These spheres can be dissoci-
ated, expanded and pooled in sufficient quantity for scientific inquiry, and lend
themselves easily to sectioning for histology or immunocytochemistry applica-
tions, and being cryopreserved [375]. Moreover, the recent finding that human
brain tumours can be similarly cultured in neurosphere conditions generates
the opportunity of understanding the stem cell hierarchy of these neoplasms
(see Section 3.2) [122]. Neurospheres contain a mix of differently committed
cells including radial glia, committed progenitors and differentiated astrocytes
40

2.2 Neural Stem Cells Introduction
Figure 2.5: The neurosphere assay used to study neural precursor cells in culture.
Cells are first isolated from embryonic or adult brain, then cultured in serum-free
conditions in the presence of EGF and FGF2 to generate floating colonies. The
primary neurospheres can then be dissociated and re-plated in EGF and FGF2 to
generate secondary neurospheres that can then be made to differentiate in the three
primary lineages of the CNS by subtracting the growth factors in adherent conditions.
Adapted from Dirks et al 2008 [122].
and neurons, that, upon removal of EGF, differentiate into the three main
lineages of the CNS with a strong preference towards astrocytes [107]. This
heterogeneity likely provides a niche that sustains 3-4% of stem cells, raising
the question as to whether it is the multipotent cells or the more differenti-
ated ones within the mixed cellular environment, that give rise to the three
lineages [107,400]. In vivo, the external signals such as secreted factors and
cell-cell interactions mediated by integral membrane proteins and the extracel-
lular matrix, control stem cell fate collectively, defining the stem cell "niche",
which has a powerful effect in maintaining the balance between quiescence,
self-renewal and cell fate commitment [165]. In vitro, the neurosphere prob-
ably provides the right mixture of cellular environments that resembles the
complex niche, sustaining NS cells in the mammalian brain. Therefore, the
neurosphere discovery is invaluable in that it demonstrates the potential in
the developing and adult CNS of rodents and primates, to give rise to stem
cells [107].
41

2.2 Neural Stem Cells Introduction
A proof-of-principle experiment performed by Conti et al [107] demonstrated
the presence of NS cells within neurospheres, and consisted in allowing passage
40 mouse neurospheres derived from fetal forebrain to attach to gelatin-coated
plastic in the presence of EGF and FGF2. Since under these conditions Conti et
al had already proven the generation of NS cells in adherent monoculture (see
Section 2.2), the appearance of bipolar cells that were indistinguishable from
NS cells and could be serially propagated as uniform RC2 + /GFAP - populations
and then induced to differentiate into astrocytes or neurons, concluded that
radial glia-like cells present in neurospheres give rise to NS cells in adherent
culture in the presence of FGF2 and EGF. Conversely, they observed that
NS cells of either ES cell or fetal brain origin readily formed neurospheres if
detached from the substratum mechanically or due to overgrowth, confirming
that NS cells and thus radial glia are likely the neurosphere forming stem cells,
although in neurospheres they constitute only a fraction of the cell population.
Analogously to the embryoid body (EB) differentiation observed in ES cell
aggregates, the differentiation observed within neurospheres is presumably due
to aggregation [125].
An important limitation of the neurosphere culture system is that, when used
to screen compounds that affect NS cell expansion, human NS cells expand
more slowly in suspension culture in vitro than do their mouse counterparts,
which makes quantification of cell proliferation harder due to variable cell
death. A second important limitation of neurosphere assays is that it is dif-
ficult to identify the precise cellular target due to the presence of restricted
progenitors and differentiated cell types, and real-time monitoring of cellu-
lar responses is not possible in aggregates. Finally, fusion of neurospheres is
a common occurrence in suspension, which confounds analyses based solely
on sphere numbers or size [404]. To summarise, the neurosphere paradigm
is invaluable in that it has demonstrated the existence of progenitors within
cultured tissues, but it is accompanied by several important shortcomings:
· the cellular complexity created by the mixed environment is a barrier for
dissecting the mechanisms responsible for the self-renewal and commit-
ment processes;
· the heterogeneity of the cellular population pollutes global expression
profiling experiments and makes it hard to identify a precise cellular tar-
get;
42

2.2 Neural Stem Cells Introduction
· neurospheres differentiate more promptly into astrocytes rather than
neurons both in vitro and when transplanted into mice;
· human NS cells cannot be properly screened for compounds affecting
them because quantification is made difficult by their slow growth in
suspension culture and their variable death rate;
· cellular responses cannot be monitored within aggregates and fusion of
neurospheres can confound results based on their number and size.
Therefore, a niche-independent environment is better suited for the growth of
stem cell cultures, in that differentiation towards a specific lineage can always
be traced back to the stem cells themselves [165]. Such an environment was
produced when the presence of FGF2 and EGF alone was discovered to be
sufficient for the continuous expansion of NS cells in adherent conditions [107].
Niche-independent NS cell derivation
The key to proper usage of ES cell-based technologies is the development of
robust, reproducible and reliable protocols for controlling propagation and dif-
ferentiation of cells, and an important goal in embryonic stem cell biology over
the past 10 years has been that of developing protocols to enable the conver-
sion of mouse and human ES cells to the neural lineage [400].
The default model of neural induction proposes that the key event is the re-
moval of BMP signaling with no positive induction required [352]. In mouse
ES cells, however, positive induction is necessary for differentiation to take
place, since these cells can be maintained in vitro through the addition of the
Leukemia inhibitory factor (LIF) and BMP extrinsic factors, as well as in-
trinsic determinants Sox2, Oct-4 and Nanog transcription factors, but require
replacement of LIF and BMP to specify the direction of differentiation. It
was initially thought that exposure to RA and serum 29 in suspension culture
was required, provided LIF retraction, to generate neurons [38,471], although
later reports showed RA was unnecessary and neural precursors could be en-
riched in a serum-free basal media [367]. During neural differentiation, ES
cells are believed to undergo progressive lineage restrictions similar to those
observed during normal fetal development, providing a means to isolate dis-
tinct neural precursor populations such as NEP cells and radial glia, as well as
29 Animal derived fluid most commonly drawn from a bovine fetus that contains hormones
and growth factors that allow cells in culture to proliferate.
43

2.2 Neural Stem Cells Introduction
a platform to study the molecular events involved in the transitions between
precursors [400]. Based on the view that neural differentiation of ES cells in
vitro does recapitulate neural development in vivo, several studies isolated an
RC2 immunoreactive radial glia-like cell as the transient neural progenitor in-
volved in the transition from NEP cells to neuronal and glial subtypes (Fig
2.6) [302,399].
Figure 2.6: Diagram to visualise the progressive lineage restriction of ES cells
differentiating toward the neural phenotype in neurospheres, showing the transition
of NEP cells to RC2 immunoreactive radial glia-like cells. Adapted from Pollard et
al 2007 [400].
Likely due to paracrine LIF signaling, many ES cell neural differentiation pro-
tocols have the drawback of fostering the generation of non-homogeneous cul-
tures that include contaminating populations of non-neural cells and residual
ES cells [400]. This effect can be overcome by adopting a "lineage selection"
strategy, in which a reporter gene or drug resistance gene is expressed as a
transgene 30 under cell type specific promoter elements, such as the Sox1-GFP
reporter construct for the isolation of NEP cells [32]. ES cell-derived cultures
engineered to express the Sox1-GFP reporter are initially enriched in Sox1 +
NEP cells but quickly differentiate to neurons and glia due to the niche environ-
ment recreated in neurospheres that mimics in vivo developmental cues [400].
In order to isolate cells capable of undergoing symmetrical stem cell divisions
without differentiation, a niche-independent protocol was devised by Conti et
al [107] that uses adherent conditions to ensure homogeneity of the stem cell
population bypassing the formation of neurospheres. The unique characteris-
tic of this protocol is the use of EGF in attached monolayer culture, which
30 A gene that does not belong to the wild type genome sequence but can be introduced
from an another organism naturally or by means of genetic engineering.
44

2.2 Neural Stem Cells Introduction
had been previously only used in the culturing of neurospheres. The proto-
col was successfully applied to the production of non-immortalized NS cells
in adherent monolayer from mouse and human ES cells, as well as mouse and
human fetal brain [107,481], and later adapted for derivation from adult mouse
brain [403], with the most recent optimisation made for drug screening appli-
cations from different regions of the fetal human CNS [198]. Other protocols
that describe adherent monolayer culturing of NS cells have been explored by
other researchers, but their approaches are all somewhat different in that they
either generate immortalised cell lines [405], are tailored to derivation from the
SGZ [34], or generate non-immortalized cell lines but have limited the charac-
terisation to primary cultures without demonstrating long-term stability and
tripotent differentiation capacity [376,536].
In the protocol described by Conti et al, prior to initiating differentiation,
mouse ES cells are plated at relatively high density and cultured for 24 hours
in standard ES cell medium containing LIF. To start monolayer differenti-
ation, undifferentiated ES cells are dissociated and resuspended directly in
N2B27 medium, a mixed formulation of basal media and supplements includ-
ing insulin and lacking LIF. Under this monoculture condition, ES cells lose
pluripotent status and predominantly commit to a neural fate [539]. The cul-
ture is re-plated after seven days in basal medium with FGF2 alone or FGF2
and EGF, during which time residual undifferentiated ES cells are eliminated
because the NS-A component of basal medium does not allow for ES cell prop-
agation. However, since in this media neural precursors associated into floating
clusters, a lineage selection strategy was adopted (Fig 2.7). Thus, after neural
commitment was induced in monolayer in Sox1-GFP reporter ES cell lines, the
neural precursors were maintained adherent in N2B27 medium, and the undif-
ferentiated ES cell and non-neural differentiation products were eliminated via
addition of puromycin. In these reporter cell lines, Sox1 is linked via an inter-
nal ribosome entry site (IRES) to a gene conferring puromycin resistance (Fig
2.8), so that the transition to NEP cells is marked by GFP fluorescence and the
selection of these cells can be completed via the addition of puromycin. The
subsequent addition of FGF2 and EGF outgrows a population of bipolar cells
called LC1, during which process the ES cell-derived NEP cells gradually ex-
tinguish GFP expression and acquire the NS cell phenotype, whereby they stop
expressing pluripotent markers Nanog, Oct-4 and Sox1 and gain the expression
of Nestin, BLBP, Sox2 and RC2 immunoreactivity and lack the expression of
GFAP. These are the same markers expressed by the NS cell lines obtained
45

2.2 Neural Stem Cells Introduction
Figure 2.7: Protocol describing conversion of ES cells into immortalised NS cell
lines. (a) Neural precursor differentiation is induced from ES cells in serum-free
adherent monoculture by subtracting LIF and detected via the Sox1-GFP reporter
transgene. Sox1 is one of the earliest neural differentiation markers. (b) After seven
days cells are re-plated in basal medium in the presence of either FGF2 alone or
FGF2 and EGF. (c) Residual undifferentiated ES cells are eliminated from the culture
through puromycin selection. (d) Neural precursors lose GFP expression as they
become Sox1 - and are re-plated in fresh medium in the presence of EGF and FGF2.
They also attach and outgrow a population of bipolar cells, termed LC1. (e) Clonogenic
NS cell lines are generated by plating single cells from the LC1 population.
Adapted from Conti et al 2005 [107].
Figure 2.8: Representation of the Sox1-GFP reporter construct used in the nicheindependent
NS cell protocol.
from mouse fetal and adult tissues using the same protocol. To establish the
presence of clonogenic NS cells, single cells were isolated from the LC1 cul-
ture and expanded as adherent cultures to show the same morphology, growth
characteristics and markers of the LC1 population. Upon withdrawal of EGF
and FGF2 and exposure to serum or BMP4, these NS cells differentiate into
astrocytes. In contrast, removal of EGF followed by FGF2 gives rise to cells
46

2.2 Neural Stem Cells Introduction
with immunochemical and electrophysiological properties of mature neurons.
Importantly, even after prolonged expansion, NS cells maintain their potential
to differentiate efficiently into neurons and astrocytes in vitro as well as upon
transplantation into the adult brain [107]. To promote oligodendroglial differ-
entiation cells were cultured on laminin coated dishes in medium containing
N2 supplement plus FGF2, PDGF and forskolin, a growth factor combination
known to enhance oligodendrocyte progenitor proliferation, which efficiently
differentiated NS cells into oligodendrocytes [165]. To summarise the charac-
terisation of the mouse NS cell lines derived by Conti et al:
· they express the astrocyte differentiation marker GFAP upon addition
of serum or BMP4 and differentiate into astrocytes;
· they express neuronal markers type III β-tubulin (TUBB3), Microtubule-
associated protein 2 (MAP2) and ERBB2 upon removal of EGF and
FGF2 (in this order) and differentiate into neurons. All NS cell lines
produced in the study by Conti et al were electrophysiologically active
and exhibited voltage-gated Na + and Ca 2+ conductance, typical of ma-
turing nerve cells;
· they show no significant decline after many passages and retain diploid
chromosome content throughout late passages, maintaining intact the
differentiation potential in both the neuronal and glial direction;
· they do not form teratomas 31 , an important step towards the confirma-
tion of the identity of NS cells. Unlike ES cells, in fact, the differentiation
potential of NS cells is incapable of teratoma formation and this obser-
vation was reproduced in an experiment by Conti et al [107] in which NS
cells were transplanted in mouse fetal and adult brain and grafted onto
mouse kidney, where they did not proliferate or give rise to teratomas.
The absence of any histological evidence of unregulated proliferation or
tumor formation was a clear confirmation of the identity of the NS cells.
Although the NS cell lines generated via the niche-independent NS cell protocol
are capable of differentiating into all three lineages of the CNS, i.e. astrocytes,
oligodendrocytes and neurons, the neuronal subtypes are limited to the gen-
eration of large amounts of GABAergic neurons 32 according to the results by
31 Encapsulated germ cell tumor derived from pluripotent cells with tissue or organ components
resembling normal derivatives of all three germ layers.
32 Neurons that release the main CNS inhibitory neurotransmitter GABA.
47

2.2 Neural Stem Cells Introduction
Conti et al [107]. This differs vastly from the neuronal subtypes identified upon
direct differentiation of ES cell with no intermediate expansion of the neural
progenitors, which are skewed towards the generation of large amounts of glu-
tamatergic neurons 33 [55]. This inconsistency shows that in vitro expansion
of NS cells may be somehow restricting the neuronal subtype differentiation
capacity of these cells and further studies will have to address whether cell
culture conditions can be altered for long-term expanded NS cells so that re-
gional identities can be re-established [400].
Previous studies have shown the derivation of glial restricted progenitors in ad-
herent culture using FGF2 for survival and expansion of the cells [76,212,276,
367]. However, these cultures change their properties over time and should not
be considered equivalent to expanded long-term stem cell lines [403]. In the
NS cell derivation from mouse ES cells described by Conti et al, the absence
of EGF causes caspase 3-lead apoptosis and an immature neuronal phenotype.
This phenomenon was found to be avoidable by culturing NS cells on laminin
in the absence of EGF, although this addressed NS cells towards neuroblast 34
commitment with differentiation upon mitogen withdrawal [290]. The NS cells
obtained in the protocol by Conti et al exhibit phenotypic similarities to ra-
dial glia in that they show no expression of neuronal or astrocyte antigens, but
uniform expression of neural precursor markers Nestin, RC2, Vimentin, 3CB2,
Lex1, Paired box gene 6 (Pax6) and Prominin (see Fig 2.2). In addition to
this set of markers considered diagnostic for neurogenic radial glia, they show
expression of the neural precursor markers Sox2, Sox3, and Emx2, and the
transcription factors Olig2 and Mash1. The absence of Sox1 and maintenance
of Sox2 is noteworthy of NS cells since the former marks all early neuroec-
todermal precursors and its absence in stem cells might indicate that Sox2
is playing the key role. In a study by Gomez-Lopez et al [169] the roles of
Sox2 and Pax6 were investigated in mouse fetal forebrain-derived NS cells,
to find that conditional deletion of either gene reduced the clonogenicity of
these cells in a gene dosage-dependent manner. Cells heterozygous for either
gene displayed moderate proliferative defects, but in the complete absence of
Sox2, cells exited the cell cycle with concomitant down-regulation of neural
progenitor markers Nestin and Blbp, and ablation of Pax6 also caused major
33 Neurons that release the main CNS excitatory neurotransmitter glutamate.
34 Neuroblasts are the descendants of NS cells in the SVZ that migrate into damaged brain
areas after strokes or other brain injuries to generate regionally appropriate new neurons
[290]
48

2.2 Neural Stem Cells Introduction
proliferative defects. However, a subpopulation of cells was able to expand
continuously without Pax6, retaining the progenitor markers but displaying
an altered capacity to differentiate into astrocytes and oligodendrocytes, high-
lighting the role of Pax6 beyond neurogenic competence. The findings by this
study, therefore, suggest that Sox2 and Pax6 are both critical for self-renewal
of differentiation-competent radial glia [169].
Time-lapse videomicroscopy of the NS cells derived by Conti et al [107] also
demonstrated that cell nuclei of those NS cells undergo interkinetic nuclear
migration, a well-characterised feature of NEP and radial glia cells in vivo.
Importantly, the mouse fetal brain, human ES cell and human fetal cortex-
derived NS cells, expressed the same radial glia and neurogenic markers as the
mouse ES cell-derived NS cells, although the human cells exhibited moderate
levels of GFAP, consistently with the known activity of the human GFAP pro-
moter in radial glia and unlike the very feeble expression observed in mouse NS
cells [310,418]. Human ES cell and fetal cortex-derived NS cells also proliferate
more slowly than the mouse-derived ones, like in neurospheres, and after se-
quential withdrawal of EGF and FGF2, generate mixed populations of TuJ1 +
neuron-like cells and GFAP + cells, with pure populations of cells with typical
astrocyte morphology and intense GFAP immunoreactivity readily produced
after exposure to serum [107].
In a separate study, Sun et al [481] report the derivation and characterisa-
tion of human NS cell lines from human fetal cortex and spinal cord using
a continuous adherent procedure that is more efficient than allowing primary
cells to form neurospheres and subsequently isolating NS cells, as described
in the protocol by Conti et al. In the protocol developed by Sun et al, pri-
mary cells are seeded onto laminin coated dishes in growth medium containing
both EGF and FGF2. In these conditions cells readily attach and produce
a morphologically heterogeneous population containing both Nestin + neural
precursors and Tuj1 + neurons. In order to enrich for undifferentiated neural
precursors, the cells are temporarily transferred onto gelatin coated dishes, in
which conditions neurons and committed neuronal progenitors fail to survive.
Three weeks after initial plating, the primary human culture is homogeneously
Nestin + and Tuj1 - [481].
Once established, human NS cells can be expanded continuously in monolayer
culture where they homogeneously express immature neural precursor markers
Nestin and Sox2. The long-term expansion of these cells can also occur suc-
49

2.2 Neural Stem Cells Introduction
cessfully in EGF alone, confirming previous findings with mouse NS cells [403],
where addition of FGF2 is essential for initial derivation but can be dispensed
with during subsequent propagation, suggesting that EGF is the major mito-
gen for NS cell self-renewal, although a contribution of autocrine FGF2 is not
excluded. However, neither human nor mouse primary cells produce stable cell
lines unless they are exposed to FGF2 during the first 2-4 weeks after plating,
suggesting that a possible contributing factor is that FGF2 may induce EGF
responsiveness in NS cells. Furthermore, when human NS cells expanded in
EGF only are exposed to the differentiation conditions described above, they
are able to generate both neurons and glial cells [481].
Immunostaining of the derived human fetal NS cells showed the expression of
the markers that are hallmarks for radial glial cells, including BLBP, 3CB2,
GLAST, Vimentin, and GFAP, as well as neural progenitor markers Nestin,
Sox2, Pax6, Olig2, and CD133, with Sox1 transiently expressed in mouse
and human neural precursor cells, but not maintained in human fetal NS
cells [107,302,481]. It takes, on average, one month to derive an adherent and
morphologically homogeneous human NS cell population with a total num-
ber of approximately 2 million cells. To the date the research article by Sun
et al was published, the group had successfully derived five human brain NS
cell lines: CB192, CB516, CB525, CB541, and CB660 [481]. The differentia-
tion potential of these human NS cells was assessed using protocols previously
developed for mouse NS cells by Conti et al [107] and Glaser et al [165]:
· Neuronal differentiation was triggered by removing EGF from growth
medium first and FGF2 successively. By the end of the fourth week of
neuronal differentiation, many cells became Tuj1 + and exhibited thin
elongated processes.
· To generate oligodendrocytes, cells were treated with basal medium sup-
plemented with insulin-cotaining N2, forskolin, FGF2, and PDGF. From
the third week, the supplements were changed to N2, PDGF, T3, and
ascorbic acid, with successive withdrawal of PDGF inducing the appear-
ance by the fifth week of 1-2% Olig-4 + cells bearing branched oligoden-
drocyte morphology.
· In the absence of EGF and FGF2 and presence of BMP4 or serum, a
morphologically homogeneous astrocyte population was derived. The
sole removal of EGF and FGF2 without addition of BMP or serum also
led to NS cell differentiation into astrocytes but with significant cell death
50

2.2 Neural Stem Cells Introduction
ensuing. Recently it was reported that different types of human astro-
cytes may express distinct GFAP isoforms: adult SVZ astroglial cells
express GFAPδ, while most other astrocytes, including the derivative as-
trocytes of the human NS cells developed in the protocol by Sun et al,
express GFAPα [432].
In the protocol described by Sun et al [481], during the first four weeks after
cells are plated, primary human cells attach on laminin substrate within 24
hours and start to proliferate in the presence of both EGF and FGF2, with
no extended cell proliferation observed in medium containing only one of the
two growth factors and the subsequent impossibility of establishing NS cell
lines successfully. In addition to EGF and FGF2, the laminin substrate is
important for efficient human NS cell derivation, since primary cells grown
on gelatin coated or uncoated dishes easily detach and tend to form neuro-
spheres resulting in slower proliferation, as described in the study by Conti
et al [107]. The laminin substrate was also found to be optimal for human
NS cell propagation, indicating that laminin may play important roles in reg-
ulating neural cell behaviour [481]. Finally, although the human fetal-derived
NS cells exhibited some features of radial glia, the artificial nature of culture
environments may result in unique cell populations in vitro, which may indi-
cate, in turn, that NS cells do not have direct in vivo counterparts. In fact, as
a further consideration, the combination of transcription factor expression in
mouse NS cells is not routinely observed during normal development [403,481].
In order to shed the light on the nature of NS cells and their relationship
to endogenous cell types, Pollard et al [403] have investigated whether cells
capable of giving rise to NS cell cultures are restricted to developmental stages
or may also be present in the mouse adult brain. Although both FGF2 and
EGF were necessary for the derivation in culture of NS cells from adult mouse
forebrain (containing the adult SVZ), once established, the stem cell lines could
be maintained in added EGF alone. As already seen in mouse ES cell-derived
NS cell cultures, the absence of EGF causes the majority of the NS cells to die
of caspase 3-activated apoptosis and the rest to start differentiating towards
the neuronal lineage, although never reaching the fully mature phenotype (Fig
2.9). On the contrary, withdrawal of FGF2 did not result in any striking
change in NS cell morphology or behaviour, with the exception of a slightly
lower doubling time for EGF-only grown colonies possibly due to a higher
cell death, although more detailed analysis are required [403]. For complete
51

2.2 Neural Stem Cells Introduction
differentiation to neurons, NS cells had to be plated on laminin-coated plates
in basal medium withdrawing first EGF and then FGF2 in this order, since so
long as EGF and FGF2 are supplied together no cell differentiation can occur
and cells continue to self-renew. For differentiation to astrocytes, NS cells were
exposed to 1% serum without EGF and FGF2 or exposed to BMP4 and LIF
[399]. For differentiation to oligodendrocytes, the culture conditions involved
re-culturing on laminin coated dishes in medium containing FGF2, PDGF
and forskolin first, adding T3 and ascorbic acid later, a procedure known to
promote the differentiation and survival of oligodendrocytes [165]. Under these
conditions, NS cells efficiently differentiated into oligodendrocytes, astrocytes
and neurons, with similar outcomes seen for NS cells derived from fetal mouse
brain, altogether demonstrating that oligodendrocyte progenitor proliferation
and differentiation seems to be preserved in adult mouse-derived NS cells.
Importantly, the ability for oligodendroglial differentiation is maintained after
transplantation. Therefore, the differentiation spectrum of these NS cells is
not restricted to neurons and astrocytes but also extends to oligodendrocytes
[165,399].
Figure 2.9: Roles of EGF and FGF2 in the derivation and maintenance of NS
cells. Fetal forebrain progenitors or ES cell-derived neural progenitors (green) can
be converted into NS cell lines (yellow) using a combination of EGF with FGF2.
Once established, NS cells can be maintained in added EGF alone, whereas in FGF2
alone, they undergo differentiation (blue) and apoptosis. Adapted from Pollard et al
2006 [403] and Conti et al 2005 [107].
52

2.2 Neural Stem Cells Introduction
In light of the successful engraftment into adult mouse brain [107], the ES
cell or primary CNS tissue-derived NS cells (fetal and adult) show the poten-
tial for delivery of cell replacement and gene therapies. Furthermore, since
homogeneous expansion of any stem cell in defined conditions has been until
now a prerogative of ES cells, these NS cells provide an accessible system for
characterisation, manipulation and analysis of the stem cell state, as well as a
resource for direct comparison with ES cells [107]. A summary of commonali-
ties and differences of lineage-restricted and pluripotent stem cells is shown in
table 2.1.
Table 2.1: Summary of commonalities and differences between ES cells and NS
cells. Adapted from Pollard et al 2006 [399]
Features ES cells NS cells
Species Rodent and primate Rodent and primate
Source Blastocyst ES cells, germinate areas
of fetal brain, SVZ of
adult brain
Growth factor dependance LIF+BMP (serum free) EGF+FGF2 (serum free)
Expansion in vitro Immortal Immortal
Clonogenic Yes Yes
Doubling time 12 hours 24 hours
Stem cell divisions Symmetrical Symmetrical
Karyotype Stable diploid Stable diploid
Niche dependence None None
In vivo counterpart Similarities to ICM Similarities to radial glia
Potency Pluripotent Multipotent
Genetic manipulation Yes Yes
Neurosphere versus adherent culture method Since NS cells offer a po-
tential source for cell and tissue replacement therapy and for drug discovery,
but no specific markers have been identified so far that distinguish them from
their more differentiated progenitor cells, both NS cells and their progenitor
cells have been isolated in a recent study by Sun et al [479] from embryonic
or adult brain and spinal cord, and maintained in suspension culture as neu-
rospheres as well as in adherent substrate-bound culture. The side-by-side
comparison of these two culture systems was called for by the necessity to
understand how processes such as cell survival, proliferation, differentiation
and passaging performed in each system for potential use in drug discovery
53

2.2 Neural Stem Cells Introduction
and cell therapy. For example, the neurosphere culture method has been used
extensively to study molecular and cellular mechanisms that control neuro-
genesis, differentiation and cancer proliferation [116,261], but a recent study
showed that long-term neurosphere cultures induce changed differentiation and
self-renewal capacities, and the occurrence of chromosomal instability [518].
Similarly, an earlier study by Pollard et al [401] had established that caution
should be exercised also when extrapolating in vitro adherent monolayer NS
cell culture findings to in vivo development because FGF2 induces a subset of
cell surface markers that are not found in vivo. In this study, microarray-based
expression profiling was used to identify a set of markers expressed by fetal
mouse NS cells but not ES cells and found the cell surface protein CD44 to
be differentially expressed with higher expression levels in fetal mouse-derived
NS cells. Although CD44 was expressed homogeneously by all NS cell lines
derived in this study, appreciable numbers of CD44 + cells could not be found
in the developing brain during neurogenic stages, nor in differentiating ES cell
cultures. Moreover, CD44 expression was found to be induced by FGF2 in
a subset of cells in primary culture and this effect did not appear to be re-
stricted to CD44, with many other NS cell markers found to be activated in
vitro, including Adam12, Cadherin20, Cx3cl1, EGFR, Frizzled9, Kitl, Olig1,
Olig2 and Vav3. Therefore, it was speculated that the self-renewing NS cell
state may be generated in vitro following transcriptional resetting induced by
FGF2 and the latter did not act simply to maintain and expand pre-existing
stem cells, but also impart significant changes to the transcriptional and cel-
lular phenotype [401].
Since the ability of NS cells to self-renew and differentiate into the three ma-
jor neural cell lineages make them ideal candidates to treat impaired cells
and tissues in neurodegenerative diseases, spinal cord injuries and stroke [544],
methods that can scale up their production need to be evaluated accurately.
Thus, the study by Sun et al [479] aimed at characterising the proliferative,
differentiation and passaging capacities of the two culture methods of election
for culturing NS cells, i.e. the neurosphere and adherent monolayer culture
methods. The starting material for this study was dissociated from E14 rat
cortical neuroepithelium, early enough in the development of the CNS that
proliferating NEP cells and NS cells coexist with their neuronal, neuroglial,
and glial progenitors, as well as newly post-mitotic cells (Fig 3.1). These cells
were expanded in serum-free media in the presence of the mitogenic growth fac-
54

2.2 Neural Stem Cells Introduction
tors FGF2 and EGF as described by several previous studies [107,165,403,481].
In their study, Sun et al [479] have shown that NS cells and their progenitors
grew on adherent substrate significantly faster in the first four passages than
the NS cells in neurospheres, but slowing down and plateauing their growth
rate after the sixth or seventh passage, while in neurospheres their growth kept
increasing slowly but steadily for more than 10 passages.
Immunocytochemical analysis using multiple lineage-specific antibodies in early
primary cultures for both neurosphere and adherent methods showed about
60% MAP2 + neurons and about 40% Nestin + progenitors, indicating that
both cultures were initially composed of neural progenitor and mature neu-
ronal populations. However, proliferating neural progenitors became dominat-
ing in adherent culture over the next few days to finish off on the ninth day
with a Nestin + progenitor percentage of 93.76%, clearly identifying the type
of cells that eventually compose the culture [479]. In order to test for the
ability of both culture systems to differentiate into neuronal and glial cells,
growth factor withdrawal was induced for seven days and differentiation was
assessed through the use of antibodies against Tuj1, GFAP and Olig-4 for the
identification of neurons, astrocytes and oligodendrocytes, respectively. The
data gathered from passages 1, 3 and 5 revealed no significant differences in
the differentiation potential of cells cultured with either method [479].
Self-renewal capacity was assessed in the two culture systems through a clono-
genic assay that evaluated the capacity of primary neurospheres to form sec-
ondary neurospheres, as well as the capacity of plating from a single cell from
the original adherent culture. The clone forming rates at passage 1 were found
to be 63% for neurospheres and 44% for adherent culture, a number signifi-
cantly higher for neurospheres than the 20% reported by Reynolds et al [525].
Interestingly, as discussed in the study by Marten et al [319], in the fetal
mouse forebrain geminal zone there seem to be two differently responding self-
renewing cells that grow neurospheres with greater diameters in the presence
of EGF alone in their culture in vitro than FGF. Since in the culture systems
developed by Sun et al both EGF and FGF were added, the two differently
responsive stem cells might have been produced together, increasing the clone
forming rates observed [479].
Finally, neurosphere cultures were found to have a greater passage potential
than adherent cultures, possibly due to the three-dimensional reproduction of
the niche environment experienced by the cells in vivo. In fact, the regulatory
55

2.2 Neural Stem Cells Introduction
signals from a niche are provided by soluble factors and the ECM, and neuro-
spheres are spheroid structures that consist of cells producing their own ECM
molecules, such as laminins, fibronectin, chondroitin sulphate proteoglycans,
as well as growth factors and beta integrins, epidermal growth factor recep-
tors, and cadherins. Therefore, the cell-cell and cell-matrix interactions in the
three-dimensional structure of neurospheres can create an environment that is
more physiologically relevant than the two-dimensional one in adherent culture
systems. Even so, the robustness of cell proliferation in adherent cultures is
probably due to the fibronectin and laminin substrates, in that fibronectin is
a ubiquitous component of various types of ECMs, and laminin was reported
as one of the five different substrates regulating neural differentiation of hu-
man ES cells, and precisely stimulating their expansion in a dose dependent
manner [303]. In the light of the increasing evidence that astrocytes [210] and
endothelial cells [354,452] might also be important components of a niche for
NS cells on top of the ECM, the adherent culture system needs to be optimised
in this regard.
56

Chapter 3
Brain Cancer Stem Cells
Contents
3.1 The Cancer Stem Cell Hypothesis . . . . . . . . . . . . . . 57
3.2 Brain Cancer Stem Cells . . . . . . . . . . . . . . . . . . . 63
3.3 Glioma Culture Systems . . . . . . . . . . . . . . . . . . . . 72
3.1 The Cancer Stem Cell Hypothesis
The cancer stem cell hypothesis proposes that organisation of cell lineage in
tumours is hierarchical and only a subpopulation of cells termed "cancer stem
cells" is responsible for tumour expansion. According to this hypothesis, stem
cells or cells that acquired the ability to self-renew, accumulate genetic changes
over long periods of time, escape from the control of their environment, and
give rise to cancerous growth. One of the postulations of the cancer stem cell
hypothesis is that a population of cells with stem cell-like features exists in tu-
mours and this population gives rise to the bulk of the tumour cells with more
differentiated phenotypes [123,457]. Two types of cells could fulfill the role of
tumour-initiating cell: adult stem cells or their progenitors, which normally
undergo limited numbers of cell divisions and aberrantly acquire the capacity
to self-renew by accumulating genetic lesions and subsequently becoming the
long-lived target (Fig 3.1). Regardless of the cell of origin, the cancer stem
cell is defined by its stem cell-like properties [486]. Normal adult stem cells
are attractive candidates because they are tissue-specific, they can self-renew
and, finally, they can differentiate into all cell types of the tissue of origin.
Each division produces at least one daughter cell that maintains the indefinite
capacity for cell division and a progenitor cell that has finite division capacity,
ultimately differentiating into the mature cell types that constitute specific
57

3.1 The Cancer Stem Cell Hypothesis Introduction
Figure 3.1: Stem cell differentiation hierarchy, in which a possible increase in plasticity
is highlighted that may be present within cancer populations. Such plasticity
would be the enabler for the bidirectional interconvertibility between cancer stem
cells and non-cancer stem cells. Adapted from Gupta et al 2009 [182].
tissues. This type of cell division is referred to as "asymmetrical division" or
"asymmetrical self-renewal" and is specifically characterised by adult stem cell
divisions that produce a new adult stem cell together with a non-stem cell sis-
ter that becomes the progenitor for short-lived, differentiating, functional cells
that in most cases mature to a terminal division arrest [420]. For the vast ma-
jority of adult tissues, it remains unclear how stem cells succeed in maintaining
a precise balance between proliferation and differentiation in steady-state. To
explain their long-term viability, it has been argued that tissue stem cells are
maintained in a long-lived quiescent state, with most divisions supported by
differentiating progenitor cells that ultimately exit the cell cycle and are re-
placed by stem cell progeny, which provides a mechanism for protecting stem
cells from damage and loss throughout adult life (Fig 3.2). Clear evidence
for asymmetric stem cell divisions is found in invertebrates C. elegans and D.
melanogaster, although in recent lineage-tracing studies in mammals it was
shown that stem cells behave as an equipotent population, in which the bal-
ance between proliferation and differentiation is achieved through frequent and
stochastic stem cell loss and replacement [240]. While the asymmetrical self-
renewing properties of adult stem cells are particularly important for homeo-
static control in adult tissues that undergo continuous cellular turnover such
as epithelium and blood [457,486], the non-stem cells of most adult tissues go
through a rapid cycle in which they are born, mature, expire, and are removed
from the tissue by apoptosis. The fact that the rate of adult cellular turnover
is much faster than that of tumour development is the basis for the hypothesis
58

3.1 The Cancer Stem Cell Hypothesis Introduction
Figure 3.2: (a) Asymmetric cell division, in which each stem cell (orange) generates
one daughter stem cell and one daughter destined to differentiate (green). (b-c)
Population strategies that provide dynamic control over the balance between stem
cells and differentiated cells, a capacity that is necessary for repair after injury or
disease. In this scheme, stem cells are defined by their "potential" to generate both
stem cells and differentiated daughters, rather than their actual production of a
stem cell and a differentiated cell at each division. (b) Symmetric cell division: each
stem cell can divide symmetrically to generate either two daughter stem cells or two
differentiated cells. (c) Combination of cell divisions: each stem cell can divide either
symmetrically or asymmetrically. Adapted from Morrison et al 2006 [345].
that non-stem cells cannot effectively initiate cancers, a concept that contrasts
with the commonly held idea that cancers may arise from any tissue cell with
equal likelihood [420]. Although adult stem cells are attractive candidates to
fulfill the role of tumour-initiating cells, two of their properties are also con-
sidered limitations to that end: firstly, asymmetric division potentially limits
the number of stem cells and therefore the incidence with which they could
drive tumourigenesis, and secondly, the immortal DNA strand co-segregation 35
process reduces the rate at which they can accumulate mutations. Through
the DNA strand co-segregation molecular manoeuvre, adult stem cells reduce
their mutation rate by more than 1000-fold, avoiding all mutations that arise
from replication errors that are not properly repaired [420].
Since all stem cells alike must self-renew and regulate the relative balance be-
tween self-renewal and differentiation, and cancer can be considered a disease
of unregulated self-renewal, understanding the regulation of normal stem cell
self-renewal is fundamental to understanding the regulation of cancer cell pro-
liferation [458]. By maintaining at least some of the properties of their tissue
35 Non random segregation of the set of chromosomes with the oldest template of the DNA
strands operated at each cell division.
59

3.1 The Cancer Stem Cell Hypothesis Introduction
of origin, cancer stem cells give rise to tumours that phenotypically resemble
their origin, either by morphology or by expression of tissue-specific genes.
However, what distinguishes cancerous tissue from normal tissue is the loss
of homeostatic mechanisms that maintain normal cell numbers, and much of
this regulation normally occurs at the stem cell level. The cancer stem cell
hypothesis raises the important experimental implication that if a population
of biologically unique cancer stem cells exists, then tumour cells lacking stem
cell properties will not be able to initiate self-propagating tumours, regardless
of their differentiation status or proliferative capacity, which has an impact
on the experimental definition of cancer stem cell. Furthermore, the cancer
stem cell hypothesis raises a clinical implication that curative therapy will re-
quire complete elimination of the cancer stem cell population, since patients
who show an initial response to treatment may ultimately relapse if even a
small number of cancer stem cells survive. On the other hand, targeted ther-
apies that eliminate the cancer stem cell population offer the potential for a
cure [486].
The concept of cancer stem cell initially arose from the observation that cancer
tissues resembled developing tissues and self-renewal mechanisms were com-
mon to cancer cells and stem cells. The definitive demonstration of cancer stem
cells in human neoplasia was first made in 1994 in leukemia [122], a non-solid
tumour found harbouring a stem cell hierarchy pattern, in which a minority
of cells within the leukemic population possessed extensive proliferation and
self-renewal capacity not found, however, in the rest of the leukemic cells.
The putative cancer stem cells in leukemia were isolated and characterised
on the basis of their phenotypical similarities to normal hematopoietic stem
cells. The principle of uncovering significant similarities between putative can-
cer stem cells and normal stem cells, was then extended to solid tumours, first
in breast cancer then in brain cancer, although normal stem cells, their differ-
entiation hierarchy and markers that identify them, are not well characterized
in most solid organs. Since then, other tissues in which the connection be-
tween stem cells and cancer has been found are mammary gland [15,288,289],
gut [177,395,414], skin [75], bladder [89] and prostate [103] .
So far, cancer stem cells have been defined on the basis of their ability to seed
tumours in animal hosts, to self-renew and to generate non-cancer stem cell
differentiated progeny. Accordingly, the number of cancer stem cells within
a population of cancer cells can be measured by the number of cells that are
60

3.1 The Cancer Stem Cell Hypothesis Introduction
required, at limiting dilutions, to seed new tumours. The pioneering studies in
leukemia and later solid tumours showed that it is possible to use cell surface
marker profiles to isolate cancer cell subpopulations that are enriched for or
depleted of cancer stem cells. Subsequent reports showed that, after implanta-
tion in vivo, cancer stem cell-enriched populations generate tumours that are
no longer enriched for cancer stem cells, implying that the cancer cells within
a single tumour are naturally found in multiple states of differentiation with
distinct tumour-seeding properties. The stemming possibility of bidirectional
interconversion between cancer stem cell and non-cancer stem cell populations
does not undermine the cancer stem cell hypothesis, since the two populations
always retain their distinct identities and they can be distinguished phenotyp-
ically and functionally at any moment (Fig 3.1) [182].
One of the emerging caveats of the cancer stem cell hypothesis is the actual
frequency of cancer stem cells. In normal tissues, somatic stem cells are inher-
ently rare, and most of this type of data in cancer is derived from xenograft
experiments, in which the frequency of human cancer stem cells is determined
upon transplantation into a mouse environment. Differences between human
and mouse stromal and support cells, cytokines as well as distinct levels of
immunological function in different immunodeficient recipient mouse strains,
have led to conflicting frequencies of cancer stem cells ranging between 1% and
25%. This discrepancy has highlighted that the stem cell properties of cancer
stem cells are inherent but might also be the result of the interaction with the
environmental milieu [393]. Thus, as suggested by recent findings, the number
of cancer stem cells in a tumour may be a function of the cell type of origin,
stromal 36 microenvironment, accumulated somatic mutations and stage of ma-
lignant progression reached by the tumour. In fact, an early report indicated
that the proportion of leukemia stem cells varies up to 500-fold between patient
samples. More recent reports have suggested that relatively undifferentiated
tumours at the histopathological level may contain higher proportions of cancer
stem cells than their more differentiated counterparts. Furthermore, the num-
ber of cancer stem cells may be different between tumour subtypes that arise
from a single tissue, indicating that the cancer stem cells may be as numerous
as the non-cancer stem cells in certain subtypes. However, the cancer stem cell
hypothesis can be adapted to state that cancer cells can exist in at least two
alternative phenotypic states with different tumour-seeding potentials, with-
36 Connective tissue supporting the parenchymal cells of an organ.
61

3.1 The Cancer Stem Cell Hypothesis Introduction
out imposing requirements on the number of cancer stem and non-stem cells
needed for different tumour subtypes or developmental stages [182]. In vitro
cultures in unattached conditions promote the growth of sphere-like cell ag-
gregates that are routinely used for enrichment and propagation of stem cells.
When tumours such as glioblastoma were cultured this way, only CD133 + cells
and not CD133 - cells, were found to successfully grow these spheres, renamed
"tumour spheres", which expressed neuronal stem cell markers, showed highly
tumourigenic potential and were resistant to radiation. Despite the success
of suspension cultures to enrich for stem cell-like cells in glioblastoma and
other tumour types found to behave similarly, it is unknown how stem cells in
this in vitro simulation find the correct niche for the support of normal tissue
stem cells, such as self-renewal, multipotency, proliferation, and differentiation.
Therefore, radiation and drug sensitivity assays performed in sphere cultures
have to be carefully designed and interpreted and comparing cells growing in
adherent and suspension cultures may yield differences irrespective of cellular
differentiation status [457].
A major paradox of the cancer stem cell model lies in the multi-step tumour
progression model, in which one precursor population of pre-malignant cells
evolves via mutation into a successor population that has a phenotypic advan-
tage, such as an increased resistance to apoptosis or growth-inhibitory signals.
The conventional depiction of the cancer stem cell model would state that
the only cells within the precursor population that are qualified to evolve into
a successor population are its stem cells, since only these cells are endowed
with the self-renewal capabilities that are required to generate unlimited num-
bers of progeny [182]. This view, however, ignores the inherent properties
of malignant cells, i.e. genomic instability and the ability to undergo rapid
evolutionary changes [457]. Also, if the percentage of stem cells in the precur-
sor cell population is small, then according to the cancer stem cell model the
number of cells that can serve as targets for genetic evolution is just as small,
and this postulation does not agree with the observation that the mutation
rate required to complete cancer formation would have to be up to two orders
of magnitude above the rates described so far for human tumour cells [182].
Thus, a major problem with this hypothesis is that it assumes stability within
the tumour and does not consider the possibility that the cancer stem cell phe-
notype can be acquired. To this end, several studies have demonstrated that
more differentiated cancer cells can acquire mutation or activate a transcrip-
62

3.2 Brain Cancer Stem Cells Introduction
tion factor and become cancer stem cells [457]. This paradox may be resolved
if the non-cancer stem cells in a precursor cell population could also serve as
targets of mutation, leading to clonal succession and, therefore, tumour pro-
gression [182]. For example, when tumour spheres derived from highly vascular
glioblastomas were transplanted into mice, the initial tumours demonstrated
low-grade glioma phenotype without any sign of angiogenesis. However, upon
serial transplantation in vivo, the tumour cells developed a highly malignant
phenotype with extensive angiogenesis and necrosis being present in the tu-
mours. This finding highlights a well-established fact that tumour cells evolve
and if more malignant and less differentiated cancer cells have growth advan-
tage, they will be selected for and expanded in the tumour. Therefore, as
tumours progress, the line between cancer stem cells and the rest of tumour
cells might gradually become blurred and can even disappear [457].
Finally, the recent finding that, depending on the tumour analysed, glioblas-
toma cancer stem cells can be CD133 + or CD133 - cells, also emphasises that
either we do not have good markers for cancer stem cells or all tumour cells
are tumourigenic at varying degree, which brings us back to what we already
know about tumours being diverse, genetically unstable, and evolving due to
the intratumoural diversity of cellular genotypes and phenotypes. Especially
in light of the recent finding that normal human differentiated cells can be
converted into functional pluripotent embryonic stem cells by expressing the
right combination of transcription factors [215,226,462], it is a possibility that
more differentiated cancer cells may acquire stem cell phenotypes. This means
that eradication of tumours will likely be achieved by the successful target-
ing of all cancer cells using a cocktail of drugs effective against all cancer cell
sub-populations and not only the stem cell type [457].
3.2 Brain Cancer Stem Cells
For many solid tumours within the CNS evidence in support of the cancer
stem cell hypothesis has emerged. In human glioblastoma, two separate stud-
ies from Bao et al [40] and Piccirillo et al [392] have separately tried to discern
the stem cell nature of this tumour, identified in a pioneering study by Singh
et al [458]. Glioblastomas are diffuse tumours that invade normal brain tissues
and frequently recur from focal masses after radiation, suggesting that only a
fraction of tumour cells is responsible for regrowth, supporting the cancer stem
cell hypothesis in solid tumours. The heterogeneity of glioblastoma starts at
63

3.2 Brain Cancer Stem Cells Introduction
the cellular level thanks to the identification of tumour-initiating cells express-
ing or not expressing Prominin as CD133 + stem cells, or more differentiated
CD133 - cells that include glioblastoma progenitor cells, respectively [121,458].
The study conducted by Singh et al [458] reports the identification and pu-
rification of a cancer stem cell exclusively isolated within the cell fraction
expressing the neural stem cell surface marker CD133, from different pheno-
types of human brain tumours. Higher-grade gliomas showed an increased
self-renewal capacity with respect to lower-grade gliomas and, importantly,
the CD133 + cells could differentiate in culture into tumour cells that pheno-
typically resembled the tumour from the patient. Neurosphere assays were
used to functionally characterize the tumour cell population and identified the
CD133 + cell as representing the minority of the tumour cell population. This
cell lacked the expression of neural differentiation markers, was necessary for
the proliferation and self-renewal of the tumour in culture, and was capable
of differentiating in vitro into cell phenotypes identical to the tumour in situ.
Since the identified cancer stem cell markers CD133 and Nestin were also iden-
tified as defining normal NS cells, it was suggested that brain tumours can be
generated from cancer stem cells that share a very similar phenotype as normal
NS cells. Such identification of a brain tumour cancer stem cell, set up the
research scene for the following investigations of the tumourigenic process in
the CNS.
In the study by Singh et al, cultures from 14 solid primary pediatric brain
tumours were set to favour stem cell growth, resulting in all tumours grow-
ing within the first 48 hours as clonally derived neurosphere-like clusters, the
"tumour spheres", that continued to proliferate and expand the tumour cell
culture over time. All primary tumour spheres were assessed for ability to
form secondary tumour spheres upon re-plating, and all successfully exhibited
self-renewal abilities. The frequency of secondary tumour sphere generation
correlated with the tumour’s tumour clinical aggressiveness and varied accord-
ing to tumour pathological subtype. Both primary and secondary tumour
spheres retained the expression of the NS cell markers Nestin and CD133,
failing instead to express the neural differentiation marker of election for as-
trocytes, GFAP, and neurons, TUBB3 [458]. In neurosphere conditions, in
fact, brain tumour cells express Nestin and CD133 but also markers of neural
precursors such as Sox2, Notch and Jagged-1 [122]. By inducing differentiation
in culture, Singh et al observed that the dissociated tumour spheres preferen-
64

3.2 Brain Cancer Stem Cells Introduction
tially differentiated down the lineage that characterised the original tumour
phenotype of the patient, losing the expression of Nestin and CD133 and gain-
ing the markers of differentiation that reflected the immunophenotype of the
original tumour [458]. This is a common finding amongst astrocytoma cells
that are grown as neurospheres, which differentiate into GFAP + astrocytes,
and glioblastoma cells, which instead differentiate into GFAP + astrocytes as
well as TUBB3 neurons, suggesting that the tumours derive from a cell with
multi-lineage differentiation capacity, i.e. a stem cell, and not a dedifferenti-
ated astrocyte as previously thought [122]. When tumour cell cultures from
the 14 pediatric tumours were sorted for CD133 expression, CD133 + tumour
cells showed growth as non-adherent tumour spheres with continuous expan-
sion of their cell population, while CD133 - cells adhered to the culture dishes
without showing proliferation and not forming spheres, demonstrating that
the stem cell properties resided in the CD133 + fraction [458]. With this study,
Singh et al demonstrated that the CD133 marker can identify an exclusive
subpopulation of brain tumour cells with NS cell activity, with three pieces of
evidence supporting this view:
1. CD133 + cells generated clusters of clonally derived cells that resembled
neurospheres, termed "tumour spheres";
2. CD133 + cells were capable of constant self-renewal, as well as prolifera-
tion;
3. CD133 + cells differentiated to recapitulate the phenotype of the tumour
from which they were derived.
It must be noted that self-renewal and proliferation differ because the former is
a cell division that must involve a cell fate decision, so that at least one of the
two daughter cells retains the full stem cell potential of the parent cell (Fig 3.2).
Exploring the molecular mechanisms involved in cell fate decisions of brain tu-
mour stem cell divisions could have important implications in understanding
clonal expansion and maintenance of these tumours [122]. Although cancer-
initiating abilities clearly reside in the CD133 + fraction, not every CD133 +
cell is capable of initiating sphere formation in vitro, demonstrating that not
every CD133 + cell has stem cell properties [122] and implying the existence of
a hierarchy, which will be functionally elucidated as more surface markers for
NS cells emerge and additional subpopulations are identified [458].
65

3.2 Brain Cancer Stem Cells Introduction
The study by Bao et al [40] aimed at assessing the resistance of the CD133 +
cells to ionization therapies caused by the more efficient and active DNA re-
pair mechanisms present in these cells with respect to the rest constituting the
bulk of the tumour. The observations in this study showed that after ionizing-
radiation treatment of glioblastoma cells grown in vitro or as grafts in mice,
the surviving fraction was enriched in CD133 + cells. These cells had the ability
to reinitiate heterogeneous tumours when transplanted into other mice, thus
demonstrating retention of their stem cell abilities. To assess the biological
relevance of the enriched CD133 + cells, xenografts with a constant number of
total cells but increasing fraction of CD133 + cells were generated that showed
a dose-dependent decrease in tumour latency, enhancement of tumour growth
and vascularity. The successive irradiation of these xenografts demonstrated
that viable tumour cells enriched for CD133 + cells could form secondary tu-
mours with decreased latencies themselves, demonstrating that enrichment of
CD133 + cells is crucial in glioma recurrence after radiotherapy. The CD133 +
tumour cells showed characteristics consistent with cancer stem cells, i.e. neu-
rosphere formation, expression of neural and cancer stem cell markers CD133,
Sox2, Musashi and Nestin, as well as multi-lineage differentiation with mark-
ers for astrocytes, neurons or oligodendrocytes. Furthermore, CD133 + cells
derived from xenografts or biopsy specimens formed neurospheres, whereas
CD133 - cells rarely did. Finally, CD133 + tumour cells were highly tumouri-
genic in brains of immunocompromised mice with characteristics of glioblas-
toma and CD133 - cells did not form detectable tumours even when implanted
with high doses of CD133 - cells, showing that CD133 + subpopulations were
enriched for characteristics of cancer stem cells, including tumourigenesis in
vivo [40].
Although ionizing radiation damages tumour cells through several mechanisms,
it kills cancer cells primarily through DNA damage, identifying DNA damage
checkpoint responses as having essential roles in cellular radio-sensitivity [40].
Progression through the mitotic cycle is driven by cyclin-CDK complexes,
which ensure that all phases of the cell cycle are executed in the correct order.
Terminally differentiated neurons cannot undergo cell cycle re-entry and CDK
activity is suppressed through interactions with two main families of inhibitory
proteins, the INK4 family that exhibits selectivity for CDK4 and CDK6, and
the CIP/KIP family that has a broader range of CDK inhibitory activity (Fig
3.3) [115]. As demonstrated in the study by Bao et al, activating phosphoryla-
tion of the checkpoint proteins Ataxia telangiectasia mutated (ATM), RAD17,
66

3.2 Brain Cancer Stem Cells Introduction
CDK1 and CDK2 was significantly higher in CD133 + cells than in CD133 - cells,
indicating that the former show greater checkpoint activation in response to
DNA damage. The fact that CD133 + glioma cells demonstrated to activate
checkpoint responses to a greater extent than CD133 - cells, suggested that the
resistance of CD133 + cells to ionizing radiation is due to preferential check-
point activation. The finding that CD133 + cells were made less resistant to
radiation if the checkpoint kinases CDK1 and CDK2, which control the pauses
in cell-cycle progression that are scheduled for DNA repair to occur (Fig 3.3),
were to be pharmacologically inhibited, provided the potential for a cure tar-
geting the resilient stem cell mass. The CD133 + resistant population should be
Figure 3.3: The cell cycle of eukaryotic cells can be divided into four successive
phases: M for mitosis, S for DNA synthesis, and two gap phases, G1 and G2. In
the G1 phase extracellular cues may induce either commitment to a further round of
cell division or withdrawal from the cell cycle into G0 to embark on a differentiation
pathway. The G1 phase is also involved in the control of DNA integrity before the
onset of DNA replication. During the G2 phase the cell checks the completion of
DNA replication and the genomic integrity before cell division starts. Adapted from
Dehay et al 2007 [115].
targeted with DNA checkpoint blockers to make these cells radiosensitive and
thus, in one therapy cycle, potentially wipe out the entire tumour mass (Fig
3.4) [123]. Together, the results exposed by Bao et al showed that CD133 +
cancer cells contributed to glioma radioresistance and tumour re-population
through preferential checkpoint response and DNA repair, and targeting of
checkpoint response in CD133 + cancer cells can overcome glioma radioresis-
tance in vitro and in vivo, which may provide a therapeutic advantage to
reduce brain tumour recurrence [40].
67

3.2 Brain Cancer Stem Cells Introduction
Figure 3.4: Glioblastoma treatment with ionizing radiation. Following radiation,
the bulk glioblastoma responds and the tumour shrinks. But CD133 + cells activate
checkpoint controls for DNA repair more strongly than CD133 - cells, resisting radiation
and prompting the tumour to regrow. These cells could be targeted with
DNA checkpoint blockers to make them radiosensitive. Adapted from Dirks et al
2006 [121].
The approach taken by Piccirillo et al [392] is based on the role that BMPs
detain as soluble factors that induce mature astrocyte differentiation in brain
tumour-initiating cells in glioblastoma. As already mentioned, these tumour-
initiating cells represent a small fraction of glioblastoma cells that belong to the
CD133 pool, display self-renewal in vitro, generate a large number of progeny,
are multipotent and can perpetuate across serial transplantation. Given that
BMPs favour the acquisition of an astroglial fate (see Section 2.2), the study
aimed to assess their effect on weakening the tumour-forming ability of CD133 +
cells by prompting their differentiation into astrocytes both in vitro and in
vivo [392]. This approach implies that tumour populations at least partially
retain a developmental hierarchy based on stem cells and remain able to re-
spond to the normal signals inducing them to mature. Treatment of cultured
glioblastoma progenitor cells or CD133 + cells with BMP, reduced the size of
the tumours grafted into mice and prolonged the survival of the animal be-
cause these cells were more mature and less invasive. Furthermore, CD133 +
cells could not be recovered from these small tumours and they were also found
to be incapable of serial engraftment, although some mice still died after three
months from treatment, showing that presumably some cancer stem cells es-
caped BMP treatment and were capable of re-iterating tumour formation (Fig
3.5) [123].
Amongst all BMPs assessed, BMP4 elicited the strongest effect by triggering
a significant reduction in the stem-like, tumour-initiating precursors of hu-
man glioblastomas. In fact, transient in vitro exposure to BMP4 abolished
the capacity of glioblastoma cells to establish intracerebral grafts, produc-
68

3.2 Brain Cancer Stem Cells Introduction
ing a pro-differentiation action predominantly in the astroglial direction and
depleting the pool of tumour-initiating cells. Based on these results, it was
inferred that the in vitro reduction in the stem-cell-like tumour-initiating cells
would correspond to a similar decline in the ability of BMP4-treated cells to
form tumours in vivo. By transiently exposing glioblastoma cells to BMP4,
in fact, the tumour-initiating stem-like population produced a significant de-
crease in the in vivo tumour-initiating ability of glioblastoma cells, since it
effectively blocked the tumour growth and associated mortality in 100% of the
mice subjected to intracerebral grafting. It is hypothesised that BMPs acti-
vate their cognate receptors and trigger the Smad signaling cascade, causing a
reduction in proliferation and increased expression of markers of neural differ-
entiation, with no effect on cell viability. In fact, blocking endogenous BMP4
reduces Smad signaling 37 and increases glioblastoma cell growth, perhaps by
regulating the balance between proliferation and differentiation, and favouring
the production of the differentiated astroglial-like cells normally found within
glioblastomas. The authors surmise that BMP4 may reduce the frequency of
tumour-initiating stem cell-like cells by decreasing symmetric cell cycles that
generate two identical cells on division, triggering differentiation of a subpop-
ulation of tumour-initiating stem cell-like cells or blocking their proliferation
and progeny, which, although not mutually exclusive events, could all con-
tribute to reducing the tumour-initiating stem cell-like population. Thus, the
signaling system constituted by BMPs and their receptors may also act as a
key inhibitory regulator of tumour-initiating, stem-like cells in glioblastomas,
other than controlling the activity of normal brain stem cells. Importantly,
the results of the study by Piccirillo et al also identified BMP4 as a novel,
non-cytotoxic therapeutic effector, which may be used to prevent growth and
recurrence of glioblastomas in humans [392].
Both these studies added depth to the cancer stem cell hypothesis, illustrat-
ing the potential of re-examining cancer under this new light and highlighting
the importance in cancer research of dissociating solid tumour samples into
single cell suspensions to purify the stem cell fraction and test its response
to treatment. Improved purification of the tumour, however, will be required
37 A signaling system that involves the activation by membrane receptor protein kinases
bound by the TGF-β ligand, of a family of receptor substrates, the Smad proteins, that
assemble into multi-subunit complexes to go into the nucleus and activate transcription.
This signaling pathway regulates a wide variety of cell-specific responses, depending on what
is "the cellular context" that the TGFβ family members and multifunctional hormones are
acting in [321].
69

3.2 Brain Cancer Stem Cells Introduction
Figure 3.5: Glioblastoma treatment with BMPs. BMPs normally cause NS cells
to differentiate into astrocytes. When used to treat isolated glioblastoma CD133 +
cells, they weaken their tumourigenicity both in vitro and, when engrafted into mice,
in vivo. The knowledge that a tumour retains a developmental hierarchy suggests
that targeting different cell populations is a promising therapeutic strategy. Adapted
from Dirks et al 2006 [121].
because the true stem cells are probably a subpopulation of the CD133 + frac-
tion, as demonstrated in the study by Bao et al by the death rates of mice
after three months from treatment [123]. Another study attempting to eluci-
date the role of CD133 was conducted in vivo on mice that were injected with
100 to 1,000 uncultured malignant brain tumour cells purified by bead sorting
for CD133 [459]. These CD133 + cells reproduced a phenotypical copy of the
patient’s original tumour and were also heterogeneous, with only a minority
of cells expressing CD133. This suggested differentiation in vivo, making dif-
ferentiation therapy look like a real option for brain tumour treatment and
denoting that brain tumours of different types are also functionally heteroge-
neous for tumour-initiating ability [122].
Several other studies have since then focused on the isolation of functional can-
cer stem cell markers in different types and stages of brain tumours. In a study
by Balenci et al [39], the mammalian IQ motif containing GTPase activating
protein 1 (IQGAP1), considered to be a scaffolding protein at the intersection
of several signaling pathways such as control of cell adhesion, polarization, di-
rectional migration and neuronal motility, was found to be a reliable marker
of Nestin + amplifying neural progenitors in rat brain. This protein is highly
70

3.2 Brain Cancer Stem Cells Introduction
abundant in rat and human glioma cell lines and it specifies a subpopulation
of amplifying tumour cells in glioblastoma-like tumours but not in tumours
with oligodendroglioma features, making it a reliable marker to distinguish
oligodendroglioma from glioblastoma. These findings suggest that the ampli-
fying IQGAP1 + cancer cells are closer to a multipotent progenitor cell and
they represent the most aggressive cancer cell population in glioblastoma [39].
In a study by Ma et al [304], the expression of the stem cell markers CD133,
Nestin, Sox2 and Musashi-1 amongst others was investigated in 72 astrocy-
tomas of different WHO grades to find out that the expression of these markers
positively correlated with an increase in the WHO grade of the astrocytomas.
Finally, in a very interesting study by Wang et al [520], the stem-cell-like
CD133 + fraction from 14 human glioblastomas was shown to include a subset
of vascular endothelial-cadherin (CD144)-expressing cells with characteristics
of endothelial progenitors capable of maturing into endothelial cells. They
conclude that a subpopulation of cells within glioblastoma can give rise to
endothelial cells via a CD133 + /CD144 + endothelial progenitor intermediate
included in the CD133 + cancer stem-cell-like fraction. This discovery opens
up important clinical options since the strong correlation of tumour grade and
neoplastic vasculature in human gliomas indicates that agents blocking the en-
dothelial transition of tumour cells may provide a novel therapeutic strategy.
Recognition of the forebrain SVZ astrocytes as the descendants of radial glia in
the adult brain and their capacity to act as NS cells in vitro raises the prospect
that this cell type might be responsible for tumour expansion, identifying it-
self with the cancer stem cell population of the cancer stem cell hypothesis.
Therefore, although it is not yet clear whether cancer-initiating events occur
in NS cells, progenitors or differentiated cells, NS cells are attractive candi-
dates. Their self-renewal abilities would allow an oncogene to more easily
initiate uncontrolled proliferation, and their potential for transformation has
been further considered based on the observations that human brain tumours
frequently arise deep in the brain near the SVZ. In p53 -/- mice, more prolifer-
ative activity is found in the SVZ and more neurospheres can be isolated from
this region, suggesting an expansion of the NS cell pool, which may make the
area more susceptible to neoplastic transformation [122]. Moreover, normal NS
cells were found in the CD133 + population of the normal human fetal brain,
again suggesting that the cell of origin of brain tumours may be a normal NS
71

3.3 Glioma Culture Systems Introduction
cell [458]. Much of the insight into brain tumour stem cells comes directly from
NS cell research because the discovery of such stem cells in the mammalian
brain has laid the foundations for designing experiments aimed at assessing
whether the hierarchy based on stem cells seen in adult healthy brain also
exists in human brain tumours and future investigations will hopefully clarify
where the brain tumour stem cell sits along the lineage hierarchy of cells [122].
In murine models for skin carcinogenesis, the tumour phenotype arising from
overexpression of HRAS, a human gene involved in the regulation of cell divi-
sion in response to growth factor stimulation, depended on the cell compart-
ment it occurred in, with suprabasal layers yielding benign tumours and hair
follicle bulge regions, the putative location for skin stem cells, yielding inva-
sive carcinomas. Along these lines, different cells of origin might give rise to
different types of brain tumours, with the more benignant ones arising from
restricted progenitors and the more aggressive ones from stem cells or early
progenitors. The cell of origin question is hampered by limited definition of
the normal NS cell hierarchy, especially by a lack of promoters that can spec-
ify gene expression in distinct compartments of the stem cell hierarchy and
of cell surface markers that can distinguish stem cells from multipotent or
lineage-restricted progenitors. Nestin, Sox2 and CD133 identify NS cells and
progenitors but no definite lineage-restricted progenitor cells have been identi-
fied. GFAP marks a rare NS cell population as well as differentiated astrocytes,
complicating the interpretation of these studies [122].
3.3 Glioma Culture Systems
Glioma Cancer Cell Lines
The historically adopted use of cancer cell lines to delineate tumour biology
and do preclinical drug screenings needs to be re-evaluated in light of the re-
cently discovered stem cell component of solid tumours, in order to assess how
well cancer cell lines reflect this characteristic amongst others with respect
to primary tumour cultures. Phenotypic characteristics and the multitude of
genetic aberrations found within repeatedly in vitro passaged cancer cell lines
often bear little resemblance to those found within the corresponding primary
human tumour and to this end, the study by Lee et al [261] attempted to find
a more biologically relevant model system for exploring glioma biology and for
72

3.3 Glioma Culture Systems Introduction
the screening of new therapeutic agents. In fact, little else is understood on the
similarities of glioma stem-like cells, human NS cells and the primary tumours,
besides the morphological similarities and the differentiation capacity of these
cells [156,458,543]. Therefore, Lee et al undertook a series of experiments to
identify and better characterise glioma tumour stem cells and their relation-
ship with the primary tumour and traditional glioma cell lines [261].
Firstly, most cancer cell lines including glioma, are grown in media containing
serum, unlike NS cell cultures that are serum-free since serum causes irre-
versible differentiation of NS cells [107,427]. In order to assess how primary
tumour cells are affected by the presence or absence of serum, single cell sus-
pensions of freshly resected and dissociated glioblastoma tissues were cultured
under conditions optimal for propagation of normal NS cells, termed "NBE"
conditions [107], as well as conditions optimal for growth of glioma cancer
cell lines, termed "serum" conditions. Under these two conditions, profound
biological differences were found in vitro:
· Cells in NBE conditions readily proliferated both as tumour spheres and
as an adherent monolayer, as is seen with normal NS cells. In contrast,
cells cultured in serum conditions formed a morphologically heteroge-
neous monolayer that within a month became homogeneous with a mor-
phology reminiscent of fibroblasts or epithelial cells.
· Cells cultured in NBE proliferated at a constant rate regardless of passage
number, whereas cells cultured in serum showed initial growth followed
by a plateau phase, only to eventually proliferate at a much greater rate
in later passages. To verify the extent of this behaviour, NBE cells were
influenced by addition of serum and recapitulated the growth pattern
seen in the serum cell population, while serum cells that were influenced
by addition of NBE culture media nearly ceased growing. This indicated
that the initial culture of primary tumour cells in serum conditions leads
to profound biological changes that cannot be subsequently reversed fol-
lowing transition to NBE conditions.
· Upon removal of EGF and FGF2 or addition of RA and serum, NBE
cells stop expressing NS cell markers Nestin, Sox2, and Stage-specific
embryonic antigen 1 (SSEA1) and start differentiation towards the glial
and neuronal lineages, although 40% of cells co-stained for glial and
neuronal markers together, suggesting that differentiation pathways in
these NBE cells are not entirely normal. Serum cells expressed few or no
73

3.3 Glioma Culture Systems Introduction
NS cell markers and did not respond to differentiation cues such as RA.
· In clonogenicity assays in which cells were tested for neurosphere forma-
tion upon single cell plating, NBE cells showed clonogenic frequencies
reminiscent of NS cell-derived neurospheres, whereas serum cells failed
to form neurosphere-like cells when plated in NBE conditions.
· The telomerase activities of NBE and serum-cultured cells differed in
that, although both cells retained telomeres as determined by fluores-
cence in situ hybridization (FISH), NBE cells had consistent telomerase
activity regardless of passage number, whereas telomerase activity was
lost when these cells were cultured in serum-containing media, consis-
tent with what occurs in normal NS cells. Likewise, early passage serum
cells did not have appreciable telomerase activity, which, however, they
gained back in later passages of exponential growth phase.
Taken together, these data demonstrate that NBE cells, as observed by the
two NBE cell lines followed in the study, contain many of the self-renewal
and differentiation characteristics of NS cells, whereas serum cultured cells do
not [261]. Other important characteristics tested on these two cell types were:
· Tumourigenic potential in vivo. Upon injection into the brains of
neonatal immunodeficient mice, NBE cells demonstrated retention of a
tumourigenic potential independent of passage number with as low as
1,000 cells, whereas serum-cultured cells at early passages did not show
any tumourigenicity. When established NBE cell-derived tumours were
dissociated and cultured under NBE conditions to be injected again into
the brains of new recipient mice, there was no loss of tumourigenic po-
tential, unlike when the same xenograft-derived cells were grown under
serum conditions and all subsequent tumourigenic potential was lost.
This suggested that the loss of tumourigenicity was due to the serum-
culture conditions. Interestingly, although early passage serum-cultured
cells did not form tumours, the late passage, exponentially growing,
telomerase-positive ones did at an increasing rate with progressive pas-
sage number.
· Phenocopy of the original human glioblastoma tumour. While
the intracranial tumours generated by NBE cells demonstrated exten-
sive infiltration along white matter tracts as observed in glioblastoma
patients, all the tumours generated from late passage serum cells were
74

3.3 Glioma Culture Systems Introduction
well delineated with little tumour cell infiltration, a characteristic pheno-
typically identical to the human tumour xenografts generated from the
standard glioma cell lines, demonstrating that only tumours derived from
NBE cells phenocopy the critically important histopathological features
of the original human glioblastoma tumours.
· Transcriptional activity. The gene expression profiles of NBE cells,
serum cells, their derived xenograft tumours, and the original glioblas-
toma tumours, show that the transcriptional landscape of NBE cells
and their derivative xenograft tumours is more closely related to that
of NS cells, and parental tumours, while the transcriptional landscape
of serum cells and their derivative xenografts is more closely related to
that of glioma cell lines and their derivative tumours. These data demon-
strate that NBE cells are remarkably similar to normal NS cells and their
derivative tumours properly maintain many biological characteristics of
the parental glioblastomas and other primary glioblastomas, whereas tra-
ditionally grown, serum-cultured cancer cell lines do not.
· Genomic changes. These were evaluated by performing SNP analy-
sis and spectral karyotyping (SKY) on NBE and serum-cultured glioma
cells. Deletion of the CDKN2A:ARF locus on chromosome 9, loss of
chromosome 10q, trisomy of chromosome 7, and local amplification of
the EGFR locus are common genomic features in primary glioblastomas
and they were found in all surgical samples. The serial genomic DNA
profiles of NBE and serum cultured cells at various passage numbers were
analysed by SNP analysis and showed that even after one year of main-
tenance (more than 70 passages) in NBE culture condition, NBE cells
largely maintained their parental tumour genotype, while serum cells
underwent significant genomic rearrangements as early as two months
of culture (less than 10 passages). Intriguingly, LOH in chromosomes 4
and 17 found in most of the late passage serum cells coincided with the
onset of increased proliferation, tumourigenicity and aneuploidy typical
of these cells, and carried the hCDC4 and p53 genes, respectively. The
fact that the remaining copy of hCDC4 was found to be down-regulated
hinted at the possibility that this E3 ubiquitin ligase involved in the
regulation of the aurora kinase STK15, up-regulated in these cells and
known to cause aneuploidy through inactivation [313,416], was partially
responsible for the increased proliferation, tumourigenicity, and aneu-
75

3.3 Glioma Culture Systems Introduction
ploidy observed in late passage serum cells. In addition, loss of the
wild-type p53 allele in late passage serum cells leaves only the mutant
p53 allele found in both the NBE cells and the parental tumours, caus-
ing a well-known genomic instability [130,153]. These genomic changes
further demonstrate the significant differences between serum cells and
their matched parental tumours [261].
In conclusion, by using a model system derived from primary glioblastomas,
Lee et al [261] have demonstrated that NBE-cultured cells derived from pri-
mary glioblastomas bear remarkable similarity to normal NS cells by retaining
the ability to form neurospheres in vitro; an indefinite self-renewal potential;
the ability to differentiate into the glial and neuronal lineages of the CNS; hav-
ing gene expression profiles similar to NS cells; bearing genetic stability over
many passages in vitro; harbour all of the genetic aberrations found within
the primary tumour; have gene expression profiles similar to the glioblastomas
they were derived from; appear to be the principal tumourigenic cell type that
can recapitulate the overall in vivo phenotype of the parental glioblastoma.
By contrast, cells derived from the same glioblastoma specimens but grown in
serum-containing media lose all the characteristics of primary tumour cultures,
although they ultimately regain the tumourigenic potential in later passages
without being able to recapitulate the tumourigenic phenotype of the original
tumour however, but rather matching the phenotypic and genotypic patterns
found in most glioma cell lines.
Since several groups have reported that tumour stem cell-like cells can be
isolated from the established tumour cell lines by culture in serum-free me-
dia with selected growth factors [242,385], experiments need to be carried out
to validate how these cells maintain their tumour stem cell-like properties in
differentiation-inducing conditions and if cells with stem-like properties may
emerge again through epigenetic reprogramming or selection of a subpopula-
tion of cells with genomic instability.
Based on their findings, Lee et al propose that the inherent tumour stem cell
population within primary glioblastomas is quickly lost in typical glioma cul-
ture conditions, and the cells found following prolonged in vitro passages are
the product of an outgrowth of a cell clone that has undergone profound de
novo genetic and/or epigenetic changes. This indicates that NBE cells may
be an optimal model system for understanding the biology of primary human
tumours, for the preclinical screening of agents and to guide personalized tu-
mour therapy. The table 3.1 below summarises the findings of this study [261].
76

3.3 Glioma Culture Systems Introduction
Table 3.1: Summary of characteristics of NBE and serum-cultured glioblastoma
cells. Adapted from Lee et al 2006.
Proliferation Constant
Clonogenicity,
tumourigenicity
Differentiation
potential
Telomerase activity Positive
tumour histology
Global gene expression
NSC-related genes
Genotype
Glioma Neural Stem Cells
NBE-cultured Serum-cultured
Limited growth, plateau, exponential
growth
Yes, regardless of passages Not at early passages
Induce to become glial
and neuronal lineages
Extensive migration, phenocopy
of primary human
GBMs
Similar to primary human
GBMs
Nestin, Sox2, CD133,
Musashi and Bmi
Same as parental tumour
regardless of passages
Do not respond to differentiation
stimuli
Negative initially, but became
positive at late passages
Fail to show infiltration like
glioma lines
Differentiation from primary
tumours but similar
to common glioma lines
-
Additional alterations not
found in parental tumour
The demonstration that the adult human brain maintains areas of radial
glia populations that have been shown to give rise to NS cells within the
SVZ, raised the prospect that these multipotent cells could be the alternate
cells responsible for glioma expansion to the differentiated glia in the brain
parenchyma [122,390,399]. As rare populations of stem cells are being discov-
ered in different tissues, the cancer stem cell hypothesis reinforces its statement
that cell lineage organization in tumours is hierarchical rather than stochastic
and only the sub-population of cancer stem cells is responsible for the expan-
sion of the tumour [123,458]. Therefore, in vitro expansion of the putative
brain cancer stem cells as stable cell lines would provide a powerful model
system to study the human disease by giving insights into the origin of tumour
heterogeneity and enable further analysis of the self-renewal, commitment, and
differentiation processes, which will hopefully lead to more targeted therapeu-
tic strategies [404].
77

3.3 Glioma Culture Systems Introduction
As mentioned in section 3.1, the neurosphere culture paradigm has been used
successfully for enrichment of tumour-initiating cells from brain tumours in-
cluding glioblastoma in serum-free media [156,261], an improvement on "clas-
sic" serum-cultured glioma cell lines that fail to model accurately the human
disease [404]. However, neurosphere culture has several limitations:
· short-lived progenitor cells also proliferate in suspension culture and true
clonal analysis is hampered by sphere aggregation;
· the spontaneous differentiation and cell death accompanying stem cell
divisions in the sphere environment limits rigorous assessment of stem
cell behaviour and marker analysis based on bulk populations;
· the true nature of the stem cell compartment across the spectrum of
gliomas and their relationship to in vivo progenitors remains unclear.
Unlike neurosphere culture, adherent culture provides uniform access to growth
factors, suppressing differentiation and enabling expansion of highly pure pop-
ulations of stem cells. The methodology reported by Conti et al [107], Pol-
lard et al [403] and Sun et al [481] in section 2.2 for deriving and expanding
mouse and human adherent NS cell lines in the presence of EGF and FGF2,
was adopted by Pollard et al [404] in an experiment designed to test whether
these same conditions enabled the isolation and expansion of stem cells from
gliomas. In these conditions, six cell lines were expanded for at least one year
and more than 20 passages, without any significant alteration in growth rate
or known GBM-related genetic aberrations. Cell lines were established from
histopathologically distinct types of tumour:
· three cases of glioblastoma multiforme: G144, G166 and GliNS2, with
the glioma sample from patient number 144 established as a biological
replicate in cell line G144ED, derived independently of G144 but using
the same initial tumour sample. In all analyses performed, no striking
differences in marker expression or behaviour were found between the
two cell lines;
· one case of giant cell glioblastoma: G179;
· one case of anaplastic oligoastrocytoma: G174.
These cell lines were phenotypically characterised by immunocytochemistry
to ensure their similarity to the fetal NS cells used as normal counterpart,
78

3.3 Glioma Culture Systems Introduction
which confirmed coherent expression of NS cell markers and neural progenitor
markers Vimentin, Sox2, Nestin, CD44 and 3CB2, and by time-lapse videomi-
croscopy, which confirmed the dynamic changes in cell shape and the high
motility typical of fetal NS cells, with G166 cells being less motile than G144
cells. Given these similarities the cell lines were termed glioma NS cell lines
or GNS cell lines [404]. In comparing the efficiency with which adherent GNS
cell lines could be established compared to suspension culture in neurospheres,
it was found that although most samples generated neurospheres upon initial
plating, only two cell lines could be passaged further, while in adherent condi-
tions the establishment of cultures was successful for all cell lines for at least 10
passages, possibly due to the increased differentiation and apoptosis processes
that take place in neurosphere culture [404].
To determine whether the GNS cells displayed genomic alterations charac-
teristic of glioblastoma, molecular cytogenetic analyses were performed us-
ing SKY, locus-specific FISH, and comparative genomic hybridization, which
showed G144 cultures exhibiting simple numerical gains of chromosomes 7 and
19, together with loss of chromosomes 6, 8, and 15. These aberrations, with
the exception of chromosome 8 deletion, are all commonly associated with
glioblastoma [326,383]. Furthermore, comparative genomic hybridization for
G144 revealed an amplification of the CDK4 locus on chromosome 12 and
deletion of PTEN on chromosome 10. However, late passage G144 cultures
did show a more complex and heterogeneous pattern of both numerical and
structural chromosomal change. Similarly, G179 exhibited a more complex
chromosomal constitution at later passages, and like in G144, polysomic gain
of whole chromosome 7 was evident. Thus, although GNS cells do not display
gross chromosome instability, alterations in whole chromosome copy number
do occur following long-term in vitro expansion, an issue that can be circum-
vented by routinely using cultures that are expanded for no more than 20
passages [404].
In order to test the capacity of GNS cells to initiate tumour formation, cells
from all GNS cell lines were injected intracranially into immunocompromised
mice that were sacrificed at 5 and 20 weeks, with the former scenario revealing
large numbers of engrafted human Nestin immunoreactive cells and the lat-
ter scenario showing the formation of large and highly vascularised tumours
histopathologically similar to human glioblastoma tumours. The cellular het-
erogeneity was highlighted by immunostaining the xenografts for Nestin and
79

3.3 Glioma Culture Systems Introduction
GFAP, and performing flow cytometry for CD133. Furthermore, transplanta-
tion of these cells after in vitro exposure to differentiation-promoting conditions
delayed tumour formation. In most transplants a striking infiltration of the
brain reminiscent of the human disease was observable with the exception of
cell line G166, which generated a more defined tumour mass. To determine
the tumour-initiating potential of the GNS cell lines, transplants were carried
using 10-fold dilutions starting with 100 cells, which were sufficient for cell en-
graftment of all GNS cells and for G144 and G174 were capable of generating
an aggressive tumour mass, which required 1,000 cells in the case of G166 and
G179. Unlike GNS cells, normal fetal NS cells never generated tumours at
any dilution. Finally, to determine whether the tumour-initiating cells could
self-renew within the xenograft, serial transplantations from the tumour mass
into secondary and tertiary recipients was carried out using G144, G144ED
and G179, showing in each case the generation of a tumour and thus demon-
strating that long-term expanded GNS cell lines remain highly tumourigenic
and are capable of forming tumours that appear to recapitulate the human
disease [404].
Since the important defining property of stem cells is their ability to gen-
erate differentiated progeny, the differentiation programs of the GNS cells
were evaluated keeping in mind that the prevalent form of glioma is the
GFAP + astrocyte-like cell-containing astrocytoma, which can also contain
anaplastic cell populations and, in some cases, an oligodendrocyte compo-
nent [301]. For all GNS cell lines, the differentiation to Octamer-binding pro-
tein 4 (Oct4) + oligodendrocytes or TuJ1 + neurons was fully suppressed in the
presence of EGF and FGF2, in contrast to what is observed in glioma neuro-
spheres. Upon growth factor withdrawal, NS cells differentiate into neurons
(see Fig 2.9), but in contrast to that, G144 and G179 GNS cells differentiated
into Oct4 + or CNPase +38 oligodendrocyte-like cells, and TuJ1 + cells, respec-
tively. Neuronal-like cells or oligodendrocytes were not apparent in G166 cul-
tures, which continued to proliferate in the absence of EGF and FGF2 without
clear differentiation. The tendency of G144 cells to differentiate into oligo-
dendrocytes was surprising because efficient oligodendrocyte differentiation of
mouse and human fetal NS cells requires exposure to thyroid hormone, ascor-
38 CNPase is a 2Õ, 3Õ-cyclic nucleotide 3Õ-phosphodiesterase that catalyses the in vitro
hydrolysis of 2Õ, 3Õ-cyclic nucleotides to produce 2Õ-nucleotides and has an in vivo function
that remains to be elucidated. High CNPase expression is seen in oligodendrocytes and
Schwann cells, accounting for roughly 4% of the total myelin protein in the CNS [137]
80

3.3 Glioma Culture Systems Introduction
bic acid, and PDGF [165]. Thus, the authors suggested that either G144 cells
represented a corrupted three-directional potent state with acquired genetic
changes influencing the lineage choice toward oligodendrocyte commitment,
or that they may have a distinct phenotype more similar to oligodendrocyte
precursor cells than NS cells. To distinguish between these two possibilities,
G144 was tested for the expression of established markers of oligodendrocyte
precursor cells, i.e. Olig2, Sox10, NG2, and PDGFRα prior to and during
differentiation, to find out that G144 cells stably exhibited an oligodendro-
cyte precursor-like phenotype expressing these markers prior to the beginning
of differentiation by growth factor withdrawal. In fact, upon re-examination
based on histopathology and CNPase staining, G144 was shown to have a
significant oligodendrocyte component even though originally diagnosed as a
glioblastoma [404].
In determining whether GNS cells could generate astrocytes, G144 and G179
were observed to undergo a striking change in cell morphology after seven days
from addition of BMP4 and removal of EGF and FGF2, the protocol used for
astrocyte differentiation of NS cells (see Fig 2.9). The majority of cells ex-
pressed high levels of GFAP, although the frequency was much lower for G166,
and cell lines G144 and G179 expressed low levels of the Doublecortin (Dcx) +
neuronal marker. This ensemble of events showed that although GNS cells
retain the capacity to differentiate, efficiency and lineage choice vary dramat-
ically between each line [404].
GFAP is expressed in radial progenitors and radial glia in the developing pri-
mate nervous system, as well as putative NS cells within the adult SVZ and,
specifically, at low levels in human fetal NS cell lines. Since an alternatively
spliced form, GFAPδ, was shown to mark human SVZ astrocytes [432], the
behaviour of the GNS cell lines with respect to GFAPδ was assessed. The
expression level of GFAPδ mRNA was five times greater in G179 than in
G144 and G166, with G179 cells also up-regulating GFAP expression following
BMP treatment and G144 cells remaining, instead, predominantly negative.
Co-expression of GFAPδ, Sox2, and Nestin was specific to G179 cells, which,
together with the ability to generate neuronal-like cells in vitro, are all fea-
tures conserved in adult SVZ astrocytes [443]. The fact that G166 cells lacked
the expression of GFAPδ and oligodendrocyte precursor markers, but differ-
entiated toward GFAP + astrocytes in vitro, suggested similarity to a more
restricted astrocyte precursor. Thus, despite their shared capacity to prolif-
erate in response to EGF and FGF2 and the widespread expression of neural
81

3.3 Glioma Culture Systems Introduction
progenitor markers, there are underlying differences between GNS cell lines
that may reflect the existence of distinct subtypes of regular neural progeni-
tors, and GFAPδ may be of use in discriminating astrocyte-like cells that have
stem cell properties [404].
To evaluate the relationship between each GNS cell line and their correspon-
dence to fetal NS cells, mRNA expression profiling was carried out using mi-
croarrays, and PCA 39 revealed that each GNS cell line had a transcriptional
state that more closely related to fetal NS cells than to adult brain tissue (Fig
3.6). Consistent with what was found about the various GNS cell lines in the
study by Pollard et al [404], the PCA analysis confirmed that G179 and G166
have a distinct expression profile, both from one another and to G174, G144,
and GliNS2. The microarray data also specifically confirmed the expression
Figure 3.6: PCA of global mRNA expression in each GNS cell line (black, G144,
G144ED, G166, G179, G174, and GliNS2), fetal NS cells (red, hf240, hf286, and
hf289), and normal adult brain tissue (blue). The "a" and "b" are biological replicates
of cell line hf240. Taken from Pollard et al 2009 [404].
of the oligodendrocyte precursor cell markers Sox8, Sox10, Olig1, Olig2, and
Nkx2.2 in G144, and the lower expression levels of the same markers observed
in G179, which, instead, showed higher levels of GFAPδ expression. The fact
that GliNS2 clustered closely with G144 and expressed the oligodendrocyte
precursor markers, suggested that the G144 phenotype may not be unique. The
G174 cell line, instead, derived from an oligoastrocytoma, clustered closely to
GliNS2 and G144 because it expressed higher levels of Olig2, although it failed
to express Sox10 or the pluripotency markers Oct4 and Nanog. Finally, G166
39 A common technique to find patterns in data of high dimension and expressing the data
in such a way as to highlight similarities and differences. Mathematically, it is a procedure
that uses an orthogonal transformation to convert a set of observations of possibly correlated
variables into a set of values of uncorrelated variables called principal components that are
less than or equal to the number of original variables.
82

3.3 Glioma Culture Systems Introduction
showed higher levels of EGFR than any other cell line, perhaps explaining its
resistance to differentiation upon EGF withdrawal or BMP treatment [404].
The CD133 and CD15 cell surface markers are expressed on fetal and adult
neural progenitors and brain tumour-initiating cells (see Section 3.2). For
G144 and G179, heterogeneity for both markers was observed, similarly to
fetal NS cells [480], whereas G166 was found to be negative consistently with
the microarray expression data. The fact that CD133 was not present on
all GNS cell lines confirmed that this marker does not universally identify tu-
mourigenic cells in malignant glioma. Unlike CD133 and CD15, the hyaluronic
acid-binding protein CD44, characterised as an astrocyte precursor marker and
recently demonstrated to also mark NS cells in vitro [401], was uniformly ex-
pressed in all GNS cell lines [404].
To demonstrate the proof of principle of the utility of GNS cells with respect to
the shortcomings of the neurosphere assay (see Section 2.2), a small-scale chem-
ical screen of known pharmaceutical drugs was carried out. Importantly, the
results of this screen extended to human brain cancer stem cells the observa-
tion that mouse neurospheres are sensitive to modulation of neurotransmitter
pathways, and future more in depth studies will have to determine whether the
drugs found in the screen that modulate the serotonin pathway of the adherent
GNS cell lines, can limit growth of xenograft tumours in vivo [404].
In their paper, Pollard et al [404] have tackled two critical issues of the brain
cancer stem cell hypothesis:
1. How to maintain and expand pure populations of cancer stem cells in
vitro by expanding them as cell lines using the adherent culture methods
previously established for fetal and human NS cells [107,481]. Specifi-
cally, Pollard et al have demonstrated that suspension culture is not a
requirement for successful long-term propagation of tumour-derived stem
cells, and that expansion in adherent conditions overcomes the limita-
tions of the neurosphere culture paradigm, such as increased levels of
differentiation and apoptosis.
2. Elucidation of the phenotypic similarities between GNS cells and the en-
dogenous progenitors within the developing and adult nervous system,
e.g. NEP cells, radial glia, glial progenitors, oligodendrocyte progeni-
tor cells, and SVZ astrocytes. For example, G144 cells strongly express
markers of the oligodendrocyte precursor cell lineage and are biased to-
ward oligodendrocyte differentiation, while G179 cells appear to be more
83

3.3 Glioma Culture Systems Introduction
similar to adult SVZ astrocytes, including expression of GFAPδ, although
more specific markers are needed in order to confirm this. This perhaps
indicates that the histological spectrum of different gliomas is dictated
by the phenotype of the underlying tumour-initiating cells.
To summarise, when the same culture conditions for NS cells were tested on
glioma tumours, NS-like cells were isolated and propagated as GNS cells. Nor-
mal NS cells and GNS cells feature many commonalities, both in their morphol-
ogy and immunoreactivity, to radial glia markers. However, unlike NS cells,
GNS cells require less exogenous growth factor for stable proliferation and reca-
pitulate the pathology of the original gliomas when xenografted into immuno-
compromised mice. This system is highly unusual in that normal and diseased
counterparts are morphologically and immunohistologically indistinguishable
and yet the differentiation behaviour of the cancer stem cells is clearly aber-
rant. The data generated using this system described by Pollard et al [404] is
consistent with tumour stem cells arising following transformation of oligoden-
droglial precursors or adult SVZ astrocytes, although it is considered equally
plausible that short-lived progenitors or differentiated cells can be converted
to a stem cell state through genetic or epigenetic disruptions. In conclusion,
GNS cells provide a versatile and renewable resource to probe the biology of
tumour-initiating cells and screen for agents that selectively and directly target
them. tumour stem cell self-renewal, migration, apoptosis, and differentiation
all represent potential therapeutic opportunities that are accessible in GNS
cell cultures [404].
84

Chapter 4
The Non-Coding RNA World
Contents
4.1 MicroRNA regulation . . . . . . . . . . . . . . . . . . . . . 86
4.2 Target Prediction and Validation . . . . . . . . . . . . . . . 90
Non-coding RNAs are a growing subset of RNAs whose sizes, regulatory func-
tions, range of regulated pathways and differential conservation within species
distinguish them from the remainder of the coding and non-coding transcripts
(mRNAs, tRNAs, rRNAs) classified as the traditional repertoire of RNAs ex-
pressed within a cell. This new set of regulatory molecules in the RNA world
is in continual expansion, as new types - unique in length, tissue specificity
and regulatory mechanisms - are discovered and classified in different species.
A broad distinction within this class of regulatory molecules is made in rela-
tion to their length, distinguishing them between long and short non-coding
RNAs (Fig 4.1). As a transcriptional class, long non-coding (nc)RNAs were
first described during the large-scale sequencing of full-length cDNA libraries
in the mouse [368], and are today arbitrarily considered longer than 200 nu-
cleotides due to a practical cut-off in RNA purification protocols that excludes
small RNAs. Their specific roles are still in the process of being unveiled,
although they already have been implicated in high-order chromosomal dy-
namics, telomere biology and sub-cellular structural organization [331]. While
this class of molecules has been most extensively studied in animal species,
studies of long non-coding RNAs in plants begin to emerge showing some con-
servation of mechanisms [31].
On the other hand, the first small non-coding RNAs discovered were endoge-
nous small-interfering RNAs (siRNAs) in plants [355,505], followed by mi-
croRNAs (microRNAs) in C. elegans [264], repeat-associated small interfering
85

4.1 MicroRNA regulation Introduction
Figure 4.1: Classes of non-coding RNAs (ncRNAs) discovered to date: siRNA,
small interfering RNA; microRNAs, microRNAs; rasiRNAs, repeat-associated small
interfering RNAs; piRNAs, piwi-associated RNAs; endosiRNAs, endogenous small
interfering RNAs; scnRNAs, scanRNAs; tasiRNAs, trans-acting RNAs.
RNAs (rasiRNAs) in D. melanogaster [24] and, only recently, piwi-associated
RNAs (piRNAs) [23,164,523] and endogenous small interfering RNAs (en-
dosiRNAs) in mouse [485,524]. Other classes are found in ciliates, such as
scanRNAs (scnRNAs) that ensure the fidelity of genome inheritance to the
next generation [336], and plants, such as trans-acting RNAs (tasiRNAs) [509].
Today siRNAs are known to act as gene silencers if exogenously introduced in
mammals and microRNAs are well-established gene regulators in plants and
animals, including the unicellular ciliate C. reinhardtii. Also, piRNAs are un-
derstood to play a role in zebrafish - widening the regulatory horizons of such
molecules - and rasiRNAs, now considered a subset of the piRNA class, have
also been found in zebrafish [199]. In this thesis the focus will be maintained
on the role of microRNA regulation in mammals.
4.1 MicroRNA regulation
The role of microRNAs has been studied in many different tissues and path-
ways, as well as through various time points in the development of organisms.
The first microRNA, lin-4, was identified in 1993 in a genetic screen for mutants
that disrupted the timing of post-embryonic development in C. elegans [129],
and today several hundred microRNAs are known to exist in the mammalian
86

4.1 MicroRNA regulation Introduction
genome regulating up to two thirds of protein coding genes [150,256]. The
way that microRNAs regulate gene expression is through an enzymatic pro-
cess that starts in the nucleus and is driven by Drosha. The biogenesis of
microRNAs in animal models starts when they are transcribed by RNA poly-
merase II as primary transcripts, termed "pri-microRNAs". The initiation
step of "cropping" is mediated by the Drosha-DGCR8 complex, also known as
the Microprocessor complex. Drosha and DGCR8 are both located mainly in
the nucleus. The product of this nuclear processing step is an approximate 70
nucleotide pre-microRNA, which possesses a short stem plus an approximate
two nucleotide 3' overhang. This structure might serve as a signature mo-
tif that is recognised by the nuclear export factor exportin-5. Pre-microRNA
constitutes a transport complex together with exportin-5 and the GTP-bound
form of cofactor Ran. Following export outside of the nucleus, the cytoplasmic
RNase III Dicer participates in the second processing step termed "dicing" to
produce microRNA duplexes. The duplex is separated and usually one strand
is selected as the mature microRNA, whereas the other strand is degraded.
When the other strand is not degraded it is indicated with the "*" mark in
microRNA target prediction algorithms [232].
Transcriptional and Post-transcriptional Regulation
MicroRNA genes are encoded either in independent transcription units, in
polycistronic clusters or within introns of protein coding genes. Indepen-
dently of Drosha, pre-microRNA hairpins can also be generated from introns
through the combined action of the spliceosome and the lariat-debranching en-
zyme [239]. Animal microRNAs have been shown to function differently from
plant microRNAs, with the former class binding to their target 3'UTRs by
imperfect matching. In fact, the RNA-Induced Silencing Complex (RISC) me-
diates target mRNA cleavage by loading siRNAs and mature microRNAs with
perfect sequence complimentarily to their target mRNA. The extent of the
mismatch region varies amongst different microRNA-mRNA duplexes in ani-
mals, but it is rarely the case that sequence complementarity spans the entire
microRNA sequence, meaning that transcript cleavage would seem to be just
as rare a mechanism of repression. Because the extent of this complementarity
is low in animals, there is a region spanning two to eight nucleotides from the
5' end of the microRNA that proves to be essential for the correct recognition
of the message target. The 5' most nucleotide (t1) of the microRNA sequence
does not have a significant role in this recognition process, as even when base
87

4.1 MicroRNA regulation Introduction
pairing occurs between t1, usually an adenine, and m1 on the messenger, usu-
ally a uridine, this feature alone does not contribute to a repressive effect on
the target [364].
Interestingly, structural studies of the Argonaute 40 (AGO)-siRNA complex,
demonstrated that the 5' most nucleotide of the guide RNA (equivalent to the
m1 base of microRNAs) is not base-paired but instead bound by AGO [291].
Due to its biological role, the region two to eight nucleotides from the 5' end
of the microRNA sequence is termed "seed" [273], which later proved an excel-
lent name choice in view of the conservation found for it across most metazoan
microRNAs [282]. A number of single nucleotide mutation studies monitoring
the effect of repression of microRNAs on their known targets also confirmed
the biological importance of the seed region [237,469]. The number and variety
of complementary seed sequences on the 3'UTRs of target genes determines
the potential for modulating their expression by different types and numbers
of microRNAs at various stages of the life of a cell [239].
Three types of seeds have been observed to date displaying a range of target-
ing efficiencies: "7mer-A1", lowest in the hierarchy, "7mer-m8", outperforming
it, and "8mer" sites with the strongest repressive effects. The 7mer-A1 type
matches over the entire seed sequence (seven positions) and carries an adenine
at position t1. The 7mer-m8 type matches over the entire seed sequence as
well but has an extra match at position 8 of the microRNA sequence, which
accounts for its outperformance of the 7mer-A1 type. Finally, the 8mer site has
a complete match over the entire seed sequence with an extra one at position
eight, but outperforms the 7mer-m8 site by carrying an adenine at position
t1 [359,446]. This relatively simplistic model, in which only the seed sequence
and its closest nucleotides play a role in determining the efficacy of repression,
has recently been refined to account for other features that seem to determine
the identity of a microRNA target. Examples of these are the presence of spe-
cific secondary structures, specific 3'UTR sequences surrounding the target site
and the extent of complementarity between the 3'UTR and the 3' end of the
microRNA sequence in order to compensate for the mismatching regions [239].
Another structural feature to account for are the G:U matches or "bulges"
that may form along the recognition sites of some microRNAs [70]. In vivo
experiments, in fact, seem to suggest that more than one G:U base pair within
40 Family of proteins that act as the catalytic component of the RNA-induced silencing
complex by using small RNA guides (microRNAs, siRNAs and piRNAs) to bind and cleave
target mRNAs.
88

4.1 MicroRNA regulation Introduction
the seed region could compromise the efficiency of target repression, a finding
recently confirmed by large-scale whole proteome microRNA impact analy-
ses [446,469]. However, extensive complementarity with the 3' end of the
microRNA sequence [469,538] and/or additional target pairing to microRNA
nucleotides 13-16 [43,180] seem to counteract the compromising effect of G:U
bulges on 7mer seed matches although these 3' compensatory and 13-16nt sup-
plementary sites, unlike the seed sequence, do not appear to be under selection
pressure [71,272]
Of equal importance to the quality of the microRNA-target sequence match, is
the overall accessibility of the target site [224], which may easily be obstructed
by RNA-binding proteins [222] or the possible un-pairing of the seed flank-
ing sequences rendered necessary by the secondary structures took on by the
3'UTR [224].
Various mechanisms have been documented to account for microRNA repres-
sion in animals: translational inhibition during the initiation or elongation of
protein synthesis, degradation of the nascent peptide chain, mRNA seques-
tration to P-bodies 41 and mRNA degradation. Even though understanding
the choice of which mechanism is used for different microRNA-mRNA du-
plexes is still a matter of controversy, it is a generally accepted concept that
features of the microRNA sequences and proteins bound to them play an im-
portant role in determining the method of suppression. The data available
today is indicative of two main types of regulation: indirect or with direct
effects on translation [359]. Amongst the transcriptionally indirect mecha-
nisms are mRNA degradation through deadenylation and decappying. The
direct effects on translation, instead, are divided into inhibition at the level of
translation initiation, in which few or no ribosomes occupy the mRNA that is
thought to be sequestered/stored in P-bodies, and inhibition post-initiation,
in which the silenced mRNAs sediment in the polyribosome fractions in a su-
crose gradient [239]. This latter type of inhibition can be further divided into
three scenarios: premature ribosome drop-off, stalled/slowed elongation and
co-translational protein degradation [359]. Data from many separate studies
support all the listed mechanisms, indicating that the inhibition scenario is a
very complex one. AGO2, for example, has been shown to bind to the m7G
cap of mRNAs through a domain similar in structure to the one on translation
initiation factor eIF4E. This observation suggests that the recruitment on the
41 Distinct foci within the cytoplasm of the eukaryotic cell consisting of many enzymes
involved in mRNA turnover.
89

4.2 Target Prediction and Validation Introduction
mRNA’s 3'UTR of AGO2 via microRNA pairing blocks m7G cap recognition
from the translation initiation factor, which is needed to start translation, thus
supporting the inhibition of initiation view of miR-mediated repression [239].
Even so, mRNAs inhibited by the action of miRs have been found to be associ-
ated with actively translating polysomes, reinforcing the idea that, depending
on the mRNA-miR duplex, a different subset of cases is selected for miR-
mediated repression [239].
4.2 Target Prediction and Validation
Initial efforts to characterise the properties of microRNAs focused on their
comprehensive identification in sequenced eukaryotic genomes, the numbers
of mRNAs targeted, and the degree of evolutionary conservation among dif-
ferent microRNA species. Whole-transcriptome and proteome analyses have
reinforced the importance of target site evolutionary conservation by demon-
strating a stronger down-regulation of the mRNAs having conserved sites
[37,139,359,446]. Although the surprising high 3'UTR homology shared among
vertebrates, very few target sites are conserved among the drosophilid and ne-
matode lineages [91] and only nine orthologous targets of miR-7 were found
to be shared between Drosophila and human (as an intersection from 97
Drosophila and 581 human predicted targets) [266]. Following the cataloguing
of microRNAs in a variety of organisms, subsequent work shifted towards the
prediction and characterisation of their mRNA targets and the functional con-
sequences of these regulatory interactions. In this context, a comprehensive
list of empirical rules for microRNA target prediction should not overlook the
importance of site accessibility.
A fundamental challenge in the target prediction field has been to successfully
predict the biologically functional targets while excluding false predictions [43].
To date, a number of sophisticated algorithms have been developed to address
these questions and at least seven of these tools are publicly available, each
one based on distinct criteria with varying false-positive rates. These tools
include: Targetscan [272,273], EMBL [470], PicTar [127,247], EIMMo [155],
miRanda [134,213], miRBase [179], PITA [224] and DIANA-microT [237,316].
Such methods differ considerably in their implementations of various predic-
tion criteria; most search for complementary sequences between microRNAs
and putative gene targets, while some consider physical and statistical hy-
bridization properties, cross-conservation of regulatory RNAs between related
90

4.2 Target Prediction and Validation Introduction
species, etc. As a result, prediction methods differ widely in their relative de-
grees of accuracy and coverage and the results produced often disagree.
With predictions in the range of 300 evolutionarily conserved targets per
mammalian microRNA family, it appears that microRNAs have the poten-
tial to modulate the expression of nearly all the mammalian mRNAs [43]. The
strong evolutionary pressure enacting the maintenance of the conserved tar-
get sites within most of the 3'UTRs [150] is avoided by some housekeeping
genes through the acquisition of exceptionally short 3'UTRs that are depleted
of target sites [71]. A fundamental step in the microRNA target prediction
pipeline is the experimental target validation of the predicted targets in order
to confirm the validity of an approach over another. Several validation meth-
ods have been employed, ranging from traditional genetic studies, rescue as-
says [70], reporter-gene constructs [237,273] and mutation studies [71,124,237].
In addition to confirming or discharging hypotheses built on networks of pre-
dicted targets, these validation approaches represent the real bottleneck of
the whole process since they are most time-consuming and expensive. High-
throughput approaches have also been developed, involving over-expression of
microRNAs in cell lines followed by microarray profiling to detect downregu-
lated targets [283] and the reverse approach of depleting microRNAs to identify
up-regulated targets [425].
Assaying only relative changes in target mRNA levels without measuring the
corresponding protein abundance is not sufficient to characterise all functional
targets [19]. Thus, it is necessary to demonstrate that in addition to medi-
ating the repression of gene expression through transcriptional degradation,
microRNAs also directly repress translation [446]. Since proteins have dif-
ferent turnover rates, a microRNA may require more or less time to change
their steady-state levels. With the new version of the Stable Isotope Label-
ing by Amino acids in Cell culture (SILAC) protocol developed by Rajwesky
et al [446], called pulse-SILAC (pSILAC), only differences in newly synthe-
sized proteins are detected by assaying the changes in their steady-state levels.
SILAC is a technique based on mass spectrometry that uses isotopic rather
than radioactive labelling of amino acids, to assay protein abundance in a cell.
In standard and pSILAC, two cell populations are grown in identical culture
media except for the presence, in one of the two cultures, of an isotope-labeled
amino acid. In standard SILAC the labeled amino acid is fed to the cell cul-
ture with the growth medium, so that it can be slowly incorporated into all
91

4.2 Target Prediction and Validation Introduction
newly synthesized proteins and the raw protein concentration is monitored.
In pSILAC, instead, the labeled amino acid is fed to the cell culture for a
short period of time (a pulse), so that only de novo protein production is
monitored [446]. Specifically, the pSILAC protocol developed by Rajwesky et
al [446] involves labelling two cell cultures (transfected and control) with a
heavy and a medium-heavy version of the same amino acid, respectively. Af-
ter 8h from transfection the cultures are pulse-labeled and the samples, united
after 32h post-transfection, are analysed by mass spectrometry. RNA from
the same samples (8h and 32h) was analyzed by Affymetrix arrays. In this
fashion, approximately 5,000 proteins were identified in HeLa cells with high
confidence. The search for six nucleotide motifs within the 3'UTRs of those
mRNAs whose protein levels decreased the most (never exceeding 4-fold) re-
vealed that the most significant motifs were exactly the seed sequences of
each respective microRNA. This showed that out of all proteins demonstrat-
ing reduced synthesis, the levels of the direct targets of microRNAs decreased
the most, and that this reduction is directly linked to the presence of the
3'UTR target site. Nevertheless, a number of repressed proteins without seed
sequences are still targeted, since their level is decreased, but prediction al-
gorithms cannot identify them because they are non-canonical and seedless.
Amongst the strongly repressed targets, a Gene Ontology search revealed that
these are mainly proteins synthesized on the Endoplasmic Reticulum. A po-
tential reason is that only mRNAs from cytosolic free ribosomes are targeted
to P-bodies for degradation [446]. Other observations were that a nine to 11
nucleotide mismatch along the mRNA-microRNA duplex was necessary for the
protein production to be repressed, being otherwise indistinguishable to that
of mRNAs lacking the seed. Although mismatches are deleterious to siRNA-
mediated mRNA cleavage, they seem to correlate with increased repression
of protein production by microRNAs. Also, the presence of multiple seeds
seemed to have multiplicative repressive effects with a higher impact when
the seeds were proximal rather than distant. On average, the repression was
found more pronounced for conserved rather than non-conserved seed sites,
indicating that there are more determinants other than the seed that mediate
efficient down-regulation of protein synthesis. Interestingly, as opposed to the
seed enrichment observed in the down-regulated genes, no seed enrichment was
observed in the up-regulated ones, strongly speaking against the microRNA-
mediated activation observed in other works [446].
92

Methods
93

Chapter 5
Methods
Contents
5.1 Tag-sequencing Data Processing . . . . . . . . . . . . . . . 94
5.2 Array Comparative Genomic Hybridization . . . . . . . . . 98
5.3 Differential Gene Expression . . . . . . . . . . . . . . . . . 100
5.4 Quantitative Real Time-PCR Validation . . . . . . . . . . . 101
5.5 Literature Mining . . . . . . . . . . . . . . . . . . . . . . . 107
5.6 Differential Isoform Expression . . . . . . . . . . . . . . . . 110
5.7 Differential Long ncRNA Expression . . . . . . . . . . . . . 112
5.8 Glioma Expression Signatures . . . . . . . . . . . . . . . . 112
5.9 External Dataset Expression Correlation . . . . . . . . . . 112
5.10 Glioblastoma Pathway Construction . . . . . . . . . . . . . 114
5.11 MicroRNA Target Prediction Analysis . . . . . . . . . . . . 117
5.1 Tag-sequencing Data Processing
Instead of the traditional microarray expression profiling, Solexa sequencing
was performed on total RNA from four GNS cell lines: G144ED, G144, G179
and G166, and two NS cell lines from fetal brain: CB541 and CB660 (Table
5.1). Tag-sequencing (Tag-seq) library preparation and sequencing was carried
out using the Illumina DGE Tag Profiling kit according to the manufacturer’s
instructions. We have submitted the data to ArrayExpress. The libraries
were specifically constructed with the longSAGE protocol, in which first and
second strand cDNA is synthesized with a biotinylated oligo deoxy-thymine
(oligo-dT) onto streptavidin beads. The bound cDNA is cleaved at the 5’ end
by an anchoring enzyme, such as NlaIII, that adds GTAC, the recognition
site for MmeI, to the poly-thymine (poly-T) strand. Sequencing adapters with
94

5.1 Tag-sequencing Data Processing Methods
Table 5.1: Summary of cell lines investigated with the Tag-seq platform. (M=Male,
F=Female)
Type of cell line Name of cell line Tissue Sex
GNS G144 GBM M
GNS G144ED GBM M
GNS G166 GBM F
GNS G179 Giant cell glioblastoma M
NS CB541 Fetal forebrain -
NS CB660 Fetal forebrain -
an MmeI recognition site are then linked to the cDNA and MmeI digestion
removes the 3’ portion of the cDNA from the beads, generating tags of con-
stant length of 17nt that are then ligated to a 3’ adaptor and sequenced, as
shown in figure 5.1 [490]. In the case of an Illumina sequencing platform, these
Figure 5.1: Step by step diagram of the longSAGE protocol used to generate reads
that contain tag sequences derived from the mRNA pool.
95

5.1 Tag-sequencing Data Processing Methods
constructs are hybridised onto the glass surface of a slide coated with comple-
mentary oligonucleotides and "bridge" amplified to then have them sequenced
with fluorescence-labeled nucleotide analogs [47]. Because each tag measures
expression levels that are not based on transcript length, a digit or "count" is
associated to it and the technology is often referred to as "digital gene expres-
sion" tag profiling [490].
The Tag-seq technology is based on the principles of solid-phase sequencing and
Serial Analysis of Gene Expression (SAGE). In this way, Tag-seq imports the
quantitative and unbiased transcriptome profiling proper of SAGE library con-
struction as well as the sequencing depth, dynamic range and cost-effectiveness
of the latest high-throughput sequencing platforms. Furthermore, Tag-seq is
not affected by the cross-hybridization and dynamic range limitations typical
of microarray technology, since no sequence-specific hybridization is required
for expression detection and the dynamic range is limited in principle only by
the sequencing depth of the platform [346,490]. Tag-seq is similar to RNA-seq
in that both technologies don’t require pre-existing knowledge of the genome
analysed and Tag-seq has proven to perform comparably in terms of gene
discovery and dynamic range [346,507]. Unlike RNA-seq, however, Tag-seq
strictly monitors the 3’ end usage of transcripts and carries no information on
their internal structure [194]. Finally, Tag-seq can be used as a strand-specific
gene expression platform, which is not always true of RNA-seq, depending on
the type of adaptor ligation performed [346].
We extracted tag sequences as the first 17nt of each read and counted the
number of occurrences of each tag in each sample. Due to sequence errors
introduced during library preparation and sequencing, highly expressed tran-
scripts might have caused a significant number of tags to differ by a single base
from the expected tag. To compensate for such errors, we used the Recount
program [528], setting the hamming distance parameter to 1. Recount uses an
expectation maximization algorithm to estimate "true" tag counts, i.e. counts
in the absence of error, based on observed tag counts and base call quality
scores.
We excluded tags matching adapters or primers used in library construction
and sequencing, as well as tags matching ribosomal RNA (rRNA) or mitochon-
drial sequence. Adapter and primer contamination was identified by running
96

