Genome-Wide Association Study in Acute Tubulointerstitial Nephritis

Abstract

Significance Statement

Polymorphisms of HLA genes may confer susceptibility to acute tubulointerstitial nephritis (ATIN), but small sample sizes and candidate gene design have hindered their investigation. The first genome-wide association study of ATIN identified two significant loci, risk haplotype DRB1*14-DQA1*0101-DQB1*0503 (DR14 serotype) and protective haplotype DRB1*1501-DQA1*0102-DQB1*0602 (DR15 serotype), with amino acid position 60 in the peptide-binding groove P10 of HLA–DRβ1 key. Risk alleles were shared among different causes of ATIN and HLA genotypes associated with kidney injury and immune therapy response. HLA alleles showed the strongest association. The findings suggest that a genetically conferred risk of immune dysregulation is part of the pathogenesis of ATIN.

Background

Acute tubulointerstitial nephritis (ATIN) is a rare immune-related disease, accounting for approximately 10% of patients with unexplained AKI. Previous elucidation of the relationship between genetic factors that contribute to its pathogenesis was hampered because of small sample sizes and candidate gene design.

Methods

We undertook the first two-stage genome-wide association study and meta-analysis involving 544 kidney biopsy-defined patients with ATIN and 2346 controls of Chinese ancestry. We conducted statistical fine-mapping analysis, provided functional annotations of significant variants, estimated single nucleotide polymorphism (SNP)-based heritability, and checked genotype and subphenotype correlations.

Results

Two genome-wide significant loci, rs35087390 of HLA-DQA1 (P=3.01×10⁻³⁹) on 6p21.32 and rs2417771 of PLEKHA5 on 12p12.3 (P=2.14×10⁻⁸), emerged from the analysis. HLA imputation using two reference panels suggested that HLA-DRB1*14 mainly drives the HLA risk association. HLA-DRB1 residue 60 belonging to pocket P10 was the key amino acid position. The SNP-based heritability estimates with and without the HLA locus were 20.43% and 10.35%, respectively. Different clinical subphenotypes (drug-related or tubulointerstitial nephritis and uveitis syndrome) seemed to share the same risk alleles. However, the HLA risk genotype was associated with disease severity and response rate to immunosuppressive therapy.

Conclusions

We identified two candidate genome regions associated with susceptibility to ATIN. The findings suggest that a genetically conferred risk of immune dysregulation is involved in the pathogenesis of ATIN.

Introduction

Acute tubulointerstitial nephritis (ATIN) is an entity that is characterized by histologic findings of interstitial inflammatory cellular infiltrate, variable degrees of tubular injury, interstitial edema, and tubulitis. Tubulointerstitial lesions that frequently accompany primary glomerulonephritis are usually not included within ATIN. In patients undergoing kidney biopsy for any indication, the incidence of ATIN ranges from 1% to 3%.^1–4 However, ATIN accounts for 15%–27% of the patients with AKI requiring kidney biopsy.^5–7 Considering different etiologies, it is reported that drug-induced ATIN (D-ATIN) accounts for more than 70% of the patients, infection-related ATIN for 15%, idiopathic forms for 10%, and tubulointerstitial nephritis and uveitis (TINU) for 4%, the remaining ones being associated with systemic disorders.^8–10 To date, routine clinical assessment, laboratory data, and imaging tests are always insufficient to make a definite diagnosis or monitor disease activity. As a result, kidney biopsy is often required to diagnose and guide accurate management. Further investigations of pathogenesis are still warranted.^11–16

The pathogenesis of ATIN is believed to have an immunologic basis, considered to involve primarily a T-cell–driven process.^10,17–20 Immune-mediated kidney injury is orchestrated by various CD4⁺ T-cell subsets and varies depending on the inciting agent, suggesting that ATIN may be the final common pathway of distinct mechanisms of injury.

Because ATIN remains a relatively rare disease in the general population even with the same exposure factors, it may suggest a role for patient-specific risk factors such as genotypic variations in susceptibility.^17,21 For example, it was reported that an incidence of proton pump inhibitor (PPI)-associated ATIN was 1.1/10,000 person-years of PPI exposure.¹⁷ In addition, there have been some patients reporting that TINU can be observed with identical histological features and similar clinical symptoms in monozygotic twins or siblings.^22–24 Several studies have reported HLA associations in TINU.^25–29 Apart from polymorphisms in the cytochrome P450 enzyme gene in D-ATIN, certain single nucleotide polymorphisms (SNPs) in HLA or cytokine genes (i.e., IL1, CTLA4) may confer susceptibility to the disease according to pathophysiological basis.^26,30–34 With 154 Chinese patients with biopsy-proven ATIN and 200 healthy controls, we previously reported that DQA1*0104, DQB1*0503, and DRB1*1405 alleles were shared to be associated with D-ATIN and TINU,²⁶ suggesting a common mechanism.

However, the pathogenesis of tubulointerstitial nephritis is still far from clear. Because the diagnosis requires a kidney biopsy, genetic discoveries have been hindered by small sample sizes and candidate gene design. To evaluate evidence for genetic association with ATIN, we undertook the first two-stage genome-wide analysis for ATIN enrolling 544 kidney biopsy-defined patients and 2346 controls of Chinese ancestry.

Methods

Inclusion Criteria

The renal lesion of ATIN defined as “an acute inflammation of the kidney characterized by cellular and fluid exudation in the interstitial tissue, accompanied by, but not dependent on, degeneration of the epithelium; at least 50% of the parenchymal area had interstitial inflammation, with <25% of the area having interstitial fibrosis.” We included the diagnostic criteria of ATIN by confirmatory histologic findings. To assess patient eligibility for entry to this study, in each pathology center, at least two nephrologists and at least two pathologists reviewed each patient. Patients who had secondary tubulointerstitial disease, glomerular diseases, hereditary kidney diseases, or malignancy were excluded from the study.

