Large-scale Genome-wide Association Study of East Asians Identifies Loci Associated With Risk for Colorectal Cancer

Abstract

Background & Aims:

Genome-wide association studies (GWASs) have associated approximately 50 loci with risk of colorectal cancer (CRC)—nearly one-third of these loci were initially associated with CRC in studies conducted in East Asian populations. We conducted a GWAS of East Asians to identify CRC risk loci and evaluate the generalizability of findings from GWAS of European populations to Asian populations

Methods:

We analyzed genetic data from 22,775 patients with CRC (cases) and 47,731 individuals without cancer (controls) from 14 studies in the Asia Colorectal Cancer Consortium. First, we performed a meta-analysis of 7 GWAS (10,625 cases and 34,595 controls) and identified 46,554 promising risk variants for replication by adding them to the Multi-Ethnic Global Array (MEGA) for genotype analysis in 6445 cases and 7175 controls. These data were analyzed, along with data from additional 5705 cases and 5961 controls genotyped using the OncoArray. We also obtained data from 57,976 cases and 67,242 controls of European descent. Variants at identified risk loci were functionally annotated and evaluated in correlation with gene expression levels.

Results:

A meta-analyses of all samples from people of Asian descent identified 13 loci and 1 new variant at a known locus (10q24.2) associated with risk of CRC at the genome-wide significance level of P<5 × 10⁻⁸. We did not perform experiments to replicate these associations in additional individuals of Asian ancestry. However, the lead risk variant in 6 of these loci was also significantly associated with risk of CRC in European descendants. A strong association (44%–75% increase in risk per allele) was found for 2 low-frequency variants: rs201395236 at 1q44 (minor allele frequency, 1.34%) and rs77969132 at 12p11.21 (minor allele frequency, 1.53%). For 8 of the 13 associated loci, the variants with the highest levels of significant association were located inside or near the protein-coding genes L1TD1, EFCAB2, PPP1R21, SLCO2A1, HLA-G, NOTCH4, DENND5B, and GNAS. For other intergenic loci, we provided evidence for the possible involvement of the genes ALDH7A1, PRICKLE1, KLF5, WWOX, and GLP2R. We replicated findings for 41 of 52 previously reported risk loci.

Conclusions:

We showed that most of the risk loci previously associated with CRC risk in individuals of European descent were also associated with CRC risk in East Asians. Furthermore, we identified 13 loci significantly associated with risk for CRC in Asians. Many of these loci contained genes that regulate the immune response, Wnt signaling to beta-catenin, prostaglandin E2 catabolism, and cell pluripotency and proliferation. Further analyses of these genes and their variants is warranted—particularly for the 8 loci for which the lead CRC risk variants were not replicated in persons of European descent.

Introduction

Colorectal cancer (CRC) is one of the most frequently diagnosed malignancies around the world. Genetic factors play a significant role in the etiology of both familial and sporadic CRC^{1, 2}. Family-based linkage studies and recent whole-exome sequencing studies have identified multiple CRC susceptibility genes, such as APC, MUTYH, MLH1, MSH2, MSH6, PMS2, PTEN, STK11, GREM1, BMPR1A, SMAD4, POLE, POLD1, NTHL1 and TP53^3-5. Deleterious mutations in these genes, however, are rare and account for less than 6% of CRC cases in the general population³. To date, genome-wide association studies (GWAS) have identified 52 independent loci in relation to CRC risk^{3, 6–9}. These common genetic risk variants, however, explain only a small fraction of the familial relative risk of CRCs^{8, 10}, particularly in non-European descendants.

In 2010, we initiated the Asia Colorectal Cancer Consortium (ACCC) to identify new genetic risk factors for CRC. Given the differences in genetic architecture and environmental exposure between Asian and European descendants, we hypothesized that our study could identify CRC genetic risk variants that are more specific to the Asian population and uncover important novel genetic risk variants that might otherwise be difficult to identify in European descendants. Over the years, we have reported findings for 13 novel CRC risk loci and two independent risk variants in known CRC risk loci^{6–8, 11}. In this study, we report the discovery of another 13 novel risk variants for CRC in a large study that includes more than 70,000 cases and controls of East Asian ancestry.

Methods

Overview of Study population and Study Design

Included in the current study are 22,775 CRC cases and 47,731 controls of East-Asian ancestry from 14 studies conducted in China, Japan, and South Korea (Supplementary Notes, Supplementary Figure 1 and Supplementary Table 1). In Stage 1, we meta-analyzed data from seven studies including 10,625 cases and 34,595 controls that were genotyped using high-density SNP arrays. We identified 46,554 variants associated with CRC risk at P < 0.005 (for common variants) or at P < 0.0005 (for rare variants). These variants were added to the Multi-Ethnic Global Array (MEGA), which includes ~2 million variants, as part of the custom content for a large study including multiple complex traits. Updated data from the Japanese CRC Study (BBJ) subsequently became available, and thus, were included in the Stage 1 meta-analysis¹². In Stage 2, we analyzed 6,445 cases and 7,175 controls from five studies genotyped using the expanded MEGA (described above) and 5,705 cases and 5,961 controls from two studies genotyped using the OncoArray, which has approximately 570,000 SNPs, including ~260,000 SNPs for the GWAS backbone. To evaluate the generalizability of our findings in Asian descendants, we used data from studies conducted in European descendants comprised of 57,976 cases and 67,242 controls that were recruited in North America, Europe and Australia. These cases and controls were included in three consortia: the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO), the Colorectal Transdisciplinary (CORECT) Study, and the Colon Cancer Family Registry (CCFR). The details of these studies have been described previously⁹. A description of participating studies is presented in the Supplementary Notes. All study protocols were approved by the relevant Institutional Review Boards.

