Abstract
Context
Observational studies investigating the role of vitamin D in renal function have yielded inconsistent results.
Objective
We tested whether 25-hydroxyvitamin D (25[OH]D) serum levels are associated with renal function, and inversely, whether altered renal function causes changes in 25(OH)D, using Mendelian randomization (MR).
Methods
In this two-sample MR study, we used single nucleotide polymorphisms (SNP) associated with 25(OH)D in 443 734 Europeans and evaluated their effects on estimated glomerular filtration rate (eGFR), blood urea nitrogen (BUN), chronic kidney disease (CKD) risk and progression in genome-wide association studies totaling over 1 million Europeans. To control for pleiotropy, we also used SNPs solely in DHCR7, CYP2R1, and GC, all genes with known roles in vitamin D metabolism. We performed a reverse MR, using SNPs for the above indices of renal function to study causal effects on 25(OH)D levels.
Results
We did not find robust evidence supporting effects of 25(OH)D on eGFR, BUN, and CKD or its progression. Our inverse variance weighted MR demonstrated a 0.56 decrease in standardized log-transformed 25(OH)D (95% CI −0.73, −0.41; P = 2.89 × 10−12) per unit increase in log-transformed eGFR. Increased BUN was associated with increased 25(OH)D (β = 0.25, 95% CI 0.15, 0.36; P = 4.12 × 10−6 per unit increase in log-transformed BUN). Finally, genetically predicted CKD conferred a 0.05 increase in standardized log-transformed 25(OH)D level (95% CI 0.04, 0.06; P = 1.06 × 10−13). Other MR methods confirmed the findings of the main analyses.
Conclusion
Genetically predicted CKD, increased BUN, and decreased eGFR are associated with increased 25(OH)D levels, but we found no causal effect of 25(OH)D on renal function in Europeans.
Keywords: Mendelian randomization, vitamin D, renal function, causal inference, chronic kidney disease, GWAS
The kidney plays an important role in vitamin D metabolism. The main sources of vitamin D (cholecalciferol or ergocalciferol) in humans are nutrition and skin synthesis from 7-dehydrocholesterol after exposure to ultraviolet irradiation. Vitamin D then undergoes 25-hydroxylation in the liver to 25-hydroxyvitamin D (25[OH]D), which is converted into the active form, 1,25-dihydroxyvitamin D (1,25[OH]2D) by 1 alpha-hydroxylation in the kidney. The 1,25(OH)2D can then be converted by the enzyme 24-hydroxylase to an inactive form, 24,25-dihydroxyvitamin D. Observational studies have reported lowered 25(OH)D levels in individuals with chronic kidney disease (CKD) (1), while other studies have shown an independent link between vitamin D deficiency, defined as 25(OH)D levels below 50 nmol/L, and mortality in CKD (2, 3). While several mechanisms limit the capacity of the kidney to maintain 1,25(OH)2D levels in CKD (4), it remains unclear whether vitamin D deficiency increases the risk of de novo CKD or causes changes in kidney function in the general population.
To date, several observational studies have investigated the effects of vitamin D status on indices of renal function in individuals with CKD with inconsistent results (5–9). These inconsistencies can be explained by the fact that observational associations might be influenced by confounding factors or reverse causation. Cross-sectional designs are particularly inappropriate when studying the relationship between vitamin D metabolism and renal function, due to the interrelation between these two. For example, individuals with both renal disease and low vitamin D levels can present with chronic inflammation or diabetes, which act as confounders when studying the association between 25(OH)D and CKD (10). On the other hand, reverse causation could lead to spurious associations between an exposure, such as 25(OH)D levels, and an outcome, such as CKD, since individuals affected by CKD tend to go outside less, have poor diet, often suffer from proteinuria, or are on medications which predispose to vitamin D deficiency. Understanding whether low 25(OH)D levels are causal in de novo CKD or its progression to end-stage CKD could be of significant public health interest, since vitamin D deficiency can be easily diagnosed and treated. The gold standard to establish causation is randomized controlled trials (RCT), but RCT evidence supporting beneficial effects of vitamin D supplementation in CKD is sparse (11–13).
Mendelian randomization (MR) is a method of causal inference, which uses genetics to test for causal associations between a biomarker, such as 25(OH)D levels, and a disease (such as CKD), or a trait (such as estimated glomerular filtration rate [eGFR] or blood urea nitrogen [BUN] levels) (14). MR offers a framework which is akin to an RCT, since according to Mendel's second law there is a natural randomization of risk-increasing disease alleles at conception of each individual. Currently, there is some evidence of a possible negative causal effect of 25(OH)D on eGFR from an MR study that has used as instruments 3 single nucleotide polymorphisms (SNPs) in the GC, CYP2R1, and DHCR7 genes (15). However, the 3 variants used in this study explain only 2% of 25(OH)D variance, and furthermore, larger genome-wide association study (GWAS) on 25(OH)D levels in European populations have been published more recently (16, 17). These studies identified up to 69 independent vitamin D loci, explaining a larger portion of the variance in 25(OH)D and offering a better set of instruments for MR studies for vitamin D.
In this project, we undertook bidirectional MR analyses to overcome the limitations of observational studies in order to unravel the relationship between 25(OH)D and kidney function. Specifically, our objective was to use MR to evaluate the causal association between genetically determined 25(OH)D levels and eGFR, BUN, and risk of CKD in a healthy population and risk of CKD progression in a population with CKD. Given the interrelation of vitamin D metabolism and kidney function, we also aimed to investigate the inverse association, that is, whether genetically altered renal function affects 25(OH)D levels.
Methods
Genetic Variants Associated With Vitamin D
To assess whether genetically altered vitamin D levels are causally associated with renal function, we used conditionally independent SNPs associated with 25(OH)D as instruments for 25(OH)D in a large GWAS meta-analysis of UK Biobank and the SUNLIGHT consortium totaling 443 734 individuals of European ancestry (16) (Fig. 1). In the MR framework, the genetic instruments (SNPs) should satisfy 3 main assumptions. The first MR assumption requires that the instruments are strongly associated with the exposure, 25(OH)D; we thus ensured that all SNPs were associated with 25(OH)D at a genome-wide significant level (P < 5 × 10−8). The second MR assumption requires that the instruments in a MR study should not be associated with confounders in the association between the exposure and the outcome. Since one of these confounders is ancestry, we limited our analyses to European populations. According to the third MR assumption, genetic variants used as instruments for an exposure should not be associated with the outcome in pathways other than those that include the exposure of interest (also known as the restriction and exclusion hypothesis) (18). In the context of MR, horizontal pleiotropy refers to a scenario in which this assumption is violated. We thus undertook sensitivity analyses using MR methods assessing for the presence of horizontal pleiotropy (18–20) and as an additional strategy, we performed analyses restricting the MR instruments to SNPs in genes directly involved in vitamin D metabolism (21).
