LAUSR

Background Optimizing treatment in early rheumatoid arthritis (ERA) improves clinical outcomes. Developing approaches that would allow for accurate outcome predictions would be useful. We examined the possibility of employing SOMAscan to identify biomarkers that predict treatment response. Objectives To define methotrexate (MTX) 6 month treatment associated response protein changes using SOMAscan. Methods Sera from 14 Disease Modifying Antirheumatic Drug (DMARD) naive ERA patients at baseline (PRE) and after six months of MTX (POST) were analyzed using SOMAscan, an aptamer based assay that offers simultaneous relative quantitation of 1310 proteins. RA activity was measured by DAS28ESR3var abbrev DAS3; RF and ACPA were measured at baseline. SOMAmer intensity data was log2 transformed and differences (D=POST-PRE) clustered using undirected hierarchical self-organization. Kolmogorov-Smirnov differential analysis determined SOMAmers contributing to these populations at p<0.05. Potential processes associated with these SOMAmer regulation groups were identified using an in-house biological enrichment tool. The potential for SOMAmers to predict treatment response was also explored; for this we defined a fractional clinical response metric dDAS3= (DAS3_POST-DAS3_PRE)/DAS3_PRE. We then selected a population of proteins (n=3 to avoid over-fitting) with PRE expression levels best correlating to dDAS3. These three PRE expression values formed a weighted average, with weighting coefficients optimized by a simple Monte-Carlo method. We included this weighted average with clinical variables in logistic regression models, where 6 month DAS3 was the dependent variable. Results Clustering gave two populations of 6 and 8 patients (POP0, POP1) with mean delta DAS3 values of -1.71 and -0.46 respectively. In POP0 compared to POP1, 113 proteins were upregulated and 121 proteins were downregulated. The upregulated proteins were involved in VEGF signalling and platelet activation. The downregulated proteins were involved in regulation of immune response, cellular response to TNF and cytokine –cytokine receptor interactions. The fractional change dDAS3 correlated well with the treatment response panel (R2=0.8645; p=6.8e-5), with the caution that expression values of the 3 best-correlating proteins exhibited low coefficients of variation (<0.1). However, these proteins did reflect RA responses or inflammation. This weighted sum was also independently associated with treatment response in regression models including baseline DAS3 (or components) and RF/ACPA. Conclusions This pilot study suggests that high content proteomic approaches such as SOMAscan may be useful for developing prediction tools of patient responses to treatment. Extension of this work into a larger patient population is ongoing. References Journal of Statistical Software, 8(18) 1–4, 2004. Nature Protocols, 4(1) 44–57, 2009. Disclosure of Interest None declared