Genome-wide association identifies multiple ulcerative colitis susceptibility loci

Abstract

Ulcerative colitis (UC) is a chronic, relapsing inflammatory condition of the gastrointestinal tract with a complex genetic and environmental etiology. We performed two distinct UC genome-wide association (GWA) studies, and analyzed these jointly with a previously published scan¹, comprising, in aggregate, 2,693 patients with UC and 6,791 controls. A total of 59 SNPs from 14 independent loci attained P < 10⁻⁵. Seven of these loci exceeded genome-wide significance (P < 5 × 10⁻⁸). After testing an independent cohort of 2009 patients with UC and 1580 controls, 14 loci were significantly associated, including novel UC associations with FCGR2A, 5p15, 2p16, CARD9 and ORMDL3. In our study we confirmed association with 14 previously identified UC susceptibility loci, while an analysis of acknowledged Crohn's disease (CD) loci showed that roughly half of known CD associations are shared with UC. These data implicate approximately 30 loci for UC, providing novel insights into disease pathogenesis.

Epidemiological studies suggest that UC and CD share some, but not all, susceptibility genes, a hypothesis supported by recent GWA studies. Meta-analysis of three CD GWA studies increased the number of susceptibility loci known for CD to over 30². UC is a condition of more modest heritability (λ_s 10–15) compared to CD (λ_s 20–35), and perhaps as a result, fewer loci have been identified for UC to date^1,3–6. Previous experience with CD implies that the currently identified UC susceptibility loci explain only a fraction of the genetic contribution to disease susceptibility. Considerations of statistical power also suggest that additional loci could be found by enlarging the number of cases and controls used in genome-wide discovery. Here, we combine data from two new UC GWA studies, and perform a meta-analysis with a recently published study¹. This brings together a discovery set of 2,693 patients with UC and 6,791 controls, all of European descent (Supplementary Table 1). Independent replication of top results from this meta-analysis was then performed in 2009 cases and 1580 controls, also of European origin.

All primary studies used similar Illumina BeadChips, allowing us to directly examine 266,047 (258,137 autosomal and 7,910 X-chromosomal) SNPs that passed QC in each study. The three studies were analyzed and corrected for population structure separately. P-values from each study were converted to Z-scores summarizing the direction and magnitude of association evidence, and combined (weighted by the relative size of each study) using standard methods⁷. A Q-Q plot (Supplementary Fig. 1) shows a significant excess of likely true positives in the tail of the distribution, against only modest overall inflation (λ_GC=1.036).

We sought to replicate 14 independent loci represented by one or more of the 59 SNPs with P < 10⁻⁵ in the meta-analysis (Table 1 and Supplementary Table 2). Several previously identified UC loci were among these identified in the meta-analysis, although not all had previously attained genome-wide significance. Loci with prior UC association included IL23R³, the HLA region (including the BTNL2 association)^1,3,8, MST1⁹, CARD9⁶, 1q32 (near IL10)⁵, 1p36 (RNF186/OTUD3/PLA2G2E)¹, DLD/LAMB1¹ (recently confirmed in a British UC GWA study¹⁰) 12q15 (neighboring to IFNG/IL26)¹, and 21q22¹¹. A second, back-up, SNP was chosen for the six regions with P < 10⁻⁶ in the discovery set. Additional SNPs (N=4) from the 1p36, 2p16 and 12q15 regions were chosen, because these SNPs appeared to be independently associated (r² < 0.2 in CEU HapMap). We also included 10 SNPs with P values between 10⁻⁴ and 10⁻⁵. Finally, previously published associations with CD and UC were included. These SNPs were typed in 2009 UC cases and 1580 controls of Dutch and Italian descent (complete data in Supplementary Table 1).

Table 1.

Loci with association with UC in GWA STUDY meta-analysis (p > 10⁻⁸) with replication cohort results

