Inherited determinants of Crohn's disease and ulcerative colitis phenotypes: a genetic association study

1Crohn's disease and ulcerative colitis, the two major forms of inflammatory bowel disease, affect about one in 200 people in developed countries, with a rising incidence and prevalence in developing countries. Many patients with inflammatory bowel disease have a lifetime of debilitating physical symptoms (eg, urgent diarrhoea, rectal bleeding, vomiting, anorexia, and lethargy), which frequently lead to poor psychosocial wellbeing with wide ranging consequences for academic attainment, employment, relationships, and sexual health. Furthermore, the financial costs of inflammatory bowel disease are substantial and are estimated at more than US$2·2 billion per year in the USA alone.

2Inflammatory bowel disease is characterised by an exaggerated mucosal immune response to luminal gut contents in genetically susceptible individuals. In Crohn's disease, inflammation can occur in any part of the gastrointestinal tract, whereas ulcerative colitis is typically confined to the colon. The universally adopted Montreal classification distinguishes clinical subphenotypes in Crohn's disease by disease location and behaviour, and age of onset, and in ulcerative colitis by disease extent and age of onset.,,,,,, Molecular studies have suggested that ileal and colonic Crohn's disease are distinct entities because variants in NOD2 are associated with small bowel disease and HLA alleles with colonic disease.,,,,,,, However, current recommendations do not advocate the use of these established markers in making treatment decisions, nor for choosing patients for clinical trials.,,,,,,,,, The natural history and clinical course of inflammatory bowel disease is very heterogeneous: up to 20% of patients with ulcerative colitis need colectomy for medically refractory disease, and more than 50% of patients with Crohn's disease need surgery within 10 years of diagnosis; however, up to 50% of patients with ulcerative colitis and 30% with Crohn's disease will have a fairly indolent disease course without the need for immunosuppression or surgery.,

3Evidence before this studyWe searched PubMed for genotype–phenotype association studies in inflammatory bowel disease, published between Jan 1, 1996, and Oct 17, 2014, with the search terms “inflammatory bowel disease” AND “genotype phenotype” AND “genetic association” AND (“disease course” or “disease extent” or “location behaviour”). We found 31 studies with sample sizes of 66 to 2804 patients, and studying between one and 163 genetic variants. Most of these studies implicated NOD2 and HLA in subphenotypes of inflammatory bowel disease. Many studies, especially the early ones, studied only NOD2 and HLA, and none included genetic variants not previously implicated in risk for inflammatory bowel disease.Added value of this studyThis study is the largest genotype–phenotype study of inflammatory bowel disease by at least a factor of ten, and is among the largest studies of genetic determinants of clinical subphenotypes of any complex disease. We have refined the known associations (for instance, the effect of NOD2 on disease behaviour is entirely driven by its association with disease location) and discovered one new associated locus (3p21/MST1 with age at diagnosis). We have explored the genetic relations between subtypes of inflammatory bowel disease with genetic risk scores for the first time, and have shown that ileal and colonic Crohn's disease are at least as genetically distinct from each other as they are from ulcerative colitis.Implications of all the available evidenceEstablished genetic factors can only explain a small fraction of the variability in subphenotype of inflammatory bowel disease, but genetic risk scores that capture all this information could be used to identify misdiagnosed patients. Future translational and clinical research should move away from a binary classification of inflammatory bowel disease into ulcerative colitis and Crohn's disease, instead considering ileal and colonic Crohn's disease as separate disease entities.

4Evidence before this study

5We searched PubMed for genotype–phenotype association studies in inflammatory bowel disease, published between Jan 1, 1996, and Oct 17, 2014, with the search terms “inflammatory bowel disease” AND “genotype phenotype” AND “genetic association” AND (“disease course” or “disease extent” or “location behaviour”). We found 31 studies with sample sizes of 66 to 2804 patients, and studying between one and 163 genetic variants. Most of these studies implicated NOD2 and HLA in subphenotypes of inflammatory bowel disease. Many studies, especially the early ones, studied only NOD2 and HLA, and none included genetic variants not previously implicated in risk for inflammatory bowel disease.

6Added value of this study

7This study is the largest genotype–phenotype study of inflammatory bowel disease by at least a factor of ten, and is among the largest studies of genetic determinants of clinical subphenotypes of any complex disease. We have refined the known associations (for instance, the effect of NOD2 on disease behaviour is entirely driven by its association with disease location) and discovered one new associated locus (3p21/MST1 with age at diagnosis). We have explored the genetic relations between subtypes of inflammatory bowel disease with genetic risk scores for the first time, and have shown that ileal and colonic Crohn's disease are at least as genetically distinct from each other as they are from ulcerative colitis.

