Maternal vitamin D levels and male reproductive health: a population-based follow-up study

Abstract

Maternal vitamin D levels during pregnancy may be important for reproductive health in male offspring by regulating cell proliferation and differentiation during development. We conducted a follow-up study of 827 young men from the Fetal Programming of Semen Quality (FEPOS) cohort, nested in the Danish National Birth Cohort to investigate if maternal vitamin D levels were associated with measures of reproductive health in adult sons. These included semen characteristics, testes volume, and reproductive hormone levels and were analysed according to maternal vitamin D (25(OH)D₃) levels during pregnancy. In addition, an instrumental variable analysis using seasonality in sun exposure as an instrument for maternal vitamin D levels was conducted. We found that sons of mothers with vitamin D levels < 25 nmol/L had 11% (95% CI − 19 to − 2) lower testes volume and a 1.4 (95% CI 1.0 to 1.9) times higher risk of having low testes volume (< 15 mL), in addition to 20% (95% CI − 40 to 9) lower total sperm count and a 1.6 (95% CI 0.9 to 2.9) times higher risk of having a low total sperm count (< 39 million) compared with sons of mothers with vitamin D levels > 75 nmol/L. Continuous models, spline plots and an instrumental variable analysis supported these findings. Low maternal vitamin D levels were associated with lower testes volume and lower total sperm count with indications of dose-dependency. Maternal vitamin D level above 75 nmol/L during pregnancy may be beneficial for testes function in adult sons.

Introduction

Vitamin D deficiency constitutes a major public health problem, both in the general population and among pregnant women [1–3]. There is currently no global consensus about the sufficient level of vitamin D, and recommendations range from 50 to 75 nmol/L [2]. The Danish Society of Obstetrics and Gynecology recommends pregnant women to maintain a plasma level of 25(OH)D at 75 nmol/L or higher [4]. However, more than half of all pregnant women in Denmark may have vitamin D levels below this recommended level [4] and this tendency is similar worldwide [1, 3].

The developing foetus relies on the mother for an adequate supply of vitamin D [3]. Low maternal vitamin D levels may affect foetal development and offspring long-term health, including skeletal, metabolic, autoimmune and neuro-developmental effects [3, 5]. Male reproductive health might be sensitive to maternal vitamin D levels during pregnancy, since the function of the hypothalamic–pituitary–gonadal (HPG) axis and reproductive organ development in the foetus are particularly vulnerable to interferences [6, 7]. The vitamin D receptor and vitamin D activating enzymes are present throughout the HPG axis [8–10], and vitamin D is metabolized in the testes during foetal development [10–13]. In the developing gonad, circulating 25(OH)D is locally activated to 1.25(OH)₂D by the enzyme CYP27B1. Following activation, 1.25(OH)₂D binds to the vitamin D receptor in foetal Leydig and Sertoli cells [11, 12] and partakes in regulation of cell proliferation, differentiation and apoptosis [14]. It has been suggested that vitamin D affects male reproductive hormone production due to the metabolism of vitamin D in Leydig cells [11], and plays a vital role in reproductive organ development and later spermatogenesis, due to the metabolism of vitamin D in foetal Sertoli and germ cells [12].

Epidemiological studies in adult males suggested that low vitamin D levels might impair their reproductive health [11, 12, 15]. Rodent studies have supported this finding and suggest that low vitamin D plasma levels may down-regulate vitamin D receptor signalling and thereby inhibit germ cell proliferation [16, 17]. Vitamin D receptor knockout mice presented with histological alterations in the testes due to increased apoptosis and reduced proliferation, resulting in impairment of the spermatogenesis [16]. Moreover, testes weight and total sperm count were reduced in mice receiving a vitamin D depleted diet for 10 weeks starting at birth [17]. No published studies have investigated the association between maternal vitamin D levels and male reproductive health.

Poor male reproductive health is of concern in Western countries [7], highlighting the importance of investigating potentially modifiable determinants for poor reproductive health such as low vitamin D levels in pregnancy. Within a triangulation framework using different methodological approaches with presumable different sources of bias, we aimed to investigate whether maternal vitamin D plasma levels during pregnancy were associated with semen characteristics, testes volume, and reproductive hormones in young adult male offspring.

Methods

This follow-up study is based on the Fetal Programming of Semen Quality (FEPOS) cohort [18], nested within the Danish National Birth Cohort (DNBC) [19]. From 1996 to 2002, Danish speaking pregnant women were recruited to the DNBC at the first antenatal visit to the general practitioners; approximately 92,000 women were enrolled in the DNBC (participation rate 60%). The pregnant women were primarily of Caucasian origin. In approximately gestational week (GW) 16, the pregnant women provided information on health behaviour and medical history by a computer-assisted telephone interview. In GW 8 and 25, the women were asked to give a blood sample. The plasma was stored at − 80 °C in the Danish National Biobank, Copenhagen, Denmark [18, 19].

In 2017, FEPOS was established as a sub-cohort within the DNBC. Male offspring were eligible for invitation if their mothers had completed the first and second interviews in the DNBC and had a blood sample stored. Between March 2017 and December 2019, 5697 adult sons aged at least 18 years and 9 months and living in the area of Aarhus or Copenhagen were randomly and consecutively invited among the 21,623 eligible for participation in FEPOS. The sons were asked to decline participation if they had undergone sterilization, orchidectomy, or chemotherapy or had only one or no testis in the scrotum. After providing informed consent, the sons participated in a clinical examination and provided a semen and a blood sample. In total, 1058 sons participated (19%). As some cryotubes in the biobank contained too little plasma to measure maternal vitamin D levels (n = 188), and as some gestational blood samples were obtained in GW 25 only (n = 43), the final study population consisted of 827 sons (14.5%) with available information on both first trimester maternal vitamin D levels and measures of male reproductive health (Fig. 1). Further details are described in the cohort profile paper [18].

Fig. 1.

Flowchart on the inclusion of participants in the Fetal Programming of Semen Quality (FEPOS) cohort, nested within the Danish National Birth Cohort, Denmark, 1998–2019

Maternal vitamin D

Maternal vitamin D levels were measured as plasm 25-hydroxyvitamin D₃ (25(OH)D₃) in the first trimester blood samples. Standard solutions were prepared in acetonitrile using 25-hydroxyvitamin D₃ and the internal standard D6-25-hydroxyvitamin D₃ (Toronto Research Chemicals, North York, ON, Canada). Two reference samples prepared in-house were included for quality control (QC1 and QC2) by pooling serum samples; two QC samples from Chromsystems Instruments & Chemicals GmbH (Mass-Check; Gräfelfing, Germany) were included in all analytical batches.

