Genome-wide association study of estimated glomerular filtration rate using repeated measurements in the Taiwan Biobank
Article information
Abstract
Background
Chronic kidney disease (CKD) is a major global public health issue, with genetic factors playing a significant role in kidney function. Although genome-wide association studies (GWAS) have identified numerous loci associated with estimated glomerular filtration rate (eGFR), most studies relied on a single time-point measurement, which limits the capacity to account for within-individual measurement variability.
Methods
We performed a repeated-measurement GWAS in the prospective Taiwan Biobank (Taiwanese ancestry; n = 25,004) using two repeated creatinine-based eGFR measurements. Repeated eGFR values were analyzed using a linear mixed-effects model with a subject-specific random intercept and time-varying covariates, providing a more precise estimate of eGFR level. Identified loci underwent functional annotation (expression quantitative trait locus, deleteriousness prediction, and epigenetic markers) and were compared with results from a single-measurement GWAS.
Results
Six loci associated with eGFR were identified, including four previously reported regions (1q22, 4q21.1, 11p14.1, and 17q21.2) and two additional loci (6p21.32 and 15q24.2). Functional annotation implicated several candidate genes—such as MUC1/EFNA1, SHROOM3, HLA-DQB1, MPPED2, NRG4, and PGAP3/FBXL20—in the regulation of kidney function.
Conclusion
Incorporating repeated eGFR measurements into GWAS may improve phenotypic precision for identifying genetic associations with kidney function. This study identified eGFR-associated loci and biologically plausible candidate genes in a Taiwanese population, which require further replication and functional validation.
Introduction
Chronic kidney disease (CKD) is a major global public health concern, affecting over 10% of adults worldwide and contributing to substantial morbidity and mortality [1]. Individuals with CKD face a substantially increased risk of adverse outcomes, including end-stage kidney disease (ESKD) and cardiovascular events, and impose a significant socioeconomic burden, underscoring the importance of early detection and intervention. Conventional risk factors for CKD include older age, male sex, smoking, diabetes mellitus (DM), and hypertension. However, these factors do not fully account for the development of the disease, underscoring the need to identify additional risk factors and biomarkers. Evidence suggests that eGFR has a substantial heritable component, while genome-wide association studies (GWAS) have identified multiple loci associated with estimated glomerular filtration rate (eGFR), a key indicator of kidney function [2]. A recent meta-analysis has shown that genetic loci account for nearly 20% of the heritable variation in eGFR [3]. This underscores the pivotal contribution of genetic variants to CKD risk, opening new avenues for improved disease prediction and management.
The KDIGO (Kidney Disease: Improving Global Outcomes) guidelines define CKD based on structural or functional kidney abnormalities that persist for at least 3 months, emphasizing the need for repeated measurements of renal function over time. Despite this, most GWAS of kidney function rely on a single eGFR measurement, which may fail to reflect an individual’s true renal function due to intrinsic biological fluctuations and measurement variability. Studies such as Pattaro et al. [4] and Wuttke et al. [3] have utilized cross-sectional eGFR measurements to map genetic loci associated with kidney function. While these analyses have been highly informative, the use of single-time-point data constrains the investigation of changes in kidney function.
Using GWAS data from the Taiwan Biobank [5], we aimed to identify genetic loci associated with eGFR through a repeated-measurement GWAS framework that incorporates repeated measurements via linear mixed-effects models. This approach leverages repeated eGFR measurements to improve phenotype precision and estimate genetic associations with eGFR levels. This repeated-measurement approach enhanced statistical power and yielded more precise estimates of genetic effects compared to traditional single-measurement GWAS using linear regression. Our results highlight additional genetic determinants of kidney function, particularly in Asian populations, and provide meaningful insights into the mechanisms that influence renal function.
