Novel Genetic Risk Variants Can Predict Anti-TNF Agent Response in Patients With Inflammatory Bowel Disease

Abstract

Background

It is important to identify patients with inflammatory bowel disease [IBD] refractory to anti-tumour necrosis factor [TNF] therapy, to avoid potential adverse effects and to adopt different treatment strategies. We aimed to identify and validate clinical and genetic factors to predict anti-TNF response in patients with IBD.

Materials and Methods

Mayo Clinic and Washington University IBD genetic association study cohorts were used as discovery and replicate datasets, respectively. Clinical factors included sex, age at diagnosis, disease duration and phenotype, disease location, bowel resection, tobacco use, family history of IBD, extraintestinal manifestations, and response to anti-TNF therapy.

Results

Of 474 patients with IBD treated with anti-TNF therapy, 41 [8.7%] were refractory to therapy and 433 [91.3%] had response. Multivariate analysis showed history of immunomodulator use (odds ratio 10.2, p = 8.73E-4) and bowel resection (odds ratio 3.24, p = 4.38E-4) were associated with refractory response to anti-TNF agents. Among genetic loci, two [rs116724455 in TNFSF4/18, rs2228416 in PLIN2] were successfully replicated and another four [rs762787, rs9572250, rs144256942, rs523781] with suggestive evidence were found. An exploratory risk model predictability [area under the curve] increased from 0.72 [clinical predictors] to 0.89 after adding genetic predictors. Through identified clinical and genetic predictors, we constructed a preliminary anti-TNF refractory score to differentiate anti-TNF non-responders (mean [standard deviation] score, 5.49 [0.99]) from responders (2.65 [0.39]; p = 4.33E-23).

Conclusions

Novel and validated genetic loci, including variants in TNFSF, were found associated with anti-TNF response in patients with IBD. Future validation of the exploratory risk model in a large prospective cohort is warranted.

1. Introduction

Inflammatory bowel diseases [IBDs], with two main subtypes of Crohn’s disease [CD] and ulcerative colitis [UC], are chronic, relapsing, intestinal inflammatory diseases affecting more than 2.5 million White people [European descent], with increasing prevalence in Asia and developing countries.^1,2 The causes of IBD have long been studied, but the advent of biologic drugs in the past two decades has substantially reshaped treatment. Tumour necrosis factor [TNF]-α is a potent proinflammatory cytokine that plays an important role in many inflammatory and autoimmune responses, as in the pathogenic mechanism of IBD.³ Anti-TNF therapy was approved for use in CD in 1998 and for UC in 2005. Several anti-TNF agents have been evaluated, including infliximab, adalimumab, certolizumab, and golimumab. However, about a third of patients with IBD were non-responders to anti-TNF agents,^4,5 and half of patients who responded initially experienced loss of response later.^6,7 Several clinical factors, including shorter duration of disease, younger age of patients, no previous IBD-related surgery or stricturing phenotype, higher C-reactive protein level, no previous failing corticosteroids or immunomodulators, no history of smoking, and optimizsd trough level of anti-TNF agent, have been associated with and predict better anti-TNF agent response in patients with IBD.^8–11

Advances in microarray-based biotechnology allowed investigators to efficiently genotype enormous numbers of single nucleotide polymorphisms [SNPs] in a short time, and have facilitated development of genome-wide association studies [GWASs] in complex human diseases. In recent years, applying interindividual genetic variations to identify potential risk for disease, drug response, and adverse reactions, has achieved varying degrees of success in several clinical fields and supported a further shift toward a more personalised, less empirical approach to health care.¹² A landmark study, combining 75 000 cases of CD, UC, and controls across 15 previously conducted adolescent- and adult-onset IBD GWASs, identified 163 IBD loci [110 shared between CD and UC, 30 CD-specific, and 23 UC-specific] and supported a connection between disease risk and host interactions with microbes, such as mycobacterial infections.¹³ A recent GWAS meta-analysis has been instrumental in defining over 230 IBD loci.¹⁴

Another recent meta-analysis suggested that carrying NOD2 mutations in CD may require aggressive therapeutic strategies [e.g., anti-TNF therapy].¹⁵ However, conflicting results were found regarding the influence of NOD2 polymorphisms on anti-TNF therapy response. In a previous study, R702W, G908R, and L1007finsC polymorphisms in NOD2 related to poorer response to anti-TNF agents,¹⁶ whereas other studies found that NOD2 mutations did not have any impact on the response to infliximab.^17,18 A recent Spanish study reported that the proportion of patients on an intensified biologic therapy was significantly higher among CD patients with a NOD2-variant, either alone or combined with an ATG16L1-variant [T300A], suggesting that there might be an influence of NOD2 on anti-TNF response.¹⁹

