Low Mutation Incidence in Polymorphic Noncoding Short Mononucleotide Repeats in Gastrointestinal Cancer of the Microsatellite Mutator Phenotype Pathway¹

A widespread genomic instability known as MSI³ is germane to the pathogenesis of some colon cancers (1, 2, 3). MSI is the landmark of sporadic and hereditary gastrointestinal cancer of the MMP pathway (4). These hundreds of thousands of microsatellite contraction-expansion mutations originate by the genetic or epigenetic inactivation of the DNA MMR machinery (5, 6). Once the mutator phenotype unfolds in precursor normal stem cells, cancer of the MMP pathway arises after mutations in genes involved in cell growth or survival (“cancer genes”) drive tumor development and progression (1, 4, 7).

Implicit in this concept is the assumption that these tumors display distinctive features in genotype and phenotype because they harbor a defined spectrum of mutated cancer genes that are specific targets for MMP (4). However, the determination of these target cancer genes has been difficult (8, 9). The exacerbated mutator phenotype of these tumors generates many neutral and irrelevant mutations, complicating the task of distinguishing cause from consequence. One of the criteria commonly used relies on the presence of a significant prevalence of frameshift mutations in SMT present in some cancer genes (10, 11). This criterion is sustained by the observation that these mutations are rare in similar noncoding SMTs and in other genes without obvious links to oncogenesis (10, 11, 12).

This interpretation has been challenged in a recent issue of Cancer Research by Zhang et al. (13). The authors analyzed the incidence of mutations in several nonfunctional short mononucleotide repeat sequences in DNA MMR-deficient colon cancer cell lines and mouse xenografts. The mutability of these noncoding SMTs exhibited high variability. Whereas some of the loci analyzed exhibited a low mutation frequency (from 0–∼5%), several of the mononucleotide loci analyzed exhibited mutation incidences that were surprisingly high (25–54.2%).

These findings have important implications for understanding the mechanisms underlying cancer pathogenesis of the MMP pathway. If some noncoding SMTs are mutated in up to 50% of the tumors, the functionality of less prevalent frameshifts in similar SMTs within some cancer genes would be seriously questioned. Zhang et al. (13) thus caution that “a significant prevalence of mutation in a given gene in MSI cancers is not a reliable indicator that such genes are targets rather than passengers, even when the mutated tract is small.”

In our experience, noncoding repeats that are eight nucleotides long display very few mutations in MMP tumors (11, 12). Consequently, we determine in our panel of primary tumors the mutation incidence of the SMTs Zhang et al. reported to undergo the highest incidence of mutations in the tumor cell lines and mouse xenografts that they analyzed (13). We have not confirmed these findings, and instead, we found a high frequency of length polymorphisms in many of these SMTs.

Tumor samples were obtained as frozen specimens from the Southern Division of the Cooperative Human Tissue Network (University of Alabama, Birmingham, AL), the National Cancer Center Research Institute (Tokyo, Japan), and Sapporo Medical University (Sapporo, Japan). The origin of the colorectal tumor samples has been described previously (11, 12). Genomic DNA was extracted with phenol-chloroform and diluted to a concentration of 20 ng/μl before PCR amplification.

The sequence of the PCR primers for seven of the chromosome 22 mononucleotide loci reported by Zhang et al. (13) to exhibit the highest mutation incidence were as follows: SMT4, 5′-CCTAGGTTGTGGGTGTATG-3′ and 5′-CCTACTCCAGTGTGGTCG-3′; SMT6, 5′-GAGAGCATAAGTCACTCAAC-3′ and 5′-CACTAGAAATTGCTGAGCCAG-3′; SMT14, 5′-CCAAGGACCACGCATCTAC-3′ and 5′-TTCCCTTGGCGCCTCACTG-3′; SMT15, 5′-GAGAGAGATAGTGGAAGG-3′ and 5′-CAGGGATGGCTACATAATTTG-3′; SMT16, 5′-CAGTGGAAATTGTTCGCC-3′ and 5′-CACCAGTGACTTACATCAC-3′; SMT28, 5′-TGCACAGGTTCCACCCTCC-3′ and 5′-CCCCATTCTGTCCTGGCC-3′; and SMT29, 5′-GAGATGTACAGCTCAACTC-3′ and 5′-TTACTATTGATGTGGCTGGG-3′.

