This was a nested retrospective study that enrolled participants from the ongoing, prospective NYU-wide Specimen and Matched Phenotype Linked Evaluation (SAMPLE) biorepository. This repository was initiated on February 14, 2014 with ongoing yearly renewal by the Institutional Review Boards of New York University Grossman School of Medicine (NYUGSoM) and Bellevue Hospital Center of the New York City Health and Hospitals Corporation. Written informed consent, which was available in English, Spanish, and Mandarin, was obtained from all participants in accordance with the Declaration of Helsinki principles. Participants were recruited from three high-volume SLE clinical sites. This nested sub-sample of eligible individuals were included, and retrospective chart reviews were completed. Inclusion criteria were individuals ≥18 years of age, of self-reported AA race, and meeting at least four of the American College of Rheumatology (ACR) revised criteria for SLE (24). Exclusion criteria were patients unwilling or unable to complete informed consent and patients with fewer than three clinical visits at least six months apart to ensure capture of longitudinal data. Self-reported ancestry was confirmed by ancestry informative markers as previously described (15), and participants without significant African admixture were excluded from analyses.

Genomic DNA was isolated from anti-coagulated whole blood collected in EDTA blood sample tubes using the Qiagen kit (Valencia, CA, USA) according to the manufacturer’s instructions. DNA was isolated from blood samples taken at clinical visits within 24–48 h of blood draw. DNA isolates were stored at −80°C. Batches of 10–15 DNA samples were evaluated and quantitated using a Nanodrop-1,000 spectrophotometer (Nanodrop Products, Wilmington, DE). One hundred nanograms (ng) of genomic DNA was used as a conventional polymerase chain reaction (PCR) template. A single 300-base-pair DNA segment containing the APOL1 polymorphisms, G1 (rs73885319 and rs60910145) and G2 (rs71785313), was amplified using AmpliTaq Gold 360 DNA Polymerase (Applied Biosystems, Foster City, CA). For quality control, DNA was elongated in both forward and reverse directions. Genotypes were analyzed using the GeneWiz online platform, as previously described (25).

We genotyped 121 study participants for APOL1; informative ancestry markers were completed, and a principal component analysis was run (26). Models were adjusted for African admixture using the Euclideandistance=(Σi=1[(PC1AA−AveragePC1AFR)2+(PC2AA−AveragePC2AFR)2])between principal components (PC) 1 and 2 of the AA samples and mean values for the African reference samples. This method excluded 10 participants who did not cluster with the African reference population. Of the 111 study participants remaining, longitudinal data capture was reviewed and supplemented with data from the electronic medical record where necessary. A study researcher (KY) blinded to the APOL1 status of the participants initially reviewed the medical charts, clinical notes, and imaging reports for documentation of evidence of ON, which was then validated by another co-researcher (AB) for accuracy. ON was defined by chart mention of a history of ON in physician notes. Mentions of Osteonecrosis in clinical notes were recorded and confirmed by imaging reports including xrays; and ultrasounds and MRIs where available. A screening workflow is presented in Figure 1.

Longitudinal, prospective data captured as part of the NYU- division wide SLE (SAMPLE) repository was reviewed, including SLE ACR criteria, physical examination findings, medication history, and disease activity measured using the the SELENA-SLEDAI HYBRID index, which comprises all elements of the SELENA-SLEDAI instrument (27) but contains a modification of the definition of proteinuria (28) such that four points for proteinuria is always captured when the UPCR is >0.5. Information on smoking history, body mass index, diabetes, dyslipidemia, blood pressure, eGFR, and proteinuria, were collected through retrospective chart review. Values for vital signs and laboratory data were obtained at three visits at least six months apart and averaged. For all study participants, four independent reviewers (KY, NL, RF, AB) reviewed each SLE-related visit, including rheumatology outpatient visits, inpatient hospitalizations, infusion clinic visits, and nephrology outpatient visits, where applicable, from 2011 (first available electronic health records) to last study visit date. The prescribed corticosteroid dose was recorded and converted to prednisone equivalent and dose duration at each visit. Mentions of “pulse dose steroids” were assumed to indicate 1,000 mg of methylprednisolone for three days. Where available, doses were corroborated with pharmacy records. Once obtained, the prednisone equivalent dose was multiplied by the number of days prescribed. The resultant values were added to obtain the cumulative prednisone dose.

Descriptive statistics [mean and standard deviation or median (25th, 75th percentiles) for continuous variables; frequencies and percentages for categorical variables] were calculated separately by group (ON present vs. ON absent) for demographic information and comorbidities. The two groups were compared using the chi-square or Fisher's exact test, as deemed appropriate, for categorical variables and the two-sample t-test or Mann–Whitney test for continuous data.

