The Genom of Homo sapiens.pdf

200 HAYASHIZAKIto known genes. As shown in Figure 6, 20% of all FAN-TOM2 FL cDNA sequences are identical to knownmouse transcripts, and 14% show various levels (85%,70%, and 50%) of homology with protein sequences fromother organisms. Surprisingly, the remaining FANTOM2sequences are novel. These novel sequences can be classifiedinto four categories. The first category consists ofsequences containing InterPro motifs (Apweiler et al.2001), MDS domains (Kawaji et al. 2002), or SCOPstructural domains (Gough et al. 2001). The second categoryconstitutes sequences with ORFs of significant size(more than 100 amino acids) that code for hypotheticalproteins but contain no recognizable motifs or domains.The sequences in the third category have no ORFs but hybridizeto ESTs reported in public databases. The finalcategory contains totally unknown sequences with noORFs, known motifs, or homology with ESTs.The first step in FANTOM annotation was identificationof ORFs in each sequence. However, sequences withno ORFs were found more frequently than expected.FANTOM2 cDNAs contained 20,487 protein-coding sequencesand 16,599 noncoding sequences; this is surprising,given that only around 100 non-coding RNAs (otherthan tRNAs) were known before FANTOM2 was developed(Fig. 6). A new “continent” of functional noncodingRNAs has been discovered; previously the protein continenthas been better explored in the examination of finalgene expression. Artifacts including unspliced cDNAs(despite efforts to prepare cytoplasmic RNA only) andgenomic contamination were present among 16,599 noncodingRNAs. A significant population of these noncodingRNAs are likely not to be “junk”: On average, eachnoncoding RNA is spliced from three exons, and CpG islandsor CG-rich sequences are preferentially located atthe 5´ ends of the noncoding RNAs. Since 32,000 protein-codingsequences are predicted from the completedhuman genome sequence, more than several thousandprotein-coding sequences are missing from FANTOM2.However, the prediction of so-called “genes” from thehuman genome sequence missed a major population (atleast several thousand) of noncoding RNAs. Analysis ofthe functions of these noncoding RNAs is an importanttask for future research.LARGE-SCALE MAPPING OF FL cDNA ONTOTHE MOUSE GENOME SEQUENCETo clarify the chromosomal distribution of the transcripts,we mapped the FL cDNAs onto the mousegenome draft sequence (MGSCv3) (Waterston et al.2002). Mapping was possible in 32,568 out of 33,047cDNA sequences. The remaining 479 sequences did notshow any homology with mouse genome sequences. Onepossible explanation for this is the exclusive use of the femalemouse in generating the genome database, thus thedatabase does not contain the Y chromosome sequence.Additionally, the mouse genome sequence database isstill only in draft form, with 3% of the genome currentlyunsequenced. The data on the chromosomal locations ofthe FL cDNAs constitute a powerful tool that will facilitatethe positional candidate approach to identifying thegenes responsible for specific phenotypes in mutant miceand in human disease. We developed a computer softwarepackage, GENOMAPPER, that can list candidategenes when given the flanking marker, expression sites,and, if possible, protein–protein interactions.SENSE AND ANTISENSE RNA PAIRSIn the FANTOM2 set, we found an unexpectedly largenumber of pairs of sense and antisense transcripts. Thesewere discovered through the mapping of cDNA onto thegenomic sequence. The pairs include all combinations ofcoding and noncoding RNAs, and spliced and unsplicedsequences. Natural sense and antisense transcripts mayregulate gene expression in various ways. For example,the antisense imprinter RNA (AIR) was reported as thenoncoding, intronless transcript controlling the imprintedexpression of Igf2r (Sleutels et al. 2002). When AIR isdisrupted, imprinted expression is heavily perturbed. Insome of the sense–antisense pairs, antisense-strand RNAmay function to repress the function of the sense-strandRNA by RNAi, the suppression of translation, or othermechanisms. Investigation of the functions of sense–antisensepairs should be an interesting direction for futureresearch.CDS ANNOTATION AND ANALYSIS OFPROTEIN-CODING SEQUENCEThe first step of annotation, the identification of thecoding sequence, addresses many questions. Is there asignificant ORF? Is this ORF protein coding or noncoding?If it is protein coding, is the transcript spliced?Which ORF is the real one? Are there any frameshift errors,or initiation and termination codon errors? Is thecDNA full-length or truncated? Which ATG is the initiationcodon? We developed a computer software package,Protein Coding Region Estimator (ProCrest), to answerthese questions (Y. Hayashizaki et al., unpubl.). Thissoftware calculates the amino acid frequency, tandemamino acid weight matrix, degeneracy of genetic code,tRNA anticodon usage (wobble rules: [U, C], A, G), biasof base contents (G,C and A,G), Kozak consensus, andpolyadenylation sites for a given sequence.Using this software, we identified sequences as proteincoding or noncoding and identified CDSs. With the resultingCDS database, functional analysis of the proteinsequences predicted by ProCrest was possible. Gene ontology(GO) analysis, motif analysis with Inter Pro and/orPfam, and gene family analysis were also performed onthe CDSs. The software developed by Matsuda et al.(Kawaji et al. 2002) was used to find new motifs in FAN-TOM2 clones, resulting in the discovery of 10 new putativemotifs (Okazaki et al. 2002).DYNAMIC VARIATION OF TRANSCRIPTSPRODUCED FROM STATIC GENOMESEQUENCE