Methods
Study participants
The Taiwan Biobank is a government-supported, community-based prospective cohort that collects extensive phenotypic and genomic data from individuals of Taiwanese ancestry. Details of the study population and recruitment procedures have been described previously [6]. Briefly, recruitment began in 2012, and by the end of 2021, the cohort included more than 140,070 participants, of whom 114,600 had completed genotyping. The Taiwan Biobank primarily enrolls generally healthy adults aged 30–70 years through non-hospital-based recruitment. At enrollment, participants provided written informed consent, completed standardized questionnaires, underwent physical examinations, and contributed blood and urine samples. In the dataset used for this study, diagnostic codes and treatment records for acute kidney injury, dialysis, and kidney transplantation were not available; therefore, these conditions could not be directly verified or excluded. In the analytic cohort, only 15 of 25,004 participants (0.06%) had eGFR <15 mL/min/1.73 m2 at baseline, indicating that CKD stage 5 was very uncommon.
Since 2016, follow-up assessments and biospecimen collections have been conducted approximately every 2 to 4 years. For this study, we included participants who had completed at least one follow-up visit, providing two eGFR measurements for repeated-measurement analysis. We defined follow-up time as the interval (in years) between baseline and follow-up visits; because the primary model did not include an explicit time term, unequal visit intervals primarily contribute to within-person averaging rather than estimation of eGFR slope. Individuals with missing data on eGFR, glycated hemoglobin (HbA1c), or systolic blood pressure (SBP) were excluded. Hypertension and DM were defined by self-reported diagnoses or SBP ≥140 mmHg and HbA1c ≥6.5%, respectively.
Genotyping, imputation, and quality control
Genomic DNA was extracted from baseline blood samples, and genotyping was performed using the TWB array (Affymetrix Inc.). Single-nucleotide polymorphisms (SNPs) that passed quality-control filters were phased using SHAPEIT2 [7] and imputed with IMPUTE2 version 2.3.1 [8], based on a population-specific reference panel comprising 1,445 whole-genome sequences from the Taiwan Biobank [9]. Quality control for GWAS analyses involved filtering out variants with a missing rate of more than 2%, minor allele frequency <1%, or deviations from Hardy-Weinberg equilibrium (p < 10–6). Non-autosomal variants were excluded. Sample-level quality control removed individuals with a missing rate of more than 2%, sex discordance, relatedness up to the third degree, or heterozygosity outliers exceeding 5 standard deviations.
Definitions of phenotype
The phenotype analyzed in this study was creatinine-based eGFR, calculated using the CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration) equation [10]. We did not apply winsorization to eGFR values because the use of repeated measurements reduces measurement error, improves phenotype precision, yields better effect-size stability, and increases sensitivity for loci with small effects.
Statistical analysis
GWAS analysis was performed using a repeated-measurement GWAS framework based on linear mixed-effects models under an additive genetic model. This approach estimates the mean effect of genotype on the biomarker while accounting for random effects and adjusting for time-varying covariates [5]. In the primary analysis, the genotype effect represents an association with the repeated-measurement eGFR level, rather than a genotype effect on rate of eGFR change. Given that most participants contributed two measurements, we focused on improving phenotype precision and reducing within-individual measurement variability rather than modeling individual-specific slopes. Using this methodology, we analyzed repeated measurements by incorporating random intercepts and adjusting for time-varying covariates. The linear mixed-effects model was chosen to account for within-individual correlation arising from repeated eGFR measurements and to improve phenotype precision by reducing random measurement error. In this framework, a subject-specific random intercept was included to capture stable inter-individual differences in baseline kidney function, while repeated measurements contributed additional information to refine the estimation of each individual’s underlying eGFR level. Accordingly, the estimated genetic effects primarily reflect differences in average eGFR levels estimated from repeated measurements rather than differences in the rate or pattern of eGFR change over time [5]. In this study, time-varying covariates included age, HbA1c, and SBP at each measurement, whereas sex and the top five genetic principal components were treated as time-invariant covariates. Each analysis incorporated fixed effects for age, sex, HbA1c, SBP, and the top five genetic principal components, with a random intercept included in the model. As a sensitivity analysis, we additionally fitted a minimally adjusted model including age, sex, and genetic principal components only. Principal component analysis (PCA) was performed using genome-wide SNP data to assess population structure. The first five principal components were included as covariates in all association analyses to account for residual population stratification. To optimize computational performance, we first applied the score test to all variants and then conducted Wald tests for variants showing the strongest evidence of association [5]. Variants with a score test p < 5 × 10–8 were further evaluated for effect-size estimation. Each genetic locus was represented by its lead variant, defined through a two-step clumping procedure: primary clumping of significant variants at r2 ≥0.6, followed by secondary clumping at r2 ≥0.1 [11]. Loci located within 500 kb of each other were merged. In addition, associations of variants within the identified loci were cross-referenced using the GWAS Catalog.
