Late-Onset Crohn’s Disease Is A Subgroup Distinct in Genetic and Behavioral Risk Factors With UC-Like Characteristics Free

Abstract

Background

Age of onset is linked to variations in clinical phenotypes and natural history in Crohn’s disease (CD). We aim to define etiologically more homogenous subgroups in CD based on ages of onset.

Methods

We examined the distribution of CD polygenetic risk score (PRS) across ages of diagnosis in a Caucasian cohort of 2344 independent CD patients. We identified subgroups with a distinct distribution of PRS and compared those groups in genetics, demographic characteristics, clinical subphenotypes, and serological markers. The results were replicated in an independent cohort of 13,065 CD patients from the International Inflammatory Bowel Diseases Genetic Consortium (IIBDGC).

Results

We identified a late-onset (LO) subgroup in CD (age at diagnosis ≥ 55 years) with significantly lower PRS compared with the intermediate group (age at diagnosis between 5 and 55 years) in both cohorts. Smoking cessation, a risk factor for ulcerative colitis (UC) and protective factor for CD, had a higher rate in this LO subgroup in comparison with the intermediate group. We also compared the LO group with the intermediate group, and, consistent with previous reports, the LO group more often had colonic CD, had less penetrating disease behavior, and had less need for surgery. Serological analysis showed that LO CD patients were more antineutrophil cytoplasmic antibody positive and less antisaccharomyces cerevisiae antibody positive compared with the intermediate group. Variance component analysis indicated that overall genetic contribution to LO CD was lower relative to the middle group, and genetic heterogeneity testing indicated that LO CD was different from the middle group in underlying genetic architecture.

Conclusions

Late-onset CD is subgroup distinct in genetic and behavioral risk factors with UC-like characteristics.

INTRODUCTION

Crohn’s disease (CD), a debilitating gastrointestinal disorder affecting more than a million people in United States,^{1, 2} is a complex disorder with a wide spectrum of observed phenotypic heterogeneity.³ Understanding this heterogeneity is important for more accurate prognosis assessment, better treatment strategies, and ultimately the development of tailored and targeted therapeutics.

Previous investigations indicate that clinical phenotypes and natural history of CD might differ according to age at diagnosis. Several studies have reported that early-onset CD patients have more severe disease,^{4, 5} more upper gastrointestinal issues,^{6, 7} and a greater need for aggressive treatment.^{5, 8, 9} In a number of studies, it has been reported that late-onset CD might differ in terms of disease location and natural history,^{6, 10–12} although the findings have been inconsistent.

In the widely applied Montreal^{13, 14} and Paris classifications,¹⁵ CD patients are classified into subgroups based on age at diagnosis. However, those classifications are largely based on the heterogeneity in clinical features, and little is known about the potentially different mechanisms contributing to onset of symptoms and age at diagnosis. Instead of focusing on the heterogeneity in clinical features, one could identify patient subsets from an etiological point of view, which might provide insight to the underlying mechanisms for this complex human diseases and lead to novel intervention strategies.

In this study, we examined the distribution of genetic factors based on 172 CD loci (including loci associated with overall IBD) identified in 2 recent large-scale association studies^{16, 17} across a European ancestry cohort of CD patients with different ages of onset. We calculated the polygenetic risk score (PRS),¹⁸ which reflects the overall genetic burden in CD, and examined its distribution across age groups. We identified a late-onset (LO) subgroup (defined as age at diagnosis ≥55 years) characterized by significantly lower PRS, which we replicated in the large International IBD Genetic Consortium (IIBDGC) cohort.^{16, 17} We further identified specific clinical and demographic features, including a significant role of smoking cessation in the LO group. Moreover, we examined differences in serological markers in the index cohort to explore the difference in innate/adaptive immunity in the identified subgroup.

