Meta-GWAS Reveals Novel Genetic Variants Associated with Urinary Excretion of Uromodulin

Significance Statement The mechanisms regulating the urinary excretion of uromodulin remain mostly unknown. A meta-GWAS conducted in 29,315 individuals from 13 cohorts identified two novel, genome-wide significant loci, KRT40 and WDR72, in addition to the previously known UMOD-PDILT locus, to be associated with urinary uromodulin. KRT40 colocalizes with uromodulin in TAL cells and functional studies showed that its expression affects the processing and apical excretion of uromodulin. WDR72, which does not colocalize with uromodulin, has been associated with kidney function, urinary acidification, and kidney stones. These studies provide novel insights into the biology of uromodulin and keratins and into the influence of the UMOD-PDILT locus on kidney function. Background Uromodulin, the most abundant protein excreted in normal urine, plays major roles in kidney physiology and disease. The mechanisms regulating the urinary excretion of uromodulin remain essentially unknown. Methods We conducted a meta-analysis of genome-wide association studies for raw (uUMOD) and indexed to creatinine (uUCR) urinary levels of uromodulin in 29,315 individuals of European ancestry from 13 cohorts. We tested the distribution of candidate genes in kidney segments and investigated the effects of keratin-40 (KRT40) on uromodulin processing. Results Two genome-wide significant signals were identified for uUMOD: a novel locus (P 1.24E–08) over the KRT40 gene coding for KRT40, a type 1 keratin expressed in the kidney, and the UMOD-PDILT locus (P 2.17E–88), with two independent sets of single nucleotide polymorphisms spread over UMOD and PDILT. Two genome-wide significant signals for uUCR were identified at the UMOD-PDILT locus and at the novel WDR72 locus previously associated with kidney function. The effect sizes for rs8067385, the index single nucleotide polymorphism in the KRT40 locus, were similar for both uUMOD and uUCR. KRT40 colocalized with uromodulin and modulating its expression in thick ascending limb (TAL) cells affected uromodulin processing and excretion. Conclusions Common variants in KRT40, WDR72, UMOD, and PDILT associate with the levels of uromodulin in urine. The expression of KRT40 affects uromodulin processing in TAL cells. These results, although limited by lack of replication, provide insights into the biology of uromodulin, the role of keratins in the kidney, and the influence of the UMOD-PDILT locus on kidney function.

UMOD monomers assemble into homopolymeric filaments, which encapsulate and aggregate uropathogens, such as type-1 fimbriated Escherichia coli, promoting their clearance in the urine. 3,4 At the level of tubular cells, UMOD regulates apical transport systems operating in the TAL and in the DCT, 5,6 modulating salt reabsorption and blood pressure control. 7,8 Multilevel evidence supports the role of UMOD, the gene coding for UMOD, in a spectrum of kidney disorders. Rare missense mutations of UMOD are the most common cause of autosomal dominant tubulointerstitial kidney disease (ADTKD), a disease entity characterized by tubular damage and interstitial fibrosis in the absence of glomerular lesions, progressing to kidney failure. 9 ADTKD-UMOD is caused by a toxic gain of function mechanism, with accumulation of intracellular aggregates of mutant UMOD in the TAL, leading to tissue damage and kidney fibrosis. [9][10][11][12] In parallel, genome-wide association studies (GWAS) have consistently associated the UMOD locus with the eGFR and the risk for developing CKD in the general population. 13,14 Remarkably, the UMOD locus has a relatively large effect size on eGFR and CKD risk, consistent across most ethnic groups studied so far. The top GWAS risk variants map in the same linkage disequilibrium (LD) block encompassing the promoter of UMOD, and they are associated with an increased expression of UMOD. 7,15 Although the roles of UMOD are increasingly recognized, our knowledge about the mechanisms regulating its production in tubular cells and its excretion into the urine remains limited. Recent studies have shown that the excretion of UMOD is functionally linked with the activity of transport processes operating in the TAL. [16][17][18] Using a meta-GWAS approach performed on 10,884 individuals of European descent from six cohorts, we previously identified common variants within the promoter of UMOD as the single genome-wide significant locus associated with the levels of UMOD in urine. 19 The top UMOD promoter variant, rs12917707, was associated in a dose-dependent fashion with the urinary levels of UMOD (uUMOD), 19 confirming the biologic link between these variants and the expression of UMOD in kidney and urine.
To gain further insights into the factors regulating the production and excretion of UMOD, and increase the statistical power to detect novel loci associated with urinary excretion of UMOD, we conducted a meta-GWAS in 29,315 individuals from 13 cohorts of European ancestry. We performed detailed expression studies and investigated the biologic relevance of the novel genome-wide significant KRT40 locus in the processing of UMOD.

Cohorts
The concentrations of urinary UMOD and creatinine were measured in 29 Table  1. UMOD and UMOD indexed to creatinine (uUCR) were inverse-normal transformed before adjusting for age, sex, and relatedness. Each study conducted linear regression analysis of the residuals of uUMOD and uUCR from genotyped and either 1000G Phase 3, HapMap, or Haplotype Reference Consortium imputed single nucleotide polymorphisms (SNPs) using an additive model with appropriate statistical software (Supplemental Table 2). 20-23 The Hg19 genome build was used for the reference panel. The genetic kinship matrix fitted for all family-based cohorts was calculated using the "ibs" function in GenABEL/ProbABEL, 22 and GWAS was performed using various software listed in Supplemental Table 2.

