Expression Levels of 4 Genes in Colon Tissue Might Be Used to Predict Which Patients Will Enter Endoscopic Remission After Vedolizumab Therapy for Inflammatory Bowel Diseases.

BACKGROUND & AIMS: We aimed to identify biomarkers that might be used to predict responses of patients with inflammatory bowel diseases (IBD) to vedolizumab therapy.
METHODS: We obtained biopsies from inflamed colon of patients with IBD who began treatment with vedolizumab (n = 31) or tumor necrosis factor (TNF) antagonists (n = 20) and performed RNA-sequencing analyses. We compared gene expression patterns between patients who did and did not enter endoscopic remission (absence of ulcerations at month 6 for patients with Crohn's disease or Mayo endoscopic subscore ≤1 at week 14 for patients with ulcerative colitis) and performed pathway analysis and cell deconvolution for training (n = 20) and validation (n = 11) datasets. Colon biopsies were also analyzed by immunohistochemistry. We validated a baseline gene expression pattern associated with endoscopic remission after vedolizumab therapy using 3 independent datasets (n = 66).
RESULTS: We identified significant differences in expression levels of 44 genes between patients who entered remission after vedolizumab and those who did not; we found significant increases in leukocyte migration in colon tissues from patients who did not enter remission (P < .006). Deconvolution methods identified a significant enrichment of monocytes (P = .005), M1-macrophages (P = .05), and CD4+ T cells (P = .008) in colon tissues from patients who did not enter remission, whereas colon tissues from patients in remission had higher numbers of naïve B cells before treatment (P = .05). Baseline expression levels of PIWIL1, MAATS1, RGS13, and DCHS2 identified patients who did vs did not enter remission with 80% accuracy in the training set and 100% accuracy in validation dataset 1. We validated these findings in the 3 independent datasets by microarray, RNA sequencing and quantitative PCR analysis (P = .003). Expression levels of these 4 genes did not associate with response to anti-TNF agents. We confirmed the presence of proteins encoded by mRNAs using immunohistochemistry.
CONCLUSIONS: We identified 4 genes whose baseline expression levels in colon tissues of patients with IBD associate with endoscopic remission after vedolizumab, but not anti-TNF, treatment. We validated this signature in 4 independent datasets and also at the protein level. Studies of these genes might provide insights into the mechanisms of action of vedolizumab.

Background

We aimed to identify biomarkers that might be used to predict response of patients with inflammatory bowel diseases (IBD) to vedolizumab therapy.

Findings

In colon tissues from patients with IBD, we identified 4 genes whose baseline expression levels were associated with remission, based on endoscopic features, after vedolizumab but not anti–tumor necrosis factor treatment. We validated this signature in 4 independent datasets and also at the protein level. Studies of these genes might provide insights into the mechanisms of action of vedolizumab.

Implications for patient care

Analysis of this gene expression patterns in colon tissues of patients with IBD might be used to identify those most likely to respond to vedolizumab therapy.

The landscape of inflammatory bowel disease (IBD) treatment has extensively changed over the past decade, with the advent of anti–tumor necrosis factor (TNF) agents, antiadhesion molecules, and anti-interleukin (IL) 12/23 compounds inducing and maintaining clinical and endoscopic remission. Nevertheless, primary nonresponse and secondary loss of response compromise the efficacy of the current available therapies. Hence, novel therapies are eagerly awaited, as well as predictive biomarkers, which can improve the likelihood of successful treatment.

Biomarkers predicting response to anti-TNF therapy are slowly emerging3, 4, 5, 6, 7 but need further validation before translation into clinical practice. In contrast, vedolizumab-specific biomarkers are even more limited, with studies focusing only on prediction of clinical response., As targets in IBD are evolving from clinical to endoscopic remission, biomarker development should focus on the prediction of endoscopic remission.

Vedolizumab, a humanized monoclonal antibody targeting the α4β7 integrin, has proven to be a safe and efficacious drug to induce and maintain clinical remission in patients with Crohn’s disease (CD) and ulcerative colitis (UC). By disturbing the interaction between mucosal vascular addressin cell adhesion molecule 1 (MAdCAM-1) on the intestinal endothelial cells and α4β7 integrin, expressed on a variety of circulating leukocytes, vedolizumab is primarily a gut-focused drug. Although it has always been considered to interfere mainly with lymphocyte trafficking to the gut, a detailed characterization of its immunological mode of action recently pointed primarily toward its influence on the innate, rather than on the adaptive immune system.

To identify the most suitable patients for vedolizumab therapy, we here studied colonic transcriptomic data of IBD patients initiating vedolizumab, performed pathway analysis and deconvolution, and searched for predictive markers of vedolizumab-specific endoscopic remission.

Materials and Methods

Patient Selection

This prospective study was carried out at the University Hospitals Leuven (Leuven, Belgium). Independent validation cohorts were recruited in the same center as well as in the IBD unit of the Hospital Clínic (Barcelona, Spain). Endoscopy-derived inflamed colonic biopsies were obtained from consecutive IBD patients initiating biologic therapy (vedolizumab, adalimumab, or infliximab). All patients had endoscopically proven active disease, and they all had to be naïve for the drug that was initiated at inclusion. Patients received vedolizumab 300 mg at baseline and weeks 2–6, with subsequent administration every 8 weeks. All CD patients received an additional infusion at week 10. In case of anti-TNF therapy, patients received infliximab (CT-P13) 5 mg/kg at baseline and week 2–6, with subsequent administration every 8 weeks. Adalimumab was administered 160 mg subcutaneously at baseline and 80 mg subcutaneously at week 2, with 40 mg subcutaneously every other week thereafter. To reduce the risk of including treatment failures secondary to immunogenicity (and not drug mechanistic failure) or non–drug-related responders, all anti-TNF treated patients had to have a good drug exposure, defined as a maintenance trough level or >3.0 μg/mL for infliximab and >5.0 μg/mL for adalimumab. Due the lack of agreement on the targeted threshold for vedolizumab, if any, we did not include an exposure requirement in the definition of (non)response for vedolizumab.

