GWAS of Hematuria

Visual Abstract

Abstract

Background and objectives

Glomerular hematuria has varied causes but can have a genetic basis, including Alport syndrome and IgA nephropathy.

Design, setting, participants, & measurements

We used summary statistics to identify genetic variants associated with hematuria in White British UK Biobank participants. Individuals with glomerular hematuria were enriched by excluding participants with genitourinary conditions. A strongly associated locus on chromosome 2 (COL4A4-COL4A3) was identified. The region was reimputed using the Trans-Omics for Precision Medicine Program followed by sequential rounds of regional conditional analysis, conditioning on previous genetic signals. Similarly, we applied conditional analysis to identify independent variants in the MHC region on chromosome 6 using imputed HLA haplotypes.

Results

In total, 16,866 hematuria cases and 391,420 controls were included. Cases had higher urinary albumin-creatinine compared with controls (women: 13.01 mg/g [8.05–21.33] versus 12.12 mg/g [7.61–19.29]; P<0.001; men: 8.85 mg/g [5.66–16.19] versus 7.52 mg/g [5.04–12.39]; P<0.001) and lower eGFR (women: 88±14 versus 90±13 ml/min per 1.72 m²; P<0.001; men: 87±15 versus 90±13 ml/min per 1.72 m²; P<0.001), supporting enrichment of glomerular hematuria. Variants at six loci (PDPN, COL4A4-COL4A3, HLA-B, SORL1, PLLP, and TGFB1) met genome-wide significance (P<5E-8). At chromosome 2, COL4A4 p.Ser969X (rs35138315; minor allele frequency=0.00035; P<7.95E-35; odds ratio, 87.3; 95% confidence interval, 47.9 to 159.0) had the most significant association, and two variants in the locus remained associated with hematuria after conditioning for this variant: COL4A3 p.Gly695Arg (rs200287952; minor allele frequency=0.00021; P<2.16E-7; odds ratio, 45.5; 95% confidence interval, 11.8 to 168.0) and a common COL4A4 intron 25 variant (not previously reported; rs58261427; minor allele frequency=0.214; P<2.00E-9; odds ratio, 1.09; 95% confidence interval, 1.06 to 1.12). Of the HLA haplotypes, HLA-B (*0801; minor allele frequency=0.14; P<4.41E-24; odds ratio, 0.84; 95% confidence interval, 0.82 to 0.88) displayed the most statistically significant association. For remaining loci, we identified three novel associations, which were replicated in the deCODE dataset for dipstick hematuria (nearest genes: PDPN, SORL1, and PLLP).

Conclusions

Our study identifies six loci associated with hematuria, including independent variants in COL4A4-COL4A3 and HLA-B. Additionally, three novel loci are reported, including an association with an intronic variant in PDPN expressed in the podocyte.

Podcast

This article contains a podcast at https://www.asn-online.org/media/podcast/CJASN/2022_04_26_CJN13711021.mp3

Introduction

Hematuria can originate from any location throughout the genitourinary tract. Glomerular hematuria has varied causes but is characteristic of Alport syndrome, IgA nephropathy, C3 nephropathy, and postinfectious GN (1). The genetic basis of Alport syndrome is well described, attributed to pathogenic variants in the type IV collagen genes, COL4A3, COL4A4, and COL4A5, encoding major glomerular basement membrane proteins (2,3). By contrast, the genetic basis of IgA nephropathy is more complex, with genome-wide association studies (GWAS) identifying numerous loci with small to moderate effects including in the MHC region of chromosome 6 (4). Recent GWAS using the UK Biobank and deCODE datasets (large genetically characterized adult cohorts intended to be a sample of the general population in the United Kingdom and Iceland, respectively) report COL4A3 and COL4A4 variants (which are adjacent) as statistically significant associated signals with hematuria or albuminuria (5). For Alport syndrome, the prevalence has been estimated to be approximately one in 50,000, accounting for the most severe cases that come to specialist attention (5). By contrast, population-based studies using deCODE and the UK Biobank data have illuminated the breadth of phenotypes in Alport syndrome (5,6). Concurrently, several family studies demonstrate that variants in COL4A3/4/5 masquerade as FSGS, once thought to be pathologically distinct from Alport syndrome (7,8).

A GWAS of dipstick hematuria by deCODE identified three loci, including a region encompassing COL4A4-COL4A3 and the MHC locus (9). Although no dedicated GWAS of hematuria has been reported in the UK Biobank, four groups have documented different variants in or near COL4A4-COL4A3 as strongly associated with albuminuria (10–13). For these reports, differing results despite using the same cohort could be accounted for by methodologic differences, including varying minor allele frequency (MAF) thresholds, using directly genotyped or differing reference datasets for imputation, and inclusion of albuminuria values below the limit of detection. Given the similar odds ratios (ORs), MAFs, and P values among the associated variants, we postulate that the different associations in COL4A4-COL4A3 were likely in strong linkage disequilibrium, tagging one variant, COL4A4 p.Ser969x (rs35138315), reported in multiple family studies of Alport syndrome.

