Genetic Characterization and Influence on Inflammatory Bowel Disease Expression in a Diverse Hispanic South Florida Cohort

Abstract

Objectives:

Hispanics represent an understudied inflammatory bowel disease (IBD) population. Prior studies examining genetic predisposition to IBD in Hispanics are limited. In this study, we examined whether European-derived IBD variants confer risk in Hispanics and their influence on IBD phenotype in Hispanics compared to non-Hispanic whites (NHW).

Methods:

Self-identified Hispanics and NHWs with IBD were included. Hispanic controls were included for our genetic analyses. We performed single-variant testing at previously identified Crohn's disease (CD) and ulcerative colitis (UC) IBD variants in Hispanic cases and controls. These risk variants were used to compute individual genetic risk scores. Genetic risk scores and phenotype associations were compared between Hispanic and NHW.

Results:

A total of 1,115 participants were included: 698 controls and 417 IBD patients (230 Hispanics). We found evidence of association within our Hispanic cohort at 22 IBD risk loci, with ~76% of the risk loci demonstrating over-representation of the European risk allele; these included loci corresponding to IL23R and NOD2 genes. CD genetic risk score for Hispanics (199.67) was similar to the score for NHW (200.33), P=0.51; the same was true in UC. Genetic risk scores did not predict IBD phenotype or complications in Hispanics or NHW except for a younger age of CD onset in Hispanics (P=0.04).

Conclusions:

This study highlights the fundamental importance of these loci in IBD pathogenesis including in our diverse Hispanic population. Future studies looking at non-genetic mechanisms of disease are needed to explain differences in age of presentation and phenotype between Hispanics and NHW.

Introduction

Despite the rising Hispanic population with inflammatory bowel disease (IBD), only a few studies have looked at disease characteristics in this group.^{1, 2, 3, 4, 5} In our prior study, we found that differences in age of presentation and IBD phenotype exist within Hispanics (US vs. foreign-born) and between Hispanics and non-Hispanic whites (NHW).³ The exact mechanism explaining differences within Hispanics remains unexplored. Studies examining immigrants arriving into IBD-prevalent countries suggest that not all migrants have the same predisposing risk to IBD onset.⁶ This alludes to genetic influence on IBD expression in spite of a similar “IBD-promoting” environment.

The first large-scale trans-ancestry association study of IBD examining East Asian, Indian and Iranian populations found genetic heterogeneity among populations at European-established risk loci (including NOD2 and IL23R);⁷ highlighting the need to study diverse populations to understand differing mechanisms of IBD pathogenesis. Thus far, genetic studies of IBD in Hispanics lag behind GWAS studies investigating North Americas, European Caucasians, Asians and middle Easterners.^{7, 8, 9, 10, 11, 12} In a recent study examining Puerto Rican IBD patients, risk allele frequencies for NOD2 and IL23R were higher in the IBD Puerto Rican population compared to Puerto Rican controls.¹³ This study suggests that at least a handful of established European-derived IBD risk loci may also confer risk to IBD in Hispanics. In addition, given the relatively superficial nature of genome-wide association studies in examining genotype–phenotype associations, studies using a composite genetic risk score across various autoimmune diseases, including IBD, have shown stronger correlations with IBD phenotype than individual single-nucleotide polymorphism (SNPs).^{14, 15, 16, 17, 18}

In this study, we have generated specific genotyping data to investigate the association of previously identified IBD risk variants with IBD in our Hispanic sample. We examined genetic risk score in the context of age of presentation and IBD phenotype, including extent of disease and surgeries. We proposed that differences between Hispanics and non-Hispanics exist in the frequency of established IBD risk alleles and in their global prediction (as a composite genetic risk score) of disease phenotype. Our study is the first to explore the genetic profile of a diverse sample of Hispanics with IBD and to incorporate the most recently described additional IBD risk alleles into a composite genetic risk score.

