Risk of Genome-Wide Association Study–Identified Genetic Variants for Colorectal Cancer in a Chinese Population

Colorectal cancer (CRC) is one of the most commonly diagnosed cancers in the world (1), and its incidence is now dramatically increasing in developing countries including China (2). As other complex diseases, CRC is a complex trait caused by both genetic and environmental factors. Twin- and family-based studies have clearly shown a strong genetic component to CRC development, which accounts for ∼35% of total CRC (3). However, germline mutations with high penetrance in a few genes (DNA mismatch repair genes, APC, SMAD4, BMPR1A, MUTYH, and STK11) are estimated to explain <5% of total CRC (4), and much of the remaining variation in genetic risk may be attributable to multiple common alleles with low penetrance.

With the advent of high-throughput genotyping technologies, it is possible to perform rapid and efficient genotyping for hundreds of thousands of genetic variants without prior knowledge of gene function through genome-wide association study (GWAS; ref. 5). Several recent GWAS on CRC have implicated 10 common single nucleotide polymorphisms (SNP) contributing to CRC risk (6). The SNP rs6983267 on chromosome 8q24.21 was the first to be identified as a common susceptibility variant for CRC by three GWAS in the United Kingdom, Canada, and Scotland (7-9). The second SNP rs4939827 on 18q21.1 harboring SMAD7, a strong candidate gene, has subsequently been identified by a GWAS in the United Kingdom and further confirmed by another GWAS in Scotland (10, 11). Then, another four SNPs located, respectively, on 15q13.3 (rs4779584), 8q23.3 (rs16892766), 10p14 (rs10795668), and 11q23 (rs3802842) were identified (11-13), of which rs3802842 is the first one showing a population difference between European and Japanese populations. Recently, a meta-analysis of two GWAS confirmed 14q22 (rs4444235, BMP4), 16q22.1 (rs9929218, CDH1), 19q13.1 (rs10411210, RHPN2), and 20p12.3 (rs961253) as susceptibility loci for CRC (14).

Despite that the associations with CRC risk of the SNPs mentioned above have been successfully replicated in the previous studies (15-21), little or nothing is known about whether these associations exist in non-Caucasian populations. The aim of this study was to examine the association between the 10 SNPs identified in GWAS, alone and in cumulation, and the risk of CRC in a large case-control cohort derived from Han Chinese population. We further explored the potential effect of age, gender, tumor site, and pathologic characteristics of CRC on these genetic variant–associated risks.

We designed a 1:1 case-control study consisting of 2,124 patients with CRC and 2,124 controls. All subjects were Han Chinese. Patients were recruited between June 2001 and May 2009 at the Cancer Hospital, Chinese Academy of Medical Sciences (Beijing). All patients with histologically confirmed colorectal adenocarcinoma were enrolled with a response rate of 93%. The detailed diagnosis of CRC patients was previously described (22). The pathologic stage of CRC at the time of diagnosis was classified into Dukes A (T_1-2N₀M₀), B (T_3-4N₀M₀), C (T_anyN_1-3M₀), and D (T_anyN_anyM₁). Tumor grade was classified into high (poorly differentiated), intermediate (moderately differentiated), and low (well differentiated) according to the WHO grade classification (23). Controls were cancer-free individuals selected from a community cancer screening program for early detection of cancer conducted in the same regions during the same time period the patients were collected, with a participation response rate of 89%. These control subjects were randomly selected from a pool of 2,800 individuals on the basis of physical examinations. The selection criteria included no history of cancer and were frequency matched to patients based on sex and age (±5 y). At recruitment, informed consent was obtained from each subject, and each participant was then interviewed to collect detailed information on demographic characteristics. This study was approved by the Institutional Review Board of the Chinese Academy of Medical Sciences Cancer Institute.

We selected 10 top SNPs from 10 chromosomal regions that have previously been reported to be most associated with CRC (6), and genotyping were done using genomic DNAs extracted from peripheral blood lymphocytes of study subjects. Genotypes of all SNPs were determined by PCR-based RFLP assays except for the SNP rs4939827, which was analyzed by the tetraprimer amplification refractory mutation system-PCR method. The PCR primers and restriction enzymes used for genotyping are shown in Supplementary Table S1. The genotypes distinguished by RFLP or the amplification refractory mutation system were further confirmed by direct DNA sequencing of PCR products. To ensure quality control, genotyping was done without knowledge of case/control status of the subjects, and a 10% random sample of cases and controls was genotyped twice by different persons; the reproducibility was >99%.

The CRC risk associated with genotype was estimated as odds ratios (OR) and 95% confidence intervals (95% CI) computed using logistic regression under an additive genetic model adjusted for age and gender. Bonferroni correction for multiple testing was applied. We also tested the cumulative effect of risk SNPs by counting the number of genotypes associated with CRC risk in each subject on the basis of the dominant or recessive genetic model from single SNP analysis (24). The ORs for patients carrying any combination of 1, 2, 3, or ≥4 risk genotypes were estimated by comparing them with patients carrying none of the risk genotypes using logistic regression model adjusted for age and gender. P < 0.05 was used as the criterion of statistical significance. All statistical analyses were carried out using the Statistic Analysis System software (version 9.0; SAS Institute).

