Abstract
OBJECTIVE: We previously described a classification system of persons with SLE based on whole blood RNA profiles and a random forest (RF) algorithm to predict individual patient endotypes. Here, we apply this algorithm prospectively in an independent set of patients to validate its use as a staging biomarker. METHODS: Whole blood from 101 patients participating in three clinical trials (NCT03626311, NCT03180021 and NCT05845593) meeting American College of Rheumatology (ACR) or Systemic Lupus Collaborating Clinics (SLICC) criteria for SLE classification was obtained at baseline, and RNA isolated and sequenced. Gene expression values were used as input to gene set variation analysis (GSVA), and the RF algorithm was applied using GSVA enrichment scores of 32 informative gene sets as input. Composite scores summarising gene expression perturbations were assigned to each patient using a ridge logistic regression algorithm. RESULTS: Patients with SLE were subset into eight endotypes identified by the algorithm. Patterns of gene enrichment in the identified endotypes mirrored those found in the previously reported endotypes. Differences in clinical characteristics, including serum complement levels, autoantibody positivity and the presence of nephritis, were observed between patients in various endotypes. Patients with active, concurrent nephritis were disproportionately assigned to the more molecularly perturbed endotypes. Composite scores were significantly, but modestly, inversely correlated with complement but not SLE Disease Activity Index (SLEDAI) or anti-double-stranded DNA antibody (anti-dsDNA) titre. CONCLUSIONS: The identification of eight molecular endotypes of lupus based on whole blood gene expression was validated in an independent data set of diverse patients. Endotyping patients with SLE based on transcriptional profiles can provide important status (presence of nephritis) information and provide novel molecular insights in support of personalised management.