Methods

Study design

Data were collected with the approval of the Institutional Review Boards at the University of Miami Miller School of Medicine and Jackson Memorial Hospital (Miami, FL, USA). Information was collected prospectively from adult Gastroenterology Clinics. Patients provided detailed demographic and medical information. Foreign-born patients were asked year of immigration US immigration. Blood samples were collected and stored within the Biorepository facility at the John P. Hussman Institute for Human Genomics (HIHG) for genetic analyses. Self-identified white Hispanics and NHWs were included. IBD patient samples were genotyped on the ImmunoChip array (a custom designed Illumina iSelect panel).^{19, 20} Control samples were self-identified Hispanics who were participating in a large genetic study of cardiovascular disease at our institution. General health questions are asked of controls in order to identify and exclude any moderate to severe cases of ulcerative colitis (UC) or Crohn's disease (CD) from this sample set. These unrelated controls were also genotyped within the HIHG using a custom Illumina genotyping array.

Phenotypic outcomes

IBD diagnosis and phenotype classification was confirmed by review of source documents and/or the clinical, endoscopic, radiologic, and pathologic findings available to the attending physician in clinic. Phenotyping for Crohn's disease and ulcerative colitis was performed according to the Montreal classification. We recorded information on surgical outcomes including abdominal surgeries (small bowel intestinal resection and subtotal or total proctocolectomies) and perianal surgeries (seton placement, mucosal advancement flaps, etc). Abdominal surgeries for CD were analyzed as a sum of outcomes and separately from perianal CD surgeries. UC surgeries (total proctocolectomies) were analyzed separately from surgeries done for CD. A history of IBD-related hospitalizations occurring from symptom onset to the time of the questionnaire were recorded.

Genotyping and genetic risk scores

There are >200 IBD susceptibility variants described in a largely European-derived ancestry population.^{7, 8} The ImmunoChip array produced raw genotype data on >192,000 SNPs, including 232 previously published IBD risk variants. These raw data were loaded into the Genome Studio software suite for genotype calling, using the PLINK²¹ software package and R (Version 3.0.2)²² for further statistical analysis. Standard genotyping quality control checks were performed to identify genotyping or sample errors. SNPs were examined for genotyping efficiency (SNPs dropped if call rate <95%) and deviations from Hardy Weinberg equilibrium (SNPs dropped if P<0.001). Controls were genotyped using the Affymetrix SNP array 6.0, with raw genotype data having undergone similar standard quality control.²³ In controls 80/232 SNPs were genotyped, with an additional 139 SNPs being successfully imputed. Imputation to 1000 Genomes was performed using IMPUTE2 (ref. 24) as described previously.²³ All the imputed SNPs included in the analyses were of good quality (INFO>0.45). The genotyping arrays utilized to produce the genetic data for both the cases and controls are designed and run using the same chemistry and the same platform (i.e., the Illumina Infinium platform). For the analyses presented in the manuscript, we simply extracted the available genetic data for the variants (or SNPs) demonstrating previous associations. Following these standard measures of quality control, we were left with 219 of 232 SNPs with high quality genotyping and/or imputation data available for both our single variant and genetic risk score analyses.

We calculated principal components (PCs) using Eigenstrat²⁵ to both remove significant population sample outliers and to use as covariates in logistic regression. To calculate the PCs, we utilized 7,900 SNPs pruned for linkage disequilibrium (pairwise r²<0.1) from 22,635 total SNPs found in common between the genotyped cases and controls (no imputed SNPs were used to calculate PCs), and excluding the known associated IBD SNPs and those across the HLA region. Eight control samples were excluded as they were more than six s.d.'s from the mean on any of the first 10 PCs. In Figure 1a we plot the first two PCs for our Hispanic cases and controls and illustrate the population substructure of our Hispanic and non-Hispanic samples relative to those of the HapMap project (Figure 1b). These figures show that Hispanic cases and controls are genetically similar.

Figure 1.