UMOD and Creatinine Measurement
Spot urine samples were collected and were frozen and stored before analysis (Supplemental Table 1). Urinary UMOD levels were measured using a well-established ELISA as described previously. 24 Human UMOD (AG 733; EMD Millipore, Temecula, CA) was used to determine the standard curve. For the capture antibody, a sheep anti-human UMOD antibody was used (K90071C; Meridian Life Science, Memphis, TN), a mouse monoclonal anti-human UMOD antibody (CL 1032A; Cedarlane Laboratories, Burlington, NC) was used as the primary antibody, and a goat antimouse IgG (H1L) horseradish peroxidase-conjugated protein (172.1011; BioRad Laboratories, Inc., Hercules, CA) as the secondary antibody. The detection range for the assay was between 3.9 ng/ml and 500 ng/ml (Supplemental Table  1). Urinary UMOD levels in samples from FHS were measured by the Rules-Based Medicine array (Rules-Based Medicine, Inc., Austin, TX) using immunoassay with a bead Luminex platform. 19 UMOD levels were expressed as uUMOD (mg/ml) or uUCR (mg/g creatinine).

Statistical Analyses (GWAS)
The GWAS summary output files from the 13 cohorts were processed for quality control using the EasyQC package, 25 which filtered out SNPs with an effect allele frequency of ,0.01 and imputation quality of ,0.4. Furthermore, we excluded SNPs with missing values in each cohort, carried out allele and marker name harmonization to the 1000 genomes reference panel, and performed frequency checks to output a clean and harmonized GWAS file for each of the cohorts, which were subsequently used for meta-analyses. The inversevariance weighted fixed-effects method implemented in METAL (v2011-03-25), 26 software was used to conduct metaanalyses of uUMOD and uUCR levels. A genomic control coefficient was computed for each cohort and was used to correct P values for any cryptic relatedness or population stratification by METAL. SNPs that were present in at least two cohorts were used to identify genome-wide significant and suggestive loci and to create Manhattan plots. Meta-analysis using sample sizes and P values was also calculated using METAL to account for any heterogeneity. Meta-analysis on urinary UMOD values adjusted for age and sex were compared against meta-analysis on urinary UMOD using age, sex, and eGFR as covariates in seven available cohorts (GS:SFHS, VIKING, CROATIA-Korcula, CROATIA-Split, CROATIA-Vis, GCKD and CoLaus, n520,620), to account for kidney function. eGFR was calculated using the CKD Epidemiology Collaboration equation.
A threshold of P#5.0E-08 was used to determine significant associations after a second correction for calculated genomic inflation factor (ø) from the meta-analysis. Annotation of association results was performed using Haploreg, 27 SNiPa, 28 and the UCSC genome browser. 29 A suggestive threshold was defined as P,1.0E-05. Locus-Zoom was used to create regional association maps, thus identifying other genome-wide significant and suggestive SNPs in LD with the SNP of interest and other genes within each locus. 30 Spearman's rank correlation was used to compare the meta-GWAS effect sizes of SNP of interest, in association with uUMOD and uUCR. Genome-wide significant SNPs and genes were investigated for the presence of candidate expression quantitative trait loci (eQTLs) using the Genotype-Tissue Expression (GTEx) database v8. 31 Versatile A Gene-based Association Study (VEGAS2) was performed on summary metadata. 32 SNPs with the lowest P value in novel genome-wide significant loci were queried for association with eGFR, urinary creatinine, and plasma creatinine levels using UK Biobank GWAS summary statistics in the Global Biobank Engine. [33][34][35]

Conditional Analyses
The GCTA COJO Slct algorithm was applied on summary metadata to select independently associated SNPs using a stepwise model selection procedure. 36 Default P value cutoff of 5.0E-08 and collinearity cutoff of 0.9 were used. The reference panel used was Phase 3 1000G (Europeans only). Conditional analysis on the SNPs identified as independent by GCTA COJO in UMOD and PDILT was carried out on uUMOD and uUCR in the GS:SFHS cohort.