PCR was carried out with Vent DNA polymerase (New England Biolabs, Beverly, MA) and AmpliTaq DNA polymerase (Applied Biosystems, Foster City, CA) for 35 cycles in the presence of 0.1 μCi of [³²P]αdCTP. A four-stage touch-down protocol was performed following the conditions described by Zhang et al. (13). PCR products were electrophoresed in denaturing 6% polyacrylamide gels (National Diagnostic, Atlanta, GA). The gels were dried on filter paper and subjected to autoradiography.

Sequencing was performed as described previously (11). The PCR products were eluted from the gels and amplified. DNA was also reamplified and purified with a QIAquick PCR purification kit (Qiagen, Valencia, CA) and sequenced with the ABI PRISMTM dye terminator cycle sequencing kit (Perkin-Elmer, Foster City, CA).

To estimate the allelic status of the SMT mutations, we assessed the extent of contaminating normal tissue in the tumor specimens by comparing the relative intensity of the bands corresponding to the wild-type and mutant alleles of BAT26 and APΔ3 (1) mononucleotide repeats (see Fig. 1 and data not shown). The implicit assumption is that, in MMP tumors, contractions of these long mononucleotide repeats occur in both alleles (1, 4, 7, 9, 10, 11, 12). The amplified wild-type allele derives thus from the normal tissue present in the surgical tumor specimens. When the intensities of bands corresponding to mutant and wild-type alleles were similar, and the tumors had little contaminating normal tissue, the mutations were considered to be clonal. When the mutant band was fainter (less than ∼20%) than the wild type and the tumor did not have much contaminating normal tissue (no more than ∼25%), the mutations were not considered to be clonal (i.e., present in only a fraction of the tumor cells).

When the extent of normal tissue contamination is significant (some tumors contained ∼50% normal tissue, i.e., as in cases 584, 612, 613, 672, and 677; see Fig. 1), the determination of mutation clonality is more difficult. Nevertheless, comparison of the relative intensity ratio of mutant/wild-type bands between long and short mononucleotide tracts can often unambiguously determine whether or not the mutation is clonal. In tumors with ∼50% normal tissue, a clonal monoallelic mutation should be present in ∼25% of the DNA molecules, which therefore would yield a band ∼1/3 the intensity of the wild-type. The situation would become more complicated if the amplification efficiency were influenced by the length of the repeat, but in the SMTs, mutated sequences differed only by one or two nucleotides, and this difference did not significantly affect amplification efficiency. Another complication arises from situations where the number of nucleotides deleted in the repeat is small and the BAT 26 PCR pattern does not resolve the wild-type and mutant alleles. In this instance, sometimes tumors that appear to be heavily contaminated with normal tissue may in fact have only minor content of normal tissue. An example is tumor 612 (Fig. 1) where the extent of contaminating normal tissue was determined to be minor by analysis of the PCR pattern of dinucleotide repeat loci (which revealed a clear expansion tendency for these repeated sequences).

Fig. 1 shows the PCR amplification of seven noncoding short mononucleotide tract loci from matched normal and tumor tissues of 14 colon cancer patients. These tumors were all positive for contractions of BAT26, a noncoding long mononucleotide tract, used for the diagnosis of MSI (Fig. 1 top). The BAT26 amplification pattern shows that the contaminating normal tissue in the tumor samples was no more than ∼50%, therefore not preventing mutation detection. Only tumor 612 exhibited a deletion mutation in each SMT6 and SMT14 (indicated by solid triangles). The rest of the loci displayed no mutations (SMT29), or mutations that, in comparison to the BAT26 pattern, were not clonal (indicated by dots): SMT4, tumors 548 and 558; SMT10, tumor 595; SMT15, tumor 672; and SMT16, tumors 558 and 611.

The mutation patterns for three polymorphic SMT loci (SMT15, SMT16, and SMT29) in 10 other colon cancers of the MMP are shown in Fig. 2. These tumors were found previously to contain frameshift mutations in BAX and TGFβRII genes (12, 14). Some of these tumors also contained frameshift mutations in the hMSH3 MMR gene (tumors 61, 211, 405, and 453). No mutations were detected in these SMT loci with the exception of two nonclonal mutations in the SMT15 locus (tumors 394 and 453). In contrast, these tumors displayed clonal mutations in TGFβRII (tumor 394) and in TGFβRII, BAX, and hMSH3 (tumor 453). Altogether, these results show that the lack of detecting clonal mutations in the noncoding SMT loci was not attributable to a selection of tumors with few mutations, or to a lack of sensitivity in mutation detection, because clonal mutations were detected in coding SMTs.

