Functional and Clinical Significance of Variants Localized to 8q24 in Colon Cancer

Multiple independent genome-wide association studies (GWAS) have localized several cancer susceptibility variants to five distinct but adjacent regions on chromosome 8q24 (128.14-128.62 Mb; refs. 1-10). The single nucleotide polymorphism (SNP) rs1447295 was the first marker at 8q24 to be identified as a risk factor for prostate cancer (1, 3, 4, 7, 11). Subsequent genome-wide association studies in colorectal cancer identified a second independent genetic susceptibility marker (rs6983267) on chromosome 8q (5, 6, 8, 12, 13). These SNPs are located in different linkage disequilibrium (LD) blocks (D′ = 0.25) and are separated by a recombination hotspot (6, 7). Markers in this region were also associated with prostate and ovarian cancer. In addition, SNPs located on two other regions seemed to be specific to prostate cancer (3, 4). Finally, SNPs in an adjacent region have been reported to be associated with breast cancer only (2). All these multiple regions with different cancer specifities were summarized in a recent report (9).

Despite the strong association reported for the various SNPs in these regions, it is unknown if these are the true causal variants. It is likely that some or all of these are in LD with the etiologically relevant markers that have not yet been identified. Additionally, these data argue for several independent risk markers mapping to this particular chromosomal region. In an effort to obtain additional fine mapping data, Yeager et al. (14) sequenced a 136-kb region, including the rs6983267 and rs1447295 variants, and generated a detailed map of genetic variation. Analyses suggested rs6983267 as the best candidate SNP for functional studies as it lies within a highly conserved region and has a regulatory potential.

The functional variants responsible for susceptibility within this 8q24 region, as well as their mechanism of action, have not yet been identified. As these markers lie in a chromosomal gene desert, the identification of specific target genes for the risk markers has been difficult. One of the known genes in close proximity is c-MYC, located telomeric to the 8q SNP cluster (∼220 Kb). However, no significant relationship has yet been observed between the 8q24 variants and c-MYC mRNA expression (6, 8, 12). Given that there are independent markers in at least five separate regions, these data suggest that altered chromatin structure may play an important role in disease susceptibility. c-MYC, as well as other closely localized genes, is frequently involved in structural chromosomal alterations in a variety of tumor types. This 8q24 region has been reported to be a frequent site of somatic amplification in both prostate (15) and colorectal cancer (16, 17). Somatic gains in copy number of 8q24 have been reported in 18% of colorectal adenomas and 34% of colorectal cancers (18).

Interestingly, in addition to amplification at c-MYC, the 8q24 chromosomal region contains another common structural alteration. Two fragile sites, FRA8C and FRA8D, lie in close proximity to the SNP cluster. This fragile site has also been reported to be one of the preferred integration sites for human papillomavirus (19). The relationship, if any, of the various SNP genotypes to 8q amplification and/or to chromosome fragility is unknown.

In this study, we directed our efforts to understand the biological mechanism and possible functional importance of this region in colon cancer. Our main goal was to determine whether there were any structural changes at 8q24 that might correlate with the variants reported for this region. We also examined the implication of these susceptibility markers for colon cancer case-control status as well as survival.

Colon cancer cases were selected from two prospective collections. The first, the Colorectal Neoplasia Repository, is an ongoing collection of biospecimens from Mayo Clinic Rochester patients with colorectal neoplasia. A subset of these patients underwent a surgical resection for colon cancer during a 3-y period, 1995 to 1998. The collection was reinitiated in 2000. Cases for the association study were selected from both of these two time periods. Of the consented patients, U.S. resident Caucasians who had only colon cancer (not rectal) were included in this study. Patient chart reviews were done to obtain clinical characteristics of the tumor, including tumor site, stage, and age at diagnosis. For tumor site, tumors of the proximal colon were defined as those occurring in the cecum, the ascending colon, or the transverse colon (including both flexures). Distal tumors were defined as those occurring in the descending or sigmoid colon.

