Novel colon cancer susceptibility variants identified from a genome-wide association study in African Americans

Abstract

Genome-wide association studies (GWAS) in ethnic/racial minority populations can help to fine-map previously identified risk regions or discover new risk loci because of the genetic diversity in these populations. We conducted a GWAS of colorectal cancer (CRC) in 6,597 African Americans (1,894 cases and 4,703 controls) (Stage 1) and followed up the most promising markers in a replication set of 2,041 participants of African descent (891 cases and 1,150 controls) (Stage 2). We identified a novel variant, rs56848936 in the gene SYMPK at 19q13.3, associated with colon cancer risk (odds ratio 0.61 for the risk allele G, P = 2.4 × 10⁻⁸). The frequency of the G allele was 0.06 in African Americans, compared to <0.01 in Europeans, Asians and Amerindians in the 1000 Genomes project. In addition, a variant previously identified through fine-mapping in this GWAS in the region 19q13.1, rs7252505, was confirmed to be more strongly associated with CRC in the African American replication set than the variant originally reported in Europeans (rs10411210). The association between rs7252505 and CRC was of borderline significance (P = 0.05) in a Hispanic population GWAS with 1,611 CRC cases and 4,330 controls. With the three datasets combined, the odds ratio was 0.84 for the risk allele A (95% confidence interval 0.79–0.89, P = 3.7 × 10⁻⁸). This study further highlights the importance of conducting GWAS studies in diverse ancestry populations.

INTRODUCTION

Globally, colorectal cancer (CRC) is the third and second most common cancer in men and women, respectively, (www.who.int; International Agency for Research on Cancer GLOBOCAN). Genetic susceptibility plays an important role in the etiology of CRC through rare, highly deleterious germline mutations (e.g., in APC, MLH1, MSH2), as well as through common, low-penetrance variants. Over 50 common variants have been identified in genome-wide association studies (GWAS) to date ¹. However, most GWAS of CRC have been conducted in Europeans and East Asians, and only limited data are available for other populations. Studies in populations of African ancestry are particularly important in that: 1) shorter linkage disequilibrium (LD) blocks in this population² facilitate pinpointing the best risk-defining variants, i.e. for fine-mapping; and 2) the greater genetic diversity in African-descent populations may lead to discovery of additional susceptibility variants.

As part of our GWAS, we previously reported fine-mapping results among 6,597 African Americans (1,894 cases and 4,703 controls) ³ for 21 published GWAS-identified CRC risk regions. To our knowledge, this was the first GWAS of CRC conducted exclusively in African Americans. Here we followed up all promising markers from this GWAS in an additional set of 2,041 African American participants (891 cases and 1,150 controls). The Hispanics is another admixed population useful for fine-mapping with shorter LD structure compared to Europeans ⁴. We therefore tried to replicate our most suggestive findings in this population.

MATERIALS AND METHODS

Participants, Genotypes and Quality Control (QC)

The 6,597 African American participants (1,894 CRC cases and 4,703 controls) included in the Stage I analysis were retained after strict quality control (QC) filtering on 7,339 participants (2,066 CRC cases, 5,273 controls). These included participants from the Multiethnic Cohort study (MEC), Colorectal Cancer Family Registry (CCFR) minority recruitment, the Southern Community Cohort Study (SCCS), MD Anderson Cancer Center, the University of North Carolina CanCORS study (UNC-CanCORS), the North Carolina Rectal Cancer Study (UNC-Rectal), and the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO). Information on basic demographics and lifestyle factors was obtained from in-person interviews and/or self-administered structured questionnaires. Details on study protocols, sample recruitment, data collection, and QC procedures have been provided before ^{3, 5}. In brief, all African American samples were genotyped using the Illumina 1M-Duo bead arrays, except for the 170 PLCO participants who were genotyped with the Illumina Omni 2.5M. Imputation of un-typed genotyped SNPs was performed with BEAGLE 3.3 ⁶ using the Europeans (EUR) and Africans (AFR) in the 1000 Genomes Project (1KGP) (Phase 1, Release 3) as reference panels, with all cases and controls combined. Markers with MAF < 0.005 in reference panels were excluded from imputation. For the African American data, 10,050,748 markers with imputation accuracy R² > 0.8 were kept for association analysis in Stage 1.