METHODS

Subjects

The details of subject recruitment in the Cedars-Sinai cohort have been described previously.^{19, 20} Briefly, CD patients were recruited at Cedars-Sinai Medical Center (CSMC) from 1985 to 2015. The diagnosis of each patient was based on standard endoscopic, histologic, and radiographic features, as previously described.^{19, 20} Blood samples were collected at the time of enrollment. The study protocol and data collection, including DNA preparation/genotyping and antibody measurement, were approved by the CSMC Institutional Review Board. Written informed consent was obtained from all study participants.

Subject recruitment in the IIBDGC cohort is documented elsewhere.^16–18 Briefly, 17,302 CD patients and controls were recruited from 15 countries in Europe, North America, and Oceania. Diagnosis of IBD was based on accepted radiological, endoscopic, and histopathological evaluation. All included cases fulfilled clinical criteria for IBD and gave written consent.

Clinical and Serologic Phenotyping

Clinical data, including patients’ current age, sex, age at diagnosis, current disease location and behavior (according to the Montreal Classification), surgical history, and smoking status at diagnosis, were collected as previously described.¹⁸ In the Cedars cohort, IBD-associated serologies, including antinuclear cytoplasmic antibody (ANCA), antisaccharomyces cerevisiae antibodies (ASCA IgG and IgA), antiflagellin (anti-CBir1), anti-outer membrane protein C (anti-OmpC), and antipseudomonas fluorescens–related protein (anti-I2), were measured by enzyme-linked immunosorbent assay, as previously described.^{21, 22} All assays were performed blinded to patient clinical characteristics. Based on the measured serological markers, we further calculated the quartile sum score (QSS) of Anti-CBir1, Anti-I2, Anti-OmpC, IgA-ASCA, and IgG-ASCA.²¹

Genotyping and Genotype Quality Control

Genotyping of the Cedars cohort was performed at CSMC using Illumina ImmunoChip (IChip) array. Individual and genotype missingness, allele frequencies, and deviations from Hardy-Weinberg equilibrium were calculated using the PLINK software package.²³ Individual-level quality control (QC) thresholds include genotyping call rate >95%, inbreeding coefficient <0.05, and lack of cryptic relatedness (Pi-hat > 0.25 in PLINK). Ethnicity outliers identified using Admixture software²⁴ were also removed. Single nucleotide polymorphisms (SNPs) with a call rate <0.95, with minor allele frequency (MAF) <0.01, and that strongly deviated from Hardy-Weinberg equilibrium (P < 1 × 10^–7) were also removed. After QC, there were 2344 CD cases and 118,611 SNPs available for analysis in this cohort.

Genotyping and QC in the IIBDGC cohort have been described elsewhere.^16–18 In brief, the IIBDGC IChip samples were genotyped in 36 batches, and genotype calling was performed separately for each batch. Similar QC was performed, which removed SNPs with a call rate lower than 98% across all genotyping batches or 90% in 1 of the genotyping batches, but not in 1000 Genomes Project Phase I, failing Hardy-Weinberg equilibrium (false discovery rate [FDR] < 1 × 10^-5 across all samples or within each genotyping batch), or monomorphic SNPs. Individuals were assigned to different populations based on principal components, and those not in the Caucasian cluster or with a low call rate (<98%), outlying heterozygosity rate (FDR < 0.01), or cryptic relatedness (identity by decent > 0.4) were removed. After QC, 152,232 SNPs and 13,065 CD cases were included in current analysis.

PRS Calculation

CD PRS were calculated as a weighted sum of the number of risk alleles carried by each individual (0, 1, or 2) at known CD loci (n = 172, including 126 loci also associated with overall IBD), with weights proportional to the effect estimates from the previously published large-scale association studies.^16–18 The PRS were then normalized separately in the Cedars and IIBDGC cohorts to have mean of 0 and standard deviation of 1. We also calculated a UC PRS, in which known UC loci (n = 157, including 126 loci associated with IBD) were included in the score calculation. As there is strong overlap in loci used to construct UC and CD PRS, we further calculated a UC-only PRS, in which variants only associated with UC but not with CD or IBD in previous reported large-scale association studies (n = 31) were included. Details of the loci included in score calculation can be found in Supplementary Table 1.