The second group was derived from a subset of the North Central Cancer Treatment Group (NCCTG) clinical trial NO147. Cases for the clinical trial were enrolled at ∼65 NCCTG sites in the United States and Canada, which included Mayo Clinic (Rochester, Arizona, and Florida). As part of the clinical trial protocol, biospecimens have been collected on an ongoing basis starting in March 2004. Patients ages ≥18 y with histologically confirmed and surgically removed stage III colon cancer were eligible. Patients undergoing chemotherapy through the multicenter NCCTG study were included if they had only colon cancer and not rectal, had surgery at most 8 wk before treatment, had no distant metastasis, and no previous radiation or chemotherapy for colon cancer.

Clinic-based controls were described in detail elsewhere (20). Briefly, the control group was obtained from patients being seen for a routine annual general physical examination in the Department of Medicine, Mayo Clinic, Rochester, Minnesota. This ongoing prospective collection of biospecimens was initiated in 2000. Control subjects had no prior history of cancer (except nonmelonoma skin cancer) at the time of enrollment.

Overall, there were 1,483 cases ascertained from the two sources described in the previous section: Colorectal Neoplasia Repository (n = 834) and NCCTG_N0147 (n = 649). After removing 110 individuals with non-Caucasian ancestry, 60 patients who did not live in the continental United States, and 6 others for miscellaneous reasons, there were 1,307 remaining Caucasian cases. Of the 1,307 available cases, 1,244 were matched on age, gender, and state of residence to the general medicine controls (all Caucasian). DNA was not available for 129 cases and 235 controls. Genotyping was done on a total of 1,178 cases and 1,009 controls that comprised the final analytic cohort.

A second independent group of cases for the survival study (n = 460) were participants of a phase III adjuvant therapy trial (NCCTG, 914653; ref. 21). DNA was available from only the proficient mismatch repair (pMMR) cases for this study. Patients had undergone complete surgical resection of high-risk stage II/III colon cancer during a 5-y period, from April 1993 to January 1998. Patients with surgically removed primary, histologically confirmed colon cancer with no evidence of residual disease but high risk of recurrence were eligible. Patients could not have prior exposure to 5-fluorouracil, any prior radiation therapy or chemotherapy, or any concurrent or previous malignant tumor within the previous 3 y, except superficial squamous or basal cell carcinoma of the skin or in situ carcinoma of the cervix. Patients ages between 31 and 85 y were randomized to receive either high dose (n = 224) or standard dose (n = 236) of treatment regimens.

Genotyping on blood DNA was carried out using multiple platforms. The TaqMan assay (Applied Biosystems) was done according to the manufacturer's guidelines, using 10 to 20 ng DNA. The following primers and probes were from Assay-by-Design (Applied Biosystems): rs1447295 (C_2160574_30), rs10090154 (C_26755480_10), rs13254738 (C_31232916_10), rs6983561 (C_1645354_10), rs16901979 (C_33280526_10), rs6983267 (C_29086771_20), and rs7000448 (C_2160622_10). Following PCR amplification, end reactions were read on the ABI Prism 7900HT using Sequence Detection Software (Applied Biosystems). The Illumina custom GoldenGate (Illumina, Inc.) genotyping protocol was done according to the manufacturer's guidelines. Arrays were imaged using an Illumina BeadArray Reader and the data analyzed using BeadStudio software 2.0 (Illumina). The microsatellite polymorphism DG8S737 was detected using ABI 3100 fragment length analysis (Applied Biosystems).

Genotyping on paraffin DNA was carried out using allele-specific PCR for the markers rs1447295, rs16901979, rs13254738, rs6983267, and DG8S737. The forward primers were fluorescently labeled by HEX or FAM (primers used are available on request). The PCR product was mixed with formamide containing ROX dye size standard (GeneScan 400HD; Applied Biosystems) and analyzed on an ABI 3100 Genetic Analyzer (Applied Biosystems). The PCR product fragment lengths were determined by GeneScan software.

