Strong Association of the HLA-DR/DQ Locus with Childhood Steroid-Sensitive Nephrotic Syndrome in the Japanese Population

Abstract

Background Nephrotic syndrome is the most common cause of chronic glomerular disease in children. Most of these patients develop steroid-sensitive nephrotic syndrome (SSNS), but the loci conferring susceptibility to childhood SSNS are mainly unknown.

Methods We conducted a genome-wide association study (GWAS) in the Japanese population; 224 patients with childhood SSNS and 419 adult healthy controls were genotyped using the Affymetrix Japonica Array in the discovery stage. Imputation for six HLA genes (HLA-A, -C, -B, -DRB1, -DQB1, and -DPB1) was conducted on the basis of Japanese-specific references. We performed genotyping for HLA-DRB1/-DQB1 using a sequence-specific oligonucleotide-probing method on a Luminex platform. Whole-genome imputation was conducted using a phased reference panel of 2049 healthy Japanese individuals. Replication was performed in an independent Japanese sample set including 216 patients and 719 healthy controls. We genotyped candidate single-nucleotide polymorphisms using the DigiTag2 assay.

Results The most significant association was detected in the HLA-DR/DQ region and replicated (rs4642516 [minor allele G], combined P_allelic=7.84×10⁻²³; odds ratio [OR], 0.33; 95% confidence interval [95% CI], 0.26 to 0.41; rs3134996 [minor allele A], combined P_allelic=1.72×10⁻²⁵; OR, 0.29; 95% CI, 0.23 to 0.37). HLA-DRB1*08:02 (Pc=1.82×10⁻⁹; OR, 2.62; 95% CI, 1.94 to 3.54) and HLA-DQB1*06:04 (Pc=2.09×10⁻¹²; OR, 0.10; 95% CI, 0.05 to 0.21) were considered primary HLA alleles associated with childhood SSNS. HLA-DRB1*08:02-DQB1*03:02 (Pc=7.01×10⁻¹¹; OR, 3.60; 95% CI, 2.46 to 5.29) was identified as the most significant genetic susceptibility factor.

Conclusions The most significant association with childhood SSNS was detected in the HLA-DR/DQ region. Further HLA allele/haplotype analyses should enhance our understanding of molecular mechanisms underlying SSNS.

Pediatric idiopathic nephrotic syndrome (INS) is the most common glomerular disease, with a prevalence of approximately 16 cases per 100,000 children.¹ In Japan, the estimated incidence of INS is 6.49 cases per 100,000 children annually.² Between 80% and 90% of patients develop steroid-sensitive nephrotic syndrome (SSNS), in which complete remission is achieved within 4 weeks after starting 60 mg/m² oral prednisolone daily, whereas 10%–20% exhibit steroid-resistant nephrotic syndrome, characterized by persistent proteinuria after 60 mg/m² oral prednisolone daily for 4 weeks (Supplemental Table 1). Therefore, elucidating the pathogenesis of the disease in order to implement optimal treatment is critical.

Previous studies have highlighted the crucial role of podocytes and the immune system in the development of INS.^3,4 The strong association between human leukocyte antigen (HLA) and susceptibility to INS has been reported in different populations.^5–22 A single-gene cause is detected in about 30% of patients with steroid-resistant nephrotic syndrome.²³ However, the pathogenesis of INS has not been completely elucidated.

Genome-wide association studies (GWASs) have been widely used to investigate common genetic variants associated with complex diseases. To date, two GWASs of nephrotic syndrome have been reported: one using genome-wide single-nucleotide polymorphism (SNP) arrays for adult-acquired nephrotic syndrome, and an exome array–based association study of childhood SSNS in patients of South Asian and European ancestry.^22,24 Because the histopathologic spectrum of INS in children differs from that in adults, and there may be ethnic differences in susceptibility genes, a GWAS examining childhood INS in the Japanese population is needed.

Here, for the purpose of identifying loci conferring susceptibility to childhood-onset INS, especially SSNS, we performed a GWAS with a replication study in the Japanese population.

Methods

Samples and Clinical Data

This study included 440 patients with childhood SSNS and 1138 adult healthy controls recruited from the Japanese population. Definitions of nephrotic syndrome are shown in Supplemental Table 1. We only included patients with SSNS. Patients with a history of steroid resistance during follow-up were excluded. There was no onset-age bias between patients in the discovery and replication sample sets (P=0.12, Supplemental Figure 1). DNA samples and clinical information regarding subjects were collected from 43 hospitals across Japan. Clinical data for the patients were collected using a simple questionnaire. Genomic DNA was extracted from peripheral blood following a standard protocol. Healthy adults (age>18 years) who have passed childhood without disease onset were selected as controls. In the discovery stage, 419 healthy adults were referred by the Department of Human Genetics, Graduate School of Medicine, The University of Tokyo. For the replication phase, 719 healthy adult controls were recruited from the Pharma SNP Consortium (Tokyo, Japan). The power of the two-stage GWAS exceeded 80% to detect common alleles (minor allele frequency [MAF] ≥5%) with genotypic relative risk >2.95, or variants with an allele frequency >50% conferring a relative risk >2.0 at a significance threshold of P=5×10⁻⁸, respectively (Supplemental Figure 2, A and B).