All included Belgian patients had given written consent to participate in the Institutional Review Board approved IBD Biobank of University Hospitals Leuven, Belgium (B322201213950/S53684). All included Spanish patients had given written consent after approval of the study by the Ethics Committee of the University Hospital Clínic Barcelona, Spain (2012/7956).

Biopsy Collection

All biopsies were taken at the most affected site, at the edge of the ulcerative surface. Biopsies were taken during endoscopy before the start of therapy, stored in RNALater buffer (Ambion, Austin, TX) and preserved at –80°C. RNA was subsequently extracted and sequenced (see Supplementary Methods). Additional biopsies were immediately fixed in formalin for up to 5 hours and then dehydrated, cleared, and paraffin-embedded for histological examination and immunohistochemistry.

Endoscopic Outcomes

Outcome was assessed objectively through ileocolonoscopy at a fixed time point. In CD patients, endoscopic remission was evaluated after 6 months, and defined as a complete absence of ulcerations, whereas in UC it was defined as a Mayo endoscopic subscore ≤1. Due to national reimbursement criteria, all UC patients were endoscopically assessed at week 8 (adalimumab) or week 14 (infliximab and vedolizumab).

Quantitative Real-Time Polymerase Chain Reaction

Gene expression of selected markers in inflamed colonic biopsies was studied through quantitative real-time polymerase chain reaction (qPCR) analysis. Complementary DNA (cDNA) was synthesized from 0.500 μg of total RNA using the RevertAid H Minus First Strand cDNA synthesis kit (Fermentas, St. Leon-Rot, Germany). The primers for the housekeeping β-actin gene were synthesized by Sigma-Genosys (Haverhill, United Kingdom) (Supplementary Table S1) and 10-μM stock solution was used to make the reaction mixture (5-μL SybrGreen, 0.2-μM forward and reverse primer, 2-μL cDNA sample, 2.8-μLRNAse-free H2O). All samples were run in duplo. Samples were analyzed with the Lightcycler 480 (Roche, Basel, Switzerland). The following amplification program was used: 5 minutes 95°C, 45 × (10 seconds 95°C, 15 seconds 60°C, 15 seconds 72°C), 5 seconds 95°C, 1 minute 60°C, 4°C.

Supplementary Table S1

Details of the Forward and Reverse Primers Used for the Beta Actin qPCR Analysis, Including the amplicon Length, Melt Temperature, 5′-3′ Sequence, and NCBI Accession Number

Primer	Amplicon length	Temperature (°C)	Sequence (5′-3′)	Gene	Accession number	Reference
Forward	108	59.9	ACAATGTGGCCGAGGACTTT	Beta actin	NM_001101.3	Own design primer BLAST3
Reverse		59.7	TGGGGTGGCTTTTAGGATGG

qPCR, quantitative real-time polymerase chain reaction.

To determine the expression of all other genes (PIWIL1, MAATS1, DCHS2, RGS13), validated target-specific primers were used for TaqMan (Thermofisher Scientific, Massachusetts) qPCR (Supplementary Table S2). A total reaction volume of 20 μL was made: 10-μL TaqMan fast advanced master mix, 1-μL TaqMan assay (containing both primers and probe), 2-μL cDNA sample, 7-μL RNAse-free H2O. Samples were analyzed with the Applied Biosystems 7500 Fast (Applied Biosystems, Foster City, CA). The following amplification program was used: 5 minutes 95°C, 40 × (3 seconds 95°C, 30 seconds 60°C), 4°C. Samples were analyzed using the comparative (ΔΔ) Ct method with normalization to the housekeeping gene β-actin.

Supplementary Table S2

Details of Target-Specific TaqMan Primers

Gene	Target-specific primer	Company
PIWIL1	Hs01041737_m1	Thermo Fisher Scientific
MAATS1	Hs00398573_m1	Thermo Fisher Scientific
DCHS2	Hs03006670_m1	Thermo Fisher Scientific
RGS13	Hs00243182_m1	Thermo Fisher Scientific

Immunohistochemistry

To localize the corresponding proteins of the predictive panel in colonic mucosa, immunohistochemical stainings were performed on 5-μm-thick step slides prepared from paraffin-embedded endoscopy-derived inflamed colonic biopsies from IBD patients, taken before vedolizumab. Endogenous peroxidase activity was blocked in deparaffinated sections by incubating the slides for 20 minutes in a 0.3% solution of H2O2 in methanol. Epitope retrieval was performed by heating the slides for 30 minutes in Tris/EDTA buffer (pH 9) at 98°C. Specific protocols for each protein are summarized in Supplementary Table S3. All procedures were conducted automatically by the BOND MAX autostainer (Leica Microsystems Ltd, Heerbrugg, Switzerland). The BOND polymer refine Detection kit (Leica Microsystems Ltd) was used for visualization of bound primary antibody according to the manufacturer’s instructions. An IBD-experienced pathologist (G.D.H.) evaluated all stains. Microscopic images were acquired with Leica Application Suite V4.1.0. software using a Leica DFC290 HD camera (Leica Microsystems Ltd) mounted on a Leica DM2000 light-emitting diode bright field microscope.