Subjects, Genotyping, and Quality Control

In the discovery stage, 299 patients with ATIN and 1991 healthy controls were all recruited from Peking University First Hospital from 1998 to 2020. All the samples were genotyped using the Illumina Infinium Global Screening Array-24 v1.0 (GSA, a panel of 700,000 tagging SNPs) BeadChip, according to the manufacturer's protocol. Genotype calling was performed using GenomeStudio software v.2.0.5 (https://emea.illumina.com/techniques/microarrays/array-data-analysis-experimental-design/genomestudio.html). After genotype calling, genetic data were processed in the following sequence using PLINK v2.0.³⁵ Samples with a sex mismatch, abnormal heterozygosity, or proportion of missing variants >2% were removed. Variants with low minor allelic frequency (<1%), missing calls >5%, deviation from the Hardy-Weinberg equilibrium (P value <1×10⁻⁴), or which were monomorphic were removed. Close relatives were identified as having an estimated identity-by-descent proportion of genome-wide shared alleles of >0.1875. Variants that showed significant differential missingness (Benjamini-Hochberg adjusted P value <0.05) between patients and controls were removed. Sample outliers were checked on the basis of the principal component analysis method using Eigensoft 7.2.1 tools.³⁶ Patients or controls lying outside ±3 standard deviations from the mean on either PC1 or PC2 were removed (Supplemental Figure 1). Following these QC steps, 285 patients and 1838 controls with 574,085 autosomal variants remained.

In the replication stage, genotyping of 276 patients from multiple Chinese centers and 878 geographically matched controls was obtained by the Illumina Infinium Asian Screening Array-24 v1.0 (ASA) BeadChip. Two hundred fifty-nine patients, 508 controls, and 447,562 autosomal genetic variants passed QC.

Clinical and laboratory data were obtained by review of clinical records. All individuals provided written informed consent. All relevant ethical guidelines have been followed, and any necessary institutional review board (IRB) and/or ethics committee approvals have been obtained. This study was approved by the IRB of Peking University First Hospital (IRB number: 2021-Y042). DNA samples genotyped were ascertained under the following ethical approvals: RJ/2011/5; HS/2016/394; and PUFH/2017/1280.

Imputation

Imputation was performed using the Michigan Imputation Server (v1.2.4) with the Genome Asian Pilot individuals as a reference panel.³⁷ Prephasing was first performed with Eagle v2.4 with array build of GRCh37/hg19. We were restricted to all imputed common variants with minor allele frequency >1%, Hardy-Weinberg equilibrium P value >1×10⁻⁴, and imputation quality (INFO) r²>0.8. Three million forty-four thousand forty-four autosomal variants in the GSA chip and 3,046,661 variants in the ASA chip passed QC for meta-analysis.

For HLA, we extracted the SNPs within the extended MHC region (chr6: 25–34 MB) and imputed classical alleles, untyped SNPs, and amino acid residues using SNP2HLA with the newly developed Asian HLA imputation panel.³⁸ We applied the same postimputation quality control criteria. To assess imputation and association validity, we also imputed using Minimac4 with a newly released multiancestry MHC reference panel (n=21,546) by the same QC procedure, which included 2069 East Asians on the basis of whole-genome sequences.³⁹

Genetic Associations

Case-control association testing was performed using logistic regression assuming an additive model. Covariates included age, sex, and the first two principal components. The first two PCs were selected on the basis of assessment of significance of eigenvalues in the Tracy-Widom statistic (P<0.1).⁴⁰ The genomic inflation factor was calculated. We analyzed both cohorts separately and then performed sample size-weighted inverse-variance meta-analysis in METAL.⁶ P≤5×10⁻⁸ was used as the threshold for genome-wide significance, with consistently significant associations with P<5×10⁻² in both discovery and replication cohorts with the same direction of effect.

The narrow-sense heritability (h²_SNP) of ATIN was estimated using LDSC.⁴¹ We estimated the variance explained by groups of associated variants using the following formula⁴²:

$${\displaystyle \sum }_{i=1}^{n}2\rho i\left(1-\rho i\right)\left({\widehat{\beta }\mathrm{i}}^2-{\tau i}^2\right)$$

where n is the number of variants, $\rho$ i is the frequency of the variant, $\widehat{\beta }$ i is the log odds ratio (OR) of the identified variant, and $\tau i$ is the estimated standard error for that variant's effect size.

We performed biallelic tests of association for all individual SNPs, classical HLA alleles, and individual amino acid residues in HLA. Stepwise conditional analysis was performed to identify independent signals within each type of variant, and the results were meta-analyzed.⁴³ This was performed by including the genotype of conditioning the top SNP under additive coding as a covariate in the outcome model.⁴⁴ The conditional analyses were performed individually within each cohort and with adjustments for age, sex, and the first two PCs. Subsequently, the conditioned summary statistics were combined using METAL, similar to our primary association analyses. The analysis was repeated until no variant reached the level of statistical significance, which we set to be P_meta≤5×10⁻⁸. We also performed an omnibus test (log-likelihood ratio test) for each multiallelic marker. An omnibus P value was calculated by a log-likelihood ratio test comparing the likelihood of the null model against the likelihood of the fitted model, as described elsewhere.^38,45 Haplotype frequency was estimated by the standard expectation maximization algorithm in the haplo/stats library in R.