Aliquots of 0.1 mL of plasma sample were added with isotopically labelled internal standards and 0.2 mL of acetonitrile to precipitate the proteins and were thereafter centrifuged. The supernatant was analysed using a triple quadrupole linear ion trap mass spectrometer (MS; QTRAP6500+, AB Sciex, Framingham, MA, USA) coupled to a liquid chromatography system (UFLCXR, Shimadzu Corporation, Kyoto, Japan; LC/MS/MS). Separation was performed using a two-dimensional system with Reprosil Gold (C4; 3 µm 20 × 4.6 mm, Dr Maisch) and Raptor Biphenyl (2.7-µm 100 × 4.6 mm; Restek). The mobile phases used was Milli-Q water and acetonitrile with 0.08% (v/v) formic acid. The MS analyses were carried out using selected reaction monitoring in positive ion mode. The m/z quantified transitions used were 383.3–297.0 for 25(OH)D₃ and 389.4–303.1 for the internal standard D6–25(OH)D₃.

The limit of detection (LOD) was 0.5 ng/mL and no samples were below the LOD. The mean levels of the two in-house prepared reference samples were 31 ng/mL for QC1 with a coefficient of variation (CV) of 8%. QC2 had a mean level of 34 ng/mL and a CV of 7%. For the two QC samples from Chromsystems, QC Level I was 14.5 ng/mL with a CV of 10%, (target value 16.7, range 13.3–20.0 ng/mL) and for the QC Level II it was 33.5 ng/mL with a CV of 8%, (target value 37.7, range 30.2–45.3 ng/mL). The 25(OH) D₃ concentrations were recalculated to nmol/L by multiplying with 2.496.

Male reproductive health

Male reproductive health included semen characteristics: semen volume, concentration, total sperm count, progressive motility, and normal morphology, in addition to testes volume and levels of reproductive hormones in plasma: Testosterone, oestradiol, sex hormone-binding globulin (SHBG), follicle stimulating hormone (FSH), luteinizing hormone (LH), free androgen index (calculated as (testosterone/SHBG) × 100), and testosterone/oestradiol ratio (calculated as free testosterone/oestradiol).

The clinical examinations took place in one of two associated clinics, at Department of Occupational Medicine at Aarhus University Hospital in Aarhus or at Department of Occupational and Environmental Medicine at Bispebjerg Hospital in Copenhagen. All semen quality measures were analysed according to World Health Organization 2010 recommendations [20]. Participants were instructed to be sexually abstinent for 48–72 h prior to sample collection. In total, 290 men (35%) had an abstinence time shorter than 48 h, and 267 men (32%) had an abstinence time longer than 72 h. The men were included in the study regardless of abstinence time, since this was taken into account in the statistical analyses. Semen samples were collected at home or at the research facility. Semen volume was derived by weighing the pre-weighted sample container. One of two experienced biomedical laboratory technicians in Aarhus or Copenhagen manually assessed motility and concentration in duplicates after liquefaction. Sperm concentration was calculated based on standard sperm dilution and manual counting. Total sperm count was calculated by multiplication of concentration and volume. Motility was assessed as proportions of progressive (PR), non-progressive (NP) and immotile (IM) spermatozoa. The proportion of strict morphologically normal spermatozoa was analysed at the Centre of Reproductive Medicine in Malmö, Sweden. The methods were continuously quality controlled and met given standards for semen quality measures [18].

Testes volume was self-measured using a Prader Orchidometer. In a validation study performed by our group, this method was valid compared with measurements performed by an experienced examiner [21].

Reproductive hormones were measured in non-fasting venous blood samples collected at the clinical examination and stored at − 80 °C until analysis at Department of Clinical Biochemistry at Aarhus University Hospital, Denmark. SHBG, FSH and LH were measured using immunoassays (Cobas^® 8000 e602; Roche Diagnostics, Mannheim, Germany) with CVs of 2.5–2.8%, 0.7–1.2% and 1.1–1.7%, respectively. LODs were 0.1 IU/L for FSH and LH, and 0.350 nmol/L for SHBG. Levels of testosterone and oestradiol were analyzed using liquid chromatography-tandem mass spectrometry (LC–MS/MS). LODs were 0.12 nmol/L for testosterone and 15 pmol/L for oestradiol [2]. Only few values were below LOD (0.1% for FSH and LH and 7.6% for oestradiol). These values were replaced by LOD/√2.

Covariates

We identified potential confounders a priori using existing literature and directed acyclic graphs [22]. We included socioeconomic status (highest of the parents), couple fecundity (including time to pregnancy and medically assisted reproduction), maternal first trimester smoking, pre-pregnancy body mass index (BMI), and age at delivery, in addition to season of the sons’ clinical visit. The latter was included to account for the potential association between season of birth, which is highly associated with maternal vitamin D levels, and season of clinical visit, because the sons were invited at the age of 18 years and 9 months (Online Resource: Supplementary Fig. 1). By use of the DNBC, which consisted primarily of Caucasian women and their offspring, ethnicity was controlled for by design. Information on maternal smoking, BMI, couple fecundity and socioeconomic status was obtained from the first trimester DNBC interview. Socioeconomic status was defined according to occupation and level of education derived from the Danish International Standard Class of Occupation and Education codes (ISCO-88 and ISCED). Information on age at delivery was obtained from the Danish Medical Birth Registry. Season of semen and blood sampling was registered at the clinical visit.

Statistical analysis

We categorised maternal vitamin D levels according to clinically relevant categories [4, 23], i.e. < 25 nmol/L (severe deficiency), 25–50 nmol/L (deficiency), 50–75 nmol/L (insufficiency), > 75 nmol/L (recommended minimum level) in line with previous research [11]. We also estimated the linear association per 10 nmol/L decrease in vitamin D to avoid introducing arbitrary cut-offs and to examine potential dose-dependency. We estimated a restricted cubic spline with three knots (at 10th percentile: 26.3 nmol/L, 50th percentile: 54.7 nmol/L, 90th percentile: 83.7 nmol/L), since the concentration at which vitamin D exhibits its function on reproductive tissues remains unknown [1]. Furthermore, vitamin D may act in a non-linear way with potential threshold, where both too low and too high levels of vitamin D may adversely interfere with foetal development of reproductive organs [24].