Figure 1.
Flowchart demonstrating the main and reverse MR study design.
Therefore, we extracted estimates of the effects of the 25(OH)D SNPs on risk of prevalent CKD from a large GWAS meta-analysis from the CKDGen consortium, comprising 23 studies of European ancestry for CKD, totaling 41 395 cases and 439 303 controls (22). We extracted effects of the 25(OH)D SNPs on eGFR and BUN from a meta-analysis of the UK Biobank with the CKDGen consortium, totaling 1 201 909 individuals of European ancestry (23). For CKD progression, we used data from an unpublished GWAS totaling 71 369 Europeans, from BioVU (24, 25) and the Million Veteran Program (26), where CKD progression was defined as the longitudinal % change per year in eGFR in patients with diagnosed CKD. In this GWAS, analyses were restricted to individuals with a minimum of 4 longitudinal eGFR measurements to optimize estimation of the slope. On average, there were 6 (SD = 4) serum creatinine values per individual, over a median 6 years of follow-up. Descriptive data for the GWAS populations and of the measurement methods for 25(OH)D, CKD, BUN, eGFR, and CKD progression in the various GWAS can be found in Supplementary Table S1. For 25(OH)D-related variants not directly present in GWAS of renal function, we selected proxy SNPs (LD r2 > 0.7) using the LDproxy function in ldlink (27) in all EUR populations from the 1000 genomes phase 3 panel. We also calculated the F-statistic for our set of 25(OH)D SNPs, as an additional metric of the strength of our MR instrument.
As shown in Fig. 1, we repeated all our two-sample MR analyses applying Steiger filtering. Given the interrelation between 25(OH)D levels and renal function, we applied Steiger filtering to exclude variants that have larger effects on the outcome (indices of renal function) than the exposure (25(OH)D). We used the Wald ratio to estimate the effect of each 25(OH)D SNP on the studied outcomes and combining the individual MR estimates for each SNP in an inverse variance weighted (IVW) random effects MR meta-analysis (28). The results of our MR analyses express the effect of 1 SD increase in the logarithmic level of 25(OH)D (corresponding to a 40.9 mmol/L change in serum 25(OH)D level in vitamin D–sufficient individuals) on log-transformed eGFR, BUN, CKD risk, and CKD progression. The direct acyclic graph of our MR analyses is presented in Fig. 2.
Figure 2.
Direct acyclic graphs (DAG) of the MR analyses. (A) Main MR analyses on the effect of 25(OHD) on eGFR, BUN, CKD, and CKD progression. (B) Reverse MR analyses on the effect of eGFR, BUN, CKD, and CKD progression on 25(OH)D levels.
MR Methods to Assess Pleiotropy
To test for the presence of horizontal pleiotropy, we applied various MR methods (weighted median (19), weighted mode (20), and MR-Egger (18)), which take into account potential pleiotropic effects of the SNP instruments.
Specifically, the weighted median approach (19) relies on the fact that estimates of SNPs without pleiotropic effects are more likely to converge toward the median, whereas we would expect pleiotropy to introduce heterogeneity and result in relative outliers. This method provides reliable results when less than 50% of the total MR effect comes from variants with pleiotropic effects. The weighted mode approach is similar to the previous one but relies on an estimation based on the mode rather than on the median, allowing the majority of SNPs to be pleiotropic (20). Finally, the MR-Egger regression addresses potentially unbalanced directional pleiotropy (18), as the regression slope provides a robust causal estimate, and the intercept estimates the directional pleiotropy. As such, MR-Egger allows a weakening of the restriction and exclusion hypothesis and requires that the association of each variant with the exposure is not correlated with its pleiotropic effect (known as the InSIDE hypothesis). As an additional control for pleiotropy, we used MR-PRESSO (global, outlier, and distortion tests) (29) to exclude outlier SNPs with potential pleiotropic effects.
Finally, we tested the heterogeneity of the SNP instruments for our exposures using the Cohran Q metric, and generated MR estimates by omitting SNPs appearing as outliers, using the Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) approach (29).
Sensitivity Analyzes With Sets of 25(OH)D SNPs
To further ensure that our estimates are not influenced by pleiotropy, we repeated the MR analyses using only genetic variants for 25(OH)D levels found in or next to DHCR7, CYP2R1, and GC, which are genes with a known role in synthesis, transport, or degradation of vitamin D (21), and as such, SNPs in these genes are less likely to be pleiotropic. We purposely did not include SNPs in the CYP24A1 gene, despite its known role in vitamin D catabolism, since this gene is expressed in the kidney and thus its SNPs can exert pleiotropic effects. Effects (betas) of the SNPs mapping in the 3 vitamin D genes on 25(OH)D levels were retrieved from the vitamin D GWAS conducted by Jiang et al (30).
Sensitivity Analysis Excluding SNPs Associated With Confounders
In a previous MR paper by our group, studying the effects of 25(OH)D SNPs on risk of type 1 diabetes, we undertook a look-up for association with potential confounders of the conditionally independent 25(OH)D SNPs used as instruments using Phenoscanner (31). We found significant GWAS associations of the some of the 25(OH)D SNPs, with traits clustering into 14 categories. The most predominant categories are serum lipids, blood traits, body composition (including BMI), and respiratory traits, while a few SNPs were associated with type 2 diabetes. We used this information to conduct a sensitivity analysis, excluding SNPs associated at a genome-wide level with body composition or type 2 diabetes, which can act as confounders when studying the association between 25(OH)D and renal function.
We used the TwoSampleMR R package (32), and its default parameters (LD clumping r2 = 0.001) to select our instruments for 25(OH)D, to harmonize them between the exposure and outcome GWAS, and to calculate 4 different MR estimates (IVW, weighted median, MR-Egger, and weighted mode) in the main analysis and sensitivity analyses with restricted sets of 25(OH)D SNPs. We elected to use random effects IVW, given the evidence of heterogeneity in some of our analyses. We used the same TwoSampleMR R package to generate scatter plots and a forest plot to visualize the MR estimates generated using different methods. Our MR-PRESSO analysis was implemented using the MR-PRESSO R package (29). To account for multiple testing, the MR P value threshold for significance was set to 0.05/4 = 0.0125 (for 4 tested outcomes: eGFR, BUN, CKD, and CKD progression).