					GWA STUDIES				REPLICATION			GWA + REPLIC	@
SNP	Chr	Position	Genes	variant (major/ minor)	CEDARS	SWEDEN	NIDDK	Combined	ITALIAN	DUTCH	Combined
rs3806308	1	19612341	RNF186 OTUD3 PLA2G2E	G/A	0.034	0.047	4.70E-08	3.28E-08	-	-	N/A	N/A	1
rs1317209	1	20012623	RNF186 OTUD3 PLA2G2E	C/T	2.43E-04	6.26E-03	0.034	8.60E-07	0.046	1.56E-04	4.41E-05	1.60E-10
rs6426833	1	20044447	RNF186 OTUD3 PLA2G2E	G/A	1.02E-04	2.79E-04	6.8E-10	2.66E-15	4.89E-04	3.87E-05	7.65E-08	1.70E-21
rs2201841	1	67466790	IL23R	T/C	0.023	4.90E-03	1.10E-06	8.91E-09	6.72E-03	9.85E-05	3.01E-06	1.28E-13	2
rs11209026	1	67478546	IL23R	G/A	7.28E-04	1.47E-03	3.10E-05	5.91E-10	0.012	1.32E-03	5.18E-05	1.89E-13
rs10800309	1	159738782	FCGR2A FCGR2C	G/A	2.64E-03	0.083	7.30E-04	2.78E-06	2.15E-04	0.14	2.52E-04	2.76E-09
rs3024505	1	205006527	IL10 IL19	C/T	3.06E-03	0.012	9.30E-03	3.24E-06	2.09E-03	0.12	1.06E-03	1.37E-08
rs6706689	2	61024549	REL CCDC139 PUS10	G/A	0.038	1.96E-04	0.031	4.40E-06	1.11E-03	0.15	8.88E-04	1.53E-08
rs13003464	2	61040333	REL CCDC139 PUS10	A/G	0.012	2.41E-03	9.50E-05	4.67E-08	0.21	0.027	0.014	7.40E-09	2
rs3197999	3	49696536	MST1	C/T	4.55E-03	2.00E-04	0.07	1.36E-06	1.90E-04	0.28	6.67E-04	3.76E-09
rs4957048	5	636180	CEP72 TPPP	C/T	0.19	5.28E-03	1.20E-03	2.28E-05	3.52E-04	7.09E-03	9.38E-06	1.18E-09
rs2395185	6	32541145	C6orf10 BTNL2	G/T	1.36E-07	4.42E-12	1.40E-06	8.75E-23	-	-	N/A	N/A	1
rs4598195	7	107290677	DLD LAMB1	A/C	0.098	4.94E-04	1.30E-05	4.18E-08	0.46	0.08	0.075	7.70E-08
rs4077515	9	138386317	CARD9	A/G	2.02E-03	0.012	4.90E-03	1.17E-06	6.34E-04	0.75	8.26E-03	5.48E-08	2
rs11190140	10	101281583	NKX2-3	C/T	8.84E-04	0.028	0.058	1.85E-05	7.61E-06	0.37	1.45E-04	1.07E-08
rs1558744	12	66790859	IFNG IL26	G/A	4.24E-03	0.028	5.50E-10	8.14E-11	2.38E-04	0.57	2.70E-03	4.18E-12
rs971545	12	66877952	IFNG IL26	A/G	2.40E-04	0.097	3.10E-04	2.44E-07	0.012	0.055	1.73E-03	2.23E-09
rs2305480	17	35315722	ORMDL3 region	C/T	3.93E-04	8.45E-04	0.18	2.06E-06	0.19	4.29E-03	3.22E-03	3.01E-08	2, 4
rs2836878	21	39387404	near PSMG1	G/A	1.92E-03	2.67E-04	2.10E-03	1.42E-08	-	-	N/A	N/A	1

@ Notes:

¹SNP assays failed to design or failed genotyping in Replication Stage

²Highly correlated proxies (r2>0.9) were typed in the Replication Stage (see Supplementary Table 2 for further details

³Region contains SULT6B1 CEBPZ C2orf56 PRKD3 QPCT

⁴Region contains STAC2 FBXL20 MED1 CRKRS NEUROD2 PPP1R1B STARD3 TCAP PNMT PERLD1 ERBB2 C17orf37 GRB7 IKZF3 ZPBP2 GSDML ORMDL3

Overall 13 loci achieved association with genome-wide statistical significance (P < 5 × 10⁻⁸) (Table 1, Supplementary Table 2). A fourteenth locus, containing the CARD9 gene, did not reach genome-wide significance but was significant after Bonferroni correction (P_nominal = 5.48 × 10⁻⁸; P_corrected = 0.014). We achieved genome-wide significance for at least 4 novel loci in populations of European origins, including 1q21 (FCGR2A/FCGR2C)(this locus was recently shown to be associated with UC in a Japanese population¹²), 2p16 (REL/PUS10), 17q12 (ORMDL3) and 5p15 (rs4957048; approximately 30 kilobases from CEP72) (Table 1). Fourteen loci either suggestively or conclusively associated with UC in prior studies, including TNFSF15, NKX2-3, IL12B, MST1, IL18RAP, HLA, IBD5 locus (5q31), RNF186/OTUD3/PLA2G2E, DLD/LAMB1, IL10, CARD9, 12q15 (IFNG/IL26), JAK2, and IL23R were replicated in our study. Finally, we demonstrated association with 10 additional loci previously associated with CD or IBD, including the IRGM, KIF21B, IKZF1, ICOSLG, CCL2/CCL7, 5p13 (near to PTGER4), 21q21, CUL2/CREM, PSMG1 and STAT3 (Table 2). Replication of these previously identified UC, CD or IBD loci was defined as an association at a level of P < 0.05, with the same risk allele as identified in the index studies.