To ensure quality control of genotyping results, one CEPH DNA and three negative controls, along with two duplicate samples, were included on each 96-well plate. Genotypes were determined blinded to phenotype status. The overall average call rate for the markers was 97% (range, 86.4-99.8%). In addition, χ² goodness of fit test for Hardy-Weinberg equilibrium was applied to the genotype proportions in the controls. All markers were in Hardy-Weinberg equilibrium, except rs10090154 (P = 0.02) that also had a low call rate of 86.4%.

Defective mismatch repair (dMMR) was defined by presence of microsatellite instability and/or the absence of protein expression for hMLH1, hMLH2, or hMSH6. pMMR was defined as tumors having normal protein expression for hMLH1, hMSH2, and hMSH6 and the presence of microsatellite stable (MSS) or low level instability (MSI-L).

Microsatellite instability in colon cancer cases was done with paired normal and tumor DNA isolated from formalin-fixed, paraffin-embedded material (22). Tumors were classified as microsatellite instability if ≥30% markers showed instability and as MSS/MSI-L if <30% showed MSI (23, 24). Immunohistochemical analysis of hMLH1, hMSH2, and hMSH6 expression was done on formalin-fixed, paraffin-embedded samples, as previously described (25).

Paraffin sections were baked and deparaffinized using a process of two 5-min incubations in Citrosolv at room temperature and two 5-min incubations in 100% ethanol at room temperature. The air dried slides were microwaved in 10 mmol/L citric acid until boiling then pepsinized for 15 min in 4-mg Pepsin/L 0.9% NaCl at 37°C followed by a dehydration series in 75%, 80%, and 100% ethanol for 2 min each. Vysis Provysion (LSI c-MYC 8q24.12-24.13 in Spectrum Green, CEP 8 in Spectrum Aqua, and LSI LPL 8p22 in Spectrum Orange) probe was applied to each slide, denatured on a HyBrite, and hybridized overnight. The slides were then washed in 2×SSC with 0.1% NP40, counterstained with 4′,6-diamidino-2-phenylindole, and coverslips were applied.

Fifty nuclei per sample were scored and the number of signals per nuclei recorded for each probe. A ratio of c-MYC copy number to CEP 8 copy number was established for each sample. The reference range was considered to be from 0.8 to 1.29. Duplication of the c-MYC region was determined to be any specimen with a ratio between 1.3 and 2.0. Amplification was defined as a ratio of >2.0 with >10 copies of c-MYC. A sample was considered to have gain of c-MYC if at least 30% of the nuclei had three or more signals for c-MYC but had a ratio in the reference range.

EBV-transformed cells were grown in RPMI 1640 with Glutamax, 10% fetal bovine serum, and 1% penicillin/streptomycin (100 units/mL, 100 μg/mL; All reagents from Invitrogen). Aphidicolin (Sigma) diluted in DMSO was added 24 h before harvest at a final concentration of 0.4 μmol/L to induce fragile sites. Cell harvest was done according to standard methods. Briefly, cells were treated with colcemid (0.075 μg/mL) for 1 h followed by a potassium chloride/sodium citrate hypotonic solution (0.075 mol/L, 0.8%) treatment for 20 min. Two changes of fixative (3:1 methanol/glacial acetic acid) were done before slide preparation.

Chromosome band 3p14.2 contains a common fragile site that is highly inducible through aphidicolin treatment. A commercially available DNA probe for centromere 3 labeled in SpectrumOrange (Abbott Molecular) was used. A DNA probe for 8q24.2 was created at the Mayo Clinic from the following BAC DNAs: RP11-382A18, RP11-327N12, RP11-367L7, RP1136L8, RP11-55J15, and RP263C20. Slides were pretreated with 2× SSC at 37°C, dehydrated in an ethanol series, and air dried. Probe solution was placed onto the slide. The probe and target DNA were codenatured using a ThermoBrite system (Abbott Molecular) set at 73°C for 3 min followed by hybridization at 37°C overnight. Slides were washed with 0.4×SSC at 72°C to remove excess probe and counterstained with 10% 4′,6-diamidino-2-phenylindolef.

