Unsupervised Clustering of Whole Blood Gene Expression Treatment Naïve Children and Adolescents with Childhood-Onset Systemic Lupus Erythematosus.

Objectives Childhood-onset systemic lupus erythematosus (cSLE) is a clinically and genetically heterogeneous disease.[1] We aimed to define subgroups of new diagnosis patients based on treatment-naïve gene expression profiles. We then examined how subgroup membership differs via demographics, clinical manifestations, and gene expression. Methods Participants were diagnosed and followed in a tertiary care lupus clinic and met SLE classification criteria. Clinical and laboratory data including disease activity and damage indices were prospectively collected and stored in a dedicated database. Participants were genotyped on a multiethnic array and ancestry was genetically inferred. Whole blood was collected prior to treatment initiation with glucocorticoids or other potent immunosuppressants, and whole blood transcriptome wide RNA sequencing was completed. We identified distinct participant clusters based on gene expression profiles using K-means clustering. Clinical and demographic feature differences between clusters were assessed using ANOVA and Fishers exact test. To gain an understanding of the genes driving cluster membership predictive modeling was used. Logistic LASSO regression and random forest supervised models were deployed and the genes most used by the models to aid in prediction of cluster membership were extracted. Results The cohort included 75 children and adolescents with cSLE with RNA sequencing on treatment-naïve whole blood samples. Initial K-means plots identified 3 clusters and one outlier which was excluded resulting in the final cohort. The group was 81% female with a median age of SLE diagnosis of 13.8 years (IQR: 12.1, 15.4). The demographic compositions of each cluster did not yield any statistically significant differences. Clusters differed significantly in the proportion of individuals with hypocomplementemia (C3 and/or C4) (Cluster 1= 40%, Cluster 2= 35%, Cluster 3= 100%, P= <0.01), fever (Cluster 1= 55%, Cluster 2= 13%, Cluster 3= 40%, P= <0.01), and Anti-Smith antibodies (Cluster 1= 70%, Cluster 2= 35%, Cluster 3= 40%, P= 0.01). Supervised models successfully identified genes predictive of cluster membership these genes include, but are not limited to IGHG1, IGHG4, INKA2, and SLC39A12-AS1. Cluster 1 was characterized by increased expression of SKA2, ARL1, and SRP68. Cluster 2 showed decreased expression of IGHG1, IGHG4, and IGLV1-44. Cluster 3 displayed increased expression of CD177, ALPL, and SLC4A1. Conclusion In a clinically heterogeneous, multiethnic cohort of patients with cSLE, treatment naïve whole blood RNAseq genome-wide expression generated 3 discrete clusters of patients. Genes characterizing cluster membership include IGHG1, SKA2, and CD177. Next steps include gene set enrichment analysis to denote differences in cluster biological processes. References [1.] Tsokos G. N Engl J Med 2011;365:2110-21.