Vitamin D Levels and Risk of Male Factor Infertility: A Mendelian Randomization Study

Abstract

Purpose

No consensus exists about the causal relationship between vitamin D (VD) and male factor infertility due to heterogeneity and confounding factors even in randomized controlled trials (RCTs). This study aimed to investigate the causal association between 25 hydroxyvitamin D (25OHD) levels and male factor infertility through Mendelian randomization (MR) and provide complementary information for optimization of future RCTs.

Materials and Methods

Two-sample MR analyses with four steps were performed. Single-nucleotide polymorphisms (SNPs) for VD were extracted from 417,580 Europeans in the UK Biobank, and the summary-level data of male factor infertility (825 cases and 85,722 controls) were extracted from the FinnGen.

Results

Totally 99 SNPs robustly associated with the 25OHD were included, and a 1-unit increase in genetically predicted natural-log transformed 25OHD levels was associated with decreased risk of male factor infertility (odds ratio [OR], 0.62; 95% confidence interval [CI], 0.44–0.89; p=0.010), which was consistent in all three sensitivity analyses (MR-Egger, weighted median, and weighted mode methods). The conclusion still stands after removing SNPs which explained more variation in the male factor infertility than the 25OHD (OR, 0.61; 95% CI, 0.42–0.88; p=0.009; n=62), and which were associated with confounders (body mass index, type 2 diabetes, smoking, and coronary artery diseases) of male factor infertility (OR, 0.58; 95% CI, 0.39–0.85; p=0.005; n=55).

Conclusions

VD supplement to increase serum 25OHD levels may be clinically beneficial for male factor infertility in the general population. The well-designed RCTs should be performed in priority to address this question.

INTRODUCTION

Male factor infertility (MFI) refers to the inability of a male to make a fertile female pregnant within at least 12 months of unprotected intercourse, which the incidence is at least 12% in the USA, bringing huge obstacles to social and economic progress [1]. Besides, since COVID-19 has been indicated somewhat reduced fertility and even led to infertility in some recovered males, it is more important to find potential effective prevention for male infertility [2].

In the last decades, the protective roles of vitamin D (VD) on human health have exploded in several diseases [3,4]. VD has been regarded closely associated with MFI, but the conclusions remained debatable in clinical observations [5,6,7,8,9,10,11,12]. Although observational studies have demonstrated positive effects of serum VD levels on seminal parameters and pregnancy outcomes [5,6,7,8,9], the current randomized controlled trials (RCTs) with VD supplements are plagued by significant heterogeneity in study populations, supplement dosage and potential confounders, thus no consensus exists about the causal relationship between VD and male factor fertility [10,11,12].

In theory, an RCT is the golden standard to answer whether increasing VD status through supplementation would improve MFI, but only under the premise of rigorous design, large sample, and sufficient duration [13]. Unfortunately, the current RCTs had limited sample sizes (n=62–269), a short follow-up time (12 weeks–150 days), and focused more on seminal parameters but less on pregnancy outcomes, which could reduce the chances of detecting a potentially beneficial effect. As an increasingly used analytical method, Mendelian randomization (MR) has been considered an ideal tool to optimize the design of subsequent randomized trials [14]. With genetic variants associated with exposure of interest as instrumental variables, MR can avoid unmeasured confounding from observational studies and investigate the causal relations between potentially modifiable risk factors and health outcomes [15]. Additionally, the effects of the genetic variants on the exposure are present since conception, which means MR can assess the effect of lifetime exposure of an exposure on the risk of an outcome [16]. Therefore, the MR can provide complementary information for RCTs, since the latter could only be conducted for a relatively short duration of time with a median duration of 40 months [14]. Besides, when RCTs produced unexpected null results, a larger than expected effect in MR analysis on the same topic might indicate the presence of an additional off-target harmful effect that counteract some of the beneficial on-target effect, which would be especially valuable to improve the further clinical trials [17].

The purpose of this study was to evaluate the causal effect of serum 25 hydroxyvitamin D (25OHD) levels (which is often used to assess VD status in the body) on male factor fertility with two-sample MR analyses in the general population and to provide references for the optimization of the future RCTs.

