Highlights
– Peripheral blood DNA methylation panels predicted response to vedolizumab and ustekinumab with validated AUCs of 0.75; adalimumab prediction failed to validate (AUC 0.25).
– Prediction accuracy declined in patients with prior anti‑TNF exposure, suggesting treatment history influences epigenetic signatures.
– Models outperformed available clinical decision support tools (CDSTs) for vedolizumab and ustekinumab and are being prospectively tested in a multicentre randomised trial.
Background and clinical need
Crohn’s disease is a chronic, relapsing inflammatory bowel disease for which biologic therapies (anti‑TNF agents, anti‑integrin agents, anti‑IL‑12/23 agents) are central to induction and maintenance of remission. Randomised trials and real‑world studies demonstrate substantial benefit, but primary non‑response (failure to achieve clinical and/or endoscopic improvement after induction) remains common and clinically important because it delays effective therapy, exposes patients to preventable morbidity, and increases costs.
Precision medicine approaches that reliably predict which patient will respond to which biologic would transform care by guiding first‑line biologic choice, avoiding futile treatment exposure, and potentially improving long‑term outcomes. DNA methylation in peripheral blood is an attractive biomarker modality: it captures stable but dynamic regulation of gene expression related to immune cell function and environmental exposures, can be assayed from routine blood draws, and may reflect systemic immune states relevant to biologic mechanism of action.
Study design and methods (EPIC‑CD)
The EPIC‑CD study (Joustra et al., Lancet Gastroenterol Hepatol 2025) is an epigenome‑wide association study using two prospectively recruited biobanks: a discovery cohort from Amsterdam University Medical Centre (n=183) and a validation cohort from the John Radcliffe Hospital, Oxford (n=90). Adults with symptomatic, endoscopically active Crohn’s disease about to start adalimumab, vedolizumab, or ustekinumab were enrolled. Key exclusions were ongoing malignancy and major concurrent inflammatory disease.
Whole peripheral blood leukocyte samples were collected before treatment. Outcomes were assessed after a median of 28 weeks (IQR 18–36) and defined as combined endoscopic improvement (≥50% reduction in the Simple Endoscopic Score for Crohn’s Disease) plus either corticosteroid‑free clinical response/remission (Harvey‑Bradshaw Index reductions) or biochemical response (CRP and faecal calprotectin reductions meeting predefined thresholds).
Epigenome‑wide DNA methylation and transcriptome‑wide gene expression analyses were performed on baseline samples. Supervised machine learning using stability‑selected gradient boosting identified methylation markers predictive of response. A post‑hoc comparison with clinical decision support tools (CDSTs) was performed in the validation cohort.
Key results
Overall, 273 participants were profiled (discovery n=183; validation n=90). In the discovery cohort the authors identified treatment‑specific panels of methylation markers associated with the combined response endpoint: 18 markers for adalimumab, 25 for vedolizumab, and 68 for ustekinumab. Classification performance in the discovery set was high: AUCs 0.86 (adalimumab), 0.87 (vedolizumab), and 0.89 (ustekinumab) — though discovery performance is vulnerable to overfitting.
External validation is the critical test. In the Oxford validation cohort the methylation panels performed as follows: adalimumab AUC 0.25 (95% CI 0.10–0.35), vedolizumab AUC 0.75 (0.65–0.85), and ustekinumab AUC 0.75 (0.65–0.87). Thus, models for vedolizumab and ustekinumab showed moderate and clinically potentially useful discrimination; the adalimumab model failed to generalise and performed worse than random classification.
When compared with published CDSTs in the validation cohort, the methylation models outperformed CDSTs for vedolizumab (methylation AUC 0.75 vs CDST 0.56) and ustekinumab (0.75 vs 0.66). Importantly, prior exposure to anti‑TNF therapy was associated with reduced methylation model accuracy for vedolizumab (AUC 0.66) and ustekinumab (AUC 0.63), indicating that treatment history modifies predictive performance.
Transcriptome analyses were performed in parallel, but the primary report emphasises the methylation classifiers and their validation performance; mechanistic links between differential methylation and expression were not the central reported outcome in the summarized data.
Interpretation and clinical implications
These findings provide proof‑of‑principle that peripheral blood DNA methylation patterns can predict response to certain biologic agents in Crohn’s disease, notably vedolizumab and ustekinumab, and may improve on currently available clinical scoring systems. If prospectively validated and implemented, such tests could inform biologic selection, particularly for biologic‑naïve patients, potentially shortening time to effective therapy and limiting exposure to ineffective treatments.
However, several caveats must be emphasised before clinical adoption:
- Validation performance was treatment specific. The failure to validate the adalimumab model suggests either a weaker or less generalisable peripheral blood methylation signal for anti‑TNF response, cohort heterogeneity between centres, or overfitting in the discovery cohort.
- Models were less accurate in patients with prior anti‑TNF exposure. This underlines the importance of accounting for treatment history when applying epigenetic biomarkers and suggests that different predictive models may be required for biologic‑experienced populations.