Conditional analysis was performed using GCTA-COJO (Genome-wide Complex Trait Analysis-Conditional and Joint analysis) to identify secondary association signals within 600 kb of the lead variants [12]. The collinearity threshold was set to 0.90, and genome-wide significance was defined as p < 5 × 10–8. Heritability is defined as the proportion of phenotypic variance attributable to genetic factors. The SNP-based heritability of eGFR was estimated using the genome-based restricted maximum likelihood approach implemented in the GCTA software [13].
To validate our findings, we generated simulated data, given the absence of an external dataset with repeated eGFR measurements in East Asian populations. Random genetic variants were simulated using LDAK, a tool widely used in genetic association studies. This simulation was designed to provide an empirical calibration of the observed association signals by comparing them with allele-frequency-matched null variants, rather than to serve as replication in an independent cohort. We employed the -make-snps function, which produces random variants under the assumptions of Hardy-Weinberg equilibrium and linkage equilibrium. For each lead variant, we simulated 1,000 random variants with matching allele frequencies for a sample of 25,004 individuals. The simulated genotypes were then combined with real phenotypic data for analysis, and the p-values for both the random variants and their corresponding lead variants were visualized using histograms for comparative evaluation.
Additionally, a conventional GWAS of eGFR was conducted to compare the robustness and sensitivity of the repeated-measurement GWAS findings. Single-measurement eGFR was analyzed using a linear regression model under an additive genetic framework, implemented in PLINK. Analyses were adjusted for age, sex, HbA1c, SBP, and the top five genetic principal components. Significance thresholds and locus definitions followed those described above.
Functional annotation
Genetic variants were annotated using FUMA (Functional Mapping and Annotation), an open-access platform for functional mapping and annotation [11]. This framework incorporates ancestry-specific linkage disequilibrium structure, ANNOVAR-derived functional annotations, and multiple deleteriousness prediction metrics to generate comprehensive variant-level annotations. Deleteriousness was evaluated using the Combined Annotation Dependent Depletion (CADD) score, which integrates diverse functional annotations by comparing naturally occurring variants with simulated mutations [14]. A CADD score greater than 12.37 was used as a threshold indicative of deleterious variation. Functional insights into the identified loci were explored using HaploReg version 4.2 [15] and NephQTL2 [16]. HaploReg version 4.2 annotates regulatory features including chromatin states, evolutionary conservation, and transcription factor motif disruptions, while NephQTL2 identifies associations between gene expression levels and the identified loci.
Ethics statement
This study was approved by the Institutional Ethics Review Board of National Taiwan University Hospital (202301212RINB). The requirement for written informed consent was waived because only de-identified data were analyzed. The study adhered to the Strengthening the Reporting of Genetic Association Studies (STREGA), an extension of the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines [17].
Results
Participant characteristics
A total of 25,004 Taiwan Biobank participants with both baseline and first follow-up eGFR measurements were included in the GWAS. The overall study workflow is depicted in Fig. 1. Among the participants, 38% were male. At baseline, the mean age was 51 ± 10 years, and the mean eGFR was 101 ± 14 mL/min/1.73 m2. The mean SBP was 117 ± 18 mmHg, and 19.3% of participants had hypertension. The mean HbA1c level was 5.7% ± 0.8%, with 9.3% of participants having DM. The median follow-up time was 4 years (interquartile range [IQR], 3.1–5.0). The median annual eGFR change was –0.87 mL/min/1.73 m2 (IQR, –2.0 to 0.1 mL/min/1.73 m2) per year. Detailed participant characteristics are presented in Table 1. PCA showed no evidence of population substructure (Supplementary Fig. 1, available online).
Study flow diagram.