This study was approved by the Research Ethics Committees of Kobe University Graduate School of Medicine and the Graduate School of Medicine, The University of Tokyo, and all of the collaborating universities and hospitals. All participants provided written informed consent for participation in this study.

Genotyping and Data Cleaning in the GWAS

In the initial GWAS, 224 patients and 419 controls were genotyped using the Affymetrix “Japonica Array”.²⁵ One control was excluded because of low call rate (<97%). The identical-by-descent test was performed and no subjects were excluded. Principal component analysis was performed using GCTA (Version 1.26.0)²⁶ for 224 patients, 418 controls, and HapMap Phase III data (113 CEU, 113 YRI, 84 CHB, and 86 JPT). Six controls were identified as outliers and excluded (Supplemental Figure 3, A–C).

We applied the following thresholds for SNP quality control (QC): SNP call rate ≥97%, MAF≥5%, and Hardy–Weinberg equilibrium (HWE) P value ≥0.001 in healthy controls. A total of 495,895 autosomal SNPs passed the filters and were used in association analyses. The inflation factor, λ, was 1.045 for all tested SNPs (Supplemental Figure 4A) and decreased to 1.040 when SNPs in the HLA region (Hg19: chr6: 29,691,116–33,054,976) were excluded (Supplemental Figure 4B).

Whole-Genome Imputation on the Basis of the 2KJPN Panel

SNP genotypes of samples in the discovery stage that passed SNP filtering were phased using SHAPEIT (v.2.r644).²⁷ Genotype imputation was performed using IMPUTE2 (ver. 2.3.1)²⁸ with a phased reference panel of 2049 healthy Japanese individuals (2KJPN panel). A total of 4,105,543 autosomal SNPs and short insertions and deletions (INDELs) passed the QC criteria after imputation (SNP/INDEL call rate ≥97%, MAF≥5%, and HWE P value ≥0.001 in healthy controls). The λ value was 1.023 for all tested variations (Supplemental Figure 5A) and decreased to 1.019 when SNPs/INDELs in the HLA region (Hg19: chr6: 29,691,116–33,054,976) were excluded (Supplemental Figure 5B).

Association Analyses in the Discovery Stage

Association analyses for typing SNPs and SNPs/INDELs after whole-genome imputation were conducted using the Cochran–Armitage trend test by PLINK 1.9. P values were corrected by genomic control (GC). All cluster plots of the SNPs with a P_GC-corrected<1×10⁻⁵ were checked by visual inspection.

Validation and Replication of Candidate SNPs

DigiTag2 genotyping assay²⁹ was used for the genotyping of candidate variants. Ten candidate SNPs/INDELs were genotyped in 224 patients in the discovery sample set for validation. Five of the ten candidate variants were genotyped successfully, with a mean concordance rate of 97% (90%–100%). Replication was performed in an independent Japanese sample set including 216 patients and 719 healthy controls.

HLA Imputation and HLA Genotyping

In the discovery stage, we conducted two-field HLA imputation for six HLA genes (HLA-A, -C, -B, -DRB1, -DQB1, and -DPB1) in 224 patients and 412 healthy controls. A total of 197 patients and 411 controls passed postimputation QC using a call-threshold >0.4.³⁰ In the replication stage, we imputed the HLA-DRB1 and -DQB1 genes using the same approach in 269 healthy controls whose HLA genotypes were not determined by direct genotyping. A total of 260 of the 269 controls passed postimputation QC (call-threshold >0.4).

Genotyping of HLA-DRB1 and -DQB1 was performed using the PCR sequence–specific oligonucleotide probing method. For validation, we performed HLA-DRB1/-DQB1 genotyping in 224 patients in the discovery sample set. A total of 409 of 412 controls in the discovery stage were previously genotyped using the same technique.³⁰ The concordance rate between the imputed (after post-imputation QC) and genotyped HLA-DRB1/-DQB1 genotypes was 99.2%. In the replication stage, we performed HLA-DRB1/-DQB1 genotyping in 216 patients, and HLA-DRB1 genotyping failed in three patients. The published HLA-DRB1/-DQB1 genotyping data for 450 controls in the replication sample set was utilized.^31,32

Statistical Methods and Software

PLINK 1.9 was used for data cleaning and SNP-based analyses. Manhattan and QQ plots were generated using the R package “qqman.” Regional plots were prepared using Locuszoom.³³ HLA haplotypes were determined using the Arlequin algorithm (Arlequin ver 3.5.2.2).³⁴ HLA allele and haplotype frequencies were compared between patient and control groups. P values were calculated using the Pearson’s chi-squared test or Fisher’s exact test in presence versus absence of each HLA allele/HLA haplotype. HLA alleles/haplotypes with frequencies<0.5% in patients or controls were excluded from the association analyses. Bonferroni correction was then performed by the standard method, in which P values were corrected for the number of alleles/haplotypes tested in each analysis (shown as P-corrected [Pc]). We considered an association significant when the P value was <0.05 after correction for multiple comparisons. The GWAS power was calculated using the R package “CATS.”³⁵ Proxy SNPs were identified with SNAP (SNP Annotation and Proxy Search)³⁶ using 1000 Genomes Pilot 1 (CHB-JPT panel). For the evaluation of disease variance explained in this study, logistic regression and the calculation of Nagelkerke’s pseudo-R² were done by R programming and R package “rcompanion.”