To explore this hypothesis, we performed GWAS and conditional regional association analysis restricted to variants in a 2-Mb window around COL4A3 and COL4A4 with hematuria in the UK Biobank (13). To identify independent (i.e., not in linkage disequilibrium) single-nucleotide polymorphisms (SNPs), analysis was sequentially repeated using significant association signals from the previous analysis as covariates (14,15). A similar approach was subsequently applied to the MHC region, which was a strongly associated locus from our broader GWAS.

Materials and Methods

Study Design

The study was approved by the Toronto General Hospital Research Ethics Board (21–5361.0). Summary statistics from PheWeb were used to identify variants associated with hematuria in White British participants from the UK Biobank. The first four principal components of genetic ancestry were derived in cases and controls, which were then included in the model as covariates to adjust for population stratification. Cases and controls show evidence of overlapping distributions in terms of principal component space (genomic control λ=1.048) (Supplemental Figure 1). Subsequently, conditional regional association analysis restricting to neighboring genes on chromosome 2q36.3, COL4A4-COL4A3, chromosome 6 encompassing the MHC locus, and all other genome-wide significant loci was performed sequentially to identify independent variants in these regions.

UK Biobank Setting

Briefly, the UK Biobank is a prospective cohort study involving approximately 500,000 UK adults between the ages of 40 and 69 years at the time of recruitment, in whom genetic and phenotypic data are collected (16). Centrally collected and analyzed clinical/genetic data are available to investigators who apply for access, and summary statistics for thousands of traits are available through various resources, including PheWeb (https://pheweb.sph.umich.edu/). UK Biobank procedures include phenotypic data and biologic samples collection for all participants at baseline through questionnaire and linkage to health records.

Whole-Genome Genotyping and Centralized Analysis

The UK Biobank performed genotyping at the Affymetrix Research Services Laboratory, and array information has been previously described (16). The Haplotype Reference Consortium (HRC) data, consisting of primarily European genetic ancestry individuals and 39,235,157 SNPs, served as the main imputation reference panel. Imputation was also performed using the merged UK10K and the 1000 Genomes phase 3 reference panels (17,18). Imputed data were combined when an SNP was present in both panels.

For our analysis, we used the genome-wide summary statistics from PheWeb for hematuria on the basis of the HRC-imputed dataset. We supplemented the analysis with conditional analyses using individual-level data using Trans-Omics for Precision Medicine Program (TOPMed)–imputed genotypes for the chromosome 2 signal as well as UK Biobank–provided HLA haplotypes for the HLA chromosome 6 signal (described below).

Trans-Omics for Precision Medicine Program Imputation of the Chromosome 2:226000000–228000000 Region (GRCh38) of the UK Biobank

We used the TOPMed imputation server (https://imputation.biodatacatalyst.nhlbi.nih.gov/#!) to impute genetic variants in the chr2:226000000–228000000 region from the UK Biobank array data in batches of 20,000 individuals at a time given that the individual-level TOPMed-imputed UK Biobank dataset is not currently available (12). The batches were subsequently merged.

Imputed Histocompatibility Leukocyte Antigen Haplotypes

The HLA region on chromosome 6 was further examined using the provided HLA-imputed haplotypes field 22182 (16). HLA imputation was conducted by the UK Biobank using HLA*IMP:02 with modified settings using a set of genetically diverse reference datasets (19). The imputation procedure and quality control steps have been described (Motyer et al., unpublished data). In brief, for each locus, only the individuals who had laboratory-based HLA types for that locus and only the SNPs that were polymorphic and were typed in at least 98% of that set of individuals were included.

Phenotype Definition for the Case-Control Study

The R package PheWAS (https://github.com/PheWAS/PheWAS/blob/master/inst/doc/PheWAS-package.pdf) was used to map ICD9 and ICD10 codes (https://phewascatalog.org/phecodes; https://phewascatalog.org/phecodes_icd10cm) to phecode 593 “hematuria” (20). Phecodes are a high-throughput phenotyping tool on the basis of ICD codes that have been previously curated to rapidly define the status of thousands of clinically meaningful diseases and conditions.

Cases had the following ICD9 or ICD10 codes: “hematuria,” 599.7 (ICD9) and R31 (ICD10); “hematuria unspecified,” 599.70 (ICD9) and R31.9 (ICD10); and “gross hematuria,” 599.71 (ICD9) and R31.0 (ICD10).

To enrich for glomerular hematuria, exclusion codes were individuals with any genitourinary phecodes (individuals who have a phecode with the range 590–593.99; ICD9 and ICD10 codes underlying each phecode in this range were extracted to get the full list), and individuals who were not designated as a case and who did not meet exclusion codes were used as controls.

Clinical characteristics for cases and controls were determined. Urine albumin-creatinine ratios were computed as (microalbumin in milligrams per liter)/(creatinine in micromoles per liter per 1000). Individuals marked as having a microalbumin value below 6.7 mg/L (field 30505) were assigned as having a microalbumin value of 6.7 mg/L. eGFR was calculated in two ways using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation or as 100/cystatin C (milligrams per liter) (21).