Earlier studies, focused on apoptosis genes and response to infliximab in CD, found that polymorphisms in the Fas ligand-843 and caspase-9 gene alleles were associated with improved response to infliximab, and interestingly, the effect of these polymorphisms on response to infliximab was cumulatively dose-dependent.^20,21 In a pilot study of ulcerative colitis, certain IL23R gene variants were found as modulators of the response to anti-TNF therapy.²² Additionally, previous studies suggested that homozygous haplotypes in the TNF-α region and carriers of TNFR1 A36G mutation were associated with poor anti-TNF responses.^18,23 A recent meta-analysis examining TNF-α polymorphisms and anti-TNF response in spondyloarthropathy, psoriasis, and CD suggested increase of anti-TNF response in TNF-α -308 G (odds ratio [OR], 2.01; p = 8.6E-05), TNF-α -238 G [OR, 2.20; p = 0.02], and TNF-α -857 C alleles [OR, 1.78; p = 0.01].²⁴ However, conflicting negative results were observed in other studies.^25,26

The lack of valid replication of the previously observed genetic associations likely resulted from the heterogeneity of study design, less stringent definition of the anti-TNF responsiveness, and relatively small sample size. A hypothesis-free approach testing polymorphisms across the genome in large, well-characterised cohorts is suggested in order to identify the novel and truly impactful genetic biomarkers. Once the novel genetic association is validated in a thorough retrospective approach, further prospective study should be warranted.

Patients who do not respond to the anti-TNF therapy could be primary non-responders during the induction treatment or secondary non-responders [i.e., lose response after the induction treatment]. However, various time frames were used in clinical trials to determine primary non-responders for different anti-TNF agents [e.g., within 14, 12, or 8 weeks following initial infusions, respectively, with infliximab, adalimumab, and certolizumab].^27,28 In the literature, the observed lower response rates of first-time anti-TNF switchers compared with those of anti-TNF naïve patients, amd the even lower response rates of second-time anti-TNF switchers, suggest that the second-time switchers may represent a unique subgroup of IBD patients.^29,30 In this study, we applied a stringent criterion [i.e., failure of two different anti-TNF agents], and aimed to identify a unique and selected subgroup of IBD patients who may not benefit from the anti-TNF therapy.

Anti-TNF therapy is very expensive and may lead to several adverse effects, including immunogenicity, infections, malignancies, heart failure, demyelinating disease, and others.³¹ It is necessary to find a way to predict the efficacy of anti-TNF agents to prevent unnecessary biologic therapy and avoid adverse effects. A model that predicts responsiveness to anti-TNF therapy, taking into account not only disease-specific factors but also genetic factors, would prevent unnecessary anti-TNF therapy. Through hypothesis-free genetic association study approaches we developed and validated a model for the influence of genetic factors on anti-TNF responsiveness, through a large retrospective discovery cohort [Mayo IBD cohort] and another independent validation cohort [Washington University IBD cohort].

2. Materials and Methods

2.1. Genetic association study datasets

Two independent genetic association study datasets were used for this study, the Mayo Clinic IBD cohort [three sites: Rochester, MM, Scottsdale, AZ, and Jacksonville, FL] and the Washington University IBD cohort [Barnes-Jewish Hospital, St Louis, MO]. All data collection and study procedures were institutional review board-approved in both cohorts. Both datasets were retrospective cohorts and recruited patients with IBD for phenotyping and biospecimen collection. Adult patients [≥18 years old] with confirmed IBD validated by medical record review were eligible. All participants gave written informed consent. Baseline surveys were administered within 30 days of initial consent and included information on demographics, tobacco use, disease phenotypic characteristics, surgical history, and current and past treatment history. Disease location, behaviour, and extent were classified according to the validated National Institutes of Diabetes and Digestive and Kidney Diseases IBD Genetics Consortium modification of the Montreal Classification as previously described.³² Extraintestinal manifestations [EIMs] included joint involvement [small joint, large joint, ankylosing spondylitis, and sacroiliitis], eye involvement [iritis, uveitis, non-specific ocular inflammation], skin involvement [erythema nodosum, pyoderma gangrenosum], and primary sclerosing cholangitis.

