GG-05 Predictive ability of SLE genetic risk factors varies across ethnicities

Abstract

Background Systemic lupus erythematosus (SLE) exhibits marked ethnic disparities. The SLE Immunochip Consortium’s transancestral association study of SLE (27 574 individuals of European (EA), African (AA) and Hispanic Amerindian (HA) ancestry) raised the number of common risk variants to >100 (Langefeld 2017). There, we proposed the cumulative hit hypothesis, where the cumulative effect of individual loci is greater than if each locus acted independently. Here, we explore the joint contribution of SLE-susceptibility loci, how it varies by ethnicity, and whether there are distinct genetic risk profiles. Methods The SLE Immunochip study design, identification of risk loci, and genetic load (risk allele count (RAC) in EA samples) were previously described (Langefeld 2017). Genetic load was tested for association with SLE in an independent set of 2000/2000 EA case/controls, and in the AA and HA cohorts. Individuals in lowest 10% of the RAC distribution were the reference sample. A logistic regression model, adjusting for admixture, computed the odds ratio (OR) comparing the reference group to samples within a moving window of 20 unweighted RAC (moving window of 4 for the weighted (SNP’s log(OR)) analysis). Lasso regression identified EA risk SNPs that maximally predict SLE status in EA, then applied prediction to AA and HA. Factor analysis identified individual genetic risk profiles. Results The OR comparing lowest versus highest 10% of RAC was ∼30,∼6, and ∼3 for EA, HA and AA, respectively (figure 1), showing EA risk loci were not highly predictive of SLE risk in HA and AA. In EA, the moving window genetic load OR showed an increase beyond that predicted by independence but did not in HA and AA due to lower predictive ability. Lasso regression identified 51 risk alleles that maximally predicted SLE in EA, and a factor analysis identified seven uncorrelated risk profiles (one driven by HLA alleles and six by non-HLA loci). The lasso identified 31 (3 HLA) and 8 (no HLA) EA risk alleles as predictive in the HA and AA, respectively. Using predicted probabilities from the lasso in EA, the area under the ROC curve was 0.75 for EA, 0.72 for HA, and 0.60 for AA. Conclusions The EA SLE risk loci are highly predictive in EA (with a greater than additive effect), modestly predictive in HA, and weakly predictive in AA. We posit that the SLE genetics of AA are meaningfully different from EA and HA, merit increased study, and will require ethnicity-informed treatment strategies. Acknowledgements We would like to thank the SLE Immunochip Consortium, the Lupus Research Alliance, NIH (NIAMS and NIAID) and RILITE Foundation.