8Implications of all the available evidence

9Established genetic factors can only explain a small fraction of the variability in subphenotype of inflammatory bowel disease, but genetic risk scores that capture all this information could be used to identify misdiagnosed patients. Future translational and clinical research should move away from a binary classification of inflammatory bowel disease into ulcerative colitis and Crohn's disease, instead considering ileal and colonic Crohn's disease as separate disease entities.

10Inflammatory bowel disease has been at the vanguard of progress in understanding the genetic framework of complex diseases, with 163 susceptibility loci identified so far. Most inflammatory bowel disease loci confer risk of both ulcerative colitis and Crohn's disease, but typically show distinct effect sizes in the two disorders. These findings suggest that genetic variation might define molecular subtypes independent of traditional and clinically defined diagnostic entities, allowing new insights into the molecular basis of these subphenotypes.

11Our international study of around 30 000 patients with inflammatory bowel disease genotyped by microarray is the largest genotype–subphenotype study in the disease done so far. We have used genetic risk scores to study genetic heterogeneity underpinning the natural history of inflammatory bowel disease. This analysis rejects the current binary classification of Crohn's disease and ulcerative colitis as distinct and homogeneous clinical entities in favour of a continuum of illness better fit by a three-category model (ie, ileal Crohn's disease, colonic Crohn's disease, and ulcerative colitis). We show that these risk scores have clinical potential (although they are currently only weak predictors), and believe they might have widespread applicability in other diseases.

12We acquired phenotype data for 34 819 patients, including 19 713 with Crohn's disease and 14 683 with ulcerative colitis. The cohort included all patients in different centres over the years who had inflammatory bowel disease as per Lennard-Jones' criteria. All inclusion criteria are included in appendix A. After quality control (appendix A), the primary analysis included 29 838 patients (16 902 with Crohn's disease, 12 597 with ulcerative colitis, 255 with indeterminate colitis, and 84 missing an exact diagnosis). This study includes patients from population-based registries, and secondary and tertiary-referral centres at 49 sites in 16 countries in Europe, North America, and Australasia, most of which have been previously described (appendix B). Confirmation of diagnosis of inflammatory bowel disease and assignment of clinical subphenotypes were done by clinicians specialising in inflammatory bowel disease or trained phenotypers through case note reviews of clinical, radiological, histopathological, and endoscopic reports, and classified per the Montreal classification criteria (see appendix A for details)., For behaviour and surgery in Crohn's disease, and colectomy in ulcerative colitis, Kaplan-Meier survival curves, stratified by location of Crohn's disease and extent of ulcerative colitis, were drawn to estimate time to first event (see appendix A for details).

13The ethical boards of each separate recruiting centre approved the study. All patients included in this study gave written informed consent.

14All cases were genotyped with the Immunochip array (Illumina, San Diego, CA, USA; appendix B) as previously described. Briefly, the Immunochip is a 195 806-polymorphism genotyping platform comprising variants identified from association studies of immune-related disorders including Crohn's disease and ulcerative colitis. Extensive quality control was performed on the dataset (appendix A), leaving 29 838 cases and 156 154 markers available for analyses. Variants in the MHC, including 23 HLA alleles that have been implicated in inflammatory bowel disease, were imputed as described in appendix A.

15All association tests were done on all genotyped variants, conditional on the first five principal components to account for population structure. Age of onset was analysed for Crohn's disease and ulcerative colitis separately and then meta-analysed; time to surgery was analysed with parametric survival-time regression models; and upper gastrointestinal involvement and perianal disease were analysed with binary logistic regression (see appendix A). For multicategory phenotypes (disease location, behaviour, and extent) we used model selection to pick the most appropriate genetic model for the phenotype (appendix A). The model selection indicated a multinomial model for Crohn's disease location (ie, three unordered categories), an ordinal logistic model for Crohn's disease behaviour (three ordered categories, B3 penetrating>B2 stricturing>B1 inflammatory), and a binary model for disease extent of ulcerative colitis (two categories: E3 extensive disease vs E2 left-sided disease and E1 proctitis). To distinguish direct associations from indirect (ie, driven by an association with a correlated phenotype), we also adjusted all regression models for the other phenotypes (age of onset, location, and behaviour for Crohn's disease; age of onset and disease extent for ulcerative colitis). Genome-wide significance (p<5 × 10−08) was required for individual single nucleotide polymorphisms (SNPs) and HLA types.