The Mayo Clinic IBD dataset was used as the discovery cohort for evaluation of genotype-phenotype [anti-TNF response] association analyses. This discovery cohort consists of 223 patients with IBD [164 CD, 45 UC, and 14 indeterminate colitis], all of self-reported non-Jewish, European ancestry, who were successfully genotyped using the Illumina Immunochip custom genotyping array [Infinium ImmunoArray 24 v2-0_A] [Illumina, Inc.] at the Broad Institute. The Washington University IBD dataset [Illumina Immunochip custom genotyping array], consisting of 251 patients with IBD [195 CD, 54 UC, and two indeterminate colitis], all of self-reported non-Jewish ancestry, was used as the replicate cohort for genetic association analyses. Immunochip is a custom-made array that includes 253,702 genetic markers which was designed to cover deep replication of major autoimmune and inflammatory diseases [e.g., IBD], and fine-mapping of established GWAS significant loci. At established GWAS loci, Immunochip covers all known SNPs from available sequencing initiatives, including dbSNP database, and this enables cost-effective fine-mapping of loci for both rare and common variants.³³

2.2. Anti-TNF response

With the evidence of the lower response rates in first-time anti-TNF switchers, and even worse response rates in second-time switchers, than those of anti-TNF naïve patients, we aimed to identify a unique subgroup of IBD patients who may not benefit from anti-TNF therapy, through a stringent criterion [i.e., failure to two anti-TNF agents]. In study surveys regarding primary outcome of response to anti-TNF therapy, ‘Yes’ was defined as being treated with an anti-TNF agent, infliximab or adalimumab, for the first time, with continued use at the time of study enrolment without failure to any biologics, and ‘No’ was defined as neither infliximab nor adalimumab being effective after treatment. Discontinuation of anti-TNF agent due to adverse effect during treatment course was excluded from the analysis.

2.3. Statistical analysis

We applied the following quality control criteria to exclude single nucleotide polymorphisms: 1] Hardy-Weinberg equilibrium test p-values <1.0E-05; 2] minor allele frequency less than 1%; and 3] call rate less than 95%. In total, 213 386 single nucleotide polymorphisms [SNPs] [i.e., 84% of the originally genotyped SNPs] passed the quality control filters. Samples with cryptic relatedness [estimated identical-by-descent, PI-HAT more than 0.25] were excluded. Principal components were computed to examine and adjust for population stratification using the first two principle components (PC1: 0.98, PC2: 0.78). We also conducted a genomic control approach to correct population stratification and the estimated inflation factor lambda λ was 1.03. For those SNPs associated with the study outcome, raw signal intensity cluster plots to confirm the accurate allele calling was conducted using Illumina Genome Studio v2011.

For genetic association analysis with anti-TNF response, SNPs which reached p <1E-03 in the discovery cohort were further validated in the replicate cohort. SNPs that reached p <1E-03 and p <0.05 in the replicate cohort were considered successfully validated and suggestive evidence, respectively. Odds ratios [ORs] and 95% confidence intervals[CIs] were calculated to estimate single locus effects for risk alleles and genotypes. For uncommon alleles, the CIs could be relatively large when it was separately analyzed in each cohort. Two sample t-tests were used for continuous variables, and nominal data were analysed using the χ² test or Fisher’s exact test. These analyses were implemented in the Golden Helix SVS software suite 8.3 [Golden Helix, Inc.]. The odds of anti-TNF response were estimated through a multivariate logistic regression model using JMP Pro 11.0.0 [SAS Institute, Inc]. The final predictive model was determined through the forward stepwise selection process. The estimate of the proportion of disease variation explained by a model can be done using the likelihood-based pseudo R-squared provided by the logistic regression analysis in JMP.

We applied multivariate logistic regression modelling to investigate the predictive accuracy of models derived from the clinical and genetic predictors for anti-TNF response in patients with IBD. Discriminative accuracy was evaluated using area under the receiver operating characteristic curves [AUCs]. AUCs were compared between models of clinical predictors only and combined and genetic predictors. Likelihood ratio test was used to compare the predictability between models of clinical factors only, genetic factors only, and combined clinical and genetic factors. We further constructed the anti-TNF refractory score by summing the selected clinical and genetic predictors, weighted by their regression coefficients derived from the combined cohort, for each study subject to predict the anti-TNF non-response rate.

3. Results

3.1. Clinical factors associated with anti-TNF response in patients with IBD

Among 474 patients with IBD [359 CD, 99 UC, 16 indeterminate colitis] of European ancestry, successfully genotyped and treated with anti-TNF therapy, 41 [8.7%] were refractory to anti-TNF agents and 433 [91.3%] had response. Mean age at diagnosis was 29.8 years [13.4] in anti-TNF responders and 26.4 years [12.1] in anti-TNF non-responders. Patients who failed one but not two anti-TNF agents and who developed intolerance were excluded from the analysis [Supplementary Table 1, available as Supplementary data at ECCO-JCC online].

