Cancer Immune Therapy Edited by G. Stuhler and P. Walden ...

4.3 Approaches to the Molecular Identification of (CTL)-defined Tumor Antigens
teins in the cytosolic compartment and transport of peptides across the membrane
of the endoplasmic reticulum. Virtually any protein that is synthesized in the cell
may be the source of peptides for association with MHC class I and b2-microglobulin
in the lumen of the endoplasmic reticulum. The rules governing the interaction between
antigenic peptides and MHC class I molecules are now well understood and
supported by abundant structural data, including relatively numerous crystal structures.
Recently, two MHC class I molecule/tumor antigenic peptide complexes have
been crystallized [34, 35]. Simple peptide-binding motifs have now been defined for
numerous MHC class I alleles. They contain three major components: a defined
length and two preferred amino acid residues at fixed positions within the short peptide.
The latter are referred to as anchor residues. Additional refinements can be introduced
such as secondary residues that influence binding either positively or negatively.
Thus, it became obvious that one could predict the set of peptides with the
ability to bind to a given MHC class I allele by scanning the gene product sequence
of interest with appropriate algorithms. This procedure had the advantage to be independent
of the availability of pairs of specific CTL clones and autologous tumor cell
lines and became known as ªreverse immunologyº [36, 37].
Although straightforward, two important limitations in the reverse immunology approach
have been identified. First, it turned out that only one in four predicted peptides
do actually bind efficiently to the MHC class I molecule of interest. Additional
refinements to the prediction algorithms allow to improve the accuracy of MHC
binding scores and to reduce the number of peptides to be tested in binding assays.
Some of these good binder peptides can efficiently induce peptide-specific CTL upon
in vitro stimulation of peripheral blood mononuclear cells (PBMC). However, the
second limitation relates to the generation of the predicted peptide by the antigenprocessing
machinery of the cell. Indeed, some predicted good binder peptides have
been shown to be either poorly generated by the proteasome or destroyed upon proteolytic
degradation by the proteasome [38, 39]. In addition, the proteasome and immunoproteasome
may significantly differ in their ability to modulate the generation
of antigenic peptides [40]. These results impose the need to incorporate the new dimension
of the ªprocessabilityº of candidate peptides in the prediction algorithms.
This is not an easy task as the factors involved are only partially understood at present.
It seems that the main parameter in this regard is the proteasome itself. Efficient
prediction of CTL epitopes from the PRAME tumor-associated protein was demonstrated
using a combination of peptide-binding motifs and experimental digestion
of long peptides by purified proteasome preparations [41]. Attempts are underway
to define algorithms incorporating these new elements in the prediction process
([42, 43] and http://www.syfpeithi.de). Clearly, this type of approaches hold great promise
to mine new tumor antigens from the human genome which can be expected
to be completed in the foreseeable future [44].
New technologies are constantly applied to the task of tumor antigen identification.
For instance, in an effort to obviate both the need for large numbers of tumor cells
(biochemical identification of the antigenic peptide) and of cloning the gene encoding
the tumor-associated antigen (genetic approaches), randomized and combinatorial
peptide libraries have been used to probe the specificity of human tumor-reactive T
43