Background: The management of Juvenile Idiopathic Arthritis (JIA) has evolved significantly over the past 15 years, driven by the widespread use of disease-modifying anti-rheumatic drugs (DMARDs), with methotrexate (MTX) remaining the cornerstone of first-line treatment for many JIA subtypes. However, the response to MTX is highly heterogeneous. Clinical outcomes often follow complex trajectories, sometimes characterised by discordant trends between physician assessments and patient- or parent-reported measures [1]. Recently, type I (IFNα) and type II (IFNγ) interferon response gene signatures in peripheral blood have been associated to MTX response [2].

Objectives: To assess whether peripheral blood IFN response scores associate with trajectories of clinical response in JIA.

Methods: This study utilised data from children and young people recruited into the Childhood Arthritis Response to Medication Study (CHARMS), a UK multicentre drug cohort collating clinical and biological data at the initiation of DMARD therapies and six months later. This cohort is integrated into the CLUSTER initiative. Participants were selected if initiating MTX for the first time, had active disease at baseline (T1), defined as an active joint count (AJC) >0 and/or a Juvenile Arthritis Disease Activity Score (JADAS10) ≥1. Furthermore, participants had to have complete clinical data at either baseline (T1) or follow-up (T2, 3-11 months after treatment initiation). Disease courses were annotated using previously-modelled and assigned multivariate group-based trajectories using shared disease courses over active joint counts, physician global assessment and patient/parent global assessment scores, and ESR [1]. Interferon (IFN) response scores measured in PBMC by bulk RNASeq were calculated using a 51-gene signature (51 IGS) and a 5-gene signature (5 IGS), as described [2], based on samples collected prior to MTX therapy. To assess representativeness, the sub-cohort with available IFN scores (n=102) was compared with the larger clinical cohort (n=699) using the Standardised Mean Difference (SMD, cut-off 0.2).

Correlation between 51 IGS and 5 IGS was tested using Pearson’s method, with results displayed as a scatter plot including the correlation coefficient and p-value. Associations between the clinical trajectory groups and IFN response scores were evaluated using the Kruskal–Wallis test. Additionally, linear regression models were used to determine if trajectory group membership was associated with differences in continuous IFN gene scores after adjusting for age, whereas logistic regression models were used to assess whether IFN gene scores predicted the likelihood of belonging to specific trajectory phenotype. Correction for multiple testing was applied using the Benjamini-Hochberg method (BH).

Results: Initial analysis included 699 children and young people (CYP) from the CHARMS cohort study in whom clinical data were available. A nested cohort of 102 CYP had available PBMC transcriptomic data for the 51-gene and 5-gene IFN response scores. While the balance between the sub-cohort of 102 CYP and the larger cohort of 699 was generally assessed with a cut-off of 0.2, it was noted that baseline variables showing imbalance had high missingness rates. Within the subcohort of 102 CYP, individuals were assigned to one of five trajectory groups derived from the model: Fast Improvers (12.4%), Slow Improvers (26.9%), Persistent Disease (37.0%), Persistent physician scores (11.9%), and Persistent patient/parent scores (12.1%). The 4-group trajectory model identified: High phyVAS (16.7%), Fast Responder (30.4%), Slow Responder (25.5%), and Non-Responder (27.5%). The 3-group trajectory model identified: Slow Responder (41.2%), Fast Responder (30.4%), and Non-Responder (28.4%). A strong correlation was observed between the 51 IGS and the 5 IGS (p value <0.001). 51- gene and 5- gene IFN scores were compared across the 3, 4, and 5 trajectory groups. There were no statistically significant differences in score distributions for either the 51 IGS (p = 0.404) or the 5 IGS (p = 0.186) between trajectory groups. Additionally, there were no significant associations between assignment to any specific trajectory group and IFN response scores after correcting via BH. Post-hoc analyses demonstrated that as the number of trajectory classes rises, to better capture clinical nuance, statistical power wanes, whereas sensitivity remains low. Conversely, reducing the number of classes increases power but results in simplistic models (dichotomising outcomes into 2 or 3 classes) that fail to capture the true variety of MTX response.

Conclusions: The findings of this study underscore the complexity of predicting treatment outcomes in JIA. The response to methotrexate is highly heterogeneous, a reality that is difficult to fully capture through standard modelling techniques. While the IFN score is effective at identifying disease activity capable of responding to methotrexate, this study indicates that it does not reflect specific disease trajectories when defined in a categorical manner.

REFERENCES: [1] Shoop-Worrall SJW, Lawson-Tovey S, Wedderburn LR, Hyrich KL, Geifman N, Consortium C. Towards stratified treatment of JIA: machine learning identifies subtypes in response to methotrexate from four UK cohorts. EBioMedicine. 2024;100:104946.

[2] Kartawinata M, Lin WY, Jebson B, O’Brien K, Ralph E, Welsh E, et al. Identification and validation of interferon-driven gene signature as a predictor of response to methotrexate in juvenile idiopathic arthritis. Ann Rheum Dis. 2025;84(8):1412-24.

Acknowledgments: NIL.

Disclosure of Interests: Laura Scagnellato: None declared, Melissa Kartawinata: None declared, Wei-Yu Lin: None declared, Stephanie Shoop-Worrall: None declared, Roberta Ramonda: None declared, Chris Wallace GSK, Lucy Wedderburn: None declared.