These PC analysis (PCA) plots illustrate the clustering of samples according to their genetic background, utilizing genotype data from 7,900 linkage disequilibrium pruned single-nucleotide polymorphism (SNPs) overlapping our Hispanic cases/controls and our non-Hispanic cases. (a) Hispanic cases and controls. Using PC1 and PC2, we highlight that our Hispanic cases and controls share a common global ancestry structure. (b) Ancestral background distribution of our data set relative to the HapMap populations. In addition to samples in our data set (“Hispanics” and “non-Hispanics”), data from the International HapMap Project populations (CEU, YRI, JPT, and CHB) are included for illustration purposes. Asian (JPT and CHB) and African (YRI) population samples cluster in distinct groups, with the vast majority of our non-Hispanic whites samples clustering with the CEU population. As anticipated, our Hispanic samples demonstrate evidence of admixture as they spread from the cluster representing European ancestry.

We calculated separate genetic risk scores for both CD and UC within our cases. These risk scores are comprised of the weighted sum of the 219 SNPs demonstrating association with either CD or UC.⁷ Scores were calculated by multiplying the number of risk alleles for each individual SNP by the natural log of the reported odds ratio for the risk allele, and then summing across all SNPs (i.e., all SNPs associated with either IBD or CD were used to compute a genetic risk score for CD cases and all SNPs associated with either IBD or UC were used to compute a score for UC cases).^{7, 14} This genetic risk score was calculated for each affected individual based on his or her specific diagnosis and used as a proxy for genetic risk burden in each individual.

Statistical analysis

Demographics, IBD phenotype and mean genetic risk scores were compared between Hispanics and NHWs using χ² and analysis of variance. A case/control dosage-based test of association for 219 previously reported IBD SNPs^{7, 8} was performed in PLINK²¹ using a logistic model and adjusting for the first three PCs. For this analysis, the genotype data for the 219 SNPs genotyped in cases and the 80 SNPs genotyped in controls was converted to dosage using GTOOL (Version 0.7.5) and merged with the imputed dosage data for the 139 SNPs that were imputed in controls. To reduce type one error and to account for multiple testing, permutation testing was performed using the max(T) permutation procedure implemented in PLINK,²¹ specifying 10,000 permutations. Both uncorrected and permuted P values for these single variant analyses are provided in Table 2 (Supplementary Table 1).

Genotype–phenotype associations were performed using R (Version 3.0.2).²² Genetic risk score, corresponding to an individual's specific diagnosis (either CD or UC), was analyzed as a continuous variable and incorporated into general linear models. We performed multivariable regression analysis of genetic risk scores with multiple outcomes (e.g., number of surgeries, etc) and adjusted for population stratification using the first three PCs and utilized the max(T) permutation procedure (permuting the phenotypes 10,000 times across the samples) to report permuted non-parametric P values (Table 3). The assumptions of linearity, independence of errors, homoscedasticity, unusual points and normality of residuals were met in our multivariable regression model.

Results

Population demographics

Our research study included 1,115 participants from the culturally and ethnic diverse population present in South Florida: 698 Hispanic controls, 230 Hispanics with IBD and 187 non-Hispanics with IBD. Most Hispanics with IBD were foreign-born (58.8%) with Cubans representing the largest proportion of foreign-born Hispanics (72 patients, 53.3%). In all, foreign-born Hispanics originated from 16 different countries (Figure 2). This patient cohort is distinct from the one described in an earlier publication on the phenotype of Hispanic patients.³ US-born Hispanics were mostly first generation (83.3%). We found that our Hispanic cases and controls are ancestrally well matched and are related as expected with ancestral populations (Figure 1).

Figure 2.

This bar graph displays the distribution of countries represented in our sample of Hispanics. US-born Hispanics represent the majority followed by Cubans.

IBD patient characteristics

In our IBD sample of Hispanics and NHWs, all groups had mostly CD (~63% Table 1). There were 261 CD patients and 156 UC patients including both Hispanic and NHWs. US-born Hispanics were diagnosed younger than foreign-born Hispanics, similar to what we found in our prior cohort.³ All demographic characteristics can be seen in Table 1. Clinical characteristics including location of luminal disease, CD-behavior (e.g. fistulizing and perforating) and surgeries were similar between Hispanics and NHWs with UC and CD (data not shown).

