Pharmacogenetics of tenofovir renal toxicity in HIV-positive Southern Africans

Objective Renal toxicity is more common with tenofovir disoproxil fumarate (TDF) than with tenofovir alafenamide fumarate (TAF). We investigated whether polymorphisms in genes relevant to tenofovir disposition affect renal toxicity among HIV-positive Southern Africans. Methods Genetic sub-study of adults randomized to initiate TAF or TDF together with dolutegravir and emtricitabine was conducted. Outcomes were changes from week 4 to 48 in the estimated glomerular filtration rate (eGFR) and from baseline to week 48 in urine retinol-binding protein and urine β2-microglobulin adjusted for urinary creatinine (uRBP/Cr and uB2M/Cr). Primary analyses prioritized 14 polymorphisms previously reported to be associated with tenofovir disposition or renal outcomes, and all polymorphisms in 14 selected genes. We also explored genome-wide associations. Results 336 participants were enrolled. Among 14 polymorphisms of primary interest, the lowest P values for change in eGFR, uRBP/Cr, and uB2M/Cr were ABCC4 rs899494 (P = 0.022), ABCC10 rs2125739 (P = 0.07), and ABCC4 rs1059751 (P = 0.0088); and in genes of interest, the lowest P values were ABCC4 rs4148481 (P = 0.0013), rs691857 (P = 0.00039), and PKD2 rs72659631 (P = 0.0011). However, none of these polymorphisms withstood correction for multiple testing. Genome-wide, the lowest P values were COL27A1 rs1687402 (P = 3.4 × 10−9), CDH4 rs66494466 (P = 5.6 × 10−8), and ITGA4 rs3770126 (P = 6.1 × 10−7). Conclusion Two ABCC4 polymorphisms, rs899494 and rs1059751, were nominally associated with change in eGFR and uB2M/Cr, respectively, albeit in the opposite direction of previous reports. COL27A1 polymorphism was genome-wide significantly associated with change in eGFR.


Introduction
Tenofovir disoproxil fumarate (TDF), a prodrug of tenofovir, is recommended by the WHO as part of first-line antiretroviral therapy (ART) regimens [1]. In 2015, the USA Food and Drug Administration approved a second tenofovir prodrug, tenofovir alafenamide fumarate (TAF), for inclusion in ART regimens [2]. Compared to TDF, TAF is as effective but has a more favourable renal and bone safety profile [3][4][5]. The prescribing shift from TDF to TAF has been driven, in part, to avoid TDF renal tubular toxicity [6][7][8][9][10]. The severity of renal dysfunction with TDF correlates with higher plasma tenofovir concentrations [11]. Plasma tenofovir concentrations with TAF are approximately 90% lower than those with TDF [3,5].
TDF undergoes esterase hydrolysis, which removes two ester groups to form tenofovir, while TAF undergoes hydrolysis by intracellular cathepsin A to form tenofovir [12]. Tenofovir is renally eliminated by glomerular filtration and tubular secretion. Clearance in the proximal tubule is controlled by membrane transport proteins. Tenofovir is actively transported into proximal renal tubular cells by organic anion transporters 1 and 2 (encoded by SLC22A6 and SLC22A7, respectively), and transported out of these cells by efflux transporters including multidrug-resistant protein 2 and 4 (encoded by ABCC2 and ABCC4, respectively) [13,14]. Genetic variants in these transporter genes have been postulated to increase risk of renal toxicity by favouring intracellular accumulation of tenofovir [11,15,16].
Studies of tenofovir toxicity in African populations have shown inconsistent results [17]. Some found tenofovir-containing regimens to be without substantial renal toxicity [18][19][20], while others reported increased renal toxicity and recommend monitoring of renal function [21,22]. Candidate gene studies from various populations have sought to associate renal transporter polymorphisms with TDF renal toxicities among individuals treated for HIV; however, most did not include African populations. A study of 115 patients in Spain receiving TDF-containing regimens found an association between renal tubular dysfunction and homozygosity for ABCC2 − 24C/C (rs717620) [23]. Similarly, an association between TDFinduced renal tubular dysfunction and two ABCC2 polymorphisms (rs717620 and rs2273697) was reported in a Japanese population [24]. Among 501 participants randomized to TDF-containing regimens in AIDS Clinical Trials Group protocol A5202 [25], a SLC22A2 polymorphism (rs3127573) was associated with more favourable creatinine clearance among African-American participants [26]. A study of Ghanaians receiving TDF-containing regimens suggested an associated between an ABCC10 polymorphism (rs2125739) and worsening renal function [27].
Because of inconsistent data from pharmacogenomic studies and the limited data regarding the pharmacogenetics of TDF toxicity in Africans, who have the highest genetic diversity worldwide, we evaluated whether laboratory markers of renal tubular injury were associated with polymorphisms in genes involved in TAF, TDF, and tenofovir metabolism and transport among ART-naive Southern Africans who had been randomized to initiate either TDFor TAF-containing regimens in a prospective clinical trial.

