Identifying Clinical Endophenotypes in Childhood-Onset Systemic Lupus Erythematosus Using Unsupervised Machine Learning

Le résumé fourni par la source

Objectives Childhood-onset systemic lupus erythematosus (cSLE) is a clinically heterogeneous autoimmune disease. We hypothesize that unsupervised machine learning may identify clinically homogeneous patient subtypes with distinct genetics. Methods We included cSLE patients diagnosed and followed at a tertiary care pediatric lupus clinic between 1992-2023 and genotyped using Illumina multiethnic arrays, imputed with TopMed. We extracted SLE manifestations, date of manifestation onset, and demographic characteristics from dedicated Lupus databases. Genetic ancestry was inferred using principal components and ADMIXTURE with 1K Genomes as a referent. We used the presence/absence and time to each SLE manifestation onset from SLE diagnosis to identify patient clusters with similarity network fusion (SNF). SNF clusters were validated with simulation-based sensitivity analysis. We ran 1000 independent simulated SNFs (simSNF). In each simSNF, we randomly used 70% of the cohort, and then the results across all iterations were aggregated. Kaplan-Meier and Cox models compared time to manifestation onset between clusters. Cluster differences in demographic and manifestation prevalences were tested using χ 2 or Fisher’s exact test. We tested autosomes (n= 17448) and chromosome X (n=713) gene-level associations between SNF-clusters, adjusted for sex and ancestry, using SKAT-O (SAIGEv1.4.5). We completed gene set enrichment pathway analysis aggregated with SKAT-O genetic associations across ’Immune system process’ Gene Ontology pathway (GO:0002376) and its decedents. We estimated the effective number of independent immune-related pathways (Galwey method, P<7×10^-5; 0.05/664 independent pathways). Results In a cohort of 442 cSLE patients (83% female, median diagnosis age 14.2 years), SNF identified 2 clusters. Cluster 1 (n=205) were predominantly of European ancestry (41%), while cluster 2 (n=237) was mainly composed of patients of East Asian (30%) and South Asian (22%) ancestry (P=3×10^-9) (Figure 1A). Cluster 2 patients had higher prevalence and earlier onset of 9 cSLE manifestations (eg, lupus nephritis III/IV, anemia, anti-dsDNA) in >900/1000 simSNFs (HR>1.8; P<6×10^-5) (Figure 1B). Simulation-based sensitivity analysis demonstrated that 95% of patients consistently clustered together over 1000 simulations. Individually, none of the tested genes were associated with cluster membership (P>5.8×10^-8). None of the tested immunology related GO pathways were significantly associated with cluster membership. Figure 1: (A) Difference in ancestral distribution between consensus SNF clusters. Difference in ancestry distribution between clusters was determined with a X 2 test. (B) Clinical and laboratory SLE manifestations with different prevalences between consensus SNF dusters. Clinical manifestations are labelled in blue and laboratory manifestations are labeled in purple. “LN” stands for lupus nephritis. Difference in manifestation prevalence between clusters was determined with a fisher’s exact test. A Bonferroni corrected threshold for significance was set at P <0.001 (0.05/38 independent tests). Conclusion In a large multiethnic cSLE cohort, SNF identified 2 robust clusters. The cluster with more severe disease and earlier onset was enriched for patients of East and South Asian ancestry. We did not find an association between genes or immunologic gene pathways and cluster membership.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.