Statistical Methods

SAIGE (version 0.44.5) was used to test variants for association with hematuria in the White British subset of the UK Biobank, where the effect allele is the nonreference allele (22).

Variables

The following covariates were included in analysis: sex, birth year, and the first four genetic ancestry principal components computed on the White British subset. Because SAIGE uses a genetic relationship matrix to properly account for relatedness, we retained related individuals. SAIGE performs QR transformation (decomposition of the covariate matrix to solve the linear least squares problem) of the covariate matrix by default, which adjusts for categorical covariates, such as birth year (23). For sex-stratified analyses on chromosome X, sex was not included as a covariate.

LocusZoom plots were generated for loci of interest (24).

deCODE Replication

Methods used in hematuria GWAS using the deCODE dataset have been previously described (9). In total, 151,677 Icelanders have been genotyped using Illumina SNP chips and genotype probabilities for untyped first- and second-degree relatives of chip-typed individuals calculated on the basis of the Icelandic genealogy database. Whole-genome sequence is also available for 15,220 Icelanders, and this was used for imputation. Hematuria phenotyping was performed with urine dipstick tests obtained from two laboratories in Iceland. Categorical analyses were performed, and cases were defined as (1) mild (at least one measurement of + or greater versus negative controls) or (2) moderate/severe (at least one measurement of ++ or greater versus negative controls).

Results

Participants

We assessed 16,866 (7074 women + 9792 men) hematuria cases and 391,420 (213,407 women + 178,013 men) controls, which includes related individuals. Individuals with any genitourinary phecodes were excluded, enriching for glomerular hematuria cases. The descriptive characteristics of the study population by genetic sex are presented in Table 1. There were significantly more men (9792) than women (7074) cases (P<0.001). The overall rate of hematuria was 4% (16,866 of 408,286) in the White British subset. The mean ages in years (baseline visit) by women and men were 61±7 years (women) and 60±7 years (men) for cases and 57±8 years (women) and 57±8 years (men) for controls, respectively. All comparisons were significantly different. The urinary albumin-creatinine ratio was significantly higher in cases (women: 13.01 mg/g [8.05–21.33]; men: 8.85 mg/g [5.66–16.19]) compared with controls (women: 12.12 mg/g [7.61–19.29]; men: 7.52 mg/g [5.04–12.39]), with a P value of P<0.001. The eGFR was significantly lower in cases (women: 88±14 ml/min per 1.72 m²; men: 87±15 ml/min per 1.72 m² by CKD-EPI) compared with controls (women: 91±13 ml/min per 1.72 m²; men: 90±13 ml/min per 1.72 m²), with similar results for 100/serum cystatin C. Systolic BP was statistically significantly higher in cases compared with controls, although the difference was nominal. The use of antihypertensive agents and body mass index were higher in cases than controls, separately by sex. Hearing aid use and self-reported hearing difficulty were more frequent in cases than controls, also separately by sex. Nonsex-specific characteristics of cases and controls are shown in Supplemental Table 1.

Table 1.

Descriptive statistics of the UK Biobank White British individuals used in the hematuria analyses by genetic sex (UK Biobank field 22001)