16All signals that showed suggestive association (p<1 × 10−5) with any of the disease subphenotypes were assessed in an independent cohort genotyped on a range of different genome-wide association study (GWAS) chips. These samples have also undergone rigorous quality control and imputation. Phenotype data for an additional and independent 2453 patients with Crohn's disease and 3729 patients with ulcerative colitis were available for these analyses. See appendix A for additional information about the replication cohort.

17To learn about the relative phenotypic variance explained by different risk factors in adult inflammatory bowel disease, we fitted a model to predict Crohn's disease location that included both demographic predictors (smoking status, age at diagnosis, and year) and genetic predictors (SNPs at NOD2, MST1, and the HLA cluster as well as the genetic risk score). Variance explained on the logit scale by each predictor was calculated with the McKelvey–Zavoina pseudo R2. Centres with a high proportion (>60%) of missing data for smoking status were removed. To reduce the effect of changes in clinical practice and smoking rates, only patients born between 1955 and 1985 were included.

18In addition to looking at single SNPs, we also combined information from 193 SNPs and 23 HLA types previously associated with inflammatory bowel disease to generate genetic risk scores (appendix A), which provide better predictive accuracy than individual SNPs. To assess classification accuracy, we re-ran the risk score analyses with a cross-validation strategy, in which models were fitted in non-UK origin samples and assessed by how well they classified UK samples.

19To assess if the risk score can be used to identify misclassified patients, we selected 97 outlier patients that fell in the extreme tail of the scores for the opposite phenotype (log Crohn's disease versus ulcerative colitis [CD vs UC] score ≤–2 for Crohn's disease outliers and log CD vs UC score ≥2 for ulcerative colitis outliers), as well as 95 randomly selected cases with non-outlier scores matched by recruitment centre. Clinicians from each centre were then asked to re-phenotype both outlier and non-outlier patients in a masked fashion. The CD versus UC risk score was chosen for this experiment because it had the strongest association with Crohn's disease location and behaviour.

20The median effect size of known inflammatory bowel disease risk variants was about OR 1·1, with a median minor allele frequency of roughly 30%. The sample size of our study gave us high power to detect an effect of equivalent magnitude of Crohn's disease location (power of 67% for ileal vs non-ileal disease), Crohn's disease behaviour (94% for complicated vs non-complicated disease) and ulcerative colitis disease extent (84% for extensive vs non-extensive disease) at genome-wide significance. Binary and linear genotype–phenotype analyses were done with PLINK version 1.07, and multinomial and ordinal regression with a custom program, Trinculo version 0.4 (appendix A). Survival analysis and risk prediction were done with R-2.15.1 using the packages “survival” and “Mangrove”, respectively. Data handling and plotting was done with R.

21The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.

22Our primary analyses were done on matched genotype and phenotype data from 29 838 patients of European ancestry (appendix A) with inflammatory bowel disease (16 902 with Crohn's disease, 12 597 with ulcerative colitis; table 1) with a total of 217 195 patient-years of follow-up (median per patient: 11 years for Crohn's disease and 10 years for ulcerative colitis). Demographic features of the study population agreed with previously published results: patients with Crohn's disease were more likely to be younger at diagnosis, female, smokers, and have affected family members than were patients with ulcerative colitis (table 1). Extensive disease was more common in those diagnosed at a younger age in both Crohn's disease and ulcerative colitis, whereas disease behaviour was relatively unaffected by age at diagnosis (appendix A). Reaffirming the progressive nature of Crohn's disease, the proportion of patients with stricturing (B2) or penetrating (B3) disease increased from less than 30% (n/N) at diagnosis to 43% (n/N) at 5 years, 56% (n/N) at 10 years, and 74% (n/N) at 30 years (figure 1, which shows the progression in B1, B2, and B3 disease individually [smoothed estimates over intervals]). By contrast, disease location showed little variation during the same period (figure 1). With the exception of the population-based cohorts from Scandinavia, survival analyses of time to development of complicated disease (B2, B3) or first surgery in Crohn's disease were highly consistent across the different countries of origin despite different health-care systems and methods of sampling (appendix A). In Crohn's disease, time from diagnosis to progression (complicated disease or surgical intervention) was significantly shorter in purely ileal (L1) compared with ileocolonic (L3) or colonic (L2) disease (p<10−100; figure 1; appendix A). Overall, 7257 (52%) of 13 862 patients with Crohn's disease had undergone surgery by the time of last follow-up. In ulcerative colitis, in which the overall rate of colectomy was 22% 10 years after diagnosis, time to surgery was shorter in patients with extensive disease (E3) than in those with left-sided disease (E2) or proctitis (E1; p=8 × 10−84; figure 1).