Reverse MR Analysis of the Effect of eGFR, BUN, and CKD on 25(OH)D Levels
To address our second objective, we used the MR approaches described before to interrogate causal effects of renal function (eGFR, BUN, and risk of CKD and its progression) on 25(OH)D levels (Fig. 1). To this end, we extracted independent SNPs significantly associated with risk of CKD, CKD progression, levels of eGFR, and BUN in the aforementioned large GWAS (22, 23). To do this, we applied LD clumping using the default parameters in TwoSampleMR. The effects of these SNPs on 25(OH)D levels were then extracted from a large GWAS on levels of 25(OH)D (16). Similar to the main MR analysis, Steiger filtering was applied to exclude SNPs with larger effects on 25(OH)D than on the exposures. We undertook sensitivity MR analyses excluding SNPs associated with body composition or type 2 diabetes in Phenoscanner. We computed IVW, weighted median, weighted mode, MR-Egger and MR-PRESSO estimates for each of the 4 exposures (eGFR, BUN, risk of CKD, and CKD progression). The direct acyclic graph of our reverse MR analyses is presented in Fig. 2.
Statistical Power Analysis
To test whether our study was adequately powered to detect clinically relevant changes of CKD risk, eGFR, BUN, and CKD progression risk in our main MR analyses, and relevant changes in 25(OH)D in our reverse MR analysis, we used a previously described MR power calculation method (33). Specifically, we computed the MR betas (for continuous outcomes, such as 25(OH)D, eGFR, BUN, CKD progression) or OR (for the binary outcome of CKD) for which we obtained a power of 80%, setting the alpha level at 0.05/4 = 0.0125, using the variance explained of each exposure by its respective genetic instruments, and a sample size of CKD of 480 698 (among which cases = 41 395), eGFR of 460 826, BUN of 852 678, and CKD progression of 71 396, and 25(OH)D of 443 734.
Results
MR Study on the Effect of Serum 25(OH)D in eGFR, BUN, CKD Risk, and CKD Progression
Among the initial list of 138 independent common and rare variants reported in the 25(OH)D GWAS, we first excluded 4 ambiguous SNPs (ie, SNPs with nonconcordant alleles eg, A/G vs A/C). We kept 87 SNPs with minor allele frequency (MAF) > 1%, and after applying LD clumping (r2 < 0.001), the SNP number was reduced to 57. All 57 SNPs were present in the CKD progression GWAS, while in the BUN/eGFR/CKD GWAS, we found 56, 55, and 56 directly matching SNPs. We were not able to find proxy SNPs with LD r2 > 0.8 for the 1 to 2 missing SNPs in the 3 GWAS.
In the main MR study using up to 57 common 25(OH)D SNPs or their proxies present in the GWAS of the outcomes, explaining up to 3.5% of the variance in 25(OH)D levels, we did not find a consistent association between 25(OH)D and the eGFR, CKD, and CKD progression at the Bonferroni-corrected level of significance (Table 1; Supplementary Tables S2-S6; Supplementary Figs. S1-S4 (34)). As shown in Table 1, our MR results remained largely unchanged before and after applying Steiger filtering. In the analysis with Steiger filtering, the results were null across different MR methods for all the outcomes (BUN, CKD, and CKD progression) with the exception of the results of the weighted mode showing a negative effect of 25(OH)D on eGFR (β = −0.004; 95% CI −0.007, −0.002; P = 0.002). However, this result was not supported by the MR-Egger, IVW, and the weighted median methods. As shown in Table 1 and Supplementary Table S6 (34), although the intercept of the MR-Egger provided no evidence of unbalanced horizontal pleiotropy except for eGFR, removal of outlier SNPs using the MR-PRESSO approach unveiled a possible negative effect of 25(OH)D on eGFR (β = −0.007, P = 0.001), and a positive association with BUN (β = 0.012, P = 0.002). Given the inconsistency of the results between the main MR methods and the MR-PRESSO, we conclude that the above findings, although suggestive, cannot unequivocally support a role of 25(OH)D in renal function.
Table 1.
MR analyses of the effect of 25(OH)D on eGFR, BUN, CKD, and CKD progression
| Analysis | eGFR | |||||||
|---|---|---|---|---|---|---|---|---|
| Beta | CI lower | CI upper | P value | Egger intercept | Intercept P value | heterogeneity Q | Q P value | |
| Analysis with all SNPs (n = 56 SNPs) | Without Steiger filtering | |||||||
| Inverse variance weighted (random) | −0.014 | −0.030 | 0.003 | 0.099 | 2709.512 | <1E-199 | ||
| Weighted median | −0.004 | −0.007 | −0.001 | 0.011 | ||||
| MR-Egger (random) | 0.000 | −0.020 | 0.020 | 0.988 | −0.001 | 0.037 | 2939.245 | <1E-199 |
| Weighted mode | −0.005 | −0.008 | −0.002 | 0.001 | ||||
| Analysis with all SNPs (n = 50 SNPs) | With Steiger filtering | |||||||
| Inverse variance weighted (random) | −0.009 | −0.017 | −0.002 | 0.014 | 476.573 | 6.56E–71 | ||
| Weighted median | −0.004 | −0.007 | −0.001 | 0.014 | ||||
| MR-Egger (random) | −0.003 | −0.012 | 0.007 | 0.582 | −0.0005 | 0.028 | 493.449 | 1.03E–73 |
| Weighted mode | −0.004 | −0.007 | −0.002 | 0.002 | ||||
| Analysis with 3 genes (n = 3 SNPs) | ||||||||
| Inverse variance weighted (random) | −0.008 | −0.012 | −0.004 | 0.0002 | 0.802 | 0.370 | ||
| Weighted median | −0.007 | −0.011 | −0.003 | 0.0004 | ||||