Data Sharing

Summarized data of typed SNPs in the initial GWAS are available through the Japan National Bioscience Database Center database (Research ID: hum0126, https://humandbs.biosciencedbc.jp/en/hum0126-v1).

Results

Subjects

Characteristics of the patients with SSNS are shown in Supplemental Table 2. The total male-to-female ratio was 2.6:1, and the median age at onset was 4.0 years. Renal biopsy was performed in 248 of the 440 patients (56%) (minimal change disease [n=227, 91.5%], FSGS [n=14, 5.7%], and mesangial proliferative GN [n=7, 2.8%]). As controls, 1138 healthy adults who had passed childhood without disease onset were included in the study.

Initial GWAS and Whole-Genome Imputation in Discovery Stage

After stringent QC and filtering steps, 224 patients and 412 healthy controls with 495,895 SNPs were retained for association analysis. Figure 1A presents the genome-wide view of the SNP association data using the Cochran–Armitage trend test. The HLA-DR/DQ region exhibited the most significant association with childhood SSNS in the initial GWAS after correction on the basis of GC (rs4642516, P_GC-corrected=5.44×10⁻¹⁰) (Figure 1B). After conditioning on rs4642516, there was no secondary signal in the HLA region (Figure 1C). No other locus outside the HLA region reached genome-wide significance (P_GC-corrected<5×10⁻⁸).

Figure 1.

The HLA-DR/DQ region exhibited the most significant association with childhood SSNS in the discovery GWAS. (A) Results of the GWAS (Manhattan plot) for childhood SSNS in the Japanese population. P values were calculated using the Cochran–Armitage trend test for 224 patients and 412 controls in the discovery stage. The horizontal red and blue lines indicate the genome-wide significant threshold (P=5×10⁻⁸) and suggestive threshold (P=1×10⁻⁵), respectively. (B) Plot of the HLA-DR/DQ region on the basis of the association analysis in the discovery stage. The purple dot indicates the top SNP in the association analysis (rs4642516, P_GC-corrected=5.44×10⁻¹⁰). (C) Plot of the HLA-DR/DQ region after conditioning on the top SNP (rs4642516).

Whole-genome imputation was performed using the 2KJPN panel. After QC, 224 patients and 412 controls with 4,105,543 autosomal SNPs and short INDELs were included for association analysis. Two loci outside of the HLA region achieved the suggestive significance (P_GC-corrected<1×10⁻⁵) using the Cochran–Armitage trend test (Supplemental Figure 6). For further validation and replication, six SNPs/INDELs in high linkage disequilibrium (LD) in the top hit of the HLA region and another four SNPs from two candidate loci outside of the HLA region (one genotyped SNP and one imputed SNP with minimum P value in each candidate locus) were selected (Supplemental Table 3A).

Replication of Candidate SNPs

In the replication stage, the associations of rs4642516 and rs3134996 in the HLA-DR/-DQ region were confirmed in an independent sample set (Supplemental Table 3, A and B, Table 1). In combined analysis including discovery and replication samples, rs4642516 (minor allele G, combined P=7.84×10⁻²³; odds ratio [OR], 0.33; 95% confidence interval [95% CI], 0.26 to 0.41, under allelic model; combined P=4.87×10⁻²³; OR, 0.29; 95% CI, 0.22 to 0.38, under dominant model) and rs3134996 (minor allele A, combined P=1.72×10⁻²⁵; OR, 0.29; 95% CI, 0.23 to 0.37, under allelic model; combined P=9.57×10⁻²¹; OR, 0.28; 95% CI, 0.21 to 0.38, under dominant model) showed significant associations with disease (Table 1). Other candidate loci didn’t achieve significance after Bonferroni correction (Supplemental Table 3A).

Table 1.

Replication and combined analysis of SNPs in HLA-DR/DQ region

CHR	SNP	A1/A2	Genetic Model	Discovery GWAS (224 Patients versus 412 Controls)				Replication (216 Patients versus 719 Controls)				Combined Analysis (440 Patients versus 1131 Controls)
				Patients	Controls	P-Fisher	OR (95% CI)	Patients	Controls	P-Fisher	OR (95% CI)	Patients	Controls	P-Fisher	OR (95% CI)
6	rs4642516	G/T	Allelic	0.11a	0.26b	1.06E−10	0.36 (0.26 to 0.50)	0.10a	0.28b	4.51E−15	0.30 (0.22 to 0.42)	0.11a	0.27b	7.84E−23	0.33 (0.26 to 0.41)
			Dominant	47 of 177	194 of 218	5.29E−11	0.30 (0.21 to 0.43)	43 of 173	336 of 381	3.43E−13	0.28 (0.20 to 0.41)	90 of 350	530 of 599	4.87E−23	0.29 (0.22 to 0.38)
6	rs3134996	A/T	Allelic	0.09a	0.23b	4.85E−11	0.32 (0.22 to 0.46)	0.12a	0.32b	1.26E−15	0.30 (0.22 to 0.41)	0.10a	0.28b	1.72E−25	0.29 (0.23 to 0.37)
			Dominant	37 of 187	168 of 244	1.86E−10	0.29 (0.19 to 0.43)	34 of 163	265 of 363	6.24E−11	0.29 (0.19 to 0.43)	71 of 350	433 of 607	9.57E−21	0.28 (0.21 to 0.38)

CHR, chromosome; A1, minor allele (test allele); A2, major allele (reference allele); P-Fisher, P value calculated using Fisher’s exact test.