Table 2.

Association between UC and known CD loci.

				GWA STUDIES				REPLICATION			GWA + REPLICATION
Gene of interest	SNP from Barrett et al.	Chr#	CD meta p-value (Barrett et al.)	CEDARS	SWEDEN	NIDDK	Combined	ITALIAN	DUTCH	Combined	P-value	Notes
IL23R	rs11465804	1p31	6.38E-34	7.28E-04	1.47E-03	3.10E-05	5.91E-10	0.012	1.32E-03	5.18E-05	1.89E-13	1,2,4,5
ATG16L1	rs3828309	2q37	2.57E-21	0.24	0.18	0.25	0.57	0.28	0.63	0.67	0.48	1
MST1	rs3197999	3p21	2.17E-07	4.55E-03	2.00E-04	0.07	1.36E-06	1.90E-04	0.28	6.67E-04	3.76E-09	4
PTGER4	rs4613763	5p13	5.02E-22	0.57	0.62	0.0049	0.11	0.048	1.58E-03	2.81E-04	4.21E-04
IBD5	rs2188962	5q31	4.58E-09	0.14	0.40	0.32	0.35	0.044	2.50E-03	3.68E-04	2.90E-03	2
IRGM	rs13361189	5q33	8.17E-11	-	-	-	N/A	0.015	4.78E-05	4.33E-06	N/A
TNFSF15	rs4263839	9q32	2.61E-07	0.024	0.42	0.56	0.035	0.2996	2.85E-04	9.72E-04	2.01E-04	1,2
ZNF365	rs10995271	10q21	1.56E-07	0.024	0.73	0.89	0.24	0.1852	0.96	0.34	0.127	1
NKX2-3	rs11190140	10q24	1.71E-10	8.84E-04	0.028	0.058	1.85E-05	7.61E-06	0.37	1.45E-04	1.07E-08
NOD2 (R702W)	rs2066844	16q12	N/A	-	-	-	N/A	0.25	0.64	0.25	N/A
NOD2 (G908R)	rs2066845	16q12	N/A	-	-	-	N/A	0.69	0.017	0.049	N/A
NOD2 (insC)	rs2066847	16q12	N/A	-	-	-	N/A	0.90	0.077	0.18	N/A
NOD2 (R459R)	rs2066843	chr16	1.20E-27	0.36	0.99	0.62	0.41	-	-	N/A	N/A	3
PTPN2	rs2542151	18p11	6.54E-11	0.14	0.57	0.75	0.17	0.98	0.17	0.32	091	1
PTPN22	rs2476601	1p13	1.81E-05	0.11	0.32	0.096	0.19	0.25	0.15	0.067	0.88
ITLN1	rs2274910	1q23	3.51E-07	0.11	0.15	0.24	0.45	0.51	0.52	0.36	0.24	2
-	rs9286879	1q24	4.01E-07	0.90	0.75	0.85	0.71	0.19	0.063	0.70	0.60
KIF21B	rs11584383	1q32	1.94E-06	0.16	0.024	8.50E-02	1.88E-03	0.020	1.08E-04	1.17E-05	2.22E-07	1
IL12B	rs10045431	5q33	8.81E-09	-	-	-	N/A	0.024	0.023	1.39E-03	N/A
CDKAL1	rs6908425	6p22	2.53E-07	0.49	0.61	0.17	0.14	0.29	0.97	0.44	0.51
ATG5	rs7746082	6q21	3.13E-04	0.069	5.68E-03	0.042	1.34E-04	0.23	0.88	0.46	6.08E-04	1
CCR6	rs2301436	6q27	3.29E-07	0.044	0.16	0.094	3.30E-03	0.38	0.28	0.89	0.029
IKZF1	rs1456893	7p12	3.16E-05	0.93	0.66	0.53	0.960	0.15	1.14E-03	9.05E-04	0.040	1
-	rs1551398	8q24	4.90E-06	0.57	0.32	0.73	0.27	4.24E-04	0.057	0.25	0.90	2
JAK2	rs10758669	9p24	6.80E-07	0.023	0.028	0.49	2.88E-03	8.17E-04	0.023	7.06E-05	1.42E-06
CUL2/CREM	rs17582416	10p11	2.21E-05	0.69	0.028	0.92	0.15	0.032	0.45	0.040	0.016	1,2
C11orf30	rs7927894	11q13	1.43E-07	0.021	0.79	0.062	0.010	0.20	0.69	0.23	0.22	1
LRRK2, MUC19	rs11175593	12q12	1.33E-07	-	-	-	N/A	0.071	0.77	0.28	N/A
-	rs3764147	13q14	1.44E-05	0.27	0.010	0.041	0.34	0.59	0.76	0.55	0.27	1
ORMDL3	rs2872507	17q21	2.13E-06	3.93E-04	8.45E-04	0.18	2.06E-06	0.19	4.29E-03	3.22E-03	3.01E-08	1,4
STAT3	rs744166	17q21	5.94E-06	0.35	0.77	0.012	0.031	0.40	0.084	0.069	4.83E-03
-	rs1736135	21q21	3.28E-05	0.36	0.053	5.90E-04	2.88E-04	0.60	8.21E-07	1.15E-04	1.54E-07	1
ICOSLG	rs762421	chr21	1.08E-05	-	-	-	N/A	1.70E-03	0.59	9.32E-03	N/A	2
-	rs4807569	19p13	1.30E-08	0.99	0.53	0.058	0.15	-	-	N/A		1,3
GCKR	rs780094	2p23	7.24E-05	0.70	0.95	0.18	0.30	0.29	0.47	0.81	0.81	2
BTNL2, SLC26A3, HLA-DRB1, HLA-DQA1	rs3763313	6p21	1.45E-08	0.015	0.20	0.25	5.05E-03	0.020	0.71	0.17	0.19
PUS10	rs13003464	2p16	7.65E-06	0.012	2.41E-03	9.50E-05	4.67E-08	0.21	0.027	0.014	7.40E-09	4
CCL2, CCL7	rs991804	17q12	4.01E-06	0.52	0.060	0.46	0.25	0.013	0.028	9.33E-04	2.88E-03	2
LYRM4	rs12529198	6p25	7.08E-07	-	-	-	N/A	0.23	0.78	0.51	N/A
SLC22A23	rs17309827	6p25	2.08E-06	-	-	-	N/A	-	-	N/A	N/A	3
-	rs7758080	6q25	7.28E-06	0.67	0.18	0.54	0.38	0.50	0.079	0.44	0.84
-	rs8098673	18q11	3.17E-05	0.97	0.86	0.92	0.89	0.16	0.29	0.81	0.96
IL18RAP	rs917997	2q11	2.17E-05	0.58	0.58	0.012	0.15	0.35	0.026	0.025	0.011