Male reproductive health measures were analysed continuously using a multivariable negative binomial regression model fitted by maximum likelihood estimation (STATA’s -nbreg- package) according to maternal Vitamin D levels. Ratios were estimated with 95% confidence intervals (CI) according to exposure groups, and relative percentage differences were then calculated: (ratio − 1) × 100%. Moreover, testes volume (< 15 mL) and semen characteristics were dichotomized according to WHO’s lower reference limits [20] to support clinical interpretation. These dichotomized variables were analysed using multivariable log-binomial regression to calculate adjusted risk ratios of low semen characteristics and low testes volume according to maternal vitamin D levels.

In addition to the potential confounders, we included variables expected to be strongly associated with the outcomes in the multivariable models to improve precision of the estimates. In models examining semen characteristics, we included abstinence time (in days), place of semen sample collection (at home or the clinic), and spillage of semen sample (yes, no); participants reporting spillage during sample collection (n = 150) were excluded from the models examining volume and total sperm count. Interval from ejaculation to analysis (in minutes) was further included in models examining motility. Though progressive motility was of main interest, we modelled non-progressive + immotile in percentage in the multivariable negative binomial regression analyses to ensure optimal model fit. In models examining testes volume, which was analysed as the average volume of the two testes, abstinence time was further included. In models examining reproductive hormones, time of the day of blood sampling (morning/afternoon/evening) was included. Information on all precision variables was recorded at the clinical visit. All continuous variables were modelled as second-order polynomials to allow for non-linearity. Complete case analyses were conducted due to the limited number of missing data on covariates.

Selection weights were estimated and applied to all models to account for potential selection bias due to selective non-participation [25]. In short, we fitted a logistic model with participation and first trimester gestational blood sample available (yes, no) as the dependent variable and the potential confounders in addition to information on intake of a prenatal vitamin D supplement as explanatory variables for participation. The latter was applied as a proxy for vitamin D levels, since this information was only available for participants. Given that all explanatory variables associated with participation were included in the calculation, and given that there was no residual confounding, we created a pseudo-population representative of the entire source population of invited young men by re-weighing all 827 participants with the selection weights [25, 26]. Due to the use of selection weights and clustering of siblings, robust standard errors were applied.

The negative binomial regression models were checked with Q–Q plots comparing the observed distributions and the model-based distributions. Further, standardized deviance residuals were plotted against model-based predictions. The model fit was considered acceptable. Data management and statistical analyses were conducted in STATA 17.0 (Statacorp, College Station, TX). All percentiles were reported as pseudo percentiles, calculated as the mean of the five values closest to the actual percentile according to local regulations (GDPR, Regulation (EU), 2016/679 of 25 May 2018).

Instrumental variable analysis

Triangulation of study results is a method to strengthen causal interpretation and represents the practice of investigating the same research question using different methods with presumably different and unrelated sources of bias [27]. If the results are comparable across these different methods, confidence in the results increase. An instrumental variable analysis allows application of a causal interpretation of results from observational data and makes it possible to obtain an unbiased estimate despite potential exposure-outcome confounding [28, 29]. The instrumental variable analysis is performed using a different set of strong assumptions compared with multivariable regression analysis using the backdoor criterion for confounder selection. In a triangulation framework, the instrumental variable analysis is therefore suitable for comparison with the main multivariable regression analyses. To strengthen causal interpretation, we thus conducted an instrumental variable analysis using seasonality in sun exposure as an instrumental variable for maternal vitamin D levels.

Seasonality in sun exposure may be an appropriate instrument for vitamin D exposure [30], since the synthetization of vitamin D₃ from 7-dehydrocolesterol in the human skin following exposure to sunlight constitutes the principal source of vitamin D [31]. The causal framework underlying the analysis is depicted in supplementary Fig. 2 (Online Resource (Supplementary Fig. 2)). The instrumental variable analysis was conducted under the core assumptions of relevance (the instrument should be strongly associated with the exposure under study), exchangeability (the instrument should be marginally or conditionally independent on unmeasured co-variates) and the exclusion restriction criteria (the instrument should be associated with the outcome under study only through its effect on the exposure) [29]. To make interferences about effect sizes and to quantify the potential causal association, a fourth identifying assumption [32, 33], the monotonicity assumption, must be applied. It refers to the condition that no study participants defied responding to the effect of the instrumental variable [29, 33]. The assumptions underlying the instrumental variable analysis are discussed in detail in the Online Resource (Supplementary text 1).

Month at GW 8 + 0 was regarded a suitable instrumental variable, since a programming window for male reproductive health at around GW 8–14 has been suggested [34]. We estimated the month where the mother was in GW 8 + 0, by subtracting the gestational age at delivery + 8 weeks from the sons’ date of birth. Information on date of birth and gestational age at delivery was obtained from the Danish Medical Birth Register [35]. In less than five participants, information on gestational age was not available, and calculations were performed assuming a gestational age of 40 + 0 weeks at delivery.

Distribution of baseline characteristics, precision variables and measures of male reproductive health were presented according to season (winter: December–February; spring: March–May; summer: June–August; and autumn: September–November) of GW 8 + 0. Associations were analysed using two-stage least square regression models (STATA’s -ivregress 2sls- package). First stage statistics were obtained and presented. This included r², a measure of the proportion of the variance in the first stage model explained by the instrumental variable, and the partial F statistics, a measure of the variance and the study size to test if the beta-coefficient for the effect of the instrumental variable on the exposure is zero [29].