- Peripheral blood methylation reflects mixed leukocyte populations. Differences in cell composition, medication use, comorbidities, and technical factors (batch effects) can confound methylation signals. Robust adjustment for cellular heterogeneity and external validation across diverse populations are essential to confirm generalisability.
- Sample size and event numbers. Although the cohorts are valuable, the absolute numbers per drug in the validation set are modest (particularly for adalimumab), limiting precision of AUC estimates and subgroup analyses.
Methodological strengths and limitations
Strengths of EPIC‑CD include prospective recruitment, clinically meaningful combined endpoints that incorporate endoscopy, blinded baseline molecular profiling, and an external validation cohort. Use of stability selection with gradient boosting is a sensible approach to limit overfitting while discovering multivariable methylation signatures.
Important limitations are the moderate validation cohort size, potential cohort differences (genetic background, prior treatment algorithms, local referral patterns), and incomplete mechanistic linkage between methylation markers and functional gene expression or cell‑type activity in most reported results. The poor external performance for adalimumab demands exploration of whether the discovery panel captures cohort‑specific confounding rather than biologically reproducible signals.
Context with prior literature
Epigenetic dysregulation has been implicated in IBD pathogenesis and in modifying gene expression responses to environmental exposures. Prior work has described disease‑associated methylation patterns in intestinal mucosa and peripheral blood, providing a rationale for methylation biomarkers in prognosis and therapeutic response prediction (Ventham et al., Gut 2016). The EPIC‑CD study advances the field by demonstrating treatment‑specific predictive panels and crucial external validation for two drug classes.
Biologic agents differ markedly in mechanism — vedolizumab targets gut‑homing α4β7 integrin, while ustekinumab blocks p40 shared by IL‑12 and IL‑23 — which may explain why peripheral immune signatures are more informative for these agents than for systemic anti‑TNF responses in this dataset. Clinical trial data show heterogeneous primary response rates across biologic classes, underscoring the unmet need for predictors to guide first‑line biologic selection (Sandborn et al., NEJM 2013; Feagan et al., NEJM 2016).
Next steps and research priorities
Key priorities before clinical implementation include:
- Prospective, multicentre validation in larger, ethnically diverse cohorts with pre‑specified endpoints and prespecified handling of prior therapy effects (the ongoing multicentre randomised trial testing these models is therefore welcome).
- Standardisation of sample collection, preprocessing, methylation assays, and analysis pipelines to reduce technical variability and promote reproducibility.
- Integration with clinical variables, other omics (transcriptomics, proteomics), and microbiome data to build multimodal predictive algorithms that are robust across practice settings.
- Mechanistic studies linking methylation marks to gene expression, immune cell function, and tissue‑level inflammation to enhance biological plausibility and possibly suggest therapeutic targets.
- Health economic modelling to evaluate cost‑effectiveness of pre‑treatment methylation testing compared with current care pathways.
Conclusion
The EPIC‑CD study is an important step toward epigenetic-guided precision medicine in Crohn’s disease. Peripheral blood DNA methylation panels achieved validated discrimination for vedolizumab and ustekinumab response (AUC 0.75) and outperformed available CDSTs in the studied cohorts, particularly in biologic‑naïve patients. However, failure to validate an adalimumab model, reduced accuracy after prior anti‑TNF exposure, and modest sample size underscore the need for larger, standardised prospective validation and mechanistic work. If the ongoing multicentre trial confirms clinical utility, methylation‑based testing could become a practical tool to personalise biologic selection in Crohn’s disease.
Funding
The EPIC‑CD study was funded by The Leona M and Harry B Helmsley Charitable Trust.
References
1. Joustra VW, Li Yim AYF, Henneman P, et al.; EPIC‑CD Consortium. 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. Lancet Gastroenterol Hepatol. 2025 Sep;10(9):818–830. doi:10.1016/S2468-1253(25)00102-5.
2. Ventham NT, Kennedy NA, Adams AT, et al. Integrative epigenome‑wide analysis demonstrates that DNA methylation may mediate genetic risk in inflammatory bowel disease. Gut. 2016;65(1):1–12. doi:10.1136/gutjnl-2015-309333.
3. Sandborn WJ, Feagan BG, Rutgeerts P, et al. Vedolizumab as induction and maintenance therapy for Crohn’s disease. N Engl J Med. 2013;369:711–723.
4. Feagan BG, Sandborn WJ, Gasink C, et al. Ustekinumab as induction and maintenance therapy for Crohn’s disease. N Engl J Med. 2016;375:1946–1960.
Clinical trial registration
The EPIC‑CD methylation models for vedolizumab and ustekinumab are being tested in a multicentre randomised clinical trial (details provided in the original manuscript).
Author note
This report summarises and critically appraises the EPIC‑CD study (Joustra et al., 2025) and places it in clinical and translational context for clinicians and researchers interested in personalised therapy for Crohn’s disease.