Notes:

¹= Highly correlated proxies were evaluated in the UC GWA studies (See Supplementary Table 2 for further details)

²= Highly correlated proxies were typed in the Replication Stage (See Supplementary Table 2 for further details)

³= SNP assays failed to design or failed genotyping in Replication Stage

⁴= also reported in Table 1

⁵= rs11465804 was captured by the highly correlated proxy rs11209026 in both scan and replication (Table q and Supplementary Table 2)

Two recently published studies performed in populations of European descent identified additional IBD associated loci^10,13. We examined these loci within our discovery set (using the exact SNP or a perfect proxy (r² = 1.0)) and were able to demonstrate association between UC and 22q12¹³ (an IBD locus)(p = 0.049), 16q22/CDH1¹⁰ (a UC locus)(p = 0.0061), as well as with the DLD/LAMB1¹⁰ locus discussed earlier. Furthermore we are able to provide further evidence of association between the 1p36 locus (rs7524102, p = 0.0015) which demonstrated strong (but not genome-wide significant) association in the WTCCC study¹⁰. We were not able to replicate the UC associations seen at 13q13¹⁰ (p = 0.15), 20q13/HNF4A¹⁰ and 2q37/GPR35 (SNPs not in our dataset and no proxy SNPs available) or the IBD associations seen at 16p11 (p = 0.17), 10q22/ZMIZ1 (p = 0.22), and 19q13 (SNP not in our dataset and no proxy SNP available). Therefore, adding these additional associations we provide evidence for at least 30 distinct risk factors for UC.

In three regions (1p36 [RNF186/OTUD3/PLA2G2E], 2p16 [PUS10/REL] and 12q15 [IFNG/IL26]), our scan identified multiple SNPs with P < 10⁻⁵ that showed low LD with each other (r² < 0.2) in HapMap CEU samples. To determine if multiple independent alleles were contributing at these loci, we performed conditional analyses in all scan and replication samples (Table 3). While the signals for each SNP conditional on the other are greatly diminished in the 2p16 region (suggesting these may both be in LD with a single, yet to be discovered, causal variant), we have extended prior evidence for multiple independent associations at 1p36 and 12q15¹. Particularly striking are three SNPs at 1p36, each of which is genome-wide significant even while conditional on the other two.