Metaphase cells were located using a fluorescence microscope equipped with a 100-W mercury lamp and a 4′,6-diamidino-2-phenylindole single-pass filter. The fluorescence in situ hybridization (FISH) signals were viewed using a ×100 oil immersion objective along with a dual-pass filter for SpectrumOrange-SpectrumGreen, a single-pass filter for SpectrumOrange, or a single-pass filter for SpectrumGreen. FISH signal patterns were documented with a Cytovision workstation (Applied Imaging). A maximum of 100 metaphases were analyzed per sample.

All categorical variables were summarized using frequencies and percents. Continuous variables were summarized using means and SDs. The case and control genotype frequencies were calculated and then tested for Hardy-Weinberg equilibrium within the controls group using χ² goodness of fit or Fisher's exact tests. Pairwise LD measures were estimated using Haploview software (v4.1). Logistic regression analyses were done to test interactions among SNPs across three regions. Estimates of relative risk (OR, odds ratio) and 95% confidence intervals (95% CI) were estimated for the association between genotypes and colon cancer using unconditional logistic regression, adjusted for age. In our analyses, an ordinal model (log-additive) was used in which homozygotes for the common allele was the reference category. Marker rs6983267 was an exception to this and the homozygotes for the rare allele was used as the reference category to be consistent with the literature. In addition to testing for the main effect of each marker, the analyses were stratified by the cases' tumor characteristics at diagnosis, mismatch repair status, tumor site, and stage.

Overall survival (OS; censored at 8 y) was calculated as the number of years from random assignment to the date of death or last contact. Disease-free survival (DFS; censored at 5 y) was calculated as the number of years from random assignment to the first of either disease recurrence or death. The distributions of OS and DFS among the 460 independent cases were estimated using Kaplan-Meier methodology. Univariate and multivariate Cox proportional hazards ratios from an additive model (26) were calculated to explore the association of clinical and genetic factors with OS and DFS. Due to rarity of homozygous DG8S737 −8/−8 and −10/−10 genotypes, they were combined with those heterozygous for DG8S737 −8 and −10, respectively, in the Cox model. The log-rank and likelihood ratio test P values were used to test the significance of each covariate in the univariate and multivariate models, respectively. According to the graphical and statistical methods used, all the underlying model assumptions were satisfied (e.g., proportional hazards; ref. 27).

All statistical tests were two sided, with a P value of ≤0.05 considered significant. P values were not adjusted for multiple comparisons. Statistical analyses were done using SAS software (SAS Institute).

Our goal was to first replicate the previously described associations at 8q24 for colon cancer. The demographic and clinical characteristics of the participants used for these studies are given in Table 1. There were 1,178 cases and 1,009 controls in the association analysis. Cases and controls were closely matched for age, gender, and state of residence at the time of diagnosis. Two-thirds of the cases had tumors of the proximal colon. The frequency of stage III cases was much higher compared with other stages because a subset of the cases were participants of the NCCTG clinical trial protocol NO147, which restricted enrollment to stage III colon cancer patients. Of the 1,073 tumors analyzed, 878 (81.8%) were pMMR (MSS/MSI-L, and/or the presence of normal protein expression) and the remaining were dMMR (microsatellite instability and/or absence of protein expression).

Eight previously reported markers, seven SNPs, and one microsatellite, in three chromosomal regions at 8q24 were evaluated for susceptibility for colon cancer risk in a case-control study. Regions 1 and 2 contain three markers each, DG8S737, rs1447295, and rs10090154, and rs13254738, rs6983561, and rs16901979, respectively; and region 3 contains two markers, rs6983267 and rs7000448 (Fig. 1). The SNPs in regions 1 and 2 were highly correlated with each other (r² >78). Region 3 SNPs presented a more moderate correlation (r² = 52). No significant interactions were observed among SNPs across these three regions (data not shown).