MATERIALS AND METHODS

This two-sample MR analysis was performed following the instructions of the STROBE-MR checklist with data from two independent databases (UK biobank [UKB] and the FinnGen research project). Sensitivity analyses and single-nucleotide polymorphism (SNP), filtering were conducted under the guidelines [18].

1. Data source

1) Vitamin D

The data of VD were collected from 417,580 Europeans based on UKB (filed ID: 30890; research resource identifier [RRID]: SCR_012815), in which genome-wide association study (GWAS) 143 independent loci were identified [19]. The VD levels were collected both at the initial assessment visit (2006–2010, n=448,271, mean=48.6132 nmol/L, standard deviation [SD]=21.1105 nmol/L) and a repeat the assessment visit (2012–2013, n=17,036, mean=47.7154 nmol/L, SD=21.9628 nmol/L). A chemiluminescent immunoassay (i.e., the Diasorin Liason^®; Diasorin, Saluggia, Italy) was applied for the quantitative determination of VD levels, which was represented by the total 25OHD concentration (i.e., 25OHD3 and 25OHD2) measured in blood samples. Finally, there were 417,580 participants included after applying the following exclusion criteria: (1) participants with non-European ancestry; (2) participants with 25OHD concentration less than 10 nmol/L or more than 375 nmol/L; (3) variants with genotype missingness more than 0.05; (4) variants with Hardy–Weinberg equilibrium test p-value more than 1×10^-5; (5) variants with minor allele frequency less than 0.01. Associations of VD with potential confounders were evaluated within covariates of age at the time of assessment, sex, body mass index (BMI), genotyping batch, assessment center, the month of testing, supplement intake, and first 40 ancestry principal components in this GWAS (Supplement Table 1). For the interpretation of all two-sample MR analyses, the MR estimates were expressed as every 1 unit change of serum 25OHD levels in results (corresponding to per natural-log nmol/L increase in genetically predicted natural-log transformed 25OHD levels), in which 1-SD was equal to 0.5 natural-log nmol/L [19]. Besides, each 1-SD increase in standardized natural-log transformed 25OHD levels has been estimated to be approximately 40.9 nmol/L increased serum 25OHD levels in VD–sufficient individuals (serum 25OHD level >70 nmol/L), 29.2 nmol/L in VD–insufficient individuals (50 nmol/L >serum 25OHD level >25 nmol/L), 14.6 nmol/L in vitamin D–deficient individuals (serum 25OHD level <25 nmol/L) [20].

2) Male factor infertility

The summary-level data on MFI (825 cases and 85,722 controls) were extracted from the FinnGen research project (RRID: SCR_022254), which combined genotype data from Finnish biobanks and digital health record data from Finnish health registries [21]. The condition of MFI was determined by doctor-diagnosed information by ICD-10 code N46, ICD-9, and ICD-8 codes 606 (including azoospermia, oligospermia, infertility due to extra testicular causes, and unspecified male infertility) from digital health record data. Males with reproductive organ cancer were excluded from the controls. The data on MFI from the FinnGen research project was not overlapped with that from UKB.