In Stage 2, we genotyped 378 SNPs using the Fluidigm platform in an independent set of African Americans from five CRC studies, the Detroit case-control study (Detroit), the North Carolina Colorectal Cancer Study (UNC-NCCCS), the National Cancer Institute-University of Maryland study (NCI-UMD), the Black Women’s Health Study (BWHS), and the MEC. Except for MEC and BWHS, the other studies are case-control studies. More details on these studies are provided in Supplementary Materials. The 378 SNPs selected for genotyping included 229 promising SNPs with p-value of association <10⁻⁴ in main effect analysis or p-value <10⁻⁵ in subgroup analyses (by anatomical subsite) in Stage 1, 119 ancestry informative markers (AIMs) that were validated for separating individuals based on continental ancestry ⁷, 27 CRC risk variants identified in GWAS up to February 2013, and three SNPs from our fine-mapping analysis of GWAS-identified CRC risk regions ³. Genotyping was performed on genomic DNA extracted from blood or mouthwash (for about one third of Detroit samples only) or on whole genome amplified DNA derived from mouthwash samples (for BWHS), according to standard protocols.

Initially, 2,053 participants and 370 SNPs with > 90% call rates were retained, including the 119 AIMs. Fourteen SNPs were then excluded for low concordance rates among duplicates, deviation from HWE (P < 10⁻⁸) among controls in the NCI-UMD, UNC-NCCCS, or Detroit studies [in which the numbers of controls were relatively large (n > 200)], minor allele frequency < 0.0025 or frequency discrepancy between Stage 1 and the 1KGP data (Figure S1). Twelve participants who were ethnicity outliers were excluded on the basis of principal components derived from the AIMs. In the end, 237 SNPs on 2,041 Stage 2 participants (891 cases, 1150 controls) were retained for association analysis. Among these remaining subjects, 660 (277 cases, 383 controls) were from Detroit, 418 (188 cases, 230 controls) from the UNC-NCCCS, 418 (137 cases, 281 controls) from NCI-UMD, 350 (166 cases, 184 controls) from BWHS, and 195 (123 cases, 72 controls) from the MEC.

Statistical Analysis

Principal components (PCs) were calculated based on ~22,000 genome-wide unrelated markers in Stage 1 and on 119 AIMs in Stage 2. Pair-wise PC plots suggested that the first two PCs were most informative for global ancestry and the distribution of the first PCs were similar in cases and controls in both stages (Figure S2). In both stages, logistic regression of CRC on allelic dosage with adjustment for age at blood draw, sex and the first 4 PCs was performed to obtain odds ratio (OR) estimates and 95% confidence interval (CI) per increase in allele count. In Stage 1, age was categorized as <55 years, 5-year intervals from 55 to 80, and ≥80 years. In Stage 2, analyses were additionally adjusted for study; we also checked that there was no strong between-study heterogeneity in Stage 2 with a likelihood-ratio test. Analysis by anatomical sub-site was performed for colon, rectal, left colon and right colon tumors in both stages.

To combine results from the two stages and for further meta-analysis with the Hispanic studies, a fixed-effect model with inverse variance weighting was implemented in METAL ⁸. Heterogeneity measure I² was calculated, and Cochran’s Q statistic was used to test for heterogeneity ⁹.

Replication in Hispanics/Latinos

The Hispanic Colorectal Cancer Study (HCCS) is a population-based study of individuals self-identified as Hispanic/Latino with a diagnosis of CRC. Details about these studies were previously published ¹⁰ and can be found in Supplementary Materials. In brief, cases aged > 21 years or older and diagnosed after January 1, 2008 were identified from the California Cancer Registry or directly from local hospitals in the Los Angeles region. Risk factor/dietary questionnaires, pathology reports, and saliva samples (for genotyping) were collected using methodologies developed in the Colon Cancer Family Registry and the MEC. The HCCS includes 1,611 Hispanic CRC cases born in Mexico, the USA, Central/South America, Cuba or the Caribbean Islands, or Europe. Hispanic individuals without a diagnosis of CRC from the MEC (n=2,106) and the Slim Initiative in Genomic Medicine for the Americas (SIGMA) Type 2 Diabetes Consortium (n=2,224) were used as controls. All study participants signed an informed consent and all procedures were approved by the Human Research Institutional Review Boards (IRB) of participating institutions. All p-values presented are two-sided, unless otherwise noted.