Variable	Women Only					Men Only
	Cases with Nonmissing Data	Controls with Nonmissing Data	Cases	Controls	t Test P Value	Cases with Nonmissing Data	Controls with Nonmissing Data	Cases	Controls	t Test P Value
Total	7074	213,407	7074	213,407	—	9792	178,013	9792	178,013	—
Age at baseline, yr, mean ± SD	7074	213,407	61±7	57±8	<0.001	9792	178,013	60±7	57±8	<0.001
No. of individuals on 2 arrays	7074	213,407	UK BiLEVE: 791 2043; UK Biobank Axiom: 6283	UK BiLEVE: 21,448; UK Biobank Axiom: 191,959	0.002 (chi square)	9792	178,013	UK BiLEVE: 1252; UK Biobank Axiom: 8540	UK BiLEVE: 21,094; UK Biobank Axiom: 156,919	0.005 (chi square)
PC1, mean ± SD	7074	213,407	−12.4±1.60	−12.36±1.61	0.14	9792	178,013	−12.3±1.62	−12.36±1.61	0.42
PC2, mean ± SD	7074	213,407	3.78±1.49	3.78±1.51	0.93	9792	178,013	3.78±1.50	3.78±1.50	0.95
Pulse rate, automated reading at baseline, bpm, mean ± SD	6565	199,182	71±11	71±11	<0.001	9166	166,507	69±13	68±12	<0.001
Systolic BP, automated reading at baseline, mm Hg, mean ± SD	6565	199,180	140±21	138±20.2	<0.001	9166	166,502	145±19.1	14±18	<0.001
Diastolic BP, automated reading at baseline, mm Hg, mean ± SD	6565	199,182	81±11	81±10.5	0.09	9166	166,507	84±11	84±11	0.63
Taking BP medication at baseline	7061	213,030	1010	21,438	<0.001 (chi square)	9781	177,726	1326	17,454	<0.001 (chi square)
Standing height at baseline, cm, mean ± SD	7046	213,014	162±6	163±6	<0.001	9752	177,618	175±7	176±7	<0.001
Weight at baseline, kg, mean ± SD	7041	212,861	72±14	72±14	<0.001	9740	177,510	88±15	86±14	<0.001
BMI at baseline, mean ± SD	7038	212,800	27.5±5.34	27.0±5.12	<0.001	9735	177,433	28.5±4.56	27.8±4.21	<0.001
Urinary albumin-creatinine ratio, mg/g, median (1st to 3rd percentile)a	6866	206,796	13.01 (8.05–21.33)	12.12 (7.61–19.29)	<0.001 (Mann–Whitney U test)	9533	173,444	8.85 (5.66–16.19)	7.52 (5.04–12.39)	<0.001 (Mann–Whitney U test)
eGFR, ml/min per 1.72 m2, mean ± SDb	7074	213,407	88±14	91±13	<0.001	9792	178,013	87±15	90±13	<0.001
eGFR computed as 100/serum cystatin C (mg/L) at baseline, mean ± SD	6771	203,329	112±19	117±18	<0.001	9351	169,805	104±17	109±16	<0.001
Hearing aid user (yes; no)	3964	120,298	267; 3697	5367; 114,931	<0.001 (chi square)	6462	115,555	542; 5920	6509; 109,046	<0.001 (chi square)
Hearing difficulty/problems via touchscreen (yes; no)	6755	203,992	1757; 4998	43,786; 160,206	<0.001 (chi square)	9464	171,974	3448; 6016	53,920; 118,054	<0.001 (chi square)

PC1, 1st principal component; PC2, 2nd principal component; BMI, body mass index.

^aIndividuals marked as having a microalbumin value below 6.7 mg/L (field 30505) were assigned as having a microalbumin value of 6.7 mg/L. Then, the ratio was computed as (microalbumin in milligrams per liter)/(creatinine in micromoles per liter per 1000).

^beGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation or as 100/serum cystatin C (milligrams per liter) (21).

Case-Control Study

Summary statistics from PheWeb, in which centrally HRC-imputed UK Biobank data were used, identified variants associated with hematuria in the White British UK subset. GWAS identified six loci (P<0.001) (Figure 1, Table 2, Supplemental Figure 2) (22,25). The following genes were closest to the strongest associated variants in each loci: PDPN, IRS1, HLA-B, SORL1, PLLP, and TGFB1 (Figure 2). Of these, three have not been previously reported (closest genes: PDPN, SORL1, and PLLP).

Figure 1.

Manhattan plot of associated signals in hematuria using centrally Haplotype Reference Consortium (HRC)–imputed UK Biobank data. Six statistically significant loci were identified. The x axis is the chromosome location from chromosome 1 to 22, and the y axis is the –log₁₀(P value). The horizontal gray dashed line represents the threshold for statistical significance at P<0.001. The figure was generated using summary statistics from https://pheweb.org/UKB-SAIGE/pheno/593 in R version 4.0.5.

Table 2.

Associated variants with hematuria in genome-wide association studies followed by conditional regional analysis at each locus

Chromosome:Position	Reference/ Alternate Allele	rs Identification	Nearest Gene(s) and Location	Minor Allele Frequency (Minor Allele)	P Value	Odds Ratio (95% Confidence Interval)	Imputation Quality
GRCh37
1:13,921,934	A/G	rs2885134	PDPN (intron 1)	0.029 (G)	5.90E-9	1.22 (1.14 to 1.31)	0.958
HLA-B*0801	—	—	—	0.14	4.41E-19	0.84 (0.82 to 0.88)	N/A
11:121,584,573	T/A	rs618048	SORL1 (downstream 3′)	0.22 (T)	1.30E-11	0.91 (0.88 to 0.94)	0.999
19:41,826,020	A/C	rs56254331	CCDC97 (intron 3), TGFB1 (downstream 3′)	0.17 (C)	8.80E-16	0.89 (0.86 to 0.91)	0.999
16:57,349,346	G/T	rs948705	PLLP (intergenic with CCL22)	0.15 (G)	2.30E-9	1.10 (1.07 to 1.14)	0.976
GRCh38
chr2:227052367 (most significant variant in locus)	G/C	rs35138315	COL4A4 (p.Ser969x, exon 32)	0.00035 (C)	7.95E-35	87.3 (47.9 to 159.0)	0.567
chr2:227077364 (results conditioned on rs35138315)	T/C	rs58261427	COL4A4 (intron 25)	0.214 (C)	2.00E-9	1.09 (1.06 to 1.12)	0.987
chr2:227277511 (results conditioned on rs35138315)	G/A	rs200287952	COL4A3 (p.Gly695Arg, exon 28)	0.00021 (A)	2.16E-7	45.5 (11.8 to 168.0)	0.268

For GRCh37, a genome-wide association study was performed using centrally Haplotype Reference Consortium–imputed UK Biobank data, in which variants are reported in reference to GRCh37. After conditioning on the most significant variant at each locus, all other associations in the region became nonsignificant. For GRCh38, reimputation of a 2-Mb region encompassing COL4A4-COLA43 was performed using the Trans-Omics for Precision Medicine Program (reference GRCh38) followed by conditional regional analysis to identify independent associated variants. Reported odds ratios refer to the alternate allele.