| MR-Egger (random) | −0.002 | −0.012 | 0.008 | 0.757 | −0.001 | 0.428 | 2.381 | 0.304 |
| Weighted mode | −0.006 | −0.011 | −0.002 | 0.107 | ||||
| BUN | ||||||||
|---|---|---|---|---|---|---|---|---|
| beta | CI lower | CI upper | P value | Egger intercept | Intercept P value | heterogeneity Q | Q P value | |
| Analysis with all SNPs (n = 55 SNPs) | Without Steiger filtering | |||||||
| Inverse variance weighted (random) | 0.012 | −0.005 | 0.028 | 0.172 | 971.604 | 4.48E–169 | ||
| Weighted median | 0.005 | 0.000 | 0.010 | 0.067 | ||||
| MR-Egger (random) | 0.001 | −0.022 | 0.023 | 0.936 | 0.001 | 0.106 | 1019.517 | 2.65E–178 |
| Weighted mode | 0.006 | 0.001 | 0.011 | 0.015 | ||||
| Analysis with all SNPs (n = 52 SNPs) | With Steiger filtering | |||||||
| Inverse variance weighted (random) | 0.010 | −0.002 | 0.023 | 0.104 | 476.573 | 6.56E–71 | ||
| Weighted median | 0.005 | −0.001 | 0.010 | 0.095 | ||||
| MR-Egger (random) | 0.003 | −0.013 | 0.020 | 0.687 | 0.0004 | 0.182 | 493.449 | 1.03E–73 |
| Weighted mode | 0.006 | 0.001 | 0.010 | 0.023 | ||||
| Analysis with 3 genes (n = 3 SNPs) | ||||||||
| Inverse variance weighted (random) | 0.006 | −0.005 | 0.017 | 0.280 | 4.512 | 0.034 | ||
| Weighted median | 0.003 | −0.003 | 0.010 | 0.342 | ||||
| MR-Egger (random) | −0.004 | −0.038 | 0.031 | 0.866 | 0.001 | 0.652 | 6.188 | 0.045 |
| Weighted mode | 0.003 | −0.004 | 0.010 | 0.484 | ||||
| CKD | ||||||||
|---|---|---|---|---|---|---|---|---|
| Odds ratio | CI lower | CI upper | P value | Egger intercept | Intercept P value | heterogeneity Q | Q P value | |
| Analysis with all SNPs (n = 56 SNPs) | Without Steiger filtering | |||||||
| Inverse variance weighted (random) | 1.121 | 0.895 | 1.405 | 0.320 | 462.029 | 3.73E-66 | ||
| Weighted median | 1.053 | 0.956 | 1.161 | 0.295 | ||||
| MR-Egger (random) | 1.002 | 0.751 | 1.337 | 0.989 | 0.007 | 0.228 | 474.762 | 3.87E–68 |
| Weighted mode | 1.055 | 0.964 | 1.155 | 0.251 | ||||
| Analysis with all SNPs (n = 52 SNPs) | With Steiger filtering | |||||||
| Inverse variance weighted (random) | 1.076 | 0.960 | 1.206 | 0.208 | 112.667 | 1.50E–06 | ||
| Weighted median | 1.053 | 0.952 | 1.164 | 0.318 | ||||
| MR-Egger (random) | 1.039 | 0.897 | 1.203 | 0.612 | 0.002 | 0.458 | 113.901 | 1.60E–06 |
| Weighted mode | 1.042 | 0.953 | 1.139 | 0.373 | ||||
| Analysis with 3 genes (n = 3 SNPs) | ||||||||
| Inverse variance weighted (random) | 1.137 | 1.009 | 1.281 | 0.035 | 0.038 | 0.845 | ||
| Weighted median | 1.126 | 0.988 | 1.284 | 0.075 | ||||
| MR-Egger (random) | 0.930 | 0.668 | 1.296 | 0.743 | 0.019 | 0.420 | 1.699 | 0.428 |
| Weighted mode | 1.091 | 0.946 | 1.259 | 0.355 | ||||
| CKD progression | ||||||||
|---|---|---|---|---|---|---|---|---|
| Beta | CI lower | CI upper | P value | Egger intercept | Intercept P value | heterogeneity Q | Q P value | |
| Analysis with all SNPs (n = 57 SNPs) | Without Steiger filtering | |||||||
| Inverse variance weighted (random) | 0.008 | −0.273 | 0.288 | 0.957 | 154.989 | 3.16E–11 | ||
| Weighted median | −0.086 | −0.287 | 0.116 | 0.404 | ||||
| MR-Egger (random) | −0.023 | −0.385 | 0.338 | 0.899 | 0.002 | 0.787 | 154.782 | 1.99E–11 |
| Weighted mode | −0.075 | −0.285 | 0.135 | 0.487 | ||||
| Analysis with all SNPs (n = 34 SNPs) | With Steiger filtering | |||||||
| Inverse variance weighted (random) | −0.018 | −0.196 | 0.161 | 0.847 | 10.803 | 1.000 | ||
| Weighted median | −0.092 | −0.302 | 0.119 | 0.395 | ||||
| MR-Egger (random) | −0.033 | −0.256 | 0.191 | 0.776 | 0.001 | 0.827 | 10.754 | 1.000 |
| Weighted mode | −0.106 | −0.312 | 0.100 | 0.321 | ||||
| Analysis with 3 genes (n = 3 SNPs) | ||||||||
| Inverse variance weighted (random) | −0.006 | −0.407 | 0.395 | 0.977 | 4.776 | 0.092 | ||
| Weighted median | −0.029 | −0.305 | 0.248 | 0.839 | ||||
| MR-Egger (random) | −0.605 | −1.296 | 0.085 | 0.336 | 0.063 | 0.316 | 1.082 | 0.298 |
| Weighted mode | −0.121 | −0.416 | 0.174 | 0.505 | ||||
Abbreviations: BUN, blood urea nitrogen; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; MR, Mendelian randomization; SNP, single nucleotide polymorphism.
Significant P-values are denoted in bold.
As demonstrated in Table 1, the results of our MR sensitivity analyses using 3 SNPs in the 3 genes with known function in vitamin D metabolism (explaining 1.03% of the variance in 25(OH)D), did not indicate any significant effect of 25(OH)D on BUN, CKD, or CKD progression, except for eGFR. We observed a possible negative effect of 25(OH)D on eGFR (IVW MR β = −0.008, P = 0.0002 and weighted median MR β = −0.007, P = 0.0004); however, these results were not supported by the MR-Egger and weighted mode. The results for BUN, CKD, and CKD progression remained null, and unchanged across the various MR methods. The P value of the MR-Egger intercept in all the above sensitivity analyses was nonsignificant, implying the absence of unbalanced horizontal pleiotropy in our instruments. Our sensitivity analysis excluding SNPs with effects on type 2 diabetes or body composition yielded similar results as those of the main analysis (Supplementary Table S7 (34)) with the exception of a significant estimate of the weighted median and weighted mode analyses for CKD when excluding SNPs associated with type 2 diabetes.