^aMAF in patients.

^bMAF in controls.

HLA Imputation in the Discovery Stage

To explore the HLA region further, imputation was performed for both HLA class I (HLA-A, -C and -B) and class II (HLA-DRB1, -DQB1, and -DPB1) genes in the discovery sample set. A total of 197 patients and 411 controls passed the post-imputation QC. Significantly associated HLA alleles were identified in all six HLA genes after multiple corrections (Supplemental Tables 4 and 5, Table 2). HLA-A*02:06 (Pc=8.84×10⁻³; OR, 1.91; 95% CI, 1.29 to 2.81), HLA-DRB1*08:02 (Pc=2.56×10⁻⁴; OR, 2.79; 95% CI, 1.72 to 4.51), and HLA-DQB1*03:02 (Pc=4.14×10⁻⁴; OR, 2.08; 95% CI, 1.46 to 2.96) were highly associated with susceptibility to childhood SSNS. HLA-A*33:03 (Pc=2.43×10⁻³; OR, 0.30; 95% CI, 0.15 to 0.60), HLA-B*44:03 (Pc=1.25×10⁻⁴; OR, 0.13; 95% CI, 0.05 to 0.37), HLA-C*14:03 (Pc=5.25×10⁻⁵; OR, 0.13; 95% CI, 0.05 to 0.36), HLA-DRB1*13:02 (Pc=7.47×10⁻⁴; OR, 0.18; 95% CI, 0.07 to 0.45), and HLA-DQB1*06:04 (Pc=3.40×10⁻⁵; OR, 0.07; 95% CI, 0.02 to 0.31) showed apparent protective effects.

Table 2.

HLA alleles and haplotypes significantly associated with childhood SSNS in the discovery stage using HLA-imputation data

HLA Alleles/Haplotypesa	Allele Frequency
	Patients (2n=394)		Controls (2n=822)		Chi-Squared Test
	No.	%	No.	%	OR (95% CI)	Pcb
A*02:06	53	13.5	62	7.5	1.91 (1.29 to 2.81)	8.84E−03
A*33:03	10	2.5	65	7.9	0.30 (0.15 to 0.60)	2.43E−03
C*14:03	4	1.0	60	7.3	0.13 (0.05 to 0.36)	5.25E−05
B*44:03	4	1.0	59	7.2	0.13 (0.05 to 0.37)	1.25E−04
DRB1*08:02	40	10.2	32	3.9	2.79 (1.72 to 4.51)	2.56E−04
DRB1*13:02	5	1.3	55	6.7	0.18 (0.07 to 0.45)	7.47E−04
DQB1*03:02	68	17.3	75	9.1	2.08 (1.46 to 2.96)	4.14E−04
DQB1*06:02	14	3.6	65	7.9	0.43 (0.24 to 0.77)	4.33E−02
DQB1*06:04	2	0.5	53	6.4	0.07 (0.02 to 0.31)	3.40E−05
DPB1*05:01	188	47.7	316	38.4	1.46 (1.15 to 1.86)	2.13E−02
A*33:03-B*44:03	4	1.0	54	6.6	0.15 (0.05 to 0.41)	6.54E−04
A*02:06-B*40:06	17	4.3	10	1.2	3.66 (1.66 to 8.07)	1.86E−02
B*44:03-C*14:03	4	1.0	59	7.2	0.13 (0.05 to 0.37)	1.59E−04
B*40:06-C*08:01	28	7.1	24	2.9	2.54 (1.45 to 4.45)	2.05E−02
A*02:06-C*08:01-B*40:06	16	4.1	9	1.1	3.82 (1.67 to 8.73)	1.94E−02
A*33:03-C*14:03-B*44:03	4	1.0	54	6.6	0.15 (0.05 to 0.41)	6.33E−04
DRB1*08:02-DQB1*03:02	33	8.4	17	2.1	4.33 (2.38 to 7.87)	4.34E−06
DRB1*13:02-DQB1*06:04	2	0.5	53	6.4	0.07 (0.02 to 0.31)	6.17E−05
DRB1*08:02-DQB1*03:02-DPB1*05:01	20	5.1	15	1.8	2.88 (1.46 to 5.68)	3.91E−02
DRB1*09:01-DQB1*03:03-DPB1*02:01	40	10.2	32	3.9	2.79 (1.72 to 4.51)	3.91E−04
DRB1*13:02-DQB1*06:04-DPB1*04:01	2	0.5	35	4.3	0.11 (0.03 to 0.48)	9.52E−03
DRB1*15:01-DQB1*06:02-DPB1*02:01	3	0.8	39	4.7	0.15 (0.05 to 0.50)	9.65E−03
A*02:06-C*08:01-B*40:06-DRB1*09:01-DQB1*03:03	12	3.0	5	0.6	5.13 (1.80 to 14.67)	6.34E−03

2n, counts of HLA alleles/haplotypes.