Statistical Analysis

Ages at diagnosis were first grouped by 5-year increments to identify distinct subgroups, and analysis of variance (ANOVA) was used to examine the difference of PRS across different groups. Structural changes in PRS as ages of diagnosis increased were evaluated using the strucchange package in R²⁵ based on the method proposed by Zeileis, Shah, and Patnaik.²⁶ We thereby divided CD patients into subgroups based on the identified changing point.

Thereafter, logistic regression was performed to compare the difference between the identified subgroups in demographic, clinical, and serological characteristics. The Cedars cohort was used as a discovery cohort, and replication was performed in the IIBDGC cohort when applicable. Serological analysis was performed only in the Cedars cohort as there were no serological data available in IIBDGC. To account for the correlation of the clinical and serological factors, identified variables in univariate analyses were put in a joint model to identify independently associated factors, and Akaike information criterion (AIC)–based stepwise model selection was used to identify variables in the final model.

Associations of Ichip SNPs and the LO CD were performed using the software PLINK²³ separately in the Cedars and IIBDGC cohorts, and a meta-analysis was performed to combine results from both cohorts. Mixed-model based association²⁷ was utilized to evaluate pathway level difference between the LO and intermediate groups, with pathways defined based on the KEGG pathway database.²⁸

We also calculated the genetic variance contribution from all SNPs genotyped on ImmunoChip for the LO and middle groups separately in the Cedars and IIBDGC cohorts using GCTA software.²⁹ The difference in underlying genetic architecture between the LO and middle groups was further examined using the pseudo-likelihood ratio (PLR) approach proposed by Liley et al.³⁰ The same approach was also utilized to examine the genetic similarity between LO CD and UC.

In all analyses, principal components from population stratification analysis³¹ were included as covariates. In the smoking analyses, current age was also included to control for potential confounding effects from patients’ age. To confirm that the association of smoking behavior with the LO group was not due to confounding of the age-cohort effect, we further performed a matched analysis in which patients from the LO group were matched based on 5-year groups in current age. Conditional logistic regression analysis was performed to examine the association of smoking behaviors with the LO CD in the matched analysis. All statistical analyses were performed using R 3.2.0.²⁵

RESULTS

Clinical characteristics for the 2 cohorts are shown in Table 1. In both cohorts, the majority of the CD patients were diagnosed between age 5 and 55 years; 106 (4.52%) and 845 CD patients were diagnosed after age 55 years in the Cedars and IIBDGC cohorts, respectively.

We first examined the distribution of PRS with different ages at diagnosis (grouped by 5 years) in the Cedars cohort (Fig. 1) and observed statistically significant differences (P = 1.17 × 10^-6) between groups. Structural change tests indicated that there is statistically significant change in the slope of PRS as age at diagnosis increases (F-value, 29.59; P = 1.23E-5). This analysis also demonstrated that there is 1 changing point in the slope of PRS in the age group ≥55 years. We thereby regrouped patients with age at diagnosis ≥55 years as the LO group. To exclude potential confounding from the very early-onset (VEO) CD patients, who have monogenic forms of IBD, we defined the patients with age at diagnosis <5 years as the VEO group and patients with age at diagnosis of 5–55 years as the intermediate group.

Thereafter we examined the mean PRS and observed a significantly lower PRS in the LO group compared with the intermediate group (0.15 vs 0.55, P = 2.99 × 10^-5) (Table 2). This difference was replicated in the IIBDGC cohort (0.25 vs 0.48, P = 1.35 × 10^-11) (Table 2). No significant difference was observed when comparing VEO CD patients PRS with the intermediate group, in either of the cohorts. Moreover, the distribution of CD PRS in patients was not observed in non-IBD patients (grouped based on current age) (Supplementary Fig. 1).