Supplementary Table S3

Overview Primary Antibodies Immunohistochemistry

Protein	Dilution primary antibody	Primary Ab	Incubation details primary Ab
PIWIL1	1:2000	Rabbit polyclonal anti-PIWIL1 Ab – HPA018798 (Sigma-Aldrich)	30 min at RT
MAATS1	1:800	Mouse monoclonal anti-MAATS1/C3orf15 Ab – MA5-26540 (Invitrogen)	30 min at RT
DCHS2	1:400	Rabbit polyclonal anti-DCHS2 Ab – HPA064159 (Sigma-Aldrich)	30 min at RT
RGS13	1:200	Rabbit polyclonal anti-RGS13 Ab – HPA044952 (Sigma-Aldrich)	30 min at RT

Ab, antibody; RT, room temperature

Statistical Analysis

All machine learning based analyses were carried out using R version 3.5.0 (R Development Core Team, Vienna, Austria). Unlike conventional statistics, for machine learning purposes the initial vedolizumab dataset (n = 31 samples) was randomly partitioned into a training (two-thirds) and validation (one-third) set. Predictive modeling was performed using the randomGLM (RGLM) package, which shares the advantages of a random forest (excellent predictive accuracy, feature importance measures, out-of-bag estimates of accuracy) with those of a forward-selected generalized linear model (interpretability). Parameter choices were optimized according to the developers suggestions, with parameters nBags = 100, nFeaturesInBag = 5, nCandidateCovariates = 5. The identified signature was validated in several independent cohorts using ConsensusClusterPlus. qPCR expression results were used in binary logistic regression analysis, whereupon predicted probabilities were used to assess performance with receiver-operating characteristic analysis. A false discovery rate (FDR) correction was applied during differential gene expression and pathway analysis, to correct for multiple testing. A 2-tailed FDR-corrected P value <.25 was considered significant. For all other analysis, a 2-tailed nominal P value <.05 was considered significant.

Results

Patient Characteristics

Thirty-one patients with endoscopically active colonic inflammatory bowel disease (11 CD, 20 UC) with a median disease duration of 8.4 (interquartile range, 4.0–15.3) years were included before their first vedolizumab administration (Table 1). One-third (n = 10, 32.3%) received vedolizumab as first-line biological therapy. In UC, an endoscopic remission rate of 65.0% was observed after 14 weeks, whereas 54.5% of CD patients achieved endoscopic remission after 6 months. Endoscopic remitters and nonremitters did not significantly differ in baseline characteristics (P > .05). Baseline features of the validation cohorts are also reported in Table 1.

Table 1

Clinical Characteristics of the Inception Cohort, Validation Cohort 2, and Validation Cohort 4

	Inception cohort discovery + validation 1 (n = 31)	Validation cohort 2 RNA sequencing (n = 16)	Validation cohort 4 qPCR (n = 37)
Diagnosis
Ulcerative colitis	20 (64.5)	7 (43.8)	30 (81.1)
Crohn’s disease	11 (35.5)	9 (56.3)	7 (18.9)
Women	17 (54.8)	7 (43.8)	25 (67.6)
Age, y	45.3 (29.6–56.3)	44.2 (26.0–55.8)	38.2 (31.0–48.0)
Disease duration, y	8.4 (4.0–15.3)	3.7 (1.6–20.7)	6.9 (1.7–11.7)
Disease locationa
L1	0 (0)	0 (0)	0 (0)
L2	2 (18.2)	4 (44.4)	3 (42.9)
L3	9 (81.8)	5 (56.6)	4 (57.1)
L4 modifier	2 (18.2)	2 (22.2)	0 (0)
E1	3 (15.0)	1 (14.3)	8 (26.7)
E2	10 (50.0)	2 (28.6)	9 (30.0)
E3	7 (35.0)	4 (57.1)	13 (43.3)
Disease behaviora
B1	6 (54.5)	4 (44.4)	6 (85.7)
B2	3 (27.3)	2 (22.2)	0 (0.0)
B3	2 (18.2)	3 (33.3)	1 (14.3)
Perianal involvement	5 (45.5)	2 (22.2)	2 (28.6)
Steroid use during induction
Topical	10 (32.3)	5 (31.3)	15 (40.1)
Systemic	8 (25.8)	6 (37.5)	7 (16.2)
Previous anti-TNF exposure
Naïve	10 (32.3)	4 (25.0)	26 (70.3)
Exposed	21 (67.7)	12 (75.0)	13 (29.7)
C-reactive protein, mg/L	2.0 (0.9–6.7)	3.8 (1.4–7.2)	1.8 (0.7–6.0)
Endoscopic remission
Yes	19 (61.3)	5 (31.1)	14 (37.8)
No	12 (38.7)	11 (68.9)	23 (62.2)

Values are n (%) or median (interquartile range).

qPCR, quantitative real-time polymerase chain reaction; TNF, tumor necrosis factor.

Montreal classification.

Additionally, colonic biopsies from 20 actively inflamed patients (6 CD, 14 UC) initiating anti-TNF therapy, of whom 17 (90.0%) were entirely anti-TNF naïve, were collected. None of them had been exposed to vedolizumab before (Supplementary Table S4).

Supplementary Table S4

Clinical Features of the Anti-TNF–Treated Cohort

Diagnosis
Ulcerative colitis	12 (60.0)
Crohn’s disease	8 (40.0)
Therapy
Adalimumab	12 (60.0)
Infliximab	8 (40.0)
Women	12 (60.0)
Age, y	33.7 (21.6–48.0)
Disease duration, y	1.4 (0.2–4.8)
Disease locationa
L1	0 (0.0)
L2	1 (16.7)
L3	5 (83.3)
L4 modifier	1 (16.7)
E1	0 (0.0)
E2	11 (83.3)
E3	3 (21.4)
Disease behaviora
B1	3 (50.0)
B2	3 (50.0)
B3	0 (0.0)
Perianal involvement	1 (16.7)
Steroid use during induction
Topical	3 (15.0)
Systemic	7 (35.0)
Immunomodulators during induction	1 (5.0)
C-reactive protein, mg/L	5.1 (1.4–14.2)
Endoscopic remission
Yes	8 (40.0)
No	12 (60.0)

Values are n (%) or median (interquartile range).

TNF, tumor necrosis factor.

Montreal classification.

Differential Gene Expression and Deconvolution

Within the inflamed colonic biopsies before vedolizumab, 186 genes were differentially expressed between remitters and nonremitters at a nominal P < .005 level (Supplementary Table S5). Among them, only 44 genes remained significantly different after applying a conservative 0.25-FDR threshold of significance. However, just 5 reached the stringent 0.05-FDR cutoff threshold of significance: KRT23, TMEM35, DCHS2, CLDN8, and IFI6 (Figure 1). None of them was differentially expressed between CD and UC samples (P > .05). Genes previously linked to anti-TNF nonresponsiveness, were not differentially expressed between vedolizumab responders and nonresponders: OSM (P = .76), IL13RA2 (P = .54), and TREM1 (P = .46). Similarly, no significant differential expression was observed in MAdCAM-1 (P = .59), integrin α4 subunit ITGA4 (P = .97), or integrin β7 subunit ITGB7 (P = .99).