Candidate Gene Analyses
A list of genes expressed in the TAL and associated with rare monogenic disorders affecting the TAL or with ADTKD was compiled (Supplemental Table 3). 9,37 All SNPs in each gene were queried in the meta-analysis results and a gene-specific threshold was calculated as previously described. 19 The gene region was defined as 61 kb from gene start and end position as listed in Ensembl for each of the genes. For each gene, region-specific multiple testing was carried out by calculating a gene-specific threshold, namely, 0.05 divided by the number of found LD blocks at r 2 50.2. Variants that had a P value lower than genespecific thresholds were declared significant. The 1000G dataset and r 2 value of 0.1 was used to define LD blocks. SNPs with minor allele frequency ,0.01 and imputation quality ,0.4 were removed before analysis.
To determine if there was a stronger association between genes expressed in the TAL compared with other segments, a simple gene query was carried out using 10 genes expressed in specific segments (Supplemental Table 4) and the genes that were close to significance in the previous meta-GWAS for UMOD (SORL1, CAB39, FAM83A, and MARCH1). 19 As above, the gene region was defined as 61 kb from the gene start and end position.
Mouse Kidney Samples and Microdissection of Nephron Segments C57BL/6J mice were housed in a light-and temperaturecontrolled environment with ad libitum access to tap water and standard chow (Diet AO3, SAFE; 25/18 GR Mucedola Srl, Settimo Milanese, Italy). Mice were sacrificed by cervical dislocation after anesthesia with isoflurane (Minrad International Inc., Orchard Park, NY, USA) for kidney collection. One kidney was used to obtain material for the primary mouse TAL (mTAL) cell culture, whereas the other was further processed for histologic analyses. Kidney tubule segments were prepared from mouse kidneys briefly digested with Liberase (Roche, Basel, Switzerland), with manual isolation of the segments according to morphologic criteria as described. 38 The samples were lysed in either Radioimmunoprecipitation assay buffer (RIPA buffer) for immunoblotting experiments or in the RNA lysis buffer from the RNAqueous Total RNA Isolation Kit (Invitrogen, Carlsbad, CA, USA) for transcript analysis. Quantitative RT-PCR was performed on pools of approximately 70 samples for each isolated fraction. The experiments were performed in accordance with the ethical guidelines at University of Zurich (Zurich, Switzerland) and the legislation of animal care and experimentation of Canton Zurich, Switzerland (Gesundheitsdirektion Veterin€ aramt; protocol ZH049/17).
Isolation, Culture, and Treatment of Primary mTAL Cells Primary cultures of mTAL cells were prepared from kidneys of C57/BL6J mice and validated as previously described. 38 Briefly, TALs were isolated under a light microscope on the basis of morphologic characteristics and cultured on permeable filter supports (Transwell-COL, pore size 0.4 mm, Corning Costar, USA) containing a DMEM:F12-based medium for 7-10 days in a humidified chamber at 37 C and 5% CO 2 until confluent monolayers were formed. For Krt40 silencing, an adenovirus expressing a short hairpin RNA against mouse Krt40 (Ad-shKrt40; targeting sequence: GGATGAGATGCGATGT-CAATA, Vector Biolabs) and a scramble (Ad-GFP-U6-scrmbl-shRNA) control were used. The transduction protocol was performed as previously described. 39 Cells were plated on filters and transduction was performed when they reached approximately 70%-80% of confluence (24 hours after plating). Cells were subsequently incubated overnight at 37 C with culture medium containing the virus at the appropriate concentration (0.2125 3 10 9 plaque forming units (PFU)/ml). Culture medium was changed every day and the cells were collected for analysis after 5 days of serum-free conditions, to ensure the best compromise between UMOD expression and the knockdown in KRT40. 38 Histologic Analysis and Immunostaining Kidneys were fixed overnight at 4 C in 4% formaldehyde (Sigma-Aldrich), dehydrated, and subsequently embedded in paraffin. Paraffin blocks were cut into 5 mm-thick sections, deparaffinized in xylene, and rehydrated in decreasing ethanol concentrations. The sections were incubated with sheep anti-UMOD (1:300; Meridian Life Science Inc., Cincinnati, OH, USA), rabbit anti-KRT40 (1:50-1:100, Life-Span Bioscience), rabbit anti-WDR72 (1:500; Sigma-Aldrich), goat anti-AQP2 (1:400, SantaCruz Biotechnology), and rabbit anti-KRT39 (1:50, ThermoFisher Scientific) overnight at 4 C. After washing steps, sections were incubated with the appropriate Alexa Fluor-labeled secondary antibody (1:400, Life Technologies, Carlsbad, CA) for 1 hour at room temperature. The use of human kidney biopsies has been approved by the UCLouvain Ethical Review Board. Monolayers of mTAL cells on polytetrafluoroethylene filters were fixed for 10 minutes in 4% formaldehyde (Sigma-Aldrich), permeabilized for 30 minutes using 0.5% saponin (Sigma-Aldrich), and blocked in 3% BSA (Sigma-Aldrich). Immunostaining procedures were similar to those used for kidney sections. After the last washing step, filters were cut and mounted on a glass slide using Prolong Gold Anti-fade reagent containing DAPI (Invitrogen Corp., Waltham, MA), and viewed under a confocal microscope (Leica Microsystems GmbH, Wetzlar, Germany) using a 363 1.4 NA oil immersion objective.
Protein Samples Preparation and Immunoblotting mTAL cells were lysed in ice-cold RIPA buffer (Sigma-Aldrich, St. Louis, MO, USA) containing protease (Roche) and phosphatase (PhosSTOP, Sigma) inhibitors. Samples (20 mg/lane) were thawed on ice, diluted in Laemmli sample buffer (BioRad), separated on a 7.5% SDS-PAGE gel, and blotted on PVDF membranes. After blocking with 5% nonfat milk (BioRad) in PBS, the membranes were incubated overnight at 4 C with primary antibody. Blots were subsequently washed and incubated with peroxidaseconjugated secondary antibodies, washed again, and visualized by Immun-Star enhanced chemiluminescence (Bio-Rad). Quantitative analysis was performed by scanning the blots and measuring the relative density of each band normalized to b-actin by using ImageJ software.