RESULTS

Characteristics of study participants are shown in Table 1. Among participants in Stage 1 (GWAS), cases were on average older than controls (mean age 68 and 62 years, respectively) and the proportion of females was higher in cases (49.6%) than in controls (35.2%) (Table 1). Little inflation of genome-wide association statistics due to population stratification was observed in Stage 1 (genomic control λ = 1.04). Cases and controls were similar with regard to global ancestry based on PCs in both stages. Stage 2 included the all-female BWHS; without it, the proportion of females was 53.4% in cases and 56.4% in controls. On average, cases and controls had a similar age distribution in Stage 2.

Table 1.

Characteristics of study participants

					% Female		Mean Age (SD)		% Colon cancer2	% advanced stage3
	Study1	Cases	Controls	Total	Cases	Controls	Cases	Controls
Stage 1	CCFR	943	19	962	50.5	63.2	59.1 (10.5)	55.5 (12.3)	80.0	—
	UNC-CanCORS	80		80	52.5		63.1 (9.9)		82.1	66.7
	UNC-Rectal	109	105	214	48.6	42.9	62.2 (10.2)	62.9 (10.0)	53.2	61.3
	MEC	342	4328	4670	41.5	33.8	73.0 (6.7)	68.6 (8.3)	85.0	47.2
	MD Anderson	185		185	50.3		56.6 (12.2)		70.3	88.6
	PLCO	76	94	170	52.6	53.2	64.8 (5.5)	66.0 (5.4)	84.0	65.3
	SCCS	159	157	316	58.5	57.3	55.0 (8.4)	55.2 (8.4)	71.9	58.8
	subtotal	1894	4703	6597	49.6	35.2	67.9 (8.7)	61.6 (11.3)	77.9	62.7
Stage 2	NCI-UMD	137	281	418	40.1	43.4	63.5 (13.1)	66.6 (7.7)	80.0	87.3
	BWHS	166	184	350	100	100	59.0 (9.9)	58.3 (9.7)	82.1	55.8
	Detroit	277	383	660	58.8	66.6	62.3 (9.8)	62.5 (9.6)	72.9	47.6
	UNC-NCCCS	188	230	418	52.7	54.3	62.0 (10.3)	65.1 (9.8)	86.2	66.7
	MEC	123	72	195	56.9	59.7	67.1 (8.9)	63.0 (7.2)	81.3	45.4
	subtotal	891	1150	2041	62.1	63.4	62.5 (10.6)	63.4 (9.5)	79.4	56.3

Abbreviation: SD, standard deviation.

¹Stage 1 studies were described in references 2 and 5; Stage 2 studies are described in Supplementary Materials. See Methods for full study names.

²Among cases with available site information and excluding cases with both a colon and rectal diagnosis.

³Advanced stage included regional or distant spread of tumor

Overall, for colorectal cancer, when all participants in both stages were included in the analyses, no p-value of association passed the Bonferroni-corrected threshold 5×10⁻⁸ for GWAS. Among the 27 GWAS-identified CRC risk variants, nine p-values for association in the combined analysis were <0.05 (Table S1). Twenty-three risk estimates were of the same direction as those originally reported (85%), and the four with opposite direction of associations were estimated with high uncertainty (P’s > 0.35). Lack of replication could be due to our relatively low power for some of the reported risk estimates (a priori power < 80% for 11 SNPs) (Table S2), or to the fact that the biologically relevant variants are correlated with the index variants in the original GWAS population, but not so in African Americans. We note that the results in Stage 1 alone for these 27 SNPs were previously reported ³.

