Identifying the dietary signatures of arthritis through metabolomics

We report results from the Versus Arthritis ‘Designa’ project, investigating the role of human biofluid metabolites, endogenous and exogenous (dietary), in the onset and progression of rheumatic diseases, specifically osteo- and rheumatoid arthritis (OA and RA). The technique of choice for examination of small molecules is high-resolution 600 MHz NMR spectroscopy, which has been used in our work to examine the polar fraction extracted from blood (serum). The initial results presented here are from 7 RA patients who have been monitored at multiple time points over a period 10 years, with timepoint = 0 corresponding to first onset/diagnosis of arthritic disease. Two different routes were used to extract and quantify peak area information from the NMR spectra. The first approach used the Chenomx software package (Chenomx Inc.,Edmonton, Canada) which is based on using spectral reference libraries to quantify a collection of annotated metabolites. The manual implementation of this approach, as used here, yields precise concentration estimates but requires substantial human oversight, a recognised bottleneck in large-scale metabolomics studies. The second approach utilised global spectral deconvolution (GSD [1]) as implemented in Mnova (Mestrelab Research S.L., Santiago de Compostela, Spain) to annotate peaks at the chemical shift level and estimate their areas and intensities. This was followed by density-based clustering applied to the chemical shifts to extract peaks found to be present at different intensities in all samples. Apart from setting some hyper-parameters, this approach is fully automated. Many of the metabolite concentrations were found to be significantly intercorrelated, forming multiple different, unconnected networks. Both data tables revealed interesting effects across the decade-long study, including some trends common to all study participants. Further, in unsupervised cluster analysis, both approaches show that a major source of systematic variance in the datasets is patient identity: metabolite profiles for individuals are distinct and persist over the decade study duration. Canonical correlation analaysis confirmed that significant information is common to both the manually and automatically prepared data tables. We conclude that the use of two complementary peak extraction methods allows for mutual validation of findings from the respective peak information tables, and the commonality of information suggests that the fast, automated GSD-approach can provide a pilot data table for useful exploration in advance of tackling the more resource-intensive metabolite quantitation.