RNA Isolation and Quantitative RT-PCR
Total RNA was extracted from mTAL cells with RNA-queousR kit (Invitrogen, Carlsbad, CA). One mg of RNA was used to perform the reverse transcription reaction with iScript TM cDNA Synthesis Kit (BioRad). Changes in target genes mRNA levels were determined by relative quantitative RT-PCR with a CFX96TM Real-Time PCR Detection System (BioRad) using iQTM SYBR Green Supermix (BioRad). The analyses were performed in duplicate with 100 nM of both sense and antisense primers in a final volume of 20 mL using iQTM SYBR Green Supermix (BioRad). Specific primers were designed using Primer3 (Supplemental Table 5). PCR conditions were 95 C for 3 minutes followed by 40 cycles of 15 seconds at 95 C, and 30 seconds at 60 C. The PCR products were sequenced with the BigDye terminator kit (Perkin Elmer Applied Biosystems) using ABI3100 capillary sequencer (Perkin Elmer Applied Biosystems). The efficiency of each set of primers was determined by dilution curves (Supplemental Table 5). The relative changes in targeted genes over Gapdh mRNAs were calculated using the 2 -DDCt formula.

Statistical Analysis (Experimental Studies)
The quantitative data were expressed as mean6SEM. Normality of the sample distribution was assessed using Shapiro-Wilk test, whereas the F test was used to check equality of the variance in the experimental groups. The difference between experimental groups were evaluated using unpaired, two-tailed t test or Mann-Whitney test in the case of non-normal distribution. The sample size (n) of each experimental group is described in the corresponding figure legends. All results are representative of more than three independent experiments. GraphPad Prism software was used for all statistical analysis. Statistical significance was set as P,0.05.
Information on the datasets and summary statistics are available in the Edinburgh Datashare repository, under the link https://doi.org/10.7488/ds/3012 created on April 7, 2021.

Summary Data and Heritability
The summary data for the 29,315 participants from the 13 cohorts with uUMOD levels are shown in Table 1. The l values in individual cohorts ranged from 0.99 to 1.04 for uUMOD and 0.99-1.05 for uUCR. Detailed information on cohorts, samples, assays, genotyping, and imputation platforms are given in Supplemental Appendix 1 and Supplemental Tables 1 and 2. The estimated heritabilities for uUMOD in the familybased cohorts ranged from 11% (CROATIA-Vis) to 45% (GS: SFHS) (Supplemental Table 6). All genome-wide significant loci associated with uUMOD and uUCR in the overall population are listed in Table 2.