5.1 Tag-sequencing Data Processing Methods
the TagDust program [258] with a target false discovery rate (FDR) of 1%.
Tags matching ribosomal sequence were identified by using the Bowtie pro-
gram [257] to align against a database consisting of all rRNA genes, including
pseudogenes, from Ensembl 56 [147] and all ribosomal repeats in the UCSC
Genome Browser RepeatMasker track for genome assembly GRCh37 [152];
only perfect matches to the extended 21nt tag sequence, consisting of the
NlaIII site CATG followed by the observed 17nt tag, were accepted. Mito-
chondrial tags were similarly excluded by searching for perfect matches to the
mitochondrial chromosome sequence. The parameters used to run the Bowtie
program were the following:
-f parameter indicates that the query input files are FASTA files;
-n parameter set to 0 means that the alignments may have no more than n
mismatches. Since we are only looking for perfect matches, n=0.
-y parameter specifies to the program to try as hard as possible to find valid
alignments when they exist, which makes running this mode much slower;
-k 2 instructs the program to report up to two valid alignments;
-m 1 instructs the program to refrain from reporting any alignments for reads
having more than one reportable alignment. This option is useful when
the user wants to guarantee that reported alignments are unique, which
is our case.
To assign tags to genes, we employed a hierarchical strategy based on the ex-
pectation that tags are most likely to originate from the 3’-most NlaIII site in
known transcripts. Tags were assigned to transcripts using virtual tag data
from the SAGE Genie database [65] and virtual tags extracted from Ensembl
transcript models. The SAGE Genie annotation consisted of 105 sets of virtual
tags obtained by scanning for NlaIII sites in cDNAs from RefSeq, Mammalian
Gene Collection (MGC) and Genbank, then Expressed Sequence Tags (ESTs),
UniGene consensus sequences and transfrags. The virtual tag sets are further
classified based on the position of the tag relative to the 3’ end of the transcript
and indicators of 3’ end reliability such as polyadenylation signal and poly-A
tail. Since SAGE Genie does not cover Ensembl transcripts, we also extracted
virtual tags from the 3’-most cut site in each Ensembl transcript. We used the
CATG recognition sequence as a "signal" that indicated where to start count-
ing 17nt ahead. This allowed us to extract that 17nt tag and know exactly from
97

5.2 Array Comparative Genomic Hybridization Methods
which transcript it was extracted. If the cut site was less than 17nt from the
end of the transcript, we extended the sequence with adenine stretches, rep-
resenting the poly-adenine (poly-A) tail. In these cases, additional upstream
tags were also extracted until one tag fully contained in the Ensembl cDNA
sequence was obtained. We disregarded Ensembl transcripts not belonging
to a gene of biotype "protein_coding" or "processed_transcript". Ensembl
annotates genes of several other biotypes, e.g. pseudogenes and small RNAs,
but those annotations are not based on full-length transcript sequences, so
we would not expect to find valid virtual tags in those transcripts. For the
majority of this study, we used a conservative subset of the virtual tags from
SAGE Genie and Ensembl comprising 25,593 unique tags assigned to 15,103
genes (Table 5.2). Specifically, we used SAGE Genie tags extracted from to the
3’-most cut site in RefSeq or MGC cDNAs having a poly-A tail or a polyadeny-
lation signal, and Ensembl tags from transcripts of type "protein_coding" or
"non_coding". Any virtual tags that mapped to multiple loci by these criteria
were excluded. For certain analyses, we made use of more comprehensive vir-
tual tag sets. In addition, we determined unique, perfect matches for tags to
the genome using Bowtie as described above. We calculated a single expression
value for each gene in each cell line by summing the counts of tags assigned to
the gene.
5.2 Array Comparative Genomic Hybridization
We re-analysed the array comparative genomic hybridization (CGH) data de-
scribed by Pollard et al [404] CGH was performed with Human Genome CGH
Microarray 4x44K arrays (Agilent), using genomic DNA from each cell line
hybridised in duplicate (dye swap) and normal human female DNA as ref-
erence (Promega). Log2 ratios were computed from processed Cy3 and Cy5
intensities reported by the software CGH Analytics (Agilent). We corrected
for effects related to GC content and restriction fragment size using a modi-
fied version of the waves array CGH correction algorithm [271]. Log2 ratios
were adjusted by sequential loess normalization on three factors: fragment GC
content, fragment size, and probe GC content. These were selected after in-
vestigating dependence of log ratio on multiple factors, including GC content
in windows of up to 500 kilobases centred around each probe. The Biocon-
ductor package CGHnormaliter [506] was then used to correct for intensity
dependence and log2 ratios scaled to be comparable between arrays using the
98

5.2 Array Comparative Genomic Hybridization Methods
Table 5.2: Classification of sequenced tags in each cell line.
G144ED G144 G166 G179 CB541 CB660
Sequenced tags 6,383,175 7,133,520 13,415,402 11,610,415 12,103,066 10,043,561
Filtered tags 751,698 11.78% 912,971 12.80% 378,009 2.82% 347,537 2.99% 327,597 2.71% 382,819 3.81%
Adapter 594,513 9.31% 765,484 10.73% 90,407 0.67% 45,318 0.39% 39,992 0.33% 248,799 2.48%
Mitochondrial 156,534 2.45% 147,245 2.06% 285,881 2.13% 302,148 2.60% 287,493 2.38% 133,935 1.33%
Ribosomal RNA 651 0.01% 242 0.00% 1,721 0.01% 71 0.00% 112 0.00% 85 0.00%
Tags assigned to a single locus 2,812,750 44.07% 2,640,558 37.02% 6,271,471 46.75% 5,603,364 48.26% 5,984,114 49.44% 3,708,853 36.93%
Reference tags, unique 1,420,009 22.25% 1,344,879 18.85% 2,970,554 22.14% 2,805,423 24.16% 2,894,539 23.92% 1,712,143 17.05%
Reference tags, best 628,707 9.85% 577,418 8.09% 1,783,685 13.30% 1,267,594 10.92% 1,485,211 12.27% 982,190 9.78%
cDNA tags, unique 146,442 2.29% 142,369 2.00% 295,764 2.20% 284,990 2.45% 261,411 2.16% 171,970 1.71%
99
cDNA tags, best 34,999 0.55% 34,472 0.48% 103,632 0.77% 105,787 0.91% 81,547 0.67% 49,374 0.49%
Other SAGE Genie tags, unique 345,759 5.42% 322,034 4.51% 652,449 4.86% 684,718 5.90% 749,653 6.19% 455,045 4.53%
Other SAGE Genie tags, best 72,658 1.14% 60,725 0.85% 158,601 1.18% 107,955 0.93% 148,854 1.23% 84,542 0.84%
Tags not mapping to known transcrip-
164,176 2.57% 158,661 2.22% 306,786 2.29% 346,897 2.99% 362,899 3.00% 253,589 2.52%
tome but uniquely to genome
Ambiguously mapping tags 205,626 3.22% 185,115 2.60% 608,978 4.54% 418,267 3.60% 437,887 3.62% 388,085 3.86%
Reference tags 165,895 2.60% 148,512 2.08% 516,134 3.85% 337,710 2.91% 366,115 3.02% 288,253 2.87%
cDNA tags 5,996 0.09% 4,464 0.06% 10,566 0.08% 6,455 0.06% 6,708 0.06% 40,908 0.41%
Other SAGE Genie tags 8,581 0.13% 6,717 0.09% 25,914 0.19% 14,668 0.13% 13,553 0.11% 12,508 0.12%
Tags not mapping to known transcrip-
25,154 0.39% 25,422 0.36% 56,364 0.42% 59,434 0.51% 51,511 0.43% 46,416 0.46%
tome but to multiple genomic locations
Unclassified tags 2,613,101 40.94% 3,394,874 47.59% 6,156,944 45.89% 5,241,247 45.14% 5,353,468 44.23% 5,563,803 55.40%

5.3 Differential Gene Expression Methods
"scale" method in the package limma [466]. Replicate arrays were averaged
and the genome (GRCh37) segmented into regions with different copy number
using the circular binary segmentation algorithm in the Bioconductor package
DNAcopy [510], with the option undo.SD set to 1. Aberrations were called
using the package CGHcall [503] [37] with the option nclass set to 4. CGH
data are available from ArrayExpress [28] under accession E-MTAB-972.
5.3 Differential Gene Expression
We called the differentially expressed genes with the Bioconductor package
DESeq that uses pseudo reference-based normalization [21]. DESeq compares
sequencing-derived expression profiles using a method that is able to account
for large variation between biological replicates, as can be expected with can-
cer samples. The R code that generated the differential expression calls is
reported below, with ## signs indicating a comment:
## Define auxiliary functions
read.gns.counts

5.4 Quantitative Real Time-PCR Validation Methods
resVarB=de$resVarB, counts)
rownames(x)

5.4 Quantitative Real Time-PCR Validation Methods
PCR. This gene set comprises 82 validation targets from Tag-seq analysis, eight
glioma and developmental markers, and three endogenous control genes - 18S
ribosomal RNA, TUBB and NDUFB10. The 18S gene was chosen because
used by ABI as a manufacturing control, while TUBB and NDUFB10 were
selected because they are present in our Tag-seq dataset and show low varia-
tion of expression across GNS and NS cell lines in an independent microarray
expression dataset [404]. The 93 genes were interrogated using 96 different
TaqMan assays (three of the validation targets required two different primer
and probe sets to cover all known transcript isoforms matching differentially
expressed tags). cDNA was generated using SuperScript III (Invitrogen) and
real-time PCR carried out using TaqMan fast universal PCR master mix. The
absence of a no-template control (NTC) amplification (horizontal slope) en-
sured that random contamination and reagent contamination were not affect-
ing our samples. In complying with the minimum information for publication
of quantitative real-time PCR experiments (MIQE) [78], a full assay list with
raw and normalised threshold cycle (Ct) values is provided in Appendix A.3.
The data analysis was performed using the R package HTqPCR [131], which
handles high-throughput qPCR data with a focus on data from Taqman low
density arrays. Ct values were normalised to the average of the three control
genes and potential outliers identified and filtered out using the plotCtBoxes
function (Fig 5.2). Figure 5.3 shows the effect of the normalisation method
Figure 5.2: Boxplot of normalised Ct values identifies outliers (empty circles).
102

5.4 Quantitative Real Time-PCR Validation Methods
Figure 5.3: Correlation scatter plot showing the effect of the normalization method
on the raw Ct values measured for the three endogenous controls.
employed on the endogenous controls rearranging the Ct values to a smaller
dynamic range. To capture biological variability within cell lines, we measured
up to four independent RNA samples per line (indicated with the letters A,
B, C, D in tables A.3 and A.4 in Appendix A.3). Figure 5.4 shows that the
concordance between all our A, B biological replicates is very high and few
outliers can be observed - identified with a gene name and highlighted in red
in the figure. Differentially expressed genes were identified by the Wilcoxon
rank sum test after averaging replicates.
103

5.4 Quantitative Real Time-PCR Validation Methods
104

5.4 Quantitative Real Time-PCR Validation Methods
105

5.4 Quantitative Real Time-PCR Validation Methods
Figure 5.4: Scatter plots of each of our A and B biological replicates shows a high
degree of concordance between samples with very few outliers (highlighted in red).
In order to assess the repeatability, or intra-assay variation, of the expression
levels measured with qRT-PCR for each of the N=96 genes, standard devia-
tions (σ) were computed for the differences in gene expression levels between
two replicates (Fig 5.5). Standard deviations ranged between 3.5

5.5 Literature Mining Methods
Figure 5.5: Dot plot of the standard deviations (blue) of the differences between
expression levels in two replicates. Standard deviations were computed for each
combination of replicate cell lines.
5.5 Literature Mining
A web script (provided in Appendix B) was implemented to automatically pro-
cess the querying of the 748 differentially expressed genes (F DR < 10%) and
therefore establish their role, if any so far had been assigned, in glioblastoma
or other cancers.
The script was modelled against a browser query frame type in the Visual
Basic.NET programming language, which I deemed the best coding choice for
this data-hefty application because of its "collection" objects, i.e. extremely
powerful hash tables that make use of an ultra-fast key and vector content
look-up. Moreover, the presence of the .NET Framework enriches this particu-
lar edition of Visual Basic with extensive client-provider and internet browsing
libraries that I expected to need for this particular application.
The code I developed implements a process of hierarchical look-up, recapit-
ulated in figure 5.6, that starts with the most glioblastoma-specific database
amongst GBMbase [162], Google Scholar, Google Search, BioGraph [280], in-
formation Hyperlinked Over Proteins (iHOP) [197] and PubMed [410], and
loops all the way to the least-specific database. In this way, the probability of
finding an association between the queried gene and glioblastoma diminishes
at every iteration and is unlikely to exist by the time the execution arrives
107

5.5 Literature Mining Methods
at the last and least-specific of the databases. This algorithm maximises the
probability of finding a reported connection between any gene and any disease,
prompted minor disease-specific details are changed such as the identity of the
first database, which needs to be the most disease-specific of the six.
If we call "case" the code that searches for the association between queried
gene and glioblastoma in one database, and "iteration" the search loop com-
pleted from first to last database, then during each case a weighted-index is
accumulated to become, at the end of one iteration, a total weighted index, i.e.
a score of how well that particular gene did in the search for its association to
glioblastoma in the six databases. Therefore, the total weighted index repre-
sents those parameters used in the look-ups that yielded the most favourable
results, i.e. which database successfully found an association between the gene
and glioblastoma, how many hits were found, and in which part of the paper
the hits were found. In fact, I assigned a greater weight to the associations
found in the title, with respect to those found in the abstract, with respect to
those found in the contents of a paper. In designing the weight structure I also
decided to favour the associations found in the same sentence to those found
more than one sentence apart. The latter, in fact, barely carry any relevance
in the value cumulation process of the total weighted index.
In this particular application it was interesting to know whether the queried
gene was implicated in any type of cancer, especially in case the literature
did not report an association with glioblastoma. Therefore, at every iteration,
the databases are also queried with parameters that identify the answer to
the question "Is this gene implicated in any cancer?" and a separate index
from the total weighted index is determined. To calculate this cancer index
I searched for an association between the gene being queried and any of the
cancers listed in a database that I compiled for this particular application. All
the cancers listed in the National Cancer Institute A to Z List of Cancers [206]
are present in the cancer database as a unique set of names. A search was per-
formed in PubMed for every combination of queried gene symbol and cancer
name from the compiled cancer database and, if the two terms appeared in the
same sentence, this was considered an association and the cancer index for that
combination incremented by one. At the end of the search, every gene-cancer
combination possessed a cancer-index and the association was reported if the
value of such index was above an arbitrarily chosen threshold value.
108

5.5 Literature Mining Methods
How many hits?
I = 0
Category 4
I ≧ 1
Category 3
Which cancers?
Yes
Is this gene associated
with cancers?
No
Is this gene
associated with GBM
in GBMbase?
No
Is this gene associated
with GBM in iHOP?
No
Is this gene associated
with GBM in
PubMed?
No
Is this gene
associated with GBM
in BioGraph?
No
Is this gene
associated with GBM
in Google Scholar?
No
Is this gene
associated with GBM
in Google Search?
Yes
Yes
Yes
Yes
Yes
Yes
How many hits?
1 ≦ I < 10
Category 2
How many hits?
1 ≦ I < 10
Category 2
How many hits?
1 ≦ I < 10
Category 2
How many hits?
1 ≦ I < 10
Category 2
How many hits?
1 ≦ I < 10
Category 2
How many hits?
1 ≦ I < 10
Figure 5.6: In this diagram the "cases" (code that searches for the association
between queried gene and glioblastoma) and "iterations" (search loop completed
from first to last database) structures in the execution code are made obvious. At
every question the answers identify the parameters that are used to assign the queried
gene to one of the four categories or skip the assignment for that particular case and
go through the rest of the iteration. I = number of associations found for the gene
symbol of the queried gene and the word "glioblastoma" or "GBM".
At the end of an iteration the total weighted index and the cancer index for
every gene-cancer combination were available and, depending on their values,
the queried gene was assigned to one of four categories, which answer one of
the following questions:
Category 1. Does extensive literature implicate the gene in GBM?
Category 2. Does a limited amount of literature implicate the gene in GBM?
Category 3. Is there no literature implicating the gene in GBM, but does it
appear to be implicated in other cancers?
109
I ≧ 10
I ≧ 10
I ≧ 10
I ≧ 10
I ≧ 10
I ≧ 10
Category 1
Category 2
Probability of finding an associa2on with GBM

5.6 Differential Isoform Expression Methods
Category 4. Is there no literature implicating the gene in any cancer?
If W = total weighted index and C = cancer index, then if W > 10 (internal
parameter, does not appear in fig 5.6), the gene is assigned to category one; if 1
< W < 10, the gene is assigned to category two; if W = 0 and C > 10 (internal
parameter, does not appear in fig 5.6) then the gene is assigned to category
three; finally, if W = 0 and C = 0 then the gene is assigned to category four.
5.6 Differential Isoform Expression
With the aim of establishing Tag-seq as a sensitive technique for the iden-
tification of 3'UTR transcript isoforms that are differentially expressed be-
tween GNS and NS cell lines, we evaluated the differential expression of 4,727
genes with multiple expressed tags using two alternative methods: the non-
parametric χ 2 test, and the logarithmic algorithm adapted from Morrissy et
al [346]. Non-parametric methods are applied when the assumptions for para-
metric methods about the underlying distribution of the data are not met,
namely that the data is normally distributed or that sample size is large enough
to support the assessment of the normality distribution. Whilst the normal-
ity assumption is satisfied when tested against the 28,351 tags mapping over
16,025 genes in the six libraries altogether, sample size drops to a range of two
to 14 tags per gene when assessing differential expression over a single tran-
script isoform. With a sample size smaller than 30, the use of non-parametric
methods is often advisable. Due to the nature of the SAGE protocol, tags map
at the 3'UTR of a gene and identify potential splicing isoforms. We conducted
a χ 2 test with the chisq.test function in R on all tags mapping on to the same
gene of the 12,794 tags expressed at higher than 10 tpm in at least one GNS
and one NS library.
In the second approach, following the logarithmic ratio method adapted by
Morrissy et al [346], the isoform expression between the GNS and NS cell lines
was evaluated by analyzing the expression of tags that mapped either twice on
the same gene or, if more than twice, for all their pair combinations. Again,
only tags with a minimum expression of 10 tpm in at least one GNS and one
NS library were considered (Table 5.3). The ratio of normalised expression
for each tag pair was calculated in each GNS and NS library, and multiplied
by the natural logarithm of the difference between the expression of each tag
in the pair, ensuring that the more highly expressed tag pairs would also be
more highly ranked (equations 5.4 and 5.5). For pairs of genes with positive
110

5.6 Differential Isoform Expression Methods
Table 5.3: Summary of statistics using the χ 2 and logarithmic tests.
Non-parametric
χ 2 test
Logarithmic
ratio
Number of tags and tag-pairs mapped (2-14/gene) 12,792 13,599
Number of genes identified by tags 4,727 4,541
Number of multiple mapping tags at p< 0.01 8,169 3,651
Number of genes differentially expressed between isoforms
identified by multiple mapping tags at p< 0.01
2,682 2,040
ratio changes, the ratio was higher in GNS cell lines compared to NS cell lines,
while pairs with negative values of ratio changes had a ratio that was higher in
NS cell lines rather than GNS cell lines (equation 5.7). Gene pair ratio-change
values ranged from -17.4 to 17.7 and no p-values were produced. The equations
used are shown below, where S1 and S2 stand for any two isoforms identified
by tag mapping:
S1
S2 = ln[ln(S1expression − S2expression) × ( S1expression
)] (5.4)
S2expression
If the second isoform is more highly expressed than the first, the ratio is cal-
culated as:
If
S1
S2 = −1 × ln[ln(S2expression − S1expression) × ( S2expression
)] (5.5)
S1expression
0 < (S1expression − S2expression)|(S2expression − S1expression) < 1 (5.6)
then the pair was omitted.
For each tag pair, the average and standard deviation was computed for all
modified means in GNS cell lines, and separately for all means in NS cell lines.
An overall measure of the change in the ratio between GNS and NS cell lines
was computed as shown below:
δ GNS
NS
= ln(µGNS ) − ln(
σGNS
µNS
) (5.7)
σNS
Dividing each mean by the standard deviation ensures that tag pairs with
lower variance in their ratios are ranked higher than tag pairs with a higher
variance [346]. The logarithmic algorithm uses the mean and standard devi-
ation statistics to describe the ratios of expression change between the GNS
and NS lines. Unlike the non-parametric method, this algorithm produces a
ratio change for each pair of tags rather than for the sum over all tags and the
111

5.7 Differential Long ncRNA Expression Methods
absolute value tells the fold change whilst the sign the direction of the change.
The 13,599 tag-pairs analysed pertained to a total of 4,541 genes, a number
closely matching the 4,727 genes assessed with the non-parametric method.
By ranking the tag-pairs according to magnitude of ratio change and setting a
threshold at two-fold, the number of tag-pairs decreased to 3,651 but the num-
ber of genes being represented was still substantially high at 2,040, a number
comparable to the 2,682 genes identified in the non-parametric method.
5.7 Differential Long ncRNA Expression
We called differential expression with the Bioconductor package DESeq for all
tags that mapped to a unique location in the reference human genome with the
Bowtie program and filtered the results to exclude protein-coding transcripts.
This analysis revealed 25 ncRNAs with strong evidence of differential expres-
sion between GNS and NS lines. It is unlikely that these transcripts represent
artifacts of the Tag-Seq procedure, because all coincide with published cDNA
sequences or ESTs.
5.8 Glioma Expression Signatures
The Tag-seq data described in Parsons et al [383] was kindly provided by
Dr. N. Papadopoulos. Following quality control, we excluded three out of 21
sequencing lanes. To treat the expression data in a similar manner as Verhaak
et al [511], Tag-seq expression values were transformed by computing log fold
change relative to the mean expression across the glioma tumour and xenograft
samples. Only those genes identified by Verhaak et al [511] as signature genes
for a subtype were used in the correlation calculations for that subtype. To
obtain robust fold change estimates, the gene sets were further limited to those
genes having a normalized expression level above 25 tags per million in any
sample. This reduced the original set of 210 signature genes per subtype to 158,
167, 183 and 164 for "proneural", "neural", "classical" and "mesenchymal"
subtypes, respectively. P-values were computed with the R function cor.test.
5.9 External Dataset Expression Correlation
We used Affymetrix Exon 1.0 ST microarray data from The Cancer Genome
Atlas (TCGA; http://cancergenome.nih.gov; [326]) for 397 glioblastomas
112

5.9 External Dataset Expression Correlation Methods
and 10 non-neoplastic brain samples (dataset GBM). In addition, we used
Affymetrix HG-U133A and HG-U133B data for 24 grade III and 50 grade IV
gliomas from Freije et al [148] and 21 grade III and 55 grade IV gliomas from
Phillips et al [390] (dataset HGG), For the survival analysis the dataset from
Gravendeel et al [176] and that from Murat et al [353] were used (Table 5.4).
All tumour microarray data were from primary glioma samples obtained at ini-
Table 5.4: Public gene expression datasets used in trying to establish tumour
expression correlations with the differentially expressed genes found through Tagseq
and with the clinical data available from TCGA in the survival analysis.
Citation Accession Microarray Number of cases
platform GBM Grade Other Grade Wt**
(Affymetrix) III III I-II brain
astrocyt* glioma glioma
TCGA [326] n.a. Exon 1.0 ST 397 0 0 0 10
Phillips et al
[390]
GSE4271 U133A,B 55 21 0 0 0
Freije
[148]
et al GSE4412 U133A,B 50 8 16 0 0
Gravendeel et
al [176]
GSE16011 U133 Plus 2.0 141 16 66 27 0
Murat
[353]
et al GSE7696 U133 Plus 2.0 70 0 0 0 0
Gravendeel et al. described 269 samples obtained at histologic diagnosis, from which
we excluded 15 containing mostly non-neoplastic tissue and four lacking survival
data; n.a.=not applicable; *astrocytoma, **Non-neoplastic.
tial diagnosis. For dataset GBM, we used processed (level 3) data from TCGA,
consisting of one expression value per gene and sample. For the survival anal-
ysis datasets and the HGG dataset, the raw microarray data was processed
with the Robust Multi-chip Average (RMA) method as implemented in the
Bioconductor package affy [161] and probe-gene mappings were retrieved from
Ensembl 68 [146]. For genes represented by multiple probesets, expression
values were averaged across probesets for randomisation tests, heatmap vi-
sualisation and GNS signature score calculation. Differential expression was
computed using limma [464]. Randomisation tests were conducted with the
limma function geneSetTest, comparing log 2(F C) for the sets of core up- or
down-regulated genes against the distribution of log 2(F C) for randomly sam-
pled gene sets of the same size.
Survival analysis was carried out with the R library survival. To combine
expression values of multiple genes for survival prediction, an approach inspired
113

5.10 Glioblastoma Pathway Construction Methods
by Colman et al [105] was taken. The normalised expression values xij, where i
represents the gene and j the sample, were first standardised to be comparable
between genes by subtracting the mean across samples and dividing by the
standard deviation, thus creating a matrix of z-scores:
zij = xij − ¯xi
SD(xi)
(5.8)
Using a set U of nU genes up-regulated in GNS lines and a set D of nD genes
down-regulated in these cells, we then computed a GNS signature score sj for
each sample j by subtracting the mean expression of the down-regulated genes
from the mean expression of the up-regulated genes:
sj =
i∈U
zij
nu
−
zij
nD
i∈D
(5.9)
IDH1 mutation calls for TCGA samples were obtained from Firehose data run
version 2012-07-07 [143] and data files from the study by Verhaak et al updated
2011-11-28 [408].
5.10 Glioblastoma Pathway Construction
The pathway map was created in Cytoscape 3.0 [451]. Cytoscape is an open-
source platform for complex network analysis and visualisation of network
data, such as molecular interaction networks or biological pathways, that can
be integrated with annotations, gene expression profiles and any other type
of useful data. Cytoscape is available for download at www.cityscape.org. In
a Cytoscape network nodes represent objects, i.e. proteins, and connecting
edges represent relationships between objects, i.e. a protein’s physical inter-
action with another protein. Once this basic network is laid out, attributes
can be assigned to nodes and edges to help the visualisation of categories of
objects and types of relationships, respectively (Fig 5.7). Cytoscape networks
can become extremely complex as layers of attributes are applied to nodes and
edges in the form of different colours, shapes, thicknesses and other graphical
features that end up representing an ever-increasing amount of biological infor-
mation pertaining to that network. Finally, cytoscape-generated networks can
be analysed with the use of "plugins", pieces of independent software devel-
oped for specific applications on Cytoscape such as graph analysis, clustering,
ontology analysis, etc. All plugins can be found at aps.cytoscape.org.
114

5.10 Glioblastoma Pathway Construction Methods
Figure 5.7: Schematisation of a typical Cytoscape network look, with nodes and
edges representing proteins and the type of interaction between them, respectively.
The colour and shape of the nodes can be attributed to different characteristics of the
proteins, such as their family, enzymatic activity, post-translational modifications,
etc. The colour and shape of the edges can be attributed to the type of interactions
between proteins, such as activating, inhibiting, coenzymatic, etc. In this example
protein A and B interact to form a complex that activates protein D, also repressed
by protein C, that once active can go ahead and activate protein E.
A series of pre-defined "layouts" available in Cytoscape, i.e. algorithms that
automatically lay a network out by generating arrangements of nodes and
edges that either make it look a specific way, such as the circular, grid and
hierarchical layouts, or use the attribute information as a guide, such as the
attribute circle, degree sorted circle, force-directed, group attributes layouts.
For our purposes we used the edge-weighed force-directed layout, also known
as "biolayout" that, once generated, was manually re-adapted to obtain a best
fit for image generation. In order to graph the network, the edge-weighed
force-directed layout, or biolayout, uses an algorithm that minimises the en-
ergy of the model by starting with an initial layout, where the positions of the
nodes are randomly assigned. Then, in every iteration, the algorithm tries to
improve the layout according to the energy model using the first derivation of
the energy function to compute a direction and a distance for the movement
of each node. Since the graphs generated are large, the minimising algorithms
do not carry a high complexity per iteration value. The algorithm of Barnes
and Hut [42] is used in Cytoscape’s biolayout for this purpose.
The glioblastoma pathway was constructed manually by integrating the infor-
mation on nodes (genes) and edges (types of interactions between genes) from
a variety of glioblastoma pathways found in the literature with the purpose of
115

5.10 Glioblastoma Pathway Construction Methods
generating an all-encompassing network containing all the known interactions
that are important in glioblastoma. A fundamental rule that I followed was to
check that every protein interaction had been experimentally validated, via lit-
erature research and the use of the BioGRID protein interaction database [62]
and the IntAct molecule interaction database [205], before adding it to the
pathway. If it was not the case then the interaction (edge) and proteins in-
volved (nodes) were not included in the final glioblastoma pathway.
The information integrated to generate the final glioblastoma pathway was
taken from four different sources:
· The pathway database of the Kyoto Encyclopedia of Genes and Genomes
(KEGG) [3], a collection of manually drawn pathway maps;
· GBMbase [162], a collection of signalling pathways in glioblastoma based
on the ones described in the TCGA project [326] including a searchable
archive of glioblastoma gene publications;
· "Automated Network Analysis Identifies Core Pathways in Glioblas-
toma" by Cerami et al 2010 [86];
· "Malignant Astrocytic Glioma: Genetics, Biology, and Paths to Treat-
ment" by Furnari et al 2007 [154].
I also included the information from important pathways in signal transduction
that are known to be specifically affected in glioma, namely:
· The Mitogen-activated protein (MAP) kinase signalling cascade, taken
from three sources:
– The Cell Signalling Technologies website [1]
– "Mitogen-activated protein (MAP) kinase pathways: regulation and
physiological functions" by Pearson et al 2001 [388]
– The pathway database of the Kyoto Encyclopedia of Genes and
Genomes [3]
· The p53 pathway, taken from three sources:
– The Panther Pathway repository [5]
– The pathway database of the Kyoto Encyclopedia of Genes and
Genomes [4]
– The Biocarta pathway archive [58]
116

5.11 MicroRNA Target Prediction Analysis Methods
· The Rb1 pathway "RB tumour suppressor/checkpoint signalling in re-
sponse to DNA damage" from the Biocarta pathway archive [60]
· The Pten pathway "PTEN dependent cell cycle arrest and apoptosis"
from the Biocarta pathway archive [59]
The complete pathway has a total of 238 nodes and the colour mapping
is based on fold change absolute values and directions taken from the dif-
ferential gene expression analysis (see Results section 6.4).The colour map-
ping was performed using the VistaClara Cytoscape plug-in [234] that assigns
colours to nodes matching expression data imported with the "Import At-
tribute/Expression Matrix File" function of Cytoscape. The assignment is
done through the sigmoid function:
y =
2
− 1 (5.10)
1 + e-sx where the value of the constant "s" determines the rate of change in the colour
gradient, with smaller values allowing for the colour to ramp very gradually,
and larger values bringing the sigmoid function closer to a step function with a
very abrupt colour change around the values −1 < y < 1. The log 2(F C) values
range between −11 < F C < 11 and a value of s=1 was applied, with darker
reds and greens indicating smaller absolute values of log 2(F C), and brighter
colours indicating greater ones. A pathway image with the same colour coding
and gradient used in the tumour correlation heatmap (Fig 7.4), has also been
generated (yellow and blue shades) to correlate the expression values of the
differentially expressed genes between the two analyses.
5.11 MicroRNA Target Prediction Analysis
In trying to answer the question "Is a subset of prediction algorithms better
at predicting than the single?" we used exon array data and microRNA mi-
croarray data from a cohort of GNS cell lines that was made available to us.
The candidate did not perform this analysis but rather used the data gath-
ered from it in the ensemble analysis of microRNA target predictions. The
exon microarray data was analysed in R using software packages from the
Bioconductor project [465]. Background correction, quantile normalization
and calculation of probe-set expression values from fluorescence data was per-
formed using the Robust Multi-chip Average (RMA) method as implemented
in the affy package [161]. We used the xmapcore system, based on the earlier
117

5.11 MicroRNA Target Prediction Analysis Methods
exonmap package and x:map database [369], to associate microarray probesets
with protein-coding genes annotated in Ensembl 58. Probeset filtering was
applied such that all probe sequences were required to map to exonic gene
components. We estimated the expression level of a gene as the median nor-
malised intensity over its associated probe-sets. Differential gene expression
was computed using the limma package [465], where statistical significance was
determined with a moderated eBayes test and the resulting p-values adjusted
using the FDR method. The microRNA microarray data were background
corrected using the method described by Edwards et al [132] and implemented
in the limma package [465]. Following least-variant set normalization with de-
fault parameters [482], differential expression was computed using the limma
package.
118

Results
119

Chapter 6
Digital Transcriptome Profiling
Contents
6.1 Clinical Data . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.2 Tag mapping . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.3 Copy Number Aberrations . . . . . . . . . . . . . . . . . . 125
6.4 Core Differentially Expressed Genes . . . . . . . . . . . . . 129
6.5 Large-scale qRT-PCR Validation . . . . . . . . . . . . . . . 136
6.6 Literature Mining for Differentially Expressed Genes . . . . 142
6.7 Isoform Differential Expression . . . . . . . . . . . . . . . . 144
6.8 Long ncRNA Differential Expression . . . . . . . . . . . . . 157
6.1 Clinical Data
The clinical data available for our cell lines are somewhat limited due to pa-
tient consent forms prohibiting surgeons from revealing it. However, we do
know that all of our GNS cell lines are clinically classified as primary glioblas-
tomas, not secondary. Consistent with this, the PCR experiments performed
by Steven Pollard on our GNS cell lines to establish the presence or absence
of IDH1 and IDH2 mutations found no evidence of any mutation in their re-
spective loci (oral communication). In the paper by Pollard et al [404] the
genetics of our GNS cell lines are described using SNP6.0 arrays, which iden-
tified "classic" mutations, in particular frequent loss of CDKN2A:ARF, gains
of CDK4 and general glioblastoma-pertinent gene instability. The age of the
patients is described in the paper and summarised in table 6.1 below.
120

6.2 Tag mapping Results
Table 6.1: Summary of the available clinical data for our GNS cell lines (M=Male,
F=Female, n.a=not applicable).
Type of Name of Tissue Type Sex IDH1 IDH2 Patient
cell line cell line mutation mutation age (years)
GNS G144 GBM Primary M No No 51
GNS G144ED GBM Primary M No No 51
GNS G166 GBM Primary F No No 74
GNS G179 Giant cell GBM Primary M No No 56
NS CB541 Fetal forebrain n.a. n.a. n.a. n.a. n.a.
NS CB660 Fetal forebrain n.a. n.a. n.a. n.a. n.a.
6.2 Tag mapping
We created one Tag-seq library per cell line and obtained between 6 and 13
million sequence reads from each (Table 6.2). Every read was formed by a first
sequencing primer, the 17nt tag, and a second sequencing primer in this order
(Fig 6.1).
Table 6.2: Summary of reads per cell line library.
Type of cell line Cell line Number of reads
GNS G144 7,133,520
GNS G144ED 6,383,175
GNS G166 13,415,402
GNS G179 11,610,415
NS CB541 12,103,066
NS CB660 10,043,561
Figure 6.1: Diagram of the construct generated by the longSAGE protocol sent to
the Illumina sequencer for the final step of Tag-seq.
Once the read sequences were received as FASTA files, we first extracted the
17nt tags they contained, then filtered these tags and, finally, aligned them
to the genome. The entire process is summarised as a diagram in figure 6.2.
Firstly, the 17nt tags were extracted out of each read, separating the primer
sequences from the tag sequences that actually interested us. Secondly, each
tag sequence was counted and the resulting counts adjusted for sequencing and
library preparation errors that might have caused highly expressed transcripts
to give rise to a significant number of tags differing from the expected tag se-
quence by just one base. Secondly, the reads that were not going to map onto
121

6.2 Tag mapping Results
Figure 6.2: Step by step diagram of the extraction, filtering and mapping phases
for reads and tags. The extracted tag population is coloured with different shades of
grey that represent a different origin (adapter, mitochondrial or ribosomal tag). In
the mapping phase the diagram to the right represents the human reference genome
and some of the filtered tags (light grey) are mapping onto some known transcripts
(gene A, gene B) and non-coding regions.
the reference genome (adapter tags formed by the ligation of two sequencing
primers, and mitochondrial RNA), as well as rRNA sequences, were filtered
out. As shown in figure 6.3, on average more than 90% of tags remained un-
filtered, meaning that most of the sequenced data were available for us to use
in subsequent analyses, the first one being alignment to the reference genome.
Finally, the tag sequences contained within the recounted and filtered reads
were mapped in two complementary ways. Notice that, in order to maximise
the ability of the aligner to effectively map our short tag sequences to the ref-
erence genome, we included the CATG recognition site of the MmeI anchoring
enzyme, at the 5’ of every tag sequence; this generated 21nt tag sequences that
were used for alignment to the reference genome. For each library, we mapped
the so-generated pool of 21nt tags to the human reference genome, as well as
to a virtual tag-to-gene library that we assembled from already existing tag-to-
gene libraries and a complementary one that we generated programmatically.
The tag mapping strategy we adopted was a hierarchical one, as summarised in
figure 6.4. This allowed us to generate sets of decreasing stringency - regarding
the number of tags mapping to known transcripts - that were best fit for differ-
122

6.2 Tag mapping Results
123
Figure 6.3: Proportion of tags after the steps that filter for adapter contamination, mitochondrial or rRNA tags are concluded. Values are
normalised across all cell lines.

6.2 Tag mapping Results
ent analyses. We first mapped all the tags obtained from the process described
in fig 6.2 to known transcripts, then we mapped the remaining tags to cDNA
libraries. Finally, we mapped the remaining tags to the genome for the poten-
tial discovery of new unannotated transcripts. As a result of this strategy, we
collected in "Ref_uniq" - our most stringent set - all the tags mapping to a
single reference gene and in "Ref_best" - our second most stringent set - all
the tags included in the former with the addition of tags mapping to multiple
genes of which, however, only one was a reference transcript. The conservative
set of tag-to-transcript mappings ("Ref_uniq") comprised 25,593 unique tags
assigned to 15,103 genes, and formed our primary dataset for further anal-
ysis. In addition, we created several more comprehensive tag mapping sets,
by including mappings to other cDNA sequences, expressed sequence tags,
internal NlaIII sites and unannotated genomic regions ("cDNA_uniq" and
"cDNA_best").
Figure 6.4: Diagram of the tag mapping strategy where each filtered tag is assigned
to a more or less stringent set used for future analyses.
As shown in figure 6.5, on average more than 55% of tags (green) in every cell
line could not be classified as belonging to any of the sets identified with the
124

6.3 Copy Number Aberrations Results
hierarchical tag assignment process described above and summarised in fig 6.4.
This showed that most of the transcriptional activity in our cell lines could not
be represented by known transcripts or cDNAs, but perhaps that much of the
regulation might be happening at the non-coding transcriptional level. On av-
erage, in every cell line, less than 45% of tags mapped perfectly to the reference
nuclear genome, i.e. the tags contained within the "Ref_uniq", "Ref_best",
"cDNA_uniq" and "cDNA_best" sets in the non-green portions of fig 6.5.
This indicated that almost half of the transcriptional activity detected in our
cell lines could be traced back to known gene functions, giving us the possi-
bility to correlate gene expression in our cell lines with, for example, known
cancer pathways or other data from similar experiments. Of these 45% tags,
on average 32% were specifically assigned to high-quality reference transcript
models from the widely used Ensembl, RefSeq and Mammalian Gene Collec-
tion (MGC) databases (i.e., belonged to the "Ref_uniq" and "Ref_best" sets).
Reassuringly, other Tag-seq studies have reported similar figures [194,346].
Given that glioblastoma is a heterogeneous tumour, and that our samples were
taken from three distinct glioblastoma patients, we performed an analysis to
observe whether any correlation existed between the cell lines (Fig 6.6) that
would reinforce or weaken that knowledge. We observed the following, calcu-
lating a single expression value for each gene in each cell line by summing the
counts of all tags assigned to that gene: the correlation between the techni-
cal replicates G144 and G144ED was the highest (Pearson R = 0.94), as was
assumable since the two cell lines were derived from the same tumour but in
different laboratories (it should be noted that differences between these repli-
cate measurements can be due to a number of factors in the procedure from
cell line establishment to tag sequencing); the correlation between the two NS
cell lines was also high (R = 0.87); any correlation between GNS cell lines
(G166, G179 and G144) showed larger differences in gene expression profiles
(R ranging from 0.78 to 0.82) than the correlation between the two biological
replicates or the two NS cell lines, as was to be expected since the GNS cell
lines originated from histologically distinct tumours while the NS cell lines
supposedly bore a wild type transcriptional activity.
6.3 Copy Number Aberrations
Since the observed differences in gene expression can be caused by a number
of factors such as chromosomal aberrations (including rearrangements, losses
125

6.3 Copy Number Aberrations Results
126
Figure 6.5: Proportion of filtered tags that, after alignment to the reference genome, cDNA libraries and virtual tag-to-gene libraries, are assigned
to a set. Values are normalised across all cell lines.

6.3 Copy Number Aberrations Results
Figure 6.6: Correlations for all combinations of cell lines: any GNS/GNS and
NS/NS cell lines.
and gains of DNA) and alterations in transcriptional regulation and mRNA
degradation, we analysed array comparative genomic hybridization (aCGH)
data for G144, G166 and G179 [404]. The candidate did not perform this
analysis but the results will be shown nonetheless to give the reader a clearer
understanding of the genomic identity of these cell lines. The aCGH method
reveals net gains and losses in a cell population over the average ploidy. Previ-
ous analysis of chromosomal alterations in G144, G179 and G166 by SKY and
CGH detected several alterations, mostly whole-chromosome gains and losses,
and found that the lines do not show gross chromosomal instability when cul-
tured. Specifically, the CGH data for G144 revealed the deletion of PTEN on
chromosome 10 [404].
These findings were confirmed by our analysis of the aCGH data (Fig 6.7).
We identified aberrations known to be common in glioblastoma, including
gain of chromosome 7 (EGFR, which is over-expressed in 40% of glioblas-
tomas [154,262], lies on chromosome 7) and losses of large parts of chromosomes
10, 13, 14 and 19 in more than one GNS line, as well as focal gain of CDK4
in G144 (arrow, chromosome 12 in figure 6.7) and focal loss of the CDKN2A-
CDKN2B locus in G179 (arrow, chromosome 9 in figure 6.7) [52,326]. We
found that both G166 and G179 had low copy numbers of chromosome 10, but
could not establish a focal PTEN loss as we did for CDK4.
127

6.3 Copy Number Aberrations Results
Figure 6.7: The image summarises the results of the CGH arrays performed on
cell lines G144, G166 and G179. Dots indicate log2 ratios for array CGH probes
along the genome, comparing each GNS line to normal female DNA. The coloured
segments indicate gain (red) and loss (green) calls, with colour intensity proportional
to mean log2 ratio over the segment. The X chromosome was called as lost in G144
and G179 because these two cell lines are from male patients; sex-linked genes were
excluded from downstream analyses of aberration calls. The red segments above the
grey line at zero indicate gains and red segments below it indicate losses over the
entire genome.
On a global level, we found a correlation between aberrations and expression
levels, but this trend was modest (Fig 6.8a). Among the 459 autosomal genes
that we found to be up-regulated in GNS relative to NS lines by Tag-seq,
there was a clear enrichment of gains (p = 0.002 with Fisher’s exact test) and
depletion of losses (p = 0.002) compared to all autosomal genes expressed in
the GNS or NS lines. Down-regulated genes showed an opposite, but weaker,
trend (Table 6.3, Fig 6.8b). We also observed that many of the 29 genes that
were found to distinguish GNS lines from NS lines by qRT-PCR had associ-
ated aberration calls that suggested their expression levels may be dictated
by a loss or a gain in their copy numbers (Fig 6.8c). Overall, these results
Table 6.3: Significance of the correlation found between CNAs and expression levels
measured with Fisher’s exact test (p-value).
Gains Losses
Up-regulated 0.002163 0.002286
Down-regulated 0.06997 0.4257
suggest that copy number changes are a significant cause of the observed gene
expression changes. However, other factors may be more important, because
only a minority of up-regulated genes (21%) showed evidence of gains. Thus,
128

6.4 Core Differentially Expressed Genes Results
regulatory changes and other alterations not detectable by aCGH, such as bal-
anced translocations and small mutations, likely play a major role in shaping
the GNS transcriptome.
Figure 6.8: (a) Curves show distributions of expression level differences between
GNS and NS lines, stratified by aberration calls. The distributions for genes in segments
without aberrations (neutral) peak near the zero mark, corresponding to an
equal expression level in GNS and NS lines. Conversely, genes in lost and gained
regions tend to be expressed at lower and higher levels, respectively. In each plot,
log2(FC) is computed between the indicated GNS line and the mean of the two NS
lines, and capped at (-8, 8) for visualisation purposes. To obtain robust FC distributions,
genes with low expression (< 25 tpm) in both GNS and NS conditions were
excluded; consequently, between 6,014 and 6,133 genes underlie each plot. (b) For
each of the three gene sets listed in the legend (inset), bars represent the percentage
of genes with the indicated copy number status. (c) Aberration calls for the 29 genes
that were found to distinguish GNS from NS lines by qRT-PCR. Circles indicate
focal (< 10 Mb) aberrations; boxes indicate larger chromosomal segments.
6.4 Core Differentially Expressed Genes
To identify genes with differing expression between GNS and NS cells, we used
the Bioconductor package DESeq on the three GNS lines G144, G166 and
G179, and the two NS lines CB541 and CB660. The DESeq package, unlike
its predecessor edgeR, uses a method whose core assumption is that the mean
is a good predictor of the variance, which implies that, for a given distribution
of genes with similar expression levels, the variance across replicates will be
similar. Given this assumption, this method developed by Simon Anders [21]
estimates a function that predicts the variance from the mean by calculating
the sample mean and variance for each gene within replicates, and then fitting
129

6.4 Core Differentially Expressed Genes Results
a curve to this data. With the diagnostic plot shown in fig 6.9 we validated that
the estimates of the single gene variance functions followed the empirical vari-
ance well enough, as indicated by the red line representing the local regression
fit, even though the spread of the single gene variance values is considerable,
as one should expect given that each variance value is estimated from just
four values. Having estimated and verified the variance-to-mean dependance,
Figure 6.9: Plot of the estimates for each gene of the variance against the base
levels, i.e. the count value for a tag divided by the total number of counts. The red
line represents the fit from the local regression.
we then proceeded to look for differentially expressed genes using the DESeq
package function nbinomTest. With this function we generated the following
values for each gene: the mean expression level, as a joint estimate between
conditions "N" (normal) and "T" (tumour) and as a separate estimate for each
condition, the fold change (FC) from condition N to T, the natural logarithm
(Ln) of the fold change, and the p-value for the statistical significance of this
change. The p-adjusted value is also computed by the nbinomTest function
to adjust the p-value for multiple testing with the Benjamin-Hochberg pro-
cedure, which controls the FDR. We first plotted the computed fold changes
against the mean values and coloured in red the genes that were significant
at 1% FDR (Fig 6.10). Interestingly, these genes seemed to cluster at higher
values of the mean and absolute value of Ln(FC), indicative of gene expression
130

6.4 Core Differentially Expressed Genes Results
levels that were coherently higher across all GNS versus NS cell lines, or all
NS versus GNS cell lines. At an FDR of 10%, this analysis revealed 485 genes
Figure 6.10: Plot of the fold change computed with the DESeq package function
nbinomTest versus the mean for the contrast of the two conditions "N" (normal) and
"T" (tumour). The red genes are significant at 1% FDR.
to be expressed at a higher average level in GNS cells (up-regulated) and 254
genes to be down-regulated (see Appendix A.1). To discern genes that are
consistently up- or down-regulated among the GNS lines compared to the NS
lines, and thus capture major gene expression changes common to G144, G166
and G179, we set strict criteria on fold changes and tag counts requiring that:
1. each GNS line show at least a two-fold change compared to each NS line,
with the direction of change being consistent among the cell lines;
2. an absolute expression level above 30 tpm in each GNS line for up-
regulated genes or each NS line for down-regulated genes.
This stringent approach yielded 32 up-regulated and 60 down-regulated genes,
in the following referred to as strictly up- and down-regulated genes, respec-
tively, or "core" differentially expressed genes (Table 6.4).
131

6.4 Core Differentially Expressed Genes Results
Table 6.4: Table of genes (alphabetical order) with large expression changes common
to the GNS lines G144, G166 and G179, relative to the normal NS lines CB541
and CB660, resulting from filtering with stringent criteria the differentially expressed
genes at 10% FDR.
Gene Entrez Differential expression results Normalised tag counts
symbol gene ID log 2(F C) FDR G144ED G144 G166 G179 CB541 CB660
ADD2 119 6.32 5.1E-04 66.4 91.9 103.1 288.2 0.0 4.1
ATP1A2 477 -6.16 1.3E-08 39.0 37.5 0.6 0.0 471.0 1346.4
BACE2 25825 7.97 5.5E-11 227.5 423.1 253.4 468.2 0.0 2.7
C10orf11 83938 3.82 1.7E-03 195.7 603.7 175.1 152.8 20.5 23.6
C5orf13 9315 -3.29 4.7E-03 230.2 162.6 55.9 234.9 1985.9 962.7
C9orf125 84302 -3.78 6.1E-04 73.9 59.0 3.6 8.0 508.5 148.1
C9orf64 84267 7.88 5.3E-10 114.7 156.8 164.1 752.5 2.9 0.0
CA12 771 -4.65 9.7E-05 45.9 28.1 22.6 30.3 945.6 415.1
CACNG8 59283 -4.77 5.7E-05 33.1 28.3 3.9 2.6 405.1 220.9
CCND2 894 -3.98 9.3E-05 158.0 125.2 0.6 0.0 827.8 496.4
CD74 972 7.11 7.9E-13 3942.1 2107.8 93.4 1427.2 8.4 9.1
CD9 928 4.87 2.7E-06 1058.3 1082.4 396.5 189.3 4.8 32.9
CEBPB 1051 3.29 5.9E-03 106.7 155.6 552.5 270.9 36.7 30.2
CHCHD10 400916 3.70 9.0E-04 844.2 713.0 523.9 457.9 28.4 58.8
CTSC 1075 3.55 1.0E-03 937.2 731.4 311.7 1499.9 10.3 133.2
CXXC4 80319 -4.84 1.2E-03 23.5 21.1 0.0 0.0 219.5 200.8
DDIT3 1649 4.40 4.6E-05 933.6 1345.3 748.9 250.5 26.9 47.3
DNER 92737 -3.89 1.8E-03 90.5 523.2 9.3 847.4 8865.3 4813.8
DTX4 23220 -5.92 4.6E-05 6.8 7.8 3.2 0.0 128.8 312.5
EDA2R 60401 -5.58 7.6E-04 1.4 6.2 9.7 0.0 306.7 208.8
EPDR1 54749 3.09 5.6E-03 310.9 269.0 739.5 319.0 20.3 82.3
FAM38B 63895 -Inf 5.5E-11 0.0 0.0 0.0 0.0 778.6 166.8
FAM69A 388650 3.68 2.2E-03 513.2 393.1 155.0 439.3 27.9 23.3
FBLN2 2199 -Inf 2.0E-07 0.0 0.0 0.0 0.0 232.1 117.5
FOXG1 2290 3.30 6.8E-03 445.0 505.5 104.5 137.4 0.0 50.8
FOXJ1 2302 -4.19 3.6E-03 4.9 3.1 4.4 27.8 254.8 175.8
FUT8 2530 3.32 4.6E-03 666.0 999.8 609.6 2352.7 95.8 168.6
GABBR2 9568 -9.50 1.2E-13 0.0 0.0 5.6 0.0 2569.7 143.1
GPR158 57512 -4.55 9.9E-06 103.8 117.4 0.0 53.3 2311.0 354.9
GRIA1 2890 -4.42 4.2E-05 33.1 68.7 2.7 73.4 1780.2 292.7
HLA-DRA 3122 5.68 6.6E-07 9744.0 5434.9 138.4 270.5 57.4 19.2
HMGA2 8091 -5.68 3.7E-05 0.0 0.0 0.0 23.7 233.3 576.3
HOXD10 3236 Inf 3.9E-11 77.0 145.9 115.3 578.2 0.0 0.0
IL17RD 54756 -4.38 8.0E-04 25.5 17.5 9.8 15.3 364.6 230.4
IRX2 153572 -4.65 7.2E-04 0.0 0.0 0.0 28.9 370.0 111.1
KALRN 8997 -5.25 1.9E-04 0.0 6.3 2.5 7.1 302.9 106.2
KCTD12 115207 -4.77 4.0E-03 48.7 16.9 10.1 54.5 1349.0 147.8
LAMA2 3908 -5.21 4.4E-06 47.3 25.2 8.0 2.6 730.2 149.0
LGALS3 3958 6.44 6.1E-07 2062.3 3254.0 2538.7 134.8 32.1 14.3
LMO2 4005 -4.82 2.5E-04 20.3 21.2 0.0 4.9 134.6 369.0
LMO3 55885 -Inf 1.0E-08 0.0 0.0 0.0 0.0 136.9 446.0
LMO4 8543 3.80 5.0E-04 553.8 1311.2 455.5 1149.8 16.7 122.5
LPAR6 10161 -3.62 5.6E-03 29.8 33.8 39.6 0.0 395.9 211.3
LUM 4060 -4.26 3.2E-03 0.0 0.0 110.9 0.0 507.4 897.7
LYST 1130 5.73 3.5E-09 126.7 180.8 284.0 548.6 10.3 2.1
MAF 4094 -4.79 9.6E-04 14.9 7.1 16.0 0.0 282.4 150.4
MAN1C1 57134 3.59 1.8E-03 556.5 478.6 107.6 378.3 7.6 45.2
MAP6 4135 -3.25 7.6E-03 2.3 32.9 0.6 111.4 629.4 288.0
MBP 4155 6.47 2.7E-09 118.6 402.5 103.3 303.0 0.0 6.3
MMP17 4326 3.58 1.0E-03 1173.5 864.3 422.6 340.8 5.1 84.8
MMRN1 22915 -8.80 1.6E-06 0.0 0.0 1.3 0.0 279.4 93.2
MN1 4330 -7.23 3.0E-07 21.6 4.2 0.6 0.0 146.8 384.1
MT2A 4502 4.13 1.3E-03 2039.2 1924.4 8308.0 5237.5 285.0 301.4
MYL9 10398 -3.85 4.3E-03 3.8 3.1 110.0 8.5 796.2 370.3
NDN 4692 -3.46 2.3E-03 105.3 84.2 0.0 0.0 214.4 405.4
NELL2 4753 -3.25 1.0E-02 117.3 173.6 1.9 380.9 1073.3 2440.6
NKX2-1 7080 -5.52 1.6E-06 4.1 17.2 0.0 0.0 425.3 103.6
NNMT 4837 3.74 5.0E-04 519.9 172.4 918.1 175.6 15.2 47.4
NPTX2 4885 -6.47 1.8E-06 0.0 0.0 6.9 4.8 422.0 272.4
NTN1 9423 -4.22 1.5E-04 95.8 239.9 22.4 250.6 5458.0 966.2
NTRK2 4915 3.61 2.6E-03 90.7 163.9 174.8 219.5 0.0 30.3
ODZ2 57451 -3.68 6.5E-03 27.0 25.0 2.5 36.0 212.8 329.5
PDE1C 5137 4.13 1.3E-04 578.2 430.3 1117.8 981.4 45.7 50.8
PDZRN3 23024 -4.40 9.4E-04 18.4 22.1 12.6 3.0 278.2 251.3
PEG3 5178 -5.14 6.0E-08 105.7 130.6 8.5 0.0 2681.0 594.6
PI15 51050 -6.45 9.3E-08 20.2 3.3 2.5 5.5 397.8 262.2
PLA2G4A 5321 7.34 1.9E-12 909.7 1226.5 483.7 755.9 0.0 10.1
PLCH1 23007 -3.41 2.4E-03 152.0 134.0 83.3 118.9 2002.9 384.3
132

6.4 Core Differentially Expressed Genes Results
Gene Entrez Differential expression results Normalised tag counts
symbol gene ID log 2(F C) FDR G144ED G144 G166 G179 CB541 CB660
PLS3 5358 3.83 3.1E-04 360.2 776.6 476.6 367.2 0.0 75.6
PMEPA1 56937 4.36 5.4E-05 240.8 123.3 621.8 297.2 11.7 22.1
PRSS12 8492 4.80 2.1E-04 218.6 119.9 151.9 156.0 0.0 10.2
PTEN 5728 -5.00 2.9E-03 2.7 0.0 14.8 0.0 191.7 126.9
RAB6B 51560 -3.51 2.2E-03 71.7 52.1 29.4 38.4 754.8 160.7
RGS5 8490 -4.82 2.9E-04 4.1 13.5 6.9 3.0 343.5 112.5
RTN1 6252 -5.34 5.7E-05 8.1 5.5 0.6 19.3 385.7 286.3
S100A6 6277 6.04 5.1E-14 2259.3 324.2 5709.9 512.7 12.3 53.4
SALL2 6297 -4.13 1.7E-03 72.9 38.6 16.5 11.6 194.2 581.9
SDC2 6383 -3.56 2.9E-03 96.7 41.3 53.0 98.9 1015.1 502.4
SEMA6A 57556 -3.48 3.3E-03 68.1 50.0 3.1 0.0 257.1 140.0
SIX3 6496 -5.50 2.9E-06 2.7 26.1 0.0 0.0 203.9 594.6
SLC4A4 8671 -4.49 6.1E-05 36.0 38.9 13.1 18.8 925.2 145.3
SLCO1C1 53919 -Inf 5.9E-07 7.7 0.0 0.0 0.0 166.2 151.8
SLCO2A1 6578 -5.61 1.0E-05 38.8 15.5 0.0 0.0 213.0 297.0
SLIT2 9353 -7.18 1.5E-03 1.4 0.0 0.7 5.2 178.8 400.8
SPARCL1 8404 -4.03 1.5E-03 44.2 39.0 4.8 282.9 1448.1 2097.6
ST6GALNAC5 81849 -8.25 3.4E-09 8.9 0.0 0.0 5.8 261.3 899.5
SULF2 55959 3.31 5.3E-03 749.0 891.0 253.5 581.6 43.3 72.4
SYNM 23336 -4.23 3.3E-03 1.4 0.0 0.0 36.4 238.8 212.8
TAGLN 6876 -7.06 4.4E-11 1.9 3.1 9.4 4.1 1089.6 384.4
TES 26136 -Inf 2.5E-09 0.0 0.0 0.0 0.0 433.0 184.2
TRAM1L1 133022 -5.13 2.5E-03 0.0 0.0 0.0 11.3 121.7 146.3
TUSC3 7991 -3.74 5.9E-03 0.0 0.0 5.9 75.0 379.0 333.7
In a wide literature search that we performed on each gene of the core set,
we found that many of both the up-and down-regulated genes appeared in
pathways known to be affected in glioma. In the up-regulated cohort we found:
· S100A6 encodes an EF-hand calcium binding protein linked to the reg-
ulation of cell proliferation, cytoskeletal organisation and metastasis.
S100A6 is up-regulated in several malignancies. In gliomas, S100A6
appears to be specifically expressed by cancer stem cells [80,186].
· HLA-DRA and HLA-DRB form one of the MHC class II receptors, which
present extracellular antigens to T-helper cells. CD74 or "invariant
chain" takes the place of the extracellular peptide before MHC class
II receptors are mature in the lumen of the endoplasmic reticulum (ER).
Increased antigen presentation can help the immune system target cancer
cells. HLA-DRA and HLA-DRB are induced by inflammatory cytokines
and up-regulated in many tumours [421].
· MBP encodes myelin basic protein, so it’s expression in GNS lines prob-
ably reflects their glial nature.
· PMEPA1 can act as an oncogene by affecting the PI3K pathway, in-
cluding PTEN down-regulation. PMEPA1 is up-regulated in multiple
tumours, but there are no reports implicating it in glioma [460,531].
· LMO4 encodes a transcriptional regulator involved in the development
of multiple organs, including the CNS. LMO4 expression is elevated in
133

6.4 Core Differentially Expressed Genes Results
several cancers. LMO4 is well-characterised as an oncogene in breast
cancer, but has not been studied in glioma. LMO4 is regulated trough
the PI3K pathway, which is known to be affected in glioma [?,340].
· The transcription factor gene CEBPB is a well-characterised glioma
oncogene, recently suggested to be a master regulator of a mesenchy-
mal gene expression signature associated with poor prognosis [85].
· CD9 is a membrane protein implicated in invasion [253] and regulation
of EGF receptor activation [455]. Its expression in astrocytic tumours
correlates with malignancy [220].
· MT2A encodes a metallothionein found to suppress apoptosis in cancer
cell lines [110], potentially by causing TP53 to misfold [411]. Inversely
correlated expression of MT2A and TP53 proteins have been observed
in glioblastoma specimens [306].
· NTRK2 encodes a receptor for brain-derived neurotrophic factor, pro-
moting differentiation, proliferation and survival [118]. In several tumour
types, its expression correlates with poor prognosis and metastasis [118]
and the protein has been detected in a subset of cells in astrocytomas [29].
As shown in figure 6.11 the Tag-seq data available for NTRK2 happens
to identify a long and a shorter isoform for this gene, potentially the one
lacking the kinase domain that has also been implicated in the regulation
of astrocyte morphology [366].
· FOXG1, a transcription factor gene involved in brain development, is
commonly amplified in the childhood brain tumour medulloblastoma [9].
FOXG1 has been proposed to act as an oncogene in glioblastoma as well,
by suppressing growth-inhibitory effects of TGFβ [448].
The set of core up-regulated genes also includes multiple genes suggested to
play a role in other neoplasias, but for which we failed to find any studies
implicating them in glioma. For five of these, previous studies have pointed to
a role in motility and invasion: PLS3 [26,163], LMO4 [475], BACE2 [243,499],
CTSC [113,530,551] and MMP17 [88,430]. Further examples of genes impli-
cated in other cancers include the putative growth factor EPDR1 [8,360], which
is highly expressed in some progenitor cell types [178], and PMEPA1, which can
act as an oncogene by affecting the PI3K pathway and PTEN stability [460].
A subset of the core up-regulated genes have to the best of our knowledge
134

6.4 Core Differentially Expressed Genes Results
Figure 6.11: NTRK2 is a brain-derived neurotrophic factor detected in the literature
in a subset of cells in astrocytomas [29] and found in our Tag-seq dataset
by mappings of tags on the shortest isoform. Three tracks are visible in the figure:
at the top, a customised track composed of 12 sub-tracks representing the expression
levels of our GNS cell lines and NS cell lines in red - for tag mappings on the
minus strand - and blue - for tag mappings on the plus strand. The level of tag
expression is indicated by the thickness of the line representing the mapped tag; at
the centre a track representing UCSC genes, RefSeq genes and Ensembl genes; at
the bottom a track showing the zoom in of the central track region where the tag
mapping takes place. The thicker lines in the bottom track indicate the 3’UTRs
of the gene. The tags representing the NTRK2 gene map onto a region that identifies
both the shorter - potentially the one that lacks the kinase domain - and a
longer isoform. Adapted from the UCSC genome browser (our Tag-seq track for
hg19 is added by pasting the following line into the custom track box http://gns:
zed9epre@www.ebi.ac.uk/\~engstrom/gns/tracks/hg19/gns\_tpm.bg.gz).
135

6.5 Large-scale qRT-PCR Validation Results
not been implicated in cancer. One of these putative oncogenes is PLA2G4A,
which encodes a cytoplasmic phospholipase (cPLA2α) involved in production
of lipid signaling molecules with mitogenic and pro-inflammatory effects [192].
Another example is PDE1C, encoding a cyclic nucleotide phosphodiesterase
that controls cellular levels of cAMP and cGMP and may regulate cell prolif-
eration [126]. PDE1C expression has been observed in other glioma cell lines,
but also in non-malignant astrocytes from rat brain [508]. Another putative
novel oncogene is C10orf11. Disruptions of this gene have been associated with
mental retardation, consistent with a developmental role [501]. Although the
human gene is poorly characterised, the orthologous gene in Ciona intestinalis
is required for embryogenesis and is a component of the Wnt/β-catenin sig-
naling pathway, which is often activated in cancer [519]. Similar observations
were made for core set of down-regulated genes:
· the TES gene (testis derived transcript) is a LIM-domain gene at frag-
ile site found to be implicated in glioblastoma as well as prostate and
head/neck cancer [294,349,444];
· the DNER gene (delta/notch-like EGF repeat containing) is a putative
growth factor receptor found to be implicated in glioblastoma neuro-
sphere formation reduce adipocyte cell proliferation [381,478];
· the PTEN gene is a tumour suppressor often down-regulated and/or
deleted in gliomas [326];
· the TUSC3 gene (tumour suppressor candidate 3) is implicated in ovarian
cancer [90].
6.5 Large-scale qRT-PCR Validation
To assess the accuracy of Tag-seq expression level estimates and investigate
gene activity in a larger panel of cell lines, we assayed 82 core differentially ex-
pressed genes in 16 GNS cell lines (derived from independent tumour samples)
and 6 normal NS cell lines (Appendix tables A.3 and A.4), by qRT-PCR, using
custom-designed TaqMan microfluidic arrays. The 82 validation targets were
selected from the 92 core differentially expressed genes based on the availabil-
ity of TaqMan probes and considering prior knowledge of gene functions. For
the cell lines assayed by both Tag-seq and qRT-PCR, measurements agreed
remarkably well between the two technologies: the median Pearson correlation
136

6.5 Large-scale qRT-PCR Validation Results
for expression profiles of individual genes was 0.91 and the differential expres-
sion calls were corroborated for all 82 genes. Figure 6.12 shows the relationship
between the qRT-PCR and the Tag-seq measurements in three different panels
for four different genes that were picked to highlight the dynamic range of the
qRT-PCR and the Tag-seq platforms: the matrix metallopeptidase is a mem-
brane protein involved in the breakdown of the ECM and a known oncogene in
glioblastoma primary tumours [183]; the damage-inducible transcript DDIT3
is induced in response to a range of compounds that sensitises glioma cells
to apoptosis [216]; MYL9 encodes for one of the myosin light chains and its
link with glioblastoma has yet to be established (possibly acting via regulation
of the cell migration process); the mannosidase gene MAN1C1 is one of the
enzymes that hydrolyse the mannose sugar but has to yet be implicated in
glioblastoma - although its sister gene MAN2C1 has been associated with a
reduction of PTEN functionality in prostate cancer [190]. In panel a of figure
6.12 the high overall correlation between the two platforms is highlighted by
the four data points that lie nearly on a straight line (R = 0.91); panel b shows
the correlation values computed as the correlation between normalised Ct val-
ues and tag counts across the five GNS cell lines (table of correlation values
available in Appendix A.4); panel c shows the mean fold-change between the
genes expressed in the GNS and NS cell lines as measured in each of the two
platforms plotted so that the influence of outliers can be reduced thanks to
the averaging.
Across the entire panel of cell lines, 29 of the 82 genes showed statistically
significant differences between GNS and NS lines at an FDR of 5% (Fig 6.13).
This set of 29 genes distinguishes GNS from NS lines and may, therefore, have
broad relevance for elucidating tumourigenic properties of GNS cells. Figure
6.14 shows a histogram for each of the 29 genes found to be relevant in distin-
guishing GNS from NS cell lines. The height of the histogram bar represents
the expression level measured via qRT-PCR for that gene and its designated
probe (see title of each histogram for probe’s name). Each expression level is
calculated as the arithmetic mean across all biological replicates for each cell
line. The histogram bar colours comply to the ones selected for GNS and NS
cell lines in the bar displayed at the top of fig 6.14, distinguishing the assayed
GNS and NS cell lines in two separate groups along the x-axis.
137

6.5 Large-scale qRT-PCR Validation Results
Figure 6.12: Expression estimates correlate well between Tag-seq and qRT-PCR.
(a) Expression values for MMP17, DDIT3, MYL9 and MAN1C1 in two GNS and
two NS lines assayed by both technologies. Note the near-perfect agreement between
Tag-seq (x-axis) and qRT-PCR (y-axis) over a wide dynamic range. (b) Histogram of
correlation between Tag-seq and qRT-PCR for each of the 82 genes measured by qRT-
PCR. (c) Fold-change estimates (indicating expression level in GNS lines relative to
NS lines) from Tag-seq and qRT-PCR for the 82 genes. qRT-PCR confirmed greater
than two-fold difference in expression (dashed lines at y = 1) for all genes.
Figure 6.13: Heatmap of 29 genes differentially expressed between 16 GNS and 6
NS cell lines. Colours indicate δδCt expression values, i.e. normalised expression
on a log 2 scale where zero corresponds to the average expression between the two
groups (GNS and NS).
138

6.5 Large-scale qRT-PCR Validation Results
139

6.5 Large-scale qRT-PCR Validation Results
140

6.5 Large-scale qRT-PCR Validation Results
141

6.6 Literature Mining for Differentially Expressed Genes Results
Figure 6.14: Expression levels of the 29 genes distinguishing GNS from NS lines,
measured by qRT-PCR and presented as percent of NS geometric mean. Two histograms
reveal the data for the two probes used to measure the expression levels of
MYL9.
6.6 Literature Mining for Differentially Expressed
Genes
An important assessment that adds value to our pathway analysis, is the clas-
sification of every differentially expressed gene (FDR< 10%) as a gene known
142

6.6 Literature Mining for Differentially Expressed Genes Results
to be associated with glioblastoma, cancer or yet unlinked to any of these. I
extracted the information held in several literature navigating resources, such
as BioGraph [280], GBMbase [162] and iHOP [197] using a script described in
detail in the Methods section and reported in Appendix B, and classified each
gene in one of the following categories:
· Extensive amount of literature implicates the gene in glioblastoma;
· Limited amount of literature implicates the gene in glioblastoma;
· Unknown to be implicated in glioblastoma, but known in other cancers;
· Unknown to be implicated in any type of cancer.
The information retrieved by the script for each of the differentially expressed
genes was assigned to one of the four categories described above (Appendix
A.2). The data outputted from this query is summarised in table 6.5.
Table 6.5: Differentially expressed genes (F DR < 10%) assigned to a four-tier
classification system. The unique total refers to the unique list of all the genes in
each category.
Category Number
Extensive amount of literature implicating the gene in GBM 238
Limited amount of literature implicating the gene in GBM 94
Unknown to be implicated in GBM, but known in other cancers 199
Unknown to be implicated in any type of cancer 232
Unique Total 748
of genes
A stripped down version of the complete table available in Appendix A.2,
highlights only the genes with limited evidence or no evidence of implication
in glioblastoma that appear in our integrated glioblastoma pathway (Table
6.6). Many of these genes had not been directly implicated in glioma, but they
participate in glioma-related pathways and they differ in expression between
GNS and NS lines. Thus, by considering expression changes in a pathway
context, we identified additional candidate glioblastoma genes, such as the
putative cell adhesion gene ITGBL1 [50], the orphan nuclear receptor NR0B1,
which is strongly up-regulated in G179 and is known to be up-regulated and
mediate tumour growth in Ewing’s sarcoma [158], and the genes PARP3 and
PARP12 which belong to the PARP family of ADP-ribosyl transferase genes
involved in DNA repair. The up-regulation of these PARP genes in GNS cells
143

6.7 Isoform Differential Expression Results
may have therapeutic relevance, as inhibitors of their homolog PARP1 are in
clinical trials for brain tumours [270]. Our comparison between GNS and NS
cell lines, thus, highlights genes and pathways that are known to be affected in
glioma as well as novel candidates, suggesting that the GNS vs NS comparison
is a promising approach for further understanding molecular aspects of glioma.
6.7 Isoform Differential Expression
To establish Tag-seq as a sensitive technique for the identification of tran-
script isoforms differentially expressed between our GNS and NS libraries, we
performed a parametric and a non-parametric test on the genes identified by
multiple tag mappings using the "Ref_best" collection (see Fig 6.5).
With the non-parametric χ 2 test, 2,682 isoforms were found to be differentially
expressed as detected by the sum over all tags for each isoform. With the para-
metric approach, in which the logarithmic ratio of expression method adapted
from Morrissy et al [346] was used, we computed a ratio change for each pair
of tags that identified an isoform, so that a total of 2,040 differentially ex-
pressed isoforms were detected. The two methods share a total of 1,454 genes
that have differentially expressed isoforms in the GNS cell lines with respect
to the NS cell lines at significant levels (p-value2 for parametric), with 1,228 genes uniquely identified by the logarithmic
method and 586 genes uniquely identified by the non-parametric method.
In the attempt to focus on isoforms relevant to the pathways affected in
glioma, we overlaid both lists of differentially expressed isoforms on the in-
tegrated glioblastoma pathway described in section 7.5 and shown in figure
7.8. This highlighted the presence of 48 matching genes from the logarithmic
(non-parametric) method and 57 matching genes from the parametric χ 2 test
method, with 35 genes shared between the two methods. Each of the 35 genes,
as well as the 13 genes identified exclusively through the logarithmic method
and the 22 genes identified exclusively through the parametric χ 2 test method,
were manually scrutinised on the University of California Santa Cruz (UCSC)
genome browser by adding our tag mappings as a custom track together with
the UCSC gene track and the gene predicted Ensembl track, both in "full"
mode for the NCBI37.2/Ensembl 64 human genome assembly. For the follow-
144

6.7 Isoform Differential Expression Results
Table 6.6: Genes with limited evidence or no evidence of implication in glioblastoma
that appear in our main glioblastoma pathway. In "Others" the genes with an
important physiological role in the brain that do not appear in our glioblastoma
pathway are listed.
Gene* Log2(FC)** Association Implications in Citations
with glioma other neoplasms
CACNA1A 7.1 None Prostate cancer (mouse model) [235]
CACNA1C -8.2 None Liver cancer [25]
CACNG7 -2.6 None None -
CACNG8 -2.6 None None -
CAMK1D -2.4 None Breast cancer [51]
CPLX2 6.4 None None -
DDIT3
(CHOP,
GADD153)
4.4 Limited General (cellular stress response) [218,227,330,415]
DUSP16 4.2 None Burkitt’s lymphoma [265]
FGF19 - None Liver, lung and colon cancer [119]
ITGA4
(CD49D)
3.0 Limited Chronic lymphocytic leukaemia,
breast cancer and others
[244,263,309]
ITGBL1 + None None -
MAP3K5
(ASK1)
5.1 Limited Gastric cancer and histiocytoma [94,189,504]
NFATC2
(NFAT1)
+ Limited Breast cancer [64,95,311,348]
NFKBIZ 5.1 None Liposarcoma [170]
NR0B1 + None Lung adenocarcinoma and Ew- [?,236]
(DAX1)
ing’s sarcoma
NR1D1 2.9 None Breast cancer [246]
PARP3, 4.1, 2.9 Homology*** The PARP gene family is in- [312,463]
PARP12
volved in DNA repair and several
other processes related to tumourigenesis
PERP 3.8 None Lung and skin cancer [92,317]
PPEF1 4.4 Limited None [296]
SNAP25 3.3 None Lung cancer [175]
SYT1 -2.5 None None -
TNFRSF14 4.0 None Follicular lymphoma [93]
TNFSF4 4.0 None Generally implicated in immune [424]
(OX40L)
response to tumours
* Aliases are listed in parentheses.** Gene expression log2FC between the GNS and
NS lines compared by Tag-seq. Some genes were detected exclusively in GNS or NS
lines (indicated in column two by + or -, respectively).*** The homolog PARP1 has
been implicated in glioma.
ing list of genes an alternative tag mapping-based isoform was observed, three
of which are reported as adapted images from the UCSC genome browser in
figures 6.15 and 6.16:
145

6.7 Isoform Differential Expression Results
· AKT2: two reads map on the 3'UTR of the longest isoform whilst one
read maps on the 3'UTR of a shorter subset of isoforms;
· AKT3: two reads map on the shortest isoform and one on the 3'UTR
and intron of the middle and longest isoforms, respectively;
· BMP7: one read maps on the 3'UTR of the longest isoform and one read
on the 3'UTR of a middle-length set isoforms;
· BRCA1: one read maps on the 3'UTR of the longest isoform, whilst one
read maps on an alternatively spliced exon, contained in a small subset
of isoforms and which is a single isoform itself;
· CANX: all six reads map onto the 3'UTR of the longest isoform, one on
the 3'UTR of a shorter isoform;
· CTSB: three reads identify three different isoforms on their 3'UTRs;
· ERBB2: two reads map on the 3'UTRs of two different isoforms (but
one is not recognised by Ensembl Gene Predictions);
· FGFR1: two reads map on the 3'UTRs of two different isoforms;
· GRIA2: three reads identify three sets of lengths of isoforms on their
3'UTRs with one identifying two of them;
· HLA-A: two reads identify two different isoforms on their 3'UTR;
· NBR1: one read maps on the 3'UTR of the longest isoform and one read
on a consistently present exon, which identifies also a shorter isoform;
· NFYC: two reads identify 3'UTRS of two different isoforms with one
mapping between the 3'UTRs and last intron;
· PTEN: two reads map on the 3'UTR of the longest isoform and on the
3'UTR of a much shorter isoform;
· RFX5: one read maps on the 3'UTR of two longer isoforms and one
identifies a smaller subset of shorter ones on the 3'UTR;
· NTRK2: two reads mapping on alternative 3'UTRs;
· FGFR10P: one read maps on the 3'UTR and one on the fifth exon, which
is displayed as a 3'UTR box on the Ensembl Prediction track;
146

6.7 Isoform Differential Expression Results
· METTL13: two reads mapping on alternative 3'UTRs;
· TSC22D2: two reads mapping on alternative 3'UTRs;
· RSRC2: two reads mapping on alternative 3'UTRs;
· TPM1: one read maps on the longest 3'UTR and the other on an internal
exon that is alternatively spliced in one isoform.
We used the GenemiR package described in chapter 8 to find out how many
genes, out of the 2,682 and 2,040 genes with differentially expressed isoforms,
were predicted to harbour the same microRNA targeting sites in their 3'UTRs.
The functionalities at the core of the GenemiR software package are to output
a list of microRNAs when a list of genes is inputted and, vice versa, to output
a list of genes when a list of microRNAs is inputted. The database used by
these core functionalities consists of all the microRNA to mRNA predictions
made by a maximum of eight leading algorithms and it varies in size depend-
ing on which algorithms the user has chosen to select as part of a specific
search. The search performed in this instance made use of the union of the
microRNA predictions from five of the most widely accepted target prediction
algorithms: PicTar [247], PITA [224], Targetscan [272], miRanda [213] and
DIANA-microT [315,316]. The choice of using the union set of the results
from the five prediction algorithms as opposed to the intersection set is justi-
fied by the fact that the intersection set for each of the two lists of over 2,000
genes is null. In fact, as explained in chapter 8, the algorithms available to
predict microRNA to mRNA interactions generate such different outputs that
it is extremely rare to have all of them agree on the predictions (granted, of
course, that the list of genes is not so large that finding an intersection set
becomes statistically possible, or that the number of prediction algorithms se-
lected are less than 3). When inputted into GenemiR, the 2,682 genes we found
with the parametric method, yielded a list of 5,016 microRNAs. Similarly, the
2,040 genes we found with the logarithmic (non-parametric) method, yielded
a list of 4,463 microRNAs. When the two lists of microRNAs were intersected
(2,358 microRNAs) and inputed again into the GenemiR package to find the
genes targeted by those microRNAs, we found 765 genes to be regulated by
the intersected list of microRNAs (Fig 6.17). Table 6.7 shows the microRNA
predictions resulting from the GenemiR query, stratified by prediction algo-
rithm and origin of the gene list - parametric vs. non-parametric method - as
well as the results that are common to both methods and unique to each.
147

6.7 Isoform Differential Expression Results
148
Figure 6.15: BMP7 encodes a member of the TGFβ superfamily and is represented in our Tag-seq data by two tags that identify transcripts of
different lengths. The expression levels across our GNS cell lines (top) are in blue (plus strand) and red (minus strand). BMP7, like other members
of the bone morphogenetic protein family, plays a key role in the transformation of mesenchymal cells into bone and cartilage. However, BMP7 has
also been recently implicated in the suppression of tumourigenicity of stem-like GBM cells when released from endogenous neural precursor cells [97].
Image adapted from the UCSC genome browser.

6.7 Isoform Differential Expression Results
149
Figure 6.16: The TPM1 gene, encoding for a member of the tropomyosin family of actin-binding proteins, is represented by two tags, the rightmost
of which shows a very high expression in the antisense strand and identifies the four longest 3'UTRs. The expression levels across our GNS cell lines
(top) are in blue (plus strand) and red (minus strand). Image adapted from the UCSC genome browser.

6.7 Isoform Differential Expression Results
Figure 6.17: This figure summarises the process of finding genes with differentially
expressed isoforms (common to both the parametric and logarithmic method) that
are predicted to harbour the same microRNA targeting sites. One way to assess this
is to find which genes are predicted to be regulated by the same microRNAs. In blue
we find the numbers of microRNAs that are predicted to regulate the 2,682 genes
(green set) and the 2,040 genes (orange set). In black we find the numbers that
refer to genes. The 2,358 microRNAs common to both sets are predicted to regulate
765 genes that are in common to the parametric and the logarithmic methods. The
prediction database used comes from the union of five prediction algorithms: PicTar
[247], PITA [224], Targetscan [272], miRanda [213] and DIANA-microT [315,316].
In order to find in more detail which of the genes with differentially expressed
isoforms that were also predicted to harbour the same microRNA targeting
sites, actually had one or more microRNA seed sequences in their 3'UTRs, we
used the Ensembl gene coordinates and the union of all microRNA seed coor-
dinates given by the same five prediction algorithms used earlier (Targetscan,
miRanda, PITA and DIANA-microT) to validate which genes harboured which
microRNA seed sequences. We found that, of the 765 genes predicted to target
the same microRNA targeting sites, 226 of the 2,682 genes with differentially
expressed isoforms (identified with the parametric method) and 340 of the
2,040 genes with differentially expressed isoforms (identified with the logarith-
mic method) hosted a microRNA seed sequence between at least two tags (Fig
6.18). These amounts clearly show that microRNA is a widely adopted mech-
anism of isoform expression modulation in GNS cell lines. Since microRNA
array data was available for four GNS cell lines (G7, G26, G144, G166) and
four NS cell lines (CB660, CB130, CB152, CB171), we used it to cross verify
150

6.7 Isoform Differential Expression Results
Table 6.7: Summary of predicted microRNAs targeting differentially expressed
isoforms of the genes identified with the parametric and logarithmic method. The
predictions from each of the five algorithms (PicTar [247], PITA [224], Targetscan
[272], miRanda [213] and DIANA-microT [315,316]) are shown in distinct rows.
Prediction
algorithms
Nonparametric
microRNAs
Parametric
microRNAs
Common
microRNAs
Unique to
nonparametric
Unique to
parametric
PicTar4 164 164 164 - -
PITA 674 673 672 miR-658 miR-937
miR-886-3p
TargetscanS 148 148 147 miR-615-3p miR-450a
miRanda 677 677 677 - -
DIANA 510 508 503 miR-151-1p miR-146b-3p
micro-T miR-324-5p miR-199b-5p
miR-369-5p miR-566
miR-423-3p miR-744
miR-602 miR-877
miR-658
miR-941
Figure 6.18: Schematisation of the localisation of the microRNA seeds given by
the five prediction algorithms within the 3'UTRs of pairs of differentially expressed
isoforms identified by two tags that map onto each one. Of the 765 genes with
differentially expressed isoforms identified by both the parametric and logarithmic
methods, 226/2,682 genes and 340/2,040 genes hosted at least one microRNA seed
sequence between at least two tags.
that the microRNA predictions found for the 765 genes with differentially ex-
pressed isoforms were reflected in experimental data (Appendix F). Of the 11
microRNAs that were predicted to regulate the 765 genes with differentially
expressed isoforms that were also available in the microRNA array dataset, we
found seven to be implicated in regulatory pathways in GBM (Table 6.8):
151

6.7 Isoform Differential Expression Results
Table 6.8: MicroRNA array results for GNS cell lines with respect to NS cell lines.
Based on a literature survey, these microRNAs would be interesting candidates to
validate experimentally in a TLDA assay.
microRNA log 2(F C) FDR
miR-128 0.9 0.000206443
miR-137 1 0.000121777
miR-34a -1.5 0.000004.43
miR-26a 1.4 0.0000143
miR-10b 2.9 1.03E-08
miR-451 1 3.15E-05
miR-129-3p 1 9.63E-05
· The levels of miR-128 have been found to be consistently lower in
glioblastoma compared with normal brain tissue [259,364]. Opposite
findings were observed in our microRNA array, in which miR-128 is up-
regulated in GNS cells with respect to NS cells (Table 6.8). miR-128
directly targets the transcription factor E2F3a, which activates genes
necessary for the progression of cell-cycle and can thus inhibit prolifera-
tion of brain cells by negatively regulating E2F3a. This microRNA also
directly targets BMI1, a gene that is thought to act as an oncogene in
glioblastoma by regulating tumour suppressors like P53 and CDKN2A.
BMI1 also promotes stem cell renewal by acting as part of a Polycomb Si-
lencing Complex to silence the expression of genes - including CDKN2A
and CDKN1A tumour suppressors - involved in differentiation and senes-
cence. Low levels of miR-128 in glioblastoma may contribute to glioma
growth by allowing the increased expression of BMI1 to promote an un-
differentiated self-renewing state. High levels of miR-128 in GNS cells
with respect to NS cells may identify a stem cell pool specific regulation
that has yet to be studied in detail. miR-128 is known to be highly
expressed in neurons but its role in the brain is still unknown and is
surmised to be the promotion of neuronal differentiation through pre-
vention of stem cell self-renewal. Our isoform data analysis with the
non-parametric method showed that miR-128 is predicted to target the
nerve growth factor receptor associated protein 1 NGFRAP1, a p75NTR-
associated cell-death executor mediated by the common neutrophin re-
ceptor p75NTR [350] that also plays a role in NGF-induced apopto-
sis in oligodendrocytes [351]. Analsysis with the logarithmic method
showed that miR-128 is predicted to target the Rab GTPase guanine nu-
152

6.7 Isoform Differential Expression Results
cleotide exchange factor GAPVD1, a gene fundamental for the activation
of RAB5A during the engulfment of apoptotic cells [238] (Fig 6.19).
· miR-137 is one of the most down-regulated microRNAs in glioblas-
toma compared with normal brain tissue. In our microRNA microarray
dataset, miR-137 is up-regulated in GNS cell lines with respect to NS cell
lines. Since miR-137 directly targets CDK6, if over-expressed in glioblas-
toma it would induce cell-cycle arrest. The level of miR-137 increases
upon differentiation of glioma neurosphere cultures and if over-expressed
in these cells it leads to the expression of markers consistent with neu-
ronal differentiation. These data suggest that the lower expression of
this microRNA in glioblastoma, and the higher expression in GNS cells
with respect to NS cells, reflects the lack of tumour cell differentiation
in the former and the presence of regulatory mechanisms downstream of
miR-137 in the latter, since cancer stem cells are the least differentiated
within the lineage hierarchy according to the cancer stem cell hypothe-
sis [364]. In our microRNA prediction analysis, miR-137 was predicted
to target the serine/threonine-protein kinase 40 STK40 gene, a nega-
tive regulator of NFKB and p53-mediated gene transcription [200] and
the transcription factor TCF4, which associated with β catenin mediates
Wnt signaling by trans-activating downstream target genes [11].
· miR-34a expression is down-regulated in glioblastoma and in p53-null
mutant gliomas since non p53-null mutants, expressed in many tumours
and that can possess gain-of-function activities, do not regulate tran-
scription of miR-34a. This suggests miR-34a as a transcriptional target
of P53 [278]. In our microRNA microarray dataset miR-34a is found to
be down-regulated as well. Furthermore, miR-34a potently inhibits the
protein expression of MET, a hepatocyte growth factor receptor encod-
ing a tyrosine-kinase, as well as MET 3'UTR reporter activity in glioma,
medulloblastoma cells and astrocytes. miR-34a also inhibits Notch-1
and Notch-2 protein expression and their 3'UTR reporter activities, as
well as CDK6 protein expression in glioma cells. Transient transfection
of miR-34a into brain tumour cell lines inhibited cell proliferation, cell
cycle progression, cell survival, and cell invasion but did not affect hu-
man astrocyte cell survival and cell cycle. miR-34a transfection, also,
did not affect the protein levels of PDGFRA in any tested cell line, al-
though miR-34a has predicted seed matches in the 3'UTR of PDGFRA.
153

6.7 Isoform Differential Expression Results
154
Figure 6.19: Isoform detection by multi tag mapping of gene GAPVD1, a GTPase guanine nucleotide exchange factor essential during engulfment
of apoptotic cells [238] and involved in the degradation of EGFR [473]. The expression levels across our GNS cell lines (top) are in blue (plus strand)
and red (minus strand). Image adapted from the UCSC genome browser.

6.7 Isoform Differential Expression Results
miR-34a transfected cells generated xenografts that were statistically
significantly smaller than control miR transfected xenografts, demon-
strating that miR-34a expression inhibits in vivo glioblastoma xenograft
growth [278]. In our microRNA prediction analysis miR-34a is predicted
to target C1orf9, an open reading frame 9 of chromosome 1 since no p53
targeting is predicted due to the deletion of the p53 genomic location
within the GNS cell lines.
· miR-26a is up-regulated in glioblastoma and targets PTEN through the
direct binding in its 3'UTR of the B2 and B3 sites, mediating transla-
tional repression and reduced steady-state levels of the protein. West-
ern blotting demonstrated that miR-26a over-expression achieved a 50%
knockdown of PTEN protein in two glioblastoma cell lines, accompa-
nied by an enhanced Akt signaling pathway. In our microRNA mi-
croarray dataset we observe up-regulation of miR-26a as well. In addi-
tion to enhancing tumourigenesis, miR-26a effectively represses endoge-
nous PTEN protein in a relevant PDGF-driven glioma model system.
The miR-26a-mediated knockdown of EZH2, a histone methylatrans-
ferase, and SMAD1, a transcription factor, was also observed in glioblas-
toma [259,364]. In our analysis miR-26a was predicted to target the
SMAD1 gene, of which two isoforms were detected through tag mapping
(Fig 6.20).
· miR-10b is up-regulated in glioblastomas but its function has not been
described yet. Increased levels of miR-10b in breast cancer correlated
with the disease’s progression [259,364]. In our microRNA microarray
dataset miR-10b is also up-regulated.
· miR-451 inhibited the growth of transfected glioblastoma cells, as de-
tected by neurosphere formation assays [259,364]. We found miR-451 to
be up-regulated in our microRNA microarray dataset, in line with its
role as a cell cycle breaker and growth inhibitor.
· miR-129-3p is found to be down-regulated in glioblastomas but its func-
tion remains unknown to date [259,364]. Interestingly, we observed miR-
129-3p to be up-regulated in our microRNA microarray dataset, possibly
indicating a different regulation in action at the stem cell level.
155

6.7 Isoform Differential Expression Results
156
Figure 6.20: Isoform detection by multi tag mapping of gene SMAD1. The expression levels across our GNS cell lines (top) are in blue (plus
strand) and red (minus strand). Image adapted from the UCSC genome browser.

6.8 Long ncRNA Differential Expression Results
6.8 Long ncRNA Differential Expression
In contrast to microarray expression profiling, Tag-seq is not limited by pre-
selected probes targeting the known transcriptome and we took advantage of
this to discover differentially expressed ncRNAs. To this end, we called differ-
ential expression for a combination of tags that mapped to the genome, the
transcriptome or virtual tags and filtered the results using coding gene anno-
tations. At an FDR of 10%, this analysis revealed 25 differentially expressed
putative non-coding RNAs, 18 of which were up-regulated and the remaining 7
down-regulated (Appendix C). Five of these are putative long antisense RNAs
that are known to be transcribed from the opposite strand of protein-coding
genes CDKN2B, CD27, PAX8, MCF2L2 and TXNRD1. Two, instead, are
known long non-coding RNAs: HOTAIRM1 [547] and NEAT1 [314] (Table
6.9). CDKN2BAS, an antisense transcript to tumour suppressor CDKN2B, is
of particular interest because of its role in Polycomb-mediated repression of
CDKN2B and CDKN2A [537]. We detected CDKN2BAS exclusively in G144,
G144ED and G166, consistent with the locus being deleted in G179 according
to aCGH data (see 6.3). The functions of the remaining differentially expressed
Table 6.9: Multiple putative long non-coding RNAs differentially expressed between
GNS and NS cells at 10% FDR.
Category Up-regulated Down-regulated
Known antisense transcripts 2 (over CDKN2B, CD27) 1 (over PAX8)
Other known ncRNAs 2 (HOTAIRM1, NEAT1) 0
Intronic RNAs 2 (in CDKN2B, TXNRD1) 1 (in FAM38B)
Intergenics RNA 9 7
non-coding RNAs are unknown, but a unifying feature of 15 of them is that
they are located in gene deserts (Appendix D). Several of these transcripts
display an expression pattern similar to a protein-coding gene near the gene
desert, suggesting that the transcripts may be functional RNAs regulating
nearby genes [372] or indicate transcription from active enhancers [230]. The
coding genes exhibiting correlated expression to these non-cancer RNAs are
cancer-related genes CTSC (Fig 6.21) [113,551] and DKK1 [461] and develop-
mental regulators IRX2, SIX3 and ZNF536 [412].
In the case of the Cathepsin C (CTSC) gene we were able to detect two dif-
ferent isoforms with Tag-seq, as well as a long non-coding RNA lying within
a 150 kilobase distance from the two isoforms. The CTSC gene encodes a
42 Megabase sized genomic segments devoid of protein-coding genes in vertebrates.
157

6.8 Long ncRNA Differential Expression Results
lysosomal cysteine protease that is part of the peptidase C1 family of proteins
and is responsible for activating serine proteases in immune and inflammatory
cells to function in processes of bone remodelling, epidermal homeostasis, and
antigen presentation [500]. During cancer progression cathepsins are secreted
into the extracellular matrix where they promote tumour invasion by cleav-
ing components of the matrix and the basement membrane, thereby creating
a passageway for the migration of cancer cells. The disruption of adherens
junctions via cleavage of E-cadherin is another example. Cathepsins can also
initiate proteolytic cascades in which they activate other proteases such as
matrix metalloproteinases, which in turn promote invasion [166,167]. Mem-
bers B, L and S of the cathepsin family have been identified as regulators
of E-cadherin function through cleavage of its N-terminus. Knockout mice
mutants for cathepsins B, L or S show obvious defects such as the decrease
of tumour cell proliferation, tumour invasion and tumour vascularity, while
CTSC - / - knockout mice have more subtle defects that consist in failing to
activate granzymes A and B in cytotoxic lymphocytes [167].
The hypothetical regulation performed by the long non-coding RNA BC038205
on CTSC isoforms could be one of maintaining the expression levels of this
gene high in GNS cell lines through a form of direct regulation which is re-
lieved when the non-coding RNA is absent or expressed at lower levels, such as
in NS cells. In figure 6.21 three panels are shown in which the location (panel
a) of the CTSC isoforms and that of the non-coding RNA is shown along the
human reference genome; the expression levels as detected by Tag-seq (panel
b) are displayed in a histogram to visually highlight the relationship between
them; a correlation plot (panel c) that shows the presence of a correlation in
both the expression trends between the first CTSC isoform and the non-coding
RNA BC038205, and the second CTSC isoform and the same non-coding RNA
gene. The correlation was calculated with the cor.test function in the R stats
package that uses the Pearson’s product moment correlation coefficient to test
for association between paired samples.
HOTAIRM1 is known to be strongly up-regulated in human NB4 promyelo-
cytic cell lines and normal hematopoietic cells upon induction of granulocytic
differentiation and is found to be up-regulated in our GNS lines as well as in
the gliomas from the Parsons et al [383] Tag-seq data with respect to our fetal
human NS cell lines and the Parsons et al primary brain samples. Interestingly,
its knock-down in NB4 cells causes down-regulation of HOXA1 and HOXA4
genes and these genes are up-regulated in our Tag-seq dataset (Fig 6.22).
158

6.8 Long ncRNA Differential Expression Results
Figure 6.21: Correlated expression of CTSC and a nearby ncRNA. (a) CTSC
(cathepsin C) is located in a gene desert harboring an uncharacterized ncRNA transcribed
in the opposite direction (cDNA BC038205). Image adapted from the Ensembl
Genome Browser [147]. (b) Both CTSC and the ncRNA have strongly elevated
expression in the GNS lines relative to the NS lines, with highest levels in G179.(c)
Correlation plots for the BC038205 ncRNA and first isoform of CTSC (grey) and
second isoform of CTSC (blue).
159

6.8 Long ncRNA Differential Expression Results
160
Figure 6.22: Histogram displaying the normalised tag counts found in our Tag-seq dataset and the Parsons et al [383] Tag-seq dataset for the
HOTAIRM1 long ncRNA and the surrounding HOX genes.

Chapter 7
Dataset Correlation Analyses
Contents
7.1 Enrichment Analysis . . . . . . . . . . . . . . . . . . . . . . 161
7.2 Glioblastoma Expression Signatures . . . . . . . . . . . . . 168
7.3 Tumour Expression Correlation . . . . . . . . . . . . . . . . 170
7.4 Survival Analysis . . . . . . . . . . . . . . . . . . . . . . . . 182
7.5 Glioblastoma Pathway Analysis . . . . . . . . . . . . . . . 187
7.1 Enrichment Analysis
The differential expression analysis revealed 485 genes to be up-regulated and
254 genes to be down-regulated between GNS and NS cell lines at an FDR of
10% (Appendix A.1). We performed Gene Set Enrichment Analysis (GSEA)
to investigate which pathways were most highly represented in our set of 739
differentially expressed genes.
In order to evaluate the enrichment of a group of genes that together define a
pathway, the GSEA method looks for those genes to follow the same trends in
the experimental dataset of interest. In fact, if a number of genes that belong
to the same pathway change expression level even moderately, it could mean
that in the evaluated experimental setting that pathway is being affected. By
having established an a priori relationship between the genes involved in the
same pathway, the GSEA method detains more statistical power to detect
smaller changes that affect the whole set as compared to a per gene statistic.
In order to achieve its goal the GSEA method first ranks the genes in the
dataset of interest according to a per gene statistic such as a p-value and
then uses the complete ranked list to assess how the genes that belong to a
161

7.1 Enrichment Analysis Results
specific pathway distribute across the ranked list, whether they are randomly
distributed throughout the ranked list or primarily found at the top or bottom.
Three key elements define the GSEA method [474]:
1. Calculating an enrichment score that measures the degree to which a set
of genes belonging to a pathway is overrepresented at the top or bottom
of the entire ranked list of differentially expressed genes, for example
(and corresponds to a weighted Kolmogorov-Smirnov like statistic).
2. Estimating the significance of the enrichment score by permuting the
phenotype labels and recomputing the enrichment score of the genes in
the pathway each time. This generates a null distribution that the p-
value of the observed original enrichment score is calculated against.
3. Adjusting the enrichment score to account for multiple hypothesis testing
when entire databases of pathways are evaluated at once, like in our case
with the KEGG and Gene Ontology (GO) databases. In this case the
enrichment score is first normalised to account for the size of the path-
way, and then the proportion of false positives, or FDR, is calculated
for each normalised enrichment score. The FDR associated to each nor-
malised enrichment score corresponds to the estimated probability that
a pathway with a given normalised enrichment score represents a false
positive finding.
The enrichment analysis using GO [106] and the KEGG pathway database [2]
confirmed the set of 739 differentially expressed genes to be enriched for path-
ways related to brain development, glioma and cancer (Table 7.2 and 7.1). We
also observed enrichment of regulatory and inflammatory genes, such as signal
transduction components, cytokines, growth factors and DNA-binding factors.
Several genes related to antigen presentation on MHC class I and II molecules
were up-regulated in GNS cells, consistent with the documented expression of
their corresponding proteins in glioma tumours and cell lines [120,174]. In line
with these findings, affected pathways from the KEGG database included Anti-
gen Processing and Presentation, Diabetes Mellitus Type I, Cytokine-cytokine
receptor interaction, Neuroactive ligand-receptor interaction, MAPK signaling
and, expectedly, Glioma, a collection of genes involved in glioma formation
(Table 7.2). The first two plots from the GSEA run that identified these path-
ways as being significantly altered in our dataset, are shown in figure 7.1. In
the top panels of the figure the green distribution represents the trend of the
162

7.1 Enrichment Analysis Results
Table 7.1: Selected Gene Ontology terms and InterPro domains enriched among differentially expressed genes.
Differentially Expressed, 739 genes Up-regulated, 485 genes Down-regulated, 254 genes
Genes p-value Genes p-value Genes p-value
A. Biological Process GO terms
Nervous system development 106 2.00E-10 62 0.005 44 2.00E-05
Cell differentiation 128 7.00E-07 74 n.s. 54 2.00E-04
Cell proliferation 86 3.00E-04 59 0.014 27 n.s.
Cell adhesion 74 1.00E-07 56 2.00E-07 18 n.s.
Cell migration 44 3.00E-04 30 0.026 14 n.s.
Immune response 70 2.00E-12 61 3.00E-16 9 n.s.
Antigen processing and presentation 17 4.00E-07 17 5.00E-10 0 n.s.
Cellular ion homeostasis 36 0.014 33 2.00E-05 3 n.s.
B. Molecular Function GO terms
Signal transducer activity 111 3.00E-07 66 0.058 45 0.002
Receptor activity 83 8.00E-07 48 n.s. 35 0.002
Cytokine activity 27 2.00E-08 25 7.00E-11 2 n.s.
Growth factor activity 20 0.011 17 0.002 3 n.s.
MHC class II receptor activity 5 0.008 5 9.00E-04 0 n.s.
Sequence-specific DNA binding 52 3.00E-04 34 0.053 18 n.s.
C. Interpro domains
Immunoglobulin-like 45 3.00E-08 32 6.00E-06 13 n.s.
MHC classes I/II-like antigen recognition protein 14 1.00E-07 14 3.00E-10 0 n.s.
Homeobox 28 8.00E-06 18 0.012 10 n.s.
P-values indicating the statistical significance of enrichment of these terms were computed with Fisher’s exact test and corrected for multiple testing
using the Bonferroni method; n.s., not significant (p-value>0.1).
163

7.1 Enrichment Analysis Results
Table 7.2: Representative KEGG pathways from signaling pathway impact analysis
of gene expression differences between GNS and NS lines.
Pathway Genes p-value Predicted status in
GNS cell lines
Cytokine-cytokine receptor interaction 29 4.00E-12 Activated
Chemokine signaling pathway 15 5.00E-06 Activated
Neuroactive ligand-receptor interaction 21 2.00E-04 Inhibited
Antigen processing and presentation 11 7.00E-04 Activated
MAPK signaling pathway 24 0.011 Activated
Glioma 10 0.013 Activated
ECM-receptor interaction 10 0.041 Inhibited
Calcium signaling pathway 15 0.041 Activated
P-values and status predictions were obtained by signaling pathway impact analysis
[26], taking fold-change estimates and pathway topology into account. P-values were
FDR-corrected for multiple testing.
enrichment score - a number that reflects the degree to which a gene set is
overrepresented at the top or bottom of a ranked list of genes - as the analysis
walks down the ranked list. Both enrichment score distributions who positive
values because they indicate gene set enrichment at the top of the ranked list.
The score at the peak of the plot (farthest from 0.0) is the enrichment score for
the gene set. The middle portions of the two plots show where the members of
the gene set appear in the ranked list of genes. The set of genes that appear in
the ranked list prior to the peak enrichment score are the ones that contribute
most to the enrichment score and are commonly referred to as "leading edge
subset". The bottom portions of the two plots show the value of the ranking
metric, which is by default the signal-to-noise ratio and has been kept such in
our GSEA runs, as you move down the list of ranked genes. The ranking met-
ric measures a gene’s correlation with a phenotype: a positive value indicates
correlation with the GNS phenotype and a negative value indicates correla-
tion with the NS phenotype. Additional information from our GSEA runs is
provided in the two plots described in figure 7.2. Surprisingly, via GSEA we
detected an up-regulation in the GNS lines of several Major Histocompatabil-
ity Complex (MHC) class II genes, as well as related genes involved in antigen
presentation on MHC class I complexes (Fig 7.3). Several works have shown
that MHC class I and II molecules are involved in aspects of human cancer
pathology such as invasion and migration [327,421,546]. These two classes of
molecules are known to be a fundamental component of the adaptive immune
response and their misregulation potentially leads to faulty antigen recognition
and processing.
164

7.1 Enrichment Analysis Results
Figure 7.1: Enrichment plots for the top two pathways revealed through GSEA of
the KEGG database of pathways [2].
Figure 7.2: (a) The plot of nominal p-values, which estimates the statistical significance
of the enrichment score for a single pathway, vs normalised enrichment (NES)
score provides a quick way to grasp the number of enriched pathways that are significant.
The FRD q-value represents the estimated probability that the normalised
enrichment score represents a false positive finding. (b) The line graph of the enrichment
scores across pathways provides a quick visual way to grasp the number
of enriched gene sets at every enrichment score value. The first two peaks are negative
global enrichment scores that represent the enrichment of their corresponding
number of pathways in the NS phenotype. The last two peaks are positive global
enrichment scores that represent the enrichment of their corresponding number of
pathways in the GNS phenotype.
The endogenous pathway is used by any nucleated cell in the body to
present endogenous, or cytosolic, fragments of proteins to cytotoxic CD8 + T
cells. The receptors that are responsible for such presentation are the Major
Histocompatibility Complex (MHC) class I molecules. By exposing cytosolic
165

7.1 Enrichment Analysis Results
proteins outside the cellular membrane, MHC class I receptors can cause the
activation of cytotoxic T cells against exogenous peptides derived from an in-
fection. In fact, healthy cells are ignored by the cytotoxic T cells thanks to the
sensitivity of the recognition system between the loaded MHC class I receptor
and the T cell receptor, while cells containing foreign proteins can be recog-
nised and, therefore, killed. MHC class I molecules are heterodimers and they
consist of two polypeptide chains, the α and the β2-microglobulin chain, of
which only the α chain is polymorphic. Loading of MHC class I receptors with
peptides occurs inside the lumen of the endoplasmic reticulum (ER) [323].
The exogenous pathway is used by a cohort of specialised cells termed pro-
fessional Antigen Presenting Cells, or APCs. These cells are macrophages,
dendritic cells and B cells and they present fragments of extracellular proteins
to helper CD4 + T cells triggering a cell-mediated (Th1) or humoral (Th2) re-
sponse for peptides deriving from extracellular pathogens. The receptors that
are responsible for such presentation are the MHC class II molecules. MHC
class II molecules, like the MHC class I molecules, are heterodimers but the
MHC class II molecules consist of two homologous peptides, the α and β chain,
both polymorphic and encoded by the HLA gene. Loading of both classes oc-
curs inside the cell, but MHC class II molecules are loaded only when the
vesicle generated at the ER fuses with a lysosome containing digested endo-
cysed extracellular proteins. In the ER vesicle, the MHC class II molecule has
its peptide-binding cleft blocked by CD74, a trimer called "invariant chain"
which prevents the binding of cellular peptides. Once this vesicle fuses with
the lysosome, the MHC class II molecules are unloaded with the invariant
chain and reloaded with a peptide from the lysosome via an MHC class II-like
structure called HLA-DM. The stable MHC class II thus formed is presented
on the cell surface. the expression of MHC class II molecules is constitutive
only on professional APCs but can be induced by cytokines such as IFNG on
other cell types, including cancer cells [492].
All genes involved in the two pathways are listed below (7.3) [327], showing all
alleles of the same Human Leukocyte Antigen (HLA) class. In a study on ovar-
ian cancer, the MHC class II receptor α polymorphic chain, encoded by the
HLA-DRA gene, was found to be the most over-expressed gene [421]. Although
also the β chain, HLA-DRB, was over-expressed, the almost complete lack of
this polypeptide at the protein level via an unknown post-transcriptional or
post-translational mechanism of regulation, precluded the formation of a ma-
ture HLA-DR receptor and a similar pattern was observed when staining brain
166

7.1 Enrichment Analysis Results
Table 7.3: Summary of all MHC class I and II genes and the FC (F DR > 10%) of
those measured in the Tag-seq dataset.
Cytosolic pathway Exogenous pathway
Gene log 2(F C) Gene log 2(F C)
HLA-A 3.04 Alpha chains
HLA-B HLA-DMA Inf
HLA-C HLA-DQA
HLA-E 2.09 DPA1 5.73
HLA-F DQA1 Inf
HLA-G DRA 5.68
HLA-K Beta chains
HLA-L HLA-DMB
B2M HLA-DOB
PSMB5 HLA-DPB1 3.83
PSMB6 HLA-DQA2 6.56
PSMB7 HLA-DQB1 4.60
PSMB8 HLA-DQB2
PSMB9 HLA-DRB1
PSMB10 HLA-DRB3
TAP1 2.36 HLA-DRB4
TAP2 HLA-DRB5 8.36
CANX Invariant chains
CALR CD74,CLIP 7.11
TAPBPL 2.57 Transcriptional co-activator
CIITA
tumour tissue [419]. It seems that compensatory mechanisms such as cyto-
plasm over-expression of Ii/CD74 and reduction of HLA-DRB are in action to
decrease the tumour’s immunogenicity by decreasing the ability of MHC class
II to present tumour-specific antigens to the host immune system. Interest-
ingly, no such mechanism is observed in our study, in which HLA-DRA, HLA-
DRB and CD74 are all over-expressed in the GNS cell lines with a fold-change
much greater than two (Table 7.3). The same magnitude of up-regulation is
observed for the MHC class II-like peptide HLA-DM that aids in the unloading
of CD74 and loading of immature MHC class II receptors with the extracel-
lular peptides present in the lysosome to form the mature molecule. Other
two MHC class II αβ heterodimer receptors are over-expressed according to
our dataset and these are HLA-DP and HLA-DQ. In fact, both homologue
chains encoded by the HLA-DPA1 and HLA-DPB1 loci and HLA-DQA1 and
HLA-DQB1 loci are over-expressed and have the potential to form a mature
MHC class II receptor. Altogether these findings suggest that in the GNS cell
lines the exogenous pathway is not affected by compensatory mechanisms of
167

7.2 Glioblastoma Expression Signatures Results
regulation and the presentation of extracellular peptides can potentially be
completed for recognition from T helper cells. As for MHC class I-mediated
immunogenicity, our study identifies two important players as strongly up-
regulated in GNS cell lines, HLA-A and TAP1. The protein encoded by TAP1
is an ATP binding cassette transporter involved in the pumping of degraded
cytosolic peptides across the ER into the vesicles where MHC class I receptors
such as HLA-A assemble. TAP1 has been show to interact with TAPBP and
HLA-A [386]. Similarly to the expression observed for MHC class II molecules,
the over-expression of these MHC class I molecules seems to suggest that part
of the endogenous pathway might also be activated in immunosurveillance
response mechanisms in GNS cell lines. We find an overall transcriptional
up-regulation of MHC class II molecules and a smaller subset of MHC class I
molecules, suggestive of the absence of transcriptionally active compensatory
mechanisms. However, it is impossible with the available data to determine
what happens at the post-transcriptional and post-translational level because
this up-regulation should be checked against protein expression data.
7.2 Glioblastoma Expression Signatures
Analysis of microarray gene expression data for hundreds of high-grade glioma
samples and a smaller number of xenografts have shown that most tumours
can be classified into a small number of subtypes correlated with survival and
response to therapy [148,390,511]. The largest such study to date identified
four glioblastoma subtypes, each characterised by a distinct gene expression
signature encompassing 210 genes [511]. The subtypes were named "proneu-
ral", "neural", "classical" and "mesenchymal" based on which genes were up-
regulated in their respective expression signatures. To investigate whether
these subtype signatures could be captured by our Tag-seq data, Tag-seq ex-
pression data for three primary glioblastoma tumours, 11 xenografts and two
normal brain samples was analysed, which had been produced with the same
Tag-seq protocol used on our GNS and NS cell lines by Parsons et al [383].
The correlations were highly significant (p < 0.01) for all tumour and xenograft
samples and both normal brain samples (Fig 7.3), confirming that Tag-seq cap-
tures the subtype expression signatures previously observed in large microarray
datasets. Specifically, of the tumour and xenograft samples, three were classi-
fied as proneural, seven as classical and three as mesenchymal. However, both
normal brain samples and none of the glioblastoma samples, were classified as
168

7.2 Glioblastoma Expression Signatures Results
neural, consistent with this subtype being the least common and characterised
by markedly better prognosis, expressing genes associated with normal brain
and neurogenesis [390,511]. In comparing the GNS line expression profiles to
the subtype signatures, we found that both G166 and G179 correlated strongly
with the mesenchymal signature. Mesenchymal subtype markers with elevated
expression in these two lines included MET, CD44, CD68 and CASP1 [511].
G144 did not correlate significantly with any of the four signatures, but showed
a slight positive correlation (R = 0.07) with the proneural one. Supporting
such classification of G144 were several of the hallmarks of the proneural sub-
type emphasised by Verhaak et al 2010, such as high expression of oligodendro-
cytic development genes PDGFRA, NKX2-2 and OLIG2, as well as ERBB3,
DCX and TCF4 genes, and low levels of tumour suppressor CDKN1A. All
in all, the Tag-seq data seems to agree with the results from the microarray
technology, which is very reassuring given these technologies are so different.
When we verified in which of the four subtypes defined by the Verhaak et
al study [511] our 29 genes fell - the genes distinguishing GNS from NS cell
lines measured via qRT-PCR - we found that only three of them, namely
CEBPB, TES and PLS3, were represented as part of the original signature
genes. Interestingly, however, each one of the three genes fell within the mes-
enchymal subtype. The mesenchymal phenotype has recently been associated
with neoplastic transformation in the CNS as a state in which cells contrive an
uncontrolled ability to invade and stimulate angiogenesis [390,497]. Since the
defining characteristics of the aggressiveness of GBM are invasion of the local
brain parenchyma and the ability to stimulate novel angiogenesis [154,219],
our findings reflect this GBM identity. Although 26 of the 29 genes were not
represented in the original 210 gene signature defined by Verhaak et al, the
three genes that were present belonged to the mesenchymal subtype, conducive
of the fact that they are core players in establishing the characteristic aggres-
siveness of GBM (as confirmed by two separate platforms).
CEBPB, in fact, which we find to be up-regulated in GNS lines with respect
to NS lines in our qRT-PCR measurements (Fig 6.14), is an important player
in the mesenchymal regulatory module together with STAT3 and is needed
for mesenchymal transformation in human glioma cells [85]. The TES gene,
which we instead find to be expressed only in NS cell lines (Fig 6.14), has been
recently identified as a tumour suppressor that is methylated in tumours and
is responsible for repressing cell growth. It is possibly inactivated by transcrip-
tional silencing resulting from CpG island methylation [493] and is interestingly
169

