Instabilotyping Reveals Unique Mutational Spectra in Microsatellite-Unstable Gastric Cancers¹

MSs³ are repetitive DNA sequences consisting of oligonucleotide units, which are distributed widely throughout the human genome. Length mutations in MSs are common among cancers with deficient DNA mismatch repair, such as hereditary nonpolyposis colorectal cancer-associated malignancies and sporadic gastric, colorectal, and endometrial cancers. MSI within coding regions causes FSMs, which result in gene inactivation. FSMs have been reported at several coding region MS loci, including loci within tumor-related genes in cancers with frequent MSI (MSI-H cancers). Known FSMs have been described in TGFβ type II receptor (TGFBR2), IGF2 receptor (IGF2R), BCL2-associated X protein (BAX), hMSH3, and hMSH6 (1, 2, 3, 4). However, FSMs are in general rare in MSI-H tumors when coding region MSs are studied without regard to their potential relationship to cancer. These findings suggest that FSMs that occur frequently in MSI-H tumors are the result of clonal selection during tumor development or progression. On the basis of this hypothesis, we previously performed a systematic genome-wide search to discover coding region MSs and FSMs to identify candidate tumor-related genes inactivated by FSMs in MSI-H colorectal cancers (5). We identified several frequently mutated genes, including the activin type II receptor (ACTRII) gene (5). In this report, we describe the results of mutational screening of 154 coding MSs in 18 MSI-H primary gastric cancers and compare the resulting FSM spectrum to the one that we had observed in MSI-H colorectal cancers. In this fashion, we identify several newly reported FSMs in candidate tumor-related genes, which provide clues to possible new carcinogenetic pathways and their uniqueness to different organs of origin.

Coding MSs consisting of homopolymeric tracts of eight nucleotides or longer were identified with a computer script as described in our previous study (5). The input of the script was Unigene Hs.seq.uniq.Z, a database containing only the clone with the longest region of high-quality sequence data among each gene cluster provided by the National Center for Biotechnology Information.⁴ We used this database as the input to decrease the likelihood of finding the same MS more than once. As a result, 300 homopolymers of eight or more nucleotides were identified and qualified for the study after manual inspection. PCR primer sets were designed to amplify each of these homopolymers. For this study, we used 154 loci at which PCR amplification using a standard human genomic DNA template was successful. All primer sequences are available on request.

Eighteen MSI-H cancers were identified from a larger group of 126 gastric cancers using five consensus loci (BAT25, BAT26, D2S123, D5S346, and D17S250) according to criteria from a National Cancer Institute Workshop in 1998 (6). Tumors were classified as MSI-H when MSI was observed at two or more of the five loci. Genomic DNA was extracted from paired normal and cancerous gastric tissues that had been frozen in liquid nitrogen after surgical resection, as described previously (7).

FSM at each locus was determined by analyses of the length of each PCR-amplified microsatellite. One primer of each pair was labeled with a fluorescent dye, i.e., Hex, Fam, or Tet. PCR reactions were performed in a total volume of 10 μl containing 20 ng of genomic DNA, 0.1 μm of each primer, 1× Taq DNA polymerase buffer (Life Technologies, Inc., Gaithersburg, MD), 0.4 mm of each deoxynucleotide phosphate, 1.5 mm MgCl₂, and 0.5 IU of Taq DNA polymerase (Life Technologies, Inc.). Conditions were as follows: an initial denaturation step at 94°C for 4 min, followed by 35 cycles of denaturation at 94°C for 30 s, annealing at 55°C for 30 s, and extension at 72°C for 30 s. After these steps, a final extension was performed at 72°C for 4 min. Products were analyzed on an automated DNA sequencer (ABI 377 or 3700; PE Biosystems, Foster City, CA) using the software programs GeneScan and Genotyper (PE Biosystems). We classified a tumor-specific alteration as an FSM only when it caused a change of >50% in peak area in the tumor sample compared with the corresponding normal sample. Biallelic alteration was defined as either two different mutations or a single mutation with absence of the normal allele.