Table 3.

Conditional analyses for three regions with multiple associated SNPs in linkage equilibrium.

SNP	Location	Conditional on	CEDARS	NIDDK	SWEDEN	ITALIAN	DUTCH	Combined-P
rs1317209	1p36	rs3806308	7.95E-05	0.043	0.00445			2.75E-06
		rs6426833	0.0002674	0.044	0.01	0.04976	0.0001907	3.10E-08
rs3806308	1p36	rs1317209	0.02201	3.40E-08	0.06			1.39E-08
		rs6426833	0.0348	2.50E-07	0.13			2.64E-07
rs6426833	1p36	rs3806308	6.37E-05	3.10E-09	0.00035			5.05E-14
		rs1317209	0.0001061	5.70E-10	0.00022	0.00055	7.48E-05	<1e-16
rs6706689	2p16	rs13003464	0.3137	8.30E-01	0.01663	0.00256	0.7145	0.0091
rs13003464	2p16	rs6706689	0.08581	7.00E-04	0.12	0.8752	0.0681	0.0011
rs1558744	12q15	rs971545	0.04439	5.00E-08	0.2	0.0018	0.9499	2.20E-08
rs971545	12q15	rs1558744	0.002071	4.60E-02	0.2	0.09206	0.06118	0.000386

We also performed interaction analysis between all pairs of SNPs listed in table 1. Among the 496 pairs of SNPs examined (Supplementary Table 3) one pair (an interaction between genome-wide significant hits at CARD9 and REL/PUS10) was significant after correction for the number of tests performed and a second interaction at this locus approached significance after replication. Few such interactions have been documented in complex disease¹⁴ and further replication is warranted before this is considered confirmed but the known functional interaction between CARD9 and REL should not be overlooked.

It is clear from these (Table 2) and other data that UC and CD share some mechanistic pathways and susceptibility genes but that some are particular to each condition. We sought to estimate the proportion of alleles with an influence on both CD and UC by calculating the likelihood of the observed UC genotype data at each CD locus, under the alternate hypotheses that the UC sample has the same allele frequency as either a CD sample or of a control sample. The maximum data set likelihood was achieved when 15 or 16 of the CD loci – essentially half of the 31 - were presumed to affect UC risk. This is consistent with the summary results in table 2, where 14 of the 31 confirmed previous findings are significant at p<.01, three have 0.01<p<.05, while the remainder fit the null distribution. The lack of complete overlap is unlikely to result entirely from limited power because our likelihood analysis limits the possibility that more than 20 of the Crohn's loci are shared, and some of the strongest CD variants (e.g., NOD2, ATG16L1) are among the loci not associated with UC, despite having increased power to detect these variants of larger effect size.

All of the genes implicated by our work could plausibly play a functional role in UC (see Supplementary Table 4 for genes located nearby to replicated UC-risk SNPs not shared with CD). A key next step in translating genetic loci to function requires that we understand gene function in the context of cell and tissue type relevant to human disease. Expression studies may help identify relevant cell types for functional studies, the nature of which are likely to differ for genes expressed in epithelial cells, macrophages or lymphocytes. We focused on five UC susceptibility loci containing genes whose expression and function are not well established. We performed quantitative real-time PCR in human intestinal and immune cDNA panels (Supplementary Fig. 2a). The RNF186/PLA2G2E/OTUD3 genes are located at 1p36 which harbors distinct UC-risk SNPs (rs1317209, rs6426833). While RNF186 and OTUD3 are proteins with unknown function, both contain protein domains that have been associated with protein ubiquitination. Ubiquitin modifications are known to regulate immune responses, since the OTU protein TNFAIP3 was identified as an important negative regulator of NFkB¹⁵. Interestingly, the TNFAIP3 locus, which has been associated with multiple autoimmune conditions^16,17, is in the vicinity of one the UC-risk regions (rs2327832) although this locus achieved association only at a suggestive level of genome-wide significance (P = 3.92 × 10⁻⁰⁵). The expression of RNF186 is higher in intestinal tissues than immune tissues. Immuno-staining indicated that RNF185 is expressed at the basal pole of epithelial cells and lamina propria within colonic tissues (Supplementary Fig. 2a and 2b). In contrast, OTUD3 transcripts had higher levels in immune tissues (spleen, lymph nodes and PBMCs) and in lymphocytes compared to CD14 positive cells (not shown). Phospholipase A2 group IIE (PLA2G2E) is a secretory PLA2 involved in the production of various types of proinflammatory lipid mediators¹⁸. PLA2G2E was undetectable in most tissues but showed weak expression in bone marrow, lymph node and thymus suggesting a role in immunity. Surprisingly, in the intestinal panel, PLA2G2E was detected very specifically in the small intestine, but not colon. RNF186, OTUD3 and PLA2G2E showed very different expression patterns suggesting that investigations of the biological functions of these candidate genes will require disparate strategies.