Genotyping and Imputation.

Details of genotyping, genotype calling and quality control are provided in the Supplementary Notes. The genotyping and quality control procedures for Stage 1 have been described previously^{6–8, 11, 12}. Genotyping for Guangzhou-2 samples was completed using the Asian ExomeChip, an expanded Illumina HumanExome Beadchip that includes 159,370 rare variants (minor allele frequency (MAF)<1%) and 49,603 common variants. Genotyping and quality control of samples in the BBJ were reported previously^{8, 11–13}. Genotyping for the Shanghai-3 and HCES-CRC samples in Stage 2 was completed using the Illumina OncoArray¹⁴, and genotyping for samples from Shanghai-4, Aichi-2, Korea-NCC, Korea-Seoul and HCES2-CRC was completed using the expanded Illumina MEGA.

We re-genotyped a subset of samples (N = 918) for all 13 index SNPs to assess imputation quality. The re-genotyping was performed with Sequenom MassARRAY. The overall concordance rates between imputed data and genotyped data were high, indicating that the imputation quality was excellent for these SNPs (Supplementary Table 12).

We evaluated population structure with principal component (PC) analysis using EIGENSTRAT¹⁵, and found little evidence of population stratification for any of the studies included in this analysis (Supplementary Notes, Supplementary Figures 2 and 3). In order to increase the genome coverage and facilitate the meta-analysis, we imputed untyped genotypes using the 1000 Genomes Project phase 3 mixed reference haplotypes. After quality control filtering, approximately 8.6~10.7 million genotyped or imputed variants on the 22 autosomes remained for the association analysis. The Guangzhou-2 study was genotyped with an exome array, and thus, no imputation was performed, given the low genome coverage of this array.

Statistical Analysis

The score test in Rvtest¹⁶ was used to evaluate the association of genotype dosage with colorectal cancer risk after adjusting for age, sex, and, when appropriate, the first five PCs in each individual study (Supplementary Notes). The likelihood ratio test in mach2dat was used to assess the association of the genotype dosage with colorectal cancer risk in the BBJ. SNPs were excluded from the analysis if they had a low imputation quality (R² < 0.3) or a low MAF (<0.1%), based on the combined cases and controls. The summary statistics from 14 case-control studies included in both Stages 1 and 2 were meta-analyzed using the inverse variance-weighted fixed effect model implemented in METAL¹⁷. Variants showing an association with CRC risk at P < 5 × 10⁻⁸ in the combined analysis of the Stage 1 and Stage 2 studies were considered to be genome-wide significant¹⁸. Each unique locus was defined as ±500 kb on either side of the most significant SNP, with a P < 5 × 10⁻⁸. We evaluated heterogeneity across studies and subgroups with a Cochran’s Q test¹⁹. No apparent inflation was noted for any of the studies included in Stage 1 and Stage 2, as the λ_GC and λ_{GC 1,000} were small in the model with the adjustment of the first five principal components (Supplementary Figure 4, Supplementary Notes).

In an attempt to identify additional association signals in the same locus, we performed conditional analyses using the GCTA-COJO approach²⁰ for meta-analysis summary statistics based on a linkage disequilibrium matrix from 6684 unrelated East Asian samples genotyped with MEGA and pruned with interindividual genetic relationships < 0.025.

The familial relative risk (λ) for the offspring of an affected individual due to a single locus was estimated using a log-additive model: λ = (pr² + q)/(pr + q)², where p is the frequency of the risk allele, q = 1−p is the frequency of the reference allele and r is the per-allele relative risk²¹. The proportion of the familial relative risk explained by this locus, assuming a multiplicative interaction between markers in the locus and other loci, was calculated as log (λ)/log (λ₀), where λ₀ is the overall familial relative risk of 2.2 for CRC (derived from a meta-analysis)²². Assuming that the risks associated with individual loci combine multiplicatively, the familial relative risks would also multiply. Thus, the combined contribution of the familial relative risks from multiple loci is equal to

$$\sum _K(log\lambda \text{k})/(log\lambda 0)$$

Using an additive genetic model, we calculated weighted polygenic risk scores (PRSs) using the natural log-odds ratios (ORs) as the variant-specific weight that was obtained from current analyses. The PRSs were calculated as the sum of the product of the weight and the number of risk alleles for each of the 57 replicated (P < 0.05) and newly identified GWAS index variants that had an overall imputation R² > 0.8 across seven studies genotyped with either OncoArray or MEGA. Using the lowest quintile of PRS as the reference, ORs associated with PRS were estimated using logistic regression for other quintile groups, with an adjustment for age, sex, study and the first five PCs (Supplementary Table 11).

Pathway Analysis

The pathway analysis was performed with MAGMA²³. This analysis mapped all quality-controlled variants onto genes, estimated the variant-wise-mean P value for individual genes, and then conduced a generalized gene-set analysis, adjusting for the gene size, gene density and other covariates. The East Asian populations in 1000 Genome Phases 3 was used as the reference dataset for linkage disequilibrium estimation. Gene sets were obtained from MsigDB v5.2 (Supplementary Table 9). The false discovery rate (FDR) was used to adjust for the multiple testing.