Meta-GWAS for Raw Urinary UMOD Levels
Meta-analysis of uUMOD levels in 29,315 individuals resulted in the identification of two genome-wide significant signals. A novel, genome-wide significant locus (P value 1.24E-08) was identified on chromosome 17 spanning the KRT40 gene. A major signal (P value 2.17E-88) was present on chromosome 16, corresponding to the previously described UMOD-PDILT locus ( Figure 1A). The quantile-quantile plot of the -log10 observed versus expected P values for raw UMOD had a ø value of 1.00, indicating there is no significant genomic inflation ( Figure 1A, inset). The full Manhattan plot without y-axis cutoff is shown in Supplemental Figure 1A.
Association of uUMOD levels and genotypes at rs8067385, the SNP with the lowest P value (1.24E-08) within the KRT40 locus, is shown in Tables 3 and 4. The effect size of the SNP on uUMOD levels was consistent in direction and similar in magnitude across most of the cohorts, as evidenced in a Forest plot ( Figure 2A). The minor allele, C, of rs8067385 was associated with lower uUMOD levels in most of the cohorts and had an average standardized effect size of -0.05 and a standard error of 0.009 (Tables 3 and 4; Figure 2A), which explained 0.1% of the variance observed. The regional association plot ( Figure 2B) showed all of the genome-wide significant SNPs in high LD with rs8067385 over the KRT40 region. SNPs in each locus with a P value ,1.0E-05 are listed in Supplemental Table 7. All KRT40 SNPs with P value ,1.0E-05 that were in high LD (r 2 $ 0.8) with rs8067385 are listed in Supplemental Table  8 and annotated using SNiPa. The exonic variants rs9908304 and rs721958 are predicted to be damaging/deleterious by SIFT and PolyPhen2 (Supplemental Table 8). KRT40 encodes keratin-40 (KRT40), a type I keratin expressed in the kidney. The 17q21.2 region also includes type I keratin genes KRT39 and KRT23.
Association of uUMOD levels and genotypes at rs12934455, the SNP with the lowest P value (2.17E-88) within the UMOD-PDILT locus, is shown in Tables 5 and 6. The average standardized effect size was 20.23 with a standard error of 0.01, which explains 1.4% of the variance observed. The Forest plot, showing effect sizes of the minor allele of rs12934455 consistent in direction for all cohorts, and the regional association plot of the UMOD-PDILT locus are shown in Supplemental Figure 2. The rs12934455 SNP had a heterogeneity I Sq value of 68.7, likely due to the lower imputation quality for that variant in CoLaus and FHS cohorts. Indeed, heterogeneity for that SNP was not significant (Het I Sq545.2, Het P value50.06) when meta-analysis of uUMOD was carried out excluding CoLaus and FHS. The UMOD-PDILT locus has been consistently associated with kidney function and levels of urinary UMOD in previous GWAS. 14,19,40 Meta-GWAS for Urinary UMOD Levels Indexed to Creatinine The meta-GWAS for uUCR yielded two genome-wide significant signals: the known UMOD-PDILT locus on chromosome www.jasn.org META-ANALYSIS Data are presented as mean6SD or median (5 th percentile, 95 th percentile) for continuous variables; and n (%) for categorical variables. Median and percentile range were used for uUMOD and uUCR as the original distribution was not normally distributed, which is why we rank transformed the phenotype before running GWAS. BMI, body mass index; eGFRcreat, eGFR on the basis of plasma creatinine, using the Chronic Kidney Disease Epidemiology      16, and a novel locus on chromosome 15 over the WDR72 gene ( Figure 1B). The minor G allele of rs9672398 in WDR72 is associated with lower levels of uUCR in most of the cohorts (Supplemental Figure 3), with an average standardized effect size of 20.05 and a standard error of 0.0089 (Supplemental Table 9), which explains 0.1% of the variance observed. The regional association plot for the chromosome 15 signal revealed that only one SNP, rs9672398, reached genome-wide significance (P value 1.65E-08) but was in LD with several other SNPs over WDR72 (Supplemental Figure 3). The association of uUMOD did not reach genome-wide significance (P value 3. 336E-05). WDR72 is highly expressed in the kidney. By GWAS, common variants in WDR72 have been associated with kidney function and CKD, 14,41 urine pH, 42 and kidney stones. 42,43 The rs13335818 SNP with the lowest P value (3.86E-118) was found within the UMOD-PDILT locus, and association to uUCR is shown in Supplemental Table  10 and Supplemental Figure 4. The regional association plot showed that rs13335818 is in high LD (r 2 50.94) with the top rs12934455, identified in meta-analysis of uUMOD levels. As for the raw UMOD level, the genome-wide signal included independent variants on UMOD (rs13335818) and PDILT (rs77924615), respectively (Supplemental Figure 4). SNPs from each locus with P,1E-05 for uUCR are shown in Supplemental Table 11.
The combined variance of rs12934455 (UMOD) and rs8067385 (KRT40) for uUMOD is 1.5% and for uUCR is 1.9%. Similar associations for uUMOD and uUCR were seen when meta-analysis was conducted using sample size and P values (Supplemental Figure 5). Spearman's rank correlation of effect sizes (P50.02) of the most significant KRT40 SNP, rs8067385, in association with uUMOD and uUCR indicated the effect sizes between the two traits were similar in each of the cohorts (Supplemental Figure 6). Analysis of the candidate genes associated with raw and indexed urinary UMOD levels revealed a number of genes expressed in the TAL, with encoded proteins playing important roles in cell homeostasis, mitochondrial function, transport, and inflammatory signaling (Supplemental Table 12).

Meta-GWAS for Urinary UMOD Levels Normalized for eGFR
As some cohorts included individuals with CKD, potentially influencing the levels of UMOD in urine, 16 we repeated the meta-analysis using UMOD normalized for eGFR, sex, and age in seven cohorts with available information. The UMOD-PDILT locus remained at genome-wide significance and the KRT40 locus at a suggestive threshold (Supplemental Table 13) whereas the WDR72 locus did not (Supplemental Table 14).

Associations of the New Loci with Creatinine and eGFR in UK Biobank
To evaluate the role of indexing urinary levels of UMOD to creatinine, we investigated potential associations of the new genome-wide significant loci for plasma and urinary creatinine and derived eGFR in the UK Biobank summary statistics database "Global Biobank Engine." 34 These analyses revealed that rs9672398 in WDR72 (index SNP for uUCR) was significantly associated with eGFR and plasma creatinine, but not with urinary creatinine levels, whereas the rs8067385 SNP in KRT40 (index for uUMOD) was not associated with any trait related to creatinine (Table 7).

Conditional Analysis
Implementing the GCTA COJO -Slct function for conditional analysis on summary statistics from meta-analysis resulted in identification of two independent loci in chromosome 16, one in UMOD (rs13335818) and one in the upstream gene, PDILT (rs11864909). A conditional GWAS using either of these two variants as covariates on the basis of individual-level genotype data from our largest cohort (GS:SFHS) confirmed that both signals remained genome-wide significant, indicating the loci are independent in the Scottish "healthy" population (Supplemental Figure 8).
Additional Genome-wide Analyses VEGAS2 is a gene-based association method that uses GWAS summary data and the associated P values and a simulation approach to calculate gene-based empirical association P values. The method tests for enrichment of multiple SNPs associated with the disease/trait that individually have a too modest effect on the phenotype to reach genome-wide significance using a per-SNP test. 32 Table 15.

