Analysis of transcriptomic features reveals molecular endotypes of SLE with clinical implications

Abstract

Abstract Background Systemic lupus erythematosus (SLE) is known to be clinically heterogeneous. Previous efforts to characterize subsets of SLE patients based on gene expression analysis have not been reproduced because of small sample sizes or technical problems. The aim of this study was to develop a robust patient stratification system using gene expression profiling to characterize individual lupus patients. Methods We employed gene set variation analysis (GSVA) of informative gene modules to identify molecular endotypes of SLE patients, machine learning (ML) to classify individual patients into molecular subsets, and logistic regression to develop a composite metric estimating the scope of immunologic perturbations. SHapley Additive ExPlanations (SHAP) revealed the impact of specific features on patient sub-setting. Results Using five datasets comprising 2183 patients, eight SLE endotypes were identified. Expanded analysis of 3166 samples in 17 datasets revealed that each endotype had unique gene enrichment patterns, but not all endotypes were observed in all datasets. ML algorithms trained on 2183 patients and tested on 983 patients not used to develop the model demonstrated effective classification into one of eight endotypes. SHAP indicated a unique array of features influential in sorting individual samples into each of the endotypes. A composite molecular score was calculated for each patient and significantly correlated with standard laboratory measures. Significant differences in clinical characteristics were associated with different endotypes, with those with the least perturbed transcriptional profile manifesting lower disease severity. The more abnormal endotypes were significantly more likely to experience a severe flare over the subsequent 52 weeks while on standard-of-care medication and specific endotypes were more likely to be clinical responders to the investigational product tested in one clinical trial analyzed (tabalumab). Conclusions Transcriptomic profiling and ML reproducibly separated lupus patients into molecular endotypes with significant differences in clinical features, outcomes, and responsiveness to therapy. Our classification approach using a composite scoring system based on underlying molecular abnormalities has both staging and prognostic relevance.

Keywords

• Systemic lupus erythematosus (SLE)
• Autoimmunity
• Inflammation
• Gene expression
• Endotype
• Machine learning (ML)