Development and validation of peripheral blood DNA methylation signatures to predict response to biological therapy in adults with Crohn's disease (EPIC-CD): an epigenome-wide association study

Abstract

Background: Biological therapeutics are widely used in Crohn's disease, with evidence of efficacy from randomised trials and real-world experience. Primary non-response is a common, poorly understood problem. We aimed to assess blood methylation as a predictor of response to adalimumab, vedolizumab, or ustekinumab in patients with Crohn's disease.

Methods: This epigenome-wide association study used data from two ongoing biobanks (one from the Amsterdam University Medical Centre, University of Amsterdam, Amsterdam, Netherlands [discovery cohort] and the other from the John Radcliffe Hospital, Oxford, UK [validation cohort]) that recruited patients between Oct 1, 2009, and June 17, 2022. Adult participants (age ≥18 years) with active symptomatic and endoscopic Crohn's disease who were scheduled to start adalimumab, vedolizumab, or ustekinumab treatment were included. Patients with ongoing malignancy or serious concomitant inflammatory diseases were excluded. Treatment response was assessed after a median of 28 weeks of treatment (IQR 18-36). Response was defined as a combination of endoscopic criteria (50% or more reduction in the Simple Endoscopic Score for Crohn's Disease) with either clinical or biochemical criteria (corticosteroid-free clinical response: ≥3 point decrease in Harvey-Bradshaw Index [HBI] score or remission [HBI ≤4] and no systemic steroids at follow up; biochemical response: C-reactive protein reduction ≥50% or ≤5 mg/L and faecal calprotectin reduction ≥50% or ≤250 μg/g) compared with baseline. Epigenome-wide DNA methylation and transcriptome-wide gene expression analyses were done on whole peripheral blood leukocyte samples that were collected before the start of treatment. To identify baseline DNA methylation markers associated with response or non-response to treatment, we performed supervised machine learning through stability selected gradient boosting. In a post-hoc analysis, we compared our DNA methylation-based prediction model with clinical decision support tools (CDSTs).

Findings: We profiled the peripheral blood DNA methylome of 273 adults with Crohn's disease scheduled to start adalimumab, vedolizumab, or ustekinumab in the discovery (Amsterdam, n=183; 108 [59·0%] female and 75 [41·0%] male) and the validation cohort (Oxford, n=90; 46 [51·1%] female and 44 [48·9%] male). In the discovery cohort, we defined a panel of DNA methylation biomarkers that were associated with combined endoscopic and clinical or biochemical response to adalimumab (18 markers), vedolizumab (25 markers), or ustekinumab (68 markers), with an area under the curve (AUC) of 0·86 (95% CI 0·58-0·97) for adalimumab, 0·87 (0·67-0·98) for vedolizumab, and 0·89 (0·76-1·00) for ustekinumab. Validation in the Oxford cohort yielded an AUC of 0·25 (0·10-0·35) for adalimumab, 0·75 (0·65-0·85) for vedolizumab, and 0·75 (0·65-0·87) for ustekinumab. In comparison, implementing the CDSTs in the validation cohort yielded an AUC of 0·56 (0·44-0·68) for vedolizumab and 0·66 (0·54-0·77) for ustekinumab. Previous anti-TNF exposure was associated with a reduction in accuracy of the methylation models for vedolizumab (0·66 [0·55-0·73]) and ustekinumab (0·63 [0·52-0·70]) when analysed in the validation cohort.

Interpretation: Our findings provide evidence for the potential use of DNA methylation as a modality for personalised medicine for Crohn's disease by predicting response to vedolizumab and ustekinumab. The models were more accurate in biologically naive patients and outperform available vedolizumab and ustekinumab CDSTs. We were unable to predict response to adalimumab. The vedolizumab and ustekinumab prediction models are currently being tested in a multicentre randomised clinical trial.

Funding: The Leona M and Harry B Helmsley Charitable Trust.