The two-stage least square regression model requires linearity between the instrumental variable and the exposure and between the exposure and the outcome to ensure model fit. Month at GW 8 + 0 was associated with vitamin D levels in a cyclic modality (Online Resource (Supplementary Fig. 3)). To approach a linear association between month and vitamin D levels, month at GW 8 + 0 was recoded and modelled as a continuous variable based on the predicted vitamin D levels for each month at GW 8 + 0, using an ordinary least square regression model. Vitamin D was kept continuous and modelled in 10 nmol/L units. Further, all outcomes were log-transformed to approximate the normal distribution and a linear association with vitamin D. This providesdrelative differences in measures of male reproductive health and may both ease interpretation and ensure that the results were directly comparable with our main results. To avoid the exclusion of sons with azoospermia when log-transforming the outcomes, we added different small values (0.1, 0.5, 1.0 and 2.0) to all zero values and ran ordinary least square regression models with vitamin D as the explanatory variable. We checked all models by Q–Q plots of residuals and by plotting the residuals against the expected values to ensure modelling assumptions were fulfilled. Adding 1.0 to all zero values yielded the best fit and narrowed the 95% CIs considerably compared with other small values; all analyses thus followed this strategy. To improve precision, we included the same baseline characteristics and precision variables in all instrumental variable models as in the main models. The same selection weights were also applied and all models were fitted with robust standard errors. Estimates were back-transformed and presented with 95% CIs.

Sensitivity analyses

To assess the robustness of the results with regard to other environmental exposures possibly affecting male reproductive health and possibly associated with vitamin D levels or fluctuating with season, we additionally adjusted all models for perfluorooctane sulfonic acid (PFOS). PFOS is a chemical substance that accumulates in the human body and readily crosses the placenta. We have previously found associations between higher maternal PFOS exposure and impaired semen characteristics in the FEPOS cohort [36]. PFOS was measured in the first trimester blood samples using triple quadrupole linear ion trap mass spectrometers equipped with TurboIonSpray sources (QTRAP 5500; AB Sciex) coupled to a liquid chromatography/tandem mass spectrometry (LC–MS/MS) system (UFLCXR; Shimadzu Corp) as described in Hærvig et al. [36].

In another sensitivity analysis, we conducted the instrumental variable analysis using the generalized method of moments (GMM) estimator (STATA’s -ivregress gmm-package). Using this estimator, we may obtain valid estimates under less strict modelling assumptions compared to the two-stage least square method [37].

Results

The median age of participating male offspring was 19 years and 2 months (range 18 years and 9 months to 21 years and 2 months). Median maternal vitamin D level was 55 nmol/L (range 9–139 nmol/L). Mothers of 73 sons (9%) had vitamin D levels < 25 nmol/L, mothers of 270 sons (33%) had levels of 25–50 nmol/L, mothers of 330 sons (40%) had levels of 50–75 nmol/L, and mothers of 154 sons (19%) had vitamin D levels > 75 nmol/L. Mothers with the lowest vitamin D levels were more likely to be overweight or obese, to have smoked during the first trimester, as well as had a lower parental socioeconomic status and a shorter time to pregnancy (Table 1).

Table 1.

Baseline and clinical characteristics in addition to measures of male reproductive health according to categorised vitamin D concentrations (nmol/L) in 827 participants from the Fetal Programming of Semen Quality (FEPOS) Cohort, 1998–2019, Denmark

	Maternal vitamin D (25(OH)D3)
	< 25 nmol/L		25–50 nmol/L		50–75 nmol/L		75–150 nmol/L		Missings
	No	%	No	%	No	%	No	%	No	%
	73	9	270	33	330	40	154	19
Vitamin Da	21 (10 to 25)		40 (27 to 49)		61 (51 to 74)		85 (75 to 120)
Baseline characteristics
Socioeconomic status (highest of the parents)									0	0
High-grade professional	19	26	91	34	122	37	51	33
Low-grade professional	25	34	87	32	107	32	55	36
Skilled or unskilled worker	>24b	>33b	83	31	87	26	38	25
Student/economically inactive	<5b	<7b	9	3	14	4	10	6
Couple fecundity incl. TTP and MAR									<5b	<lb
Unplanned pregnancy	12	16	58	21	51	15	24	16
TTP 0–5 months	<48b	<66	155	57	208	63	<86b	<56
TTP 6–12 months	7	10	26	10	25	8	23	15
TTP >12 months or MAR	6	8	31	11	46	14	21	14
Maternal age at delivery in mean years (SD)	30.5 (4.6)		31.0(4.0)		31.0 (4.3)		31.5 (4.0)		<5b	<lb
Maternal pre-pregnancy BMI (kg/m2)									22	3
< 18.5	6	8	15	6	20	6	5	3
18.5–24.9	< 40b	<55b	193	71	238	72	<126b	<82b
25–29.9	21	29	42	16	48	15	18	12
> 30	6	8	11	4	14	4	<5b	<3b
Maternal smoking 1. trimester (cigarettes/day)									0	0
0	51	70	204	76	262	79	120	78
1–10	> 17b	> 23b	52	19	60	18	> 29b	> 19b
> 10	< 5b	< 7b	14	5	8	2	< 5b	< 3b
Clinical characteristics
Season at clinical visit									0	0
Winter	15	21	53	20	78	34	26	17
Spring	8	11	49	18	56	17	29	19
Summer	17	23	66	24	73	22	40	26
Autumn	33	45	102	38	123	37	59	38
Abstinence time (days)c	2.0 (1.1)		2.3 (1.5)		2.4 (1.7)		2.3 (1.4)		< 5b	< lb
Place of semen sample collection									8	1
Home	11	15	36	13	39	12	24	16
Clinic	<62b	<85b	<234b	< 87b	<291b	<88b	<130b	<84b
Spillage									8	1
No	<54b	<74b	<222b	<82b	<279b	<85b	<130b	<84b
Yes	19	26	48	18	51	15	24	16
Interval ejaculation—analysis (min)d	50 (21)		49 (21)		49(18)		52 (19)		9	1
Time a day of blood sampling									9	1
Morning (< 12 p.m.)	29	40	89	33	124	38	50	32
Afternoon (12–18 p.m.)	36	49	< 149b	<55b	< 174b	<53b	<88b	<57b
Evening (>18 p.m.)	8	11	32	12	32	10	16	10
Semen characteristics a
Volume (mL)e	2.8 (1.0 to 5.5)		2.6 (1.1 to 5.3)		2.7 (1.0 to 5.6)		2.7 (1.1 to 5.1)		150	18
Concentration (mill/mL)	33 (3 to 121)		37 (2 to 140)		41 (4 to 134)		38 (1 to 151)		<5b	<lb
Total sperm count (mill)e	83 (5 to 380)		96 (9 to 404)		110(8 to 407)		101 (6 to 431)		150	18
Progressive motility (PR %)f	63 (26 to 83)		65 (39 to 84)		63 (31 to 84)		64 (26 to 84)		16	2
Morphology (% normal)f	7(1 to 17)		6(0 to 15)		6 (Oto 15)		6(1 to 14)		21	3
Testes volume
Average testes volume (mL)	14 (7 to 24)		15 (7 to 25)		15 (7 to 25)		18 (8 to 25)		<5b	<lb
Reproductive hormone levels a
Testosterone (nmol/L)	19(9 to 31)		18 (11 to 29)		18 (10 to 30)		18 (10 to 29)		9	1
Oestradiol (pmol/L)g	59 (13 to 126)		52 (11 to 108)		52 (11 to 102)		53 (11 to 125)		9	1
Sex-hormone binding globulin (nmol/L)	32 (15 to 68)		33 (18 to 56)		32 (16 to 58)		34 (20 to 62)		9	1
Luteinizing hormone (IU/L)g	4.9 (2.8 to 9.2)		4.9 (2.8 to 9.0)		5.1 (2.5 to 8.7)		4.8 (2.5 to 9.1)		9	1
Follicle stimulating hormone (IU/L)g	3.7 (1.2 to 9.8)		3.7 (1.2 to 8.6)		3.4 (1.4 to 7.7)		3.0 (1.3 to 8.6)		9	1
Free androgen index (%)h	58(29 to 99)		55 (34 to 87)		56 (31 to 99)		53 (31 to 92)		9	1
Testosterone/oestradiol ratio	6.5 (3.2 to 23.4)		7.2 (3.8 to 29.8)		7.3 (4.0 to 23.2)		7.0 (3.4 to 21.9)		9	1