We also compared the allele frequencies of known CD SNPs^{16, 17} between the LO and intermediate groups (details in Supplementary Table 2). The NOD2 frameshift mutation (rs5743293) is the only known CD-associated SNP that was associated, after correction for multiple testing, with the LO CD group when compared with the intermediate group (OR, 0.58; P = 3.98 × 10^-6 in the meta-analysis), with the direction of association in the inverse direction of the reported association with CD. Interestingly, the log of odds ratios (ORs) for the known CD variants in the LO vs intermediate analysis are negatively correlated (r = –0.67; P = 3.54 × 10^-25) with the association observed in the intermediate group vs non-IBD controls (Fig. 2). This negative correlation remained when SNPs with strong effects (including variants in NOD2, ATG16L1, and IL23R) were excluded (r = –0.43; P = 9.90 × 10^-7).

We further examined sex and smoking behavior in the LO and intermediate groups (Table 3). There was no statistically significant difference between the LO and intermediate groups in sex in the Cedars cohort (OR, 0.87; P = 0.49) or the IIBDGC cohort (OR, 0.97; P = 0.76). In the Cedars cohort, there were significantly fewer current smokers (OR, 0.03; P = 4.60 × 10^-4), and there were more ex-smokers (OR, 4.38; P = 3.30 × 10^-6) in the LO group, even after controlling for current age. A similar phenomenon was observed in the IIBDGC cohort (OR, 0.72 and 1.65; P = 5.26 × 10^-3 and 2.17 × 10^-5, respectively). There was no statistically significant difference for ever-smoker in either cohort. In the matched analysis in which patients from the LO group were matched based on 5-year age groups in current age, a similar trend was observed (for ex-smoking, hazard ratio [HR], 4.22 and 1.34; P = 7.39 × 10^-5 and 5.91 × 10^-3; in the Cedars and IIBDGC cohorts, respectively; for current-smoking, HR = 0.08 and 0.75; P = 0.02 and 0.01; in the Cedars and IIBDGC cohorts, respectively).

Distribution of disease location in the LO and intermediate groups is shown in Table 4. There was no statistically significant difference in L1 between the LO and intermediate groups in the Cedars cohort, although the difference was marginally significant in the IIBDGC cohort (OR, 1.21; P = 0.013). The proportion of L2 was higher in the LO group compared with the intermediate group in the Cedars cohort (OR, 2.00; P = 2.58 × 10^-3) and also in the IIBDGC cohort (OR, 2.48; P = 7.74 × 10^-32). In contrast, the proportion of L3 was lower for LO in both the Cedars cohort (OR, 0.57; P = 7.90 × 10^-3) and the IIBDGC cohort (OR, 0.32; P = 7.93 × 10^-36). There was no statistically significant difference for L4 and perianal location in the Cedars cohort, but the results were significant in IIBDGC (OR, 0.31; P = 2.22 × 10^-8; and OR, 0.38; P = 6.11 × 10^-18; respectively, for upper gastrointestinal [GI] and perianal location).

In Table 4, we also compared disease behavior (measured based on Montreal classification^{13, 14}) and need for surgery between the LO and intermediate groups. In both cohorts, the difference in disease behavior was observed in B3 (penetrating disease) vs B1 (nonstricturing nonpenetrating disease), with an OR of 0.26 (P = 3.84 × 10^-4) in the Cedars cohort and 0.40 (P = 1.30 × 10^-16) in the IIBDGC cohort. The difference in B2 (stricturing disease) vs B1 was not significant in the Cedars cohort but showed borderline significance in IIBDGC (OR, 0.84; P = 0.047). In both cohorts, patients had less need for surgery in the LO group (OR, 0.56; P = 6.30 × 10^-3; in the Cedars cohort; OR, 0.47; P = 7.65 × 10^-22; in IIBDGC). Similar results were observed if the length of follow-up was included as a covariate in both cohorts (details not shown).