Ethics
The present study was conducted in accordance with the Declaration of Helsinki and the ADVANCE protocol WRHI 060 (NCT03122262) received ethics and regulatory approvals from the Wits Human Research Ethics Committee (REF 160606B) and the South African Health Products Regulatory Authority (REF 20160620), respectively. Ethics approval for the pharmacogenetics sub-study was also granted by the University of Cape Town Health Sciences Human Research Ethics (REF 571/2019). Written informed consent for genetic research was obtained from study participants.

Study population
The ADVANCE study was a 96-week, randomized, phase 3 non-inferiority clinical trial which enrolled 1053 HIVpositive participants living in Johannesburg, South Africa. The ADVANCE trial has been previously described in detail [28]. In brief, participants in ADVANCE were randomly assigned 1 : 1 : 1 to receive: TAF, dolutegravir, and emtricitabine; TDF, dolutegravir, and emtricitabine; or TDF, efavirenz, and emtricitabine [28]. Eligible participants were at least 12 years of age, weighed at least 40 kg, had creatinine clearances greater than 60 ml/min, were ARTnaive within 6 months before entry, were not pregnant, and did not have tuberculosis at the study start [28]. Among 702 participants randomized to receive either TAF, dolutegravir, and emtricitabine, or TDF, dolutegravir, and emtricitabine, 340 (48%) consented for genetic research.
Beyond these polymorphisms, we used a stepwise approach to prioritize sets of polymorphisms to interrogate to decrease the burden of multiple comparisons. We surmised that polymorphisms previously and strongly associated with at least one drug-related phenotype, or that have been significantly genome-wide associated with any trait, are most likely to be true associations. We explored all polymorphisms from the Pharmacogenomics Knowledgebase that were associated with any drug at levels of evidence of 1 (preponderance of evidence shows an association, replicated in multiple cohorts, and preferably with strong effect size) or 2 (moderate evidence of association, replicated but some studies may not show statistical significance, or with small effect size) (accessed 10 February 2021) [34], and all a curated collection of all human genome-wide association studies (NHGRI-EBI GWAS) Catalog polymorphisms with P values less than 5.0 × 10 −8 for any trait (accessed 10 February 2021) [35]. We prioritized polymorphisms common to both Pharmacogenomics Knowledgebase and the GWAS Catalog, considering these to have the most robust evidence for true associations. We secondarily explored all polymorphisms from Pharmacogenomics Knowledgebase and the GWAS Catalog (based on criteria described above), and all polymorphisms in our imputed genome-wide genotype data.

Genetic polymorphisms
Whole blood labelled with coded identifiers was stored and DNA extraction was performed at the Sydney Brenner Institute for Molecular Bioscience at the University of the Witwatersrand, using the salting out method as described elsewhere [36]. Genotyping with the Illumina Infinium Multi-Ethnic Global BeadChip (MEGA EX , Illumina, Inc, San Diego, California, USA) was done at Vanderbilt Technologies for Advanced Genomics in Nashville, Tennessee, USA. Post-genotype quality control included sex checks, call rates by marker and sample, identity by descent plots, assessment for batch effects, concordance between duplicate samples, and HapMap controls.
Quality control steps were performed using PLINK version 1.9, Cambridge, Massachusetts, USA [37]. Genotyping efficiency per participant was greater than 95% in all samples. Markers with genotyping efficiency less than 95% were excluded, as were those with minor allele frequencies (MAFs) < 5%. We excluded 21 samples with overall genotyping call rates less than 95%. After quality control, data were imputed using the TOPMed reference panel after transforming to genome build 38 using liftOver and stratification by chromosome to parallelize the imputation process [38,39]. For each chromosome in each phase, 100% concordance with genotyped data was assessed. Polymorphisms with imputation scores <0.3, genotyping call rates <99%, MAF < 0.05, or Hardy-Weinberg Equilibrium (HWE) P values <1.0 × 10 −8 were excluded. To control for population stratification, we used Eigenstrat/Eigensoft package (Cambridge, Massachusetts) 6.0.1 to estimate principal components.

Clinical and laboratory data
Clinical data from the ADVANCE trial were collected at screening, enrolment, weeks 4 and 12, and every 12 weeks until week 96 [28]. eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration formula [40]. Change in eGFR from week 4 to week 48 (ΔeGFR 4-48 ) was used in analysis. Week 4 was used as baseline because dolutegravir increases serum creatinine concentrations by inhibiting tubular secretion of creatinine, which plateaus by week 4 [41,42]. The low molecular weight proteins RBP and B2M were selected because their increases in urine are early indicators of renal tubular dysfunction or toxicity [42]. These markers were calculated as ratios to urine creatinine (ΔRBP/Cr 0-48 and ΔB2M/Cr 0-48 , respectively).
The Shapiro-Wilk test was used to assess normality, with P < 0.05 considered statistically significant. Clinical data not normally distributed were described with frequencies, medians, and interquartile ranges (IQR). Natural logarithmic transformations were done on uRBP/Cr and uB2M/Cr to approximate normality.