In Silico Functional Characterization of Novel Loci

To identify potential functional variants and target genes at novel risk loci uncovered in this study, we used ANNOVAR²⁴ to annotate all variants that are in linkage disequilibrium (LD) (r² ≥ 0.8, EAS in the 1000 Genomes Project) with the lead variants revealed in our GWAS within 500kb (189 variants in total, Supplementary Table 6). The functional impact of variants were assessed with GERP²⁵, SiPhy²⁶ and PolyPhen2²⁷. We further characterized these 189 variants using the web-based HaploReg v4²⁸. The data from the Roadmap Epigenomic project (ChromHMM states corresponding to enhancer or promoter elements, histone modification ChIP-seq peaks and DNase hypersensitivity data peaks) were extracted, along with the regulatory protein binding from the ENCODE project and the regulatory motifs based on the commercial, literature and motif-finding analysis of the ENCODE project. We also examined these variants using the web-based RegulomeDB²⁹.

cis-Expression Quantitative Trait Loci (cis-eQTL) Analysis

We performed cis-eQTLs analyses for newly identified risk variants using data from tumor-adjacent normal samples obtained from 133 East Asian CRC patients^{7, 8}. For these analyses, we defined cis- as a ±1mb region flanking each risk variant. Germline DNA of these patients was genotyped using the MEGA. We evaluated expression levels of genes using RNA sequencing. Associations between gene expression levels and SNP genotypes were analyzed using linear regression model, adjusting for sex and the top two PCs. We also evaluated the association of these variants with expression of genes, using data from transverse colon tissues (n=246) in the Genotype-Tissue Expression (GTEx) database³⁰.

Results

In the combined analysis of 22,775 CRC cases and 47,731 controls, we identified 13 novel risk loci for CRC at the genome-wide significance level (P < 5×10⁻⁸) (Table 1, Supplementary Figure 5). In two of these loci, the lead SNPs have a low MAF: rs201395236 at 1q44 (MAF = 1.34%, allelic OR = 1.75 for the major allele) and rs77969132 at 12p11.21 (MAF = 1.53%, allelic OR = 1.44 for the minor allele). The risk associated with these two low-frequency variants was substantially higher than all common risk variants reported in previous GWAS. The lead SNPs in the remaining 11 were common (MAF > 5%), with MAFs ranging from 14 to 43% and ORs ranging from 1.08 to 1.16. Interestingly, we also identified a new risk variant, rs6584283, at a known CRC locus, 10q24.2, that was reported initially in a GWAS conducted in European descendants. However, this variant was not in LD with either of the two previously reported variants (r² _{EUR or EAS} < 0.01 with rs1035209³¹ and rs11190164³². The association for rs6584283 was genome-wide significant (ORcondition (95%CI) for C allele = 1.08 (1.06 – 1.11), P_condition = 5.9×10⁻¹⁰) after adjustments for these two variants.

Table 1.

Summary results of newly identified genetic risk loci/variants associated with colorectal cancer risk at P < 5 × 10⁻⁸: Results from the Asia Colorectal Cancer Consortium

Locus	SNP	Position (hgl9)	Nearby genes & Annotation	Alleles	RAF (%)	OR (95% CI)	P value	PHet	I2
Novel risk loci
1p31.3	rs7542665	62,673,037	missense of L1TD1	C/T	27.3	1.08 (1.05 – 1.11)	3.51 × 10−8	0.98	0.0
1q44	rs201395236	245,181,421	intronic of EFCAB2	T/C	98.7	1.75 (1.43 – 2.13)	4.63 × 10−8	0.52	0.0
2p16.3	rs7606562	48,686,695	intronic of PPP1R21	T/A	81.3	1.10 (1.07 – 1.14)	1.21 × 10−8	0.88	0.0
3q22.2	rs113569514	133,748,789	5'UTR of SLCO2A1	T/C	62.0	1.10 (1.07 – 1.13)	2.45 × 10−12	0.92	0.0
5q23.2	rs12659017	125,988,175	intergenic (ALDH7A1, PHAX)	G/A	23.2	1.09 (1.06 – 1.12)	4.45 × 10−8	0.11	33.5
6p22.1	rs1476570	29,809,860	intronic of HLA-G	A/G	37.6	1.12 (1.08 – 1.16)	6.71 × 10−9	0.01	58.0
6p21.32	rs3830041	32,191,339	intronic of NOTCH4	T/C	14.0	1.16 (1.10 – 1.21)	1.65 × 10−8	0.03	46.2
12p11.21	rs77969132	31,594,813	intronic of DENND5B	T/C	1.5	1.44 (1.27 – 1.65)	4.85 × 10−8	0.18	26.2
12q12	rs2730985	43,130,624	147kb 5' of PRICKLE1	G/A	62.9	1.08 (1.05 – 1.11)	1.23 × 10−8	0.20	23.9
13q22.1	rs1886450	73,986,628	intergenic (KLF5, KLF12)	G/A	59.3	1.09 (1.07 – 1.12)	6.28 × 10−12	0.60	0.0
16q23.2	rs4341754	80,039,621	intergenic (WWOX, MAF)	G/C	57.0	1.09 (1.06 – 1.12)	1.73 × 10−9	0.64	0.0
17p12	rs1078643	10,707,241	19kb 3' of PIRT	A/G	77.1	1.13 (1.09 – 1.17)	8.40 × 10−13	0.57	0.0
20q13.32	rs13831	57,475,191	intronic of GNAS	G/A	68.4	1.08 (1.05 – 1.11)	2.06 × 10−8	0.68	0.0
Novel risk variant in known loci
10q24.2	rs6584283	101,290,301	2kb 5' of NKX2–3	C/T	56.4	1.09 (1.06 – 1.12)	1.21 × 10−10	0.65	0.0

hg19: genome build 37; Alleles: risk allele/reference allele; RAF, risk allele frequency calculated with the combined stage 1 and stage 2 samples; OR, odds ratio; CI, confidence interval; I² and P_Het: derived from the heterogeneity test across all 14 studies included in Stage 1 and Stage 2.