Numbers in the table correspond to mean (SD); frequencies (%); or p50 (p5 to p95). Percentages may not add up to 100% due to rounding of numbers

TTP, time to pregnancy; MAR, medically assisted reproduction; SD, standard deviation; p5, 5th pseudo percentile; p50, 50th pseudo percentile; p95, 95th pseudo percentile; BMI, body mass index

^aReported as 50th percentile (5th–95th). All percentiles are pseudo percentiles calculated from the average of five values

^bDue to local data regulations it is not allowed to report numbers smaller than five, why the numbers in the table have been modified to mask the numbers smaller than five

^cIn total, 290 participants, corresponding to 35% of all participants in this study, had an abstinence time of less than the recommended 48 h

^dIn total, 200 participants, corresponding to 24% of all participants in this study, had their semen sample analysed more than 60 min after ejaculation

^eExcluding samples from participants reporting spillage. In total n = 150

^fExcluding azoosperme samples n = 11

^gIn total n = 64 samples, corresponding to 8% of all samples in this study, was below limit of detection for oestradiol, and n < 5, corresponding to 0% of all samples in this study, was below limit of detection for luteinizing hormone and follicle-stimulating hormone

^hCalculated as (total testosterone/Sex-hormone binding globulin) × 100%

Sons of mothers with vitamin D levels below the reference level of 75 nmol/L had lower average testes volume compared to sons of mothers with vitamin D levels above 75 nmol/L (Table 2). Sons of mothers with vitamin D levels < 25 nmol/L had lower testes volume of − 11% (95% CI − 19 to − 2); sons of mothers with levels of 25–50 nmol/L had lower testes volume of − 8% (95% CI − 14 to − 2); and sons of mothers with levels of 50–75 nmol/L had lower testes volume of − 7% (95% CI − 13 to − 1). Sons of mothers with vitamin D levels < 25 nmol/L had 20% (95% CI − 40 to 9) lower total sperm count compared to sons of mothers with vitamin D levels > 75 nmol/L. For the remaining semen characteristics and reproductive hormones, we did not observe clear associations.

Table 2.

Crude and adjusted^a (95% confidence intervals) relaive percentage differences in measures of male reproductive health according to categorised maternal vitamin D concentrations and continuously modelled maternal vitamin D (nmol/L) in 827 participants from the Fetal Programming of Semen Quality (FEPOS) Cohort, 1998–2019, Denmark