2. Selection of instrumental variables and Mendelian randomization assumptions

The MR analysis should meet three core assumptions to obtain unbiased estimates: (1) the variant is significantly associated with the exposure; (2) the variant is not associated with any confounder of the exposure-outcome association; (3) the variant does not affect the outcome, except possibly via other biological pathways (i.e., horizontal pleiotropic effect) with the exposure [18] (Fig. 1A, 1B). To guarantee the robustness of the MR analysis, we performed two-sample MR analyses with four steps (Fig. 1C). Firstly, we applied SNP as the genetic instrumental variable. To assure the extracted SNPs were robustly associated with the serum 25OHD levels (assumption 1), we only included genome-wide significant associated (p<5×10^-8) index SNPs with F statistic >10 to minimize the possibility of weak instrumental variable bias among participants of Europeanancestry in UKB. Secondly, we removed variants in potential linkage disequilibrium (LD) with r²≥0.01 and LD distance <10,000 kb and harmonized the datasets, after which we performed the first MR. Thirdly, we additionally removed the 25OHD-associated SNPs that are also genome-wide significant associated (p<5×10^-8) with MFI and applied the Steiger filter to remove SNPs with a larger R-Squared in MFI than 25OHD (assumptions 2 & 3). Then we performed the second MR analysis. Finally, we exclude SNPs that were associated at a genome-wide significance level (p<5×10^-8) with any of the following confounders of MFI: BMI, type 2 diabetes, smoking, and coronary artery diseases, and we performed the third MR analysis (assumptions 2 & 3) [22,23,24]. Besides, we additionally applied 3 MR methods (MR-Egger, weighted median, and weighted mode methods) with relatively robust estimates to horizontal pleiotropy as sensitivity analyses, which can determine how robust MR results are to the assumption 2 & 3 that genetic variants have no pleiotropic effects on the outcome under different alternative methods [18]. The details about the confounders were available in Supplement Table 2.

Fig. 1. (A) Schematic diagram of the principles of MR. (B) Study flow graph. The dashed lines refer to potential directional pleiotropic and direct causal effects which could violate MR assumptions. (C) The procedures of MR analysis. BMI: body mass index, CAD: coronary artery disease, IVW: the inverse variance weighted method, MR, Mendelian randomization, SNP: single-nucleotide polymorphism, T2D: type 2 diabetes.

3. Statistical methods of Mendelian randomization analysis

In these serial two-sample MR analyses, the inverse variance weighted (IVW) random effects method, a meta-analysis of the variant-specific Wald ratios for each variant (i.e., the β coefficient of the SNP for MFI divides by the β coefficient of the SNP for 25OHD), was used to provide a combined estimate of the causal estimate from each SNP for each direction of potential influence. The IVW method assumes that the genetic variants are independent of each other and are valid instrumental variables [25]. However, it can ignore the other risk factors’ mediated effects or potential pleiotropic, and bias which can influence the outcome of interest via causal pathways besides the exposure could occur when horizontal pleiotropy exists among the instrument SNPs, leading to the violence of instrumental variable assumptions of MR [26]. Therefore, we additionally applied MR-Egger, weighted median, and weighted mode methods (Fig. 1B) [18]. The MR-Egger method presumes that the magnitude of the pleiotropic effect is not associated with the strength of the association between the genetic variant and the phenotype of interest among all the instruments [26]. The weighted mode requires the largest subset of instrumental variables that recognize the same causal effect to be valid. In contrast, the weighted median requires 50% of the weight assigned to variables from valid instruments [27]. These additional analyses can determine how robust MR results are to the assumption that genetic variants have no pleiotropic effects on the outcome under different alternative specifications. To detect the possible heterogeneities, we used the Cochran Q statistic and the p-value <0.05 was regarded as significant heterogeneity. Besides, the intercept obtained from the MR Egger analysis was applied to evaluate the horizontal pleiotropy (Fig. 1B), in which a p-value <0.05 indicated the presence of horizontal pleiotropy. We conducted Power calculation via an online tool (https://cnsgenomics.com/shiny/mRnd/; RRID: SCR_022156) according to the analytical guidelines (Supplement Table 3). All analyses were performed in R 3.6.0 using the TwoSampleMR package (https://github.com/MRCIEU/TwoSampleMR; RRID: SCR_019010). Besides, we reported all the estimates with two-tailed p-values. The suggestive causal associations were regarded as p<0.05 in IVW models and the direction of the association remained consistent in the other three models (MR-Egger, weighted median, and weighted mode).

RESULTS

We initially included 143 genome-wide significant associated (p<5×10^-8) index SNPs. The F-statistic value is all >10 in every filtering step (minimum: 61.98; Supplement Table 3), indicating strong instrumental variables. After removing variants in potential LD, there were 99 SNPs included in the first MR analysis. There were a total of 37 SNPs excluded before the second MR analysis since they were found genome-wide significant associated (p<5×10^-8) with MFI and had a larger R-Squared in MFI than 25OHD via Steiger filter, which means these SNPs explained more variation in the MFI than the 25OHD. Finally, the third MR analysis was performed after filtering 7 SNPs associated with confounders of MFI.