All 14 risk variants showed a consistent association between Stage 1 and Stage 2, and 12 of them were statistically significant at P < 0.05 in both stages (Supplementary Table 2). Although the associations for rs201395236 and rs3830041 were not significant at P < 0.05 in Stage 1, they were both statistically significant in the two components (OncoArray and MEGA) of Stage 2 (Supplementary Table 3). Associations with CRC risk for each of the 14 new risk variants were consistent across all studies included in both Stages 1 and 2, with little evidence of heterogeneity (Table 1). Only two variants (rs1476570 and rs3830041) showed some heterogeneity across 14 participating studies in their association with CRC risk (P = 0.01 and 0.03, respectively). However, the difference was no longer statistically significant after adjusting for multiple comparisons. No statistically significant difference in the associations was found across ethnicity groups (Supplementary Table 4).

Next, we evaluated the generalization of associations for newly identified CRC risk variants in populations of European ancestry, using data from 57,976 cases and 67,242 controls. Imputed genotype data were available for 12 variants (Table 2). Six variants were found to be associated with CRC risk at P < 0.05 in the same direction as observed in the East-Asian population (rs113569514, rs6584283, rs2730985, rs1886450, rs4341754 and rs1078643), including five (rs113569514, rs6584283, rs2730985, rs4341754 and rs1078643) that showed an association at P < 4 × 10⁻³, the Bonferroni-corrected significance level (0.05/12 variants). However, there are considerable heterogeneities in the strength of the associations of these SNPs with CRC risk between Asian and European descendants, even for SNPs that showed a significant association in both populations. Furthermore, the frequencies of risk variants also differ considerably in these two populations (Tables 1 and 2). These results are consistent with our hypothesis and provide supports for genetic association studies in non-European populations.

Table 2.

Associations of newly identified risk variants with colorectal cancer risk in populations of European Ancestry

Locus	SNP	Position (hgl9)	Alleles	RAF (%)	OR (95% CI)	P value	PHet-ASN vs EUR	I2
1p31.3	rs7542665	62,673,037	C/T	66.2	1.01 (0.99–1.03)	0.23	6.69 × 10−5	93.7
1q44	rs201395236	245,181,421	T/C	100	NAb	-	-	-
2p16.3	rs7606562	48,686,695	T/A	65.8	1.01 (0.99–1.03)	0.35	4.83 × 10−6	95.2
3q22.2	rs113569514	133,748,789	T/C	84.7	1.06 (1.03–1.08)	1.6×10−5	0.02	82.2
5q23.2	rs12659017	125,988,175	G/A	73.1	0.99 (0.97–1.01)	0.46	5.95 × 10−7	96.0
6p22.1	rs1476570	29,809,860	A/G	27.6	1.01 (0.99–1.03)	0.44	1.62 × 10−6	95.7
6p21.32	rs3830041	32,191,339	T/C	9.1	1.00 (0.97–1.03)	0.92	1.75 × 10−6	95.6
10q24.2	rs6584283	101,290,301	C/T	53.3	1.03 (1.01–1.05)	3.6×10−4	8.21 × 10−4	91.1
12p11.21	rs77969132	31,594,813	T/C	0	NAb	-	-	-
12q12	rs2730985	43,130,624	G/A	52.3	1.05 (1.03–1.07)	5.5×10−9	0.10	63.9
13q22.1	rs1 886450	73,986,628	G/A	69.1	1.02 (1.01–1.04)	9.7×10−3	5.04 × 10−5	93.9
16q23.2	rs4341754	80,039,621	G/C	43.6	1.05 (1.03–1.07)	5.2×10−8	0.04	77.0
17p12	rs1078643	10,707,241	A/G	75.9	1.08 (1.05–1.10)	6.6×10−12	0.02	81.7
20q13.32	rs13831	57,475,191	G/A	70.5	1.01 (0.99–1.03)	0.32	4.39 × 10−5	94.0

hg19: genome build 37; Alleles: risk allele/reference allele as defined in the Asia Colorectal Cancer Consortium (See Table 1); RAF, risk allele frequency; OR, odds ratio; CI, confidence interval; P_Het-EUR: derived from the heterogeneity test across all studies included in the European meta-analysis. I² and P_{Het-ASN vs EUR}: heterogeneity between East-Asian and European populations calculated with Cochran’s Q test.

^a, Risk allele frequency derived from European populations of 1000 Genome Project (Phase 3).

^b, NA, not estimated due to low imputation quality in European populations.

Evaluation of Previously Reported CRC Risk Variants

We investigated 62 previously reported CRC risk variants at 52 independent loci (Table 3 and Supplementary Table 5)^{9, 33}. There were 19 independent risk variants at 18 loci initially reported in GWAS conducted in East Asian populations. All of them were replicated at P < 0.05, and 18 of them showed an association at P < 0.005, the Bonferroni-corrected significance level (0.10/19, one-side tests) (Table 3). Of the 43 risk variants at the 37 loci that were initially reported in European populations, 26 variants at 24 loci were associated with CRC risk at P < 0.05 in the same direction as reported initially, including 17 variants at P < 0.002 (0.10/43, one-side test). Only five of the tests for heterogeneity were statistically significant at P < 0.05, indicating little heterogeneity as judged using the multiple comparison-adjusted significance level of 9 × 10⁻⁴ (0.05/53).