Supplementary Table S5

All Differentially Expressed Genes at the Nominal Significance P < .005 Level

Gene	Base mean	log2FoldChange	Nominal P value	FDR-Adjusted P value
KRT23	16.7180859	–2.0242052	1.09 × 10–8	.000168888
TMEM35	22.0073144	1.02387191	5.24 × 10–7	.004044778
DCHS2	14.2168032	1.56822472	2.75 × 10–6	.014128976
CLDN8	90.0608422	4.39610438	4.96 × 10–6	.019127255
IFI6	334.629287	–0.7472599	6.36 × 10–6	.019615927
APOBEC3A	70.4624261	–1.6813582	2.93 × 10–5	.075378873
PCOLCE2	7.46167868	2.22476748	3.85 × 10–5	.084915322
CXCL6	367.065443	–1.8601495	5.42 × 10–5	.096168704
P2RX2	3.98803679	2.33736209	5.61 × 10–5	.096168704
HEPHL1	7.19784773	–2.7120633	6.93 × 10–5	.106875332
GZMB	179.696681	–1.235948	.000105257	.126064653
CCL3	110.886222	–1.370798	.000106858	.126064653
DCBLD1	335.800147	–0.4339657	.000109094	.126064653
IL18RAP	88.2525652	–0.9455257	.000114389	.126064653
IL32	1239.13865	–0.516138	.000191042	.196506067
RGS13	26.713778	1.00558131	.000209322	.196608037
C16orf89	27.4676645	0.79870715	.000216627	.196608037
RASGRP4	66.4823444	–0.8669479	.000229439	.196667853
GLRA2	12.8801446	2.03544389	.000259165	.207049407
NCF2	451.245388	–0.8127664	.000276283	.207049407
PPP1R3C	50.4672113	0.77759178	.000281809	.207049407
PTGER1	5.95195593	–1.8184337	.000319712	.209795607
AFF3	85.7959497	0.72082547	.000327476	.209795607
SERPINA9	5.57881637	3.07311416	.000328424	.209795607
ITPRIPL2	719.964066	–0.3225183	.000339937	.209795607
SHC1	1492.18376	–0.2762233	.000370729	.212595281
MAATS1	16.8498269	1.35939925	.000393874	.212595281
PRF1	175.562754	–0.7783421	.000408969	.212595281
RIPK2	223.222424	–0.4889622	.000415642	.212595281
C7	206.902141	1.13251488	.000431099	.212595281
VASN	134.723245	–0.9239671	.000469609	.212595281
LILRB2	378.028889	–1.0345434	.00050092	.212595281
OCA2	5.77409323	1.79797243	.00050103	.212595281
GDF6	5.72769633	1.73964914	.000509621	.212595281
HMG20A	436.390639	0.2244065	.000515674	.212595281
ARRB2	608.61371	–0.4538602	.000518642	.212595281
MMP1	3428.22966	–2.0023607	.000527115	.212595281
IFNG	39.9970933	–1.7218582	.00053709	.212595281
C21orf88	75.0439577	2.48943313	.000537379	.212595281
CEBPB	427.693647	–0.8978266	.000657776	.24482901
CSF2	22.9106571	–1.7392829	.000659524	.24482901
ZNF587B	108.09995	0.32214911	.000668663	.24482901
PIWIL1	1.67138853	3.85503952	.000682457	.24482901
EVA1B	59.2723597	–0.9274526	.000698197	.24482901
SLC13A2	65.2167772	1.48745286	.000751791	.254142023
HSD11B1	70.3742699	–1.0479977	.000783544	.254142023
SHISA3	12.4227235	1.4228452	.000796262	.254142023
NTRK3	10.7751902	1.20335112	.00079719	.254142023
S100A3	32.1489835	–1.1512714	.000810185	.254142023
NPTX2	112.143113	–1.7394179	.000823586	.254142023
CMIP	1060.93813	–0.2499559	.000847772	.256475996
WBSCR27	13.6274888	1.30468994	.000930061	.275959815
APOL1	3402.48609	–0.8000499	.000968958	.282076494
LILRA5	142.345939	–1.2109718	.001013086	.289461169
TENM2	4.01279765	2.20373098	.001039446	.29159288
TBC1D1	837.906313	–0.2292631	.00112296	.299758073
KCNH2	74.3991527	0.64204441	.001154201	.299758073
HSPA12B	77.7690355	–0.7681986	.001154923	.299758073
PFKFB4	177.110178	–0.634147	.001188816	.299758073
SOX18	75.6968692	–1.0378603	.001217785	.299758073
FCGR2A	590.678255	–1.0078412	.001242733	.299758073
OSCAR	51.8112129	–0.6944374	.00124279	.299758073
RP11-812E19.9	21.3315831	–1.08004	.001243239	.299758073
MMP3	4282.89856	–1.863898	.001298899	.299758073
ADNP2	329.918377	0.19372723	.001305613	.299758073
PDPN	555.884207	–0.9337707	.001322834	.299758073
PTAFR	604.274834	–0.6096574	.001323439	.299758073
PLAU	1885.1355	–1.0021618	.001329836	.299758073
MMP14	2445.67364	–0.5253513	.001344761	.299758073
HCAR2	118.63972	–1.6680141	.001360294	.299758073
FFAR2	158.960936	–1.2633905	.001379404	.299758073
SLC43A2	499.956189	–0.4968154	.001422941	.301417508
GNAI1	286.287253	–0.3617917	.001436971	.301417508
GBP5	1201.6973	–1.0456035	.001458182	.301417508
KRTAP13-2	7.31385049	4.80719515	.001478671	.301417508
ZNF525	93.3827561	0.44302787	.001579396	.301417508
CLGN	10.7331527	1.03788853	.001585583	.301417508
WARS	7144.39929	–0.8301275	.001626954	.301417508
PXDN	1131.81879	–0.6205878	.001628192	.301417508
NRCAM	80.9563399	–1.1639303	.001639771	.301417508
TBX2	211.98147	–0.5767591	.00165562	.301417508
CCR1	328.757107	–0.7665532	.001700304	.301417508
GPRASP1	96.6515105	0.43756646	.001703022	.301417508
TIMM10	242.756012	0.41371367	.001710192	.301417508
FCN3	84.2750831	–1.407743	.001752829	.301417508
DRAXIN	3.8240981	–1.4868304	.001762212	.301417508
RAB31	1034.09613	–0.5151868	.001763017	.301417508
IL7R	1452.38809	–0.626916	.001778229	.301417508
FAM26F	223.730022	–0.9277878	.00179398	.301417508
CA1	2884.81851	2.07665726	.001794694	.301417508
PRELP	149.260382	1.06887891	.001826146	.301417508