Association analyses
Multivariable linear regression analyses were performed to evaluate associations between genetic polymorphisms and each renal outcome. Covariates included baseline age, sex, study arm (TAF or TDF), screening BMI, log 10 plasma HIV-1 RNA, CD4 T-cell count, and the first three principal components to account for population substructure. The inclusion of cotrimoxazole use as a covariate was also examined. Our primary objective was to identify genetic associations with renal toxicity, rather than to characterize toxicity differences between TDF and TAF on kidney biomarkers. We therefore combined both study arms in our analyses while adjusting for randomization arm. Bonferroni correction was used to account for multiple testing, with a cut-off of 0.05 divided by number of polymorphisms tested in each targeted analysis, and P = 5.0 × 10 −8 for genome-wide analyses. For genomewide significant associations we examined how well these would cross-replicate internally by randomly dividing the dataset into five groups of near equal size. We tested for association on four-fifths (training set), and for replication on the remaining one-fifth (testing set). We repeated this a total of five times, once for each training-testing set pair, and assessed how consistently the association replicated.

Results
Of the 702 ART-naive participants who were randomly assigned to the two dolutegravir-containing arms of ADVANCE, 336 were included in our study: 173 (51.5%) in the TAF group and 163 (48.5%) in the TDF group. All participants were Black Southern Africans (55.6% South African, 38.4%, Zimbabwean, and 6% from Lesotho, Malawi, Mozambique, Swaziland, or Zambia), the median age was 32 years, most were female, and median BMI was 23.4 kg/m 2 . Baseline characteristics were similar between the TAF and TDF groups (Table 1).
As a sensitivity analysis, we repeated the above analyses but based on the change in eGFR from screening (rather than week 4) to week 48. In this analysis, no associations withstood correction for multiple testing. Inclusion of cotrimoxazole use as a covariate did not substantially alter genetic associations.

Genetic associations with change in log-transformed urinary β 2 -microglobulin/creatinine ratio
We next characterized associations with ΔB2M/Cr 0-48 . Among 14 candidate polymorphisms previously reported to be associated with TAF, TDF and tenofovir disposition or renal toxicity, the lowest P value for association with Manhattan plots of genome-wide associations with change in estimated glomerular filtration rate from week 4 to week 48. The figure shows −log 10 P values for association among 300 individuals who were evaluable for genetic associations. The black arrow indicates the lowest P value. The lowest P value was COL27A1 rs168402 (P = 3.4 × 10 −9 ). Two other COL27A1 polymorphisms in strong linkage disequilibrium with rs168402 were also genome-wide significant.   Manhattan plots of genome-wide associations with change in log-transformed urinary retinol-binding protein/creatinine ratio from baseline to week 48. The figure shows −log 10 P values for association among 262 individuals who were evaluable for genetic associations. The black arrow indicates the lowest P value. The lowest P value was CDH4 rs66494466 (P = 5.6 × 10 −8 ). No polymorphisms withstood correction for multiple testing. previously associated with any GWAS Catalog trait, the lowest P value for association with ΔB2M/Cr 0-48 was ITG4 rs1143676 (β = −0.52, P = 2.2 × 10 −5 ). Considering genome-wide associations regardless of the GWAS Catalog, the lowest P value for association with ΔB2M/ Cr 0-48 was ITG4 rs3770126 (β = −0.79, P = 6.1 × 10 −7 ) (Fig. 3). In the above analyses for ΔB2M/Cr 0-48 , no polymorphism withstood correction for multiple testing. The five lowest P values for association with ΔB2M/Cr 0-48 from each prioritized analysis are presented in Table 4. Inclusion of cotrimoxazole use as a covariate did not substantially alter genetic associations.

Associations by randomized study arm
Our study included randomized arm in ADVANCE (TAF or TDF) as a covariate. In our multivariable analyses, randomized study arm was not independently associated with any of the renal outcomes examined.