	nb	nmol/L	Crude	Adjusteda (95% CI)
Semen characteristics
Volume (mL)c	653	> 75	Ref	Ref
		50–75	2%	2% (− 7 to 12)
		25–50	3%	1% (− 9 to 11)
		< 25	− 1%	0% (− 14 to 17)
Per 10 nmol/L lower vitamin D			− 0%	0% (− 2 to 1)
Concentration (mill/mL)	787	> 75	Ref	Ref
		50–75	− 1%	− 6% (− 20 to 12)
		25–50	− 6%	− 4% (− 20 to 15)
		< 25	− 17%	− 12% (− 32 to 13)
Per 10 nmol/L lower vitamin D			− 2%	− 2% (− 4 to 1)
Total sperm count (mill)c	653	> 75	Ref	Ref
		50–75	− 5%	− 3% (− 20 to 17)
		25–50	− 5%	− 4% (− 20 to 17)
		< 25	− 26%	− 20% (− 40 to 9)
Per 10 nmol/L lower vitamin D			− 3%	− 3% (− 5 to 0)
Motility (% NP + IM)d	771	> 75	Ref	Ref
		50–75	1%	2% (− 7 to 11)
		25–50	− 3%	− 5% (− 13 to 3)
		< 25	7%	3% (− 8 to 11)
Per 10 nmol/L lower vitamin D			0%	− 1% (− 2 to 1)
Morphology (% normal)	770	> 75	Ref	Ref
		50–75	0%	− 4% (− 16 to 9)
		25–50	0%	0% (− 13 to 14)
		< 25	7%	7% (− 13 to 31)
Per 10 nmol/L lower vitamin D			0%	0% (− 2 to 2)
Testes volume
Average testes volume (mL)	794	> 75	Ref	Ref
		50–75	− 7%	− 7% (− 13 to −1)
		25–50	− 7%	− 8% (− 14 to −2)
		< 25	− 11%	− 11% (− 19 to −2)
Per 10 nmol/L lower vitamin D			− 1%	− 1% (− 2 to 0)
Reproductive hormones
Testosterone (nmol/L)	791	> 75	Ref	Ref
		50–75	− 2%	− 4% (− 10 to 2)
		25–50	− 0%	0% (− 6 to 6)
		< 25	1%	− 1% (− 10 to 8)
Per 10 nmol/L lower vitamin D			0%	0% (− 1 to 1)
	nb	nmol/L	Crude	Adjusteda (95% CI)
Oestradiol (pmol/L)	791	> 75	Ref	Ref
		50–75	− 6%	− 8% (− 16 to 2)
		25–50	− 4%	− 2% (− 13 to 10)
		< 25	5%	5% (− 9 to 21)
Per 10 nmol/L lower vitamin D			1%	1% (− 1 to 3)
SHBG (nmol/L)	791	> 75	Ref	Ref
		50–75	− 5%	− 6% (− 13 to 2)
		25–50	− 3%	− 3% (− 10 to 4)
		< 25	0%	1% (− 11 to 14)
Per 10 nmol/L lower vitamin D			0%	0% (− 1 to 1)
LH (IU/L)	791	> 75	Ref	Ref
		50–75	5%	3% (− 4 to 11)
		25–50	3%	2% (− 6 to 10)
		< 25	3%	− 2% (− 12 to 8)
Per 10 nmol/L lower vitamin D			0%	0% (− 1 to 1)
FSH (IU/L)	791	> 75	Ref	Ref
		50–75	4%	1% (− 10 to 14)
		25–50	7%	5% (− 7 to 18)
		< 25	10%	6% (− 11 to 27)
Per 10 nmol/L lower vitamin D			2%	2% (0 to 3)
Free androgen index (%)	791	> 75	Ref	Ref
		50–75	5%	4% (− 3to 11)
		25–50	4%	4% (− 4 to 12)
		< 25	4%	− 1% (− 10 to 8)
Per 10 nmol/L lower vitamin D			0%	0% (− 1 to 1)
Testosterone/oestradiol ratio	791	> 75	Ref	Ref
		50–75	9%	8% (− 5 to 22)
		25–50	17%	13% (− 1 to 29)
		< 25	− 4%	− 5% (− 21 to 14)
Per 10 nmol/L lower vitamin D			0%	0% (− 2 to 2)

SHBG, sex-hormone binding globulin; LH, luteinizing hormone; FSH, follicle stimulating hormone

^aAll analyses are adjusted for socioeconomic status (highest of the parents), couple fecundity, maternal age at delivery; maternal pre-pregnancy body mass index; maternal first trimester smoking and season of clinical visit. Semen characteristics are further adjusted for abstinence time; place of semen sample collection and spillage. Testes volume are further adjusted for abstinence time. Reproductive hormones are further adjusted for time of blood sampling

^bNumbers correspond to the adjusted models. The number of sons contributing to each analysis vary due to missingness (due to spillage or azoospermia)

^cExcluding samples with spillage

^dFurther adjusted for interval between ejaculation and analysis (minutes). Due to model fit, tables show results for NP (Non-progressive) + IM (Immotile) spermatozoa. Therefore, positive estimates should be interpreted as a decrease in progressive motility and vice versa

Overall, the linear associations supported these findings and gave some indications of dose-dependency (Table 2). Lower levels of maternal vitamin D were associated with lower average testes volume, concentration, total sperm count, and higher FSH.

The adjusted spline plots (Fig. 2) showed that maternal vitamin D levels were associated with concentration, total sperm count, and average testes volume, in addition to an inverse association with FSH in a dose-dependent manner.

Fig. 2.

Restricted cubic spline plots (three knots) of measures of male reproductive health according to maternal vitamin D levels (solid lines) with 95% confidence intervals (dotted lines). The estimated measures are presented for a reference son, whose parents’ highest socioeconomic position was a high-grade professional and had a time to pregnancy of 0–5 months, whose mother was 30 years at the delivery, was normal weight, and a non-smoker. The reference son had an abstinence time of 2.5 days, delivered his semen sample at the clinic, did not report any spillage of the semen sample, attended the clinic during the autumn season, had his motility assessment done 30 min after ejaculation and had blood drawn for assessment of reproductive hormone levels from 12 to 18 p.m. Abbreviations: NP, non-progressive motility; IM, Immotile motility; SHBG, Sex-hormone binding globulin; FSH, Follicle-stimulating hormone; LH, Luteinizing hormone

The risk of having low semen characteristics and testes volume was higher for sons of mothers with lower vitamin D levels compared to sons of mothers with higher vitamin D levels (Table 3). The adjusted risk of having low testes volume was 1.4 (95% CI 1.0 to 1.9) in sons of mothers with vitamin D levels < 25 nmol/L compared to sons of mothers with vitamin D levels > 75 nmol/L. Sons of mothers with vitamin D levels < 25 nmol/L had an increased risk of low total sperm count of 1.6 (95% CI 0.9 to 2.9) compared to sons of mothers with vitamin D levels > 75 nmol/L.

Table 3.

Crude and adjusted^a (95% confidence intervals) relative risks of low semen characteristics according to WHO’s lower reference limits and low testes volume according to categorised maternal vitamin D concentrations and continuously modelled maternal vitamin D (nmol/L) in 827 participants from the Fetal Programming of Semen Quality (FEPOS) Cohort, 1998–2019, Denmark