We further examined serological markers in the LO and intermediate groups in the Cedars cohort (Table 5). In the LO group, a higher proportion of ANCA+ (OR, 1.57; P = 0.034) and a lower proportion of both anti-CBir1+ (OR, 0.61; P = 0.020) and both IgA-ASCA+ (OR, 0.21; P = 2.27 × 10^-6) and IgG-ASCA+ (OR, 0.29; P = 1.43 × 10^-5) were observed. We also calculated the QSS of serological markers using Anti-CBir1, Anti-I2, Anti-OmpC, IgA-ASCA, and IgG-ASCA, and lower QSS was observed in the LO group (11.67 ± 2.96 in the LO group vs 12.68 ± 3.85 in the intermediate group; P = 0.027).

We subsequently examined the difference between the LO and intermediate groups with demographic, serological (only in the Cedars cohort), and clinical characteristics (B3, L2, L3, and surgery) jointly in a multivariate model. Interestingly, in the Cedars cohort, differences in 3 factors (ex-smoking, ASCA positivity, and CD PRS) were observed in the joint model (P = 5.64 × 10^-7, 3.87 × 10^-4, and 0.046, respectively), whereas the difference in B3 and L2 was not significant at all. To mimic the situation in the IIBDGC cohort, in which serological markers were not available, we re-examined the joint model after excluding the serological markers, and marginal differences in 2 more variables, B3 and L2 (P = 0.036 and 0.099, respectively), were observed in addition to ex-smoking and CD PRS in the joint model. A similar pattern was observed in the IIBDGC joint model, with differences in ex-smoking, B3, L2, and CD PRS observed (P = 2.0 × 10^-3, 3.14 × 10^-9, 5.04 × 10^-15, and 2.10 × 10^-7, respectively).

With the UC-like features of the LO CD group, we calculated a UC GRS in which all known SNPs associated with UC (with SNPs associated with both CD and UC included) were included and weighted based on association with UC. A similar pattern was observed with lower UC PRS in the LO CD group (Supplementary Fig. 2). We further constructed a UC-only PRS, in which only SNPs that are associated with UC but not CD or IBD were included (based on SNPs reported to be associated only with UC in Jostin et al.¹⁶ and Liu et al.¹⁷), and there was no statistically significant difference between the LO and intermediate groups in UC-only PRS in the Cedars cohort (P = 0.29) or in the IIBDGC cohort (P = 0.75).

After excluding the known loci, no statistically significant signal survived the genome-wide significance threshold (data not shown) for single SNPs in iChip comparing the LO group with the intermediate group. The pathway-level association suggests that the NOD-like receptor signaling pathway (P = 1.07 × 10^-4) (Supplementary Table 3) might contribute to the difference between the LO and intermediate groups.

We also examined the correlation in the association signals of the LO group vs the intermediate group, and those in UC vs CD. A surprisingly strong correlation in log(ORs) of the 2 independent associations was observed in SNPs with P values <1.0E-3 in UC vs CD (r = 0.57; P = 1.68E-13) (Fig. 3; details in Supplementary Table 4) after pruning for SNPs in linkage disequilibrium (LD) with each other. Similar patters were observed when using different P value cutoffs from 0.05 to 5.0E-8 in UC vs CD (details not shown).

We used the software GCTA to estimate the genetic contribution from all SNPs in ImmunoChip to the LO CD and middle groups. After control for principal component analysis, the genetic contributions to the LO and middle groups were estimated to be 6.2% (P = 0.15) and 10.24% (P = 2.02E-90) for the 109,525 variants that passed the QC procedures in the Cedars cohort and 11.5% (P = 2.15E-17) and 17.83% (P < 4.94E-324) for the 156,499 variants that passed the QC procedures in the IIBDGC cohort.

