Prevalence of somatic alterations in the colorectal cancer cell genome

To explore the frequency of somatic alterations in coding regions of tumor DNA, we selected a total of 504 genes for analysis. By using a combination of the public (19) and private (Celera; ref. 20) genome databases, we extracted available genomic data for a total of 2,206 exons contained in these genes and designed primers for PCR amplification and sequencing. The sequences of these genes were determined in a panel of 12 colorectal cancer cell lines. Cell lines rather than primary tumors were used to ensure that nonneoplastic cells within the tumors did not complicate the sequence analysis. These cell lines were generally of early passage, and each was extensively examined with several different kinds of microsatellites to ensure that no MIN was present, thus excluding mismatch repair deficiency as a source of potential somatic alterations in these cases. The cell lines also were chosen to contain regions of loss of heterozygosity (LOH) in areas containing genes of interest to increase the sensitivity of detecting somatic mutations by DNA sequencing, as any alteration observed would appear as a homozygous change. The exons described above were amplified from tumor DNA samples using PCR and were subjected to dideoxy sequencing. A total of 1,811 exons were successfully amplified from 470 genes and sequenced from an average of 11 different tumor DNA samples (Table 1). The average size of examined exons was 150 bp and ranged up to 699 bp in length. Both intronic splice acceptor and donor regions were analyzed, with the exception of cases where the exon size was greater than the length of readable sequencing product. A total of ≈6.5 Mb of sequence data was obtained in this way (Table 1).

Sequencing results were analyzed by assembling data of related exons from different tumors and comparing them in silico to reference sequences (Fig. 1). We confined our analysis to 3.2 Mb that encoded exons and focused on protein altering (nonsynonymous) changes, as these provided the best indicators of alterations most likely to have an effect on tumorigenesis. A total of 320 alterations resulting in nonsynonymous changes were identified in the 3.2-Mb assembly. Observed changes first were compared with polymorphism databases to determine whether they corresponded to previously recognized single nucleotide polymorphisms (SNPs). A total of 90 SNPs were identified and excluded in this manner. The remaining changes, representing potential somatic alterations, then were analyzed by sequencing of DNA from normal tissues of the corresponding patients. Two hundred and twenty seven of these changes were present in the corresponding patient's normal tissues. These changes thus represented previously unidentified SNPs within exons.

After analysis of all potential changes by sequencing of both tumor and matched normal DNA samples, only three somatic alterations were identified, each in a different gene (Table 2). Two of these corresponded to missense changes in well characterized genes: a G to T transversion at nucleotide position 799 of the eukaryotic translation initiation factor 3 subunit 1 gene, resulting in an aspartate to tyrosine change at amino acid position 267 (Fig. 2), and a C to T transition at nucleotide position 46 of the neurofilament 3 gene, resulting in an arginine to tryptophan change at amino acid position 16 (Fig. 3). The third alteration corresponded to a 14-bp deletion in a gene of unknown function (Celera hCT1843352) at nucleotide position 154, resulting in a frameshift and introduction of premature translational termination codon at amino acid position 54 (Fig. 4). All three changes occurred in different tumor samples, and no mutations in the remaining exons of these genes were detected in the ≈10 additional tumors examined.

The identification of three somatic alterations from 3.2 Mb of analyzed DNA corresponds to a somatic mutation frequency of approximately one alteration (0.22–2.5 alterations, 95% probability interval) per Mb of tumor DNA, or approximately 3,000 alterations (668–7450 alterations, 95% probability interval) per haploid genome. However, this represents an underestimate of the number of somatic changes in a tumor genome, as we examined only nonsynonymous changes. On the basis of genome-wide polymorphism studies (21, 22), it has been shown that nonsynonymous and synonymous changes have similar frequencies in coding DNA, and that the frequency of coding and noncoding changes is approximately the same. Therefore, one would expect that our somatic mutation frequency would be increased approximately 2-fold. This probably represents the higher end of such an estimate, because approximately two-thirds of random coding mutations alter an amino acid, and the selection pressure on these nonsynonymous changes is likely to be substantially less at the single-cell level than at the organismal level. Accordingly, the majority of cancers would be expected to carry less than 6,000 alterations per genome.