Table 3.

Association results in East Asian populations for risk variants reported in previous GWAS of colorectal cancer, Results from the Asia Colorectal Cancer Consortium^a

Locus	SNP	Nearby genes	Alleles	RAF (1KGP)	RAF (ACCC)	OR (95% CI)	P value	PHet-ASN	I2
		Loci initially identified in GWAS of East-Asian-ancestry populationsb
5q31.1	rs647161	PITX1	A/C	0.30/0.66	0.33	1.12 (1.09 – 1.15)	3.10 × 10−15	0.04	45.2
6p21.1	rs4711689	TFEB	A/G	0.80/0.54	0.85	1.09 (1.05 – 1.13)	6.17 × 10−6	0.53	0.0
8q23.3	rs2450115	EIF3H	T/C	0.54/0.81	0.56	1.11 (1.08 – 1.14)	2.39 × 10−14	0.92	0.0
8q23.3	rs6469656	EIF3H	A/G	0.67/0.89	0.65	1.09 (1.06 – 1.12)	2.06 × 10−10	0.41	3.3
10q22.3	rs704017	ZMIZ1	G/A	0.28/0.56	0.33	1.09 (1.06 – 1.12)	1.17 × 10−9	0.66	0.0
10q24.32	rs4919687	CYP17A1	G/A	0.75/0.70	0.83	1.07 (1.03 – 1.11)	2.94 × 10−4	0.81	0.0
10q25.2	rs12241008	VTI1A	C/T	0.30/0.10	0.27	1.11 (1.07–1.14)	1.77 × 10−11	0.39	5.9
10q25.2	rs11196172	TCF7L2	A/G	0.65/0.12	0.69	1.12 (1.09 – 1.16)	3.19 × 10−15	0.41	3.8
11q12.2	rs174537	MYRF	G/T	0.43/0.65	0.62	1.10 (1.07 – 1.13)	7.52 × 10−12	0.36	8.4
12p13.32	rs10774214	CCND2	T/C	0.32/0.38	0.40	1.09 (1.06 – 1.13)	2.60 × 10−9	0.41	3.5
12p13.31	rs10849432	PLEKHG6	T/C	0.81/0.90	0.82	1.08 (1.04 – 1.12)	1.95 × 10−5	0.27	16.7
12p13.31	rs11064437	SPSB2	C/T	0.72/0.99	0.72	1.04 (1.01 – 1.07)	3.47 × 10−3	0.43	1.6
12pl3.2	rs2238126	ETV6	G/A	0.47/0.19	0.49	1.03 (1.00 – 1.06)	0.03	0.45	0.0
16q24.1	rs847208	FENDRR	A/C	0.68/0.62	0.67	1.08 (1.05 – 1.11)	2.62 × 10−7	0.04	44.2
17p13.3	rs12603526	NXN	C/T	0.20/0.01	0.32	1.06 (1.03 −1.09)	6.40 × 10−5	0.04	43.2
18q21.1	rs7229639	SMAD7	A/G	0.13/0.10	0.17	1.21 (1.17 – 1.25)	2.49 × 10−27	0.68	0.0
19q13.2	rs1800469	TMEM91	G/A	0.45/0.69	0.50	1.06 (1.03 −1.09)	1.61 × 10−5	0.42	2.8
20p12.3	rs2423279	HAO1	C/T	0.33/0.27	0.31	1.10 (1.06 −1.13)	2.87 × 10−10	0.20	24.2
20q13.12	rs6065668	TOX2	T/C	0.47/0.25	0.40	1.07 (1.04 – 1.10)	3.65 × 10−7	0.09	36.9
		Loci initially identified in GWAS of European-ancestry populationsc
1q25.3	rs10911251	LAMC1	A/C	0.48/0.55	0.54	1.05 (1.03–1.08)	1.03 × 10−4	0.42	2.9
2q35	rs992157	TMBIM1	A/G	0.62/0.58	0.59	1.04 (1.01 – 1.07)	6.26 × 10−3	0.79	0.0
3p22.1	rs35360328	CTNNB1	A/T	0.08/0.16	0.06	1.09 (1.02 – 1.16)	0.01	0.04	50.0
3p14.1	rs812481	LRIG1	G/C	0.79/0.56	0.78	0.97 (0.94 – 1.00)	0.07	0.25	19.4
3q26.2	rs10936599	MYNN	C/T	0.42/0.76	0.39	1.06 (1.04 – 1.09)	2.41 × 10−6	0.56	0.0
4q22.2	rs1370821	SMARCAD1	T/C	0.33/0.40	0.26	1.01 (0.98 – 1.04)	0.53	0.30	13.9
5p15.33	rs2735940	TERT	G/A	0.52/0.49	0.61	1.07 (1.04 – 1.10)	6.00 × 10−6	0.35	9.3
5p13.1	rs58791712	PTGER4	GT/G	0.04/0.26	0.04	1.12 (1.02 – 1.22)	0.02	0.61	0.0
6p21.31	rs6906359	FKBP5	C/T	0.93/0.90	0.94	1.04 (0.98 – 1.10)	0.20	0.69	0.0
6p21.2	rs1321311	CDKN1A	A/C	0.17/0.22	0.15	1.09 (1.05 – 1.13)	1.35 × 10−6	0.21	22.3
6p12.1	rs62404968	BMP5	C/T	0.94/0.75	0.96	1.08 (1.01 – 1.15)	0.03	0.34	10.7
8q24.21	rs6983267	POU5F1B	G/T	0.39/0.50	0.39	1.17 (1.14 – 1.20)	1.38 × 10−31	0.05	42.5
10p14	rs10795668	GATA3	G/A	0.63/0.68	0.61	1.14 (1.11 – 1.17)	5.20 × 10−24	0.27	17.2
10q11.23	rs10994860	A1CF	C/T	0.95/0.80	0.95	1.04 (0.97 – 1.10)	0.25	0.91	0.0
10q24.2	rs1035209	SLC25A28	T/C	0.18/0.19	0.18	1.08 (1.04 – 1.12)	1.12 × 10−5	0.64	0.0
11q13.4	rs3824999	POLD3	G/T	0.42/0.52	0.41	1.07 (1.04 −1.09)	3.16 × 10−6	1.00	0.0
11q23.1	rs3802842	COLCA2	C/A	0.40/0.27	0.39	1.07 (1.04 – 1.10)	5.00 × 10−7	0.25	18.6
12p13.32	rs3217810	CCND2	T/C	0.01/0.12	0.01	1.11 (0.97 – 1.29)	0.14	0.73	0.0
12q13.12	rs7136702	LARP4	T/C	0.44/0.34	0.53	1.02 (0.99 – 1.05)	0.20	0.90	0.0
12q13.12	rs11169552	ATF1	C/T	0.60/0.75	0.67	1.04 (1.01 – 1.07)	8.30 × 10−3	0.97	0.0
12q24.21	rs72013726	MED13L	C/CACAA	0.36/0.50	0.39	1.05 (1.01 – 1.09)	0.01	0.29	16.1
14q22.2	rs4444235	BMP4	C/T	0.47/0.49	0.55	1.04 (1.02 – 1.07)	7.85 × 10−4	0.68	0.0
14q22.2	rs1957636	BMP4	T/C	0.66/0.41	0.59	0.99 (0.97 – 1.02)	0.57	0.20	23.9
15q13.3	rs16969681	SCG5	T/C	0.37/0.07	0.46	1.08 (1.05 – 1.11)	1.31 × 10−8	0.06	41.8
15q13.3	rs4779584	SCG5	T/C	0.81/0.20	0.83	1.04 (1.01 – 1.08)	0.02	0.19	24.1
16q22.1	rs9929218	CDH1	G/A	0.77/0.71	0.83	1.05 (1.01 – 1.09)	9.22 × 10−3	0.65	0.0
16q24.1	rs2696839	FOXF1	G/C	0.75/0.50	0.75	1.03 (1.0 – 1.06)	0.04	0.50	0.0
18q21.1	rs4939827	SMAD7	T/C	0.31/0.53	0.24	1.13 (1.09 – 1.16)	5.44 × 10−15	0.84	0.0
19q13.11	rs10411210	RHPN2	C/T	0.81/0.90	0.83	1.10 (1.06 – 1.14)	1.22 × 10−7	0.17	27.0
20p12.3	rs961253	BMP2	A/C	0.11/0.36	0.10	1.08 (1.03 −1.12)	6.93 × 10−4	0.68	0.0
20p12.3	rs4813802	BMP2	G/T	0.23/0.32	0.21	1.10 (1.06 – 1.13)	1.44 × 10−8	0.32	0.0
20q13.13	rs6066825	PREX1	A/G	0.71/0.62	0.71	1.10 (1.07 – 1.13)	4.34 × 10−11	0.15	29.7
20q13.13	rs1810502	PTPN1	C/T	0.39/0.55	0.45	1.07 (1.04 – 1.10)	1.38 × 10−6	0.02	21.4
20q13.33	rs4925386	LAMA5	C/T	0.74/0.67	0.78	1.02 (0.99 – 1.05)	0.22	0.12	32.3