Discussion
We characterized genetic associations with renal tubular dysfunction among ART-naive Southern African participants who were randomized to the two dolutegravir arms (plus either TAF and emtricitabine, or TDF and emtricitabine) of the ADVANCE clinical trial. Among 14 polymorphisms that we prioritized based on reported associations with TDF renal toxicity or plasma tenofovir exposure, we found nominal associations with two polymorphisms previously reported -ABCC4 rs899494 with ΔeGFR 4-48 , and ABCC4 rs1059751 with ΔB2M/Cr 0-48 ; however, these associations did not withstand correction for multiple comparisons and were in the opposite direction to those previously reported, as discussed below. In addition, three polymorphisms in COL27A1, which encodes collagen type XXVII alpha 1 chain, were genome-wide associated with change in eGFR.
Our primary analyses focused on polymorphisms previously reported to be associated with TDF-related renal toxicity in other populations. Prior studies have implicated multidrug resistance proteins, encoded by ABCC family members, in tenofovir clearance and TDFrelated renal toxicity [14,23,30,43]. We found that ABCC4 rs1059751 was nominally associated with ΔB2M/Cr 0-48 (P = 0.009). A previous report associated rs1059751 T→C (at position 4976 of ABCC4) with β 2 -microglobulinuria in Thai patients treated with TDF-containing ART [44]. In that study, the rs1059751 C allele was associated with β 2 -microglobulinuria, with allele frequency of 0.60 in cases and 0.48 in controls. In contrast, the present study associated the rs1059751 G allele (equivalent to C allele in the Thai study) with a decrease in urinary β 2 -microglobulin (β = −0.37), with a MAF of 0.19. The negative association indicates that participants with the minor G allele in our study were less likely to experience renal tubular dysfunction (increased β 2 -microglobulinuria). We also found that ABCC4 rs899494 G→A was nominally associated ΔeGFR 4-48 (P = 0.02). This polymorphism was previously associated with renal phosphorus wasting in Spanish patients treated with TDF-containing ART [23]. In that study, the CC genotype was associated with worsening of the renal marker; however, for the present study, the A allele (equivalent to T allele in the Thai study) was associated with greater eGFR decline, which may be indicative of renal tubular dysfunction [45]. We did not replicate associations with ABCC2 rs2273697 and ABCC10 rs2125739, which were previously reported to be associated with altered amino acid excretion [23] and with kidney tubular dysfunction, respectively [30].
In the present study, several COL27A1 polymorphisms were genome-wide associated with change in ΔeGFR  . To help assess whether this was by chance alone, we performed internal cross-replication, which showed that the association replicated in most, but not all, paired Manhattan plots of genome-wide associations with change in log-transformed urinary β 2 -microglobulin/creatinine ratio from baseline to week 48. The figure shows −log 10 P values for association among 290 individuals who were evaluable for genetic associations. The black arrow indicates the lowest P value. The lowest P value was ITGA4 rs3770126 (P = 6.1 × 10 −7 ). No polymorphisms withstood correction for multiple testing.
training-testing datasets. This gene encodes collagen type XXVII alpha 1 chain [46], and mutations in COL27A1 have been associated with Steel syndrome, knee osteoarthritis and Achilles tendinopathy [47][48][49]. Although COL27A1 is expressed in renal tubules [50], it is unclear how this gene might contribute mechanistically to tenofovir renal toxicity.
To identify novel polymorphisms associated with renal outcomes, we leveraged imputed genome-wide data against the considerable knowledge generated by prior genetic association studies represented in Pharmacogenomics Knowledgebase and the GWAS Catalog. We reasoned that those polymorphisms associated with at least one drug-related phenotype in the Pharmacogenomics Knowledgebase with levels of evidence of 1 or 2 (as described in Methods) or with any trait in the GWAS Catalog at P < 5.0 × 10 −8 in at least one published study, would most likely be true positives in the present study. We did not identify any such associations that withstood correction for multiple testing.
Our study has some limitations. A larger sample size would have increased power to detect associations with modest effect size, or associations with infrequent variants. We did not adjust for concomitant drugs that are potentially nephrotoxic, which may have obscured genetic associations. The study was performed exclusively in Black people living with HIV, who were previously shown to be at lower risk of renal tubular dysfunction [51] and TDF-associated renal tubulopathy [45]; our results can therefore not be extrapolated to other populations. The relatively short exposure (48 weeks) and use of an unboosted third agent in this young population may have contributed to the minimal effects of tenofovir on renal tubular function, and thus our ability to identify genetic associations with the three measures of renal tubular dysfunction. In our relatively young cohort, we were unable to adjust for pre-existing hypertension, diabetes or cardiovascular disease.
In conclusion, we found associations with two polymorphisms previously reported to be associated with TDF-induced renal toxicity, although in our study the associations did not withstand correction for multiple comparisons and were in the opposite directions than previously reported. We also identified genome-wide significant associations with change in eGFR and polymorphisms in COL27A1. These findings contribute to our understanding of tenofovir-related renal toxicity. Significance threshold was 5 × 10 −8 for the genome-wide analysis. f We used a P value cut-off of P < 1 × 10 −6 for GWAS SNP filtering. Only 1SNP was associated with ΔB2M/Cr 0-48.