In univariate analysis, immunomodulatory use (OR, 10.7 [1.46–79.1]; p = 5.0E-04) and IBD-related bowel resection (OR, 3.38 [1.75–6.53]; p = 2.42E-04) were associated with a refractory response to anti-TNF agents. Age at diagnosis, disease duration, family history and phenotypes of IBD, advanced disease behaviour, perianal disease involvement, and EIMs were not associated with anti-TNF response. Although not statistically significant, male patients were less likely to be refractory to anti-TNF agent (OR, 0.55 [0.27–1.11]; p = 0.09). Multivariate analysis showed immunomodulatory use [OR, 10.2; p = 8.73E-4] and IBD bowel resection [OR, 3.24; P = 4.38E-4] were associated with refractory response to anti-TNF [Table 1].

Table 1.

Clinical and demographic factors between TNF antagonist responders and non-responders.

Clinical factors	Anti-TNF responders [n= 433]	Anti-TNF non-responders [n= 41]	OR [95% CI] p-Value	Adjusted OR [95% CI] p-Value
Sex, male, no. [%]	184 [42.5]	12 [29.3]	0.55 [0.27–1.11]	--
			p = 0.09
Age at diagnosis, years [SD]	29.8 [13.4]	26.4 [12.1]	1.35 [0.70–2.58]	--
<25 years, no. [%]	178 [41.1]	20 [48.8]	p = 0.37
Disease duration, years [SD]	11.9 [9.1]	12.5 [7.4]	p = 0.71	--
Family history of IBD, no. [%]	68 [15.7]	5 [12.2]	0.75 [0.28–1.97]	--
			p = 0.50
Smoking at diagnosis, no. [%]	74 [17.1]	8 [19.5]	1.18 [0.5–2.64]	--
			p = 0.70
IBD phenotype, no. [%]			p = 0.70	--
CD	327 [75.5]	32 [78.0]
UC	91 [21.0]	8 [19.5]
IC	15 [3.5]	1 [2.4]
Immunomodulator, no. [%]	339 [78.3]	41 [100]	10.7 [1.46–79.1]	10.2 [1.38–75.00]
			p = 5.0E-04	p = 8.73E-04
Bowel resection, no. [%]	137 [31.6]	25 [61.0]	3.38 [1.75–6.53]	3.24 [1.67–6.30]
			p = 2.42E-04	p = 4.38E-04
Advanced IBD behaviour [stricturing/penetrating, extensive colitis], no. [%]	245 [56.6]	26 [63.4]	1.32 [0.67–2.60]	--
			p = 0.42
Perianal abscess/fistula, no. [%]	74 [17.1]	10 [24.4]	1.78 [0.82–3.84]	--
			p = 0.16
Extra-GI manifestation, no. [%]			p = 0.28
Eye	17 [3.9]	3 [7.3]	p = 0.13	--
Skin	19 [4.4]	4 [9.8]	p = 0.75	--
Joint	83 [19.2]	8 [19.5]	p = 0.66	--
PSC	7 [1.6]	1 [2.4]		--

OR >1.00 suggests no response to anti-TNF agent.

CD, Crohn’s disease; GI, gastrointestinal; IBD, inflammatory bowel disease; IC, indeterminate colitis; OR, odds ratio; PSC, primary sclerosing cholangitis; TNF, tumour necrosis factor; UC, ulcerative colitis; CI, confidence interval; SD, standard deviation; GI, gastrointestinal..

3.2. Genetic variants associated with anti-TNF response in patients with IBD

Among genetic loci associated with anti-TNF response, which reached p <1E-03 in the discovery cohort, two were successfully replicated with p <1E-03 in the replicate cohort, and another four with suggestive evidence of p <0.05. The most significant genetic variant associated with refractory response to anti-TNF agents was rs116724455 [minor/risk allele C, 6% in anti-TNF non-responders and 0.3% in responders; OR_additive, 19.9 [4.57–86.7]; p = 4.79E-08], located at chromosome 1 genetic loci of TNF-superfamily [TNFSF], TNFSF4, and TNFSF18, reaching genome-wide significance level [p <5.0E-08] [Table 2]. The other successfully replicated genetic variant was rs2228416 [minor/risk allele T, 12% in anti-TNF non-responders and 3% in responders; OR_additive, 5.25 [2.33–11.8]; p = 5.24E-06], located at chr9 genetic loci of PLIN2 and HAUS6.

Table 2.

Genetic variants associated with anti-TNF response in patients with IBD and their effect size.