To further confirm genetic association, we also included kidney disease control. An original genome-wide association study (GWAS) dataset from IgA nephropathy with the similar sample size (1112 Chinese patients with IgA nephropathy and 1240 controls) was checked. The IgA nephropathy GWAS took the same genotyping array (GSA chip) and was also recruited from the same center. These samples have been described more recently.⁴⁶ To check whether loci were disease-specific, colocalization of summary statistics in the HLA-II region (chr6, 32–33 MB) was conducted and visualized by LocusCompare.⁴⁷ Statistics on membranous nephropathy (1632 patients and 3209 controls) for HLA from East Asian (ebi-a-GCST010005, https://gwas.mrcieu.ac.uk/) were also retrieved.⁴⁴

Variant Annotations

We used VARAdb v1.0 to perform functional annotations of our top associated variants.⁴⁸ 3DSNP v2.0 was checked to link noncoding variants to their interacting genes through 3D chromatin loops.⁴⁹ To identify regulatory effects of genetic variants on mRNA transcript (eQTL), we used gene expression data from blood (eQTLGen, n=31,684), kidney tissue (NephQTL n=187; or Human Kidney eQTL Atlas, n=686), and different tissue sites (GTEx V8, n=948).^50–52 Genetic associations with protein (pQTL) abundance were checked against the data from the Systematic and Combined AnaLysis of Olink Proteins consortium (n=70,000).⁵³

Statistical Analyses

For clinical data, statistical analysis was performed using SPSS 23.0 statistical software (SPSS, Chicago, IL). Normally distributed variables were expressed as the mean±SD and were compared using the t test. Nonparametric variables were expressed as the median and interquartile range and were compared using the Mann-Whitney U test.

Results

Summary of the Studied Subjects

Five hundred seventy-five individuals were recruited with a clinical diagnosis of ATIN from several centers in China. In total, 544 patients were included in the GWAS after quality control (Figure 1). Of these, 318 (58.5%) were female and 226 (41.5%) male. The mean age of the patients was 49±15 years (range 12–80 years). Clinical information can be referred to in Table 1. All involved controls (2869 individuals) were healthy population controls from blood donors, and any individuals with a known diagnosis of kidney disease (serum creatinine [sCr] range 0.47–1.22 mg/dl), immune disease, or current infection were excluded. Two thousand three hundred forty-six controls were left in analysis after quality control (45.5% female, age 36±10 years).

Figure 1.

A flow diagram of building the study cohort. Five hundred seventy-five individuals were previously recruited with a clinical diagnosis of ATIN. Three of the largest renal pathology centers in China (North of China, Peking University First hospital; South of China, Ruijin Hospital and Huashan Hospital) provided diagnostic pathology. To assess patient eligibility for entry to this study, in each center, at least two nephrologists and at least two pathologists reviewed each patient. After quality controls, 544 patients were included in the final GWAS analysis.

Table 1.

Clinical data for patients with acute tubulointerstitial nephritis in this genome-wide association study

Clinical Parameters	GSA Chip N=285	ASA Chip N=259
Age (yr)	47.88±14.24	49.69±14.89
Sex, men (%)	111 (38.9)	115 (44.4)
Previous underlying diseases, n (%)
Hypertension	100 (35.1)	98 (37.8)
Diabetes mellitus	57 (20.0)	32 (12.4)
sCr (mg/dl)
At kidney biopsy	2.47 (1.71–3.77)	2.77 (1.75–4.83)
At peak	3.14 (2.13–5.32)	3.10 (1.98–5.86)
AKI/AKD, n (%)	244 (85.6)	237 (91.5)
Urine NAG (U/L)	23 (14–39)	13 (9–26)
Urine α1-MG (mg/L)	129 (44–210)	77 (28–159)
24 h proteinuria	0.64 (0.26–1.13)	0.44 (0.20–0.97)
The most likely cause, n (%)
D-ATIN	158 (55.4)	101 (39.0)
Autoimmunity-related ATIN	69 (24.2)	14 (5.4)
IgG4-related disease	6 (2.1)	4 (1.5)
Sjogren syndrome	16 (5.6)	5 (2.0)
TINU syndrome	37 (13.0)	3 (1.2)
Other ADs	10 (3.5)	2 (0.8)
Systemic diseases	2 (0.7)	11 (4.2)
Infection-associated	7 (2.5)	5 (1.9)
Idiopathic or other undetermined causes	49 (17.2)	128 (49.4)
Steroid or immunosuppressants, n (%)	238 (83.5)	NA
Renal outcome at discharge, n (%)
Complete remission/Partial remission	195 (68.4)	NA
No recovery at 6 mo	90 (31.6)	NA

Data are presented as mean±SD, n (%), or median (interquartile range). Serum creatinine: normal range, 0.5–1.5 mg/dl; N-acetyl-β-D-glucosaminidase: normal range, 0–21 U/L; α1-microglobulin: normal range, 0–12 mg/L. Kidney biopsy is generally required to make a definitive diagnosis. Multiple causes of acute tubulointerstitial nephritis exist, including drugs, infections, and systemic and idiopathic diseases. Drugs, particularly antibiotics, proton pump inhibitors, and nonsteroidal anti-inflammatory drugs are the most common cause of acute tubulointerstitial nephritis. Autoimmune diseases include IgG4 TIN, Sjogren syndrome, tubulointerstitial nephritis and uveitis, and others (sarcoidosis, SLE, vasculitis). Systemic diseases include lymphoproliferative diseases and paraproteinemia. Immunosuppressants includes cyclophosphamide and mycophenolate mofetil. Complete remission is defined as serum creatinine recovered back to baseline; partial remission is defined as serum creatinine decreased over 30% of the peak value. GSA, global screening array-24 v1.0; ASA, Asian screening array-24 v1.0; sCr, serum creatinine; AKD, acute kidney disease; NAG, N-acetyl-β-d-glucosaminidase; α1-MG, α1-microglobulin; D-ATIN, drug-induced ATIN; ATIN, acute tubulointerstitial nephritis; TINU, tubulointerstitial nephritis and uveitis; AD, autoimmune disease.

