Written informed consent was obtained from all study participants, and the institutional review board at each collaborating center approved the study (University of California, San Francisco; Oklahoma Medical Research Foundation, Monash Health; and University of California, Los Angeles).

Participants were patients from established SLE cohorts and included individuals of European, African American, and Mestizo race and ethnicities. A total of 1,139 SLE cases were obtained from the United States, Australia, Spain, and Mexico. All participants fulfilled the American College of Rheumatology revised classification criteria for SLE.⁹ Participants were recruited from a variety of settings, including academic medical centers and community hospitals.

Four clinical manifestations, including anti-dsDNA autoantibody status, anti-Smith autoantibody status, lupus nephritis, and age of SLE onset, were studied. Lupus nephritis was defined as fulfilling the American College of Rheumatology renal classification criterion (>0.5 grams of proteinuria per day or 3+ protein on urine dipstick analysis) or having evidence of lupus nephritis on kidney biopsy.

DNA was collected from blood or saliva (Oragene DNA sample collection kits, DNAGenotek) from all study participants. Samples were genotyped using the Affymetrix LAT1 World array at the University of California, San Francisco, Institute of Human Genetics Genomics Core Facility. The Affymetrix LAT1 World array covers 817,810 SNP markers across the genome and was designed to include coverage for multiple diverse ethnic populations. Quality control procedures have been previously described.² Briefly, samples were filtered for a genotyping call rate less than 95%, discrepancies between reported and genetically assessed sex, and evidence of relatedness (one of each first-degree relative pairs removed, defined by identity by descent pi-hat >0.25). Departure from Hardy-Weinberg equilibrium was assessed using European and Asian patients without lupus nephritis or anti-dsDNA autoantibodies. SNPs were removed from analysis if evidence of departure from Hardy-Weinberg equilibrium was present (P < 5 E-08 in self-identified European patients and P < 1 E-05 in self-identified Asian patients) or if genotyping call rates were below 95%. Standard Affymetrix Axiom metrics were also applied (Dish QC ≥ 0.82 and default cluster metrics of SNPolisher). After applying these quality control assessments, 1,139 participants and 722,240 SNPs were available for analysis.

We used 1000 Genomes (1000G)¹⁰ reference panel for individuals of European, African American, and Mestizo ancestry. To improve our reference on Indigenous American ancestry, we used an external dataset consisting of 43 individuals with more than 95% Indigenous American ancestry to identify the subset of 1000G with more than 95% Indigenous American ancestry.¹¹ ADMIXTURE¹² was run unsupervised including all Amerindian 1000G patients combined with the 43 nonadmixed patients, assuming 2 subpopulations (K = 2, Indigenous American and other). Nonadmixed 1000G Indigenous American patients were used for downstream ancestry inference.

We used ADMIXTURE to determine the percentage of the overall genome belonging to each ancestral population. We first combined our sample data with 1000G genotype data and pruned SNPs for linkage disequilibrium, removing each SNP with an R² value greater than 0.1 in a 50 SNP sliding window advanced by 10 SNPs each time, as recommended.¹² This left 162,159 SNPs for global ancestry estimation. We then ran ADMIXTURE unsupervised assuming 5 subpopulations (K = 5, European, African, East Asian, South Asian, and Indigenous American). We used known labels from 1000G to determine the ancestry of the estimated proportions for each patient with SLE and placed them into European, African American, and Mestizo subgroups.

To ensure a high density within the MHC region of the genome for local ancestry estimation, HIBAG¹³ was used to impute classical HLA alleles (HLA- A, B, C, DRB1, DQB1, DQA1, DQB1, and DPB1). HIBAG includes pretrained models for African American and Mestizo populations and is well suited for HLA allele imputation in the admixed SLE samples. Imputed alleles with more than 0.5 posterior probability were retained for downstream analysis, as recommended. Genotype data were phased with all populations together using BEAGLEv5.1,¹⁴ 1000G as a reference panel, and GRCh37 genetic map positions in centimorgans.

Genome-wide local ancestry inference was estimated separately in African American and Mestizo patients using a machine learning algorithm, RFMixv2.0.¹⁵ For each population, RFMixv2.0 requires four inputs: (1) phased haplotypes admixed patients, (2) phased reference haplotypes, (3) a file with labels for the reference populations, and (4) genetic map positions. For African American samples, the reference population consists of all nonadmixed African and European 1000G patients. For Mestizo samples, the reference population consists of the 24 nonadmixed Indigenous American 1000G patients, as well as 24 randomly sampled African and European 1000G patients. RFMix was run using recommended input parameters of five minimum number of reference haplotypes per tree node and three expectation maximization iterations. The number of generations since admixture occurred that were used as input parameters were 6 and 11 for African American and Mestizo patients, respectively, according to previous estimates for populations in the United States.¹⁶

Global ancestry estimation identified a total of 236 African American, 306 Mestizo, and 597 European patients with SLE. Admixed patients (African American and Mestizo patients) were included in downstream analyses. On average, African American patients were 80.7% African and 19.3% European, and Mestizo patients were 48.9% Indigenous American, 6.1% African, and 45.0% European (Table 1).

Within each racial and ethnic group, we tested for association between global ancestry proportions and SLE manifestations. There were no significant associations between global ancestry and SLE manifestations in African American or Mestizo patients.

We investigated the association between local ancestry and SLE manifestations. We investigated whether local ancestry spanning chromosome 6:28477797–33448345, which included HLA class I, II, and III regions, was associated with lupus nephritis, age at onset, anti-dsDNA autoantibody status, and anti-Smith autoantibody status, separately for each admixed population. In African American and Mestizo patients, RFMix estimated 50 and 85 windows of local ancestry, respectively.

Local ancestry differences between patients with SLE with and without the manifestations of interest were small throughout all the windows in the MHC region. On average, the difference in European ancestry in African American patients was 3.62% for lupus nephritis, 0.58% for age at onset, 0.20% for anti-Smith autoantibodies, and 0.01% for anti-dsDNA autoantibodies. In Mestizo patients, the average difference in European ancestry between patients with and without SLE manifestations was 3.30% for lupus nephritis, 0.14% for age at onset, 8.1% for anti-Smith autoantibodies, and 6.3% for anti-dsDNA autoantibodies. After correction for multiple hypothesis testing, no significant differences were observed in local ancestry between patients with SLE with and without the manifestations of interest in African American or Mestizo populations.

Results for anti-dsDNA autoantibodies showed the smallest nominal P values. Although no results reached statistical significance in our analysis, we report the top 10 results for differences in local ancestry with anti-dsDNA autoantibodies in both African American patients and Mestizo patients in Table 2. Figures 1A and 1B show the small differences in local ancestry observed in African American and Mestizo patients with SLE with and without anti-dsDNA autoantibodies.