Results of the association study with risk of colon cancer are provided in Table 2. We observed two significant associations with case-control status. At region 2, SNP rs13254738 had an ordinal OR of 0.82 (95% CI, 0.072-0.94; P_trend = 0.0037); the frequency of the risk allele C was 0.34 in controls. The association of the SNP rs6983267 at region 3 showed significance with an increased risk of 1.17 (95% CI, 1.03-1.32; P_trend = 0.013).

In subset analysis, the associations between these markers were evaluated with a number of tumor characteristics. Results of the analysis for SNP rs6983267 showed statistically significant associations based on the tumor MMR status, tumor site, and tumor stage. Significant associations were observed in patients whose tumors were pMMR, occurred in the distal colon and were higher stage (Table 2). No statistically significant associations were observed for the remaining markers tested.

In addition to the role of these markers in tumor etiology, we examined their potential role in tumor progression. To perform this analysis, a separate set of 460 cases were analyzed (Table 1). Subjects were dominantly Caucasians with a mean age of 62 years. The gender distribution slightly favored males. Fifty-six percent of the cases had tumors of the distal colon (44% proximal). About three-fourths of the cases had tumor-node-metastasis stage III, the remaining were stage II. These cases were a subset of the participants of a phase III adjuvant therapy (NCCTG, 914653), which recruited stage II and III colon cancer patients. All 460 tumors analyzed were pMMR.

Survival analysis was done for five of the eight markers. All three chromosomal regions were covered by using markers that were in LD with adjacent markers (Fig. 1). Univariate analysis did not detect any significant differences in OS or DFS for any of the markers, although the rare alleles showed a trend for decreased survival rate. As expected, patients with stage III tumors had significantly decreased OS and DFS compared with stage II cases (OS OR, 2.08; 95% CI, 1.38-3.15; P = 4 × 10⁻⁴; Table 3).

The functional variant responsible for susceptibility to colorectal cancer within this 8q24 region remains unknown. As the region of interest is very gene poor, our goal was to determine whether there were any structural changes on this chromosomal segment that might be correlated with the variants reported.

First, we tested for abnormalities at or near the c-MYC locus using fluorescently labeled DNA probes for FISH analysis (Fig. 2). A summary of results for the 380 cases that had both FISH and 8q24 marker genotype data are shown in Table 4. Seventy-one cases (19%) showed normal copy number at or near c-MYC, 129 cases (34%) had gain, 164 cases (43%) had c-MYC duplication, and 16 cases (4%) had c-MYC amplification. Only histologic grade and tumor site were significantly associated with the four FISH categories (P = 0.02 and P = 0.03, respectively; Table 4). No differences were found among the four c-MYC categories for either the allele or genotype frequencies for any of the 8q markers tested (Table 4).

Second, we evaluated whether cases with a certain genotype were more susceptible to fragile site induction at FRA8C and FRA8D (Fig. 2). The fragility test was conducted on cases homozygous for the candidate SNP demonstrating the most significant association results, rs6983267. Overall, 10 samples were tested, five cases that were homozygous for the common G allele, and five cases that were homozygous for the rare T allele. Four of the five cases in each group were pMMR. A FISH probe for the centromere of chromosome 3 was included as a locator of an internal control for breakage assay (28). As expected, all 10 cases exhibited 3p14 breakage (proportion range, 6-32%). Of the 10 cases tested, 3 TT and 2 GG homozygous samples exhibited low-level fragility in 1% to 4% of cells (Table 5). None of the samples in either group had breakage in both 3p and 8q in the same metaphase spread.

Polymorphisms on chromosome 8q24 have been studied in relation to risk for a range of cancers, including prostate, colon, breast, and ovarian (1, 3, 6-9, 29). These studies have identified a number of independent risk-markers that cluster within a fairly narrow region at 8q24.