Figure 2.

LocusZoom plot at six loci showing associated hematuria variants using centrally HRC-imputed UK Biobank data (and Trans-Omics for Precision Medicine Program–imputed data at chromosome 2). The following genes were closest to the strongest associated variants at these loci: PDPN, COL4A4, HLA-B, SORL1, PLLP, and TGFB1. PDPN, SORL1, and PLLP are novel associations. This potentially highlights immune regulation (HLA-B) and podocytes in mediating hematuria. The x axis is the chromosome location, and the y axis is the –log₁₀(P value). The horizontal gray dashed lines represent the threshold for statistical significance at P<0.001. Blue lines above the plots denote locations of variants present in the European Bioinformatics Institute–National Human Genome Research Institute genome-wide association studies catalog. The figure was generated using summary statistics from https://pheweb.org/UKB-SAIGE/pheno/593.

The strongest association was observed for rs71431010 (MAF=0.00055; P<0.001; OR, 105; 95% confidence interval [95% CI], 52 to 231), which is intergenic, and the closest gene is IRS1. This variant was also associated with albuminuria in a GWAS using the UK Biobank (13). The variant is 1.4 Mb from the COL4A4-COL4A3 region.

The associated region on chromosome 2q36.3, where COL4A-COL4A3 resides, was examined more closely. COL4A4 p.Ser969x (rs35138315) is directly genotyped on the UK Biobank arrays, but HRC imputation loses it. This variant is important because it is associated with albuminuria in the UK Biobank, and it is reported in multiple cases of autosomal recessive Alport syndrome and autosomal dominant FSGS (11,26–29). Thus, we utilized TOPMed for regional imputation of this region of chromosome 2, enabling prediction of variants with lower frequency, including putative loss-of-function alleles (17,30).

We assessed the 2 Mb surrounding COL4A4-COL4A3 using TOPMed imputation in order to capture the rare variants in the locus, which identified 13 variants (P<0.001). The strongest associated variant was COL4A4 p.Ser969x (rs35138315; MAF=0.00035; P<0.001; OR, 87.3; 95% CI, 47.9 to 159.0) (Table 2). We then performed conditional regional association analysis conditioned on rs35138315, leading to most variants no longer being significantly associated; this represents strong linkage disequilibrium among them. Two additional variants, however, remained significant (P<0.001) and were not previously reported in UK Biobank studies (Figure 2, Table 2). These include COL4A3 p.Gly695Arg (rs200287952; MAF=0.00021; P<0.001; OR, 45.5; 95% CI, 11.8 to 168.0) and a common variant in intron 25 of COL4A4 (rs58261427; MAF=0.214; P<0.001; OR, 1.09; 95% CI, 1.06 to 1.12), which is ∼500 base pairs from the nearest exon (exon 25). Using pheweb.org, we interrogated results to examine the association of the three COL4A4-COL4A3 variants with other traits (Supplemental Table 2). Sex-stratified analysis of COL4A5 variants also did not identify any statistically significant associations.

Of the other hematuria GWAS signals, the second strongest association was in HLA-B. We further performed conditional regional analysis using UK Biobank–provided imputed HLA haplotypes. Of 362 haplotypes, there were six associated at P<0.001, but none remained significant after conditional analysis with HLA-B*08:01 (MAF=0.14; P<0.001; OR, 0.84; 95% CI, 0.82 to 0.88) (Supplemental Table 3). HLA haplotypes are not single-nucleotide sequence variants, but rather, they are several changes in a haplotype and are denoted numerically (31). In the only other reported hematuria GWAS using deCODE data, association for a variant in the immediate neighboring gene on chromosome 6, HLA-C (6:31271845C>A; hg38; the authors stated that this is p.Asp33Tyr, but it is residue 9; MAF[C allele]=0.4275; P<0.001; OR, 0.92; 95% CI, 0.90 to 0.94), was reported (Table 3). The variant identified by deCODE is not available in either the HRC-imputed or HLA-imputed data in the UK Biobank. We replicated three loci (two of which are the same variant) in the deCODE hematuria GWAS (Table 3).

Table 3.