RGS3	603.119819	–0.313215	.001894436	.301417508
GSDMC	13.6827704	–1.3250678	.001897953	.301417508
TYMP	1464.76898	–0.6961674	.001937775	.301417508
MYO1F	606.425133	–0.5323932	.001947213	.301417508
EDN1	151.13679	–0.4851836	.00195317	.301417508
MT2A	754.065	–0.8008857	.001957318	.301417508
CD300C	38.6480185	–0.6878881	.001965729	.301417508
CNTN1	34.8207164	1.20250742	.001979141	.301417508
SLC9A9	151.514629	0.39011774	.00198252	.301417508
APOL2	1380.80307	–0.618163	.001991339	.301417508
GLT1D1	30.5540448	–1.3354594	.001992649	.301417508
KIAA2022	8.39545959	1.11065981	.002032918	.303055283
KANK4	11.9996694	1.24284523	.002050125	.303055283
EMR2	368.717865	–0.8265429	.002062402	.303055283
CHMP4C	264.861881	0.5013719	.00211717	.307508524
CXXC4	27.2122769	0.52250071	.002132569	.307508524
PDGFB	178.670281	–0.6208979	.002153947	.307715287
PRKCDBP	113.367496	–0.7947516	.002195238	.310736989
HCAR3	166.644571	–1.7971432	.002252748	.312805681
MEFV	69.9932486	–1.0795521	.002255787	.312805681
ITIH1	3.76511485	1.80120388	.002270674	.312805681
ZNF180	96.8109938	0.28440763	.00232086	.31688978
ACSL1	769.356769	–0.6292047	.002375655	.321526177
SERPINH1	1448.64285	–0.4357825	.002401686	.322222743
KCNJ15	74.2169719	–1.3884412	.00247467	.329152443
FPR2	152.266592	–1.5619318	.002610446	.343491034
PML	856.765038	–0.4504262	.002626997	.343491034
CRIP3	9.85846253	1.13141354	.002653799	.344079565
PAK3	14.2995118	0.87116381	.002741388	.352095927
S100A8	516.102614	–1.4489693	.002761268	.352095927
AGTRAP	303.595404	–0.3661557	.002851262	.357943211
NRBF2	254.204633	–0.2179738	.00285909	.357943211
LCN8	2.66174082	4.5263242	.002886973	.357943211
LCP2	700.802206	–0.5292133	.002903855	.357943211
CD97	1624.92009	–0.2937652	.002923122	.357943211
RNF149	966.868431	–0.3855569	.003016884	.359954883
FAM20C	554.457254	–0.7291105	.003027376	.359954883
IL6	137.460364	–1.8728125	.003039422	.359954883
MAMDC2	77.1436418	1.077587	.00304496	.359954883
FRMD5	42.4566262	–0.5517891	.003056199	.359954883
PRSS23	1269.57438	–0.518664	.003105654	.36300863
ADAMTS2	404.668272	–0.8236445	.003170801	.363452743
CXCR2	198.069766	–1.5049112	.003175255	.363452743
CGNL1	107.06744	0.55456682	.003210693	.363452743
SLC26A7	7.97938116	1.15536571	.003215226	.363452743
MAPK4	8.79019065	1.59986179	.003274881	.363452743
GPR68	145.288283	–0.6319797	.003320694	.363452743
CCDC85B	70.387112	–1.0722359	.003334462	.363452743
TFB2M	208.454153	0.30508172	.003341185	.363452743
FPR3	538.077495	–0.4912119	.003355985	.363452743
IL10	27.9763111	–0.748586	.00336366	.363452743
GZMH	45.6742582	–0.9784585	.003368575	.363452743
TWIST1	22.0625386	–1.1686077	.003429279	.367432956
KYNU	313.530786	–0.8871439	.003465275	.368729158
P2RY6	118.969069	–0.6825427	.003508725	.370795351
DUOXA1	41.7844729	1.16498649	.003605027	.375315993
THNSL1	71.4019298	0.48108439	.003614439	.375315993
GAS1	52.0746488	–1.3020567	.003624479	.375315993
AJAP1	22.6129557	–0.9033464	.003684437	.376071973
TARBP1	393.023524	0.27379753	.003688553	.376071973
ECEL1	6.27832952	–1.744911	.003711289	.376071973
TRPM5	25.125889	0.86827164	.003729277	.376071973
TAP1	4573.54696	–0.4324256	.003804006	.381116896
XIRP1	5.95863417	–1.9154595	.003852525	.381611932
SPON2	436.782172	–0.6820999	.003868974	.381611932
RND3	742.304639	–0.3040314	.003883147	.381611932
PI15	236.012309	–1.6421623	.003944315	.383224026
GBP1	2082.25661	–0.688434	.003961081	.383224026
CDC25B	1152.1294	–0.4573937	.003974065	.383224026
COL19A1	14.0999984	1.16153197	.004058959	.383658079
UCN2	15.8223209	–1.7066439	.004067746	.383658079
KIAA1199	892.903884	–1.4640356	.004114575	.383658079
FAM65C	246.664958	–0.805228	.004122964	.383658079
TNFAIP6	103.975943	–1.2385583	.004124454	.383658079
SPI1	442.202474	–0.6064343	.004127762	.383658079
IL22	12.1184065	–1.5552835	.004226837	.39051419
LOX	233.733809	–0.500771	.00427652	.392752577
CLEC4E	72.0366822	–1.0876011	.004314872	.392778101
HSD3B2	9.88903122	2.61567965	.004327713	.392778101
HAL	17.0775854	–1.1779315	.004384927	.395643489
GAPT	86.1978148	0.5772995	.004463871	.396779869
SNAP29	538.224499	–0.2050471	.004487674	.396779869
SEC22A	156.855161	0.21470733	.004515121	.396779869
SPHK1	149.208726	–0.8540513	.004534048	.396779869
FCGR3A	666.452688	–0.8194133	.004563509	.396779869
CENPF	452.425366	0.36369156	.004563724	.396779869
KLF10	769.779455	–0.3019486	.004608633	.396779869
TWIST2	6.12323293	–1.2439755	.004640758	.396779869
CYP27C1	11.7654784	0.87578612	.004652416	.396779869
TMEM132A	295.03094	–0.7747564	.004654686	.396779869
CEBPD	375.693399	–0.8544119	.004682067	.396920927
TNFAIP3	1311.60335	–0.5350924	.004740862	.398333069
TMEM255A	37.6285225	0.84079489	.004750359	.398333069
FHL1	1126.47582	0.52699071	.004880715	.407051615
MNDA	354.070638	–0.9214579	.004912218	.40747638