Summary of the GWAS

With power calculation, using GWAS, we would have provided >80% power to detect association with common variants (Supplemental Figure 2). In both stages, there were no significant population stratifications, with the genomic inflation factor lambda of 1.03 in the discovery stage, 1.01 in the replication stage (Supplemental Figure 3), and 1.03 in the meta-analysis. The results of genome-wide meta-analyses are presented in Figure 2 and Table 2. With four covariates corrected, we discovered two genome-wide significant loci in meta-analysis: a locus on chromosome 6p21.32 encoding MHC class II genes spanning HLA-DRB1, DQA1, and DQB1 (rs35087390, OR, 4.23, 95% confidence interval [CI], 3.41 to 5.25, P_meta=3.01×10⁻³⁹) and a locus on chromosome 12p12.3 encoding PLEKHA5 (rs2417771, OR, 1.64, 95% CI, 1.38 to 1.96, P_meta = 2.14×10⁻⁸).

Figure 2.

The genome-wide association results in ATIN. (A) Manhattan plot. Observed −log₁₀P values versus chromosomal location. The y-axis represents the −log₁₀P value for association of SNPs with ATIN. The threshold for genome-wide significance (P<5×10⁻⁸) is represented by a horizontal red line. Genome-wide significant variants in the two genome-wide significant loci are in blue. (B) Quantile-quantile plot. Quantile-quantile plot of the association before (red, λ=1.028) and after (blue, λ=1.021) removing HLA (–chr 6 –from-kb 25,000 –to-kb 34,000). Figure 2 can be viewed in color online at www.jasn.org.

Table 2.

Genetic associations with acute tubulointerstitial nephritis with genome-wide significance (P<5×10⁻⁸) in the genome-wide association study meta-analysis

SNP	Chr: Position (hg19)	Loci	Risk Allele	Discovery Cohort (GSA)				Replication Cohort (ASA)				Meta-Analysis
				RAF				RAF
				Case	Control	P	OR	Case	Control	P	OR	P meta	OR	I2	P Q
rs35087390	6:32610437	HLA-DQA1	A	26.14	5.67	2.92×10−36	5.62 (4.29–7.35)	16.67	5.73	4.32×10−7	2.54 (1.77–3.64)	3.01×10−39a	4.23 (3.41–5.25)a	91.7	5.38×10−4
rs2417771	12:19350081	PLEKHA5	A	79.9	70.61	1.86×10−5	1.64 (1.30–2.04)	76.88	69.39	3.04×10−4	1.67 (1.27–2.17)	2.14×10−8a	1.64 (1.38–1.96)a	0	0.91

Variant is presented as both dbSNP rs number and chr:position in build37/hg19. The genetic association for the single variant was conducted using logistic regression assuming an additive model. Age, sex, and the first two principal components were included as covariates. Meta-analysis was conducted using METAL. SNP, single nucleotide polymorphism; RAF, risk allele frequency; GSA, global screening array-24 v1.0; ASA, Asian screening array-24 v1.0; MAF, minor allele frequency; OR, odds ratio; P_Q, P value of heterogeneity test. All the P values were from analysis corrected for age, sex, and the first two PCs.

^aIndicate results from meta-analysis.

HLA Associations

Imputation resulted in 80 two-digit classical alleles, 120 four-digit classical alleles, 540 amino acids in HLA proteins, and 46,535 SNPs for fine-mapping analysis. There were 931 SNPs significantly associated with ATIN in the MHC region (P_meta<5×10⁻⁸). The lead SNP rs35087390 was a missense variant (c.664G>A, p.Ala222Thr) in HLA-DQA1. In the analysis conditioning on rs35087390, no variant reached genome-wide (<5×10⁻⁸) or regional significance (<1.06×10⁻⁶, 0.05/47,195 after Bonferroni correction) (Figure 3 and Supplemental Table 1).

Figure 3.

Results of meta-analyses of HLA signals and location of associated amino acid positions in ATIN. (A) Association results of the unconditional analysis (upper) of SNPs (gray), HLA classical alleles (orange), and amino acid residues (AA, red) and results of the same variants after conditioning (below) on lead SNP of rs35087390. Each point corresponds to the P_Meta value in each type of variant. The figures presented were drawn by HLA-TAPAS (https://github.com/immunogenomics/HLA-TAPAS).³⁹ (B) The key amino acid positions identified by association analysis are highlighted in three-dimensional ribbon models for the HLA-DR proteins. All protein structures are positioned to accommodate the view of the peptide binding groove and the associated amino acid residues using UCSF Chimera. The amino acids at positions 60 and 57 are located in the peptide-binding groove of HLA-DRB1. Figure 3 can be viewed in color online at www.jasn.org.

On the basis of peptide presentation, associations of HLA alleles may offer a more relevant biological explanation. For the classical alleles, the association signal was identified in HLA class II genes including HLA-DQB1 (P_omnibus=2.68×10⁻¹³), HLA-DRB1 (P_omnibus=1.04×10⁻¹¹), and HLA-DQA1 (P_omnibus=7.94×10⁻¹⁰) (Supplemental Table 2). Condition on any one of them will diminish the independent role of other loci. For specific risk alleles, the strongest associated one was HLA-DQB1*0503 (P_meta=1.27×10⁻³⁷, OR, 4.61, 95% CI, 3.65 to 5.82), followed by HLA-DRB1*14 (P_meta=9.90×10⁻³⁶, OR, 3.71, 95% CI, 3.02 to 4.56) and HLA-DQA1*0101 (P_meta=8.95×10⁻²⁸, OR, 2.72, 95% CI, 2.27 to 3.25) (Supplemental Table 3). Negative associations were observed with alleles DQB1*06 (P_meta=8.52×10⁻⁸, OR, 0.60, 95% CI, 0.49 to 0.72), DQA1*0102 (P_meta=1.68×10⁻⁶, OR, 0.60, 95% CI, 0.49 to 0.74), and DRB1*15 (P_meta=2.90×10⁻⁶, OR, 0.58, 95% CI, 0.46 to 0.73). The lead SNP rs35087390 was in linkage disequilibrium (LD) with HLA-DQB1*0503 (r²=0.72, D′=0.95), HLA-DRB1*14 (r²=0.88, D′=0.99), and HLA-DQA1*0101 (r²=0.61, D′=1.00). Haplotype analysis showed similar results, producing a risk haplotype DRB1*14-DQA1*0101-DQB1*0503 (DR14 serotype, frequency in case 18.09% versus control 4.22%, P=9.48×10⁻⁵⁷, OR, 5.03, 95% CI, 4.03 to 6.28) and a protective DRB1*1501-DQA1*0102-DQB1*0602 haplotype (DR15 serotype, frequency in case 5.87% versus control 9.22%, P=4.10×10⁻⁶, OR, 0.60, 95% CI, 0.46 to 0.79). In the conditional analysis on HLA-DQB1*0503 or HLA-DRB1*14, no additionally classical allele passed the genome-wide (5×10⁻⁸) or regional P value threshold (1.06×10⁻⁶).