7.3 Tumour Expression Correlation Results
located on the long arm of chromosome 7, which we know is gained in our GNS
cell lines. Thus, very effective transcriptional silencing mechanisms must be
actively suppressing the TES gene. Finally, PLS3 encodes for an actin-binding
protein that has yet to become associated with any form of neoplasia and that
we find to be up-regulated in our GNS cell lines with respect to NS cell lines
as measured by qRT-PCR.
Figure 7.3: Correlation with glioblastoma subtype expression signatures for tissue
samples and cell lines interrogated by Tag-seq. (a) Shows the heatmap from
the Verhaak et al [511] paper and in (b) colours indicate the correlation (Pearson
R) between subtype-specific centroid values determined by Verhaak et al [511] and
gene expression in our indicated Tag-seq measured samples. Cells showing positive
correlation are labeled with p-values indicating the significance of the correlation.
7.3 Tumour Expression Correlation
To investigate whether the identified core set of differentially expressed genes
that included 32 up-regulated and 60 down-regulated genes, showed similar
expression patterns in primary tumours as in GNS lines, we made use of a
cohort of public microarray data (see Methods, table 5.4).
The core set of differentially expressed genes included genes with established
roles in glioblastoma, such as PTEN [326] and CEBPB [85], as well as others
not previously implicated in the disease. It was our interest to investigate
170

7.3 Tumour Expression Correlation Results
how each of these genes behaved in different settings: stem cell component
of the tumour (GNS cells), primary tumour and lower grade gliomas, given
they differ significantly from one another under a biological perspective. In
fact, the tissues in the primary tumours comprise a heterogeneous mixture
of cell types, whilst our GNS culture system has been designed to maintain
the very specific stem cell component of the tumour in culture. Thus, we did
not expect a perfect agreement between tissue and cell based results. Fur-
thermore, assessing the behaviour of these same genes in lower grade gliomas
could give us insights into the key players that determine the severity of the
disease. Considering that our GNS cell lines have all been classified as primary
GBMs, the differences observed from the lower grade glioma datasets will be
especially meaningful in telling us how the same genes behave in the two dif-
ferent categories (if we approximate that a significant percentage of the lower
grade gliomas transform into higher grade gliomas such as GBMs).
Panels a and c of figure 7.4 compare the GBM data in the TCGA dataset to
the non-neoplastic brain tissue data in the TCGA dataset (altogether referred
to as dataset GBM in Methods), for the core up-regulated (a) and core down-
regulated (c) genes. Panels b and d, on the other hand, compare the GBM
(grade IV glioma) data to the grade III glioma data in the combined Phillips
and Freije datasets (altogether referred to as dataset HGG in Methods), for
the core up-regulated (b) and core down-regulated (d) genes. From this figure
we observe that there is a clear trend for the core up-regulated genes to be
more highly expressed in glioblastoma tumours than in non-neoplastic brain
tissue (Fig 7.4a; p = 0.02, randomisation test) and an opposite trend for the
core down-regulated genes (Fig 7.4c; p = 3x10 -5 ). This means that, out of all
the core differentially expressed genes that we measured with Tag-seq, a signif-
icant part of the ones we found to be more highly expressed in GNS cell lines
with respect to NS cell lines are also more highly expressed in primary GBM
with respect to non-neoplastic brain tissue (histogram bars in the direction of
positive average log2(FC) identify these genes in panel a of figure 7.4).
Similarly, a significant part of the core differentially expressed genes that we
found to be down-regulated in our GNS cell lines with respect to our NS cell
lines, are also down-regulated in primary GBM with respect to non-neoplastic
brain tissue (histogram bars in the direction of negative average log2(FC) iden-
tify these genes in panel c of figure 7.4). The colour of the histogram bars
represents the level of significance of the comparison with black indicating a
p-value0.01.
171

7.3 Tumour Expression Correlation Results
Figure 7.4: Core gene expression changes in GNS lines are mirrored in glioblastoma
tumours. Gene expression in tumours for the core up-regulated (a, b) and downregulated
(c, d) genes. The gene sets were identified by comparison of Tag-seq
expression profiles for GNS and NS cell lines. Bars depict average FC between
glioblastoma and non-neoplastic brain tissue (a, c) and between glioblastoma and
grade III astrocytoma (b, d). Black bars indicate genes with significant differential
expression in the microarray data (p < 0.01). The heatmaps show expression in
individual samples relative to the average level in non-neoplastic brain (a, c) or
grade III astrocytoma (b, d). The side dots point to the genes that belong to the
cohort of 29 differentially expressed genes measured via qRT-PCR and have the same
colour gradient adopted in figure 6.13. One gene (CHCHD10) not quantified in the
TCGA dataset is omitted from panel a.
172

7.3 Tumour Expression Correlation Results
The orientation of the bar represents the direction of FC and the height of
the bar the magnitude of the FC (−3 < F C > 3). We hypothesised that ex-
pression of these genes might also differ between glioblastoma and less severe
gliomas and, upon examination of the expression patterns for grade III gliomas
and glioblastoma in the dataset HGG, we found the core up-regulated genes to
be more highly expressed in glioblastoma than in grade III glioma (Fig 7.4b;
p < 10 -6 ), while the core down-regulated genes showed the opposite pattern
(Fig 7.4d; p < 10 -5 ). Thus, of the core differentially expressed genes we mea-
sured with Tag-seq, a significant portion of the highly expressed ones in GNS
cell lines with respect to NS cell lines, were also more highly expressed in grade
IV gliomas (GBMs) with respect to grade III gliomas (histogram bars in the
direction of positive average log2(FC) identify these genes in panel b of figure
7.4). This trend may be indicative of the fact that the core up-regulated genes
in our GNS cell lines that are also up-regulated in GBM primary tumours,
belong to a cohort of genes involved in regulatory networks already activated
in grade III gliomas. The very fact that the same genes are similarly over ex-
pressed in our GNS cell lines and in grade III gliomas may be an indication of a
specific prognosis, namely the progression to a higher grade glioma. Of course,
at this level of analysis this is only a hypothesis and it would be interesting to
test it further and try to establish if it can be used as a pre-diagnostic tool.
Similarly to the matching trends observed for the core up-regulated genes,
a significant portion of the differentially expressed genes that we found to
be down-regulated in GNS cell lines with respect to NS cell lines, were also
down-regulated in grade IV gliomas (GBMs) with respect to grade III gliomas
(histogram bars in the direction of negative average log2(FC) identify these
genes in panel d of figure 7.4). This opposite trend is not surprising consid-
ering the hypothesis mentioned before. If the regulatory networks involved in
the progression of the disease from a grade III to a grade IV have already been
activated in grade III gliomas, then a set of complementary genes and regu-
latory networks must be silenced and rendered transcriptionally inactive. In
this scenario postulating a diagnostic tool that derives from the combination
of the two oppositely moving groups of genes may serve a pre-diagnostic tool
purpose well. Of course, a specific study would need to be carried to verify the
validity of this hypothesis.
The dots next to the gene names on the sides of the heatmaps in figure 7.4 are
there to highlight whether a gene also belongs to the cohort of the 29 genes
found via qRT-PCR to distinguish GNS cell lines from NS cell lines. The colour
173

7.3 Tumour Expression Correlation Results
of the dot indicates the direction (red or blue) and the magnitude (intensity
of red or blue) of the FC as measured via qRT-PCR and in doing so adopts
the same colour scheme of the heatmap in 7.4. Table 7.4 summarises all the
information gathered in the literature searches for the 29 genes differentially
expressed as measured via qRT-PCR, information that is expanded upon in
the paragraphs below.
Up-regulated in GNS cell lines. The genes HOXD10, CD9, PLA2G4A,
MT2A, SULF2, DDIT3, PLS3, CEBPB, PRSS12 and LYST are all highly ex-
pressed in both GNS cell lines and primary GBMs. On the contrary, genes
FOXG1, LMO4, ADD2 and PDE1C are up-regulated in GNS cell lines but are
down-regulated in primary GBMs.
Interestingly, FOXG1 is proposed to act as an oncogene in GBM by suppress-
ing the growth inhibitory effects of TGFβ [448], so that the up-regulation we
observe in GNS cell lines (Fig 7.4a) could be a cell type specific regulation that
especially affects the stem cell component of the tumour. This would corrob-
orate the hypothesis that NS cells are candidates for tumour-initiating cells
in GBM. Also this up-regulation is not mirrored in grade III astrocytomas,
indicating different oncogenic factors in action (Fig 7.4b).
The transcriptional regulator LMO4 is of particular interest as it is involved in
the development of multiple organs, including the CNS, and its expression is
elevated in several cancers [202,339,475,488,542]. LMO4 is especially well stud-
ied as an oncogene in breast cancer and regulated through the phosphoinosi-
tide 3-kinase pathway [340], which is commonly affected in glioblastoma [326].
Similarly to FOXG1, the up-regulation of LMO4 in GNS cell lines but not
in primary GBM (Fig 7.4a) could be reflective of its oncogenic role in GBM
tumour-initiating cells. Interestingly, LMO4 could be an oncogene early at
work in the progression of the disease from a lower to a higher grade, as its
up-regulation is observed in grade III astrocytomas (Fig 7.4b).
The ADD2 gene encodes a cytoskeletal protein that interacts with FYN, a
tyrosine kinase promoting cancer cell migration [456,533]. The up-regulation
of this gene in GNS cell lines is reflective of the invasiveness that charac-
terises GBM, and is possibly lost through differentiation since ADD2 appears
to be down-regulated in primary GBM data (Fig 7.4a). The small level of
up-regulation observed for this gene in grade III astrocytomas (Fig 7.4b) is a
reflection of its activity early on in the progression of the disease.
PDE1C is a cyclic nucleotide phosphodiesterase gene that we observe to be up-
regulated in GNS cell lines and down-regulated in primary GBM (Fig 7.4a).
174

7.3 Tumour Expression Correlation Results
Table 7.4: Literature survey for the 29 genes found to distinguish GNS from NS lines across a panel of 21 cell lines. The table details whether each
gene has previously been implicated in glioma or other neoplasias, and includes references to relevant publications.
Gene (aliases) Category Evidence in Evidence Selected references (PubMed IDs)
glioma? other cancers?
ADD2 Up-regulated in GNS No No Shima et al. 2001 (11526103), Pariser et al. 2005 (16105548), Ferrandi et al. 2010 (19838659)
CD9 Up-regulated in GNS Yes Yes Shi et al. 2000 (10662783), Kawashima et al. 2002 (12185197), Zöller 2009 (19078974), Kolesnikova et al. 2009
(19107234), Lafleur and Hemler 2009 (19211836)
CEBPB Up-regulated in GNS Yes Yes Homma et al. 2006 (16465418), Nerlov 2007 (17658261), Zahnov 2009 (19351437), Carro et al. 2010 (20032975)
DDIT3 Up-regulated in GNS Limited Yes Oyadomari and Mori 2004 (14685163), Ragel et al. 2007 (17486380), Kaul and Maltese 2009 (19724676), Meng et
al. 2009 (19549908)
FOXG1 Up-regulated in GNS Yes Yes Seoane et al. 2004 (15084259), Adesina et al. 2007 (17522785)
HMGA2 Down-regulated in GNS Limited Yes Akai et al. 2004 (15497774), Young and Narita 2007 (17473167), Cleynen and Van de Ven 2008 (18202751), Liu et
al. 2010 (20368557)
HOXD10 Up-regulated in GNS Limited Yes Osborne et al. 1998 (9773404), Reddy et al. 2008 (18922890), Ma et al. 2007 (17898713), Baffa et al. 2009
(19593777), Sasayama et al. 2009 (19536818), Sun et al. 2011 (21419107)
IRX2 Down-regulated in GNS No Yes Adamowicz et al. 2006 (16752383)
LMO4 Up-regulated in GNS No Yes Sum et al. 2002 (11751867), Mizunuma et al. 2003 (12771919), Sum et al. 2005 (15897450), Taniwaki et al. 2006
(16865272), Yu et al. 2008 (19099607), Montanez-Wiscovich et al. 2009 (19648968)
LYST Up-regulated in GNS No No Kaplan et al. 2008 (18043242)
MAF Down-regulated in GNS No Yes Rasmussen et al. 2003 (14692531), Hurt et al. 2004 (14998494), Murakami et al. 2007 (18059226), Peng et al. 2007
(17823980), Natkunam et al. 2009 (19687312)
MAP6 Down-regulated in GNS No Yes Vater et al. 2009 (19016712)
MT2A Up-regulated in GNS Yes Yes Maier et al. 1997 (9444362), Cui et al. 2003 (12646258), Yamasaki et al. 2007 (17914565), Krona et al. 2007
(17982672), Lim et al. 2009 (19062161), Puca et al. 2009 (18996371)
MYL9 Down-regulated in GNS No Yes Medjkane et al. 2009 (19198601), Lu et al. 2010 (21139803)
NDN Down-regulated in GNS Limited Yes Aizawa et al. 1992 (1394972), Hu et al. 2003 (12913118), Chapman and Knowles 2009 (19626646), Jörnsten et al.
2011 (21525872)
NELL2 Down-regulated in GNS Limited Yes Kuroda et al. 1999 (10548494), Maeda et al. 2001 (11803583), DiLella et al. 2001 (11304808), Nelson et al. 2004
(15183717), Fassunke et al. 2008 (18819986)
PDE1C Up-regulated in GNS No No Das and Sharma 2005 (16142372), Vatter et al. 2005 (15816855), Dolci et al. 2006 (16455054)
PLA2G4A Up-regulated in GNS Limited Yes Hernandez et al. 2000 (10838595), Moolwaney and Igwe 2005 (15950779), Linkous et al. 2010 (20729478), Jeong et
al. 2010 (20944117), Han et al. 2010 (20683962), Caiazza et al. 2011 (21119660)
PLCH1 Down-regulated in GNS No No Stewart et al. 2007 (17895620), Kim et al. 2011 (21262355)
PLS3 Up-regulated in GNS No Yes Lin et al. 1993 (8428952), Arpin et al. 1994 (7806577), Su et al. 2003 (14612505), Kari et al. 2003 (12782714),
Ikeda et al. 2005 (16142308), Capriotti et al. 2008 (18569641)
PRSS12 Up-regulated in GNS No No Mitsui et al. 2007 (17223089), Matsumoto-Miyai et al. 2009 (19303856)
PTEN Down-regulated in GNS Yes Yes The Cancer Genome Atlas Research Network 2008 (18772890), Hollander et al. 2011 (21430697)
SDC2 Down-regulated in GNS Limited Yes Fears et al. 2006 (16574663), Watanabe et al. 2006 (16132527), Theocaris et al. 2008 (20840587)
ST6GALNAC5 Down-regulated in GNS Yes Yes Bos et al. 2009 (19421193), Kroes et al. 2010 (20616019), Oster et al. 2011 (21400501)
SULF2 Up-regulated in GNS Yes Yes Johansson et al. 2005 (15750623), Morimoto-Tomita et al. 2005 (16331886), Dai et al. 2005 (16192265), Lemjabbar-
Alaoui et al. 2010 (19855436), Lai et al. 2010 (20725905), Phillips et al. 2012 (22293178)
SYNM Down-regulated in GNS Yes Yes Jing et al. 2005 (15657940), Pan et al. 2008 (18509200), Noetzel et al. 2010 (20543860), Sun et al. 2010 (19853601),
Liu et al. 2011 (21144834), Pitre et al. 2012 (22337773)
TAGLN Down-regulated in GNS Limited Yes Gunnersen et al. 2000 (11008214), Shields et al. 2002 (11773051), Yeo et al. 2006 (16402363), Assinder et al. 2009
(18378184), Zhao et al. 2009 (19329940), Kim et al. 2010 (20705054), Yeo et al. 2010 (20336793), Prasad et al.
2010 (20012321)
TES Down-regulated in GNS Yes Yes Tatarelli et al. 2000 (10950921), Tobias et al. 2001 (11420696), Mueller et al. 2007 (16909125), Martinez et al.
2009 (19550145), Gunduz et al. 2009 (19289703), Ma et al. 2010 (20626849), Weeks et al. 2010 (20573277), Qui et
al. 2010 (20180808)
TUSC3 Down-regulated in GNS Limited Yes MacGrogan et al. 1996 (8661104), Bookstein et al. 1997 (9088270), Ahuja et al. 1998 (9850084), Li et al. 1998
(9671399), Bashyam et al. 2005 (16036106), Pils et al. 2005 (16270321), Guervos et al. 2007 (17641416), Cooke et
al. 2008 (18840272), Arasaradnam et al. 2010 (20505342), Bui et al. 2010 (19812376)
175

7.3 Tumour Expression Correlation Results
Up-regulation of PDE1C has been associated with proliferation in other cell
types through hydrolysis of cAMP and cGMP [126,437]. Up-regulation in
GNS cell lines may foster the proliferation that characterises these cells, which
may be lost in more differentiated progeny belonging to the primary tumour
sample. In figure 7.4b we observe a down-regulation of PDE1C in grade III
astrocytomas.
Altogether, FOXG1, LMO4, ADD2 and PDE1C may be part of the GNS ex-
pression profile that defines the stem cell identity of GBM and is therefore lost
through the differentiation pathways undertaken by the rest of the tumour
cells that do not retain their stem cell identities, according to the cancer stem
cell hypothesis.
Of the ten genes that are highly expressed in both GNS cell lines and primary
GBMs, HOXD10 encodes a protein with a homeobox DNA-binding domain
that is known to be involved in limb development and differentiation [82].
HOXD10 protein levels are suppressed by a microRNA (miR-10b) which is
highly expressed in gliomas, and it has been suggested that HOXD10 suppres-
sion by miR-10b promotes invasion [476]. Interestingly, the HOXD10 mRNA
up-regulation we observe in GNS cell lines and GBM tumours is not mirrored
in grade III astrocytoma, perhaps reflective of the less invasive phenotype. In
fact, miR-10b is present at higher levels in glioblastoma compared to gliomas
of lower grade [476].
The CD9 gene encodes a cell-surface glycoprotein, or antigen, that has been
previously implicated in glioma with a role in adhesion and migration. In the
CNS CD9 is expressed in the myelin sheath and is believed to suppress the
metastatic potential of some human tumours including gliomas [221]. We find
up-regulation of CD9 in GNS cell lines as well as primary GBM (Fig 7.4a),
but a down-regulation in grade III astrocytoma (Fig 7.4b), perhaps reflective
of the lower invasive potential of the latter.
The PLA2G4A gene encodes a phospholipase enzyme that catalyses the hy-
drolysis of membrane phospholipids to lipid-based cellular hormones that then
regulate a variety of intracellular pathways. PLA2G4A has not been linked to
GBM but has been implicated in other neoplasias [192,285,341]. We observed
an up-regulation of PLA2G4A in GNS cell lines as well as primary GBM (Fig
7.4a), but a down-regulation in grade III astrocytoma (Fig 7.4b).
The MT2A gene encodes a metallothionein that has yet to be implicated in
gliomas, but has been observed to interact with the kinase domain of a member
of the PKC family in prostate cancer [422] and TP53 in breast cancer epithe-
176

7.3 Tumour Expression Correlation Results
lial cells to possibly regulate apoptosis in the latter [373]. Also in the case of
MT2A we observed an up-regulation in GNS cell lines as well as primary GBM
(Fig 7.4a), but a down-regulation in grade III astrocytoma (Fig 7.4b).
The SULF2 gene encodes a sulfatase that edits the sulfation status of heparan
sulfate proteoglycans on the outside of cells and, in this way, regulates criti-
cal signaling pathways [433]. Disregulation of SULF2 has been implicated in
non-small cell lung cancer [268], pancreatic cancer [357], hepatocellular car-
cinoma [254], breast cancer [344], and gliomas, in which knock-down of the
SULF2 gene resulted in decreased GBM growth in vivo in mice. Molecu-
larly, ablation of SULF2 resulted in decreased PDGFRα phosphorylation and
decreased downstream MAPK signaling activity. Interestingly, of this obser-
vation on the proneural GBM subtype defined by Verhaak et al [511] that is
characterized by aberrations in PDGFRα, showed the strongest SULF2 expres-
sion [391]. In our observations SULF2 was up-regulated in GNS cell lines and
primary GBM (Fig 7.4a), in line with the observations of Phillips et al [391]
that made it a candidate oncogene. We observed SULF2 to also be strongly
up-regulated in grade III astrocytoma (Fig 7.4b) indicating the possibility of a
regulatory impact on behalf of SULF2 early in the progression of the disease.
The DDIT3 gene encodes the pro-apoptotic protein CHOP that is known to
drive the down-regulation of the anti-apoptotic mitochondrial protein Bcl-2,
thereby favouring apoptosis through the activation of cytochrome c and cas-
pase 3. Studies in hepatoma and pheochromocytoma cell lines have shown that
the transcription factor encoded by CEBPB (C/EBPβ) promotes the expres-
sion of DDIT3 [324] and thus of CHOP, which in turn can inhibit C/EBPβ by
dimerizing with it and acting as a dominant negative [324]. This interplay be-
tween CEBPB and DDIT3 may be relevant for glioma therapy development, as
DDIT3 induction in response to a range of compounds sensitises glioma cells to
apoptosis [217]. In line with the pro-apoptotic role of DDIT3 described, we ob-
served up-regulation of the gene in GNS cell lines and primary GBM (Fig 7.4a)
and down-regulation in grade III astrocytoma (Fig 7.4b); similarly, we found
CEBPB to be up-regulated in GNS cell lines and primary GBM (Fig 7.4a),
and down-regulated in grade III astrocytoma (Fig 7.4b). PLS3 (T-plastin)
encodes a regulator of actin organisation and its over-expression in the CV-1
fibroblast-like cell line resulted in partial loss of adherence [27]. The elevated
levels of PLS3 expression we observe in GNS cell lines and primary GBM may
thus be relevant to the invasive phenotype. Accordingly, the less invasive lower
grade III astrocytoma shows down-regulation of PLS3 (Fig 7.4b).
177

7.3 Tumour Expression Correlation Results
The PRSS12 gene encodes a protease that can activate tissue plasminogen acti-
vator (tPA) [335], an enzyme which is highly expressed by glioma cells and has
been suggested to promote invasion [168]. We observe a slight up-regulation
in the GNS cell lines and primary GBM of PRSS12 (Fig 7.4a), and a slight
down-regulation in grade III astrocytomas (Fig 7.4b).
The LYST gene encodes a vesicular transport protein called the "lysosomal
trafficking regulator" that so far has not been implicated in any neoplasia but
is known to be associated with a rare recessive disorder (Chédiak-Higashi syn-
drome) caused by a microtubule polymerisation defect that decreases phago-
cytosis ability [87]. In line with the hypothetical increase of cellular activities
and rates, and thus of vesicular transport, in GNS cells, we observed LYST to
be up-regulated in GNS cell lines and primary GBM (Fig 7.4a), and slightly
down-regulated in grade III astrocytomas (Fig 7.4b).
Down-regulated in GNS cell lines. The panels c and d of figure 7.4 the core
down-regulated genes found through Tag-seq are used to evaluate the compar-
ison of the GBM data to the non-neoplastic tissue data (TCGA dataset), and
the comparison of the GBM (grade IV) to the grade III astrocytoma (HGG
dataset). In this comparison the genes that were also identified via qRT-PCR
as being differentially expressed between GNS cell lines and NS cell lines, are
highlighted by the blue colour gradient dots. The trends highlighted in figure
7.4 show matching down-regulation for genes NELL2, TUSC3, ST6GALNAC5,
PLCH1, NDN, MAP6 and PTEN in GNS cell lines and primary GBM, and non
matching down-regulation in GNS cell lines and up-regulation in primary GBM
for genes MAF, MYL9, HMGA2, SDC2, SYNM, IRX2, TES and TAGLN.
The MAF gene encodes a transcription factor and oncoprotein that belongs to
the same AP-1 super-family of JUN and FOS. MAF has been associated with
multiple myeloma whereby its up-regulation is suggested to enhance myeloma
proliferation and adhesion to the bone marrow [406]. Although not previously
linked to glioma, we find MAF to be down-regulated in GNS cell lines but
slightly up-regulated in primary GBM, perhaps as an indicator of its cell type
restricted proliferation enhancing functions (Fig 7.4c). Accordingly, we ob-
served a slight down-regulation of MAF in grade III astrocytoma with respect
to primary GBM (Fig 7.4d).
The protein encoded by the MYL9 gene is a myosin light chain that has
previously been associated with the stem cell component of lung adenocar-
cinoma [447] and medullary breast cancer [54] in gene expression profiling
studies. MYL9 has never been implicated in glioma and we observed down-
178

7.3 Tumour Expression Correlation Results
regulation of its expression in GNS cell lines opposed to its up-regulation in
primary GBM (Fig 7.4c), and a strong down-regulation in grade III astrocy-
toma (Fig 7.4d).
The HMGA2 gene encodes a transcriptional regulator that belongs to the non-
histone family of structural proteins. The members of this family act as chro-
matin architectural factors and contain structural DNA-binding domains that
allow them to act as transcriptional factors. We find HMGA2 to be down-
regulated in GNS lines and up-regulated in primary GBM (Fig 7.4c), and
slightly down-regulated in grade III astrocytoma (Fig 7.4d). Low or absent
protein expression of HMGA2 has been observed in GBM compared to low
grade gliomas [14] and HMGA2 polymorphisms have been associated with
survival time in GBM [295].
The SDC2 gene encodes a transmembrane heparan sulfate proteoglycan that
participates as an extracellular matrix receptor in the processes of cell prolifera-
tion, cell migration and cell-matrix interaction. An altered expression of SDC2
has been detected in esophageal carcinoma [201], colon carcinoma [184], fi-
brosarcoma [380], prostate cancer [108,407] and gliomas, where over-expression
of SDC2 promotes membrane protrusion, migration, capillary tube formation
and cell-cell interactions in microvascular endothelial cells [141]. We found
SDC2 to be down-regulated in GNS cell lines and grade III astrocytoma, but
slightly up-regulated in primary GBM (Fig 7.4).
The SYNM gene is a type IV intermediate filament that has recently been
shown to interact with the LIM domain protein Zyxin, thereby possibly mod-
ulating cell adhesion and cell motility. Aberrant SYNM promoter methylation
has been associated with early breast cancer relapse [362]. In gliomas SYNM
has been found to promote AKT-dependent GBM cell proliferation by antago-
nising protein phosphatase PP2A, the major regulator of Akt dephosphoryla-
tion [396]. We found SYNM to be down-regulated in GNS cell lines and grade
III astrocytoma, but slightly up-regulated in primary GBM, although with a
relatively high p-value (Fig 7.4). The IRX2 gene is a iroquois-class homeobox
genes that has been associated with development of the vertebrate embryo
and in humans specifically in the development of brain [322] and breast [274].
Over-expression of IRX2 has been detected in soft tissue sarcomas [7]. IRX2
has been proposed to enhance antitumour immune responses in that ex vivo
pre-treatment of CD8 + T cells with IRX-2 provided protection from tumour-
induced apoptosis [112]. In line with this suggested role of IRX2, we observed
the gene to be strongly down-regulated in GNS cell lines and grade III astro-
179

7.3 Tumour Expression Correlation Results
cytoma, but up-regulated in primary GBM (Fig 7.4).
Finally, we found the TAGLN and TES genes to be absent or low in most GNS
lines, but displaying the opposite trend in GBM tissue compared to normal
brain (Fig 7.4c) or grade III astrocytoma (Fig 7.4d). Interestingly, the TES
gene is located on chromosome 7, which is known to be gained in all our GNS
cell lines (see 6.3 section). A strong regulatory effect must be in action in
GNS cell lines to effect the down-regulation of TES. Similarly to the SYNM
gene, TES has been shown to interact with the LIM domain protein Zyxin and
therefore is believed to have a role in cell motility and is often found in focal
adhesions [109]. The TAGLN gene encodes an actin protein found in fibrob-
lasts and smooth muscle, the expression of which is down-regulated in many
cell lines and may be an early marker for the onset of transformation [260].
Both TAGLN and TES have been characterised as tumour suppressors in ma-
lignancies outside the brain and TES is known to often be silenced by promoter
hypermethylation in GBM [30,349].
Altogether, MAF, MYL9, HMGA2, SDC2, SYNM, IRX2, TES and TAGLN
may be part of a GNS expression profile that helps define the stem cell identity
of these cells and the tumour that derives from them according to the cancer
stem cell hypothesis. Interestingly, the expression pattern of most of these
genes is mirrored in grade III astrocytomas.
Of the seven genes that are down-regulated in both GNS cell lines and pri-
mary GBMs, TUSC3 is a candidate tumour suppressor known to be silenced
by promoter methylation in GBM, particularly in patients over 40 years of
age. Loss or down-regulation of TUSC3 has been found in other cancers, such
as colon cancer, where its promoter becomes increasingly methylated with age
in the healthy mucosa [12]. These data suggest that transcriptional changes
in healthy aging tissue, such as TUSC3 silencing, may contribute to the more
severe form of glioma in older patients. In line with its role as tumour suppres-
sor, we found TUSC3 to be down-regulated in GNS cell lines, primary GBM
and grade III astrocytoma (Fig 7.4).
The PLCH1 gene is a member of the phospholipase C family of enzymes that
cleave phosphatidylinositol 4,5-bisphosphate to generate second messengers in-
ositol 1,4,5-trisphosphate and diacylglycerol. PLCH1 is thus involved in phos-
phoinositol signaling [228], just like the frequently mutated phosphoinositide
3-kinase complex [326]. We found PLCH1 to be down-regulated in GNS cell
lines and primary GBM (Fig 7.4c) and slightly up-regulated in grade III as-
trocytoma (Fig 7.4d).
180

7.3 Tumour Expression Correlation Results
The PTEN lipid phosphatase gene is a well known tumour suppressor that
is frequently deleted or mutated in GBM and its down-regulation affects the
correct regulation of the RTK/PI3K/PTEN pathway in which it acts as an
instrumental player [326]. PTEN is mutated in a variety of other cancers be-
sides gliomas, including prostate, breast, endometrial cancer and melanoma
[377,378,439,514]. We observe the expression of PTEN to be down-regulated
in GNS cell lines and primary GBM (Fig 7.4c) but up-regulated in grade III
astrocytoma (Fig 7.4d).
The ST6GALNAC5 gene belongs to the sialyltransferase family of proteins
that modify proteins and ceramides on the cell surface to alter cell-cell or cell-
extracellular matrix interactions [498]. ST6GALNAC5 facilitates the transmi-
gration of cancer cells through the blood-brain barrier and is known to mediate
breast cancer metastasis to the brain [67]. In GBM ST6GALNAC5 has already
been observed to be down-regulated with respect to normal brain tissue [250],
and we observed it to be down-regulated in GNS cell lines as well as primary
GBM and slightly up-regulated in grade III astrocytoma (Fig 7.4).
The NDN gene is an imprinted gene expressed exclusively from the pater-
nal allele that has no introns and acts as a growth suppressor. Studies in
mice suggest that the Necdin protein suppresses growth in postmitotic neu-
rons [209,487]. Necdin has also been suggested to interact with and negatively
regulate HIF1α, a factor that mediates cellular homeostatic responses like
angiogenesis to reduce O2 availability [342]. The expression of NDN is also
known to be down-regulated due to the transcriptional control implemented
by STAT3 in human melanoma, prostate cancer and breast cancer [188]. We
found NDN to be strongly down-regulated in GNS cell lines and primary GBM
but strongly up-regulated in grade III astrocytoma (Fig 7.4).
The MAP6 gene encodes a microtubule-associated protein that is calmodulin-
binding and calmodulin-regulated and is therefore involved in microtubule sta-
bilisation. MAP6 has yet to be associated with any neoplasia and we observed
it to be down-regulated in GNS cell lines and primary GBM, and slightly
up-regulated in grade III astrocytoma (Fig 7.4). Finally, the NELL2 gene en-
codes a glycoprotein containing several EGF-like repeats that is found in the
cytoplasm and is involved in neural cell growth and differentiation as well as
oncogenesis [521]. Together with NELL1, NELL2 is predominantly expressed
in neuroblastoma and other embryonal neuroepithelial tumours [305], but has
yet to be characterised in glioblastoma. We observed its expression to be down-
regulated in GNS cell lines and primary GBM, as well as grade III astrocytoma
181

7.4 Survival Analysis Results
(Fig 7.4). Overall, we can confirm that the set of core differentially expressed
genes identified by Tag-seq defines an expression signature characteristic of
glioblastoma and related to glioma histological grade.
7.4 Survival Analysis
To explore the relevance in glioma of the patterns observed for the GNS vs. NS
cell line comparison, we decided to integrate clinical information with tumour
expression data. Although this analysis was not performed by the candidate it
is included in this thesis because it is very relevant to the rest of the material
and analyses performed by the candidate and will therefore help paint for the
reader a more complete picture of the relevance of GNS cell lines in glioblas-
toma cancer research.
We first tested for associations between gene expression and survival time using
the TCGA dataset, consisting of 397 glioblastoma cases (see Methods section
5.9 for table 5.4). For each gene, we fitted a Cox proportional hazards model
with gene expression as a continuous explanatory variable and computed a
p-value by the score test (Table 7.5). The set of 29 genes found to distinguish
GNS cells from NS cells across the 22 cell lines assayed by qRT-PCR was en-
riched for low p-values compared to the complete set of 18,632 genes quantified
in the TCGA dataset (p = 0.02, one-sided Kolmogorov-Smirnov test), demon-
strating that expression analysis of GNS and NS lines had enriched for genes
associated with patient survival. Seven of the 29 genes had a p-value

7.4 Survival Analysis Results
Table 7.5: Survival tests for the 29 genes found via qRT-PCR to distinguish GNS
cell lines from NS cell lines.
Gene Category TCGA dataset Gravendeel dataset (GBM cases)
Coefficient* p-value Probeset** Coefficient* p-value
ADD2 Up -0.13 0.2858 237336_at -0.17 0.1420
CD9 Up 0.18 0.0731 201005_at 0.17 0.0689
CEBPB Up 0.19 0.1028 212501_at 0.17 0.0651
DDIT3 Up 0.17 0.0128 209383_at 0.09 0.2777
FOXG1 Up 0.13 0.0861 206018_at 0.11 0.0380
HMGA2 Down 0.13 0.1456 1561633_at -0.84 0.2459
HOXD10 Up 0.12 0.0108 229400_at 0.15 0.0021
IRX2 Down -0.19 0.2346 228462_at -0.20 4.4x10 -4
LMO4 Up 0.24 0.1046 209205_s_at 0.20 0.1435
LYST Up 0.05 0.5590 203518_at 0.10 0.4151
MAF Down 0.10 0.5873 209348_s_at 0.38 0.0074
MAP6 Down 0.16 0.3063 235672_at -0.30 0.0087
MT2A Up 0.16 0.1554 212185_x_at 0.27 0.0127
MYL9 Down 0.08 0.3764 201058_s_at 0.15 0.0252
NDN Down -0.04 0.4874 209550_at -0.22 6.0x10 -5
NELL2 Down 0.08 0.1021 203413_at 0.14 0.0215
PDE1C Up 0.20 0.0105 236344_at 0.21 0.0134
PLA2G4A Up -0.06 0.3198 210145_at 0.30 2.9x10 -4
PLCH1 Down 0.10 0.3165 214745_at 0.45 0.0094
PLS3 Up 0.13 0.0381 201215_at 0.30 0.0069
PRSS12 -0.11 0.1865 213802_at 0.20 0.0296
PTEN Down -0.53 0.0047 228006_at -0.40 0.0062
SDC2 Down 0.22 0.0044 212158_at 0.28 5.8x10 -4
ST6GALNAC5 Down 0.01 0.9116 220979_s_at 0.08 0.2416
SULF2 -0.11 0.1525 233555_s_at -l0.15 0.0930
SYNM Down -0.06 0.5620 212730_at 0.08 0.2613
TAGLN Down 0.03 0.5947 205547_s_at 0.17 0.0030
TES Down -0.05 0.5759 202720_at 0.07 0.5499
TUSC3 Down -0.14 0.0079 209227_at -0.18 0.0060
*Fitted coefficient from Cox model; a positive coefficient indicates that higher expression
is associated with poor survival and a negative coefficient indicates the
opposite. ** For the Gravendeel dataset, the result for the most significant probeset
interrogating the gene is shown; Up=Up-regulated; Down=Down-regulated.
To test whether these findings generalise to independent clinical sample groups,
we examined the glioblastoma datasets described by Gravendeel et al [176] and
Murat et al [353], consisting of 141 and 70 cases, respectively (see Methods
section 5.9 for table 5.4). The GNS signature score was correlated with pa-
tient survival in both of these datasets (p = 3x10 5 and p = 0.006, respectively;
Fig 7.5a). At the level of individual GNS signature genes, five genes were
significantly associated with survival (p < 0.05) in both of the two largest
183

7.4 Survival Analysis Results
Figure 7.5: Association between GNS signature and other survival predictors. (a)
Scatter plots demonstrate the correlation between GNS signature score and age at
diagnosis for the TCGA (left) and Gravendeel (right) datasets. The regression line,
Pearson correlation coefficient (r) and p-value indicating statistical significance of
the correlation are shown. (b) GNS signature score for samples in the Gravendeel
dataset, stratified by IDH1 mutation status and histological grade. Blue circles
represent individual samples (independent cases) and grey boxplots summarise their
distribution. Only cases with known IDH1 status are shown (127 mutated, 77 wild
type).
glioblastoma datasets we investigated (TCGA and Gravendeel): HOXD10,
PDE1C, PLS3, PTEN and TUSC3 (Table 7.5). In addition to glioblastoma
(grade IV) tumours, Gravendeel et al also characterised 109 grade I-III glioma
cases (see Methods section 5.9). Inclusion of these data in survival analyses
made the association with GNS signature even more apparent (Fig 7.5b). This
is consistent with the observation made in section 7.3 whereby core transcrip-
tional alterations in GNS cells correlated with histological grade of primary
tumours. Analysis of data from the studies of Phillips et al [390] and Freije et
al [148], which profiled both grade III and IV gliomas (see Methods section 5.9
for table 5.4), further confirmed the correlation between GNS signature and
survival (Figure 7.5b). In summary, the association between GNS signature
and patient survival was reproducible in five independent datasets comprising
184

7.4 Survival Analysis Results
867 glioma cases in total (see Methods section 5.9 for table 5.4).
In trying to investigate the presence of a relationship to known predictors of
survival in glioma, we noted that the GNS signature scores correlated with
patient age at diagnosis, suggesting that the GNS-related expression changes
were associated with the more severe form of the disease observed in older pa-
tients (Figure 7.6a). Of the genes contributing to the GNS signature, HOXD10,
PLS3, PTEN and TUSC3 correlated with age both in the TCGA and Graven-
deel datasets.
Figure 7.6: Association between GNS signature score and patient survival. Kaplan-
Meier plots illustrate the association between signature score and survival for (a)
three independent glioblastoma datasets and (b) three datasets that include gliomas
of lower grade (see Methods section 5.9). Higher scores indicate greater similarity
to the GNS expression profile. Hazard ratios and log-rank p-values were computed
by fitting a Cox proportional hazards model to the data. Percentile thresholds were
chosen for illustration; the association with survival is statistically significant across
a wide range of thresholds and the p-values given in the text and Table 7.5 were
computed without thresholding, using the score as a continuous variable.
IDH1 mutation affecting codon 132 of the IDH1 gene is present in most grade
III astrocytomas and a minority of glioblastomas, resulting in an amino acid
change (R132H, R132S, R132C, R132G, or R132L). The presence of this muta-
tion is associated with lower age at disease onset and better prognosis [383,517].
As already mentioned in section 6.1, all 16 GNS lines profiled in this study
were derived from glioblastoma tumours, and the IDH1 locus was sequenced in
each cell line (data not shown) and none of the cell lines appeared to harbour
the mutation. We therefore wanted to investigate whether the GNS signature
was characteristic of IDH1 wild-type glioblastomas or not. We could perform
this analysis thanks to the fact that IDH1 status had been determined for most
cases in the TCGA and Gravendeel datasets (Table 7.5) [176,326,511]. As ex-
pected, we found that gliomas with the IDH1 mutation tend to have lower GNS
185

7.4 Survival Analysis Results
Table 7.6: Significance of survival association for GNS signature and IDH1 status.
dataset Number
of cases
Single covariate Two covariates (GBM cases)
GNS signature IDH1 status GNS signature IDH1 status
TCGA 270 5.3x10-5 0.0015 0.0091 0.1489
Gravendeel,
GBM cases
118 2.7x10-5 0.0031 9.2x10-4 0.0840
Gravendeel
et al, grade
I-III cases
86 6.5x10-4 0.5776 6.3x-4 0.5408
Wald test p-values, indicating association with survival, for each covariate in a Cox
proportional hazards model with one or two covariates (GNS signature, IDH1 status
or both). Cases with unknown IDH1 mutation status were excluded.
signature scores than IDH1 wild-type gliomas of the same histological grade
(Fig 7.6b). However, we also found that the GNS signature bore a stronger
survival association than the IDH1 status (Table 7.6). In fact, the signature
remained a significant predictor of patient survival when controlling for IDH1
status (Table 7.6), demonstrating that it contributes independent information
to the survival model and does not simply represent a transcriptional state of
IDH1 wild-type tumours. This was evident in glioblastomas as well as grade
I-III gliomas; the effect is thus not limited to grade IV tumours.
Finally, to investigate whether the correlation between GNS signature and
age could be explained by the higher proportion of cases with IDH1 mutation
among younger patients, we repeated the correlation analysis described above
(Fig 7.6a), limiting the data to glioblastoma cases without IDH1 mutation. For
the TCGA dataset, the correlation was decreased somewhat (Pearson R = 0.25
compared to R = 0.36 for the full dataset) but still highly significant (p =
6x10 5 ), demonstrating that the correlation with age is only partially explained
by IDH1 status. This result was confirmed in the Gravendeel dataset, where
the effect of controlling for IDH1 status and grade was negligible (R = 0.38
compared to R = 0.39 for the full dataset including grade I-III samples).
Among the individual signature genes, both HOXD10 and TUSC3 remained
correlated with age in both datasets when limiting the analysis to IDH1 wild-
type glioblastoma cases.
186

7.5 Glioblastoma Pathway Analysis Results
7.5 Glioblastoma Pathway Analysis
In order to identify differentially expressed genes within pathways known to be
perturbed in glioma and pathways related to the progenitor cell state, I man-
ually built and curated an integrated pathway map by gathering information
about interactions and interactors from the literature and the information data
bases described in the Methods chapters. The glioblastoma pathways that al-
ready exist in the literature, such as the KEGG "Glioma" pathway [3], are not
as comprehensive as the present level of understanding of the disease requires
of them. These existing pathways are shortcoming in four specific ways:
1. they do not take into account the contribution from the stem cell com-
ponent of the tumour and, thus, the progenitor cell state remains unrep-
resented;
2. core cell-cycle regulatory pathways that are very relevant in cancer, such
as the MAPK cascade and the Integrin signal transduction network, are
presented as zoomed out blocks in which the regulatory genes involved
are omitted;
3. they do not take into account the contribution from cancer-specific path-
ways such as apoptosis, angiogenesis and invasion, or glioblastoma-specific
phenotypes such as antigen processing and presentation and mesenchy-
mal transformation;
4. they are not available in formats other than poorly editable pdf/jpg/giff
images, so they cannot be used to overlay custom expression data and
highlight, in a very quick way, the relevant gene networks that are turned
on/off.
In order to address these shortcomings and be able to visualise gene expression
differences in a pathway context, I compiled our own glioblastoma integrated
pathway map that tries to gather all the relevant information missing so far
from the literature. Therefore, our map includes the pathways most commonly
affected in glioblastoma: (i) RTK/PI3K/PTEN signaling, (ii) p53 signaling
and (iii) Rb-mediated control of cell cycle progression [154,326], as well as core
cell-cycle regulatory pathways: (i) MAPK cascade and (ii) RTK signaling, and
finally cancer-specific and glioblastoma-specific pathways: (i) antigen process-
ing and presentation, (ii) apoptosis, (iii) angiogenesis, invasion, motility and
(iii) mesenchymal transformation [85].
187

7.5 Glioblastoma Pathway Analysis Results
Figure 7.7: The integrated glioblastoma pathway is subdivided into rough sections
contained within the orange and blue boxes identifying the gene networks that participate
in the pathway. In the orange boxes the "classic" glioblastoma pathways
are highlighted: TP53, RB1 and PTEN signaling; in the blue boxes the cancer and
glioblastoma-specific pathways are represented.
188

7.5 Glioblastoma Pathway Analysis Results
Figure 7.7 shows the integrated pathway in its default colours, without any
overlaid expression data that would colour the gene nodes. Sections pertaining
to different pathways are coloured in blue if the pathways are glioblastoma-
specific or cancer-specific and orange if they reflect the cross-talk between the
three well-known pathways commonly affected in glioblastoma. The colour,
line and endpoints of the edges represent the type of interactions between the
nodes: activation (solid, green, arrow point), transcriptional activation (solid,
green, diamond point), tentative activation (dashed, green, arrow point), inhi-
bition (solid, red, T point), transcriptional inhibition (solid, red, delta point),
includes (solid, black, circle point), becomes (solid, black, top half arrow point),
simple interaction (solid, grey, no endpoint), leading to (dashed, grey, arrow),
lets in (dashed, grey, no endpoint). The shape of the node represents the type:
gene (round), complex (hexagon), family (hexagon), molecule (fee), process
(round rectangle). Finally, the beige colour distinguishes genes (grey if not
overlaid with colour-coded expression data) from non genes (beige). All the
interactions described by the pathway can be found in Appendix D.1.
In order to ensure that every single interaction described in the pathway had
been experimentally validated I checked every one of them in the interactions
databases described in detail in the Methods section, such as BioGrid [62] and
Intact [205]. Thus, this glioblastoma pathway is a representation of physically
interacting proteins at work in the specific disease context. When the interac-
tion needs to occur with chromatin to describe the regulation properly, such as
in the case of a transcription factor, either a node named "DNA" is described
as one of the two interactors, or a special edge named "activates transcription
of" or "inhibits transcription of" - that can be identified by the diamond or
delta endpoint, respectively - connects the two interactors. The entire pathway
can be recreated from the list of interactions and interactors described in Ap-
pendix D.1, but an editable version - with extension CYS - can be downloaded
from www.ebi.ac.uk/~diva/GBM-pathway.cys, where the latest and previous
versions of the pathway are available. The CYS extension allows the pathway
to be viewed and/or edited with any network editing software, such as Cy-
toscape [451]. The availability of the pathway for download and use by other
researchers, as well as its constant curation, attempts to address one of the
four main shortcomings of already existing GBM pathways: the unavailability
of formats other than images. Hopefully this resource will now be used by all
the researchers who want to overlay expression data on a new comprehensive
integrated GBM pathway to observe changes in the wider disease context.
189

7.5 Glioblastoma Pathway Analysis Results
Naturally, we wanted to make use of the resource ourselves and used it to
visualise gene expression changes in the context of the molecular networks
represented. The complete pathway has a total of 245 nodes, of which 57 are
not genes and the remaining 188 are single genes. The nodes representing
genes can be recognised by their distinct circular shapes that are unique to
this category. Of the 57 non gene nodes: 22 are families of proteins, rep-
resented by a diamond shape; eight are complexes of proteins, represented
by a hexagon shape; six are non-proteic molecules, represented by a vee-like
shape; the remaining 20 define end-processes such as "apoptosis", "cell signal-
ing" and "protein synthesis" and are represented by round rectangular shapes.
The only isoforms represented in the pathway are the two isoforms of gene
CDKN2A labeled as CDKN2A and CDKN2A:ARF (Table 7.7). The dataset
Table 7.7: Node assignment in the glioblastoma pathway. The colour of the nodes
representing genes and isoforms is only grey at the default pathway level. When
expression data are overlaid on the network the gene nodes will become coloured
according to a customisable colour palette.
Node type Number of nodes Node shape Node colour
Gene families 22 Diamond Beige
Cellular Processes 20 Round Rectangle Beige
Non-proteic molecules 6 Vee Beige
Complexes 8 Hexagon Beige
Genes 186 Circle Grey, default
Isoforms 2 Circle Grey, default
Total 245
of differentially expressed genes from the Tag-seq expression dataset contains
739 genes at an FDR of 10%. By overlaying the expression data for the 739
differentially expressed genes over the 188 gene nodes in the pathway, we found
that 70 genes were differentially expressed between GNS and NS cell lines at
10% FDR. Of these 70 genes, 51 were up-regulated in GNS cell lines (red colour
gradient), and 19 were down-regulated in GNS cell lines (blue colour gradient)
(Fig 7.8). The intensity of the colour reflects the value of the log2(FC) be-
tween GNS and NS cell lines and the values defining the red and blue colour
gradients were specifically taken from the same colour gradients defined in
the qRT-PCR heatmap of figure 6.13 and the tumour expression correlation
heatmap of figure 7.4.
190

7.5 Glioblastoma Pathway Analysis Results
Figure 7.8: Integrated GBM pathway with Tag-seq GNS dataset overlaid. The
map is composed of 245 nodes, of which 188 nodes represent individual genes, shown
as circles of intensities that correlate with FC expression changes.
191

7.5 Glioblastoma Pathway Analysis Results
Figure 7.9 displays a simplified and condensed version of the integrated GBM
pathway of figure 7.8 that is focused on the pathways most frequently affected
in glioblastoma, meant to highlight that we did capture some of the most com-
mon alterations in glioblastoma with the GNS/NS comparison. Such changes
include up-regulation of EGFR and down-regulation of the tumour suppres-
sor PTEN [326]. Similarly to the complete pathway in figure 7.8, the colour
gradient of the condensed pathway in figure 7.9 is taken from the same colour
gradients defined in the qRT-PCR heatmap of figure 6.13 and the tumour
expression correlation heatmap of figure 7.4.
Figure 7.9: Affected p53, RB1 and PTEN/PI3K pathways. Genes are represented
by circles coloured according to the expression FC measured between GNS and NS
cell lines by Tag-seq.
Given the high heterogeneity typical of glioblastoma tumours, we envisioned
another application for our GBM integrated pathway. Since every cell line
originated from a separate patient (for the exception of G144ED and G144)
192

7.5 Glioblastoma Pathway Analysis Results
and the heterogeneity of the tumour has recently been emphasised further by
signature microarray studies like those of Verhaak et al [511] and Phillips et
al [390], we mapped the Tag-seq measured expression levels for every GNS cell
line on our GBM integrated pathway to try and visualise differences poten-
tially unique to any GNS cell line and, therefore, to the tumour it originated
from (single pathways in Appendix D.2). Figure 7.10 highlights the genes in
our GBM pathway that are expressed in each individual GNS cell line, as mea-
sured via Tag-seq. The green nodes reflect the genes that have been measured
and the intensity of the green colour correlates with the level of expression ob-
served, where lighter greens signify smaller expression values and darker greens
signify higher expression values (normalised digital tag counts range from 0 to
16,500) (Fig 7.10). For a clearer view of each of these pathways refer to Ap-
pendix D.2 Pathway Images.
By having the GNS cell line-specific GBM pathways next to each other in fig-
ure 7.10, it is possible to observe differences as well as commonalities in gene
expression patterns. For example, in each GNS cell line, tumour suppressor
PTEN is not expressed (blue shaded boxes) and the antigen presentation-
related genes light up with different intensities, highlighting a more highly
activated antigen presentation process at the transcriptional level in cell line
G144 than in cell lines G166 and G179 (red shaded boxes). Also, since the
CDKN2A-CDKN2B tumour suppressor locus is deleted in cell line G179 but
retained in cell lines G144 and G166 - possibly in response to an oncogenic
signal - the expression of its CDKN2A, CDKN2A:ARF and CDKN2B genes is
very low (darker green nodes) in the G179 pathway and much higher (lighter
green nodes) in the G144 and G166 pathways (green shaded boxes). Finally,
we can see that the orphan nuclear receptor NR0B1, which is known to be
up-regulated and drive tumour growth in Ewing’s sarcoma [158], is also highly
expressed (lighter green nodes) in our G179 pathway as opposed to the G144
and G166 pathways (yellow shaded boxes). The PARP3 and PARP12 genes,
located below NR0B1 in the pathways and included within the yellow shaded
box of figure 7.10, are highly expressed across all GNS pathways, consistent
with the potential therapeutic role of these genes in GNS cells as inhibitors of
their homolog PARP1, for which characteristic they are now being studied in
brain tumour clinical trials [270]. In order to consider other sources of ex-
pression data that could highlight interesting patterns when compared to our
Tag-seq dataset, we overlaid the exon and microarray expression data from the
TCGA and HGG datasets (described in Methods section 5.9; all genes overlaid
193

7.5 Glioblastoma Pathway Analysis Results
194
Figure 7.10: Four integrated GBM pathways overlaid with Tag-seq expression level measures for each GNS cell line. Lighter green nodes indicate
lower expression than darker green nodes. The blue box highlights the location of PTEN; the red box of the antigen presentation network of genes;
the green box of the location of the CDKN2A-CDKN2B locus genes; the yellow box of the location of genes NR0B1, PARP3 and PARP12.

7.5 Glioblastoma Pathway Analysis Results
195

7.5 Glioblastoma Pathway Analysis Results
were found to be differentially expressed at FDR

7.5 Glioblastoma Pathway Analysis Results
CDKN2A/CDKN2B locus deletion - observable in the TCGA pathway by the
red-orange colours of nodes representing CDKNA genes (Fig 7.11). Amplifi-
cation of the CDK4 locus was also a common finding in the TCGA dataset,
CDK4 forming complexes with CDKN2A and CDKN2B to maintain cell cycle
progression, which can be observed in the TCGA pathway as an orange node
(Fig 7.11).
Other interesting patterns that can be observed by the comparison of the three
pathways in figure 7.13, are those highlighted with the other coloured boxes.
The HIF1A gene, responsible for the regulation of poor-oxygen level responses
in the cell, is up-regulated in the TCGA and HGG pathway but undetected in
the GNS pathway (light blue shaded boxes). The SERPINB2 and SERPINE
genes, responsible for the inhibition of angiogenesis and metastasis, seem to be
strongly up-regulated in the GNS pathway but slightly down-regulated in the
HGG pathway and undetected in the TCGA pathway (brown shaded boxes).
The green shaded boxes highlight the regulatory loop for mesenchymal trans-
formation that has been found to be active in GBM tumours. In this loop,
CEBPB activates STAT3 and FOSL2 that together activate RUNX1 - also
activated by DEC1 that is activated itself by FOSL2, which in turn activates
JUN leading to mesenchymal transformation [85]. This regulatory loop is
clearly active and up-regulated in the TCGA and HGG pathways, although
only CEBPB appears as detected and up-regulated in the GNS pathway. The
pink shaded boxes highlight the regulation of motility-relevant genes, such
as those belonging to the SNARE complex, CALM1 and calcium channels
of the CACN family, that appear to be strongly up-regulated in the GNS
pathway but down-regulated in the TCGA and HGG pathways. Finally, the
purple shaded boxes highlight the behaviour of the antigen processing and
presentation-related genes such as HOXA, that are consistently up-regulated
in all three pathways, although at different levels with the GNS pathway de-
taining the highest levels of up-regulation. This confirms the importance of the
immune-evasion phenotype that is transcriptionally active in all three datasets.
Overall, this pathway approach allowed us to identify differentially expressed
genes that participate in glioma-related pathways and highlight patterns be-
tween different tumour cell types. However, it should be made clear that the
GBM integrated pathway is not a flawless representation of the gene networks
and regulations that occur inside GBM tumour cells, but rather the widest
yet representation of the known interactions. Since the pathway was manually
built there is always a chance to expand specific sub-networks of interest by
197

7.5 Glioblastoma Pathway Analysis Results
writing to the curator at dt322@cam.ac.uk or downloading the latest version
of the pathway at www.ebi.ac.uk/~diva/GBM-pathway.cys and editing the
pathway yourself. The manual curation of the pathway is nonetheless ongoing
and new additions are released on a regular basis at the same address.
198

7.5 Glioblastoma Pathway Analysis Results
Figure 7.11: Integrated GBM pathway overlaid with TCGA dataset. The map
is composed of 245 nodes and contains 188 nodes representing individual genes that
correlate with FC expression changes (http://cancergenome.nih.gov; [326]).
199

7.5 Glioblastoma Pathway Analysis Results
Figure 7.12: Integrated GBM pathway overlaid with HGG dataset. The map
composed of 245 nodes and contains 188 nodes representing individual genes that
correlate with FC expression changes [148,390].
200

7.5 Glioblastoma Pathway Analysis Results
Figure 7.13: Three integrated GBM pathways overlaid with the GNS, TCGA
and HGG datasets (FDR

Chapter 8
MicroRNA Target Prediction
Ensemble Software
Contents
8.1 Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
8.2 Workflows . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
8.3 Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8.4 Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.5 Target Prediction Ensemble Analysis . . . . . . . . . . . . 211
8.1 Principles
Hundreds of microRNA sequences are now annotated throughout the human
genome and are predicted to modulate the expression levels of thousands of
mRNAs. Families of microRNA sequences can be identified through evolution-
ary conservation of sequence patterns in related species, and reciprocally, genes
targeted by microRNAs are under selection pressure to retain the recognition
sites needed for the annealing of microRNA to mRNA duplexes. Numerous
regulatory target prediction algorithms have been developed to exploit these
properties, but they vary widely in terms of criteria, accuracy and prediction
coverage. Most prediction methods search for complementary sequences be-
tween microRNAs and putative gene targets, while some consider physical and
statistical hybridization properties, cross-conservation of regulatory RNAs be-
tween related species, etc. As a result, there is very little overlap between
the microRNA:mRNA annealing predicted by different algorithms. It is of-
ten the case that researchers have asked themselves "which target prediction
algorithm will predict the highest number of genes in my list?". Needless to
202

8.1 Principles Results
say, this is a flawed approach triggered by the absence of a unifying common
theme in the prediction algorithms that would make them more accurate in
their predictions. Rationally, if several algorithms can be developed that yield
vastly different results, then we are missing an important variable in our equa-
tion. In this perspective, the existing prediction algorithms are not capable of
simulating accurately what is happening in Nature time after time in the same
reproducible way. Having said this, we were interested in finding out if a single
prediction algorithm or a superset of several prediction algorithms was best at
predicting microRNA to mRNA annealing. During this analysis we stumbled
upon an observation that might be a hint towards the identity of the missing
variable in the prediction algorithm equation.
In order to determine which set of prediction algorithms were best at predict-
ing microRNA:mRNA annealing, we needed experimentally validated data to
help us screen the positives from the false positives, which contribute greatly
to the pool of predictions. We used exon array data and microRNA array
data that was available to us from our GNS cell lines (Appendix E and F) to
validate our findings. Since a large number of prediction algorithms are com-
monly used in research and they each predict a vast number of interactions,
we built a tool called "GenemiR" to allow molecular biologists to generally
access and manage microRNA predictions on a large scale and across differ-
ent algorithms that would also help us address our research question. The
GenemiR software exists as a command line binary executable for Unix-based
systems or as a compiled local installation for Windows. Due to its length,
the code of the binary executable could not be added to the appendix of this
thesis, but it is available for anyone to view or download at www.ebi.ac.
uk/~diva/GenemiR/genemir-code.txt. The executable for Windows can be
downloaded at www.ebi.ac.uk/~diva/GenemiR/GenemiR.zip.
In order to address the need for unified search and presentation of microRNA
to mRNA target interactions predicted across multiple algorithms, GenemiR
relates human and mouse microRNAs with their predicted target genes as
reported by eight leading predictors. At the core of the program are two
primitives that allow it to be extremely fast in extrapolating lists of gene tar-
gets given a microRNA and, performing the reciprocal analysis, extrapolating
lists of microRNAs predicted to repress one or more genes of interest (Fig 8.1).
Many other auxiliary functions and filters allow the expert user to parametrize
203

8.2 Workflows Results
and optimise the analysis to make it as stringent or as loose as necessary. The
target predictions in GenemiR’s database can be queried through a variety of
selectable filters, allowing the user to include or exclude any combination of
algorithms, group multiple microRNAs into the same query, and retrieve vari-
able subsets of targeted genes (Fig 8.1). Finally, GenemiR allows the matching
of prediction data to any gene expression dataset of interest, and provides a fa-
cility for loading external datasets determined by the user. Graphical plotting
functions display target expression levels relative to the conditions of the ex-
periment, such as tissue-specific transcriptional profiling or time course series.
The software itself does not contain internal experimental data, but simply
aggregates the output of prediction algorithms in a manner suitable to explo-
ration and hypothesis testing, reducing the complexity of this approach by
integrating microRNA and gene annotations, genome-wide expression data,
and regulatory target predictions in a common analysis framework (Fig 8.1).
8.2 Workflows
The GenemiR software allows for the discovery of systemic or tissue-specific
patterns that may be hidden in the microRNA targeting prediction data from
eight leading algorithms. In order to do this, two flows of information are estab-
lished that address reciprocal biological questions: "which genes are targeted
by one or more specific microRNAs?" and "which microRNAs are predicted
to target a given set of genes?". These issues are intimately related within
the context of the same biological system, but are organised as two opposite
flows of information in terms of program operation and layout (Fig 8.2). De-
pending on the type of query executed against the internal databases, a list
of genes predicted to be targeted by a number of microRNAs according to a
customer-defined selection of prediction algorithms, is generated (Workflow 1,
figure 8.3). In the opposite workflow, a list of microRNAs is generated start-
ing from a list of gene symbols or Genbank, RefSeq, EMBL, ENSG or ENST
identifiers, according to the customer-defined selection of prediction algorithms
(Workflow 2, figure 8.3). Conversion from the identifiers contained originally
in the prediction files is constantly active in both workflows to always translate
the queries to a human intelligible list of gene symbols.
204

8.3 Databases Results
Figure 8.1: Overview of GenemiR. (a) Queries based on various data types (the
input section: microRNAs, expression data, genes, and identifiers) are executed
against internal or external databases to generate other data types (the output section:
genes, microRNAs, graphs and converted identifiers). The shaded boxes distinguish
the core primitive functions from the auxiliary ones. (b) Relationships between
databases used by GenemiR and the functions that operate on them. Solid lines depict
databases of fixed size over the lifetime of the program’s execution, whereas the
dashed lines indicate a dataset of variable dimensions according to the files that are
inputted by the user.
8.3 Databases
All microRNAs and genes that could constitute a query are stored in the form
of three internal databases:
1. Identifiers for all known human and mouse microRNAs;
2. Genbank, RefSeq [409], EMBL [252], ENSG, ENST [147] identifiers for
all annotated human and mouse genes;
205

8.3 Databases Results
Figure 8.2: Internal organisation of the target prediction database of GenemiR.
From top to bottom, the two conceptual workflows are consecutively displayed.
Each of eight algorithms (circles) is represented by a different acronym:
Diana-microT=DN-T; miRanda=MR; miRBase=MC;ElMMo=EM; PITA=PA; Pic-
Tar=PT5/6; TargetscanS=TSS. MicroRNAs (diamonds) and target genes (squares)
are assigned random labels for illustrative purposes.
3. Gene symbols as determined by the HUGO Gene Nomenclature Commit-
tee for [203] and the MGI Mouse Genome Informatics database resource
for mouse [333].
All expression datasets that the user wants to upload in the software are re-
ferred to as "external databases" because they are not part of the internal
databases, they are user-defined and variable in size. Gene annotation data
are derived from the Ensembl database [147], and a utility script is provided
with the program to allow automatic updating of files, via the Ensembl Perl
API, to reflect the latest annotation release of the human and mouse genomes.
The internal databases of human and mouse microRNAs and gene identifiers
for all annotated human and mouse genes are linked in a meaningful way by
the algorithms that predict which of the microRNAs regulate which of the
gens and, vice versa, which of the genes are regulated by which microRNAs.
Thus, these linked databases consists of the union of prediction results gen-
erated from the eight most widely used algorithms (considering PicTar 5-way
and PicTar 6-way as separate algorithms) (Table 8.1), with the possibility of
increasing this number as new algorithms are produced with new sets of pre-
dictions, thanks to the modular internal structure with which GenemiR was
designed:
1. PicTar [127,247] identifies putative microRNA targets in vertebrates, C.
elegans and Drosophila, using the principle of cross-conservation. The
two versions, PicTar 5-way and PicTar 6-way, refer to the number of
species in the comparison: 5-way includes human, chimp mouse, rat
and dog genomes, and 6-way adds the chicken genome. PicTar employs
206

8.3 Databases Results
Figure 8.3: Workflows at the core of the primitive functions of the GenemiR software.
The first workflow answers the question: "which microRNAs target this list
of genes?" and returns a list of microRNAs filtered according to the Minimum or
Cumulative filter or both. The second workflow answers the question: "which genes
are targeted by this list of microRNAs?" and returns a list of genes filtered according
to the settings of the Minimum or Cumulative filter or both.
pair-wise alignments to filter seed sequences conserved in 7 species of
Drosophila and 8 species of vertebrates. In addition to sequence conser-
vation, PicTar considers secondary evidence, such as co-expression and
clustering of microRNAs, as well as target identification based on onto-
logical parameters, such as expression in the common cell types or devel-
207

8.3 Databases Results
opmental stages. Importantly, it also takes into account the number of
multiple seeds occurring within the same 3'UTR. The false-positive rate
for this algorithm is estimated to be approximately 30% [48].
2. TargetScanS is the second version of the mammalian microRNA target
prediction algorithm developed by Lewis and colleagues following the
publication of their original TargetScan approach [272,273]. The earlier
program searched for a 7nt-long seed region starting from the second
nucleotide of the 5' end miRNA sequence, calculated the free energy of
hybridization using RNAFold [57], and computed a score based on the
presence of multiple seeds capable of annealing to the same microRNA.
The criteria used by TargetScanS are more specific, requiring a 6nt-long
seed sequence preceded by an adenosine, positioned in a conserved region
that, in turn, is surrounded by areas of lesser conservation. TargetScanS
omits the use of free-energy calculations to predict hybridization affinity,
and is estimated to have a false-positive rate of between 22 and 31% [48].
3. DIANA-microT [237,316] version 3.0 recognises only single seed se-
quences having a central stem-loop secondary structure, instead of select-
ing for near-perfect complementarity between the 5' seed region of the
microRNA and the 3'UTR of the target message. In addition to search-
ing for the classical annealing pattern in the 5' seed region, DIANA-
microT also requires complementarity at the 3' end of the microRNA
sequence [48].
4. miRanda [134,213] identifies putative targets throughout the human and
Drosophila genomes. It implements a dynamic programming approach
to compute local alignments of complementary microRNA and mRNA
sequences. The algorithm assigns scores to putative seed sequences using
a position matrix, giving a higher weight to those closer to the 5' end
of the microRNA sequence. However, it is not constrained by the re-
quirement for exact seed matches. The program takes into consideration
free-energy annealing calculations (via RNAfold), and seed-sequence con-
servation. To account for tandem annealing of microRNAs to multiple
consensus sequences in the targeted transcript, miRanda assigns a high
score to alignment matches to the same as well as different microRNAs.
5. ElMMo [155] uses a general Bayesian method that scores the conser-
vation of microRNA binding sites according to an evolutionary model.
208

8.4 Filters Results
Bayesian methods are based on prior probability of an observation and
are updated as new data keeps entering. This model assumes a phyloge-
netic relationship among several species.
6. miRBase [179] uses the miRanda algorithm to identify potential binding
sites of a microRNA. Dynamic programming alignment is used to identify
highly complementary sites that also require strict complementarity at
the 5' seed region and thermodynamic stability, which is estimated for
each target site. For inclusion in the database, conservation of the target
site at the exact same position in at least two species is needed.
7. PITA [224] takes into consideration the strength of microRNA repression
given target site accessibility and for each target site, an energy-based
measure representing the difference between the free energy gained by
the binding of the microRNA to the target, and the free energy lost
by un-pairing the nucleotides within the target site, is calculated. The
energy used to un-pair additional nucleotides flanking the target sites is
also taken into account.
Table 8.1: microRNA target prediction algorithms used by GenemiR with number
of microRNA:3'UTR interactions predicted. The original target identifiers refer to
the identifiers used by a prediction algorithm to identify the targeted genes. The
final target identifiers refer to the identifiers that are returned by any query of any
prediction algorithm database.
Prediction algorithm Number
microRNAs
Number of
targets
Original target
identifiers
Final target
identifiers
PITA 678 22,974 RefSeq HGNC/MGI
PicTar 5-way 178 9,334 RefSeq HGNC/MGI
PicTar 6-way 130 3,585 RefSeq HGNC/MGI
DIANA-microT 555 18,986 ENSG/ENSMUSG HGNC/MGI
Elmmo 1206 31,303 RefSeq,EMBL HGNC/MGI
Miranda 1100 32,641 EMBL HGNC/MGI
Microcosm 694 34,507 ENST/ENSMUST HGNC/MGI
TargetscanS 967 17,725 HGNC/MGI HGNC/MGI
8.4 Filters
Critical to the successful outcome of user queries is the appropriate use of the
built-in filtering functions. The purpose of the filter auxiliary functions (Fig
209

8.4 Filters Results
8.3) is to allow the user to specify an appropriate range of return results from
each query, depending on either the numbers of different microRNAs expected
to play a role in gene regulation (adjusted according to the Minimum and
Cumulative filters), or the level of concordance the user enforces between the
combination of prediction algorithms defined. The user can at all times pref-
erentially select a desired subset of algorithms upon which prediction results
will be based and the two types of filters available - Minimum and Cumulative
- will be applied on the prediction results of those algorithms.
The consequence of setting the Minimum filter to a given value depends on
the type of query he user makes. If the query starts with a list of microRNAs
and therefore asks which genes are target by these microRNAs, according to a
number n of prediction algorithms, then setting the Minimum filter to a given
value returns a set of genes targeted by at least that number of microRNAs,
as reported by all of the selected algorithms. For example, if the inputted list
contains m = 10 microRNAs and n = 3 prediction algorithms are selected,
then the Minimum filter will range from one to 10 and setting it on four will
cause it to report only those genes that are targeted by at least eight microR-
NAs as predicted by each of the three algorithms alone. If the query were to
start from a list containing g = 30 genes and n = 3 prediction algorithms were
selected, then the Minimum filter would range from one to 30 and setting it
on 24 would cause it to report only those microRNAs that target 24 genes as
predicted by each of the prediction algorithms alone (Fig 8.3). The number of
genes returned from the query will therefore depend greatly on the prediction
algorithms selected.
The Cumulative filter acts in an alternative fashion: a gene is reported if
the total number of microRNA predictions, according to all of the chosen algo-
rithms, is at least equal to or greater than the minimum value set by the user.
This alleviates the requirement that every algorithm independently predict a
set number of microRNA interactions; rather, the aggregate sum of all results
in determining which target genes to report, will be considered. Using the
Cumulative filter is a less stringent approach and, all things equal, returns a
higher number of outputted results. For example, if the inputted list contains
m = 10 microRNAs and n = 3 prediction algorithms are selected, then the Cu-
mulative filter will range from one to 10x3 = 30 and setting it on 20 will cause
it to report the genes that are targeted by at least 20 microRNAs according
to any combination of prediction algorithms selected, without the requirement
that any particular algorithm predict a set threshold of microRNAs. If the
210

8.5 Target Prediction Ensemble Analysis Results
query were to start from a list containing g = 30 genes and n = 3 prediction
algorithms were selected, then the Cumulative filter would range from one to
30x3 = 90 and setting it on 56 would cause it to report only those microRNAs
that target 56 genes as predicted by any combination of the prediction algo-
rithms (Fig 8.3).
While the Minimum and Cumulative filters operate on microRNAs, the possi-
bility of selecting a subset of the eight prediction algorithms affords the user
control over the number of algorithms required to report a given prediction,
and thus define the criteria by which predictions are deemed accurate. For
example, certain applications - such as experimental validation and cloning of
microRNAs involved in particular pathways of interest - involve a significant
amount of effort to perform. If this is the ultimate goal of the investigation,
then it is desirable to focus on a small set of highly-scoring microRNA targets
unanimously reported by several algorithms. Other searches though, such as
those performed for comparative genomic analyses, are limited only by com-
putational feasibility and can therefore afford to include a greater number of
putative regulators whose involvement may only be predicted by one or two
algorithms.
Any combination of the above filtering methods is possible, keeping in mind
that: (i) they are always applied to the original set of data, and that (ii) a
given gene is included in the final results only if it passes the criteria imposed
by each individual filter set by the user. Therefore, the utility of applying dif-
ferent filter combinations depends on the relevance of the biological question
they help to answer.
8.5 Target Prediction Ensemble Analysis
A question we wanted to ask was whether any prediction algorithm fared bet-
ter than any other combination of prediction algorithms. Since the divergence
between prediction results from any algorithm is large, finding whether any
combination of a subset of these algorithms fares better than any algorithm
alone, is a necessary step. Using our software tool GenemiR we tried to eluci-
date how different combinations of prediction algorithms fared with respect to
the single algorithm, and thus address the hypothesis suggested by Alexiou et
al [16] that combinations of algorithms predict more accurately than the single
algorithm alone. The way we addressed this problem was using relevant ex-
perimental data from exon arrays and microRNA microarrays generated from
211

8.5 Target Prediction Ensemble Analysis Results
the same GNS cell lines of which we had digital gene expression profiles. With
this set of data we could link the down-regulation of expression observed for
the 2,290 genes at F DR < 1% as measured in the exon arrays, to the up-
regulation of expression observed for the 258 microRNAs at F DR < 1% as
measured in the microRNA microarrays.
Our first interesting finding was that the final score of our analysis, represent-
ing how well the subset of algorithms at stake fared when asked to predict
which microRNAs targeted the list of 2,290 down-regulated genes, worsened
considerably if an initial filter was not applied that made our analysis tissue-
specific. When I mentioned earlier that the ensemble analysis gave us a hint as
to the identity of the variable missing from prediction algorithms, it was tissue
specificity. So far prediction algorithms have not been designed with tissue-
specificity in mind. However, microRNA regulation is highly tissue specific
and there is no way around having to factor this in when designing an algo-
rithm that wil accurately predict the interaction between a microRNA and an
mRNA. Current prediction algorithms consider different cohorts of variables
that all refer to a main concept: sequence complementarity. Secondary con-
siderations are factors such as secondary structure and number of microRNA
seeds in a 3'UTR. There is no easy way to implement tissue-specificity variables
within a target prediction algorithm because of the empirical nature of such
variables. We have yet to understand what exactly triggers tissue-specificity in
microRNA regulation of mRNAs, but any tuning implemented by epigenetic
mechanisms or feedback regulatory loops may be too computationally inten-
sive to simulate still.
One way to work around this problem in the present is to filter the target
predictions by a background gene list that filters out most non tissue-specific
predictions. Of course, this strategy will only yield partially accurate results
but it can surely improve the use of prediction results at the present time. As a
result of this understanding, we used a background gene list ourselves in order
to minimise the number of false positives caused by the brain tissue-specificity
of the query. The background gene list was retrieved from the exon array
dataset and consisted of all 6,254 genes with an expression level that was mea-
sured at F DR < 1%. Running the algorithm without this background gene
list worsened the final score by ten fold.
The method we used to evaluate the performance and accuracy of each pre-
diction algorithm in the GenemiR internal database is described below and
212

8.5 Target Prediction Ensemble Analysis Results
summarised in figure 8.4. We first run the method on the predictions from
each algorithm alone and then extended it to all combinations of prediction
algorithm results. The score E reflecting the accuracy of each prediction or
combination of predictions, ranges between 0 and 1 (0 > E > 1), with 1 re-
flecting a perfectly matching set of predictions between the algorithm being
evaluated and our experimental data. The score is calculated following these
successive steps:
1. filter out, for each prediction algorithm and for each microRNA exper-
imentally observed to be up-regulated, the genes that are predicted to
be target by that microRNA but are not present in the gene background
list;
2. generate a union list from the genes filtered in step 1 for the entire cohort
of up-regulated microRNAs, for each prediction algorithm;
3. iterate over each gene in the union list and increase a cumulative counter
every time the gene is predicted to be targeted by one of the up-regulated
microRNAs, for each prediction algorithm. This counter is normalised
to the number of microRNAs that the prediction contains from the list
of 258 experimentally measured microRNAs;
4. sort the union gene list by the count calculated in step 3 to generate a
"hit" list, representative of each prediction algorithm;
5. iterate over the hit list summing in a cumulative counter C1, the counts
associated to all the genes in the list, which have been predicted by the
prediction algorithm ;
6. iterate over the hit list summing in a cumulative counter C2, the counts
associated to the genes that are experimentally observed to be down-
regulated in our exon array data;
7. generate a score value by dividing the cumulative counters C1 and C2,
representing a measure of how well a prediction algorithm fared based
on how many genes it correctly predicted to be targeted in our GNS cell
lines, considering the same cell line expression profile as a background
for calculations.
Scores for each single prediction algorithm are listed in table 8.2 below. To
evaluate whether combinations of different algorithms were more accurate at
213

8.5 Target Prediction Ensemble Analysis Results
Figure 8.4: Step by step diagram of the ensemble method adopted to find the
score E (=C2/C1) of prediction accuracy for prediction algorithms. The method
was applied to the predictions from all combinations of target prediction algorithms
and an E-score was generated for each one. The red set of gene predictions varied
in size depending on how many and which algorithms were being considered for a
particular round.
predicting than the single, we performed the same steps 1-7 but with lists that
resulted as the union of all the user-defined prediction algorithm genes (vary-
ing size of the predicted genes set of figure 8.4). The results are listed in table
8.5. The purpose of the hit list is to "weight" the importance to the predicted
genes. For example, if gene A has been predicted to be targeted by only nine
of the 258 microRNAs, its weight when added to the cumulative score C1 will
be only 9/258th the weight added by gene B, if gene B were predicted to be
214

8.5 Target Prediction Ensemble Analysis Results
Table 8.2: Single prediction algorithm ensemble analysis results. Displayed in
descending order of E-score.
Prediction name E-score
ElMMo 0,3181
Diana-microT 0,3122
PITA 0,3088
TargetscanS 0,3073
PicTar 6 0,3020
miRBase 0,3008
PicTar 5 0,2989
miRanda 0,2979
targeted by all 258 microRNAs. This scoring system takes into consideration
the possibility that not all prediction algorithms necessarily predict the target-
ing for all the 258 up-regulated microRNAs, by always measuring the accuracy
of a prediction set over the number of genes of the background gene list that
are predicted by a particular algorithm.
This analysis highlights the fact that very little improvement, in the order of
the thousandth, is achieved by combining prediction algorithms together. The
highest scoring prediction algorithms as-singles are ElMMo and Diana-microT
and when used in combination with other prediction algorithms the E-score
of ElMMo always decreases, while the E-score of Diana-microT, in combina-
tion with one or two other prediction algorithms, seems to slightly increase.
This analysis reveals that the best performing combination of algorithms does
not outperform the best scoring as-single algorithm ElMMo. However, this
initial analysis should be followed by a series of other analysis in which other
approaches for the evaluation of the score are taken, and the Minimum and
Cumulative filters are taken into consideration as well. This approach would
increase the number of combinations from the current 256 (2 8 ) because each
combination of filters would have to be evaluated for all the 256 microRNAs
and would therefore require a global energy combinatorial optimisation algo-
rithm such as a genetic algorithm, whereby a search that mimics the process
of natural evolution would be performed on a population of candidate solu-
tions that evolve towards the best solution starting from a randomly selected
population.
215

8.5 Target Prediction Ensemble Analysis Results
Table 8.3: All combinations of prediction algorithms in descending order of E-score. The Escores
for the single algorithms are also present and highlighted along the list in bold characters.
Prediction algorithms involved in the combination set E-score
ElMMo 0.3181
ElMMo PicTar6 0.3173
ElMMo PicTar5 0.3162
ElMMo miRBase 0.3160
Diana-microT ElMMo 0.3159
Diana-microT ElMMo PicTar6 0.3155
ElMMo PicTar5 PicTar6 0.3155
ElMMo miRBase PicTar6 0.3154
Diana-microT ElMMo PicTar5 0.3148
Diana-microT ElMMo miRBase 0.3147
ElMMo miRBase PicTar5 0.3145
Diana-microT ElMMo PicTar5 PicTar6 0.3144
ElMMo TargetscanS 0.3144
Diana-microT ElMMo miRBase PicTar6 0.3143
ElMMo miRBase PicTar5 PicTar6 0.3139
ElMMo PicTar6 TargetscanS 0.3139
Diana-microT ElMMo TargetscanS 0.3138
Diana-microT ElMMo miRBase PicTar5 0.3137
Diana-microT ElMMo PicTar6 TargetscanS 0.3135
Diana-microT ElMMo miRBase PicTar5 PicTar6 0.3134
ElMMo PicTar5 TargetscanS 0.3133
ElMMo miRBase TargetscanS 0.3132
Diana-microT ElMMo miRBase TargetscanS 0.3130
Diana-microT ElMMo PicTar5 TargetscanS 0.3130
ElMMo miRBase PicTar6 TargetscanS 0.3129
ElMMo PicTar5 PicTar6 TargetscanS 0.3129
Diana-microT ElMMo miRBase PicTar6 TargetscanS 0.3127
Diana-microT ElMMo PicTar5 PicTar6 TargetscanS 0.3127
ElMMo miRBase PicTar5 TargetscanS 0.3123
Diana-microT ElMMo miRBase PicTar5 TargetscanS 0.3123
Diana-microT 0.3122
ElMMo PITA 0.3122
Diana-microT ElMMo PITA 0.3122
Diana-microT ElMMo miRBase PicTar5 PicTar6 TargetscanS 0.3121
ElMMo PITA PicTar6 0.3120
Diana-microT ElMMo PITA PicTar6 0.3120
ElMMo miRBase PicTar5 PicTar6 TargetscanS 0.3120
Diana-microT ElMMo miRBase PITA 0.3118
Diana-microT ElMMo PITA PicTar5 0.3118
ElMMo miRBase PITA 0.3117
ElMMo PITA PicTar5 0.3117
Diana-microT ElMMo miRBase PITA PicTar6 0.3116
Diana-microT ElMMo PITA PicTar5 PicTar6 0.3116
ElMMo miRBase PITA PicTar6 0.3115
ElMMo PITA PicTar5 PicTar6 0.3115
Diana-microT ElMMo PITA TargetscanS 0.3115
ElMMo PITA TargetscanS 0.3114
Diana-microT ElMMo PITA PicTar6 TargetscanS 0.3114
Diana-microT ElMMo miRBase PITA PicTar5 0.3113
Diana-microT PicTar6 0.3113
ElMMo PITA PicTar6 TargetscanS 0.3113
ElMMo miRBase PITA PicTar5 0.3112
Diana-microT ElMMo miRBase PITA PicTar5 PicTar6 0.3112
Diana-microT ElMMo miRBase PITA TargetscanS 0.3112
Diana-microT ElMMo PITA PicTar5 TargetscanS 0.3112
ElMMo miRBase PITA PicTar5 PicTar6 0.3110
ElMMo miRBase PITA TargetscanS 0.3110
ElMMo PITA PicTar5 TargetscanS 0.3110
Diana-microT ElMMo miRBase PITA PicTar6 TargetscanS 0.3110
Diana-microT ElMMo PITA PicTar5 PicTar6 TargetscanS 0.3110
ElMMo PITA PicTar5 PicTar6 TargetscanS 0.3109
Diana-microT ElMMo miRBase PITA PicTar5 TargetscanS 0.3108
ElMMo miRBase PITA PicTar6 TargetscanS 0.3108
Diana-microT ElMMo miRBase PITA PicTar5 PicTar6 TargetscanS 0.3107
Diana-microT ElMMo miRanda 0.3106
ElMMo miRBase PITA PicTar5 TargetscanS 0.3106
ElMMo miRBase PITA PicTar5 PicTar6 TargetscanS 0.3105
Diana-microT ElMMo miRanda PicTar6 0.3104
Diana-microT miRBase 0.3100
ElMMo miRanda 0.3100
Diana-microT ElMMo miRBase miRanda 0.3100
Diana-microT PicTar5 0.3100
216

8.5 Target Prediction Ensemble Analysis Results
Prediction algorithms involved in the combination set E-score
Diana-microT ElMMo miRanda PicTar5 0.3100
Diana-microT ElMMo miRanda TargetscanS 0.3099
Diana-microT ElMMo miRanda PITA 0.3098
Diana-microT ElMMo miRBase miRanda PicTar6 0.3098
Diana-microT ElMMo miRanda PicTar5 PicTar6 0.3098
Diana-microT TargetscanS 0.3098
Diana-microT ElMMo miRanda PicTar6 TargetscanS 0.3098
Diana-microT PITA 0.3097
ElMMo miRanda PicTar6 0.3097
Diana-microT ElMMo miRanda PITA PicTar6 0.3097
Diana-microT ElMMo miRBase miRanda PITA 0.3095
Diana-microT ElMMo miRBase miRanda PicTar5 0.3095
Diana-microT ElMMo miRanda PITA PicTar5 0.3095
Diana-microT PITA PicTar6 0.3095
Diana-microT ElMMo miRBase miRanda TargetscanS 0.3095
Diana-microT ElMMo miRanda PITA TargetscanS 0.3095
Diana-microT ElMMo miRanda PicTar5 TargetscanS 0.3095
Diana-microT PicTar6 TargetscanS 0.3095
ElMMo miRanda PITA 0.3094
Diana-microT miRBase PicTar6 0.3094
Diana-microT ElMMo miRBase miRanda PITA PicTar6 0.3094
Diana-microT PicTar5 PicTar6 0.3094
Diana-microT ElMMo miRanda PITA PicTar5 PicTar6 0.3094
Diana-microT ElMMo miRBase miRanda PicTar6 TargetscanS 0.3094
Diana-microT ElMMo miRanda PITA PicTar6 TargetscanS 0.3094
Diana-microT ElMMo miRanda PicTar5 PicTar6 TargetscanS 0.3094
ElMMo miRBase miRanda 0.3093
ElMMo miRanda PicTar5 0.3093
ElMMo miRanda PITA PicTar6 0.3093
Diana-microT ElMMo miRBase miRanda PicTar5 PicTar6 0.3093
ElMMo miRanda TargetscanS 0.3093
Diana-microT miRBase PITA 0.3092
Diana-microT PITA PicTar5 0.3092
Diana-microT ElMMo miRBase miRanda PITA PicTar5 0.3092
Diana-microT PITA TargetscanS 0.3092
Diana-microT ElMMo miRBase miRanda PITA TargetscanS 0.3092
Diana-microT ElMMo miRanda PITA PicTar5 TargetscanS 0.3092
ElMMo miRanda PicTar6 TargetscanS 0.3092
Diana-microT ElMMo miRBase miRanda PITA PicTar6 TargetscanS 0.3092
Diana-microT ElMMo miRanda PITA PicTar5 PicTar6 TargetscanS 0.3092
ElMMo miRBase miRanda PITA 0.3091
ElMMo miRanda PITA PicTar5 0.3091
ElMMo miRBase miRanda PicTar6 0.3091
ElMMo miRanda PicTar5 PicTar6 0.3091
Diana-microT ElMMo miRBase miRanda PITA PicTar5 PicTar6 0.3091
ElMMo miRanda PITA TargetscanS 0.3091
Diana-microT ElMMo miRBase miRanda PicTar5 TargetscanS 0.3091
Diana-microT PITA PicTar6 TargetscanS 0.3091
Diana-microT miRBase PITA PicTar6 0.3090
ElMMo miRBase miRanda PITA PicTar6 0.3090
Diana-microT PITA PicTar5 PicTar6 0.3090
ElMMo miRanda PITA PicTar5 PicTar6 0.3090
Diana-microT ElMMo miRBase miRanda PITA PicTar5 TargetscanS 0.3090
ElMMo miRanda PITA PicTar6 TargetscanS 0.3090
Diana-microT ElMMo miRBase miRanda PicTar5 PicTar6 TargetscanS 0.3090
Diana-microT ElMMo miRBase miRanda PITA PicTar5 PicTar6 TargetscanS 0.3089
PITA 0.3088
ElMMo miRBase miRanda PITA PicTar5 0.3088
Diana-microT miRBase TargetscanS 0.3088
ElMMo miRBase miRanda TargetscanS 0.3088
Diana-microT miRBase PITA TargetscanS 0.3088
ElMMo miRBase miRanda PITA TargetscanS 0.3088
Diana-microT PicTar5 TargetscanS 0.3088
ElMMo miRanda PicTar5 TargetscanS 0.3088
Diana-microT PITA PicTar5 TargetscanS 0.3088
ElMMo miRanda PITA PicTar5 TargetscanS 0.3088
ElMMo miRBase miRanda PicTar5 0.3087
Diana-microT miRBase PITA PicTar5 0.3087
ElMMo miRBase miRanda PITA PicTar5 PicTar6 0.3087
ElMMo miRBase miRanda PicTar6 TargetscanS 0.3087
Diana-microT miRBase PITA PicTar6 TargetscanS 0.3087
ElMMo miRBase miRanda PITA PicTar6 TargetscanS 0.3087
ElMMo miRanda PicTar5 PicTar6 TargetscanS 0.3087
Diana-microT PITA PicTar5 PicTar6 TargetscanS 0.3087
217

8.5 Target Prediction Ensemble Analysis Results
Prediction algorithms involved in the combination set E-score
ElMMo miRanda PITA PicTar5 PicTar6 TargetscanS 0.3087
PITA PicTar6 0.3086
Diana-microT miRBase PITA PicTar5 PicTar6 0.3086
ElMMo miRBase miRanda PITA PicTar5 TargetscanS 0.3086
Diana-microT miRBase PicTar5 0.3085
ElMMo miRBase miRanda PicTar5 PicTar6 0.3085
PITA TargetscanS 0.3085
Diana-microT miRBase PITA PicTar5 TargetscanS 0.3085
Diana-microT miRBase PicTar6 TargetscanS 0.3085
Diana-microT PicTar5 PicTar6 TargetscanS 0.3085
ElMMo miRBase miRanda PITA PicTar5 PicTar6 TargetscanS 0.3085
ElMMo miRBase miRanda PicTar5 TargetscanS 0.3084
PITA PicTar6 TargetscanS 0.3083
Diana-microT miRBase PITA PicTar5 PicTar6 TargetscanS 0.3083
miRBase PITA 0.3082
PITA PicTar5 0.3082
ElMMo miRBase miRanda PicTar5 PicTar6 TargetscanS 0.3082
miRBase PITA PicTar6 0.3081
Diana-microT miRBase PicTar5 PicTar6 0.3080
PITA PicTar5 PicTar6 0.3080
miRBase PITA TargetscanS 0.3080
Diana-microT miRBase PicTar5 TargetscanS 0.3080
PITA PicTar5 TargetscanS 0.3080
miRBase PITA PicTar6 TargetscanS 0.3079
PITA PicTar5 PicTar6 TargetscanS 0.3079
Diana-microT miRBase PicTar5 PicTar6 TargetscanS 0.3078
miRBase PITA PicTar5 0.3077
miRBase PITA PicTar5 TargetscanS 0.3076
miRBase PITA PicTar5 PicTar6 0.3075
miRBase PITA PicTar5 PicTar6 TargetscanS 0.3075
TargetscanS 0.3073
Diana-microT miRanda PITA TargetscanS 0.3071
Diana-microT miRanda PITA 0.3070
Diana-microT miRanda PITA PicTar6 TargetscanS 0.3070
Diana-microT miRanda PITA PicTar6 0.3069
Diana-microT miRBase miRanda PITA TargetscanS 0.3068
Diana-microT miRanda PITA PicTar5 TargetscanS 0.3068
PicTar6 TargetscanS 0.3068
Diana-microT miRBase miRanda PITA 0.3067
Diana-microT miRanda PITA PicTar5 0.3067
Diana-microT miRBase miRanda PITA PicTar6 0.3067
Diana-microT miRBase miRanda PITA PicTar6 TargetscanS 0.3067
Diana-microT miRanda PITA PicTar5 PicTar6 TargetscanS 0.3067
Diana-microT miRanda PITA PicTar5 PicTar6 0.3066
Diana-microT miRBase miRanda PITA PicTar5 TargetscanS 0.3066
Diana-microT miRBase miRanda PITA PicTar5 0.3065
Diana-microT miRBase miRanda PITA PicTar5 PicTar6 TargetscanS 0.3065
Diana-microT miRBase miRanda PITA PicTar5 PicTar6 0.3064
miRanda PITA TargetscanS 0.3060
miRanda PITA PicTar6 TargetscanS 0.3060
miRBase TargetscanS 0.3059
miRanda PITA 0.3058
miRBase miRanda PITA TargetscanS 0.3058
PicTar5 TargetscanS 0.3058
miRanda PITA PicTar5 TargetscanS 0.3058
miRanda PITA PicTar6 0.3057
miRBase miRanda PITA PicTar6 TargetscanS 0.3057
miRanda PITA PicTar5 PicTar6 TargetscanS 0.3057
miRBase miRanda PITA PicTar5 TargetscanS 0.3056
miRBase PicTar6 TargetscanS 0.3056
miRBase miRanda PITA 0.3055
miRanda PITA PicTar5 0.3055
PicTar5 PicTar6 TargetscanS 0.3055
miRBase miRanda PITA PicTar5 PicTar6 TargetscanS 0.3055
miRBase miRanda PITA PicTar6 0.3054
miRanda PITA PicTar5 PicTar6 0.3054
Diana-microT miRanda TargetscanS 0.3053
miRBase miRanda PITA PicTar5 0.3052
miRBase miRanda PITA PicTar5 PicTar6 0.3052
Diana-microT miRanda PicTar6 TargetscanS 0.3052
Diana-microT miRBase miRanda TargetscanS 0.3050
miRBase PicTar5 TargetscanS 0.3049
Diana-microT miRanda PicTar5 TargetscanS 0.3049
Diana-microT miRBase miRanda PicTar6 TargetscanS 0.3049
218

8.5 Target Prediction Ensemble Analysis Results
Prediction algorithms involved in the combination set E-score
Diana-microT miRanda PicTar5 PicTar6 TargetscanS 0.3049
Diana-microT miRBase miRanda PicTar5 TargetscanS 0.3047
miRBase PicTar5 PicTar6 TargetscanS 0.3047
Diana-microT miRBase miRanda PicTar5 PicTar6 TargetscanS 0.3046
Diana-microT miRanda 0.3045
Diana-microT miRanda PicTar6 0.3044
Diana-microT miRBase miRanda 0.3041
Diana-microT miRanda PicTar5 0.3040
Diana-microT miRBase miRanda PicTar6 0.3040
Diana-microT miRanda PicTar5 PicTar6 0.3039
Diana-microT miRBase miRanda PicTar5 0.3037
Diana-microT miRBase miRanda PicTar5 PicTar6 0.3037
PicTar6 0.3020
miRanda TargetscanS 0.3020
miRanda PicTar6 TargetscanS 0.3020
miRBase miRanda TargetscanS 0.3019
miRBase miRanda PicTar6 TargetscanS 0.3019
miRanda PicTar5 TargetscanS 0.3018
miRanda PicTar5 PicTar6 TargetscanS 0.3018
miRBase miRanda PicTar5 TargetscanS 0.3017
miRBase miRanda PicTar5 PicTar6 TargetscanS 0.3017
miRBase PicTar6 0.3011
miRBase 0.3008
miRBase PicTar5 PicTar6 0.3003
miRBase PicTar5 0.3000
PicTar5 PicTar6 0.2999
PicTar5 0.2989
miRBase miRanda PicTar5 PicTar6 0.2987
miRBase miRanda PicTar6 0.2986
miRBase miRanda PicTar5 0.2985
miRBase miRanda 0.2984
miRanda PicTar5 PicTar6 0.2983
miRanda PicTar6 0.2982
miRanda PicTar5 0.2981
miRanda 0.2979
219

Chapter 9
Discussion
9.1 Digital Profiling of GNS Cell Lines
Gliobastoma multiforme is the most common primary brain tumour and the
most aggressive glioma in adults. No effective solution has been embodied
as a treatment yet, causing the prognosis for this disease to be very poor,
with a median survival time of 15 months [472]. The extensive cellular het-
erogeneity typical of glioblastomas is confirmed by the consistently different
molecular signatures and copy number variations observed in the subclasses
present within the primary and secondary subtypes [154,334,450]. These ob-
servations are very important since they point at the need for approaching
treatment research from a more molecular standpoint that allows for patient
treatment diversification, in which the patient’s genome is specifically tailored
to obtain the most effective results. An important finding within the cancer
stem cell field in the context of glioblastomas, was the observation that these
tumours contain a population of cells with similarities to NS cells. Before
then, gliomas were studied as cancer cell lines, which hid the underlying stem
cell component of the tumour by causing it to differentiate and could therefore
never be specifically targeted. NS cells give rise to both neurons and glia during
the development of the nervous system and are present in restricted regions of
the adult human brain, where they constitute proliferating germinative zones
that are active throughout adulthood [248]. According to the cancer stem cell
hypothesis, such stem cell-like cell populations are responsible for maintaining
cancers, as well as giving rise to the differentiated progeny responsible for the
cellular diversity of many neoplasias, including glioblastoma [379]. If this is
the case, isolating and characterizing the glioma cancer stem cells will be key
to developing efficient therapies for glioblastoma multiforme.
220

9.1. Digital Profiling of GNS Cell Lines Discussion
Before Pollard et al [404] adapted the protocol developed by Conti et al [107]
for the derivation of NS cells in adherent serum-free culture, to glioma tu-
mours, glioma-related research was conducted on glioma cell lines (see section
1.3) and neurosphere cultures (see section 2.2). Unlike the derivation method
for cancer cell lines, which involves serum-containing media that is selective
for proliferative cells which therefore lose their differentiation identity, the
NS cell line derivation method uses a defined medium with the addition of
growth hormones EGF and FGF2 that rather supports the self-renewal pro-
cess. While neurosphere cultures have been instrumental in the detection and
quantification of the presence of a stem cell component in gliomas, the imme-
diate differentiation that is observed when neurospheres start adhering, makes
them a suboptimal tool for the long term characterisation and manipulation
of glioma stem cells. When it was discovered that, although adult human NS
cells are difficult to study, fetal human NS cells can be isolated and main-
tained as untransformed cell lines in serum-free medium supplemented with
growth factors, the link was made that the same protocol could be used to ob-
tain NS-like cells from gliomas and maintain them in vitro [404]. These cells,
termed Glioma Neural Stem Cells (GNS), have similar morphology as NS cells
obtained in the same manner, when grown as adherent cultures and feature
other similarities including expression of progenitor cell markers and capacity
to differentiate into multiple neural lineages. In contrast to NS cells, however,
GNS cells harbour genetic aberrations characteristic of glioblastoma and form
glioma-like tumours when transplanted into immunocompromised mice [404].
Thus, the derivation of NS cells from brain tissue forms the basis for culturing
GNS cells from gliomas and by using the same protocol for the isolation of
fetal human NS cells, the isolation of NS-like cells in adherent culture [404] set
the best research platform for conducting research that lends itself better to
the achievement of patient tailored therapies.
In this thesis I show several lines of support for GNS lines being suitable models
for the understanding of the molecular basis of glioma:
1. GNS/NS differential expression analysis highlights pathways in glioma;
2. GNS expression profiles capture molecular signatures of glioblastoma
subtypes established by large microarray studies;
3. many core DE genes identified in the GNS/NS comparison show similar
changes in expression data from primary glioblastomas and xenografts.
221

9.1. Digital Profiling of GNS Cell Lines Discussion
In order to reveal transcriptional changes that underlie glioblastoma, I per-
formed an in-depth analysis of gene expression in malignant stem cells derived
from patient tumours in relation to untransformed, karyotypically normal neu-
ral stem cells. These cell types are closely related and it has been hypothe-
sised that gliomas arise by mutations in NS cells or in glial cells that have
reacquired stem cell features [404]. We measured gene expression by high-
throughput RNA tag sequencing (Tag-seq), a method which features high sen-
sitivity and reproducibility compared to microarrays [490]. qRT-PCR valida-
tion further demonstrates that Tag-seq expression values are highly accurate.
Other cancer samples and cell lines have recently been profiled with the same
method [346,383], which should make these samples comparable to ours and
the analyses presented in this thesis. Through Tag-seq expression profiling
of normal and cancer stem cells followed by qRT-PCR validation in a wider
panel of 22 cell lines, we identified 29 genes strongly discriminating GNS from
NS cells. Some of these genes have previously been implicated in glioma, in-
cluding four with a role in adhesion and/or migration, CD9, ST6GALNAC5,
SYNM and TES [63,241,251,477], and two transcriptional regulators, FOXG1
and CEBPB. This observation is in line with the gene ontology analysis, which
revealed "Cell adhesion” and "Cell migration” as relevant biological processes
amongst our set of differentially expressed genes (Fig 7.1), and the fact that,
although infiltrative spread is a common feature of all diffuse astrocytic tu-
mours, glioblastoma is particularly notorious for its rapid invasion of neigh-
boring brain structures [301]. Activation of the TGFβ and AKT pathways
have been described as possible molecular mediators of this invasion [245,527]
and a number of other expression profiling studies have identified a subset of
tumours with elevated expression of ECM components as well as intracellular
proteins associated with cell motility [148,390]. FOXG1, which has been pro-
posed to act as an oncogene in glioblastoma by suppressing growth-inhibitory
effects of TGFβ [448], showed remarkably strong expression in all 16 GNS lines
assayed by qRT-PCR. CEBPB was recently identified as a master regulator
of a mesenchymal gene expression signature associated with poor prognosis in
glioblastoma [85]. Studies in hepatoma and pheochromocytoma cell lines have
shown that the transcription factor encoded by CEBPB (C/EBPβ) promotes
expression of DDIT3 [140], another transcriptional regulator that we found
to be up-regulated in GNS cells. DDIT3 encodes the protein CHOP, which
in turn can inhibit C/EBPβ by dimerizing with it and acting as a dominant
negative [85]. This interplay between CEBPB and DDIT3 may be relevant for
222

9.1. Digital Profiling of GNS Cell Lines Discussion
glioma therapy development, as DDIT3 induction in response to a range of
compounds sensitises glioma cells to apoptosis (see e.g. [216]).
Our results also corroborate a role in glioma for several other genes with limited
prior links to the disease. This list includes PLA2G4A, HMGA2, TAGLN and
TUSC3, all of which have been implicated in other neoplasias (Appendix A.2).
PLA2G4A encodes a phospholipase that functions in the production of lipid
signaling molecules with mitogenic and pro-inflammatory effects. In a subcu-
taneous xenograft model of glioblastoma, expression of PLA2G4A by the host
mice was required for tumour growth [285]. For HMGA2, a transcriptional
regulator down-regulated in most GNS lines, low or absent protein expression
has been observed in glioblastoma compared to low-grade gliomas [285], and
HMGA2 polymorphisms have been associated with survival time in glioblas-
toma [295]. TAGLN, another gene down-regulated in most GNS lines, encodes
the actin-binding protein transgelin. TAGLN has been characterised as a tu-
mour suppressor with lost expression in prostate, breast and colon cancers [30],
but we only found one prior study on TAGLN in glioma, showing low expression
in a glioma cell line from rats [181]. TUSC3 is commonly silenced by promoter
methylation in glioblastoma, in particular in patients above 40 years of age.
Loss or down-regulation of TUSC3 has been found in several other cancers,
e.g. of the colon where TUSC3 becomes hypermethylated with age [13]. The
function of TUSC3 is unknown, but may relate to protein glycosylation [337].
The set of 29 genes found to generally distinguish GNS from NS cells also in-
cludes multiple genes implicated in other neoplasias, but without direct links
to glioma, such as SULF2, NNMT and LMO4 (Appendix A.2). Of these, the
transcriptional regulator LMO4 may be of particular interest, as it is well stud-
ied as an oncogene in breast cancer and regulated through the phosphoinosi-
tide 3-kinase pathway [340], which is commonly affected in glioblastoma [326].
SULF2 encodes an extracellular sulfatase that modulates interactions between
growth factors and their receptors, with effects on multiple signaling pathways.
It is up-regulated in several cancers, including a mouse model of glioma [211]
and, as shown in our differential expression analysis, in many GNS lines.
Five of these 29 genes have not been directly implicated in cancer. This
list comprises one gene down-regulated in GNS lines (PLCH1) and four up-
regulated (ADD2, LYST, PLA2G4A, PDE1C and PRSS12). PLCH1 is in-
volved in phosphoinositol signaling [228], like the frequently mutated phospho-
223

9.1. Digital Profiling of GNS Cell Lines Discussion
inositide 3-kinase complex [326]. Interestingly, both PDE1C and PLCH1 are
activated by intracellular Ca 2+ [126,228], in line with these genes being involved
in PI3K and cAMP regulation, which are both Ca 2+ regulated pathways (see
pathway figure 7.8). PLA2G4A encodes a cytoplasmic phospholipase involved
in production of lipid signaling molecules with mitogenic and pro-inflammatory
effects [192]. ADD2 encodes a cytoskeletal protein that interacts with FYN,
a tyrosine kinase promoting cancer cell migration [456,533]. For PDE1C, a
cyclic nucleotide phosphodiesterase gene, we found higher expression to cor-
relate with shorter survival after surgery. Up-regulation of PDE1C has been
associated with proliferation in other cell types through hydrolysis of cAMP
and cGMP [126,437]. PRSS12 encodes a protease that can activate tissue plas-
minogen activator (tPA) [335], an enzyme which is highly expressed by glioma
cells and has been suggested to promote invasion [168]. Indeed, our Tag-seq
data shows that the tissue plasminogen activator gene PLAT is expressed in all
the assayed GNS lines and strongly up-regulated in all except one (Appendix
A.1). LYST, which encodes a cytoplasmic protein involved in lysosomal traf-
ficking, has no clear role in cancer-related processes, although there is evidence
of altered protein kinase C levels in LYST-deficient cells [168]. The GNS versus
NS transcriptome comparison thus identified multiple genes known to play a
role in glioma as well as several other genes likely to do so.
I have compared gene expression and non-coding RNA expression between can-
cer stem cells from glioblastoma and NS cells, as well as evaluated correlations
amongst sets of highly significant differentially expressed genes found in our
dataset and other glioblastoma cell lines established by The Cancer Genome
Atlas project [326] and other lower grade gliomas from studies by Freije et
al [148] and Phillips et al [390]. I have also verified how gene expression might
be affected by the aberrations known to exist in glioblastoma cell genomes to
conclude that there was only a modest trend between the presence of aberra-
tions and the gene expression observed and no adjustment needed to be made
to our differential expression calls. Interestingly, the GNS cell lines tend to
have very low genomic instability outside the one carried in lieu of them being
cancer-derived cell lines, which typically increases over a short period of time
to then stabilise later on in classic cancer cell lines. GNS cell lines start accu-
mulating chromosomal and genomic aberrations very late, after having passed
the one hundred passage mark [404]. The four GNS lines that we profiled have
been thoroughly phenotypically characterised and all give rise to glioma-like
224

9.1. Digital Profiling of GNS Cell Lines Discussion
tumours when transplanted into immunocompromised mice [404]. In addition,
the lines were established from tumours with differing histology, allowing us
to sample the breadth of the disease. In fact, our correlation analysis (Fig
6.6) showed that when compared with each other the GNS cell lines fared
the worst with respect to the high scoring correlations observed, instead, be-
tween our biological replicates (the two cell lines established from the same
parental tumour, but in different laboratories) and the two NS cell lines. This
is expected, as G144, G166 and G179 originate from different tumours with
histologically distinct properties and further confirms the presence of molecular
variability amongst different glioblastoma patients: G144 was derived from a
glioblastoma with a significant oligodendrocyte component, G166 from a case
of glioblastoma multiforme and G179 from a giant cell glioblastoma.
Furthermore, by considering expression changes in a pathway context (Fig
7.8), we identified several genes that have not been directly implicated in
glioma, but participate in glioma-related pathways, such as the putative cell
adhesion gene ITGBL1 [49], the orphan nuclear receptor NR0B1, which is
strongly up-regulated in G179 and is known to be up-regulated and mediate
tumour growth in Ewing’s sarcoma [158], and the genes PARP3 and PARP12
(Table 6.6). Other examples include three down-regulated calcium channel
genes, CACNA1C, CACNG7 and CACNG8, and one up-regulated, CACNA1A.
Deregulation of these genes could affect cellular calcium levels, which influence
the activation status of protein kinase C and indirectly the MAPK pathway.
Another example is ITGBL1, an integrin-related gene that is expressed by
G166 and G179 and may play a role in cell adhesion [49]. The pathway analy-
sis also highlighted the up-regulated genes PARP3 and PARP12, which belong
to the PARP family of ADP-ribosyl transferase genes involved in DNA repair.
The up-regulation of these genes may have relevance for therapy, as PARP
inhibitors currently are in clinical trials for several cancers and there is evi-
dence that PTEN loss, which is common in glioblastoma, may sensitise cells
to PARP inhibitors [325].
Transcriptome analysis thus identified multiple genes of known significance
in glioma pathology as well as several novel candidate genes and pathways.
These results are further corroborated by survival analysis, which revealed a
GNS expression signature associated with patient survival time in five inde-
pendent datasets. This finding is compatible with the notion that gliomas
225

9.1. Digital Profiling of GNS Cell Lines Discussion
contain a GNS component of relevance for prognosis. Five individual GNS
signature genes were significantly associated with survival of glioblastoma pa-
tients in both of the two largest data sets: PLS3, HOXD10, TUSC3, PDE1C
and the well-studied tumour suppressor PTEN. PLS3 (T-plastin) regulates
actin organisation and its over-expression in the CV-1 cell line resulted in
partial loss of adherence [26]. Elevated PLS3 expression in GNS cells may
thus be relevant for the invasive phenotype. The association between tran-
scriptional up-regulation of HOXD10 and poor survival is surprising, because
HOXD10 protein levels are suppressed by a microRNA (miR-10b), which is
highly expressed in gliomas, and it has been suggested that HOXD10 suppres-
sion by miR-10b promotes invasion [476]. Notably, in our microRNA microar-
ray dataset microRNA miR-10b is up-regulated three-fold in GNS cell lines
with respect to NS cell lines (Table 6.8; Appendix F) and the HOXD10 mRNA
up-regulation we observe in GNS lines also occurs in glioblastoma tumours, as
shown by comparison with grade III astrocytoma (Fig 7.4b). Similarly to
our findings, miR-10b is present at higher levels in glioblastoma compared to
gliomas of lower grade [476]. It is conceivable that HOXD10 transcriptional
up-regulation and post-transcriptional suppression is indicative of a regulatory
program associated with poor prognosis in glioma.
Also, from the survival analysis it was clear that tumours from older patients
featured an expression pattern more similar to the GNS signature. One of the
genes contributing to this trend, TUSC3, is known to be silenced by promoter
methylation in glioblastoma, particularly in patients over 40 years of age [277].
Loss or down-regulation of TUSC3 has been found in other cancers, e.g. of
the colon, where its promoter becomes increasingly methylated with age in the
healthy mucosa [13]. Taken together, these data suggest that transcriptional
changes in healthy aging tissue, such as TUSC3 silencing, may contribute to the
more severe form of glioma in older patients. Thus, the molecular mechanisms
underlying the expression changes described here are likely to be complex and
varied. To capture these effects and elucidate their causes, transcriptome anal-
ysis of cancer samples will benefit from integration of diverse genomic data,
including structural and nucleotide-level genetic alterations, as well as DNA
methylation and other chromatin modifications.
To identify expression alterations common to most glioblastoma cases, other
studies have profiled tumour biopsies in relation to non-neoplastic brain tis-
sue [196,383,497]. While such comparisons have been revealing, their power
226

9.1. Digital Profiling of GNS Cell Lines Discussion
is constrained by discrepancies between reference and tumour samples; for in-
stance the higher neuronal content of normal brain tissue compared to tumours.
Gene expression profiling of tumour biopsies further suffers from mixed signal
due to a stromal cell component and heterogeneous populations of cancer cells,
only some of which contribute to tumour progression and maintenance [379].
Part of a recent study bearing a closer relationship to our analysis examined
gene expression in another panel of glioma-derived and normal NS cells [299],
but included neurosphere cultures which often contain a heterogeneous mix-
ture of self-renewing and differentiating cells. We have managed to circumvent
these issues by profiling uniform cultures of primary malignant stem cell lines
that can reconstitute the tumour in vivo [402], in direct comparison to normal
counterparts of the same cell type [107,481]. While the resulting expression
patterns largely agree with those obtained from glioblastoma tissues, there are
notable differences. For example, we found the breast cancer oncogene LMO4
(discussed above) to be up-regulated in most GNS lines, although its average
expression in glioblastoma tumours is low relative to normal brain tissue (Fig
7.4 3a). Similarly, TAGLN and TES were absent or low in most GNS lines, but
displayed the opposite trend in glioblastoma tissue compared to normal brain
(Fig 7.4c) or grade III astrocytoma (Fig 7.4d). Importantly, both TAGLN and
TES have been characterised as tumour suppressors in malignancies outside
the brain and the latter is often silenced by promoter hypermethylation in
glioblastoma [30,349]. A very interesting hypothesis about the TES gene that
we could explore further with specific biochemical assays, is that, assuming the
methylation mark observed in glioblastoma literature is maintained in our GNS
cell lines, there must be a very robust mechanism to keep this gene silenced.
In fact, TES is found on chromosome 7, which is nearly always present in very
high copy number, for example in cell line G144, which carries more than 10
copies at late passages (Fig 6.7). If we could verify that the methylation mark
in our GNS cell lines was maintained, this would be consistent with TES being
silenced very early on in the disease, before the chromosomal gains and ane-
uploidy, which may make it a good candidate as a glioblastoma initiating event.
In assigning each GNS cell line to one of the four expression signature profiles
as defined by the latest study by Verhaak et al [511], each match was reflective
of their known histopathological features and this further supported the need
for a patient stratification approach in assigning therapies. G166 and G179,
both glioblastoma cell lines, were assigned to the mesenchymal signature, asso-
227