CEP72/TPPP genes are located close to the UC-risk SNP rs4957048. CEP72 and TPPP are both involved in microtubule organization and are expressed ubiquitously (Supplementary Fig. 2a and 2b). LAMB1 and DLD are located near the UC risk SNPs rs4598195 and rs2237686 at 7q31 and were both expressed ubiquitously (Supplementary Fig. 2a). LAMB1 is an extracellular matrix glycoprotein constituent of basement membranes. These data together with recently reported associations between UC and CHD1 and CHD3¹⁰ further implicate defects in barrier integrity in the development of colonic inflammation. Importantly future studies will need to integrate disease associations and the consequences of risk allele specific expression to uncover functional role for genes in diseases such as UC.

GSDMB and ORMDL3 genes are located nearby to the UC risk SNPs rs2305480 and rs8067378. Both ORMDL3 and GSDMB showed higher expression in immune tissues in comparison to intestinal tissues (Supplementary Fig. 2a). We selected the ORMDL3 region for functional analysis not only because of the novel association with UC presented herein but also because ORMDL3 has been implicated in many diseases involving dysregulated immune responses, although the underlying mechanisms of this association remain unclear^2,19–21. ORMDL3 is suggested to be involved in protein folding and growing evidence demonstrate interactions between the Unfolded Protein Response (UPR) and immune responses^22,23. The ORMDL3-GFP protein is localized in the ER (Fig. 1a), confirming previous published data²⁴. We next investigated whether ORMDL3 expression is involved in the UPR in epithelial cells. Over-expression of ORMDL3 decreased both the basal and ER-stress induced UPR (Fig. 1b). Knockdown of ORMDL3 expression induced a higher UPR following tunicamycin or thapsigargin stimulation (Fig. 1c), indicating that ORMDL3 expression levels can regulate UPR and that ORMDL3 maybe an important factor to ensure ER homeostasis.

Figure 1.

(a) Cos-7 cells were transfected with pCMV-ORMDL3-GFP. After 24h, cells were fixed using 4% paraformaldehyde and subjected to immunostaining using anti-PDI antibodies as an ER marker. Fluorescence signals were acquired with a confocal microscope. Displayed images represent a single confocal section.

(b) HEK293T cells were co-transfected with a firefly luciferase reporter p5xUPRE-GL3, renilla luciferase as transfection control and pCMV-3XHA or pCMV-ORMDL3-3XHA. After 24h, cells were left untreated or treated using 5μg/ml of tunicamycin or 10μM of thapsigargin for 6h. ORMDL3-3XHA was detected using an anti-HA antibody. Luciferase activities were measured to detect UPRE transcription activation.

(c). HEK293T cells were co-transfected with plasmids encoding shRNA scramble or directed against ORMDL3, firefly luciferase reporter p5xUPRE-GL3 and renilla luciferase as transfection control. After 48h, cells were left untreated or treated using 5μg/ml of tunicamycin or 10μM of thasigargin. ORMDL3 expression was determined by quantitative RT-PCR (left). Luciferase activities were measured to detect UPRE transcription activation (right)

The data presented herein significantly increase our understanding of the pathogenesis of UC and its relationship to CD, as well as providing novel expression and functional data for some of the implicated genes. The genetic associations described in this report taken together with recently published papers^10,12,13 highlight the importance of barrier function, cell specific innate responses (ER stress, microbe elicited responses ROS production, NF-kB activation), gene sets that coordinately regulate key functional programs in adaptive immunity and resolution of inflammation in UC pathogenesis. Taken together we have explained less than 10% of the variance of UC and the challenge now is to both identify additional genetic factors and to translate these advances into real benefits for patients.

Methods

Two unpublished UC GWAS were combined with the published NIDDK UC GWAS¹ (Supplementary Table 1).