Candidate Gene and Sensitivity Analyses
As UMOD is essentially produced in the TAL cells, we tested whether common variants within genes causing Mendelian disorders affecting the TAL may also influence the urinary excretion of UMOD. SLC12A1, KCNJ1, CLDN19, HNF1B, and MUC1 showed at least one SNP with a P value below the gene-specific threshold associated with uUMOD and/or uUCR (Supplemental Figure 9; Supplemental Table 16). We should note the association for genes expressed in the TAL was not stronger than that identified for genes expressed in other nephron segments, when queried against the meta-GWAS results (Supplemental Table 17). The suggestive loci (SORL1, CAB39, FAM83A, and MARCH1) identified in the first meta-GWAS for UMOD, 19 also did not reach the suggestive threshold (Supplemental Table 17). The effect sizes of the most significant SNP within UMOD and PDILT are similar, as shown in Supplemental  Table 18. The sensitivity analysis removing the FHS cohort, which measured UMOD using a distinct immunoassay, showed the UMOD/PDILT locus remained genome-wide significant and both the KRT40 and WDR72 loci remained within the suggestive threshold limit, despite the reduced sample size (Supplemental Table 19).

Segmental Distribution of KRT40 and WDR72 in relation to UMOD
To substantiate the biologic relevance of the newly identified KRT40 and WDR72 loci, we first evaluated the segmental distribution of KRT40, WDR72, and UMOD in mouse kidney (Figure 3). KRT40 is widely distributed in the kidney, more abundant in distal tubular segments and overlapping with UMOD in the TAL-both at the mRNA and www.jasn.org META-ANALYSIS protein levels (Figure 3, A-C). In isolated TAL segments, the expression of KRT40 was at least two orders of magnitude lower than that of UMOD and NKCC2 (Slc12a1) (Figure 3B). In situ hybridization evidenced a weak, selective expression of Krt40 in Umod-positive segments of the mouse kidney, with no signal for Krt39 (Supplemental Figure 10). Immunostaining confirmed a signal for KRT40 in UMOD-positive tubules, particularly at the apical pole of cells lining the TAL, whereas no colocalization between UMOD and WDR72 was observed ( Figure 3C). Both KRT40 and WDR72 were detected in AQP2-positive segments of the mouse kidney (Supplemental Figure 11). In the human kidney, KRT40 was detected in both UMODpositive and negative tubules, whereas KRT39 did not show any signal ( Figure 3D).

Modulation of KRT40 Expression Influences UMOD Excretion by mTAL Cells
The codistribution of KRT40 and UMOD led us to test whether the level of KRT40 expression may modulate the processing and excretion of UMOD by TAL cells. This hypothesis was supported by the existence of at least two exonic variants in high LD with the index KRT40 variant rs8067385, predicted to be damaging/deleterious by SIFT and PolyPhen2 (Supplemental Table 8). Furthermore, in the GTEx portal, the minor, C allele of rs8067385 is associated with a significant, dose-dependent decrease in the expression of KRT40 in a variety of epithelial tissues including the testis, pancreas, esophagus, and colon (no eQTL data for kidney medulla tissue available) (Supplemental Figure 12). Characterization of mTAL cells verified that KRT40 and UMOD were both endogenously expressed (Figure 4, A  and B). Transduction of mTAL cells with an adenovirus expressing a short hairpin RNA against mouse Krt40 (Ad-shKrt40) induced a specific silencing of KRT40, compared with cells treated with a scramble adenovirus (Ad-Scrmbl) ( Figure 4C). In these conditions, the downregulation of KRT40 was reflected by a significant accumulation of UMOD and a sharp decrease in the amount of excreted UMOD in the apical medium of mTAL cells ( Figure 4C). Confocal microscopy indicated the silencing of KRT40 in mTAL cells resulted in perinuclear accumulation of www.jasn.org META-ANALYSIS UMOD, contrasting with the control signal in cells transduced with Ad-Scrmbl ( Figure 4D). The trafficking defect induced by KRT40 downregulation was confirmed by Z-stack image analysis, with a perinuclear staining for UMOD contrasting with the diffuse signal observed in control conditions (Supplemental Figure 13A). The silencing of KRT40 had also an effect on the trafficking of ROMK (Supplemental Figure 13B), but did not modify the expression of TAL genes including Slc12a1, Kcnj1, Hnf1b, and Muc1 in mTAL cells (data not shown). Taken together, these data suggest the expression of the cytokeratin KRT40 regulates the processing and excretion of UMOD in TAL cells ( Figure 4E).