Figure 1

Top 5 differentially expressed genes. Visual representation of the top differentially expressed genes in mucosal biopsies of patients responding and not responding to vedolizumab. FDR, false discovery rate–corrected P value; logFC, log fold change. Top 5 differentially expressed genes: (A) KRT23, (B) TMEM35, (C) DCHS2, (D) CLDN8, and (E) IFI6.

Pathway analysis on the 186 differentially expressed genes using ingenuity pathway analysis revealed rather unspecific top canonical pathways (granulocyte adhesion and diapedesis [P = 9.4 × 10–5] and role of cytokines in mediating communication between immune cells [P = 1.1 × 10–3]). Similarly, a more focused gene enrichment analysis using gene set enrichment analysis, looking at Gene Ontology (GO) gene sets covering leukocyte migration and cell adhesion confirmed the GO leukocyte migration gene set, among many other trafficking and adhesion gene sets, indeed significantly enriched in nonremitters (P = .006) (Supplementary Table S6, Supplementary Figure S1). Predicted upstream regulators in vedolizumab nonremitters included TNF (P = 2.2 × 10–10), nuclear factor kappa B (P = 4.6 × 10–9), and interleukin 1β (P = 1.8 × 10–8). Deconvolution methods showed a significant enrichment of effector memory CD4+T cells (P = .008), monocytes (P = .005), M1 macrophages (P = .05), and regulatory T cells (P = .05) in nonremitters before vedolizumab initiation. In contrast, naïve B cells were significantly enriched in colonic biopsies of remitters (P = .03) (Figure 2).

Supplementary Table S6

Gene Set Enrichment Analysis Results Focused on the Leukocyte Migration and Cell Adhesion Gene Ontology Gene Sets, Derived From the MSigDB

GO gene set	Nominal P value	FDR corrected P value
Leukocyte migration	.006	.017
Leukocyte adhesion to vascular endothelial cells	.032	.090
Cellular extravasation	.038	.060
Leukocyte cell adhesion	.089	.098
Integrin-mediated signaling pathway	.117	.145
Integrin binding	.171	.166
Cell adhesion molecule binding	.201	.221
Cell substrate adhesion	.232	.208

All gene sets are enriched in the nonresponder group.

FDR, false discovery rate; GO, Gene Ontology; TNF, tumor necrosis factor.

Supplementary Figure S1

Gene set enrichment analysis enrichment of the Gene Ontology (GO) leukocyte migration gene set in the colonic transcriptomic dataset. The bar-code plot indicates the position of the genes on the expression data rank, sorted by its association with vedolizumab-induced endoscopic remission (P < .001).

Figure 2

Cellular deconvolution. Visual representation of the enrichment scores for the individual cells types identified being differentially represented between vedolizumab (A) nonremitters and (B) remitters, according to deconvolution techniques on the baseline transcriptome. T regs, regulatory T cells; T em, effector memory cells.

A 4-Gene Based Model Predicting Endoscopic Outcome to Vedolizumab Therapy

The initial dataset containing 31 inflamed colonic IBD biopsies, was randomly split into discovery (n = 20) and validation (n = 11) sets. Within the dataset of all 44 differentially expressed genes (at the 0.25-FDR level), we identified a 4-gene signature predicting endoscopic remission to vedolizumab using randomized general linear regression. A model containing RGS13, DCHS2, MAATS1, and PIWIL1 expression could accurately (accuracy 80.0%) predict endoscopic remission in the discovery cohort. Similarly, the same model could accurately differentiate vedolizumab remitters from nonremitters in validation cohort 1 (3 nonremitters, 8 remitters) (Table 2). Importantly, RGS13, DCHS2, MAATS1, and PIWIL1 expression was not significantly different between anti-TNF naïve and anti-TNF–exposed patients (Ps = .96, .96, .99, and .98, respectively) (Supplementary Table S7).