In the analysis of polymorphic amino acid sites, only HLA-DRB1 and HLA-DQB1 had genome-wide significant amino acid residue. Among them, HLA-DRB1 residue 57-Ala and 60-His showed the strongest signal (both P=4.25×10⁻²⁶) (Supplemental Table 4). Both r² and D′ between 57-Ala and 60-His were 1.0. After conditioning amino acid residue 60-His in the model, HLA-DRB1 residue 140-Thr was observed to be independently associated (P=4.84×10⁻⁹) (Supplemental Table 5). Furthermore, stepwise conditional analysis identified no additionally independent residue. The results together suggested that the HLA-DRB1 gene may be mainly associated with susceptibility to ATIN. Because of the great diversity of amino acids at many positions, we analyzed the combined significance (P_omnibus) of all amino acids at each position (Supplemental Table 6). There were 14 of 37 polymorphic residue positions in HLA-DRB1, and seven of 39 polymorphic residue positions in HLA-DQB1 reached genome-wide significance. When conditioning on the specific HLA amino acid position(s), we included the multiallelic variants of the amino acid residues. HLA-DRB1-60 (P_omnibus=8.97×10⁻¹⁴) and HLA-DRB1-57 (P_omnibus=6.38×10⁻¹³) were the top associated multiallelic amino acid positions. Condition on DRB1 residue 60 diminished significant associations of any other residues (Supplemental Table 7). For DRB1 residue 60, the presence of histidine (H) (P=4.25×10⁻²⁶, OR, 3.30, 95% CI, 2.64 to 4.11) conferred a high risk, whereas the presence of serine (S) (P=3.57×10⁻⁶, OR, 0.69, 95% CI, 0.59 to 0.81) or tyrosine (Y) (P=0.03, OR, 0.85, 95% CI, 0.74 to 0.99) was protective against ATIN (Supplemental Table 8).

Imputation using a second reference panel resulted in genotype concordance >90% and similar associations (r²>0.99) (Supplemental Methods and Results, Supplemental Table 9, Supplemental Figure 4). The amino acids at positions 60 and 57 were located in the peptide-binding groove of HLA-DRβ1 (Figures 3B and 4). Both of them have been shown to be associated with rheumatoid arthritis, type 1 diabetes, and leprosy in Asian populations.^54–57 Although position 60 displayed a more significant association compared with 57 in some studies, 57 was more often considered to be the causal one in the literature.⁵⁸ Owing to perfect LD between them, we just highlighted position 60 here. The substitution of Tyr60 to His at position 60 strengthened the positive charge distribution and created an empty volume causing conformational changes on the molecular surface (Figure 4).

Figure 4.

Representation of molecular surface change for amino acid at position 60 of HLA-DRB1. The molecular graphics program PyMOL was used to analyze the changes caused by the variant and the molecular surface charge distribution and to display the results (with PDB code 3pdo). Right upper for the structure for Tyr60 and lower for the His60 variant. The electrostatic surface was shown in color varying from blue (positive) to red (negative). Protein structural elements involved were shown as ribbons and Tyr60 and His60 as sticks with the rest of the side chains as lines. The substitution of Tyr60 to His60 strengthened the positive charge distribution on the molecular surface. The shorter His side chain created an empty volume causing conformational changes of the molecular surface. PDB, protein data bank. Figure 4 can be viewed in color online at www.jasn.org.

HLA rs35087390 (c.664G>A, p.Ala222Thr) was also a missense variant located in exon four of HLA-DQA1. It was predicted as damaging by SIFT and CADD. It was located in the helical transmembrane domain predicted by AlphaFold, but without Protein Data Bank structure known. Thus, it may not be a better candidate compared with DRB1 residue 60 in the role of antigen presentation. However, rs35087390 was reported to exert a trans-pQTL effect on plasma cathepsin L1 (CTSL1), CXC-type chemokine ligand 16 (CXCL16), and IL6 levels (P<5×10⁻⁸) (Supplemental Table 10) and was associated with tens of diseases, including asthma, membranous nephropathy, sarcoidosis, and SLE.

Novel Association with PLEKHA5

The second genome-wide significant locus in meta-analysis was an intronic variant rs2417771 of PLEKHA5 (Pleckstrin Homology Domain Containing A5). PLEKHA5 was highly expressed in immune and renal systems (Supplemental Figure 5). Condition on rs2417771 revealed no additional independent associations (Supplemental Figure 6). It was annotated as a regulatory variant. We assessed its effects on mRNA (eQTL) and protein (pQTL) levels in whole blood and tubulointerstitium. It was associated with plasma vascular endothelial growth factor-A and CX3CL1 levels (Supplemental Table 10), which were reported to be elevated in AKI.^59,60 Its genomic location was in the enhancer state, colocalized with epigenetic markers such as H3K27ac and Assay for Transposase-Accessible Chromatin peaks in several cell types including renal cortical epithelial cells, CD14-positive monocytes, T helper 2 cells, and kidney myeloid cells (Supplemental Figure 7), suggesting that PLEKHA5 might be functionally relevant for inflammation.