The total somatic alterations in a cancer represent those that accumulated during normal somatic cell division as well as in the tumor during successive waves of clonal expansion and growth. Each of the bottlenecks that the tumor cell passes through (including initiation plus each wave of clonal expansion) leads to fixation of all mutations that have previously occurred in the tumor cell's progenitors. Colorectal cancers are thought to develop from epithelial stem cells residing in colonic crypts (23–25). As the entire intestinal epithelium replaces itself every few days, a typical colon epithelial stem cell is thought to undergo >100 cell divisions per year and >4,000 divisions before the onset of most adenomas at age 40. Additionally, as cell turnover far exceeds net tumor growth (2, 26, 27), during the 20-yr progression of a colorectal carcinoma, at least another 2,000 cell divisions would occur. By using the estimated mutation rate in normal somatic human cells of 1 × 10⁻¹⁰–1 × 10⁻⁹ nucleotides per cell per division (1, 28, 29), one would expect that a 20-year old tumor removed from a 60-year old individual would have accumulated 1,800–18,000 alterations per haploid genome. This number of mutations is similar to the prevalence of alterations we observed in sporadic cancers and suggests that normal mutation rates are sufficient to explain tumor progression in most cancers.

Like all large-scale studies, our approach has several limitations. First, the majority of genes analyzed, including the three that contained somatic alterations, were present in regions that had undergone LOH in the tumor DNA. Although there is no current evidence to suggest a different rate of point mutations in regions of LOH, it is possible that our conclusions would be altered by analyses of additional genes from heterozygous regions. Second, our analysis was performed on cell lines rather than primary tumors, which may allow for accumulation of mutations during in vitro passaging. However, all of the mutations observed were clonal in the tumors analyzed, making it unlikely that they developed during the few in vitro passages of the cell lines. This possibility, even if true, would only decrease the actual number of mutations calculated to be present in the tumors and, therefore, would not impact on the major conclusions of this study. Third, if we assume a hierarchical model of epithelial cell generation in colonic crypts rather than the conventional single stem cell model (30), only ≈50 instead of 4,000 cell divisions would occur in an average tumor progenitor cell, and the number of expected mutations per tumor genome would be reduced from a range of 1,800–18,000 to 600–6,000. Although the current experimental evidence does not support such a model (25), the reduced number of mutations remains within several-fold of our observed mutation rate within tumors and would suggest at best a weak increase in mutation rates in cancer cells.

Despite these potential limitations, it is clear that these results have several significant implications. One implication is that they do not disprove the argument that genetic instability is an inherent feature of tumorigenesis. Instead, they emphasize that, although a subtle instability at the nucleotide level does not play a role in most colorectal cancers, a different form of genetic instability generally occurs in these tumors. In fact, extensive studies of polymorphic markers in our panel of tumors showed that they each had numerous allelic losses, consistent with widespread chromosomal changes (31). Additionally, the results highlight the importance of specific mutations in the genes that drive neoplasia. Because the studies described above show that passenger mutations (i.e., random mutations, with no functional significance) are found so rarely in these tumors, great confidence can be placed in the significance of somatic mutations in genes like APC and p53, which occur in nearly all of the colorectal cancers represented in our panel. Lastly, the results are important for interpreting the significance of mutations in candidate tumor-suppressor genes. The search for other tumor-suppressor genes involved in common cancers continues to be active, and the best evidence to validate such candidates comes from mutational data (32). On the one hand, our data show that mutations in any given gene are likely to be uncommon unless selected for during tumorigenesis. On the other hand, even random genes like those studied here are occasionally mutated in these cancers as a result of normal mutational processes and bottlenecks during the neoplastic process. A prudent conclusion from these results, therefore, would be to invoke the requirement for a significantly higher mutation frequency—for example, at least 2 independent examples of functionally altering mutations from a panel of no more than 20 tumors—to implicate a gene as a candidate tumor suppressor. To help investigators rigorously determine whether the number of observed mutations in a gene of interest is significantly above the expected background mutation frequency, we have developed simple statistical tools, freely available at http://astor.som.jhmi.edu/∼gp/trab/.

We thank members of the Molecular Genetics Laboratory for helpful discussions. This work was supported by the Maryland Cigarette Restitution Fund, the Clayton Fund, and National Institutes of Health Grants CA62924, CA57345, and CA43460.