Table 1. Patient demographics according to ethnicity.

	Hispanics with IBD N=230			Controls N=698	NHW with IBD N=187
	US born N=94	Foreign-born N=135	Combined N=229a			Hispanics vs. NHW P value	US vs. foreign-born Hispanic P value
Crohn's disease n (%)	58 (62)	84 (62)	142 (63)	NA	118.0 (63)	1.00	1.00
Female n (%)	49 (52)	63 (47)	112(49)	226 (32)	99.0 (53)	0.43	0.42
Age of diagnosis mean (years, s.d.)	22.7 (8)	36.6 (14)	30.9 (20)	NA	29.2 (16)	0.22	<1.00E-03
Age of symptom onset mean (years, s.d.)	22.0 (9)	36.2 (21)	30.3 (15)	NA	27.3 (18)	0.03	<1.00E-03
Disease duration from symptom onset mean (years, s.d.)	8.8 (10)	8.6 (10)	8.7 (10)	NA	13.5 (17)	<1.00E-03	0.86
Family history, n (%)b	10.0 (11)	20.0 (15)	30.0 (13)	–	39.0 (21)	0.02	0.04
Active smoking at diagnosisb	10.0 (11)	16.0 (15)	26.0 (13)	NA	32.0 (20)	0.11	0.5
CD, n (%)	8.0 (15)	12.0 (17)	20.0 (16)	–	24.0 (23)	0.22	0.53
UC, n (%)	2.0 (6)	4.0 (11)	6.0 (8)	–	8.0 (15)	0.43	0.39

IBD, inflammatory bowel disease; NHW, non-hispanic whites; s.d., standard deviation; UC, ulcerative colitis.

^aOne patient was missing country of birth.

^bData was incomplete for some UC patients

The bold highlights the significant P-values.

Associations with previously identified IBD risk loci in Hispanic patients

We tested 219 previously identified IBD susceptibility variants for association in our data set. We found nominally significant (P<0.05) evidence of association for 22 variants (6 IBD, 10 CD, and 6 UC; Table 2). Results for the remaining 197 variants tested can be found in Supplementary Table 1. After correcting for multiple testing, our most significant single-variant associations for CD and IBD were both found within the IL23R gene (encodes a subunit of the receptor for the pro-inflammatory cytokine interleukin-23) locus on chromosome one (rs7517847 and rs80174646 respectively), with our top UC association found on chromosome 16 nearest the KIR2DL1/2 locus. Our top two most significant variants are also the most significant variants seen by Liu et al. for all known associations in Europeans.⁷ A total of 166 out of 219 (76%) variants tested have over-representation of the European risk allele in IBD cases (Table 2; Supplementary Table 1). We did however also find statistically significant evidence for two SNPs (rs11739663 for UC and rs7438704 for CD) demonstrating effects in the opposite direction (i.e., protective for our cohort) compared to previous reports.^{7, 8}

Table 2. Association results of established IBD risk variants in Hispanics.