23We tested genetic variants for association with age at diagnosis and time to surgery in all patients with inflammatory bowel disease; disease location and behaviour in Crohn's disease; and disease extent in ulcerative colitis (table 2 and table 3). Across all analyses, three loci achieved genome-wide significance (p<5 × 10−8): 3p21 (MST1), NOD2, and the MHC. No additional signals were noted after replication of suggestive loci (p<1 × 10−5) in an independent GWAS cohort (appendix B). Although NOD2 was strongly associated with Crohn's disease location, behaviour, and age at diagnosis, adjustment for the other phenotypes showed that the association of NOD2 with behaviour was driven almost entirely by its phenotypic correlation with location and age at diagnosis (figure 2).

24We noted complex and correlated HLA signals for susceptibility to inflammatory bowel disease overall and age at onset, as well as Crohn's disease location and behaviour, extent of ulcerative colitis, and surgery (figure 2; appendix A; appendix B). In-depth analysis of the MHC region including classical HLA alleles showed that the strongest signal for disease location was a colonic association with HLA-DRB1*01:03 (p=1·47 × 10−23; ileal vs colonic odds ratio [OR] 0·32, 95% CI 0·29–0·41; ileocolonic vs colonic OR 0·47, 0·39–0·57), which is also the strongest shared risk allele for Crohn's disease and ulcerative colitis, followed by HLA-DRB1*07:01 (figure 2; appendix A). rs77005575 was independently associated with Crohn's disease behaviour (p=1·56 × 10−09; figure 2; appendix A). Notably, alleles associated with susceptibility to ulcerative colitis were better predictors of colonic disease location in Crohn's disease than alleles associated with susceptibility to Crohn's disease (appendix A). The top signal for extent of ulcerative colitis was rs3115674 (p=5·11 × 10−17 OR 0·70, 0·64–0·76; appendix A), which correlates with HLA-B*08 (R2=0·66), found mostly on the ancestral 8.1 HLA haplotype. HLA-DRB1*13:01 was the top signal for age at diagnosis of ulcerative colitis (3·50 × 10−09; figure 2; appendix A).

25On the basis of sample size, our primary analysis had similar power to detect associations to disease location (ileal vs colonic) as the first International Inflammatory Bowel Disease Genetics Consortium (IIBDGC) meta-analysis on Crohn's disease. However, with the exception of NOD2, MHC, and MST1, we do not report significant associations between subphenotypes and individual SNPs, including those robustly associated with disease susceptibility. We noted, however, that many known risk loci showed nominal evidence for association to a range of subphenotypes, so we posited that genetic risk scores representing the combined effect of many individually weak signals might be a more powerful approach to study the genetic underpinnings of subphenotypes for inflammatory bowel disease. We calculated different inflammatory bowel disease risk scores constructed from all available data (strength and direction of association) on the lead SNPs from each of the 163 known inflammatory bowel disease susceptibility loci. Although all of the risk scores were associated with Crohn's disease and ulcerative colitis subphenotypes, the most powerful score used the differences between Crohn's disease and ulcerative colitis (CD vs UC score; figure 2). Importantly, this CD versus UC score retained significance even after NOD2, MHC, and MST1 were removed (appendix A), lending support to the notion that the genetic risk score offers more information about the genetic substructure of inflammatory bowel disease than individual SNP associations alone. The strongest correlations in our study were between the CD versus UC risk score and Crohn's disease location and behaviour (figure 2; p=1·65 × 10−78, or p=9·23 × 10−18 after genome-wide significant loci were removed). Risk scores that incorporated imputed HLA types that have been implicated in risk for inflammatory bowel disease significantly improved the genetic risk scores compared with those using SNPs only (appendix A).