Genetic variants [minor/ major allele]	Discovery cohort MAF [NR vs R]	Replicate cohort MAF [NR vs R]	Combined cohort MAF [NR vs R]	OR [95% CI]	Adjusted OR [95% CI]
Position	Allelic OR [95% CI]	Allelic OR [95% CI]	Allelic OR [95% CI]	[Additive mode]
Candidate gene	p-Value	p-Value	p-Value	p-Value	p-Value
rs116724455 [C/T]	0.07 vs 0.005	0.06 vs 0.002	0.06 vs 0.003
chr01:173,191,085	13.8 [2.26–85.4]	27.3 [2.42–309]	18.6 [4.38–79.6]	19.9 [4.57–86.7]	43.1 [6.98–266]
TNFSF4, TNFSF18	p = 2.43E-04	p = 6.28E-06	p = 5.37E-08	p = 4.79E-08	p = 1.14E-04
rs2228416 [T/C]	0.11 vs 0.02	0.14 vs 0.03	0.12 vs 0.03
chr09: 19,126,281	5.97 [1.87–19.1]	5.21 [1.76–15.4]	5.33 [2.43–11.7]	5.25 [2.33–11.8]	5.42 [2.05–14.3]
PLIN2,HAUS6	p = 7.17E-04	p = 9.87E-04	p = 3.76E-06	p = 5.24E-06	p = 1.62E-03
rs762787 [T/C]	0.17 vs 0.04	0.19 vs 0.07	0.18 vs 0.06
chr03: 46,488,936	4.74 [1.92–11.7]	3.39 [1.37–8.35]	3.82 [2.03–7.17]	4.27 [2.16–8.48]	5.04 [2.23–11.4]
LTF, CCR5, CCRL2	p = 2.46E-04	p = 5.22E-03	p = 9.51E-06	p = 6.47E-06	p = 1.95E-04
rs9572250 [G/A]	0.45 vs 0.19	0.39 vs 0.21	0.43 vs 0.20
chr13: 70,302,178	3.52 [1.87–6.63]	2.39 [1.18–4.84]	2.94 [1.84–4.70]	3.00 [1.84–4.89]	3.24 [1.80–5.85]
KLHL1	p = 4.30E-05	p = 0.01	p = 2.84E-06	p = 3.19E-06	p = 7.71E-05
rs144256942 [G/A]	0.07 vs 0.003	0.06 vs 0.01	0.06 vs 0.007
chr01: 213,732,082	27.8 [2.83–273]	5.42 [1.01–28.9]	9.30 [2.77–31.2]	9.88 [2.88–33.9]	10.2 [2.26–46.4]
PROX1,RPS6KC1	p = 1.97E-05	p = 0.02	p= 1.27E-05	p = 1.14E-05	p = 4.41E-03
rs523781 [G/C]	0.20 vs 0.06	0.14 vs 0.06	0.17 vs 0.06
chr09:	3.81 [1.65-	2.73 [0.98-	3.36 [1.77-	3.97	4.90 [2.02-
77,323,239	8.80]	7.60]	6.39]	[1.95–8.07]	11.9]
RORB,TRPM6	p = 8.84E-04	p = 0.04	p = 9.79E-05	p = 5.28E-05	p = 7.81E-04

Abbreviations: A, adenine; C, cytosine; chr, chromosome; G, guanine; MAF, minor allele frequency; NR, no response; OR, odds ratio; R, response; T, thymine; TNF, tumour necrosis factor; CI, confidence interval,IBD, inflammatory bowel disease.

Another four genetic variants were found with suggestive evidence of an association with responsiveness to anti-TNF agents in IBD: 1] rs762787 [minor/risk allele T, 18% in anti-TNF non-responders and 6% in responders; OR_additive, 4.27 [2.16–8.48]; p = 6.74E-06], located at chr3 genetic loci of LTF, CCR5, and CCRL2; 2] rs9572250 [minor/risk allele G, 43% in anti-TNF non-responders and 20% in responders; OR_additive, 3.00 [1.84–4.89]; p = 3.19E-06], located at chr13 genetic loci of KLHL1; 3] rs144256942 [minor/risk allele G, 6% in anti-TNF non-responders and 0.7% in responders; OR_additive, 9.88 [2.88–33.9]; p = 1.14E-05], located at chr1 genetic loci of PROX1 and RPS6KC1; and 4] rs523781 [minor/risk allele G, 17% in anti-TNF non-responders and 6% in responders; OR_additive, 3.97 [1.95–8.07]; p = 5.28E-05], located at chr9 genetic loci of RORB and TRPM6 [Table 2]. We further examined the raw signal intensity cluster plots of the top replicated six SNPs [Supplementary Figures S1, S2, S3, S4, S5, S6, available as Supplementary data at ECCO-JCC online] and confirmed the accurate allele calling.

We assessed the model predictability [AUC] of clinical predictors [72%], genetic predictors [85%], and combined clinical and genetic predictors [89%] in the combined cohort. The AUC increased from 72% for clinical predictors only to 89% [p <1E-4] after adding the genetic predictors, and the explained anti-TNF response variance increased from 11% to 35.7% [p <1E-4] [Figure 1].