The significant Cohran Q P value in all main analyses but not in the sensitivity analyses with 3 SNPs for eGFR and CKD, suggested presence of heterogeneity among the 25(OH)D SNPs used as instruments (Table 1), notably in MR analyses using a greater number of SNPs as instruments. All 25(OH)D SNPs used as instruments in our MR analyses had an F-statistic >10 (average F-statistic was 234; Supplementary Tables S2-S6 (34)).
MR Study on the Effect of eGFR, BUN, CKD Risk, and CKD Progression on Serum 25(OH)D
As shown in Table 2, the results of our reverse MR analysis with and without Steiger filtering remained consistent throughout the exposures (eGFR, BUN, CKD, and CKD progression). Importantly, after applying Steiger filtering, our reverse analysis showed strong evidence of a negative effect of genetically determined eGFR on 25(OH)D levels. Using 330 independent SNPs as MR instruments for eGFR (explaining 5.3% of its variance), we observed that a genetically predicted 1-unit increase in eGFR level in the logarithmic scale was associated to a 0.57 SD decrease in log-transformed 25(OH)D (β = −0.57; 95% CI −0.73, −0.41; P = 2.89 × 10−12). Using pleiotropy-robust MR methods, we found results in the same direction and the magnitude of the effect was comparable across the various MR methods (Supplementary Tables S8 and S12; Supplementary Fig. S5 (34)).
Table 2.
MR analyses of the effect of eGFR, BUN, CKD, and CKD progression on 25(OH)d levels
| Analysis | beta | CI lower | CI upper | P value | Egger intercept | Intercept P value | heterogeneity Q | Q P value |
|---|---|---|---|---|---|---|---|---|
| MR of the effect of eGFR on 25OHD (N = 332 SNPs) | ||||||||
| Inverse variance weighted (random) | −0.645 | −0.822 | −0.469 | 7.63E–13 | 1601.612 | 2.08E–165 | ||
| Weighted median | −0.676 | −0.828 | −0.525 | 2.03E–18 | ||||
| MR-Egger (random) | −0.832 | −1.224 | −0.441 | 3.97E–05 | 0.0007 | 0.2849 | 1603.799 | 9.32E–165 |
| Weighted mode | −0.721 | −0.938 | −0.504 | 2.71E–10 | ||||
| MR of the effect of eGFR on 25OHD (N = 330 SNPs) with Steiger filtering | ||||||||
| Inverse variance weighted (random) | −0.567 | −0.726 | −0.408 | 2.89E–12 | 1312.102 | 2.32E–117 | ||
| Weighted median | −0.672 | −0.829 | −0.516 | 3.57E–17 | ||||
| MR-Egger (random) | −0.705 | −1.062 | −0.348 | 1.32E–04 | 0.000 | 0.416 | 1309.626 | 2.93E–117 |
| Weighted mode | −0.727 | −0.949 | −0.504 | 5.29E–10 | ||||
| MR of the effect of BUN on 25OHD (N = 192 SNPs) | ||||||||
| Inverse variance weighted (random) | 0.269 | 0.132 | 0.406 | 1.20E–04 | 1014.450 | 2.64E–112 | ||
| Weighted median | 0.282 | 0.160 | 0.405 | 6.49E–06 | ||||
| MR-Egger (random) | 0.200 | −0.102 | 0.503 | 0.196 | 0.0004 | 0.6539 | 1019.162 | 4.77E–112 |
| Weighted mode | 0.311 | 0.118 | 0.503 | 0.002 | ||||
| MR of the effect of BUN on 25OHD (N = 188 SNPs) with Steiger filtering | ||||||||
| Inverse variance weighted (random) | 0.254 | 0.146 | 0.363 | 4.12E–06 | 632.539 | 9.38E–50 | ||
| Weighted median | 0.283 | 0.167 | 0.399 | 1.74E–06 | ||||
| MR-Egger (random) | 0.161 | −0.081 | 0.402 | 0.19 | 0.001 | 0.399 | 629.754 | 1.36E–49 |
| Weighted mode | 0.317 | 0.112 | 0.523 | 0.0028 | ||||
| MR of the effect of CKD on 25OHD (N = 22 SNPs)* | ||||||||
| Inverse variance weighted (random) | 0.048 | 0.035 | 0.061 | 1.06E–13 | 29.332 | 0.1063 | ||
| Weighted median | 0.055 | 0.039 | 0.071 | 5.40E–12 | ||||
| MR-Egger (random) | 0.083 | 0.055 | 0.112 | 1.33E–05 | −0.0035 | 0.0096 | 21.816 | 0.3506 |
| Weighted mode | 0.062 | 0.043 | 0.043 | 2.69E–06 | ||||
| MR of the effect of CKD progression on 25OHD (N = 5 SNPs)* | ||||||||
| Inverse variance weighted (random) | −0.024 | −0.048 | 0.000 | 0.051 | 15.894 | 0.0032 | ||
| Weighted median | −0.002 | −0.026 | 0.021 | 0.838 | ||||
| MR-Egger (random) | −0.070 | −0.102 | −0.038 | 0.024 | 0.0094 | 0.0308 | 3.827 | 0.2808 |
| Weighted mode | −0.048 | −0.072 | −0.024 | 0.017 | ||||
Abbreviations: 25(OH)D, 25-hydroxyvitamin D; BUN, blood urea nitrogen; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; MR, Mendelian randomization; SNP, single nucleotide polymorphism.
*No SNPs removed after Steiger filtering.
Significant P-values are denoted in bold.
When studying the effect of BUN on 25(OH)D levels after applying Steiger filtering, we found a significant positive effect. We used 188 SNPs as instruments for BUN, explaining 4% of its variance. Our MR analysis showed that a genetically predicted 1-unit increase in log-transformed BUN level was associated with an increase in standardized log-transformed 25(OH)D levels (β = 0.254; 95% CI 0.146, 0.363; P = 4.12 × 10−6). The estimates from the other MR methods were generally in the same direction as the IVW result (Table 2; Supplementary Tables S9 and S12; Supplementary Fig. S6 (34)).
Finally, our CKD analysis using 22 SNPs explaining 6.5% of its variance, demonstrated that genetically predicted CKD was associated with an average increase in 25(OH)D levels by 0.05 SD on the logarithmic scale (IVW β = 0.047; 95% CI 0.035, 0.061; P = 1.06 × 10−13). Pleiotropy-robust MR methods also supported effects in the same direction. We did not observe a significant effect of CKD progression on 25(OH)D levels using different MR methods and a set of 5 SNPs, explaining 5.6% of the variance in CKD progression (Table 2; Supplementary Tables S10-S12; Supplementary Fig. S7 (34)).