The summary of these results for a panel of 41 colon cancers of the MMP is shown in Table 1. The results are compared with the data from Zhang et al. (13). The mutation frequencies of noncoding SMT loci reported by Zhang et al. to reach 25–50% were significantly lower in our panel of tumors. Three of these loci, including the two with the highest mutation frequencies reported by Zhang et al. (SMT15 and SMT16, 45% and 54%, respectively) displayed no mutations at all, and the mutation frequency in the other five was much lower (2.9–6.7%).

These experiments revealed that four of these loci (SMT6, SMT15, SMT16, and SMT29) are polymorphic in the human population because their PCR patterns were typical of heterozygosity of loci differing in length by one nucleotide. In the experiment shown in Fig. 1, SMT6 was heterozygous for the (G)_8/7 alleles in cases 558, 590, 611, 668, and 677, whereas the rest of the cases were homozygous for only the (G)₈ allele. SMT15 was heterozygous for the (G)_8/9 alleles in individuals 602, 612, 613, and 672, whereas the rest were homozygous for only the (G)₈ allele. SMT16 was also polymorphic: case 547 was homozygous for the longer (G)₉ allele; individuals 611, 668, and 677 were heterozygous for the (G)_8/9 alleles; and the rest of the individuals were homozygous for the smaller (G)₈ allele. SMT29 was homozygous for the shorter (A)₈ allele in cases 547, 584, 590, 595, 611, and 612; homozygous for the longer (A)₉ allele in cases 558, 603, and 668; and heterozygous in individuals 548, 602, 613, 672, and 677.

As shown in Fig. 2, SMT15 was homozygous for the (G)₉ allele in case 353; homozygous for the (G)₈ allele in cases 61, 197, 315, 394, 437, and 453; and heterozygous for the (G)_9/8 alleles in individuals 211, 405, and 441. Three individuals (353, 394, and 453) were homozygous for the SMT16 (G)₉ allele, two were heterozygous for the (G)_9/8 alleles (61 and 211), and the rest were homozygous for the most prevalent (G)₈ allele. SMT29 was heterozygous for the (A)_9/8 alleles in three individuals (61, 315, and 441). The rest were homozygous for the (A)₈ allele.

The frequency of clonal somatic mutations in SMTs that we have observed in our primary tumors is significantly lower than that reported by Zhang et al. (13) for cell lines and mouse xenografts, which we suspect represent an overestimation of mutation frequency. In this particular context, a negative result (absence of mutations) is stronger evidence than the opposite positive result (presence of mutations). Evidence for such “mutations” may be attributable to various spurious reasons (9), including the mistaken identification of polymorphisms for mutations when no matching normal tissue DNA is analyzed. Of note, the differences between the reported mutations by Zhang et al. and the mutations combined with polymorphisms in our study lost statistical significance (Table 1).

In contrast to the polymorphic loci described above, we found other loci to be monomorphic in the nearly 50 individuals analyzed (Table 1): SMT4, SMT10, SMT14 (Fig. 1), and SMT28 (data not shown). The frequency of somatic mutations was not higher in polymorphic loci than in monomorphic loci, ruling out a possible link between germline and somatic hypermutability. Somatic mutations in noncoding SMTs were scarce in both the (A) and (G) tracts, and there was no significant difference in mutation frequency regardless of clonality status (Table 1).

The mutation frequency of these monomorphic SMTs was also significantly lower than that reported by Zhang et al. (13). We cannot explain the variance between mutation frequencies in these loci, variances that appear to be statistically significant (Table 1). Whether the discrepancy is attributable to the different types of specimens used (primary tumors versus cell lines and mouse xenografts) remains to be determined. For instance, enrichment attributable to “random drift” during the propagation in culture or growth in nude mice of nonclonal mutations initially present in primary tumors may account for these differences. The mutator phenotype persists in these tumor cells, which continue to rapidly accumulate mutations during propagation in vitro and in vivo (15, 16). Mutation frequencies higher in cell lines than in primary tumors has been described for some coding SMTs, and it has been suggested that cell lines cannot be used to asses the level of instability (17). A criterion for distinguishing clonal from nonclonal mutations, such as the one we have used by comparing the detected mutations with the PCR pattern of BAT26, seems thus pertinent for the classification of putatively relevant target gene mutations in MMP tumors.