At the first MR, the results supported that each 1-unit increase in genetically predicted natural-log transformed 25OHD levels can significantly decrease the risk of MFI up to 38% (IVW: odds ratio [OR], 0.62; 95% confidence interval [CI], 0.44–0.89; p=0.010) (Fig. 2A). The conclusion remained consistent in all three sensitivity analyses (MR Egger: OR, 0.53; 95% CI, 0.33–0.85; p=0.010; weighted median: OR, 0.56; 95% CI, 0.33–0.95; p=0.031; weighted mode: OR, 0.54; 95% CI, 0.35–0.83; p=0.005) (Table 1). There was limited evidence that could prove the existence of heterogeneity by Cochran Q statistic (p=0.78) and horizontal pleiotropy by analysis of MR-egger intercept (p=0.30) (Supplement Table 4).

Fig. 2. (A, B) The scatter plots show the causal effect of each 1-unit change in vitamin D (serum 25OHD) on male factor infertility in (A) the first MR analysis; (B) the second MR analysis; (C) the third analysis. These results suggest that each 1-unit increase in genetically predicted natural-log transformed 25OHD levels can significantly decrease the risk of male factor infertility. The consistent results in sensitivity analyses (MR-Egger, weighted median, and weighted mode methods) suggested little bias in the IVW estimates. (A) Heterogeneity Q-value for IVW=86.97 (p=0.78); (B) heterogeneity Q-value for IVW=14.01 (p=0.99); (C) heterogeneity Q-value for IVW=11.15 (p=0.99). (D) The three-stage analysis results of associations of vitamin D in the MR analysis with male factor infertility. CI: confidence interval, IVW: the inverse variance weighted method, MR, Mendelian randomization, OR: odds ratio, SNP: single-nucleotide polymorphism, 25OHD: 25 hydroxyvitamin D.

Table 1. Results of two-sample MR.

Exposure		Outcome	SNPa	MR method	OR	95% LCI	95% UCI	p-value
The first MR
	Vitamin D	Male factor infertility	99	IVW	0.62	0.44	0.89	0.010
				MR Egger	0.53	0.33	0.85	0.010
				Weighted median	0.56	0.33	0.95	0.031
				Weighted mode	0.54	0.35	0.83	0.005
The second MR
	Vitamin D	Male factor infertility	62	IVW	0.61	0.42	0.88	0.009
				MR Egger	0.56	0.35	0.90	0.021
				Weighted median	0.56	0.33	0.94	0.028
				Weighted mode	0.55	0.35	0.85	0.009
The third MR
	Vitamin D	Male factor infertility	55	IVW	0.58	0.39	0.85	0.005
				MR Egger	0.56	0.34	0.90	0.021
				Weighted median	0.56	0.33	0.94	0.028
				Weighted mode	0.55	0.35	0.86	0.011

MR: Mendelian randomization, SNP: single-nucleotide polymorphism, OR: odds ratio, LCI: lower confidence interval, UCI: upper confidence interval, IVW: inverse variance weighted.

^aNumber of SNPs retained for this analysis.

At the second MR, increased 25OHD levels were also found associated with decreased risk of MFI for approximately 39% (IVW: OR, 0.61; 95% CI, 0.42–0.88; p=0.009; n=62) (Fig. 2B, Table 1). All three sensitivity analyses provided results similar to those of IVW. No evidence of heterogeneity (p=0.99) or horizontal pleiotropy (p=0.60) were found in the second analysis (Supplement Table 4). At the third MR, a stronger protective effect of 25OHD with statistical significance was demonstrated after excluding SNPs associated with confounders of MFI (IVW: OR, 0.58; 95% CI, 0.39–0.85; p=0.005; n=55) and similar results were also found in all three sensitivity analyses (Fig. 2C, Table 1). Neither heterogeneity (p=0.99) nor horizontal pleiotropy (p=0.80) was indicated (Supplement Table 4).

