UBD modifies APOL1-induced kidney disease risk

We genotyped DNA from 253 African American (AA) individuals with either FSGS or nondiabetic ESRD to identify potential loci that interact with and modify the APOL1-associated phenotype(s). We used admixture mapping to discover genetic modifier(s) for the APOL1 high-risk genotype individuals (G1G1, G2G2, or G1G2) with FSGS, then compared the candidate loci with APOL1 high-risk individuals with ESRD and high-risk people without known kidney disease, conditioned on each person’s genome-wide admixture background. We also compared the genome-wide ancestries between different groups. We replicated our findings by genotyping an additional 62 FSGS samples (Fig. S1 and Table S1).

We examined FSGS as our primary phenotype of interest. Although most forms of nondiabetic CKD are associated with the APOL1 risk genotype, risk is much stronger with FSGS, a histologically defined (and presumably a more uniform) phenotype. We developed a ranked list of genomic loci, ordered by the degree of ancestry bias. We ranked loci by the number of SDs above (African ancestry enriched) or below (European ancestry enriched) the average African ancestry. For each SNP in the genome, we also applied a log likelihood ratio test (11) by comparing the disease-associated model with the null model to estimate the ancestry disease risk.

For admixture mapping, local ancestry was inferred by RFMix (12) and HAPMIX (13) programs. Both methods use data from ancestral populations (here African and European) as references to train a statistical model and then estimate the most likely ancestry states of the sequence of each query sample. We used the phased sequencing data of 99 YRI and 99 CEU from 1000 Genomes Project Phase 3 (14) as reference for RFMix. We also used the genotyping data of 113 YRI and 112 CEU individuals from HapMap (15) as reference for HAPMIX to replicate the ancestry inferring process. As the results of them were highly concordant, we present the results obtained using RFMix in this paper (Fig. S2).

For each individual, after local ancestry inference, we determined genome-wide ancestry and local ancestry at loci throughout the genome. The APOL1 high-risk genotype individuals studied included individuals with FSGS, AA individuals with ESRD, AA individuals from the Jackson Heart Study with or without CKD, and high-risk individuals imputed from two collections of samples available through the dbGAP database without kidney phenotype information but genotyped using the same platform as we used for the FSGS discovery and ESRD groups (phs000507 and phs000560) (Table S1).

We first compared the overall percentage of European and African ancestry in these different sample sets. Individuals with a high-risk APOL1 genotype had similar fractions of recent European ancestry except for the FSGS groups. The FSGS groups exhibited markedly higher proportions of recent European ancestry, despite sharing the high-risk APOL1 genotype, which is exceedingly rare except in people self-identified as AA. This difference (Fig. 1A) may reflect the methods of ascertainment and sample selection rather than any underlying biology as members of this FSGS cohort were chosen on the basis of APOL1 genotype, whereas self-identified AA race was a prerequisite for entry into the other study groups.

We then estimated the locus-specific African ancestry for each group after normalizing global ancestry person by person, following the method in Patterson et al. (16) to control the global ancestry difference between study groups. Using two complementary local ancestry inference algorithms and two sources of reference data (12, 13), we sought loci with ancestral bias greater than 3 SDs from the mean relative African ancestry in the FSGS group. We illustrate this approach by plotting the results for the Chr6p21.33–22.2 locus that we term the UBD locus (Fig. 1B). This locus has African ancestry 3.2 SDs beyond the mean in the discovery group and 2.3 SDs in replication group.

Other than the APOL1 locus itself on Chr22, only two loci reached the 3 SD threshold: Chr6p21.33–22.2 (Table 1) and Chr18q22.3–23 in the discovery group. The deviation at the Chr6 locus was replicated in a second smaller FSGS group (Fig. 1B and Fig. S2), whereas the locus on Chr18 did not replicate. For the Chr6 locus, a P value = 4.8 × 10⁻⁵ [false discovery rate (FDR) = 0.0061] representing the African ancestry bias was obtained using the combined data from the FSGS groups. Despite having more European ancestry at a genome-wide level, the high-risk APOL1 FSGS individuals were nevertheless enriched for African ancestry at the chromosome 6 locus (Fig. 1 and Fig. S1). Thus, we prioritized this locus for further evaluation.