Again, we detected significant heterogeneity in our instruments using the Cohran Q test in all our analyses except the one for CKD (Table 2). We also observed significant P values in the MR-Egger intercepts of the analyses for CKD risk and CKD progression, implying the presence of horizontal pleiotropy; nevertheless the MR-PRESSO did not detect any outliers in these analyses (Supplementary Table S12 (34)). The results of the sensitivity analyses excluding eGFR, BUN, and CKD SNPs associated with type 2 diabetes or body composition were consistent with those of the main reverse MR analyses (Supplementary Table S13 (34)). Finally, the average F-statistic of the SNPs used as instruments was 159 for eGFR, 176 for BUN, 898 for CKD risk, and 666 for CKD progression, and all SNPs had F-statistics >10 (Supplementary Tables S8-S11 (34)).
MR Power Analysis
Our main MR study had 80% power to detect effects of 25(OH)D levels as small as an OR of 1.09 for CKD, and betas as small as of −0.0095 for eGFR, 0.059 for BUN, 0.17 for CKD progression per 1 SD increase in standardized log-transformed 25(OH)D (Supplementary Table S14 (34)). Moreover, our reverse MR study had 80% power to detect effects of CKD, BUN, eGFR, and CKD progression on 25(OH)D ranging from betas of −0.008 to 0.008 per unit increase in the above exposures in the logarithmic scale (Supplementary Table S14 (34)). We conclude that all our analyses were well-powered, with the exception of the main analysis on the effect of vitamin D on CKD progression.
Discussion
Using MR, a method allowing for causal inference, and limiting bias from confounding and reverse causation, we did not observe clear evidence for a causal association between 25(OH)D levels and indices of renal function. Conversely, our reverse MR analysis suggested that genetically predicted CKD is linked to increased serum 25(OH)D. In the same direction, a genetically predicted increase in eGFR was associated to decreased 25(OH)D levels, while there was a positive association between BUN and 25(OH)D level. One of the main strengths of our study is the use of the largest available European GWAS for both 25(OH)D levels and indices or renal function, which provided us power for discovery of associations. Also, we undertook sensitivity analyses which with a few exceptions provided results in the same direction as the main IVW analyses. This approach offers further validation to our findings and ensures that our estimates were not influenced by pleiotropy. Moreover, we applied Steiger filtering as an additional measure to ensure that our MR findings are not influenced by reverse causation. Finally, the bidirectional design of our MR approach unraveled the presence of an inverse association between vitamin D and renal function.
Specifically, our findings suggest that the low 25(OH)D levels observed in individuals with CKD are not likely to be causal for their deteriorated renal function, but they are rather driven by confounding. Indeed, individuals with diabetes, obesity, and inflammation are at increased risk for both vitamin D deficiency and decreased in renal function (35). As such, these conditions can be regarded as confounders in the association between 25(OH)D levels and CKD. Also, it is known that 25(OH)D levels decline with age, since older individuals have decreased vitamin D synthesis in the skin (35). The prevalence of CKD also increases with aging, and thus age could be also considered as a confounder in the association of vitamin D and renal function. On the other hand, CKD patients tend to go out less, are less exposed to sunlight, and may have decreased skin synthesis of vitamin D (35, 36). Further, they often have poor diet and suboptimal vitamin D intake (37). One of the common causes of CKD is diabetic nephropathy, which is associated with proteinuria, which leads to depletion of vitamin D binding protein, and a subsequent decrease in serum 25(OH)D levels (38). All the above mechanisms support the presence of confounders in the association between vitamin D and renal function.
The results of our reverse MR study, suggesting an increased 25(OH)D in the context of genetically predicted decline in renal function are in contrast with the aforementioned observed decrease of 25(OH)D in CKD (5, 7–9). A plausible explanation for our MR result could be the downregulation of the one-alpha hydroxylase in the context of CKD. Also, increased FGF-23 and phosphorus levels, which are common in CKD, inhibit the activity of one-alpha hydroxylase, leading to an accumulation of the substrate, serum 25(OH)D (35). Finally, the decline in eGFR could limit the filtration of 25(OH)D, resulting in a buildup of 25(OH)D levels (35, 39–41). The observed decrease in 25(OH)D in CKD patients (42–44), despite the opposite direction of the effect indicated by our MR study, can be explained by the aforementioned confounders.
Our MR study has some limitations. One main limitation is that we did not address causal effects on renal function of altered 1,25(OH)2D levels, the active form of vitamin D, due to lack of available large GWAS on 1,25(OH)2D allowing to extract strong instruments for 1,25(OH)2D. However, our group has previously reported estimates of the 25(OH)D SNPs on 1,25(OH)2D (16), in 1591 participants in the Ely Study (16). While some 25(OH)D SNPs reached nominal significance, the lowest P value of 0.0003 was found for a SNP in the CYP24A1, which is pleiotropic when studying the association of 1,25(OH)2D with CKD. Hence, it is not feasible to perform such two-sample MR study. Another limitation of our study is that the SNPs which were used as instruments for 25(OH)D explain only 3.5% of variance in 25(OH)D levels in the main analysis, and only 1.03% of its variance in the sensitivity analysis with 3 SNPs, which may have affected the power of our study to identify very small effects of 25(OH)D on renal function. Moreover, in the present study, we did not undertake a stratified MR analysis to assess nonlinear effects, as we do not have access to the individual level data on renal function of participants in the 25(OH)D GWAS. This means that effects of extremely low or high 25(OH)D levels on renal function cannot be excluded. Interestingly, a recent study using a stratified MR approach has demonstrated a nonlinear causal effect of eGFR on 25(OH)D levels (45). Our MR studies were conducted using SNPs from European GWAS, and thus our results cannot be generalized to non-European populations. Future large GWAS on 25(OH)D and indices of renal function in non-European ancestries will enable to interrogate their associations in other ethnic populations. The samples where the MR instruments [“gene-25(OH)D” or “gene-renal function index”] and their causal effects [“25(OH)D-outcome” or “renal-function index-25(OH)D”] were estimated included partially overlapping GWAS participants from UK Biobank in the BUN and eGFR analyses, and as such the MR estimates would be biased toward the confounding effect (46). Nevertheless, we have repeated the same bidirectional MR analyses using data for BUN and eGFR from a previous GWAS not including UK Biobank (22), and we obtained similar results (Supplementary Table S15 (34)). Furthermore, our findings have not been validated with clinical or biochemical studies, as this exceeds the scope of this study.