DISCUSSION

In this two-sample MR analysis with data from large sample sizes (VD: n=417,580; MFI: 825 cases and 85,722 controls) in two independent databases, the results provided evidence that each 1 unit increase in genetically predicted natural-log transformed 25OHD levels confers a decrease in MFI, which were consistent and robust in all three sensitivity analyses. Therefore, our study indicated increased serum 25OHD levels through extra VD intake in the general population may help reduce the incidence of MFI. Besides, these results provided complementary information for the design and implementation of future RCTs.

To our acknowledge, there were no previous MR analyses evaluating the causal effect of VD on male factor fertility in the general population. The results of this MR study support the protective role of VD in MFI, which complements the previous RCTs. Blomberg Jensen’s team found no significant improvement in sperm parameters in VD deficient men (<25 nmol/L, n=66) treated with VD [12]. Although spontaneous pregnancies and live birth rates tended to be higher in the overall treatment group (n=269) but with no statistical significance. The different results between Blomberg Jensen et al’s [12] and our study can be explained by the relatively small sample size included in the final analysis and short intervention period (the truly effective intervention period could be less than 150 days after serum 25OHD reached an effective concentration). Another study indicated VD intervention for 3 months can improve sperm motility in men with asthenozoospermia and serum 25OHD <30 ng/mL (n=86), but no pregnancy outcomes were reported [10,28]. Reversely, the study conducted by Amini et al [11] showed no association between VD supplement and MFI, but they included fewer participants (n=62) excluding patients with azoospermia for a short intervention (12 weeks). After all, significant heterogeneity existed among the interventional studies about the intervention period, the dosage of VD supplement, elevated level of serum 25OHD level after treatment, population studied, and primary outcomes, which led to no consensus about the role of VD in male reproduction.

Compared to the above RCTs, our study included genetic variants from the largest sample size of participants with MFI (n=825), to reduce the imbalances in the randomization resulting from a small sample size of the previous RCTs [14]. Moreover, our MR analyses represent the results of a significantly longer intervention of VD supplement since RCTs only estimate the effect of a particular intervention over the timeframe of the study [18]. Besides, the definition of MFI in our study (including azoospermia, oligospermia, infertility due to extra testicular causes, and unspecified male infertility) is comprehensive, which can guarantee the generalizability of our results.

Theoretically, VD can have an impact on the progression of MFI since VD receptors have been found in the male reproductive system, and the potential protective role of VD on MFI has been demonstrated by in vitro studies [29]. In idiopathic infertility men with VD deficiency, the methylation of the VDR gene was found significantly higher, which was negatively associated with sperm concentration and progressive motility [30]. Further, another study demonstrated that incubation with 1,25(OH)2D for 30 minutes could enhance sperm motility, probably by promoting the synthesis of adenosine triphosphate through the cAMP/PKA pathway [5]. Recently, the role of oxidative stress markers in VD and MFI has received widespread attention. Sperm are susceptible to oxidative stress due to their reduced inherent antioxidant defenses and DNA repair mechanisms, while VD has a potential scavenger capacity as a membrane antioxidant. Shahid and colleagues found that 4-Hydroxynonenal, an oxidative stress marker, was significantly elevated when sperm parameters were altered, and was inversely correlated with VD in a cross-sectional study [31]. As a second messenger of reactive oxygen species, 4-Hydroxynonenal has also been reported a strong negative correlation with sperm motility and morphology [32], but would be reduced when treated with VD [33] in other publications. Besides, an RCT conducted in asthenozoospermia infertile males demonstrated that VD supplementation increased total antioxidant capacity levels and decreased Malondialdehyde levels (an indicator of lipid peroxidation) in serum and semen, as well as the total and progressive sperm motility compared to the placebo group [28]. These results indicated the possible mechanism of VD’s effect on MFI may be reducing oxidative stress.

RCTs and MR studies could answer different but complementary questions. The clinical significance of our MR study is that we provided information for the optimization of future RCTs in the following aspects.