The χ² test for goodness of fit (χ²) was performed to test the null hypothesis, i.e., that there was no difference in mutation frequency between gastric and colorectal MSI-H cancers. χ² value was calculated as follows:

where Ei is (number of informative samples) × (mean of mutation rates for gastric and colorectal cancers). Whereas Oi represented the observed number of mutated samples, Ei is the expected number of mutated samples. k is the number of genes analyzed. The degree of freedom for this χ² value was (k − 2).

To perform comprehensive screening of FSMs in coding region mononucleotide repeat MSs, 154 coding homopolymers consisting of eight or more nucleotides were examined in 18 MSI-H gastric cancers. These 154 loci were selected by using an automated coding region mononucleotide repeat database search as described in “Materials and Methods.” It should be emphasized that in order to identify potential involvement of less well-understood genes in carcinogenesis, we included many loci lacking known links to proliferation, differentiation, or cell death. Moreover, as positive controls, genes reported previously to undergo FSM were also included among the 154 loci, i.e., TGFBR2, IGF2R, BAX, hMS3, hMSH6, BRCA1, BRCA2, the retinoblastoma protein-interacting zinc finger protein gene (RIZ), and methyl-CpG binding protein 4 (MBD4; Refs. 1, 2, 3, 4, 8, 9, 10). Fig. 1 shows examples of electropherograms for FSM-positive and FSM-negative samples.

Among 154 coding region loci tested, 19 loci showed frequent mutation (i.e., in >20% of tumors). Loci mutated in >30% of tumors were TGFBR2 (11 of 18; 61.1%), ACTRII (8 of 18; 44.4%), DKFZp564K112 (7 of 17; 41.2%), SEC63 (6 of 16; 37.5%), MBD4 (6 of 18; 33.3%), BAX (4 of 12; 33.3%), and MAC30 (3 of 9; 33.3%). Among these loci, ACTRII, SEC63, DKFZp564K112, and MAC30 have not been reported previously by other groups as showing mutations or other alterations in human cancer. A BLAST search revealed that DKFZp564K112 had homology at the amino acid level with the Drosophila tumor suppressor gene, multi-sex-combs (msx). In addition, loci mutated in >20% of tumors were proline 4-hydroxylase β polypeptide (P4HB, thyroid hormone binding protein p55, protein disulfide isomerase), TTK, KIAA0896 (a human homologue of Drosophila hyperplastic discs, hHYD), RIZ, absent in melanoma 2 (AIM2), hMSH3, interleukin-1β converting enzyme (ICE/caspase-1), KIAA0905 (SEC31), monocarboxylate transporter member 4 (MCT4), hnRNP H, calnexin, and hKF-1. Two separate loci were analyzed for the tetratricopeptide containing Down syndrome gene 1 (TPRD1), and four of 18 samples (22.2%) exhibited FSMs in one of these loci. Forty-two loci with FSM rates of 10% or higher in MSI-H tumors are listed in Table 1.

The process of coding region mutation appeared to be selective, in that high mutation rates were observed in only a small percentage of loci. Mutation frequencies varied widely across the complete set of 154 loci (i.e., from 0 to 61.1%; mean, 7.7%; geometric mean, 3.0%), but the majority of loci did not undergo mutation at all. Sixty-nine loci were not mutated in any of the tumors, and 43 loci were mutated in <10% of tumors. Moreover, many genes showing high FSM rates had known or potential links to carcinogenesis. Longer repeats tended to mutate more frequently, as reported previously (data not shown; Refs. 5, 11). Eight loci were located in close proximity to the 3′ ends of coding regions (i.e., within 50 bases upstream of the last coding region nucleotide). Four of these eight loci (MAC30, TTK, AIM2, and MCT4) were mutated in >20% of gastric tumors. Mutations near the 3′ ends of genes do not usually result in major truncations of their protein products. However, mutations near the 3′ ends of genes may still cause dysregulation or dysfunction of protein products. Therefore, we did not eliminate them from our analyses.