Alleles, risk/reference allele; RAF, risk allele frequency derived from the 1000 Genome Project (Phase 3) East Asians/Europeans (1KGP) or Asia Colorectal Cancer Consortium (ACCC); OR, odds ratio; CI, confidence interval; I² and P_Het-ASN: derived from the heterogeneity test across all studies included in Stage 1 and Stage 2.

^aVariants not replicated in previous studies were not evaluated in our study. Also not evaluated in our study were 9 variants that were initially reported GWAS conducted in European populations and have a very low MAF in Asians (<0.01%), including rs72647484 at 1p36.12, rs6691170 at 1q41, rs140355816 at 8q23.3, rs16892766 at 8q23.3, rs76316943 at 8q23.3, rs3184504 at 12q24.12, rs73208120 at 12q24.12, and rs17094983 at 14q23.1, or being on the X chromosome (rs5934683 at Xp22.2) (see Supplementary Table 5 for references). Sample size included in these analyses ranged from 62,256 to 70,506.

^bVariants at each locus with r² _EAS < 0.2 were listed; two loci (EIF3H at 8q23.3 and SMAD7 at 18q21.1) were initially reported in European populations, but variants identified in East Asian populations were not in LD with variants identified in European populations.

^cVariants at each locus with r² _EUR < 0.2 were listed; one locus (CCND2 at 12p13.32) was initially reported in East Asian populations, but variants identified in European populations were not in LD with variants identified in East Asian populations.