	nb	nmol/L	Crude	Adjusted (95% CI)
Semen characteristics
Volume < 1.5 mLc	653	> 75	Ref	Ref
		50–75	1.4	1.2 (0.6 to 2.2)
		25–50	1.2	1.1 (0.6 to 2.0)
		< 25	1.9	1.6 (0.7 to 3.5)
Per 10 nmol/L lower vitamin D			1.1	1.1 (1.0 to 1.2)
Concentration < 15 mill/mL	787	> 75	Ref	Ref
		50–75	0.7	0.7 (0.5 to 1.0)
		25–50	0.9	1.0 (0.6 to 1.4)
		< 25	0.9	0.9 (0.5 to 1.5)
Per 10 nmol/L lower vitamin D			1.0	1.0 (0.9 to 1.1)
Total sperm count < 39 millc	653	> 75	Ref	Ref
		50–75	1.2	1.1 (0.7 to 1.7)
		25–50	1.2	1.2 (0.8 to 1.9)
		< 25	1.9	1.5 (0.9 to 2.7)
Per 10 nmol/L lower vitamin D			1.1	1.1 (1.0 to 1.2)
Progressive motility < 32%d	771	> 75	Ref	Ref
		50–75	0.8	0.9 (0.4 to 1.9)
		25–50	0.3	0.3 (0.1 to 0.9)
		< 25	1.3	1.2 (0.5 to 3.2)
Per 10 nmol/L lower vitamin D			0.9	1.0 (0.8 to 1.1)
Morphology < 4% normal	770	> 75	Ref	Ref
		50–75	1.2	1.4 (1.0 to 2.0)
		25–50	1.0	1.2 (0.8 to 1.7)
		< 25	1.1	1.2 (0.7 to 2.0)
Per 10 nmol/L lower vitamin D			1.0	1.0 (1.0 to 1.1)
Testes volume
TAverage testes volume < 15 mL	794	> 75	Ref	Ref
		50–75	1.2	1.2 (0.9 to 1.6)
		25–50	1.2	1.2 (0.9 to 1.6)
		< 25	1.4	1.4 (1.0 to 1.9)
Per 10 nmol/L lower vitamin D			1.0	1.0 (1.0 to 1.1)

^aAll analyses are adjusted for socioeconomic status (highest of the parents), couple fecundity, maternal age at delivery; maternal pre-pregnancy body mass index; maternal first trimester smoking and season of clinical visit. Semen characteristics are further adjusted for abstinence time; place of semen sample collection and spillage. Testes volume are further adjusted for abstinence time

^bNumbers correspond to the adjusted models. The number of sons contributing to each analysis vary due to missingness (due to spillage or azoospermia)

^cExcluding samples with spillage

^dFurther adjusted for interval between ejaculation and analysis (minutes)

Instrumental variable analysis

Distribution of baseline characteristics did not differ across levels of season of GW 8 + 0 (Online Resource (Supplementary Table 1)), with the exception of season at clinical visit. However, vitamin D concentrations differed with the highest median concentration during the summer season [p50: 63 nmol/L, p5 to p95: (36 to 101)] and the lowest during the spring (p50; 45 nmol/L, p5 to p95: (17 to 82)). Month at GW 8 + 0 could explain approximately 20–22% of total variability in all models and partial F-statistics were > 126 in all models (Table 4).

Table 4.

Adjusted^a (95% confidence intervals) relative percentage differences in measures of male reproductive health per 10 nmol/L lower predicted maternal vitamin D concentrations in gestational week 8 + 0 in 827 participants from the Fetal Programming of Semen Quality (FEPOS) Cohort, 1998–2019, Denmark

Semen characteristics	n	Adjusteda (95% CI)
Volume (mL)b	653	− 4% (− 8 to 0)
Concentration (mill/mL)c	787	− 1% (− 10 to 8)
Total sperm count (mill)b	653	− 6% (− 15 to 4)
Motility (% NP + IM)d	771	− 1% (− 4 to 2)
Morphology (% normal)e	770	− 2% (− 9 to 4)
Testes volume
Testes volume (mL)f	794	− 2% (− 5 to 1)
Reproductive hormones g
Testosterone (nmol/L)	791	0% (− 3 to 2)
Oestradiol (pmol/L)	791	2% (− 2 to 6)
SHBG (nmol/L)	791	1% (− 2 to 4)
LH (IU/L)	791	1% (− 2 to 4)
FSH (IU/L)	791	5% (1 to 10)
Free androgen index (%)	791	− 1% (− 3 to 2)
Testosterone/oestradiol ratio	791	− 2% (− 5 to 1)

^aAdjusted for socioeconomic status (highest of the parents), couple fecundity, maternal age at delivery, maternal pre-pregnancy body mass index, maternal first trimester smoking, season at clinical visit. Semen characteristics further adjusted for abstinence time, place of semen sample collection and spillage; testes volume further adjusted for abstinence time; reproductive hormones further adjusted for time of the day of blood sampling

^bExcluding samples with spillage. r² = 18.82% F-statistics = 124.683

^cr² = 20.86% F-statistics = 164.889

^dFurther adjusted for interval between ejaculation and analysis (minutes). Due to model fit, tables show results for NP (Non-progressive) + IM (Immotile) spermatozoa. Therefore, positive estimates should be interpreted as a decrease in progressive motility and vice versa. r² = 21.20% F-statistics = 162.635

^er² = 21.20% F-statistics = 165.579

^fr² = 21.34% F-statistics = 169.694

^gr² = 20.95%. F-statistics = 165.209

Lower vitamin D levels predicted from the season in GW 8 + 0 was associated with lower semen volume of − 4% (95% CI − 8 to 0), lower total sperm count of − 6% (95% CI − 15 to 4), lower testes volume of − 2% (95% CI − 5 to 1), and higher FSH of 5% (95% CI 1 to 10) per 10 nmol/L decrease in vitamin D levels (Table 4).

Sensitivity analysis

When further adjusting the main analysis for PFOS, results remained essentially the same (Online Resource (Supplementary Table 2)). When conducting the instrumental variable analysis using the gmm estimator, results did not change (Online Resource (Supplementary Table 3)).

Discussion

Key results

This large follow-up study investigated the association between maternal vitamin D levels and male reproductive health using different methodological strategies to strengthen the causal interpretation for this previously unstudied association. Overall, lower maternal vitamin D levels in pregnancy were associated with lower testes volume, lower total sperm count, and higher FSH in the adult male offspring.

Strengths and limitations

We used unique data from the DNBC and FEPOS, the largest male-offspring cohort to date [18]. The major strengths of this study included the supplementary instrumental variable analysis to support the main regression analysis, the use of quality controlled and valid measures of maternal vitamin D levels in pregnancy and measures of reproductive health in male offspring 19 years later. Moreover, we were able to adjust for multiple important, potential confounders, and applied inverse probability of selection weights in all analyses.