This criterion does not disregard nonclonal coding mutations, but only considers the relevance of distinguishing causal from consequential mutations. Although nonclonal mutations in cancer genes or genes involved in genome integrity cannot be causally linked to tumor development, these mutations may still be relevant in tumor progression and in some aspects of the biology of these tumors. For instance, they may have an impact on response to chemotherapy treatment.

The results reported by Zhang et al. (13) with noncoding SMT mutations cast doubts over the significance in MMP tumorigenesis of coding SMT mutations. But our inability to reproduce these results show that the importance of somatic frameshifts in cancer genes with short mononucleotide repeats in tumors of the MMP pathway is not diminished. Mutations in genes such as BAX must be under selective pressure during tumorigenesis because equivalent neutral SMTs display much lower mutation incidences. However, we agree with Zhang et al. in that the criterion for functionality based solely on mutation incidence does not apply to tumors of the mutator phenotype.

Considerations based on mutation frequency alone (8, 17) lead to the need of setting arbitrary cutoff points: mutations are determined to be nonfunctional below the cutoff point, whereas functional above. As shown by Zhang et al., this cannot be ascertained a priori because the background mutation incidence depends upon each specific sequence context. The criticism is especially valid when putative “target” genes for the mutator phenotype exhibit relatively low mutation frequencies (17, 18, 19). We further agree with Zhang et al. in that diagnostic classification of potential target genes in MMP tumors needs to take into account functional criteria (20, 21, 22).

However, we disagree with the implication that a low incidence of mutations in MSI tumors negates their functionality. Cancer gene frameshifts in tumors of the MMP pathway may not be disregarded as nonfunctional (“passengers” in the nomenclature by Zhang et al.) even if their incidence is low. Because of their exacerbated mutator phenotype, tumors of the MMP pathway may develop and/or progress by the accumulation of mutations in multiple genes of the same oncogenic networks (14, 17, 18, 19), each of which may be infrequently mutated (Fig. 3).

A further complication in tumorigenesis of the mutator pathway arises from another paradoxical feature displayed by these tumors. Whereas noncoding (intronic) long mononucleotide repeats such as BAT 26 or APΔ3 are universally mutated by biallelic mutations, frameshift mutations in cancer genes with proven oncogenic roles in tumorigenesis, such as TGFβRII and BAX (Fig. 1) or p53 (1) and APC (unpublished data), are often monoallelic. However, the hypothesis that mutations need to be biallelic to be functional (Knudson’s two-hit model; Ref. 23) may not always apply to tumors of the mutator pathway, perhaps with the exception of the initial mutator mutations (i.e., hMLH1 or hMSH2). As shown in Fig. 3, these tumors may also progress by the accumulation in the same cells of monoallelic mutations in each of several genes of the same oncogenic networks (18, 19). The accumulation of heterozygous mutations presumably reduces the threshold amount of the corresponding negative cancer gene products below their suppressor levels.

Because this model needs experimental validation, criteria for diagnosis of proven target genes for the mutator phenotype are difficult to define at present, in the absence of functional in vitro and in vivo evidence. This is a forced conclusion because of the peculiar features of tumors with MMR deficiency, both in genotype and phenotype. But there is no reason to disregard mutations in genes with a proven oncogenic role, just because they occur with relatively low frequencies. Otherwise, this would lead to the incorrect conclusion that mutations in some genes with proven oncogenic functionality would be considered to be irrelevant in tumors with the mutator phenotype, but not in tumors without. For instance, it would be unsound to dismiss the oncogenic role in MMP tumorigenesis of mutant APC, K-ras, or p53, the prototypical cancer genes for colorectal tumorigenesis, all of which have paradoxically lower mutation incidences in MMP tumors compared with tumors without the mutator phenotype (1, 4, 16, 25). To make an exception with other genes because they have not been found frequently mutated in tumors without the mutator phenotype (e.g., β-catenin, axin, Tcf-4, or BAX) would amount to a (bio)logical fallacy.