^aAlleles/haplotypes, HLA alleles/haplotypes with frequencies <0.5% in patients or controls were excluded from the association analyses.

^bPc, P values for allele/haplotype frequency comparisons between patients and controls using the Pearson’s chi-squared test or Fisher’s exact test and then corrected for the multiplicity of testing on the basis of the number of comparisons.

In HLA haplotype analyses, significantly associated haplotypes were detected after multiple corrections (Supplemental Tables 6–9, Table 2). HLA-DRB1*08:02-DQB1*03:02 (Pc=4.34×10⁻⁶; OR, 4.33; 95% CI, 2.38 to 7.87) and HLA-DRB1*13:02-DQB1*06:04 (Pc=6.17×10⁻⁵; OR, 0.07; 95% CI, 0.02 to 0.31) demonstrated the most significant positive and negative associations, respectively. HLA-DRB1*08:02-DQB1*03:02 even showed a stronger and more robust association than HLA-DRB1*08:02 or HLA-DQB1*03:02 alone. The two significant HLA-DRB1-DQB1 haplotypes (HLA-DRB1*08:02-DQB1*03:02 and HLA-DRB1*13:02-DQB1*06:04) showed more significant associations than their related HLA-DRB1-DQB1-DPB1 haplotypes (HLA-DRB1*08:02-DQB1*03:02-DPB1*05:01, Pc=3.91×10⁻²; OR, 2.88; 95% CI, 1.46 to 5.68; HLA-DRB1*13:02-DQB1*06:04-DPB1*04:01, Pc=9.52×10⁻³; OR, 0.11; 95% CI, 0.03 to 0.48).

Replication of HLA-DRB1/-DQB1

In the replication stage, HLA-DRB1/-DQB1 genotypes of 213 patients and 710 controls were obtained by HLA genotyping or high-accuracy HLA imputation. Almost all of the significant HLA-DRB1 and -DQB1 alleles and HLA-DRB1-DQB1 haplotypes identified in the discovery stage were replicated except HLA-DQB1*03:02 (Supplemental Tables 10–12, Table 3). Although HLA-DQB1*03:02 did not show any significance in the replication stage (P=3.07×10⁻¹ before multiple correction, Supplemental Table 11), HLA-DRB1*08:02-DQB1*03:02 was still the most significant susceptibility haplotype in the same sample set (Pc=1.25×10⁻⁴; OR, 3.15; 95% CI, 1.87 to 5.32) (Supplemental Table 12).

Table 3.

Validation, replication and combined analyses of the significant HLA-DRB1/-DQB1 alleles and HLA-DRB1-DQB1 haplotypes

HLA Alleles/Haplotypesa	Validation				Replication				Combined Analysis
	Patients (2n=448) No. (%)	Controls (2n=818) No. (%)	Chi-Squared Test		Patients (2n=426) No. (%)	Controls (2n=1420) No. (%)	Chi-Squared Test		Patients (2n=874) No. (%)	Controls (2n=2238) No. (%)	Chi-Squared Test
			OR (95% CI)	Pcb			OR (95% CI)	Pcb			OR (95% CI)	Pcb
DRB1*08:02	46 (10.3)	32 (3.9)	2.81 (1.76 to 4.48)	1.24E−04	44 (10.3)	62 (4.4)	2.52 (1.69 to 3.77)	6.64E−05	90 (10.3)	94 (4.2)	2.62 (1.94 to 3.54)	1.82E−09
DRB1*13:02	5 (1.1)	55 (6.7)	0.16 (0.06 to 0.39)	1.28E−04	7 (1.6)	125 (8.8)	0.17 (0.08 to 0.37)	9.32E−06	12 (1.4)	180 (8.0)	0.16 (0.09 to 0.29)	7.31E−11
DQB1*03:02	80 (17.9)	77 (9.4)	2.09 (1.49 to 2.93)	1.44E−04	48 (11.3)	136 (9.6)	1.20 (0.85 to 1.70)	NS	128 (14.6)	213 (9.5)	1.63 (1.29 to 2.06)	4.25E−04
DQB1*06:02	15 (3.4)	65 (8.0)	0.40 (0.23 to 0.71)	1.43E−02	14 (3.3)	108 (7.6)	0.41 (0.23 to 0.73)	1.81E−02	29 (3.3)	173 (7.7)	0.41 (0.27 to 0.61)	7.84E−05
DQB1*06:04	2 (0.5)	53 (6.5)	0.06 (0.02 to 0.27)	5.26E−06	5 (1.2)	117 (8.2)	0.13 (0.05 to 0.33)	2.89E−07	7 (0.8)	170 (7.6)	0.10 (0.05 to 0.21)	2.09E−12
DRB1*08:02-DQB1*03:02	36 (8.0)	17 (2.1)	4.12 (2.28 to 7.42)	8.76E−06	28 (6.6)	31 (2.2)	3.15 (1.87 to 5.32)	1.25E−04	64 (7.3)	48 (2.1)	3.60 (2.46 to 5.29)	7.01E−11
DRBQ*13:02-DQB1*06:04	2 (0.5)	53 (6.5)	0.06 (0.02 to 0.27)	1.00E−05	5 (1.2)	117 (8.2)	0.13 (0.05 to 0.33)	5.25E−06	7 (0.8)	170 (7.6)	0.10 (0.05 to 0.21)	4.18E−12
DRB1*15:01-DQB1*06:02	15 (3.4)	65 (8.0)	0.40 (0.23 to 0.71)	2.74E−02	14 (3.3)	107 (7.5)	0.42 (0.24 to 0.74)	3.77E−02	29 (3.3)	172 (7.7)	0.41 (0.28 to 0.62)	1.85E−04

2n, counts of HLA alleles/haplotypes; NS, Pc≥0.05.

