POS1045 MASS SPECTROMETRY-BASED PROTEOMICS FOR THE IDENTIFICATION OF CANDIDATE SERUM PROTEIN BIOMARKERS THAT MAY PREDICT TREATMENT RESPONSE IN PATIENTS WITH PSORIATIC ARTHRITIS

Abstract

BackgroundThe Group for Research and Assessment of Psoriasis and Psoriatic Arthritis (GRAPPA) has identified a need for biomarkers to predict which patients (pts) with psoriatic arthritis (PsA) are most likely to respond to a specific therapy. Failure to identify effective treatments early on results in sub-optimal PsA disease management. Tofacitinib is an oral JAK inhibitor for the treatment of PsA. The efficacy and safety of tofacitinib 5 and 10 mg twice daily (BID) in pts with PsA have been demonstrated.1,2ObjectivesTo identify protein biomarker candidates, which may identify responders (R) vs non-responders (NR) to treatment of PsA, using mass spectrometry-based proteomics.MethodsBaseline (BL) serum samples from pts with PsA receiving tofacitinib 5 or 10 mg BID, adalimumab or placebo in OPAL Broaden (NCT01877668)1 were analysed. Pts were identified as R and NR based on the Psoriatic Arthritis Disease Activity Score (PASDAS) at Month 3; pts with lowest PASDAS ≤3.2 were defined as R, those with highest PASDAS >3.2 as NR. Two proteomic strategies were employed for analysis of BL serum samples: (1) targeted mass spectrometric multiple reaction monitoring analysis of an in-house panel (‘PAPRICA’) comprising of 206 proteins, originally developed to distinguish between different arthropathies, and (2) unbiased discovery liquid chromatography-tandem mass spectrometry (LC-MS/MS). PAPRICA data were normalised using two methods: normalisation to stable isotopically labelled peptide spike-ins (SIL; corrects for fluctuations in sample injections/mass spectrometry loading amounts), and normalisation to an endogenous peptide panel representing total serum protein abundance (TSPA; corrects for different amounts of total serum protein across samples). Univariate analyses (Student’s t-test) and multivariate machine learning Random Forest (RF) modelling3 were performed. Univariate analysis of the PAPRICA panel of proteins was performed on R vs NR, and within each treatment arm, with no adjustment for multiplicity.Results96 pts were identified as 47 R and 49 NR based on PASDAS scores. Of pts receiving tofacitinib 5 or 10 mg BID (data pooled), adalimumab or placebo, there were 26 R vs 26 NR, 13 R vs 13 NR and 8 R vs 10 NR, respectively. Results from univariate analysis identified 110 differentially expressed PAPRICA peptides between R vs NR (p≤0.05). RF multivariate analysis of all data (n=96) revealed a set of PAPRICA peptide signatures with the ability to differentiate between R and NR. Two RF models generated from the PAPRICA peptide data had training area under curves (AUCs) 0.956 [95% CI 0.93, 0.99] (TSPA) and 0.959 [95% CI 0.94, 0.98] (SIL). In total, 115 PAPRICA peptides representing 87 proteins were identified as potential biomarkers for predicting treatment response. Using unbiased discovery LC-MS/MS, univariate analysis of all data revealed one candidate peptide biomarker (p≤0.05). RF modelling revealed peptides that contributed to two prediction models with training AUCs of 0.903 [95% CI 0.86, 0.96] and 0.928 [95% CI 0.89, 0.96]. In total, from unbiased discovery LC-MS/MS, 66 peptides representing 39 proteins that may act as potential peptide biomarkers were identified in univariate and multivariate analyses.ConclusionUsing two complementary proteomic approaches and a combination of univariate and machine learning models, a total of 181 candidate biomarker peptides corresponding to 106 proteins have been identified that may act as potential biomarkers for predicting response to treatment of PsA. Further study is required to verify and evaluate these candidate biomarkers, and we will report how these proteins map to biological processes, pathways and networks.