CHR	SNP	Position (bp)	Trait	Candidate genes	RA (EUR)	OR (EUR)	RAF (hisp cases)	RAF (hisp controls)	RAF (EUR)	RAF (AMR)	RAF (AFR)	OR	SE	P valuea
1	rs3806308	20,142,866	UC		G	1.19	0.70	0.61	0.63	0.56	0.71	1.57	0.179	9.50E-03
1	rs7517847	67,681,669	CD	IL23R,IL12RB2	A	1.40	0.67	0.55	0.58	0.43	0.84	1.66	0.140	<0.0001b
1	rs80174646	67,708,155	IBD	IL23R,IL12RB2	C	1.88	0.96	0.92	0.93	0.94	0.96	2.31	0.262	8.00E-04
1	rs2651244	70,995,562	UC		G	1.06	0.77	0.66	0.60	0.78	0.91	1.83	0.204	2.60E-03
1	rs4845604	151,801,680	UC	RORC	G	1.18	0.90	0.85	0.86	0.87	0.71	2.27	0.374	2.72E-02
2	rs925255	28,614,794	CD	FOSL2,BRE	G	1.11	0.66	0.60	0.54	0.72	0.75	1.37	0.143	2.77E-02
2	rs7608910	61,204,856	IBD	REL,C2orf74	G	1.13	0.45	0.37	0.36	0.27	0.38	1.35	0.112	7.60E-03
			KIAA1841,AHSA2
2	rs17229285	199,523,122	UC		G	1.10	0.62	0.54	0.50	0.58	0.91	1.41	0.171	4.63E-02
4	rs7438704	48,363,245	CD	TXK,TEC,SLC10A4	G	1.09	0.53	0.60	0.64	0.54	0.46	0.73	0.132	1.66E-02
4	rs13126505	102,865,304	CD		A	1.20	0.09	0.05	0.06	0.03	0.00	1.88	0.245	8.70E-03
5	rs11739663	594,083	UC	SLC9A3	A	1.08	0.70	0.79	0.77	0.82	0.52	0.63	0.177	6.80E-03
5	rs11742570	40,410,584	CD	PTGER4	G	1.28	0.65	0.57	0.59	0.44	0.61	1.37	0.137	2.25E-02
5	rs10051722	130,104,076	CD		A	1.09	0.73	0.66	0.68	0.55	0.64	1.34	0.150	4.74E-02
6	rs116392568	31,274,380	CD	HLA-C,PSORS1C1	G	1.16	0.45	0.35	0.36	0.30	0.27	1.48	0.138	3.90E-03
				NFKBIL1,MICB
8	rs13277237	130,604,563	IBD		G	1.06	0.51	0.44	0.43	0.44	0.56	1.33	0.111	9.60E-03
9	rs10781499	139,266,405	IBD	CARD9,PMPCA,	A	1.17	0.47	0.40	0.40	0.57	0.25	1.31	0.110	1.53E-02
			SDCCAG3,INPP5E
13	rs9525625	43,018,030	CD	AKAP1, TFSF11	A	1.08	0.61	0.53	0.46	0.54	0.71	1.47	0.141	6.50E-03
16	rs2066844	50,745,926	CD	NOD2,ADCY7	A	2.13	0.07	0.05	0.05	0.02	0.00	2.01	0.311	2.26E-02
18	rs7240004	46,395,022	IBD	SMAD7	A	1.07	0.63	0.58	0.63	0.58	0.42	1.33	0.128	2.46E-02
19	rs516246	49,206,172	CD	DBP,SPHK2	A	1.12	0.53	0.44	0.44	0.34	0.49	1.46	0.139	5.60E-03
				IZUMO1,FUT2
19	rs17771967	55,380,214	UC	NLRP7,NLRP2	G	1.07	0.55	0.43	0.44	0.41	0.37	1.73	0.178	1.40E-03
				KIR2DL1,LILRB4
20	rs6087990	31,349,908	IBD	DNMT3B	G	1.05	0.47	0.42	0.37	0.63	0.76	1.30	0.115	2.04E-02

AFR, African; AMR, Ad Mixed American; CD, Crohn's disease; CHR, chromosome; EUR, European; IBD, inflammatory bowel disease; RA, risk allele; RAF, risk allele frequency; SNP, single-nucleotide polymorphism; OR, odds ratio; UC, ulcerative colitis.

^aPermuted P value.

^bThis result was the most significant after 10,000 permutations.

The bold indicates those effects that are opposite in direction for our Hispanic cohort. Position is based on human genome 19 and dbSNP 137. Trait=the phenotype with the largest MANTRA Bayes factor from Liu et al.,⁷ Supplementary Table 1. Candidate gene(s) as previously defined (Jostins et al.;⁸ Liu et al.⁷) RA (Based on Liu et al.⁷) Risk allele frequencies (RAFs) for EUR, AMR, and AFR are phase 3 allele frequencies from 1000 Genomes (www.1000genomes.org).

Permuted P value was calculated using max (T) permutation as implemented in PLINK (using 10,000 permutations).