Functional Characterization of the Novel Risk Loci and cis-eQTL Analyses

To identify putative causal variants and genes underlying the observed associations, we used functional genomics data to annotate each of the lead variants we identified at the 13 novel loci and their correlated variants (r² ≥ 0. 80). Aligning these risk variants with histone methylation/acetylation marks and DNase hypersensitivity sites²⁸ revealed that variants at 8 loci (2p16.3, 3q22.2, 5q23.2, 12q12, 13q22.1, 16q23.2, 17p12 and 20q13.32) overlapped with the promoter/enhance histone marks or DNase hypersensitivity sites in gastrointestinal tissues (Supplementary Table 6). This suggests that these variants may be involved in regulating gene expressions in gastrointestinal tissues. To identify potential target genes, we performed cis-eQTL analyses using data from adjacent normal colon tissues, obtained from 133 CRC patients of East Asian ancestry (Supplementary Table 7), as well as transverse colon tissues obtained from 246 individuals predominantly of European ancestry in the GTEx (Supplementary Table 8). Significant correlations at P < 0.05 were found for 21 and 37 SNP-gene pairs in the East Asian and GTEx data set, respectively. For three of these SNPgene pairs, an identical association direction was found in both data sets. The risk A allele of rs1476570 was associated with reduced expressions of HLA-G and HLA-V; the risk T allele of rs3830041 was associated with reduced MICA expressions.

Pathway analysis of genome-wide summary statistics of CRC risk

We performed pathway analyses based on genome-wide summary statistics of CRC risk. The top enriched pathways were related to mesenchymal cell proliferation, Smad protein phosphorylation, pluripotent states of reprogrammed somatic cells, embryonic development, MHC (HLA) protein complex, gland morphogenesis and epithelial cell migration (FDR P value < 0.05, Supplementary Table 9).

Familial Relative Risk of CRC Explained and CRC risk of Weighted PRS

We estimated that the 14 novel risk variants identified in this study combined explain approximately 3.5% of the familial relative risk of CRC in East Asian populations, comparable to the familial relative risk (4.1%) explained by the 19 risk variants identified previously in the Asian population (Supplementary Table 9). An additional 4.1% of the familial relative risk in East Asian populations can be explained by the 24 risk variants initially identified in studies conducted among European-ancestry populations and replicated in this study. Together, 11.7% of the familial relative risk of CRC in individuals of East Asian ancestry can be explained by the 57 CRC risk variants newly identified (n = 14) and replicated (n = 43) in our study.

There was a clear dose-response association between weighted PRS and CRC risk. Individuals in the highest PRS quintile group (≥ 4.80) had a 3.2-fold increased CRC risk (OR (95%CI) = 3.16 (2.88 – 3.47)) when compared with individuals in the lowest PRS group (< 4.10) (Supplementary Table 11).

Discussion

In this large GWAS, with a sample size of 22,775 CRC cases and 47,731 controls of East Asian descent, we identified, at P<5 × 10^–8, 13 new risk loci for CRC and an independent risk variant in a locus previously reported in a GWAS conducted in European descendants. We replicated all of the 19 independent CRC risk variants identified in previous GWAS conducted in East Asian populations and 26 of the 43 risk variants initially identified in GWAS conducted in European-ancestry populations. Using functional genomics data, we showed that most of the newly identified risk variants, or their highly correlated variants, are located in functional regions of the genome, contributing to the regulation of genes with established roles in colorectal tumorigenesis or cellular functions involved in cancer development pathways. Our cis-eQTL analyses provide additional evidence supporting a possible role of several risk variants identified in our study in regulating expression of cancer-related genes. The pathway analysis corroborated with the known biological roles of Smad protein regulation and cellular pluripotency in CRC carcinogenesis, and suggested the HLA protein complex in regulating CRC risk. Our study provides substantial novel information towards the understanding of the genetic and biological basis for CRC.

Of the 13 novel loci identified in our study, seven of the lead risk variants are located inside a protein-coding gene, including a nonsynonymous SNP (rs7542665) located in the L1TD1 gene (1p31.3). However, using the PolyPhen2 prediction, rs7542665 and two other nonsynonymous SNPs (rs7533274 and rs11207933) that are in high LD with rs7542665 (r² EAS ≥ 0. 80) are classified as “benign”, and thus may not affect the function of the protein encoded by L1TD1. Interestingly, the risk allele of rs7542665 was associated with increased L1TD1 expression in multiple tissues investigated in the GTEx (Supplementary Figure 7). L1TD1 is a RNA-binding protein, and this protein is one of the most specific and abundant proteins in pluripotent stem cells, which is essential for the maintenance of pluripotency in human cells³⁴. It was found that elevated L1TD1 levels might increase the risk of cancer by increasing cellular pluripotency³⁵.

The other six lead risk variants located inside a protein-coding gene are all intronic variants. Findings for rs1476570 and rs3830041 are particularly interesting, as they are located inside the HLA-G and NOTCH4 genes, respectively. These genes play a central role in immune regulation and response. Our eQTL analyses of East Asian CRC samples showed that rs1476570 was associated with the expression of multiple HLA genes, including HLA-F, HLA-G and HLA-V, and rs3830041 was associated with the expression of HLA-DQB2, LTB and MICA (Supplementary Table 7). The risk allele of rs3830041 was associated with reduced MICA expressions. Reduced MICA expressions on the cancer cell surface were found to inhibit NK cell–mediated antitumor immunity³⁶. Genetic variants in this region are associated with infectious diseases, Crohn’s diseases in the Ashkenazi Jewish population, and blood cell phenotypes^37–39. To our knowledge, no previous GWAS has identified any CRC risk loci in genes with a primary role in immune regulation. Our study provides the first piece of evidence that germline variation in immune response genes may play a significant role in the pathogenesis of CRC.