A total of 2,124 CRC patients and 2,124 frequency-matched controls were included in this study. No significant differences were found between patients and controls in the distribution of age and sex (Table 1). Males were 59.8% among patients compared with 58.3% among controls (P = 0.35). The mean age (±SD) was 56.9 (±11.8) years for patients and 56.4 (±11.3) years for controls (P = 0.16). Of the patients, 820 had colon cancer and 1,304 had rectal cancer. Regarding tumor stage, 269, 742, 783, and 305 patients were classified as Dukes A, B, C, or D stage at the time of diagnosis, respectively, whereas 25 patients had unknown stage. For histologic differentiation, 170, 1,494, and 365 patients were classified as low, intermediate, or high grade, respectively, whereas 95 patients had missing data.

Among the 10 SNPs investigated, rs16892766 was not polymorphic in our study population and was thus excluded in the analysis. The genotyping results are shown in Table 2. The average call rate for the remaining nine SNPs genotyped was >99% (ranging from 99.1-100%), and the genotypes for all SNPs in controls were in Hardy-Weinberg equilibrium. Multivariate logistic regression analyses showed that the rs6983267GG genotype had an OR of 1.54 (95% CI, 1.29-1.83) compared with the rs6983267TT genotype. However, the rs6983267TG genotype was not associated with the risk (OR, 1.11; 95% CI, 0.97-1.27), suggesting a recessive effect of the polymorphism. Pooled analysis with the rs6983267 TT and TG genotypes as reference group resulted in an OR of 1.45 (95% CI, 1.24-1.69; P = 2.00 × 10⁻⁶) for the GG genotype. The rs4939827CT (OR, 1.25; 95% CI, 1.09-1.43) but not TT genotype (OR, 1.10; 95% CI, 0.79-1.54) had an increased risk compared with the rs4939827CC genotype. Analysis with pooled 4939827CT and rs4939827TT genotypes yielded an OR of 1.23 (95% CI, 1.08-1.41) compared with the 4939827CC genotype. The rs10795668GG genotype had an OR of 1.44 (95% CI, 1.18-1.75) compared with the rs10795668AA genotype, whereas the rs10795668AG genotype was not associated with the risk (OR, 1.11; 95% CI, 0.92-1.35), suggesting a recessive allele effect. Subjects with the rs3802842AC or rs3802842CC genotype had an OR of 1.38 (95% CI, 1.20-1.58) or 1.60 (95% CI, 1.34-1.91) compared with subjects with the rs3802842AA genotype, suggesting a dominant effect of this SNP. For rs961253 SNP, the AC or AA genotype had an OR of 1.33 (95% CI, 1.13-1.56) or 2.96 (95% CI, 1.38-6.37) compared with the CC genotype. No significant associations were detected between CRC risk and rs4779584, rs4444235, rs9929218, and rs10411210 SNPs, despite that the statistical significance for the rs4444235 and rs10411210 SNPs was marginal (P = 0.07 and P = 0.09, respectively).

We next examined the cumulative effect of risk SNPs by counting the number of genotypes associated with CRC risk in each subject on the basis of the best-fitting genetic model from single SNP analysis (Table 3). Patients carrying 1, 2, 3, or ≥4 risk genotypes had increased risks of CRC compared with patients carrying none of the risk genotypes of the five SNPs, and there was a gradual increase in the OR with increased number of risk genotypes. Specifically, compared with patients carrying none of the risk genotypes, patients carrying 1, 2, 3, or ≥4 risk genotypes had an adjusted OR of 1.39 (95% CI, 1.09-1.76), 1.95 (95% CI, 1.55-2.46), 2.59 (95% CI, 2.01-3.34), or 3.25 (95% CI, 2.24-4.72), respectively (P = 1.32 × 10⁻²¹ for trend test).

The association between the five risk SNPs and site, grade, or Dukes stage of CRC was further evaluated (Table 4). Although genotype frequencies for the rs6983267, rs3802842, rs4939827, and rs961253 SNPs were not significantly different between colon cancer and rectal cancer, we found a significant site-specific difference in risk for the rs10795668 SNP, which was significantly associated with increased risk of rectal cancer (OR, 1.44; 95% CI, 1.25-1.66) but not colon cancer (OR, 1.14; 95% CI, 0.97-1.34). Among the five SNPs, rs3802842 seemed to be associated with increased risk for being more aggressive CRC. When the rs3802842 AA genotype was taken as the reference group for analysis, the pooled AC and CC genotypes were significantly associated with tumor stage, with an OR being 1.29 (95% CI, 1.07-1.55) for Dukes C and D stages against Dukes A and B. No significant associations were found between the SNPs and other clinical characteristics including age and sex (data not shown).