To test the association between APOL1 genotype and ON, we treated the number of APOL1 risk variants (additive model) as the predictive variable. The primary outcomes were time from SLE disease diagnosis to first ON occurrence and the number of ON sites. The analysis of “time from SLE disease diagnosis to first ON occurrence” was accomplished b computing the Kaplan–Meier product limit curves, where the data were stratified by the number of APOL1 risk variants. In cases where the endpoint event had not yet occurred, data were censored at the time of death or at the date of the last study visit. Comparison across APOL1 genotypes were made using the log-rank test. Multiple imputations were performed using the Fully Conditional Specification approach as implemented in the MICE R package to create 10 imputed datasets (29, 30). Cox penalized regressions were fitted using each imputed dataset independently starting with an initial model that included the following variables: disease duration (D3_Disease_Duration), age (D1_Age), sex (D1_Sex), BMI (D2_BMI), eGFR (D2_eGFR), history of nephritis (D3_Nephritis), end stage kidney disease (D3_ESRD) average complement C3 level) (D3_C3), average complement C4 level (D3_C4), average SLEDAI score (D3_AVE_SLEDAI), APOL1 genotype (APOL1_Risk_allele), and cumulative prednisone dose (Total_pred_divided_by_visits) (31, 32). Variables consistently selected across the 10 imputed datasets were retained. A final model that only included variables that were consistently selected in the penalized cox regression across the 10 imputed datasets was fitted without the penalty term to provide unbiased parameter estimates. These variables were lupus nephritis, age, and APOL1 genotypes. We also included biological sex in the model due to the known sex differences in SLE. Ten sets of estimates, one for each imputed dataset, were obtained. These estimates were combined using Rubin's approach for combined results from multiple imputations (33–35). Model results are reported as hazard ratios with corresponding 95% confidence intervals. Unless otherwise specified, results were considered statistically significant at the 0.05 level of significance. Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC), SPSS version 25, or R 3.6.3 platforms.

One hundred and eleven consented participants met the inclusion criteria. Demographic characteristics are shown in Table 1. Consistent with the reported higher incidence of SLE in women vs. men, 93% of study participants were female. The mean age was 43 ± 14 years, and mean BMI was 28.8 kg/m². The mean age of SLE onset was 29 ± 12 years, and median disease duration was 10 (5, 18) years. Among those who had received glucocorticoids, the median cumulative prednisone equivalent dose was 9.2 g (3.7, 21.2). Fifty-six percent of study participants had a history of SLE nephritis, with 17% having received a course of cyclophosphamide therapy. The clinical characteristics of the cohort are described in Table 1. As shown in Figure 1, 37 participants carried 0RVs, 59 carried 1RV, and 15 carried 2RV. The allele frequencies of the ancestral allele, G0, and variant alleles, G1 and G2, were 0.6, 0.22, and 0.18, respectively.

In our cohort 18 participants developed ON. Participants with ON had longer disease duration and carried more APOL1 RV copies with trends toward lower eGFR (Table 1). Across APOL1 genotype, 2 of the 37 0RV, 11 of the 59 1RV, and 5 of the 15 2RV participants developed ON (Figure 1). About 24% of the participants had no documented courses of glucocorticoids, both among those with and without ON. In participants who had received a glucocorticoid dose, the median cumulative prednisone equivalent dose was 18.6 g (2.0, 34.8) in those with ON and 9.0 g (3.8, 18.6) in those without ON. This trend did not meet statistical significance. The median disease duration in patients with vs. without ON was 9 (4, 16) and 20.5 (14, 30) years, respectively (Figure 2).

In time-to-event analysis, the proportion of individuals free from ON throughout SLE disease was assessed, and participants were censored at the age of the last follow-up visit or death. In an unadjusted ON-free survival model, 1RV and 2RV carriers developed ON earlier in SLE disease course, as presented by the Kaplan Meier Curves in Figure 3A. Across APOL1 genotype groups, the mean time to developing ON was 27 years in 0RV carriers, 22 years in 1RV carriers, and 18 years in 2 RV carriers (χ2 = 9.7; p = 0.005). The adjusted Cox regression model, 1RV and 2RV carriers exhibited shorter ON-free survival at hazard ratios of 3.1 (95% CI 1.6–6.2) and 9.6 (95% CI 2.4–37.8), respectively (Figure 3B).

We lastly assessed the number of body sites and outcomes of ON in each of the APOL1 genotype groups. Figures 4A,B show that APOL1 variant carriers were more likely to have multiple and bilateral disease sites. Among the 18 individuals with ON, 40 joint sites were affected. The most reported sites were the femoral head, followed by the distal femur, proximal tibia, and ankles. There was one report of ON in the wrist. All 3 ON-affected joints in 0RV individuals occurred at a unilateral site. By contrast, 9 of 23 and 7 of 14 ON-affected joints in 1RV and 2RV individuals occurred at bilateral sites. Regarding ON frequency, 0RV carriers developed ON in 1–2 (mean = 1.5) joints compared to 1–3 (mean = 1.9) and 1–5 (mean = 2.8) joints in 1RV and 2RV carriers, respectively (χ² = 19.0, p = 0.04, Figure 4B). The proportion of joints replaced was no different across the genotype groups.