9.1. Digital Profiling of GNS Cell Lines Discussion
ciated with the poorest prognosis. G144, instead, with its high oligodendrocyte
component known to positively correlate with patient survival rates in glioblas-
toma, was assigned to the proneural signature, linked with neural markers and
therefore the best prognosis of all. In comparing the GNS line expression pro-
files to the subtype signatures, we found that both G166 and G179 correlated
strongly with the mesenchymal signature with worst prognosis. Mesenchymal
subtype markers with elevated expression in these two lines included MET,
CD44, CD68 and CASP1 [511]. G144 did not correlate significantly with any
of the four signatures, but showed a slight positive correlation with the proneu-
ral one. Supporting such classification of G144 were several of the hallmarks
of the proneural subtype emphasised by Verhaak et al 2010 [511]: high expres-
sion of oligodendrocytic development genes PDGFRA, NKX2-2 and OLIG2,
as well as ERBB3, DCX and TCF4 genes and low levels of tumour suppressor
CDKN1A. These expression signature profile studies should be performed on
an always greater number of glioblastoma samples in order for them to be able
to capture even finer differences than the ones proposed by the study by Ver-
haak et al, which have already proven the existence of different glioblastoma
classes and different prognoses associated with each class. If each class were
to be targeted by a tailored molecular therapy, this class would find it most
beneficial as a treatment and patients would obtain better results.
A Gene Ontology term analysis confirmed that the sets of differentially ex-
pressed genes were enriched for genes involved in processes related to brain de-
velopment and cancer biology. We also observed enrichment of genes encoding
regulatory and inflammatory proteins, such as signal transduction components,
cytokines, growth factors and DNA binding proteins. In line with these find-
ings, affected pathways from the KEGG database included Cytokine-cytokine
receptor interaction, Neuroactive ligand-receptor interaction, MAPK signaling
and, expectedly, Glioma, a collection of genes involved in glioma formation.
GSEA analysis revealed a consistent up-regulation of inflammatory genes in
the GNS lines belonging especially to the MHC class II family, suggesting an
immune-evasion phenotype that has already been called upon by a small num-
ber of early glioma studies [468,484]. The up-regulation in the GNS lines was
seen in several MHC class II genes, as well as related genes involved in antigen
presentation on MHC class I complexes. Several works have already shown
that MHC class I and II molecules are involved in aspects of human cancer
pathology such as invasion and migration [327,421,546]. We find an overall
228

9.1. Digital Profiling of GNS Cell Lines Discussion
transcriptional up-regulation of MHC class II molecules and a smaller subset
of MHC class I molecules, suggestive of the absence of transcriptionally active
compensatory mechanisms. Follow-up proteomic studies could be carried out
by us to further explore the dynamics of this immunoevasion mechanism in ac-
tion. The identification of specific protein expression levels, in fact, would help
us understand the level at which this regulation, which is bound to exist given
the up-regulation observed especially in MHC class II molecules, takes place.
It would help us answer questions such as whether an excess of MHC class
II molecules is also present on the surface of GNS cells, and if there are any
correlations with the expression signature profiles and the diagnosis associated
with them. In fact, it is possible that in the proneural signature would fall
those glioblastoma samples that carry higher protein expression levels of MHC
class II molecules, which by exposing aberrant extracellular antigens, would
be responsible for a more efficient immunological activation against those cells,
in line with the kinder prognosis associated with this signature. On the other
hand, it is also possible that little or no sample-dependent variation exists in
the up-regulation of MHC class molecules and therefore no correlation between
prognosis and MHC member protein levels.
In the attempt to observe whether non-coding RNA regulation plays an impor-
tant role in the levels of gene expression observed for GNS cell lines, differential
microRNA-targeted isoform expression and long non-coding RNA expression
were evaluated. Assuming that the differential isoform expression is due to
microRNA regulation exerted on a specific isoform, after having identified
approximately 2,000 isoform candidates for differential expression detected by
multiple tag mappings, we verified which predicted microRNA sequences (from
five of the leading algorithms) aligned on their 3'UTRs. An interesting candi-
date for isoform microRNA regulation is gene NTRK2, which encodes a recep-
tor for brain-derived neurotrophic factor, promoting differentiation, prolifera-
tion and survival [118]. In several tumour types, NTRK2 expression correlates
with poor prognosis and metastasis [118] and the protein has been detected
in a subset of cells in astrocytomas [29]. The Tag-seq data demonstrates that
GNS lines express a short NTRK2 isoform, potentially the one that lacks the
kinase domain (Fig 6.11), which has been implicated in regulation of astrocyte
morphology [366]. Other interesting candidate genes for microRNA regulation
of isoform expression, are the tumour suppressor gene BRCA1, involved in p53
signaling [370], the genes AKT2 and AKT3, involved in glioblastoma tumour
evasion phenotypes [245], BMP7, a member of the TGF-β superfamily of bone
229

9.1. Digital Profiling of GNS Cell Lines Discussion
morphogenetic proteins that plays a key role in the transformation of mes-
enchymal cells into bone and cartilage [245], and ERBB2, HLA-A and PTEN.
We found that 226 of the approx. 2,000 differentially expressed isoforms hosted
at least one microRNA seed sequence between at least two tags, which identi-
fied microRNA regulation as a widely adopted mechanism of regulation of gene
and isoform expression in GNS cell lines. An important follow-up experiment
for which data came in only recently in our laboratory, is the measurement of
microRNA expression levels in our GNS cell lines on a microRNA microarray
platform. Analysis of this data revealed that, of the 226 microRNAs identified
on a prediction basis as regulators of isoforms within our GNS study, miR-
10b (see discussion above) and miR-26a were consistently up-regulated in our
GNS cell lines, and miR-137, miR 128, miR-34a, miR-129-3p, and miR-451
consistently down-regulated, in line with the glioblastoma microRNA litera-
ture [96,278,364].
In the assessment of the regulation potentially performed by long non-coding
RNAs, we found 18 up-regulated and 7 down-regulated putative ncRNAs, three
of which are known to be long antisense RNAs: CDKN2BAS, HOTAIRM1
and NEAT1. Although long-non coding RNA regulation still needs to be
thoroughly elucidated, an interesting pattern was observed with the levels of
expression of HOTAIRM1 in our GNS cell lines and the expression trend found
in literature in human NB4 promyelocytic cell lines. In fact, upon induction
of granulocytic differentiation, HOTAIRM1 becomes strongly up-regulated in
human NB4 promyelocytic cell lines and normal hematopoietic cells, and its
knock-down in NB4 cells causes down-regulation of the HOXA1 and HOXA4
genes [547]. In line with these observations, HOTAIRM1 is found to be up-
regulated in our GNS cell lines compared to our NS lines, and the HOXA1 and
HOXA4 genes are also up-regulated (Fig 6.22), which potentially identifies
them as conserved targets of HOTAIRM1, although an immuno-precipitation
assay should be performed in order to conclude that.
Overall, in this thesis I demonstrate that our results support GNS lines as
suitable model for understanding the molecular basis of glioblastoma, and use
of NS cell lines as controls in this setting. By this approach, we have identified
several likely oncogenes and tumour suppressors that have not previously been
associated with glioblastoma. With the advent of GNS cell lines, the glioblas-
toma research field is bound to be enriched by experiments that will represent
230

9.2. MicroRNA Target Prediction Analysis Discussion
in their results also the stem cell component of the cancer and therefore enable
more accurate targeting therapeutic strategies. Outside of the glioblastoma
research field, similar concepts will be applied for the adherent culturing of
other solid cancers with a preponderant stem cell component, such as other
brain cancers and breast as well as colon, pancreatic and lymphoma cancers.
The ability to work in an adherent culture that maintains intact the stem cell
component of the cancer for long periods of time is an enabling feature for the
entire cancer research field that will enable researchers to move forward faster
towards the achievement of effective cancer therapeutic strategies.
9.2 MicroRNA Target Prediction Analysis
In this thesis I have also constructed a microRNA target prediction analysis
tool for the manipulation of microRNA target prediction data and combined
the exon array expression data (Appendix E) with the microRNA array ex-
pression data (Appendix E) to produce, using the GenemiR software tool, an
ensemble analysis that would help me answer the question whether any com-
bination of prediction algorithms together were more effective and accurate at
predicting mRNA targets that any prediction algorithm alone.
microRNA target prediction is the result of the factoring of several different
variables that are each treated differently, depending on the algorithm at work,
in terms of the importance that is assigned to each within the algorithm itself.
Although several robust prediction algorithms have been developed, they all
have the problem of predicting with a very high percentage of false positives.
Furthermore, the poor agreement that is so often observed between sets of
results originating from exactly the same input list but processed by different
algorithms, tells us that there needs to be more to be then account of. Thus,
the characterisation of microRNA function and regulation of mRNAs is de-
pendent on a robust strategy for managing disparate results. Some prediction
algorithms generate a large number of putative regulatory targets, many of
which are exclusive to a particular method. Moreover, those results that do
agree will not necessarily exhibit higher prediction accuracy. Even in the best
case where the degree of overlap is approximately 70% (between TargetScanS
and PicTar), it is not obvious which algorithm produces the optimal set of
target predictions. When predictions are made, they must then be viewed
231

9.3. Concluding Remarks Discussion
in context with other genomic information in order to elucidate regulatory
function. In an attempt to elucidate the answer to the research question of
interest, the hypothesis put forward by Alexiou et al [16] that combinations of
algorithms predict more accurately than the single algorithm alone was chal-
lenged. Through the analysis described in this thesis it is clear that the target
prediction algorithm ElMMo [155]is at the head of the list sorted by accuracy
score out of the 258 combinations tested for. Although this analysis shows
that ElMMo in particular has the most accurate target prediction algorithm
within the pool of eight tested prediction algorithms, it must be noted that
there is a substantial discrepancy between the score achieved by ElMMo and
the next best score for a single target prediction algorithm. Before Diana-
microT [237,316], the second highest solo score, many combinations of two to
three target prediction algorithms appear earlier in the list, demonstrating the
potential for better accuracy than another solo prediction algorithm. Although
it is refreshing to see that one target prediction algorithm can fare better than
the rest, the question remains as to whether the variables and factors that are
taken into consideration, and the ones that are not, are fairly judged in the
algorithm so as to simulate the intricate regulations happening within the cell.
In fact, during this analysis the accuracy score of each prediction algorithm
decreased greatly when a background gene list was not used to filter out all
the genes that were predicted but were not tissue specific.
While the results of this analysis indicate that some algorithms alone and some
group of algorithms are better than others, it highlights at the same time the
importance of re-assessing the factors that are taken into consideration to
make the microRNA target predictions. If we have reached the limit of what
sequence-based algorithms can achieve, we ought to start thinking of adding
another dimension to the field of microRNA target prediction algorithms that
takes factors in components of the tissue-specific regulatory actions of microR-
NAs.
9.3 Concluding Remarks
In this thesis I aimed to characterise the transcriptional landscape of Glioma
Neural Stem Cells (GNS) in the most comprehensive way possible. Several
approaches were taken to examine the expression profiling data that gave in-
sights into the stem cell component of the biology of these cells, which had
232

9.3. Concluding Remarks Appendix
previously been overlooked due to the suboptimal culturing systems for the
maintenance of pluripotency.
To this end I have approached the analysis from the coding transcriptome and
the non-coding transcriptome point of view, trying to draw as many parallels
as possible and as many comparisons as possible with other external datasets
that could give insights in the characteristics the are unique to the GNS sys-
tem.
Pairing the available expression data with new DNA methylation data would
allow us to gain insights into the epigenetic mechanisms at work. It is, in
fact, important that the future works in this field and with these cell lines
are directed towards a deeper understanding of the mechanisms that elect
and maintain them as tumour-initiating agents of the glioblastoma primary
tumour.
233

Appendix A
Differentially Expressed Genes
A.1 Differential Expression
The differentially expressed genes generated with the DESeq Bioconductor
package are listed below in alphabetical order. The values for each gene are
reported in each cell line in order to give the reader an idea on the directionality
of the fold change between diseased and normal counterpart. Rows with mul-
tiple gene Ensembl and/or Entrez gene IDs separated by comma correspond
to cases where there was not a one-to-one correspondence between Entrez and
Ensembl genes.
234

A.1 Differential Expression Appendix
Table A.1: Classification of differentially expressed genes at 10% FDR.
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
ABAT 18 ENSG00000183044 -2.61 2.69E-04 1.07E-02 769.4 1682.2 48.1 223.8 6470.4 1513.8
AC007405.8,LOC285141 285141 ENSG00000204334 3.05 1.26E-03 3.56E-02 16.2 23.0 120.1 184.0 0.0 25.9
AC012215.1,UNC5D 137970 ENSG00000156687,ENSG00000233863 -Inf 1.06E-08 1.71E-06 0.0 0.0 0.0 0.0 206.3 57.2
AC012354.2 ENSG00000225301 -Inf 4.53E-07 4.98E-05 0.0 0.0 0.0 0.0 26.6 170.3
AC026410.6 ENSG00000227087 3.52 2.06E-03 5.09E-02 56.7 96.6 93.1 19.3 0.0 11.9
AC034102.1 ENSG00000237493 8.09 9.42E-10 1.95E-07 79.2 227.3 187.2 0.0 0.0 1.0
AC067930.1 ENSG00000228381 4.86 1.80E-03 4.69E-02 28.4 50.3 71.3 0.0 0.7 1.9
AC068399.1 ENSG00000204089 4.25 1.80E-05 1.16E-03 1.4 7.4 346.5 22.9 0.0 12.7
AC092296.1 ENSG00000229449 4.90 2.63E-04 1.05E-02 140.1 87.2 79.9 13.6 0.0 4.2
ACIN1 22985 ENSG00000100813 5.44 4.58E-04 1.65E-02 65.1 0.0 133.1 0.0 0.0 2.0
ACTA2 59 ENSG00000107796 -2.67 3.27E-03 7.27E-02 0.0 0.0 114.4 146.7 618.4 486.3
ACTC1 70 ENSG00000159251 -Inf 1.14E-09 2.33E-07 1.4 0.0 0.0 0.0 0.0 422.8
ADAMTS1 9510 ENSG00000154734 2.88 1.30E-03 3.62E-02 103.3 405.7 57.5 107.6 0.0 51.1
ADAMTS10 81794 ENSG00000142303 -3.00 7.32E-04 2.32E-02 87.9 58.1 12.5 1.7 165.5 218.6
ADAMTS4 9507 ENSG00000158859 Inf 4.45E-04 1.62E-02 31.1 40.2 3.1 54.1 0.0 0.0
ADCYAP1R1 117 ENSG00000078549 -3.35 1.56E-03 4.18E-02 6.1 16.8 0.0 0.0 108.9 8.8
ADD2 119 ENSG00000075340 5.18 1.62E-03 4.29E-02 27.6 18.5 30.6 60.5 0.0 2.0
ADRA1B 147 ENSG00000170214 5.86 2.94E-03 6.68E-02 0.0 0.0 87.6 1.2 0.0 1.0
ADRA2A 150 ENSG00000150594 Inf 7.29E-05 3.65E-03 65.0 130.0 0.0 0.0 0.0 0.0
AFAP1L2 84632 ENSG00000169129 3.56 4.23E-03 8.93E-02 211.6 214.7 1.3 0.0 0.0 12.3
AGPAT9 84803 ENSG00000138678 Inf 3.58E-15 3.32E-12 2.6 0.0 484.9 582.3 0.0 0.0
AGT 183 ENSG00000135744 10.27 3.91E-16 5.28E-13 1585.7 1784.3 4.5 85.8 0.0 1.0
AIDA 64853 ENSG00000186063 2.25 2.85E-03 6.53E-02 349.3 401.7 533.4 334.1 3.2 173.7
AKR1B10 57016 ENSG00000198074 Inf 8.11E-08 1.07E-05 0.0 0.0 27.5 219.6 0.0 0.0
ALDH1A3 220 ENSG00000184254 4.77 1.59E-03 4.23E-02 0.0 0.0 9.9 114.8 0.0 3.0
AMMECR1 9949 ENSG00000101935 3.30 3.46E-03 7.58E-02 42.1 33.4 65.7 96.5 0.0 13.0
ANGPTL1 9068 ENSG00000116194 -7.69 9.68E-06 6.95E-04 5.1 0.0 1.3 0.0 88.4 84.2
ANGPTL2 23452 ENSG00000136859 3.74 2.26E-03 5.48E-02 55.6 128.1 8.0 81.8 3.8 6.7
AP000280.1,C21orf62 56245 ENSG00000205929,ENSG00000239565 -4.76 4.65E-06 3.71E-04 0.0 16.4 0.0 0.0 236.6 47.3
APLN 8862 ENSG00000171388 -7.41 1.52E-06 1.43E-04 1.4 1.6 0.0 0.0 133.9 44.7
APOD 347 ENSG00000189058 3.18 3.52E-04 1.34E-02 558.9 1217.7 12.5 6.6 0.0 90.9
AQP4 361 ENSG00000171885 -Inf 2.09E-03 5.14E-02 5.4 0.0 0.0 0.0 7.8 39.4
ARC 23237 ENSG00000198576 -4.03 1.22E-05 8.48E-04 31.7 27.8 6.8 5.0 417.6 19.0
ARHGAP20 57569 ENSG00000137727 Inf 6.40E-04 2.11E-02 156.7 74.6 0.8 18.0 0.0 0.0
ARHGAP8,PRR5,PRR5-
23779,553158,55615 ENSG00000186654,ENSG00000241484 -5.71 1.48E-03 4.03E-02 0.0 0.0 3.1 0.0 5.1 103.5
235
ARHGAP8
ARHGEF7 8874 ENSG00000102606 -3.05 9.53E-05 4.54E-03 120.8 94.7 18.6 28.3 713.6 79.4
ASNS 440 ENSG00000070669 2.33 4.47E-03 9.28E-02 619.4 1064.7 1962.9 478.9 207.0 256.1
ASPN 54829 ENSG00000106819 -Inf 4.57E-03 9.45E-02 0.0 0.0 0.0 0.0 0.0 39.5
ATAD3C 219293 ENSG00000215915 6.90 1.30E-04 5.85E-03 0.0 13.8 118.7 3.0 1.0 0.0
ATOH8 84913 ENSG00000168874 -3.96 3.00E-05 1.74E-03 129.9 36.2 0.0 18.8 226.6 339.9
ATP10B 23120 ENSG00000118322 Inf 7.54E-13 2.95E-10 669.9 823.1 0.0 8.1 0.0 0.0
ATP1A2 477 ENSG00000018625 -6.16 4.66E-11 1.33E-08 39.0 37.5 0.6 0.0 471.0 1346.4
ATP1B2 482 ENSG00000129244 -2.46 6.07E-04 2.02E-02 337.6 303.2 6.3 69.3 816.5 581.6
AZGP1 563 ENSG00000160862 Inf 1.37E-14 8.49E-12 1174.8 1213.2 23.8 0.0 0.0 0.0
B3GNT9 84752 ENSG00000237172 2.96 5.06E-04 1.77E-02 84.4 78.1 85.8 504.6 10.7 46.6
B4GALNT1 2583 ENSG00000135454 3.56 1.99E-04 8.35E-03 1771.8 1329.5 54.1 93.5 62.7 21.1
BACE2 25825 ENSG00000182240 7.97 1.22E-13 5.48E-11 227.5 423.1 253.4 468.2 0.0 2.7
BAMBI 25805 ENSG00000095739 5.49 4.12E-09 7.20E-07 1880.4 3430.1 25.3 173.3 23.2 30.4
BATF3 55509 ENSG00000123685 5.12 9.08E-05 4.34E-03 212.0 123.3 5.0 83.6 0.0 3.8
BC008001 4.94 1.32E-03 3.67E-02 25.7 171.5 2.7 0.0 3.8 0.0

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
BEX5 340542 ENSG00000184515 4.39 1.64E-04 7.07E-03 0.0 0.0 159.3 65.7 0.0 6.8
BGN 633 ENSG00000182492 -5.65 8.87E-05 4.27E-03 20.3 4.5 0.0 0.0 0.0 156.2
BID 637 ENSG00000015475 2.53 4.78E-03 9.74E-02 103.7 66.6 117.1 429.8 42.6 28.3
BIRC3 330 ENSG00000023445 Inf 1.20E-13 5.48E-11 0.0 0.0 48.2 739.4 0.0 0.0
BMP7 655 ENSG00000101144 -1.95 4.86E-03 9.82E-02 301.5 145.7 0.0 3.3 197.9 185.3
BMP8B 656 ENSG00000116985 7.34 3.19E-06 2.74E-04 198.7 244.2 2.0 0.0 0.0 1.0
BMPER 168667 ENSG00000164619 Inf 2.97E-03 6.69E-02 81.9 65.6 0.0 6.6 0.0 0.0
BST2 684 ENSG00000130303 7.02 5.19E-10 1.12E-07 0.0 1.6 263.7 229.0 1.1 0.9
BTBD11 121551 ENSG00000151136 -6.75 7.12E-04 2.27E-02 0.0 0.0 1.3 0.0 45.1 44.3
BTC 685 ENSG00000174808 Inf 7.52E-07 7.76E-05 5.4 35.9 37.9 126.8 0.1 0.0
BTG1 694 ENSG00000133639 5.34 7.29E-08 9.85E-06 223.8 302.7 228.7 24.6 0.0 8.6
C10orf11 83938 ENSG00000148655 3.82 2.80E-05 1.65E-03 195.7 603.7 175.1 152.8 20.5 23.6
C10orf116 10974 ENSG00000148671 6.23 4.44E-06 3.59E-04 0.0 0.0 220.6 8.3 0.0 2.0
C10orf81 79949 ENSG00000148735 Inf 2.49E-06 2.19E-04 67.8 57.9 9.7 115.1 0.0 0.0
C10orf90 118611 ENSG00000154493 Inf 2.12E-04 8.75E-03 53.9 106.7 9.4 0.0 0.0 0.0
C14orf143 90141 ENSG00000140025 Inf 3.76E-06 3.10E-04 3.6 0.0 81.2 93.4 0.0 0.0
C1S 716 ENSG00000182326 3.25 4.97E-04 1.75E-02 10.2 27.1 25.1 440.9 15.2 19.1
C1orf133 574036 ENSG00000203706 2.57 2.45E-03 5.81E-02 8.2 35.9 68.2 358.3 0.0 51.7
C1orf187 374946 ENSG00000162490 3.07 3.73E-03 8.08E-02 152.5 112.4 1.4 142.9 0.0 20.2
C1orf94 84970 ENSG00000142698 Inf 3.01E-04 1.18E-02 9.6 53.0 0.0 46.0 0.0 0.0
C20orf103 24141 ENSG00000125869 5.00 3.62E-03 7.87E-02 8.2 7.9 0.0 89.5 0.0 2.3
C2orf80 389073 ENSG00000188674 4.69 1.55E-03 4.18E-02 264.9 154.4 0.0 1.6 0.0 3.8
C3 718 ENSG00000125730 Inf 2.15E-12 7.99E-10 0.0 0.0 24.4 585.6 0.0 0.0
C3orf58 205428 ENSG00000181744 -2.70 1.58E-03 4.22E-02 41.9 53.1 39.3 95.7 489.1 327.6
C4orf32 132720 ENSG00000174749 5.37 6.36E-04 2.11E-02 0.0 0.0 125.8 0.0 0.0 1.7
C5orf13 9315 ENSG00000134986 -3.29 1.01E-04 4.75E-03 230.2 162.6 55.9 234.9 1985.9 962.7
C5orf38 153571 ENSG00000186493 -5.52 9.66E-04 2.88E-02 2.7 1.5 0.0 0.5 54.6 18.2
C5orf41 153222 ENSG00000164463 4.56 3.81E-03 8.23E-02 29.7 25.6 82.8 0.0 0.0 3.0
C6orf138 442213 ENSG00000178729,ENSG00000244694 -Inf 7.73E-04 2.40E-02 1.4 0.0 0.0 0.0 40.5 13.0
C6orf15 29113 ENSG00000204542 Inf 2.61E-18 1.24E-14 0.0 0.0 2446.3 0.0 0.0 0.0
C7orf16 10842 ENSG00000106341 -Inf 1.64E-03 4.35E-02 0.0 0.0 0.0 0.0 2.3 48.4
C7orf40 285958 ENSG00000232956 4.02 7.73E-04 2.40E-02 39.2 63.9 133.2 0.0 0.0 8.1
C8orf4 56892 ENSG00000176907 4.75 2.23E-08 3.31E-06 12.6 20.3 613.3 2647.7 27.4 53.3
C9orf125 84302 ENSG00000165152 -3.78 8.38E-06 6.10E-04 73.9 59.0 3.6 8.0 508.5 148.1
C9orf64 84267 ENSG00000165118 7.88 1.40E-12 5.34E-10 114.7 156.8 164.1 752.5 2.9 0.0
C9orf95 54981 ENSG00000106733 4.40 3.54E-03 7.75E-02 15.9 15.6 98.1 14.8 3.0 1.0
CA12 771 ENSG00000074410 -4.65 9.71E-07 9.69E-05 45.9 28.1 22.6 30.3 945.6 415.1
CABP7 164633 ENSG00000100314 Inf 1.51E-05 1.02E-03 0.0 6.1 6.5 135.5 0.0 0.0
CACNA1A 773 ENSG00000141837 7.13 1.84E-14 1.10E-11 1.3 4.7 22.1 1252.5 0.0 5.7
CACNA1C 775 ENSG00000151067 -8.15 2.57E-05 1.54E-03 1.4 0.0 0.6 0.0 103.0 15.9
CACNG7 59284 ENSG00000105605 -2.58 4.27E-03 8.99E-02 89.2 41.8 12.6 2.5 149.8 78.7
CACNG8 59283 ENSG00000142408 -4.77 5.28E-07 5.68E-05 33.1 28.3 3.9 2.6 405.1 220.9
CALM1 801 ENSG00000198668 -2.30 3.99E-03 8.49E-02 2159.3 2745.1 824.4 1635.9 14162.8 2936.0
CAMK2B 816 ENSG00000058404 Inf 3.26E-06 2.79E-04 124.2 187.0 5.5 0.0 0.0 0.0
CAPN6 827 ENSG00000077274 -Inf 5.42E-07 5.75E-05 0.0 0.0 0.0 0.0 0.0 199.0
CARD17,CASP1,CARD16 114769,440068,834 ENSG00000137752,ENSG00000204397 5.04 2.68E-05 1.60E-03 4.1 10.9 121.7 166.1 6.0 0.0
CBLC 23624 ENSG00000142273 Inf 1.79E-03 4.67E-02 0.0 0.0 76.2 0.0 0.0 0.0
CCDC129 223075 ENSG00000180347 -Inf 3.70E-07 4.29E-05 0.0 0.0 0.0 0.0 169.0 2.0
CCDC48 79825 ENSG00000114654 -4.17 1.39E-03 3.81E-02 4.1 9.0 0.0 0.0 32.7 79.7
CCDC64 92558 ENSG00000135127 6.38 2.45E-04 9.86E-03 0.0 0.0 38.2 88.5 0.0 1.0
CCKBR 887 ENSG00000110148 4.37 3.25E-03 7.22E-02 50.3 187.2 0.6 0.0 6.1 0.0
CCL2 6347 ENSG00000108691 8.81 4.58E-19 6.80E-15 13.3 0.0 826.6 13567.4 0.0 21.1
236

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
CCL26 10344 ENSG00000006606 Inf 8.09E-11 2.07E-08 0.0 0.0 355.9 78.5 0.0 0.0
CCL7 6354 ENSG00000108688 Inf 1.80E-05 1.16E-03 0.0 0.0 6.9 136.2 0.0 0.0
CCND2 894 ENSG00000118971 -3.98 9.23E-07 9.27E-05 158.0 125.2 0.6 0.0 827.8 496.4
CCNO 10309 ENSG00000152669 Inf 2.09E-06 1.88E-04 2.7 17.2 143.8 22.1 0.0 0.0
CCNY 219771 ENSG00000108100 -2.29 3.93E-03 8.44E-02 265.6 136.3 189.6 199.7 1270.9 448.1
CCR1 1230 ENSG00000163823 Inf 1.75E-03 4.58E-02 33.8 81.5 1.3 0.0 0.0 0.0
CD248 57124 ENSG00000174807 -Inf 1.69E-03 4.45E-02 2.4 0.0 0.0 0.0 4.6 45.7
CD55 1604 ENSG00000196352 6.82 1.95E-05 1.24E-03 0.0 0.0 101.3 71.7 0.0 1.0
CD58 965 ENSG00000116815 3.26 2.05E-03 5.09E-02 40.4 36.8 120.4 165.3 17.5 5.0
CD68 968 ENSG00000129226 3.41 1.53E-03 4.12E-02 4.1 2.4 107.8 180.8 11.8 6.5
CD70 970 ENSG00000125726 Inf 1.78E-16 2.78E-13 0.0 0.0 1301.8 114.4 0.0 0.0
CD74 972 ENSG00000019582 7.11 6.35E-16 7.87E-13 3942.1 2107.8 93.4 1427.2 8.4 9.1
CD9 928 ENSG00000010278 4.87 1.74E-08 2.74E-06 1058.3 1082.4 396.5 189.3 4.8 32.9
CD97 976 ENSG00000123146 3.56 5.05E-04 1.77E-02 78.3 70.3 116.2 101.6 0.0 16.4
CDH13 1012 ENSG00000140945 Inf 2.65E-11 8.17E-09 847.0 494.6 4.8 73.0 0.0 0.0
CDH19 28513 ENSG00000071991 Inf 1.50E-03 4.05E-02 143.8 86.4 0.0 0.0 0.0 0.0
CDH6 1004 ENSG00000113361 -3.05 2.29E-03 5.53E-02 74.7 22.5 3.1 1.0 139.1 4.3
CDHR1 92211 ENSG00000148600 6.68 5.61E-05 2.90E-03 23.9 48.4 46.3 61.5 0.0 1.5
CDKN2A 1029 ENSG00000147889 7.23 1.35E-14 8.49E-12 815.1 848.2 1206.3 0.0 3.0 6.6
CDKN2C 1031 ENSG00000123080 2.55 2.50E-03 5.90E-02 345.8 242.3 258.6 826.0 83.7 67.2
CEBPB 1051 ENSG00000172216 3.29 1.33E-04 5.94E-03 106.7 155.6 552.5 270.9 36.7 30.2
CGREF1 10669 ENSG00000138028 4.77 4.80E-03 9.75E-02 34.5 26.9 20.7 56.6 1.5 1.0
CHCHD10 400916 ENSG00000241579,ENSG00000242131 3.70 1.31E-05 9.02E-04 844.2 713.0 523.9 457.9 28.4 58.8
CHODL 140578 ENSG00000154645 -Inf 4.70E-03 9.64E-02 0.0 0.0 0.0 0.0 24.3 10.4
CHRDL1 91851 ENSG00000101938 3.00 2.10E-03 5.16E-02 1410.3 1413.1 0.7 42.9 75.0 46.4
CILP 8483 ENSG00000138615 -Inf 3.58E-04 1.36E-02 0.0 0.0 0.0 0.0 0.0 71.5
CITED4 163732 ENSG00000179862 6.87 1.58E-05 1.05E-03 0.0 0.0 176.1 2.7 0.0 1.0
CLDN10 9071 ENSG00000134873 Inf 2.43E-04 9.79E-03 25.7 15.6 57.0 27.9 0.0 0.0
CLDN11 5010 ENSG00000013297 5.73 7.86E-11 2.05E-08 2597.6 1809.3 60.7 948.0 0.0 35.3
CLDN3 1365 ENSG00000165215 Inf 4.40E-03 9.18E-02 0.0 0.0 55.8 9.0 0.0 0.0
CMAH 8418 ENSG00000168405 5.82 2.97E-03 6.69E-02 0.0 0.0 3.1 83.2 0.0 1.0
CMPK2 129607 ENSG00000134326 4.21 6.11E-05 3.13E-03 8.1 27.9 53.9 228.5 0.0 11.1
CMTM5 116173 ENSG00000166091 5.35 1.50E-03 4.07E-02 70.1 116.5 0.0 6.6 0.0 2.0
CMTM8 152189 ENSG00000170293 2.76 4.87E-03 9.83E-02 6.8 10.9 217.8 19.4 0.0 24.5
CNKSR2 22866 ENSG00000149970 4.52 1.66E-04 7.15E-03 129.7 154.6 6.8 108.9 3.8 4.4
CNTN1 1272 ENSG00000018236 Inf 5.11E-04 1.77E-02 0.0 0.0 10.5 81.1 0.0 0.0
CNTN6 27255 ENSG00000134115 -Inf 6.87E-06 5.15E-04 0.0 0.0 0.0 0.0 78.3 46.6
COL1A2 1278 ENSG00000164692 -3.11 4.70E-03 9.64E-02 95.9 3.0 642.9 200.5 0.0 4846.9
COL21A1 81578 ENSG00000124749 4.23 8.43E-07 8.58E-05 0.0 0.0 6.7 880.4 0.0 31.3
COL3A1 1281 ENSG00000168542 -6.48 1.64E-09 3.25E-07 7.7 6.5 0.0 33.9 0.8 2365.8
COL4A6 1288 ENSG00000197565 -3.00 2.27E-03 5.50E-02 0.0 21.9 45.6 8.2 232.0 170.8
COL8A1 1295 ENSG00000144810 -3.07 4.73E-04 1.69E-02 249.9 79.2 227.1 1177.6 6989.8 1335.1
CPAMD8 27151 ENSG00000160111 -Inf 1.24E-03 3.53E-02 1.4 0.0 0.0 0.0 32.8 16.2
CPLX2 10814 ENSG00000145920 6.38 2.39E-04 9.74E-03 13.1 18.7 0.0 108.2 0.0 1.1
CPNE2 221184 ENSG00000140848 -2.19 2.61E-03 6.09E-02 487.9 433.9 77.3 138.2 1114.3 863.3
CPNE5 57699 ENSG00000124772 -4.49 4.42E-04 1.61E-02 0.0 0.0 0.0 10.7 156.7 3.3
CRB2 286204 ENSG00000148204 -4.11 2.21E-03 5.38E-02 8.1 3.7 0.6 7.5 64.3 63.5
CREB3L1 90993 ENSG00000157613 -1.99 3.99E-03 8.49E-02 225.2 267.1 119.2 51.6 989.8 173.2
CRIP2 1397 ENSG00000182809 -2.56 1.91E-03 4.86E-02 362.8 95.2 16.9 45.8 277.6 345.0
CRYBB1 1414 ENSG00000100122 -Inf 9.38E-04 2.82E-02 0.0 0.0 0.0 0.0 42.3 8.4
CRYBB2 1415 ENSG00000244752 Inf 1.85E-04 7.86E-03 1.4 0.0 106.5 0.0 0.0 0.0
CRYM 1428 ENSG00000103316 -7.77 5.80E-10 1.23E-07 0.0 0.0 0.6 4.1 19.2 665.5
237

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
CSGALNACT1 55790 ENSG00000147408 7.68 2.19E-07 2.65E-05 359.0 256.3 0.6 54.1 0.0 1.3
CTLA4 1493 ENSG00000163599 -Inf 1.89E-03 4.84E-02 0.0 0.0 0.0 0.0 0.0 49.8
CTNNA2 1496 ENSG00000066032 6.39 2.30E-06 2.05E-04 59.1 95.6 0.0 159.1 0.0 2.2
CTSC 1075 ENSG00000109861 3.55 1.55E-05 1.03E-03 937.2 731.4 311.7 1499.9 10.3 133.2
CXCL1 2919 ENSG00000163739 7.23 2.34E-13 9.96E-11 0.0 0.0 1.2 912.1 0.0 3.8
CXCL12 6387 ENSG00000107562 2.86 3.24E-03 7.22E-02 2.7 0.0 62.2 215.3 0.0 24.7
CXCL14 9547 ENSG00000145824 2.59 7.14E-04 2.27E-02 0.0 0.0 7.0 2853.2 0.0 313.6
CXCL2 2920 ENSG00000081041 Inf 5.32E-11 1.44E-08 0.0 0.0 17.3 446.4 0.0 0.0
CXCL3 2921 ENSG00000163734 7.86 1.87E-16 2.78E-13 0.0 0.0 29.7 2452.6 0.0 7.0
CXCL6 6372 ENSG00000124875 6.13 3.56E-09 6.46E-07 0.0 0.0 5.6 529.8 0.0 5.1
CXXC4 80319 ENSG00000168772 -5.50 1.94E-05 1.24E-03 18.1 5.8 0.0 0.0 78.3 110.7
CXorf38 159013 ENSG00000185753 5.08 2.32E-03 5.57E-02 1.4 10.9 72.7 19.6 0.0 2.0
CYB5R2 51700 ENSG00000166394 6.33 2.99E-04 1.17E-02 9.5 15.6 91.0 16.5 0.0 1.0
DCBLD2 131566 ENSG00000057019 2.43 3.57E-03 7.77E-02 145.0 145.2 612.9 5664.5 263.6 529.0
DCHS1 8642 ENSG00000166341 -3.37 1.38E-03 3.80E-02 91.8 21.1 3.5 10.7 73.8 176.8
DCN 1634 ENSG00000011465 -Inf 1.87E-09 3.65E-07 0.0 0.0 0.0 0.0 0.0 401.7
DDAH2 23564 ENSG00000213722 -3.46 1.51E-03 4.09E-02 128.6 23.3 49.4 2.7 37.0 510.7
DDIT3 1649 ENSG00000175197 4.40 4.13E-07 4.59E-05 933.6 1345.3 748.9 250.5 26.9 47.3
DDIT4L 115265 ENSG00000145358 -6.41 1.15E-03 3.31E-02 1.4 0.0 1.3 0.0 51.5 18.9
DDO 8528 ENSG00000203797 Inf 2.79E-05 1.65E-03 0.0 17.1 29.5 88.4 0.0 0.0
DHRS3 9249 ENSG00000162496 5.20 2.00E-09 3.87E-07 44.5 20.3 592.3 173.7 0.0 14.1
DIAPH2 1730 ENSG00000147202 Inf 1.10E-04 5.09E-03 31.0 57.8 33.9 27.6 0.0 0.0
DKFZp434H1419,AC012513.4 150967 ENSG00000226052 -2.12 4.68E-03 9.64E-02 34.6 71.8 0.0 28.0 239.5 50.3
DKK1 22943 ENSG00000107984 Inf 1.16E-04 5.31E-03 0.0 0.0 46.5 64.9 0.0 0.0
DLK1 8788 ENSG00000185559 -7.95 2.69E-11 8.17E-09 0.0 6.3 0.0 0.0 11.1 1017.5
DNER 92737 ENSG00000187957 -3.89 3.03E-05 1.75E-03 90.5 523.2 9.3 847.4 8865.3 4813.8
DNM3 26052 ENSG00000197959 4.45 4.19E-05 2.28E-03 115.8 128.5 4.6 233.7 6.1 5.0
DOCK10 55619 ENSG00000135905 3.79 2.43E-05 1.48E-03 513.1 652.2 2.2 257.8 4.6 39.3
DOCK5 80005 ENSG00000147459 -4.03 3.76E-04 1.42E-02 0.0 0.0 0.0 21.1 191.6 40.5
DPY19L1 23333 ENSG00000173852 6.17 9.15E-06 6.64E-04 129.0 142.9 86.9 16.3 2.3 0.0
DRD2 1813 ENSG00000149295 8.58 2.39E-10 5.55E-08 0.0 0.0 20.1 414.9 1.1 0.0
DTX4 23220 ENSG00000110042 -5.92 4.13E-07 4.59E-05 6.8 7.8 3.2 0.0 128.8 312.5
DUSP16 80824 ENSG00000111266 4.16 3.36E-03 7.40E-02 62.7 95.2 40.7 0.0 0.0 5.2
DUSP5 1847 ENSG00000138166 3.87 4.66E-05 2.48E-03 130.7 410.4 181.5 18.5 10.4 16.9
DYNC1I1 1780 ENSG00000158560 3.40 1.83E-03 4.73E-02 14.9 60.9 39.5 167.7 6.9 9.9
ECHDC2 55268 ENSG00000121310 Inf 1.20E-05 8.39E-04 29.2 48.6 78.6 24.4 0.0 0.0
EDA2R 60401 ENSG00000131080 -4.02 6.46E-04 2.12E-02 6.8 9.1 4.8 0.0 134.7 20.7
EEF1A2 1917 ENSG00000101210 2.74 3.81E-03 8.23E-02 53.7 53.1 166.2 157.1 15.5 22.4
EEF1D 1936 ENSG00000104529 5.12 3.08E-05 1.77E-03 669.4 175.8 0.0 87.7 0.0 4.9
EFEMP1 2202 ENSG00000115380 2.85 2.27E-04 9.29E-03 4.1 3.1 551.2 613.9 2.4 104.6
EFHD2 79180 ENSG00000142634 2.54 2.77E-03 6.38E-02 275.7 251.5 676.0 84.9 75.7 40.3
EFNA5 1946 ENSG00000184349 -4.73 2.34E-03 5.61E-02 1.4 0.0 4.5 2.5 45.6 75.1
EGFR 1956 ENSG00000146648 2.94 4.40E-03 9.18E-02 49.9 132.3 90.2 48.6 2.3 21.3
ELMO1 9844 ENSG00000155849 4.43 4.57E-06 3.67E-04 642.9 731.1 0.6 128.0 11.4 15.1
ELMO2 63916 ENSG00000062598 -4.56 1.78E-04 7.62E-03 5.5 0.0 11.4 3.3 224.0 4.7
ELMOD1 55531 ENSG00000110675 3.24 2.81E-03 6.45E-02 273.5 272.5 1.9 124.8 21.9 6.1
ELN 2006 ENSG00000049540 4.26 2.36E-06 2.09E-04 915.0 1024.5 2.9 136.1 0.0 40.7
ELOVL2 54898 ENSG00000197977 -2.68 2.04E-04 8.54E-03 176.2 207.2 7.2 89.1 1134.0 167.8
ELTD1 64123 ENSG00000162618 -Inf 1.52E-07 1.92E-05 0.0 0.0 0.0 0.0 184.9 5.0
EML2 24139 ENSG00000125746 4.53 8.68E-05 4.20E-03 137.2 161.5 138.8 7.4 6.9 2.1
EPAS1 2034 ENSG00000116016 -2.84 8.37E-05 4.09E-03 184.1 372.0 51.3 68.8 2186.6 166.7
EPB41L3 23136 ENSG00000082397 2.93 4.86E-04 1.72E-02 243.2 268.9 0.0 487.4 2.3 63.5
238

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
EPDR1 54749 ENSG00000086289 3.09 1.22E-04 5.55E-03 310.9 269.0 739.5 319.0 20.3 82.3
EPHB3 2049 ENSG00000182580 -2.78 1.26E-04 5.69E-03 204.6 268.5 14.4 14.6 902.5 469.4
ERBB4 2066 ENSG00000178568 -4.00 8.10E-05 3.97E-03 8.1 29.7 0.7 5.2 59.4 314.9
ESRRG 2104 ENSG00000196482 Inf 4.35E-05 2.35E-03 36.2 39.4 1.4 86.1 0.0 0.0
ETS1 2113 ENSG00000134954 5.23 1.13E-05 7.98E-04 121.4 201.8 19.4 92.2 1.3 4.0
ETS2 2114 ENSG00000157557 2.83 7.06E-04 2.27E-02 169.8 124.6 235.9 202.8 0.0 52.4
EVC2 132884 ENSG00000173040 Inf 2.87E-03 6.55E-02 8.1 3.1 39.5 28.7 0.0 0.0
F12 2161 ENSG00000131187 4.30 3.11E-05 1.78E-03 181.6 247.2 84.6 66.2 2.3 10.9
F2RL1 2150 ENSG00000164251 4.11 5.83E-04 1.95E-02 0.0 0.0 74.7 129.5 0.8 7.1
FAM126A 84668 ENSG00000122591 2.84 1.16E-03 3.33E-02 1061.8 1338.9 374.2 276.1 82.2 102.7
FAM129A 116496 ENSG00000135842 2.81 1.72E-03 4.52E-02 43.9 71.2 250.9 489.4 64.0 13.1
FAM134B 54463 ENSG00000154153 9.61 1.37E-14 8.49E-12 886.3 1073.7 39.0 78.7 0.0 1.0
FAM150B 285016 ENSG00000189292 -5.14 3.09E-04 1.19E-02 0.0 4.7 0.6 0.0 55.4 69.6
FAM176A 84141 ENSG00000115363 3.37 2.00E-03 5.01E-02 6.0 34.8 47.6 165.8 3.8 12.4
FAM181A 90050 ENSG00000140067 -Inf 2.48E-03 5.86E-02 4.1 0.0 0.0 0.0 0.8 45.6
FAM184B 27146 ENSG00000047662 Inf 1.05E-03 3.11E-02 93.6 82.5 5.5 0.0 0.0 0.0
FAM189A1 23359 ENSG00000104059 -Inf 3.09E-05 1.77E-03 0.0 0.0 0.0 0.0 78.2 18.2
FAM196A,C10orf141 642938 ENSG00000188916 4.89 6.80E-06 5.13E-04 602.4 580.3 1.3 13.8 11.3 2.3
FAM38B2,FAM38B,C18orf58 63895 ENSG00000154864,ENSG00000168738,ENSG00000175388 -Inf 1.16E-13 5.48E-11 0.0 0.0 0.0 0.0 778.6 166.8
FAM55C,NFKBIZ 64332,91775 ENSG00000144802,ENSG00000144815 3.08 4.25E-04 1.58E-02 36.5 91.7 90.7 783.8 48.0 28.1
FAM5B 57795 ENSG00000198797 2.65 2.97E-03 6.69E-02 13.7 34.4 0.6 356.7 0.0 41.3
FAM69A 388650 ENSG00000154511 3.68 3.96E-05 2.18E-03 513.2 393.1 155.0 439.3 27.9 23.3
FAM70A 55026 ENSG00000125355 6.01 1.99E-06 1.84E-04 202.7 114.1 15.7 163.8 0.0 3.5
FAM84A,LOC653602 151354,653602 ENSG00000162981 5.14 4.32E-04 1.59E-02 74.7 110.8 5.4 44.3 0.0 2.7
FBLN2 2199 ENSG00000163520 -Inf 9.47E-10 1.95E-07 0.0 0.0 0.0 0.0 232.1 117.5
FBN2 2201 ENSG00000138829 -4.63 2.08E-04 8.65E-03 2.7 0.0 16.3 8.2 12.2 389.9
FBXO27 126433 ENSG00000161243 5.09 1.64E-05 1.07E-03 0.0 0.0 64.5 195.6 0.0 4.9
FBXO32 114907 ENSG00000156804 5.32 1.07E-10 2.71E-08 30.4 88.9 572.3 1235.4 17.5 14.1
FCGR2B,FCGR2C,FCGR2A 2212,2213,9103 ENSG00000072694,ENSG00000143226,ENSG00000244682 6.26 1.33E-05 9.12E-04 338.7 219.8 13.2 0.0 0.0 2.2
FCRLA 84824 ENSG00000132185 Inf 2.46E-15 2.61E-12 2134.8 1514.3 6.1 0.0 0.0 0.0
FERMT3 83706 ENSG00000149781 4.18 2.53E-03 5.95E-02 2.5 5.2 13.9 119.9 0.0 5.0
FEZF2 55079 ENSG00000153266 -Inf 1.45E-07 1.86E-05 0.0 0.0 0.0 0.0 29.7 196.7
FGF19 9965 ENSG00000162344 -Inf 3.78E-07 4.32E-05 26.0 0.0 0.0 0.0 0.0 208.3
FGFBP2 83888 ENSG00000137441 8.40 6.75E-14 3.72E-11 1699.4 939.1 0.6 84.3 0.0 2.0
FGFR1 2260 ENSG00000077782 -1.90 4.51E-03 9.34E-02 1177.0 1288.2 147.2 311.0 3067.1 1294.6
FMNL1 752 ENSG00000184922 3.74 4.24E-03 8.94E-02 0.0 0.0 47.7 95.0 0.0 6.6
FOXG1 2290 ENSG00000176165 3.30 1.57E-04 6.83E-03 445.0 505.5 104.5 137.4 0.0 50.8
FOXJ1 2302 ENSG00000129654 -4.19 7.16E-05 3.60E-03 4.9 3.1 4.4 27.8 254.8 175.8
FOXQ1 94234 ENSG00000164379 -Inf 4.65E-04 1.66E-02 0.0 0.0 0.0 0.0 49.7 9.1
FUT8 2530 ENSG00000033170 3.32 9.74E-05 4.63E-03 666.0 999.8 609.6 2352.7 95.8 168.6
FXYD1 5348 ENSG00000221857 7.79 8.28E-12 2.86E-09 269.6 673.5 0.0 0.0 0.0 2.0
FXYD3 5349 ENSG00000089356 Inf 6.70E-06 5.08E-04 110.9 180.1 0.0 0.0 0.0 0.0
FXYD5 53827 ENSG00000089327 3.06 3.46E-04 1.32E-02 45.6 1.6 489.3 0.0 1.9 37.7
FXYD7 53822 ENSG00000221946 Inf 1.11E-16 2.06E-13 1650.8 1205.4 6.9 705.9 0.0 0.0
FZD3 7976 ENSG00000104290 -2.63 2.29E-03 5.53E-02 98.9 57.5 11.2 52.7 240.7 262.9
GABBR2 9568 ENSG00000136928 -9.50 5.68E-17 1.21E-13 0.0 0.0 5.6 0.0 2569.7 143.1
GABRA5 2558 ENSG00000186297 -6.00 1.45E-03 3.99E-02 8.1 1.7 0.0 0.0 0.9 65.5
GABRQ 55879 ENSG00000147402 Inf 8.54E-05 4.15E-03 87.7 126.9 0.0 0.0 0.0 0.0
GAL 51083 ENSG00000069482 Inf 1.01E-03 3.00E-02 0.0 0.0 0.0 82.6 0.0 0.0
GALNT5 11227 ENSG00000136542 4.71 5.17E-04 1.78E-02 8.1 32.5 121.1 5.7 0.0 3.8
GALR1 2587 ENSG00000166573 Inf 7.97E-04 2.45E-02 86.2 96.5 0.0 0.0 0.0 0.0
GAS7 8522 ENSG00000007237 Inf 6.98E-08 9.52E-06 294.5 288.3 0.0 3.6 0.0 0.0
239

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
GBP1 2633 ENSG00000117228 3.90 2.24E-04 9.22E-03 1.5 0.0 64.1 266.1 7.6 6.9
GBP2 2634 ENSG00000162645 4.06 6.94E-06 5.18E-04 0.0 4.7 468.3 654.3 41.3 4.0
GBP3 2635 ENSG00000117226 3.47 1.53E-04 6.69E-03 0.0 0.0 49.3 720.9 37.5 9.1
GBP4 115361 ENSG00000162654 Inf 3.54E-04 1.35E-02 7.5 0.0 15.3 82.2 0.0 0.0
GBP5 115362 ENSG00000154451 Inf 3.88E-09 6.87E-07 0.0 0.0 52.6 276.2 0.0 0.0
GCNT1 2650 ENSG00000187210 Inf 4.77E-03 9.74E-02 5.5 1.6 15.0 46.2 0.0 0.0
GDF15 9518 ENSG00000130513 4.37 3.15E-07 3.72E-05 288.9 1402.4 1358.2 108.3 22.8 69.9
GDPD2 54857 ENSG00000130055 -3.12 7.12E-04 2.27E-02 32.7 37.2 0.0 4.2 133.1 108.5
GEM 2669 ENSG00000164949 4.64 1.68E-06 1.57E-04 57.7 61.0 37.7 610.9 15.7 3.0
GFAP 2670 ENSG00000131095 5.34 1.43E-06 1.37E-04 468.5 336.4 0.0 94.7 0.0 7.2
GFPT2 9945 ENSG00000131459 2.25 4.84E-03 9.82E-02 24.3 15.6 247.6 243.2 0.0 70.6
GGH 8836 ENSG00000137563 2.45 4.70E-03 9.64E-02 899.6 886.6 7220.3 4880.1 475.4 1096.6
GJA1 2697 ENSG00000152661 -3.49 1.89E-04 8.00E-03 334.7 130.2 764.6 116.4 6363.9 1206.7
GJB2 2706 ENSG00000165474 2.80 1.20E-03 3.42E-02 1.4 0.0 18.4 448.1 0.0 44.0
GJC3 349149 ENSG00000176402 Inf 1.50E-03 4.05E-02 31.1 86.2 0.0 0.0 0.0 0.0
GLYATL2 219970 ENSG00000156689 -3.99 2.64E-03 6.15E-02 39.2 9.5 0.0 0.0 25.8 73.6
GMPR 2766 ENSG00000137198 3.72 2.86E-03 6.54E-02 108.1 140.1 33.5 35.5 4.5 6.3
GNG11 2791 ENSG00000127920 -2.33 3.49E-03 7.64E-02 393.2 316.9 345.3 1039.6 4802.2 908.6
GPC3 2719 ENSG00000147257 -5.79 1.24E-06 1.19E-04 0.0 0.0 35.1 0.0 11.6 1268.7
GPNMB 10457 ENSG00000136235 6.74 1.68E-11 5.42E-09 7688.5 16473.1 127.8 224.6 82.2 23.2
GPR158 57512 ENSG00000151025 -4.55 7.39E-08 9.89E-06 103.8 117.4 0.0 53.3 2311.0 354.9
GPR98 84059 ENSG00000164199 -3.99 2.40E-04 9.74E-03 30.5 20.0 4.4 0.0 76.9 183.8
GRB14 2888 ENSG00000115290 5.96 1.66E-03 4.39E-02 31.2 30.3 4.6 60.7 0.0 1.2
GRIA1 2890 ENSG00000155511 -4.42 3.62E-07 4.24E-05 33.1 68.7 2.7 73.4 1780.2 292.7
GRIA3 2892 ENSG00000125675 7.17 7.45E-05 3.71E-03 85.8 129.9 1.4 32.7 0.8 0.0
GRIK3 2899 ENSG00000163873 5.24 3.84E-07 4.35E-05 580.1 631.6 0.0 0.0 0.0 11.1
GRM3 2913 ENSG00000198822 -Inf 4.70E-08 6.47E-06 0.0 0.0 0.0 0.0 162.4 66.0
GYG2 8908 ENSG00000056998 -2.89 5.71E-04 1.93E-02 52.6 51.3 3.8 0.8 223.4 55.4
H1F0 3005 ENSG00000189060 3.59 9.61E-04 2.87E-02 93.7 46.8 89.0 103.7 0.0 13.5
HAND2 9464 ENSG00000164107 Inf 1.11E-03 3.22E-02 75.0 84.1 1.0 2.9 0.0 0.0
HAPLN1 1404 ENSG00000145681 6.40 4.55E-07 4.98E-05 521.8 383.7 0.0 0.0 0.0 3.0
HCP5 10866 ENSG00000206337 Inf 8.81E-09 1.49E-06 0.0 0.0 180.4 114.3 0.0 0.0
HEPACAM 220296 ENSG00000165478 Inf 3.91E-03 8.39E-02 82.3 65.6 0.0 3.0 0.0 0.0
HIF3A 64344 ENSG00000124440 -Inf 1.27E-03 3.56E-02 0.0 0.0 0.0 0.0 15.2 36.3
HLA-A 3105 ENSG00000206503 3.04 3.02E-04 1.18E-02 294.8 469.0 840.5 535.7 83.8 65.7
HLA-DMA 3108 ENSG00000204257 Inf 8.98E-07 9.08E-05 383.9 196.7 13.7 7.4 0.0 0.0
HLA-DPA1 3113 ENSG00000231389 5.73 5.76E-11 1.53E-08 1554.1 1033.9 49.5 243.3 4.6 12.4
HLA-DPB1 3115 ENSG00000223865 3.83 4.38E-04 1.61E-02 1007.2 359.8 0.6 12.3 2.4 15.1
HLA-DQA1 3117 ENSG00000196735 Inf 1.25E-03 3.53E-02 198.2 89.6 0.0 0.0 0.0 0.0
HLA-DQA2 3118 ENSG00000237541 6.56 5.45E-07 5.75E-05 0.0 0.0 8.1 280.8 0.0 2.4
HLA-DQB1 3119 ENSG00000179344 4.60 3.26E-05 1.84E-03 348.2 347.0 16.8 31.7 3.4 7.4
HLA-DRA 3122 ENSG00000204287 5.68 3.67E-09 6.58E-07 9744.0 5434.9 138.4 270.5 57.4 19.2
HLA-DRB3,HLA-DRB1,HLA- 3123,3125,3126 ENSG00000196126 Inf 2.04E-10 4.89E-08 1105.3 381.8 3.1 87.8 0.0 0.0
240
DRB4
HLA-DRB5 3127 ENSG00000198502 8.36 1.38E-13 6.04E-11 2885.9 995.1 0.0 0.0 0.0 2.0
HMGA2 8091 ENSG00000149948 -5.68 3.13E-07 3.72E-05 0.0 0.0 0.0 23.7 233.3 576.3
HOMER1 9456 ENSG00000152413 6.06 4.09E-05 2.24E-03 94.6 173.2 28.9 0.0 0.0 2.0
HOXA1 3198 ENSG00000105991 Inf 2.21E-03 5.39E-02 13.5 10.9 12.5 48.3 0.0 0.0
HOXA10 3206 ENSG00000153807 Inf 8.63E-14 4.42E-11 479.4 631.7 19.5 273.5 0.0 0.0
HOXA4 3201 ENSG00000197576 Inf 4.31E-11 1.26E-08 63.1 100.0 2.5 382.0 0.0 0.0
HOXA5 3202 ENSG00000106004 Inf 3.20E-03 7.17E-02 22.7 33.6 2.5 33.1 0.0 0.0
HOXA7 3204 ENSG00000122592 Inf 5.79E-04 1.94E-02 51.7 102.1 0.7 0.0 0.0 0.0

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
HOXB6 3216 ENSG00000108511 Inf 1.08E-03 3.16E-02 0.0 0.0 23.8 59.8 0.0 0.0
HOXB7 3217 ENSG00000120087 Inf 2.51E-14 1.43E-11 0.0 0.0 795.3 109.6 0.0 0.0
HOXB9 3219 ENSG00000170689 Inf 2.50E-10 5.72E-08 0.0 0.0 338.7 55.6 0.0 0.0
HOXC10 3226 ENSG00000180818 Inf 1.74E-06 1.62E-04 90.5 160.7 47.1 0.0 0.0 0.0
HOXC13 3229 ENSG00000123364 Inf 2.48E-03 5.86E-02 3.5 76.4 0.5 0.0 0.0 0.0
HOXD10 3236 ENSG00000128710 Inf 7.25E-14 3.85E-11 77.0 145.9 115.3 578.2 0.0 0.0
HOXD13 3239 ENSG00000128714 Inf 2.33E-05 1.44E-03 27.2 45.4 0.0 93.3 0.0 0.0
HOXD3 3232 ENSG00000128652 Inf 4.05E-03 8.61E-02 49.4 40.5 1.5 25.7 0.0 0.0
HOXD9 3235 ENSG00000128709 Inf 7.80E-04 2.42E-02 49.7 61.6 23.2 7.4 0.0 0.0
HPRT1 3251 ENSG00000165704 2.62 6.73E-04 2.18E-02 28.4 82.7 614.7 246.9 4.3 96.9
HPSE 10855 ENSG00000173083 6.28 4.17E-04 1.56E-02 10.8 16.0 18.2 84.8 0.0 1.0
HRCT1 646962 ENSG00000196196 6.40 1.67E-03 4.42E-02 0.0 0.0 89.8 6.6 0.8 0.0
HTATIP2 10553 ENSG00000109854 7.51 1.09E-14 7.73E-12 38.0 50.0 3.8 1331.9 0.0 4.7
ICAM1 3383 ENSG00000090339 2.75 7.47E-04 2.35E-02 138.3 272.9 100.5 848.2 28.1 92.5
ID4 3400 ENSG00000172201 -2.33 4.89E-04 1.73E-02 1342.6 1447.2 17.6 161.5 4504.8 958.5
IFI27 3429 ENSG00000165949 3.86 2.55E-05 1.54E-03 11.8 7.4 446.9 38.2 2.3 20.5
IFI30 10437 ENSG00000216490 4.27 2.84E-06 2.47E-04 25.7 40.6 470.2 277.2 16.3 11.1
IFI6 2537 ENSG00000126709 3.36 2.69E-05 1.60E-03 110.0 57.7 2294.7 801.8 29.0 175.0
IFITM2 10581 ENSG00000185201 -4.89 3.84E-05 2.13E-03 0.0 0.0 31.2 8.3 0.0 775.8
IFITM8P ENSG00000215096 -Inf 2.27E-03 5.50E-02 0.0 0.0 0.0 0.0 0.6 46.3
IGF2,INS,INS-
3481,3630,723961 ENSG00000129965,ENSG00000167244,ENSG00000240801 -7.34 3.37E-08 4.77E-06 0.0 0.0 0.6 4.5 2.8 505.9
241
IGF2,AC132217.2
IGFBP5 3488 ENSG00000115461 -2.23 4.78E-03 9.74E-02 840.8 936.7 30.1 1081.5 3234.4 3148.3
IGLON5 402665 ENSG00000142549 5.38 1.58E-07 1.98E-05 750.3 630.9 0.0 0.0 0.0 10.1
IGSF11 152404 ENSG00000144847 3.49 7.56E-04 2.37E-02 453.7 549.5 0.0 0.0 14.7 18.3
IGSF3 3321 ENSG00000143061 3.78 5.11E-04 1.77E-02 301.4 260.7 41.9 45.7 6.9 10.0
IKBKE 9641 ENSG00000143466 -4.25 2.18E-03 5.35E-02 7.4 2.3 2.1 3.3 82.9 22.2
IL13RA2 3598 ENSG00000123496 5.66 2.22E-11 7.02E-09 0.0 0.0 451.0 628.7 0.0 14.3
IL17RD 54756 ENSG00000144730 -4.38 1.14E-05 8.03E-04 25.5 17.5 9.8 15.3 364.6 230.4
IL1B 3553 ENSG00000125538 Inf 1.93E-08 2.96E-06 0.0 0.0 115.5 175.1 0.0 0.0
IL1R1 3554 ENSG00000115594 2.49 1.56E-03 4.18E-02 25.7 15.0 67.6 977.0 19.4 104.7
IL1RAPL1 11141 ENSG00000169306 5.30 1.20E-03 3.42E-02 81.1 75.0 1.4 43.5 0.0 2.3
IL33 90865 ENSG00000137033 -Inf 1.60E-03 4.26E-02 0.0 0.0 0.0 0.0 0.0 51.2
IL4I1 259307 ENSG00000104951 6.01 1.36E-03 3.77E-02 2.7 2.1 12.6 84.2 0.0 1.0
IL6 3569 ENSG00000136244 Inf 7.89E-07 8.09E-05 44.6 25.3 110.4 64.6 0.0 0.0
IL8 3576 ENSG00000169429 7.68 4.39E-08 6.10E-06 0.0 0.0 26.4 286.9 0.0 1.5
INHBA 3624 ENSG00000122641 7.38 1.07E-06 1.05E-04 86.5 172.1 66.0 16.5 0.0 1.0
INHBB 3625 ENSG00000163083 5.89 2.05E-06 1.87E-04 240.6 361.5 0.0 0.0 0.0 4.1
INPP5D 3635 ENSG00000168918 -3.50 4.59E-03 9.47E-02 6.8 10.9 3.8 0.0 61.6 49.5
INSIG1 3638 ENSG00000186480 2.68 1.89E-03 4.84E-02 780.8 1475.9 1531.6 3387.9 282.2 383.3
IRAK1 3654 ENSG00000184216 2.95 4.63E-04 1.66E-02 292.6 279.4 764.0 375.0 75.1 46.1
IRX1 79192 ENSG00000170549 -3.75 4.11E-03 8.72E-02 30.5 11.4 0.0 0.0 5.1 92.7
IRX2 153572 ENSG00000170561 -4.65 1.01E-05 7.21E-04 0.0 0.0 0.0 28.9 370.0 111.1
ISYNA1 51477 ENSG00000105655 -2.54 3.24E-03 7.22E-02 308.0 123.3 127.2 149.7 668.8 883.1
ITGA4 3676 ENSG00000115232 2.99 9.76E-04 2.91E-02 70.2 63.0 11.9 594.6 35.7 20.2
ITGBL1 9358 ENSG00000198542 Inf 1.12E-03 3.24E-02 0.0 0.0 33.4 48.5 0.0 0.0
ITIH5 80760 ENSG00000123243 5.87 3.70E-10 8.21E-08 1336.6 1302.0 2.8 1.4 3.6 10.7
ITM2A 9452 ENSG00000078596 -Inf 4.47E-05 2.40E-03 0.0 0.0 0.0 0.0 0.0 105.8
JAM2 58494 ENSG00000154721 -2.11 3.94E-03 8.44E-02 196.8 310.6 12.5 199.1 780.0 726.7
JPH1 56704 ENSG00000104369 6.70 5.47E-05 2.84E-03 36.3 77.3 31.3 50.7 0.0 1.0
KALRN 8997 ENSG00000160145 -5.25 2.09E-06 1.88E-04 0.0 6.3 2.5 7.1 302.9 106.2
KANK1 23189 ENSG00000107104 2.97 8.79E-04 2.67E-02 374.9 583.2 79.6 135.3 14.5 53.9

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
KCNA2 3737 ENSG00000177301 Inf 4.29E-04 1.58E-02 149.2 103.9 0.0 2.4 0.0 0.0
KCNJ12 3768 ENSG00000184185 Inf 4.57E-05 2.44E-03 60.5 139.1 0.0 0.0 0.0 0.0
KCNK1 3775 ENSG00000135750 5.95 1.82E-03 4.72E-02 0.0 0.0 90.8 3.3 0.0 1.0
KCNK12 56660 ENSG00000184261 -Inf 1.37E-03 3.79E-02 0.0 0.0 0.0 0.0 39.5 6.8
KCNMB2 10242 ENSG00000197584 5.32 1.85E-05 1.19E-03 2.6 0.0 38.9 204.5 0.0 4.5
KCTD12 115207 ENSG00000178695 -4.06 2.20E-06 1.97E-04 105.3 71.6 25.5 58.7 1598.8 150.0
KIAA1217 56243 ENSG00000120549 4.49 1.62E-03 4.29E-02 35.7 52.3 81.5 4.0 0.0 3.6
KLF6 1316 ENSG00000067082 4.56 5.25E-04 1.80E-02 58.2 113.0 67.2 0.0 0.0 5.4
L3MBTL4 91133 ENSG00000154655 Inf 2.11E-10 4.97E-08 58.0 67.9 11.6 341.6 0.0 0.0
LAMA2 3908 ENSG00000196569 -5.21 3.06E-08 4.42E-06 47.3 25.2 8.0 2.6 730.2 149.0
LAMA4 3910 ENSG00000112769 2.81 2.23E-03 5.43E-02 885.8 984.7 31.7 116.1 38.0 69.6
LEMD1 93273 ENSG00000186007 2.85 3.79E-04 1.43E-02 0.0 6.3 630.6 4.1 0.0 58.9
LFNG 3955 ENSG00000106003 -2.21 2.21E-03 5.38E-02 66.1 96.7 6.9 49.2 429.5 42.4
LGALS3 3958 ENSG00000131981 2.80 1.27E-03 3.56E-02 2898.0 3802.5 3936.6 643.9 219.8 578.3
LHFPL3 375612 ENSG00000187416 Inf 2.25E-09 4.29E-07 485.5 402.6 0.0 0.0 0.0 0.0
LIF 3976 ENSG00000128342 3.97 5.11E-06 3.99E-04 25.7 28.0 98.9 891.5 15.9 27.3
LIFR 3977 ENSG00000113594 -2.95 1.06E-03 3.15E-02 181.0 124.8 38.0 727.7 1803.6 2775.9
LINGO1 84894 ENSG00000169783 5.52 9.07E-04 2.74E-02 113.3 139.7 0.0 0.0 0.0 2.0
LMO2 4005 ENSG00000135363 -4.82 2.89E-06 2.50E-04 20.3 21.2 0.0 4.9 134.6 369.0
LMO3 55885 ENSG00000048540 -Inf 3.43E-11 1.02E-08 0.0 0.0 0.0 0.0 136.9 446.0
LMO4 8543 ENSG00000143013 3.80 6.59E-06 5.03E-04 553.8 1311.2 455.5 1149.8 16.7 122.5
LNX1 84708 ENSG00000072201 4.67 2.34E-03 5.61E-02 2.7 3.1 12.6 101.0 0.0 3.0
LOC283070,CAMK1D 283070,57118 ENSG00000183049 -2.40 5.40E-04 1.84E-02 109.8 142.2 0.6 35.0 578.0 52.7
LOXL4 84171 ENSG00000138131 4.94 1.10E-04 5.09E-03 186.4 184.3 32.9 15.6 0.0 4.7
LPAR6 10161 ENSG00000139679 -3.62 1.24E-04 5.61E-03 29.8 33.8 39.6 0.0 395.9 211.3
LPHN2 23266 ENSG00000117114 -2.05 2.04E-03 5.09E-02 268.8 205.3 10.0 12.7 446.7 183.5
LRAT 9227 ENSG00000121207 5.42 6.40E-06 4.93E-04 0.0 0.0 1.9 290.9 1.5 3.0
LRP4 4038 ENSG00000134569 -2.68 2.41E-03 5.73E-02 29.6 38.9 5.6 18.6 187.5 86.0
LRRC2 79442 ENSG00000163827 Inf 4.14E-06 3.38E-04 0.0 0.0 169.1 0.0 0.0 0.0
LRRC55 219527 ENSG00000183908 -Inf 1.89E-03 4.84E-02 0.0 0.0 0.0 0.0 0.0 49.6
LRRN2 10446 ENSG00000170382 -4.91 1.54E-04 6.71E-03 7.8 5.5 0.0 0.0 97.9 27.3
LTF 4057 ENSG00000012223 Inf 2.03E-07 2.47E-05 0.0 0.0 222.4 0.0 0.0 0.0
LUM 4060 ENSG00000139329 -4.26 6.32E-05 3.23E-03 0.0 0.0 110.9 0.0 507.4 897.7
LXN 56925 ENSG00000079257 -5.76 8.55E-05 4.15E-03 0.0 0.0 0.0 3.7 139.4 9.0
LY96 23643 ENSG00000154589 7.53 1.62E-07 2.01E-05 3.4 59.4 106.5 115.6 0.0 1.0
LYST 1130 ENSG00000143669 5.22 1.71E-07 2.10E-05 85.4 117.1 79.3 454.5 10.3 1.0
MACROD2 140733 ENSG00000172264 -3.07 2.85E-03 6.53E-02 20.6 21.8 0.0 0.0 90.0 33.4
MAF 4094 ENSG00000178573 -4.79 1.41E-05 9.55E-04 14.9 7.1 16.0 0.0 282.4 150.4
MAL 4118 ENSG00000172005 6.24 2.59E-10 5.83E-08 22.8 23.3 11.9 655.2 0.0 6.0
MALL 7851 ENSG00000144063 4.75 3.34E-08 4.77E-06 0.0 0.0 0.0 1117.4 5.4 22.2
MAN1C1 57134 ENSG00000117643 3.59 3.15E-05 1.79E-03 556.5 478.6 107.6 378.3 7.6 45.2
MAP3K5 4217 ENSG00000197442 5.06 1.46E-05 9.83E-04 43.3 28.1 110.6 178.0 5.0 1.0
MAP6 4135 ENSG00000171533 -3.25 1.78E-04 7.62E-03 2.3 32.9 0.6 111.4 629.4 288.0
MAP7 9053 ENSG00000135525 4.45 6.64E-04 2.17E-02 0.0 1.6 25.8 139.8 0.0 5.0
MAPT 4137 ENSG00000186868 4.55 8.34E-06 6.10E-04 353.1 572.9 3.3 34.4 5.3 12.2
MARCH1 55016 ENSG00000145416 6.53 1.27E-04 5.70E-03 18.9 37.1 61.6 41.7 0.0 1.1
MARS 4141 ENSG00000166986 2.80 1.53E-03 4.12E-02 1181.8 771.8 272.0 1.3 39.5 60.2
MARVELD3 91862 ENSG00000140832 -Inf 3.77E-04 1.42E-02 0.3 0.0 0.0 0.0 57.1 3.3
MATN2 4147 ENSG00000132561 3.44 4.46E-04 1.62E-02 326.2 720.9 35.9 6.6 22.8 24.1
MBP 4155 ENSG00000197971 6.47 7.66E-12 2.71E-09 118.6 402.5 103.3 303.0 0.0 6.3
MEIS2 4212 ENSG00000134138 3.05 6.37E-04 2.11E-02 69.3 85.9 36.6 331.3 0.0 36.3
MEST 4232 ENSG00000106484 -2.13 3.82E-03 8.25E-02 1932.6 895.3 66.5 334.9 1665.0 2122.4
242

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
MET 4233 ENSG00000105976 9.09 1.36E-18 1.01E-14 1.4 0.0 43.8 9851.7 12.2 0.0
MFAP2 4237 ENSG00000117122 -3.26 2.73E-03 6.32E-02 12.0 6.2 124.4 55.3 3.4 1174.2
MGC87042 256227 ENSG00000105889 Inf 4.14E-03 8.76E-02 65.1 67.1 1.3 0.0 0.0 0.0
MGLL 11343 ENSG00000074416 5.05 1.96E-04 8.25E-03 191.4 165.8 30.8 17.4 2.6 2.0
MIA 8190 ENSG00000213054 5.64 5.51E-04 1.87E-02 21.5 151.9 0.0 0.0 0.0 2.0
MICA 4276 ENSG00000204520 3.69 1.56E-05 1.03E-03 0.0 3.1 1713.4 98.0 65.6 28.8
MICB 4277 ENSG00000204516 4.20 4.73E-03 9.68E-02 0.0 0.0 27.2 85.6 0.0 4.0
MID1IP1 58526 ENSG00000165175 6.08 8.98E-04 2.72E-02 12.1 1.6 101.3 0.0 0.0 1.0
MIPOL1 145282 ENSG00000151338 -Inf 7.14E-04 2.27E-02 0.0 0.0 0.0 0.0 6.3 54.5
MKI67 4288 ENSG00000148773 -Inf 2.65E-04 1.06E-02 0.0 0.0 0.0 0.0 63.7 0.0
MLC1 23209 ENSG00000100427 -2.52 6.39E-04 2.11E-02 207.8 109.7 22.5 12.3 495.8 59.5
MLPH 79083 ENSG00000115648 2.92 2.65E-03 6.15E-02 0.0 0.0 27.7 267.6 1.5 24.7
MMP17 4326 ENSG00000198598 3.58 1.55E-05 1.03E-03 1173.5 864.3 422.6 340.8 5.1 84.8
MMP7 4316 ENSG00000137673 5.74 1.01E-04 4.75E-03 0.0 0.0 17.5 146.3 0.0 2.3
MMRN1 22915 ENSG00000138722 -8.80 1.01E-08 1.65E-06 0.0 0.0 1.3 0.0 279.4 93.2
MN1 4330 ENSG00000169184 -7.23 1.51E-09 3.03E-07 21.6 4.2 0.6 0.0 146.8 384.1
MOCOS 55034 ENSG00000075643 4.47 3.85E-06 3.16E-04 14.1 75.8 201.7 195.0 0.0 14.0
MOSC2 54996 ENSG00000117791 4.40 1.96E-04 8.25E-03 39.5 69.7 77.7 77.9 0.0 7.1
MT1A 4489 ENSG00000205362 Inf 7.71E-04 2.40E-02 1.8 17.4 41.7 25.4 0.0 0.0
MT2A 4502 ENSG00000125148 4.13 2.03E-05 1.28E-03 2039.2 1924.4 8308.0 5237.5 285.0 301.4
MTTP 4547 ENSG00000138823 -3.36 8.32E-04 2.54E-02 6.5 6.3 7.2 27.0 249.5 25.2
MX1 4599 ENSG00000157601 7.05 5.50E-06 4.28E-04 45.1 82.6 86.3 31.8 0.0 0.9
MXRA5 25878 ENSG00000101825 -3.18 1.27E-03 3.56E-02 46.1 23.4 0.0 23.0 123.2 158.0
MYBL1 4603 ENSG00000185697 3.44 4.68E-05 2.49E-03 87.8 81.2 162.7 765.9 17.4 44.4
MYC 4609 ENSG00000136997 3.95 7.75E-06 5.72E-04 155.1 341.8 478.0 32.0 11.4 24.8
MYH3 4621 ENSG00000109063 -4.54 2.06E-05 1.30E-03 18.9 17.2 16.5 51.9 11.8 1298.9
MYL1 4632 ENSG00000168530 -Inf 1.75E-08 2.74E-06 0.0 0.0 0.0 0.0 6.6 310.5
MYL9 10398 ENSG00000101335 -3.85 8.93E-05 4.28E-03 3.8 3.1 110.0 8.5 796.2 370.3
MYO1B 4430 ENSG00000128641 -3.03 1.10E-03 3.19E-02 13.5 0.0 55.3 111.5 514.9 387.9
NAMPT 10135 ENSG00000105835 2.92 5.09E-04 1.77E-02 499.3 678.3 484.4 1970.5 117.5 158.3
NBL1 4681 ENSG00000158747 2.84 7.32E-04 2.32E-02 584.5 491.1 763.1 2068.2 156.4 151.9
NCALD 83988 ENSG00000104490 -2.50 4.45E-03 9.25E-02 53.2 51.5 0.0 4.9 86.7 126.4
NCAM1 4684 ENSG00000149294 3.13 8.63E-04 2.63E-02 3171.6 3747.4 36.2 1733.8 232.8 187.5
NDE1 54820 ENSG00000072864 3.42 4.84E-04 1.72E-02 195.9 200.2 200.6 102.0 24.5 6.9
NDN 4692 ENSG00000182636 -3.46 4.28E-05 2.32E-03 105.3 84.2 0.0 0.0 214.4 405.4
NEFM 4741 ENSG00000104722 -Inf 1.35E-10 3.34E-08 0.0 0.0 0.0 0.0 0.0 536.3
NELL2 4753 ENSG00000184613 -3.24 2.99E-04 1.17E-02 117.2 174.9 0.7 378.4 1070.5 2418.9
NFASC 23114 ENSG00000163531 5.94 2.04E-03 5.08E-02 16.7 25.2 11.9 56.8 0.0 1.0
NFATC2 4773 ENSG00000101096 Inf 1.18E-03 3.37E-02 74.3 83.5 4.4 0.0 0.0 0.0
NFE2L3 9603 ENSG00000050344 3.89 1.43E-04 6.37E-03 43.2 51.1 49.9 215.7 0.0 14.3
NFIA 4774 ENSG00000162599 2.95 3.06E-03 6.87E-02 220.0 429.2 10.7 9.8 1.5 37.4
NID1 4811 ENSG00000116962 2.90 4.46E-04 1.62E-02 954.4 470.8 328.9 2577.4 96.8 204.4
NKX2-1 7080 ENSG00000136352 -5.52 1.01E-08 1.65E-06 4.1 17.2 0.0 0.0 425.3 103.6
NKX6-2 84504 ENSG00000148826 -Inf 1.47E-03 4.02E-02 0.0 0.0 0.0 0.0 0.0 52.2
NLGN4X 57502 ENSG00000146938 3.12 4.96E-04 1.75E-02 570.9 700.9 8.1 162.0 8.4 58.2
NMU 10874 ENSG00000109255 4.32 5.32E-04 1.82E-02 8.1 20.7 153.8 8.9 0.0 6.1
NNAT 4826 ENSG00000053438 -Inf 4.86E-09 8.40E-07 0.0 0.0 0.0 0.0 0.0 363.2
NNMT 4837 ENSG00000166741 3.74 6.59E-06 5.03E-04 519.9 172.4 918.1 175.6 15.2 47.4
NOP16 51491 ENSG00000048162 2.60 1.91E-03 4.86E-02 305.8 448.6 493.2 198.3 57.0 67.7
NOTCH3 4854 ENSG00000074181 -4.24 5.14E-04 1.78E-02 0.0 0.0 0.0 25.1 0.0 317.9
NOV 4856 ENSG00000136999 Inf 4.28E-06 3.47E-04 265.6 174.9 9.8 0.0 0.0 0.0
NPFFR2 10886 ENSG00000056291 -Inf 9.86E-09 1.65E-06 0.0 0.0 0.0 0.0 250.3 5.0
243

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
NPTX2 4885 ENSG00000106236 -6.47 1.11E-08 1.78E-06 0.0 0.0 6.9 4.8 422.0 272.4
NR0B1 190 ENSG00000169297 Inf 6.79E-15 5.60E-12 0.0 0.0 0.0 1039.8 0.0 0.0
NR1D1 9572 ENSG00000126368 2.86 1.07E-03 3.15E-02 80.9 136.8 469.2 220.4 51.7 24.2
NR4A2 4929 ENSG00000153234 Inf 4.54E-04 1.64E-02 86.9 79.7 4.4 14.6 0.0 0.0
NRBF2 29982 ENSG00000148572 -Inf 2.85E-04 1.12E-02 0.0 0.0 0.0 0.0 12.9 59.5
NRG1 3084 ENSG00000157168 Inf 2.27E-04 9.29E-03 2.7 9.7 60.4 30.0 0.0 0.0
NRN1 51299 ENSG00000124785 Inf 6.59E-09 1.13E-06 317.7 235.8 49.5 48.5 0.0 0.0
NRP2 8828 ENSG00000118257 2.86 4.26E-04 1.58E-02 33.7 60.1 169.0 901.9 26.6 76.3
NTN1 9423 ENSG00000065320 -4.22 1.58E-06 1.48E-04 95.8 239.9 22.4 250.6 5458.0 966.2
NTRK2 4915 ENSG00000148053 3.61 4.89E-05 2.57E-03 90.7 163.9 174.8 219.5 0.0 30.3
NXPH1 30010 ENSG00000122584 4.57 4.70E-05 2.49E-03 325.0 414.1 0.0 26.6 8.4 4.0
OAS1 4938 ENSG00000089127 6.47 1.46E-04 6.41E-03 0.0 0.0 53.9 81.2 0.0 1.4
OAS2 4939 ENSG00000111335 Inf 2.18E-05 1.36E-03 0.0 0.0 73.4 65.7 0.0 0.0
OAS3 4940 ENSG00000111331 3.88 1.48E-03 4.03E-02 2.7 6.5 55.6 162.4 9.1 1.0
OASL 8638 ENSG00000135114 Inf 2.20E-08 3.30E-06 1.2 7.8 102.4 180.2 0.0 0.0
ODZ2 57451 ENSG00000145934 -3.68 1.49E-04 6.51E-03 27.0 25.0 2.5 36.0 212.8 329.5
OGN 4969 ENSG00000106809 -Inf 3.58E-04 1.36E-02 0.0 0.0 0.0 0.0 0.0 71.6
OLFM1 10439 ENSG00000130558 4.72 2.31E-04 9.44E-03 200.9 227.7 11.3 0.0 0.0 6.0
OTOR 56914 ENSG00000125879 9.92 3.31E-15 3.28E-12 1169.3 1474.1 0.0 0.0 0.0 1.0
OTX2 5015 ENSG00000165588 -Inf 6.59E-04 2.16E-02 0.0 0.0 0.0 0.0 40.1 14.8
OXTR 5021 ENSG00000180914 3.67 1.99E-03 5.00E-02 28.0 111.2 36.0 94.6 7.6 4.6
P2RX7 5027 ENSG00000089041 7.72 1.34E-11 4.42E-09 691.5 611.8 25.5 2.2 0.0 2.0
PARP12 64761 ENSG00000059378 2.92 1.36E-03 3.77E-02 18.2 21.9 196.2 300.2 24.5 21.2
PARP3 10039 ENSG00000041880 4.12 3.46E-05 1.93E-03 5.4 25.0 145.0 267.2 7.7 9.3
PAX8 7849 ENSG00000125618 -2.80 1.78E-03 4.66E-02 0.0 0.0 40.1 66.0 438.3 53.5
PCDH20 64881 ENSG00000197991 6.16 1.08E-05 7.67E-04 65.1 7.8 1.3 208.6 0.0 2.0
PCDHB12 56124 ENSG00000120328 -7.05 2.05E-03 5.09E-02 2.7 0.0 0.7 0.0 31.9 23.1
PCDHB3 56132 ENSG00000113205 5.39 1.08E-03 3.16E-02 90.5 107.6 0.0 35.2 2.3 0.0
PCDHB4 56131 ENSG00000081818 5.88 2.01E-06 1.85E-04 337.8 351.5 5.6 0.0 0.0 4.0
PCOLCE2 26577 ENSG00000163710 5.77 4.31E-03 9.05E-02 0.0 0.0 3.7 79.4 0.0 1.3
PDE1C 5137 ENSG00000154678 4.13 1.38E-06 1.32E-04 578.2 430.3 1117.8 981.4 45.7 50.8
PDE4B 5142 ENSG00000184588 -5.35 2.42E-03 5.74E-02 9.5 1.6 0.6 0.0 42.5 17.2
PDGFA 5154 ENSG00000197461 3.58 1.02E-04 4.80E-03 1540.7 1178.9 41.4 32.8 20.8 49.3
PDGFRA 5156 ENSG00000134853 7.42 1.03E-14 7.67E-12 2309.4 2706.9 66.4 355.6 0.0 12.1
PDGFRB 5159 ENSG00000113721 -3.44 1.87E-03 4.81E-02 34.3 18.8 29.9 3.3 58.0 316.0
PDPN 10630 ENSG00000162493 4.68 4.08E-08 5.72E-06 1470.9 1283.2 107.6 2041.3 12.9 75.3
PDZRN3 23024 ENSG00000121440 -4.40 1.38E-05 9.38E-04 18.4 22.1 12.6 3.0 278.2 251.3
PEA15 8682 ENSG00000162734 -2.44 2.01E-03 5.04E-02 1189.8 1377.8 590.8 1060.2 9727.1 1297.6
PERP 64065 ENSG00000112378 3.83 4.28E-04 1.58E-02 1.4 0.0 200.3 108.2 11.4 3.3
PHACTR3 116154 ENSG00000087495 Inf 2.70E-04 1.07E-02 44.0 117.1 0.0 0.0 0.0 0.0
PHLPP1 23239 ENSG00000081913 -2.64 8.54E-04 2.61E-02 469.3 331.2 166.3 100.8 1557.1 924.7
PI15 51050 ENSG00000137558 -6.45 4.25E-10 9.29E-08 20.2 3.3 2.5 5.5 397.8 262.2
PIGA 5277 ENSG00000165195 3.43 1.79E-03 4.67E-02 13.9 10.3 182.7 97.2 16.1 2.0
PION 54103 ENSG00000186088 Inf 7.59E-05 3.75E-03 7.3 18.7 13.2 87.4 0.0 0.0
PITPNC1 26207 ENSG00000154217 3.72 5.25E-04 1.80E-02 293.0 215.2 39.5 54.9 1.7 13.7
PITX1 5307 ENSG00000069011 6.80 7.33E-13 2.94E-10 1.4 3.1 480.6 531.7 6.2 0.0
PKNOX2 63876 ENSG00000165495 -2.77 3.36E-03 7.40E-02 62.9 32.3 1.5 1.9 109.0 53.4
PLA2G2A 5320 ENSG00000188257 Inf 3.34E-18 1.24E-14 1227.8 4419.1 0.6 6.3 0.0 0.0
PLA2G4A 5321 ENSG00000116711 7.34 1.68E-15 1.92E-12 909.7 1226.5 483.7 755.9 0.0 10.1
PLAC9 219348 ENSG00000189129 2.81 2.96E-03 6.69E-02 182.5 578.2 0.6 18.7 0.0 56.6
PLCB1 23236 ENSG00000182621 -2.27 2.87E-03 6.55E-02 103.0 73.1 8.4 53.8 366.7 70.9
PLCH1 23007 ENSG00000114805 -3.41 4.41E-05 2.38E-03 152.0 134.0 83.3 118.9 2002.9 384.3
244

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
PLD3 23646 ENSG00000105223 -3.31 2.42E-04 9.76E-03 228.5 33.1 87.5 9.8 700.5 162.0
PLEKHA6 22874 ENSG00000143850 -5.40 3.21E-03 7.18E-02 6.8 0.0 0.6 1.6 39.9 24.7
PLS1 5357 ENSG00000120756 2.94 9.39E-04 2.82E-02 67.5 149.4 218.6 344.8 36.5 25.3
PLS3 5358 ENSG00000102024 3.83 3.67E-06 3.06E-04 360.2 776.6 476.6 367.2 0.0 75.6
PLSCR1 5359 ENSG00000188313 2.45 4.58E-03 9.47E-02 37.0 137.7 307.6 533.6 81.2 38.1
PMEPA1 56937 ENSG00000124225 4.36 4.99E-07 5.41E-05 240.8 123.3 621.8 297.2 11.7 22.1
PMP2 5375 ENSG00000147588 Inf 9.08E-15 7.10E-12 948.1 1312.7 0.0 0.0 0.0 0.0
PNMA2 10687 ENSG00000240694 -2.28 1.93E-03 4.90E-02 155.7 183.7 8.8 368.1 1517.6 305.5
PNP 4860 ENSG00000198805 3.92 3.87E-03 8.32E-02 36.5 20.7 115.6 8.3 2.4 3.7
PPAP2C 8612 ENSG00000141934 7.37 4.02E-07 4.52E-05 21.6 6.3 242.9 3.1 0.0 1.2
PPCS 79717 ENSG00000127125 2.94 3.03E-04 1.18E-02 57.1 106.2 771.9 214.9 32.7 61.6
PPEF1 5475 ENSG00000086717 4.44 6.87E-04 2.21E-02 12.2 165.1 9.6 23.0 0.0 6.4
PPHLN1 51535 ENSG00000134283 3.56 7.59E-04 2.37E-02 98.5 165.8 119.1 6.6 2.6 14.1
PPL 5493 ENSG00000118898 2.45 4.84E-03 9.82E-02 2.7 1.6 90.7 358.0 11.2 43.8
PPM1K 152926 ENSG00000163644 5.15 1.93E-03 4.90E-02 16.2 28.1 56.3 23.6 0.0 2.4
PPP4R1 9989 ENSG00000154845 -2.16 2.30E-03 5.54E-02 168.6 215.1 42.9 218.4 1239.4 180.3
PRIMA1 145270 ENSG00000175785 6.84 2.97E-09 5.51E-07 865.4 465.5 0.0 54.9 0.0 3.0
PROCR 10544 ENSG00000101000 4.09 1.87E-03 4.81E-02 5.5 0.0 181.8 0.0 6.3 1.5
PRSS12 8492 ENSG00000164099 3.66 1.95E-03 4.92E-02 128.0 49.8 55.6 87.1 0.0 10.1
PSORS1C1 170679 ENSG00000204540 Inf 1.11E-07 1.43E-05 0.0 0.0 212.3 29.0 0.0 0.0
PSPH 5723 ENSG00000146733 3.45 1.87E-03 4.81E-02 40.0 60.0 168.5 1.9 4.0 10.2
PTEN 5728 ENSG00000171862 -5.00 5.70E-05 2.94E-03 2.7 0.0 14.8 0.0 191.7 126.9
PTGS1 5742 ENSG00000095303 -3.91 4.31E-03 9.05E-02 12.2 6.8 1.6 0.0 61.9 20.2
PTPRD 5789 ENSG00000153707 -3.18 2.97E-03 6.69E-02 2.8 20.1 7.0 0.0 101.0 63.7
PTPRH 5794 ENSG00000080031 2.67 3.36E-03 7.40E-02 13.5 2.0 162.7 371.3 35.8 19.7
PTPRR 5801 ENSG00000153233 5.05 2.78E-03 6.42E-02 0.0 0.0 12.8 88.6 0.0 2.0
PXDN 7837 ENSG00000130508 3.01 6.69E-04 2.18E-02 1025.9 1801.8 163.8 146.9 41.9 132.8
PYGL 5836 ENSG00000100504 3.71 6.76E-05 3.41E-03 569.0 837.9 179.4 74.3 36.2 19.4
RAB11FIP1 80223 ENSG00000156675 -3.57 4.79E-03 9.75E-02 19.1 6.9 19.0 0.0 56.8 140.1
RAB38 23682 ENSG00000123892 3.17 2.79E-04 1.10E-02 137.3 221.7 53.7 963.8 58.6 33.4
RAB6B 51560 ENSG00000154917 -3.75 8.86E-05 4.27E-03 36.5 22.8 10.7 8.2 303.4 76.3
RAB7L1 8934 ENSG00000117280 6.23 5.37E-04 1.83E-02 82.1 62.9 52.1 0.0 0.0 1.0
RAD50 10111 ENSG00000113522 3.49 3.57E-03 7.77E-02 25.7 49.4 70.3 84.2 6.3 5.7
RADIL 55698 ENSG00000157927 3.00 4.70E-03 9.64E-02 113.9 92.3 13.5 137.8 0.0 20.3
RARRES1 5918 ENSG00000118849 6.00 5.06E-11 1.42E-08 5.6 0.0 16.6 1153.4 6.1 6.3
RARRES3 5920 ENSG00000133321 3.53 2.08E-04 8.65E-03 64.8 26.5 64.7 471.8 24.3 8.1
RASGEF1C 255426 ENSG00000146090 4.10 3.55E-03 7.75E-02 471.5 143.6 2.8 9.8 0.0 6.0
RASGRP3 25780 ENSG00000152689 Inf 2.81E-03 6.45E-02 0.0 5.2 26.1 38.5 0.0 0.0
REC8 9985 ENSG00000100918 -2.64 1.11E-03 3.22E-02 99.8 91.8 23.9 125.9 633.6 373.3
RGAG4 340526 ENSG00000242732 -3.26 7.97E-04 2.45E-02 21.3 18.6 28.2 19.8 267.2 157.6
RGL3 57139 ENSG00000205517 Inf 3.90E-04 1.47E-02 10.2 4.7 61.0 30.3 0.0 0.0
RGS17 26575 ENSG00000091844 4.76 7.40E-04 2.33E-02 84.3 107.6 5.6 50.7 0.0 4.2
RGS20 8601 ENSG00000147509 -3.16 4.95E-03 9.97E-02 0.0 0.0 7.1 25.6 136.0 56.5
RGS5 8490 ENSG00000143248 -4.82 3.40E-06 2.87E-04 4.1 13.5 6.9 3.0 343.5 112.5
RHBDF2 79651 ENSG00000129667 3.41 3.12E-05 1.78E-03 9.5 18.8 214.9 956.3 10.7 63.6
RHOBTB3 22836 ENSG00000164292 2.77 2.60E-03 6.09E-02 3292.5 2727.3 146.6 130.6 90.0 203.9
RHPN1 114822 ENSG00000158106 -4.18 2.88E-03 6.56E-02 21.7 5.9 0.0 0.0 40.3 35.3
RIPK4 54101 ENSG00000183421 6.16 6.76E-04 2.19E-02 2.4 0.0 50.8 58.0 0.0 1.0
RNASEH1 246243 ENSG00000171865 2.43 4.37E-03 9.14E-02 68.9 165.9 213.0 440.5 48.5 52.6
RNF175 285533 ENSG00000145428 6.81 5.25E-05 2.74E-03 225.4 165.0 5.1 0.0 0.0 1.1
RP11-473I1.1 ENSG00000220793 2.89 5.13E-04 1.78E-02 73.0 177.2 882.0 3.2 48.9 46.0
RP11-93B14.2,hCG_2018279 100127888 ENSG00000232803 Inf 3.11E-03 6.99E-02 41.3 72.1 1.3 0.0 0.0 0.0
245

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
RP5-955M13.1,KCNG1 3755 ENSG00000026559,ENSG00000242964 3.17 4.85E-03 9.82E-02 31.5 43.7 73.1 75.4 0.0 14.1
RP9 6100 ENSG00000164610 2.59 3.34E-03 7.40E-02 569.1 650.0 335.5 82.3 64.7 53.8
RPSAP52 204010 ENSG00000241749 4.40 2.37E-03 5.66E-02 0.0 0.0 12.2 116.6 0.0 4.2
RRAD 6236 ENSG00000166592 Inf 2.54E-09 4.79E-07 0.0 0.0 84.0 253.0 0.0 0.0
RRP7A 27341 ENSG00000189306 3.00 2.73E-03 6.32E-02 78.1 35.6 263.9 2.0 9.9 15.1
RRS1 23212 ENSG00000179041 4.41 8.14E-04 2.50E-02 72.2 64.0 98.5 0.0 0.0 5.5
RTN1 6252 ENSG00000139970 -5.34 5.31E-07 5.68E-05 8.1 5.5 0.6 19.3 385.7 286.3
RTP4 64108 ENSG00000136514 Inf 7.77E-06 5.72E-04 0.0 0.7 98.1 58.2 0.0 0.0
S100A6 6277 ENSG00000197956 6.04 2.05E-17 5.08E-14 2259.3 324.2 5709.9 512.7 12.3 53.4
S100B 6285 ENSG00000160307 -2.18 5.35E-04 1.83E-02 3679.1 1302.1 6.3 66.5 3663.5 496.5
S1PR3 1903 ENSG00000213694 -4.14 4.96E-03 9.98E-02 29.7 6.3 0.0 0.0 4.6 69.1
SALL1 6299 ENSG00000103449 -2.39 2.66E-03 6.19E-02 126.2 81.2 0.0 112.8 409.7 267.7
SALL2 6297 ENSG00000165821 -4.13 2.94E-05 1.71E-03 72.9 38.6 16.5 11.6 194.2 581.9
SCN11A 11280 ENSG00000168356 -Inf 3.20E-03 7.17E-02 0.0 0.0 0.0 0.0 0.0 43.4
SCN1B 6324 ENSG00000105711 4.36 3.42E-06 2.87E-04 654.7 1438.9 141.0 25.4 31.9 20.2
SCN5A 6331 ENSG00000183873 -Inf 4.18E-03 8.84E-02 0.0 0.0 0.0 0.0 0.0 40.4
SDC2 6383 ENSG00000169439 -3.28 7.59E-04 2.37E-02 18.2 11.0 14.3 47.7 253.9 216.2
SDK2 54549 ENSG00000069188 -3.29 3.28E-03 7.27E-02 44.4 23.9 0.0 4.1 0.0 178.5
SELENBP1 8991 ENSG00000143416 -3.91 7.83E-04 2.42E-02 25.5 5.2 20.1 6.7 110.1 203.2
SEMA3D 223117 ENSG00000153993 -6.30 5.75E-04 1.94E-02 0.0 2.3 0.0 0.0 0.0 82.0
SEMA4G 57715 ENSG00000095539 -3.43 4.43E-03 9.22E-02 29.7 13.2 4.0 0.0 69.2 47.4
SEMA6A 57556 ENSG00000092421 -3.48 6.56E-05 3.32E-03 68.1 50.0 3.1 0.0 257.1 140.0
SERPINE2 5270 ENSG00000135919 3.55 1.21E-04 5.51E-03 1392.0 2436.4 127.5 513.6 92.8 83.1
SEZ6L 23544 ENSG00000100095 6.22 1.07E-03 3.15E-02 103.7 109.6 0.0 4.0 0.0 1.0
SFRP1 6422 ENSG00000104332 -2.10 7.59E-04 2.37E-02 91.7 535.5 0.0 12.3 1178.7 396.2
SFTA3 253970 ENSG00000229415 -Inf 9.82E-07 9.73E-05 0.0 0.0 0.0 0.0 125.0 32.3
SGCD 6444 ENSG00000170624 4.73 1.20E-05 8.39E-04 172.6 443.4 0.0 0.0 0.0 11.0
SH3BGR 6450 ENSG00000185437 3.88 4.80E-06 3.79E-04 47.5 40.0 989.7 672.0 39.6 37.6
SHC4 399694 ENSG00000185634 2.94 4.92E-03 9.93E-02 50.1 30.0 184.2 63.3 12.9 11.1
SHISA2 387914 ENSG00000180730 -Inf 1.15E-03 3.31E-02 0.0 0.0 0.0 0.0 0.0 55.9
SHOX2 6474 ENSG00000168779 Inf 8.30E-04 2.54E-02 41.9 23.4 64.0 0.0 0.0 0.0
SHROOM3 57619 ENSG00000138771 -3.30 3.19E-04 1.23E-02 95.0 29.6 72.1 132.7 807.8 726.4
SIX3 6496 ENSG00000138083 -5.50 1.88E-08 2.91E-06 2.7 26.1 0.0 0.0 203.9 594.6
SIX6 4990 ENSG00000184302 -10.35 4.28E-15 3.74E-12 1.4 1.6 0.0 0.0 1346.6 17.9
SKAP2 8935 ENSG00000005020 4.09 4.19E-05 2.28E-03 52.5 58.8 26.6 360.6 8.4 8.9
SLC15A2 6565 ENSG00000163406 -6.36 1.23E-05 8.53E-04 1.4 1.2 1.9 0.0 150.4 38.3
SLC16A3 9123 ENSG00000141526 Inf 1.99E-03 5.01E-02 0.0 0.0 21.9 52.7 0.0 0.0
SLC26A2 1836 ENSG00000155850 2.94 2.54E-03 5.96E-02 590.4 1552.1 12.7 95.7 96.6 48.5
SLC2A5 6518 ENSG00000142583 Inf 7.62E-05 3.75E-03 1.7 3.1 4.5 109.7 0.0 0.0
SLC38A1 81539 ENSG00000111371 3.22 3.45E-04 1.32E-02 150.8 124.7 320.3 98.7 14.5 24.5
SLC38A5 92745 ENSG00000017483 Inf 8.90E-04 2.70E-02 63.4 20.6 61.7 4.0 0.0 0.0
SLC46A1 113235 ENSG00000076351 Inf 3.44E-03 7.54E-02 12.1 25.0 10.0 34.8 0.0 0.0
SLC4A11 83959 ENSG00000088836 Inf 1.75E-05 1.14E-03 15.5 41.5 87.7 11.6 0.0 0.0
SLC4A4 8671 ENSG00000080493 -4.49 5.85E-07 6.12E-05 36.0 38.9 13.1 18.8 925.2 145.3
SLCO1C1 53919 ENSG00000139155 -Inf 3.19E-09 5.86E-07 7.7 0.0 0.0 0.0 166.2 151.8
SLCO2A1 6578 ENSG00000174640 -5.61 7.54E-08 1.00E-05 38.8 15.5 0.0 0.0 213.0 297.0
SLITRK5 26050 ENSG00000165300 -4.56 7.21E-04 2.29E-02 13.7 0.0 16.3 0.0 46.4 207.6
SMAGP 57228 ENSG00000170545 3.32 4.27E-04 1.58E-02 17.6 11.6 176.0 226.3 8.5 19.2
SMOC2 64094 ENSG00000112562 -Inf 2.12E-04 8.75E-03 2.7 0.0 0.0 0.0 0.0 79.9
SNAP25 6616 ENSG00000132639 3.34 1.14E-04 5.26E-03 306.2 396.7 79.9 513.8 20.5 44.6
SNCAIP 9627 ENSG00000064692 -5.66 3.64E-03 7.90E-02 4.4 1.6 0.0 0.0 2.8 49.4
SNHG12 85028 ENSG00000197989 2.75 1.36E-03 3.77E-02 29.7 29.7 357.1 235.4 25.1 36.6
246

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
SNX10 29887 ENSG00000086300 4.39 2.53E-03 5.95E-02 82.6 39.4 24.4 63.7 0.0 4.3
SNX22 79856 ENSG00000157734 7.52 1.08E-06 1.05E-04 162.4 276.1 1.9 0.0 0.0 1.0
SOD2 6648 ENSG00000112096 3.20 1.06E-04 4.94E-03 45.8 51.4 187.6 1755.6 53.2 90.9
SOD3 6649 ENSG00000109610 Inf 1.66E-10 4.05E-08 112.8 3.1 401.2 0.8 0.0 0.0
SORCS2 57537 ENSG00000184985 -9.04 6.02E-13 2.49E-10 1.5 1.6 0.6 0.0 690.7 77.4
SORCS3 22986 ENSG00000156395 Inf 1.63E-04 7.07E-03 82.4 81.1 2.6 30.0 0.0 0.0
SORL1 6653 ENSG00000137642 3.93 3.16E-04 1.22E-02 586.9 293.8 1.4 52.7 3.0 12.1
SOX10 6663 ENSG00000100146 Inf 1.63E-17 4.85E-14 2350.8 3216.5 0.0 0.0 0.0 0.0
SOX3 6658 ENSG00000134595 3.74 1.67E-04 7.18E-03 2.5 3.1 256.8 75.8 1.8 15.2
SPAG4 6676 ENSG00000061656 5.25 1.07E-03 3.15E-02 6.8 4.7 81.5 30.5 0.0 2.4
SPARCL1 8404 ENSG00000152583 -4.03 2.40E-05 1.48E-03 44.2 39.0 4.8 282.9 1448.1 2097.6
SPINK2 6691 ENSG00000128040 Inf 9.14E-04 2.76E-02 0.0 3.8 79.9 0.0 0.0 0.0
SPINT1 6692 ENSG00000166145 4.31 1.01E-06 9.90E-05 0.0 0.0 1009.1 7.2 19.9 13.7
SPP1 6696 ENSG00000118785 3.58 1.10E-04 5.10E-03 75.6 194.5 2549.4 7850.3 335.4 254.9
SPRED1 161742 ENSG00000166068 -3.32 6.80E-04 2.20E-02 15.4 17.7 6.9 24.1 217.2 101.8
SPTBN5 51332 ENSG00000137877 -5.93 3.39E-05 1.90E-03 0.0 0.0 5.0 0.0 151.8 52.0
SQRDL 58472 ENSG00000137767 3.58 1.07E-03 3.15E-02 31.7 90.8 183.0 52.3 18.4 0.0
SRGAP3 9901 ENSG00000196220 -3.35 5.76E-04 1.94E-02 21.5 14.4 0.0 38.0 214.2 138.7
SRP9 6726 ENSG00000143742 -2.29 2.21E-03 5.38E-02 443.5 726.1 195.4 161.4 1982.7 1557.9
SRPX 8406 ENSG00000101955 2.86 1.23E-03 3.49E-02 1070.0 1473.9 771.8 4757.1 321.8 319.2
ST6GAL1 6480 ENSG00000073849 4.93 4.81E-04 1.71E-02 184.6 171.7 0.7 12.0 0.0 4.0
ST6GALNAC3 256435 ENSG00000184005 -4.73 1.69E-03 4.46E-02 0.0 4.7 0.0 0.0 23.6 59.7
ST6GALNAC5 81849 ENSG00000117069 -8.25 1.01E-11 3.42E-09 8.9 0.0 0.0 5.8 261.3 899.5
STAMBPL1 57559 ENSG00000138134 3.81 7.34E-05 3.66E-03 0.0 0.0 86.7 334.9 3.5 15.9
STC2 8614 ENSG00000113739 3.31 1.18E-03 3.37E-02 0.0 0.0 230.2 26.8 0.0 17.2
STEAP1 26872 ENSG00000164647 6.19 2.40E-07 2.87E-05 6.2 35.9 219.6 104.7 2.5 1.4
STEAP2 261729 ENSG00000157214 3.94 2.25E-05 1.40E-03 0.0 0.0 319.9 414.3 25.9 6.1
STRA6 64220 ENSG00000137868 -5.09 6.26E-07 6.51E-05 0.0 38.8 7.5 9.8 16.7 1255.8
STX3 6809 ENSG00000166900 Inf 5.56E-04 1.88E-02 4.1 0.0 72.6 19.5 0.0 0.0
STXBP5L 9515 ENSG00000145087 -Inf 1.60E-05 1.05E-03 0.0 0.0 0.0 0.0 103.4 0.0
SULF1 23213 ENSG00000137573 -3.79 3.40E-03 7.47E-02 0.0 11.3 0.0 0.0 13.7 86.9
SULF2 55959 ENSG00000196562 3.31 1.15E-04 5.28E-03 749.0 891.0 253.5 581.6 43.3 72.4
SYNGR1 9145 ENSG00000100321 2.48 2.40E-03 5.73E-02 127.2 130.6 487.6 414.2 52.8 71.1
SYNM 23336 ENSG00000182253 -4.23 6.56E-05 3.32E-03 1.4 0.0 0.0 36.4 238.8 212.8
SYT1 6857 ENSG00000067715 -2.46 6.89E-04 2.22E-02 87.5 110.3 0.0 16.1 374.6 93.7
SYTL4 94121 ENSG00000102362 3.14 2.93E-03 6.67E-02 37.0 54.7 71.1 142.0 5.9 14.1
SYTL5 94122 ENSG00000147041 3.34 1.88E-03 4.82E-02 0.0 0.0 14.8 277.9 12.2 7.3
TAGLN 6876 ENSG00000149591 -7.06 8.97E-14 4.44E-11 1.9 3.1 9.4 4.1 1089.6 384.4
TAPBPL 55080 ENSG00000139192 2.57 3.86E-03 8.32E-02 123.3 118.8 114.3 780.6 93.5 20.3
TBC1D8 11138 ENSG00000204634 5.89 5.01E-05 2.62E-03 13.8 15.7 73.7 90.9 0.0 2.4
TCEAL2 140597 ENSG00000184905 Inf 2.51E-08 3.69E-06 1.4 3.3 27.2 248.9 0.0 0.0
TCF7 6932 ENSG00000081059 3.00 6.69E-04 2.18E-02 59.5 70.8 87.7 368.1 8.4 35.1
TCF7L2 6934 ENSG00000148737 -2.60 3.98E-03 8.49E-02 33.8 34.3 69.4 42.3 343.9 248.4
TECRL 253017 ENSG00000205678 Inf 3.34E-06 2.83E-04 256.3 195.2 0.0 0.0 0.0 0.0
TES 26136 ENSG00000135269 -Inf 6.77E-12 2.45E-09 0.0 0.0 0.0 0.0 433.0 184.2
TESC 54997 ENSG00000088992 -3.42 2.19E-03 5.38E-02 21.6 13.6 0.0 0.0 92.8 8.6
TF 7018 ENSG00000091513 3.70 4.33E-03 9.07E-02 140.3 205.9 0.0 14.5 5.5 6.1
TFAP2A 7020 ENSG00000137203 3.71 4.55E-04 1.64E-02 476.7 357.8 42.5 104.2 26.0 0.0
TFCP2 7024 ENSG00000135457 5.47 6.46E-04 2.12E-02 47.7 99.6 20.3 14.7 0.0 2.0
TGFA 7039 ENSG00000163235 4.55 3.44E-05 1.92E-03 216.7 272.7 99.7 54.3 12.1 0.0
TGM2 7052 ENSG00000198959 3.93 9.46E-06 6.83E-04 0.0 62.7 290.3 418.6 11.5 22.5
THBS2 7058 ENSG00000186340 3.55 2.42E-05 1.48E-03 237.0 73.4 39.0 1655.3 38.9 61.5
247