As one would expect, the reported 1000 Genomes super population allele frequencies for each of the variants we tested differs between the ancestral populations (AFR, AMR, and EUR) that comprise the diversely admixed Hispanic population.²⁶ Most interestingly, our two variants (rs11739663 and rs7438704) demonstrating protective effects have a much lower frequency for the risk allele within the African super population compared to either the European or admixed American population frequencies. These data suggest that our Hispanic cohort may have more African derived haplotypes across these loci that potentially confer protective effects. Within our control population, we find that the risk allele frequencies of all but two variants (rs9525625 and rs7240004 nearest SMAD7) track most closely to the reported European allele frequencies. Notably, the frequency of these two variants within controls most closely resemble the AMR super population and may be a reflection of local ancestry differences and admixture proportions found within the Hispanic population of largely Caribbean origin within South Florida.

Genetic risk score and onset of IBD in Hispanics and non-Hispanic whites

The mean genetic risk score for Hispanics with CD 199.67 (range 50.57; s.d. 9.06) was similar to the mean genetic risk score for NHWs with CD 200.33 (range 95.10; s.d. 10.29), P=0.51. The same was true for UC genetic risk scores (Hispanics 161.58 (range 39.228 s.d. 8.39) vs. non-Hispanics 162.23 (range 36.58 s.d. 7.95), P=0.50). We found no difference in mean genetic risk scores between US and foreign-born Hispanics (data not shown). In addition, because our cohort is mainly Cuban, we compared genetic risk score between Cubans and non-Cuban Hispanics and found that there was no difference in mean scores (data not shown). Multivariable regression analysis demonstrated that a higher genetic risk score was associated with a younger age of IBD symptom onset in Hispanic patients with CD (P=0.04) but not Hispanics with UC (P=0.76); see Table 3. Among NHWs, a higher CD genetic risk score did not predict a younger age of onset (P value=0.14). We did not see a correlation between genetic risk score and age of diagnosis in either ethnicity with UC (Table 3).

Table 3. Multivariable regression analysis of genetic risk scores with CD and UC outcomes.

Phenotypic outcome	Ethnicity	Genetic risk score beta coefficient	R-squared	Permuted P value
Crohn's disease
Age of diagnosis	Hispanics	−0.229	0.059	0.046
	NHW	−0.162	0.063	0.191
Age of symptom onset	Hispanics	−0.231	0.060	0.039
	NHW	−0.170	0.054	0.144
Location of disease
Ileal disease	Hispanics	−0.026	0.047	0.184
	NHW	0.031	0.035	0.126
Ileocolonic disease	Hispanics	−0.002	0.011	0.946
	NHW	−0.028	0.041	0.253
Colonic disease	Hispanics	0.015	0.023	0.443
	NHW	−0.007	0.054	0.737
Isolated upper GI tract	Hispanics	0.204	0.080	0.098
	NHW	−0.826	0.014	0.602
Crohn's behavior
Stricturing	Hispanics	0.023	0.022	0.233
	NHW	0.022	0.017	0.274
Penetrating	Hispanics	0.017	0.015	0.536
	NHW	−0.039	0.039	0.156
Perianal disease	Hispanics	0.029	0.033	0.158
	NHW	−0.004	0.029	0.841
History of hospitalizations	Hispanics	0.067	0.038	0.079
	NHW	−0.023	0.069	0.690
Abdominal surgeries	Hispanics	0.026	0.036	0.050
	NHW	0.011	0.013	0.317
Perianal surgeries	Hispanics	0.049	0.026	0.082
	NHW	−0.007	0.016	0.786
Ulcerative colitis
Age of diagnosis	Hispanics	0.068	0.013	0.720
	NHW	0.136	0.073	0.550
Age of symptom onset	Hispanics	0.059	0.013	0.760
	NHW	0.147	0.500	0.530
Location of disease
Proctitis	Hispanics	0.053	0.041	0.390
	NHW	−0.063	0.018	0.395
Left-sided colitis	Hispanics	0.010	0.045	0.716
	NHW	−0.037	0.119	0.327
Pan-colitis	Hispanics	−0.014	0.046	0.592
	NHW	0.051	0.132	0.175
History of hospitalizations	Hispanics	−0.077	0.020	0.270
	NHW	0.108	0.066	0.150
Colectomy	Hispanics	−0.018	0.060	0.720
	NHW	0.176	0.04	0.090

CD, Crohn's disease; IBD, inflammatory bowel disease; NHW, non-hispanic whites; UC, ulcerative colitis.