Confirmation of Association by Taking IgA Glomerulonephritis as Disease Controls

For further confirmation, we included a set of participants with known IgA nephropathy as additional controls. For data consistency and comparability, we only examined the IgA nephropathy GWAS cohort recruited from the same center with similar sample size and was genotyped by the same array (GSA) in the same laboratory (Supplemental Methods and Results, Supplemental Table 11). Compared with healthy controls, only HLA was associated with IgA nephropathy within this dataset, which was similar to findings in early GWASs in IgA nephropathy^61,62 (Supplemental Figures 8 and 9). By taking IgA nephropathy as controls (285 ATIN patients versus 1112 IgA nephropathy controls), the most significant locus was still the HLA region (rs112025116, 2.5kb 5′ of HLA-DQA1, P=4.48×10⁻⁵¹, OR, 23.73, 95% CI, 15.70 to 35.86), but with a larger effect size. The associations signals colocalized well between using healthy controls and using IgA nephropathy disease controls (Supplemental Figure 10A). ATIN association signals that colocalization also applied in the PLEKHA5 locus (rs34495079, intronic variant, P=1.67×10⁻⁴, OR, 2.25, 95% CI, 1.48 to 3.43) (Supplemental Figure 10B). Because rs35087390 and rs2417771 were absent in the IgA nephropathy dataset, we checked the LD and found that both variants (rs112025116 and rs34495079) were in high LD with the above two top SNPs (both D′=1.0).

We also checked whether the HLA signals in ATIN were shared with other kidney diseases. Compared with healthy controls, associations were not colocalized well among ATIN, IgA nephropathy, and membranous nephropathy (Supplemental Figure 10, C and D). It may suggest certain disease-specific effects. However, a single cohort for each disease, low density of shared variants, and moderate sample size may weaken this conclusion.

Heritability

To quantify the genetic influence on ATIN, the “SNP heritability” (h²_SNP)⁶³ captured by all the SNPs was estimated: The SNP-based heritability estimates with and without the HLA locus were 20.43% (all SNPs) and 10.35% (non-HLA SNPs). The lead HLA variant rs35087390 alone accounts for 3.90% of the explained heritability. At the genome-wide scale, ATIN showed no genetic correlation with IgA nephropathy (genetic correlation coefficient −0.39, P=0.41).

Subphenotype Analysis and Clinical Correlations

GWAS analysis for clinically defined subphenotype, including D-ATIN (n=259), and autoimmunity-associated ATIN (AD-ATIN, n=83) is described in Supplemental Figure 11. When compared among D-ATIN, AD-ATIN, or TINU at the genome-wide level, no statistically differential signals can be observed (Supplemental Table 12).

We also evaluated the clinical correlations of associated variants in ATIN. We only observed significant correlations for the HLA variant. Patients of risk genotypes of rs35087390 showed higher sCr both at biopsy and peak and higher urinary biomarker levels including proteinuria, N-acetyl-β-d-glucosaminidase, and α1-microglobulin (Figure 5). Patients carrying the risk genotypes tended to be treated with more aggressive therapy using corticosteroids (with/without mycophenolate mofetil and cyclophosphamide) (AA genotype 100%) because of the higher sCr (AA genotype: median sCr at peak 4.24 mg/dl). However, patients carrying the risk genotype also showed a higher frequency of renal function recovery (AA genotype: remission rate 87.5%). The higher remission rate seemed not entirely attributable to a higher proportion of use of immunosuppressants. When only focusing on patients who had received immunosuppressive therapy, patients with the HLA risk genotype also showed a higher remission rate (AA 87.5% versus GA 89.1% versus GG 59.8%, P<0.001). For individuals carrying non-risk HLA genotype GG, they also partially benefited from immunosuppressants (remission rate: 31.6% without steroid versus 59.8% with steroid) (Supplemental Table 13).

Figure 5.

Clinical correlations stratified by different genotypes of HLA variant rs35087390. (A–F) Comparisons were on the basis of baseline data collection in pooling. (G–I) Comparisons were on the basis of available therapy and prognosis data in the discovery cohort. Detailed statistics could be referred to in Supplemental Table 13. *Significant P<0.05. AKD, acute kidney disease. Figure 5 can be viewed in color online at www.jasn.org.

Discussion

Identifying genetic risk factors may not only aid in increased understanding of the pathological mechanisms but may also enhance diagnostic capabilities, facilitating earlier identification of individuals prone to ATIN for specific therapy. In this study, we reported the results of the first GWAS meta-analysis of ATIN. The diagnosis of ATIN was supported in every patient by kidney biopsy. Rigorous genetic criteria were used, and no evidence of population stratification was observed. Analysis of the frequency of common genetic variants across the genome identified two loci (HLA and PLEKHA5) achieving genome-wide significance of P<5×10⁻⁸.