In our previous fine-mapping analysis of 27 GWAS regions using the Stage 1 data, rs7252505 at the 19q13.1 locus was more strongly associated with CRC than the published index hit rs10411210 (P = 1.8×10⁻⁴ and 0.11, respectively)³. We replicated this association in Stage 2 with p-value of 9.9×10⁻⁵ for rs7252505, compared to the p-value of 0.78 for rs10411210 (Table 2). When we combined the results from African Americans in Stage 2 with those from the Hispanic GWAS, the common A allele was associated with a reduction in CRC risk at the p-value threshold 5 ×10⁻⁸ (OR = 0.84, 95% CI: 0.79–0.89, P = 3.7×10⁻⁸) (Table 2). In 1KGP, rs7252505 and rs10411210 are correlated in EUR (r² = 0.77, D´ = 0.88) and in Amerindians (r² = 0.65, D´ =0.81), but not in Africans (r² = 0.05, D´ = 0.44). Consistent with this LD pattern, the association between rs7252505 and CRC was not altered after adjustment for rs10411210 in the African American studies (data not shown).

Table 2.

Associations between rs7252505 and the originally reported index SNP, rs10411210, at 19q13, with CRC in African Americans and Hispanics

SNP	BP	Allelea	African Americans						Hispanics			Combined
			Stage 1			Stage 2
			Freqb	OR (95% CI)	P	Freqb	OR (95% CI)	P	Freqb	OR (95% CI)	P	OR (95% CI)	P	I2
rs7252505	33575064	A/G	0.62	0.85 (0.78–0.92)	1.8×10−4	0.64	0.76 (0.67–0.87)	9.9×10−5	0.15	0.89 (0.79–1.00)	0.05	0.84 (0.79–0.89)	3.7×10−8	26
rs10411210	33532300	T/C	0.41	0.94 (0.86–1.02)	0.11	0.41	0.98 (0.86–1.12)	0.78	0.15	0.87 (0.77–0.99)	0.03	0.93 (0.88–0.99)	0.02	0

Abbreviations: CI, confidence interval; OR, odds ratio; BP, base pair.

^aEffect/other allele

^bEffect allele frequency

In the anatomical subsite analysis, an association was found between rs56848936 (G/A) in the 19q13.3 region and colon cancer risk (Table 3) at the p-value threshold 5 ×10⁻⁸, with consistent directions of association in both stages (odds ratio for the risk allele G = 0.61, 95% confidence interval: 0.52–0.73, P = 2.4×10⁻⁸, I² =0). The risk allele G’s frequency was 0.06 in African Americans. The association was similar for left-sided (odds ratio for the risk allele G = 0.68, 95% confidence interval: 0.52–0.89, P = 0.0045) and right-sided colon tumors (odds ratio for the risk allele G = 0.56, 95% confidence interval: 0.45 −0.70, P = 4.6 ×10⁻⁷). Genome-wide association statistics for colon cancer in Stage 1 also did not show deviation from the expected (Figure S3), indicating the result in Stage 1 was not due to bias caused by population stratification. The risk allele was not observed in the 1KGP EUR or East Asian populations and was rare in Amerindians (frequency < 0.01). We therefore could not replicate this finding in other studies, including in the Hispanic GWAS. Associations between rs56848936 and rectal cancer (P=0.23) and overall CRC (P= 9.0×10⁻⁷) are shown in Table S3. All associations with p-values < 0.05 in stage 2 are shown in Table S4.

Table 3.

Association between rs56848936 at 19q13.3 and colon cancer in African Americans

BP (HG19)	Effect/Other	Group	Case	Control	Frequency (G)		OR (95% CI)	P
					Case	Control
46321507	G/A	Stage 1 1	1451	4703	0.047	0.067	0.62 (0.50, 0.75)	2.4×10−6
		Stage 2	607	1146	0.042	0.067	0.60 (0.43, 0.84)	0.0028
		Combined (I2 = 0)	2058	5849			0.61 (0.52, 0.73)	2.4×10−8

¹Imputed with R² = 0.96 in Stage 1.

DISCUSSION

It is important to assess the strength and robustness of GWAS-identified association signals in diverse racial/ethnic populations to gain insight into the pathogenesis of population-specific variants. By genotyping a second independent sets of studies in African Americans, we confirmed the association between CRC and rs7252505, a candidate risk SNP identified in our previous fine-mapping study of the 19q13 locus. When we combined the results from both stages and a Hispanic CRC GWAS, the p-value for the association was 3.7 × 10⁻⁸. Further, we identified a novel risk variant, rs56848936, for colon cancer in African Americans.