We further tested the hypothesis that the overall underlying genetic architecture is different between the LO and middle groups using the PLR approach.³⁰ We observed a PLR of 44.57 in the Cedars cohort for the heterogeneity between the LO CD and middle groups, and after 2000 permutations, we observed a P value of 0.17. In IIBDGC, the PLR and P value after permutation were 771.56 and 5.05E-6, respectively.

We also compared the underlying genetic architecture of LO and UC using the PLR approach. In the Cedars cohort, the PLR was 32.52 with a P value of 0.19. In IIBDGC cohort, the PLR and P value were 3374.34 and 2.40E-6, respectively, for the genetic heterogeneity between LO CD and UC.

DISCUSSION

In the current study, we examined the CD PRS in CD patients with different ages at diagnosis and identified that LO patients with age at diagnosis ≥55 years have lower PRS. LO CD patients differ in disease location, disease behavior, and need for surgery, consistent with previous reports. Moreover, we demonstrated UC-like smoking behaviors (more ex-smokers, fewer current smokers) and UC-like serological patterns (high ANCA, low ASCA) in the LO group. Our analyses also illustrated that in an LO group vs intermediate group comparison, there is a parallel pattern in single-SNP level signals to UC vs CD. Our findings clearly demonstrate that the late-onset CD group is a distinct subgroup with UC-like features that could impact future clinical practice.

For more than 30 years, clinicians have observed that late-onset CD patients might have different characteristics than their younger peers. There have been a number of studies comparing LO with other CD patients,^{6, 10–12} with conflicting results, probably due to small sample sizes and inconsistent and often arbitrary definitions of “late onset.” In the current study, we started from an etiological point of view by examining the genetic burden in patients with different ages at diagnosis, and further expanded the comparison to include clinical and immunological features. The main findings were replicated in an independent, large cohort, indicating that we’ve identified a distinct subgroup of CD. The way we identified this subgroup can be viewed as a model mechanism for patient stratification in other complex diseases.

The observed lower PRS in the LO group, in both cohorts, indicates differences in underlying pathogenic mechanisms in this subgroup. The cause of CD remains unclear, but it’s generally agreed that the interplay of genetic factors, innate/adaptive immunity, microbiome, and environmental triggers are contributing factors.^32–34 The much lower PRS in the LO group indicates that known genetic variants, which have been demonstrated to be critical in disease development in most CD cases, might play much less important roles in the subgroup of LO CD. This may imply distinct underlying mechanisms in LO patients. Those results are further supported by the heterogeneity test that indicates strong difference in genetic structure underlying the LO and middle groups, with significant results in the PLR test in the IIBDGC cohort, although the test for difference in genetic structure in the Cedars cohort was not significant, probably due to a relatively small sample size.

It is important to determine whether the difference of PRS in LO is due to a subset of the known associated SNPs. When examining the differences of allele frequencies of single known SNPs between the LO and intermediate groups, the only variant that was statistically significant was the NOD2 frameshift mutation, with a much lower frequency in the LO group. Moreover, we observed a strong negative correlation between log(ORs) in the LO vs intermediate groups and the intermediate group vs non-IBD controls, even after excluding variants like NOD2. Thus, for the majority of known variants, the LO group has lower allele frequencies for risk variants and higher frequencies for protective variants, although our sample size may not to be large enough to capture, with high statistical confidence, any association with other SNPs beyond NOD2. This suggests that the low PRS in the LO group is not due to a single or a few known SNPs, but an overall weaker genetic burden, which is further supported by the lower overall genetic contribution from all SNPs on the ImmunoChip in the LO group.