Cedars-Sinai (CS) UC GWAS

The CS UC scan consisted of 852 cases and 3271 controls. Samples with more than 1% missing genotype data (1 case, 278 controls) were removed. Standard searches for relatedness (using PLINK –genome) and population structure (using PLINK –mds)²⁵ were performed. Identity-by descent of >20% (catching half-sibs and above) was identified in 81 pairs (primarily controls) and one member from each pair was removed. The first ten principal components: PC1 represented the expected N. Europe – S. Europe – Ashkenazi axis were detected. The second PC identified outliers that were removed as they were not readily matched in a case-control sense. For the final analysis of 723 cases and 2880 controls a logistic regression analysis correcting for MDS covariates 1 and 3–10 was performed. This QC process reduced the λ_GC from ~2.5 to 1.074. This residual inflation was genomic-control corrected before inclusion in the meta-analysis.

Swedish UC GWAS

Cases were enrolled at three sites in Sweden. All genotyping was performed on the Illumina Hap550 array and Illumina Quad-610 at the Genome Institute of Singapore. Genotype data from 640 previously QC'd controls, all free of inflammatory diseases, was combined with common controls from the same Epidemiological Investigation of Rheumatoid Arthritis genotyped on the Illumina Hap550 (N=460) and Quad-610 (N=378) BeadChips (total of 1,478 controls). Alleles were called in the Illumina BeadStudio software by reclustering with cases and controls included for each chip type. Sample call-rates exceeded 97.29%, and none were removed for this reason. Four samples with low heterozygosity were removed from further analysis. Fifteen samples (10 cases and 5 controls) were excluded on the basis of sex discrepancies between database records and X chromosome zygosity. 112 (106 cases and 6 controls) sample duplicates and first-degree relatives were identified by excess allele sharing as calculated in PLINK, and subsequently excluded. 138 population outliers (82 cases and 56 controls) were identified by a principal components analysis implemented in Eigenstrat.

SNPs with call-rates <95%; minor allele frequency < 0.005, and a Hardy-Weinberg equilibrium P < 10⁻⁷ in controls together with non-autosomal SNPs were removed. Finally, the common set of directly genotyped SNPs across all three chip types were used, resulting in a post-QC dataset of 948 cases and 1,408 controls for 297,031 SNPs. Trend tests of association were calculated in PLINK. λ_GC for the entire dataset was 1.04, uncorrected. Eigenstrat correction based on the top ten PC's (bringing λ_GC to 1.03), followed by genomic control correction was used to remove the small residual stratification.

Meta-analysis

As all studies utilized compatible Illumina platforms, we combined 266,047 (258,137 autosomal and 7,910 X chromosome) SNPs passing QC in all three studies. P-values from the population structure corrected analyses were converted to Z-scores, consistently oriented to the combined minor allele. These consistently oriented Z-scores were then combined, and the square of those scores evaluated as a chi-square. The Q-Q plot is shown in Supplementary Figure 2 – while overall there is only very modest inflation (λ_GC=1.036), there is a significant excess of true positive results in the tail of the distribution. There were 59, 126 and 511 SNPs from the 266,047 that exceed .00001, .0001 and .001 respectively (~20, 5 and 2-fold in excess of null expectation).

Replication cohorts

Two independent case-control cohorts were examined for the replication phase. An Italian study population composed of 1094 UC patients and 908 controls collected at the S. Giovanni Rotondo “CSS” (SGRC) Hospital in Italy and secondly a Dutch study population composed of 1090 UC patients of Caucasian ethnicity recruited through the Inflammatory Bowel Disease unit of the University Medical Center Groningen, Groningen; the Academic Medical Center, Amsterdam; the Leiden University Medical Center, Leiden and the Radboud University Medical Center, Nijmegen. Healthy controls (N=804) of self-declared European ancestry were drawn from volunteers at the University Medical Center, Utrecht. Patients were diagnosed according to accepted clinical, endoscopic, radiological and histological findings. All patients and controls gave informed consent and the study was approved by the ethics review committees of each participating hospital. All DNA samples and data in this study were handled anonymously.

Replication Genotyping and quality control

Selected SNPs were designed into multiplex assays, and genotyped using primer extension chemistry and mass spectrometric analysis (iPlex assay, Sequenom, San Diego, California, USA) on the Sequenom MassArray at the Laboratory for Genetics and Genomic Medicine of Inflammation (www.inflammgen.org) of the Universite de Montreal. In cases where SNPs did not design into multiplex assays or failed the quality control thresholds, a proxy was selected. Samples showing >10% missing data, as well as SNPs with >10% missing data or significantly out of Hardy-Weinberg equilibrium (P < 0.001), were excluded from the analyses. The overall genotyping call rate in the replication dataset following QC was >99%, and consisted of 993 UC cases and 826 controls (Italian), and 1016 UC cases and 754 controls (Dutch).