Other intronic risk variants identified in this study also reveal potential novel pathways for CRC pathogenesis. For example, the protein encoded by PPP1R21 at 2p16.3 (rs7606562) regulates the dephosphorylation activity of the ubiquitous and conserved protein phosphatase 1 enzyme. This protein and its regulatory proteins are involved in signaling pathways that regulate cellular growth, cell cycle and apoptosis⁴⁰. SNP rs77969132 is located in the intron of the DENND5B gene, encoding GDP-GTP exchange factors 5B, putatively acting on Rab39 in membrane trafficking control⁴¹. SNP rs13831 is an intronic variant for the GNAS gene, encoding the Gsα subunit of the heterotrimeric G proteins. Activating mutations in this gene was thought to promote intestinal tumorigenesis through the activation of the Wnt and ERK1/2 MAPK pathways⁴².

Two of the intronic risk variants identified in our study have a low-frequency allele. SNP rs201395236 is located in the intron 2 of the EFCAB2 gene (MAF = 1.3%, OR = 1.75 for the major allele), while rs77969132 is located in the intron 8 of the DENND5B gene (MAF = 1.5%, OR = 1.44 for the minor allele). These are the first two low-frequency variants identified as having an association with CRC risk, and the risk associated with these variants is substantially higher than that of the common risk variants identified for CRC risk. These variants will be invaluable in estimating individual CRC risk.

We identified six lead risk variants located in intergenic regions. SNP rs113569514, at 3q22.2, is mapped to the 5’ region of SLCO2A1. This gene encodes a principal transmembrane prostaglandin transporter that mediates the uptake and clearance of extracellular prostaglandins⁴³. Germline mutations in SLCO2A1 were found to contribute to familial CRC risk⁴⁴. Using data from the GTEx project, we found that the risk allele of rs113569514 was associated with a reduced expression level of SLCO2A1 in multiple tissues³⁰, perhaps leading to a reduced prostaglandin catabolism, and subsequently, an increased CRC risk⁴³. Our eQTL analyses also identified a significant correlation of rs1886450 at13q22.1 with KLF5 expression³⁰, rs1078643 at 17p12 with SHISA6, DNAH9 and GLP2R expression, and rs2730985 at 12q12 with PRICKLE1 expression (Supplementary Table 7 and 8). The chromosome region containing KLF5 is significantly amplified in CRC⁴⁵. Kruppel-like factor 5 is a zinc-finger transcription factor that is indispensable for the integrity and oncogenic transformation of intestinal stem cells⁴⁶. GLP2R is expressed predominantly within the gastrointestinal tract^{47, 48}, and has been shown to maintain the integrity of the intestinal epithelium by both stimulating cell proliferation and inhibiting apoptotic cell death in the crypt compartment^48–52. The PRICKLE1 gene encodes a Dishevelled-associated protein that negatively regulates the Wnt/β-catenin signaling pathway⁵³.

For the other two intergenic risk variants, we could not find any significant correlation with any genes. SNP rs2659017 at 5q23.2 is located 57 kb upstream of ALDH7A1. This gene encodes an enzyme of the aldehyde dehydrogenase superfamily, which degrades and detoxifies aldehydes generated by alcohol metabolism and lipid peroxidation. SNP rs4341754 at 16q23.2 is located downstream of the proto-oncogene MAF (405 kb) and the tumor suppressor gene WWOX (793 kb). Several variants in this region are associated with oral cavity and pharyngeal cancer risk (rs4284656)⁵⁴ and breast cancer risk (rs13329835)⁵⁵ in European-ancestry populations. However, rs4341754 is not in LD with these GWAS-identified risk variants (r²_{EUR or EAS} = 0.00). The region containing WWOX is recurrently deleted in multiple cancers⁵⁶, including CRC⁴⁵. The WW domain containing oxidoreductase, encoded by WWOX, is involved in maintaining genomic stability, and its loss leads to genomic instability and cancer development^{57, 58}

Efforts to replicate the associations for these 13 new risk loci on CRC risk in additional Asian descendants were not undertaken in this study. However, the lead risk variants in five of these loci also showed a significant association with CRC risk in European descendants. Given the differences in genetic architectures between Asian and European descendants, it is not surprising that some of the risk variants identified in one population cannot be directly replicated in another population. Fine-mapping of these loci will be helpful to identify risk variants that may be more specifically related to CRC risk in European descendants. On the other hand, we could not completely rule out the possibility of false positive findings for some of these newly identified risk variants. Replication of these findings in additional Asian descendants should be helpful to validate our finding, especially for the 8 loci that were not replicated in European descents in relation to CRC risk.

In summary, we identified 14 novel risk variants for CRC, including 13 in loci not yet reported in previous studies. These risk variants, along with the common risk variants replicated in our study, explain about 11.7% of the familial relative risk of CRC in the East Asian population. Some of the putative target genes suggested by results from our studies are located in established pathways for colorectal tumorigenesis, such as the Wnt/β-catenin signaling (PRICKLE1) and prostaglandin E2 catabolism (SLCO2A1). Our study also suggests that genes involved in maintaining colon stem cells (L1TD1 and KLF5) or regulating the proliferation of an intestinal epithelium cell (GLP2R) may play a significant role in the pathogenesis of CRC. Finally, we provided evidence, for the first time, that genetic variations in the HLA gene region may contribute to the susceptibility of CRC, supporting a critical role in immune regulation and response in the development and progression of CRC.

Figure 1.

Manhattan plot showing the joint meta-analysis association statistics (-log₁₀ (P values)) of Stage 1 and Stage 2 (N case = 22,775, N control = 47,731). Known loci in blue and novel loci in red.

In a study of more than 60,000 East Asian descendants, 13 new genetic loci were discovered that are associated with an increased risk of colorectal cancer.