Fig. 1

Identification of the optimal number of lupus endotypes using five datasets

K-means clustering of GSVA scores of the 32 features in A 1620 adult female lupus patients from GSE88884 (ILL-1 & ILL-2) yielded six clusters using baseline gene expression, B 266 adult lupus patients from GSE45291 yielded six clusters, C 137 pediatric lupus patients from GSE65391 yielded five clusters, and D 160 adult lupus patients from GSE116006 yielded four clusters using baseline gene expression. Of note, only 28 features were used in GSE65391 because of the microarray chip restrictions. E Cosine similarity analysis and F hierarchical clustering of the endotypes identified in these five datasets led to a final designation of eight transcriptionally distinct endotypes. Endotypes were considered similar if their cosine similarity was > 0.7. Endotypes underlined in red in E indicate the unique endotypes between datasets. Hierarchical clustering using complete agglomeration and a cut height of 1.8 is displayed in F. If available, ancestry, disease activity (where active indicates SLEDAI ≥ 6), lymphopenia (< 1 billion lymphocytes/L), and leukopenia (< 3.8 billion leukocytes/L) are annotated with color bars below each heatmap. Heatmaps were generated in GraphPad Prism v. 9.4.0 (673). In E, heatmaps were generated in R using the plot.matrix package and edited in Adobe Illustrator. Dendrogram in F was generated in R using the ggplot2 package