Although ATIN occurs through different mechanisms,^21,64 evidence suggests that genetics may play a role in etiology. Consistent with our previous data,²⁶ we replicated the association of ATIN with classical alleles of the HLA with statistical significance at the genome-wide level. There was an overlap of 105 patients between this study (n=299) and our previous report on HLA-II (total n=154), with no controls overlapped.²⁶ The reasons for patients being excluded were inadequate DNA (n=47) and failure in quality control (n=2). Only 20% of the patients have been overlapped with the GSA chip, and we also included an independent cohort by a different genotyping ASA array. We thus confirmed the HLA association on the same genes and the same associated alleles. Furthermore, we provided more information to refine the associations. We found that the association at this locus was mainly driven by HLA-DRB1 alleles. The HLA-DRB1*14 allele (OR, 3.71), nevertheless, had been consistently reported to be associated with ATIN in both Chinese and White patients.²⁵ It seemed that ATIN had different HLA allele associations compared with glomerulonephritis (Supplemental Table 14). In comparison, a protective effect of the DR15 serotype (DRB1*1501-DQA1*0102-DQB1*0602 haplotype) was observed in ATIN, which had been suggested to be of risk in antiglomerular basement membrane disease, membranous nephropathy, and ANCA-associated vasculitis. Our study was also the first to identify associations of key amino acid positions in ATIN HLA genes, HLA-DRB1 residue 60 and its correlated position 57, both of which belonged to pocket P10. Although HLA rs35087390 (c.664G>A, p.Ala222Thr) was also a missense variant locating in exon four of HLA-DQA1. In the analysis of polymorphic amino acid sites, HLA-DQA1 did not have genome-wide significant amino acid residue and the protein structure was unknown on rs35087390. Considering functional relevance, especially for antigen presentation, we thus only highlighted the DRB1 residue. These variants might participate in antigen binding, suggesting that conformational changes in HLA molecules as a consequence might influence peptide presentation to T cells, thus modulating the risk of ATIN. The position 57 in HLA-DRβ1 had been associated with the development of rheumatoid arthritis,⁵⁴ type 1 diabetes,^55,57 and leprosy.⁵⁶ Of note, it had been also denoted as the major drive in HLA for these diseases. Exploring the effect of the amino acid variant identified in this study on the crystal structure of the HLA molecules may provide invaluable information about the molecular pathogenesis of ATIN. In the future, we need further extensive work by involving more populations, larger sample size, and functional investigation on the responses through the T cell receptor of the responding T cells.

PLEKHA was reported to be involved in T-cell immunity and related autoimmune diseases. Using probabilistic transcriptome-wide association, a likely causal relationship between PLEKHA5 expression and myeloid cell count was suggested (P=8.15×10⁻⁵).⁶⁵ In infants at risk of type 1 diabetes, it was suggested that near-birth expression of PLEKHA5 along with the HLA risk score could predict islet-antibody seroconversion.⁶⁶ In melanoma, it was reported that PLEKHA5 interacted with PDCD4 (programmed cell death 4) to influence brain metastasis, which associated with increased immune cell infiltration.⁶⁷ Thus, it was possible that PLEKHA5 may interact with HLA risk in shaping auto/immunity occurrence and progression.

We took patients with IgA nephropathy as disease controls and confirmed the associations in ATIN were even more evident when compared with patients with glomerulonephritis. It may suggest that the associations may be ATIN-specific. However, because this study was not aiming to determine genome-wide pleiotropy among different kidney diseases and we presently had inadequate matched GWAS datasets and statistical power, we suggested that a phenome-wide association study approach would be needed in the future.

We also observed patients with different clinical causes, including D-ATIN, AD-ATIN, or TINU, had similar genetic susceptibility in the top associated alleles. It thus suggested that there should be common immunologic pathogenesis in triggering tubulointerstitial nephritis. However, some difference could be observed in risk effect size, which warranted more widespread replications and further functional investigations. In clinical correlations, we observed significant associations between risk genotype of HLA variant and higher levels of markers of kidney injury. Nonetheless, the at-risk individuals showed a higher remission rate under immune therapy.²⁶ We, therefore, assumed that HLA risk allele/protein enhanced the antigen-presenting ability of renal tubular cells and thus facilitated the development of tubulointerstitial nephritis under some specific pathogenic stimuli. Assessment of the HLA genotype might aid in identifying patients with higher risk of immunity-mediated AKI, who would possibly show better response to steroid or immunosuppressive therapy.⁶⁸ Because specific T-cell subset-relevant cytokine IL9 and TNF-α have been shown to be helpful in ATIN diagnosis and prognosis,^11,12 future inclusion of HLA genotype may further improve precise understanding of the disease mechanism, guide management decisions, and help enroll appropriate patients in therapeutic trials.

Our study had several limitations. First, owing to rarity of the disease, we pooled data across a heterogeneous patient cohort with diverse primary disease causes. More widespread replications and further functional investigations will be needed in the future. Genetic association of PLEKHA5 still needs validation in different populations. We should also keep in mind that genetic association may not directly tell us the exact gene due to linkage disequilibrium. Targeted functional assay may help us know how these candidates are involved in ATIN. Second, we were also not able to evaluate the contribution of rare variants in this study, and future sequencing studies will be needed to evaluate relative contributions of rare and common variants to the overall disease risk. Last but not the least, further validation of these clinical correlations in prospective cohorts will be needed. The expected external validity of these findings outside the Chinese population is also warranted.

In conclusion, we identified two candidate regions associated with susceptibility to ATIN. The findings suggest that a genetically conferred risk of immune dysregulation is involved in the pathogenesis of ATIN.

Supplementary Material

jasn-34-895-s001.pdf ^{(10.1MB, pdf)}

Disclosures

G. Chen reports Employer: WeGene. C.M. Hao reports Consultancy: AstraZeneca, Fibrogen, and Nicoya Therapeutics (Shanghai) Co., Ltd.; Honoraria: AstraZeneca, Bayer, Eli Lilly, and Fibrogen; Advisory or Leadership Role: Associate Editor of Kidney Disease and Editorial Board of the American Journal of Physiology-Renal Physiology; and Other Interests or Relationships: Member of ASN, Member of ISN, President of the Chinese Society of Physiology—Renal Physiology, and Standing Committee Member of the Chinese Society of Nephrology. L.C. Tsoi reports Research Funding: Galderma, Janssen, and Novartis. L.-Z. Wang reports Employer: WeGene; and Honoraria: WeGene. H. Zhang reports Consultancy: Calliditas, Chinook, Novartis, OMEROS, and Ostuka. The remaining authors have nothing to disclose.