Beta coefficient: the beta value is a measure of how strongly each predictor variable (in this case genetic risk score) influences the dependent variable (in this case IBD phenotype).

R-squared: is a number that indicates the proportion of the variance in the dependent variable (IBD phenotype) that is predictable from the independent variable (genetic risk score).

Bold means significant p-values (P<0.05).

The mean genetic risk score for the control group was 185.24 (range 59.64; s.d. 9.01), using the CD risk alleles and 157.28 (range 51.88; s.d. 8.06), using the UC risk alleles. When comparing the genetic risk scores between the Hispanic IBD cases and controls, the risk scores were significantly higher in the CD and UC cases compared to controls (P<0.0001).

Genetic risk score and phenotypic manifestations of disease in Hispanics and non-Hispanic whites

Regression analyses using genetic risk score as an independent predictor of CD disease location and behavior was not significant (Table 3). No association existed between UC genetic risk score and extent of disease in UC was observed (Table 3). A higher genetic risk score was weakly associated with an increased risk of abdominal surgeries in Hispanics with CD (P=0.05), but was not significant in non-Hispanics (Table 3). No relationship was seen between colectomies for UC patients or predictive of number of IBD-related hospitalizations (Table 3).

Discussion

To our knowledge, this is the first study to examine genetic burden in Hispanics and its relationship to age of disease onset, diagnosis and detailed IBD phenotype. It is also the first study to examine the genetic profile of 200 established IBD risk variants in a diverse sample of Hispanics. Although the prevalence of IBD is lower in Latin America, we found that the composite genetic risk score of established European risk loci was similar between Hispanics (US and foreign-born) and NHWs with IBD. These findings do not explain differences in age of IBD presentation observed between US and foreign-born Hispanics, and why US-born Hispanics have a similar age of presentation to their NHW counterparts; we have replicated these phenotypic findings in this new cohort.³ Our findings highlight the critical factor played by the environment and substantiate that IBD onset occurring after migration may be strongly influenced by exposure to an environmental trigger that cannot be circumvented by genetic burden as measured herein.

An important aspect of our study was to determine whether the IBD risk loci identified in large genome-wide association studies of European populations also confer risk in the diverse Hispanic population. We observed nominally significant associations for 22 (6 IBD, 10 CD, and 6 UC) previously established risk variants in our Hispanic population data set. For 20/22 variants, the direction of effect was consistent with previously meta-analyses.⁷ Within the 22 risk loci, we found significant associations across SNPs corresponding to IL23R, NOD2, and Smad7 genes. A recent study in the African American population found that many IBD risk variants, including those in IL23R and NOD2, also demonstrate evidence for association in African Americans.²⁷ In addition, our findings parallel the results reported in the Puerto Rican IBD study, even though our sample is comprised of only 3.9% Puerto Ricans. In a recent study of Puerto Rican IBD patients, the haplotype structure for both the IL23R and NOD2 regions were found to be predominantly of European ancestry. We can extrapolate from their study that the same (European) haplotypic structure may be seen at these loci in our admixed Hispanic data set, composed largely of Cubans.

