Detection of Genetic Overlap Between Rheumatoid Arthritis and Systemic Lupus Erythematosus Using GWAS Summary Statistics

Abstract

BackgroundClinical and epidemiological studies have suggested systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA) are comorbidities and common genetic etiologies can partly explain such coexistence. However, shared genetic determinations underlying the two diseases remain largely unknown.MethodsOur analysis relied on summary statistics available from genome-wide association studies of SLE (N = 23,210) and RA (N = 58,284). We first evaluated the genetic correlation between RA and SLE through the linkage disequilibrium score regression (LDSC). Then, we performed a multiple-tissue eQTL (expression quantitative trait loci) weighted integrative analysis for each of the two diseases and aggregated association evidence across these tissues via the recently proposed harmonic mean P-value (HMP) combination strategy, which can produce a single well-calibrated P-value for correlated test statistics. Afterwards, we conducted the pleiotropy-informed association using conjunction conditional FDR (ccFDR) to identify potential pleiotropic genes associated with both RA and SLE.ResultsWe found there existed a significant positive genetic correlation (rg = 0.404, P = 6.01E-10) via LDSC between RA and SLE. Based on the multiple-tissue eQTL weighted integrative analysis and the HMP combination across various tissues, we discovered 14 potential pleiotropic genes by ccFDR, among which four were likely newly novel genes (i.e., INPP5B, OR5K2, RP11-2C24.5, and CTD-3105H18.4). The SNP effect sizes of these pleiotropic genes were typically positively dependent, with an average correlation of 0.579. Functionally, these genes were implicated in multiple auto-immune relevant pathways such as inositol phosphate metabolic process, membrane and glucagon signaling pathway.ConclusionThis study reveals common genetic components between RA and SLE and provides candidate associated loci for understanding of molecular mechanism underlying the comorbidity of the two diseases.

Machine learning approaches for the genomic prediction of rheumatoid arthritis and systemic lupus erythematosus

“…Previous GWASs have demonstrated that several SNPs are associated with both RA and SLE [ 31 ]. For example, rs7574865 in STAT4 and rs2476601 in PTPN22 are well-known genetic variants associated with both diseases [ 32 , 33 ].…”

“…For example, rs7574865 in STAT4 and rs2476601 in PTPN22 are well-known genetic variants associated with both diseases [ 32 , 33 ]. In patients with RA, rs6457617 and rs9275406 in HLA-DQA1; rs9275406 and rs12525220 in HLA-DQB1; and rs6457620, rs615672, rs7765379, rs660895, rs13192471, rs6910071, rs9268839, rs9271348, rs3104413, rs9269234, rs9268839 and rs112112734 in HLA-DRB1 were common genetic loci [ 31 ]. Previous studies had reported rs2647012 and rs2187668 in HLA-DQA1; rs3129716 and rs114092478 in HLA DQB1; and rs9271100, rs9270984, and rs3135394 in HLA DRB1 to be associated with SLE [ 31 ].…”

“…In patients with RA, rs6457617 and rs9275406 in HLA-DQA1; rs9275406 and rs12525220 in HLA-DQB1; and rs6457620, rs615672, rs7765379, rs660895, rs13192471, rs6910071, rs9268839, rs9271348, rs3104413, rs9269234, rs9268839 and rs112112734 in HLA-DRB1 were common genetic loci [ 31 ]. Previous studies had reported rs2647012 and rs2187668 in HLA-DQA1; rs3129716 and rs114092478 in HLA DQB1; and rs9271100, rs9270984, and rs3135394 in HLA DRB1 to be associated with SLE [ 31 ]. However, such GWASs primarily compared genetic variants in patients with RA or SLE with those in healthy controls.…”

“…ML and AI applications in the context of autoimmune diseases classification have been widely investigated [ 31 ]. However, the most prevalent ML methods in prior RA and SLE studies have been LR, RFs, and SVM [ 31 ].…”

“…ML and AI applications in the context of autoimmune diseases classification have been widely investigated [ 31 ]. However, the most prevalent ML methods in prior RA and SLE studies have been LR, RFs, and SVM [ 31 ]. GTB and XGB have rarely been used in related investigations, but our results indicate the remarkable prediction performance of these two models in the classification of RA and SLE.…”