In conclusion, our MR findings, demonstrating that genetically predicted CKD, decreased eGFR, and increased BUN are associated with higher, rather than lower, 25(OH)D levels, provide a further step in understanding a potential causal relationship between these traits, and providing support to decision-making processes, such as conducting clinical trials or interventions. Having elevated 25(OH)D levels due to reduced 1 hydroxylase activity would be expected to be associated with low 1,25(OH)2D levels. As the latter is the active form of vitamin D, this implies that elevated 25(OH)D levels may be associated with active vitamin D deficiency in individuals with CKD. Our findings do not support the use of vitamin D supplements to prevent de novo CKD in the general population or a decline in renal function in patients with CKD. However, our findings cannot exclude effects on renal function in individuals with frank vitamin D deficiency. Overall, the above findings have potential important public health implications suggesting that efforts to prevent the impact of confounders in the association of vitamin D and renal function, such as diabetes and inflammation, might render more promising results than vitamin D supplements in individuals with 25(OH)D levels within the normal range.
Abbreviations
- 1,25(OH)2D
1,25-dihydroxyvitamin D
- 25(OH)D
25-hydroxyvitamin D
- BUN
blood urea nitrogen
- CKD
chronic kidney disease
- eGFR
estimated glomerular filtration rate
- GWAS
genome-wide association study
- IVW
inverse variance weighted
- MR
Mendelian randomization
- RCT
randomized controlled trial
- SNP
single nucleotide polymorphism
Contributor Information
Manel Adi, Department of Biochemistry and Molecular Medicine, University of Montreal, Montreal, QC H3T1J4, Canada.
Faegheh Ghanbari, Research Center of the Sainte-Justine University Hospital, University of Montreal, Montreal, QC H3TAC5, Canada.
Mallory L Downie, Department of Renal Medicine, University College London, London NW32PF, UK.
Adriana Hung, Department of Medicine, Vanderbilt University Medical Center, Veterans Administration Tennessee Valley Healthcare System, Nashville, TN 37212, USA.
Cassiane Robinson-Cohen, Division of Nephrology, Vanderbilt University Medical Center, Nashville, TN 37232, USA.
Despoina Manousaki, Department of Biochemistry and Molecular Medicine, University of Montreal, Montreal, QC H3T1J4, Canada; Research Center of the Sainte-Justine University Hospital, University of Montreal, Montreal, QC H3TAC5, Canada; Department of Pediatrics, University of Montreal, Montreal, QC H3T1C5, Canada.
Funding
D.M. is a Fonds de Recherche du Quebec-Sante (FRQS) Junior 1 Scholar and has received a career development award from the Canadian Child Health Clinician Scientist Program (CCHCSP). M.L.D.'s work is supported by St. Peter's Trust for Kidney, Bladder, and Prostate Research and the KRESCENT Post-Doctoral Fellowship from the Kidney Foundation of Canada. C.R.-C.'s work is supported by R01DK122075 from the National Institutes of Diabetes and Digestive and Kidney Diseases (NIDDK).
Author Contributions
D.M. conceived the design of the experiment. D.M., M.A., and C.R.C. collected the data. M.A., D.M., and F.G. undertook the analyses and wrote the first draft of the manuscript. All authors interpreted the results and reviewed the manuscript. D.M. is the guarantor of the manuscript.
Disclosures
The authors have nothing to disclose.
Data Availability
All codes used in the current study are available upon request to the corresponding author.
References
- 1. Hruska K, Mathew S, Lund R, Fang Y, Sugatani T. Cardiovascular risk factors in chronic kidney disease: does phosphate qualify? Kidney Int Suppl. 2011;79(121):S9‐S13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Pilz S, Tomaschitz A, Friedl C, et al. Vitamin D status and mortality in chronic kidney disease. Nephrol Dial Transplant. 2011;26(11):3603‐3609. [DOI] [PubMed] [Google Scholar]
- 3. Jayedi A, Soltani S, Shab-Bidar S. Vitamin D status and all-cause mortality in patients with chronic kidney disease: a systematic review and dose–response meta-analysis. J Clin Endocrinol Metab. 2017;102(7):2136‐2145. [DOI] [PubMed] [Google Scholar]
- 4. Levin A, Bakris GL, Molitch M, et al. Prevalence of abnormal serum vitamin D, PTH, calcium, and phosphorus in patients with chronic kidney disease: results of the study to evaluate early kidney disease. Kidney Int. 2007;71(1):31‐38. [DOI] [PubMed] [Google Scholar]
- 5. de Boer IH, Katz R, Chonchol M, et al. Serum 25-hydroxyvitamin D and change in estimated glomerular filtration rate. Clin J Am Soc Nephrol. 2011;6(9):2141‐2149. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Damasiewicz MJ, Magliano DJ, Daly RM, et al. Serum 25-hydroxyvitamin D deficiency and the 5-year incidence of CKD. Am J Kidney Dis. 2013;62(1):58‐66. [DOI] [PubMed] [Google Scholar]
- 7. Guessous I, McClellan W, Kleinbaum D, et al. Serum 25-hydroxyvitamin D level and kidney function decline in a Swiss general adult population. Clin J Am Soc Nephrol. 2015;10(7):1162‐1169. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Rebholz CM, Grams ME, Lutsey PL, et al. Biomarkers of vitamin D status and risk of ESRD. Am J Kidney Dis. 2016;67(2):235‐242. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Keyzer CA, Lambers-Heerspink HJ, Joosten MM, et al. Plasma vitamin D level and change in albuminuria and eGFR according to sodium intake. Clin J Am Soc Nephrol. 2015;10(12):2119‐2127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Pilz S, Verheyen N, Grubler MR, Tomaschitz A, Marz W. Vitamin D and cardiovascular disease prevention. Nat Rev Cardiol. 2016;13(7):404‐417. [DOI] [PubMed] [Google Scholar]
- 11. Bhan I, Dobens D, Tamez H, et al. Nutritional vitamin D supplementation in dialysis: a randomized trial. Clin J Am Soc Nephrol. 2015;10(4):611‐619. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Meireles MS, Kamimura MA, Dalboni MA, Giffoni de Carvalho JT, Aoike DT, Cuppari L. Effect of cholecalciferol on vitamin D-regulatory proteins in monocytes and on inflammatory markers in dialysis patients: a randomized controlled trial. Clin Nutr. 2016;35(6):1251‐1258. [DOI] [PubMed] [Google Scholar]