Recent GWAS have identified many genetic variants associated with CRC risk; however, most initial scan and replication studies were conducted in Caucasian populations (6). Because there are significant differences in the prevalence of CRC and the frequencies of genetic variations among different ethnic populations (25, 26), it is greatly important to explore the effects of these variations in non-Caucasian populations. We therefore systematically evaluated the 10 reported risk SNPs for CRC in a hospital-based, case-control study in a Chinese population. We confirmed that SNPs at 8q24.21, 18q21.1, 10p14, 11q23, and 20p12.3, but not 15q13.3, 14q22, 16q22.1, and 19q13.1, were associated with risk of CRC in our study population. In addition, a cumulative effect of the five risk SNPs at 8q24.21, 18q21.1, 10p14, 11q23, and 20p12.3 was observed.

Several studies in Caucasians have shown that an increased risk of CRC is associated with the rs6983267G variant allele in an allele-dose–dependent manner (15-19). However, we did not observe this dose effect in our study population. A similar study in Japanese also reported that only the rs6983267GG genotype is significantly associated with CRC risk (20). These findings suggest that this SNP is more likely to act as a recessive manner in Asian populations. The association between rs6983267T>G SNP and risk of CRC is biologically plausible because two recent studies have shown that rs6983267 is located in an enhancer region that interacts with the MYC proto-oncogene in a long distance way (27, 28). The rs6983267T and G alleles differentially bind transcription factor 7–like 2 (TCF7L2) or transcription factor 4 (TCF4), suggesting that this SNP may have an effect on MYC expression and thus contributes to colorectal carcinogenesis.

In this study, we found that rs4939827 SNP is a CRC susceptibility locus in a Chinese population, which is consistent with the GWAS by Broderick et al. (10). Another GWAS by Tenesa et al. (11) showed that risk for the rs4939827 SNP was greater for rectal cancer than for colon cancer; however, in a replication study, this SNP was associated with distal colon cancer but not proximal or rectal cancer (19). In contrast, a study reported by Thompson et al. (29) failed to replicate the association of rs4939827 with CRC risk. These inconsistent results emphasize a need of additional replication studies and tumor site–specific investigations. More importantly, functional study of the polymorphism in addition to epidemiologic replication is crucial to clarify whether the association is causative. The rs4939827 SNP is located in SMAD7, an important member of the SMAD family that functions as a negative feedback regulator of transforming growth factor-β responses (30). It has been shown that the overexpression of SMAD7 can induce tumorigenesis by blocking transforming growth factor-β–induced growth inhibition and apoptosis, which suggests a possible linkage of the association (31).

We did not observe any association between other four SNPs, rs4779584, rs4444235, rs9929218, and rs10411210, and risk of CRC. Among these SNPs, rs4779584 has been replicated in a Dutch familial CRC cohort (21). The remaining three SNPs were identified from a meta-analysis of GWAS (14) and have never been confirmed before our study. Failure to replicate the results from meta-analysis of GWAS does not seem to be rare as seen in studies of Parkinson's disease and type 2 diabetes (32, 33). The inconsistent results obtained by the present study and previous meta-analysis of GWAS may attribute to the ethnic difference in study populations because clear differences in the allele frequencies of these SNPs exist between Chinese population and Caucasian population (Table 5). Because of the different pattern of linkage disequilibrium in different populations, it is possible that genuine associations in different populations are not on the very top of the list in data arising from GWAS. A good example is the replication study of SNPs at 15q25 and lung cancer risk in a Chinese population, in which the risk SNPs are totally different from those identified in Caucasian populations (34). Another possible reason for skipping these SNPs as susceptibility variants for CRC might lie in the possibility of insufficient statistical power in the present study. The estimated powers were 100.0%, 92.1%, 64.5%, and 95.4% at α<0.05 and 100.0%, 71.5%, 31.8%, and 79.9% at α<0.005 (significance level after Bonferroni correction) for rs4779584, rs4444235, rs9929218, and rs10411210, respectively.

Our study has several strengths. We systematically evaluated all the 10 risk variants in a relatively larger case-control study derived from a Han Chinese population, which may better reduce the possible effects of population stratification. Furthermore, we assessed the cumulative effect of risk SNPs by counting the number of genotypes associated with CRC risk in each subject on the basis of the best-fitting genetic model from single SNP analysis. However, our study also has some limitations. We only genotyped the top tagSNPs identified in GWAS in Caucasians, which might provide incomplete linkage disequilibrium information of our study population. In addition, despite of a relatively large sample size, the statistical power for some rare SNPs is limited, and therefore, caution should be taken in interpreting the results such as for the rs4444235 and rs10411210 SNPs.

In conclusion, this association study replicating all newly identified CRC loci confirmed SNPs at 8q24, 18q21, 10p14, 11q23, and 20q12.3 but not 15q13.3, 14q22, 16q22.1, and 19q13.1 as genetic susceptibility factors for CRC in a Chinese population. Although these findings might provide new insights into colorectal tumorigenesis and have potential implications in early detection and target treatment of CRC, the causal variants have yet to be identified at any of these loci and further studies are needed to characterize the functional sequences that cause CRC.

No potential conflicts of interest were disclosed.

Grant Support: State Key Basic Research Program grants 2004CB518701.

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.