Funding

Support was provided by the National Key R&D Program of China (2022YFC2502500, 2022YFC2502502), the National Science Foundation of China (82130021, 82022010, 82131430172, 81970613, 8211001014, 81870460), the Beijing Young Scientist Program (BJJWZYJH01201910001006), the Capital's Funds for Health Improvement and Research (CFH2022-1-4071), the Beijing Natural Science Foundation (Z190023), and the Fok Ying Tung Education Foundation (171030). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Author Contributions

Conceptualization: X.J. Zhou

Data curation: G. Chen, J.-W. Huang, Y. Jia, G. Li, Y. Liu, T. Su, J. Xie, Q.-H. Xie, L. Yang, X.-J. Yu, X.J. Zhou

Formal analysis: C.C. Berthier, G. Chen, Y. Hu, Y. Liu, S.K. Nath, M.T. Patrick, L.C. Tsoi, L.-Z. Wang, X.J. Zhou

Funding acquisition: X.J. Zhou

Investigation: C.C. Berthier, L.C. Tsoi, X.J. Zhou

Methodology: M.T. Patrick, L.C. Tsoi, X.J. Zhou

Project administration: X.J. Zhou

Software: X.J. Zhou

Supervision: N. Chen, C.-M. Hao, G. Liu, S.-X. Wang, H. Xu, L. Yang, H. Zhang, X.J. Zhou

Validation: C.C. Berthier, H. Zhang

Visualization: H. Zhang

Writing – original draft: L. Yang, X.J. Zhou

Writing – review & editing: H. Xu, X.J. Zhou, H. Zhang

Data Sharing Statement

Genome-wide summary statistics for this study have been made publicly available. Summary statistics could be referred to as supplementary in the journal website.

It contains three files, including “ATIN_ASA.glm.logistic” for ASA chip, “ATIN_GSA.glm.logistic” for GSA chip, and “ATIN_metal.TBL” for meta-analysis.

“.glm.logistic” file is the result from plink association; it contains information on CHROM (chromosome), POS (position in hg19), ID (variant ID/SNP identifier), REF (reference allele), ALT (alternative allele), A1 (test A1/minor allele for the purposes of regression), TEST (test model), OBS_CT (observed sample count), BETA (effect size/beta for A1 allele), SE (standard error of effect size), L95 (lower limit of 95% CI), U95(upper limit of 95% CI), Z_STAT (Z-stat), and P (P value).

“.TBL” file is the result from METAL in meta-analysis; it contains information on MarkerName (variant ID/SNP identifier in hg19), Allele1 (Allele1), Allele2 (Allele2), Effect (beta for A2 allele), StdErr (standard error of effect size), P-value (P value), Direction (direction of effect size in each study, reported with respect to A2 allele; +, -, or ? denoting positive, negative, or missing effect, respectively), HetISq (Heterogeneity I^2 parameter), HetChiSq (Heterogeneity Chi^2 statistic), HetDf (Heterogeneity degrees of freedom), HetPVal (Heterogeneity P value, on the basis of above Chi^2 statistic), and TotalSampleSize (total sample size).

The following Supplemental Tables 1, 3, 4, and 9 from METAL were in the same format.

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/JSN/D744.

Supplemental Methods and Results.

Supplemental Figure 1. Plots of the two significant principal components (PC1 and PC2) in the two cohorts.

Supplemental Figure 2. Power calculation.

Supplemental Figure 3. Manhattan plot of P values and quantile-quantile plot for SNP associations with ATIN in both discovery and replication cohorts.

Supplemental Figure 4. Regional plot for association results of HLA variants using the newly released multiancestry MHC reference panel and comparisons with data using Pan Asian reference.

Supplemental Figure 5. RNA and protein baseline expression for PLEKHA5.

Supplemental Figure 6. The regional plots in single variant associations for the non-HLA gene regions.

Supplemental Figure 7. The genomic location of rs2417771 of PLEKHA5 colocalized with epigenetic markers such as H3K27ac and ATAC peaks in several epithelial cell types by 3DSNP v2.0.

Supplemental Figure 8. Manhattan plot and quantile-quantile plot for SNP associations in the IgAN cohort.

Supplemental Figure 9. Miami plot showing genome-wide association test statistics for ATIN (top) and IgAN (bottom) compared with healthy controls.

Supplemental Figure 10. LocusCompare visualizations of colocalization between ATIN GWAS using healthy controls and ATIN GWAS using IgAN disease controls at HLA and PLEKHA5.

Supplemental Figure 11. Manhattan plots for genome-wide associations in different causes of ATIN defined by clinicians.

Supplemental Table 1. Biallelic HLA SNP association. Logistic regression results for all SNPs were meta-analyzed.

Supplemental Table 2. Omnibus association tests and conditional analysis for classical HLA alleles.

Supplemental Table 3. HLA classical allele association in ATIN.

Supplemental Table 4. Unconditional biallelic amino acid site analysis in HLA region.

Supplemental Table 5. Conditional biallelic amino acid site analysis in HLA region.

Supplemental Table 6. Omnibus association for amino acid polymorphic sites in HLA region.

Supplemental Table 7. Conditional analysis of amino acid polymorphic sites in Omnibus tests in HLA region.

Supplemental Table 8. Association results of the top HLA classical and amino acid variants for ATIN.

Supplemental Table 9. Summary statistics for the association results of HLA variants using a multiancestry MHC reference panel.

Supplemental Table 10. Functional annotations of genome-wide significant and suggestive ATIN loci.

Supplemental Table 11. Associations of reported IgAN loci in the current IgAN GSA cohort dataset.

Supplemental Table 12. Lead SNPs in different causes of ATIN.

Supplemental Table 13. Comparison of clinical data among patients with different genotypes of HLA variant rs35087390.

Supplemental Table 14. Association of classical HLA alleles with kidney phenotypes.

Supplemental Table 15. The six candidate genes were associated with kidney phenotypes in UK biobank PheWAS data.

Summary Statistics. Summary statistics contain three files, including “ATIN_ASA.glm.logistic” for the ASA chip, “ATIN_GSA.glm.logistic” for the GSA chip, and “ATIN_metal.TBL” for meta-analysis.