- 13. de Boer IH, Zelnick LR, Ruzinski J, et al. Effect of vitamin D and Omega-3 fatty acid supplementation on kidney function in patients with type 2 diabetes: a randomized clinical trial. JAMA. 2019;322(19):1899‐1909. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey Smith G. Mendelian Randomization: using genes as instruments for making causal inferences in epidemiology. Stat Med. 2008;27(8):1133‐1163. [DOI] [PubMed] [Google Scholar]
- 15. Teumer A, Gambaro G, Corre T, et al. Negative effect of vitamin D on kidney function: a Mendelian randomization study. Nephrol Dial Transplant. 2018;33(12):2139‐2145. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Manousaki D, Mitchell R, Dudding T, et al. Genome-wide association study for vitamin D levels reveals 69 independent loci. Am J Hum Genet. 2020;106(3):327‐337. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Revez JA, Lin T, Qiao Z, et al. Genome-wide association study identifies 143 loci associated with 25 hydroxyvitamin D concentration. Nat Commun. 2020;11(1):1647. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol. 2015;44(2):512‐525. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol. 2016;40(4):304‐314. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol. 2017;46(6):1985‐1998. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Mokry LE, Ross S, Ahmad OS, et al. Vitamin D and risk of multiple sclerosis: a Mendelian randomization study. PLoS Med. 2015;12(8):e1001866. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Wuttke M, Li Y, Li M, et al. A catalog of genetic loci associated with kidney function from analyses of a million individuals. Nat Genet. 2019;51(6):957‐972. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Stanzick KJ, Li Y, Schlosser P, et al. Discovery and prioritization of variants and genes for kidney function in >1.2 million individuals. Nat Commun. 2021;12(1):4350. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Ritchie MD, Denny JC, Crawford DC, et al. Robust replication of genotype-phenotype associations across multiple diseases in an electronic medical record. Am J Hum Genet. 2010;86(4):560‐572. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Pendergrass S, Dudek SM, Roden DM, Crawford DC, Ritchie MD. Visual integration of results from a large DNA biobank (BioVU) using synthesis-view. In: Pacific Symposium on Biocomputing. World Scientific; 2011:265‐275. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Gaziano JM, Concato J, Brophy M, et al. Million Veteran Program: a mega-biobank to study genetic influences on health and disease. J Clin Epidemiol. 2016;70:214‐223. [DOI] [PubMed] [Google Scholar]
- 27. https://ldlink.nci.nih.gov. Accessed.
- 28. Burgess S, Butterworth A, Thompson SG. Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol. 2013;37(7):658‐665. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet. 2018;50(5):693‐698. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Jiang X, O'Reilly PF, Aschard H, et al. Genome-wide association study in 79,366 European-ancestry individuals informs the genetic architecture of 25-hydroxyvitamin D levels. Nat Commun. 2018;9(1):260. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Manousaki D, Harroud A, Mitchell RE, et al. Vitamin D levels and risk of type 1 diabetes: a Mendelian randomization study. PLoS Med. 2021;18(2):e1003536. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Hemani G, Zheng J, Elsworth B, et al. The MR-base platform supports systematic causal inference across the human phenome. Elife. 2018;7:e34408. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Brion MJ, Shakhbazov K, Visscher PM. Calculating statistical power in Mendelian randomization studies. Int J Epidemiol. 2013;42(5):1497‐1501. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Adi M GF, Downie M, Hung A, Robinson-Cohen C, Manousaki D. Supplemental material to be deposited. 2022. https://www.dropbox.com/s/zg9zpbrw1ugmha4/Supplemental_Tables.pdf?dl=0 [DOI] [PMC free article] [PubMed]
- 35. Melamed ML, Chonchol M, Gutierrez OM, et al. The role of vitamin D in CKD stages 3 to 4: report of a scientific workshop sponsored by the National Kidney Foundation. Am J Kidney Dis. 2018;72(6):834‐845. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Jacob AI, Sallman A, Santiz Z, Hollis BW. Defective photoproduction of cholecalciferol in normal and uremic humans. J Nutr. 1984;114(7):1313‐1319. [DOI] [PubMed] [Google Scholar]
- 37. National Kidney Foundation . K/DOQI clinical practice guidelines for bone metabolism and disease in chronic kidney disease. Am J Kidney Dis. 2003;42(4 Suppl 3):S1-201. [PubMed] [Google Scholar]
- 38. Sato KA, Gray RW, Lemann J. Urinary excretion of 25-hydroxyvitamin D in health and the nephrotic syndrome. J Lab Clin Med. 1982;99(3):325‐330. [PubMed] [Google Scholar]
- 39. Dusso AS, Brown AJ, Slatopolsky E. Vitamin D. Am J Physiol Renal Physiol. 2005;289(1):F8‐F28. [DOI] [PubMed] [Google Scholar]
- 40. Kim CJ, Kaplan LE, Perwad F, et al. Vitamin D 1alpha-hydroxylase gene mutations in patients with 1alpha-hydroxylase deficiency. J Clin Endocrinol Metab. 2007;92(8):3177‐3182. [DOI] [PubMed] [Google Scholar]
- 41. Usatii M, Rousseau L, Demers C, et al. Parathyroid hormone fragments inhibit active hormone and hypocalcemia-induced 1,25(OH)2D synthesis. Kidney Int. 2007;72(11):1330‐1335. [DOI] [PubMed] [Google Scholar]
- 42. Cunningham J, Zehnder D. New vitamin D analogs and changing therapeutic paradigms. Kidney Int. 2011;79(7):702‐707. [DOI] [PubMed] [Google Scholar]
- 43. Bouchard J, Ouimet D, Vallee M, et al. Comparison of the prevalence of calcidiol insufficiency in predialysis and osteoporotic populations. Int Urol Nephrol. 2009;41(4):983‐988. [DOI] [PubMed] [Google Scholar]
- 44. Chonchol M, Scragg R. 25-Hydroxyvitamin D, insulin resistance, and kidney function in the Third National Health and Nutrition Examination Survey. Kidney Int. 2007;71(2):134‐139. [DOI] [PubMed] [Google Scholar]
- 45. Park S, Lee S, Kim Y, et al. Mendelian Randomization reveals causal effects of kidney function on various biochemical parameters. Commun Biol. 2022;5(1):713. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two-sample Mendelian randomization. Genet Epidemiol. 2016;40(7):597‐608. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
All codes used in the current study are available upon request to the corresponding author.