Table 2

Accuracy of the 4-Gene Signature in Vedolizumab and Anti-TNF–Treated Patients

	Discovery dataset RNA-seq (n = 20; 9 NR, 11 R)	Validation dataset 1 RNA-seq (n = 11; 3 NR, 8 R)	Validation dataset 2 RNA-seq (n = 16; 11 NR, 5 R)	Validation dataset 3 Microarray (n = 13; 9 NR, 4 R)	Anti-TNF dataset RNA-seq (n = 20; NR 12, R 8)
Accuracy, %	80.0	100.0	81.3	76.9	55.0
Sensitivity, %	81.8	100.0	66.7	100.0	75.0
Specificity, %	77.8	100.0	90.0	70.0	41.7
Positive predictive value, %	81.8	100.0	80.0	50.0	46.2
Negative predictive value, %	77.8	100.0	81.8	100.0	71.4
Positive likelihood ratio	3.7	∞	6.67	3.3	1.3
Negative likelihood ratio	0.2	0	0.3	0	0.6

NR, nonresponder; R, responder; RNA-seq, RNA sequencing; TNF, tumor necrosis factor.

Supplementary Table S7

Differential Gene Expression Between anti-TNF–Naïve and Anti-TNF–Exposed Vedolizumab Patients

Gene	Nominal P value	FDR-Corrected P value	Log2 fold change
RGS13	.13	.96	0.32
DCHS2	.13	.96	0.40
MAATS1	.62	.99	0.22
PIWIL1	.18	.98	1.77

FDR, false discovery rate; TNF, tumor necrosis factor.

Subsequently, we recruited another 16 consecutive patients initiating vedolizumab (validation cohort 2) (Table 1), in whom we could accurately predict remission (accuracy 81.3%) through unsupervised consensus clustering based on the expression of the 4 identified genes (Table 2). Combining validation cohort 1 and 2 together (14 nonresponders, 13 responders) ultimately resulted in an 88.9% accuracy (positive likelihood ratio [LR+] 11.1 and negative likelihood ratio [LR–] 0.15). All 4 genes were significantly upregulated in remitters, with PIWIL1 not at all expressed in any of the nonremitters (Supplementary Figure S2). In contrast, the 4-gene signature was not accurate to predict response in inflamed ileal biopsies (accuracy 50.0%). Furthermore, the expression of these genes did not correlate with mucosal TNF/IL6 or C-reactive protein, suggesting that they do simply not reflect the inflammatory burden.

Supplementary Figure S2

Differential expression of the 4 genes in the predictive panel. Visual representation of the differential gene expression in mucosal biopsies of patients responding and not responding to vedolizumab therapy of the 4 genes included in the predictive panel. FDR P value, false discovery rate corrected P value; logFC, log fold change. Differential expression of the 4 genes in the predictive panel: (A) PIWIL1, (B) MAATS1, (C) RGS13, and (D) DCHS2.

Validation of the 4-Gene Model in a Publicly Available Dataset From the GEMINI Long-Term Extension Program

Publicly available transcriptomic data in vedolizumab treated patients are limited. Therefore, we could validate our signature only in a small independent cohort of 13 UC patients (validation cohort 3), treated during the GEMINI long-term extension program (GSE73661). Those patients received vedolizumab according to the standard dosing (weeks 0, 2, and 6), and were endoscopically assessed at week 14. In this historic cohort, the 4-gene panel could accurately identify those patients who would not benefit from vedolizumab therapy (LR– 0.0, overall accuracy 76.9%) (Table 2).

The combination of validation cohorts 1, 2, and 3 did confirm a predictive accuracy >80.0% in both CD and UC patients separately.

Validation of the 4-Gene Model in an Independent Cohort Using qPCR

This 4-gene panel was tested in an additional Belgian-Spanish cohort using qPCR (30 UC, 7 CD) (Table 1), accurately differentiating remitters from nonremitters with an area under the curve (AUC) of 78.6% (95% confidence interval, 63.8–93.3%; P = .003) (Figure 3). In contrast, the predictive accuracy of the individual genes was clearly lower: PIWIL1 AUC 69.6% (P = .05), MAATS1 AUC 61.2% (P = .20), RGS13 AUC 49.0% (P = .69), and DCHS2 AUC 50.4% (P = .80).

Figure 3

Receiver-operating characteristic statistics predicting vedolizumab-induced endoscopic remission based on the colonic 4-gene predictive panel in an independent Belgian-Spanish validation cohort. AUC, area under the curve.

A Vedolizumab-Specific Signature

To confirm the vedolizumab specificity of this panel, its predictive accuracy was tested in a cohort of 20 patients initiating anti-TNF therapy. In contrast to vedolizumab, this signature could not predict endoscopic outcome in anti-TNF–treated patients (accuracy 55.0%, LR+ 1.3, LR– 0.6) (Table 2).

PIWIL1 expression could not at all be observed in regenerating epithelium (Supplementary Figure S3), whereas it was clearly expressed in goblet cells and to some extent in stromal cells in inflamed tissue (Figure 4A). In contrast, MAATS1 was predominantly identified in endothelial cells and only weakly in epithelium and smooth muscle cells (Figure 4B). Likewise, DCHS2 was found in endothelial cells (Figure 4C). Finally, RGS13 was expressed solely in the epithelial barrier, cytoplasmic just above the cell nucleus (Figure 4D).

Supplementary Figure S3

Immunohistochemical PIWIL1 staining in regenerating colonic epithelium (original magnification ×50).

Figure 4

(A) Immunohistochemical PIWIL1 staining in inflamed inflammatory bowel disease (IBD) colon (original magnification [OM] ×100). (B) Immunohistochemical MAATS1 staining in inflamed IBD colon (OM ×100). (C) Immunohistochemical DCHS2 staining in inflamed IBD colon (OM ×200). (D) Immunohistochemical RGS13 staining in inflamed IBD colon (OM ×200).

Discussion

Despite the therapeutic success of emerging drugs in IBD,, endoscopic remission rates are still not exceeding 30%. Besides a better patient selection and individualized dosing schemes using population pharmacokinetic-pharmacodynamic modeling, therapy outcomes could be further improved using predictive biomarkers. In this study, we identified and validated a 4-gene colonic expression panel predicting endoscopic success of vedolizumab therapy specifically.