CADD, Combined Annotation Dependent Depletion; cis-eQTL, cis-expression quantitative trait locus; eGFR, estimated glomerular filtration rate; GCTA-COJO, Genome-wide Complex Trait Analysis–Conditional and Joint Analysis; GWAS, genome-wide association study; HbA1c, glycated hemoglobin; HWE, Hardy-Weinberg equilibrium; MAF, minor allele frequency; PC, principal component; QC, quality control; SBP, systolic blood pressure; SD, standard deviation; SNP, single-nucleotide polymorphism.
Genome-wide association study for estimated glomerular filtration rate
Using repeated eGFR measurements, we performed a GWAS to identify genetic loci associated with eGFR levels while accounting for within-individual repeated observations. A slight genomic inflation was observed in the quantile–quantile plot (λGC = 1.07) (Supplementary Fig. 2, available online), for which genomic control was applied. After adjustment, six loci reached genome-wide significance, located at 1q22, 4q21.1, 6p21.32, 11p14.1, 15q24.2, and 17q21.2. In comparison, the conventional GWAS based on single-time-point eGFR identified only four loci (Table 2, Fig. 2). In a minimally adjusted sensitivity analysis including age, sex, and the top five genetic principal components, all six lead loci remained genome-wide significant (Supplementary Table 1, available online). All p-values for the simulated random variants exceeded 10–4, clearly distinguishing them from the corresponding lead variants (Fig. 3), supporting the non-random nature of the associations and the robustness of the results. Conditional analyses within the identified loci revealed no evidence of secondary association signals. Using the GCTA framework, SNPs with a minor allele frequency >0.01 explained 13.7% of the SNP-based heritability for eGFR.
Associations of independent genetic variants with eGFR using repeated-measurement and conventional GWAS
Manhattan plots of repeated-measurement GWAS (A) and conventional GWAS (B) for eGFR.
The repeated-measurement GWAS (A) identified six eGFR-associated loci, while the conventional GWAS (B) found four, with three loci overlapping between the two methods. The Manhattan plot shows the –log10 p-value obtained from the score test for the genetic effect on eGFR by chromosome base position (GRCh37). The red dashed line marks genome-wide significance (5 × 10–8).
GWAS, genome-wide association study; eGFR, estimated glomerular filtration rate.
Distribution of p-values for lead variants and simulated random variants.
The x-axis represents the –log10 p-value derived from the Wald test of repeated-measurement GWAS. The histograms illustrate the distribution of p-values for simulated random variants, while arrows indicate the p-values for lead variants. The distribution of p-values for both lead variants and their corresponding simulated random variants confirmed a significant association between lead variants and eGFR.
GWAS, genome-wide association study; eGFR, estimated glomerular filtration rate.
Four of the loci identified, 1q22, 4q21.1, 11p14.1, and 17q21.2, have been previously catalogued as eGFR-associated regions in the GWAS Catalog. These results are concordant with findings from both the Biobank Japan study and a recent transethnic GWAS meta-analysis [3]. The two additional loci detected in our analysis, 6p21.32 and 15q24.2, have been previously associated with related traits, such as serum uric acid levels and hemoglobin concentrations (Supplementary Table 2, available online).
Functional annotation of loci associated with estimated glomerular filtration rate
NephQTL eQTL analysis (Table 3) revealed several kidney-relevant genes linked to the identified loci. At 1q22, rs7366775 showed significant associations with MUC1, YY1AP1, and SCAMP3 expression in the glomerulus, and with MUC1, SSR2, SLC50A1, THBS3, MSTO1, ADAM15, and FDPS in the tubulointerstitium. At 4q21.1, rs28394165 was associated with expression of SHROOM3, CDKL2, STBD1, and CCDC158 in the tubulointerstitium. At 6p21.32, rs3830060 showed strong associations with HLA-class II genes (HLA-DQA2, HLA-DQA1, HLA-DRB1, HLA-DQB1) across both compartments. At 15q24.2, rs79063526 correlated with PPCDC expression in the glomerulus and NRG4 in the tubulointerstitium. Finally, rs667239 at 17q21.2 was associated with PNMT expression in the glomerulus and PGAP3 and FBXL20 in the tubulointerstitium.
