OP0257 A PERIPHERAL BLOOD CHROMOSOME CONFORMATION SIGNATURE CAN PREDICT FLARE FOLLOWING DMARD CESSATION IN RHEUMATOID ARTHRITIS (2024)

Abstract

Background: In rheumatoid arthritis (RA), achievement of drug-free remission is limited by significant likelihood of flare (~50%) following DMARD withdrawal[1]. BIO-FLARE (ISRCTN16371380) is a prospective longitudinal study involving complete DMARD cessation and offers the possibility for biomarker development for prediction of flare. Chromosome conformation signatures, based on chromatin loops and reflecting 3D genomic architecture, have shown potential as epigenomic biomarkers for prediction and prognosis, including response to methotrexate in RA[2].

Objectives: To develop a baseline peripheral blood chromosome conformation signature for classification of future flare versus remission in RA prior to DMARD cessation, and to compare performance as a predictive biomarker against i) clinical parameters alone and ii) a combined clinical and epigenomic model.

Methods: 107 participants were selected from the BIO-FLARE study, in which RA patients in remission (DAS28-CRP <2.4) on conventional synthetic DMARDs stopped all therapy and were followed up for 24 weeks[3]. The primary outcome was flare occurrence. This binary outcome creates 2 natural comparator groups at baseline, namely future flare (F) versus maintenance of remission (R). Participants were split into discovery (n=12; 6F, 6R), training (n=50; 25F, 25R) and test/validation (n=45; 24F, 21R) sets. Chromosome conformation capture (3C) library prep was carried out on baseline peripheral blood mononuclear cells, and chromosome conformation compared between groups. For the discovery set, a custom microarray comprising ~10⁶ probes (EpiSwitch Explorer, Oxford Biodynamics) was used for genome-wide screening to identify stable differentially abundant chromatin loops as candidates for inclusion in the final signature. Top ranked loops by fold change and significance were selected for translation to qPCR, with the aim of generating a robust, clinically applicable test. qPCR takeoff point values from the training set were used for feature reduction/ signature refinement (penalised logistic regression), with final signature loop data used to train the predictive model (XGBoost algorithm). Predictive performance was internally validated using hitherto unseen test set data. To determine added value over routine data, performance was compared against predictive models trained and validated using the same populations and algorithms with i) clinical parameters alone, and ii) combined clinical plus chromatin loop data.

Results: Baseline clinical parameters are shown in Table 1. 55/107 participants (51.4%) flared, while 52/107 (48.6%) remained in remission. In the discovery set, 625 differentially abundant stable loops were identified (fold change >1.1 and p-value <0.05). Of 22 candidates translated to qPCR, the top 10 following feature reduction were used to train the predictive model. Using the test set, the final 10-loop epigenomic model correctly called 21/24 flare and 17/21 remission participants, with classification accuracy (for flare) 84.2%, positive predictive value (PPV) 84.0%, and negative predictive value (NPV) 85.0%. A model using clinical parameters alone had accuracy 78.5%, PPV 93.8%, NPV 53.3%, while a combined clinical/ epigenomic model had accuracy 85.3%, PPV 93.8%, NPV 73.3% (see Table 2). The epigenomic model was optimal for guiding DMARD withdrawal decisions, as high NPV minimises risk of flare, potentially reducing flares per number of DMARD withdrawals from 1 in 2 (no test) to 1 in 7.

Conclusion: A peripheral blood 10-loop chromosome conformation signature shows promise as an epigenomic biomarker for predicting flare following DMARD withdrawal. For the specific clinical context, the epigenomic model appears to offer enhanced predictive performance compared with clinical data, suggesting a simple blood test might be feasible. Such an approach could enable stratified decision making, to reduce unnecessary treatment exposure while minimising risk of flare.

Acknowledgements: NIL.

Disclosure of Interests: Andrew Melville MRC GSK EMINENT PhD Studentship, Tarun Naithani Employee of Oxford Biodynamics PLC, Amy E Anderson: None declared, Fiona Rayner: None declared, Sean Kerrigan Lectures for Vifor, Andrew McGucken: None declared, Bernard Dyke: None declared, Shaun Hiu: None declared, Jonathan Prichard: None declared, Jayne Green Employee of Oxford Biodynamics, Ryan Powell Employee of Oxford Biodynamics PLC, Catharien Hilkens Research funding from GSK, Christopher D Buckley: None declared, Iain B. Mc Innes Honoraria or research support from Abbvie, Janssen, Novartis, Eli Lilly, Astra Zeneca, GSK, BMS, Moonlake, Evelo, Causeway THerapeutics, Cabaletta, Roche, Pfizer and Compugen., Wan-Fai Ng: None declared, Andrew Filer: None declared, Karim Raza Research grant support from Bristol Myers Squibb and personal fees for lectures/ consultancy from Abbvie and Sanofi, Arthur Pratt: None declared, Kenneth F Baker Consulting fees from Modern Biosciences, Research support from Genentech and clinical improvement funding from Pfizer, Ewan Hunter Employee of Oxford Biodynamics PLC, John Isaacs Speaker fees from AbbVie, Consulting fees from Anaptys Bio, Annexon Biosciences, AstraZeneca, BMS, Cyxone AB, Eli Lilly, Galapagos NV, Gilead Sciences Ltd, GSK, Istesso Ltd, Janssen, Kenko International, Kira Biotech, Ono Pharma, Pfizer, Revelo Biotherapeutics, Roche and Sanofi, Research grants from Pfizer, Janssen and GSK, Stefan Siebert Speaker or consulting fees from AbbVie, Amgen, AstraZeneca, Eli Lilly, GSK, Janssen, UCB, Institutional research grants from Amgen (previously Celgene), Boehringer Ingelheim, Bristol Myers Squibb, Eli Lilly, GSK, Janssen and UCB, Carl S Goodyear: None declared.