Admixture analysis of Cubans using ancestry informative markers illustrates that Cubans have largely a European contribution, estimated at 72%.²⁸ Therefore, these data suggest that the contribution of NOD2 and IL23R to the development of CD in our Hispanics is likely similar to that of European patients. The IL23R locus demonstrates a complex set of association signals with IBD, with both risk and protective effects having been previously shown potentially resulting from the effects of the genetic variants on differential splicing.²⁹ Moreover, our most significant variant (rs7517847) found in IL23R has not only previously demonstrated highly significant association with CD, but has also been found to be associated with plasma omega-6 polyunsaturated fatty acid levels (linoleic acid) in a recent genome-wide association study.³⁰ Although many studies suggest that a high dietary ratio of ω6/ω3 polyunsaturated fatty acids is potentially risk-conferring for CD, evidence across studies is equivocal but suggests that the modern western diet is highly likely to play a role in chronic human disease.^{31, 32, 33} While our present results are modest, taken together these findings may begin to help explain our observations of increased numbers of CD Hispanic patients in our clinics compared to the expected prevalence of CD within this admixed population.

We also find evidence of association near the Smad7 (SMAD family member 7) gene, which encodes a protein that is overexpressed in IBD patients and is responsible for suppressing anti-inflammatory signaling and is associated with colon cancer^{34, 35} (Table 2). Our study suggests that this locus is relevant to IBD risk in the Hispanic population. Another interesting finding is the association to a gene involved in DNA methylation. CD Hispanics were 1.3x more likely to carry the risk allele for rs6087990 corresponding to the DNMT3B (DNA methyltransferase 3B) gene, which encodes for a DNA methyltransferase responsible for de novo methylation. A recent systematic review of cancer found a decreased risk between rs6087990 and colon cancer in the Asian population.³⁶ Further studies are needed to better delineate the potential role this gene may play in colon cancer risk for Hispanic IBD patients and the downstream epigenetic effects altering gene expression.

In this study, we found that a high genetic risk score predicted a younger age of Crohn's disease onset in Hispanics. While we may not have replicated this finding in our non-Hispanic cohort,¹⁶ we were able to augment these findings with data on Hispanics. Predicting risk of complications in CD or UC remains imperfect. In this study, we observe that our composite genetic risk score does not correlate with IBD phenotypes or surgeries in either UC or CD. However, other studies differ from our findings.^{37, 38} The largest genotype–phenotype association study, using 193 SNPs and 23 HLA types to generate genetic risk scores, examined phenotypes across 34,819 patients of European ancestry and found that IBD “subphenotype” (i.e., ileal Crohn's, colonic Crohn's) may be determined in part by genetics.¹⁸ Our study findings may differ from the former study either as a result of a different population examined or because our sample size could not detect associations of this small effect size. Although our findings were not significant, the possibility of a genetic risk score as a prediction tool for disease phenotype remains enticing, particularly in phenotypes that warrant further classification, such as IBD- unclassified. The genetic risk score as a measure of overall genetic burden is a contemporary model used in cardiovascular and autoimmune diseases, including IBD that has shown stronger correlations with phenotype than individual SNPs.^{14, 16, 17, 39}

One of the limitations of our study is the size of our data set and for this reason we chose to focus on the known genetic associations to ask a very specific question regarding where or not previously identified variants confer risk in Hispanic IBD. The smaller sample size in NHWs, may explain the negative associations noted between genetic risk score and age of onset and phenotype, including ileal disease–observed in prior recent studies.^{16, 18} However, even with a small sample size and using sound statistical methodology, including permutation testing, we were able to replicate previous findings for several European risk alleles in Hispanic cases and found that a composite weighted score of these established European risk alleles is still applicable to Hispanics with IBD. Last, since our focus was on Hispanics, we did not include a control population of NHWs. Genetic association studies comparing NHWs with IBD to NHW controls already exist and are larger than the cohort of non-Hispanics in our database, thereby abating the need for us to replicate these findings.^{8, 9, 10, 11}

In summary, our results provide biological insight into the pathogenesis of IBD in Hispanics and a better understanding of the genetic architecture across this diverse population. Our findings highlight the potential applicability of a genetic risk score in the prediction of age of onset in the Hispanic IBD population. When immigrating, time in the US is necessary to manifest IBD even in the face of a similar genetic burden. These findings provide the framework for future studies to focus on gene-environment interactions and unidentified environmental exposures that may explain the rising incidence of IBD in US Hispanic populations.

Study Highlights