Cigarette smoking is one of the few widely accepted environmental risk factors for CD^{35, 36} and is associated with more complicated diseases,^{32, 37–43} need for surgery,^{32, 37–39, 42, 43} and relapse after treatments.^{32, 39–41, 44} Conversely, smoking cessation decreases the risk of developing CD but is associated with increased risk of UC.^{35, 36} Unexpectedly, we observed a lower prevalence of current smokers and more ex-smokers in the LO group of CD patients, which is a UC-like feature. As age was included as a covariate in smoking behavior analysis, the observed effects of smoking behaviors cannot be explained by the age-cohort effect. This is also supported by the similar results observed when examining age-matched analysis. Smoking cessation can lead to profound changes in the composition of the gut microbiome,⁴⁵ and the gut microbiome is well known to play important roles in the development of CD.^{46, 47} Still, it is intriguing to observe opposite effects of smoking behaviors in CD at different ages of onset, indicating, further, a complex role for smoking as an environmental factor in IBD. Further investigation is warranted to demonstrate the role of smoking in the pathogenesis of this subgroup and, moreover, whether it shares similar etiology as UC. Furthermore, it would be worthwhile to explore the potential treatment options related to this unique and modifiable environmental factor.

Distinction of LO CD is also reflected by our analysis examining the serological markers in the LO group, with higher ANCA and lower anti-CBir1 and ASCA observed. Higher ANCA has long been associated with UC^{44, 48} and with UC-like features in CD, including L2 disease location.⁴⁹ The observed higher ANCA in LO CD indicates a common intestinal mucosal inflammatory process between LO CD and UC. Higher ANCA has also been linked to nonresponse to anti–tumor necrosis factor (anti-TNF) treatments in both CD and UC,^{50, 51} and given the higher adverse event rate that has been reported in the elderly on anti-TNF therapies, these data, collectively, suggest that a dedicated trial of the risks and benefits of anti-TNF therapy in late-onset Crohn’s disease is warranted. Anti-CBir1 and ASCA reflect innate/adaptive immunity to selected microbiome agents,^{52, 53} and both are associated with a more complicated CD phenotype. Lower anti-CBir1 and much lower ASCA clearly indicate an alternative route of shifted genetic/environmental factors leading to abnormal immune response in the gut microbiome, which in turn influences disease onset, leading to different clinical phenotypes, and defines a separate disease subgroup. However, the observed serological characteristics of LO CD, which were identified in the Cedars cohort, cannot be replicated in the IIBDGC cohort as there are no serology measures in the IIBDGC phenotype data set. Independent replications on the serological measures in LO CD are warranted to validate the serological findings in the current study.

In the current study, we have demonstrated that there is more isolated colonic (L2) and less ileocolonic disease in the LO group. This is consistent with previous observations in elderly-onset CD patients.^{6, 11, 12, 54, 55} We also observed that LO CD patients tend to have less need for surgery and less penetrating behavior, consistent with previous reports that isolated colonic CD tends to be less severe.^{6, 10–12, 55} It is also critical to develop specific intervention strategies for this subgroup. For example, the LO CD cohort may be a subgroup of CD patients who respond to mesalazine or even to anti-MADCAM therapy, both of which seem to be more effective therapies for UC than CD.^{56, 57}

With more colon-only affection, UC-like smoking behaviors (more ex-smokers, fewer current smokers), and UC-like serological patterns (high ANCA and low ASCA), the LO group is likely a distinct subgroup of CD with UC-like characteristics in both clinical presentation and disease etiology. This is also supported by the strong correlation of single-SNP signals in LO vs intermediate and CD vs UC comparisons. These findings further illustrate the heterogeneous nature of CD and the need for more personalized treatment strategies in clinical practice.

Interestingly, even with the UC-like clinical and environmental characteristics, heterogeneity analysis using a PLR approach³⁰ indicates that the overall underlying genetic structure is different between LO CD and UC. The seemingly contradicting results are not very surprising as the LO CD patients were diagnosed as CD but not UC based on standard clinical and endoscopic criteria, indicating submucosal or transmural inflammations in those patients, which might be related to different genetic mutations. One possibility is that UC and LO CD might have related or overlapping but still distinct pathogenesis mechanisms; more research is needed to solve this complex puzzle.