Loci from a hematuria genome-wide association study using deCODE data were replicated (9)

Gene or Loci	deCODE							UK Biobank
	Variant	Minor Allele	Minor Allele Frequency	P Value	Odds Ratio	95% Confidence Interval	Phenotype	Variant	Minor Allele	Minor Allele Frequency	P Value	Odds Ratio	95% Confidence Interval
Previously published by deCODE (9)
COL4A3	Gly695Arg	A	0.0003	<0.001	5.46	2.94 to 10.16	≥++ versus −	Gly695Arg	A	0.00021	<0.001	45.5	11.8 to 168
COL4A3	2.5-kb deletion (Gly289_Lys330del)	—	0.0006	<0.001	11.78	7.26 to 19.11	≥++ versus −	Not examined	—	—	—	—	—
COL4A3	rs760545501	T	0.0017	<0.001	2.36	1.80 to 3.09	≥++ versus −	Not examined	—	—	—	—	—
TGFB1	rs56254331	C	0.18	<0.001	0.92	0.89 to 0.94	≥+ versus −	rs56254331	C	0.17	<0.001	0.89	0.86 to 0.91
MHC	Chr 6:31271845(hg38) HLA-C	C	0.43	<0.001	0.92	0.90 to 0.94	≥+ versus −	HLA-B*08:01	—	0.14	<0.001	0.84	0.82 to 0.88
Not previously reported by deCODE
SORL1	rs618048	T	0.15	<0.001	1.08	1.05 to 1.11	≥+ versus −	rs618048	T	0.22	<0.001	1.10	1.08 to 1.11
PDPN	rs2885134	G	0.03	0.006	1.09	1.02 to 1.15	≥++ versus −	rs2885134	G	0.029	<0.001	1.22	1.14 to 1.31
PLLP	rs948705	G	0.14	<0.001	0.93	0.90 to 0.96	≥+ versus −	rs948705	G	0.15	<0.001	0.91	0.89 to 0.92

Variants, minor allele frequencies, odds ratios, and 95% confidence intervals at associated loci are shown. Reported odds ratios refer to the minor allele. Hematuria is defined in deCODe by dipstick and quantitated as ≥+ or ≥++. —, minor allele (allele with frequency <50%).

For the remaining four loci, the same conditional regional approach was performed using the HRC-imputed White British UK Biobank subset with the strongest associated variant as a covariate (Table 2). After conditioning on the top variant at the locus, all other associations became nonsignificant (P<0.001).

Of these, three were associations with hematuria that have not been previously reported. Thus, the association was analyzed in the deCODE dataset and was found to be replicated with the same direction of effect, although nominally for the PDPN variant (Table 3, Supplemental Table 4). Cell- and tissue-specific expression along with associated phenotypes from PheWeb are shown in Supplemental Table 5.

Discussion

Here, we present the first dedicated GWAS of hematuria using the UK Biobank data with several novel findings. Six loci that are genome-wide significantly associated with hematuria are identified, including three that have not been previously reported (nearest genes: PDPN, SORL1, and PLLP). For the known loci residing on chromosome 2q36.3 encompassing COL4A4-COL4A3, regional reimputation and sequential association analysis were performed, leading to the identification of three independent SNPs that include COL4A4 p.Ser969x (rs35138315), COL4A3 p.Gly695Arg (rs200287952), and a common variant not previously reported in intron 25 of COL4A4 (rs58261427). Similar fine-mapping analysis was performed for chromosome 6 encompassing the MHC region, identifying one previously unreported common variant in HLA-B*08:01 as driving the association with hematuria. Of the four remaining associated loci, three have not been previously reported, but we show that they are replicated in the deCODE dataset.

Hematuria is a typical manifestation observed with glomerular basement membrane defects as evidenced in Alport syndrome but variably in primary podocyte disorders. Of the three novel loci, one is an intronic variant in PDPN (rs2885134) (Table 2), which encodes podoplanin, reported to anchor to the podocyte cytoskeleton through binding with ezrin (32). PDPN is also reported to be expressed in podocytes by a single-cell sequencing atlas (http://humphreyslab.com/SingleCell/displaycharts.php). In Genotype-Tissue Expression (GTEx) V8, this SNP is a cis-eQTL for PDPN in aorta (P<0.001; with the minor allele G having lower expression; https://gtexportal.org/home/snp/rs2885134) (33). This finding thus potentially informs filtration barrier biology, highlighting a role of podocytes in protecting against glomerular hematuria.

The variant with the closest gene PLLP (rs948705) (Table 2) is a cis-eQTL for CX3CL1 (P<0.001; with the minor allele G having higher expression) in adrenal tissue from GTEx (https://gtexportal.org/home/snp/rs948705) (33). It is also in strong linkage disequilibrium with a pQTL for CX3CL1 in plasma (34).

For chromosome 2, the strongest associated variant was intergenic, closest to IRS1 with the HRC-imputed dataset, but reimputing the region with TOPMed reveals that the association is actually driven by COL4A4 p.Ser969x (rs35138315). Previous GWAS of albuminuria in the UK Biobank identified different SNPs in the region, with two studies reporting COL4A4 p.Ser969x (rs35138315) as a strongly associated signal (11,12). For these studies, direct genotyping or imputed data on the basis of TOPMed were used. The available HRC-imputed dataset that we used in our GWAS loses rs35138315, despite the fact that it was directly genotyped on the array. Our fine-mapping strategy demonstrates that the 13 associated variants on chromosome 2 are in linkage disequilibrium, revealing that only three are actually independently associated with hematuria. These three variants are observed to have an association with kidney disease–relevant or Alport syndrome traits (Supplemental Table 2).