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
THY1 7070 ENSG00000154096 3.29 2.41E-04 9.76E-03 2994.5 2506.5 150.3 2389.9 150.0 195.0
TM4SF1 4071 ENSG00000169908 5.56 2.08E-08 3.16E-06 117.3 242.3 308.7 4.9 1.0 7.4
TMCO4 255104 ENSG00000162542 5.51 1.45E-04 6.41E-03 2.7 1.6 42.3 129.9 1.2 1.0
TMEM100 55273 ENSG00000166292 4.02 1.48E-03 4.03E-02 186.9 150.7 0.0 47.0 0.0 8.1
TMEM132D 121256 ENSG00000151952 -Inf 1.01E-04 4.75E-03 0.0 0.0 0.0 0.0 57.0 23.2
TMEM158 25907 2.67 1.85E-03 4.77E-02 535.5 466.7 154.3 1586.1 140.7 91.0
TMEM176B 28959 ENSG00000106565 6.03 2.07E-03 5.11E-02 20.3 93.6 0.0 5.5 0.0 1.0
TMEM200B 399474 ENSG00000187975 4.13 9.61E-04 2.87E-02 2.7 1.6 58.5 119.9 0.9 6.5
TMEM38A 79041 ENSG00000072954 3.09 1.80E-03 4.67E-02 205.9 608.9 70.7 5.1 33.3 20.2
TMEM71 137835 ENSG00000165071 -3.20 2.42E-03 5.74E-02 1.4 12.6 20.2 1.0 194.7 10.1
TMSB15A 11013 ENSG00000158164 -2.98 1.29E-03 3.62E-02 0.0 0.0 171.6 265.9 1121.9 1173.9
TMSL1 ENSG00000223551 -2.64 4.07E-03 8.65E-02 25.6 9.8 177.7 36.9 595.0 336.6
TNC 3371 ENSG00000041982 2.99 3.08E-04 1.19E-02 215.9 188.2 31.9 1597.5 59.8 91.8
TNFAIP2 7127 ENSG00000185215 6.89 5.19E-11 1.43E-08 0.0 0.0 28.1 513.3 0.0 3.0
TNFAIP3 7128 ENSG00000118503 2.86 3.34E-03 7.40E-02 16.2 17.2 83.3 177.3 0.0 25.7
TNFAIP6 7130 ENSG00000123610 4.95 8.74E-08 1.14E-05 621.2 458.0 6.8 288.5 0.0 15.8
TNFRSF14 8764 ENSG00000157873 4.01 2.55E-04 1.02E-02 12.7 7.9 67.4 260.7 12.9 1.0
TNFSF4 7292 ENSG00000117586 3.98 3.69E-06 3.06E-04 35.1 21.4 55.2 2851.5 58.6 64.6
TNNC2 7125 ENSG00000101470 -Inf 1.46E-04 6.41E-03 0.0 0.0 0.0 0.0 0.0 85.9
TNNI1 7135 ENSG00000159173 -Inf 2.55E-05 1.54E-03 0.0 0.0 0.0 0.0 0.0 116.0
TNNI2 7136 ENSG00000130598 -Inf 2.84E-05 1.67E-03 0.0 0.0 0.0 0.0 0.0 114.0
TOX 9760 ENSG00000198846 3.31 1.07E-03 3.15E-02 197.4 221.1 5.6 134.9 3.2 21.2
TPM1 7168 ENSG00000140416 -2.94 1.95E-03 4.92E-02 85.7 78.5 269.4 215.4 955.6 1911.5
TPMT 7172 ENSG00000137364 2.85 1.09E-03 3.17E-02 179.6 367.0 563.7 1733.5 198.9 47.9
TRAF1 7185 ENSG00000056558 3.92 2.00E-03 5.01E-02 17.9 36.3 19.5 145.7 6.8 2.0
TRAM1L1 133022 ENSG00000174599 -5.13 4.75E-05 2.50E-03 0.0 0.0 0.0 11.3 121.7 146.3
TRIB2 28951 ENSG00000071575 -2.05 1.95E-03 4.92E-02 434.9 352.2 34.9 55.8 919.1 307.4
TRIB3 57761 ENSG00000101255 3.42 5.71E-05 2.94E-03 83.9 109.6 4203.7 837.1 158.2 161.8
TRIM14 9830 ENSG00000106785 3.38 2.02E-03 5.05E-02 54.4 63.7 73.3 155.4 13.5 5.3
TRIM47 91107 ENSG00000132481 3.88 7.17E-06 5.33E-04 106.8 49.3 265.0 527.1 7.6 29.9
TRIM48 79097 ENSG00000150244 Inf 1.84E-04 7.84E-03 21.6 124.2 0.0 0.0 0.0 0.0
TRPM8 79054 ENSG00000144481 4.14 2.35E-05 1.45E-03 156.8 802.9 10.6 75.3 20.5 13.1
TSHZ3 57616 ENSG00000121297 4.45 2.11E-05 1.32E-03 136.6 142.5 38.5 184.1 0.0 11.1
TSLP 85480 ENSG00000145777 5.71 1.24E-04 5.63E-03 4.1 4.6 155.0 0.0 0.0 2.0
TSPAN13 27075 ENSG00000106537 2.62 1.25E-03 3.53E-02 126.1 234.5 127.9 748.4 28.1 91.9
TSPAN7 7102 ENSG00000156298 3.70 6.53E-05 3.32E-03 737.7 849.2 1.3 1.7 0.0 43.7
TSPAN9 10867 ENSG00000011105 2.72 2.51E-03 5.92E-02 78.4 90.4 99.0 4588.1 416.6 68.8
TSTD1 100131187 ENSG00000215845 4.79 1.38E-03 3.80E-02 0.0 0.0 127.5 0.0 0.0 2.7
TTF2 8458 ENSG00000116830 3.05 6.65E-04 2.17E-02 55.3 112.4 65.1 472.7 25.1 27.3
TTN 7273 ENSG00000155657 -Inf 2.68E-03 6.22E-02 0.0 0.0 0.0 0.0 0.0 45.5
TTYH1 57348 ENSG00000167614 -1.91 2.34E-03 5.61E-02 1960.4 834.0 0.0 21.2 1183.5 958.4
TUBA4A 7277 ENSG00000127824 4.85 2.91E-05 1.70E-03 54.0 47.9 185.7 29.1 0.0 6.1
TUBB4 10382 ENSG00000104833 4.59 1.43E-03 3.94E-02 150.5 113.6 32.7 0.0 0.0 3.6
TUSC3 7991 ENSG00000104723 -3.74 1.32E-04 5.89E-03 0.0 0.0 5.9 75.0 379.0 333.7
UGT8 7368 ENSG00000174607 4.64 4.89E-06 3.85E-04 155.7 274.5 169.2 38.9 3.8 8.6
UNC80 285175 ENSG00000144406 -1.98 4.34E-03 9.08E-02 113.5 162.2 0.0 38.7 322.5 207.8
VAX1 11023 ENSG00000148704 -Inf 2.74E-06 2.39E-04 0.0 0.0 0.0 0.0 32.7 121.6
VCAM1 7412 ENSG00000162692 4.34 1.49E-07 1.89E-05 0.0 0.0 46.0 2398.4 8.4 71.4
VIPR1 7433 ENSG00000114812 -4.43 4.74E-06 3.77E-04 6.5 3.1 66.7 6.8 1096.8 1.0
VIPR2 7434 ENSG00000106018 -Inf 2.65E-08 3.86E-06 0.0 0.0 0.0 0.0 216.4 15.9
VIT 5212 ENSG00000205221 6.07 1.83E-05 1.18E-03 14.9 12.4 0.0 193.0 0.0 2.1
VSNL1 7447 ENSG00000163032 3.76 1.35E-04 5.99E-03 0.0 1.6 16.2 332.9 0.0 16.7
248

A.1 Differential Expression Appendix
Gene symbol(s) Entrez gene ID(s) Ensembl 56 gene ID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
WBSCR17 64409 ENSG00000185274 -Inf 3.18E-05 1.80E-03 0.0 0.0 0.0 0.0 17.6 90.9
XXbac-BPG116M5.1,C2,CFB 629,717 ENSG00000166278,ENSG00000243649,ENSG00000244255 3.95 4.11E-04 1.54E-02 5.4 4.0 98.3 132.4 0.0 10.6
XYLT1 64131 ENSG00000103489 4.78 1.10E-03 3.20E-02 174.3 164.9 0.0 0.8 0.0 4.0
ZEB1 6935 ENSG00000148516 -2.40 2.63E-03 6.14E-02 504.6 290.3 85.0 57.0 508.8 1006.3
ZIC2 7546 ENSG00000043355 -5.61 2.65E-05 1.59E-03 1.4 6.3 0.8 0.0 0.0 223.2
ZIC3 7547 ENSG00000156925 -Inf 2.12E-04 8.75E-03 0.0 0.0 0.0 0.0 0.0 79.8
ZIC4 84107 ENSG00000174963 -7.51 5.84E-04 1.95E-02 0.0 0.0 0.6 0.0 28.5 47.4
ZIC5 85416 ENSG00000139800 -7.19 6.13E-06 4.74E-04 0.0 1.6 0.6 0.0 0.0 211.7
ZNF281 23528 ENSG00000162702 5.45 4.39E-04 1.61E-02 60.6 51.6 81.7 0.0 0.0 2.0
ZNF423 23090 ENSG00000102935 -7.23 3.72E-07 4.29E-05 0.0 0.0 0.0 3.0 80.2 247.2
ZNF454 285676 ENSG00000178187 4.41 7.62E-05 3.75E-03 24.3 118.3 19.6 152.8 0.0 9.2
ZNF536 9745 ENSG00000198597 Inf 3.95E-05 2.18E-03 41.7 100.5 3.8 37.9 0.0 0.0
ZNF710 374655 ENSG00000140548 -4.55 3.99E-03 8.49E-02 4.1 2.1 5.3 0.0 33.8 68.7
ZNF714 148206 ENSG00000160352 Inf 2.07E-03 5.12E-02 86.5 59.1 18.8 0.0 0.0 0.0
ZNF747 65988 ENSG00000169955 Inf 4.62E-04 1.66E-02 12.2 9.5 14.6 71.1 0.0 0.0
249

A.2 Classified Differential Expression Appendix
A.2 Classified Differential Expression
The differentially expressed genes generated by the Bioconductor R package
DESeq at a FDR

A.2 Classified Differential Expression Appendix
Table A.2: Classification of differentially expressed genes based on literature mining analysis.
Implicated in glioma Limited evidence in glioma Not implicated in glioma but other cancers Cancer type Not implicated in cancers
ACIN1 ABAT ADAMTS1 non-small lung carcinoma AC012354.2
ACTC1 ADCYAP1R1 ADAMTS10 glaucoma AC026410.6
ADAMTS4 ADD2 AKR1B10 cervical, endometrial AC034102.1
AGT ADRA1B AMMECR1 renal cell carcinoma AC067930.1
ALDH1A3 BGN ANGPTL2 ovarian cancer AC068399.1
APLN BST2 ARC breast cancer AC092296.1
APOD CNTN6 ARHGAP20 breast, BCLL ACTA2
AQP4 DDAH2 ARHGAP8 colorectal ADRA2A
ASNS DDIT3 ARHGEF7 gastric AFAP1L2
ASPN FCGR2B ATAD3C lung adenocarcinoma AGPAT9
ATOH8 FUT8 AZGP1 breast, pancreatic AIDA
ATP10B FXYD1 BACE2 breast ANGPTL1
ATP1B2 GABRA5 BEX5 breast ATP1A2
BAMBI GBP1 BTBD11 neuroblastoma B3GNT9
BID GBP2 BTG1 ovarian B4GALNT1
BIRC3 GFPT2 C5orf13 esophageal BATF3
BMP7 GJB2 C6orf15 blood BMP8B
BTC HAND2 C8orf4 thyroid BMPER
C1S HAPLN1 CAPN6 HeLa C10orf11
C2 HLA-DMA CARD17 head, neck C10orf116
C3 HLA-DPA1 CDH19 head, neck C10orf81
CA12 HLA-DRB5 CDH6 renal cell carcinoma C10orf90
CALM1 HOXA5 CILP breast C14orf143
CAMK2B HOXA7 CLDN10 hepatocelluar carcinoma C1orf133
CASP1 HOXB6 CLDN11 gastric cancer C1orf187
CBLC HOXB9 CMPK2 chronic myelogenous leukemia and CLL C1orf94
CCKBR HOXC10 CMTM5 prostate,pancreas,carcinoma C20orf103
CCL2 HOXC13 CMTM8 HeLa C21orf62
CCL26 IFITM2 CNKSR2 HL60 leukemia cells C2orf80
CCL7 IGSF3 COL8A1 hepatocarcinoma C3orf58
CCND2 IL17RD CRIP2 nasopharyngeal epithelial cell line C4orf32
CCNY IL1RAPL1 CRYBB2 rhabdoid tumour (kidney) C5orf38
CCR1 INSIG1 CRYM prostate C5orf41
CD248 IRX1 CTNNA2 endometrial C6orf138
CD55 ITGA4 CTSC esophageal cancer C7orf16
CD58 KCNMB2 CXXC4 renal carcinoma C7orf40
CD68 LAMA2 DCBLD2 gastric,lung C9orf125
CD70 LFNG DDIT4L melanoma C9orf64
CD74 LHFPL3 DHRS3 neuroblastoma C9orf95
CD9 MAP3K5 DOCK10 melanoma CABP7
CD97 MEST DOCK5 osteosarcoma CACNA1A
CDH13 MFAP2 DUSP16 Burkitt’s lymphoma CACNA1C
CDKN2A MMP17 DYNC1I1 myeloma CACNG7
CDKN2C MOSC2 EDA2R colorectal CACNG8
CEBPB MX1 EEF1A2 ovarian, pancreatic, lung CAMK1D
CFB NCALD EPDR1 colorectal CARD16
CITED4 NEFM EPHB3 non small cell lung cancer, colon CCDC129
CLDN3 NELL2 ETS2 prostate,colon,breast CCDC48
CNTN1 NFATC2 F12 small cell lung CCDC64
COL1A2 NFKBIZ FAM84A colon CCNO
COL3A1 NR0B1 FBLN2 breast CDHR1
COL4A6 NRN1 FBN2 squamous cell carcinoma CGREF1
251

A.2 Classified Differential Expression Appendix
Implicated in glioma Limited evidence in glioma Not implicated in glioma but other cancers Cancer type Not implicated in cancers
CPAMD8 NXPH1 FCGR2B follicular lymphoma CHCHD10
CREB3L1 OAS3 FCRLA B cell lymphoma CHODL
CTLA4 OLFM1 FERMT3 lung CHRDL1
CXCL1 OXTR FGF19 liver,lung,colon CMAH
CXCL12 PCDHB3 FOXJ1 pancreatic COL21A1
CXCL14 PDE4B FOXQ1 colorectal,breast CPLX2
CXCL2 PERP FXYD5 gastric,pancreatic,esophageal,cervical CPNE2
CXCL3 PIGA GCNT1 prostate CPNE5
CXCL6 PITPNC1 GJB2 breast,head,neck,pancreatic,endometrial CRB2
DCN PKNOX2 GJC3 HeLa CRYBB1
DKK1 PMP2 GMPR Human promyelocytic leukemia-cells CSGALNACT1
DLK1 PNP GPC3 hepatocelluar carcinoma,liver CXorf38
DNER PPEF1 GRB14 breast,prostate CYB5R2
DRD2 PSPH HAND2 neuroblastoma DCHS1
DUSP5 PTPRR HLA-DPA1 pilocytic astrocytoma DDO
EEF1D PXDN HLA-DQA1 breast, gastric,lung adenocarcinoma DIAPH2
EFEMP1 PYGL HLA-DRB4 childhood ALL DKFZp434H1419
EFNA5 RAD50 HOXA4 ovarian,AML DNM3
EGFR RARRES3 HTATIP2 medulloblastoma,neuroblastoma DPY19L1
ELMO1 RASGEF1C IFI27 skin,epithelial,breast DTX4
EPAS1 RRAD IFI6 hepatocellular,gastric,fibrosarcoma ECHDC2
EPB41L3 SALL1 IGSF11 hepatocellular,gastrointestinal EFHD2
ERBB4 SCN1B IL4I1 B-cell lymphoma ELMO2
ESRRG SCN5A INS colon,breast, ELMOD1
ETS1 SIX6 IRX2 soft tissue sarcoma ELN
F2RL1 SLC4A4 ITIH5 breast cancer ELOVL2
FAM38B SLITRK5 ITM2A alveolar rhabdomyosarcoma ELTD1
FCGR2A SNAP25 JPH1 colorectal carcinoma EML2
FGFBP2 SNX10 KANK1 renal cell carcinoma EPDR1
FGFR1 SORL1 KCNJ12 prostate,stomach,breast EVC2
FMNL1 SQRDL KIAA1217 non-small cell lung cancer FAM126A
FOXG1 SRGAP3 L3MBTL4 breast cancer FAM129A
FXYD3 SRPX LEMD1 prostate,colorectal FAM134B
GAL STAMBPL1 LMO2 gastrointestinal,hepatocellular,BLL,T-ALL FAM150B
GALR1 SULF2 LMO4 squamous cell carcinoma,breast,pancreatic FAM176A
GAS7 SYT1 LOXL4 head/neck carcinoma,bladder FAM181A
GDF15 TAGLN LPHN2 breast cancer FAM184B
GFAP TM4SF1 LRAT colon,colorectal,renal,mammary FAM189A1
GGH TMEM71 LRP4 CLL FAM196A
GJA1 TNFAIP6 LRRC55 pancreatic FAM5B
GPNMB TUSC3 LUM cervical,colorectal FAM69A
GRIA1 LXN prostate,gastric,colon,melanoma,medulloblastoma FAM70A
GRIA3 MAF cervical squamous FBXO27
GRM3 MALL oligodendrogliomas FBXO32
HEPACAM MAN1C1 malignant fibrous histiocytomas FCGR2C
HIF3A MAP7 colon FEZF2
HLA-A MARVELD3 pancreatic FXYD7
HLA-DPB1 MEIS2 ovarian,embryonal carcinoma,myeloid leukemia FZD3
HLA-DQB1 MGLL breast GABBR2
HLA-DRA MICB cervical,osteosarcoma,squamous cell carcinoma GABRQ
HLA-DRB1 MIPOL1 nasopharyngeal carcinoma,prostate GALNT5
HLA-DRB3 MLPH lung,breast GBP3
HMGA2 MMRN1 non-small cell lung cancer GBP4
252

A.2 Classified Differential Expression Appendix
Implicated in glioma Limited evidence in glioma Not implicated in glioma but other cancers Cancer type Not implicated in cancers
HOMER1 MN1 meningioma,ALL GBP5
HOXA1 MT1A breast GDPD2
HOXA10 MXRA5 ovarian,colon GEM
HOXB7 MYBL1 leukemias GLYATL2
HOXD10 MYH3 adenocarcinoma GNG11
HOXD13 NBL1 pancreatic,colon GPR158
HOXD3 NDE1 AML GPR98
HOXD9 NFE2L3 pancreatic,T-cell lymphoblastic lymphoma GRIK3
HPRT1 NID1 colon,gastric GYG2
HPSE NOP16 murine breast H1F0
ICAM1 NPFFR2 Primary CNS lymphoma HCP5
ID4 NR0B1 lung adenocarcinoma,sarcoma HLA-DQA2
IFI30 NR1D1 breast HRCT1
IGF2 NTN1 colorectal,neuroblastoma IFITM8P
IGFBP5 OAS2 prostate,breast IGLON5
IKBKE ODZ2 lymphoma IL33
IL13RA2 OGN hepatocarcinoma INS-IGF2
IL1B PARP3 breast cancer ISYNA1
IL1R1 PARP12 breast cancer ITGBL1
IL6 PCDH20 non-small-cell lung cancer KALRN
IL8 PHACTR3 non-small-cell lung cancer KCNG1
INHBA PITX1 breast,lung,adenocarcinoma KCNK12
INHBB PKI55 cancer KCTD12
INPP5D PLD3 breast LINGO1
IRAK1 PLS1 lung adenocarcinoma LOC100127888
JAM2 PLS3 HeLa LOC285141
KCNA2 PLSCR1 ovarian,colorectal LRRC2
KCNK1 PMEPA1 renal cell carcinoma LYST
KLF6 PNMA2 gastrointestinal neuroendocrine carcinomas MACROD2
LAMA4 PPAP2C MAP6
LGALS3 PPL esophageal MARCH1
LIF PPP4R1 breast,gastric MID1IP1
LIFR PRIMA1 thyroid,non small cell lung cancer cells,prostate MOCOS
LMO3 PROCR breast MYL1
LNX1 PTPRD gastrointestinal MYL9
LPAR6 PTPRH stomach,colorectal MYO1B
LRRN2 RAB11FIP1 breast NKX6-2
LTF RAB38 melanoma NRBF2
LY96 RAB6B breast,neuroendocrine OASL
MAL RARRES1 melanoma,nasopharyngeal carcinoma,prostate OTOR
MARS RARRES3 ovarian,breast PCDHB12
MATN2 REC8 gastrointestinal stromal tumour,lymphoma PCDHB4
MBP RGS17 lung,prostate,ovarian PCOLCE2
MET RGS5 gastric,renal cell carcinoma PDE1C
MIA RHOBTB3 ganglioglioma PDE1C
MICA RNASEH1 prostate PDZRN3
MKI67 RNF175 pediatric AML,non-Hodgkinslymphoma PION
MLC1 RTN1 neuroblastoma PLA2G4A
MMP7 SALL2 synovial sarcoma PLAC9
MT2A SEMA4G colorectal PLCH1
MTTP SEZ6L lung PLCH1
MYC SHOX2 lung PLEKHA6
NAMPT SIX3 Chondrosarcomas PPCS
253

A.2 Classified Differential Expression Appendix
Implicated in glioma Limited evidence in glioma Not implicated in glioma but other cancers Cancer type Not implicated in cancers
NCAM1 SKAP2 pancreatic PPHLN1
NDN SLC16A3 bladder cancer PPM1K
NFASC SLC26A2 gastric,colon carcinoma PRSS12
NFIA SLC2A5 breast,adenocarcinoma,colon,liver,lymphomas PRSS12
NKX2-1 SLCO2A1 colorectal PSORS1C1
NKX6-2 SMAGP metastatic cell lines RAB7L1
NMU SPAG4 various cancers RADIL
NNAT SPARCL1 colorectal,pancreatic,lung RGAG4
NNMT SPINK2 lymphomas RGL3
NOTCH3 SPINT1 colorectal,gastric,breast,stomach,ovarian RGS20
NOV SPRED1 juvenile myelomonocytic leukemia, RHBDF2
NPTX2 SRP9 colorectal RHPN1
NR4A2 ST6GALNAC3 renal cancer RIPK4
NRG1 ST6GALNAC5 breast RP11-473I1.1
NRP2 STC2 esophageal,squamous-cell carcinoma,ovarian RP9
NTRK2 STEAP1 breast,prostate RPSAP52
OAS1 STEAP2 prostate RRS1
OTX2 STRA6 colorectal RTP4
P2RX7 SULF1 colorectal,ovarian,hepatocellular,gastric SCN11A
PAX8 SULF2 lung adenocarcinoma,squamous cell carcinoma SDK2
PDGFA SYTL4 breast SEMA6A
PDGFRA SYTL5 breast SFTA3
PDGFRB TBC1D8 pancreatic SGCD
PDPN TCEAL2 ovarian SH3BGR
PEA15 TCF7 colon cancer,acanthoma SHISA2
PHLPP1 TM4SF1 pancreatic SHROOM3
PI15 TNFAIP2 nasopharyngeal carcinoma SLC38A5
PLA2G2A TNFRSF14 pancreatic SLC46A1
PLCB1 TNFSF4 Hodgkin’s lymphoma,breast SLC4A11
PTEN TPMT acute and lymphoblastic leukemia SMOC2
PTGS1 TRIB2 lung,AML,melanoma SNHG12
PTPRD TRIB3 breast,colorectal cancer SNX22
RASGRP3 TSHZ3 breast,prostate SORCS2
S100A6 TSPAN13 prostate,breast SORCS3
S100B TSPAN7 neuroblastoma,ALL SPTBN5
S1PR3 TSPAN9 ovarian STEAP1B
SDC2 TTN melanoma STX3
SELENBP1 ZIC3 meningioma STXBP5L
SEMA3D ZIC5 meningioma SYNGR1
SERPINE2 ZNF423 neuroblastoma,CML TAPBPL
SFRP1 TBC1D8
SHC4 TECRL
SLC15A2 TESC
SLC38A1 TMCO4
SLCO1C1 TMEM100
SLITRK5 TMEM132D
SNCAIP TMEM158
SOD2 TMEM176B
SOD3 TMEM200B
SOX10 TMEM38A
SOX3 TMSB15A
SPP1 TMSL1
ST6GAL1 TNNC2
254

A.2 Classified Differential Expression Appendix
Implicated in glioma Limited evidence in glioma Not implicated in glioma but other cancers Cancer type Not implicated in cancers
ST6GALNAC5 TNNI1
SYNM TNNI2
TCF7L2 TOX
TES TRAM1L1
TF TRIM14
TFAP2A TRIM48
TFCP2 TSTD1
TGFA TTF2
TGM2 TUBB4
THBS2 TUSC3
THY1 UNC5D
TNC UNC80
TNFAIP3 VAX1
TPM1 VIPR2
TRAF1 VIT
TRIM47 WBSCR17
TRPM8 XYLT1
TSLP ZNF281
TTYH1 ZNF454
TUBA4A ZNF536
UGT8 ZNF710
VCAM1 ZNF714
VIPR1 ZNF747
255
VSNL1
ZEB1
ZIC2
ZIC4

A.3 Quantitative RT-PCR Appendix
A.3 Quantitative RT-PCR
Ct values were normalised to the mean of three endogenous control genes (18S
rRNA, TUBB and NDUFB10). Values were capped at 37 prior to normaliza-
tion. Each sample was then normalised by subtracting the mean Ct for the
sample, of the three control assays, and adding the global mean Ct of the con-
trol assays. The subtraction corresponds to a standard deltaCt normalisation
to adjust for differences in the amount of RNA among samples. The addition
ensures that normalised values occupy approximately the same range as the
original values; this is only done for convenience and does not affect the statis-
tical tests since all Ct values are incremented by the same amount. Replicated
samples have suffix (A) - (D). Raw and normalised Ct values are listed in the
tables below.
256

A.3 Quantitative RT-PCR Appendix
Table A.3: Raw Ct values. Abbreviations: "down" for down-regulated, "up" for up-regulated, and "Norm" for Normalisation control.
Well Applied Biosystems assay ID Gene CB130 CB130 CB130 CB152 CB152 CB152 CB152 CB171 CB171 CB171 CB192 CB541 CB660 CB660 G2 G2 G7 G7 G7 G7 G9 G9 G14 G14
Category A B C A B C D A B C A B A B A B C D A B A B
1 MYL9-Hs00382913_m1 Core down 27.9 27.5 28.0 22.9 25.4 26.2 25.1 28.9 31.9 29.6 26.8 24.4 27.1 25.8 37.0 35.9 31.4 36.2 31.2 28.7 26.7 27.4 34.1 34.8
2 CHCHD10-Hs01369775_g1 Core up 26.7 25.3 25.9 25.8 25.5 25.3 24.3 26.7 28.1 26.2 28.8 29.2 29.1 28.6 25.7 25.6 25.8 25.9 26.4 24.0 25.8 27.1 26.2 26.6
3 RGS5-Hs00186212_m1 Core down 30.9 30.0 30.2 29.0 31.9 31.0 29.0 32.5 31.1 30.9 32.6 27.0 26.3 26.7 31.4 32.0 31.1 31.1 29.2 26.9 30.9 32.6 31.6 31.8
4 ST6GALNAC5-
Core down 28.5 28.3 28.5 33.4 34.0 33.0 31.7 29.4 32.2 29.5 25.6 26.2 23.3 24.0 36.1 35.9 36.0 40.0 35.7 34.0 28.9 29.7 30.2 29.5
Hs00229612_m1
5 CEBPB-Hs00270923_s1 Core up 32.8 30.3 30.5 30.2 32.0 31.8 31.1 32.4 32.0 31.4 30.7 31.5 32.4 30.4 32.6 35.2 28.6 30.0 30.5 28.9 29.9 31.0 29.8 29.7
6 C5orf13-Hs00854282_g1 Core down 28.1 27.3 28.0 25.4 27.6 27.0 25.3 28.0 26.8 26.7 29.0 24.3 25.0 25.3 40.0 27.8 29.9 29.2 23.5 26.2 27.7 40.0 26.9 31.1
7 PDGFRA-Hs00998026_m1 Marker 40.0 37.0 34.7 33.1 40.0 28.6 28.2 40.0 40.0 40.0 32.4 35.2 29.7 31.3 25.2 26.9 22.8 22.4 22.2 21.8 24.3 24.9 25.0 25.8
8 CCND2-Hs00922419_g1 Core down 32.0 30.4 30.9 31.5 33.9 29.0 28.3 31.9 32.7 30.9 25.1 23.7 23.9 23.9 23.5 23.8 22.3 22.2 24.7 22.5 30.7 30.7 33.2 36.9
9 NKX2-1-Hs00163037_m1 Core down 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 28.3 28.8 29.0 40.0 40.0 32.3 34.8 33.0 31.7 40.0 40.0 30.4 30.9
10 CTSC-Hs00175188_m1 Core up 23.1 24.2 24.5 24.4 24.2 25.3 24.1 23.1 25.9 24.0 23.9 24.6 25.3 22.9 21.5 22.5 24.9 24.6 27.2 24.4 23.3 24.3 22.5 22.8
11 18S-Hs99999901_s1 Norm 13.3 11.6 11.4 12.1 14.0 11.5 11.1 13.0 12.4 11.3 12.5 12.3 12.0 11.2 14.9 18.9 12.1 13.0 11.1 10.9 12.9 14.8 11.9 12.2
12 DNER-Hs00294564_m1 Core down 31.9 31.9 32.7 32.6 31.8 28.8 27.3 31.7 31.5 32.1 30.4 21.8 22.3 21.9 26.4 25.9 27.4 25.4 24.3 24.3 26.8 24.8 31.7 31.1
13 LMO4-Hs01086790_m1 Core up 26.6 26.3 26.8 26.6 27.7 27.2 25.2 26.4 27.9 27.2 25.6 25.6 25.3 25.9 23.5 23.9 21.8 22.6 24.7 23.0 24.7 26.5 24.3 24.7
14 PLCH1-Hs00392783_m1 Core down 26.1 25.9 26.1 26.3 27.0 27.9 26.1 25.7 28.1 27.6 26.8 26.5 27.7 27.9 33.6 34.8 29.1 30.1 29.9 28.2 28.2 29.9 26.3 26.5
15 SPARCL1-Hs00949886_m1 Core down 33.3 30.8 30.9 26.9 29.8 24.8 22.0 35.2 29.9 30.4 28.4 24.8 22.6 24.8 25.5 25.6 24.0 27.8 29.8 29.0 22.7 26.2 28.1 27.7
16 SYNM-Hs00322391_m1 Core down 28.9 27.8 28.0 26.7 27.6 27.9 25.5 28.2 28.9 28.7 27.0 27.1 26.9 26.4 29.5 30.2 28.5 29.7 29.4 27.9 25.5 28.6 32.8 34.1
17 FBLN2-Hs00157482_m1 Core down 40.0 35.7 40.0 35.9 40.0 40.0 34.9 40.0 35.2 35.5 33.8 28.6 27.3 28.3 34.4 36.0 35.0 32.9 36.9 34.6 33.5 40.0 36.0 35.8
18 SALL2-Hs00413788_m1 Core down 40.0 33.0 33.4 30.9 37.0 30.7 27.0 37.0 30.3 33.2 32.7 26.0 24.4 27.0 31.4 33.8 27.7 28.7 29.6 29.9 34.9 40.0 30.8 29.9
19 TUSC3-Hs00185147_m1 Core down 24.0 22.4 22.9 23.2 23.8 23.3 21.6 24.0 25.2 23.3 24.3 22.9 23.3 23.0 40.0 40.0 35.3 33.9 37.0 33.5 23.7 23.7 23.1 23.5
20 GPR158-Hs00393109_m1 Core down 34.4 34.1 34.8 36.9 36.5 32.1 32.4 35.6 36.1 33.4 26.9 25.1 27.0 25.7 27.2 29.0 27.7 26.8 28.7 25.9 34.9 35.6 31.7 31.9
21 PEG3-Hs00377844_m1 Core down 30.7 30.9 31.1 29.9 30.9 31.0 29.2 29.9 31.2 31.0 29.8 27.6 25.0 27.0 28.2 29.6 26.1 24.8 26.3 25.6 33.6 36.6 28.1 28.7
22 FUT8-Hs00189535_m1 Core up 26.3 26.3 26.7 26.1 27.0 28.0 26.2 26.5 28.7 27.2 26.3 27.4 28.2 26.2 26.1 26.6 27.5 27.5 27.3 26.6 26.0 27.3 27.6 28.0
23 HLA-DRA-Hs00219575_m1 Core up 27.5 26.1 26.3 28.2 30.1 26.0 22.1 28.6 27.0 27.4 22.3 26.7 30.7 25.5 21.4 23.1 24.3 24.2 24.7 22.9 21.4 26.0 26.8 27.6
24 ASCL1-Hs00269932_m1 Marker 34.0 32.0 31.8 32.7 36.0 28.1 26.1 34.5 32.1 35.1 32.7 28.9 27.0 33.6 24.4 24.9 25.5 24.3 24.4 25.1 40.0 40.0 33.4 40.0
25 PI15-Hs00210658_m1 Core down 31.6 30.3 30.1 23.0 28.6 27.1 24.4 31.8 28.6 31.0 29.6 24.5 24.4 23.9 26.3 28.6 31.8 27.6 29.8 29.3 27.0 33.7 31.1 30.1
26 EPDR1-Hs00378148_m1 Core up 28.6 27.9 28.6 26.5 27.5 29.4 26.7 29.0 30.5 29.1 30.8 28.5 29.8 29.1 27.0 27.8 28.2 29.0 29.5 27.5 26.9 28.9 27.9 28.4
27 DTX4-Hs00392288_m1 Core down 32.9 30.6 31.2 28.2 31.6 29.0 27.0 33.2 29.1 30.3 32.9 26.0 24.9 25.9 34.2 37.0 30.3 30.8 32.0 32.6 29.8 33.1 27.5 27.6
28 CXXC4-Hs00228693_m1 Core down 30.9 30.8 30.7 34.1 40.0 30.8 29.4 31.6 31.4 33.2 28.2 25.5 25.6 27.1 28.6 27.4 27.0 26.7 26.3 26.8 40.0 40.0 28.5 29.9
29 IRX2-Hs01383002_m1 Core down 30.1 30.5 30.6 29.1 29.2 31.4 29.9 29.2 31.9 31.1 30.6 26.1 26.4 25.5 29.8 29.8 40.0 40.0 40.0 40.0 30.3 30.3 40.0 40.0
30 MAP6-Hs01023152_s1 Core down 27.6 24.6 25.3 25.8 27.6 24.7 21.8 27.0 25.8 24.7 28.0 26.4 25.7 27.2 26.8 26.6 27.0 26.9 27.5 26.5 28.6 30.6 28.4 28.7
31 SIX3-Hs00193667_m1 Core down 40.0 40.0 40.0 40.0 40.0 40.0 34.6 40.0 40.0 40.0 40.0 25.0 23.8 24.1 40.0 35.2 32.0 33.8 30.5 31.6 32.6 33.7 34.2 34.3
32 NKX2-2-Hs00159616_m1 Marker 40.0 40.0 40.0 37.0 40.0 33.3 33.6 40.0 40.0 40.0 40.0 28.1 30.7 28.7 25.9 26.2 26.7 26.3 29.0 29.3 26.8 27.2 28.0 27.8
33 S100A6-Hs00170953_m1 Core up 19.7 20.0 20.6 20.8 19.2 20.7 20.6 19.0 23.3 20.6 21.9 20.6 25.6 22.1 19.7 19.8 21.3 22.9 22.4 19.1 19.8 18.6 18.8 18.8
34 MMP17-Hs01108847_m1 Core up 27.9 28.2 28.5 28.2 27.7 30.3 28.9 27.5 31.0 30.1 28.6 31.8 30.2 29.0 26.6 30.0 30.0 30.0 29.8 30.1 27.7 27.3 28.0 28.1
35 FAM38B-Hs00926225_m1 Core down 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 23.8 26.8 25.2 40.0 40.0 40.0 40.0 40.0 40.0 24.7 25.3 28.4 28.7
36 OLIG2-Hs00300164_s1 Marker 29.8 32.6 33.1 29.9 30.2 28.9 28.2 32.1 36.0 35.5 29.2 25.6 28.0 26.9 24.3 26.5 21.8 22.2 22.3 22.7 30.5 32.8 30.5 36.0
37 PMEPA1-Hs00375306_m1 Core up 25.7 23.6 23.9 24.7 25.1 25.8 25.8 25.6 28.1 26.0 27.6 26.0 29.0 26.3 26.5 27.8 27.3 27.9 28.3 26.6 26.2 25.1 23.1 24.2
38 HOXD10-Hs00157974_m1 Core up 40.0 37.0 40.0 40.0 40.0 33.1 32.2 40.0 40.0 40.0 40.0 40.0 40.0 40.0 30.1 29.9 27.0 27.9 28.9 26.7 28.3 28.9 27.8 28.3
39 NDN-Hs00267349_s1 Core down 24.8 24.5 25.1 24.0 24.4 25.4 23.1 24.5 26.9 25.2 25.4 23.3 23.7 23.8 25.9 25.6 24.6 25.6 26.6 25.4 23.7 24.9 23.8 23.9
40 FOXG1-Hs01850784_s1 Core up 40.0 40.0 40.0 33.6 34.9 25.9 25.1 40.0 32.7 35.9 35.3 27.3 27.4 27.9 25.9 26.1 23.5 24.1 22.1 20.7 24.3 25.2 23.7 24.0
41 MN1-Hs00159202_m1 Core down 32.0 31.3 31.4 29.3 30.6 31.8 27.6 33.2 33.4 34.1 31.0 25.8 25.8 26.1 29.6 31.0 29.7 28.4 29.6 27.8 34.3 33.1 32.6 32.9
42 NDUFB10-Hs00605903_m1 Norm 24.8 25.0 25.9 25.5 25.2 26.5 24.8 25.1 27.6 25.9 25.3 25.1 25.8 25.2 25.5 25.0 25.0 25.7 27.2 25.5 25.5 25.2 24.2 25.0
43 HMGA2-Hs00171569_m1 Core down 31.5 27.8 28.4 33.5 30.5 29.7 29.5 29.9 29.9 28.9 28.9 26.9 27.0 26.0 32.0 32.0 30.9 30.5 29.6 28.2 34.4 31.3 29.5 30.7
44 MMRN1-Hs00201182_m1 Core down 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 26.0 27.2 27.7 40.0 40.0 36.6 36.4 40.0 36.0 40.0 36.9 32.4 33.9
45 RAB6B-Hs00981572_m1 Core down 34.4 32.0 32.2 32.8 35.3 32.4 29.5 34.1 32.8 32.5 30.4 27.8 28.5 28.1 29.5 30.6 27.2 27.9 28.1 27.7 29.7 30.2 29.3 28.9
46 NTRK2-Hs00178811_m1 Core up 28.9 30.5 31.0 23.2 24.9 26.2 22.9 29.2 30.1 31.0 29.0 28.5 26.7 27.8 22.4 23.3 29.1 25.0 27.6 26.8 22.7 25.0 24.3 23.6
47 KALRN-Hs00610179_m1 Core down 29.5 29.1 30.0 28.0 29.7 31.0 28.8 29.5 30.0 30.1 28.3 26.0 25.7 25.7 28.6 31.0 28.3 28.0 28.1 27.5 27.1 29.7 29.9 30.2
48 CD74-Hs00269961_m1 Core up 27.3 27.3 27.4 27.9 28.9 28.2 24.0 27.4 27.4 28.9 22.1 26.9 30.1 25.5 21.8 24.1 25.2 23.9 26.5 24.8 21.9 27.1 26.9 27.9
257

A.3 Quantitative RT-PCR Appendix
Well Applied Biosystems assay ID Gene CB130 CB130 CB130 CB152 CB152 CB152 CB152 CB171 CB171 CB171 CB192 CB541 CB660 CB660 G2 G2 G7 G7 G7 G7 G9 G9 G14 G14
Category A B C A B C D A B C A B A B A B C D A B A B
49 LUM-Hs00158940_m1 Core down 29.4 28.9 29.2 30.3 31.9 36.9 32.0 31.6 33.4 32.8 29.9 25.5 24.6 24.6 40.0 40.0 32.5 29.0 33.0 31.0 23.0 27.3 29.6 29.7
50 NELL2-Hs00196254_m1 Core down 24.5 24.3 24.7 28.0 29.3 29.1 24.9 24.3 25.6 25.3 25.3 24.6 23.8 25.0 30.5 30.9 30.2 30.7 31.7 30.7 28.7 30.0 34.2 34.0
51 MAP6-Hs01929835_s1 Core down 27.0 25.3 26.1 25.6 26.6 25.0 22.4 26.5 26.0 25.1 27.4 26.1 25.6 27.9 25.8 25.9 26.8 26.4 28.5 27.7 27.3 29.3 26.3 26.5
52 PDE1C-Hs01095694_m1 Core up 29.9 32.4 32.9 27.0 27.9 29.0 28.0 29.2 31.4 30.8 30.4 26.2 27.9 25.9 26.4 28.0 24.3 24.6 25.5 23.3 23.8 27.1 24.2 24.4
53 SULF2-Hs00378697_m1 Core up 28.5 29.7 30.1 25.9 26.8 29.1 25.4 27.9 30.9 31.3 26.8 26.4 27.8 26.7 23.0 23.9 23.3 23.1 25.3 24.4 24.0 23.7 23.9 23.8
54 TUBB-Hs00962420_g1 Norm 20.0 21.3 21.5 21.2 20.6 23.1 22.7 19.8 24.0 22.2 21.3 20.4 21.0 20.2 21.5 23.0 20.8 21.1 23.1 20.6 21.7 21.5 20.5 20.8
55 LMO2-Hs00277106_m1 Core down 36.8 37.1 33.7 35.3 34.1 30.0 29.1 36.0 34.9 33.1 30.4 26.1 25.7 26.3 26.8 27.2 25.4 26.6 27.1 26.8 28.1 30.8 31.7 31.9
56 MAN1C1-Hs00220595_m1 Core up 27.8 26.9 27.0 25.9 29.0 27.8 25.7 27.2 27.2 27.0 27.1 31.1 29.0 28.9 26.8 29.0 26.0 26.8 27.5 26.4 25.2 27.4 27.6 27.4
57 TAGLN-Hs00162558_m1 Core down 21.2 18.8 18.5 20.5 21.9 20.2 21.9 22.0 20.7 20.0 21.8 21.2 23.6 21.8 27.4 29.7 40.0 36.9 37.2 29.3 21.6 22.1 28.6 28.9
58 RTN1-Hs00382515_m1 Core down 26.4 23.0 23.2 29.7 28.8 28.9 25.9 27.2 27.6 26.9 23.8 23.8 25.5 25.3 26.1 26.1 23.4 24.4 26.3 24.2 26.1 27.3 24.7 25.2
59 MAF-Hs00193519_m1 Core down 26.5 25.9 26.7 26.4 25.8 27.4 26.0 26.4 29.7 27.6 27.1 25.2 26.4 26.0 27.4 27.0 27.8 29.4 27.9 25.2 30.1 30.5 27.9 28.2
60 NPTX2-Hs00383983_m1 Core down 35.2 30.6 30.5 30.5 26.8 32.8 29.8 40.0 36.9 35.3 33.6 22.9 25.5 23.6 40.0 40.0 26.2 29.4 31.4 28.5 31.1 26.8 29.6 29.2
61 EDA2R-Hs00939736_m1 Core down 27.9 28.0 28.0 27.6 27.2 28.9 26.8 27.5 29.6 28.2 28.5 26.6 28.4 27.0 27.8 29.2 30.3 34.3 33.5 30.9 28.2 28.9 26.2 26.7
62 SOX10-Hs00366918_m1 Marker 40.0 40.0 40.0 40.0 40.0 35.2 32.8 40.0 40.0 40.0 40.0 40.0 35.1 40.0 27.9 28.0 40.0 37.0 36.0 36.0 40.0 40.0 40.0 40.0
63 SEMA6A-Hs00221174_m1 Core down 25.0 25.9 25.9 24.3 25.2 26.2 24.3 24.6 26.8 26.4 22.7 22.9 23.0 23.3 25.6 26.4 24.9 25.2 26.7 26.3 26.7 30.2 26.2 27.3
64 DDIT3-Hs00358796_g1 Core up 28.0 29.0 29.2 28.4 27.6 27.9 27.3 28.5 30.3 28.6 28.7 27.6 27.9 27.5 27.4 27.2 25.9 25.9 28.8 26.2 26.6 27.1 26.4 26.9
65 CA12-Hs01080909_m1 Core down 30.1 32.2 32.6 30.9 28.3 29.5 31.0 30.9 32.6 31.5 21.5 22.1 23.4 22.5 25.4 26.1 23.4 23.1 24.9 22.3 24.1 26.4 30.4 30.8
66 C9orf125-Hs00260558_m1 Core down 32.9 32.2 32.8 36.7 40.0 34.6 34.8 33.7 35.4 36.3 29.4 28.5 28.1 29.0 28.0 30.2 29.0 28.5 30.2 29.4 28.3 28.2 30.4 31.0
67 SDC2-Hs00299807_m1 Core down 24.7 24.5 24.7 24.1 24.5 25.0 23.3 25.3 26.4 25.2 25.5 25.1 25.8 25.5 27.4 27.0 26.6 28.4 28.3 26.7 26.4 27.9 26.1 26.5
68 SLIT2-Hs00191193_m1 Core down 29.3 33.1 32.9 28.9 28.7 31.4 29.3 29.8 33.0 32.9 31.8 24.0 24.2 24.4 40.0 36.4 27.8 27.6 31.9 30.7 26.4 28.3 25.1 25.5
69 MT2A-Hs02379661_g1 Core up 25.2 23.6 24.2 22.9 22.4 20.9 20.4 25.5 25.3 23.9 23.9 23.3 24.4 24.2 22.5 22.2 20.3 21.4 22.4 20.0 19.1 19.0 22.5 22.3
70 IL17RD-Hs00296982_m1 Core down 26.8 25.7 25.7 25.8 26.9 27.1 26.2 25.8 28.3 26.9 26.3 23.9 24.4 23.7 27.4 28.7 27.7 27.0 27.6 26.9 24.3 24.8 25.9 26.0
71 PDZRN3-Hs00392900_m1 Core down 29.0 28.9 29.0 27.2 28.8 29.6 26.6 28.4 27.7 28.8 28.4 26.4 26.7 26.8 29.8 32.4 27.1 29.1 29.3 29.6 27.9 30.9 29.3 29.6
72 LGALS3-Hs00173587_m1 Core up 25.7 24.5 25.4 24.8 24.0 24.5 23.0 25.6 26.6 25.1 26.5 25.8 25.6 25.0 25.0 24.0 26.4 27.7 26.2 25.1 22.7 23.5 25.5 25.3
73 FAM69A-Hs00961685_m1 Core up 25.3 25.0 25.2 23.8 24.6 26.6 23.5 24.8 26.5 26.3 26.4 27.6 29.7 28.0 25.6 26.0 25.7 26.5 27.3 25.9 24.9 25.9 25.3 25.3
74 TERT-Hs00972656_m1 Marker 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 36.7 36.5 35.5 33.6 34.4 32.4 32.2 32.7 32.0 33.4 33.4 31.0 34.0
75 PLS3-Hs00418605_g1 Core up 24.6 27.6 27.9 25.0 24.1 26.7 26.4 24.4 29.7 28.9 22.9 26.1 25.6 24.1 23.4 24.4 25.3 26.6 26.9 24.2 22.5 23.1 21.8 22.0
76 PTEN-Hs02621230_s1 Core down 30.0 27.4 27.6 29.4 30.5 28.1 26.4 28.8 28.3 27.5 29.5 27.6 27.6 27.3 40.0 40.0 29.9 30.9 27.9 27.3 40.0 40.0 27.8 28.0
77 SALL2-Hs00826674_m1 Core down 34.4 31.7 31.9 32.2 34.0 33.1 29.7 34.8 32.6 33.0 34.8 32.5 32.4 32.6 31.5 32.2 30.8 31.4 33.2 28.7 33.1 34.9 34.1 33.9
78 KCTD12-Hs00540818_s1 Core down 30.6 30.2 30.3 33.3 33.9 32.7 33.2 29.7 29.7 29.8 26.5 25.3 27.3 26.2 27.8 30.0 30.8 32.4 35.5 32.1 27.2 29.7 33.6 32.4
79 TES-Hs00210319_m1 Core down 25.7 25.6 26.1 27.7 27.3 28.9 28.4 25.7 29.1 26.9 32.6 25.1 26.1 25.6 40.0 36.6 40.0 40.0 40.0 37.0 33.4 40.0 35.9 40.0
80 SOX2-Hs01053049_s1 Marker 25.2 21.3 21.7 24.1 26.2 22.9 20.4 24.3 22.5 22.0 24.4 23.1 23.4 23.5 24.7 24.9 23.0 23.4 22.6 21.7 25.3 27.9 23.7 23.8
81 LPAR6-Hs00271758_s1 Core down 30.9 29.0 29.4 34.3 34.3 32.2 30.7 31.2 30.8 30.1 29.8 27.9 28.5 28.5 31.9 31.9 28.1 29.8 29.4 27.8 27.7 29.6 29.2 29.8
82 ODZ2-Hs00393060_m1 Core down 28.3 27.3 27.5 26.7 28.0 28.1 26.8 28.8 29.9 29.4 25.7 26.9 25.0 24.7 30.8 31.6 24.7 26.4 27.8 26.1 27.8 30.1 25.4 25.8
83 NNMT-Hs00196287_m1 Core up 31.0 27.5 27.9 28.8 32.2 30.1 29.6 32.0 30.9 28.5 29.9 27.7 29.4 27.9 30.9 32.6 23.7 26.1 24.5 22.2 22.3 23.7 28.9 27.0
84 CACNG8-Hs01100182_m1 Core down 40.0 33.0 32.7 32.6 40.0 33.7 34.5 33.6 34.8 35.3 28.7 27.3 26.1 27.3 29.7 30.4 26.8 28.7 30.1 28.8 29.6 32.4 32.0 32.4
85 PRSS12-Hs00186221_m1 Core up 29.9 34.6 34.8 32.1 30.8 34.2 33.4 29.7 33.4 33.8 40.0 29.9 29.6 28.0 27.4 29.4 28.5 30.2 34.2 33.1 25.6 26.8 23.8 24.3
86 FOXJ1-Hs00230964_m1 Core down 35.7 32.9 32.7 26.7 33.7 30.5 26.3 34.8 30.2 33.4 35.5 30.5 29.4 32.2 30.0 31.3 31.2 32.5 31.3 31.0 37.0 40.0 33.0 33.1
87 NTN1-Hs00180355_m1 Core down 30.2 29.3 29.3 26.4 30.4 27.7 25.9 29.5 28.6 29.0 26.9 25.2 25.8 25.5 29.6 30.9 28.9 28.9 28.7 29.5 33.4 32.9 27.5 29.1
88 LMO3-Hs00375237_m1 Core down 29.3 28.5 29.0 34.5 40.0 35.4 30.7 29.6 29.0 30.4 33.9 28.7 25.5 28.4 40.0 40.0 31.2 35.2 28.0 27.9 35.6 36.9 40.0 40.0
89 CHI3L1-Hs00609691_m1 Marker 34.1 31.7 32.4 34.1 35.5 31.4 33.3 40.0 36.0 34.1 32.0 31.8 30.8 29.8 28.7 32.2 22.4 27.3 29.4 26.4 23.0 25.4 30.0 31.1
90 CD9-Hs01124025_g1 Core up 27.5 30.4 30.9 24.9 24.3 26.1 24.1 27.7 29.8 29.9 27.1 27.3 28.0 26.7 21.1 22.2 22.4 22.4 22.8 21.2 24.1 23.8 22.9 23.4
91 ADD2-Hs00242289_m1 Core up 25.9 27.4 27.8 27.1 27.3 29.2 27.8 24.9 29.4 28.8 28.1 37.0 30.3 31.8 27.2 28.9 24.6 25.0 27.7 26.7 28.2 27.9 26.3 26.6
92 MYL9-Hs00697086_m1 Core down 25.4 24.2 24.8 20.9 22.6 24.2 23.6 26.5 30.1 28.8 24.1 22.5 25.5 23.7 31.7 35.8 28.4 33.2 28.6 25.9 22.3 24.2 32.4 33.0
93 LYST-Hs00179814_m1 Core up 31.0 30.1 30.3 30.6 30.6 31.4 29.3 30.3 31.7 31.5 31.1 30.7 32.7 31.5 28.2 28.7 28.4 28.7 30.5 28.5 28.9 30.0 29.1 29.9
94 LAMA2-Hs00166308_m1 Core down 35.3 40.0 37.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 31.1 28.6 26.7 28.7 32.0 35.2 27.7 32.8 27.9 25.2 25.8 29.3 30.6 30.4
95 PLA2G4A-Hs00233352_m1 Core up 29.2 33.2 33.1 36.0 40.0 33.0 33.1 29.2 33.6 31.8 29.9 34.9 32.1 32.4 29.0 30.2 28.6 29.6 31.0 30.7 29.3 28.9 28.4 28.5
96 BACE2-Hs00273238_m1 Core up 26.5 26.7 27.1 26.7 26.2 28.3 26.8 26.2 29.1 28.2 34.0 31.7 31.1 30.5 26.0 26.5 26.6 26.7 28.8 26.3 26.3 25.5 24.5 25.0
258

A.3 Quantitative RT-PCR Appendix
Well Applied Biosystems assay ID Gene G19 G19 G21 G21 G23 G23 G24 G24 G25 G25 G26 G26 G30 G30 G31 G31 G32 G32 G144 G144 G166 G166 G179 G179
Category A B A B A B A B A B A B A B A B A B A B A B A B
1 MYL9-Hs00382913_m1 Core down 34.0 36.1 33.8 33.1 29.1 31.9 27.7 27.4 30.7 29.9 28.8 29.9 34.1 33.6 35.1 34.8 31.1 32.2 34.2 33.9 28.3 28.6 31.9 31.4
2 CHCHD10-Hs01369775_g1 Core up 26.2 27.2 26.9 27.3 25.8 25.9 26.1 26.0 26.5 25.6 25.2 26.3 26.6 26.7 26.5 26.5 27.0 26.6 27.6 26.7 25.3 24.8 24.6 24.0
3 RGS5-Hs00186212_m1 Core down 31.0 31.6 32.1 32.9 30.0 29.9 31.3 31.5 30.3 30.0 30.2 32.3 30.0 30.1 30.0 30.5 30.1 30.6 36.2 33.0 30.4 31.4 31.1 30.1
4 ST6GALNAC5-
Core down 29.7 31.1 25.5 26.1 32.9 30.3 33.1 40.0 35.9 34.3 29.9 31.4 33.5 33.7 27.1 27.0 28.8 28.6 36.0 35.9 30.9 33.5 34.5 33.2
Hs00229612_m1
5 CEBPB-Hs00270923_s1 Core up 30.6 30.4 30.0 30.7 28.2 27.8 29.9 30.2 29.8 29.5 29.7 30.5 29.9 29.7 29.1 29.2 29.7 29.9 32.2 31.5 28.4 28.7 28.1 28.2
6 C5orf13-Hs00854282_g1 Core down 29.1 28.1 27.2 27.8 28.2 25.2 21.6 24.4 26.3 26.7 40.0 29.4 26.4 25.0 40.0 26.3 40.0 27.3 26.9 40.0 26.1 28.0 28.7 28.2
7 PDGFRA-Hs00998026_m1 Marker 24.7 25.0 22.5 22.3 25.3 27.3 24.9 24.4 26.5 26.0 23.8 26.9 23.6 23.6 27.5 27.8 27.3 28.9 26.7 26.2 25.2 28.6 26.5 25.9
8 CCND2-Hs00922419_g1 Core down 35.0 30.5 21.6 22.0 27.9 27.4 30.0 32.4 27.4 26.8 24.5 26.0 22.1 22.0 20.7 20.9 20.6 21.2 25.1 25.3 32.4 32.0 32.4 31.6
9 NKX2-1-Hs00163037_m1 Core down 31.1 31.5 35.3 35.3 40.0 34.8 35.8 40.0 40.0 40.0 40.0 40.0 40.0 40.0 31.2 31.7 32.2 31.0 34.7 36.9 40.0 40.0 40.0 40.0
10 CTSC-Hs00175188_m1 Core up 21.9 23.2 24.2 25.5 21.6 20.7 26.0 25.5 22.5 22.2 20.5 22.1 24.0 23.6 24.4 24.5 24.3 24.5 23.9 22.8 21.2 22.8 24.3 23.1
11 18S-Hs99999901_s1 Norm 12.9 13.1 12.0 12.4 11.9 12.2 11.7 13.0 12.3 11.9 11.7 12.2 12.1 11.9 11.4 11.9 11.6 12.0 14.5 13.8 12.0 13.8 11.8 11.4
12 DNER-Hs00294564_m1 Core down 31.4 28.4 27.3 26.4 32.7 32.4 27.7 27.2 26.6 26.7 25.3 27.9 24.2 24.0 23.9 24.5 25.0 26.1 31.3 28.6 27.0 30.3 27.9 27.0
13 LMO4-Hs01086790_m1 Core up 24.5 25.3 21.9 22.3 23.4 23.6 24.4 24.1 24.4 23.8 23.3 25.4 22.8 22.5 22.4 22.6 22.9 22.9 25.7 24.7 23.7 24.1 24.4 23.2
14 PLCH1-Hs00392783_m1 Core down 26.0 26.3 27.1 27.7 27.0 28.7 28.1 28.5 31.6 30.5 29.4 29.9 29.6 29.3 27.3 27.4 27.5 27.0 32.5 31.5 28.8 29.6 31.2 31.0
15 SPARCL1-Hs00949886_m1 Core down 26.9 25.6 26.4 27.4 26.2 28.9 25.1 24.0 27.5 27.3 24.8 29.3 24.3 24.0 26.5 26.4 27.0 28.9 31.2 31.3 27.9 30.5 26.3 25.2
16 SYNM-Hs00322391_m1 Core down 34.0 33.4 27.9 28.2 27.7 27.6 27.2 28.3 36.0 40.0 29.4 29.3 40.0 36.1 31.9 33.3 31.7 32.0 40.0 40.0 36.8 40.0 29.6 29.0
17 FBLN2-Hs00157482_m1 Core down 40.0 35.5 31.0 31.9 30.1 30.7 36.7 35.9 32.8 33.9 34.3 36.2 34.1 34.6 34.3 34.5 33.9 35.4 34.3 35.4 36.5 34.7 28.7 28.6
18 SALL2-Hs00413788_m1 Core down 29.9 28.1 26.6 26.3 34.3 36.9 35.1 33.2 32.0 31.0 29.5 33.1 26.3 26.2 26.6 26.5 26.8 27.2 30.2 28.6 40.0 40.0 33.9 33.9
19 TUSC3-Hs00185147_m1 Core down 23.2 23.5 40.0 40.0 24.4 25.0 34.1 32.9 37.0 35.0 40.0 40.0 24.9 24.9 40.0 40.0 30.0 30.2 40.0 40.0 32.0 28.8 26.5 25.3
20 GPR158-Hs00393109_m1 Core down 30.6 30.0 27.9 28.2 40.0 40.0 32.3 32.7 31.1 31.6 28.2 30.4 30.5 31.2 28.7 29.3 29.1 29.2 28.2 28.9 31.4 33.3 31.0 30.9
21 PEG3-Hs00377844_m1 Core down 28.9 28.9 23.6 24.9 32.9 33.9 27.8 28.5 27.3 27.3 27.6 30.1 25.2 25.6 40.0 40.0 29.1 32.0 29.5 28.7 29.8 31.4 40.0 40.0
22 FUT8-Hs00189535_m1 Core up 27.4 28.2 25.1 25.1 27.0 27.2 27.1 27.3 26.0 26.0 27.3 28.5 26.2 26.3 26.5 26.2 26.1 26.5 28.3 27.4 25.0 26.0 27.3 26.2
23 HLA-DRA-Hs00219575_m1 Core up 27.0 26.1 33.4 35.0 35.3 40.0 27.9 27.7 22.4 22.9 23.7 24.5 24.0 23.9 33.7 32.0 28.0 30.5 24.4 22.8 28.3 26.6 22.2 21.1
24 ASCL1-Hs00269932_m1 Marker 36.9 30.7 23.1 23.4 28.7 30.4 29.4 33.9 35.4 40.0 36.4 40.0 23.4 23.4 21.9 22.5 22.2 22.6 28.1 26.5 35.1 40.0 31.6 31.3
25 PI15-Hs00210658_m1 Core down 28.9 30.8 32.3 28.9 30.6 32.2 28.9 26.6 27.5 27.6 27.8 26.9 31.1 30.6 36.0 37.0 27.7 34.6 32.6 34.1 27.4 30.0 28.2 26.7
26 EPDR1-Hs00378148_m1 Core up 27.6 29.2 30.2 30.8 28.5 29.2 27.6 28.2 26.9 25.8 26.9 28.0 27.9 27.8 29.2 29.1 29.1 28.9 30.8 29.8 27.1 26.8 29.5 28.6
27 DTX4-Hs00392288_m1 Core down 27.8 27.7 28.5 28.0 32.8 33.8 30.4 30.5 30.0 29.4 34.5 34.2 29.8 29.5 26.4 26.4 26.5 27.6 30.4 30.7 31.0 31.7 28.5 28.1
28 CXXC4-Hs00228693_m1 Core down 29.7 28.2 24.3 24.7 29.1 30.9 28.2 29.9 36.9 40.0 34.9 40.0 26.9 26.7 25.4 26.1 25.4 26.9 30.7 29.3 30.4 32.8 36.9 35.0
29 IRX2-Hs01383002_m1 Core down 40.0 40.0 40.0 40.0 28.4 28.7 36.4 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 37.1 34.9 40.0 40.0 40.0 40.0 30.5 30.4
30 MAP6-Hs01023152_s1 Core down 27.9 28.3 26.0 26.7 28.3 29.6 30.9 29.7 30.4 30.8 32.4 32.6 28.4 28.2 29.1 29.4 28.6 29.8 33.7 31.8 30.7 33.7 27.5 26.5
31 SIX3-Hs00193667_m1 Core down 33.0 34.0 40.0 40.0 29.9 30.5 34.5 34.8 36.0 35.9 40.0 40.0 35.1 36.4 34.7 36.1 40.0 34.9 31.6 30.8 33.7 33.1 40.0 40.0
32 NKX2-2-Hs00159616_m1 Marker 28.7 28.4 26.5 27.8 27.5 27.9 26.4 26.8 25.5 24.8 25.8 27.4 24.9 24.6 25.0 25.3 24.8 24.5 27.1 25.9 28.4 28.8 31.0 31.0
33 S100A6-Hs00170953_m1 Core up 18.5 20.1 29.2 27.3 18.3 17.8 19.9 20.0 20.3 19.2 21.4 21.2 21.1 20.9 23.0 22.9 22.7 21.8 24.0 23.8 17.5 17.9 19.3 18.3
34 MMP17-Hs01108847_m1 Core up 28.4 29.6 30.3 30.1 28.6 27.7 28.9 28.8 25.1 24.9 30.1 31.3 26.2 25.9 29.4 28.8 29.2 29.2 29.2 26.4 26.6 26.4 28.6 28.4
35 FAM38B-Hs00926225_m1 Core down 29.0 29.4 40.0 40.0 36.9 36.4 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 32.7 31.6
36 OLIG2-Hs00300164_s1 Marker 34.9 29.1 22.1 22.8 26.8 27.5 25.4 26.1 23.1 22.9 23.1 24.9 21.8 21.2 22.6 22.8 22.6 22.7 24.8 23.6 31.0 33.2 30.5 30.5
37 PMEPA1-Hs00375306_m1 Core up 24.1 25.1 27.9 27.6 25.5 26.4 22.4 23.0 25.4 24.3 26.7 27.3 25.2 24.6 27.1 26.9 27.4 26.9 27.3 27.7 22.6 25.0 26.0 25.7
38 HOXD10-Hs00157974_m1 Core up 28.5 28.5 40.0 40.0 27.7 28.6 29.4 29.5 29.5 28.9 29.5 31.2 28.9 28.6 28.8 28.9 28.8 29.0 30.7 28.8 26.5 27.0 27.9 27.2
39 NDN-Hs00267349_s1 Core down 23.8 24.2 29.4 30.1 40.0 40.0 25.9 25.8 25.9 25.7 27.1 27.1 36.9 40.0 25.4 25.1 25.5 25.0 28.6 27.0 35.7 40.0 40.0 35.2
40 FOXG1-Hs01850784_s1 Core up 24.4 24.2 22.9 23.0 24.6 25.3 24.3 25.3 25.5 24.8 26.3 26.3 23.5 23.5 21.4 21.6 21.4 21.8 26.3 24.7 23.8 25.5 21.8 21.3
41 MN1-Hs00159202_m1 Core down 32.2 34.9 28.0 27.9 31.8 33.4 26.4 27.3 30.5 28.9 33.7 35.5 28.0 27.1 26.5 27.2 26.8 26.4 31.6 30.1 34.7 35.0 31.5 30.8
42 NDUFB10-Hs00605903_m1 Norm 25.0 25.6 24.5 25.2 25.3 25.3 25.1 25.7 26.0 25.4 24.7 25.9 25.3 24.9 25.2 25.5 25.4 25.2 28.5 26.6 24.5 25.0 26.4 25.3
43 HMGA2-Hs00171569_m1 Core down 29.9 32.0 33.0 33.2 27.5 27.2 35.4 35.9 32.4 31.8 32.1 33.6 32.4 31.5 33.1 32.8 32.6 32.3 35.8 33.8 31.3 32.1 29.4 28.7
44 MMRN1-Hs00201182_m1 Core down 33.7 31.6 40.0 40.0 33.0 34.6 40.0 40.0 40.0 40.0 32.0 31.5 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 33.7 35.0 33.6 32.2
45 RAB6B-Hs00981572_m1 Core down 28.9 30.7 29.8 29.3 33.2 33.0 27.9 28.0 29.5 29.0 30.7 32.4 28.0 27.4 26.6 26.8 26.7 27.1 31.7 30.2 29.6 30.5 31.0 29.9
46 NTRK2-Hs00178811_m1 Core up 24.0 24.6 29.5 29.7 31.6 32.6 33.2 33.1 23.4 23.4 23.3 26.3 24.3 24.0 34.6 37.0 27.5 31.8 26.4 25.1 22.5 24.8 28.1 27.4
47 KALRN-Hs00610179_m1 Core down 30.5 30.6 27.5 27.4 30.3 31.8 28.5 28.3 30.7 30.1 30.6 31.1 27.2 27.5 27.4 27.5 26.7 27.4 32.2 30.8 27.9 30.9 31.2 31.0
48 CD74-Hs00269961_m1 Core up 27.5 26.4 32.1 35.1 32.7 34.2 27.9 28.2 22.2 23.0 24.2 25.6 24.7 24.4 36.5 33.5 28.3 30.3 24.9 24.7 28.7 26.6 22.3 22.0
49 LUM-Hs00158940_m1 Core down 30.1 31.9 40.0 40.0 23.4 26.0 32.8 27.5 40.0 40.0 31.4 31.2 40.0 40.0 32.3 31.2 34.9 40.0 40.0 40.0 28.8 27.5 30.4 29.1
50 NELL2-Hs00196254_m1 Core down 33.1 33.2 33.2 29.0 31.6 32.1 29.8 29.6 36.9 37.0 32.5 34.4 30.8 31.4 27.0 27.4 28.2 29.1 30.1 28.0 29.1 31.8 32.2 31.1
259

A.3 Quantitative RT-PCR Appendix
Well Applied Biosystems assay ID Gene G19 G19 G21 G21 G23 G23 G24 G24 G25 G25 G26 G26 G30 G30 G31 G31 G32 G32 G144 G144 G166 G166 G179 G179
Category A B A B A B A B A B A B A B A B A B A B A B A B
51 MAP6-Hs01929835_s1 Core down 25.7 26.5 25.0 25.4 28.9 30.1 29.7 28.7 29.5 30.3 31.4 31.9 26.9 27.0 29.7 29.8 28.3 30.0 34.8 31.0 30.9 33.9 27.4 26.4
52 PDE1C-Hs01095694_m1 Core up 24.4 25.4 27.3 28.5 24.9 24.1 23.6 23.6 24.4 23.6 25.5 25.9 26.5 25.8 28.0 27.9 28.0 27.8 27.5 26.4 22.9 23.5 26.3 25.7
53 SULF2-Hs00378697_m1 Core up 23.6 24.8 26.1 25.7 24.0 26.3 22.6 22.4 24.6 23.8 23.6 25.7 23.2 22.5 24.9 24.4 24.9 25.1 27.5 24.7 25.2 25.4 26.8 26.2
54 TUBB-Hs00962420_g1 Norm 20.4 21.8 20.0 20.5 21.2 21.3 21.0 21.7 21.8 20.8 20.9 22.1 20.7 20.5 19.7 20.0 19.9 19.8 23.8 22.4 19.7 20.0 23.2 22.8
55 LMO2-Hs00277106_m1 Core down 30.3 30.5 26.3 25.9 31.1 35.5 30.8 30.3 34.3 34.0 27.3 30.5 25.9 26.8 26.0 25.9 25.8 25.4 32.6 32.6 33.3 35.4 29.8 29.0
56 MAN1C1-Hs00220595_m1 Core up 27.6 27.3 24.9 24.9 31.8 34.1 29.3 29.2 27.2 26.3 24.2 26.2 25.3 25.2 24.9 24.3 25.3 24.8 28.9 27.4 27.1 27.3 26.9 26.7
57 TAGLN-Hs00162558_m1 Core down 28.1 30.1 40.0 33.0 24.7 27.2 21.0 21.5 26.5 25.0 29.1 29.3 25.7 24.4 32.7 30.4 28.5 28.7 29.9 31.4 30.4 29.7 27.7 26.8
58 RTN1-Hs00382515_m1 Core down 24.8 25.3 32.7 32.5 27.5 28.6 29.3 29.1 32.6 31.9 24.9 26.5 25.5 25.2 28.0 27.1 29.0 29.6 29.5 29.3 28.1 30.7 31.6 30.7
59 MAF-Hs00193519_m1 Core down 27.9 28.0 27.8 28.6 27.2 30.1 28.0 28.7 31.4 31.5 26.7 28.8 27.5 27.3 26.9 27.4 27.7 26.8 33.9 33.3 27.1 28.5 35.5 34.6
60 NPTX2-Hs00383983_m1 Core down 29.8 28.3 28.1 31.8 28.7 29.2 34.1 37.0 40.0 34.7 27.2 26.8 31.9 31.2 30.3 28.7 33.4 29.5 34.2 33.3 27.4 29.2 32.6 32.4
61 EDA2R-Hs00939736_m1 Core down 27.2 27.8 27.9 28.2 28.9 26.7 28.0 28.1 35.5 34.8 32.0 32.5 28.5 28.2 34.5 34.4 29.8 31.5 37.0 34.0 31.0 30.1 35.3 34.5
62 SOX10-Hs00366918_m1 Marker 40.0 40.0 40.0 40.0 40.0 40.0 36.5 40.0 24.7 24.9 40.0 40.0 35.1 35.0 40.0 40.0 40.0 40.0 27.1 25.7 40.0 40.0 40.0 40.0
63 SEMA6A-Hs00221174_m1 Core down 26.6 26.3 22.4 23.3 25.6 26.1 23.6 24.8 27.7 27.2 24.6 26.8 23.9 23.8 22.4 22.9 22.8 23.0 27.2 26.2 25.3 27.1 28.0 27.7
64 DDIT3-Hs00358796_g1 Core up 26.7 27.2 27.1 27.7 24.9 22.3 27.1 27.2 27.1 25.9 27.6 29.0 27.0 26.6 21.0 21.1 21.0 21.1 25.0 24.4 24.2 23.7 27.2 26.1
65 CA12-Hs01080909_m1 Core down 30.1 28.6 22.4 23.3 25.5 26.5 31.3 30.4 27.9 26.9 23.5 26.3 26.7 27.8 23.9 23.8 24.0 23.2 28.3 26.3 25.9 26.8 26.3 26.1
66 C9orf125-Hs00260558_m1 Core down 30.7 31.6 40.0 40.0 34.5 37.0 28.7 28.9 32.1 31.9 29.8 31.6 27.0 27.3 40.0 40.0 36.5 36.6 32.0 30.4 31.2 33.9 35.8 34.9
67 SDC2-Hs00299807_m1 Core down 26.0 27.0 35.1 36.0 25.9 26.2 25.2 25.6 25.9 24.9 25.8 26.7 29.0 28.5 28.0 27.6 28.8 28.9 29.2 29.0 26.9 28.5 26.1 25.4
68 SLIT2-Hs00191193_m1 Core down 25.1 24.2 33.3 31.9 25.9 26.1 32.3 30.9 33.8 32.3 31.7 32.2 33.3 34.2 28.9 29.0 26.8 27.2 32.9 33.3 32.1 33.3 31.4 30.7
69 MT2A-Hs02379661_g1 Core up 22.6 23.0 21.7 21.5 20.9 21.9 22.8 23.1 24.0 22.5 20.0 21.9 21.8 22.0 21.4 21.4 21.4 21.7 24.3 22.9 18.8 18.5 20.0 18.9
70 IL17RD-Hs00296982_m1 Core down 26.3 27.0 25.5 26.6 23.7 24.4 26.3 27.0 28.0 27.0 26.6 28.5 26.1 25.9 26.2 26.4 26.2 26.6 29.0 27.8 28.0 28.1 28.0 27.6
71 PDZRN3-Hs00392900_m1 Core down 29.0 28.3 26.4 26.7 33.9 40.0 35.8 40.0 26.7 26.0 27.7 28.6 28.6 28.9 26.3 26.3 26.1 26.0 33.1 32.9 30.8 31.6 29.0 29.0
72 LGALS3-Hs00173587_m1 Core up 24.8 24.8 24.3 24.6 23.2 23.3 26.6 25.6 24.7 24.6 22.5 24.1 24.4 24.3 25.4 24.8 24.4 24.6 27.8 25.9 24.1 23.1 24.5 23.3
73 FAM69A-Hs00961685_m1 Core up 25.0 24.8 23.7 25.2 24.9 27.7 25.3 24.0 25.7 25.5 24.3 25.9 26.4 26.3 25.2 25.4 25.0 25.0 28.1 26.5 26.1 26.0 27.4 26.0
74 TERT-Hs00972656_m1 Marker 31.5 33.0 28.1 26.8 30.4 30.2 31.4 31.7 31.4 30.4 30.5 32.0 29.8 29.3 29.8 29.6 29.5 29.2 31.9 29.4 29.2 29.8 34.7 34.8
75 PLS3-Hs00418605_g1 Core up 21.6 22.7 24.4 25.4 22.8 22.9 21.6 22.8 24.2 23.3 22.5 23.9 24.9 24.4 24.4 24.8 24.8 24.3 26.5 24.6 21.0 22.4 26.3 25.1
76 PTEN-Hs02621230_s1 Core down 28.9 29.7 32.9 30.3 27.9 28.6 28.6 28.8 29.9 29.9 30.3 31.4 28.1 28.0 30.4 30.9 30.4 30.7 40.0 40.0 29.1 31.1 37.0 34.9
77 SALL2-Hs00826674_m1 Core down 33.8 32.6 31.7 33.1 33.6 37.0 33.4 32.5 33.1 32.7 32.8 34.6 32.5 31.7 31.0 31.5 31.1 31.1 35.3 34.6 32.4 31.7 31.4 30.8
78 KCTD12-Hs00540818_s1 Core down 31.9 32.0 29.5 32.0 30.2 31.8 25.8 26.3 27.9 28.1 28.2 29.3 27.7 27.6 29.9 31.1 29.4 30.8 30.8 31.0 29.4 30.0 26.7 26.6
79 TES-Hs00210319_m1 Core down 40.0 40.0 40.0 40.0 34.2 36.5 34.1 36.8 34.1 34.2 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 40.0 36.2 36.1 35.8 33.5
80 SOX2-Hs01053049_s1 Marker 23.7 23.9 22.0 22.7 24.5 24.5 24.2 24.7 23.5 23.5 22.8 24.9 22.7 22.5 22.2 22.4 22.0 22.5 25.9 25.1 24.5 26.9 22.7 21.9
81 LPAR6-Hs00271758_s1 Core down 29.6 29.9 28.9 29.3 29.7 31.2 28.7 29.1 33.3 32.9 30.5 32.3 28.2 28.0 29.0 28.7 29.0 28.7 34.1 32.9 28.8 30.2 32.0 31.4
82 ODZ2-Hs00393060_m1 Core down 25.6 26.4 26.7 28.6 24.8 25.9 28.1 26.9 31.9 31.0 24.8 26.9 30.6 30.3 27.1 26.3 27.9 27.1 29.6 29.4 30.0 32.2 28.0 27.6
83 NNMT-Hs00196287_m1 Core up 27.9 28.3 30.0 30.6 23.8 27.1 32.7 30.3 28.2 26.9 23.7 24.6 27.3 27.1 29.0 28.1 29.9 28.3 27.5 28.0 23.0 23.9 24.3 23.5
84 CACNG8-Hs01100182_m1 Core down 31.7 30.9 27.3 28.6 30.2 29.8 30.5 37.0 33.6 32.9 33.7 34.4 27.0 26.9 30.9 32.6 29.6 29.9 30.3 28.9 30.4 31.8 32.8 33.4
85 PRSS12-Hs00186221_m1 Core up 24.2 25.7 30.3 33.5 24.9 24.5 31.0 31.8 27.8 26.6 25.3 26.9 40.0 37.0 27.4 27.8 26.9 26.8 27.3 25.8 23.3 24.3 31.5 31.3
86 FOXJ1-Hs00230964_m1 Core down 32.5 30.5 30.9 30.9 33.5 36.4 30.6 30.9 30.2 30.8 29.6 31.3 27.9 27.6 30.5 30.7 29.9 30.4 32.9 32.1 35.2 40.0 31.9 31.9
87 NTN1-Hs00180355_m1 Core down 29.2 29.0 26.6 26.2 34.2 35.6 25.4 25.0 28.0 28.0 28.1 30.5 27.1 26.8 25.5 25.9 25.7 25.6 28.0 27.9 30.5 35.0 29.4 30.0
88 LMO3-Hs00375237_m1 Core down 40.0 40.0 32.1 34.8 31.7 33.6 40.0 40.0 37.0 40.0 28.1 28.9 32.1 30.6 27.0 27.9 27.4 28.6 40.0 40.0 40.0 40.0 35.9 34.3
89 CHI3L1-Hs00609691_m1 Marker 31.2 25.9 29.9 27.8 33.1 36.9 24.3 24.0 29.3 30.7 23.8 27.4 21.8 22.8 27.6 27.6 28.3 29.2 40.0 37.0 23.0 28.2 22.3 21.8
90 CD9-Hs01124025_g1 Core up 23.1 24.0 22.4 23.5 24.4 24.3 25.6 25.0 22.4 22.5 22.9 24.6 22.5 22.3 22.6 22.9 22.6 23.0 25.2 23.9 22.3 23.0 24.2 23.0
91 ADD2-Hs00242289_m1 Core up 26.5 27.7 24.0 25.2 26.5 25.2 26.7 27.8 28.6 27.9 30.3 32.1 25.8 25.8 24.9 24.9 24.8 24.8 28.9 27.7 25.6 25.4 29.0 28.8
92 MYL9-Hs00697086_m1 Core down 32.6 33.0 31.4 31.3 27.3 29.7 22.8 23.3 28.6 27.7 26.2 28.0 28.4 27.3 33.8 32.2 30.4 30.8 33.0 32.8 25.7 26.1 27.3 26.9
93 LYST-Hs00179814_m1 Core up 29.8 30.3 29.8 30.6 29.6 28.4 29.3 29.0 28.9 29.2 29.3 31.6 28.3 28.7 28.8 29.2 29.2 29.9 30.1 29.4 28.1 28.0 28.9 27.9
94 LAMA2-Hs00166308_m1 Core down 29.9 29.4 36.9 36.9 28.6 29.7 40.0 36.0 31.5 33.0 32.2 32.5 30.6 31.2 29.6 28.8 35.3 34.7 29.4 28.2 32.7 34.0 34.0 35.0
95 PLA2G4A-Hs00233352_m1 Core up 28.1 29.2 29.3 28.9 29.4 29.4 30.1 29.8 27.1 27.0 31.1 32.1 26.5 26.1 26.5 26.2 26.4 26.0 29.0 28.4 28.6 27.3 29.7 28.9
96 BACE2-Hs00273238_m1 Core up 24.8 25.6 34.7 36.9 29.4 30.4 26.4 26.5 26.0 25.5 26.6 29.2 27.4 27.7 29.9 29.8 29.6 30.0 28.9 27.5 25.8 25.6 27.7 27.1
260

A.3 Quantitative RT-PCR Appendix
Table A.4: Normalised Ct values. Abbreviations: "down" for down-regulated, "up" for up-regulated, and "Norm" for Normalisation control.
Well Applied Biosystems assay ID Gene CB130 CB130 CB130 CB152 CB152 CB152 CB152 CB171 CB171 CB171 CB192 CB541 CB660 CB660 G2 G2 G7 G7 G7 G7 G9 G9 G14 G14
Category A B C A B C D A B C A B A B A B C D A B A B
1 MYL9-Hs00382913_m1 Core down 28.2 27.9 28.2 23.0 25.2 25.6 25.3 29.4 30.2 29.5 26.8 24.9 27.2 26.7 36.1 33.4 31.9 36.0 30.5 29.4 26.3 26.6 35.0 35.2
2 CHCHD10-Hs01369775_g1 Core up 27.1 25.7 26.1 25.9 25.3 24.6 24.5 27.2 26.5 26.1 28.8 29.7 29.3 29.5 24.7 23.1 26.2 25.7 25.6 24.7 25.5 26.3 27.1 26.9
3 RGS5-Hs00186212_m1 Core down 31.3 30.4 30.4 29.2 31.7 30.4 29.3 32.9 29.5 30.8 32.6 27.4 26.5 27.6 30.4 29.4 31.5 30.9 28.4 27.6 30.5 31.8 32.4 32.1
4 ST6GALNAC5-
Core down 28.9 28.8 28.6 33.5 33.7 32.4 31.9 29.8 30.5 29.5 25.6 26.7 23.4 24.9 35.2 33.3 36.4 36.8 34.9 34.7 28.6 28.9 31.1 29.9
Hs00229612_m1
5 CEBPB-Hs00270923_s1 Core up 33.1 30.7 30.6 30.3 31.8 31.2 31.3 32.9 30.3 31.3 30.7 32.0 32.6 31.3 31.6 32.7 29.1 29.8 29.8 29.6 29.6 30.2 30.7 30.1
6 C5orf13-Hs00854282_g1 Core down 28.5 27.7 28.1 25.5 27.3 26.4 25.6 28.5 25.2 26.6 29.1 24.8 25.2 26.2 36.1 25.2 30.3 29.0 22.8 26.9 27.3 36.2 27.8 31.5
7 PDGFRA-Hs00998026_m1 Marker 37.4 37.4 34.8 33.2 36.8 27.9 28.4 37.5 35.4 36.9 32.4 35.7 29.9 32.1 24.2 24.3 23.2 22.2 21.5 22.5 24.0 24.1 25.9 26.2
8 CCND2-Hs00922419_g1 Core down 32.3 30.8 31.0 31.6 33.7 28.4 28.5 32.4 31.1 30.9 25.1 24.1 24.1 24.7 22.5 21.3 22.7 22.0 23.9 23.2 30.4 29.9 34.1 37.3
9 NKX2-1-Hs00163037_m1 Core down 37.4 37.4 37.1 37.1 36.8 36.4 37.2 37.5 35.4 36.9 37.0 28.8 29.0 29.8 36.1 34.4 32.8 34.6 32.2 32.4 36.7 36.2 31.2 31.3
10 CTSC-Hs00175188_m1 Core up 23.5 24.6 24.6 24.6 24.0 24.7 24.3 23.5 24.3 24.0 23.9 25.1 25.5 23.7 20.6 20.0 25.3 24.4 26.5 25.1 23.0 23.5 23.4 23.2
11 18S-Hs99999901_s1 Norm 13.7 12.0 11.5 12.2 13.8 10.9 11.3 13.4 10.8 11.2 12.5 12.8 12.1 12.1 14.0 16.4 12.5 12.8 10.4 11.6 12.6 14.0 12.7 12.6
12 DNER-Hs00294564_m1 Core down 32.3 32.3 32.8 32.7 31.5 28.1 27.5 32.2 29.9 32.0 30.4 22.3 22.4 22.7 25.5 23.3 27.8 25.2 23.5 25.0 26.5 24.0 32.5 31.5
13 LMO4-Hs01086790_m1 Core up 27.0 26.8 26.9 26.7 27.5 26.6 25.4 26.9 26.3 27.1 25.7 26.1 25.5 26.7 22.5 21.3 22.2 22.4 24.0 23.7 24.4 25.7 25.1 25.0
14 PLCH1-Hs00392783_m1 Core down 26.5 26.3 26.2 26.4 26.8 27.3 26.3 26.2 26.4 27.5 26.8 27.0 27.8 28.7 32.6 32.2 29.5 29.9 29.1 28.9 27.9 29.1 27.2 26.8
15 SPARCL1-Hs00949886_m1 Core down 33.6 31.2 31.0 27.0 29.5 24.1 22.2 35.6 28.3 30.3 28.5 25.3 22.8 25.6 24.5 23.1 24.4 27.6 29.1 29.7 22.4 25.4 28.9 28.1
16 SYNM-Hs00322391_m1 Core down 29.3 28.2 28.1 26.8 27.3 27.3 25.7 28.6 27.3 28.6 27.1 27.6 27.1 27.2 28.6 27.7 29.0 29.5 28.7 28.6 25.2 27.8 33.7 34.5
17 FBLN2-Hs00157482_m1 Core down 37.4 36.2 37.1 36.0 36.8 36.4 35.1 37.5 33.6 35.5 33.8 29.1 27.5 29.2 33.5 33.4 35.4 32.7 36.2 35.3 33.2 36.2 36.8 36.2
18 SALL2-Hs00413788_m1 Core down 37.4 33.4 33.5 31.0 36.7 30.1 27.2 37.4 28.7 33.1 32.8 26.5 24.5 27.9 30.5 31.2 28.1 28.5 28.9 30.6 34.6 36.2 31.6 30.3
19 TUSC3-Hs00185147_m1 Core down 24.4 22.8 23.0 23.3 23.6 22.7 21.8 24.4 23.5 23.3 24.4 23.4 23.4 23.9 36.1 34.4 35.7 33.7 36.3 34.2 23.3 22.9 24.0 23.8
20 GPR158-Hs00393109_m1 Core down 34.8 34.5 34.9 37.0 36.3 31.4 32.6 36.0 34.5 33.3 26.9 25.6 27.1 26.6 26.3 26.4 28.2 26.6 27.9 26.6 34.6 34.8 32.6 32.3
21 PEG3-Hs00377844_m1 Core down 31.0 31.4 31.2 30.0 30.7 30.3 29.4 30.4 29.6 30.9 29.8 28.1 25.2 27.9 27.2 27.0 26.5 24.6 25.6 26.3 33.3 35.8 29.0 29.0
22 FUT8-Hs00189535_m1 Core up 26.6 26.7 26.8 26.3 26.7 27.4 26.4 27.0 27.1 27.1 26.4 27.9 28.4 27.1 25.2 24.0 27.9 27.3 26.5 27.3 25.6 26.5 28.5 28.3
23 HLA-DRA-Hs00219575_m1 Core up 27.8 26.5 26.4 28.3 29.8 25.3 22.3 29.1 25.4 27.4 22.3 27.2 30.8 26.4 20.5 20.6 24.7 24.0 23.9 23.6 21.1 25.2 27.6 28.0
24 ASCL1-Hs00269932_m1 Marker 34.3 32.4 31.9 32.8 35.7 27.5 26.3 35.0 30.5 35.0 32.7 29.4 27.1 34.4 23.5 22.4 25.9 24.1 23.7 25.8 36.7 36.2 34.3 37.4
25 PI15-Hs00210658_m1 Core down 32.0 30.7 30.3 23.1 28.3 26.5 24.6 32.3 27.0 30.9 29.6 25.0 24.5 24.8 25.4 26.0 32.3 27.4 29.1 30.0 26.6 32.9 32.0 30.5
26 EPDR1-Hs00378148_m1 Core up 29.0 28.3 28.8 26.6 27.3 28.8 26.9 29.5 28.9 29.0 30.8 29.0 30.0 30.0 26.0 25.2 28.6 28.8 28.7 28.2 26.6 28.1 28.8 28.8
27 DTX4-Hs00392288_m1 Core down 33.2 31.1 31.3 28.3 31.4 28.3 27.2 33.6 27.4 30.2 32.9 26.5 25.0 26.7 33.3 34.4 30.8 30.6 31.3 33.3 29.5 32.3 28.4 28.0
28 CXXC4-Hs00228693_m1 Core down 31.3 31.2 30.8 34.2 36.8 30.2 29.6 32.1 29.7 33.1 28.3 25.9 25.7 27.9 27.7 24.9 27.4 26.5 25.5 27.5 36.7 36.2 29.3 30.3
29 IRX2-Hs01383002_m1 Core down 30.4 30.9 30.7 29.3 28.9 30.8 30.1 29.7 30.3 31.1 30.7 26.6 26.6 26.3 28.9 27.3 37.4 36.8 36.3 37.7 29.9 29.5 37.9 37.4
30 MAP6-Hs01023152_s1 Core down 28.0 25.1 25.4 25.9 27.4 24.1 22.0 27.5 24.1 24.7 28.1 26.9 25.8 28.0 25.9 24.1 27.4 26.7 26.8 27.2 28.3 29.8 29.2 29.1
31 SIX3-Hs00193667_m1 Core down 37.4 37.4 37.1 37.1 36.8 36.4 34.8 37.5 35.4 36.9 37.0 25.5 24.0 25.0 36.1 32.6 32.5 33.6 29.8 32.3 32.3 32.9 35.1 34.7
32 NKX2-2-Hs00159616_m1 Marker 37.4 37.4 37.1 37.1 36.8 32.6 33.8 37.5 35.4 36.9 37.0 28.6 30.8 29.6 25.0 23.7 27.2 26.1 28.2 30.0 26.5 26.4 28.9 28.2
33 S100A6-Hs00170953_m1 Core up 20.0 20.4 20.8 20.9 18.9 20.0 20.8 19.5 21.7 20.5 21.9 21.1 25.7 23.0 18.8 17.3 21.7 22.7 21.6 19.8 19.5 17.8 19.7 19.2
34 MMP17-Hs01108847_m1 Core up 28.3 28.7 28.6 28.4 27.5 29.7 29.1 28.0 29.3 30.0 28.6 32.3 30.3 29.9 25.7 27.4 30.4 29.8 29.1 30.8 27.4 26.4 28.9 28.5
35 FAM38B-Hs00926225_m1 Core down 37.4 37.4 37.1 37.1 36.8 36.4 37.2 37.5 35.4 36.9 37.0 24.3 27.0 26.1 36.1 34.4 37.4 36.8 36.3 37.7 24.4 24.5 29.3 29.1
36 OLIG2-Hs00300164_s1 Marker 30.2 33.1 33.2 30.1 30.0 28.3 28.4 32.5 34.3 35.4 29.2 26.1 28.1 27.8 23.4 23.9 22.2 22.0 21.5 23.4 30.2 32.0 31.4 36.4
37 PMEPA1-Hs00375306_m1 Core up 26.0 24.0 24.1 24.8 24.9 25.2 26.0 26.1 26.5 25.9 27.6 26.5 29.1 27.2 25.5 25.3 27.7 27.7 27.5 27.3 25.9 24.3 24.0 24.6
38 HOXD10-Hs00157974_m1 Core up 37.4 37.4 37.1 37.1 36.8 32.5 32.4 37.5 35.4 36.9 37.0 37.5 37.1 37.9 29.1 27.4 27.4 27.7 28.1 27.4 28.0 28.1 28.7 28.7
39 NDN-Hs00267349_s1 Core down 25.1 25.0 25.2 24.2 24.1 24.7 23.4 25.0 25.3 25.1 25.4 23.8 23.9 24.7 25.0 23.1 25.0 25.4 25.9 26.1 23.4 24.1 24.7 24.3
40 FOXG1-Hs01850784_s1 Core up 37.4 37.4 37.1 33.7 34.7 25.3 25.3 37.5 31.1 35.8 35.3 27.8 27.6 28.8 25.0 23.5 23.9 23.9 21.4 21.4 24.0 24.4 24.6 24.4
41 MN1-Hs00159202_m1 Core down 32.3 31.8 31.5 29.4 30.3 31.2 27.8 33.6 31.8 34.1 31.0 26.3 25.9 27.0 28.7 28.5 30.1 28.2 28.8 28.5 34.0 32.3 33.4 33.3
42 NDUFB10-Hs00605903_m1 Norm 25.1 25.5 26.0 25.6 25.0 25.8 25.0 25.5 26.0 25.8 25.4 25.6 26.0 26.0 24.6 22.4 25.5 25.5 26.4 26.2 25.2 24.4 25.0 25.4
43 HMGA2-Hs00171569_m1 Core down 31.8 28.3 28.5 33.6 30.3 29.1 29.7 30.3 28.2 28.8 29.0 27.3 27.1 26.8 31.0 29.5 31.3 30.3 28.9 28.9 34.1 30.5 30.4 31.0
44 MMRN1-Hs00201182_m1 Core down 37.4 37.4 37.1 37.1 36.8 36.4 37.2 37.5 35.4 36.9 37.0 26.4 27.3 28.6 36.1 34.4 37.0 36.2 36.3 36.7 36.7 36.1 33.2 34.3
45 RAB6B-Hs00981572_m1 Core down 34.8 32.4 32.3 32.9 35.1 31.7 29.7 34.6 31.1 32.4 30.4 28.3 28.6 29.0 28.5 28.1 27.6 27.7 27.3 28.4 29.3 29.4 30.2 29.3
46 NTRK2-Hs00178811_m1 Core up 29.3 30.9 31.1 23.4 24.7 25.5 23.1 29.7 28.5 30.9 29.0 29.0 26.9 28.6 21.5 20.8 29.5 24.8 26.9 27.5 22.4 24.2 25.2 24.0
47 KALRN-Hs00610179_m1 Core down 29.9 29.5 30.1 28.1 29.5 30.3 29.0 30.0 28.4 30.0 28.3 26.5 25.9 26.5 27.7 28.5 28.7 27.8 27.4 28.2 26.8 28.9 30.8 30.6
48 CD74-Hs00269961_m1 Core up 27.6 27.7 27.5 28.0 28.7 27.6 24.2 27.8 25.7 28.8 22.2 27.4 30.3 26.3 20.8 21.6 25.7 23.7 25.7 25.5 21.6 26.3 27.8 28.3
261

A.3 Quantitative RT-PCR Appendix
Well Applied Biosystems assay ID Gene CB130 CB130 CB130 CB152 CB152 CB152 CB152 CB171 CB171 CB171 CB192 CB541 CB660 CB660 G2 G2 G7 G7 G7 G7 G9 G9 G14 G14
Category A B C A B C D A B C A B A B A B C D A B A B
49 LUM-Hs00158940_m1 Core down 29.7 29.3 29.3 30.4 31.7 36.3 32.2 32.1 31.8 32.7 29.9 26.0 24.7 25.5 36.1 34.4 32.9 28.8 32.3 31.7 22.7 26.5 30.4 30.0
50 NELL2-Hs00196254_m1 Core down 24.8 24.8 24.8 28.1 29.1 28.5 25.1 24.8 24.0 25.3 25.4 25.0 23.9 25.8 29.6 28.4 30.6 30.5 31.0 31.4 28.4 29.2 35.1 34.4
51 MAP6-Hs01929835_s1 Core down 27.4 25.7 26.2 25.7 26.4 24.4 22.6 27.0 24.4 25.0 27.5 26.6 25.7 28.7 24.9 23.4 27.2 26.2 27.8 28.4 27.0 28.5 27.2 26.9
52 PDE1C-Hs01095694_m1 Core up 30.3 32.8 33.1 27.1 27.6 28.4 28.2 29.7 29.8 30.7 30.4 26.7 28.0 26.7 25.4 25.5 24.7 24.4 24.7 24.0 23.5 26.3 25.1 24.8
53 SULF2-Hs00378697_m1 Core up 28.9 30.1 30.3 26.0 26.6 28.5 25.6 28.4 29.3 31.2 26.9 26.9 27.9 27.5 22.0 21.3 23.7 22.9 24.6 25.1 23.7 22.9 24.8 24.2
54 TUBB-Hs00962420_g1 Norm 20.3 21.7 21.7 21.3 20.4 22.5 22.9 20.2 22.4 22.2 21.3 20.8 21.1 21.1 20.6 20.4 21.2 20.9 22.3 21.3 21.4 20.7 21.4 21.2
55 LMO2-Hs00277106_m1 Core down 37.2 37.4 33.8 35.5 33.8 29.3 29.3 36.4 33.3 33.1 30.5 26.6 25.9 27.2 25.8 24.6 25.8 26.4 26.4 27.5 27.7 30.0 32.6 32.3
56 MAN1C1-Hs00220595_m1 Core up 28.1 27.4 27.2 26.0 28.7 27.1 25.9 27.7 25.6 26.9 27.2 31.6 29.1 29.8 25.9 26.4 26.4 26.6 26.8 27.1 24.9 26.6 28.4 27.8
57 TAGLN-Hs00162558_m1 Core down 21.5 19.3 18.6 20.6 21.7 19.5 22.1 22.4 19.1 19.9 21.9 21.6 23.7 22.6 26.4 27.1 37.4 36.7 36.3 30.0 21.3 21.3 29.5 29.3
58 RTN1-Hs00382515_m1 Core down 26.7 23.4 23.3 29.8 28.6 28.3 26.2 27.7 26.0 26.8 23.8 24.3 25.6 26.2 25.2 23.5 23.8 24.2 25.6 24.9 25.8 26.5 25.6 25.6
59 MAF-Hs00193519_m1 Core down 26.9 26.3 26.8 26.5 25.6 26.8 26.2 26.9 28.1 27.6 27.1 25.7 26.6 26.9 26.5 24.4 28.2 29.1 27.2 25.9 29.7 29.7 28.8 28.6
60 NPTX2-Hs00383983_m1 Core down 35.6 31.0 30.6 30.6 26.6 32.1 30.0 37.5 35.3 35.3 33.6 23.4 25.7 24.5 36.1 34.4 26.6 29.2 30.7 29.2 30.7 26.0 30.5 29.6
61 EDA2R-Hs00939736_m1 Core down 28.2 28.4 28.1 27.7 27.0 28.3 27.0 28.0 28.0 28.1 28.6 27.1 28.5 27.8 26.9 26.6 30.7 34.1 32.8 31.6 27.9 28.1 27.1 27.1
62 SOX10-Hs00366918_m1 Marker 37.4 37.4 37.1 37.1 36.8 34.6 33.0 37.5 35.4 36.9 37.0 37.5 35.2 37.9 27.0 25.4 37.4 36.8 35.2 36.7 36.7 36.2 37.9 37.4
63 SEMA6A-Hs00221174_m1 Core down 25.4 26.4 26.1 24.4 25.0 25.6 24.5 25.1 25.2 26.4 22.7 23.3 23.1 24.2 24.7 23.8 25.4 25.0 25.9 27.0 26.3 29.4 27.1 27.7
64 DDIT3-Hs00358796_g1 Core up 28.3 29.4 29.3 28.5 27.4 27.3 27.5 29.0 28.7 28.5 28.7 28.1 28.1 28.3 26.4 24.6 26.3 25.7 28.1 26.9 26.3 26.3 27.3 27.3
65 CA12-Hs01080909_m1 Core down 30.5 32.7 32.7 31.0 28.1 28.9 31.2 31.4 31.0 31.4 21.5 22.6 23.6 23.4 24.5 23.5 23.8 22.9 24.2 23.0 23.8 25.6 31.3 31.2
66 C9orf125-Hs00260558_m1 Core down 33.2 32.7 32.9 36.8 36.8 33.9 35.0 34.2 33.7 36.3 29.4 29.0 28.3 29.9 27.0 27.6 29.4 28.3 29.4 30.1 28.0 27.4 31.3 31.4
67 SDC2-Hs00299807_m1 Core down 25.1 24.9 24.8 24.3 24.3 24.4 23.5 25.7 24.8 25.1 25.5 25.5 25.9 26.3 26.5 24.4 27.0 28.2 27.5 27.4 26.1 27.1 27.0 26.9
68 SLIT2-Hs00191193_m1 Core down 29.7 33.5 33.0 29.0 28.5 30.8 29.6 30.3 31.4 32.8 31.8 24.5 24.3 25.2 36.1 33.9 28.3 27.4 31.2 31.4 26.1 27.5 26.0 25.9
69 MT2A-Hs02379661_g1 Core up 25.5 24.0 24.4 23.1 22.2 20.2 20.6 26.0 23.7 23.9 23.9 23.8 24.6 25.0 21.6 19.7 20.8 21.2 21.7 20.7 18.8 18.2 23.4 22.7
70 IL17RD-Hs00296982_m1 Core down 27.1 26.1 25.8 25.9 26.7 26.5 26.4 26.3 26.7 26.8 26.3 24.3 24.5 24.6 26.5 26.2 28.1 26.8 26.8 27.6 24.0 24.0 26.8 26.4
71 PDZRN3-Hs00392900_m1 Core down 29.3 29.3 29.1 27.3 28.6 28.9 26.8 28.8 26.1 28.7 28.4 26.9 26.8 27.6 28.8 29.9 27.5 28.9 28.6 30.2 27.6 30.1 30.2 30.0
72 LGALS3-Hs00173587_m1 Core up 26.1 24.9 25.5 24.9 23.8 23.8 23.2 26.1 25.0 25.1 26.6 26.3 25.7 25.8 24.1 21.4 26.8 27.5 25.5 25.8 22.4 22.7 26.4 25.7
73 FAM69A-Hs00961685_m1 Core up 25.7 25.5 25.3 23.9 24.4 25.9 23.7 25.2 24.8 26.2 26.4 28.1 29.8 28.8 24.7 23.4 26.2 26.3 26.6 26.6 24.6 25.1 26.2 25.7
74 TERT-Hs00972656_m1 Marker 37.4 37.4 37.1 37.1 36.8 36.4 37.2 37.5 35.4 36.9 37.0 37.2 36.6 36.3 32.6 31.8 32.8 32.0 32.0 32.6 33.1 32.6 31.9 34.4
75 PLS3-Hs00418605_g1 Core up 25.0 28.0 28.0 25.1 23.9 26.1 26.6 24.9 28.1 28.8 22.9 26.6 25.7 25.0 22.5 21.8 25.7 26.4 26.2 24.9 22.2 22.3 22.7 22.4
76 PTEN-Hs02621230_s1 Core down 30.4 27.9 27.7 29.5 30.3 27.4 26.6 29.3 26.7 27.5 29.6 28.1 27.7 28.1 36.1 34.4 30.3 30.7 27.1 28.0 36.7 36.2 28.7 28.4
77 SALL2-Hs00826674_m1 Core down 34.7 32.2 32.0 32.3 33.7 32.4 29.9 35.3 31.0 33.0 34.9 33.0 32.5 33.4 30.6 29.6 31.2 31.2 32.5 29.4 32.8 34.1 35.0 34.3
78 KCTD12-Hs00540818_s1 Core down 31.0 30.7 30.4 33.4 33.7 32.1 33.4 30.1 28.1 29.7 26.5 25.8 27.4 27.0 26.8 27.4 31.2 32.1 34.8 32.8 26.9 28.9 34.4 32.7
79 TES-Hs00210319_m1 Core down 26.0 26.0 26.2 27.8 27.1 28.2 28.6 26.2 27.5 26.8 32.6 25.6 26.2 26.5 36.1 34.0 37.4 36.8 36.3 37.7 33.0 36.2 36.7 37.4
80 SOX2-Hs01053049_s1 Marker 25.6 21.7 21.8 24.3 26.0 22.2 20.6 24.8 20.8 22.0 24.4 23.6 23.6 24.3 23.7 22.4 23.4 23.2 21.9 22.4 25.0 27.1 24.6 24.1
81 LPAR6-Hs00271758_s1 Core down 31.3 29.4 29.6 34.5 34.0 31.5 30.9 31.6 29.2 30.0 29.9 28.4 28.6 29.4 31.0 29.3 28.6 29.6 28.7 28.5 27.4 28.8 30.1 30.2
82 ODZ2-Hs00393060_m1 Core down 28.7 27.8 27.6 26.8 27.7 27.4 27.0 29.2 28.3 29.4 25.7 27.4 25.1 25.6 29.8 29.0 25.1 26.2 27.1 26.8 27.5 29.3 26.3 26.2
83 NNMT-Hs00196287_m1 Core up 31.3 27.9 28.0 29.0 31.9 29.5 29.8 32.5 29.3 28.4 30.0 28.2 29.6 28.8 30.0 30.0 24.1 25.9 23.7 22.9 21.9 22.9 29.7 27.4
84 CACNG8-Hs01100182_m1 Core down 37.4 33.4 32.8 32.8 36.8 33.1 34.7 34.0 33.1 35.3 28.7 27.8 26.3 28.2 28.8 27.8 27.2 28.5 29.3 29.5 29.2 31.6 32.9 32.7
85 PRSS12-Hs00186221_m1 Core up 30.2 35.0 35.0 32.2 30.6 33.6 33.6 30.2 31.8 33.7 37.0 30.4 29.7 28.9 26.5 26.8 28.9 30.0 33.5 33.8 25.3 26.0 24.7 24.7
86 FOXJ1-Hs00230964_m1 Core down 36.1 33.3 32.8 26.8 33.4 29.8 26.5 35.3 28.5 33.4 35.6 31.0 29.5 33.0 29.1 28.7 31.6 32.3 30.5 31.7 36.7 36.2 33.9 33.5
87 NTN1-Hs00180355_m1 Core down 30.6 29.7 29.4 26.5 30.2 27.1 26.1 30.0 27.0 28.9 27.0 25.7 25.9 26.4 28.7 28.3 29.3 28.7 27.9 30.2 33.1 32.1 28.4 29.5
88 LMO3-Hs00375237_m1 Core down 29.6 28.9 29.1 34.7 36.8 34.8 30.9 30.1 27.4 30.3 33.9 29.2 25.7 29.2 36.1 34.4 31.6 35.0 27.3 28.6 35.3 36.1 37.9 37.4
89 CHI3L1-Hs00609691_m1 Marker 34.4 32.1 32.5 34.2 35.3 30.8 33.5 37.5 34.4 34.0 32.0 32.3 30.9 30.7 27.7 29.6 22.8 27.1 28.6 27.1 22.7 24.6 30.9 31.5
90 CD9-Hs01124025_g1 Core up 27.8 30.8 31.1 25.0 24.1 25.5 24.4 28.2 28.2 29.8 27.1 27.8 28.1 27.6 20.1 19.6 22.8 22.2 22.1 21.9 23.8 23.0 23.8 23.8
91 ADD2-Hs00242289_m1 Core up 26.2 27.8 27.9 27.2 27.1 28.6 28.0 25.4 27.8 28.8 28.1 37.5 30.4 32.6 26.3 26.3 25.0 24.8 26.9 27.4 27.8 27.1 27.2 27.0
92 MYL9-Hs00697086_m1 Core down 25.8 24.7 24.9 21.1 22.4 23.5 23.8 27.0 28.4 28.7 24.1 23.0 25.6 24.5 30.7 33.3 28.8 33.0 27.9 26.6 22.0 23.4 33.3 33.4
93 LYST-Hs00179814_m1 Core up 31.4 30.5 30.4 30.8 30.3 30.8 29.5 30.8 30.1 31.4 31.1 31.1 32.8 32.4 27.3 26.1 28.8 28.5 29.7 29.2 28.5 29.2 29.9 30.3
94 LAMA2-Hs00166308_m1 Core down 35.6 37.4 37.1 37.1 36.8 36.4 37.2 37.5 35.4 36.9 31.1 29.1 26.9 29.5 31.1 32.6 28.1 32.6 27.1 25.9 25.5 28.5 31.4 30.7
95 PLA2G4A-Hs00233352_m1 Core up 29.6 33.6 33.2 36.1 36.8 32.3 33.3 29.7 32.0 31.7 29.9 35.4 32.3 33.3 28.1 27.7 29.0 29.4 30.2 31.4 28.9 28.1 29.3 28.9
96 BACE2-Hs00273238_m1 Core up 26.9 27.1 27.2 26.9 26.0 27.7 27.1 26.6 27.5 28.1 34.1 32.2 31.3 31.3 25.0 23.9 27.0 26.5 28.0 27.0 26.0 24.7 25.4 25.4
262

