Problems encountered in compiling the census

在癌症统计调查时会遇到的问题

We have attempted to be comprehensive in compilation of the cancer-gene list, but have also been conservative in our criteria for inclusion. Several genes that were considered were ultimately excluded. Below, we have reviewed some common problems that we encountered in determining whether a gene should be included as a cancer gene.

Genes with few reported mutations. All classes of mutation occur as more or less random processes in both normal and neoplastic cells. Indeed, it is likely that most somatic mutations do not confer a clonal growth advantage. So, cancer genomes carry several 'PASSENGER' OR 'BYSTANDER' MUTATIONS, in addition to the mutations that cause the neoplastic phenotype. Unfortunately, the prevalence of passenger mutations in the genomes of most cancer types is not known. Therefore, if somatic mutations in a putative cancer gene are found in a very small proportion of tumours, it is difficult to exclude the possibility that they represent chance clustering of passenger alterations. Consequently, we have excluded genes in which fewer than five unambiguous somatic mutations have been reported in primary neoplasms.

Cancers with large numbers of mutations. A particular problem arises in cancer cells that have genomes with a high prevalence of somatic mutations. For example, cancers with mismatch-repair (MMR) gene defects carry tens of thousands of small insertions and deletions in short tandem repeats (known as microsatellites)⁹. Most of these occur in intronic or intergenic DNA and are, therefore, almost certainly passenger mutations. However, several hundred are in coding sequences and will often result in translational frameshifts, with the consequence of premature protein termination or nonsense-mediated RNA decay. It is highly likely that a small percentage of these are causally involved in oncogenesis. It is equally probable, however, that most are not. In these circumstances, distinguishing mutated cancer genes from genes with clusters of passenger mutations is particularly problematic.

Several genes with small insertions and deletions of coding microsatellites/mononucleotide repeats that lead to protein truncation in MMR-deficient tumours have been proposed as cancer genes⁹. Attempts have been made to distinguish mutations that confer a clonal growth advantage from bystander mutations in MMR-deficient cells¹⁰. Transforming growth factor- type II receptor (TGF- RII), which mediates the effects of the growth-suppressing ligand TGF-, is one example of a potential cancer gene that was discovered in MMR-deficient cells. Mutations in TGF-RII are believed to be more than simple bystander mutations because of the biological function of its gene product, the high frequency of biallelic mutations in MMR-deficient cells and the presence of some biologically relevant somatic mutations in MMR-proficient cancers¹¹. However, the distinction is still unclear for most genes that have microsatellite instability in their coding sequence. Therefore, genes in which somatic mutations occur exclusively or predominantly in MMR-deficient cells have been excluded from the cancer-gene census.

Putative cancer genes with particularly high mutation rates in cancer cells. A similar problem arises when there is an increased mutation rate restricted to a small region of the genome. For example, the fragile histidine triad gene (FHIT) and WW-domain-containing oxidoreductase (WWOX) genes each straddle a fragile site (FRA3B and FRA16D, respectively^{12, 13}). Fragile sites are genomic regions that are associated with a high frequency of chromosome breakage, and many were originally identified through chemical stress of normal lymphocytes¹⁴. The intrinsic fragility of these regions is frequently expressed in cancer cells as rearrangements and deletions, mutational patterns similar to those found in certain recessive cancer genes (for example, CDKN2A and PTEN^{15, 16}). Therefore, based simply on mutation clustering and pattern, it is not clear whether FHIT and WWOX are recessive cancer genes (and therefore causally implicated in oncogenesis) or fragile sites that are frequently rearranged in cancer cells because of defects in the DNA-repair process, or both. Most well-established recessive cancer genes also contain somatic base substitutions and small insertions/deletions that cause protein truncation. These are uncommon in FHIT and WWOX. So, although there are additional biological data that might support a role for genes such as FHIT and WWOX in oncogenesis, genes overlying fragile sites are not included in this census of cancer genes.

Mutations that encompass many genes. Most classes of mutations (such as base substitutions) only affect a single gene, or, at most, two genes (such as reciprocal chromosomal translocations). However, copy-number changes, such as gene amplification, can affect several megabases of DNA and encompass many genes. Therefore, solely on the basis of genetic evidence, it is not always clear which gene is the crucial target of the amplification, and it is conceivable that, in some cases, there is more than one target. Additional evidence that might be invoked in assessing which gene(s) on an amplicon mediates oncogenesis includes increased expression levels in cancer cells and functional effects. Chromosome defects, such as trisomy, that involve even larger regions of the genome pose even greater problems, as they might alter the copy number of thousands of genes. We have therefore only included in the census amplified genes for which there is reasonable consensus that the cancer-causing gene has been identified, although even these might be controversial. Many amplified genes that have not been included in the cancer-gene census clearly exist and might be included in the future.

Low-penetrance cancer-susceptibility genes. There is usually little ambiguity in the identification of mutated genes that are responsible for high-penetrance (high-risk) cancer-susceptibility syndromes or of mutated genes that are associated with characteristic non-neoplastic manifestations in addition to cancer predisposition. However, germline variants of many genes have recently been proposed as low-penetrance (low-risk) cancer-susceptibility alleles without additional non-neoplastic features. In some cases - for example, for the variants APC^*I1307K¹⁷ and CHK2^*1100delC¹⁸ - the evidence for this effect is strong. For most genes, however, the evidence is statistically weak and often conflicting. Therefore, for purposes of clarity, we have not included this type of cancer-susceptibility gene in the current census.