The UK Biobank is not a random sample, with only a 6% participation rate, and the true prevalence of diseases could be higher (16,35). The COL4A4 p.Ser969x (rs35138315) has been reported as causing autosomal dominant FSGS and autosomal recessive Alport syndrome (26–29). In one report, rs35138315 was the most common pathogenic variant detected in a series of autosomal recessive Alport syndrome, accounting for 23% of cases (27). Some of these patients were homozygous for the variant, whereas others were compound heterozygotes with another COL4A4 variant. Five of six patients with this variant were recorded to have sensorineural hearing loss, and the ages of kidney failure, where known, were 18, 22, and 28 years.

Another independently associated variant is COL4A3 p.Gly695Arg (rs200287952), which has been implicated in thin basement nephropathy and Alport syndrome by several groups (6,29,36,37). Recently, this variant has been shown to have a higher frequency in hematuria cases (0.13%) than controls (0.048%) from the 100,000 Genomes Project (6). Additionally, it is reported in a GWAS of dipstick hematuria using the deCODE dataset (MAF=0.00021; P<0.001; OR, 5.46; 95% CI, 2.94 to 10.16) (9). The third independent signal is a common COL4A4 intron 25 variant, which has not been previously associated with kidney phenotypes. Analysis of this variant in GTEx (https://gtexportal.org/home/snp/rs58261427) suggests that it is producing an alternative COL4A4 transcript from the fifth last exon (exon 44) in thyroid tissue. However, there is no independent evidence for this. Furthermore, it is also possible that this variant could be in linkage disequilibrium with a pathogenic coding sequence variant not genotyped or imputed, which can occur for a number of reasons, including if it is a structural variant.

The second most strongly associated locus in our hematuria GWAS was a variant within the MHC locus on chromosome 6 (HLA-B*08:01; MAF=0.14; P<0.001; OR, 0.84; 95% CI, 0.82 to 0.88). At this locus, we also performed conditional analysis using imputed HLA haplotypes, going from six associated variants to only one (as a result of linkage disequilibrium) when the most statistically significant haplotype is conditioned on (38). In the only other reported hematuria GWAS using deCODE data, association for a variant in a neighboring gene on chromosome 6, HLA-C, was described (Table 2) (9). This specific variant is chromosome 6:31271845 (hg38), where the reference is C and also the minor allele with the alternate allele A. The variant identified by deCODE is not available in either the HRC-imputed or HLA-imputed data in the UK Biobank. Instead, the most significant association in the UK Biobank is with the immediate neighbor gene HLA-B*0801, which is approximately 80 kb away. The reported stratified analysis revealed nominally significant sex differences in effect sizes with OR of 0.92 (95% CI, 0.90 to 0.94; P<0.001) for men and OR of 0.90 (95% CI, 0.87 to 0.92; P<0.001) for women. Additionally, three GWAS of IgA nephropathy have revealed associations with variants in the chromosome 6 MHC region (39–41). These papers report associations with HLA-DQA1*0101, HLA-DQA1*0102, HLA-DQB1*0201, and HLA-DQB1*0301 (41,42). In the UK Biobank White British subset, there is no evidence for association of HLA-DQA1*0101 and HLA-DQ*0102 with hematuria (Supplemental Table 3). The initial association for HLA-DQB1*0201 (OR, 0.88; 95% CI, 0.86 to 0.91) and HLA-DQB1*0301 (OR, 1.07; 95% CI, 1.04 to 1.10) is lost after conditioning on HLA-B*0801, suggesting that it is due to linkage disequilibrium (Supplemental Table 3). The r², D’, and haplotype frequencies formed by pairs of these haplotypes are shown in Supplemental Tables 6–10. HLA-B*0801 and HLA-DQB1*0201 are positively correlated, whereas HLA-B*0801 and HLA-DQB1*0301 are negatively correlated, and, as expected, HLA-DQB1*0201 and HLA-DQB1*0301 are negatively correlated. Identifying the causal variant(s) at MHC will benefit from higher-resolution imputation and/or direct HLA sequencing (43).

We replicated six loci in the deCODE hematuria GWAS. One locus is at the MHC region on chromosome 6 (Table 2). Two are the same variant, including an intronic variant in TGFB1 (rs56254331; MAF[C]=0.1833; P<0.001; OR, 0.92; 95% CI, 0.89 to 0.94) with the same direction of effect and COL4A3 p.Gly695Arg (rs200287952; MAF, 0.00021; P<0.001; OR, 3.41; 95% CI, 1.18 to 9.89) (Table 2) (9). TGFB1 rs56254331 (MAF[C]=0.1833; P<0.001; OR, 0.92; 95% CI, 0.89 to 0.94) has previously been associated with urinary albumin-creatinine ratio in the UK Biobank (45).