A.3 Quantitative RT-PCR Appendix
Well Applied Biosystems assay ID Gene G19 G19 G21 G21 G23 G23 G24 G24 G25 G25 G26 G26 G30 G30 G31 G31 G32 G32 G144 G144 G166 G166 G179 G179
Category A B A B A B A B A B A B A B A B A B A B A B A B
1 MYL9-Hs00382913_m1 Core down 34.3 35.7 34.7 33.4 29.4 32.1 28.2 27.1 30.4 30.2 29.5 29.6 34.4 34.2 36.0 35.4 31.9 32.9 31.7 32.8 29.3 28.7 31.1 31.3
2 CHCHD10-Hs01369775_g1 Core up 26.6 26.7 27.8 27.6 26.0 26.1 26.5 25.6 26.2 26.0 25.8 26.0 27.0 27.4 27.4 27.1 27.8 27.3 25.1 25.5 26.3 24.9 23.9 23.9
3 RGS5-Hs00186212_m1 Core down 31.3 31.2 33.0 33.2 30.3 30.0 31.8 31.1 30.0 30.4 30.8 32.0 30.3 30.7 31.0 31.1 30.9 31.3 33.6 31.8 31.4 31.5 30.3 30.0
4 ST6GALNAC5-
Core down 30.0 30.7 26.4 26.5 33.2 30.4 33.6 36.6 35.6 34.7 30.6 31.0 33.9 34.3 28.1 27.6 29.5 29.3 33.5 34.7 31.9 33.7 33.7 33.1
Hs00229612_m1
5 CEBPB-Hs00270923_s1 Core up 30.9 30.0 30.9 31.1 28.5 27.9 30.4 29.9 29.5 29.8 30.3 30.2 30.3 30.3 30.0 29.8 30.5 30.6 29.6 30.4 29.4 28.8 27.3 28.0
6 C5orf13-Hs00854282_g1 Core down 29.4 27.6 28.1 28.1 28.5 25.3 22.1 24.0 26.0 27.1 37.6 29.1 26.8 25.6 37.9 26.9 37.8 28.1 24.4 35.8 27.1 28.1 27.9 28.0
7 PDGFRA-Hs00998026_m1 Marker 25.0 24.6 23.4 22.7 25.5 27.4 25.4 24.0 26.2 26.4 24.4 26.6 23.9 24.2 28.4 28.3 28.1 29.6 24.2 25.0 26.2 28.7 25.8 25.8
8 CCND2-Hs00922419_g1 Core down 35.3 30.0 22.5 22.4 28.2 27.5 30.4 32.0 27.1 27.2 25.1 25.6 22.5 22.7 21.7 21.5 21.4 21.9 22.5 24.2 33.4 32.2 31.7 31.4
9 NKX2-1-Hs00163037_m1 Core down 31.4 31.0 36.2 35.7 37.3 35.0 36.3 36.6 36.7 37.4 37.6 36.7 37.4 37.6 32.1 32.3 33.0 31.8 32.2 35.7 38.0 37.1 36.2 36.9
10 CTSC-Hs00175188_m1 Core up 22.2 22.8 25.1 25.8 21.9 20.9 26.5 25.1 22.2 22.5 21.2 21.8 24.3 24.2 25.3 25.1 25.1 25.2 21.4 21.6 22.2 22.9 23.5 23.0
11 18S-Hs99999901_s1 Norm 13.2 12.7 12.9 12.7 12.1 12.3 12.2 12.6 12.0 12.2 12.3 11.8 12.4 12.5 12.3 12.5 12.4 12.7 12.0 12.6 13.0 13.9 11.0 11.2
12 DNER-Hs00294564_m1 Core down 31.7 27.9 28.2 26.7 33.0 32.5 28.2 26.8 26.3 27.1 25.9 27.6 24.5 24.6 24.9 25.1 25.8 26.8 28.7 27.4 28.0 30.4 27.1 26.8
13 LMO4-Hs01086790_m1 Core up 24.8 24.8 22.8 22.6 23.7 23.7 24.9 23.8 24.1 24.2 23.9 25.1 23.2 23.2 23.3 23.2 23.7 23.7 23.2 23.5 24.7 24.2 23.6 23.1
14 PLCH1-Hs00392783_m1 Core down 26.3 25.9 28.0 28.0 27.2 28.8 28.5 28.1 31.3 30.9 30.0 29.6 29.9 29.9 28.2 28.0 28.3 27.7 30.0 30.3 29.8 29.7 30.5 30.8
15 SPARCL1-Hs00949886_m1 Core down 27.3 25.2 27.3 27.7 26.5 29.0 25.5 23.6 27.2 27.7 25.4 28.9 24.6 24.6 27.5 27.0 27.7 29.7 28.7 30.2 29.0 30.6 25.5 25.1
16 SYNM-Hs00322391_m1 Core down 34.3 33.0 28.8 28.6 28.0 27.7 27.7 27.9 35.7 37.4 30.0 29.0 37.4 36.7 32.9 33.9 32.5 32.7 34.5 35.8 37.8 37.1 28.8 28.9
17 FBLN2-Hs00157482_m1 Core down 37.3 35.0 31.9 32.2 30.4 30.8 37.2 35.5 32.5 34.3 34.9 35.8 34.5 35.2 35.2 35.1 34.7 36.2 31.7 34.2 37.5 34.8 27.9 28.5
18 SALL2-Hs00413788_m1 Core down 30.2 27.7 27.5 26.6 34.6 37.1 35.5 32.8 31.7 31.4 30.1 32.7 26.6 26.8 27.5 27.1 27.6 27.9 27.7 27.5 38.0 37.1 33.1 33.7
19 TUSC3-Hs00185147_m1 Core down 23.5 23.0 37.9 37.3 24.7 25.1 34.6 32.5 36.6 35.4 37.6 36.7 25.3 25.5 37.9 37.6 30.8 30.9 34.5 35.8 33.1 28.9 25.7 25.1
20 GPR158-Hs00393109_m1 Core down 30.9 29.5 28.8 28.5 37.3 37.1 32.8 32.3 30.8 32.0 28.8 30.1 30.8 31.9 29.6 29.9 29.9 29.9 25.7 27.7 32.4 33.4 30.2 30.8
21 PEG3-Hs00377844_m1 Core down 29.3 28.5 24.5 25.2 33.2 34.0 28.3 28.1 27.0 27.6 28.2 29.8 25.6 26.2 37.9 37.6 29.8 32.7 27.0 27.6 30.8 31.6 36.2 36.9
22 FUT8-Hs00189535_m1 Core up 27.7 27.8 26.0 25.4 27.3 27.3 27.6 27.0 25.6 26.3 28.0 28.1 26.6 27.0 27.5 26.7 26.8 27.3 25.8 26.3 26.0 26.1 26.5 26.1
23 HLA-DRA-Hs00219575_m1 Core up 27.3 25.7 34.3 35.4 35.6 37.1 28.4 27.4 22.1 23.3 24.4 24.2 24.3 24.5 34.6 32.6 28.8 31.2 21.9 21.6 29.3 26.7 21.4 21.0
24 ASCL1-Hs00269932_m1 Marker 37.2 30.3 24.0 23.7 29.0 30.6 29.9 33.5 35.1 37.4 37.0 36.7 23.8 24.0 22.8 23.1 22.9 23.4 25.6 25.3 36.1 37.1 30.9 31.2
25 PI15-Hs00210658_m1 Core down 29.2 30.4 33.2 29.3 30.9 32.3 29.3 26.2 27.2 27.9 28.4 26.5 31.5 31.3 36.9 37.6 28.5 35.4 30.1 32.9 28.4 30.1 27.5 26.6
26 EPDR1-Hs00378148_m1 Core up 28.0 28.7 31.1 31.2 28.7 29.3 28.1 27.8 26.6 26.2 27.6 27.7 28.3 28.4 30.2 29.7 29.9 29.6 28.3 28.6 28.1 26.9 28.8 28.4
27 DTX4-Hs00392288_m1 Core down 28.1 27.3 29.4 28.4 33.1 33.9 30.9 30.2 29.7 29.7 35.1 33.8 30.2 30.1 27.4 27.0 27.3 28.3 27.8 29.5 32.0 31.8 27.7 27.9
28 CXXC4-Hs00228693_m1 Core down 30.0 27.8 25.2 25.1 29.4 31.0 28.7 29.5 36.6 37.4 35.5 36.7 27.3 27.4 26.3 26.7 26.2 27.7 28.2 28.1 31.4 32.9 36.1 34.9
29 IRX2-Hs01383002_m1 Core down 37.3 36.6 37.9 37.3 28.6 28.8 36.9 36.6 36.7 37.4 37.6 36.7 37.4 37.6 37.9 37.6 37.8 35.6 34.5 35.8 38.0 37.1 29.8 30.2
30 MAP6-Hs01023152_s1 Core down 28.2 27.9 26.9 27.0 28.6 29.7 31.4 29.3 30.1 31.2 33.1 32.3 28.8 28.8 30.1 30.0 29.4 30.5 31.2 30.6 31.7 33.8 26.7 26.4
31 SIX3-Hs00193667_m1 Core down 33.4 33.6 37.9 37.3 30.2 30.6 35.0 34.5 35.7 36.3 37.6 36.7 35.4 37.1 35.6 36.7 37.8 35.6 29.1 29.7 34.7 33.2 36.2 36.9
32 NKX2-2-Hs00159616_m1 Marker 29.1 28.0 27.4 28.1 27.8 28.0 26.8 26.4 25.2 25.2 26.4 27.0 25.3 25.3 25.9 25.9 25.5 25.2 24.6 24.8 29.4 28.9 30.3 30.9
33 S100A6-Hs00170953_m1 Core up 18.8 19.7 30.1 27.6 18.6 17.9 20.4 19.6 20.0 19.6 22.0 20.9 21.4 21.5 23.9 23.5 23.5 22.5 21.4 22.6 18.5 18.0 18.5 18.1
34 MMP17-Hs01108847_m1 Core up 28.7 29.1 31.2 30.5 28.9 27.8 29.3 28.5 24.8 25.3 30.7 31.0 26.5 26.6 30.3 29.4 30.0 29.9 26.7 25.2 27.6 26.6 27.8 28.3
35 FAM38B-Hs00926225_m1 Core down 29.3 28.9 37.9 37.3 37.2 36.5 37.5 36.6 36.7 37.4 37.6 36.7 37.4 37.6 37.9 37.6 37.8 37.7 34.5 35.8 38.0 37.1 31.9 31.5
36 OLIG2-Hs00300164_s1 Marker 35.2 28.7 23.0 23.2 27.1 27.6 25.8 25.8 22.8 23.3 23.7 24.6 22.1 21.8 23.5 23.4 23.4 23.4 22.3 22.4 32.1 33.3 29.7 30.4
37 PMEPA1-Hs00375306_m1 Core up 24.4 24.7 28.8 28.0 25.8 26.5 22.8 22.6 25.1 24.7 27.4 27.0 25.6 25.3 28.0 27.5 28.1 27.7 24.8 26.5 23.6 25.1 25.3 25.6
38 HOXD10-Hs00157974_m1 Core up 28.8 28.0 37.9 37.3 28.0 28.7 29.8 29.1 29.2 29.3 30.1 30.9 29.3 29.2 29.7 29.5 29.6 29.8 28.1 27.6 27.5 27.1 27.1 27.1
39 NDN-Hs00267349_s1 Core down 24.1 23.7 30.3 30.5 37.3 37.1 26.4 25.4 25.6 26.1 27.8 26.8 37.3 37.6 26.4 25.7 26.3 25.7 26.1 25.8 36.8 37.1 36.2 35.0
40 FOXG1-Hs01850784_s1 Core up 24.7 23.8 23.8 23.3 24.8 25.5 24.8 24.9 25.1 25.1 26.9 25.9 23.8 24.1 22.3 22.2 22.2 22.5 23.8 23.6 24.8 25.7 21.0 21.2
41 MN1-Hs00159202_m1 Core down 32.5 34.5 28.9 28.3 32.1 33.6 26.9 26.9 30.2 29.3 34.3 35.1 28.4 27.7 27.4 27.8 27.6 27.1 29.1 28.9 35.7 35.1 30.7 30.6
42 NDUFB10-Hs00605903_m1 Norm 25.3 25.2 25.4 25.6 25.6 25.4 25.6 25.3 25.7 25.7 25.3 25.6 25.7 25.5 26.2 26.1 26.2 25.9 26.0 25.4 25.5 25.1 25.7 25.2
43 HMGA2-Hs00171569_m1 Core down 30.2 31.6 33.9 33.5 27.7 27.3 35.8 35.5 32.1 32.2 32.7 33.3 32.8 32.1 34.0 33.4 33.3 33.1 33.3 32.6 32.3 32.2 28.6 28.5
44 MMRN1-Hs00201182_m1 Core down 34.0 31.1 37.9 37.3 33.3 34.7 37.5 36.6 36.7 37.4 32.6 31.2 37.4 37.6 37.9 37.6 37.8 37.7 34.5 35.8 34.8 35.1 32.8 32.1
45 RAB6B-Hs00981572_m1 Core down 29.2 30.3 30.7 29.6 33.5 33.1 28.4 27.6 29.2 29.4 31.4 32.1 28.3 28.1 27.6 27.3 27.4 27.8 29.1 29.0 30.7 30.7 30.2 29.7
46 NTRK2-Hs00178811_m1 Core up 24.3 24.1 30.4 30.0 31.9 32.7 33.6 32.7 23.1 23.8 23.9 26.0 24.7 24.6 35.6 37.5 28.2 32.5 23.8 23.9 23.5 24.9 27.3 27.3
47 KALRN-Hs00610179_m1 Core down 30.8 30.2 28.4 27.7 30.6 31.9 29.0 27.9 30.4 30.5 31.2 30.8 27.6 28.1 28.3 28.1 27.4 28.2 29.6 29.6 28.9 31.1 30.5 30.8
48 CD74-Hs00269961_m1 Core up 27.9 25.9 33.0 35.4 33.0 34.3 28.4 27.8 21.9 23.3 24.8 25.2 25.1 25.0 37.5 34.1 29.0 31.0 22.4 23.5 29.7 26.7 21.5 21.8
49 LUM-Hs00158940_m1 Core down 30.4 31.4 37.9 37.3 23.6 26.1 33.3 27.1 36.7 37.4 32.0 30.9 37.4 37.6 33.3 31.8 35.7 37.7 34.5 35.8 29.8 27.6 29.6 29.0
50 NELL2-Hs00196254_m1 Core down 33.4 32.8 34.1 29.3 31.8 32.2 30.3 29.2 36.6 37.4 33.1 34.1 31.2 32.0 27.9 28.0 29.0 29.8 27.5 26.8 30.1 32.0 31.4 31.0
263

A.3 Quantitative RT-PCR Appendix
Well Applied Biosystems assay ID Gene G19 G19 G21 G21 G23 G23 G24 G24 G25 G25 G26 G26 G30 G30 G31 G31 G32 G32 G144 G144 G166 G166 G179 G179
Category A B A B A B A B A B A B A B A B A B A B A B A B
51 MAP6-Hs01929835_s1 Core down 26.0 26.1 25.9 25.7 29.1 30.2 30.2 28.3 29.2 30.7 32.0 31.6 27.3 27.6 30.6 30.4 29.1 30.7 32.2 29.8 31.9 34.0 26.6 26.3
52 PDE1C-Hs01095694_m1 Core up 24.8 25.0 28.2 28.9 25.2 24.2 24.1 23.3 24.1 24.0 26.1 25.5 26.9 26.5 28.9 28.5 28.7 28.5 25.0 25.2 23.9 23.6 25.5 25.5
53 SULF2-Hs00378697_m1 Core up 23.9 24.3 27.0 26.1 24.3 26.4 23.1 22.0 24.3 24.1 24.3 25.4 23.5 23.2 25.9 25.0 25.7 25.8 25.0 23.5 26.2 25.5 26.0 26.0
54 TUBB-Hs00962420_g1 Norm 20.7 21.3 20.9 20.9 21.5 21.4 21.4 21.3 21.5 21.2 21.5 21.8 21.1 21.2 20.7 20.6 20.6 20.5 21.2 21.2 20.7 20.1 22.5 22.7
55 LMO2-Hs00277106_m1 Core down 30.6 30.1 27.2 26.2 31.3 35.6 31.3 29.9 34.0 34.4 27.9 30.2 26.3 27.4 26.9 26.5 26.6 26.1 30.0 31.4 34.3 35.6 29.0 28.9
56 MAN1C1-Hs00220595_m1 Core up 27.9 26.9 25.8 25.2 32.1 34.2 29.7 28.8 26.9 26.7 24.8 25.9 25.6 25.9 25.8 24.9 26.0 25.5 26.3 26.3 28.1 27.4 26.2 26.6
57 TAGLN-Hs00162558_m1 Core down 28.4 29.7 37.9 33.3 24.9 27.3 21.4 21.1 26.2 25.4 29.7 29.0 26.1 25.0 33.7 31.0 29.3 29.4 27.4 30.2 31.4 29.8 26.9 26.7
58 RTN1-Hs00382515_m1 Core down 25.1 24.8 33.6 32.9 27.8 28.8 29.8 28.7 32.2 32.3 25.5 26.2 25.8 25.8 28.9 27.7 29.8 30.3 27.0 28.1 29.1 30.8 30.8 30.6
59 MAF-Hs00193519_m1 Core down 28.2 27.5 28.7 28.9 27.5 30.2 28.5 28.3 31.1 31.9 27.3 28.4 27.8 28.0 27.8 28.0 28.4 27.5 31.4 32.1 28.1 28.7 34.7 34.5
60 NPTX2-Hs00383983_m1 Core down 30.1 27.8 29.0 32.2 28.9 29.3 34.5 36.6 36.7 35.0 27.8 26.5 32.3 31.8 31.2 29.3 34.2 30.2 31.7 32.1 28.4 29.3 31.8 32.3
61 EDA2R-Hs00939736_m1 Core down 27.5 27.4 28.8 28.6 29.2 26.9 28.4 27.8 35.2 35.2 32.7 32.2 28.8 28.8 35.4 35.0 30.6 32.2 34.4 32.8 32.0 30.2 34.6 34.3
62 SOX10-Hs00366918_m1 Marker 37.3 36.6 37.9 37.3 37.3 37.1 37.0 36.6 24.4 25.2 37.6 36.7 35.4 35.6 37.9 37.6 37.8 37.7 24.5 24.5 38.0 37.1 36.2 36.9
63 SEMA6A-Hs00221174_m1 Core down 26.9 25.9 23.3 23.6 25.9 26.2 24.1 24.5 27.4 27.6 25.2 26.4 24.3 24.4 23.4 23.4 23.6 23.7 24.7 25.0 26.3 27.2 27.2 27.6
64 DDIT3-Hs00358796_g1 Core up 27.0 26.8 28.0 28.1 25.2 22.4 27.6 26.8 26.8 26.3 28.3 28.7 27.3 27.2 22.0 21.7 21.8 21.8 22.5 23.2 25.3 23.8 26.4 25.9
65 CA12-Hs01080909_m1 Core down 30.4 28.1 23.3 23.6 25.8 26.6 31.8 30.0 27.6 27.3 24.1 25.9 27.1 28.4 24.8 24.4 24.8 23.9 25.8 25.1 26.9 26.9 25.6 25.9
66 C9orf125-Hs00260558_m1 Core down 31.0 31.1 37.9 37.3 34.8 37.1 29.1 28.5 31.8 32.3 30.4 31.3 27.3 27.9 37.9 37.6 37.3 37.3 29.4 29.2 32.3 34.1 35.1 34.8
67 SDC2-Hs00299807_m1 Core down 26.3 26.5 36.0 36.3 26.2 26.4 25.6 25.2 25.6 25.3 26.4 26.3 29.4 29.2 29.0 28.2 29.6 29.6 26.7 27.8 27.9 28.6 25.4 25.3
68 SLIT2-Hs00191193_m1 Core down 25.4 23.8 34.2 32.3 26.2 26.3 32.8 30.6 33.5 32.7 32.3 31.8 33.7 34.8 29.8 29.6 27.6 27.9 30.4 32.1 33.1 33.5 30.7 30.6
69 MT2A-Hs02379661_g1 Core up 22.9 22.6 22.6 21.8 21.2 22.0 23.3 22.7 23.7 22.9 20.6 21.5 22.2 22.7 22.3 22.0 22.1 22.4 21.8 21.7 19.8 18.6 19.2 18.8
70 IL17RD-Hs00296982_m1 Core down 26.6 26.5 26.4 26.9 24.0 24.5 26.8 26.6 27.7 27.4 27.2 28.2 26.5 26.6 27.1 27.0 26.9 27.4 26.4 26.6 29.0 28.2 27.2 27.5
71 PDZRN3-Hs00392900_m1 Core down 29.3 27.9 27.3 27.0 34.2 37.1 36.2 36.6 26.4 26.4 28.3 28.3 29.0 29.5 27.3 26.9 26.9 26.7 30.5 31.7 31.8 31.8 28.2 28.9
72 LGALS3-Hs00173587_m1 Core up 25.1 24.3 25.2 25.0 23.4 23.4 27.1 25.2 24.4 25.0 23.1 23.8 24.7 24.9 26.3 25.4 25.1 25.3 25.2 24.8 25.1 23.2 23.8 23.2
73 FAM69A-Hs00961685_m1 Core up 25.3 24.4 24.6 25.6 25.1 27.8 25.7 23.6 25.4 25.9 24.9 25.6 26.8 26.9 26.1 26.0 25.7 25.8 25.6 25.3 27.1 26.1 26.6 25.9
74 TERT-Hs00972656_m1 Marker 31.8 32.5 29.0 27.2 30.7 30.3 31.9 31.3 31.0 30.8 31.2 31.7 30.2 29.9 30.7 30.2 30.3 29.9 29.4 28.2 30.2 29.9 33.9 34.7
75 PLS3-Hs00418605_g1 Core up 21.9 22.3 25.3 25.8 23.1 23.1 22.1 22.4 23.9 23.7 23.1 23.6 25.3 25.1 25.3 25.4 25.6 25.0 24.0 23.4 22.1 22.5 25.5 24.9
76 PTEN-Hs02621230_s1 Core down 29.2 29.2 33.8 30.6 28.1 28.7 29.1 28.5 29.6 30.3 31.0 31.1 28.5 28.7 31.3 31.5 31.1 31.4 34.5 35.8 30.1 31.2 36.2 34.8
77 SALL2-Hs00826674_m1 Core down 34.2 32.2 32.6 33.4 33.8 37.1 33.9 32.1 32.8 33.1 33.5 34.2 32.8 32.4 32.0 32.1 31.8 31.8 32.8 33.4 33.4 31.9 30.6 30.7
78 KCTD12-Hs00540818_s1 Core down 32.2 31.6 30.4 32.3 30.4 31.9 26.3 25.9 27.6 28.5 28.9 29.0 28.0 28.2 30.9 31.7 30.2 31.5 28.3 29.8 30.4 30.1 25.9 26.5
79 TES-Hs00210319_m1 Core down 37.3 36.6 37.9 37.3 34.5 36.6 34.5 36.4 33.8 34.6 37.6 36.7 37.4 37.6 37.9 37.6 37.8 37.7 34.5 35.8 37.2 36.2 35.0 33.3
80 SOX2-Hs01053049_s1 Marker 24.0 23.5 22.9 23.0 24.8 24.6 24.6 24.4 23.2 23.9 23.4 24.5 23.0 23.2 23.2 23.0 22.8 23.3 23.4 23.9 25.5 27.0 21.9 21.8
81 LPAR6-Hs00271758_s1 Core down 29.9 29.5 29.7 29.7 30.0 31.4 29.2 28.7 33.0 33.2 31.1 31.9 28.6 28.6 29.9 29.3 29.8 29.4 31.6 31.7 29.8 30.3 31.3 31.2
82 ODZ2-Hs00393060_m1 Core down 25.9 26.0 27.6 28.9 25.0 26.1 28.6 26.5 31.6 31.4 25.5 26.5 31.0 31.0 28.0 26.9 28.7 27.9 27.1 28.2 31.0 32.3 27.2 27.4
83 NNMT-Hs00196287_m1 Core up 28.3 27.9 30.9 30.9 24.1 27.2 33.2 29.9 27.9 27.3 24.3 24.2 27.6 27.8 29.9 28.7 30.6 29.0 25.0 26.8 24.0 24.1 23.5 23.4
84 CACNG8-Hs01100182_m1 Core down 32.0 30.5 28.1 28.9 30.4 29.9 31.0 36.6 33.3 33.3 34.3 34.0 27.3 27.6 31.8 33.2 30.4 30.7 27.7 27.7 31.4 31.9 32.0 33.3
85 PRSS12-Hs00186221_m1 Core up 24.5 25.3 31.2 33.8 25.1 24.6 31.5 31.5 27.4 27.0 25.9 26.6 37.4 37.6 28.4 28.3 27.7 27.5 24.8 24.6 24.3 24.4 30.7 31.1
86 FOXJ1-Hs00230964_m1 Core down 32.8 30.1 31.8 31.2 33.8 36.5 31.1 30.5 29.9 31.2 30.3 30.9 28.2 28.3 31.4 31.3 30.6 31.2 30.3 30.9 36.2 37.1 31.1 31.8
87 NTN1-Hs00180355_m1 Core down 29.5 28.5 27.5 26.6 34.5 35.7 25.9 24.6 27.7 28.4 28.7 30.2 27.4 27.4 26.4 26.5 26.5 26.3 25.5 26.8 31.5 35.1 28.6 29.8
88 LMO3-Hs00375237_m1 Core down 37.3 36.6 33.0 35.2 31.9 33.7 37.5 36.6 36.7 37.4 28.7 28.6 32.4 31.2 28.0 28.5 28.2 29.3 34.5 35.8 38.0 37.1 35.1 34.2
89 CHI3L1-Hs00609691_m1 Marker 31.6 25.5 30.7 28.1 33.4 37.1 24.8 23.7 29.0 31.1 24.4 27.1 22.1 23.4 28.5 28.2 29.0 29.9 34.5 35.8 24.0 28.4 21.5 21.7
90 CD9-Hs01124025_g1 Core up 23.4 23.6 23.3 23.9 24.7 24.4 26.0 24.6 22.1 22.9 23.5 24.2 22.8 23.0 23.5 23.5 23.4 23.8 22.6 22.7 23.4 23.2 23.4 22.9
91 ADD2-Hs00242289_m1 Core up 26.8 27.3 24.9 25.5 26.8 25.3 27.2 27.4 28.3 28.3 30.9 31.7 26.2 26.5 25.8 25.5 25.6 25.6 26.4 26.5 26.6 25.5 28.2 28.7
92 MYL9-Hs00697086_m1 Core down 33.0 32.5 32.3 31.6 27.6 29.8 23.3 22.9 28.3 28.1 26.8 27.6 28.7 27.9 34.7 32.8 31.1 31.5 30.5 31.6 26.7 26.2 26.6 26.8
93 LYST-Hs00179814_m1 Core up 30.1 29.8 30.7 31.0 29.9 28.5 29.7 28.6 28.6 29.6 29.9 31.3 28.7 29.3 29.8 29.8 29.9 30.6 27.6 28.2 29.1 28.1 28.1 27.7
94 LAMA2-Hs00166308_m1 Core down 30.2 28.9 37.8 37.3 28.8 29.9 37.5 35.6 31.2 33.4 32.8 32.2 31.0 31.8 30.5 29.4 36.0 35.4 26.8 27.0 33.7 34.1 33.2 34.9
95 PLA2G4A-Hs00233352_m1 Core up 28.4 28.7 30.2 29.3 29.6 29.5 30.6 29.4 26.8 27.4 31.8 31.8 26.9 26.7 27.4 26.8 27.1 26.8 26.5 27.2 29.6 27.5 29.0 28.7
96 BACE2-Hs00273238_m1 Core up 25.1 25.2 35.6 37.3 29.7 30.5 26.9 26.1 25.7 25.9 27.2 28.9 27.8 28.3 30.8 30.4 30.3 30.8 26.3 26.3 26.8 25.7 27.0 26.9
264

A.4 Tag-seq vs qRT-PCR Correlation Appendix
A.4 Tag-seq vs qRT-PCR Correlation
The correlation between the qRT-PCR measurements from the 21 cell lines
(16 GNS and 5 NS cell lines) and the Tag-seq measurements from the five cell
lines (three GNS and two NS cell lines) was found by taking the normalised
Ct values and tag counts for each of the 82 core differentially expressed genes
- as determined by Tag-seq on the three GNS cell lines and two NS cell lines -
and applying the cor function from the stats R package.
Table A.5: Pearson correlation values between the normalised Ct values measured
through qRT-PCR and the tag counts measured across the five GNS and NS cell
lines assayed via Tag-seq.
Gene name Pearson Gene name Pearson
correlation correlation
PTEN 0.9973 MAN1C1 0.9953
NDN 0.9942 HMGA2 0.9932
HOXD10 0.9925 TUSC3 0.9888
TES 0.9848 SYNM 0.9823
SIX3 0.9800 LYST 0.9797
MYL9 0.9777 PLA2G4A 0.9757
IRX2 0.9749 DDIT3 0.9748
MT2A 0.9741 MMP17 0.9735
BACE2 0.9684 MAF 0.9672
CA12 0.9653 LMO3 0.9652
NKX2-1 0.9625 IL17RD 0.9571
SEMA6A 0.9560 CXXC4 0.9520
FAM69A 0.9475 ADD2 0.9451
SPARCL1 0.9416 LMO2 0.9391
CHCHD10 0.9388 PLCH1 0.9362
PEG3 0.9345 CCND2 0.9344
KALRN 0.9310 FAM38B 0.9278
EDA2R 0.9270 GPR158 0.9222
LMO4 0.9203 CD9 0.9187
ODZ2 0.9168 ST6GALNAC5 0.9139
MAP6 0.9116 NPTX2 0.9047
CACNG8 0.9045 SLIT2 0.9016
MMRN1 0.8906 CD74 0.8906
TAGLN 0.8874 PDE1C 0.8832
CEBPB 0.8813 DNER 0.8811
FUT8 0.8804 C9orf125 0.8772
RAB6B 0.8709 PI15 0.8654
NTN1 0.8620 SULF2 0.8602
KCTD12 0.8384 PMEPA1 0.8322
LUM 0.8263 SALL2 0.8078
NNMT 0.8034 HLA-DRA 0.7884
RTN1 0.7777 MN1 0.7714
RGS5 0.7679 CTSC 0.7661
C5orf13 0.7595 PLS3 0.7439
NELL2 0.7146 NTRK2 0.7115
LPAR6 0.7090 EPDR1 0.7053
LAMA2 0.6889 SDC2 0.6595
S100A6 0.6580 FOXG1 0.6051
PDZRN3 0.6032 FBLN2 0.5795
PRSS12 0.5211 LGALS3 0.5204
DTX4 0.4964 FOXJ1 0.4454
265

Appendix B
Literature Mining Script
It should be noted that the text preceded by the single quotation mark sign
(’) that appears in italic typeface is a comment and not code that should be
executed.
Public Class Form1 ’GBMbase iHOP PubMed BioGraph Google Scholar Google Search
Dim genes() As String = {"SFTA3", "SLC4A4", "STRA6", "IL6", "SNX22", "COL21A1",
"CA12", "CCND2", "SPINT1", "INHBA", "GFAP", "GPC3", "HOXC10", "PDE1C",
"FAM70A", "GEM", "KALRN", "NTN1", "INHBB", "CCNO", "C10orf81", "CTNNA2",
"APLN", "VAX1", "PCDHB4", "KCTD12", "ELN", "IFI30", "BMP8B", "CAMK2B",
"LMO2", "TECRL", "RGS5", "NOV", "MOCOS", "SCN1B", "TNFSF4", "C14orf143",
"LRRC2", "PLS3", "AP000280.1", "C21orf62", "C10orf116", "VIPR1", "ELMO1", "UGT8",
"SH3BGR", "LIF", "MX1", "LRAT", "FXYD3", "CNTN6", "NNMT", "ZIC5", "GBP2",
"TRIM47", "FAM196A", "C10orf141", "LMO4", "MYC", "RTP4", "C9orf125", "DPY19L1",
"FCGR2B", "FCGR2C", "FCGR2A", "MAPT", "SLC15A2", "TGM2", "IRX2", "PCDH20",
"ANGPTL1", "IL17RD", "ARC", "CHCHD10", "MAF", "CD55", "FBXO27", "PDZRN3",
"ETS1", "SGCD", "CITED4", "MAP3K5", "STXBP5L", "ECHDC2", "CTSC", "MICA",
"MMP17", "AC068399.1", "VIT", "TSHZ3", "CXXC4", "CABP7", "KCNMB2", "MT2A",
"MYH3", "HOXD13", "STEAP2", "SLC4A11", "CCL7", "IFI27", "SPARCL1", "TNNI1",
"OAS2", "TRPM8", "THBS2", "DOCK10", "ZIC2", "IFI6", "TNNI2", "CARD17", "CASP1",
"CARD16", "ATOH8", "DNER", "FAM189A1", "SALL2", "C10orf11", "EEF1D", "WB-
SCR17", "TUBA4A", "RHBDF2", "PARP3", "MAN1C1", "DDO", "F12", "SPTBN5",
"TGFA", "ZNF536", "HLA-DQB1", "IFITM2", "DNM3", "JPH1", "FAM69A", "CACNA1C",
"ESRRG", "ITM2A", "SKAP2", "TRAM1L1", "NDN", "PLCH1", "KCNJ12", "NXPH1",
"MYBL1", "DUSP5", "NTRK2", "RNF175", "HOMER1", "PTEN", "CMPK2", "TBC1D8",
"TRIB3", "CDHR1", "SYNM", "LUM", "SEMA6A", "ADRA2A", "STAMBPL1", "TSPAN7",
"PYGL", "FOXJ1", "ZNF454", "EML2", "GABRQ", "EPAS1", "ERBB4", "RAB6B",
"LXN", "MYL9", "BGN", "FUT8", "GRIA3", "ARHGEF7", "SLC2A5", "MMP7", "TMEM132D",
"PION", "C5orf13", "SOD2", "PDGFA", "LOXL4", "SPP1", "BATF3", "SNAP25", "SULF2",
"LPAR6", "EPDR1", "EPHB3", "TSLP", "TMCO4", "SERPINE2", "TUSC3", "VSNL1",
"ATAD3C", "MARCH1", "DKK1", "CEBPB", "NFE2L3", "TNNC2", "ODZ2", "OAS1",
266

Appendix
"BEX5", "DIAPH2", "GBP3", "SORCS3", "SOX3", "FOXG1", "LRRN2", "ELMO2",
"MAP6", "TRIM48", "CNKSR2", "MOSC2", "CRYBB2", "GJA1", "ELOVL2", "B4GALNT1",
"C10orf90", "FBN2", "SMOC2", "ZIC3", "RARRES3", "GBP1", "EFEMP1", "GPR98",
"PLD3", "OLFM1", "CCDC64", "MGLL", "THY1", "CPLX2", "FAM150B", "MKI67",
"AC092296.1", "PHACTR3", "TNFRSF14", "NRG1", "RAB38", "ABAT", "NRBF2", "CLDN10",
"C1orf94", "NELL2", "CYB5R2", "TNC", "HLA-A", "PPCS", "SORL1", "SHROOM3",
"SLC38A1", "FXYD5", "CILP", "OGN", "MARVELD3", "APOD", "LEMD1", "DOCK5",
"GBP4", "KCNA2", "FAM55C", "NFKBIZ", "RGL3", "HLA-DPB1", "XXbac-BPG116M5.1",
"C2", "CFB", "NRP2", "HPSE", "CPNE5", "SMAGP", "FAM84A", "LOC653602", "NR4A2",
"PERP", "ZNF281", "NID1", "ACIN1", "FOXQ1", "MATN2", "IRAK1", "NDE1", "COL8A1",
"TFAP2A", "C1S", "ST6GAL1", "CD97", "ID4", "IGSF3", "NAMPT", "RP11-473I1.1",
"EPB41L3", "NOTCH3", "GALNT5", "NLGN4X", "PITPNC1", "KLF6", "B3GNT9",
"RIPK4", "S100B", "GYG2", "LOC283070", "CAMK1D", "ZNF747", "HOXA7", "NMU",
"RAB7L1", "STX3", "SRGAP3", "CNTN1", "ATP1B2", "MAP7", "ADAMTS4", "F2RL1",
"MIA", "MLC1", "ARHGAP20", "C4orf32", "EDA2R", "OTX2", "MEIS2", "SPRED1",
"PPEF1", "TCF7", "TFCP2", "HPRT1", "TTF2", "SLITRK5", "PXDN", "MIPOL1",
"SYT1", "CXCL14", "ETS2", "PPHLN1", "NBL1", "GDPD2", "SDC2", "ADAMTS10",
"C6orf138", "ICAM1", "SELENBP1", "C7orf40", "GALR1", "SFRP1", "IGSF11", "MTTP",
"RRS1", "RGAG4", "BTBD11", "PHLPP1", "RGS17", "NCAM1", "MID1IP1", "TMEM200B",
"CRYBB1", "KANK1", "MT1A", "SHOX2", "PLS1", "H1F0", "SEMA3D", "ITGA4",
"ZIC4", "SPINK2", "SLC38A5", "LINGO1", "FAM184B", "NR1D1", "SQRDL", "SPAG4",
"HOXD9", "HAND2", "LIFR", "MYO1B", "TPMT", "PCDHB3", "SHISA2", "XYLT1",
"REC8", "NFATC2", "TOX", "STC2", "FAM126A", "GAL", "CPAMD8", "HLA-DQA1",
"SRPX", "GJB2", "HOXB6", "HIF3A", "MXRA5", "ITGBL1", "LGALS3", "C5orf38",
"IL1RAPL1", "TSPAN13", "AC007405.8", "LOC285141", "TMSB15A", "ADAMTS1", "KCNK12",
"PARP12", "DCHS1", "TSTD1", "SEZ6L", "SNHG12", "IL4I1", "CCDC48", "NKX6-2",
"OAS3", "TUBB4", "CDH19", "GJC3", "TMEM100", "DDIT4L", "ARHGAP8", "PRR5",
"PRR5-ARHGAP8", "DDAH2", "ADCYAP1R1", "C3orf58", "IL33", "IL1R1", "CD68",
"MARS", "ALDH1A3", "C2orf80", "C7orf16", "KIAA1217", "AC067930.1", "CD248", "CMTM5",
"HRCT1", "CCR1", "FAM129A", "PIGA", "PAX8", "GRB14", "ADD2", "CTLA4", "DYNC1I1",
"KCNK1", "LRRC55", "PDGFRB", "SYTL5", "TMEM158", "PROCR", "PSPH", "CRIP2",
"CBLC", "TMEM38A", "INSIG1", "ST6GALNAC3", "NOP16", "TPM1", "FAM176A",
"PPM1K", "PNMA2", "OXTR", "TRAF1", "TRIB2", "TRIM14", "AQP4", "PEA15",
"PRSS12", "LPHN2", "CD58", "NFASC", "CHRDL1", "AC026410.6", "IKBKE", "CRB2",
"SRP9", "IFITM8P", "SLC16A3", "ANGPTL2", "COL4A6", "LFNG", "FZD3", "CDH6",
"PPP4R1", "LAMA4", "LNX1", "EFNA5", "TMEM71", "RPSAP52", "SYNGR1", "GABRA5",
"TESC", "TTYH1", "FAM181A", "HOXC13", "C1orf133", "LRP4", "FERMT3", "CDKN2C",
"TSPAN9", "CXorf38", "HOXA1", "TTN", "TMEM176B", "CPNE2", "SALL1", "SLC26A2",
"ZEB1", "GLYATL2", "RHOBTB3", "EFHD2", "MLPH", "MFAP2", "PTPRR", "RRP7A",
"SNX10", "ZNF714", "MACROD2", "GMPR", "BMPER", "AIDA", "FAM5B", "PLCB1",
"ADRA1B", "ELMOD1", "RP11-93B14.2", "hCG_2018279", "SYTL4", "NFIA", "CMAH",
"PLAC9", "EVC2", "SCN11A", "RASGRP3", "RHPN1", "PTPRD", "PDE4B", "CXCL12",
"CCKBR", "ISYNA1", "SDK2", "TNFAIP3", "ACTA2", "PCDHB12", "AMMECR1",
"PKNOX2", "PTPRH", "DUSP16", "SULF1", "RP9", "GNG11", "C5orf41", "C9orf95",
"DCBLD2", "RAD50", "RASGEF1C", "PCOLCE2", "PNP", "PLEKHA6", "HEPACAM",
267

Appendix
"TAPBPL", "MEST", "EEF1A2", "C1orf187", "CALM1", "CREB3L1", "CCNY", "MGC87042",
"TCF7L2", "ZNF710", "JAM2", "SCN5A", "SLC46A1", "TMSL1", "C20orf103", "HOXA5",
"FMNL1", "CACNG7", "TF", "RNASEH1", "UNC80", "EGFR", "PTGS1", "ASPN",
"SEMA4G", "ASNS", "FGFR1", "CHODL", "IRX1", "INPP5D", "PLSCR1", "NCALD",
"CLDN3", "COL1A2", "GGH", "MICB", "FNDC5", "RAB11FIP1", "DKFZp434H1419",
"AC012513.4", "CGREF1", "RP5-955M13.1", "KCNG1", "GFPT2", "IGFBP5", "CCDC8",
"RADIL", "BID", "PPL", "RGS20", "BMP7"}
Dim dbUrls() As String = {"www.gbmbase.org", "www.ncbi.nlm.nih.gov/pubmed", "www.ihop-
net.org/UniPub/iHOP/", "biograph.be/", "scholar.google.com/scholar", "www.google.com"}
Dim pageLoadTimes, nDB, ngene As Integer
Public Event nextGene()
Public Event nextDB()
Private Sub ButtonStart_Click(sender As Object, e As EventArgs) Handles ButtonStart.Click
ngene = -1
RaiseEvent nextGene()
End Sub
Private Sub nextGene_() Handles Me.nextGene
ngene += 1
If ngene < genes.Count Then
nDB = 4 ’0 ’-1+
TextBox3.Text = genes(ngene)
RaiseEvent nextDB()
Else
’ End data collection
End If
End Sub
Private Sub nextDB_() Handles Me.nextDB
nDB += 1
If nDB < dbUrls.Count Then
pageLoadTimes = 0
Select Case nDB
Case 0, 3 : WebBrowser1.Navigate(dbUrls(nDB))
Case 1 : WebBrowser1.Navigate(dbUrls(nDB) & "?term=" & genes(ngene) & " glioblas-
toma")
Case 2 : WebBrowser1.Navigate(dbUrls(nDB) & "?search=" & genes(ngene) & " &field=all
&ncbi_tax_id=9606&organism_syn=")
Case 4 : WebBrowser1.Navigate(dbUrls(nDB) & "?as_vis=0&q=" & genes(ngene) & "
+glioblastoma&hl=en&as_sdt=0,5")
Case 5 : WebBrowser1.Navigate(dbUrls(nDB) & "#hl=en&gs_nf=1&cp=10&gs_id=r&xhr=t
&q=" & genes(ngene) & "+NEAR+glioblastoma&fp=1")
268

End Select
End If
End Sub
Appendix
Private Sub WebBrowser1_DocumentCompleted(sender As Object, e As WebBrowserDoc-
umentCompletedEventArgs) Handles WebBrowser1.DocumentCompleted
If nDB = 0 Then ’ GMBbase
Select Case pageLoadTimes
Case 0 ’ Search and fill the "search textbox"
For Each elem As HtmlElement In WebBrowser1.Document.All
Dim NameStr As String = elem.GetAttribute("name")
If ((NameStr IsNot Nothing) AndAlso (NameStr.Length 0)) Then
If NameStr.ToLower().Equals("search") Then
elem.InnerText = genes(ngene)
End If
End If
Next
WebBrowser1.Document.GetElementById("title_submit").InvokeMember("click")
Case 1 ’ Search the link to Gene name
For Each elem As HtmlElement In WebBrowser1.Document.Links
Dim NameStr As String = elem.InnerText
If ((NameStr IsNot Nothing) AndAlso (NameStr.Length 0)) Then
If Trim(elem.InnerText) = genes(ngene) Then
elem.InvokeMember("Click")
Exit Select
End If
End If
Next
Case 2 ’ Get the results from page
Dim st As String = WebBrowser1.DocumentText.ToString
Dim st1 As Integer = st.IndexOf("Glioblastoma multiforme Gene Publications:")
If st1 > 0 Then
TextBox3.Text = Convert.ToInt32(Trim(st.Substring(st1 + 49, 5)))
If TextBox3.Text "0" Then RaiseEvent nextGene() Else RaiseEvent nextDB()
Exit Sub
End If
End Select
ElseIf nDB = 1 Then ’ PubMed
269

Select Case pageLoadTimes
Case 0 ’ Get the results from page
Dim st As String = WebBrowser1.DocumentText.ToString
Dim stp As Integer = st.IndexOf("class=""result_count"">Results: ")
If stp > 0 Then
Appendix
Dim stR As String = st.Substring(stp + 28, 20) ’ ex. : 1 to 20 of 87

If NameStr.ToLower().Equals("query") Then
elem.InnerText = genes(ngene) : Exit For
End If
End If
Next
Appendix
For Each elem As HtmlElement In WebBrowser1.Document.All ’ Search and click submit
button
Dim NameStr As String = elem.GetAttribute("name")
If ((NameStr IsNot Nothing) AndAlso (NameStr.Length 0)) Then
If NameStr.ToLower().Equals("commit") Then
elem.InvokeMember("click") : Exit For
End If
End If
Next
Case 1 ’ Get the result page
Dim st As String = WebBrowser1.DocumentText.ToString
End Select
ElseIf nDB = 4 Then ’ Google Scholar
Select Case pageLoadTimes
Case 0 ’ Get the results from page
Dim st As String = WebBrowser1.DocumentText.ToString
Dim stp As Integer = st.IndexOf("About ")
If stp > 0 Then
Dim stR As String = st.Substring(stp + 24, 20) ’ ex. 18,700 results (0.03 sec)...
stp = stR.IndexOf("results")
st = stR.Substring(0, stp).Replace(",", "") ’ ex. 18,700 –>18700
TextBox3.Text = Convert.ToInt32(Trim(st))
If TextBox3.Text "0" Then RaiseEvent nextGene() Else RaiseEvent nextDB()
Exit Sub
End If
End Select
ElseIf nDB = 5 Then ’ Google Search
Select Case pageLoadTimes
Case 0 ’ Get the results from page
Dim st As String = WebBrowser1.DocumentText.ToString
Dim stp As Integer = st.IndexOf("About ")
If stp > 0 Then
Dim stR As String = st.Substring(stp + 24, 20) ’ex. 18,700 results (0.03 sec)...
271

stp = stR.IndexOf("results")
st = stR.Substring(0, stp).Replace(",", "") ’ex. 18,700 –>18700
TextBox3.Text = Convert.ToInt32(Trim(st))
If TextBox3.Text "0" Then RaiseEvent nextGene()
Exit Sub
End If
End Select
End If
pageLoadTimes += 1
End Sub
End Class
272
Appendix

Appendix C
Long ncRNAs
We detected 25 differentially expressed long non-coding RNAs. Several of these
display an expression pattern similar to a neighbouring protein-coding gene,
including cancer-associated genes DKK1 and CTSC [113,461,551] and devel-
opmental regulators IRX2, SIX3 and ZNF536 [412], suggesting that they may
be functional RNAs regulating nearby genes [372] or represent transcription
from active enhancers [230].
273

C. Long ncRNAs Appendix
Table C.1: Differentially expressed non-coding RNAs.
Tag Accession Description In gene desert? PubMedID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
HOTAIRM1, a ncRNA from the
HOXA locus. There is evidence
CCGCCTTAATAAATGTA BC031342
No 19144990 Inf 7.9E-10 4.3E-06 174.7 158.7 129.8 206.3 0.0 0.0
that it might regulate HOXA gene
expression.
NEAT1, a ncRNA involved in
TACATAATTACTAATCA NR_028272 paraspeckle formation and regu- Yes 20137068; 21170033 4.32 1.6E-06 2.2E-03 123.4 627.4 775.4 1927.1 18.5 91.8
lated during NS cell differentiation.
CDKN2BAS, an antisense transcript
to CDKN2B that appears
TGGATAAACAAAATGAA NR_003529 to function in Polycomb-mediated No 18185590; 20541999 Inf 1.4E-05 1.3E-02 18.0 33.6 125.8 0.0 0.0 0.0
repression of tumor suppressors
CDKN2A and CDKN2B.
Transcript from the opposite strand
TATCCCCAAATAAACAA NR_015382
No None 4.42 3.2E-05 2.4E-02 151.4 123.2 332.4 135.3 9.3 9.4
of CD27.
Transcript from the opposite strand
ATGGCACCATATTGTGT NR_015377
No None -7.08 2.5E-05 2.0E-02 0.0 0.0 2.2 0.0 136.9 58.3
of PAX8.
Transcript from the opposite strand
CCTGACCCTGCATCCCT BC013229
No None Inf 1.4E-04 7.5E-02 1.7 0.0 36.9 68.3 0.0 0.0
of MCF2L2.
Transcript from the opposite strand
TAGAACGGTGTTCCTCC BM977209
No None Inf 1.7E-05 1.5E-02 0.0 0.0 121.6 24.7 0.0 0.0
of TXNRD1.
Transcript from CDKN2B intron
ACAAACAAAAGCCTTCC DB099927
No None Inf 1.4E-05 1.3E-02 41.9 105.5 51.8 0.0 0.0 0.0
(sense strand).
Transcript from gene desert near
ATTACATTTATGTCCTT BC015429 DKK1. Correlated with DKK1 ex- Yes None Inf 2.8E-09 1.1E-05 0.0 0.0 389.5 18.0 0.0 0.0
pression.
TAACTGATCCTTAGATA BQ957425 Transcript from gene desert. Yes None Inf 2.0E-04 9.9E-02 0.0 0.0 66.8 30.8 0.0 0.0
GAAACATTCCAAACCTA AK094154 Transcript from gene desert. Yes None Inf 1.3E-10 9.9E-07 0.0 0.0 328.5 243.9 0.0 0.0
CAAATAAACTTTATACC BC063641 Transcript downstream of PRRG4. No None 7.53 2.4E-11 2.3E-07 4.2 68.0 267.3 1381.8 0.0 5.9
Transcript between CTSC and
GCTTTATTTTTTCTGCT BC038205 GRM5. Correlated with CTSC Yes None 7.46 5.4E-07 8.7E-04 54.0 96.2 49.6 126.3 0.0 1.0
expression.
GAGCCCAGACTAGATGG BF031226 Transcript from gene desert. Yes None Inf 3.9E-05 2.8E-02 1.4 0.0 30.7 99.7 0.0 0.0
274
Yes 12082533; 15872005 -Inf 3.5E-07 6.3E-04 0.0 0.0 0.0 0.0 21.8 283.9
The tag is in last intron of
ncRNA NCRMS, perhaps detecting
an NCRMS isoform. NCRMS might
be a host transcript for mir-1251
and mir-135a-2. The tag also coincides
with a SINE repeat.
AAATATTAGTTTTTCTT CX868766

C. Long ncRNAs Appendix
Tag Accession Description In gene desert? PubMedID(s) Differential expression results Normalised tag counts
log 2(F C) p-value FDR G144ED G144 G166 G179 CB541 CB660
Yes None 5.95 5.4E-05 3.7E-02 4.2 0.0 96.3 95.6 0.0 2.1
AAATATGGATAAATGTA CA425887
Yes None Inf 9.3E-10 4.9E-06 263.9 566.0 8.8 3.7 0.0 0.0
AAATTGGTGCTGTTGCT BM679519
No None -4.65 1.6E-05 1.4E-02 34.0 27.7 4.5 2.4 354.0 224.1
GAAGGTCCCCCAGGGGT AK131287
Yes None Inf 7.3E-06 7.5E-03 12.5 126.1 0.0 50.3 0.0 0.0
Transcript from gene desert around
FOXG1. Overlaps an EST similar
to a region of downstream of the
pseudoautosomal gene SPRY3.
Transcript from gene desert harboring
ZNF536 and TSHZ3. Correlated
with ZNF536 expression.
Intergenic transcript, or possibly a
very long CACNG8 3’-UTR extension.
Might be a host transcript for
mir-935.
Transcript from gene desert around
KCNF1.
Transcript from gene desert upstream
of SIX3. Correlated with
SIX3 expression.
Transcript from gene desert downstream
of SIX3. Correlated with
SIX3 expression.
Transcript from gene desert downstream
of SIX3. Correlated with
SIX3 expression.
Transcript from gene desert near
DCBLD2.
Transcript from gene desert near
IRX2. Correlated with IRX2 ex-
GAATACAGATTAATCCT BG201257
Yes 17084678 -Inf 6.9E-08 1.8E-04 0.0 0.0 0.0 0.0 162.1 173.2
ATCATCACGTGAGAGAT AK126832
Yes 17084678 -Inf 4.2E-07 7.2E-04 0.0 0.0 0.0 0.0 123.7 142.3
TATAATAATAATGCTTA GD259214
275
Yes 17084678 -Inf 1.3E-05 1.2E-02 0.0 0.0 0.0 0.0 44.3 125.7
GAGAGTGAATGTTTAAA CR623536
Yes None 6.14 1.3E-06 1.9E-03 0.0 0.0 103.5 225.2 0.0 3.1
TACACAATAAATATTTA BG166405
Yes None -Inf 2.9E-05 2.2E-02 0.0 0.0 0.0 0.0 80.4 35.2
ATAATAAAAGTATTTTT AA993778
pression.
No (1688605) Inf 1.2E-04 6.9E-02 4.2 6.1 73.4 27.6 0.0 0.0
Transcript downstream of HLA-
F, on opposite strand. Overlaps
an antisense transcript to HLA-F
(NR_026972).
TTTTTCATCAAGAGGAA DB349183

Appendix D
Glioblastoma Pathway
D.1 Pathway Interactions
Table D.1: Network interaction data for the integrated glioblastoma pathway. The
first column identifies the first interactor; the second column identifies the interaction
type; the third column identifies the second interactor. Sorted in alphabetical order
on the first column interactor.
First interactor Interaction Second interactor
AKT1 activates MAP2K7
AKT1 activates MDM2
AKT1 activates TSC2
AKT1 inhibits CDKN1A
AKT1 inhibits CDKN1B
AKT1 inhibits FOXO3
AKT1 inhibits TSC-complex
APAF1 leads-to APOPTOSIS
ATM activates CHEK1
ATM activates PRKDC
ATM activates TP53
BASC-complex includes BRCA1
BASC-complex includes MSH6
BASC-complex leads-to APOPTOSIS
BASC-complex leads-to DNA-REPAIR
BAX inhibits BCL2
BCL2 inhibits APAF1
BDNF activates NTRK2
BID activates BAX
BID activates CYCS
BRCA1 interacts MSH6
BTRC inhibits NFKB
BTRC inhibits PHLPP1
BUB1B inhibits IRS1
BUB1B leads-to CELL-GROWTH
BUB1B leads-to PROTEIN-SYNTHESIS
CACN activates PKC
CACN includes CACNA1A
CACN includes CACNA1C
CACN includes CACNG7
CACN includes CACNG8
CACN lets-in Calcium
CALM1 activates CAMK1D
CALM1 activates CAMK2B
CALM1 activates PPEF1
CALM1 interacts MAPT
CALM1 interacts SNCA
CALM1 interacts SYT1
CAMK1D activates AKT1
CAMK2A activates RAS
CAMK2B interacts CAMK2A
276

D.1 Pathway Interactions Appendix
First interactor Interaction Second interactor
CASP1 activates CASP3
CASP1 activates IL1B
CASP3 activates PARP
CASP8 activates BID
CBL inhibits RTK
CBL interacts CRK
CCND-CDK4/6-complex activates RB1
CCND-CDK4/6-complex includes CCND1
CCND-CDK4/6-complex includes CCND2
CCND-CDK4/6-complex includes CDK4
CCND-CDK4/6-complex includes CDK6
CCNE-CDK2-complex activates RB1
CCNE-CDK2-complex includes CCNE1
CCNE-CDK2-complex includes CDK2
CDC25C activates CDK1
CDK1 leads-to G2-M-PROGRESSION
CDKN1A inhibits CCND-CDK4/6-complex
CDKN1A inhibits CCNE-CDK2-complex
CDKN1B inhibits CCNE-CDK2-complex
CDKN2A inhibits CCND-CDK4/6-complex
CDKN2A:ARF inhibits MDM2
CDKN2C inhibits CCND-CDK4/6-complex
CEBPB activates DEC1
CEBPB activates FOSL2
CEBPB activates STAT3
CEBPB activates_transcription CEBPB
CHEK1 inhibits CDC25C
CPLX2 inhibits SNARE-complex
CPLX2 interacts STX1A
CREBBP inhibits TP53
CREBBP interacts CREB1
CREBBP interacts JUN
CREBBP interacts MYC
CREBBP interacts NFATC2
CRK interacts GAB1
CYCS activates CASP9
Calcium activates CALM1
Calcium activates SYT1
Cytosolic-antigen activates HLA-A
DAG activates PKC
DDIT3 leads-to APOPTOSIS
DEC1 activates RUNX1
DNA-DAMAGE activates ATM
DNA-DAMAGE activates GADD45
DOCK1 interacts CBL
DOCK1 interacts CRK
DOCK1 interacts ERBB2
DOCK1 interacts PIK3R1
DUSP16 inhibits MAPK14
DUSP16 inhibits MAPK8
DUSP5 inhibits MAPK14
DUSP5 inhibits MAPK8
E2F1 leads-to G1-S-PROGRESSION
EGF interacts EGFR
EGFR activates CASP1
EIF4E leads-to CELL-GROWTH
EIF4E leads-to PROTEIN-SYNTHESIS
EIF4EBP1 inhibits EIF4E
ELK1 interacts ELK4
ELK1 interacts SRF
ELK4 interacts MYC
ELK4 interacts SRF
EP300 activates TP53
EPHB2 activates CCND-CDK4/6-complex
EPHB2 inhibits TSC-complex
EPHB2 leads-to CELL-CYCLE-PROGRESSION
ERBB2 interacts EGFR
ERBB2 interacts PIK3R1
ERRFI1 inhibits EGFR
Endocytosed-antigen activates MHC-classII
FAS leads-to APOPTOSIS
FASLG activates FAS
FGF activates FGFR
FGF includes FGF19
FGF includes FGF23
FGF23 interacts CBL
277

D.1 Pathway Interactions Appendix
First interactor Interaction Second interactor
FGFR includes FGFR1
FOSL2 activates DEC1
FOSL2 activates RUNX1
FOXO includes FOXO3
FOXO leads-to APOPTOSIS
FOXO3 activates-transcription CDKN1B
FOXO3 activates-transcription FASLG
GAB1 activates PI3K-class1a
GADD45 activates CCND-CDK4/6-complex
GADD45 activates MAP3K4
GADD45 interacts PCNA
GRB2 activates GAB1
GRB2 activates SOS1
HIF1A leads-to HYPOXIA
HLA-A leads-to CD8-T-CELL-ACTIVATION
HLA-DM activates MHC-classII
HLA-DM includes HLA-DMA
HLA-DM includes HLA-DMB
HLA-DM inhibits CD74
HLA-DRA interacts CBL
IFI30 activates Endocytosed-antigen
IFNG activates IFI30
IFNG activates Immunoproteasome
IGF1 activates IGF1R
IGF2 leads-to APOPTOSIS
IGFBP5 inhibits IGF2
IKBKE activates NFKB
IKBKE interacts MAP3K14
IL1B activates IL1R1
IL1B leads-to APOPTOSIS
IL1B leads-to DIFFERENTIATION
IL1B leads-to INFLAMMATION
IL1B leads-to PROLIFERATION
IL1R1 activates PIK3R1
IL1R1 activates TRAF2
IL1R1 activates TRAF6
ILK activates AKT1
IP3 activates ITPR1
IRAK1 activates MAP3K14
IRAK1 activates MAP3K1
IRS1 activates PI3K-class1a
ITGB activates ILK
ITGB activates PTK2
ITGB activates SHC
ITGB includes ITGA4
ITGB includes ITGBL1
ITPR1 lets-in Calcium
Immunoproteasome activates Cytosolic-antigen
JUN leads-to DIFFERENTIATION
JUN leads-to INFLAMMATION
JUN leads-to PROLIFERATION
MAP2K1/2 activates MAPK1/2
MAP2K4 activates MAPK14
MAP2K6 activates MAPK14
MAP2K7 activates MAPK8
MAP3K14 tentatively-activates NFKB
MAP3K4 activates MAP2K4
MAP3K4 interacts TRAF4
MAP3K5 activates AKT1
MAP3K7 activates MAP3K14
MAPK1 activates ELK1
MAPK1 activates MAPT
MAPK1/2 activates RPS6KA3
MAPK1/2 inhibits MAP2K1/2
MAPK1/2 inhibits SOS1
MAPK1/2 inhibits TSC2
MAPK14 activates DDIT3
MAPK8 activates JUN
MAPK8 inhibits NFATC2
MAPT interacts S100B
MAPT interacts SNCA
MAPT leads-to MICROTUBULE-DISASSEMBLY
MDM2 inhibits RB1
MDM2 inhibits TP53
MDM4 inhibits TP53
MHC-classI includes HLA-A
278

D.1 Pathway Interactions Appendix
First interactor Interaction Second interactor
MHC-classII includes HLA-DPA1
MHC-classII includes HLA-DPB1
MHC-classII includes HLA-DQA1
MHC-classII includes HLA-DQA2
MHC-classII includes HLA-DQB1
MHC-classII includes HLA-DRA
MHC-classII includes HLA-DRB5
MHC-classII interacts CD74
MHC-classII leads-to CD4-T-CELL-ACTIVATION
NFATC2 interacts EP300
NFKB leads-to ANTI-APOPTOSIS
NFKB leads-to INFLAMMATION
NFKB leads-to PROLIFERATION
NFKBIZ inhibits NFKB
NGF activates NTRK1
NR activates MAP2K6
NR includes NR0B1
NR includes NR1D1
NR includes NR4A2
NTRK1 activates TRAF6
NTRK1 interacts GRB2
NTRK1 interacts SHC1
NTRK2 activates TRAF6
NTRK2 interacts SHC1
PARP includes PARP12
PARP includes PARP3
PARP leads-to APOPTOSIS
PDGF activates PDGFR
PDGFR includes PDGFRA
PDGFR includes PDGFRB
PDGFR interacts CBL
PDGFR interacts CRK
PDGFR interacts PI3K-class1a
PDK1 activates AKT1
PDPK1 activates AKT1
PEA15 interacts CASP8
PEA15 interacts MAPK1/2
PERP activates CYCS
PHLPP1 inhibits AKT1
PI3K includes PI3K-class1a
PI3K includes PI3K-class3
PI3K-class1a activates PIP2
PI3K-class1a interacts RB1
PI3K-class3 activates TORC1-complex
PIP2 becomes DAG
PIP2 becomes IP3
PIP2 becomes PIP3
PIP3 activates ILK
PIP3 activates PDK1
PIP3 becomes PIP2
PIP3 interacts AKT1
PKC activates MAP2K1/2
PKC activates RAF1
PKC activates RAS
PLC activates PIP2
PLCG activates PIP2
PLCG activates PKC
PRKAB1 activates TSC-complex
PRKDC activates TP53
PTEN inhibits AKT1
PTEN inhibits PIP3
PTEN inhibits PTK2
PTEN inhibits SHC
PTEN inhibits TP53
PTEN interacts GAB1
PTEN interacts PIK3CA
PTEN interacts PIK3R1
PTK2 activates MAP3K4
PTK2 activates PI3K-class1a
PTK2 inhibits TSC-complex
PTK2 interacts GRB2
PTK2 leads-to ACTIN-ORGANIZATION
RAF1 activates MAP2K1/2
RAS activates PI3K
RAS activates RAF1
RAS includes HRAS
279

D.1 Pathway Interactions Appendix
First interactor Interaction Second interactor
RAS includes KRAS
RAS includes NRAS
RB1 inhibits E2F1
RHEB activates TORC1-complex
RPS6KA3 activates BUB1B
RPS6KA3 activates STK11
RPS6KA3 inhibits TSC2
RPS6KA3 interacts CREBBP
RPS6KA3 interacts PEA15
RTK activates IRS1
RTK activates PI3K
RTK activates PLCG
RTK activates SHC
RTK activates SRC
RTK includes EGFR
RTK includes FGFR
RTK includes IGF1R
RTK includes MET
RTK includes PDGFR
RTK interacts CRK
RUNX1 activates JUN
RUNX1 leads-to MESENCHYMAL-TRANSFORMATION
S100B inhibits TP53
SERPINB2 leads-to INHIBITION-ANGIOGENESIS
SERPINB2 leads-to INHIBITION-METASTASIS
SERPINE2 interacts SERPINB2
SHC activates GRB2
SHC includes SHC1
SHC includes SHC4
SHC4 interacts NTRK2
SHISA5 interacts PERP
SNARE-complex includes SNAP25
SNARE-complex includes STX1A
SNCA interacts CAMK2B
SNCA interacts SNCAIP
SOS1 activates RAS
SPRY2 inhibits CBL
SPRY2 inhibits RAS
SRC activates GRB2
SRC activates PI3K-class1a
SRC inhibits PTEN
SRF interacts DNA
STAT3 activates FOSL2
STAT3 activates RUNX1
STAT3 activates_transcription STAT3
STK11 activates PRKAB1
SYT1 activates SNARE-complex
TAB2 activates MAP3K7
TGFA activates EGFR
TGFB activates TGFBR
TGFBR activates NR
TLR4 activates TRAF6
TNF activates TNFR
TNF includes TNFSF4
TNFR activates TRAF2
TNFR includes TNFRSF14
TORC1-complex activates BUB1B
TORC1-complex includes MLST8
TORC1-complex includes MTOR
TORC1-complex includes RPTOR
TORC1-complex inhibits EIF4EBP1
TORC2-complex activates AKT1
TORC2-complex includes MAPKAP1
TORC2-complex includes MLST8
TORC2-complex includes MTOR
TORC2-complex includes RICTOR
TORC2-complex leads-to ACTIN-ORGANIZATION
TP53 activates BRCA1
TP53 activates HIF1A
TP53 activates MDM2
TP53 activates S100B
TP53 activates-transcription BAX
TP53 activates-transcription CDKN1A
TP53 activates-transcription FAS
TP53 activates-transcription GADD45
TP53 activates-transcription IGFBP5
280

D.1 Pathway Interactions Appendix
First interactor Interaction Second interactor
TP53 activates-transcription PTEN
TP53 activates-transcription SERPINE2
TP53 activates-transcription SHISA5
TP53 inhibits CDK1
TP53 inhibits CDK2
TP53 inhibits-transcription BCL2
TP53 inhibits-transcription TIMP3
TP53 interacts CDKN2C
TP53 leads-to APOPTOSIS
TRAF2 activates MAP3K5
TRAF4 activates MAPK8
TRAF6 activates IRAK1
TRAF6 interacts TAB2
TSC-complex includes TSC1
TSC-complex includes TSC2
TSC2 inhibits RHEB
281

D.2 Pathway Images Appendix
D.2 Pathway Images
Figure D.1: Integrated GBM pathway overlaid with Tag-seq G144 expression data.
The colour intensity of the nodes (green) indicates the magnitude of the expression.
282

D.2 Pathway Images Appendix
Figure D.2: Integrated GBM pathway overlaid with Tag-seq G144ED expression
data. The colour intensity of the nodes (green) indicates the magnitude of the
expression.
283

D.2 Pathway Images Appendix
Figure D.3: Integrated GBM pathway with Tag-seq G166 expression data. The
colour intensity of the nodes (green) indicates the magnitude of the expression.
284

D.2 Pathway Images Appendix
Figure D.4: Integrated GBM pathway with Tag-seq G179 expression data. The
colour intensity of the nodes (green) indicates the magnitude of the expression.
285

Appendix E
Exon Array Data
Table E.1: Fold-changes measured by exon array and filtered at FDR

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
4.92 -2.647 0.000 0.000000 ENSG00000141682 PMAIP1
7.47 -1.782 0.000 0.000000 ENSG00000144730 IL17RD
7.37 -1.397 0.000 0.000000 ENSG00000145012 LPP
7.8 -1.172 0.000 0.000000 ENSG00000145431 PDGFC
4.45 -2.471 0.000 0.000000 ENSG00000145708 CRHBP
5.27 -1.439 0.000 0.000000 ENSG00000147234 FRMPD3
7.75 -4.124 0.000 0.000000 ENSG00000149591 TAGLN
5.48 -3.438 0.000 0.000000 ENSG00000151150 ANK3
6.55 -3.912 0.000 0.000000 ENSG00000151388 ADAMTS12
5.76 -4.527 0.000 0.000000 ENSG00000151572 ANO4
7.52 -5.064 0.000 0.000000 ENSG00000151892 GFRA1
8.11 -2.063 0.000 0.000000 ENSG00000152154 TMEM178A
3.22 -1.658 0.000 0.000000 ENSG00000153993 SEMA3D
5.1 -2.672 0.000 0.000000 ENSG00000155966 AFF2
5.84 -2.395 0.000 0.000000 ENSG00000156675 RAB11FIP1
5.63 -1.037 0.000 0.000000 ENSG00000157110 RBPMS
5.49 -2.472 0.000 0.000000 ENSG00000159217 IGF2BP1
5.97 -1.422 0.000 0.000000 ENSG00000159433 STARD9
6.62 -1.718 0.000 0.000000 ENSG00000162849 KIF26B
4.79 -3.556 0.000 0.000000 ENSG00000163071 SPATA18
7.66 -1.821 0.000 0.000000 ENSG00000163110 PDLIM5
6.5 -1.878 0.000 0.000000 ENSG00000163297 ANTXR2
7.09 -3.748 0.000 0.000000 ENSG00000163814 CDCP1
4.72 -1.731 0.000 0.000000 ENSG00000164002 DEM1
5.01 -1.766 0.000 0.000000 ENSG00000164932 CTHRC1
8.12 -1.339 0.000 0.000000 ENSG00000164938 TP53INP1
6.06 -5.065 0.000 0.000000 ENSG00000165588 OTX2
7.09 -2.589 0.000 0.000000 ENSG00000167081 PBX3
7.26 -1.998 0.000 0.000000 ENSG00000167693 NXN
8.27 -1.438 0.000 0.000000 ENSG00000169439 SDC2
6.61 -1.004 0.000 0.000000 ENSG00000170561 IRX2
4.48 -2.288 0.000 0.000000 ENSG00000172123 SLFN12
5.14 -4.167 0.000 0.000000 ENSG00000172260 NEGR1
7.6 -1.24 0.000 0.000000 ENSG00000172667 ZMAT3
5.4 -1.874 0.000 0.000000 ENSG00000173068 BNC2
6.66 -3.789 0.000 0.000000 ENSG00000173530 TNFRSF10D
4.85 -1.134 0.000 0.000000 ENSG00000173535 TNFRSF10C
6.29 -1.907 0.000 0.000000 ENSG00000174099 MSRB3
8.45 -1.552 0.000 0.000000 ENSG00000174136 RGMB
3.84 -1.408 0.000 0.000000 ENSG00000176040 TMPRSS7
7.2 -1.425 0.000 0.000000 ENSG00000176720 BOK
8.22 -1.108 0.000 0.000000 ENSG00000177119 ANO6
6.03 -1.552 0.000 0.000000 ENSG00000178573 MAF
6.78 -4.433 0.000 0.000000 ENSG00000184613 NELL2
4.67 -1.519 0.000 0.000000 ENSG00000184809 C21orf88
6.28 -1.561 0.000 0.000000 ENSG00000184985 SORCS2
5.2 -2.054 0.000 0.000000 ENSG00000185046 ANKS1B
6.45 -4.02 0.000 0.000000 ENSG00000185274 WBSCR17
8.51 -2.037 0.000 0.000000 ENSG00000185567 AHNAK2
7.01 -2.956 0.000 0.000000 ENSG00000189184 PCDH18
6.47 -2.842 0.000 0.000000 ENSG00000196730 DAPK1
7.77 -1.504 0.000 0.000000 ENSG00000196923 PDLIM7
5.87 -3.559 0.000 0.000000 ENSG00000198796 ALPK2
3.59 -2.326 0.000 0.000000 ENSG00000204764 RANBP17
5.82 -2.638 0.000 0.000000 ENSG00000204767 FAM196B
8.15 -1.501 0.000 0.000000 ENSG00000205213 LGR4
5.06 -1.915 0.000 0.000000 ENSG00000206538 VGLL3
4.48 -1.527 0.000 0.000000 ENSG00000213186 TRIM59
5.66 -2.628 0.000 0.000000 ENSG00000244694 PTCHD4
9.77 -1.446 0.000 0.000010 ENSG00000026025 VIM
8.93 -1.491 0.000 0.000010 ENSG00000035403 VCL
7.97 -3.845 0.000 0.000010 ENSG00000079931 MOXD1
5.7 -1.554 0.000 0.000010 ENSG00000080546 SESN1
11.03 -1.357 0.000 0.000010 ENSG00000099194 SCD
9.2 -1.406 0.000 0.000010 ENSG00000100345 MYH9
7.34 -1.425 0.000 0.000010 ENSG00000100918 REC8
4.65 -1.337 0.000 0.000010 ENSG00000107614 TRDMT1
7.08 -1.026 0.000 0.000010 ENSG00000109654 TRIM2
7.28 -1.459 0.000 0.000010 ENSG00000112144 ICK
5.52 -0.778 0.000 0.000010 ENSG00000115648 MLPH
8.11 -3.461 0.000 0.000010 ENSG00000117114 LPHN2
6.19 -1.353 0.000 0.000010 ENSG00000125257 ABCC4
8.36 -2.308 0.000 0.000010 ENSG00000129038 LOXL1
7.12 -2.728 0.000 0.000010 ENSG00000131378 RFTN1
4.4 -2.19 0.000 0.000010 ENSG00000134516 DOCK2
8.18 -1.64 0.000 0.000010 ENSG00000134871 COL4A2
287

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
5.66 -1.108 0.000 0.000010 ENSG00000135299 ANKRD6
6.65 -2.293 0.000 0.000010 ENSG00000139211 AMIGO2
6.01 -0.834 0.000 0.000010 ENSG00000143466 IKBKE
6.94 -1.227 0.000 0.000010 ENSG00000143772 ITPKB
5.93 -3.525 0.000 0.000010 ENSG00000147459 DOCK5
7.08 -1.588 0.000 0.000010 ENSG00000149269 PAK1
4.75 -1.397 0.000 0.000010 ENSG00000149948 HMGA2
8.76 -2.143 0.000 0.000010 ENSG00000150551 LYPD1
5.98 -1.944 0.000 0.000010 ENSG00000150687 PRSS23
8.52 -1.26 0.000 0.000010 ENSG00000150938 CRIM1
7.07 -1.701 0.000 0.000010 ENSG00000151474 FRMD4A
7.81 -1.827 0.000 0.000010 ENSG00000152104 PTPN14
7.53 -1.628 0.000 0.000010 ENSG00000153707 PTPRD
6.11 -1.021 0.000 0.000010 ENSG00000166016 ABTB2
6.79 -1.103 0.000 0.000010 ENSG00000166444 ST5
6.42 -0.938 0.000 0.000010 ENSG00000168140 VASN
6.31 -1.634 0.000 0.000010 ENSG00000171533 MAP6
6.1 -1.588 0.000 0.000010 ENSG00000171843 MLLT3
7.3 -1.553 0.000 0.000010 ENSG00000172175 MALT1
7.77 -1.621 0.000 0.000010 ENSG00000172638 EFEMP2
7.19 -1.756 0.000 0.000010 ENSG00000173848 NET1
6 -1.068 0.000 0.000010 ENSG00000180592 SKIDA1
6.47 -1.659 0.000 0.000010 ENSG00000182985 CADM1
6.33 -2.532 0.000 0.000010 ENSG00000187720 THSD4
9.32 -1.343 0.000 0.000010 ENSG00000188042 ARL4C
7.93 -1.344 0.000 0.000010 ENSG00000197702 PARVA
9.34 -1.438 0.000 0.000020 ENSG00000002586 CD99
7.95 -1.593 0.000 0.000020 ENSG00000065923 SLC9A7
5.65 -1.606 0.000 0.000020 ENSG00000071205 ARHGAP10
7.14 -1.517 0.000 0.000020 ENSG00000072401 UBE2D1
7.46 -1.853 0.000 0.000020 ENSG00000085377 PREP
6.82 -1.943 0.000 0.000020 ENSG00000103460 TOX3
9.2 -1.742 0.000 0.000020 ENSG00000112972 HMGCS1
5.82 -1.582 0.000 0.000020 ENSG00000114861 FOXP1
7.28 -0.983 0.000 0.000020 ENSG00000128739 SNRPN
7.58 -1.104 0.000 0.000020 ENSG00000129474 AJUBA
5.5 -1.192 0.000 0.000020 ENSG00000129682 FGF13
7.08 -1.836 0.000 0.000020 ENSG00000137942 FNBP1L
8.55 -1.31 0.000 0.000020 ENSG00000154380 ENAH
5.64 -2.757 0.000 0.000020 ENSG00000165659 DACH1
9.13 -1.323 0.000 0.000020 ENSG00000166033 HTRA1
7.58 -2.2 0.000 0.000020 ENSG00000168672 FAM84B
7.44 -1.906 0.000 0.000020 ENSG00000170175 CHRNB1
7.08 -0.917 0.000 0.000020 ENSG00000182197 EXT1
4.37 -1.48 0.000 0.000020 ENSG00000183778 B3GALT5
7.62 -3.42 0.000 0.000020 ENSG00000187955 COL14A1
10.42 -1.09 0.000 0.000020 ENSG00000196924 FLNA
3.77 -1.626 0.000 0.000020 ENSG00000203995 ZYG11A
8.82 -1.245 0.000 0.000020 ENSG00000213625 LEPROT
7.54 -1.127 0.000 0.000020 ENSG00000250588 IQCJ-SCHIP1
4.94 -1.757 0.000 0.000020 ENSG00000255994 C14orf162
8.77 -1.205 0.000 0.000030 ENSG00000065150 IPO5
7.74 -1.274 0.000 0.000030 ENSG00000073712 FERMT2
6.31 -1.635 0.000 0.000030 ENSG00000100592 DAAM1
8.63 -1.933 0.000 0.000030 ENSG00000125266 EFNB2
6.64 -3.548 0.000 0.000030 ENSG00000126010 GRPR
7.09 -4.473 0.000 0.000030 ENSG00000132429 POPDC3
8.83 -0.845 0.000 0.000030 ENSG00000137076 TLN1
5.45 -1.938 0.000 0.000030 ENSG00000137831 UACA
9.29 -1.364 0.000 0.000030 ENSG00000138448 ITGAV
7.34 -2.1 0.000 0.000030 ENSG00000139687 RB1
7.03 -2.562 0.000 0.000030 ENSG00000145536 ADAMTS16
8.41 -1.569 0.000 0.000030 ENSG00000148484 RSU1
7.15 -0.853 0.000 0.000030 ENSG00000150457 LATS2
7.54 -1.712 0.000 0.000030 ENSG00000153976 HS3ST3A1
6.68 -2.064 0.000 0.000030 ENSG00000154556 SORBS2
5.6 -1.892 0.000 0.000030 ENSG00000166450 PRTG
7 -1.416 0.000 0.000030 ENSG00000169047 IRS1
5.16 -1.289 0.000 0.000030 ENSG00000183840 GPR39
6.2 -1.112 0.000 0.000030 ENSG00000203772 SPRN
6.92 -0.86 0.000 0.000040 ENSG00000013364 MVP
6.65 -1.951 0.000 0.000040 ENSG00000069869 NEDD4
5.28 -1.552 0.000 0.000040 ENSG00000079102 RUNX1T1
6.07 -1.116 0.000 0.000040 ENSG00000085741 WNT11
6.6 -1.686 0.000 0.000040 ENSG00000099284 H2AFY2
7.93 -1.989 0.000 0.000040 ENSG00000101670 LIPG
288

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
6.88 -0.837 0.000 0.000040 ENSG00000107758 PPP3CB
7.91 -1.277 0.000 0.000040 ENSG00000128829 EIF2AK4
8.37 -0.912 0.000 0.000040 ENSG00000136478 TEX2
6.12 -1.235 0.000 0.000040 ENSG00000137502 RAB30
7.56 -1.128 0.000 0.000040 ENSG00000139668 WDFY2
3.8 -1.027 0.000 0.000040 ENSG00000150672 DLG2
7.69 -0.971 0.000 0.000040 ENSG00000163625 WDFY3
10.31 -1.446 0.000 0.000040 ENSG00000168615 ADAM9
7.77 -1.029 0.000 0.000040 ENSG00000181827 RFX7
8.07 -0.637 0.000 0.000040 ENSG00000198752 CDC42BPB
7.51 -0.797 0.000 0.000040 ENSG00000214717 ZBED1
6.19 -1.808 0.000 0.000050 ENSG00000066468 FGFR2
5.66 -2.079 0.000 0.000050 ENSG00000081803 CADPS2
7.49 -0.96 0.000 0.000050 ENSG00000087088 BAX
6.32 -1.122 0.000 0.000050 ENSG00000100968 NFATC4
8.22 -1.297 0.000 0.000050 ENSG00000102893 PHKB
5.64 -1.108 0.000 0.000050 ENSG00000108239 TBC1D12
9 -2.672 0.000 0.000050 ENSG00000112276 BVES
5.95 -2.589 0.000 0.000050 ENSG00000115232 ITGA4
4.59 -1.69 0.000 0.000050 ENSG00000124134 KCNS1
7.48 -2.329 0.000 0.000050 ENSG00000136928 GABBR2
8.18 -1.075 0.000 0.000050 ENSG00000150347 ARID5B
6.01 -2.228 0.000 0.000050 ENSG00000162745 OLFML2B
6.77 -1.239 0.000 0.000050 ENSG00000166073 GPR176
4.66 -3.976 0.000 0.000050 ENSG00000166342 NETO1
4.69 -1.97 0.000 0.000050 ENSG00000172403 SYNPO2
8.1 -0.895 0.000 0.000050 ENSG00000176014 TUBB6
8.3 -1.04 0.000 0.000060 ENSG00000100403 ZC3H7B
7.95 -1.065 0.000 0.000060 ENSG00000143344 RGL1
7.39 -1.509 0.000 0.000060 ENSG00000171862 PTEN
7.99 -0.895 0.000 0.000060 ENSG00000179820 MYADM
8.3 -0.826 0.000 0.000060 ENSG00000187079 TEAD1
5.29 -2.689 0.000 0.000070 ENSG00000101311 FERMT1
9.82 -0.993 0.000 0.000070 ENSG00000104549 SQLE
7.35 -0.867 0.000 0.000070 ENSG00000112655 PTK7
7.33 -0.735 0.000 0.000070 ENSG00000130338 TULP4
10.17 -0.996 0.000 0.000070 ENSG00000131236 CAP1
8.78 -1.538 0.000 0.000070 ENSG00000134824 FADS2
6.55 -0.792 0.000 0.000070 ENSG00000137936 BCAR3
9.79 -0.973 0.000 0.000070 ENSG00000150093 ITGB1
6.93 -1.288 0.000 0.000070 ENSG00000168502 SOGA2
8.38 -1.506 0.000 0.000070 ENSG00000179431 FJX1
8.32 -1.283 0.000 0.000070 ENSG00000197043 ANXA6
5.8 -1.297 0.000 0.000070 ENSG00000198113 TOR4A
7.19 -1.449 0.000 0.000070 ENSG00000205269 TMEM170B
9.7 -1.29 0.000 0.000080 ENSG00000071127 WDR1
7.75 -1.188 0.000 0.000080 ENSG00000073792 IGF2BP2
7.37 -0.762 0.000 0.000080 ENSG00000100139 MICALL1
6.25 -1.627 0.000 0.000080 ENSG00000140557 ST8SIA2
4.93 -2.465 0.000 0.000080 ENSG00000148798 INA
7.75 -1.245 0.000 0.000080 ENSG00000151612 ZNF827
6.56 -1.118 0.000 0.000080 ENSG00000151718 WWC2
7.26 -1.995 0.000 0.000080 ENSG00000174721 FGFBP3
7.34 -1.183 0.000 0.000080 ENSG00000175115 PACS1
7.06 -1.147 0.000 0.000080 ENSG00000182287 AP1S2
7.79 -1.122 0.000 0.000080 ENSG00000187239 FNBP1
6.27 -0.958 0.000 0.000080 ENSG00000229619 AC106722.1
10.55 -1.573 0.000 0.000090 ENSG00000101608 MYL12A
5.59 -1.066 0.000 0.000090 ENSG00000101665 SMAD7
6.15 -1.436 0.000 0.000090 ENSG00000107518 ATRNL1
5.53 -3.929 0.000 0.000090 ENSG00000144227 NXPH2
5.63 -0.965 0.000 0.000090 ENSG00000158246 FAM46B
6.62 -1.761 0.000 0.000090 ENSG00000165323 FAT3
7.74 -1.946 0.000 0.000100 ENSG00000102996 MMP15
8.65 -1.862 0.000 0.000100 ENSG00000104290 FZD3
6.36 -2.28 0.000 0.000100 ENSG00000151136 BTBD11
7.57 -2.996 0.000 0.000100 ENSG00000152661 GJA1
7.28 -0.967 0.000 0.000100 ENSG00000166912 MTMR10
7.44 -1.18 0.000 0.000110 ENSG00000058091 CDK14
8.62 -0.924 0.000 0.000110 ENSG00000073921 PICALM
5.74 -1.788 0.000 0.000110 ENSG00000152128 TMEM163
7.49 -1.413 0.000 0.000110 ENSG00000158966 CACHD1
5.56 -1.457 0.000 0.000110 ENSG00000165449 SLC16A9
3.3 -2.026 0.000 0.000110 ENSG00000170571 EMB
6.66 -0.922 0.000 0.000110 ENSG00000174307 PHLDA3
5.19 -1.959 0.000 0.000110 ENSG00000183691 NOG
289

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
6.73 -0.823 0.000 0.000120 ENSG00000011114 BTBD7
6.86 -0.624 0.000 0.000120 ENSG00000037757 MRI1
4.78 -1.277 0.000 0.000120 ENSG00000068781 STON1-GTF2A1L
9.49 -1.337 0.000 0.000120 ENSG00000075213 SEMA3A
4.78 -1.205 0.000 0.000120 ENSG00000108187 PBLD
7.01 -1.596 0.000 0.000120 ENSG00000109586 GALNT7
6.73 -0.668 0.000 0.000120 ENSG00000110237 ARHGEF17
6.37 -1.075 0.000 0.000120 ENSG00000126790 L3HYPDH
6.21 -1.532 0.000 0.000120 ENSG00000165996 PTPLA
6.16 -1.903 0.000 0.000120 ENSG00000180914 OXTR
9.41 -1.204 0.000 0.000120 ENSG00000196576 PLXNB2
5.41 -1.323 0.000 0.000120 ENSG00000197646 PDCD1LG2
6.14 -0.846 0.000 0.000130 ENSG00000090975 PITPNM2
6.87 -0.981 0.000 0.000130 ENSG00000148737 TCF7L2
7.63 -1.404 0.000 0.000130 ENSG00000177508 IRX3
6.23 -2.345 0.000 0.000140 ENSG00000104332 SFRP1
8.2 -1.368 0.000 0.000140 ENSG00000140682 TGFB1I1
7.55 -0.981 0.000 0.000140 ENSG00000148468 FAM171A1
6.12 -0.812 0.000 0.000140 ENSG00000162804 SNED1
7.73 -1.008 0.000 0.000140 ENSG00000196865 NHLRC2
7.36 -1.145 0.000 0.000150 ENSG00000055163 CYFIP2
6.64 -0.859 0.000 0.000150 ENSG00000072422 RHOBTB1
6.33 -1.765 0.000 0.000150 ENSG00000130176 CNN1
8.66 -0.711 0.000 0.000150 ENSG00000130638 ATXN10
8.32 -0.937 0.000 0.000150 ENSG00000139793 MBNL2
5.8 -1.718 0.000 0.000150 ENSG00000173320 STOX2
7.41 -1.382 0.000 0.000150 ENSG00000176788 BASP1
5.41 -4.49 0.000 0.000150 ENSG00000187714 SLC18A3
6.97 -0.57 0.000 0.000160 ENSG00000061273 HDAC7
8.41 -1.415 0.000 0.000160 ENSG00000163430 FSTL1
6.58 -1.178 0.000 0.000160 ENSG00000169436 COL22A1
8.98 -1.848 0.000 0.000170 ENSG00000135048 TMEM2
8.41 -1.185 0.000 0.000170 ENSG00000137710 RDX
4.5 -1.657 0.000 0.000170 ENSG00000154027 AK5
8.87 -0.936 0.000 0.000170 ENSG00000167460 TPM4
7.64 -1.238 0.000 0.000170 ENSG00000182319 PRAGMIN
8.22 -0.609 0.000 0.000170 ENSG00000182534 MXRA7
8.7 -1.291 0.000 0.000170 ENSG00000186575 NF2
7.62 -2.017 0.000 0.000180 ENSG00000084710 EFR3B
7.84 -0.796 0.000 0.000180 ENSG00000108091 CCDC6
8.51 -1.519 0.000 0.000180 ENSG00000148848 ADAM12
9.53 -0.645 0.000 0.000180 ENSG00000168175 MAPK1IP1L
7.32 -1.041 0.000 0.000180 ENSG00000170525 PFKFB3
7.31 -1.108 0.000 0.000180 ENSG00000178764 ZHX2
6.15 -0.992 0.000 0.000190 ENSG00000116675 DNAJC6
6.52 -1.174 0.000 0.000190 ENSG00000131370 SH3BP5
6.08 -0.978 0.000 0.000190 ENSG00000143816 WNT9A
6.26 -1.511 0.000 0.000190 ENSG00000255103 KIAA0754
6.71 -1.362 0.000 0.000200 ENSG00000122335 SERAC1
5.2 -3.22 0.000 0.000200 ENSG00000123119 NECAB1
5.45 -1.76 0.000 0.000200 ENSG00000138696 BMPR1B
6.28 -2.18 0.000 0.000200 ENSG00000146426 TIAM2
8.62 -0.877 0.000 0.000200 ENSG00000152558 TMEM123
6.89 -0.815 0.000 0.000210 ENSG00000068383 INPP5A
7.22 -1.283 0.000 0.000210 ENSG00000105974 CAV1
9.12 -1.876 0.000 0.000210 ENSG00000106484 MEST
7.38 -1.482 0.000 0.000210 ENSG00000113328 CCNG1
6.45 -1.103 0.000 0.000210 ENSG00000119681 LTBP2
6.12 -0.72 0.000 0.000210 ENSG00000132334 PTPRE
6.68 -1.403 0.000 0.000210 ENSG00000188158 NHS
7.53 -0.946 0.000 0.000230 ENSG00000107819 SFXN3
7.73 -0.772 0.000 0.000230 ENSG00000120254 MTHFD1L
7.79 -0.638 0.000 0.000230 ENSG00000122863 CHST3
3.92 -2.068 0.000 0.000230 ENSG00000158164 TMSB15A
7.94 -0.812 0.000 0.000230 ENSG00000185950 IRS2
7.35 -1.124 0.000 0.000240 ENSG00000067064 IDI1
7.4 -0.902 0.000 0.000240 ENSG00000124788 ATXN1
5.98 -0.83 0.000 0.000240 ENSG00000138131 LOXL4
7.87 -1.371 0.000 0.000240 ENSG00000187498 COL4A1
7.87 -0.797 0.000 0.000240 ENSG00000198951 NAGA
4.77 -1.178 0.000 0.000250 ENSG00000100285 NEFH
7.33 -0.837 0.000 0.000250 ENSG00000103335 PIEZO1
8.26 -2.268 0.000 0.000250 ENSG00000132470 ITGB4
8.65 -1.241 0.000 0.000250 ENSG00000181019 NQO1
5.67 -0.561 0.000 0.000250 ENSG00000198832 RP3-412A9.11
6.86 -0.871 0.000 0.000260 ENSG00000070269 C14orf101
290

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
5.37 -1.251 0.000 0.000260 ENSG00000113448 PDE4D
7.86 -0.688 0.000 0.000260 ENSG00000139645 ANKRD52
4.6 -1.255 0.000 0.000260 ENSG00000165626 BEND7
6.96 -0.585 0.000 0.000260 ENSG00000182541 LIMK2
6.14 -0.788 0.000 0.000270 ENSG00000014914 MTMR11
8.85 -1.086 0.000 0.000270 ENSG00000064666 CNN2
7.38 -0.682 0.000 0.000270 ENSG00000128283 CDC42EP1
8.39 -1.216 0.000 0.000270 ENSG00000162909 CAPN2
8.41 -0.967 0.000 0.000270 ENSG00000176903 PNMA1
5.62 -1.998 0.000 0.000270 ENSG00000182732 RGS6
6.79 -0.841 0.000 0.000280 ENSG00000151553 FAM160B1
7.41 -1.128 0.000 0.000280 ENSG00000170365 SMAD1
8.41 -1.273 0.000 0.000280 ENSG00000183722 LHFP
3.46 -1.447 0.000 0.000290 ENSG00000022556 NLRP2
2.61 -0.607 0.000 0.000290 ENSG00000133640 LRRIQ1
7.3 -1.36 0.000 0.000300 ENSG00000185339 TCN2
5.04 -1.653 0.000 0.000300 ENSG00000215475 SIAH3
9 -1.175 0.000 0.000310 ENSG00000107779 BMPR1A
8.88 -1.175 0.000 0.000310 ENSG00000117298 ECE1
8.39 -0.598 0.000 0.000310 ENSG00000198954 KIAA1279
5.41 -2.252 0.000 0.000320 ENSG00000124610 HIST1H1A
7.1 -1.379 0.000 0.000320 ENSG00000205726 ITSN1
7.71 -0.927 0.000 0.000330 ENSG00000102753 KPNA3
7 -0.851 0.000 0.000330 ENSG00000143553 SNAPIN
9.78 -0.783 0.000 0.000330 ENSG00000147416 ATP6V1B2
5.61 -2.025 0.000 0.000330 ENSG00000188517 COL25A1
6.51 -2.126 0.000 0.000340 ENSG00000152977 ZIC1
7.7 -0.908 0.000 0.000340 ENSG00000154001 PPP2R5E
4.97 -0.768 0.000 0.000340 ENSG00000203877 RIPPLY2
6.95 -1.345 0.000 0.000350 ENSG00000080298 RFX3
8.79 -1.528 0.000 0.000350 ENSG00000082781 ITGB5
8.19 -0.882 0.000 0.000350 ENSG00000175662 TOM1L2
4.95 -2.667 0.000 0.000360 ENSG00000089472 HEPH
9.11 -1.105 0.000 0.000360 ENSG00000102898 NUTF2
7.58 -1.796 0.000 0.000360 ENSG00000125430 HS3ST3B1
9.2 -1.289 0.000 0.000360 ENSG00000172380 GNG12
6.85 -1.643 0.000 0.000370 ENSG00000139263 LRIG3
8.35 -0.864 0.000 0.000370 ENSG00000165476 REEP3
6.9 -1.305 0.000 0.000380 ENSG00000040199 PHLPP2
8.79 -0.701 0.000 0.000380 ENSG00000120733 KDM3B
5 -1.693 0.000 0.000380 ENSG00000143473 KCNH1
6.27 -2.298 0.000 0.000380 ENSG00000173917 HOXB2
6.76 -0.673 0.000 0.000380 ENSG00000205011 AC073082.1
6.92 -1.127 0.000 0.000390 ENSG00000042493 CAPG
7.91 -0.553 0.000 0.000390 ENSG00000204138 PHACTR4
9.42 -0.581 0.000 0.000400 ENSG00000143742 SRP9
8.05 -1.162 0.000 0.000410 ENSG00000107249 GLIS3
4.6 -0.65 0.000 0.000410 ENSG00000172238 ATOH1
8.33 -1.572 0.000 0.000420 ENSG00000152767 FARP1
4.7 -2.758 0.000 0.000420 ENSG00000163762 TM4SF18
7.29 -1.321 0.000 0.000420 ENSG00000179242 CDH4
4.32 -1.336 0.000 0.000420 ENSG00000188316 ENO4
7.53 -0.977 0.000 0.000430 ENSG00000039560 RAI14
8.64 -1.21 0.000 0.000430 ENSG00000128791 TWSG1
7.73 -1.201 0.000 0.000430 ENSG00000130147 SH3BP4
6.59 -1.229 0.000 0.000440 ENSG00000113721 PDGFRB
7.62 -0.694 0.000 0.000440 ENSG00000150760 DOCK1
7.79 -0.655 0.000 0.000450 ENSG00000148411 NACC2
4.02 -1.254 0.000 0.000450 ENSG00000154760 SLFN13
6.46 -1.117 0.000 0.000450 ENSG00000188153 COL4A5
4.84 -1.145 0.000 0.000460 ENSG00000109339 MAPK10
9.11 -2.027 0.000 0.000460 ENSG00000118523 CTGF
7.55 -1.017 0.000 0.000460 ENSG00000137693 YAP1
7.01 -0.558 0.000 0.000460 ENSG00000186174 BCL9L
6.04 -1.369 0.000 0.000460 ENSG00000197565 COL4A6
7.85 -0.791 0.000 0.000460 ENSG00000198856 OSTC
3.62 -0.988 0.000 0.000470 ENSG00000155974 GRIP1
6.5 -0.997 0.000 0.000470 ENSG00000171055 FEZ2
6.12 -0.539 0.000 0.000470 ENSG00000171680 PLEKHG5
10.02 -1.536 0.000 0.000490 ENSG00000041982 TNC
7.6 -0.459 0.000 0.000490 ENSG00000166454 ATMIN
6.43 -0.794 0.001 0.000500 ENSG00000197261 C6orf141
8.15 -1.104 0.001 0.000510 ENSG00000119655 NPC2
5.83 -1.663 0.001 0.000510 ENSG00000146373 RNF217
3.77 -1.114 0.001 0.000510 ENSG00000150540 HNMT
7.94 -0.855 0.001 0.000510 ENSG00000159842 ABR
291

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
7.84 -0.72 0.001 0.000510 ENSG00000214265 SNURF
3.66 -1.512 0.001 0.000510 ENSG00000214919 AC104472.1
5.04 -1.875 0.001 0.000520 ENSG00000102935 ZNF423
6.22 -0.928 0.001 0.000530 ENSG00000106991 ENG
7.3 -0.798 0.001 0.000530 ENSG00000119977 TCTN3
8.03 -1.528 0.001 0.000530 ENSG00000120437 ACAT2
6.41 -1.877 0.001 0.000530 ENSG00000138829 FBN2
3.03 -0.791 0.001 0.000530 ENSG00000162614 NEXN
6.94 -2.214 0.001 0.000540 ENSG00000121005 CRISPLD1
6.77 -1.284 0.001 0.000540 ENSG00000164465 DCBLD1
4.6 -1.496 0.001 0.000540 ENSG00000172671 ZFAND4
5.97 -0.955 0.001 0.000550 ENSG00000162520 SYNC
6.25 -0.89 0.001 0.000550 ENSG00000163485 ADORA1
4.05 -1.925 0.001 0.000550 ENSG00000165983 PTER
7.78 -1.38 0.001 0.000550 ENSG00000180537 RNF182
7.99 -0.599 0.001 0.000560 ENSG00000166333 ILK
5.29 -0.835 0.001 0.000570 ENSG00000139832 RAB20
8.27 -1.102 0.001 0.000580 ENSG00000122026 RPL21
6.95 -0.781 0.001 0.000580 ENSG00000167123 CERCAM
5.84 -2.436 0.001 0.000580 ENSG00000205664 RP11-706O15.1
7.91 -1.209 0.001 0.000590 ENSG00000122884 P4HA1
4.18 -1.509 0.001 0.000590 ENSG00000172716 SLFN11
8.52 -0.9 0.001 0.000600 ENSG00000149485 FADS1
6.61 -0.601 0.001 0.000610 ENSG00000087903 RFX2
5.67 -0.674 0.001 0.000610 ENSG00000095539 SEMA4G
7.5 -0.85 0.001 0.000610 ENSG00000160584 SIK3
7.03 -0.509 0.001 0.000610 ENSG00000166507 NDST2
7.78 -0.812 0.001 0.000620 ENSG00000074181 NOTCH3
6.24 -0.729 0.001 0.000630 ENSG00000157322 CLEC18A
9.2 -1.574 0.001 0.000640 ENSG00000112186 CAP2
7.07 -2.615 0.001 0.000650 ENSG00000087510 TFAP2C
7.91 -0.832 0.001 0.000650 ENSG00000142173 COL6A2
8.58 -0.86 0.001 0.000650 ENSG00000197965 MPZL1
5.92 -0.73 0.001 0.000660 ENSG00000124831 LRRFIP1
6.9 -1.51 0.001 0.000670 ENSG00000104783 KCNN4
3.47 -0.881 0.001 0.000670 ENSG00000146038 DCDC2
5.8 -0.887 0.001 0.000670 ENSG00000198768 APCDD1L
6.88 -0.976 0.001 0.000680 ENSG00000148429 USP6NL
8.6 -1.479 0.001 0.000680 ENSG00000169855 ROBO1
6.76 -0.914 0.001 0.000680 ENSG00000181649 PHLDA2
7.55 -0.953 0.001 0.000690 ENSG00000014216 CAPN1
5.39 -0.948 0.001 0.000690 ENSG00000112183 RBM24
7.14 -0.869 0.001 0.000690 ENSG00000118263 KLF7
7.46 -0.671 0.001 0.000700 ENSG00000148498 PARD3
5.7 -1.198 0.001 0.000710 ENSG00000135824 RGS8
6.54 -1.323 0.001 0.000720 ENSG00000161958 FGF11
6.12 -1.585 0.001 0.000720 ENSG00000172059 KLF11
8.33 -0.669 0.001 0.000720 ENSG00000172081 MOB3A
6.15 -0.95 0.001 0.000720 ENSG00000239887 C1orf226
6.85 -1.301 0.001 0.000730 ENSG00000123096 SSPN
6.8 -0.95 0.001 0.000740 ENSG00000196935 SRGAP1
6.28 -0.714 0.001 0.000750 ENSG00000157335 CLEC18C
6.9 -2.327 0.001 0.000760 ENSG00000146411 SLC2A12
7.01 -1.299 0.001 0.000760 ENSG00000197093 GAL3ST4
7.62 -0.757 0.001 0.000770 ENSG00000137216 TMEM63B
5.99 -1.143 0.001 0.000770 ENSG00000144218 AFF3
8.19 -1.003 0.001 0.000780 ENSG00000173457 PPP1R14B
6.95 -1.347 0.001 0.000790 ENSG00000065534 MYLK
9.56 -1.273 0.001 0.000790 ENSG00000124942 AHNAK
8.75 -1.094 0.001 0.000790 ENSG00000160285 LSS
6.52 -1.65 0.001 0.000790 ENSG00000172469 MANEA
6.01 -0.921 0.001 0.000790 ENSG00000173715 C11orf80
4.28 -1.027 0.001 0.000790 ENSG00000197140 ADAM32
4.65 -1.509 0.001 0.000800 ENSG00000169946 ZFPM2
7.97 -1.128 0.001 0.000800 ENSG00000183098 GPC6
5.92 -1.632 0.001 0.000810 ENSG00000003436 TFPI
7.2 -0.498 0.001 0.000810 ENSG00000070366 SMG6
6.43 -0.515 0.001 0.000830 ENSG00000172346 CSDC2
4.78 -0.836 0.001 0.000830 ENSG00000228120 AP001631.10
8.58 -0.967 0.001 0.000840 ENSG00000145349 CAMK2D
8.74 -0.995 0.001 0.000850 ENSG00000068793 CYFIP1
7.38 -0.478 0.001 0.000850 ENSG00000197324 LRP10
6.39 -0.685 0.001 0.000860 ENSG00000065665 SEC61A2
6.27 -0.754 0.001 0.000860 ENSG00000140839 CLEC18B
4.96 -1.366 0.001 0.000870 ENSG00000150630 VEGFC
7.31 -0.864 0.001 0.000870 ENSG00000151929 BAG3
292

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
5.94 -1.001 0.001 0.000870 ENSG00000165406 Mar-08
7.37 -0.827 0.001 0.000880 ENSG00000126773 PCNXL4
8.75 -0.652 0.001 0.000890 ENSG00000138107 ACTR1A
8.13 -0.645 0.001 0.000900 ENSG00000132142 ACACA
2.65 -0.596 0.001 0.000900 ENSG00000151338 MIPOL1
7.65 -2.89 0.001 0.000910 ENSG00000145623 OSMR
4.78 -1.089 0.001 0.000930 ENSG00000164484 TMEM200A
8.17 -0.961 0.001 0.000930 ENSG00000180776 ZDHHC20
6.74 -1.118 0.001 0.000940 ENSG00000165512 ZNF22
3.35 -0.987 0.001 0.000950 ENSG00000174792 C4orf26
6.79 -1.011 0.001 0.000950 ENSG00000183580 FBXL7
7.28 -0.97 0.001 0.000960 ENSG00000109536 FRG1
5.53 -0.609 0.001 0.000960 ENSG00000123191 ATP7B
8.79 -0.678 0.001 0.000960 ENSG00000152022 LIX1L
6.78 -1.182 0.001 0.000960 ENSG00000164251 F2RL1
7.8 -0.836 0.001 0.000970 ENSG00000147100 SLC16A2
7.29 -0.674 0.001 0.000990 ENSG00000130643 CALY
6.22 -0.585 0.001 0.000990 ENSG00000198546 ZNF511
4.79 2.559 0.000 0.000000 ENSG00000005020 SKAP2
5.52 1.043 0.000 0.000000 ENSG00000008056 SYN1
8.25 3.134 0.000 0.000000 ENSG00000010278 CD9
4.2 2.503 0.000 0.000000 ENSG00000064763 FAR2
6.63 0.942 0.000 0.000000 ENSG00000085117 CD82
4.32 2.793 0.000 0.000000 ENSG00000086300 SNX10
4.71 2.29 0.000 0.000000 ENSG00000087303 NID2
7.32 0.863 0.000 0.000000 ENSG00000105379 ETFB
3.96 1.434 0.000 0.000000 ENSG00000105499 PLA2G4C
3.11 1.41 0.000 0.000000 ENSG00000105889 STEAP1B
5.57 1.454 0.000 0.000000 ENSG00000106789 CORO2A
5.47 1.651 0.000 0.000000 ENSG00000114948 ADAM23
5.72 0.898 0.000 0.000000 ENSG00000119431 HDHD3
7.21 4.479 0.000 0.000000 ENSG00000124785 NRN1
7.81 0.974 0.000 0.000000 ENSG00000128563 PRKRIP1
5.96 1.277 0.000 0.000000 ENSG00000128709 HOXD9
6.11 3.085 0.000 0.000000 ENSG00000128710 HOXD10
5.82 1.922 0.000 0.000000 ENSG00000128713 HOXD11
6.64 1.409 0.000 0.000000 ENSG00000130203 APOE
5.62 4.15 0.000 0.000000 ENSG00000134853 PDGFRA
3.79 1.54 0.000 0.000000 ENSG00000137727 ARHGAP20
5.02 1.678 0.000 0.000000 ENSG00000140022 STON2
5.67 2.034 0.000 0.000000 ENSG00000142494 SLC47A1
7.69 1.182 0.000 0.000000 ENSG00000147883 CDKN2B
4.8 2.142 0.000 0.000000 ENSG00000153029 MR1
3.98 1.624 0.000 0.000000 ENSG00000154493 C10orf90
7.05 1.693 0.000 0.000000 ENSG00000154537 FAM27C
4.8 1.635 0.000 0.000000 ENSG00000156395 SORCS3
5.16 1.341 0.000 0.000000 ENSG00000160179 ABCG1
5.79 2.664 0.000 0.000000 ENSG00000163235 TGFA
3.18 1.87 0.000 0.000000 ENSG00000164647 STEAP1
4.75 1.82 0.000 0.000000 ENSG00000167216 KATNAL2
5.26 2.322 0.000 0.000000 ENSG00000167306 MYO5B
5.67 2.245 0.000 0.000000 ENSG00000168234 TTC39C
5.37 1.902 0.000 0.000000 ENSG00000168779 SHOX2
8.11 1.311 0.000 0.000000 ENSG00000169919 GUSB
7.18 1.878 0.000 0.000000 ENSG00000170215 FAM27B
5.26 1.276 0.000 0.000000 ENSG00000171729 TMEM51
4.33 2.548 0.000 0.000000 ENSG00000173083 HPSE
6.17 1.425 0.000 0.000000 ENSG00000173805 HAP1
6.57 3.093 0.000 0.000000 ENSG00000175197 DDIT3
6.62 3.734 0.000 0.000000 ENSG00000176165 FOXG1
7.24 1.431 0.000 0.000000 ENSG00000178381 ZFAND2A
3.53 1.663 0.000 0.000000 ENSG00000178538 CA8
7.2 1.913 0.000 0.000000 ENSG00000182368 FAM27A
4.39 2.726 0.000 0.000000 ENSG00000186960 C14orf23
7.12 1.148 0.000 0.000000 ENSG00000187244 BCAM
5.4 1.328 0.000 0.000000 ENSG00000187764 SEMA4D
6.42 4.099 0.000 0.000000 ENSG00000189058 APOD
3.76 1.495 0.000 0.000000 ENSG00000196954 CASP4
5.08 1.747 0.000 0.000000 ENSG00000204634 TBC1D8
5.43 1.793 0.000 0.000000 ENSG00000214575 CPEB1
4.75 2.361 0.000 0.000000 ENSG00000223572 CKMT1A
4.81 2.497 0.000 0.000000 ENSG00000237289 CKMT1B
5.34 2.004 0.000 0.000000 ENSG00000258518 AC112502.1
6.52 1.423 0.000 0.000010 ENSG00000003096 KLHL13
5.29 2.135 0.000 0.000010 ENSG00000019549 SNAI2
4.56 1.944 0.000 0.000010 ENSG00000056736 IL17RB
293

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
8.68 0.866 0.000 0.000010 ENSG00000073578 SDHA
6.03 3.092 0.000 0.000010 ENSG00000089127 OAS1
3.57 1.6 0.000 0.000010 ENSG00000092607 TBX15
4.82 1.835 0.000 0.000010 ENSG00000102445 KIAA0226L
4.8 1.261 0.000 0.000010 ENSG00000103723 AP3B2
7.97 0.683 0.000 0.000010 ENSG00000106263 EIF3B
7.77 1.071 0.000 0.000010 ENSG00000121083 DYNLL2
6.83 1.603 0.000 0.000010 ENSG00000123146 CD97
6.29 0.672 0.000 0.000010 ENSG00000125746 EML2
7.19 0.883 0.000 0.000010 ENSG00000130305 NSUN5
7.03 2.957 0.000 0.000010 ENSG00000146072 TNFRSF21
4.7 1.327 0.000 0.000010 ENSG00000151117 TMEM86A
4.81 1.072 0.000 0.000010 ENSG00000153790 C7orf31
5.74 2.749 0.000 0.000010 ENSG00000166741 NNMT
5.11 1.937 0.000 0.000010 ENSG00000186088 PION
4.71 1.77 0.000 0.000010 ENSG00000187098 MITF
3.72 1.906 0.000 0.000010 ENSG00000204397 CARD16
7.31 1.009 0.000 0.000010 ENSG00000232098 AC012313.1
6.94 1.111 0.000 0.000020 ENSG00000072071 LPHN1
5.36 2.843 0.000 0.000020 ENSG00000089041 P2RX7
7.79 0.858 0.000 0.000020 ENSG00000089057 SLC23A2
5.99 1.358 0.000 0.000020 ENSG00000096433 ITPR3
8.6 0.812 0.000 0.000020 ENSG00000101474 APMAP
6.01 0.937 0.000 0.000020 ENSG00000103241 FOXF1
7.4 0.773 0.000 0.000020 ENSG00000105518 TMEM205
4.7 3.12 0.000 0.000020 ENSG00000106688 SLC1A1
4.11 1.188 0.000 0.000020 ENSG00000107864 CPEB3
6.82 1.511 0.000 0.000020 ENSG00000110841 PPFIBP1
6.66 1.511 0.000 0.000020 ENSG00000117280 RAB7L1
8.48 1.567 0.000 0.000020 ENSG00000125148 MT2A
5.93 2.858 0.000 0.000020 ENSG00000125820 NKX2-2
6.31 2.012 0.000 0.000020 ENSG00000134202 GSTM3
3.83 1.725 0.000 0.000020 ENSG00000134716 CYP2J2
4.06 1.958 0.000 0.000020 ENSG00000136237 RAPGEF5
7.69 2.759 0.000 0.000020 ENSG00000136842 TMOD1
6.32 0.922 0.000 0.000020 ENSG00000141026 MED9
6.37 1.279 0.000 0.000020 ENSG00000147889 CDKN2A
5.91 1.256 0.000 0.000020 ENSG00000158856 EPB49
7.98 1.855 0.000 0.000020 ENSG00000162873 KLHDC8A
6.97 0.685 0.000 0.000020 ENSG00000167106 FAM102A
6.31 3.628 0.000 0.000020 ENSG00000170689 HOXB9
5.8 1.171 0.000 0.000020 ENSG00000172183 ISG20
4.23 1.273 0.000 0.000020 ENSG00000172345 STARD5
7.02 1.316 0.000 0.000020 ENSG00000176490 DIRAS1
6.55 0.867 0.000 0.000020 ENSG00000177556 ATOX1
5.87 2.507 0.000 0.000020 ENSG00000196361 ELAVL3
5.22 0.873 0.000 0.000020 ENSG00000196476 C20orf96
5.9 1.137 0.000 0.000020 ENSG00000196683 TOMM7
8.22 1.421 0.000 0.000020 ENSG00000205362 MT1A
4.9 1.465 0.000 0.000030 ENSG00000018869 ZNF582
6.66 0.818 0.000 0.000030 ENSG00000101181 GTPBP5
5.69 1.972 0.000 0.000030 ENSG00000113296 THBS4
5.74 1.581 0.000 0.000030 ENSG00000119922 IFIT2
6.83 0.596 0.000 0.000030 ENSG00000130813 C19orf66
6.99 1.327 0.000 0.000030 ENSG00000145216 FIP1L1
5.26 1.11 0.000 0.000030 ENSG00000149927 DOC2A
4.84 1.528 0.000 0.000030 ENSG00000151090 THRB
7.77 1.094 0.000 0.000030 ENSG00000157259 GATAD1
3.99 0.896 0.000 0.000030 ENSG00000164398 ACSL6
4.64 2.353 0.000 0.000030 ENSG00000165443 PHYHIPL
5.03 1.62 0.000 0.000030 ENSG00000170941 AC135352.1
4.99 2.142 0.000 0.000030 ENSG00000182326 C1S
2.77 0.976 0.000 0.000030 ENSG00000188906 LRRK2
7.76 1.085 0.000 0.000030 ENSG00000198874 TYW1
4.54 1.754 0.000 0.000040 ENSG00000073910 FRY
6.26 2.104 0.000 0.000040 ENSG00000104419 NDRG1
7.3 1.434 0.000 0.000040 ENSG00000130066 SAT1
4.96 1.985 0.000 0.000040 ENSG00000132436 FIGNL1
3.52 0.689 0.000 0.000040 ENSG00000143226 FCGR2A
4.62 1.31 0.000 0.000040 ENSG00000143320 CRABP2
5.93 4.882 0.000 0.000040 ENSG00000156298 TSPAN7
5.53 2.923 0.000 0.000040 ENSG00000160862 AZGP1
6.28 2.132 0.000 0.000040 ENSG00000163638 ADAMTS9
4.94 1.336 0.000 0.000040 ENSG00000189164 ZNF527
6.19 1.071 0.000 0.000050 ENSG00000105649 RAB3A
5.66 1.003 0.000 0.000050 ENSG00000136002 ARHGEF4
294

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
6.21 1.475 0.000 0.000050 ENSG00000136243 NUPL2
5.64 0.992 0.000 0.000050 ENSG00000151093 OXSM
4.85 1.082 0.000 0.000050 ENSG00000155629 PIK3AP1
7.07 0.716 0.000 0.000050 ENSG00000157778 PSMG3
8.48 1.368 0.000 0.000050 ENSG00000162545 CAMK2N1
4.72 1.871 0.000 0.000050 ENSG00000164741 DLC1
5.24 1.051 0.000 0.000060 ENSG00000058866 DGKG
5.61 1.415 0.000 0.000060 ENSG00000073464 CLCN4
9.27 0.763 0.000 0.000060 ENSG00000086232 EIF2AK1
3.97 2.144 0.000 0.000060 ENSG00000108176 DNAJC12
7.3 1.402 0.000 0.000060 ENSG00000112852 PCDHB2
4.68 1.736 0.000 0.000060 ENSG00000137393 RNF144B
6.16 1.387 0.000 0.000070 ENSG00000116574 RHOU
8.42 1.11 0.000 0.000070 ENSG00000124226 RNF114
5.72 0.79 0.000 0.000070 ENSG00000129282 MRM1
6 1.143 0.000 0.000070 ENSG00000159228 CBR1
7.25 1.234 0.000 0.000070 ENSG00000166770 AC004696.2
6.42 0.938 0.000 0.000070 ENSG00000172840 PDP2
6.38 0.76 0.000 0.000070 ENSG00000174672 BRSK2
7.22 0.934 0.000 0.000070 ENSG00000179304 FAM156B
6.01 2.099 0.000 0.000070 ENSG00000182217 HIST2H4B
6.01 2.099 0.000 0.000070 ENSG00000183941 HIST2H4A
7.93 2.197 0.000 0.000070 ENSG00000196532 HIST1H3C
6.17 0.703 0.000 0.000070 ENSG00000196741 CXorf24
4.98 1.47 0.000 0.000070 ENSG00000214106 AC093726.6
5.49 0.968 0.000 0.000080 ENSG00000036672 USP2
5.53 1.341 0.000 0.000080 ENSG00000088035 ALG6
5.15 1.343 0.000 0.000080 ENSG00000156500 FAM122C
5.74 1.177 0.000 0.000080 ENSG00000163884 KLF15
6.04 1.258 0.000 0.000080 ENSG00000166402 TUB
7.17 0.939 0.000 0.000080 ENSG00000182646 FAM156A
6.64 4.408 0.000 0.000080 ENSG00000250349 RP5-972B16.2
6.97 0.713 0.000 0.000090 ENSG00000101189 C20orf20
3.92 1.212 0.000 0.000090 ENSG00000119514 GALNT12
3.71 0.824 0.000 0.000090 ENSG00000125804 FAM182A
7.55 0.989 0.000 0.000090 ENSG00000163156 SCNM1
3.42 1.18 0.000 0.000090 ENSG00000181016 C7orf53
6.35 0.891 0.000 0.000090 ENSG00000184205 TSPYL2
4.35 1.373 0.000 0.000090 ENSG00000189283 FHIT
5.51 1.475 0.000 0.000090 ENSG00000198270 TMEM116
7.84 1.309 0.000 0.000100 ENSG00000101224 CDC25B
5.52 0.686 0.000 0.000100 ENSG00000126950 TMEM35
4.4 2.366 0.000 0.000100 ENSG00000148735 PLEKHS1
5.04 1.951 0.000 0.000100 ENSG00000163644 PPM1K
7.83 1.683 0.000 0.000100 ENSG00000172115 CYCS
5.24 1.512 0.000 0.000100 ENSG00000204807 FAM27E2
6.23 1.594 0.000 0.000100 ENSG00000251369 AC003682.1
6.76 1.991 0.000 0.000100 ENSG00000255986 MT1JP
5.41 1.122 0.000 0.000110 ENSG00000131242 RAB11FIP4
7.78 0.733 0.000 0.000110 ENSG00000146676 PURB
6.69 2.507 0.000 0.000110 ENSG00000149294 NCAM1
4.24 0.751 0.000 0.000110 ENSG00000175697 GPR156
4.15 0.854 0.000 0.000110 ENSG00000214290 C11orf93
4.27 1.558 0.000 0.000120 ENSG00000082482 KCNK2
4.89 0.887 0.000 0.000120 ENSG00000100307 CBX7
2.71 0.772 0.000 0.000120 ENSG00000146856 AGBL3
5.69 0.606 0.000 0.000120 ENSG00000169136 ATF5
4.36 0.938 0.000 0.000120 ENSG00000175170 FAM182B
6.37 0.774 0.000 0.000120 ENSG00000255098 RP11-481A20.11
3.52 1.432 0.000 0.000130 ENSG00000003147 ICA1
4.87 1.498 0.000 0.000130 ENSG00000007944 MYLIP
5.81 0.772 0.000 0.000130 ENSG00000101104 PABPC1L
5 1.218 0.000 0.000130 ENSG00000102362 SYTL4
6.13 1.354 0.000 0.000130 ENSG00000158186 MRAS
4.22 0.902 0.000 0.000130 ENSG00000176273 SLC35G1
6.5 1.029 0.000 0.000130 ENSG00000256073 C21orf119
5.73 2.686 0.000 0.000140 ENSG00000117602 RCAN3
7.14 1.232 0.000 0.000140 ENSG00000141753 IGFBP4
5.52 0.559 0.000 0.000140 ENSG00000149599 DUSP15
4.73 1.741 0.000 0.000140 ENSG00000154734 ADAMTS1
3.69 1.648 0.000 0.000150 ENSG00000064225 ST3GAL6
5.21 1.771 0.000 0.000150 ENSG00000106004 HOXA5
6.07 1.193 0.000 0.000150 ENSG00000111364 DDX55
6.02 1.5 0.000 0.000150 ENSG00000120318 ARAP3
5.89 0.804 0.000 0.000150 ENSG00000178809 TRIM73
7.59 0.945 0.000 0.000150 ENSG00000184371 CSF1
295