In this study, we replicate the findings reported by others showing an association between two of the eight polymorphic markers, rs6983267 and rs13254738, on 8q24 region and colon cancer risk. The results for rs6983267 were similar to those first observed from the genome-wide association studies (6, 8) and were in agreement with the other replication studies (5, 9, 13, 30-32). Our data also agree with studies using a different SNP (rs10505477) that is in high LD with the rs6983267 (8, 12). The observed minor allele frequency (T allele) of 0.49 among controls was similar to the reported study populations. However, it should be stressed that the association observed in our study is for the G allele, which differs from other markers as we used the rare allele as the reference (T allele for this specific SNP). The analyses were done in this manner to be consistent with other published studies.

The SNP, rs13254738, has not been extensively evaluated by other groups. Available data from the three recent publications did not show a significant association between this SNP and risk of colorectal cancer (5, 9, 13). Earlier studies have different case and control selection criteria. Cases were selected for our study only if they had colon but not rectal cancer. It might be important to test rs13254738 on rectal only cases as this may help us understand whether this SNP is specific to colon only cases.

We tested all eight markers stratified by tumor characteristics and detected a significant association of rs6983267 with MMR status, stage, and site. No significant evidence of association was reported in the previous studies with respect to these tumor characteristics (5, 6, 12, 30) except that a trend was found toward the association of rs6983267 with MSS colorectal cancer cases in a Finnish population (OR, 1.37; 95% CI, 1.02-1.85; P = 0.04; ref. 33). Similarly, our findings in the pMMR cases showed an association for SNP rs6983267, with an OR of 1.17. In addition, pMMR cases with stage III to IV distal tumors were at increased colon cancer risk for SNP rs6983267 (Table 2). Although, the overall effects are modest, the data suggests that this SNP has an effect in the development of colon cancer among subjects with distinct tumor characteristics profiles. Additional studies will be required to confirm these findings and to fully understand the precise mechanism.

In addition to their potential etiologic role, the clinical significance of these 8q24 variants in colon cancer was examined. Although there was a trend toward an association between the rare alleles and decreased survival, none of the associations were statistically significant. No survival differences between carriers and the noncarriers of the rare allele were also reported by Gruber et al. (12) in a dominantly Jewish population.

Finally, we examined the possible functional importance of this region. Our main goal was to determine whether there were any structural changes on this chromosomal segment that might correlate with the variants reported for this region. The functional and structural role of the 8q24 markers is not well understood. Defining their role has been particularly challenging as these markers lie in a chromosomal region in which there are no known genes. The c-MYC oncogene is the only well-known gene that is in close proximity. c-MYC is frequently deregulated by structural chromosomal alterations in various tumor types. In addition, chromosomal region 8q24 contains two fragile sites, FRA8C and FRA8D, which suggests that 8q24 region, may be prone to structural alterations due to its fragile nature. However, we were not able to detect any correlation between the 8q24 variants and any structural change in this region. Similarly, the susceptibility to breakage was not strongly correlated with the SNP variant.

There are several limitations to our study. First, our association study was restricted to Caucasians that had colon cancer only and thus our results may not be generalized to other ethnicities or rectal cancers. Second, the low event rate (30-35%) in our survival data limited our power for associations. Third, our fragility data were derived from a small number of samples and the prevalence of 8q24 fragility in at risk patients should be interpreted with caution.

Clearly, additional functional studies will be needed to understand the importance of this 8q24 region. In vitro studies may help to investigate the role of these SNPs in regulating gene expression. Although such a mechanism has not yet been implicated for genes located nearby, one cannot rule out an interaction for genes located at some distance away either in cis or trans. In addition, searching and identifying sequence conservation within and flanking regions might be important in determining the region's role in regulation and transcription.

In summary, we have further confirmed the association of two 8q24 markers with colon cancer risk. Our data, however, suggest that the 8q variants do not influence patient survival and are not implicated in the two chromosomal changes tested—c-MYC amplification in tumor tissue and induction of fragility at FRA8C and FRA8D.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We wish to thank the generous support of the Richard M. Schulze Family Foundation.