^aAlleles/haplotypes, significant HLA-DRB1/-DQB1 alleles and HLA-DRB1-DQB1 haplotypes identified in the discovery stage using HLA-imputation data.

^bPc, P values for haplotype frequency comparisons between patients and controls using the Pearson’s chi-squared test or Fisher’s exact test and then corrected for multiplicity of testing on the basis of the number of comparisons.

Combined analyses were performed using the HLA-DRB1/-DQB1 genotype data in validation and replication steps (Supplemental Tables 10–12, Table 3). HLA-DRB1*08:02 (Pc=1.82×10⁻⁹; OR, 2.62; 95% CI, 1.94 to 3.54), HLA-DRB1*13:02 (Pc=7.31×10⁻¹¹; OR, 0.16; 95% CI, 0.09 to 0.29), and HLA-DQB1*06:04 (Pc=2.09×10⁻¹²; OR, 0.10; 95% CI, 0.05 to 0.21) were identified as the most significantly associated HLA alleles. HLA-DRB1*08:02-DQB1*03:02 (Pc=7.01×10⁻¹¹; OR, 3.60; 95% CI, 2.46 to 5.29) and HLA-DRB1*13:02-DQB1*06:04 (Pc=4.18×10⁻¹²; OR, 0.10; 95% CI, 0.05 to 0.21) were highly associated with childhood SSNS. Considering the tight LD between HLA-DRB1 and -DQB1, we performed reciprocal conditional analyses using logistic regression in the combined dataset, aiming to elucidate which allele in the HLA-DRB1-DQB1 haplotype exerts an independent effect on the disease. When conditioned on HLA-DQB1*03:02, the significance of HLA-DRB1*08:02 decreased from P=2.88×10⁻¹⁰ to conditional P=3.13×10⁻⁷, but still showed a robustly independent effect. However, when we conditioned on HLA-DRB1*08:02, the significance of HLA-DQB1*03:02 disappeared (P=5.62×10⁻⁵ to conditional P=0.10). Similarly, HLA-DQB1*06:04 showed some independent effect when we conditioned on HLA-DRB1*13:02 (P=2.63×10⁻⁹ to conditional P=7.22×10⁻⁵), whereas HLA-DRB1*13:02 did not show any significance when we conditioned on HLA-DQB1*06:04 (P=2.32×10⁻⁹ to conditional P=0.37). Both HLA-DRB1*08:02 (P=2.88×10⁻¹⁰ to conditional P=1.14×10⁻⁸ when conditioned on HLA-DQB1*06:04) and HLA*DQB1*06:04 (P=2.63×10⁻⁹ to conditional P=8.47×10⁻⁹ when conditioned on HLA-DRB1*08:02) still showed strikingly significant associations in the reciprocal conditional analyses (Table 4).

Table 4.

HLA allele reciprocal conditional analyses

HLA Allele	Test	OR (95% CI)	P
HLA-DRB1*08:02	Logistic regression	2.70 (1.98 to 3.67)	2.88E−10
	Condition on DQB1*03:02	2.41 (1.72 to 3.37)	3.13E−07
HLA-DQB1*03:02	Logistic regression	1.62 (1.28 to 2.04)	5.62E−05
	Condition on DRB1*08:02	1.25 (0.96 to 1.62)	0.10 (NS)
HLA-DRB1*13:02	Logistic regression	0.17 (0.09 to 0.30)	2.32E−09
	Condition on DQB1*06:04	1.68 (0.53 to 5.30)	0.37 (NS)
HLA-DQB1*06:04	Logistic regression	0.10 (0.05 to 0.21)	2.63E−09
	Condition on DRB1*13:02	0.06 (0.01 to 0.24)	7.22E−05
HLA-DRB1*08:02	Logistic regression	2.70 (1.98 to 3.67)	2.88E−10
	Condition on DQB1*06:04	2.48 (1.82 to 3.39)	1.14E−08
HLA-DQB1*06:04	Logistic regression	0.10 (0.05 to 0.21)	2.63E−09
	Condition on DRB1*08:02	0.11 (0.05 to 0.23)	8.47E−09

NS, P≥0.05.