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
6.4 0.769 0.000 0.000150 ENSG00000184857 TMEM186
5.64 1.271 0.000 0.000150 ENSG00000198814 GK
4.72 1.253 0.000 0.000150 ENSG00000213903 LTB4R
5.17 1.876 0.000 0.000160 ENSG00000008277 ADAM22
3.51 1.09 0.000 0.000160 ENSG00000118777 ABCG2
5.85 1.024 0.000 0.000160 ENSG00000164048 ZNF589
6.16 0.956 0.000 0.000160 ENSG00000175175 PPM1E
6.99 1.096 0.000 0.000160 ENSG00000232838 PET117
6.64 0.794 0.000 0.000170 ENSG00000102934 PLLP
6.78 0.622 0.000 0.000170 ENSG00000141582 CBX4
8.37 0.573 0.000 0.000170 ENSG00000143543 JTB
3.96 1.576 0.000 0.000170 ENSG00000158315 RHBDL2
6.45 1.095 0.000 0.000170 ENSG00000204524 ZNF805
6.43 1.7 0.000 0.000180 ENSG00000006652 IFRD1
5.02 1.218 0.000 0.000180 ENSG00000092068 SLC7A8
4.53 2.247 0.000 0.000180 ENSG00000152527 PLEKHH2
6.11 1.012 0.000 0.000180 ENSG00000164620 RELL2
5.74 1.411 0.000 0.000180 ENSG00000165795 NDRG2
6.38 2.236 0.000 0.000190 ENSG00000105409 ATP1A3
8.02 1.715 0.000 0.000190 ENSG00000106537 TSPAN13
6.58 1.894 0.000 0.000190 ENSG00000125144 MT1G
5.6 1.454 0.000 0.000190 ENSG00000164684 ZNF704
6.47 0.985 0.000 0.000200 ENSG00000065717 TLE2
3.93 1.031 0.000 0.000200 ENSG00000122862 SRGN
4.01 1.439 0.000 0.000200 ENSG00000137203 TFAP2A
6.21 1.126 0.000 0.000200 ENSG00000178878 APOLD1
5.99 0.934 0.000 0.000200 ENSG00000180953 ST20
8.5 1.354 0.000 0.000200 ENSG00000198743 SLC5A3
5.6 0.559 0.000 0.000200 ENSG00000251357 AP000350.10
7.33 0.633 0.000 0.000210 ENSG00000130733 YIPF2
5.89 0.896 0.000 0.000210 ENSG00000142528 ZNF473
5.73 1.626 0.000 0.000210 ENSG00000143162 CREG1
6.19 0.828 0.000 0.000210 ENSG00000146909 NOM1
5.34 0.95 0.000 0.000210 ENSG00000149150 SLC43A1
3.29 0.995 0.000 0.000210 ENSG00000165923 AGBL2
6.6 0.893 0.000 0.000210 ENSG00000175898 S1PR2
4.07 0.944 0.000 0.000210 ENSG00000188732 FAM221A
7.44 1.401 0.000 0.000210 ENSG00000231997 FAM27D1
5.25 0.581 0.000 0.000220 ENSG00000126583 PRKCG
6.21 1.077 0.000 0.000220 ENSG00000173875 ZNF791
3.79 1.662 0.000 0.000230 ENSG00000071073 MGAT4A
6.82 0.743 0.000 0.000230 ENSG00000106638 TBL2
7.47 1.573 0.000 0.000230 ENSG00000189060 H1F0
6.2 0.673 0.000 0.000240 ENSG00000106404 CLDN15
6.5 0.814 0.000 0.000240 ENSG00000106479 ZNF862
5.81 0.871 0.000 0.000240 ENSG00000106608 URGCP
5.85 0.819 0.000 0.000240 ENSG00000108852 MPP2
6.65 2.284 0.000 0.000240 ENSG00000154096 THY1
9.19 1.685 0.000 0.000240 ENSG00000166986 MARS
6.7 2.826 0.000 0.000240 ENSG00000175445 LPL
7.67 0.515 0.000 0.000240 ENSG00000184787 UBE2G2
5.33 1.325 0.000 0.000240 ENSG00000237198 FAM27E1
3.64 1.397 0.000 0.000240 ENSG00000250305 KIAA1456
7.42 0.921 0.000 0.000250 ENSG00000130254 SAFB2
4.71 0.878 0.000 0.000250 ENSG00000152049 KCNE4
4.05 1.314 0.000 0.000250 ENSG00000175906 ARL4D
3.91 0.779 0.000 0.000250 ENSG00000204086 RPA4
5.35 2.281 0.000 0.000250 ENSG00000205364 MT1M
4.05 1.053 0.000 0.000260 ENSG00000076554 TPD52
8.37 1.379 0.000 0.000260 ENSG00000085662 AKR1B1
8.61 0.814 0.000 0.000260 ENSG00000104687 GSR
6.29 1.082 0.000 0.000260 ENSG00000105926 MPP6
9.84 0.823 0.000 0.000260 ENSG00000146701 MDH2
5.59 0.957 0.000 0.000260 ENSG00000168301 KCTD6
5.9 0.865 0.000 0.000260 ENSG00000174939 ASPHD1
5.92 0.645 0.000 0.000260 ENSG00000204536 CCHCR1
6.69 0.769 0.000 0.000270 ENSG00000106785 TRIM14
6.99 0.808 0.000 0.000270 ENSG00000125875 TBC1D20
5.82 1.073 0.000 0.000270 ENSG00000164742 ADCY1
4.02 1.271 0.000 0.000270 ENSG00000196932 TMEM26
4.66 0.86 0.000 0.000270 ENSG00000197933 ZNF823
6 1.557 0.000 0.000270 ENSG00000198053 SIRPA
5.43 0.934 0.000 0.000270 ENSG00000215897 ZBTB8B
4.79 0.711 0.000 0.000280 ENSG00000124613 ZNF391
6.98 1.247 0.000 0.000280 ENSG00000126368 NR1D1
4.95 0.642 0.000 0.000290 ENSG00000125510 OPRL1
296

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
4.21 1.238 0.000 0.000290 ENSG00000166387 PPFIBP2
5.58 0.709 0.000 0.000290 ENSG00000186951 PPARA
7.3 1.284 0.000 0.000290 ENSG00000187372 PCDHB13
3.61 0.734 0.000 0.000290 ENSG00000196167 C11orf92
5.69 1.412 0.000 0.000300 ENSG00000068650 ATP11A
7.51 0.631 0.000 0.000300 ENSG00000106070 GRB10
5.19 0.553 0.000 0.000300 ENSG00000170379 FAM115C
5.03 0.896 0.000 0.000300 ENSG00000257755 orphan
4.33 1.526 0.000 0.000310 ENSG00000104324 CPQ
5.18 0.844 0.000 0.000310 ENSG00000107957 SH3PXD2A
6.49 0.752 0.000 0.000310 ENSG00000114859 CLCN2
7.06 1.16 0.000 0.000310 ENSG00000135457 TFCP2
5.35 1.852 0.000 0.000310 ENSG00000153714 LURAP1L
10.2 0.876 0.000 0.000310 ENSG00000171867 PRNP
3.1 1.481 0.000 0.000320 ENSG00000106560 GIMAP2
7.45 1.354 0.000 0.000320 ENSG00000150893 FREM2
5.65 0.632 0.000 0.000320 ENSG00000163132 MSX1
8 0.759 0.000 0.000320 ENSG00000178741 COX5A
5.45 1.098 0.000 0.000320 ENSG00000211445 GPX3
6.16 0.599 0.000 0.000330 ENSG00000083812 ZNF324
7.47 0.735 0.000 0.000330 ENSG00000145882 PCYOX1L
5.9 0.732 0.000 0.000330 ENSG00000155428 TRIM74
6.82 0.92 0.000 0.000340 ENSG00000106415 GLCCI1
5.56 0.658 0.000 0.000340 ENSG00000116039 ATP6V1B1
4.71 0.686 0.000 0.000340 ENSG00000132801 ZSWIM3
6.32 0.688 0.000 0.000340 ENSG00000134086 VHL
7.86 0.575 0.000 0.000350 ENSG00000113269 RNF130
5.23 1.823 0.000 0.000350 ENSG00000131015 ULBP2
4.27 1.325 0.000 0.000350 ENSG00000197757 HOXC6
7.01 1.393 0.000 0.000350 ENSG00000231360 AL592284.2
3.41 0.946 0.000 0.000360 ENSG00000025423 HSD17B6
5.67 1.773 0.000 0.000360 ENSG00000165118 C9orf64
4.77 0.903 0.000 0.000360 ENSG00000175911 AC127496.1
4.45 1.145 0.000 0.000360 ENSG00000232956 SNHG15
5.73 1.061 0.000 0.000370 ENSG00000135245 HILPDA
6.24 1.297 0.000 0.000370 ENSG00000135472 FAIM2
6.44 0.912 0.000 0.000370 ENSG00000154930 ACSS1
7.65 1.07 0.000 0.000370 ENSG00000168303 MPLKIP
5.48 0.809 0.000 0.000370 ENSG00000205358 MT1H
4.75 1.489 0.000 0.000380 ENSG00000112379 KIAA1244
4.21 1.081 0.000 0.000380 ENSG00000119547 ONECUT2
3.38 0.996 0.000 0.000380 ENSG00000144290 SLC4A10
5.43 0.988 0.000 0.000380 ENSG00000146833 TRIM4
7.01 1.023 0.000 0.000380 ENSG00000173276 ZNF295
8.3 0.683 0.000 0.000380 ENSG00000175193 PARL
4.84 0.881 0.000 0.000390 ENSG00000204856 FAM216A
6.52 1.323 0.000 0.000400 ENSG00000107829 FBXW4
6.84 0.91 0.000 0.000400 ENSG00000124228 DDX27
7.24 0.665 0.000 0.000410 ENSG00000101407 TTI1
6.79 0.747 0.000 0.000410 ENSG00000106245 BUD31
6.3 0.592 0.000 0.000410 ENSG00000142700 DMRTA2
6.5 0.915 0.000 0.000410 ENSG00000189046 ALKBH2
5.51 2.199 0.000 0.000410 ENSG00000215247
5.9 0.698 0.000 0.000420 ENSG00000006194 ZNF263
7.11 1.686 0.000 0.000420 ENSG00000120328 PCDHB12
7.42 0.68 0.000 0.000420 ENSG00000136213 CHST12
5.42 1.536 0.000 0.000420 ENSG00000151364 KCTD14
7.23 0.928 0.000 0.000420 ENSG00000198171 DDRGK1
6.44 0.8 0.000 0.000430 ENSG00000131368 MRPS25
8.34 0.747 0.000 0.000440 ENSG00000111678 C12orf57
6.57 0.838 0.000 0.000440 ENSG00000165661 QSOX2
5.88 1.608 0.000 0.000440 ENSG00000186868 MAPT
5.39 0.894 0.000 0.000440 ENSG00000253293 HOXA10
7.35 0.627 0.000 0.000450 ENSG00000156928 MALSU1
5.28 1.124 0.000 0.000450 ENSG00000172006 ZNF554
4.69 0.646 0.000 0.000450 ENSG00000177335 C8orf31
5.68 1.476 0.000 0.000450 ENSG00000206535 LNP1
6.92 0.94 0.000 0.000460 ENSG00000197608 ZNF841
3.25 1.454 0.000 0.000470 ENSG00000168811 IL12A
4.52 0.897 0.000 0.000470 ENSG00000176593 CTD-2368P22.1
3.63 1.342 0.000 0.000480 ENSG00000113389 NPR3
3.35 1.071 0.000 0.000480 ENSG00000176907 C8orf4
5.37 0.494 0.000 0.000480 ENSG00000177380 PPFIA3
7.44 0.526 0.000 0.000480 ENSG00000229212 RP11-561C5.4
4.52 1.355 0.000 0.000490 ENSG00000120915 EPHX2
4.77 1.002 0.000 0.000490 ENSG00000125388 GRK4
297

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
6.05 0.737 0.001 0.000500 ENSG00000103024 NME3
7.42 0.557 0.001 0.000500 ENSG00000105583 WDR83OS
4.38 0.804 0.001 0.000500 ENSG00000133401 PDZD2
6.69 0.801 0.001 0.000500 ENSG00000196652 ZKSCAN5
7.71 1.671 0.001 0.000510 ENSG00000070669 ASNS
6.84 0.567 0.001 0.000510 ENSG00000096070 BRPF3
5.69 1.106 0.001 0.000510 ENSG00000101187 SLCO4A1
5.48 0.832 0.001 0.000510 ENSG00000120158 RCL1
9.84 0.767 0.001 0.000510 ENSG00000163399 ATP1A1
5.71 1.081 0.001 0.000510 ENSG00000196981 WDR5B
4.19 1.925 0.001 0.000520 ENSG00000136514 RTP4
6.56 2.056 0.001 0.000520 ENSG00000156011 PSD3
4.5 0.937 0.001 0.000520 ENSG00000159882 ZNF230
5.22 0.836 0.001 0.000520 ENSG00000163362 C1orf106
5.09 2.391 0.001 0.000530 ENSG00000106511 MEOX2
5.05 1.189 0.001 0.000530 ENSG00000128604 IRF5
6.86 1.183 0.001 0.000530 ENSG00000163938 GNL3
6.95 0.971 0.001 0.000540 ENSG00000127952 STYXL1
7.12 1.905 0.001 0.000540 ENSG00000164649 CDCA7L
9.33 0.772 0.001 0.000540 ENSG00000187145 MRPS21
7.6 0.708 0.001 0.000550 ENSG00000068305 MEF2A
5.5 0.907 0.001 0.000550 ENSG00000124772 CPNE5
6.17 0.721 0.001 0.000550 ENSG00000129347 KRI1
4.03 0.802 0.001 0.000550 ENSG00000196482 ESRRG
5.88 0.673 0.001 0.000560 ENSG00000123358 NR4A1
4.08 0.7 0.001 0.000560 ENSG00000203815 AL358813.2
4.84 0.817 0.001 0.000570 ENSG00000110031 LPXN
6.18 0.467 0.001 0.000580 ENSG00000104731 KLHDC4
5.1 0.815 0.001 0.000580 ENSG00000144115 THNSL2
5 1.87 0.001 0.000590 ENSG00000104611 SH2D4A
5.53 0.752 0.001 0.000590 ENSG00000117266 CDK18
7.58 1.016 0.001 0.000590 ENSG00000171223 JUNB
6.64 0.746 0.001 0.000600 ENSG00000068400 GRIPAP1
5.05 0.96 0.001 0.000600 ENSG00000184208 C22orf46
8 0.613 0.001 0.000610 ENSG00000130717 UCK1
3.53 1.117 0.001 0.000610 ENSG00000135678 CPM
6.86 1.135 0.001 0.000610 ENSG00000257921 RP11-571M6.15
6.36 0.507 0.001 0.000610 ENSG00000258720
3.41 0.725 0.001 0.000620 ENSG00000140009 ESR2
5.09 1.82 0.001 0.000620 ENSG00000144847 IGSF11
3.89 2.44 0.001 0.000620 ENSG00000157214 STEAP2
7.27 0.911 0.001 0.000620 ENSG00000174243 DDX23
5.36 0.76 0.001 0.000620 ENSG00000213983 AP1G2
4.77 0.842 0.001 0.000630 ENSG00000095380 NANS
4.07 2.206 0.001 0.000630 ENSG00000152078 TMEM56
5.1 0.773 0.001 0.000640 ENSG00000188818 ZDHHC11
6.82 1.812 0.001 0.000650 ENSG00000120324 PCDHB10
5.77 1.023 0.001 0.000650 ENSG00000122877 EGR2
6.72 1.068 0.001 0.000650 ENSG00000203814 HIST2H2BF
4.4 1.535 0.001 0.000660 ENSG00000111846 GCNT2
4.62 1.74 0.001 0.000660 ENSG00000174007 CEP19
5.05 1.321 0.001 0.000660 ENSG00000180818 HOXC10
8.34 0.553 0.001 0.000660 ENSG00000214331 RP11-252A24.2
4.6 1.204 0.001 0.000670 ENSG00000185015 CA13
6.4 0.578 0.001 0.000680 ENSG00000148290 SURF1
6.12 1.141 0.001 0.000680 ENSG00000214357 NEURL1B
5.77 3.086 0.001 0.000690 ENSG00000154529 CNTNAP3B
6.46 0.53 0.001 0.000690 ENSG00000160299 PCNT
7.31 0.729 0.001 0.000700 ENSG00000087250 MT3
6.03 2.89 0.001 0.000700 ENSG00000106714 CNTNAP3
6.23 0.712 0.001 0.000700 ENSG00000181191 PJA1
8.21 0.906 0.001 0.000710 ENSG00000101210 EEF1A2
5.14 0.814 0.001 0.000710 ENSG00000160207 HSF2BP
5.97 0.818 0.001 0.000710 ENSG00000160908 ZNF394
5.69 0.548 0.001 0.000710 ENSG00000165655 ZNF503
5.24 2.121 0.001 0.000710 ENSG00000174607 UGT8
7.63 0.754 0.001 0.000720 ENSG00000104980 TIMM44
4.83 0.918 0.001 0.000720 ENSG00000162757 C1orf74
5.19 0.796 0.001 0.000730 ENSG00000062282 DGAT2
6.04 1.083 0.001 0.000730 ENSG00000064300 NGFR
3.4 1.382 0.001 0.000730 ENSG00000165071 TMEM71
3.53 1.144 0.001 0.000730 ENSG00000227471 AKR1B15
6.81 1.233 0.001 0.000740 ENSG00000111276 CDKN1B
6.24 0.731 0.001 0.000740 ENSG00000172366 FAM195A
6.03 1.004 0.001 0.000740 ENSG00000188283 ZNF383
5.05 0.755 0.001 0.000750 ENSG00000111199 TRPV4
298

E. Exon Array Data Appendix
Average expression across samples Fold-change p-value FDR Ensembl Gene ID Symbol
7.26 1.393 0.001 0.000750 ENSG00000116991 SIPA1L2
8.23 0.568 0.001 0.000780 ENSG00000196367 TRRAP
6.49 1.202 0.001 0.000780 ENSG00000198729 PPP1R14C
4.34 0.843 0.001 0.000790 ENSG00000135127 CCDC64
6.54 1.345 0.001 0.000800 ENSG00000106689 LHX2
4.92 0.897 0.001 0.000800 ENSG00000221994 ZNF630
8.31 0.722 0.001 0.000810 ENSG00000101365 IDH3B
7.15 0.669 0.001 0.000810 ENSG00000114767 RRP9
6.95 0.503 0.001 0.000810 ENSG00000198258 UBL5
7.8 0.771 0.001 0.000820 ENSG00000078043 PIAS2
7.41 0.87 0.001 0.000820 ENSG00000111266 DUSP16
6.56 0.92 0.001 0.000820 ENSG00000112715 VEGFA
6.95 1.222 0.001 0.000830 ENSG00000123080 CDKN2C
5.81 0.756 0.001 0.000830 ENSG00000130158 DOCK6
5 0.858 0.001 0.000840 ENSG00000125352 RNF113A
4.69 2.173 0.001 0.000840 ENSG00000138642 HERC6
4.04 2.745 0.001 0.000860 ENSG00000138646 HERC5
6.46 0.684 0.001 0.000860 ENSG00000197037 ZNF498
8.15 0.797 0.001 0.000870 ENSG00000071462 WBSCR22
5.13 1.572 0.001 0.000870 ENSG00000133321 RARRES3
7.79 0.901 0.001 0.000870 ENSG00000160633 SAFB
7.56 0.741 0.001 0.000870 ENSG00000164933 SLC25A32
7.65 0.65 0.001 0.000880 ENSG00000125844 RRBP1
7.59 0.595 0.001 0.000880 ENSG00000164896 FASTK
4.47 1.076 0.001 0.000880 ENSG00000197857 ZNF44
7.31 1.007 0.001 0.000890 ENSG00000077348 EXOSC5
7.23 0.664 0.001 0.000900 ENSG00000101040 ZMYND8
3.75 1.724 0.001 0.000900 ENSG00000137752 CASP1
4.91 0.534 0.001 0.000900 ENSG00000143578 CREB3L4
6.86 1.846 0.001 0.000900 ENSG00000154856 APCDD1
4.11 0.919 0.001 0.000910 ENSG00000105708 ZNF14
4.74 1.522 0.001 0.000910 ENSG00000116761 CTH
7.33 0.778 0.001 0.000910 ENSG00000196365 LONP1
7.82 1.045 0.001 0.000920 ENSG00000013374 NUB1
6.08 0.799 0.001 0.000920 ENSG00000077713 SLC25A43
3.54 0.864 0.001 0.000930 ENSG00000157224 CLDN12
7.3 0.662 0.001 0.000930 ENSG00000160200 CBS
5.12 3.1 0.001 0.000930 ENSG00000227921 AL353791.1
4.02 1.202 0.001 0.000940 ENSG00000152433 ZNF547
6.71 0.572 0.001 0.000950 ENSG00000105559 PLEKHA4
6.45 1.93 0.001 0.000960 ENSG00000113205 PCDHB3
3.5 0.926 0.001 0.000970 ENSG00000136167 LCP1
2.97 0.557 0.001 0.000970 ENSG00000136839 OR13C9
3.02 0.567 0.001 0.000980 ENSG00000130783 CCDC62
4.96 1.806 0.001 0.000980 ENSG00000132498 ANKRD20A3
7.62 0.717 0.001 0.000980 ENSG00000146834 MEPCE
7.66 2.08 0.001 0.000980 ENSG00000196963 PCDHB16
4.6 1.779 0.001 0.000990 ENSG00000110436 SLC1A2
8.16 0.75 0.001 0.000990 ENSG00000137817 PARP6
7.33 1.172 0.001 0.000990 ENSG00000141576 RNF157
4.53 2.338 0.001 0.001000 ENSG00000018236 CNTN1
5.8 2.56 0.001 0.001000 ENSG00000100095 SEZ6L
7.36 0.859 0.001 0.001000 ENSG00000126947 ARMCX1
3.88 1.307 0.001 0.001000 ENSG00000138741 TRPC3
5.29 0.642 0.001 0.001000 ENSG00000163013 FBXO41
299

Appendix F
MicroRNA Array Data
The data generated with the Agilent microRNA array platform is reported be-
low, with the name of each microRNA that was probed listed in first column -
the microRNAs with an asterisk in their name represent sequences that origi-
nated from the opposite strand in the pri-microRNA secondary structure [71].
The expression value measured is displayed in each GNS and NS cell line, and
a fold change measurement is calculated - with an FDR associated - to rep-
resent the differential expression for each microRNA in GNS versus NS cell
lines. Negative fold change values identify greater average expression across
the NS cell lines with respect to GNS cell lines, whilst positive fold change
values identify greater average expression across GNS cell lines with respect
to NS cell lines. The FDR values are reported in scientific notation.
300

F. MicroRNA Array Data Appendix
Table F.1: Differentially expressed microRNAs in GNS vs NS cell lines at FDR

F. MicroRNA Array Data Appendix
GNS NS
microRNAs G7A G7B G26A G26B G144A G144B G166A G166B CB660A CB660B CB130A CB130B CB152A CB152B CB171A CB171B LogFC p-value FDR
hsa-miR-514b-5p 2.54 2.61 2.74 3.48 1.98 1.98 1.33 1.73 2.5 2.45 1.41 0.81 0.77 1.57 1.22 1.1 0.8 3.91E-04 1.50E-03
hsa-miR-130b* 3.91 3.92 2.5 2.23 2.52 2.8 -2.89 -2.71 -2.58 -2.93 -1.7 0.06 2.14 0.86 0.08 2.66 1.7 3.75E-04 1.45E-03
hsa-miR-187* 1.14 1.44 0.17 1.1 -0.68 0.37 0.72 -0.49 0.08 -1.87 -0.83 -0.97 -0.7 -1.62 -1.08 -1.28 1.5 2.99E-04 1.19E-03
hsa-miR-365 8.58 8.42 9.03 9.18 8.76 8.66 6.75 7.27 7.8 7.45 7.42 7.34 7.8 7.17 8 7.92 0.7 2.85E-04 1.15E-03
hsa-miR-340* 6.71 6.84 6.21 6.37 7.33 7.43 4.33 4.6 5.66 5.54 5.43 4.89 6.15 6.03 5.52 4.92 0.7 2.71E-04 1.10E-03
hsa-miR-4323 2.11 2.23 2.44 2.22 1.64 1.69 0.23 0.18 1.15 0.91 0.28 0.4 0.98 1.38 1.32 1.33 0.6 2.61E-04 1.07E-03
hsa-miR-374c 5.06 5.32 4.4 4.56 3.8 4 -1.71 1.04 3.15 1.86 -1.03 -1.74 2.12 0.16 0.04 2.78 2.4 2.26E-04 9.49E-04
hsa-miR-214 -1.04 0.09 -0.39 -0.36 2.27 2.83 -2.15 -1.93 -1.62 -1.05 -1.96 -0.91 -1.52 -0.71 -0.94 -2.02 1.3 2.24E-04 9.47E-04
hsa-miR-4299 6.17 5.65 7.74 7.81 6.77 6.66 6.79 6.93 6.83 6.81 5.95 5.56 5.96 5.84 6.06 6.14 0.7 2.22E-04 9.42E-04
hsa-miR-3176 0.34 0.36 0.87 0.83 0.1 -0.73 -1.08 -1.83 -0.36 -0.72 -1.29 -1.13 -0.9 -1.25 -1.06 -1.27 0.9 2.09E-04 9.01E-04
hsa-miR-129* 3.47 3.71 3.6 3.42 4.69 4.23 3.06 3.14 2.99 2.93 2.56 2.75 3.4 3.24 3.17 2.95 0.7 2.09E-04 9.01E-04
hsa-miR-1290 3.42 3.61 2.68 3.42 3.9 3.81 2.38 3.01 3.18 3.13 2.31 2.25 2.43 2.23 2.32 1.98 0.8 2.04E-04 8.87E-04
hsa-miR-185 6.96 7.11 7.17 7.07 7.77 7.55 6.38 6.66 6.15 5.53 6.3 6.36 6.91 6.52 6.59 6.44 0.7 1.81E-04 7.99E-04
hsa-miR-671-5p 3.05 3.16 2.93 3.26 2.9 3.24 3.2 2.62 2.51 2.41 0.83 2.12 2.24 1.32 2.18 1.98 1.1 1.68E-04 7.51E-04
hsa-miR-3607-5p 3.6 3.93 2.58 3.88 2.29 3.57 1.86 3 2.55 2.79 1.44 2.11 -0.08 1.35 1.2 1.06 1.5 1.57E-04 7.07E-04
hsa-miR-545 2.15 2.84 1.45 2.48 4.14 4.37 0.92 1.48 -1.76 -3.67 0.37 1.57 3.6 2.97 1.87 2.26 1.6 1.43E-04 6.56E-04
hsa-miR-590-5p 9.23 9.6 8.57 8.77 8.83 8.57 7.55 8.01 6.99 5.48 8.03 7.8 8.21 7.89 8.13 7.87 1.1 1.43E-04 6.56E-04
hsa-miR-1973 4.94 5 3.38 3.62 2.97 3.63 2.15 3.3 2.61 1.92 -1.6 1.02 1.96 -0.3 1.43 3.28 2.3 1.31E-04 6.14E-04
hsa-miR-135b 11.25 11.09 8.22 8.68 7.12 6.78 10.79 11.45 7.9 7.18 9.73 8.29 6.56 5.95 9.92 8.53 1.4 1.32E-04 6.14E-04
hsa-miR-149* 0.51 0.77 0.41 0.2 -0.8 -0.51 -1.02 -1.06 -1.07 -1.1 -1.22 -1.54 -1.21 -0.83 -1.93 -0.6 1 1.32E-04 6.14E-04
hsa-miR-20b* -0.99 -1.17 -1.15 -0.77 -1.02 -1.3 -1.97 -2.21 -2.03 -2.19 -2.19 -1.81 -2.14 -2.06 -1.86 -2.25 0.7 1.31E-04 6.14E-04
hsa-miR-3162 8.59 9.17 8.29 9.92 8.68 8.85 9.9 9.76 8.57 8.32 7.31 7.26 8.34 8.28 7.45 8.13 1.2 1.26E-04 6.00E-04
hsa-miR-935 1.9 1.58 -0.41 -0.58 0.41 -0.67 -1.31 -0.43 -0.44 0.85 -1.77 -1.11 -1.78 -2.33 -2.34 -1.53 1.4 1.20E-04 5.74E-04
hsa-miR-3200-5p 0.94 0.05 -0.24 -0.25 -0.45 -0.25 -1.06 -0.71 -1.57 -1.74 -0.74 -0.99 -0.74 -0.77 -1.09 -1.01 0.8 1.11E-04 5.42E-04
hsa-miR-3065-3p 2.39 2.87 3.33 3.4 3.79 3.87 -0.19 0.64 1.39 1.69 1.46 -0.23 1.94 2.28 0.79 0.69 1.3 1.06E-04 5.21E-04
hsa-miR-3621 0.96 0.43 -0.05 0.36 -0.18 0.31 -0.96 -1.11 -0.17 -0.82 -1.15 -1.1 -1.35 -0.5 -1.47 -1.47 1 1.04E-04 5.10E-04
hsa-miR-3937 3.09 3.31 3.12 3.56 2.58 2.94 3.03 3.19 2.71 1.95 1.74 1.33 0.88 2.58 1.57 1.81 1.3 9.89E-05 4.90E-04
hsa-miR-34b 3.17 2.7 4.35 4.14 -0.19 -0.4 4.32 4.6 2.73 2.11 -0.76 0.49 3.5 3.43 0.2 1.29 1.2 9.87E-05 4.90E-04
hsa-miR-3188 3.36 3.59 3.17 4.1 2.97 3.21 2.98 3.16 3.13 3.15 1.64 1.26 3.07 3.01 1.76 2.14 0.9 6.69E-05 3.47E-04
hsa-miR-760 2.29 2.4 1.89 2.48 2.11 2.25 1.66 1.4 0.22 0.2 1.93 1.27 1.08 0.7 1.99 1.27 1 6.02E-05 3.16E-04
hsa-miR-423-3p 3.85 3.45 2.93 2.74 2.6 2.22 1.71 2.04 1.28 0.64 1.71 0.82 2.27 1.58 2.39 1.97 1.1 5.54E-05 2.93E-04
hsa-miR-3679-5p 5.72 5.81 5.62 6.16 5.16 5.83 5.01 5.18 6.06 5.87 3.78 3.89 4.28 4.18 4.08 4.76 1 5.21E-05 2.77E-04
hsa-miR-4322 3.44 3.36 2.44 2.81 1.97 2.07 1.69 1.89 2.49 2.23 1.17 1.28 0.8 1.53 1.23 0.06 1.1 4.72E-05 2.54E-04
hsa-miR-26b 10.61 10.75 9.79 10.17 9.35 9.39 8.08 8.74 9.27 8.81 8.44 7.89 8.83 8.86 8.77 8.42 1 4.44E-05 2.41E-04
hsa-miR-769-5p 7.58 7.53 7.21 7.09 6.89 6.74 5.68 5.75 6.1 6.03 6.02 5.56 6.33 6.28 6.09 5.59 0.8 4.33E-05 2.36E-04
hsa-miR-340 8.66 9.06 7.59 7.91 8.32 8.26 5.66 5.97 6.74 5.9 6.7 6.17 7.36 7.27 6.78 6.59 1 4.24E-05 2.34E-04
hsa-miR-125a-3p 5.18 5.19 4.76 4.88 4.33 4.19 4.16 4.06 4.44 3.98 3.52 3.51 4.04 3.86 3.71 3.61 0.8 4.25E-05 2.34E-04
hsa-miR-208b 3.99 4.4 1.91 2.17 4.5 4.53 1.47 1.46 1.23 1.65 2.66 2.36 2.76 2.54 2.37 2.32 0.8 4.01E-05 2.22E-04
hsa-miR-128 8.13 8.18 7.63 7.59 6.6 6.88 5.28 5.67 6.19 5.84 5.67 6.1 6.56 6.22 6.1 6.35 0.9 3.68E-05 2.06E-04
hsa-miR-4304 1.79 2.27 1.8 1.95 1.17 1.24 0.32 0.41 1.4 1.37 0.08 -0.35 -0.02 0.2 0.17 0.81 0.9 3.67E-05 2.06E-04
hsa-miR-29b 12.87 12.94 13.11 13.16 13.07 13.08 13.02 13.56 10.64 8.35 11.83 12.41 11.41 11.23 12.47 13 1.7 3.59E-05 2.04E-04
hsa-miR-95 7.72 8.24 1.52 2.55 4.85 5.34 -2.35 -2.21 4.95 4.86 -0.66 -1.64 4.17 3.78 0.06 -0.82 1.4 2.89E-05 1.69E-04
hsa-miR-124* 3.06 3.15 -0.19 -0.76 2.86 2.95 0.61 0.16 -1.08 -0.02 -1.12 -0.05 -0.32 -0.14 1.1 -0.9 1.8 2.80E-05 1.65E-04
hsa-miR-101 8.43 8.68 6.61 6.9 7.96 8.21 6.77 7.29 6.6 4.83 6.18 5.9 6.75 6.51 6.31 6.35 1.4 2.48E-05 1.49E-04
hsa-miR-570 1.5 1.49 2.86 2.31 2.67 2.39 1.92 2.3 0.92 -0.62 0.44 0.24 1.42 1.16 1.17 1.75 1.4 2.48E-05 1.49E-04
hsa-miR-3196 6.87 7.28 6.59 7.25 6.68 7.05 6.58 6.66 6.54 6.45 5.48 5.42 5.89 5.78 5.76 6.08 0.9 2.27E-05 1.40E-04
hsa-miR-4324 5.64 5.7 4.11 4.73 4.92 5.24 -3.79 -3.54 4.33 3.5 -1.02 -3 2.44 1.73 0.63 0.69 1.7 2.03E-05 1.27E-04
hsa-miR-484 7.33 7.51 7.34 7.01 6.8 6.76 6.04 6 6.17 6.11 5.88 5.77 6.22 6.15 5.96 6.17 0.8 2.00E-05 1.26E-04
hsa-miR-629* 4.88 4.91 4.07 4.08 3.87 4.03 4.31 4.1 4.12 3.91 3.36 3.21 3.52 3.48 3.25 2.88 0.8 2.00E-05 1.26E-04
hsa-miR-371-5p 5.66 5.85 5.93 6.68 5.06 4.99 5.64 5.9 4.75 4.57 4.7 3.9 4.74 4.83 4.8 4 1.2 1.97E-05 1.25E-04
hsa-miR-137 4.51 4.44 6.22 6.54 6.65 6.75 4.03 4.46 7.1 6.31 3.72 3.56 3.75 3.94 3.71 3.3 1 1.91E-05 1.22E-04
hsa-miR-4271 6.58 6.84 6.29 6.94 5.5 5.27 5.83 5.95 5.6 5.34 4.84 3.64 4.65 5.29 4.92 4.63 1.3 1.89E-05 1.22E-04
302

F. MicroRNA Array Data Appendix
GNS NS
microRNAs G7A G7B G26A G26B G144A G144B G166A G166B CB660A CB660B CB130A CB130B CB152A CB152B CB171A CB171B LogFC p-value FDR
hsa-miR-378* 4.12 3.99 0.27 2.1 3.84 3.32 -0.89 -0.46 -0.84 0.35 -1.52 0.85 -0.35 -0.52 -0.93 0.12 2.4 1.86E-05 1.21E-04
hsa-miR-450b-5p 3.48 3.73 2.31 2.74 3.18 3.66 2.32 1.99 2.26 1.63 1.67 2.03 2.66 2.41 1.08 1.3 1 1.73E-05 1.13E-04
hsa-miR-30e 9 9.36 9.93 9.9 8.07 8.21 7.51 8.09 8.44 7.28 7.03 6.87 8.07 8.07 7.49 7.41 1.2 1.61E-05 1.06E-04
hsa-miR-3665 8.04 8.35 8.04 8.83 7.64 7.58 8.09 8.32 7.6 7.72 6.54 5.59 6.66 7.18 6.85 5.91 1.4 1.57E-05 1.04E-04
hsa-miR-129-3p 4.48 4.46 3.52 3.2 4.47 4.63 2.27 2.44 2.93 2.99 2.59 2.7 2.77 2.19 2.9 2.16 1 1.44E-05 9.63E-05
hsa-miR-23a 12.81 12.59 12.67 12.85 11.48 11.41 10.52 11.15 11.61 11.2 10.54 10.26 11.44 11.2 10.78 10.67 1 1.40E-05 9.44E-05
hsa-miR-320b 9.63 9.55 9.06 8.96 8.92 8.72 8.56 8.86 8.22 7.98 8.12 7.78 8.4 8.44 8.25 8.22 0.9 1.39E-05 9.40E-05
hsa-miR-4317 6.09 5.79 5.62 5.85 4.06 4.09 0.37 2.73 4.18 3 -0.34 -1.82 3.32 2.32 0.9 2.49 2.6 1.25E-05 8.61E-05
hsa-miR-629 1.44 2.01 0.67 1.35 0.16 1.05 -0.81 0.29 0.67 0.68 -1.01 -1.9 -1.54 -1.47 -0.77 -1.56 1.6 1.03E-05 7.24E-05
hsa-miR-148a* 0.91 0.65 0.76 0.78 -1.05 -1.08 -1.04 -0.51 -1.2 -1.59 -2.04 -0.8 -1.7 -1.71 -1.61 -0.86 1.4 1.04E-05 7.24E-05
hsa-miR-140-5p 10.97 10.71 9.25 9.58 12.19 12.22 8.34 8.5 9.39 8.74 9.18 8.81 9.47 9.48 9.36 8.85 1.1 9.76E-06 6.96E-05
hsa-miR-146b-5p 7.41 7.49 4.98 5.54 5.51 5.33 4.87 4.97 4.96 4.76 4.87 4.87 4.75 4.99 4.82 4.58 0.9 9.75E-06 6.96E-05
hsa-miR-711 1.63 1.83 1.27 1.41 0.34 1.27 1 0.47 -0.96 -0.43 -0.59 0.19 -0.23 0.26 -0.28 0.22 1.4 9.61E-06 6.95E-05
hsa-miR-660 7.54 7.79 7.55 7.76 8.26 7.84 6.92 7.3 7.06 6.71 6.54 6.43 6.8 6.78 6.51 6.4 1 8.62E-06 6.31E-05
hsa-miR-1207-5p 7.02 7.09 7.22 7.8 7.18 7.59 7.01 7.05 7.41 7.1 5.67 5.15 5.93 6.11 6.06 6.1 1.1 8.14E-06 5.99E-05
hsa-miR-222 6.83 6.41 7.62 7.68 3.66 3.63 6.95 7.09 5.92 5.76 3.8 4.21 4.92 4.73 4.63 5.82 1.3 8.05E-06 5.96E-05
hsa-miR-598 8 8 7.43 7.39 7.68 7.5 6.84 7.01 6.63 6.33 6.69 6.55 6.66 6.52 6.82 6.69 0.9 8.00E-06 5.95E-05
hsa-miR-181a 11.59 11.76 9.88 9.97 11.25 11.66 7.97 8.27 10.01 9.73 8.93 8.21 9.32 9.43 9.08 8.42 1.2 7.90E-06 5.91E-05
hsa-miR-105 1.46 1.97 -0.62 -0.72 2.53 2.71 0.13 -0.56 -0.43 -0.95 -0.2 0.17 -0.44 -0.88 -0.01 -0.84 1.3 7.68E-06 5.81E-05
hsa-miR-449a 3.88 3.85 3.6 3.79 2.56 2.38 1.74 3.52 2.17 1.65 1.24 0.91 0.57 -0.1 1.75 -0.19 2.2 7.41E-06 5.65E-05
hsa-miR-139-3p 2.16 2.02 1.21 1.61 3.89 3.47 -1.75 -0.13 -0.59 -0.68 -0.01 -1.51 -1.15 0.1 -0.31 0.11 2.1 7.43E-06 5.65E-05
hsa-miR-1225-5p 6.66 6.74 6.51 7.16 6.54 6.89 7 7.24 6.43 6.13 5.21 4.39 5.78 6.13 5.55 4.98 1.3 7.35E-06 5.65E-05
hsa-miR-490-5p 0.87 0.79 -0.74 0.86 0.19 -0.25 -1.61 -1.55 -1.29 -1.67 -1.87 -1.55 -1.41 -1.83 -2.25 -1.97 1.6 6.47E-06 5.04E-05
hsa-miR-3663-3p 6.36 6.41 6.14 6.45 5.8 5.79 5.6 5.91 5.37 5.01 4.09 3.42 5.5 5.59 4.65 3.3 1.4 6.11E-06 4.78E-05
hsa-miR-200a 4.01 4.31 3.1 3.61 1.92 0.89 0.47 0.35 0.91 0.9 0.97 0.98 1.26 -0.34 -0.37 0.31 1.8 5.99E-06 4.72E-05
hsa-miR-15b* 4.2 4.12 3.42 3.24 2.24 2.3 2.1 2.34 -0.58 0.13 0.97 1.38 1.06 -0.61 0.61 2.79 2.3 5.80E-06 4.59E-05
hsa-miR-195 11.3 11.31 8.79 9.23 8.48 8.53 7.06 7.67 7.76 6.7 7.62 6.93 8.43 8.19 7.82 7.09 1.5 5.71E-06 4.55E-05
hsa-miR-16 13.44 13.4 12.5 12.61 11.98 11.81 11.64 12.16 11.44 11 11.41 11.21 11.68 11.4 11.67 11.51 1 5.70E-06 4.55E-05
hsa-miR-15b 14.51 14.17 13.91 13.8 12.97 12.7 13.2 13.72 12.27 12.01 12.75 12.46 12.82 12.52 12.91 12.52 1.1 5.20E-06 4.27E-05
hsa-miR-29a* 2.3 2.61 2.6 2.96 3.57 3.69 2.13 2.61 1.55 -0.4 -0.63 0.57 1.11 1 1.5 1.79 2 4.73E-06 3.95E-05
hsa-miR-3652 2.86 3.37 2.52 3.89 2.72 2.61 3.38 3.42 2.19 1.93 0.68 -0.49 2.13 2.23 1.35 1.39 1.7 4.28E-06 3.72E-05
hsa-miR-1226* 4.44 4.62 4.5 4.7 3.6 3.23 3.38 3.58 3.38 3.03 3 2.57 2.7 2.87 2.41 3 1.1 3.94E-06 3.49E-05
hsa-miR-4306 7.84 7.92 7.91 7.87 8.3 8.21 6.92 7.2 6.9 6.34 6.51 6.56 7.11 6.82 6.77 6.95 1 3.90E-06 3.48E-05
hsa-miR-193b* 1.98 1.82 2.36 2.7 2.56 2.42 -0.83 -0.33 0.49 -0.14 -1.62 0.09 -0.56 -1.35 -0.21 0.6 1.9 3.82E-06 3.45E-05
hsa-let-7f 15.52 15.6 15.18 15.47 14.52 14.37 14.19 14.86 14.18 13.51 13.43 13.35 14.1 13.87 13.85 13.99 1.2 3.64E-06 3.33E-05
hsa-miR-3174 4.63 4.83 4.02 3.74 2.41 1.77 0.74 1.71 1.76 1.53 0.86 1.65 1.19 1.67 1.46 1.8 1.5 3.60E-06 3.32E-05
hsa-miR-642b 6.32 6.37 6.02 6.89 6.03 6.16 6.55 6.56 5.25 4.85 4.7 4.22 5.18 5.72 4.94 3.86 1.5 3.53E-06 3.31E-05
hsa-miR-451 4.5 4.32 2.86 2.74 3.32 3.32 2.81 2.63 2.55 2.46 2.39 2.15 2.32 2.46 1.79 2.28 1 3.34E-06 3.15E-05
hsa-miR-7 6.39 6.54 3.65 4.32 6.93 7.22 4.68 4.49 3.24 3.51 3.18 4.38 3.55 3.01 3.35 4.85 1.9 3.20E-06 3.03E-05
hsa-miR-146a 4.58 4.84 2.46 2.99 6.05 5.67 1.66 1.47 2.06 2.47 2.51 2.52 2.05 2.04 3.03 1.94 1.4 3.15E-06 3.03E-05
hsa-let-7a 14.87 14.84 14.45 14.5 13.97 13.82 13.33 14.02 13.03 12.59 12.99 12.68 13.41 13.24 13.38 12.86 1.2 3.17E-06 3.03E-05
hsa-miR-3610 3.31 3.33 3.33 4.91 2.71 3.31 3.68 3.61 3.08 2.2 0.48 0.53 1.4 2.48 1.35 1.22 1.9 3.03E-06 2.94E-05
hsa-miR-4281 8.13 8.21 7.77 8.42 8.14 8.31 8.42 8.49 7.72 7.49 6.41 6.25 7.37 7.64 6.72 7.18 1.1 2.92E-06 2.85E-05
hsa-miR-3651 6.93 6.57 5.82 5.87 5.95 5.51 6.65 7.31 5.59 4.77 4.89 4.52 5.2 5.04 5.04 4.77 1.3 2.88E-06 2.83E-05
hsa-miR-3141 5.22 5.53 5.11 5.48 4.37 4.46 4.41 4.62 4.23 4.11 3.37 3.44 3.53 3.48 3.3 4.15 1.2 2.76E-06 2.73E-05
hsa-miR-29a 12.2 12.39 12.19 12.19 11.55 11.41 11.25 11.85 10.56 9.41 10.62 10.7 10.38 10.24 11.11 10.96 1.4 2.65E-06 2.64E-05
hsa-miR-193b 6.4 6.32 7.28 7.15 6.82 6.6 4.68 5.07 5.12 4.65 5.11 5.12 4.91 4.11 5.82 5.53 1.2 2.45E-06 2.46E-05
hsa-let-7g 12.74 12.74 12.26 12.47 11.31 11.13 10.74 11.34 10.71 10.23 10.52 10.16 11.05 10.94 10.88 10.29 1.2 2.38E-06 2.41E-05
hsa-miR-425 9.31 9 9.46 9.46 8.7 8.54 9.16 9.76 7.76 7.61 8.1 7.92 8.39 8.3 8.33 8.31 1.1 2.28E-06 2.32E-05
hsa-miR-3934 2.46 2.51 2.25 1.83 0.98 1.39 0.4 1.41 0.94 0.51 -0.87 -0.59 0.03 -0.05 0.4 1.06 1.5 2.20E-06 2.27E-05
hsa-miR-503 5.77 5.65 4.71 5.13 4.72 5.49 3.27 3.87 4.78 2.96 1.69 0.98 5.46 4.69 0.18 0.64 2.2 2.05E-06 2.14E-05
hsa-miR-4270 4.8 4.88 4.8 5.28 4.32 4.93 4.27 4.35 4.36 4.11 3.07 2.61 3.82 3.95 2.98 3.05 1.2 1.88E-06 1.98E-05
303

F. MicroRNA Array Data Appendix
GNS NS
microRNAs G7A G7B G26A G26B G144A G144B G166A G166B CB660A CB660B CB130A CB130B CB152A CB152B CB171A CB171B LogFC p-value FDR
hsa-miR-762 5.94 6.4 5.5 6.29 5.68 5.74 5.03 5 5.45 5.68 3.69 2.78 4.47 4.76 3.59 3.62 1.4 1.79E-06 1.90E-05
hsa-miR-106b 13.58 13.74 12.99 12.97 12.23 12.14 10.91 11.37 10.44 9.66 11.5 11.39 11.85 11.57 11.84 11.7 1.2 1.51E-06 1.61E-05
hsa-miR-93 12.56 12.52 12.49 12.25 11.35 11.13 10.08 10.31 9.69 9.46 10.7 10.63 10.81 10.43 10.8 11.19 1.1 1.48E-06 1.59E-05
hsa-miR-1285 3.13 3.55 2.55 2.9 1.58 1.7 -0.07 0.87 -1.02 -1.13 -1.59 0.15 0.09 -0.66 -1.37 0.56 2.6 1.31E-06 1.44E-05
hsa-miR-630 7.37 7.13 7.81 7.71 6.67 8.1 4.32 4.61 7.43 7.17 3.79 3.73 5.18 4.9 3.9 4.38 1.7 1.30E-06 1.44E-05
hsa-miR-26a 11.67 11.8 9.99 10.19 12.63 12.51 8.54 9.09 9.77 9.52 9.48 8.59 9.23 9.48 9.69 9.14 1.4 1.27E-06 1.43E-05
hsa-miR-29c 10.44 10.67 10.87 10.79 9.45 9.3 10 10.6 8.67 7.33 8.65 8.65 8.85 8.65 9.13 9.17 1.6 1.04E-06 1.18E-05
hsa-miR-452 6.57 6.02 -0.35 0.35 0.28 1.01 0.14 -0.12 -0.7 -0.55 0.92 0.77 -0.59 0.3 -0.62 0.17 1.8 9.05E-07 1.05E-05
hsa-miR-188-5p 4.27 4.18 4.62 4.98 4.06 5.02 3.98 4.17 3.55 3.16 2.58 2.03 3.31 3.26 2.57 2.86 1.5 9.02E-07 1.05E-05
hsa-miR-483-5p 4.39 4.58 4.16 4.36 4.23 4.97 2.39 2.68 3.86 3.67 2.09 1.4 1.53 1.61 1.81 2.75 1.6 7.87E-07 9.49E-06
hsa-miR-1915* 1.48 1.38 0.81 0.77 0.11 0.1 0.13 0.61 -0.07 -0.07 -1.02 0.05 -1.03 -1.16 -1.43 -1.8 1.5 7.89E-07 9.49E-06
hsa-miR-589* 0 -0.11 -0.26 0.13 -1.23 -1.25 -0.61 -0.39 -2.07 -1.86 -1.75 -2.53 -1.32 -1.88 -2.12 -1.7 1.4 7.64E-07 9.35E-06
hsa-miR-664 6.32 6.25 5.29 5.7 5.05 4.98 3.68 3.76 4.76 4.64 3.65 3.45 3.98 4.14 3.8 3.6 1.1 7.06E-07 8.72E-06
hsa-miR-224 8.65 7.99 -0.01 -0.12 2.47 2.55 1.1 0.3 0.65 0.62 2.73 1.99 1.29 1.37 0.68 1.05 1.6 6.73E-07 8.47E-06
hsa-miR-550a* 6.74 6.7 5.88 5.58 4.63 4.43 5.17 5.4 3.48 3.74 4.4 4.88 4.28 4.04 4.64 4.97 1.3 6.09E-07 7.81E-06
hsa-miR-345 2.07 2.14 2.47 2.11 2.27 1.98 -0.24 0.64 -0.08 -0.89 -1.35 -1.19 0.16 -0.87 -0.42 0.85 2.2 5.78E-07 7.55E-06
hsa-miR-30b 9.44 9.35 10.26 10.06 8.15 7.81 8.54 9.23 7.63 7.22 7.53 6.77 8.09 7.83 7.8 7.21 1.6 5.83E-07 7.55E-06
hsa-miR-584 4.79 4.62 2.79 2.95 4.2 4.17 3.25 3.31 3.72 3.59 2.26 2.47 2.51 2.58 2.11 1.69 1.1 5.77E-07 7.55E-06
hsa-miR-4318 3.83 3.59 3.01 2.77 0.81 1.43 0.29 0.45 0.85 0.59 0.08 1.01 0.05 0.55 0.25 -0.01 1.6 5.25E-07 7.14E-06
hsa-miR-30d 10.35 10.36 11.76 11.52 9.9 9.71 9.85 10.28 9.65 9.67 8.76 8.04 9.35 9.52 9.01 8.57 1.4 5.13E-07 7.05E-06
hsa-miR-3607-3p 3.84 4.19 2.65 2.8 2.99 3.24 1.81 2.52 2.57 2.37 1.09 0.24 0.76 1.21 1.35 0.59 1.7 4.78E-07 6.64E-06
hsa-miR-3615 1.41 1.66 0.49 2.03 -1 -0.69 -1.13 -0.99 -0.88 -1.04 -2.26 -2.16 -1.5 -1.11 -2.12 -2.15 1.9 4.03E-07 5.78E-06
hsa-miR-30b* 3.86 4.01 4.77 4.68 3.11 2.8 4.27 4.48 1.68 1.65 2.31 2.07 3.26 2.19 2.91 2.03 1.7 4.12E-07 5.78E-06
hsa-miR-20b 11.79 11.98 12.66 12.56 11.47 11 9.71 10.31 9.53 8.79 10.12 10.32 9.38 9.01 10.32 10.94 1.6 4.06E-07 5.78E-06
hsa-miR-3194 2.54 3.02 2.44 3.35 2.56 2.62 2.27 2.32 1.4 1.44 0.36 0.19 1.39 1.95 0.56 1.04 1.6 4.12E-07 5.78E-06
hsa-miR-4298 4.97 5.17 4.64 5.02 3.79 4.53 3.44 3.42 3.58 3.61 2.05 1.99 2.48 2.56 2.37 3.35 1.6 4.10E-07 5.78E-06
hsa-miR-150* 5.95 6.28 5.47 6.35 5.13 5.06 5.64 5.59 5.05 5 3.62 3.41 4 4.31 4.05 3.43 1.6 3.34E-07 5.00E-06
hsa-miR-124 10.04 10.47 0.4 0.44 10.54 10.53 -0.55 -2.15 -0.75 -0.49 3 0.55 3.87 4.27 -1.24 0.93 3.7 3.25E-07 4.92E-06
hsa-miR-1915 8.53 8.9 7.81 8.78 8.27 8.24 9.02 8.95 7.8 7.7 6.14 5.3 6.86 7.44 6.35 5.59 1.9 3.21E-07 4.92E-06
hsa-miR-30d* 2.17 2.45 3.97 3.75 1.08 -0.17 2.72 3.96 -1.92 -3.21 -1.89 -2.27 1.98 -0.18 0.55 -0.8 3.5 2.97E-07 4.60E-06
hsa-miR-1273e 2.2 2.46 1.67 2.65 0.12 0.58 1.9 1.91 -1.25 -0.76 -0.47 -1 -0.08 -0.63 -0.9 0.54 2.3 2.89E-07 4.53E-06
hsa-miR-362-3p 6.21 6.42 5.81 5.88 6.59 6.13 3.96 4.66 4.6 2.9 2.81 1.56 4.23 3.44 2.04 2.72 2.7 2.75E-07 4.40E-06
hsa-miR-454 7.64 7.96 6.42 7.02 7.19 7.37 4.42 5.09 5.78 5.08 4.02 3.17 5.47 5.74 4.58 4.35 1.9 2.30E-07 3.73E-06
hsa-miR-877 1.92 2.27 1.23 2.56 0.53 1.25 1.31 1.16 -0.47 -0.81 -0.68 -1.04 -0.64 0.01 -0.78 -0.33 2.1 2.22E-07 3.65E-06
hsa-miR-500a* 5.75 5.85 5.83 5.87 6.44 5.94 4.86 5.05 5.08 5.16 3.73 3.81 4.6 4.62 4.07 4.32 1.3 2.22E-07 3.65E-06
hsa-miR-320c 8.03 7.86 7.77 7.7 7.12 7.2 6.9 7.17 6.81 6.46 5.75 5.26 6.16 6.24 5.92 6.25 1.4 2.11E-07 3.60E-06
hsa-miR-199a-3p 8.68 8.85 2.18 2.55 8.3 7.86 -0.76 -0.23 3.58 3.37 2.11 1.17 0.26 0.91 2.71 0.5 2.9 2.04E-07 3.52E-06
hsa-miR-423-5p 6.41 6.33 5.65 5.44 4.11 4.01 3.62 4.26 3.69 3.28 2.63 1.54 3.78 3.59 3.59 3.5 1.8 1.88E-07 3.29E-06
hsa-let-7b 14.64 14.61 14.02 13.92 13.3 13.16 12.82 13.47 11.75 11.16 12.1 11.67 12.29 12.13 12.49 11.41 1.9 1.84E-07 3.27E-06
hsa-miR-488* 2.77 3 3.28 3.89 3.88 4.41 -1.71 -1.63 -1.74 -1.57 1.4 0.02 -1.87 -1.09 1.77 0.39 2.6 1.69E-07 3.03E-06
hsa-miR-3065-5p 3.53 3.62 4.44 4.66 4.88 4.88 0.87 2 2.21 1.68 0.63 0.67 3.15 3.29 1.5 0.82 1.9 1.69E-07 3.03E-06
hsa-miR-32 4.12 3.99 0.34 1.75 4.14 4.43 -0.52 2.51 -3.73 -4.55 -0.98 -2.44 -1.46 -2.5 1.72 1.61 4.1 1.60E-07 2.96E-06
hsa-miR-320d 10.53 10.41 10.04 9.87 9.59 9.44 9.25 9.58 8.89 8.55 8.3 7.81 8.55 8.7 8.44 8.33 1.4 1.48E-07 2.80E-06
hsa-miR-4327 5.09 5.34 4.84 5.45 4.16 4.07 4.5 4.57 4.13 4 2.18 1.28 2.58 3.18 2.97 2.2 1.9 1.30E-07 2.53E-06
hsa-miR-361-3p 6.09 5.96 6.01 6.41 4.37 4.46 3.56 3.78 4.17 3.96 3.11 2.32 3.88 3.71 3.59 3.33 1.6 1.29E-07 2.53E-06
hsa-miR-664* 3.63 3.52 2.54 3.26 2.81 2.82 -0.19 0.11 2.11 2.34 -1.16 -1.32 0.55 0.33 0.37 -0.94 2 1.18E-07 2.38E-06
hsa-miR-492 3.09 3.1 1.97 2.97 2.41 2.41 2.91 2.83 1.16 0.44 0.52 -0.52 1.01 1.17 0.66 0.25 2.1 1.11E-07 2.26E-06
hsa-miR-424 12.87 12.79 11.31 11.95 12.07 12.87 9.77 10.36 11.39 9.36 7.33 6.82 11.52 11.17 6.66 6.88 2.9 1.01E-07 2.10E-06
hsa-miR-200b 5.36 5.23 4.31 4.7 3.76 2.92 1.85 2.79 1.95 2.39 1.72 0.64 1.96 2 0.63 1.19 2.3 1.01E-07 2.10E-06
hsa-miR-1181 7.14 7.72 6.28 6.94 5.95 5.79 6.61 6.53 5.36 5.38 4.6 4.5 4.75 5.29 4.81 4.28 1.7 9.04E-08 1.98E-06
hsa-miR-320e 9.82 9.74 9.31 9.2 8.9 8.77 8.46 8.79 8.17 7.83 7.46 6.98 7.7 7.89 7.61 7.75 1.5 9.16E-08 1.98E-06
hsa-miR-542-3p 8.8 9.12 6.99 8.06 8.24 8.95 6.1 6.4 7.57 6.34 3.77 2.8 8.08 7.74 2.97 2.63 2.6 8.39E-08 1.91E-06
304

F. MicroRNA Array Data Appendix
GNS NS
microRNAs G7A G7B G26A G26B G144A G144B G166A G166B CB660A CB660B CB130A CB130B CB152A CB152B CB171A CB171B LogFC p-value FDR
hsa-miR-34c-3p 0.51 0.34 0.45 0.72 -1.75 -2.1 -1.58 -1.28 -1.98 -2.25 -2.4 -2.82 -1.66 -2.5 -2.5 -2.38 1.7 8.50E-08 1.91E-06
hsa-miR-3609 3.73 4.15 3.19 3.56 2.37 2.59 0.16 0.8 2.05 1.95 -0.71 -0.03 0.07 0.31 1.01 0.21 2 7.33E-08 1.70E-06
hsa-miR-25 13.1 13.05 12.06 12.09 11.86 11.63 9.99 10.44 9.88 9.54 10.44 10.44 10.6 10.29 10.65 10.74 1.5 7.07E-08 1.67E-06
hsa-miR-885-5p 6.59 6.58 1.25 1.99 5.06 4.58 -0.48 -0.72 1.21 0.84 0.33 -0.77 -0.51 -1 -0.12 1.66 2.9 6.18E-08 1.51E-06
hsa-miR-140-3p 9.2 8.7 7.21 7.27 9.37 9.37 5.29 5.58 6.43 6.14 6.02 5.74 6.37 6.17 6.38 6.2 1.6 3.99E-08 9.94E-07
hsa-miR-2276 5.71 6.43 3.34 4.75 4.85 4.87 4.99 4.7 3.26 3.2 -0.45 1.43 1.39 2.05 1.39 1.29 3.3 3.62E-08 9.29E-07
hsa-miR-338-3p 8.47 8.76 3.96 4.84 9.63 10.78 2.73 3.02 3.11 2.81 4.8 3.77 2.74 2.67 5.22 3.99 2.9 3.65E-08 9.29E-07
hsa-miR-502-5p 3.6 3.83 3.81 3.93 4.25 3.73 2.27 3.19 2.15 1.96 0.38 -0.44 2 0.92 0.16 0.91 2.6 3.54E-08 9.29E-07
hsa-miR-155 7.97 7.44 6.73 6.98 5.48 5.2 6.99 7.33 5.33 5.83 5.89 6.46 1.16 0.73 6.64 6.88 1.9 3.18E-08 8.56E-07
hsa-miR-542-5p 7.26 7.37 6.24 6.43 6.74 7.36 4.02 4.35 5.96 4.95 1.97 0.8 6.44 6.21 1.9 1.3 2.5 3.00E-08 8.25E-07
hsa-miR-362-5p 6.34 6.36 6.14 6.38 6.79 6.31 5.45 6.04 5.02 5.1 3.75 2.71 4.6 4.72 3.7 3.41 2.1 2.66E-08 7.45E-07
hsa-miR-4257 4.55 5.02 4.38 5.85 3.7 4.39 5.3 5.43 3.78 3 -0.77 -2.4 1.95 3.18 -0.18 -1.54 4 2.33E-08 6.82E-07
hsa-miR-186 7.64 7.91 7.63 8.02 6.22 6.45 5 5.23 6.1 5.84 4.05 4.45 5.53 5.39 4.59 4.81 1.7 2.21E-08 6.61E-07
hsa-miR-532-5p 5.94 6.13 5.81 6.05 6.54 6 5.01 5.69 4.49 4.39 3.08 2.29 4.39 4.37 3.6 2.98 2.2 1.65E-08 5.18E-07
hsa-miR-139-5p 4.27 4.37 1.34 1.36 6.36 6.21 -0.11 -0.38 -0.55 -0.17 -1.54 -0.77 -0.79 1.06 -1.36 -0.03 3.4 1.22E-08 4.00E-07
hsa-miR-192 7.99 8.26 6.54 6.97 6.23 6.13 5.47 5.69 5.43 5.24 5.13 4.95 4.99 4.91 4.67 4.68 1.7 1.13E-08 3.81E-07
hsa-miR-663 6.22 6.99 4.14 5.28 5.31 5.39 5.5 5.43 3.95 4.15 2.59 1.41 2.59 3.02 1.97 1.9 2.8 9.73E-09 3.36E-07
hsa-miR-29c* 4.32 4.33 5.6 5.47 2.58 2.59 4.21 4.51 2.57 1.53 0.55 1.21 3.03 3 1.9 2.01 2.2 9.57E-09 3.36E-07
hsa-miR-1972 2.46 3.57 0.98 2.45 0.4 0.91 1.29 2.52 -0.6 -1.88 -2.86 -2.53 -2.49 -2.18 -2.29 -2.09 3.9 5.28E-09 1.98E-07
hsa-miR-1208 4.58 4.72 3.55 4.49 4 3.75 4.5 4.67 2.59 1.82 1.53 1.46 2.53 2.85 1.21 1.35 2.4 5.32E-09 1.98E-07
hsa-miR-191 2.92 2.6 3.34 3.53 0.08 0.48 1.23 2.24 -0.52 -0.12 -1.04 -2.35 -1.68 -1.06 -1.58 -0.82 3.2 4.89E-09 1.95E-07
hsa-miR-339-5p 1.41 0.96 0.95 1.04 -1.15 -0.41 -1.43 -1.43 -2.38 -2.36 -2.41 -2.22 -1.76 -2 -2.24 -1.72 2.1 4.93E-09 1.95E-07
hsa-miR-501-3p 2.7 3.11 2.6 3.31 3.52 3.02 -1.01 -0.5 -0.65 0.1 -1.72 -1.98 0.06 -1.43 -1.86 -1.55 3.2 4.37E-09 1.84E-07
hsa-miR-424* 4.62 4.34 3.36 3.73 1.93 2.94 -0.62 -0.14 1.82 0.15 -2.63 -2.35 -0.24 -0.77 -2.08 -1.78 3.5 2.91E-09 1.27E-07
hsa-miR-488 6.31 6.56 6.22 6.64 6.88 7.3 -2.47 -2.52 -1.39 -1.59 5.08 4.09 -2.39 -1.8 5.27 4.24 2.9 2.65E-09 1.19E-07
hsa-miR-718 2.78 3.08 2.24 2.9 1.66 2.12 0.39 0.96 0.23 -0.8 -2.01 -2.28 -0.07 -1.32 -1.89 -2.09 3.3 2.26E-09 1.05E-07
hsa-miR-199b-5p 8.7 8.77 1.52 1.24 8.02 7.27 1.96 2.09 3.25 2.81 2.32 2.27 1.95 2.01 1.27 2.23 2.7 1.30E-09 6.50E-08
hsa-miR-10b* 7.24 7.18 0.71 0.53 3.88 4.1 2.21 2.49 -0.46 0.52 0.79 0.72 0.67 0.98 0.28 1.25 2.9 8.25E-10 4.27E-08
hsa-miR-532-3p 4.65 4.71 4.82 4.91 5.51 4.88 2.66 3.6 3.17 3.32 -0.71 -2.7 2.23 1.83 -2.24 -2.56 4.2 7.74E-10 4.17E-08
hsa-miR-135a* 3.87 4.15 3.08 3.76 2.86 2.68 3.38 3.03 2.8 1.96 -1.5 -2.16 -0.79 0.02 -0.38 -1.38 3.5 6.72E-10 3.77E-08
hsa-miR-215 6.72 6.99 5.17 5.69 4.99 4.88 3.79 4.02 3.52 3.18 2.87 2.57 2.53 2.89 2.82 2.86 2.4 4.79E-10 2.80E-08
hsa-miR-339-3p 1.65 1.69 2.03 2.15 0.97 0.42 -1.98 -0.75 -2.39 -2.82 -2.42 -2.9 -2.36 -2.83 -2.06 -2.14 3.3 4.13E-10 2.53E-08
hsa-miR-874 6.65 6.82 4.79 5.75 4.59 4.57 4.6 4.5 4.86 4.73 -1.2 -2.4 2.94 3.55 1.05 0.76 3.5 3.30E-10 2.12E-08
hsa-miR-10b 12.01 11.95 4.74 4.94 8.18 8.22 6.15 6.92 1.92 1.97 5.53 5.61 6.13 6.14 5.94 6.7 2.9 1.53E-10 1.03E-08
hsa-miR-625 5.69 5.92 4.53 4.94 3.78 3.84 4.83 5.15 -3.28 -3.06 3.82 4.1 3.8 3.84 3.99 4.53 2.6 1.46E-10 1.03E-08
hsa-miR-96 9.43 8.84 11.22 10.47 9.1 9.79 10.24 10.87 6.63 4.82 4.42 4.56 5.98 6.15 4.81 4.78 4.7 1.11E-10 8.33E-09
hsa-miR-450a 8.77 8.89 6.75 7.14 6.89 7.59 3.98 4.43 6.05 4.53 -1.22 -2.24 6.31 5.94 -1.08 -0.88 4.6 7.33E-11 5.81E-09
hsa-miR-1469 2.29 2.37 2.47 2.57 0.59 1.31 -1.04 0.19 -1.91 -2.45 -2.93 -2.7 -2.33 -2.22 -2.5 -2.79 3.8 5.98E-11 5.13E-09
hsa-miR-129-5p 3.05 3.21 1.45 1.72 4.88 4.14 -1.13 -0.49 -0.88 -0.41 -1.51 -1.32 -1.06 -0.83 -1.43 -1.42 3.2 6.09E-11 5.13E-09
hsa-miR-363 10.05 10.11 11.75 11.63 10.9 10.67 7.46 7.96 7.92 7.03 7.71 8.06 3.92 3.98 7.9 8.26 3.2 5.76E-11 5.13E-09
hsa-miR-378 6.65 6.4 3.26 4.84 7.16 6.8 -0.15 0.94 -2.19 -2.12 -2.12 -2.27 -1.57 -1.33 -2.01 -0.22 6.2 2.18E-11 2.49E-09
hsa-miR-183* 0.43 -0.16 3.65 2.92 1.27 2.35 3.42 4.08 -1.7 -2.32 -2.33 -2.41 -2.21 -2.35 -2.23 -1.97 4.4 2.00E-11 2.49E-09
hsa-miR-183 3.63 2.03 6.54 5.61 1.16 3.84 3.46 4.62 -4.24 -4.58 -4.79 -4.79 -4.44 -4.41 -4.63 -4.54 8.4 8.49E-12 1.27E-09
hsa-miR-194 5.86 6.18 4.32 4.75 3.78 3.57 0.69 1.93 0.94 0.47 -1.84 -3.02 -1.25 -1.49 -1.94 -2.22 5.2 7.16E-12 1.20E-09
hsa-miR-196a 8.87 9.09 5.05 4.64 5.86 5.37 6.27 6.68 0.42 0.32 -0.65 1.03 1.71 2.32 1.46 1.51 5.5 4.81E-12 9.25E-10
hsa-miR-138 4.97 5.22 5.48 4.16 5.2 4.91 -2.52 -2.22 -2.63 -2.61 -3.95 -3.78 4.31 3.81 -3.78 -3.46 4.7 4.18E-12 9.25E-10
hsa-miR-3648 4.88 4.92 4.35 4.93 4.06 4.87 4.18 4.35 3.87 3.07 -2.62 -2.85 -1.67 -0.53 -2.01 -2.07 5.2 3.29E-12 8.86E-10
hsa-miR-502-3p 4.53 4.56 4.46 4.67 4.9 4.3 1.96 2.49 2.78 2.4 -2.36 -2.05 0.39 0.87 -2.42 -1.88 4.3 1.98E-12 6.66E-10
hsa-miR-182 4.95 4.03 6.91 6.15 3.72 4.84 4.96 5.48 -0.98 -2 -3.14 -2.42 -2.19 -2.12 -2.96 -2.53 7.4 2.74E-13 1.23E-10
hsa-miR-148a 10.36 10.32 10.3 10.87 9.02 8.9 8.89 9.5 6.6 5.95 0 0.66 0.39 1.51 0.36 0.63 7.8 1.63E-14 1.10E-11
hsa-miR-196b 9.79 9.81 3.26 3.39 6.5 6.32 2.45 3.53 -3.08 -3.1 -3.77 -3.41 -2.81 -2.23 -3.57 -3.06 8.8 7.07E-16 9.52E-13
305

List of Abbreviations
5-bromo-2’-deoxyuridine BrdU
AGO Argonaute
AKT v-akt murine thymoma viral oncogene
APC Antigen Presenting Cell
ARF Alternate Reading Frame
ASR Age Standardized Rate
ATM Ataxia telangiectasia mutated
BLBP Brain lipid binding protein
BMP Bone Morphogenetic Protein
Ca 2+ Calcium
CCDS Consensus Coding Sequence
CD133 Prominin
CD144 Vascular endothelial-cadherin
CDK Cyclyn-Dependent Kinase
CDKN Cyclyn-Dependent Kinase Inhibitor
CGH Comparative genomics hybridization
CHI3L1 Chitinase 3-like 1
CNA Copy Number Aberrations
CNS Central Nervous System
CO2
Carbon dioxide
CSF Cerebro Spinal Fluid
Ct Cycle threshold
CTL Cytotoxic Lymphocyte
CTMP C-terminal modulator protein
CpG Cytosine-phosphate-Guanine
DNA Deoxyribonucleic Acid
Dcx Doublecortin
ECM Extracellular matrix
EGF Epidermal Growth Factor
EGFR Epidermal Growth Factor Receptor
ES Embryonic Stem
EST Expressed Sequence Tags
FACS Fluorescent Activated Cell Sorting
FC Fold change
FDR False Discovery Rate
FGF2 Fibroblast Growth Factor 2
FGF2 Fibroblast growth factor 2
FISH Fluorescence In Situ Hybridization
FOXO Forkhead box O
G-CIMP Glioma CpG Island Methylator Phenotype
GABA Gamma-aminobutyric acid
GBM Glioblastoma Multiforme
GEMM Genetically Engineered Mouse Model
GFAP Glial Fibrillary Acidic Protein
GFAP Glial Fibrillary Acidic Protein
GFP Green Fluorescent Protein
GLAST Glutamate Aspartate Transporter
GO Gene Ontology
GSEA Gene Set Enrichment Analysis
GTP Guanosine-5’-triphosphate
GTPase Guanosine-5’-triphosphate hydrolase
HGNC HUGO gene nomenclature committee
HIF1 Hypoxia Inducible Factor 1
HIF1A Hypoxia Inducible Factor 1, subunit α
ICM Inner Cell Mass
IDH1 Isocitrate Dehydrogenase 1
IQGAP1 IQ motif containing GTPase activating protein 1
IRES Internal Ribosome Entry Site
KEGG Kyoto Encyclopedia of Genes and Genomes
LIF Leukemia Inhibitory Factor
306

Ln Natural logarithm
LOH Loss of Heterozygosity
MAP2 Microtubule-associated protein 2
MAPK Mitogen Activated Protein Kinase
MDM2 Mdm2 p53 binding protein homolog
MDM4 Mdm4 p53 binding protein homolog
MGC Mammalian Gene Collection
MGI Mouse Genome Informatics
MGMT O-6-methylguanine-DNA methyltransferase
MIQE Minimum Information for publication of Quantitative real-time PCR Experiments
MMR Mismatch Repair
MSH6 MutS homolog 6
NADPH Nicotinamide Adenine Dinucleotide Phosphate
NCBI National Center for Biotechnology Information
NEP Neuroepithelial progenitor
NF1 Neurofibromin 1
NS Neural Stem
OCT4 Octamer-binding protein 4
OLIG2 Oligodendrocyte lineage transcription factor 2
ORF Open Reading Frame
PCA Principal Component Analysis
PDGF Platelet-Derived Growth Factor
PDGFR Platelet-Derived Growth Factor Receptor
PDPK1 3-phosphoinositide dependent protein kinase-1
PHLPP PH domain and leucine rich repeat protein phosphatase 1
PI3K Phosphoinositide-3-Kinase
PI3KR Phosphoinositide-3-Kinase Receptor
PIP3
phosphatidylinositol (3,4,5)-trisphosphate
PTEN Phosphatase and tensin homolog
Pax6 Paired box gene 6
RA Retinoic Acid
RAS Rat Sarcoma
RB1 Retinoblastoma 1
RISC RNA Induced Silencing Complex
RNA Rybonucleic Acid
RTK Receptor Tyrosine Kinase
SAGE Serial Analysis of Gene Expression
SCV Squared Coefficient of Variation
SGZ Subgranular Zone
SHH Sonic hedge hog
SILAC Stable Isotope Labeling by Amino acids in Cell culture
SKY Spectral Karyotyping
SNP Single Nucleotide Polymorphism
SOX1 Sex determining region Y-box 1
SOX2 Sex determining region Y-box 2
SSEA1 Stage-specific embryonic antigen 1
SVZ Subventricular Zone
TCGA The Cancer Genome Atlas
TGFβ Tumour Growth Factor β
TP53 Tumor Protein 53
TUBB type III β-tubulin
Tag-seq Tag sequencing
UCSC University California Santa Cruz
UTR Untranslated Region
VZ Ventricular Zone
WHO World Health Organization
aCGH array comparative genomics hybridization
bp base pair
iHOP information Hyperlinked Over Proteins
mRNA messenger RNA
mm millimeter
ncRNAs non-coding RNAs
nt nucleotide
oligo-dT oligo deoxy-thymine
poly-A poly-adenine
poly-T poly-thymine
qRT-PCR quantitative Real-Time PCR
rRNA ribosomal RNA
tpm tags per million
UCSC University of California Santa Cruz

List of Figures
1.1 Estimates of survival amongst GBM patients treated with radio-
therapy alone or radiotherapy with the alkylating agent temo-
zolomide. Taken from Stupp et al 2005 [472]. . . . . . . . . . . . 7
1.2 KEGG Glioma Pathway. . . . . . . . . . . . . . . . . . . . . . . 23
1.3 Visualisation generated from list of 345 interactors (orange) of
TP53 (yellow) from the BioGRID 3.1 [62] repository for inter-
action datasets. . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4 The Biocarta pathway for Rb signaling. . . . . . . . . . . . . . . 30
2.1 Cross-section through the neural tube. . . . . . . . . . . . . . . 35
2.2 Surface markers of radial glia are expressed by NS cell lines,
indicating that these cells may provide the biological context to
work with progenitors of the CNS. . . . . . . . . . . . . . . . . . 37
2.3 Sources of NS cells: (a) ICM; (b) SVZ. . . . . . . . . . . . . . . 38
2.4 (a,b) Contrast microscopy images of early phase neurosphere
formation. (c,d) Immunofluorescence microscopy images of EGFR
and Nestin detected on an intact neurosphere. . . . . . . . . . . 39
2.5 Schematisation of the neurosphere assay used to study neural
precursor cells in culture. . . . . . . . . . . . . . . . . . . . . . . 41
2.6 Diagram of the progressive lineage restriction of ES cells differ-
entiating toward the neural phenotype. . . . . . . . . . . . . . . 44
2.7 Protocol describing conversion of ES cells into immortalised NS
cell lines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.8 Representation of the Sox1-GFP reporter construct used in the
niche-independent NS cell protocol. . . . . . . . . . . . . . . . . 46
2.9 Roles of EGF and FGF2 in the derivation and maintenance of
NS cells. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.1 Stem cell differentiation hierarchy. . . . . . . . . . . . . . . . . . 58
3.2 Diagram of asymmetric and symmetric cell division. . . . . . . . 59
308

3.3 Schematisation of cell cycle phases. . . . . . . . . . . . . . . . . 67
3.4 Glioblastoma treatment with ionizing radiation. . . . . . . . . . 68
3.5 Glioblastoma treatment with BMPs. . . . . . . . . . . . . . . . 70
3.6 PCA diagram of global mRNA expression in GNS cell lines. . . 82
4.1 Classes of non-coding RNAs discovered to date. . . . . . . . . . 86
5.1 Schematisation of the longSAGE protocol. . . . . . . . . . . . . 95
5.2 Boxplot of normalised Ct values. . . . . . . . . . . . . . . . . . . 102
5.3 Correlation scatter plot of the raw Ct values vs endogenous
controls. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.4 Scatter plots of our A and B biological replicates. . . . . . . . . 106
5.5 Dot plot of the standard deviations of the differences between
expression levels in two replicates. . . . . . . . . . . . . . . . . . 107
5.6 Literature mining diagram of code functions. . . . . . . . . . . . 109
5.7 Schematisation of a typical Cytoscape network. . . . . . . . . . 115
6.1 Sequencing construct schematisation. . . . . . . . . . . . . . . . 121
6.2 Diagram of the extraction, filtering and mapping phases for
reads and tags. . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.3 Pie charts for the proportion of filtered tags. . . . . . . . . . . . 123
6.4 Diagram of the tag mapping strategy. . . . . . . . . . . . . . . . 124
6.5 Pie charts for the assignment of tags to genes. . . . . . . . . . . 126
6.6 Correlations for all combinations of cell lines. . . . . . . . . . . . 127
6.7 CGH array analysis. . . . . . . . . . . . . . . . . . . . . . . . . 128
6.8 Curves show distributions of expression level differences between
GNS and NS lines. . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.9 Plot of the estimates of the variance against the base levels for
each gene. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.10 Plot of the fold change versus the mean for normal vs tumour
samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.11 Tag mapping of NTRK2 on UCSC genome browser. . . . . . . . 135
6.12 Expression estimates correlate well between Tag-seq and qRT-
PCR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.13 Heatmap of 29 genes differentially expressed between 16 GNS
and 6 NS cell lines . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.14 Expression levels of the 29 genes distinguishing GNS from NS
lines as percent of NS geometric mean. . . . . . . . . . . . . . . 142
6.15 Tag mapping of BMP7 on UCSC genome browser. . . . . . . . . 148

6.16 Tag mapping of TPM1 on UCSC genome browser. . . . . . . . . 149
6.17 Schematisation of the process of finding genes with differentially
expressed isoforms. . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.18 Schematisation of the localisation of the microRNA seeds on the
mapped isoforms. . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.19 Isoform detection by multi tag mapping of gene GAPVD1 on
UCSC genome browser. . . . . . . . . . . . . . . . . . . . . . . . 154
6.20 Isoform detection by multi tag mapping of gene SMAD1 on
UCSC genome browser. . . . . . . . . . . . . . . . . . . . . . . . 156
6.21 Correlated expression of CTSC and a nearby ncRNA. . . . . . . 159
6.22 Histogram of expression levels of HOTAIRM1 and surrounding
HOX genes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
7.1 Enrichment plots for the top two pathways revealed through
GSEA of the KEGG database of pathways [2]. . . . . . . . . . . 165
7.2 GSEA plots of (a) nominal p-values vs normalised enrichment
score and (b) line graph of the enrichment scores across pathways.165
7.3 Correlation of Tag-seq interrogated cell lines with glioblastoma
subtype expression signatures. . . . . . . . . . . . . . . . . . . . 170
7.4 Core gene expression changes in GNS lines are mirrored in
glioblastoma tumours. . . . . . . . . . . . . . . . . . . . . . . . 172
7.5 Association between GNS signature and other survival predictors.184
7.6 Association between GNS signature score and patient survival. . 185
7.7 The integrated glioblastoma pathway subdivided into sections
identifying the gene networks that participate in the pathway. . 188
7.8 Integrated GBM pathway used to overlay the Tag-seq GNS
dataset. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.9 Affected p53, RB1 and PTEN/PI3K pathways. . . . . . . . . . 192
7.10 Four integrated GBM pathways overlaid with Tag-seq expres-
sion level measures for each GNS cell line. . . . . . . . . . . . . 194
7.11 Integrated GBM pathway with the TCGA dataset overlaid. . . . 199
7.12 Integrated GBM pathway with the HGG dataset overlaid. . . . 200
7.13 Three integrated GBM pathways overlaid with the GNS, TCGA
and HGG datasets. . . . . . . . . . . . . . . . . . . . . . . . . . 201
8.1 Overview of GenemiR software. . . . . . . . . . . . . . . . . . . 205
8.2 Internal organisation of the target prediction database of Gen-
emiR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

8.3 Workflows at the core of the primitive functions of the GenemiR
software. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
8.4 Step by step diagram of the ensemble method adopted to find
the score E (=C2/C1) of prediction accuracy for prediction al-
gorithms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
D.1 Integrated GBM pathway with G144 cell line Tag-seq expres-
sion data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
D.2 Integrated GBM pathway with G144ED cell line Tag-seq ex-
pression data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
D.3 Integrated GBM pathway with G166 cell line Tag-seq expres-
sion data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
D.4 Integrated GBM pathway with G179 cell line Tag-seq expres-
sion data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

List of Tables
1.1 Histological Types and Prognosis of Gliomas (y, years). Taken
from Doyle et al 2005 [128]. . . . . . . . . . . . . . . . . . . . . 6
2.1 Summary of commonalities and differences between ES cells and
NS cells. Adapted from Pollard et al 2006 [399] . . . . . . . . . 53
3.1 Summary of characteristics of NBE and serum-cultured glioblas-
toma cells. Adapted from Lee et al 2006. . . . . . . . . . . . . . 77
5.1 Summary of cell lines investigated with Tag-seq. . . . . . . . . . 95
5.2 Classification of sequenced tags in each cell line. . . . . . . . . . 99
5.3 Summary of statistics using the χ 2 and logarithmic tests. . . . . 111
5.4 Public gene expression datasets used in thesis. . . . . . . . . . . 113
6.1 Summary of the available clinical data for our GNS cell lines. . . 121
6.2 Summary of reads per cell line library. . . . . . . . . . . . . . . 121
6.3 Significance of the correlation found between CNAs and expres-
sion levels measured with Fisher’s exact test (p-value). . . . . . 128
6.4 Table of genes with large expression changes common to the
GNS cell lines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
6.5 Differentially expressed genes assigned to a four-tier classifica-
tion system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.6 Genes with limited or no evidence of implication in glioblastoma
that appear in our pathway. . . . . . . . . . . . . . . . . . . . . 145
6.7 Summary of predicted microRNAs targeting differentially ex-
pressed isoforms. . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.8 MicroRNA array results for GNS cell lines with respect to NS
cell lines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.9 Differentially expressed ncRNAs. . . . . . . . . . . . . . . . . . 157
312

7.1 Selected Gene Ontology terms and InterPro domains enriched
among differentially expressed genes. . . . . . . . . . . . . . . . 163
7.2 Representative KEGG pathways from signaling pathway impact
analysis of gene expression differences between GNS and NS lines.164
7.3 Summary of all MHC class I and II genes. . . . . . . . . . . . . 167
7.4 Literature survey for the 29 genes found to distinguish GNS
from NS lines across a panel of 21 cell lines . . . . . . . . . . . . 175
7.5 Survival tests for the 29 genes found via qRT-PCR to distinguish
GNS cell lines from NS cell lines. . . . . . . . . . . . . . . . . . 183
7.6 Significance of survival association for GNS signature and IDH1
status. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
7.7 Node assignment in the glioblastoma pathway. . . . . . . . . . . 190
8.1 microRNA target prediction algorithms used by GenemiR with
number of microRNA:3'UTR interactions predicted. The origi-
nal target identifiers refer to the identifiers used by a prediction
algorithm to identify the targeted genes. The final target iden-
tifiers refer to the identifiers that are returned by any query of
any prediction algorithm database. . . . . . . . . . . . . . . . . 209
8.2 Single prediction algorithm ensemble analysis results. Displayed
in descending order of E-score. . . . . . . . . . . . . . . . . . . . 215
8.3 All combinations of prediction algorithms in descending order
of E-score. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
A.1 Classification of differentially expressed genes at 10% FDR. . . . 235
A.2 Classification of differentially expressed genes based on litera-
ture mining analysis. . . . . . . . . . . . . . . . . . . . . . . . . 251
A.3 Raw Ct values. Abbreviations: "down" for down-regulated,
"up" for up-regulated, and "Norm" for Normalisation control. . 257
A.4 Normalised Ct values. Abbreviations: "down" for down-regulated,
"up" for up-regulated, and "Norm" for Normalisation control. . 261
A.5 Pearson correlation values between the normalised Ct values
measured through qRT-PCR and the tag counts measured across
the five GNS and NS cell lines assayed via Tag-seq. . . . . . . . 265
C.1 Differentially expressed non-coding RNAs. . . . . . . . . . . . . 274
D.1 GBM pathway interaction data. . . . . . . . . . . . . . . . . . . 276

E.1 Fold-changes measured by exon array for GNS cell lines. . . . . 286
F.1 Differentially expressed microRNAs in GNS vs NS cell lines at
FDR

Bibliography
[1] Cell signaling technology. MAPK-ERK signaling cascade.
http://www.cellsignal.com/reference/pathway/mapk_erk_growth.html.
[2] Kegg pathways. http://www.genome.jp/kegg/pathway.html.
[3] Kyoto encyclopedia of genes and genomes. Glioma pathway.
http://www.genome.jp/kegg-bin/_pathway?hsa05214.
[4] Kyoto encyclopedia of genes and genomes. P53 pathway.
http://www.genome.jp/kegg/pathway/hsa/hsa04115.html.
[5] Panter pathway. P53 pathway. http://www.pantherdb.org/pathway/pathwaydiagram.
[6] S. Acharya, T. Wilson, S. Gradia, M. F. Kane, S. Guerrette, G. T. Marsischky,
R. Kolodner, and R. Fishel. hMSH2 forms specific mispair-binding complexes with
hMSH3 and hMSH6. Proceedings of the National Academy of Sciences of the United
States of America, 93(24):13629–13634, Nov. 1996.
[7] M. Adamowicz, B. Radlwimmer, R. Rieker, D. Mertens, M. Schwarzbach, P. Schraml,
A. Benner, P. Lichter, G. Mechtersheimer, and S. Joos. Frequent amplifications and
abundant expression of trio, nkd2, and irx2 in soft tissue sarcomas. Genes, Chromosomes
and Cancer, 45(9):829–838, 2006.
[8] D. Adams, B. Hasson, A. Boyer-Boiteau, A. El-Khishin, and V. Shashoua. A peptide
fragment of ependymin neurotrophic factor uses protein kinase c and the mitogenactivated
protein kinase pathway to activate c-jun n-terminal kinase and a functional
ap-1 containing c-jun and c-fos proteins in mouse nb2a cells. Journal of neuroscience
research, 72(3):405–416, 2003.
[9] A. Adesina, Y. Nguyen, V. Mehta, H. Takei, P. Stangeby, S. Crabtree, M. Chintagumpala,
and M. Gumerlock. Foxg1 dysregulation is a frequent event in medulloblastoma.
Journal of neuro-oncology, 85(2):111–122, 2007.
[10] C. Agulhon, J. Petravicz, A. McMullen, E. Sweger, S. Minton, S. Taves, K. Casper,
T. Fiacco, and K. McCarthy. What is the role of astrocyte calcium in neurophysiology?
Neuron, 59(6):932–946, 2008.
[11] E. Ah Cho and G. Dressler. Tcf-4 binds [beta]-catenin and is expressed in distinct
regions of the embryonic brain and limbs. Mechanisms of development, 77(1):9–18,
1998.
[12] N. Ahuja, Q. Li, A. Mohan, S. Baylin, and J. Issa. Aging and dna methylation in
colorectal mucosa and cancer. Cancer research, 58(23):5489–5494, 1998.
[13] N. Ahuja, Q. Li, A. Mohan, S. Baylin, and J. Issa. Aging and dna methylation in
colorectal mucosa and cancer. Cancer research, 58(23):5489, 1998.
[14] T. Akai, Y. Ueda, Y. Sasagawa, T. Hamada, T. Date, S. Katsuda, H. Iizuka, Y. Okada,
and K. Chada. High mobility group ic protein in astrocytoma and glioblastoma.
Pathology-Research and Practice, 200(9):619–624, 2004.
[15] M. Al-Hajj, M. Wicha, A. Benito-Hernandez, S. Morrison, and M. Clarke. Prospective
identification of tumorigenic breast cancer cells. Proceedings of the National Academy
of Sciences, 100(7):3983, 2003.
[16] P. Alexiou, M. Maragkakis, G. Papadopoulos, M. Reczko, and A. Hatzigeorgiou. Lost
in translation: an assessment and perspective for computational microrna target identification.
Bioinformatics, 25(23):3049, 2009.
315

[17] N. Allen and B. Barres. Signaling between glia and neurons: focus on synaptic plasticity.
Current opinion in neurobiology, 15(5):542–548, 2005.
[18] A. Alvarez-Buylla and D. Lim. For the Long Run: Maintaining Germinal Niches in
the Adult Brain. Neuron, 41(5):683–686, 2004.
[19] V. Ambros and X. Chen. The regulation of genes and genomes by small RNAs.
Development, 134(9):1635–41, 2007.
[20] M. Amiry-Moghaddam and O. Ottersen. The molecular basis of water transport in
the brain. Nature Reviews Neuroscience, 4(12):991–1001, 2003.
[21] S. Anders and W. Huber. Differential expression analysis for sequence count data.
Genome Biol, 11(10):R106, 2010.
[22] P. Andrews. Retinoic acid induces neuronal differentiation of a cloned human embryonal
carcinoma cell line in vitro* 1. Developmental biology, 103(2):285–293, 1984.
[23] A. Aravin, D. Gaidatzis, S. Pfeffer’t, M. Lagos-Quintana, P. Landgraf, and T. Tuschl.
A novel class of small rnas bind to mili protein in mouse testes. Nature, 442:203–207,
2006.
[24] A. Aravin, N. Naumova, A. Tulin, V. Vagin, Y. Rozovsky, and V. Gvozdev. Doublestranded
rna-mediated silencing of genomic tandem repeats and transposable elements
in the d. melanogaster germline. Current Biology, 11(13):1017–1027, 2001.
[25] K. Archer, V. Mas, K. David, D. Maluf, K. Bornstein, and R. Fisher. Identifying genes
for establishing a multigenic test for hepatocellular carcinoma surveillance in hepatitis
c virus-positive cirrhotic patients. Cancer Epidemiology Biomarkers & Prevention,
18(11):2929–2932, 2009.
[26] M. Arpin, E. Friederich, M. Algrain, F. Vernel, and D. Louvard. Functional differences
between l-and t-plastin isoforms. The Journal of cell biology, 127(6):1995, 1994.
[27] M. Arpin, E. Friederich, M. Algrain, F. Vernel, and D. Louvard. Functional differences
between l-and t-plastin isoforms. The Journal of cell biology, 127(6):1995–2008, 1994.
[28] ArrayExpress. www.ebi.ac.uk/arrayexpress.
[29] M. Assimakopoulou, M. Kondyli, G. Gatzounis, T. Maraziotis, and J. Varakis. Neurotrophin
receptors expression and jnk pathway activation in human astrocytomas.
BMC cancer, 7(1):202, 2007.
[30] S. Assinder, J. Stanton, and P. Prasad. Transgelin: an actin-binding protein and tumour
suppressor. The international journal of biochemistry & cell biology, 41(3):482–
486, 2009.
[31] P. Au, Q. Zhu, E. Dennis, and M. Wang. Long non-coding rna-mediated mechanisms
independent of the rnai pathway in animals and plants. RNA biology, 8(3), 2011.
[32] J. Aubert, M. Stavridis, S. Tweedie, M. O’Reilly, K. Vierlinger, M. Li, P. Ghazal,
T. Pratt, J. Mason, D. Roy, et al. Screening for mammalian neural genes via
fluorescence-activated cell sorter purification of neural precursors from sox1-gfp knockin
mice. Proceedings of the National Academy of Sciences of the United States of
America, 100(Suppl 1):11836, 2003.
[33] F. Azevedo, L. Carvalho, L. Grinberg, J. Farfel, R. El Ferreti, R. Leite, W. Jacob filho,
R. lent, and S. herculano houzel. Equal numbers of neuronal and non-neuronal cells
make the human brain an isometrically scaled-up primate brain. brain, 513:532–541,
2009.
[34] H. Babu, G. Cheung, H. Kettenmann, T. Palmer, and G. Kempermann. Enriched
monolayer precursor cell cultures from micro-dissected adult mouse dentate gyrus
yield functional granule cell-like neurons. PLoS One, 2(4):e388, 2007.
[35] A. Bader, S. Kang, and P. Vogt. Cancer-specific mutations in PIK3CA are oncogenic
in vivo. Proceedings of the National Academy of Sciences of the United States of
America, 103(5):1475, 2006.
[36] A. Bader, S. Kang, L. Zhao, and P. Vogt. Oncogenic PI3K deregulates transcription
and translation. Nature reviews cancer, 5(12):921–929, 2005.
[37] D. Baek, J. Villen, C. Shin, F. D. Camargo, S. P. Gygi, and D. P. Bartel. The impact
of microRNAs on protein output. Nature, 455(7209):64–71, 2008.

[38] G. Bain, D. Kitchens, M. Yao, J. Huettner, and D. Gottlieb. Embryonic stem cells
express neuronal properties in vitro. Developmental biology, 168(2):342–357, 1995.
[39] L. Balenci. IQGAP1 Protein Specifies Amplifying Cancer Cells in Glioblastoma Multiforme.
Cancer Research, 66(18):9074–9082, Sept. 2006.
[40] S. Bao, Q. Wu, R. McLendon, Y. Hao, Q. Shi, A. Hjelmeland, M. Dewhirst, D. Bigner,
and J. Rich. Glioma stem cells promote radioresistance by preferential activation of
the DNA damage response. Nature, 444(7120):756–760, 2006.
[41] I. Barani, S. Benedict, and P. Lin. Neural stem cells: implications for the conventional
radiotherapy of central nervous system malignancies. International Journal of
Radiation Oncology* Biology* Physics, 68(2):324–333, 2007.
[42] J. Barnes and P. Hut. A hierarchical 0 (n log iv) force-calculation algorithm. nature,
324:4, 1986.
[43] D. P. Bartel. MicroRNAs: target recognition and regulatory functions. Cell,
136(2):215–33, 2009.
[44] N. Baumann. Biology of oligodendrocyte and myelin in the mammalian central nervous
system. Physiological Reviews, 2001.
[45] A. Bellacosa, C. C. Kumar, A. Di Cristofano, and J. R. Testa. Activation of AKT
kinases in cancer: implications for therapeutic targeting. Advances in cancer research,
94:29–86, 2005.
[46] A. Bellacosa, J. R. Testa, R. Moore, and L. Larue. A portrait of AKT kinases: human
cancer and animal models depict a family with strong individualities. Cancer biology
& therapy, 3(3):268–275, Mar. 2004.
[47] D. R. Bentley. Whole-genome re-sequencing. Current opinion in genetics and development,
16(6):545–52, 2006.
[48] I. Bentwich. Prediction and validation of micrornas and their targets. FEBS letters,
579(26):5904–5910, 2005.
[49] R. Berg, E. Leung, S. Gough, C. Morris, W. Yao, S. Wang, J. Ni, and G. Krissansen.
Cloning and characterization of a novel-beta integrin-related cdna coding for the protein
tied. Genomics, 56(2):169–178, 1999.
[50] R. Berg, E. Leung, S. Gough, C. Morris, W. Yao, S. Wang, J. Ni, and G. Krissansen.
Cloning and characterization of a novel β integrin-related cdna coding for the protein
tied (ÂŞten β integrin egf-like repeat domainsÂŤ) that maps to chromosome band
13q33: a divergent stand-alone integrin stalk structure. Genomics, 56(2):169–178,
1999.
[51] A. Bergamaschi, Y. Kim, K. Kwei, Y. La Choi, M. Bocanegra, A. Langerød, W. Han,
D. Noh, D. Huntsman, S. Jeffrey, et al. Camk1d amplification implicated in epithelialmesenchymal
transition in basal-like breast cancer. Molecular oncology, 2(4):327–339,
2008.
[52] R. Beroukhim, G. Getz, L. Nghiemphu, J. Barretina, T. Hsueh, D. Linhart, I. Vivanco,
J. Lee, J. Huang, S. Alexander, et al. Assessing the significance of chromosomal
aberrations in cancer: methodology and application to glioma. Proceedings of the
National Academy of Sciences, 104(50):20007–20012, 2007.
[53] M. Berry, Z. Ahmed, B. Lorber, M. Douglas, and A. Logan. Regeneration of axons in
the visual system. Restorative neurology and neuroscience, 26(2):147–174, 2008.
[54] F. Bertucci, P. Finetti, N. Cervera, E. Charafe-Jauffret, E. Mamessier, J. Adélaïde,
S. Debono, G. Houvenaeghel, D. Maraninchi, P. Viens, et al. Gene expression profiling
shows medullary breast cancer is a subgroup of basal breast cancers. Cancer Research,
66(9):4636–4644, 2006.
[55] M. Bibel, J. Richter, K. Schrenk, K. Tucker, V. Staiger, M. Korte, M. Goetz, and
Y. Barde. Differentiation of mouse embryonic stem cells into a defined neuronal lineage.
Nature neuroscience, 7(9):1003–1009, 2004.
[56] L. Biesecker. Exome sequencing makes medical genomics a reality. Nature genetics,
42(1):13, 2010.

[57] E. Bindewald and B. Shapiro. Rna secondary structure prediction from sequence
alignments using a network of k-nearest neighbor classifiers. Rna, 12(3):342–352,
2006.
[58] Biocarta. P53 signaling pathway. http://www.biocarta.com/pathfiles/h_p53pathway.asp.
[59] Biocarta. Pten dependent cell cycle arrest and apoptosis.
http://www.biocarta.com/pathfiles/h_ptenpathway.asp.
[60] Biocarta. Rb tumor suppressor/checkpoint signaling in response to dna damage.
http://www.biocarta.com/pathfiles/h_rbpathway.asp.
[61] Biocarta. www.biocarta.com.
[62] BioGRID. Database of protein and genetic interactions. www.thebiogrid.org.
[63] B. Boëda, D. Briggs, T. Higgins, B. Garvalov, A. Fadden, N. McDonald, and M. Way.
Tes, a specific mena interacting partner, breaks the rules for evh1 binding. Molecular
cell, 28(6):1071–1082, 2007.
[64] S. Bonnet, S. Archer, J. Allalunis-Turner, A. Haromy, C. Beaulieu, R. Thompson,
C. Lee, G. Lopaschuk, L. Puttagunta, S. Bonnet, et al. A mitochondria-k+ channel
axis is suppressed in cancer and its normalization promotes apoptosis and inhibits
cancer growth. Cancer cell, 11(1):37–51, 2007.
[65] K. Boon, E. Osório, S. Greenhut, C. Schaefer, J. Shoemaker, K. Polyak, P. Morin,
K. Buetow, R. Strausberg, S. De Souza, et al. An anatomy of normal and malignant
gene expression. Proceedings of the National Academy of Sciences, 99(17):11287, 2002.
[66] B. Borrell. How accurate are cancer cell lines? Nature, 463(7283):858, Feb. 2010.
[67] P. Bos, X. Zhang, C. Nadal, W. Shu, R. Gomis, D. Nguyen, A. Minn, M. van de
Vijver, W. Gerald, J. Foekens, et al. Genes that mediate breast cancer metastasis to
the brain. Nature, 459(7249):1005–1009, 2009.
[68] R. Bourgo, U. Ehmer, J. Sage, and E. Knudsen. RB deletion disrupts coordination
between dna replication licensing and mitotic entry in vivo. Molecular biology of the
cell, 22(7):931, 2011.
[69] C. Brennan, H. Momota, D. Hambardzumyan, T. Ozawa, A. Tandon, A. Pedraza,
and E. Holland. Glioblastoma subclasses can be defined by activity among signal
transduction pathways and associated genomic alterations. PLoS ONE, 4(11):e7752,
2009.
[70] J. Brennecke, D. R. Hipfner, A. Stark, R. B. Russell, and S. M. Cohen. bantam
encodes a developmentally regulated microRNA that controls cell proliferation and
regulates the proapoptotic gene hid in Drosophila. Cell, 113(1):25–36, 2003.
[71] J. Brennecke, A. Stark, R. B. Russell, and S. M. Cohen. Principles of microRNA-target
recognition. PLoS Biol, 3(3):e85, 2005.
[72] J. Briscoe and J. Ericson. Specification of neuronal fates in the ventral neural tube.
Current opinion in neurobiology, 11(1):43–49, 2001.
[73] P. Brodal. The Central Nervous System: Structure and Function. Oxford Univ Pr,
2010.
[74] J. Brognard, E. Sierecki, T. Gao, and A. C. Newton. PHLPP and a second isoform,
PHLPP2, differentially attenuate the amplitude of Akt signaling by regulating distinct
Akt isoforms. Molecular cell, 25(6):917–931, Mar. 2007.
[75] K. Brown, D. Strathdee, S. Bryson, W. Lambie, and A. Balmain. The malignant
capacity of skin tumours induced by expression of a mutant h-ras transgene depends
on the cell type targeted. Current biology, 8(9):516–524, 1998.
[76] O. Brustle, K. Jones, R. Learish, K. Karram, K. Choudhary, O. Wiestler, I. Duncan,
and R. McKay. Embryonic stem cell-derived glial precursors: A source of myelinating
transplants. Science, 285(5428):754–756, 1999.
[77] J. Buckner, P. Brown, and B. O’Neill. Central nervous system tumors. Mayo Clinic
Proceedings, 2007.

[78] S. Bustin, V. Benes, J. Garson, J. Hellemans, J. Huggett, M. Kubista, R. Mueller,
T. Nolan, M. Pfaffl, G. Shipley, et al. The miqe guidelines: minimum information for
publication of quantitative real-time pcr experiments. Clinical chemistry, 55(4):611–
622, 2009.
[79] D. Cahill, K. Levine, R. Betensky, P. Codd, C. Romany, L. Reavie, T. Batchelor,
P. Futreal, M. Stratton, W. Curry, et al. Loss of the mismatch repair protein MSH6
in human glioblastomas is associated with tumor progression during temozolomide
treatment. Clinical cancer research, 13(7):2038, 2007.
[80] I. Camby, N. Nagy, M. Lopes, B. Schäfer, C. Maurage, M. Ruchoux, P. Murmann,
R. Pochet, C. Heizmann, J. Brotchi, et al. Supratentorial pilocytic astrocytomas,
astrocytomas, anaplastic astrocytomas and glioblastomas are characterized by a differential
expression of s100 proteins. Brain pathology, 9(1):1–19, 1999.
[81] M. Carlén, K. Meletis, C. Göritz, V. Darsalia, E. Evergren, K. Tanigaki, M. Amendola,
F. Barnabé-Heider, M. S. Y. Yeung, L. Naldini, T. Honjo, Z. Kokaia, O. Shupliakov,
R. M. Cassidy, O. Lindvall, and J. Frisén. Forebrain ependymal cells are Notchdependent
and generate neuroblasts and astrocytes after stroke. Nature neuroscience,
12(3):259–267, Mar. 2009.
[82] E. Carpenter, J. Goddard, A. Davis, T. Nguyen, and M. Capecchi. Targeted disruption
of hoxd-10 affects mouse hindlimb development. Development, 124(22):4505–4514,
1997.
[83] M. Carpenter, X. Cui, Z. Hu, J. Jackson, S. Sherman, Å. Seiger, and L. Wahlberg. In
vitro expansion of a multipotent population of human neural progenitor cells. Experimental
neurology, 158(2):265–278, 1999.
[84] A. Carracedo, A. Alimonti, and P. P. Pandolfi. PTEN Level in Tumor Suppression:
How Much Is Too Little? Cancer Research, 71(3):629–633, Feb. 2011.
[85] M. Carro, W. Lim, M. Alvarez, R. Bollo, X. Zhao, E. Snyder, E. Sulman, S. Anne,
F. Doetsch, H. Colman, et al. The transcriptional network for mesenchymal transformation
of brain tumours. Nature, 463(7279):318–325, 2009.
[86] E. Cerami, E. Demir, N. Schultz, and B. Taylor. Automated network analysis identifies
core pathways in glioblastoma. PLoS ONE, 2010.
[87] S. Certain, F. Barrat, E. Pastural, F. Le Deist, J. Goyo-Rivas, N. Jabado, M. Benkerrou,
R. Seger, E. Vilmer, G. Beullier, et al. Protein truncation test of lyst reveals
heterogenous mutations in patients with chediak-higashi syndrome. Blood, 95(3):979–
983, 2000.
[88] V. Chabottaux, S. Ricaud, L. Host, S. Blacher, A. Paye, M. Thiry, A. Garofalakis,
C. Pestourie, K. Gombert, F. Bruyere, et al. Membrane-type 4 matrix metalloproteinase
(mt4-mmp) induces lung metastasis by alteration of primary breast tumour
vascular architecture. Journal of cellular and molecular medicine, 13(9b):4002–4013,
2009.
[89] K. Chan, I. Espinosa, M. Chao, D. Wong, L. Ailles, M. Diehn, H. Gill, J. Presti,
H. Chang, M. Van De Rijn, et al. Identification, molecular characterization, clinical
prognosis, and therapeutic targeting of human bladder tumor-initiating cells. Proceedings
of the National Academy of Sciences, 106(33):14016–14021, 2009.
[90] C. Cheadle, M. Nesterova, T. Watkins, K. Barnes, J. Hall, A. Rosen, K. Becker,
and Y. Cho-Chung. Regulatory subunits of pka define an axis of cellular proliferation/differentiation
in ovarian cancer cells. BMC Medical Genomics, 1(1):43, 2008.
[91] K. Chen and N. Rajewsky. Deep conservation of microRNA-target relationships and
3’UTR motifs in vertebrates, flies, and nematodes. Cold Spring Harb Symp Quant
Biol, 71:149–56, 2006.
[92] L. Z. L. Y. Z. Q. Chen K, Luo Z. Perp gene therapy attenuates lung cancer xenograft
via inducing apoptosis and suppressing vegf. Cancer biology and therapy, 12, 2011.
[93] A. J. S. T. S. C. B.-N. S. S. J. M. R. M. R. S. S. Q. H. P. J. T. A. R. T. L. W. S.
K. C. J. B. C. M. M. G. R. H. D. Cheung KâĂŘJJ, Johnson NA. Acquired tnfrsf14
mutations in follicular lymphoma are associated with worse prognosis. Cancer Res,
70:9166–9174, 2010.