Of note, conditioning on etiological factors (smoking behavior and CD PRS) and serological markers reflecting host response to the microbiome, the difference in clinical presentations (disease location and behavior) is no longer significant in the Cedars cohort. This observation seems to indicate that the observed clinical characteristics in the LO group are mainly due to the difference in underlying pathogenesis, which again suggests that LO CD may be a unique IBD subgroup. This observation cannot be replicated in the IIBDGC cohort, though, as serological markers are not measured in this cohort, and additional cohorts are needed to replicate this particular observation.

We did not know the exact age of disease onset for each CD patient in this current study. Instead, we used age at diagnosis as a proxy for age of onset. The diagnosis of CD often presents a challenge for clinicians, probably due to similar presenting symptoms as functional digestive pathologies such as irritable bowel disease.^{58, 59} This delay in diagnosis has been reported to be 18–24 months on average^{58, 60} and in certain cases can be multiple years. This could potentially bias our results, in particular in the IIBDGC cohort, in which centers from different areas of world are involved. However, this delay of diagnosis would only make patients with age of onset before age 55 years classified as late-onset in our analysis, which would likely bias our results toward the null hypothesis. In this sense, our findings may be more conservative.

Note that in the previously established Montreal and Paris classification of CD,^13–15 the A3 group was defined as CD patients with age at diagnosis >40 years. The identified LO subgroup, with its distinct clinical and etiological characteristics, indicates that adding an additional subgroup for a later age of onset is warranted. It is possible that defining disease subgroups based on age of onset or age at diagnosis, which is probably a proxy to the difference in underlying disease-causing mechanisms, cannot be exact by nature. Better characterization of the underlying disease-causing mechanisms might help to identify more homogenous subgroups in CD.

It is worth noting that in the current study, the Cedars cohort mainly consists of CD patients from the southwestern United States, whereas the IIBDGC cohort (after excluding overlapping samples) consists of patients from different locations in North America, Europe, and Australia, with the majority of the samples from Europe.¹⁶ Between the 2 cohorts, the source populations will likely have distinct demographic characteristics with different disease diagnosis/treatment guidelines applied in clinical practice. This might explain the observation that the effect sizes vary between the 2 cohorts in the comparison between the LO and intermediate groups. This is particularly true for smoking behavior–based analysis as previous epidemiology studies have indicated a lower smoking prevalence rate in the United States in comparison with Europe.^{61, 62} Moreover, the smoking prevalence rate in California, where most subjects in the Cedars cohort are from, was reported to be the lowest in the United States.⁶³ This partially explain the dramatic difference in the proportion of current smokers and ever smokers between the Cedars and IIBDGC cohorts. Despite these differences, we still observed a highly consistent pattern in the 2 cohorts, strongly suggesting that the distinct characteristics we observed in the LO group are genuine.

In summary, we identified late-onset CD patients as a distinct subgroup with different genetic, clinical, environmental, and serological characteristics. The features in this subgroup are more UC-like, and further investigations on the underlying mechanism(s) and specific treatment strategies are warranted.

SUPPLEMENTARY DATA

Supplementary data are available at Inflammatory Bowel Diseases online.

Conflicts of interest: None.

Supported by: U01DK062413 (D.P.B.M.), P01DK046763 (S.R.T., D.P.B.M.), R01HS021747 (D.P.B.M.), U01AI067068 (D.P.B.M.), and R21DK108140 (D.L.). The International IBD Genetic Consortium Data Coordinating Center is supported by National Institute of Diabetes and Digestive and Kidney Diseases grant DK062429.

Author contributions: D.L. proposed and performed the analysis. D.P.B.M., S.R.T., and M.D. supervised the whole project. T.H. performed the genotyping. T.H., A.P., and D.L. performed the QC of the genotype data. S.Y. and D.P.B.M. collected the clinical phenotypes. C.L. measured the serological markers. D.L. and H.H. calculated the genetic risk score. Primary drafting of the manuscript: D.L., S.R.T., D.P.B.M. All authors read, had the opportunity to comment on, and approved the final draft.

REFERENCES