Figure 1.

Model predictability of clinical factors only, genetic factors only, and combined clinical and genetic factors.

We further constructed an anti-TNF refractory score: Score= bowel resection [Yes= 1, No= 0] × 1.10 + immunomodulatory use [Yes= 1, No= 0] × 1.78 + rs116724455 [number of risk allele] × 3.93 + rs144256942 [number of risk allele] × 1.84 + rs762787 [number of risk allele] × 1.49+ rs523781 [number of risk allele] × 1.50 + rs2228416 [number of risk allele] × 1.34 + rs9572250 [number of risk allele] × 1.13. Mean scores in anti-TNF non-responders and responders were 5.49 [1.99] and 2.65 [1.39] [p = 4.33E-23], respectively [Figure 2]. If only using clinical information without genotype data, the mean risk score was 2.45 [0.54] in anti-TNF non-responders and 1.74 [0.91] in anti-TNF responders [p<1.0E-04]. Based on the receiver operating characteristic [ROC] curve, a cut-off at 2.88 of score, this model has a sensitivity of 61%, specificity of 75%, positive predictive value of 19%, and a negative predictive value of 95%. However, if genotype data are available, with a cut-off at 5.35 of score, this model will have a sensitivity of 56%, specificity of 95%, positive predictive value of 50%, and a negative predictive value of 96%.

Figure 2.

Box plot of anti-TNF refractory risk score between responders and non-responders. TNF indicates tumour necrosis factor.

4. Discussion

Through a hypothesis-free, genetic association study in a long-term follow-up IBD cohort validated by another large independent IBD cohort, genome-wide significant signal in TNFSF4/18 loci and several other suggestive loci linked to TNF-α were confirmed to be associated with anti-TNF response in patients with IBD. We explored a model integrating clinical and genetic predictors with good predictability [AUC, 89%] of anti-TNF response in patients with IBD. A preliminary anti-TNF refractory score, including clinical predictors [bowel resection and immunomodulatory use] and genetic predictors [rs116724455, rs2228416, rs144256942, rs762787, rs523781, rs9572250], was highly differentiable between anti-TNF non-responders (5.49 [1.99]) and responders (2.65 [1.39]; p = 4.33E-23).

In our study, the most significant genetic variant associated with refractory response to anti-TNF agents was rs116724455 [minor/risk allele C, 6% in anti-TNF non-responders and 0.3% in responders; OR_additive, 19.9 [4.57–86.7]; p = 4.79E-08], located at chromosome 1 [173 191 085 bp] genetic loci of TNF-super family TNFSF4 and TNFSF18, and reached genome-wide significance level [p <5.0E-08]. Interestingly, in the recent landmark study of 163 IBD loci, TNFSF4/TNFSF18 receptors [chromosome 1, rs12103] and TNFSF18 [chromosome 1, rs9286879] were associated with the susceptibility risk of IBD.¹³ In addition to NOD2, among the IBD risk loci in the human genome, TNF-receptor super family [TNFRSF18] [directional selection, p = 8.9E-05] and its ligand, TNFSF18 [directional selection, p = 5.2E-04],¹³ were found significantly under natural selection force, mostly driven by parasites during the human-microbe co-evolution history.³⁴ Although variants in TNFSF4/TNFSF18 and its receptors were previously found associated with the IBD susceptibility risk, to the best of our knowledge our finding in TNFSF4/TNFSF18 is the first work to show the association between variant in TNFSF4 and TNFSF18 and the refractoriness to the anti-TNF agents.

We further explored the clinical significance through the association of rs116724455 genotype status with different clinical characteristics and outcomes [Supplementary Table 2, available as Supplementary data at ECCO-JCC online]. Except for the expected association with anti-TNF response (OR, 19.9 [4.57–86.6]; p= 4.79E-08), there was no significant association between rs116724455 and any other clinical characteristics, including gender, age at diagnosis, disease duration, IBD phenotype, advanced disease behaviour, bowel resection, tobacco use, family history of IBD, and EIMs.