We also analyzed biallelic alterations that should have resulted in gene inactivation, suggesting involvement of these mutations in carcinogenesis. Biallelic alteration was inferred when a coding region locus exhibited either two different mutations in the same MS or absence of the normal allele. Four of 11 tumors with FSM in TGFBR2 contained biallelic alterations. Similarly, two biallelic alterations were observed among eight tumors with ACTRII mutation, five tumors with TTK mutation, and five tumors with AIM2 mutation, respectively. Other novel loci with biallelic alterations were DKFZp564K112 (an msx homologue), SEC63, KIAA0896 (hHYD), MCT4, and calnexin. Moreover, biallelic alterations may have occurred more frequently than we observed; our sensitivity of detecting loss of the normal allele may have been low in some tumors because of normal cell contamination. In addition, it is possible that biallelic alterations in some genes occurred by aberrant promoter methylation or by mutations in regions not analyzed in this study.

To address the organ specificity of coding MS mutational spectra, a comparison of these spectra in gastric and colorectal cancers was also performed, based partially on our previous data on colorectal cancers (Fig. 2; Ref. 5). One hundred forty-eight of 154 loci in the current study had mutation frequency data available for both gastric and colorectal cancers (Table 2). Many loci were mutated at similar rates in both types of cancer. Among these were BAX, TTK, KIAA0896 (hHYD), and hMSH3, as depicted in Table 2, group c, and Fig. 2. In contrast, loci with markedly different mutation frequencies between gastric and colorectal cancers are listed in Table 2, groups a and b. Loci with the highest mutation frequencies are also illustrated in Fig. 2. Although the number of gastric cancers evaluated in this study was relatively small, the following loci were mutated significantly more frequently in gastric than in colorectal cancers: DKFZ564K112 (an msx homologue, 41.2% in gastric cancers versus 6.8% in colorectal cancers), MAC30 (33.3% versus 4.9%, respectively), RIZ (27.8% versus 4.5%), and ICECASP1 (27.8% versus 2.4%); P < 0.05, Fig. 2 and Table 2, group a. Conversely, TGFBR2 (61.1% versus 79.1%) and AIM2 (27.8% versus 47.6%) tended to mutate more frequently in colorectal cancers (Table 2, group b). Statistical significance was not achieved at these latter two loci by Fisher’s exact test because of the imbalance between the number of gastric (n = 18) and colorectal (n = 46) cases. The difference between mutational spectra of gastric and colorectal MSI-H cancers was also evaluated using the χ² test for goodness of fit, as described in “Materials and Methods.” This calculation was performed based on data for 46 genes with mutation rates of 10% or higher in at least one of the two organ sites. Our analysis of colorectal cancers showed that P for the calculated χ² value (73.9) was 0.0032 (the number of gastric cancers was too small to reliably use the χ² test for goodness of fit). Thus, because colorectal cancers showed a highly significant deviation from the mean mutation frequencies between gastric and colorectal cancers, our data disprove the null hypothesis (that mutation frequencies are the same in both tumor types). This result suggests that the mutational profiles of colorectal and gastric cancers are significantly different, in support of the belief that the pathophysiologies underlying these two types of MSI-H cancers are distinct.