26Having shown the genetic risk score to be a useful measurement of inflammatory bowel disease subphenotype, we used it to study the genetic relation between ileal Crohn's disease, colonic Crohn's disease, and ulcerative colitis. The CD versus UC risk score placed colonic Crohn's disease between ileal Crohn's disease and ulcerative colitis (figure 3). Several other risk scores supported this relation, and although partly driven by the highly location-specific NOD2 variants, a risk score with NOD2 removed showed a similar pattern (appendix A). Additionally, statistical model selection across SNPs and HLA types strongly favoured a model in which colonic Crohn's disease is intermediate between ileal Crohn's disease and ulcerative colitis over one that grouped both Crohn's disease subphenotypes as a single category (appendix A). To test whether this finding extends to other subtypes of inflammatory bowel disease, we applied the genetic risk score to two intermediate forms of the disease: ileocolonic Crohn's disease (L3), in which the disease affects both small and large bowel, and colonic inflammatory bowel disease unclassified, in which the clinical and histological appearances are indistinguishable between Crohn's disease and ulcerative colitis. The CD versus UC score placed ileocolonic Crohn's disease as intermediate between ileal (L1) and colonic (L2) Crohn's disease, and colonic inflammatory bowel disease unclassified between ulcerative colitis and colonic Crohn's disease (figure 3).

27Despite the statistical significance of the associations between genetic risk score and subphenotype, the small effect sizes translated into fairly low predictive accuracy when tested by cross-validation (appendix A). The risk score that was most significantly associated with location of Crohn's disease in the primary analysis (CD vs UC) gave an area under the receiver-operating characteristic (ROC) curve of only 0·60 (95% CI 0·57–0·63) for distinguishing between colonic (L2) and ileal (L1) Crohn's disease in cross-validation, and even a specifically constructed ileal versus colonic score achieved an area under the curve of only 0·63 (0·59–0·66; appendix A). Comparison of the clinical characteristics of patients with Crohn's disease that fell into the extreme tails of the genetic risk score (low: log[CD vs UC] ≤–2; high: log[CD vs UC] ≥3) showed significant differences in disease location and behaviour (appendix A), suggesting that although a genetic risk score might not be able to classify all patients, it could be informative at the extremes.

28Overall, we can only explain a little of the phenotypic variance in the adult population with either classical or genetic predictors. The combination of smoking and the strongest genetic predictors explains only 6·8% of the variance for disease location in Crohn's disease (table 4 and table 5), and 1·1% for disease extent in ulcerative colitis (table 5).

29To assess the possible clinical usefulness of the genetic risk score, we reassessed the cases of Crohn's disease with a low CD versus UC risk score (more ulcerative colitis-like), and cases of ulcerative colitis with a high CD versus UC risk score (more Crohn's disease-like; figure 4). Masked re-phenotyping of these cases raised doubts about the original diagnosis in 27% of the outlier cases compared with 8% of non-outlier cases (corrected for disease location, p=6·8 × 10−4; appendix A). This finding suggests that we can indeed use genetics to identify small numbers of misclassified patients.

30The successful identification of genetic variants associated with complex diseases such as inflammatory bowel disease has raised the exciting possibility of a more personalised approach to clinical management. In inflammatory bowel disease, this quest is particularly urgent because of the substantial heterogeneity in disease course, and individual response to therapy. Past studies in inflammatory bowel disease have established a genetic component of disease subphenotype,,,,, but these studies have been limited to a handful of candidate regions in modest numbers of patients. Our study, involving the universal application of standardised phenotyping by trained personnel on nearly 30 000 patients with inflammatory bowel disease from 49 centres worldwide, combined with matching genotypes from more than 150 000 variants, represents the definitive investigation to date into the genetic basis of subphenotypes of inflammatory bowel disease.

31The only genome-wide significant associations we noted were between age of onset and disease location with variants at NOD2, MHC, and 3p21. Although the associations between ileal Crohn's disease and NOD2, and those between colonic Crohn's disease and the MHC, have been previously described,,,,,, our study has dissected the phenotype–location associations for the first time. Importantly, in Crohn's disease, we showed that NOD2 is not associated with stricturing disease after accounting for disease location. These findings, and the rarity of long-term change in disease location compared with behaviour, suggest that location is a fundamental biological aspect of a patient's disease, whereas behaviour (like surgery or treatment history) is a marker of disease progression.