A difference between the UK Biobank and the deCODE hematuria GWAS is the method of phenotyping. For deCODE, hematuria was defined by dipstick measurements (quantitated as ≥+ or ≥++), but for the UK Biobank, it was by ICD9/10 codes. In the UK Biobank, the hematuria rate was 4% (16,866 of 408,286). In deCODE, the rate was significantly higher, with 22% (18,839 of 87,742) having mild hematuria and 42% having (49,212 of 118,115) moderate or severe hematuria. Notably, the ORs in replicated loci in the UK Biobank compared with deCODE were higher, likely speaking to the increased specificity in using ICD9/10 codes compared with dipstick measurements for phenotyping.

The clinical characteristics in Table 1 also support that the approach of using ICD9/10 codes to enrich for glomerular hematuria cases is specific, with cases compared with controls having higher urinary albumin-creatinine ratios and lower eGFR measurements.

In the future, sequencing as opposed to genotyping and imputation in the UK Biobank will provide more resolution at associated regions, overcoming the limitations of this study. For power, we only included data for the largest genetic ancestry group (White British), which limits generalizability to other ancestral groups.

In summary, the UK Biobank is a genetically characterized population-based cohort that has facilitated a range of large-scale GWAS. We present results of an analysis focused on glomerular-enriched hematuria, revealing novel loci and insights into glomerular filtration barrier biology. Additionally, we provide independent associated SNPs, including novel ones, within the strongest associated regions, the region that encompasses COL4A4-COL4A3 and the MHC region. This work is important to understand how to prioritize which variants to study, a successful approach in cystic fibrosis where different classes of therapies are dictated by the pathogenic variant and purported mechanism of action (45).

This manuscript was written to conform to STREGA recommendations (46).

Disclosures

M. Barua reports ownership interest in AstraZeneca, honoraria from Natera, serving on the editorial board of Glomerular Diseases, and speakers bureau for Sanofi. D.F. Gudbjartsson, K. Stefansson, P. Sulem, and G. Sveinbjornsson report employment by deCODE Genetics, which is fully owned by Amgen. All remaining authors have nothing to disclose.

Funding

M. Barua and A.D. Paterson were coprincipal investigators on a grant funded by the Alport Syndrome Foundation in 2020 and are supported by the Canadian Institutes of Health Research Project Grants 437528 and 427165. Funding has also been donated by the Toronto General Hospital Foundation. This work was also supported by FRQS.

Author Contributions

A.D. Paterson conceptualized the study; M. Barua, S.A. Gagliano Taliun, D.F. Gudbjartsson, K. Stefansson, P. Sulem, and G. Sveinbjornsson were responsible for data curation; M. Barua, S.A. Gagliano Taliun, D.F. Gudbjartsson, K. Stefansson, P. Sulem, and G. Sveinbjornsson were responsible for investigation; M. Barua and S.A. Gagliano Taliun were responsible for formal analysis; M. Barua, S.A. Gagliano Taliun, and A.D. Paterson were responsible for visualization; M. Barua wrote the original draft; and S.A. Gagliano Taliun and A.D. Paterson reviewed and edited the manuscript.

Data Sharing Statement

All data used in this study are available in this article.

Supplemental Material

This article contains the following supplemental material online at http://cjasn.asnjournals.org/lookup/suppl/doi:10.2215/CJN.13711021/-/DCSupplemental.

Supplemental Figure 1. PCA1-4 in cases and controls in the UK Biobank (available at UK Biobank data field 22009).

Supplemental Figure 2. Q-Q plot for centrally HRC-imputed hematuria GWAS.

Supplemental Table 1. Descriptive characteristics of the UK Biobank White British subset used in the hematuria analyses.

Supplemental Table 2. Association of three independent COL4A4-COL4A3 variants with other kidney disease–relevant or Alport syndrome traits using the TOPMed-imputed UK Biobank dataset.

Supplemental Table 3. GWAS via SAIGE using HLA haplotypes provided by the UK Biobank on White British for ICD-based case-control hematuria.

Supplemental Table 4. Novel hematuria loci in the UK Biobank are replicated in the deCODE dataset, although association with variant and nearest gene PDPN is nominal.

Supplemental Table 5. Cell- and tissue-specific expression of novel loci in the hematuria GWAS.

Supplemental Table 6. Estimated pairwise linkage disequilibrium between UK Biobank–imputed haplotypes HLA-B*0801, HLA-DQB1*0201, and HLA-DQB1*0301 for the White British hematuria cases and controls.

Supplemental Table 7. HLA haplotype frequencies.

Supplemental Table 8. Estimated haplotype frequencies for HLA-B*0801 and HLA-DQB1*0201 in White British in the UK Biobank.

Supplemental Table 9. Estimated haplotype frequencies for HLA-B*0801 and HLA-DQB1*0301 in White British in the UK Biobank.

Supplemental Table 10. Estimated haplotype frequencies for HLA-DQB1*0201 and HLA-DQB1*0301 in White British in the UK Biobank.