Many of the genes in which we found high FSM rates have known potential links to carcinogenesis. For example, ACTRII is a member of the TGF-β receptor family that is involved in the induction of differentiation, growth suppression, and apoptosis. Mutant ACTRII inhibits activin-mediated induction of differentiation (12). DKFZ564K112 is a human homologue of the Drosophila tumor suppressor gene multi-sex-comb (mxc), a member of the homeobox gene transcription repressor family Polycomb group. Mxc is ubiquitously expressed, and its loss of function provokes uncontrolled malignant cell growth (13). ICE/caspase-1, a human homologue of Caenorhabditis elegans CED3, is a mediator of Fas-mediated apoptosis. Down-regulation of ICE/caspase-1 is observed in various human cancers (14), and overexpression of ICE/caspase-1 inhibits the growth of renal cell carcinoma cell lines in vivo (15). KIAA0896 (hHYD) is a human homologue of the Drosophila tumor suppressor gene hyperplastic discs. hHYD is expressed at moderately high levels in various tissues, and it acts as a ubiquitin ligase that coordinates DNA topoisomerase IIb-binding protein 1 during the DNA damage response (16). TPRD1 is a candidate gene for Down’s syndrome, a syndrome known to predispose to cancer, and its gene product is involved in mediating protein-protein interactions (17). It is expressed ubiquitously during mouse embryogenesis (18), suggesting a role in controlling cell differentiation.

Interestingly, among the loci mutated most frequently in gastric tumors, there was a subgroup of four known ER chaperone protein genes, which included SEC63. The other members of this subgroup were calnexin, P4HB (p55), and SEC31. In conjunction with known individual functions of these genes, this result suggests the involvement of chaperone proteins in gastric carcinogenesis. For example, calnexin is involved in ER accumulation of various proteins, including MHC class I antigen and the T cell receptor (19), which may affect the immunogenicity or intercellular interactions of tumor cells. Indeed, reduced expression of calnexin, correlated with anchorage-independent growth, was observed in a human colon cancer cell line (20). P4HB (p55) plays a crucial role in collagen synthesis, and the same gene is also known as an inducer of growth arrest, protein disulfide isomerase (21). SEC63 is involved in the process of protein folding and translocation, including the nuclear translocation of nucleoproteins (22). Finally, SEC31 has a role in ER-Golgi protein transport (23).

Evaluating the functional significance of coding MS mutations is a somewhat difficult task. Systematic mutation frequency surveys at short homopolymers with eight nucleotides located within the 3′ untranslated region (5) and intronic (24) regions in MSI-H tumors have revealed that some of these loci show mutation rates (up to 50%). Because these mutations are unlikely to have effects on gene function or to be objects of selection pressure during carcinogenesis, the high frequency of mutation in these noncoding tracts may be promoted by surrounding nucleotide sequences or genomic structure. Therefore, frequent coding MS mutations per se may not connote direct involvement in carcinogenesis. Additional evidence is needed to prove that disruption of these genes is involved in carcinogenesis. In the current study, we addressed this dilemma by exploring published known or potential functional links to cancer and complete inactivation of genes caused by biallelic alteration. In the future, additional molecular genetic findings supporting involvement of candidate genes in carcinogenesis could include mutations occurring outside of the MS tracts we studied, particularly in non-MSI-H tumors.

Unlike single candidate gene studies of human cancers, instabilotyping offers the potential advantage of removing preconceived biases regarding roles of particular genes in cancers. This unbiased approach identifies some coding region MSs that are not mutated at all. However, the apparent rarity of mutations found by this approach may itself constitute an advantage; frequently mutated loci identified using this approach often have known or potential links to cancer. Moreover, biallelic mutations at these coding region MSs suggest that total inactivation of at least some of these genes is occurring, possible creating a growth or survival advantage for these gastric cancer cells. Thus, our findings suggest that comprehensive instabilotyping is a valid strategy to identify cancer-related genes. The above data imply that instabilotyping is an advantageous approach not only to discover novel mutations among coding MS loci in human cancer but also to identify tissue specificity patterns in mutation rates at these loci. These observed intertissue differences in mutational spectra may reflect known differences in etiology and biology between gastric and colorectal cancers. Finally, these findings imply that our instabilotyping strategy should be applied to other tumors showing frequent MSI, such as endometrial cancers.