32Our results for ulcerative colitis accord with the previously reported independent associations between the MHC and both extensive disease and colectomy., Notably, the strongest associations with extensive ulcerative colitis are variants on the ancestral 8.1 HLA haplotype. This haplotype is a known recessive risk for primary sclerosing cholangitis,, a disease often associated with an extensive but quiescent form of ulcerative colitis. Although variants associated with susceptibility to Crohn's disease and those associated with susceptibility to ulcerative colitis are both predictive of disease location, the ulcerative colitis-associated variants are the most predictive. The variants associated with disease susceptibility are also slightly predictive of age at diagnosis.

33Composite genetic risk scores from all 163 independent susceptibility signals were strongly associated with all our main subphenotypes, and these findings remained significant after excluding NOD2 and MHC. This result accords with a similar finding for genetic risk scores in bipolar disorder, and hints at the possibility that such approaches might be broadly applicable for studying clinical heterogeneity of common diseases. This finding suggests that many or most risk variants for inflammatory bowel disease do contribute weakly to subphenotype. The relative dearth of individual single-nucleotide polymorphism associations with subphenotypes in our study, by contrast with those reported in similarly powered studies of inflammatory bowel disease susceptibility, suggests that the genetic variants studied here have a small effect, and that environmental factors (such as diet, microbiota, and smoking) might be strong contributors to the subphenotypes. However, an intriguing possibility, supported by the notable absence of any functional or pathway enrichment in the components of the genetic risk scores, is that current phenotypic classifications do not correspond strongly to underlying molecular entities. Of particular note is the genetic distinction seen between ileal Crohn's disease, colonic Crohn's disease, and ulcerative colitis. These disease types were identified as equally distinct entities on a genetic continuum: on multiple risk scores, colonic Crohn's disease was genetically intermediate between ileal Crohn's disease and ulcerative colitis, a finding that remained significant after excluding NOD2. This result substantiates the view that colonic versus ileal disease, rather than disease extent, is the primary clinical unit of Crohn's disease classification, and is further supported by the finding that both the genetic risk score and clinical complication rate in patients with ileocolonic disease is intermediate between that of patients with ileal disease and colonic disease. It will be of great interest, and potential clinical use, to see if application of these risk scores helps to classify the 10% of patients who are currently designated colonic inflammatory bowel disease unclassified.

34The composite scores were also able to identify small numbers of patients with outlier scores who were much more likely to be misdiagnosed than a typical patient. These findings support the possible clinical usefulness of composite scores of multiple genetic variants, each of small effect. For example, genetic outliers could be excluded or specifically targeted for clinical trials to select more homogeneous groups. If these data were readily available, they might affect clinical decisions or inform risk–benefit discussions with patients. For example, the type of surgery offered to patients with refractory colitis crucially depends on whether they have ulcerative colitis or colonic Crohn's disease; poor outcomes (including pelvic sepsis, incontinence, and sexual dysfunction) of ileal pouch–anal anastomosis reconstruction surgery are much more common in patients with colonic Crohn's disease. Use of genetic risk scores to identify possible misdiagnoses in this group of patients could help to reduce this problem. Our data also suggest that genetic risk scores could augment biomarkers, such as faecal calprotectin, currently used for patient stratification in inflammatory bowel disease.

35A limitation of our study is that the genetic variants tested were restricted to those present (and that passed quality control) on the Immunochip platform, designed for replication and fine mapping of potential immune-mediated disease loci. Therefore, there still might be important loci that determine disease behaviour, location, and age at onset but are independent of those that confer risk for inflammatory bowel disease (or other immune-mediated diseases), and had limited or absent coverage on the Immunochip. A high-powered GWAS, designed to assess possibly overlooked genetic determinants for these outcomes of phenotype expressivity, which uses our large collection of cooperatively phenotyped cases of inflammatory bowel disease with genomic DNA (and the pre-existing Immunochip genotypes), will likely be of great value.

36In summary, our research represents the largest genotype–phenotype study in inflammatory bowel disease done so far. Associations achieving genome-wide significance were identified at only three loci, suggesting that new clinical phenotypic classifications might need to be explored for inflammatory bowel disease, and the relation examined between subphenotypes and other omic profiles and environmental factors, including the microbiota. However, our data suggest that on the basis of genetic factors, inflammatory bowel disease is better classified into three distinct groups (ileal Crohn's disease, colonic Crohn's disease, and ulcerative colitis), and we would recommend that clinicians adopt this nomenclature in regular practice. We also show that, although genetic risk scores do not yet have widespread clinical use, they are already valuable in some contexts in inflammatory bowel disease, and their study in other complex disease phenotypes is warranted.