An exploratory analysis of circulating cytokine profiles, pain-associated transcripts and autoantibody responses associated with treatment response in knee osteoarthritis

Background: Pain response to treatment in knee osteoarthritis (OA) varies substantially between individuals, but the biological factors underlying this heterogeneity remain poorly understood. We investigated whether circulating protein, transcriptional and autoantibody features were associated with variation in pain response to lifestyle interventions in people with knee OA. Methods: We performed an exploratory multi-modal analysis using participants from the INSPIRE randomised controlled trial and, for proteomic analyses, an independent WebEx cohort. Circulating inflammatory proteins were measured using Olink proximity extension assay at baseline and 6 weeks. Associations between baseline protein concentrations, longitudinal protein changes and pain response were assessed using multivariable regression, with multivariate protein patterns examined using supervised and sparse partial least squares approaches. Whole-blood transcriptomic data from INSPIRE were analysed using limma, restricted a priori to pain-associated genes represented in the Human Pain Genetics Database (HPGDB), with a responder-by-visit interaction used to assess differential longitudinal expression. Finally, exploratory IgG binding to murine satellite glial cells (SGCs) was assessed in selected INSPIRE participants with contrasting clinical responses to exercise. Results: No circulating protein or transcriptomic association remained significant after correction for multiple testing. Several nominal associations were observed in the proteomic analyses, including baseline IL-10RA and CST5 in INSPIRE and OSM and TNFSF14 in WebEx. Longitudinal changes in IL-17A and TSLP were nominally associated with improvement in pain in INSPIRE, with a similar nominal association between IL-17A and pain improvement observed in WebEx. In the transcriptomic analysis, MMP9 showed the strongest nominal association with differential longitudinal expression according to pain response under both the ΔNRS ≤ −1 (log2 fold change = 0.91, p = 0.008) and ΔNRS ≤ −2 (log2 fold change = 1.02, p = 0.004) definitions; neither association remained significant following FDR correction. Exploratory SGC-binding analysis identified markedly increased baseline IgG binding in one participant with worsening pain, but this pattern was not consistently observed among other participants with poor clinical responses. Paired analyses provided no evidence of a consistent increase in SGC-binding IgG following exercise. Conclusions: This exploratory multi-modal analysis found no robust molecular signature of the pain response that survived correction for multiple testing. Nominal associations involving several circulating proteins, including IL-10RA, IL-17A and MMP9 provide potential leads for future investigation, but these findings should be interpreted cautiously given the multiple comparisons and limited sample sizes. Exploratory autoantibody analyses likewise provided no evidence of a consistent SGC-binding signature. Larger, independently powered studies are required to determine whether these candidate molecular features are reproducibly associated with treatment-related variation in pain response.