Association analysis of the replication phase

Association testing of single SNPs in the replication cohorts was performed by standard χ² test carried out on a 2 × 2 contingency table (PLINK). Combination of the results from the two replication cohorts, as well as combination of the results from the screen and replication, was achieved by the calculation of a combined weighted z -score. The threshold for significant independent replication was set at P-values of < 0.05 in the combined Italian and Dutch datasets.

Antibodies

Experiments were performed using the following antibodies: anti-RNF186 (Abnova), anti-CEP72 (Novus Biological NB100-60661), anti-Troma-1 (developmental Studies Hybridoma Bank, Iowa City) and anti-PDI (Stressgen).

Reverse transcription and real time polymerase chain reaction

For expression maps, immune and intestinal cDNA panels were obtained from Clontech (two independent panels of cDNA). For ORMDL3 functional studies, RNA extraction was performed using RNeasy kit (Qiagen) according to the manufacturer's instructions. One μg of total RNA was reverse-transcribed using an iScriptTM cDNA synthesis kit (Bio-Rad). The displayed gene expression is representative of 3 independent experiments. Real time quantitative PCR was performed in duplicate in a Bio-Rad iCycler thermal cycler equipped of an iQ5 optical module using the iQTMSYBR®Green super mix (Bio-Rad). Briefly, 100 ng of reverse transcribed cDNA were used for each PCR with 250 nM forward and reverse primers. The thermal cycling conditions were 4 min at 95°C, followed by 40 cycles at 94°C for 15 s, 59°C for 1 minute. Values were normalized to GAPDH and the condition containing the lowest mRNA content was defined as 1 arbitrary unit. All the PCR products were analyzed with a 2% agarose gel to check that the size of the amplicons were as predicted.

Patient intestinal biopsies

Following informed consent and the approval of the Human Research Committee of the Massachusetts General Hospital, colonic biopsies from non-inflamed regions were obtained from IBD patients undergoing colonoscopy. Tissues were embedded and frozen in OCT compound (Fisher).

Immunostaining and fluorescence microscopy

Cells or frozen tissue sections were fixed using 4% paraformaldehyde and permeabilized PBS-Triton X-100 0.1%. After washing with PBS, the sections were incubated 30 minutes in PBS containing 3M glycine to block the reactive groups of paraformaldehyde. The sections were incubated (1 hour) with a blocking solution containing 10% of donkey serum (Rockland Immunochemicals) and 10% of Human Fc block reagent (Miltenyi Biotec). The preparations were then incubated with the primary antibodies (1 hour), washed using PBS, incubated with fluorescent secondary antibodies (Jackson Immunoresearch)(1 hour), washed using PBS and incubated with PBS containing 100μg/ml of DABCO (Sigma) as antifading reagent before mounting in Glycergel medium (Dako). Fluorescence signals were captured using a laser confocal microscope (model Radiance 2000 Bio-Rad). Image acquisition was performed with LaserSharpScanning software (Bio-Rad).

Plasmids

ORMDL3 constructs for mammalian expression: a clone of ORMDL3 in pCMV-SPORT6 was obtained from Open Biosystems. ORMDL3 coding sequence was amplified by PCR using forward and reverse primers (supplementary table 7) that contain the enzyme restriction sites EcoRI and NotI, respectively. Using EcoRI and NotI digestion, the ORMDL3 coding sequence was sub-cloned into C-terminal tagged pCMV-3xHA vector derived from the Clontech pCMV-Myc vector (Cat. # 631604) and into a pcDNA4/TO-GFP-C vector derived from Invitrogen's pcDNATM4/TO vector by inserting EGFP sequence from XhoI and ApaI sites in pcDNATM4/TO vector.

ShRNA constructs. ShRNA hairpins directed against the human transcript of ORMDL3 were designed using the tools of the RNAi consortium from the Broad Institute (http://www.broadinstitute.org/rnai/trc/lib). The pairs of oligonucleotides purchased to construct the hairpins are listed in supplementary table 7.

After annealing and according to the RNAi consortium instructions, the paired oligonculeotides were inserted using AgeI EcoRI enzyme restriction sites in TRC22 vector derived from pLKO1.

ER-Stress Reporter

HEK293T cells were plated on 24-well plates at a density of 2 × 10⁵ cells per well. After 24 hours, cells of each well were transfected with 2 ng of the ER-stress firefly luciferase reporters p5xUPRE-GL3²⁶ and 0.025ng of renilla luciferase (Promega) using TransFectin Lipid Reagent (Bio Rad). Luciferase activities were measured using the Dual-Luciferase Reporter Assay System (Promega) and were normalized to the internal transfection control of renilla luciferase activity.