[94] F. Chibon, O. Mariani, J. Derré, A. Mairal, J. Coindre, L. Guillou, X. Sastre, F. Pédeutour,
and A. Aurias. Ask1 (map3k5) as a potential therapeutic target in malignant
fibrous histiocytomas with 12q14–q15 and 6q23 amplifications. Genes, Chromosomes
and Cancer, 40(1):32–37, 2004.
[95] S. Chigurupati, R. Venkataraman, D. Barrera, A. Naganathan, M. Madan, L. Paul,
J. Pattisapu, G. Kyriazis, K. Sugaya, S. Bushnev, et al. Receptor channel trpc6 is a
key mediator of notch-driven glioblastoma growth and invasiveness. Cancer research,
70(1):418–427, 2010.
[96] E. Chiocca. The many functions of microRNAs in glioblastoma. World neurosurgery,
2010.
[97] S. Chirasani, A. Sternjak, P. Wend, S. Momma, B. Campos, I. Herrmann, D. Graf,
T. Mitsiadis, C. Herold-Mende, D. Besser, et al. Bone morphogenetic protein-7 release
from endogenous neural precursor cells suppresses the tumourigenicity of stem-like
glioblastoma cells. Brain, 133(7):1961–1972, 2010.
[98] M. Choi, U. Scholl, W. Ji, T. Liu, I. Tikhonova, P. Zumbo, A. Nayir, A. Bakkaloğlu,
S. Özen, S. Sanjad, et al. Genetic diagnosis by whole exome capture and massively parallel
dna sequencing. Proceedings of the National Academy of Sciences, 106(45):19096–
19101, 2009.
[99] L. M. Chow and S. J. Baker. PTEN function in normal and neoplastic growth. Cancer
letters, 241(2):184–196, Sept. 2006.
[100] L. M. Chow, R. Endersby, X. Zhu, S. Rankin, C. Qu, J. Zhang, A. Broniscer, D. W.
Ellison, and S. J. Baker. Cooperativity within and among Pten, p53, and Rb pathways
induces high-grade astrocytoma in adult brain. Cancer Cell, 19(3):305–16, 2011.
[101] M. J. Clark, N. Homer, B. D. O’Connor, Z. Chen, A. Eskin, H. Lee, B. Merriman, and
S. F. Nelson. U87MG decoded: the genomic sequence of a cytogenetically aberrant
human cancer cell line. PLoS genetics, 6(1):e1000832, 2010.
[102] N. M. Cohen, E. Kenigsberg, and A. Tanay. Primate CpG islands are maintained by
heterogeneous evolutionary regimes involving minimal selection. Cell, 145(5):773–786,
May 2011.
[103] A. Collins, P. Berry, C. Hyde, M. Stower, and N. Maitland. Prospective identification
of tumorigenic prostate cancer stem cells. Cancer research, 65(23):10946, 2005.
[104] V. Collins. Amplified genes in human gliomas. In Seminars in cancer biology, volume 4,
page 27, 1993.
[105] H. Colman, L. Zhang, E. Sulman, J. McDonald, N. Shooshtari, A. Rivera, S. Popoff,
C. Nutt, D. Louis, J. Cairncross, et al. A multigene predictor of outcome in glioblastoma.
Neuro-oncology, 12(1):49–57, 2010.
[106] G. O. Consortium et al. Gene ontology: tool for the unification of biology. Nature
genetics, 25(1):25–29, 2000.
[107] L. Conti, S. M. Pollard, T. Gorba, E. Reitano, M. Toselli, G. Biella, Y. Sun, S. Sanzone,
Q.-L. Ying, E. Cattaneo, and A. Smith. Niche-Independent Symmetrical Self-Renewal
of a Mammalian Tissue Stem Cell. PLoS Biology, 3(9):e283, 2005.
[108] H. Contreras, R. Ledezma, J. Vergara, F. Cifuentes, C. Barra, P. Cabello, I. Gallegos,
B. Morales, C. Huidobro, and E. Castellón. The expression of syndecan-1 and-2
is associated with gleason score and epithelial-mesenchymal transition markers, ecadherin
and β-catenin, in prostate cancer. In Urologic Oncology: Seminars and
Original Investigations, volume 28, pages 534–540. Elsevier, 2010.
[109] A. Coutts, E. MacKenzie, E. Griffith, and D. Black. Tes is a novel focal adhesion
protein with a role in cell spreading. Journal of cell science, 116(5):897–906, 2003.
[110] Y. Cui, J. Wang, X. Zhang, R. Lang, M. Bi, L. Guo, and S. Lu. Ecrg2, a novel
candidate of tumor suppressor gene in the esophageal carcinoma, interacts directly
with metallothionein 2a and links to apoptosis* 1,* 2,* 3. Biochemical and biophysical
research communications, 302(4):904–915, 2003.
[111] M. Cully, H. You, and A. Levine. Beyond PTEN mutations: the PI3K pathway as an
integrator of multiple inputs during tumorigenesis. Nature Reviews Cancer, 2006.

[112] M. Czystowska, J. Han, M. Szczepanski, M. Szajnik, K. Quadrini, H. Brandwein,
J. Hadden, K. Signorelli, and T. Whiteside. Irx-2, a novel immunotherapeutic, protects
human t cells from tumor-induced cell death. Cell Death & Differentiation, 16(5):708–
718, 2009.
[113] C. Dang, M. Gottschling, K. Manning, E. O’Currain, S. Schneider, W. Sterry,
E. Stockfleth, and I. Nindl. Identification of dysregulated genes in cutaneous squamous
cell carcinoma. Oncology reports, 16(3):513–519, 2006.
[114] L. De Filippis, G. Lamorte, E. Snyder, A. Malgaroli, and A. Vescovi. A novel, immortal,
and multipotent human neural stem cell line generating functional neurons and
oligodendrocytes. Stem cells, 25(9):2312–2321, 2007.
[115] C. Dehay and H. Kennedy. Cell-cycle control and cortical development. Nature Reviews
Neuroscience, 8(6):438–450, 2007.
[116] L. Deleyrolle and B. Reynolds. Isolation, expansion, and differentiation of adult mammalian
neural stem and progenitor cells using the neurosphere assay. Methods Mol Biol,
549:91–101, 2009.
[117] G. Denning, B. Jean-Joseph, C. Prince, D. Durden, and P. Vogt. A short n-terminal
sequence of pten controls cytoplasmic localization and is required for suppression of
cell growth. Oncogene, 26(27):3930–3940, 2007.
[118] C. Desmet and D. Peeper. The neurotrophic receptor trkb: a drug target in anti-cancer
therapy? Cellular and molecular life sciences, 63(7):755–759, 2006.
[119] L. Desnoyers, R. Pai, R. Ferrando, K. Hötzel, T. Le, J. Ross, R. Carano, A. D’Souza,
J. Qing, I. Mohtashemi, et al. Targeting fgf19 inhibits tumor growth in colon cancer
xenograft and fgf19 transgenic hepatocellular carcinoma models. Oncogene, 27(1):85–
97, 2007.
[120] T. Di Tomaso, S. Mazzoleni, E. Wang, G. Sovena, D. Clavenna, A. Franzin, P. Mortini,
S. Ferrone, C. Doglioni, F. Marincola, et al. Immunobiological characterization
of cancer stem cells isolated from glioblastoma patients. Clinical Cancer Research,
16(3):800–813, 2010.
[121] P. Dirks. Stem cells and brain tumours. Nature, 444(7120):687–688, 2006.
[122] P. Dirks. Brain tumour stem cells: the undercurrents of human brain cancer and their
relationship to neural stem cells. Philosophical Transactions of the Royal Society B:
Biological Sciences, 363(1489):139, 2008.
[123] P. B. Dirks. Cancer: stem cells and brain tumours. Nature, 444(7120):687–8, 2006.
[124] J. G. Doench and P. A. Sharp. Specificity of microRNA target selection in translational
repression. Genes Dev, 18(5):504–11, 2004.
[125] T. Doetschman, H. Eistetter, M. Katz, W. Schmidt, and R. Kemler. The in vitro
development of blastocyst-derived embryonic stem cell lines: formation of visceral
yolk sac, blood islands and myocardium. Journal of embryology and experimental
morphology, 87(1):27, 1985.
[126] S. Dolci, A. Belmonte, R. Santone, M. Giorgi, M. Pellegrini, E. Carosa, E. Piccione,
A. Lenzi, and E. Jannini. Subcellular localization and regulation of type-1c
and type-5 phosphodiesterases. Biochemical and biophysical research communications,
341(3):837–846, 2006.
[127] G. Dominic, W. Yi-Lu, L. David, and R. Nikolaus. microrna target predictions across
seven drosophila species and comparison to mammalian targets. 2005.
[128] D. Doyle, G. Hanks, and N. Cherny. Oxford textbook of palliative medicine. Oxford
University Press, USA, 2005.
[129] T. Du. microPrimer: the biogenesis and function of microRNA. Development,
132(21):4645–4652, Sept. 2005.
[130] A. Duensing and S. Duensing. Guilt by association? p53 and the development of aneuploidy
in cancer. Biochemical and biophysical research communications, 331(3):694–
700, 2005.
[131] H. Dvinge and P. Bertone. Htqpcr: high-throughput analysis and visualization of
quantitative real-time pcr data in r. Bioinformatics, 25(24):3325–3326, 2009.

[132] D. Edwards. Non-linear normalization and background correction in one-channel cdna
microarray studies. Bioinformatics, 19(7):825, 2003.
[133] A. Efeyan and M. Serrano. p53: guardian of the genome and policeman of the oncogenes.
Cell Cycle, 6(9):1006–1010, 2007.
[134] A. J. Enright, B. John, U. Gaul, T. Tuschl, C. Sander, and D. S. Marks. MicroRNA
targets in Drosophila. Genome Biol, 5(1):R1, 2003.
[135] P. Eriksson, E. Perfilieva, T. Björk-Eriksson, A. Alborn, C. Nordborg, D. Peterson,
and F. Gage. Neurogenesis in the adult human hippocampus. Nature medicine,
4(11):1313–1317, 1998.
[136] J. Erlichman and J. Leiter. Glia modulation of the extracellular milieu as a factor in
central CO2 chemosensitivity and respiratory control. Journal of Applied Physiology,
108(6):1803, 2010.
[137] C. Esposito, M. Scrima, A. Carotenuto, A. Tedeschi, P. Rovero, G. D’Errico, A. Malfitano,
M. Bifulco, and D. Anna Maria. Structures and micelle locations of the
nonlipidated and lipidated c-terminal membrane anchor of 2’, 3’-cyclic nucleotide-
3’-phosphodiesterase. Biochemistry, 47(1):308–319, 2008.
[138] S. Falcon and R. Gentleman. Using gostats to test gene lists for go term association.
Bioinformatics, 23(2):257, 2007.
[139] K. K. Farh, A. Grimson, C. Jan, B. P. Lewis, W. K. Johnston, L. P. Lim, C. B.
Burge, and D. P. Bartel. The widespread impact of mammalian MicroRNAs on mRNA
repression and evolution. Science, 310(5755):1817–21, 2005.
[140] T. Fawcett, H. Eastman, J. Martindale, and N. Holbrook. Physical and functional
association between gadd153 and ccaat/enhancer-binding protein beta during cellular
stress. Journal of Biological Chemistry, 271(24):14285–14289, 1996.
[141] C. Fears, C. Gladson, and A. Woods. Syndecan-2 is expressed in the microvasculature
of gliomas and regulates angiogenic processes in microvascular endothelial cells.
Journal of Biological Chemistry, 281(21):14533–14536, 2006.
[142] R. Feil and F. Berger. Convergent evolution of genomic imprinting in plants and
mammals. Trends in Genetics, 23(4):192–199, 2007.
[143] B. G. Firehose. Broad gdac firehose.
[144] A. Fischer and R. Bongini. Turning müller glia into neural progenitors in the retina.
Molecular neurobiology, pages 1–11, 2010.
[145] J. Flax, S. Aurora, C. Yang, C. Simonin, A. Wills, L. Billinghurst, M. Jendoubi,
R. Sidman, J. Wolfe, S. Kim, et al. Engraftable human neural stem cells respond to
developmental cues, replace neurons, and express foreign genes. Nature biotechnology,
16:1033–1039, 1998.
[146] P. Flicek, M. Amode, D. Barrell, K. Beal, S. Brent, D. Carvalho-Silva, P. Clapham,
G. Coates, S. Fairley, S. Fitzgerald, et al. Ensembl 2012. Nucleic acids research,
40(D1):D84–D90, 2012.
[147] P. Flicek, M. Amode, D. Barrell, K. Beal, S. Brent, Y. Chen, P. Clapham, G. Coates,
S. Fairley, S. Fitzgerald, et al. Ensembl 2011. Nucleic acids research, 39(suppl 1):D800,
2011.
[148] W. Freije, F. Castro-Vargas, Z. Fang, S. Horvath, T. Cloughesy, L. Liau, P. Mischel,
and S. Nelson. Gene expression profiling of gliomas strongly predicts survival. Cancer
research, 64(18):6503, 2004.
[149] R. Fricker, M. Carpenter, C. Winkler, C. Greco, M. Gates, and A. Björklund. Sitespecific
migration and neuronal differentiation of human neural progenitor cells after
transplantation in the adult rat brain. The Journal of neuroscience, 19(14):5990, 1999.
[150] R. Friedman, K. Farh, C. Burge, and D. Bartel. Most mammalian mrnas are conserved
targets of micrornas. Genome research, 19(1):92–105, 2009.
[151] M. Frolov and N. Dyson. Molecular mechanisms of E2F-dependent activation and
pRB-mediated repression. Journal of cell science, 117(11):2173, 2004.
[152] P. Fujita, B. Rhead, A. Zweig, A. Hinrichs, D. Karolchik, M. Cline, M. Goldman,
G. Barber, H. Clawson, A. Coelho, et al. The ucsc genome browser database: update
2011. Nucleic acids research, 39(suppl 1):D876, 2011.

[153] T. Fujiwara, M. Bandi, M. Nitta, E. Ivanova, R. Bronson, and D. Pellman. Cytokinesis
failure generating tetraploids promotes tumorigenesis in p53-null cells. Nature,
437(7061):1043, 2005.
[154] F. B. Furnari, T. Fenton, R. M. Bachoo, A. Mukasa, J. M. Stommel, A. Stegh, W. C.
Hahn, K. L. Ligon, D. N. Louis, C. Brennan, L. Chin, R. A. DePinho, and W. K.
Cavenee. Malignant astrocytic glioma: genetics, biology, and paths to treatment.
Genes and development, 21(21):2683–710, 2007.
[155] D. Gaidatzis, E. Van Nimwegen, J. Hausser, and M. Zavolan. Inference of mirna
targets using evolutionary conservation and pathway analysis. BMC bioinformatics,
8(1):69, 2007.
[156] R. Galli, E. Binda, U. Orfanelli, B. Cipelletti, A. Gritti, S. De Vitis, R. Fiocco,
C. Foroni, F. Dimeco, and A. Vescovi. Isolation and characterization of tumorigenic,
stem-like neural precursors from human glioblastoma. Cancer research, 64(19):7011,
2004.
[157] G. Gallia, V. Rand, I. Siu, et al. PIK3CA gene mutations in pediatric and adult
glioblastoma multiforme. Molecular cancer research, 4(10):709, 2006.
[158] E. Garcia-Aragoncillo, J. Carrillo, E. Lalli, N. Agra, G. Gomez-Lopez, A. Pestana,
and J. Alonso. Dax1, a direct target of ews/fli1 oncoprotein, is a principal regulator
of cell-cycle progression in ewing’s tumor cells. Oncogene, 27(46):6034–6043, 2008.
[159] M. Gardiner-Garden and M. Frommer. CpG islands in vertebrate genomes. Journal
of molecular biology, 196(2):261–282, July 1987.
[160] A. Gartel and S. Radhakrishnan. Lost in transcription: p21 repression, mechanisms,
and consequences. Cancer research, 65(10):3980, 2005.
[161] L. Gautier, L. Cope, B. Bolstad, and R. Irizarry. affyÂŮanalysis of affymetrix genechip
data at the probe level. Bioinformatics, 20(3):307–315, 2004.
[162] GBMbase. A bioinformatics resource for glioblastoma multiforme.
http://beta.gbmbase.org/page/welcome/display.
[163] A. Giganti, J. Plastino, B. Janji, M. Van Troys, D. Lentz, C. Ampe, C. Sykes, and
E. Friederich. Actin-filament cross-linking protein t-plastin increases arp2/3-mediated
actin-based movement. Journal of cell science, 118(6):1255, 2005.
[164] A. Girard, R. Sachidanandam, G. J. Hannon, and M. A. Carmell. A germline-specific
class of small RNAs binds mammalian Piwi proteins. Nature, 442(7099):199–202,
2006.
[165] T. Glaser, S. M. Pollard, A. Smith, and O. Brüstle. Tripotential Differentiation of
Adherently Expandable Neural Stem (NS) Cells. PLoS ONE, 2(3):e298, Mar. 2007.
[166] V. Gocheva and J. Joyce. Cysteine cathepsins and the cutting edge of cancer invasion.
Cell cycle, 6(1):60–64, 2007.
[167] V. Gocheva, W. Zeng, D. Ke, D. Klimstra, T. Reinheckel, C. Peters, D. Hanahan,
and J. Joyce. Distinct roles for cysteine cathepsin genes in multistage tumorigenesis.
Genes & development, 20(5):543–556, 2006.
[168] K. Goh, W. Poon, D. Chan, and C. Ip. Tissue plasminogen activator expression in
meningiomas and glioblastomas. Clinical neurology and neurosurgery, 107(4):296–300,
2005.
[169] S. Gomez-Lopez, O. Wiskow, R. Favaro, S. Nicolis, D. Price, S. Pollard, and S. A.
Sox2 and pax6 maintain the proliferative and developmental potential of gliogenic
neural stem cells in vitro. Glia, 59:1588–1599, 2011.
[170] M. Göransson, M. Andersson, C. Forni, A. Ståhlberg, C. Andersson, A. Olofsson,
R. Mantovani, and P. Åman. The myxoid liposarcoma fus-ddit3 fusion oncoprotein
deregulates nf-κb target genes by interaction with nfkbiz. Oncogene, 28(2):270–278,
2008.
[171] E. Gould, A. Reeves, M. Graziano, and C. Gross. Neurogenesis in the neocortex of
adult primates. Science, 286(5439):548, 1999.
[172] A. Gourine and S. Kasparov. Astrocytes as brain interoceptors. Experimental Physiology,
96(4):411, 2011.

[173] A. Gourine, V. Kasymov, N. Marina, F. Tang, M. Figueiredo, S. Lane,
A. Teschemacher, K. Spyer, K. Deisseroth, and S. Kasparov. Astrocytes control
breathing through pH-dependent release of ATP. Science, 329(5991):571, 2010.
[174] M. Graeber, C. Tran, P. Wolz, R. Egensperger, S. Kösel, Y. Imai, K. Bise, S. Kohsaka,
and P. Mehraein. Differential expression of mhc class ii molecules by microglia and
neoplastic astroglia: relevance for the escape of astrocytoma cells from immune surveillance.
Neuropathology and applied neurobiology, 24:293–301, 1998.
[175] B. M. H. H. G. M. Graff L, Castrop F. Expression of vesicular monoamine transporters,
synaptosomalâĂŘassociated protein 25 and syntaxin1: a signature of human small cell
lung carcinoma. Cancer Res, 61:2138–2144, 2001.
[176] L. Gravendeel, M. Kouwenhoven, O. Gevaert, J. de Rooi, A. Stubbs, J. Duijm, A. Daemen,
F. Bleeker, L. Bralten, N. Kloosterhof, et al. Intrinsic gene expression profiles of
gliomas are a better predictor of survival than histology. Cancer research, 69(23):9065–
9072, 2009.
[177] A. Gregorieff and H. Clevers. Wnt signaling in the intestinal epithelium: from endoderm
to cancer. Genes & development, 19(8):877–890, 2005.
[178] C. Gregorio-King, J. McLeod, F. Collier, G. Collier, K. Bolton, G. Van Der Meer,
J. Apostolopoulos, and M. Kirkland. Merp1: a mammalian ependymin-related protein
gene differentially expressed in hematopoietic cells. Gene, 286(2):249–257, 2002.
[179] S. Griffiths-Jones, H. K. Saini, S. van Dongen, and A. J. Enright. MiRBase: tools for
microRNA genomics. Nucleic Acids Res, 36(Database issue):D154–8, 2008.
[180] A. Grimson, K. K. Farh, W. K. Johnston, P. Garrett-Engele, L. P. Lim, and D. P. Bartel.
MicroRNA targeting specificity in mammals: determinants beyond seed pairing.
Mol Cell, 27(1):91–105, 2007.
[181] J. Gunnersen, V. Spirkoska, P. Smith, R. Danks, and S. Tan. Growth and migration
markers of rat c6 glioma cells identified by serial analysis of gene expression. Glia,
32(2):146–154, 2000.
[182] P. Gupta, C. Chaffer, and R. Weinberg. Cancer stem cells: mirage or reality? Nature
medicine, 15(9):1010–1012, 2009.
[183] C. Hagemann, J. Anacker, S. Haas, D. Riesner, B. Schömig, R. Ernestus, and G. Vince.
Comparative expression pattern of matrix-metalloproteinases in human glioblastoma
cell-lines and primary cultures. BMC research notes, 3(1):293, 2010.
[184] I. Han, H. Park, and E. Oh. New insights into syndecan-2 expression and tumourigenic
activity in colon carcinoma cells. Journal of Molecular Histology, 35(3):319–326, 2004.
[185] K. L. Harms and X. Chen. The C terminus of p53 family proteins is a cell fate
determinant. Molecular and cellular biology, 25(5):2014–2030, Mar. 2005.
[186] M. Harris, H. Yang, B. Low, J. Mukherje, A. Guha, R. Bronson, L. Shultz, M. Israel,
and K. Yun. Cancer stem cells are enriched in the side population cells in a mouse
model of glioma. Cancer research, 68(24):10051–10059, 2008.
[187] E. Hartfuss, R. Galli, N. Heins, and M. Götz. Characterization of cns precursor
subtypes and radial glia. Developmental biology, 229(1):15–30, 2001.
[188] R. Haviland, S. Eschrich, G. Bloom, Y. Ma, S. Minton, R. Jove, and W. Cress.
Necdin, a negative growth regulator, is a novel stat3 target gene down-regulated in
human cancer. PloS one, 6(10):e24923, 2011.
[189] Y. Hayakawa, Y. Hirata, H. Nakagawa, K. Sakamoto, Y. Hikiba, H. Kinoshita,
W. Nakata, R. Takahashi, K. Tateishi, M. Tada, et al. Apoptosis signal-regulating
kinase 1 and cyclin d1 compose a positive feedback loop contributing to tumor growth
in gastric cancer. Proceedings of the National Academy of Sciences, 108(2):780–785,
2011.
[190] L. He, C. Fan, A. Kapoor, A. Ingram, A. Rybak, R. Austin, J. Dickhout, J. Cutz,
J. Scholey, and D. Tang. [alpha]-mannosidase 2c1 attenuates pten function in prostate
cancer cells. Nature Communications, 2:307, 2011.
[191] C. Heldin, A. Ostman, A. Eriksson, A. Siegbahn, L. Claesson-Welsh, and B. Westermark.
Platelet-derived growth factor: Isoform-specific signalling via heterodimeric or
homodimeric receptor complexes. Kidney Int, 41(3):571–574, 1992.

[192] M. Hernandez, M. Nieto, and M. Sanchez Crespo. Cytosolic phospholipase a2 and
the distinct transcriptional programs of astrocytoma cells. Trends in neurosciences,
23(6):259–264, 2000.
[193] K. Herrup and Y. Yang. Cell cycle regulation in the postmitotic neuron: oxymoron
or new biology? Nature Reviews Neuroscience, 8(5):368–378, 2007.
[194] M. S. Hestand, A. Klingenhoff, M. Scherf, Y. Ariyurek, Y. Ramos, W. van Workum,
M. Suzuki, T. Werner, G. J. van Ommen, J. T. den Dunnen, M. Harbers, and P. A.
t Hoen. Tissue-specific transcript annotation and expression profiling with complementary
next-generation sequencing technologies. Nucleic acids research, 38(16):e165,
2010.
[195] S. Hockfield and R. McKay. Identification of major cell classes in the developing
mammalian nervous system. The Journal of neuroscience, 5(12):3310, 1985.
[196] J. Hodgson, R. Yeh, A. Ray, N. Wang, I. Smirnov, M. Yu, S. Hariono, J. Silber,
H. Feiler, J. Gray, et al. Comparative analyses of gene copy number and mrna expression
in glioblastoma multiforme tumors and xenografts. Neuro-oncology, 11(5):477–
487, 2009.
[197] R. Hoffmann and A. Valencia. A gene network for navigating the literature. Nature
genetics, 36(7):664–664, 2004.
[198] L. Hook, J. Vives, N. Fulton, M. Leveridge, S. Lingard, M. Bootman, A. Falk, S. Pollard,
T. Allsopp, D. Dalma-Weiszhausz, et al. Non-immortalized human neural stem
(ns) cells as a scalable platform for cellular assays. Neurochemistry international, 2011.
[199] S. Houwing, L. Kamminga, E. Berezikov, D. Cronembold, A. Girard, H. van den Elst,
D. Filippov, H. Blaser, E. Raz, C. Moens, et al. A role for piwi and pirnas in germ
cell maintenance and transposon silencing in zebrafish. Cell, 129(1):69–82, 2007.
[200] S. Huang, L. Shu, M. Dilling, J. Easton, F. Harwood, H. Ichijo, and P. Houghton. Sustained
activation of the jnk cascade and rapamycin-induced apoptosis are suppressed
by p53/p21cip1. Molecular cell, 11(6):1491–1501, 2003.
[201] X. Huang, D. Xiao, L. Xu, H. Zhong, L. Liao, Z. Xie, and E. Li. Prognostic significance
of altered expression of sdc2 and cyr61 in esophageal squamous cell carcinoma.
Oncology reports, 21(4):1123, 2009.
[202] Z. Huang, Y. Kawase-Koga, S. Zhang, J. Visvader, M. Toth, C. Walsh, and T. Sun.
Transcription factor lmo4 defines the shape of functional areas in developing cortices
and regulates sensorimotor control. Developmental biology, 327(1):132–142, 2009.
[203] HUGO. Gene nomenclature committee. http://www.genenames.org/.
[204] C. Hunter, R. Smith, D. Cahill, P. Stephens, C. Stevens, J. Teague, C. Greenman,
S. Edkins, G. Bignell, H. Davies, et al. A hypermutation phenotype and somatic MSH6
mutations in recurrent human malignant gliomas after alkylator chemotherapy. Cancer
research, 66(8):3987, 2006.
[205] E. B. Institute. Intact database. http://www.ebi.ac.uk/intact/main.xhtml.
[206] N. C. Institute. A to z list of cancers.
[207] N. Ishii, D. Maier, A. Merlo, M. Tada, Y. Sawamura, A. Diserens, and E. Meir.
Frequent co-alterations of TP53, p16/CDKN2A, p14ARF, PTEN tumor suppressor
genes in human glioma cell lines. Brain pathology, 9(3):469–479, 1999.
[208] K. Jabbari and G. Bernardi. Cytosine methylation and CpG, TpG, (CpA) and TpA
frequencies. Gene, 333:143–149, 2004.
[209] P. Jay, C. Rougeulle, A. Massacrier, A. Moncla, M. Mattel, P. Malzac, N. Roëckel,
S. Taviaux, J. Lefranc, P. Cau, et al. The human necdin gene, ndn, is maternally imprinted
and located in the prader-willi syndrome chromosomal region. Nature genetics,
17(3):357–361, 1997.
[210] J. Jiao and D. Chen. Induction of neurogenesis in non-conventional neurogenic regions
of the adult cns by niche astrocyte-produced signals. Stem Cells, pages 2007–0513v1,
2008.
[211] F. Johansson, H. Göransson, and B. Westermark. Expression analysis of genes involved
in brain tumor progression driven by retroviral insertional mutagenesis in mice.
Oncogene, 24(24):3896–3905, 2005.

[212] K. Johe, T. Hazel, T. Muller, M. Dugich-Djordjevic, and R. McKay. Single factors
direct the differentiation of stem cells from the fetal and adult central nervous system.
Genes & development, 10(24):3129, 1996.
[213] B. John, A. J. Enright, A. Aravin, T. Tuschl, C. Sander, and D. S. Marks. Human
MicroRNA targets. PLoS Biol, 2(11):e363, 2004.
[214] P. Jones and P. Laird. Cancer-epigenetics comes of age. Nature genetics, 21(2):163–
167, 1999.
[215] A. Kanawaty and J. Henderson. Genomic analysis of induced pluripotent stem (ips)
cells: routes to reprogramming. Bioessays, 31(2):134–138, 2009.
[216] Y. Kang, I. Kim, E. Kim, M. Yoon, S. Kim, T. Kwon, and K. Choi. Paxilline enhances
trail-mediated apoptosis of glioma cells via modulation of c-flip, survivin and dr5.
Experimental & molecular medicine, 43(1):24, 2011.
[217] Y. Kang, I. Kim, E. Kim, M. Yoon, S. Kim, T. Kwon, and K. Choi. Paxilline enhances
trail-mediated apoptosis of glioma cells via modulation of c-flip, survivin and dr5.
Experimental & molecular medicine, 43(1):24–34, 2011.
[218] A. Kaul and W. Maltese. Killing of cancer cells by the photoactivatable protein kinase
c inhibitor, calphostin c, involves induction of endoplasmic reticulum stress. Neoplasia
(New York, NY), 11(9):823, 2009.
[219] B. Kaur, F. Khwaja, E. Severson, S. Matheny, D. Brat, and E. Van Meir. Hypoxia
and the hypoxia-inducible-factor pathway in glioma growth and angiogenesis. Neurooncology,
7(2):134, 2005.
[220] M. Kawashima, K. Doh-ura, E. Mekada, M. Fukui, and T. Iwaki. Cd9 expression
in solid non-neuroepithelial tumors and infiltrative astrocytic tumors. Journal of
Histochemistry & Cytochemistry, 50(9):1195, 2002.
[221] M. Kawashima, K. Doh-ura, E. Mekada, M. Fukui, and T. Iwaki. Cd9 expression
in solid non-neuroepithelial tumors and infiltrative astrocytic tumors. Journal of
Histochemistry & Cytochemistry, 50(9):1195–1203, 2002.
[222] M. Kedde, M. J. Strasser, B. Boldajipour, J. A. Oude Vrielink, K. Slanchev, C. le Sage,
R. Nagel, P. M. Voorhoeve, J. van Duijse, U. A. Orom, A. H. Lund, A. Perrakis, E. Raz,
and R. Agami. RNA-binding protein Dnd1 inhibits microRNA access to target mRNA.
Cell, 131(7):1273–86, 2007.
[223] G. Kempermann, S. Jessberger, B. Steiner, and G. Kronenberg. Milestones of neuronal
development in the adult hippocampus. Trends in Neurosciences, 27(8):447–452, 2004.
[224] M. Kertesz, N. Iovino, U. Unnerstall, U. Gaul, and E. Segal. The role of site accessibility
in microrna target recognition. Nature genetics, 39(10):1278–1284, 2007.
[225] I. Kil, S. Kim, S. Lee, and J. Park. Small interfering RNA-mediated silencing of
mitochondrial NADP + -dependent isocitrate dehydrogenase enhances the sensitivity
of HeLa cells toward tumor necrosis factor-α and anticancer drugs. Free Radical
Biology and Medicine, 43(8):1197–1207, 2007.
[226] D. Kim, C. H. Kim, J. I. Moon, Y. G. Chung, M. Y. Chang, B. S. Han, S. Ko, E. Yang,
K. Y. Cha, R. Lanza, and K. S. Kim. Generation of human induced pluripotent stem
cells by direct delivery of reprogramming proteins. Cell Stem Cell, 4(6):472–6, 2009.
[227] I. Kim, Y. Kang, M. Yoon, E. Kim, S. Kim, T. Kwon, I. Kim, and K. Choi. Amiodarone
sensitizes human glioma cells but not astrocytes to trail-induced apoptosis via chopmediated
dr5 upregulation. Neuro-oncology, 13(3):267–279, 2011.
[228] J. Kim, J. Choi, S. Lim, O. Kwon, J. Seo, S. Ryu, and P. Suh. Phospholipase c-eta 1 is
activated by intracellular ca2+ mobilization and enhances gpcrs/plc/ca2+ signaling.
Cellular Signalling, 2011.
[229] S. Kim, M. Seong, B. Jeon, H. Ko, J. Kim, and S. Park. Phase analysis identifies
compound heterozygous deletions of the PARK2 gene in patients with early-onset
Parkinson disease. Clinical Genetics.
[230] T. Kim, M. Hemberg, J. Gray, A. Costa, D. Bear, J. Wu, D. Harmin, M. Laptewicz,
K. Barbara-Haley, S. Kuersten, et al. Widespread transcription at neuronal activityregulated
enhancers. Nature, 465(7295):182–187, 2010.

[231] T.-M. Kim, W. Huang, R. Park, P. J. Park, and M. D. Johnson. A developmental
taxonomy of glioblastoma defined and maintained by MicroRNAs. Cancer Research,
71(9):3387–3399, May 2011.
[232] V. N. Kim. MicroRNA biogenesis: coordinated cropping and dicing. Nature Reviews
Molecular Cell Biology, 6(5):376–385, May 2005.
[233] W. Kim and N. Sharpless. The regulation of INK4/ARF in cancer and aging. Cell,
127(2):265–275, 2006.
[234] R. Kincaid, A. Kuchinsky, and M. Creech. Vistaclara: an expression browser plug-in
for cytoscape. Bioinformatics, 24(18):2112–2114, 2008.
[235] S. Kinney, D. Smiraglia, S. James, M. Moser, B. Foster, and A. Karpf. Stage-specific
alterations of dna methyltransferase expression, dna hypermethylation, and dna hypomethylation
during prostate cancer progression in the transgenic adenocarcinoma
of mouse prostate model. Molecular Cancer Research, 6(8):1365–1374, 2008.
[236] L. S. Kinsey M, Smith R. Nr0b1 is required for the oncogenic phenotype mediated by
ews/fli in ewing’s sarcoma. Mol Cancer Res, 4:851–859, 2006.
[237] M. Kiriakidou, P. T. Nelson, A. Kouranov, P. Fitziev, C. Bouyioukos, Z. Mourelatos,
and A. Hatzigeorgiou. A combined computational-experimental approach predicts
human microRNA targets. Genes Dev, 18(10):1165–78, 2004.
[238] M. Kitano, M. Nakaya, T. Nakamura, S. Nagata, and M. Matsuda. Imaging of rab5 activity
identifies essential regulators for phagosome maturation. Nature, 453(7192):241–
245, 2008.
[239] C. Klattenhoff and W. Theurkauf. Biogenesis and germline functions of piRNAs.
Development, 135(1):3–9, 2008.
[240] A. Klein and B. Simons. Universal patterns of stem cell fate in cycling adult tissues.
Development, 138(15):3103, 2011.
[241] T. Kolesnikova, A. Kazarov, M. Lemieux, M. Lafleur, S. Kesari, A. Kung, and M. Hemler.
Glioblastoma inhibition by cell surface immunoglobulin protein ewi-2, in vitro and
in vivo. Neoplasia (New York, NY), 11(1):77, 2009.
[242] T. Kondo, T. Setoguchi, and T. Taga. Persistence of a small subpopulation of cancer
stem-like cells in the c6 glioma cell line. Proceedings of the National Academy of
Sciences of the United States of America, 101(3):781, 2004.
[243] K. Kondoh, N. Tsuji, C. Kamagata, M. Sasaki, D. Kobayashi, A. Yagihashi, and
N. Watanabe. A novel aspartic protease gene, alp56, is up-regulated in human breast
cancer independently from the cathepsin d gene. Breast cancer research and treatment,
78(1):37–44, 2003.
[244] A. Korshunov, R. Sycheva, and A. Golanov. Genetically distinct and clinically relevant
subtypes of glioblastoma defined by array-based comparative genomic hybridization
(array-cgh). Acta neuropathologica, 111(5):465–474, 2006.
[245] D. Koul, R. Shen, S. Bergh, Y. Lu, J. de Groot, T. Liu, G. Mills, and W. Yung.
Targeting integrin-linked kinase inhibits Akt signaling pathways and decreases tumor
progression of human glioblastoma. Molecular cancer therapeutics, 4(11):1681, 2005.
[246] C. R. E. C. B. M. C. D. Kourtidis A, Jain R. An rna interference screen identifies
metabolic regulators nr1d1 and pbp as novel survival factors for breast cancer cells
with the erbb2 signature. Cancer Res, 70:1783–1792, 2010.
[247] A. Krek, D. Grün, M. Poy, R. Wolf, L. Rosenberg, E. Epstein, P. MacMenamin,
I. Da Piedade, K. Gunsalus, M. Stoffel, et al. Combinatorial microrna target predictions.
Nature genetics, 37(5):495–500, 2005.
[248] A. Kriegstein and A. Alvarez-Buylla. The glial nature of embryonic and adult neural
stem cells. Annual review of neuroscience, 32:149, 2009.
[249] A. Kriegstein and M. Götz. Radial glia diversity: a matter of cell fate. Glia, 43(1):37–
43, 2003.
[250] R. Kroes, G. Dawson, and J. Moskal. Focused microarray analysis of glyco-gene
expression in human glioblastomas. Journal of neurochemistry, 103(s1):14–24, 2007.

[251] R. Kroes, H. He, M. Emmett, C. Nilsson, F. Leach, I. Amster, A. Marshall,
and J. Moskal. Overexpression of st6galnacv, a ganglioside-specific alpha 2, 6sialyltransferase,
inhibits glioma growth in vivo. Proceedings of the National Academy
of Sciences, 107(28):12646, 2010.
[252] T. Kulikova, P. Aldebert, N. Althorpe, W. Baker, K. Bates, P. Browne, A. Van
Den Broek, G. Cochrane, K. Duggan, R. Eberhardt, et al. The embl nucleotide
sequence database. Nucleic Acids Research, 32(suppl 1):D27–D30, 2004.
[253] M. Lafleur, D. Xu, and M. Hemler. Tetraspanin proteins regulate membrane type-1
matrix metalloproteinase-dependent pericellular proteolysis. Molecular Biology of the
Cell, 20(7):2030–2040, 2009.
[254] J. Lai, D. Sandhu, C. Yu, T. Han, C. Moser, K. Jackson, R. Guerrero, I. Aderca,
H. Isomoto, M. Garrity-Park, et al. Sulfatase 2 up-regulates glypican 3, promotes
fibroblast growth factor signaling, and decreases survival in hepatocellular carcinoma.
Hepatology, 47(4):1211–1222, 2008.
[255] Y. Lam, E. di Tomaso, H.-K. Ng, J. Pang, M. Roussel, and N. Hjelm. Expression of
p19 ink4d, cdk4, cdk6 in glioblastoma multiforme. British journal of neurosurgery,
14(1):28–32, 2000.
[256] P. Landgraf, M. Rusu, R. Sheridan, A. Sewer, N. Iovino, A. Aravin, S. Pfeffer, A. Rice,
A. Kamphorst, M. Landthaler, et al. A mammalian microrna expression atlas based
on small rna library sequencing. Cell, 129(7):1401–1414, 2007.
[257] B. Langmead, C. Trapnell, M. Pop, and S. Salzberg. Ultrafast and memory-efficient
alignment of short dna sequences to the human genome. Genome Biol, 10(3):R25,
2009.
[258] T. Lassmann, Y. Hayashizaki, and C. Daub. TagdustÂŮa program to eliminate artifacts
from next generation sequencing data. Bioinformatics, 25(21):2839, 2009.
[259] S. Lawler and E. Chiocca. Emerging functions of micrornas in glioblastoma. Journal
of neuro-oncology, 92(3):297–306, 2009.
[260] D. Lawson, M. Harrison, and C. Shapland. Fibroblast transgelin and smooth muscle
sm22α are the same protein, the expression of which is down-regulated in many cell
lines. Cell motility and the cytoskeleton, 38(3):250–257, 1998.
[261] J. Lee, S. Kotliarova, Y. Kotliarov, A. Li, Q. Su, N. M. Donin, S. Pastorino, B. W.
Purow, N. Christopher, W. Zhang, J. K. Park, and H. A. Fine. Tumor stem cells
derived from glioblastomas cultured in bFGF and EGF more closely mirror the phenotype
and genotype of primary tumors than do serum-cultured cell lines. Cancer
Cell, 9(5):391–403, May 2006.
[262] J. Lee, I. Vivanco, R. Beroukhim, J. Huang, W. Feng, R. DeBiasi, K. Yoshimoto,
J. King, P. Nghiemphu, Y. Yuza, et al. Epidermal growth factor receptor activation
in glioblastoma through novel missense mutations in the extracellular domain. PLoS
medicine, 3(12):e485, 2006.
[263] K. Lee, W. Han, J. Kim, I. Shin, E. Ko, I. Park, D. Lee, K. Oh, and D. Noh. The
cd49d+/high subpopulation from isolated human breast sarcoma spheres possesses
tumor-initiating ability. International journal of oncology, 40(3):665, 2012.
[264] R. Lee, R. Feinbaum, and V. Ambros. The c. elegans heterochronic gene lin-4 encodes
small rnas with antisense complementarity to lin-14. Cell, 75(5):843–854, 1993.
[265] S. Lee, N. Syed, J. Taylor, P. Smith, B. Griffin, M. Baens, M. Bai, K. Bourantas,
J. Stebbing, K. Naresh, et al. Dusp16 is an epigenetically regulated determinant of
jnk signalling in burkitt’s lymphoma. British journal of cancer, 103(2):265–274, 2010.
[266] Y. S. Lee, S. Pressman, A. P. Andress, K. Kim, J. L. White, J. J. Cassidy, X. Li,
K. Lubell, H. Lim do, I. S. Cho, K. Nakahara, J. B. Preall, P. Bellare, E. J. Sontheimer,
and R. W. Carthew. Silencing by small RNAs is linked to endosomal trafficking. Nat
Cell Biol, 11(9):1150–6, 2009.
[267] S. J. Leevers, B. Vanhaesebroeck, and M. D. Waterfield. Signalling through phosphoinositide
3-kinases: the lipids take centre stage. Current opinion in cell biology,
11(2):219–225, Apr. 1999.

[268] H. Lemjabbar-Alaoui, A. van Zante, M. Singer, Q. Xue, Y. Wang, D. Tsay, B. He,
D. Jablons, and S. Rosen. Sulf-2, a heparan sulfate endosulfatase, promotes human
lung carcinogenesis. Oncogene, 29(5):635–646, 2009.
[269] G. Lemke. Glial control of neuronal development. Annual review of neuroscience,
24:87–105, 2001.
[270] C. Leonetti, A. Biroccio, G. Graziani, and L. Tentori. Targeted therapy for brain
tumours: Role of parp inhibitors. Current cancer drug targets, 12(3):218, 2012.
[271] F. Leprêtre, C. Villenet, S. Quief, O. Nibourel, C. Jacquemin, X. Troussard, F. Jardin,
F. Gibson, J. Kerckaert, C. Roumier, et al. Waved acgh: to smooth or not to smooth.
Nucleic acids research, 38(7):e94–e94, 2010.
[272] B. Lewis, C. Burge, and D. Bartel. Conserved seed pairing, often flanked by adenosines,
indicates that thousands of human genes are microrna targets. Cell, 120(1):15–20,
2005.
[273] B. P. Lewis, I. H. Shih, M. W. Jones-Rhoades, D. P. Bartel, and C. B. Burge. Prediction
of mammalian microRNA targets. Cell, 115(7):787–98, 2003.
[274] M. Lewis, S. Ross, P. Strickland, C. Snyder, and C. Daniel. Regulated expression
patterns of irx-2, an iroquois-class homeobox gene, in the human breast. Cell and
tissue research, 296(3):549–554, 1999.
[275] A. Li, J. Walling, Y. Kotliarov, A. Center, M. E. Steed, S. J. Ahn, M. Rosenblum,
T. Mikkelsen, J. C. Zenklusen, and H. A. Fine. Genomic changes and gene expression
profiles reveal that established glioma cell lines are poorly representative of primary
human gliomas. Molecular cancer research : MCR, 6(1):21–30, 2008.
[276] M. Li, L. Pevny, R. Lovell-Badge, and A. Smith. Generation of purified neural precursors
from embryonic stem cells by lineage selection. Current biology, 8(17):971–974,
1998.
[277] Q. Li, A. Jedlicka, N. Ahuja, M. Gibbons, S. Baylin, P. Burger, J. Issa, et al. Concordant
methylation of the er and n33 genes in glioblastoma multiforme. Oncogene,
16(24):3197, 1998.
[278] Y. Li, F. Guessous, Y. Zhang, C. DiPierro, B. Kefas, E. Johnson, L. Marcinkiewicz,
J. Jiang, Y. Yang, T. Schmittgen, et al. Microrna-34a inhibits glioblastoma growth
by targeting multiple oncogenes. Cancer research, 69(19):7569, 2009.
[279] S. Liebner, C. Czupalla, and H. Wolburg. current concepts of blood-brain barrier
development. The International journal of developmental biology, 2011.
[280] A. Liekens, J. De Knijf, W. Daelemans, B. Goethals, P. De Rijk, and J. Del-Favero.
Biograph: unsupervised biomedical knowledge discovery via automated hypothesis
generation. Genome biology, 12(6):R57, 2011.
[281] K. Ligon, J. Alberta, A. Kho, J. Weiss, M. Kwaan, C. Nutt, D. Louis, C. Stiles,
and D. Rowitch. The oligodendroglial lineage marker OLIG2 is universally expressed
in diffuse gliomas. Journal of Neuropathology & Experimental Neurology, 63(5):499,
2004.
[282] L. P. Lim, M. E. Glasner, S. Yekta, C. B. Burge, and D. P. Bartel. Vertebrate
microRNA genes. Science, 299(5612):1540, 2003.
[283] L. P. Lim, N. C. Lau, P. Garrett-Engele, A. Grimson, J. M. Schelter, J. Castle,
D. P. Bartel, P. S. Linsley, and J. M. Johnson. Microarray analysis shows that some
microRNAs downregulate large numbers of target mRNAs. Nature, 433(7027):769–73,
2005.
[284] L. Linares, A. Hengstermann, A. Ciechanover, S. Müller, and M. Scheffner. Hdmx
stimulates Hdm2-mediated ubiquitination and degradation of p53. Proceedings of the
National Academy of Sciences, 100(21):12009, 2003.
[285] A. Linkous, E. Yazlovitskaya, and D. Hallahan. Cytosolic phospholipase a2 and
lysophospholipids in tumor angiogenesis. Journal of the National Cancer Institute,
102(18):1398–1412, 2010.
[286] M. M. Lino and A. Merlo. PI3Kinase signaling in glioblastoma. Journal of neurooncology,
103(3):417–427, July 2011.

[287] R. Lister, M. Pelizzola, R. Dowen, R. Hawkins, G. Hon, J. Tonti-Filippini, J. Nery,
L. Lee, Z. Ye, Q. Ngo, et al. Human DNA methylomes at base resolution show
widespread epigenomic differences. Nature, 462(7271):315–322, 2009.
[288] B. Liu, Y. Kim, V. Leatherberry, P. Cowin, and C. Alexander. Mammary gland development
requires syndecan-1 to create a β-catenin/tcf-responsive mammary epithelial
subpopulation. Oncogene, 22(58):9243–9253, 2003.
[289] B. Liu, S. McDermott, S. Khwaja, and C. Alexander. The transforming activity of
wnt effectors correlates with their ability to induce the accumulation of mammary
progenitor cells. Proceedings of the National Academy of Sciences of the United States
of America, 101(12):4158, 2004.
[290] F. Liu, Y. You, X. Li, T. Ma, Y. Nie, B. Wei, T. Li, H. Lin, and Z. Yang. Brain injury
does not alter the intrinsic differentiation potential of adult neuroblasts. The Journal
of Neuroscience, 29(16):5075, 2009.
[291] J. Liu, M. A. Valencia-Sanchez, G. J. Hannon, and R. Parker. MicroRNA-dependent
localization of targeted mRNAs to mammalian P-bodies. Nat Cell Biol, 7(7):719–23,
2005.
[292] W. Liu, Y. Fu, S. Xu, F. Ding, G. Zhao, K. Zhang, C. Du, B. Pang, and Q. Pang.
c-Met expression is associated with time to recurrence in patients with glioblastoma
multiforme. Journal of clinical neuroscience : official journal of the Neurosurgical
Society of Australasia, 18(1):119–121, 2011.
[293] X. Liu, A. Bolteus, D. Balkin, O. Henschel, and A. Bordey. Gfap-expressing cells
in the postnatal subventricular zone display a unique glial phenotype intermediate
between radial glia and astrocytes. Glia, 54(5):394–410, 2006.
[294] X. Liu, M. Cicek, S. Plummer, E. Jorgenson, G. Casey, and J. Witte. Association
of testis derived transcript gene variants and prostate cancer risk. The Journal of
urology, 177(3):894–898, 2007.
[295] Y. Liu, S. Shete, C. Etzel, M. Scheurer, G. Alexiou, G. Armstrong, S. Tsavachidis,
F. Liang, M. Gilbert, K. Aldape, et al. Polymorphisms of lig4, btbd2, hmga2, and
rtel1 genes involved in the double-strand break repair pathway predict glioblastoma
survival. Journal of Clinical Oncology, 28(14):2467–2474, 2010.
[296] D. Q. S. P. C. K. L. O. T. J. L. J. K. V. C. T. M. P. L. T. L. L. N. S. T. C.-L.
Liu Q, Nguyen DH. Molecular properties of cd133+ glioblastoma stem cells derived
from treatment-refractory recurrent brain tumors. J Neurooncol, 94:1–19, 2009.
[297] C. Lois and A. Alvarez-Buylla. Proliferating subventricular zone cells in the adult
mammalian forebrain can differentiate into neurons and glia. Proceedings of the National
Academy of Sciences, 90(5):2074, 1993.
[298] B. Lorber, A. Guidi, J. Fawcett, and K. Martin. Activated retinal glia mediated axon
regeneration in experimental glaucoma. Neurobiology of Disease, 2011.
[299] C. Lottaz, D. Beier, K. Meyer, P. Kumar, A. Hermann, J. Schwarz, M. Junker,
P. Oefner, U. Bogdahn, J. Wischhusen, et al. Transcriptional profiles of cd133+
and cd133- glioblastoma-derived cancer stem cell lines suggest different cells of origin.
Cancer research, 70(5):2030–2040, 2010.
[300] D. Louis, H. Ohgaki, and O. Wiestler. The 2007 WHO classification of tumours of
the central nervous system. Acta Neuropathologica, 2007.
[301] N. Louis. WHO Classification of Tumours of the Central Nervous System. International
Agency for Research on Cancer, 2007.
[302] S. Lowell, A. Benchoua, B. Heavey, and A. Smith. Notch promotes neural lineage
entry by pluripotent embryonic stem cells. PLoS biology, 4(5):e121, 2006.
[303] W. Ma, T. Tavakoli, E. Derby, Y. Serebryakova, M. Rao, and M. Mattson. Cellextracellular
matrix interactions regulate neural differentiation of human embryonic
stem cells. BMC developmental biology, 8(1):90, 2008.
[304] Y.-H. Ma, R. Mentlein, F. Knerlich, M.-L. Kruse, H. M. Mehdorn, and J. Held-
Feindt. Expression of stem cell markers in human astrocytomas of different WHO
grades. Journal of neuro-oncology, 86(1):31–45, 2008.

[305] K. MAEDA, S. MATSUHASHI, K. TABUCHI, T. WATANABE, T. KATAGIRI,
M. OYASU, N. SAITO, and S. KURODA. Brain specific human genes, nell1 and
nell2, are predominantly expressed in neuroblastoma and other embryonal neuroepithelial
tumors. Neurologia medico-chirurgica, 41(12):582–589, 2001.
[306] H. Maier, C. Jones, B. Jasani, D. Öfner, B. Zelger, K. Schmid, and H. Budka. Metallothionein
overexpression in human brain tumours. Acta neuropathologica, 94(6):599–
604, 1997.
[307] S. Maira, I. Galetic, D. Brazil, S. Kaech, E. Ingley, M. Thelen, and B. Hemmings.
Carboxyl-terminal modulator protein (ctmp), a negative regulator of pkb/akt and
v-akt at the plasma membrane. Science, 294(5541):374, 2001.
[308] C. L. Maire and K. L. Ligon. Glioma Models: New GEMMs Add “Class” with Genomic
and Expression Correlations. Cancer Cell, 19(3):295–297, Mar. 2011.
[309] A. Majid, T. Lin, G. Best, K. Fishlock, S. Hewamana, G. Pratt, D. Yallop, A. Buggins,
S. Wagner, B. Kennedy, et al. Cd49d is an independent prognostic marker that is
associated with cxcr4 expression in cll. Leukemia research, 35(6):750–756, 2011.
[310] P. Malatesta, E. Hartfuss, and M. Gotz. Isolation of radial glial cells by fluorescentactivated
cell sorting reveals a neuronal lineage. Development, 127(24):5253, 2000.
[311] M. Mancini and A. Toker. Nfat proteins: emerging roles in cancer progression. Nature
Reviews Cancer, 9(11):810–820, 2009.
[312] B. A. Mangerich A. How to kill tumor cells with inhibitors of poly(adp-ribosyl)ation.
Int J Cancer, 128:251–265, 2011,.
[313] J. Mao, J. Perez-losada, D. Wu, R. DelRosario, R. Tsunematsu, K. Nakayama,
K. Brown, S. Bryson, and A. Balmain. Fbxw7/cdc4 is a p53-dependent, haploinsufficient
tumour suppressor gene. Nature, 432(7018):775–779, 2004.
[314] Y. Mao, H. Sunwoo, B. Zhang, and D. Spector. Direct visualization of the cotranscriptional
assembly of a nuclear body by noncoding rnas. Nature cell biology,
13(1):95–101, 2010.
[315] M. Maragkakis, P. Alexiou, G. Papadopoulos, M. Reczko, T. Dalamagas, G. Giannopoulos,
G. Goumas, E. Koukis, K. Kourtis, V. Simossis, et al. Accurate microrna
target prediction correlates with protein repression levels. BMC bioinformatics,
10(1):295, 2009.
[316] M. Maragkakis, M. Reczko, V. Simossis, P. Alexiou, G. Papadopoulos, T. Dalamagas,
G. Giannopoulos, G. Goumas, E. Koukis, K. Kourtis, et al. Diana-microt web
server: elucidating microrna functions through target prediction. Nucleic acids research,
37(suppl 2):W273–W276, 2009.
[317] I. R. B. R. A. L. Marques MR, Horner JS. Mice lacking the p53/p63 target gene perp
are resistant to papilloma development. Cancer Res, 65:6551–6556, 2005.
[318] C. Marshall. RAS and RHO GTPases in G1-phase cell-cycle regulation. Nature
Reviews Molecular Cell Biology, 5(5):355–366, 2004.
[319] D. Martens, V. Tropepe, and D. van der Kooy. Separate proliferation kinetics of fibroblast
growth factor-responsive and epidermal growth factor-responsive neural stem
cells within the embryonic forebrain germinal zone. The Journal of Neuroscience,
20(3):1085, 2000.
[320] D. L. Masica and R. Karchin. Correlation of somatic mutation and expression identifies
genes important in human glioblastoma progression and survival. Cancer Research,
71(13):4550–4561, July 2011.
[321] J. Massagué and D. Wotton. Transcriptional control by the tgf-beta/smad signaling
system. The EMBO Journal, 19:1745–1754, 2000.
[322] K. Matsumoto, S. Nishihara, M. Kamimura, T. Shiraishi, T. Otoguro, M. Uehara,
Y. Maeda, K. Ogura, A. Lumsden, and T. Ogura. The prepattern transcription factor
irx2, a target of the fgf8/map kinase cascade, is involved in cerebellum formation.
Nature neuroscience, 7(6):605–612, 2004.
[323] M. Matsumura, D. Fremont, P. Peterson, I. Wilson, et al. Emerging principles for the
recognition of peptide antigens by mhc class i molecules. Science (New York, NY),
257(5072):927, 1992.

[324] K. McCullough, J. Martindale, L. Klotz, T. Aw, and N. Holbrook. Gadd153 sensitizes
cells to endoplasmic reticulum stress by down-regulating bcl2 and perturbing the
cellular redox state. Science Signalling, 21(4):1249, 2001.
[325] B. McEllin, C. Camacho, B. Mukherjee, B. Hahm, N. Tomimatsu, R. Bachoo, and
S. Burma. Pten loss compromises homologous recombination repair in astrocytes:
implications for glioblastoma therapy with temozolomide or poly (adp-ribose) polymerase
inhibitors. Cancer research, 70(13):5457–5464, 2010.
[326] R. McLendon, A. Friedman, D. Bigner, and E. Van Meir. Comprehensive genomic
characterization defines human glioblastoma genes and core pathways. Nature, 2008.
[327] M. Mehling, P. Simon, M. Mittelbronn, R. Meyermann, S. Ferrone, M. Weller, and
H. Wiendl. WHO grade associated downregulation of MHC class I antigen-processing
machinery components in human astrocytomas: does it reflect a potential immune
escape mechanism? Acta neuropathologica, 114(2):111–9, 2007.
[328] K. Meletis, F. Barnabé-Heider, M. Carlén, E. Evergren, N. Tomilin, O. Shupliakov,
and J. Frisén. Spinal cord injury reveals multilineage differentiation of ependymal
cells. PLoS biology, 6(7):e182, 2008.
[329] I. Mellinghoff, M. Wang, I. Vivanco, D. Haas-Kogan, S. Zhu, E. Dia, K. Lu, K. Yoshimoto,
J. Huang, D. Chute, et al. Molecular determinants of the response of glioblastomas
to egfr kinase inhibitors. New England Journal of Medicine, 353(19):2012–2024,
2005.
[330] X. Meng, M. Leyva, M. Jenny, I. Gross, S. Benosman, B. Fricker, S. Harlepp,
P. Hébraud, A. Boos, P. Wlosik, et al. A ruthenium-containing organometallic compound
reduces tumor growth through induction of the endoplasmic reticulum stress
gene chop. Cancer research, 69(13):5458–5466, 2009.
[331] T. Mercer, M. Dinger, and J. Mattick. Long non-coding rnas: insights into functions.
Nature Reviews Genetics, 10(3):155–159, 2009.
[332] F. Merkle, A. Tramontin, J. García-Verdugo, and A. Alvarez-Buylla. Radial glia give
rise to adult neural stem cells in the subventricular zone. Proceedings of the National
Academy of Sciences of the United States of America, 101(50):17528, 2004.
[333] MGI. Mouse genome informatics database. http://www.informatics.jax.org/.
[334] P. S. Mischel, S. F. Nelson, and T. F. Cloughesy. Molecular analysis of glioblastoma:
pathway profiling and its implications for patient therapy. Cancer biology and therapy,
2(3):242–7, 2003.
[335] S. Mitsui, N. Yamaguchi, Y. Osako, and K. Yuri. Enzymatic properties and localization
of motopsin (prss12), a protease whose absence causes mental retardation. Brain
research, 1136:1–12, 2007.
[336] K. Mochizuki, N. Fine, T. Fujisawa, and M. Gorovsky. Analysis of a piwi-related gene
implicates small rnas in genome rearrangement in tetrahymena. Cell, 110(6):689–699,
2002.
[337] E. Mohorko, R. Glockshuber, and M. Aebi. Oligosaccharyltransferase: the central
enzyme of n-linked protein glycosylation. Journal of inherited metabolic disease, pages
1–10, 2011.
[338] K. Mokhtari, S. Paris, L. Aguirre-Cruz, N. Privat, E. Criniere, Y. Marie, J. Hauw,
M. Kujas, D. Rowitch, K. Hoang-Xuan, et al. Olig2 expression, gfap, p53 and 1p loss
analysis contribute to glioma subclassification. Neuropathology and applied neurobiology,
31(1):62–69, 2005.
[339] M. Montanez-Wiscovich, D. Seachrist, M. Landis, J. Visvader, B. Andersen, and
R. Keri. Lmo4 is an essential mediator of erbb2/her2/neu-induced breast cancer cell
cycle progression. Oncogene, 28(41):3608–3618, 2009.
[340] M. E. Montanez-Wiscovich, D. D. Seachrist, M. D. Landis, J. Visvader, B. Andersen,
and R. A. Keri. Lmo4 is an essential mediator of erbb2/her2/neu-induced breast
cancer cell cycle progression. Oncogene, 28:3608–3618, 2009.
[341] A. Moolwaney and O. Igwe. Regulation of the cyclooxygenase-2 system by interleukin-
1β through mitogen-activated protein kinase signaling pathways: A comparative study
of human neuroglioma and neuroblastoma cells. Molecular brain research, 137(1):202–
212, 2005.

[342] H. Moon, M. Ahn, J. Park, K. Min, Y. Kwon, and K. Kim. Negative regulation of
hypoxia inducible factor-1α by necdin. FEBS letters, 579(17):3797–3801, 2005.
[343] T. Mori, A. Buffo, and M. Götz. The novel roles of glial cells revisited: the contribution
of radial glia and astrocytes to neurogenesis. Current topics in developmental biology,
69:67–99, 2005.
[344] M. Morimoto-Tomita, K. Uchimura, A. Bistrup, D. Lum, M. Egeblad, N. Boudreau,
Z. Werb, and S. Rosen. Sulf-2, a proangiogenic heparan sulfate endosulfatase, is
upregulated in breast cancer. Neoplasia (New York, NY), 7(11):1001, 2005.
[345] S. Morrison and J. Kimble. Asymmetric and symmetric stem-cell divisions in development
and cancer. Nature, 441(7097):1068–1074, 2006.
[346] A. S. Morrissy, R. D. Morin, A. Delaney, T. Zeng, H. Mcdonald, S. Jones, Y. Zhao,
M. Hirst, and M. A. Marra. Next-generation tag sequencing for cancer gene expression
profiling. Genome Research, 19(10):1825–1835, Oct. 2009.
[347] C. M. Morshead and D. van der Kooy. Disguising adult neural stem cells. Current
opinion in neurobiology, 14(1):125–131, Feb. 2004.
[348] G. Mosieniak, B. Pyrzynska, and B. Kaminska. Nuclear factor of activated t cells
(nfat) as a new component of the signal transduction pathway in glioma cells. Journal
of neurochemistry, 71(1):134–141, 1998.
[349] W. Mueller, C. Nutt, M. Ehrich, M. Riemenschneider, A. Von Deimling, D. Van
Den Boom, and D. Louis. Downregulation of runx3 and tes by hypermethylation in
glioblastoma. Oncogene, 26(4):583–593, 2006.
[350] J. Mukai, T. Hachiya, S. Shoji-Hoshino, M. Kimura, D. Nadano, P. Suvanto,
T. Hanaoka, Y. Li, S. Irie, L. Greene, et al. Nade, a p75ntr-associated cell death
executor, is involved in signal transduction mediated by the common neurotrophin
receptor p75ntr. Journal of Biological Chemistry, 275(23):17566, 2000.
[351] J. Mukai, P. Suvant, and T. Sato. Nerve growth factor-dependent regulation of nadeinduced
apoptosis. Vitamins & Hormones, 66:385–402, 2003.
[352] I. Muñoz-Sanjuán and A. H. Brivanlou. Neural induction, the default model and
embryonic stem cells. Nature Reviews Neuroscience, 3(4):271–280, apr 2002.
[353] A. Murat, E. Migliavacca, T. Gorlia, W. Lambiv, T. Shay, M. Hamou, N. De Tribolet,
L. Regli, W. Wick, M. Kouwenhoven, et al. Stem cell-related self-renewal signature
and high epidermal growth factor receptor expression associated with resistance
to concomitant chemoradiotherapy in glioblastoma. Journal of Clinical Oncology,
26(18):3015–3024, 2008.
[354] N. Nakagomi, T. Nakagomi, S. Kubo, A. Nakano-Doi, O. Saino, M. Takata,
H. Yoshikawa, D. Stern, T. Matsuyama, and A. Taguchi. Endothelial cells support
survival, proliferation, and neuronal differentiation of transplanted adult ischemiainduced
neural stem/progenitor cells after cerebral infarction. Stem Cells, 27(9):2185–
2195, 2009.
[355] C. Napoli, C. Lemieux, and R. Jorgensen. Introduction of a chimeric chalcone synthase
gene into petunia results in reversible co-suppression of homologous genes in trans.
The Plant Cell Online, 2(4):279, 1990.
[356] K. Nave. Axon-glial signaling and the glial support of axon function. Annual review
of neuroscience, 2008.
[357] R. Nawroth, A. Van Zante, S. Cervantes, M. McManus, M. Hebrok, and S. Rosen.
Extracellular sulfatases, elements of the wnt signaling pathway, positively regulate
growth and tumorigenicity of human pancreatic cancer cells. PLoS One, 2(4):e392,
2007.
[358] S. Ng, K. Buckingham, C. Lee, A. Bigham, H. Tabor, K. Dent, C. Huff, P. Shannon,
E. Jabs, D. Nickerson, et al. Exome sequencing identifies the cause of a mendelian
disorder. Nature Genetics, 42(1):30–35, 2009.
[359] C. B. Nielsen, N. Shomron, R. Sandberg, E. Hornstein, J. Kitzman, and C. B. Burge.
Determinants of targeting by endogenous and exogenous microRNAs and siRNAs.
RNA, 13(11):1894–910, 2007.

[360] I. Nimmrich, S. Erdmann, U. Melchers, S. Chtarbova, U. Finke, S. Hentsch, I. Hoffmann,
M. Oertel, W. Hoffmann, and O. Muller. The novel ependymin related gene
ucc1 is highly expressed in colorectal tumor cells. Cancer letters, 165(1):71–79, 2001.
[361] S. Noctor, A. Flint, T. Weissman, W. Wong, B. Clinton, and A. Kriegstein. Dividing
precursor cells of the embryonic cortical ventricular zone have morphological and
molecular characteristics of radial glia. The Journal of neuroscience, 22(8):3161, 2002.
[362] E. Noetzel, M. Rose, E. Sevinc, R. Hilgers, A. Hartmann, A. Naami, R. Knüchel,
and E. Dahl. Intermediate filament dynamics and breast cancer: Aberrant promoter
methylation of the synemin gene is associated with early tumor relapse. Oncogene,
29(34):4814–4825, 2010.
[363] H. Noushmehr, D. Weisenberger, and K. Diefes. Identification of a CpG island methylator
phenotype that defines a distinct subgroup of glioma. Cancer Cell, 2010.
[364] J. Novakova, O. Slaby, and R. Vyzula. Microrna involvement in glioblastoma pathogenesis.
Biochemical and biophysical research communications, 2009.
[365] H. Ohgaki and P. Kleihues. Epidemiology and etiology of gliomas. Acta neuropathologica,
109(1):93–108, 2005.
[366] K. Ohira, N. Funatsu, K. Homma, Y. Sahara, M. Hayashi, T. Kaneko, and S. Nakamura.
Truncated trkb-t1 regulates the morphology of neocortical layer i astrocytes in
adult rat brain slices. European Journal of Neuroscience, 25(2):406–416, 2007.
[367] S. Okabe, K. Forsberg-Nilsson, A. Spiro, M. Segal, and R. McKay. Development of
neuronal precursor cells and functional postmitotic neurons from embryonic stem cells
in vitro. Mechanisms of development, 59(1):89–102, 1996.
[368] Y. Okazaki, M. Furuno, T. Kasukawa, J. Adachi, H. Bono, S. Kondo, I. Nikaido,
N. Osato, R. Saito, H. Suzuki, et al. Analysis of the mouse transcriptome based on
functional annotation of 60,770 full-length cdnas. Nature, 420(6915):563–573, 2002.
[369] M. Okoniewski, T. Yates, S. Dibben, and C. Miller. An annotation infrastructure
for the analysis and interpretation of affymetrix exon array data. Genome biology,
8(5):R79, 2007.
[370] P. Ongusaha, T. Ouchi, K. Kim, E. Nytko, J. Kwak, R. Duda, C. Deng, and S. Lee.
Brca1 shifts p53-mediated cellular outcomes towards irreversible growth arrest. Oncogene,
22(24):3749–3758, 2003.
[371] P. C. Orban, D. Chui, and J. D. Marth. Tissue- and site-specific DNA recombination
in transgenic mice. Proceedings of the National Academy of Sciences of the United
States of America, 89(15):6861–6865, Aug. 1992.
[372] U. Ørom, T. Derrien, M. Beringer, K. Gumireddy, A. Gardini, G. Bussotti, F. Lai,
M. Zytnicki, C. Notredame, Q. Huang, et al. Long noncoding rnas with enhancer-like
function in human cells. Cell, 143(1):46–58, 2010.
[373] E. Ostrakhovitch, P. Olsson, S. Jiang, and M. Cherian. Interaction of metallothionein
with tumor suppressor p53 protein. FEBS letters, 580(5):1235–1238, 2006.
[374] T. Ozawa, C. W. Brennan, L. Wang, M. Squatrito, T. Sasayama, M. Nakada, J. T.
Huse, A. Pedraza, S. Utsuki, Y. Yasui, A. Tandon, E. I. Fomchenko, H. Oka, R. L.
Levine, K. Fujii, M. Ladanyi, and E. C. Holland. PDGFRA gene rearrangements are
frequent genetic events in PDGFRA-amplified glioblastomas. Genes & Development,
24(19):2205–2218, Oct. 2010.
[375] L. Pacey, J. Stead, A. Gleave, K. Tomczyk, and L. Doering. Neural stem cell culture:
neurosphere generation, microscopical analysis and cryopreservation. Nat. Protoc,
215:1–14, 2006.
[376] T. Palmer, P. Schwartz, P. Taupin, B. Kaspar, S. Stein, and F. Gage. Cell culture:
Progenitor cells from human brain after death. Nature, 411(6833):42–43, 2001.
[377] P. Pandolfi. Breast cancerÂŮloss of pten predicts resistance to treatment. New England
Journal of Medicine, 351(22):2337–2338, 2004.
[378] K. Paraiso, Y. Xiang, V. Rebecca, E. Abel, Y. Chen, A. Munko, E. Wood, I. Fedorenko,
V. Sondak, A. Anderson, et al. Pten loss confers braf inhibitor resistance
to melanoma cells through the suppression of bim expression. Cancer research,
71(7):2750–2760, 2011.

[379] D. Park and J. Rich. Biology of glioma cancer stem cells. Molecules and cells, 28(1):7–
12, 2009.
[380] H. Park, I. Han, H. Kwon, and E. Oh. Focal adhesion kinase regulates syndecan-
2–mediated tumorigenic activity of ht1080 fibrosarcoma cells. Cancer research,
65(21):9899–9905, 2005.
[381] J. Park, J. Jung, M. Seo, S. Kang, Y. Lee, and K. Kang. Dner modulates adipogenesis
of human adipose tissue-derived mesenchymal stem cells via regulation of cell
proliferation. Cell proliferation, 43(1):19–28, 2010.
[382] D. Parry, D. Mahony, K. Wills, and E. Lees. Cyclin d-cdk subunit arrangement is
dependent on the availability of competing ink4 and p21 class inhibitors. Molecular
and cellular biology, 19(3):1775, 1999.
[383] D. Parsons, S. Jones, X. Zhang, J. Lin, and R. Leary. An integrated genomic analysis
of human glioblastoma multiforme. Science, 2008.
[384] H. Pasantes-Morales and A. Schousboe. Role of taurine in osmoregulation in brain
cells: mechanisms and functional implications. Amino Acids, 12(3):281–292, 1997.
[385] L. Patrawala, T. Calhoun, R. Schneider-Broussard, J. Zhou, K. Claypool, and D. Tang.
Side population is enriched in tumorigenic, stem-like cancer cells, whereas abcg2+ and
abcg2- cancer cells are similarly tumorigenic. Cancer research, 65(14):6207, 2005.
[386] K. Paulsson, M. Kleijmeer, J. Griffith, M. Jevon, S. Chen, P. Anderson, H. Sjögren,
S. Li, and P. Wang. Association of tapasin and copi provides a mechanism for the
retrograde transport of major histocompatibility complex (mhc) class i molecules from
the golgi complex to the endoplasmic reticulum. Journal of Biological Chemistry,
277(21):18266, 2002.
[387] G. Paxinos, J. Mai, and S. O. service). The Human Nervous System. Elsevier Academic
Press London, 2004.
[388] G. Pearson, F. Robinson, T. Gibson, B. Xu, M. Karandikar, K. Berman, and M. Cobb.
Mitogen-activated protein (map) kinase pathways: regulation and physiological functions.
Endocrine reviews, 22(2):153–183, 2001.
[389] L. Pevny and M. Placzek. Sox genes and neural progenitor identity. Current opinion
in neurobiology, 15(1):7–13, 2005.
[390] H. S. Phillips, S. Kharbanda, R. Chen, W. F. Forrest, R. H. Soriano, T. D. Wu,
A. Misra, J. M. Nigro, H. Colman, L. Soroceanu, P. M. Williams, Z. Modrusan, B. G.
Feuerstein, and K. Aldape. Molecular subclasses of high-grade glioma predict prognosis,
delineate a pattern of disease progression, and resemble stages in neurogenesis.
Cancer Cell, 9(3):157–173, Mar. 2006.
[391] J. Phillips, E. Huillard, A. Robinson, A. Ward, D. Lum, M. Polley, S. Rosen, D. Rowitch,
and Z. Werb. Heparan sulfate sulfatase sulf2 regulates pdgfrα signaling and
growth in human and mouse malignant glioma. The Journal of clinical investigation,
122(3):911, 2012.
[392] S. Piccirillo, B. Reynolds, N. Zanetti, G. Lamorte, E. Binda, G. Broggi, H. Brem,
A. Olivi, F. Dimeco, A. Vescovi, et al. Bone morphogenetic proteins inhibit the
tumorigenic potential of human brain tumour-initiating cells. Nature, 444(7120):761,
2006.
[393] S. G. Piccirillo, E. Binda, R. Fiocco, A. L. Vescovi, and K. Shah. Brain cancer stem
cells. Journal of molecular medicine, 87(11):1087–95, 2009.
[394] R. Piet, L. Vargová, E. Syková, D. Poulain, and S. Oliet. Physiological contribution
of the astrocytic environment of neurons to intersynaptic crosstalk. Proceedings of the
National Academy of Sciences of the United States of America, 101(7):2151, 2004.
[395] D. Pinto and H. Clevers. Wnt control of stem cells and differentiation in the intestinal
epithelium. Experimental cell research, 306(2):357–363, 2005.
[396] A. Pitre, N. Davis, M. Paul, A. Orr, and O. Skalli. Synemin promotes akt-dependent
glioblastoma cell proliferation by antagonizing pp2a. Molecular biology of the cell,
23(7):1243–1253, 2012.
[397] S. Pleasure, C. Page, and V. Lee. Pure, postmitotic, polarized human neurons derived
from ntera 2 cells provide a system for expressing exogenous proteins in terminally
differentiated neurons. The Journal of neuroscience, 12(5):1802, 1992.

[398] E. Poch, R. Miñambres, E. Mocholí, C. Ivorra, A. Pérez-Aragó, C. Guerri, I. Pérez-
Roger, and R. M. Guasch. RhoE interferes with Rb inactivation and regulates the
proliferation and survival of the U87 human glioblastoma cell line. Experimental cell
research, 313(4):719–731, Feb. 2007.
[399] S. Pollard, A. Benchoua, and S. Lowell. Neural stem cells, neurons, and glia. Methods
in enzymology, 418:151–169, 2006.
[400] S. Pollard and L. Conti. Investigating radial glia in vitro. Progress in neurobiology,
83(1):53–67, 2007.
[401] S. Pollard, R. Wallbank, S. Tomlinson, L. Grotewold, and A. Smith. Fibroblast growth
factor induces a neural stem cell phenotype in foetal forebrain progenitors and during
embryonic stem cell differentiation. Molecular and Cellular Neuroscience, 38(3):393–
403, 2008.
[402] S. Pollard, K. Yoshikawa, I. Clarke, D. Danovi, S. Stricker, R. Russell, J. Bayani,
R. Head, M. Lee, M. Bernstein, et al. Glioma stem cell lines expanded in adherent
culture have tumor-specific phenotypes and are suitable for chemical and genetic
screens. Cell Stem Cell, 4(6):568–580, 2009.
[403] S. M. Pollard. Adherent Neural Stem (NS) Cells from Fetal and Adult Forebrain.
Cerebral Cortex, 16(Supplement 1):i112–i120, July 2006.
[404] S. M. Pollard, K. Yoshikawa, I. D. Clarke, D. Danovi, S. Stricker, R. Russell, J. Bayani,
R. Head, M. Lee, M. Bernstein, J. A. Squire, A. Smith, and P. Dirks. Glioma stem cell
lines expanded in adherent culture have tumor-specific phenotypes and are suitable
for chemical and genetic screens. Cell Stem Cell, 4(6):568–80, 2009.
[405] K. Pollock, P. Stroemer, S. Patel, L. Stevanato, A. Hope, E. Miljan, Z. Dong,
H. Hodges, J. Price, and J. Sinden. A conditionally immortal clonal stem cell line
from human cortical neuroepithelium for the treatment of ischemic stroke. Experimental
neurology, 199(1):143–155, 2006.
[406] R. Popovic and J. Licht. Mek and maf in myeloma therapy. Blood, 117(8):2300–2302,
2011.
[407] A. Popovi&cacute, A. Demirovi&cacute, B. Spaji&cacute, G. Štimac, and
D. B Krušlin. Expression and prognostic role of syndecan-2 in prostate cancer. Prostate
cancer and prostatic diseases, 13(1):78–82, 2009.
[408] T. D. Portal. Tcga data portal: An integrated genomic analysis identifies clinically
relevant subtypes of glioblastoma characterized by abnormalities in pdgfra, idh1, egfr
and nf1.
[409] K. Pruitt, T. Tatusova, and D. Maglott. Ncbi reference sequences (refseq): a curated
non-redundant sequence database of genomes, transcripts and proteins. Nucleic acids
research, 35(suppl 1):D61–D65, 2007.
[410] PubMed. Pubmed. http://www.ncbi.nlm.nih.gov/pubmed.
[411] R. Puca, L. Nardinocchi, G. Bossi, A. Sacchi, G. Rechavi, D. Givol, and G. D’Orazi.
Restoring wtp53 activity in hipk2 depleted mcf7 cells by modulating metallothionein
and zinc. Experimental cell research, 315(1):67–75, 2009.
[412] Z. Qin, F. Ren, X. Xu, Y. Ren, H. Li, Y. Wang, Y. Zhai, and Z. Chang. Znf536, a
novel zinc finger protein specifically expressed in the brain, negatively regulates neuron
differentiation by repressing retinoic acid-induced gene transcription. Molecular and
cellular biology, 29(13):3633, 2009.
[413] A. Quiñones Hinojosa and N. Sanai. Cellular composition and cytoarchitecture of
the adult human subventricular zone: a niche of neural stem cells. The Journal of
comparative neurology, 2006.
[414] F. Radtke and H. Clevers. Self-renewal and cancer of the gut: two sides of a coin.
Science, 307(5717):1904, 2005.
[415] B. Ragel, W. Couldwell, D. Gillespie, and R. Jensen. Identification of hypoxia-induced
genes in a malignant glioma cell line (u-251) by cdna microarray analysis. Neurosurgical
review, 30(3):181–187, 2007.
[416] H. Rajagopalan, P. Jallepalli, C. Rago, V. Velculescu, K. Kinzler, B. Vogelstein,
and C. Lengauer. Inactivation of hcdc4 can cause chromosomal instability. Nature,
428(6978):77–81, 2004.

[417] P. Rakic. Guidance of neurons migrating to the fetal monkey neocortex. Brain research,
33(2):471, 1971.
[418] P. Rakic. Elusive radial glial cells: historical and evolutionary perspective. Glia,
43(1):19–32, 2003.
[419] S. Ramaswamy, K. Ross, E. Lander, and T. Golub. A molecular signature of metastasis
in primary solid tumors. Nature genetics, 33(1):49–54, 2002.
[420] L. Rambhatla, S. Ram-Mohan, J. Cheng, and J. Sherley. Immortal dna strand cosegregation
requires p53/impdh–dependent asymmetric self-renewal associated with adult
stem cells. Cancer research, 65(8):3155, 2005.
[421] L. B. Rangel, R. Agarwal, C. A. Sherman-Baust, V. Mello-Coelho, E. S. Pizer, H. Ji,
D. D. Taub, and P. J. Morin. Anomalous expression of the HLA-DR alpha and beta
chains in ovarian and other cancers. Cancer biology and therapy, 3(10):1021–7, 2004.
[422] P. Rao, M. Jaggi, D. Smith, G. Hemstreet, and K. Balaji. Metallothionein 2a interacts
with the kinase domain of pkcµ in prostate cancer. Biochemical and biophysical
research communications, 310(3):1032–1038, 2003.
[423] S. K. Rao, J. Edwards, A. D. Joshi, I.-M. Siu, and G. J. Riggins. A survey of glioblastoma
genomic amplifications and deletions. Journal of neuro-oncology, 96(2):169–179,
2010.
[424] W. A. Redmond WL, Ruby CE. The role of ox40-mediated co-stimulation in t-cell
activation and survival. Crit Rev Immunol, 29:187–201, 2009.
[425] J. Rehwinkel, J. Raes, and E. Izaurralde. Nonsense-mediated mRNA decay: Target
genes and functional diversification of effectors. Trends Biochem Sci, 31(11):639–46,
2006.
[426] K. Reilly, D. Loisel, and R. Bronson. Nf1; Trp53 mutant mice develop glioblastoma
with evidence of strain-specific effects. Nature Genetics, 2000.
[427] B. Reynolds and S. Weiss. Generation of neurons and astrocytes from isolated cells
of the adult mammalian central nervous system. Science, 255(5052):1707, 1992.
[428] S. Riaz, E. Jauniaux, G. Stern, and H. Bradford. The controlled conversion of human
neural progenitor cells derived from foetal ventral mesencephalon into dopaminergic
neurons in vitro. Developmental brain research, 136(1):27–34, 2002.
[429] M. J. Riemenschneider, R. Büschges, M. Wolter, J. Reifenberger, J. Boström, J. A.
Kraus, U. Schlegel, and G. Reifenberger. Amplification and overexpression of the
MDM4 (MDMX) gene from 1q32 in a subset of malignant gliomas without TP53
mutation or MDM2 amplification. Cancer Research, 59(24):6091–6096, Dec. 1999.
[430] A. Rizki, V. Weaver, S. Lee, G. Rozenberg, K. Chin, C. Myers, J. Bascom, J. Mott,
J. Semeiks, L. Grate, et al. A human breast cell model of preinvasive to invasive
transition. Cancer research, 68(5):1378, 2008.
[431] D. Robinson, Y. Wu, and S. Lin. The protein tyrosine kinase family of the human
genome. biomedical science, 19:5548–5557, 2000.
[432] R. Roelofs, D. Fischer, S. Houtman, J. Sluijs, W. Van Haren, F. Van Leeuwen, and
E. Hol. Adult human subventricular, subgranular, and subpial zones contain astrocytes
with a specialized intermediate filament cytoskeleton. Glia, 52(4):289–300, 2005.
[433] S. Rosen and H. Lemjabbar-Alaoui. Sulf-2: an extracellular modulator of cell signaling
and a cancer target candidate. Expert opinion on therapeutic targets, 14(9):935–949,
2010.
[434] L. Rosso and J. Mienville. Pituicyte modulation of neurohormone output. Glia,
57(3):235–243, 2009.
[435] A. Rousseau, C. Nutt, R. Betensky, A. Iafrate, M. Han, K. Ligon, D. Rowitch, and
D. Louis. Expression of oligodendroglial and astrocytic lineage markers in diffuse
gliomas: use of ykl-40, apoe, ascl1, and nkx2-2. Journal of Neuropathology & Experimental
Neurology, 65(12):1149, 2006.
[436] N. Roy, A. Benraiss, S. Wang, R. Fraser, R. Goodman, W. Couldwell, M. Nedergaard,
A. Kawaguchi, H. Okano, and S. Goldman. Promoter-targeted selection and
isolation of neural progenitor cells from the adult human ventricular zone. Journal of
neuroscience research, 59(3):321–331, 2000.

[437] S. Rybalkin, C. Yan, K. Bornfeldt, and J. Beavo. Cyclic gmp phosphodiesterases and
regulation of smooth muscle function. Circulation research, 93(4):280–291, 2003.
[438] T. Saito, K. Shibasaki, M. Kurachi, S. Puentes, M. Mikuni, and Y. Ishizaki. Cerebral
capillary endothelial cells are covered by the VEGF-expressing foot processes of
astrocytes. Neuroscience Letters, 2011.
[439] A. Sakurada, H. Hamada, S. Fukushige, T. Yokoyama, K. Yoshinaga, T. Furukawa,
S. Sato, A. Yajima, M. Sato, S. Fujimura, et al. Adenovirus-mediated delivery of
the pten gene inhibits cell growth by induction of apoptosis in endometrial cancer.
International journal of oncology, 15(6):1069–1074, 1999.
[440] Y. Samuels, Z. Wang, A. Bardelli, N. Silliman, J. Ptak, S. Szabo, H. Yan, A. Gazdar,
S. Powell, G. Riggins, et al. High frequency of mutations of the PIK3CA gene in
human cancers. Science, 304(5670):554, 2004.
[441] D. San, J. Ray, and F. Gage. Bipotent progenitor cell lines from the human cns.
Nature biotechnology, 15(6):574–580, 1997.
[442] N. Sanai, A. Alvarez-Buylla, and M. Berger. Neural stem cells and the origin of
gliomas. New England Journal of Medicine, 353(8):811–822, 2005.
[443] N. Sanai, A. Tramontin, A. Quiñones-Hinojosa, N. Barbaro, N. Gupta, S. Kunwar,
M. Lawton, M. McDermott, A. Parsa, J. Verdugo, et al. Unique astrocyte ribbon
in adult human brain contains neural stem cells but lacks chain migration. Nature,
427(6976):740–744, 2004.
[444] M. Sarti, C. Sevignani, G. Calin, R. Aqeilan, M. Shimizu, F. Pentimalli, M. Picchio,
A. Godwin, A. Rosenberg, A. Drusco, et al. Adenoviral transduction of testin gene
into breast and uterine cancer cell lines promotes apoptosis and tumor reduction in
vivo. Clinical cancer research, 11(2):806–813, 2005.
[445] E. Schmidt, K. Ichimura, H. Goike, A. Moshref, L. Liu, and V. Collins. Mutational
profile of the pten gene in primary human astrocytic tumors and cultivated xenografts.
Journal of Neuropathology & Experimental Neurology, 58(11):1170, 1999.
[446] M. Selbach, B. Schwanhausser, N. Thierfelder, Z. Fang, R. Khanin, and N. Rajewsky.
Widespread changes in protein synthesis induced by microRNAs. Nature,
455(7209):58–63, 2008.
[447] D. Seo, J. Sung, H. Cho, H. Yi, K. Seo, I. Choi, D. Kim, J. Kim, A. El-Aty, H. Shin,
et al. Gene expression profiling of cancer stem cell in human lung adenocarcinoma
a549 cells. Mol Cancer, 6(1):75, 2007.
[448] J. Seoane, H. Le, L. Shen, S. Anderson, and J. Massagué. Integration of smad and
forkhead pathways in the control of neuroepithelial and glioblastoma cell proliferation.
Cell, 117(2):211–223, 2004.
[449] B. Seri, J. Garcı a Verdugo, B. McEwen, and A. Alvarez-Buylla. Astrocytes give rise
to new neurons in the adult mammalian hippocampus. The Journal of Neuroscience,
21(18):7153, 2001.
[450] R. Shai, T. Shi, T. J. Kremen, S. Horvath, L. M. Liau, T. F. Cloughesy, P. S. Mischel,
and S. F. Nelson. Gene expression profiling identifies molecular subtypes of gliomas.
Oncogene, 22(31):4918–4923, July 2003.
[451] P. Shannon, A. Markiel, O. Ozier, N. Baliga, J. Wang, D. Ramage, N. Amin,
B. Schwikowski, and T. Ideker. Cytoscape: a software environment for integrated
models of biomolecular interaction networks. Genome research, 13(11):2498–2504,
2003.
[452] Q. Shen, Y. Wang, E. Kokovay, G. Lin, S. Chuang, S. Goderie, B. Roysam, and
S. Temple. Adult svz stem cells lie in a vascular niche: a quantitative analysis of niche
cell-cell interactions. Cell Stem Cell, 3(3):289–300, 2008.
[453] W. Shen, A. Balajee, J. Wang, H. Wu, C. Eng, P. Pandolfi, and Y. Yin. Essential role
for nuclear pten in maintaining chromosomal integrity. Cell, 128(1):157–170, 2007.
[454] C. Sherr and J. Roberts. CDK inhibitors: positive and negative regulators of G1-phase
progression. Genes & development, 13(12):1501, 1999.
[455] S. L. Shi W, Fan H and D. R. The tetraspanin cd9 associates with transmembrane
tgf-alpha and regulates tgf-alpha-induced egf receptor activation and cell proliferation.
J. Cell Biol, (148):591–602, 2000.

[456] T. Shima, N. Okumura, T. Takao, Y. Satomi, T. Yagi, M. Okada, and K. Nagai.
Interaction of the sh2 domain of fyn with a cytoskeletal protein, β-adducin. Journal
of Biological Chemistry, 276(45):42233–42240, 2001.
[457] M. Shipitsin and K. Polyak. The cancer stem cell hypothesis: in search of definitions,
markers, and relevance. Laboratory Investigation, 88(5):459–463, 2008.
[458] S. Singh, I. Clarke, M. Terasaki, V. Bonn, C. Hawkins, J. Squire, and P. Dirks. Identification
of a cancer stem cell in human brain tumors. Cancer Research, 63(18):5821,
2003.
[459] S. Singh, C. Hawkins, I. Clarke, and J. Squire. Identification of human brain tumour
initiating cells. Nature, 2004.
[460] P. Singha, I. Yeh, M. Venkatachalam, P. Saikumar, et al. Transforming growth factor
beta-inducible gene tmepai converts tgf-beta from a tumor suppressor to a tumor
promoter in breast cancer. Cancer research, 70(15):6377–6383, 2010.
[461] D. Smadja, C. d’Audigier, L. Weiswald, C. Badoual, V. Dangles-Marie, L. Mauge,
S. Evrard, I. Laurendeau, F. Lallemand, S. Germain, et al. The wnt antagonist
dickkopf-1 increases endothelial progenitor cell angiogenic potential. Arteriosclerosis,
thrombosis, and vascular biology, 30(12):2544–2552, 2010.
[462] K. Smith, M. Luong, and G. Stein. Pluripotency: toward a gold standard for human
es and ips cells. Journal of cellular physiology, 220(1):21–29, 2009.
[463] B. J. M. D. C. B. G. R. Smith SJ, Long A. Pediatric high-grade glioma: identification
of poly(adp-ribose) polymerase as a potential therapeutic target. Neuro-oncology,
13:1171–1177, 2011.
[464] G. Smyth. Linear models and empirical bayes methods for assessing differential expression
in microarray experiments. Statistical applications in genetics and molecular
biology, 3(1):3, 2004.
[465] G. Smyth. Limma: linear models for microarray data. Bioinformatics and computational
biology solutions using R and Bioconductor, pages 397–420, 2005.
[466] G. Smyth and T. Speed. Normalization of cdna microarray data. Methods, 31(4):265–
273, 2003.
[467] D. A. Solomon, J.-S. Kim, W. Jean, and T. Waldman. Conspirators in a capital
crime: co-deletion of p18INK4c and p16INK4a/p14ARF/p15INK4b in glioblastoma
multiforme. Cancer Research, 68(21):8657–8660, Nov. 2008.
[468] J. Soos, J. Krieger, O. Stüve, C. King, J. Patarroyo, K. Aldape, K. Wosik, A. Slavin,
P. Nelson, J. Antel, et al. Malignant glioma cells use mhc class ii transactivator
(ciita) promoters iii and iv to direct ifn-γ-inducible ciita expression and can function
as nonprofessional antigen presenting cells in endocytic processing and cd4+ t-cell
activation. Glia, 36(3):391–405, 2001.
[469] A. Stark, J. Brennecke, N. Bushati, R. B. Russell, and S. M. Cohen. Animal MicroR-
NAs confer robustness to gene expression and have a significant impact on 3’UTR
evolution. Cell, 123(6):1133–46, 2005.
[470] A. Stark, J. Brennecke, R. B. Russell, and S. M. Cohen. Identification of Drosophila
MicroRNA targets. PLoS Biol, 1(3):E60, 2003.
[471] C. StrÃÂijbing, G. Ahnert-Hilger, J. Shan, B. Wiedenmann, J. Hescheler, and A. M.
Wobus. Differentiation of pluripotent embryonic stem cells into the neuronal lineage
in vitro gives rise to mature inhibitory and excitatory neurons. Mechanisms of Development,
53(2):275–287, 1995.
[472] R. Stupp, W. P. Mason, M. J. van den Bent, M. Weller, B. Fisher, M. J. B. Taphoorn,
K. Belanger, A. A. Brandes, C. Marosi, U. Bogdahn, J. Curschmann, R. C. Janzer,
S. K. Ludwin, T. Gorlia, A. Allgeier, D. Lacombe, J. G. Cairncross, E. Eisenhauer,
R. O. Mirimanoff, E. O. for Research, T. of Cancer Brain Tumor, R. Groups, and
N. C. I. of Canada Clinical Trials Group. Radiotherapy plus concomitant and adjuvant
temozolomide for glioblastoma. The New England journal of medicine, 352(10):987–
996, Mar. 2005.
[473] X. Su, C. Kong, and P. Stahl. Gapex-5 mediates ubiquitination, trafficking, and
degradation of epidermal growth factor receptor. Journal of Biological Chemistry,
282(29):21278–21284, 2007.

[474] A. Subramanian, P. Tamayo, V. Mootha, S. Mukherjee, B. Ebert, M. Gillette,
A. Paulovich, S. Pomeroy, T. Golub, E. Lander, et al. Gene set enrichment analysis:
a knowledge-based approach for interpreting genome-wide expression profiles.
Proceedings of the National Academy of Sciences of the United States of America,
102(43):15545–15550, 2005.
[475] E. Sum, D. Segara, B. Duscio, M. Bath, A. Field, R. Sutherland, G. Lindeman, and
J. Visvader. Overexpression of lmo4 induces mammary hyperplasia, promotes cell
invasion, and is a predictor of poor outcome in breast cancer. Proceedings of the
National Academy of Sciences of the United States of America, 102(21):7659–7664,
2005.
[476] L. Sun, W. Yan, Y. Wang, G. Sun, H. Luo, J. Zhang, X. Wang, Y. You, Z. Yang, and
N. Liu. Microrna-10b induces glioma cell invasion by modulating mmp-14 and upar
expression via hoxd10. Brain research, 1389:9–18, 2011.
[477] N. Sun, T. Huiatt, D. Paulin, Z. Li, and R. Robson. Synemin interacts with the lim
domain protein zyxin and is essential for cell adhesion and migration. Experimental
cell research, 316(3):491–505, 2010.
[478] P. Sun, S. Xia, B. Lal, C. Eberhart, A. Quinones-Hinojosa, J. Maciaczyk, W. Matsui,
F. DiMeco, S. Piccirillo, A. Vescovi, et al. Dner, an epigenetically modulated gene,
regulates glioblastoma-derived neurosphere cell differentiation and tumor propagation.
Stem Cells, 27(7):1473–1486, 2009.
[479] T. Sun, X. Wang, S. Xie, D. Zhang, X. Wang, B. Li, W. Ma, and H. Xin. A comparison
of proliferative capacity and passaging potential between neural stem and progenitor
cells in adherent and neurosphere cultures. International Journal of Developmental
Neuroscience, 2011.
[480] Y. Sun, W. Kong, A. Falk, J. Hu, L. Zhou, S. Pollard, and A. Smith. Cd133 (prominin)
negative human neural stem cells are clonogenic and tripotent. PloS one, 4(5):e5498,
2009.
[481] Y. Sun, S. Pollard, L. Conti, M. Toselli, G. Biella, G. Parkin, L. Willatt, A. Falk,
E. Cattaneo, and A. Smith. Long-term tripotent differentiation capacity of human
neural stem (ns) cells in adherent culture. Molecular and Cellular Neuroscience,
38(2):245–258, 2008.
[482] C. Suo, A. Salim, K. S. Chia, Y. Pawitan, and S. Calza. Modified least-variant set
normalization for miRNA microarray. RNA, 16(12):2293–303, 2010.
[483] C. Svendsen, M. ter Borg, R. Armstrong, A. Rosser, S. Chandran, T. Ostenfeld, and
M. Caldwell. A new method for the rapid and long term growth of human neural
precursor cells. Journal of neuroscience methods, 85(2):141–152, 1998.
[484] Y. Takamura, H. Ikeda, T. Kanaseki, M. Toyota, T. Tokino, K. Imai, K. Houkin, and
N. Sato. Regulation of mhc class ii expression in glioma cells by class ii transactivator
(ciita). Glia, 45(4):392–405, 2004.
[485] O. Tam, A. Aravin, P. Stein, A. Girard, E. Murchison, S. Cheloufi, E. Hodges,
M. Anger, R. Sachidanandam, R. Schultz, et al. Pseudogene-derived small interfering
rnas regulate gene expression in mouse oocytes. Nature, 453(7194):534, 2008.
[486] B. Tan, C. Park, L. Ailles, and I. Weissman. The cancer stem cell hypothesis: a work
in progress. Laboratory investigation, 86(12):1203–1207, 2006.
[487] N. Taniguchi, H. Taniura, M. Niinobe, C. Takayama, K. Tominaga-Yoshino, A. Ogura,
and K. Yoshikawa. The postmitotic growth suppressor necdin interacts with a
calcium-binding protein (nefa) in neuronal cytoplasm. Journal of Biological Chemistry,
275(41):31674–31681, 2000.
[488] M. Taniwaki, Y. Daigo, N. Ishikawa, A. Takano, T. Tsunoda, W. Yasui, K. Inai,
N. Kohno, and Y. Nakamura. Gene expression profiles of small-cell lung cancers:
molecular signatures of lung cancer. International journal of oncology, 29(3):567–576,
2006.
[489] A. Tarca, S. Draghici, P. Khatri, S. Hassan, P. Mittal, J. Kim, C. Kim, J. Kusanovic,
and R. Romero. A novel signaling pathway impact analysis. Bioinformatics, 25(1):75,
2009.

[490] P. A. C. t’Hoen, Y. Ariyurek, H. H. Thygesen, E. Vreugdenhil, R. H. A. M. Vossen,
R. X. De Menezes, J. M. Boer, G.-J. B. Van Ommen, and J. T. Den Dunnen. Deep
sequencing-based expression analysis shows major advances in robustness, resolution
and inter-lab portability over five microarray platforms. Nucleic Acids Research,
36(21):e141–e141, Oct. 2008.
[491] C. Thomas, G. Ely, C. D. James, R. Jenkins, M. Kastan, A. Jedlicka, P. Burger, and
R. Wharen. Glioblastoma-related gene mutations and over-expression of functional
epidermal growth factor receptors in SKMG-3 glioma cells. Acta neuropathologica,
101(6):605–615, June 2001.
[492] J. Ting and J. Trowsdale. Genetic control of mhc class ii expression. Cell, 109(2):S21–
S33, 2002.
[493] E. Tobias, A. Hurlstone, E. MacKenzie, R. McFarlane, D. Black, et al. The tes gene at
7q31. 1 is methylated in tumours and encodes a novel growth-suppressing lim domain
protein. Oncogene, 20(22):2844, 2001.
[494] V. Tropepe, M. Sibilia, B. Ciruna, J. Rossant, E. Wagner, and D. Kooy. Distinct neural
stem cells proliferate in response to egf and fgf in the developing mouse telencephalon.
Developmental biology, 208(1):166–188, 1999.
[495] L. Trotman, X. Wang, A. Alimonti, Z. Chen, J. Teruya-Feldstein, H. Yang,
N. Pavletich, B. Carver, C. Cordon-Cardo, H. Erdjument-Bromage, et al. Ubiquitination
regulates pten nuclear import and tumor suppression. Cell, 128(1):141–156,
2007.
[496] A. B. Trovó-Marqui and E. H. Tajara. Neurofibromin: a general outlook. Clinical
genetics, 70(1):1–13, July 2006.
[497] C. Tso, P. Shintaku, J. Chen, Q. Liu, J. Liu, Z. Chen, K. Yoshimoto, P. Mischel,
T. Cloughesy, L. Liau, et al. Primary glioblastomas express mesenchymal stem-like
properties. Molecular cancer research, 4(9):607–619, 2006.
[498] A. Tsuchida, T. Okajima, K. Furukawa, T. Ando, H. Ishida, A. Yoshida, Y. Nakamura,
R. Kannagi, M. Kiso, and K. Furukawa. Synthesis of disialyl lewis a (lea) structure in
colon cancer cell lines by a sialyltransferase, st6galnac vi, responsible for the synthesis
of α-series gangliosides. Journal of Biological Chemistry, 278(25):22787–22794, 2003.
[499] N. Tsuji, K. Kondoh, M. Furuya, D. Kobayashi, A. Yagihashi, Y. Inoue, T. Meguro,
S. Horita, H. Takahashi, and N. Watanabe. A novel aspartate protease gene, alp56,
is related to morphological features of colorectal adenomas. International journal of
colorectal disease, 19(1):43–48, 2004.
[500] V. Turk, B. Turk, G. Guncar, D. Turk, and J. Kos. Lysosomal cathepsins: structure,
role in antigen processing and presentation, and cancer. Advances in enzyme
regulation, 42:285, 2002.
[501] A. Tzschach, A. Bisgaard, M. Kirchhoff, L. Graul-Neumann, H. Neitzel, S. Page,
A. Ahmed, I. Müller, F. Erdogan, H. Ropers, et al. Chromosome aberrations involving
10q22: report of three overlapping interstitial deletions and a balanced translocation
disrupting c10orf11. European Journal of Human Genetics, 18(3):291–295, 2009.
[502] N. Uchida, D. Buck, D. He, M. Reitsma, M. Masek, T. Phan, A. Tsukamoto, F. Gage,
and I. Weissman. Direct isolation of human central nervous system stem cells. Proceedings
of the National Academy of Sciences, 97(26):14720, 2000.
[503] M. Van De Wiel, K. Kim, S. Vosse, W. Van Wieringen, S. Wilting, and B. Ylstra.
Cghcall: calling aberrations for array cgh tumor profiles. Bioinformatics, 23(7):892–
894, 2007.
[504] J. Van Den Boom, M. Wolter, R. Kuick, D. Misek, A. Youkilis, D. Wechsler, C. Sommer,
G. Reifenberger, and S. Hanash. Characterization of gene expression profiles
associated with glioma progression using oligonucleotide-based microarray analysis
and real-time reverse transcription-polymerase chain reaction. The American journal
of pathology, 163(3):1033–1043, 2003.
[505] A. Van der Krol, L. Mur, M. Beld, J. Mol, and A. Stuitje. Flavonoid genes in petunia:
addition of a limited number of gene copies may lead to a suppression of gene
expression. The Plant Cell Online, 2(4):291, 1990.

[506] B. van Houte, T. Binsl, H. Hettling, and J. Heringa. Cghnormaliter: a bioconductor
package for normalization of array cgh data with many cnas. Bioinformatics,
26(10):1366–1367, 2010.
[507] F. van Ruissen and F. Baas. Serial analysis of gene expression (SAGE). Methods in
molecular biology, 383:41–66, 2007.
[508] S. Vatter, G. Pahlke, J. Deitmer, and G. Eisenbrand. Differential phosphodiesterase
expression and cytosolic ca2+ in human cns tumour cells and in non-malignant and
malignant cells of rat origin. Journal of neurochemistry, 93(2):321–329, 2005.
[509] F. Vazquez, H. Vaucheret, R. Rajagopalan, C. Lepers, V. Gasciolli, A. Mallory,
J. Hilbert, D. Bartel, and P. Crété. Endogenous trans-acting sirnas regulate the
accumulation of arabidopsis mrnas. Molecular Cell, 16(1):69–79, 2004.
[510] E. Venkatraman and A. Olshen. A faster circular binary segmentation algorithm for
the analysis of array cgh data. Bioinformatics, 23(6):657–663, 2007.
[511] R. G. W. Verhaak, K. A. Hoadley, E. Purdom, V. Wang, Y. Qi, M. D. Wilkerson,
C. R. Miller, L. Ding, T. Golub, J. P. Mesirov, G. Alexe, M. Lawrence, M. O’Kelly,
P. Tamayo, B. A. Weir, S. Gabriel, W. Winckler, S. Gupta, L. Jakkula, H. S. Feiler,
J. G. Hodgson, C. D. James, J. N. Sarkaria, C. Brennan, A. Kahn, P. T. Spellman,
R. K. Wilson, T. P. Speed, J. W. Gray, M. Meyerson, G. Getz, C. M. Perou, D. N.
Hayes, and Cancer Genome Atlas Research Network. Integrated genomic analysis
identifies clinically relevant subtypes of glioblastoma characterized by abnormalities
in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell, 17(1):98–110, Jan. 2010.
[512] A. Vescovi, E. Parati, A. Gritti, P. Poulin, M. Ferrario, E. Wanke, P. Frölichsthal-
Schoeller, L. Cova, M. Arcellana-Panlilio, A. Colombo, et al. Isolation and cloning of
multipotential stem cells from the embryonic human cns and establishment of transplantable
human neural stem cell lines by epigenetic stimulation. Experimental neurology,
156(1):71–83, 1999.
[513] A. Villa, E. Snyder, A. Vescovi, and A. Martínez-Serrano. Establishment and properties
of a growth factor-dependent, perpetual neural stem cell line from the human
cns. Experimental neurology, 161(1):67–84, 2000.
[514] R. Vlietstra, D. van Alewijk, K. Hermans, G. van Steenbrugge, and J. Trapman.
Frequent inactivation of pten in prostate cancer cell lines and xenografts. Cancer
Research, 58(13):2720–2723, 1998.
[515] B. Vogelstein, D. Lane, A. Levine, et al. Surfing the p53 network. Nature,
408(6810):307–310, 2000.
[516] A. Volterra. Astrocytes, from brain glue to communication elements: the revolution
continues. Nature Reviews Neuroscience, 2005.
[517] A. Von Deimling, A. Korshunov, and C. Hartmann. The next generation of glioma
biomarkers: Mgmt methylation, braf fusions and idh1 mutations. Brain Pathology,
21(1):74–87, 2011.
[518] V. Vukicevic, A. Jauch, T. Dinger, L. Gebauer, V. Hornich, S. Bornstein, M. Ehrhart-
Bornstein, and A. Müller. Genetic instability and diminished differentiation capacity in
long-term cultured mouse neurosphere cells. Mechanisms of ageing and development,
131(2):124–132, 2010.
[519] S. Wada, M. Hamada, K. Kobayashi, and N. Satoh. Novel genes involved in canonical
wnt beta-catenin signaling pathway in early ciona intestinalis embryos. Development,
growth & differentiation, 50(4):215–227, 2008.
[520] R. Wang, K. Chadalavada, J. Wilshire, U. Kowalik, K. Hovinga, A. Geber, B. Fligelman,
M. Leversha, C. Brennan, and V. Tabar. Glioblastoma stem-like cells give rise
to tumour endothelium. Nature, 468(7325):829–833, 2010.
[521] T. Watanabe, T. Katagiri, M. Suzuki, F. Shimizu, T. Fujiwara, N. Kanemoto, Y. Nakamura,
Y. Hirai, H. Maekawa, and E. Takahashi. Cloning and characterization of two
novel human cdnas (nell1 and nell2) encoding proteins with six egf-like repeats. Genomics,
38(3):273–276, 1996.
[522] T. Watanabe, S. Nobusawa, P. Kleihues, and H. Ohgaki. Idh1 mutations are early
events in the development of astrocytomas and oligodendrogliomas. The American
journal of pathology, 174(4):1149, 2009.

[523] T. Watanabe, A. Takeda, T. Tsukiyama, K. Mise, T. Okuno, H. Sasaki, N. Minami,
and H. Imai. Identification and characterization of two novel classes of small rnas
in the mouse germline: retrotransposon-derived sirnas in oocytes and germline small
rnas in testes. Genes & development, 20(13):1732, 2006.
[524] T. Watanabe, Y. Totoki, A. Toyoda, M. Kaneda, S. Kuramochi-Miyagawa, Y. Obata,
H. Chiba, Y. Kohara, T. Kono, T. Nakano, et al. Endogenous sirnas from naturally
formed dsrnas regulate transcripts in mouse oocytes. Nature, 453(7194):539–543, 2008.
[525] S. Weiss, C. Dunne, J. Hewson, C. Wohl, M. Wheatley, A. Peterson, and B. Reynolds.
Multipotent cns stem cells are present in the adult mammalian spinal cord and ventricular
neuroaxis. The Journal of neuroscience, 16(23):7599, 1996.
[526] P. Wen and S. Kesari. Malignant gliomas in adults. New England Journal of Medicine,
359(5):492–507, 2008.
[527] W. Wick, U. Naumann, and M. Weller. Transforming growth factor-beta: A molecular
target for the future therapy of glioblastoma. Current pharmaceutical design,
12(3):341–349, 2006.
[528] E. Wijaya, M. Frith, Y. Suzuki, and P. Horton. Recount: expectation maximization
based error correction tool for next generation sequencing data. In Genome Inform,
volume 23, pages 189–201, 2009.
[529] R. Williams and K. Herrup. The Control of Neuron Number. Annual Review of
Neuroscience, 11(1):423–453, 1988.
[530] P. Wolters, M. Laig-Webster, and G. Caughey. Dipeptidyl peptidase i cleaves matrixassociated
proteins and is expressed mainly by mast cells in normal dog airways.
American journal of respiratory cell and molecular biology, 22(2):183, 2000.
[531] L. Xu, Y. Shi, G. Petrovics, C. Sun, M. Makarem, W. Zhang, I. Sesterhenn,
D. McLeod, L. Sun, J. Moul, et al. Pmepa1, an androgen-regulated nedd4-binding protein,
exhibits cell growth inhibitory function and decreased expression during prostate
cancer progression. Cancer research, 63(15):4299, 2003.
[532] X. Xu, J. Zhao, Z. Xu, B. Peng, Q. Huang, E. Arnold, and J. Ding. Structures of human
cytosolic nadp-dependent isocitrate dehydrogenase reveal a novel self-regulatory
mechanism of activity. Journal of Biological Chemistry, 279(32):33946–33957, 2004.
[533] V. Yadav and M. Denning. Fyn is induced by ras/pi3k/akt signaling and is required
for enhanced invasion/migration. Molecular carcinogenesis, 50(5):346–352, 2011.
[534] K. Yamada and M. Watanabe. Cytodifferentiation of Bergmann glia and its relationship
with Purkinje cells. Anatomical science international, 77(2):94–108, 2002.
[535] H. Yan, D. Parsons, G. Jin, R. McLendon, B. Rasheed, W. Yuan, I. Kos, I. Batinic-
Haberle, S. Jones, G. Riggins, et al. Idh1 and idh2 mutations in gliomas. New England
Journal of Medicine, 360(8):765–773, 2009.
[536] J. Yan, L. Xu, A. Welsh, G. Hatfield, T. Hazel, K. Johe, and V. Koliatsos. Extensive
neuronal differentiation of human neural stem cell grafts in adult rat spinal cord. PLoS
medicine, 4(2):e39, 2007.
[537] K. Yap, S. Li, A. Muñoz-Cabello, S. Raguz, L. Zeng, S. Mujtaba, J. Gil, M. Walsh,
and M. Zhou. Molecular interplay of the noncoding rna anril and methylated histone
h3 lysine 27 by polycomb cbx7 in transcriptional silencing of ink4a. Molecular cell,
38(5):662–674, 2010.
[538] S. Yekta, I. Shih, et al. Microrna-directed cleavage of hoxb8 mrna. Science,
304(5670):594, 2004.
[539] Q. Ying and A. Smith. Defined conditions for neural commitment and differentiation.
Methods in enzymology, 365:327–341, 2003.
[540] A. Yool. Aquaporins: multiple roles in the central nervous system. The Neuroscientist,
13(5):470, 2007.
[541] H. You, K. Yamamoto, and T. Mak. Regulation of transactivation-independent
proapoptotic activity of p53 by foxo3a. Proceedings of the National Academy of Sciences,
103(24):9051, 2006.

[542] J. Yu, K. Ohuchida, K. Nakata, K. Mizumoto, L. Cui, H. Fujita, H. Yamaguchi,
T. Egami, H. Kitada, M. Tanaka, et al. Lim only 4 is overexpressed in late stage
pancreas cancer. Mol Cancer, 7:93, 2008.
[543] X. Yuan, J. Curtin, Y. Xiong, G. Liu, S. Waschsmann-Hogiu, D. Farkas, K. Black, and
J. Yu. Isolation of cancer stem cells from adult glioblastoma multiforme. Oncogene,
23(58):9392–9400, 2004.
[544] S. Yun, K. Byun, J. Bhin, J. Oh, L. Nhung, D. Hwang, and B. Lee. Transcriptional
regulatory networks associated with self-renewal and differentiation of neural stem
cells. Journal of cellular physiology, 225(2):337–347, 2010.
[545] Z. Zador, O. Bloch, X. Yao, and G. Manley. Aquaporins: role in cerebral edema and
brain water balance. Progress in brain research, 161:185–194, 2007.
[546] D. Zagzag, K. Salnikow, L. Chiriboga, H. Yee, L. Lan, M. A. Ali, R. Garcia, S. Demaria,
and E. W. Newcomb. Downregulation of major histocompatibility complex
antigens in invading glioma cells: stealth invasion of the brain. Laboratory investigation;
a journal of technical methods and pathology, (3):328–41.
[547] X. Zhang, Z. Lian, C. Padden, M. Gerstein, J. Rozowsky, M. Snyder, T. Gingeras,
P. Kapranov, S. Weissman, and P. Newburger. A myelopoiesis-associated regulatory
intergenic noncoding rna transcript within the human hoxa cluster. Blood,
113(11):2526–2534, 2009.
[548] S. Zhao, Y. Lin, W. Xu, W. Jiang, Z. Zha, P. Wang, W. Yu, Z. Li, L. Gong, Y. Peng,
et al. Glioma-derived mutations in idh1 dominantly inhibit idh1 catalytic activity and
induce hif-1{alpha}. Science Signalling, 324(5924):261, 2009.
[549] H. Zheng, H. Ying, H. Yan, A. Kimmelman, and D. Hiller. p53 and Pten control
neural and glioma stem/progenitor cell renewal and differentiation. Nature, 2008.
[550] Y. Zhu, P. Ghosh, P. Charnay, D. Burns, and L. Parada. Neurofibromas in nf1:
Schwann cell origin and role of tumor environment. Science, 296(5569):920, 2002.
[551] Z. Zhuang, P. Jian, L. Longjiang, H. Bo, and X. Wenlin. Oral cancer cells with
different potential of lymphatic metastasis displayed distinct biologic behaviors and
gene expression profiles. Journal of Oral Pathology & Medicine, 39(2):168–175, 2010.
[552] G. Zupanc and S. Clint. Potential role of radial glia in adult neurogenesis of teleost
fish. Glia, 43(1):77–86, 2003.