Investigating the metabolomics of treatment response in patients with inflammatory rheumatic diseases

Background: Rheumatic and musculoskeletal diseases (RMDs) are autoimmune-mediated chronic diseases affecting the joints around the body, involving an inappropriate immune response being launched against the tissues of the joint. These devastating diseases include rheumatoid arthritis (RA) and psoriatic arthritis (PsA). If insufficiently managed – or indeed in severe cases – these diseases can substantially impact a patient’s quality of life, leading to joint damage, dysfunction, and disability. However, numerous treatments exist for these diseases that control the immune-mediated factors driving disease, described as disease modifying anti-rheumatic drugs (DMARDs). Despite the success of these drugs for patients in achieving remission, they are not effective in all patients, and those who do not respond well to first-line treatments will typically be given an alternative drug on a trial-and-error basis until they respond successfully. Given the rapid and irreversible damage these diseases can induce even in the early stages, the need for early and aggressive treatment is fundamental for reaching a good outcome for the patient. Biomarkers can be employed to identify the most suitable drug to administer on a patient-to-patient basis, using these to predict who will respond to which drug. Incorporating biomarkers into the clinical management of these diseases is expected to be fundamental for precision medicine. These may come from multiple molecular sources. For example, currently used biomarkers include autoantibodies while this project primarily focuses on discovering biomarkers from the metabolome. Methodology: This project involved the secondary analyses of metabolomic and transcriptomic datasets generated from patients enrolled on multiple clinical studies. These include data from the Targeting Synovitis in Early Rheumatoid Arthritis (TaSER) (n=72), Treatment in the Rotterdam Early Arthritis Cohort (tREACH) (n=82), Characterising the Centralised Pain Phenotype in Chronic Rheumatic Disease (CENTAUR) (n=50) and Mayo Clinic - Hur et al. (2021) (n=64) – cohorts. The metabolic findings' translatability across cohorts was evaluated by incorporating datasets from various regions, including the United Kingdom, the Netherlands, and the United States of America. These multi-omic datasets were analysed using an in-house workflow developed throughout this project’s duration, involving the use of the R environment to perform exploratory data analysis, supervised machine learning and an investigation of the biological relevance of the findings. Other methods were also employed, notably an exploration and evaluation of data integration methods. Supervised machine learning was included to generate molecular profiles of treatment responses from multiple datasets. Doing so showed the value of combining multiple weakly-associated analytes in a model that could predict patient responses. However, an important component, the validation of these models, could not be performed in this work, although suggestions were made throughout of possible next steps. Results and Discussion: The analysis of the TaSER metabolomic data showed metabolites associated with methotrexate response after 3 months of treatment. Tryptophan and argininerelated metabolites were included in the metabolic model predictive of the 3-month response. While the model was not directly validated using subsequent datasets, including the tREACH and Mayo Clinic cohorts, additional features from these pathways were associated with treatment response. Included across cohorts were several tryptophan metabolites, including those derived from indole. Since these are largely produced via the gut microbiome it was suggested that the gut microbiome may influence the effectiveness of RMD treatments. Since RA and PsA were considered in this work as two archetypal RMDs, part of the project intended to investigate whether there were shared metabolic features found in association to treatment response in both diseases. These common metabolites were not clearly identified, although arginine-related metabolites were observed in models generated from the TaSER and CENTAUR cohorts in association with response to treatment in both conditions. Owing to the limitations of the untargeted metabolomic approach, this work was expected to provide an initial step in understanding the involvement of arginine and tryptophan related pathways in influencing treatment response in RMDs. Not performed in this work, it was expected that targeted metabolomics would provide clearer insights into these metabolites, providing absolute quantification with the identification of these features of interest in the patient samples. It was expected that expanding the cohort sizes and incorporating other omics platforms would provide a greater understanding of the mechanisms of the resolution of RMDs and inform future therapeutic targets. An important output from this project was the analytical pipeline developed and employed throughout for the omics analysis to inform biomarker discovery. Later work will involve generating a package in the R environment called markerHuntR. The R scripts for the functions with example datasets can be found at https://github.com/cambest202/markerHuntR.git. It is anticipated that the package will soon be described in more detail in a publication. The package will be available for researchers familiar with R to perform similar analyses as those described in this work