Table 4 lists putative functional variants in high linkage disequilibrium with the eGFR-associated loci. At 1q22, three variants (rs12904, rs4745, and rs9297) localized to enhancer regions defined by kidney histone marks. The missense variant rs4745 results in a p.D137V substitution in the EFNA1 protein, whereas rs12904 and rs9297, both located in the 3′-untranslated region (UTR) of EFNA1, are predicted to disrupt protein-binding and regulatory motifs. At 4q21.1, intronic variants of SHROOM3 (rs7654754, rs7656186) mapped to regions of accessible chromatin in kidney tissue, suggesting potential regulatory effects. At 6p21.32, intronic HLA-DQB1 variants (rs28746835, rs3830059, rs3830060) overlapped promoter-associated histone marks across kidney and multiple other tissues, consistent with altered regulatory motif activity. At 15q24.2, the NRG4 intronic variant rs77452453 was predicted to modify transcription factor binding motifs. At 17q21.2, five variants exhibited evidence of deleteriousness: intronic variants in FBXL20 (rs2302073, rs4795355) and CDK12 (rs7503705), as well as 3′-UTR variants in MED1 (rs6503513) and FBXL20 (rs3744348). These variants were enriched in histone- and chromatin-associated regulatory elements and are predicted to affect protein-binding or regulatory motif activity.
Discussion
Using repeated eGFR measurements in a large Taiwanese cohort, we identified six loci, 1q22, 4q21.1, 6p21.32, 11p14.1, 15q24.2, and 17q21.2, significantly associated with kidney function. Through comprehensive functional annotation integrating eQTL data, deleteriousness metrics, and epigenetic signatures, we highlighted several biologically plausible candidate genes, including MUC1/EFNA1, SHROOM3, HLA-DQB1, MPPED2, NRG4, and PGAP3/FBXL20. These findings add to current knowledge of the genetic architecture of kidney function and suggest candidate biological pathways that warrant further investigation. Previous GWAS investigating kidney function decline using serial eGFR measurements have often required large multi-cohort meta-analyses to achieve adequate statistical power, such as those conducted by the CKDGen consortium [18]. In contrast, our study demonstrates that incorporating repeated eGFR measurements within a single population-based cohort can improve phenotype precision and provide valuable insights into the genetic determinants of kidney function. In addition, we applied the conventional genome-wide significance threshold (p < 5 × 10–8), and the association statistics showed minimal inflation (λGC = 1.07), supporting adequate calibration of the GWAS results. Nevertheless, futher replication in diverse populations and experimental validation will be necessary to clarify the causal roles of these loci.
Previous large-scale GWAS have primarily used cross-sectional eGFR values, identifying multiple loci associated with kidney function in European and transethnic populations [3,19]. By contrast, our analysis incorporated repeated eGFR measurements, which more effectively capture within-individual variability and improve the precision of genetic effect estimation. Using this approach, we replicated four loci previously reported in European [3,19] and Japanese populations, and identified two additional loci at 6p21.32 and 15q24.2. In parallel, GWAS of kidney function decline using serial eGFR measurements in predominantly European populations have been conducted through multi-cohort meta-analyses, such as those from the CKDGen consortium [18]. More recently, a biobank-scale trajectory-based genetic study applied trajectory-based GWAS methods to model temporal changes in kidney function, including analyses of eGFR decline using repeated measurements [20]. These studies suggest that trajectory-informed phenotypes can identify genetic signals related to kidney function decline that are partly distinct from those identified using cross-sectional eGFR. Together, these findings place our repeated-measurement approach within the broader landscape of trajectory-based GWAS approaches and highlight its complementary role in elucidating genetic influences on kidney function.
The 6p21.32 locus is situated within the major histocompatibility complex region and annotated to genes such as HLA-DQA1, HLA-DQA2, HLA-DQB1, and HLA-DRB1. These genes are central to immune regulation and have been implicated in autoimmune kidney diseases, including immunoglobulin A nephropathy [21,22] and lupus nephritis, both of which can result in impaired kidney function. Previous research has also demonstrated associations between specific HLA alleles and an increased risk of ESKD, highlighting their importance in renal health.