Very little is known about the role of the 4 identified genes, PIWIL1, MAATS1, RGS13, and DCHS2 in IBD, and even in normal colonic mucosa to a larger extent. Based on our results, they do not reflect the mucosal inflammatory burden. Piwi-like protein 1 (PIWIL1) encodes a member of the PIWI subfamily of Argonaute proteins. PIWI proteins and PIWI-interacting RNAs participate in many vital biological processes, including cell proliferation, migration, survival, and inflammation., Hence, their involvement in wound healing and tissue regeneration does not come as a surprise. Although its highest expression is observed in germline tissue, PIWIL1 has been reported along the gastrointestinal tract. Existing studies mainly focused on the aberrant expression of PIWIL1 in tumors, but the biological role of PIWIL1 in IBD has never been elucidated. As PIWIL1 is upregulated in vedolizumab remitters, it may suggest that those patients have an a priori higher likelihood of stem cell renewal/survival, as compared with nonresponders. PIWIL1 immunohistochemistry on the other hand pointed toward the contribution of goblet cells, which are fully differentiated and hence not expected to represent a more proliferative state. Whether PIWIL1 affects goblet cell function is currently unknown, and how this is linked to vedolizumab efficacy in particular cannot be answered based on the current study.

In contrast, MAATS1 (or C3orf15) and DCHS2 (or Cadherin J) were mainly found on endothelial cells, which could suggest that both may interfere with diapedesis and cell migration, key processes in the mode of action of vedolizumab. Overall, MAATS1 is predominantly expressed in the fallopian tube and testis, but expression along the gastrointestinal tract has been reported. However, MAATS1 function is entirely unknown so far. In contrast, DCHS2 is implicated in cell adhesion, considered an unconventional cadherin, and mainly expressed in the reproductive system, the gastrointestinal tract and the brain.,

Finally, RGS13 was mainly observed in our staining in the epithelial barrier. Apart from its abundant expression in innate and adaptive immune cells, it is indeed expressed throughout the digestive system. Interestingly, RGS13 expression impacts CD4+ T cell migration through the RGS13-induced unresponsiveness to CXCL12, despite high levels of its receptor CXCR4 on T cells. As CXCL12 and CXCR4 are upregulated and constitutively expressed by intraepithelial cells (IECs) in patients with active IBD, a positive feedback loop has been suggested: increased expression and secretion of CXCL12 by IECs result in an accumulation of CXCR4+ monocytes and T cells,, which on their turn contribute to additional CXCL12 expression by IECs. But, increased RGS13 expression results in impaired CXCL12 responsiveness, implying less leukocyte trafficking. Additionally, CXCL12 itself improves the adhesion of α4β7+ cells to MadCAM-1 by increasing the α4β7 affinity, without affecting the subcellular distribution of α4β7. Whether this also affects vedolizumab efficacy remains unknown.

The reduced a priori leukocyte trafficking in vedolizumab endoscopic remitters, for instance also reflected by an increased RGS13 expression in the 4-gene model, was also observed in our unsupervised transcriptome-wide analysis. This raises the question whether many more escape mechanisms exist to maintain leukocyte trafficking and subsequent intestinal inflammation in nonremitters, regardless of α4β7 blocking. Using deconvolution techniques, we identified an enrichment of proinflammatory M1 macrophages (M1ϕ) in nonresponders, before vedolizumab. In contrast to nonclassical monocytes essential for intestinal wound healing mediated by M2ϕ (which are blocked by vedolizumab therapy), classical monocytes can still migrate via the αLβ2-ICAM1 pathway, differentiate in proinflammatory M1ϕ and maintain intestinal inflammation. Vedolizumab may also affect the innate immune system, as described by Zeissig et al. They demonstrated a switch from an M1ϕ to a M2ϕ environment, but only in vedolizumab clinical responders. Our data now demonstrate that endoscopic nonremitters have an a priori abundance of M1ϕ already, together with an increased proportion of monocytes and effector memory CD4+T cells as compared with remitters. As vedolizumab is not able to reduce the abundance of effector memory CD4+T cells, the additional abundance of regulatory T cells in nonresponders is not able to dampen the proinflammatory environment, despite vedolizumab therapy.

Finally, we observed a significant baseline enrichment of naïve B cells in vedolizumab endoscopic remitters. The data by Zeissig et al also pointed toward the B cell compartment, as B cell receptor signaling was significant downregulated upon vedolizumab exposure. However, the role of the complex B cell biology in IBD pathogenesis is very poorly understood, and mainly disregarded after the failure of the anti-CD20 rituximab in randomized trial in UC, which obviously acts differently than vedolizumab. Why we observe a significant enrichment of naïve B cells in vedolizumab endoscopic remitters cannot be fully answered based on our findings, and raises the question whether a vedolizumab induced depletion of this cell population is key to its therapeutic success. Indeed, recent data on a small cohort of HIV-infected IBD patients demonstrated that vedolizumab therapy importantly reduced naïve B cells in intestinal mucosa, preventing subsequent priming by dendritic cells, who are surveying the mucosal barrier for invading pathogens.

Although we demonstrated the vedolizumab specificity of this 4-gene panel as compared with anti-TNF treated patients, the predictive accuracy in ustekinumab- or tofacitinib-treated patients is currently unknown. Furthermore, the rather limited sample size in this pilot project warrants caution. However, the validation in several independent, heterogeneous datasets suggests its true clinical and biological relevance. Finally, the enrichment scores derived through xCell deconvolution techniques raise novel hypotheses, but the need further microscopic or flow cytometric validation as the enrichment scores cannot be interpreted as proportions.

In conclusion, we identified and validated a 4-gene vedolizumab specific signature predicting therapeutic success in IBD, highlighting novel pathways previously unrecognized in vedolizumab efficacy. Additionally, our transcriptome wide unbiased analysis of inflamed colonic biopsies before vedolizumab therapy suggested an a priori increased leukocyte trafficking in endoscopic nonremitters, and provided novel insights in the vedolizumab mode of action, including the involvement of B cell compartment in vedolizumab response.