TNFSF contains about 19 ligands, including TNF-α [TNFSF2], TNF-β [TNFSF1], OX40L [TNFSF4], glucocorticoid-induced TNFR family-related gene ligand [GITRL, or TNFSF18].³⁵ TNFSF binds to TNFRSF, a family of 30 structurally similar receptors, and their interactions provide costimulatory signals that control proliferation and differentiation of immune cells involved in regulating the pathology of autoimmune diseases. TNFSF4 gene encodes a ligand protein, also known as OX40L, which is found predominantly in activated CD4-positive regulatory T cells. In a experimental colitis mouse model, OX40L-IgG fusion protein ameliorated the inflammatory process by reducing TNF-α, interleukin [IL]-1, IL-12, and interferon [IFN]-γ and reduced T cell infiltrate in intestinal mucosa.³⁶TNFSF18 gene encodes a ligand protein, also known as glucocorticoid-induced TNFR family-related gene ligand [GITRL], which can regulate T cell activation and play a role in interactions between activated T lymphocytes and endothelial cells. Evidence that administration of a soluble form of GITR:Fc blocks the GITR/GITRL pathway, preventing the clinical, histological, and immunological signs of chemical‐induced colitis in wild‐type mice, further supports the role of the GITR/GITRL pathway in regulating both acquired and innate mucosal immune responses.³⁷ This finding may suggest that the GITR/GITRL pathway is a target for therapeutic intervention in inflammatory disorders, such as IBD.

The other successfully replicated genetic variant was rs2228416 [minor/risk allele T, 12% in anti-TNF non-responders and 3% in responders; OR_additive, 5.25 [2.33–11.8]; p = 5.24E-06], located at chromosome 9 [19 126 281 bp] genetic loci of PLIN2. PLIN2 [perilipin-2, also known as adipose differentiation-related protein or adipophilin] gene belongs to the perilipin family, which forms intracellular lipid storage droplets in adipocytes, fibroblasts, endothelial, and epithelial cells. Adipophilin involves metabolism and regulation of lipids by peroxisome proliferator-activated receptor [PPAR]-α.³⁸ Recent studies and clinical trials have reported that PPAR-γ ligands, such as rosiglitazone, can suppress inflammatory response in UC by inhibiting the activity of macrophages, cytokine production, and NF-κB transcriptional activities.^39,40

In our study, another four genetic variants were found with suggestive evidence of an association with responsiveness of anti-TNF agents in IBD: 1] rs762787(, 4.27 [2.16–8.48]; p = 6.74E-06), located at chromosome 3 [46 488 936 bp] genetic loci of LTF, CCR5, and CCRL2; 2] rs9572250 (OR, 3.00 [1.84–4.89]; p = 3.19E-06), located at chromosome 13 [70 302 178 bp] genetic loci of KLHL1; 3] rs144256942 (OR, 9.88 [2.88–33.9]; p = 1.14E-05), located at chromosome 1 [213 732 082 bp] genetic loci of PROX1 and RPS6KC1; and 4] rs523781 (OR, 3.97 [1.95–8.07]; P = 5.28E-05), located at chromosome 9 [77 323 239 bp] genetic loci of RORB and TRPM6. Several of the involved candidate genes in these loci were either recently found GWAS IBD risk alleles or involved in TNF-α-related pathways or mechanisms.

LTF gene, a previously identified GWAS IBD risk locus,⁴¹ is a member of the transferrin family of genes, and its protein product is found in the secondary granules of neutrophils. The protein is a major iron-binding protein found in milk and body secretions, with a broad spectrum of properties, including regulation of iron homeostasis, anti-inflammatory activity, regulation of cellular growth and differentiation, and protection against cancer development and metastasis. The protein has an antimicrobial effect on several Gram-negative and Gram-positive bacteria, enveloped and non-enveloped viruses, and various types of fungi and parasites. In mouse models, LTF was found to significantly lower serum TNF-α level, induced by lipopolysaccharide, and further supported the preventive activity of LTF in infection.⁴²RPS6KC1 [ribosomal protein S6 kinase δ-1] encodes a protein to bind to sphingosine kinase and phosphatidylinositol 3-phosphate, suggesting a role in a lipid cellular messenger, sphingosine-1-phophate [S1P], and its signalling pathway. S1P has been implicated in the trafficking of lymphocytes in the inflamed intestine, and subsequently intensifies inflammation by increasing proinflammatory cytokines, such as TNF-α and IL-6.⁴³ Oral small-molecule S1P receptor modulators [e.g. ozanimod], which can downregulate S1P receptors expressed on lymphocytes, are under clinical investigations for inflammatory diseases such as IBD.⁴⁴TRPM6 [transient receptor potential cation channel subfamily M member 6] gene encodes a protein crucial for magnesium ion channel domain and plays an essential role in epithelial magnesium transport and active magnesium absorption in the gut. In IBD, TRPM6 expression was decreased in the lamina propria of the inflamed intestinal mucosa,⁴⁵ but increased in peripheral leukocytes, which correlates with the recent finding of upregulation of TRPM6 expression by TNF-α stimulation.⁴⁶

Among the clinical factors associated with anti-TNF responsiveness in IBD, the landmark SONIC trial in CD and the UC SUCCESS trial in UC demonstrated the better outcome of combined anti-TNF [infliximab] with immunomodulatory [azathioprine] than monotherapy.^47,48 In our study, the history of immunomodulatory use (OR, 10.7 [1.46–79.1]; p = 5.0E-04) was found associated with a refractory response to anti-TNF agents. This likely resulted from the limitation of the retrospective study design. The detailed temporal relationship of immunomodulatory agents relative to the use of anti-TNF agents is not available in our study. The observed association of immunomodulatory agents with refractory response to anti-TNF agents might have reflected the underlying complex and severe disease behaviour.