Firstly, it is necessary and worthy to continue RCTs for determining the role of supplement VD on MFI. Our results have demonstrated a lifetime (including intrauterine) exposure to VD indeed has a causal effect on the risk of MFI, which provided a solid theoretical basis for conducting similar RCTs [17].

Secondly, our study can help determine how much the serum 25OHD levels must be changed to produce the important clinical difference in the male factor fertility. Our results indicated that each 1 unit increase in genetically predicted natural-log transformed 25OHD levels confers a significant decrease in MFI, in which 1-SD was 0.5 natural-log nmol/L (i.e., 14.6–40.9 nmol/L as described in the method section) and 5.3 nmol/L increase in 25OHD levels conferred by taking daily 100 IU of cholecalciferol [19,20]. Therefore, our findings emphasize the importance of closely monitoring the change of serum 25OHD levels in the intervention and control groups in the subsequent RCTs. Besides, it is necessary to determine the time point that the serum 25OHD levels were significantly different (i.e., 14.6 nmol/L to 40.9 nmol/L) compared with the baseline level, as well as between the control group and the experimental group, and whether the difference can be maintained stably after this time-point.

Thirdly, our results suggest that future RCTs should consider a longer effective intervention of VD supplement than the present ones. Since MFI would be diagnosed after at least 12 months of unprotected intercourse according to the definition, it somewhat requires a relatively longer intervention and follow-up. Besides, due to the principles of MR analysis, our results support the protective causal effect of long-term serum 25OHD elevation on MFI [14]. Therefore, our study supports that prolonging the intervention time of VD in future RCTs may lead to a greater chance of detection of the potential beneficial effect of VD on MFI.

Lastly, after a comprehensive comparison with existing interventional studies, our results support the existence of an additional off-target harmful effect that can offset the beneficial effect of VD on MFI [17]. The off-target effect may be the detrimental effects of hypervitaminosis D on spermatogenesis because the subsequent hypercalcemia could damage cytoplasmic activity at the mitochondrial level [34]. Unfortunately, because the MR assumes a linear causal association between the exposure of interest and outcome, despite our results supporting an SD increase of serum 25OHD levels would decrease the risk, we provided limited evidence for the proper dosage of VD in the improvement of male infertility. Therefore, future RCTs should focus on the most appropriate dose of serum 25OHD.

The following limitations should be taken into consideration when interpreting our results. Firstly, the GWAS for VD did not provide sex-stratified data, which may lead to potential bias [19]. Secondly, the MR assumes a linear association between VD and MFI, while the potential nonlinear association requires further investigation. Thirdly, detailed data including characteristic and semen analysis parameters were not available in the FinnGen research project. Finally, this MR analysis was applied to European ancestry. More studies are warranted to find whether the conclusions of our research apply to other ancestries.

CONCLUSIONS

Our study provided genetic evidence to support increased levels of serum 25OHD in the general population as a causal protective factor for MFI. These results raise the possibility that elevating serum 25OHD level via VD supplement is a potentially effective way to treat or prevent MFI, which provided complementary information for the optimization of future RCTs.

Data Sharing Statement

The data used in this study were publicly available and can be accessed via the following links:

• 1) The genetic instruments come from the published GWAS study as described in the method section.

• 2) The GWAS summary statistics for the male factor infertility are available at the FinnGen research project (https://finngen.gitbook.io/documentation/). The RRID for the Finn-Gen research project is SCR_022254.

• 3) Analyses are performed using the TwoSampleMR package and the code is available on the website (https://mrcieu.github.io/TwoSampleMR/reference/index.html). The RRID for the TwoSampleMR package is SCR_019010.

Supplementary Materials

Supplementary materials can be found via https://doi.org/10.5534/wjmh.220109.

Supplement Table 1

Summary information of vitamin D data

Supplement Table 2

Summary information of used summary-level data study

Supplement Table 3

Statistical power to detect the difference in VD

Supplement Table 4

Results of heterogeneity and MR Egger-intercept