Single-Tag SNP for Capturing HLA-DRB1*08:02 and HLA-DRB1*08:02-DQB1*03:02

rs3129888 (G) was reported as a tag SNP for HLA-DRB1*08:02 and HLA-DRB1-DQB1 haplotypes involving HLA-DRB1*08:02 (sensitivity of 92.3% and specificity of 98.9%) in the Japanese population.^37,38 In the discovery stage of our GWAS, rs3129888 showed a suggestive significant association with childhood SSNS (P_GC-corrected=4.42×10⁻⁶ by the Cochran–Armitage trend test). We genotyped rs3129888 in our combined sample set, and this tag SNP showed high sensitivity and specificity for capturing both HLA-DRB1*08:02 (sensitivity of 96.2% and specificity of 97.8%) and HLA-DRB1*08:02-DQB1*03:02 (sensitivity of 98.2% and specificity of 95.7%).

Variance Explained by HLA Association

Logistic regression was done in the replication sample set using case-control status as outcome with the genotype of rs4642516 as covariate, and Nagelkerke’s pseudo-R² was calculated to measure the proportion of variance explained by the genetic factor. The top SNP in the HLA region (rs4642516) explained 9.7% of variance (Nagelkerke’s pseudo-R²) in Japanese childhood SSNS. We also conducted the logistic regression using the two independent classic HLA alleles, and 8.1% of the disease variance (Nagelkerke’s pseudo-R²) can be explained by HLA-DRB1*08:02 and HLA-DQB1*06:04.

Discussion

To identify genetic variants influencing childhood SSNS, we performed the first GWAS using the population-specific SNP array “Japonica array” followed by whole-genome imputation in the Japanese population. The most significant association was identified in the HLA-DR/DQ region, and the top SNPs were replicated in an independent Japanese sample set.

The HLA region is known as the most polymorphic genetic system in humans. SNPs in the HLA-DR region do not conform to the HWE because of the marked copy number polymorphism of HLA-DRB genes. Thus, commercially available SNP genotyping arrays are insufficient for analyzing SNPs in the HLA-DR region. Therefore, even the HLA-DQ region may appear to be the most significantly associated genetic factor governing disease susceptibility in SNP-based GWASs; the HLA-DR region, which is in strong LD with the HLA-DQ region, must also be considered. HLA fine-mapping is necessary to identify the primary disease-associated HLA genes and alleles.³⁹ Because four-digit HLA alleles correspond to specific HLA molecules, we performed imputation for four-digit classic HLA alleles and high-resolution typing at the four-digit level to dissect the HLA associations. Consistent with a previous Japanese report,²⁰ HLA-DRB1*08:02 was identified as a significant risk allele and further identified as a primary genetic factor on the basis of our reciprocal conditional analyses. An increased allele frequency of HLA-DQB1*03:02⁷ was also observed in our study (14.6% in patients versus 9.5% in healthy controls), which could have been caused by the LD with HLA-DRB1*08:02. HLA-DRB1*08:02-DQB1*03:02 was identified as the most significant susceptibility haplotype. More importantly, HLA-DRB1*08:02-DQB1*03:02 showed a stronger and more significant association than HLA-DRB1*08:02 or HLA-DQB1*03:02 alone, indicating that the HLA-DRB1-DQB1 haplotype may play a more important role than the single HLA-DRB1/-DQB1 allele in the pathogenesis of Japanese childhood SSNS. Moreover, HLA-DRB1*13:02-DQB1*06:04 was reported as the most significant protective haplotype, and HLA-DQB1*06:04 was identified as the primary genetic factor on the basis of our reciprocal conditional analyses.

A strong association between HLA (especially HLA-DR/DQ genes) and INS has been frequently reported in candidate gene studies involving small sample sizes.^5–21 The significantly associated HLA alleles detected in our study differ from those of previous studies in other populations. An increased frequency of the HLA-DR7 antigen in INS has been reported in several populations,^{5,6,9,10,12,14,18,19} but it was not replicated in a previous Japanese study²⁰ or this study. On the basis of reported frequencies in the Allele Frequency Net Database (http://www.allelefrequencies.net),⁴⁰ HLA-DR7 is common in European, African, Chinese, and Saudi Arabian populations, with an allele frequency of 10.3%–18%.^41–45 By contrast, the frequency of HLA-DR7 is only 0.3%–0.8% in the Japanese population.^46,47 However, the significant susceptibility allele in our study, HLA-DRB1*08:02, is common in Japan (4.2%)⁴⁸ but rare in European and Chinese populations (0%–0.9%)⁴⁹ and relatively rare in South Koreans (2.4%).⁵⁰ The susceptibility haplotype HLA-DRB1*08:02-DQB1*03:02 seems to be extremely rare in Europeans, because there is no haplotype frequency reported. These observations indicate that the associated HLA alleles may vary depending on geographic and ethnic origin.