We constructed a preliminary anti-TNF refractory score based on the identified significant clinical factors [history of bowel resection, history of immunomodulatory use] and genotype status of rs116724455, rs144256942, rs762787, rs523781, rs9572250, and rs222841. For clinicians seeing an IBD patient who had history of bowel resection and previous use of an immunomodulatory drug, but unknown genotype status of risk variants, based on the exploratory risk model [clinical predictors only, AUC 0.72] the probability of this patient being refractory to anti-TNF agents is only 18%. However, for a patient without previous history of bowel resection or immunomodulatory use, carrying one copy at rs116724455, based on the model [clinical and genetic predictors, AUC 0.89] the probability of refractoriness to anti-TNF agents becomes 26% and will be further higher to 63% if carrying one copy at rs116724455 and one copy at rs144256942. Therefore, a different treatment strategy [e.g. anti-integrin agent] in such high-risk patients may be considered.

Some limitations of our study warrant consideration. First, this is a retrospective study, and the stringent criteria we used to define anti-TNF non-response were based on unsuccessful treatment with both infliximab and adalimumab, which may not be applicable to patients who were only exposed to one anti-TNF agent. Second, the defined anti-TNF non-response was based on study questionnaire surveys because of a lack of objective markers or physician reports of non-response. It did not differentiate primary non-response from secondary non-response, which could be clinically relevant during follow-up. However, this defined surrogate outcome may be used for gene discovery and should not prevent discovery of large effect-size variants. Third, the exact time interval between the use of anti-TNF and its treatment response is not available. Among patients who have failed to respond to anti-TNF, we cannot differentiate between those who have been on treatment for years and those who have just started the treatment for a few months. Fourth, patient anti-TNF drug levels, which can play an important role in the success of treatment, were not included in our study and may underestimate the explainable variability; however, we do not think this would confound our results, because the identified genetic loci were not known to be in association with the metabolites of anti-TNF drugs or influencing the anti-TNF drug concentrations.

Our study has several strengths. To our knowledge, this is the first validated genetic association study using two independent, large, long-term follow-up IBD cohorts [discovery and replicate datasets] focused on the anti-TNF response in patients with IBD. Furthermore, the study was conducted in multiple medical centres with high volumes of IBD patients and established long-term follow-up data collection. In addition, the included studies had high quality, according to the methodological quality assessment in IBD clinical behaviour and phenotyping and genotyping data.

To conclude, we successfully identified novel genome-wide significant signals in TNFSF4/18 loci and several other suggestive loci linked to TNF-α, which may predict anti-TNF response in patients with IBD. An exploratory predictive risk model and risk score integrating clinical and genetic predictors has the potential to identify patients who may not respond to anti-TNF agents, and may be a useful personalised medicine-based tool to guide different treatment strategies. Future work to differentiate primary non-response from secondary non-response to anti-TNF in our cohort, and further validation of this model in an independent prospective cohort, is warranted.

Funding

This study was supported by the Helmsley Charitable Trust. M-H W was supported by the Mayo Clinic Intramural Grant [Department of Medicine and Center of Individualized Medicine]. KCC was supported by the T32DK007130 grant and the UL1TR000448 grant. The funding bodies had no role in the study design, the collection, analysis, and interpretation of data, or the writing of the article and the decision to submit it for publication. Mayo Clinic does not endorse specific products or services included in this article.

Conflict of Interest

None.

Author Contributions

M-H W contributed to the concept and design, analysis and interpretation of the data, drafting and critical revision of the article, and generation/collection of figures. JJF contributed to the experiments and collection of data. LER contributed to the concept and design and critical revision of the article. JAL contributed to the concept and design and critical revision of the article. SFP contributed to the concept and design and critical revision of the article. MFP contributed to the concept and design and critical revision of the article. KCC contributed to the experiments and collection of data. KM contributed to the experiments and collection of data. BDN contributed to the experiments and collection of data. RDN contributed to the concept and design and critical revision of the article. WAF contributed to the concept and design and critical revision of the article. All authors approved the final article.