DISCUSSION
To gain novel insights into the mechanisms regulating UMOD excretion, we performed a meta-GWAS on urinary UMOD levels in 29,315 individuals of European ancestry, three times more than in our previous analysis. 19 We identified two novel, genome-wide significant loci, KRT40 and WDR72, in addition to the previously known UMOD-PDILT locus to be associated with uUCR and uUMOD. Mechanistic studies in primary mTAL cells demonstrated that modulating the expression of KRT40 affects the processing and apical excretion of UMOD. These studies provide insights into the biology of UMOD and keratins, and into the links between the UMOD-PDILT locus and kidney function.
The UMOD-PDILT locus has been consistently among the strongest associated loci with eGFR and CKD. 14,40 The relevance of the UMOD variants, which are associated with the levels of UMOD in the kidney and urine, is immediate because the gene is kidney specific and involved in a spectrum of kidney diseases. 1,7,40 In our meta-analysis, the variant showing the strongest association with uUCR is rs13335818 (P value 3.86E-118), a synonymous variant within UMOD, in high LD (r 2 5 0.98) with the top SNP in our previous study, rs12917707, and with UMOD promoter variants associated with eGFR and CKD and with expression of UMOD. 7,40 In a previous study of genetic associations with urinary UMOD levels, 14 two independently associated variants in the UMOD-PDILT locus were identified in conditional analyses: rs77924615, mapping into an intron of the upstream gene PDILT, and rs34262842, mapping to an intron of UMOD. Similarly, our conditional analysis in a large cohort with individual-level genotype data identified two independent loci in that region, one in UMOD (SNP rs13335818 in high LD with rs34262842, r 2 50.94) and one in PDILT (rs11864909, in almost complete LD with rs77924615, r 2 50.98).
The lead SNP in PDILT from our meta-analysis, rs77924615, had the strongest association with CKD and eGFR in the GWAS performed by the CKDGen Consortium. 14 Of interest, the intronic PDILT rs77924615 maps to open chromatin regions identified from various kidney cell types. Because PDILT is not expressed in the human kidney and rs77924615 was significantly associated with both differential expression of UMOD in kidney tissue and urine UMOD levels (obtained in the GCKD cohort), it was considered a regulatory SNP. 14 Collectively, these results substantiate the independent association between UMOD and PDILT variants and the levels of UMOD in urine. A recent Mendelian randomization study clarified the causality between uUMOD levels and kidney function in individuals of European descent: genetically driven levels of UMOD have a direct, causal, and adverse effect on kidney function outcome in the general population. 44 The KRT40 locus on chromosome 17 is a novel, genome-wide significant locus associated with UMOD levels in the urine. The association at the KRT40 locus is on the basis of multiple genome-wide significantly associated SNPs in high LD with the index SNP, rs8067385. Individuals homozygous for the minor, C allele of rs8067385 had lower levels of uUMOD compared with individuals homozygous for the G allele. Of note, the effect sizes of rs8067385 for uUMOD and uUCR are only marginally different, as indicated by Spearman's rank correlation analysis. The KRT40 signal remained above the suggestive threshold (P,5.17E-07) in the meta-analysis on urinary UMOD corrected for eGFR, and in the VEGAS analysis. In contrast, the rs8067385 SNP in KRT40 was not significantly associated with eGFR and plasma or urinary creatinine levels in the UK Biobank. Together, these results support the value of the KRT40 signal in relation to urinary UMOD levels.
KRT40 encodes KRT40, a type I keratin that belongs to the family of intermediate filament-forming keratins that form the cytoskeleton in epithelial cells. Types I and II keratins form obligate heterodimers and are regulated in a pairwise fashion in epithelia, depending on the tissue, the differentiation state, and the biologic context. 45 KRT40 belongs to a cluster of type I KRT genes located on chromosome 17q21.2, close to KRT39. As cytoskeletal proteins, keratins are involved in maintaining the physical integrity, mechanical stability, and shape of epithelial cells. They are also important for intracellular organization and transport within cells, for example, trafficking of proteins to the plasma membrane. 46 Keratins are considered as cytoprotective, undergoing dynamic upregulation in disease states, and potentially affecting migration, growth, proliferation, and protein synthesis. 47 Several inherited keratinopathies (e.g., skin disorders) have been reported, but none involving KRT40.
Little is known about the role of keratins in the kidney. Recent studies evidenced robust changes in the expression and subcellular localization of KRT7-8 and KRT18-19 in response to kidney stress, with KRT18 in urine being a potential biomarker for tubular cell injury. 48 Our studies in mouse and human kidney reveal that KRT40 is weakly expressed in the TAL, where it colocalizes with UMOD. The prediction of missense variants in KRT40 in LD with rs8067385 and the association in GTEx of the minor allele of rs8067385 with a decreased expression of KRT40 in epithelial tissues suggest a possible loss of function of the KRT40 variant. We tested this hypothesis in the mTAL cells, which endogenously express both KRT40 and UMOD. Specific silencing of KRT40 in mTAL cells was reflected by a sharp decrease in the apical excretion of UMOD, causing the intracellular accumulation of the protein. The silencing was also reflected by altered apical targeting of ROMK in these cells. That altered expression of KRT40 affects UMOD and ROMK processing in TAL cells may suggest a role of KRT40 on the polarized sorting of proteins to the apical membrane, affecting the release of UMOD in urine ( Figure 4E).
