Background/Purpose: Over the past 15 years, more than 10,000 whole tissue biopsies from patients with rheumatic diseases have been deposited into publicly available gene expression databases. Often rheumatic diseases are studied in isolation, but these data can be harnessed to characterize the full catalog of shared molecular patterns perturbed in these disease states. We present plans for a rheumatic disease transcriptomic compendium (Fig 1) and demonstrate feasibility through a case study in SLE whole blood gene expression data.

Methods: We curated experiments from studies of rheumatic diseases from ArrayExpress. For our SLE  whole blood case study, we selected 8 experiments from multiple platforms, including data from 2 clinical trials examining the effects of treatments modulating IFN (Lauwerys, et al. Arthritis Rheumatol. 2013; Welcher, et al. Arthritis Rheumatol. 2015.). Cross-platform normalization using quantile normalization was performed. Interferon module gene sets from Chiche, et al. Arthritis Rheumatol. 2014. and unsupervised machine learning algorithms were used to examine the change in IFN signatures during treatment and the overall data structure during the integration process. This is, to our knowledge, the first application of the Chiche, et al. whole blood modular framework to these trials.

Results: We demonstrate that it is possible to integrate SLE whole blood data from multiple platforms and studies and retain underlying biology. We find that expression of Type I IFN module genes are altered following the treatment with the therapeutic vaccine IFN-alpha-kinoid in patients with high baseline Type I signatures, consistent with the therapeutics mechanism of action and the original study (Fig 2). We also find that only putative Type II modules are altered during blockade of IFN-gamma.

Conclusion: We have established the feasibility of a rheumatic disease gene expression compendium that is an order of magnitude larger than any single publicly available experiment. Our results further support the utility of data-driven cross-disease modules and suggest that unsupervised approaches can yield insight into complex molecular patterns altered in rheumatic diseases (Fig 1).

Fig 1. Overview of the Rheumatic Disease Data Refinery.

Fig 2. Summarized expression values (change from baseline) of IFN modules during treatment with IFN-_-kinoid (data from Lauwerys, et al. Arthritis Rheumatol. 2013.). Patients were stratified into the following groups: placebo, those with a low Type I (M1.2) signature at base (IFN-negative), and those with a high M1.2 signature at baseline.