A previous exome array association study identified four exome-wide significant variants in or around HLA-DQA1 and -DQB1 in South Asian children with SSNS. Two missense variants in HLA-DQA1 (C34Y [rs1129740, allele A], allelic P=1.187×10⁻⁶; OR, 2.11; 95% CI, 1.56 to 2.86; F41S [rs1071630, allele G], allelic P=1.187×10⁻⁶; OR, 2.11; 95% CI, 1.56 to 2.86) were further replicated in children of European ancestry.²² As hypothesis-free approaches, both our GWAS and the exome-wide association study identified the most significantly associated variants in the HLA-DR/DQ region, providing solid evidence that the HLA-DR/DQ region plays the predominant role in susceptibility to childhood SSNS. Although the two reported SNPs were not included in our genotyping array or imputation reference, another two SNPs that are in LD with the missense variants achieved suggestive significant associations in our discovery GWAS (rs9272346, r²=0.76, D’=1 with rs1129740, allele A, allelic P_GC-corrected=9.54×10⁻⁸; OR, 1.97; 95% CI, 1.54 to 2.53; rs1063355, r²=0.65, D’=1 with rs1071630, allele G, allelic P_GC-corrected=1.60×10⁻⁷; OR, 1.93; 95% CI, 1.51 to 2.47), indicating that the susceptibility loci within the HLA region detected by SNP-based analyses may be shared across different populations. Notably, refinement of the HLA-DQA1 association was recently done by HLA imputation using exome array data in the same South Asian samples.⁵¹ HLA-DRB1*07:01 (P=4.6×10⁻⁶; OR, 2.51), HLA-DQA1*02:01 (P=5.5×10⁻⁶; OR, 2.49), HLA-DQB1*02:01 (P=1.0×10⁻⁶; OR, 2.28), and HLA-DQA1*01:01 (P=3.3×10⁻⁴; OR, 0.47) were significantly associated with childhood SSNS. However, HLA-DRB1*07:01 and HLA-DQB1*02:01 were not replicated in our study, because they are at a low allele frequency in the Japanese population (HLA-DRB1*07:01 [0.375%], HLA-DQB1*02:01 [0.134%], http://hla.or.jp). Considering that the LD patterns and allele frequencies in the HLA region are highly differentiated across populations, HLA fine-mapping after SNP-based association is essential for understanding the molecular mechanism of the disease.

A GWAS including patients with both primary and secondary adult-onset nephrotic syndrome was conducted in the Japanese population.²⁴ rs16946160 (allele A, P=6.0×10⁻¹¹ under recessive model; OR, 2.54; 95% CI, 1.91 to 3.40) in GPC5 was reported as a significant genetic factor contributing to the common pathway of massive proteinuria in acquired nephrotic syndrome, whereas the HLA region did not show significant association. The reported SNP was not replicated in Japanese children with SSNS in our study (rs16946160, P=0.60 under recessive model in the discovery stage). One possible explanation is differences in the disease spectra in the two studies. As the most common pathologic type in childhood SSNS, minimal change disease accounted for 91.5% of patients with renal biopsy in our study, whereas it accounted for only 18% of patients in the GWAS of adult nephrotic syndrome.

For autoimmune diseases with a main association signal coming from HLA class II region, such as type 1 diabetes, celiac disease, multiple sclerosis, SLE, inflammatory bowel disease, and narcolepsy, the reported variance explained by HLA alleles varies from 2% to 58%.^52,53 In this study, rs4642516 (the top SNP in HLA region) explained 9.7% (Nagelkerke’s pseudo-R²) of the disease variance in Japanese childhood SSNS. Considering the complicated LD structure within the HLA region, there may be strong intercorrelations between the index SNP in GWAS and the two independent classic HLA alleles (rs4642516 and HLA-DRB1*08:02, r²=0.02, D’=1; rs4642516 and HLA-DQB1*06:04, r²=0.21, D’=1 in combined dataset). We also built the logistic regression model using HLA-DRB1*08:02 and HLA-DQB1*06:04 and 8.1% of the disease variance was explained. The lower disease variance explained by the two HLA alleles compared to that explained by the top GWAS hit (rs4642516) indicated that there could be more independent HLA alleles contributing to the disease pathogenesis.

However, we did not identify any other association outside of the HLA region. The sample size placed limits on the ability to detect modest associations. GWASs with larger sample sizes will be necessary for identifying more candidate loci (especially in non-HLA regions).

Interestingly, the single-tag SNP rs3129888 exhibited high sensitivity and specificity for capturing HLA-DRB1*08:02 and HLA-DRB1*08:02-DQB1*03:02 in our study. The single-tag SNP rs3129888 could be an economic clinical biomarker for capturing susceptibility genetic factors.

Disclosures

K. Iijima received grants from Novartis Pharma K.K., the Japan Blood Product Organization, AbbVie LLC, JCR Pharmaceuticals Co., Ltd, Daiichi Sankyo Co., Ltd, Teijin Pharma Ltd, CSL Behring, Novo Nordisk Pharma Ltd, Air Water Medical Inc., Astellas Pharma Inc., Takeda Pharmaceutical Co., Ltd, Taisho Toyama Pharmaceutical Co., Ltd, Eisai Co. Ltd, and Biofermin Pharmaceutical Co., Ltd; lecture fees and/or consulting fees from Zenyaku Kogyo Co., Ltd, Novartis Pharma K.K., Chugai Pharmaceutical Co., Ltd, Astellas Pharma Inc., Springer Japan KK, Meiji Seika Pharma Co., Ltd, Asahi Kasei Pharma Corporation, Medical Review Co., Ltd, Nippon Boehringer Ingelheim Co., Ltd, Baxter Limited, Ono Pharmaceutical Co., Ltd, Sanwa Kagaku Kenkyusho Co., Ltd, Sanofi K.K., Alexion Pharma LLC, and Kyowa Hakko Kirin Co., Ltd.