The additional locus at 15q24.2 is annotated to NRG4, which encodes neuregulin 4, a member of the neuregulin family that activates the epidermal growth factor receptor. Prior studies have shown markedly reduced NRG4 levels in patients with diabetic nephropathy [23] and in relevant animal models [23,24]. Moreover, administration of exogenous Nrg4 in diabetic nephropathy rat models improved renal function, reduced tubulointerstitial fibrosis, and attenuated inflammation [24]. In our study, the allele associated with lower eGFR correlated with reduced NRG4 expression, consistent with emerging evidence indicating an inverse association between circulating NRG4 levels and CKD risk [25]. Further investigation into the mechanisms linking NRG4 to kidney function may reveal a promising therapeutic molecule in metabolic and renal diseases.
The candidate causal genes at the 1q22 locus may include MUC1 and EFNA1. MUC1 is a well-established gene implicated in autosomal dominant tubulointerstitial kidney disease (ADTKD-MUC1), which is characterized by adult-onset, slowly progressive kidney failure [26]. The most common pathogenic MUC1 mutations are frameshifts that generate a truncated protein unable to reach the apical membrane, leading to its accumulation in the cytoplasm and subsequent tubular injury [27]. In the present study, we observed that the eGFR-decreasing allele was associated with reduced MUC1 expression, suggesting that insufficient MUC1 levels may impair tubular function and contribute to long-term decline in kidney function. EFNA1 encodes ephrin-A1, a membrane-bound ligand that interacts with the receptor tyrosine kinase EphA2. This interaction activates downstream pathways that regulate key cellular processes, including cell polarity, migration, proliferation, and differentiation [28]. Experimental studies have shown that EphA2 and its ephrin ligands are upregulated in the kidney in response to ischemia-reperfusion injury, where they contribute to cytoskeletal repair and epithelial recovery [29]. In our study, we identified a missense variant in EFNA1, suggesting that altered ephrin-A1 signaling may impact renal epithelial integrity. Together, these findings support MUC1 and EFNA1 as biologically plausible contributors to the association between the 1q22 locus and kidney function; however, further functional studies will be necessary to clarify their causal roles.
The loci at 4q21.1 and 11p14.1, mapped to SHROOM3 and MPPED2, have been consistently identified in previous GWAS of both creatinine- and cystatin C-based eGFR [19]. Experimental studies have demonstrated that SHROOM3 is essential for maintaining the glomerular filtration barrier and podocyte structural integrity [30–32]. In our study, the eGFR-decreasing allele was associated with increased SHROOM3 expression, a finding consistent with prior evidence that SHROOM3 overexpression enhances transforming growth factor beta 1 signaling, promotes interstitial fibrosis, and contributes to kidney allograft dysfunction [33]. Collectively, these observations support a potential pathogenic mechanism linking regulatory variation at SHROOM3 to progressive kidney dysfunction. MPPED2, which encodes a metallophosphoesterase-domain–containing protein, has been implicated in neurodevelopment and tumor biology [34]. Although its renal function is less well characterized, knockdown experiments in zebrafish have demonstrated podocyte morphological abnormalities, suggesting a potential role for MPPED2 in maintaining podocyte health and, consequently, kidney function [35].
Based on the eQTL analysis, PGAP3 and FBXL20 emerge as potential candidate causal genes at 17q21.2. PGAP3 encodes a glycophosphatidylinositol (GPI)-specific phospholipase A2 that plays a key role in GPI-anchor remodeling, a process essential for the proper localization and function of GPI-anchored proteins within lipid rafts. Impairment of PGAP3 leads to Mabry syndrome, a congenital disorder of glycosylation characterized by hyperphosphatasia and intellectual disability [36]. In the present study, the eGFR-decreasing allele was associated with reduced PGAP3 expression, consistent with experimental evidence showing that Pgap3 knockout mice develop larger glomeruli with immunoglobulin deposition [37]. These findings collectively support a potential role of PGAP3 in maintaining glomerular integrity. Although published evidence directly linking FBXL20 to kidney function is limited, its expression includes renal tissue, and it participates in ubiquitination pathways that regulate protein turnover [38]. Given the central importance of ubiquitin-mediated degradation in podocyte and tubular homeostasis, FBXL20 remains a plausible candidate gene within this locus. Further studies will be needed to clarify the mechanistic contributions of PGAP3 and FBXL20 to kidney function and their potential roles in eGFR regulation.