We performed fine mapping in an effort to isolate causal variants and genes at the Chr6 locus. Based on the log likelihood ratio score estimated by Eq. S1 (ADM test), which maximizes the likelihood for estimation of ancestry disease risk, we estimated the relative probability of each SNP (or its highly linked sites) within the Chr6p locus as being causal, and also the 95% confidence interval for the causal allele of this locus (Fig. 1D), following the method of Freedman et al. (17). The 95% confidence interval is located from 26,378,681 to 30,951,367 bp (hg19), containing 148 genes. The most ancestry-biased region is located from 29,543,646 to 30,434,900 bp (hg19) and referred to as the high-risk region hereafter.

G1 and G2 both originated recently (18). We hypothesized that variability in the APOL1 associated phenotype may be influenced by differences in expression of modifier genes because most GWAS loci are located in regulatory regions (19, 20). By using the data from the Genetic European Variation in Health and Disease, A European Medical Sequencing Consortium (GEUVADIS) Project (21) and normalizing the RNA sequencing data to control for confounding factors (22, 23), we first checked for genes with cis-eQTL peaks (FDR ≤ 0.2) located in this high-risk region, where the peak was defined as the most significant SNP that correlated with gene expression. The eQTLs were estimated using data available from the YRI population to determine the effect of African ancestry at the Chr6p locus. There are 33 genes that meet this criterion. We compared the population-level gene expression for these genes between YRI and CEU populations (t test), as we hypothesized that gene expression difference was the driving force for ancestry bias at this locus. Only two genes meet both criteria, UBD and PPP1R18 (FDR < 0.2 for expression difference). We focused our attention on the UBD gene because UBD is among the three most highly up-regulated genes in glomeruli from nephrotic syndrome patients with a high-risk APOL1 genotype compared with other genotypes (10). The most highly associated SNPs in this region overlap with the SNPs that represent the strongest eQTLs for UBD (Fig. 1 and Table S2). As shown in Fig. 2, the YRI individuals have, on average, significantly lower UBD expression than those from the CEU population (P value = 8.4 × 10⁻⁴, FDR = 0.015). Taken together, this implies that higher African ancestry at this locus is associated with lower levels of UBD expression and prompted us to test the effect of UBD expression on APOL1 in a cell-based system.

We note that there are coding variants in UBD. Combinations of these variants form haplotypes with different frequencies in Africans and Europeans. However, these coding variants are not located in the region that shows the greatest ancestry bias in the FSGS samples. (We also performed functional testing of the major haplotypes defined by these variants between the YRI and CEU populations and saw no difference in the effect of these haplotypes on APOL1-associated cell toxicity as below.)

UBD is a ubiquitin-like protein modifier (6). Both UBD and APOL1 are up-regulated in response to inflammatory stimuli and may play roles in the innate immune system (10, 24, 25). Sampson et al. (10) showed that UBD is among the three most up-regulated genes in glomeruli from nephrotic syndrome patients with a high-risk APOL1 genotype. We therefore explored this interaction further using cellular assays.

We previously generated T-REx-293 stable cell lines that express Flag-tagged APOL1 G0, G1, and G2 isoforms under the control of tetracycline (tet) (26). An empty vector (EV) control cell line contains only the plasmid backbone. Induction of G1 and G2 expression (but not G0) leads to a marked increase in UBD mRNA (Fig. 3). In contrast to UBD mRNA, endogenous UBD protein level was reduced in G1- and G2-expressing cells compared with untreated cells and G0 cells (Fig. 3B). For comparison, we quantified the transcript level of the ubiquitin gene UBB as well as the GABBR1 gene that overlaps in genomic location with UBD. The induction of APOL1 expression did not affect the mRNA level of either UBB or GABBR1 (Fig. S3).