Though the participation rate was low (19%), it was comparable to most studies on volunteers providing a semen sample [26]. Participation of the sons is unlikely to be associated with reproductive health, since they—given their young age—were most likely unaware of their fertility status [38]. We cannot exclude the possibility that participation of the sons was associated with levels of maternal vitamin D, since we did not have information on maternal vitamin D levels among non-participants. However, sons of mothers with available information on vitamin D were of similar age, had similar levels of potential confounders and male reproductive health compared to sons with no available information on maternal vitamin D levels. Mothers of participants were more likely to report alcohol consumption, to be non-smokers and to be underweight than mothers of non-participants. However, in a validation study, we found that the risk of selection bias was low in aetiological studies examining prenatal exposures and male reproductive health using the FEPOS cohort [26]. The risk of selection bias in this study is therefore considered low.

We used a quantitative measure of vitamin D, i.e., plasma levels of 25(OH)D₃, which is a major strength. Levels of 25(OH)D₃ were measured in blood samples obtained in the first trimester, and with a plasma half-life of approximately 2–3 weeks [23], 25(OH)D₃ was considered a useful biomarker for determining exposure in early pregnancy, which may be a relevant exposure period regarding male reproductive health [34]. Most 25(OH)D₃ is synthetized endogenously in the skin following exposure to sunlight [1]. By measuring plasma 25(OH)D₃ we therefore considered the contributions from both sunlight and intake of 25(OH)D₃ from dietary sources. We did not measure 25-hydroxyvitamin D₂ (25(OH)D₂), which is present in some foods, such as mushrooms. Only few women in this study confirmed they were vegetarians (n < 5), and the concentration of 25(OH) D₂ is often below the detection limit in most samples from a Danish population [39]. We therefore expect the contribution from 25(OH)D₂ in the total pool of bioavailable vitamin D to be of minor importance. Potential measurement errors are expected to be minor, independent, and non-differential with regard to participant outcomes.

Outcome assessment was blinded to the participants’ exposure status. Potential measurement errors on male reproductive health are therefore most likely non-differential. Moreover, semen characteristics and reproductive hormone levels were measured using state-of-the-art techniques and methods were continuously quality controlled, limiting potential measurement errors. We only obtained one semen sample, however, the with-in individual variation in semen characteristics is not suspected to introduce systematic errors [40]. Though some participants may have underestimated their testes volume [21], this is likely independent of and non-differential regarding participants’ exposure status, thus not introducing any systematic errors. The daily fluctuations in reproductive hormone levels will introduce some measurement error; we adjusted for time of day of blood sampling to accommodate this.

Though we adjusted for several important potential confounders, our results from the main regression analyses may still be affected by residual confounding. However, the results from the instrumental variable analysis overall confirmed the results from the main analysis. This increases our confidence in the results, since the two different methods are conducted under different sets of assumptions [27]. In the instrumental variable analysis, it may be possible to obtain a valid estimate despite potential unmeasured or residual exposure-outcome confounding, given that the core assumptions of relevance, exchangeability and the exclusion restriction criteria were fulfilled. As thoroughly discussed in the Online Resource (Supplementary Text 1), we had a strong instrument with a partial F-statistics > 126 in all models; hence, we consider the relevance assumption to be fulfilled. Further, we obtained exchangeability by adjusting the instrumental variable analysis for many covariates, including parental couple fecundity (Online Resource (Supplementary Fig. 4)). A violation of the exclusion restriction criteria may pose the main threat to the validity of the results obtained in the instrumental variable analysis. If month at GW 8 + 0 is associated with another environmental factor, e.g., air pollution, high temperature or use of sunscreen containing chemicals that may also affect male reproductive health, this would potentially induce bias of the results in either direction as described in Online Resources (Supplementary Text 1).

Interpretation

Our results supported the hypothesis that low maternal vitamin D levels in pregnancy may be detrimental to reproductive health in male offspring [5–7]. We found associations between vitamin D levels and lower testes volume, lower total sperm count and higher FSH, supporting the hypothesis that vitamin D may play a vital role in reproductive organ development and later spermatogenesis due to the metabolism of vitamin D in foetal Sertoli and germ cells [12]. Low vitamin D levels in pregnancy were not associated with other markers of male reproductive health in this study. We did not find any support of the hypothesis that vitamin D may affect male reproductive hormone production [11].

Testes volume is correlated with total sperm count, since testes volume is a measure of the capacity of the testes to produce spermatozoa [41]. High FSH is a marker of impaired spermatogenesis [42] and may reflect dysfunction at the testicular rather than the hypothalamic level, e.g., in the seminiferous tubules. This suggests that vitamin D may have exerted an effect on male reproductive development primarily on the testes [16], potentially by impairing Sertoli cell proliferation [12]. However, this must be further investigated. No studies have investigated the association between maternal vitamin D levels in pregnancy and reproductive health in adult sons, and our findings should thus be independently replicated. Moreover, additional proof of concept in animal studies investigating the effect of prenatal vitamin D depletion on reproductive health is warranted.

At the time of inclusion of the pregnant women into the DNBC, there were no general recommendations regarding supplementary vitamin D intake. This may explain the high proportion of women with vitamin D levels below the present recommended level in our study (81%). Endogenous skin synthesis of vitamin D occurs only during the summer at northern latitudes [1], and food fortification with vitamin D was and is not employed in Denmark. Therefore, the prevalence of women with vitamin D levels below recommended may be higher compared with other populations. Today, all pregnant women in Denmark are advised to maintain a plasma level of 25(OH)D above 75 nmol/L by consuming a healthy, balanced diet and by a daily supplementation of 400 IU of vitamin D [4]. A recent Danish study found only 42% of pregnant women having vitamin D plasma levels > 75 nmol/L, despite high adherence (86%) to the current recommendations, [43] highlighting the possibility that the current recommended daily intake may be too low.

We investigated Danish pregnant women with vitamin D levels within the low to normal range and their young, healthy sons. Since our findings are independent on social or cultural settings, our results may be generalizable to populations outside the Nordic countries.

In conclusion, we found lower testes volume, lower total sperm count and higher FSH in sons of mothers with low vitamin D levels in pregnancy compared to sons of mothers with vitamin D levels above the recommended level of 75 nmol/L. Our results suggest that it may be beneficial to strengthen antenatal guidance to ensure that women maintain a recommended level of vitamin D during pregnancy.

Data availability

The dataset analysed in the study is not publicly available due to national data security legislation on sensitive personal data. Researchers may apply for access to data from the DNBC. Please see https://www.dnbc.dk/data-available or write to dnbc-research@ssi.dk for additional information.