Strengths of this study include the focus on a well-characterized Taiwanese population and the use of repeated eGFR measurements, which enabled a more accurate assessment of genetic influences on kidney function than conventional single-time-point analyses. However, several limitations should be noted. First, the study included only individuals of Taiwanese ancestry, which may limit the generalizability of our findings to other populations. Replication in diverse ancestral groups will be important to establish the broader relevance of these loci. Second, although the community-based design of the Taiwan Biobank makes inclusion of individuals receiving dialysis or with prior kidney transplantation unlikely, these conditions, as well as acute kidney injury, could not be directly verified or excluded because diagnostic codes and treatment records were unavailable. CKD stage 5 was very uncommon in the analytic cohort, with only 15 participants (0.06%) having eGFR <15 mL/min/1.73 m2 at baseline; however, dialysis status and transplantation history could not be confirmed. Third, the analyses were limited to common variants; rare variants with potentially large effects were not captured, underscoring the need for larger sample sizes and additional repeated eGFR measurements. Fourth, the lack of an independent external cohort with repeated eGFR measurements, particularly in East Asian populations, limits external validation of our findings. Although large international biobanks such as UK Biobank and the All of Us Research Program have expanded genomic resources, East Asian representation remains relatively limited, and kidney function measurements suitable for repeated-measurement GWAS are still scarce [39,40]. Although allele-frequency–matched random-variant simulations provide an internal benchmark for expected null behavior, they do not constitute independent replication and cannot fully address cohort-specific biases; alternative internal validation strategies (e.g., permutation or split-sample analyses) may be considered when suitable cohorts are available. Fifth, our GWAS focused exclusively on autosomal variants, leaving open the possibility that meaningful associations on the sex chromosomes were overlooked. Sixth, information on medication use, particularly renin-angiotensin system inhibitors, was not available in the Taiwan Biobank, which prevented the assessment of potential confounding or mediating effects. Future studies incorporating multi-ethnic cohorts, rare variant analyses, sex chromosome data, and detailed medication information, together with experimental validation, will be essential to clarify the biological roles of the identified variants and genes. Finally, because most participants had only two eGFR measurements, our study could not characterize nonlinear trajectories or rates of decline, which should be addressed in future studies with denser follow-up.
In conclusion, this GWAS leveraging repeated eGFR measurements in a Taiwanese cohort identified six genetic loci associated with kidney function. Through comprehensive functional annotation, we highlighted MUC1/EFNA1, SHROOM3, HLA-DQB1, MPPED2, NRG4, and PGAP3/FBXL20 as promising candidate causal genes influencing eGFR. These findings may inform future studies of kidney function genetics but require replication in independent cohorts and functional validation to clarify their biological relevance.
Supplementary Materials
Supplementary data are available at Kidney Research and Clinical Practice online (https://doi.org/10.23876/j.krcp.25.430).
Notes
Conflicts of interest
All authors have no conflicts of interest to declare.
Funding
This study was funded by the National Taiwan University Hospital (Grant No: 113-X0014), the Mrs. Hsiu-Chin Lee Kidney Research Foundation, and the National Science and Technology Council (Grant No: 111-2314-B-038-112-MY3), Taipei, Taiwan. The funding agency had no role in the study design, data collection, data analysis, data interpretation, writing of the report, or the decision to submit the report for publication.
Acknowledgments
We acknowledge the Taiwan Biobank for providing data and support for this research. The authors thank the Data Science Statistical Cooperation Centre of Academia Sinica for their support.
Data sharing statement
Summary data are available in the supplementary materials. The individual-level data supporting the findings of this study originate from the Taiwan Biobank. Owing to restrictions on data availability, these data are not publicly accessible and can only be used under a license for the current study. However, these data can be made available upon reasonable request and with permission from the Taiwan Biobank.
Authors’ contributions
Conceptualization, Methodology: YCL, HMT, JSH, YCC, YHJS, VCW, SLL, TSL
Data curation, Formal analysis, Investigation, Resources: YCL, HMT, YHC, TSL
Funding acquisition: YCL, TSL
Supervision: TSL
Writing–original draft: YCL, HMT
Writing–review & editing: All authors
All authors read and approved the final manuscript.