Previous studies noted that UBD overexpression (27–29) may lead to apoptosis. We transiently transfected the APOL1 T-REx-293 stable cell lines with UBD (pCMV6-UBD-HA) or EV control (pCMV6-entry) for 24 h followed by 24 h tet induction of APOL1. The cytotoxicity/viability ratio decreased significantly in G1 and G2 cell lines in the presence of UBD-HA overexpression (Fig. 4A). To confirm whether the cytotoxicity/viability ratio correlated with UBD expression, we transiently transfected different amounts of pCMV6-UBD-HA plasmid, balanced by EV plasmid to keep constant the total amount of transfected cDNA. As the UBD protein level increased, APOL1-associated cytotoxicity in G1 and G2 cell decreased (Fig. 4). This relationship was not observed in EV or G0 cell lines.

To investigate the association between the dose of UBD and APOL1-induced cytotoxicity, we inactivated endogenous UBD via both siRNA and CRISPR. For knockdown of UBD by siRNA, we transiently transfected T-REx-293 cells either with siUBD or with control siRNA and allowed the cells to grow for 48 h. In G1- and G2-expressing UBD knockdown cells, but not in cells treated with control siRNA, there was a substantial increase in APOL1 G1- and G2-mediated cytotoxicity (Fig. 4D). Previous studies have shown that in HeLa cells, knockdown of UBD by siRNA leads to cell death (30). However, here we did not find any significant changes in cytotoxicity in EV and G0 knockdown cells. We note that the observed change in cytotoxicity did not require particularly high knockdown efficiency (53%).

For UBD gene inactivation by CRISPR, we infected the T-REx-293 cells with virus and appropriate sgRNAs and controls and then selected single colonies. Immunoblots were done to identify cell colonies in which UBD was inactivated (Fig. 4F). In UBD-inactivated cells, G1- and G2-induced cytotoxicity became severe within 18 h of tet induction of APOL1 (Fig. 4E), more than with siRNA knockdown, consistent with more complete UBD inactivation by CRISPR. In the EV and G0 cells, UBD inactivation did not alter cytotoxicity.

To test whether UBD overexpression altered APOL1 levels in T-REx-293 stable cell lines, we immunoblotted whole-cell lysates prepared after transient transfection with UBD (pCMV6-UBD-HA), which overexpresses UBD with a hemagglutinin (HA) tag, or empty vector as a control, followed by tet induction for 24 h. Even with high levels of APOL1 induction in the stable cell lines, UBD expression led to a clear decrease in G1 and G2 protein but no noticeable change in G0 protein (Fig. 5A). Immunoblotting with anti-HA showed that UBD protein level is much lower in APOL1 G1- and G2-expressing cells compared with G0 cells, consistent with the results of endogenous UBD protein level (Fig. 3B). This suggests increased degradation of both APOL1 and UBD. We performed RT-PCR to see if UBD overexpression affects APOL1 protein level directly or rather influences APOL1 RNA level. UBD overexpression had no effect on the mRNA level of APOL1 (Fig. 5B).

These results imply that UBD interacts with APOL1 and targets it for degradation in the presence of G1 or G2 alleles. To confirm an interaction between UBD and APOL1 in vivo and the effect of proteasome inhibition on any such interaction, we performed coimmunoprecipitation (co-IP) experiments. T-REx-293 cells were tet-induced (50 ng/mL) for 9 h, with or without proteasome inhibitor MG132 (or DMSO as negative control) during the final 4 h. Lysates were subjected to IP with anti-FLAG (APOL1) and immunoblotted with anti-UBD. The interaction between UBD and each APOL1 variant was more easily detected with proteasome inhibition (Fig. 6). To confirm the interaction of APOL1 and UBD, we performed bidirectional co-IP. We observed that the G0, G1, and G2 forms of APOL1 all coimmunoprecipitated with anti-UBD. Endogenous UBD protein was detectable by APOL1 IP as well.