We detected a genome-wide significant association between variants in WDR72 and the urinary UMOD level indexed to creatinine (uUCR). WDR72 encodes a protein with eight WD40 (or b-transducin) repeats, which fold to form two circular, b-propeller structures, and an a-solenoid tail at the C-terminus. This combination of domains is conserved among membrane-coating proteins, which serve as a docking site for protein-protein interactions and stabilize membrane curvature. 49 WDR72 is highly expressed in the kidney, although we found it clustered in the collecting ducts. Recessive mutations in WDR72 have been associated with amelogenesis imperfecta, 50 and distal renal tubular acidosis. 51,52 By GWAS, variants in WDR72 have been associated with kidney function and CKD, 14,41,53 urine pH, 42 and risk of kidney stones. 42,43 In CKDGen, the index SNP at WDR72 was associated with blood urea nitrogen. 14 Variants in WDR72 are strongly associated with eGFR on the basis of serum creatinine or cystatin C and with BUN. 41 The fact that rs9672398 is significantly associated with eGFR and plasma creatinine levels in UK Biobank, that the WDR72 locus does not reach any suggestive threshold in the meta-analysis using UMOD normalized for eGFR, and that WDR72 does not colocalize with UMOD suggest the WDR72 signal, only detected for uUCR, is most likely related to its effect on eGFR.
Although limited by power, our candidate gene analysis revealed that a few common variants in genes causing rare Mendelian disorders targeting the TAL are weakly associated with UMOD levels. These results support the functional interactions operating in TAL cells, including the transcription factor HNF1-b, known to be an essential transcriptional regulator of UMOD, 1,9 and ROMK, which directly regulates processing and release of UMOD by TAL cells. 17 Analysis of the candidate genes from the loci associated with uUMOD and uUCR with a suggestive P value (,1.0E-05) revealed a number of genes expressed in the TAL/DCT, with encoded proteins playing roles in cell homeostasis, mitochondrial function and transport. Future studies will address the relevance and biologic mechanisms that underlie these genetic associations.
Our study combines the advantages of the largest to date meta-GWAS on urinary UMOD, measured with a robust assay in various types of cohorts, and complemented with detailed expression studies in mouse and human kidneys, and functional investigations in TAL cells. Limitations of this study include the availability of data only for individuals of European descent and the lack of replication due to limited availability of additional cohorts with available uUMOD measurements. We noted some variability of UMOD levels that were measured in different cohorts, even when using the same assay and apparently unrelated to sample processing and/or storage conditions. 19,24 Variations in the physiology excretion of UMOD have been reported, potentially linked to dietary habits, tubular transport activities, or level of residual kidney function. 1,16 Common, independent variants in KRT40, UMOD, and PDILT influence the levels of UMOD in urine. The expression of the type I keratin KRT40 affects UMOD processing in TAL cells. These results advance our understanding of the biology of UMOD, the role of keratins in the kidney and substantiate the association of UMOD-PDILT variants with kidney function.
Supplemental Table 1. Study sample characteristics for UMOD measurement.
Supplemental Table 2. Genotyping and imputation platforms. Supplemental Table 3. List of candidate genes associated with rare Mendelian disorders affecting the TAL.
Supplemental Table 4. List of genes associated with diseases in specific kidney segments and candidate genes previously published by Olden et al. (2014).
Supplemental Table 5. Primers for quantitative RT-PCR analyses. Supplemental Table 6. Heritability estimates for family-based cohorts. Supplemental Table 7. Most significant SNP from each locus with P value ,1E-05 from urinary UMOD (uUMOD) meta-analysis.
Supplemental Table 8. List of KRT40 variants, in high LD with rs8067385, and with P value ,1E-05 in association with uUMOD.
Supplemental Table 13. Effect size and P values of rs8067385 from meta-analysis of uUMOD concentration and uUMOD_eGFR using seven cohorts.
Supplemental Table 14. Effect size and P values of rs9672398 from meta-analysis of uUMOD concentration and uUMOD_eGFR using seven cohorts.
Supplemental Table 16. Candidate gene analysis for uUMOD and uUCR levels.
Supplemental Table 17. Look-up analyses of genes associated with specific segments of the kidney and candidate genes previously published by Olden et al. (2014).
Supplemental Table 18. Effect sizes of the most significant SNP in UMOD and PDILT in association with uUMOD and uUCR.
Supplemental Table 19. Effect sizes of the most significant SNP in UMOD, KRT40 and WDR72 in association with uUMOD and uUCR in 12 cohorts (FHS excluded).
Supplemental Appendix 1. Summary characteristics of the study cohorts.
Supplemental Figure 1. Genetic loci associated with uUMOD and uUCR.
Supplemental Figure 2. Effect size of rs12934455 and regional association plot of UMOD-PDILT locus from raw UMOD levels.
Supplemental Figure 3. Effect size of rs9672398 and regional association plot of WDR72 locus for uUCR meta-analysis.
Supplemental Figure 4. Effect size of rs13335818 and regional association plot of UMOD-PDILT locus from uUCR meta-analysis.
Supplemental Figure 5. Manhattan plot of meta-GWAS of uUMOD and uUCR using sample size and P values for analysis of the 13 study cohorts.
Supplemental Figure 8. Manhattan plot showing GWAS results in association with uUMOD and uUMOD conditioned for rs12934455 or for rs11864909 using GS:SFHS. Supplemental Figure 9. Candidate genes influencing the urinary excretion of UMOD.
Supplemental Methods. Supplemental References.