We found that, in the known 19q13 risk locus, rs7252505 was more strongly associated with CRC in African Americans than the original index SNP, rs10411210, identified in a European GWAS ¹¹. Rs10411210 is intronic to RHPN2 (rhophilin, Rho GTPase binding protein 2), and rs7252505 is intronic to the neighboring gene GPATCH1 (G-patch domain containing 1). RHPN2 is a master regulator of actin cytoskeleton rearrangement with a role in various actin-modulating protein targets ¹². Rs10411210 and the SNPs in LD with it have been studied further since the original report. Rs10411210 is located in a region with DNase I hypersensitivity in many cell lines and with strong evidence of promoter histone marks (H3K9ac, H3K4me3) in colon and rectal mucosa cells, according to the ENCODE/Roadmap epigenomic data summarized by HaploReg 4.1^{13, 14}. A non-synonymous SNP in RHPN2, rs28626308 (R70Q), which is weakly correlated with rs10411210 in Europeans (r² = 0.21), was suggested as a functional candidate in this region ¹⁵. Rs28626308, and the only SNP in high LD with it in AFR in 1KGP (r²>0.8), rs28840750, were not genotyped and not imputed with high accuracy (R²>0.8) in our African American data, perhaps owing to their relatively lower frequencies in this population (MAF ≤ 0.06 in AFR in 1KGP). The functional basis for the rs7252505 association signal observed here is also not clear. Although GPATCH1 is expressed in the colon, little is known about its function other than the fact that it contains a G-patch domain, a domain typically associated with RNA processing. Rs7252505 itself maps to putative regulatory elements marked by histone modification in a few cell types, but not in colon or rectal cells ¹³. No SNP in the 1KGP AFR data is in high LD (r² ≥ 0.8) with rs7252505.

Rs56848936 (chr19, 46321507 bp), the SNP in the 19q13.3 region identified as being associated with colon cancer in the present study is intronic to the SYMPK genes encoding symplekin. This nuclear protein is thought to participate in the regulation of polyadenylation ¹⁶ and gene expression and in 3′-end maturation of histone mRNAs, which do not undergo polyadenylation ¹⁷. Rs56848936 maps to DNase hypersensitive regions in stem cells and muscle cells and to promoter regions in stem cells (ES, ES-derived and iPS cells) with strong histone modification marks (H3K4me3 and H3K9ac), although evidence for such marks is weak in colon or rectal cancer cell lines ^{13, 14}. Among the SNPs (r² ≥ 0.8) in high LD with rs56848936 in AFR, rs142542461 (chr19, 46366539 bp, r² = 0.82 with rs56848936) in the 5′-UTR region of SYMPK is a potential functional variant. It is in the promoter region marked by DNase sensitivity and histone marks (H3K4me3) in almost all cell types, including colon and rectal tumors, as described in the Roadmap Epigenomics project¹⁴. Interestingly, symplekin, which is found in tight junctions of the epithelial cells, has been shown to enhance cell proliferation and inhibit cell differentiation in colorectal cancer cell lines^18–20. Nonetheless, we are not aware of any molecular features in the region including rs56848936 that may explain the specific association with colon cancer and not rectal cancer. Rs142542461 was not available in our African American data. Like rs56848936, rs142542461 is monomorphic in the EUR and East Asian data and has a minor allele frequency of 0.01 in Amerindians in 1KGP.

In summary, the present study confirmed that rs7252505 may be a better index risk variant in African Americans, further characterizing the 19q13 CRC susceptibility locus. The study also identified a novel risk SNP (rs56848936) in the 19q13.3 region for colon cancer. It is possible that our results are due to chance and further work is needed to validate and better characterize these findings. If confirmed, the association between rs56848936 and colon cancer observed in African Americans would be population-specific since this variation is not observed in Europeans or East Asians.

Novelty and Impact.

The heritability of sporadic colorectal cancer is not fully explained by known genetic risk variants. Association studies in African Americans can help in identification of ethnic-specific risk variants or in precise definition of known risk loci. From a genome-wide association study exclusively in African Americans, we identified a novel risk variant for colon cancer risk.

Abbreviations

CRC

Colorectal cancer

GWAS

Genome-wide association studies

1KGP

The 1000 Genomes Project

LD

Linkage disequilibrium

MAF

Minor allele frequency

OR

Odds ratio

CI

Confidence interval

PC

Principal component