The associations between exposure to mixed environmental endocrine disruptors and sex steroid hormones in men: a comparison of different statistical models

Abstract

In recent years, worldwide fertility rates have continued to decrease. Humans are frequently exposed to a combination of environmental endocrine disruptors, which can cause male reproductive disorders. The study employed three distinct analytical models to examine the correlation between exposure to a combination of 25 chemicals and sex steroid hormone levels in adult males. This involved evaluating 12 chemicals and their metabolites from personal care and consumer products, as well as 13 metabolites linked to phthalates and plasticisers. The study analysed 25 chemicals and 3 measured sex steroid hormone outcomes, as well as two calculated hormonal outcomes (free androgen index, TT/E2 ratio) in 1262 adult men who participated in the National Health and Nutrition Examination Survey (NHANES) 2013–2016 in the United States. The study employed several statistical methods to estimate the relationships between single chemicals or chemical blends and sex hormones. These methods included linear regression, weighted quantile sum (WQS) regression, and Bayesian kernel machine (BKMR) regression. The results of the linear regression analysis indicate that chemical exposure has a negative correlation with E2, TT, and FAI, and a positive correlation with SHBG and TT/E2. The mixture effect analyses using the WQS and BKMR models further confirmed that BP3, MECPP, and MECOP were the most highly weighted chemical mixtures. The analyses also suggested that there were differences in the effects of different concentrations of EDCs on sex steroid hormones. Exposure to environmental endocrine-disrupting chemicals (EDCs) has been found to have a negative correlation with estradiol and total testosterone, as well as FAI. Conversely, this exposure has been found to have a positive correlation with sex hormone binding globulin and the TT/E2 ratio. The study also revealed differences in the effects of different concentrations of EDCs.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-76972-z.

Introduction

Environmental endocrine-disrupting compounds (EDCs) are substances that disrupt endocrine homeostasis by mimicking or competing with endogenous hormones, thereby causing adverse health effects¹. These include phenols, parabens, and phthalates². Evidence from both experimental and observational studies indicates that exposure to EDCs can damage spermatogenic and supportive cells, leading to abnormalities in testicular development and function. This can cause a decrease in sperm count and activity and an increase in the number of abnormal spermatozoa, which ultimately affects male fertility³. In recent years, male infertility has been increasing globally. Several studies have shown a significant decline in sperm quantity and quality in men over the past few decades. One study reported that the sperm concentration in men in Western countries declined by approximately 50–60% between 1973 and 2011. These EDCs are present in everyday life in the form of various consumer products, and several epidemiological investigations have detected EDCs in humans^4–6.

Assessing exposure to EDCs through the detection of human-excreted environmental endocrine disruptors or their representative metabolites is highly feasible compared with that of assessing exposure from sources and routes. Several previous studies have demonstrated a correlation between urinary concentrations of these chemicals and male steroid hormones. However, most of these studies have focused on single chemicals or combinations of several simple chemicals, and only weighted linear regression has been used to assess the relationship between exposure and sex steroid hormones^7–9.

In the real world, people are often exposed to multiple chemicals at the same time, with complex interactions and high correlations among them. Recent studies have not effectively estimated the effects of EDCs on steroid hormones in men, and more comprehensive strategies are needed to assess the confounding health effects of multiple chemical exposures.

This study analyzed demographic data from adult male NHANES participants in the United States from 2013 to 2016 to identify EDCs that may be associated with steroid hormones in adult males. We included 12 chemicals and their metabolites found in personal care and consumer products, as well as 13 metabolites of phthalates and plasticizers that, based on previous studies, suggest an association with male steroid hormones^10–12. The impact of these chemicals was assessed via linear regression, WQS regression, and BKMR models. The strengths and weaknesses of each model were considered, and the results from all three methods were interpreted together. These findings may provide insights for future longitudinal epidemiological investigations and experimental studies.

Methods

Study demographics

The NHANES is a cross-sectional study conducted biennially that is designed to assess the health and nutritional status of both adults and children within the United States.

The study is designed to be nationally representative. The analysis utilized data released in 2013–2014 and 2015–2016 and was confined to male participants aged 20 years or older who had not used any of the sex hormone medications listed on the NHANES questionnaire, such as testosterone, progesterone, estrogen, or ‘other sex hormones.’ Additionally, measurements of participants’ urinary environmental phenols, parabens, triclocarban, and phthalate metabolite concentrations, as well as serum measurements of sex hormones were included.

The NHANES interview included demographic, socioeconomic, dietary, and health-related questions. The NCHS (National Center for Health Statistics) research ethics review board (ERB) approved the NHANES study protocol, and participants provided written informed consent at enrollment. The NCHS IRB/ERB protocol numbers of the 2007–2016 National Health and Nutrition Survey are ‘#2005-06’ and ‘#2011-17’ (the website is https://www.cdc.gov/nchs/nhanes/irba98.htm). Since this study was based on publicly available deidentified data and informed consent was waived, ethical approval and consent were not required.

Demographic covariates

Trained interviewers completed a standardized questionnaire at the participants’ residences to capture demographic variables, including age, racial/ethnic background, gender, marital status, educational level, and poverty level. Monthly expenses for take-out food were documented to calculate the potential exposure to chemicals from disposable food packaging¹³. To account for variations in sex steroid hormone levels influenced by seasonal changes, the statistical analysis was adjusted for the season blood was drawn¹⁴. Sample collection seasons were coded as the 6-month period at the time of the examination and were divided into two categories: from November 1-April 30 and from May 1-October 31.

Exposure information

Spot urine samples were collected at mobile examination centers (MECs). The NHANES Laboratory Procedures Manual (LPM) provides detailed collection and processing protocols for various samples. The samples were stored in secure vials at an optimal freezing temperature of -20 °C until they were shipped to the National Center for Environmental Health for analysis. We included 25 metabolites that had undergone complete testing.

A novel analytical technique for the quantification of bisphenol A (BPA), benzophenone-3, triclosan, a suite of four parabens, two dichlorophenols, and triclocarban has been devised, building upon a previously reported methodology. Detailed information about the chemical measurement methods and quality control procedure can be obtained from the NHANES laboratory methods (https://wwwn.cdc.gov/Nchs/Nhanes/2013-2014/EPHPP_H.htm). This technique employs online solid-phase extraction (SPE) interfaced with high-performance liquid chromatography (HPLC) and tandem mass spectrometry (MS/MS) to facilitate the sensitive and efficient analysis of target compounds. The incorporation of isotopically labeled internal standards ensures a low detection limit, ranging from 0.1 to 1.7 micrograms per liter (µg/L) in a 100 µL urine sample. This level of sensitivity is adequate for the assessment of the urinary concentrations of phenolic compounds, parabens, and triclocarban in subjects not occupationally exposed to these chemicals.

High-performance liquid chromatography‒electrospray ionization‒tandem mass spectrometry (HPLC‒ESI‒MS/MS) was used to quantitatively detect several phthalate metabolites and alternative phthalates in the urine samples. The resulting urine samples were subjected to enzymatic hydrolysis to analyze the glycoconjugates, followed by online solid-phase extraction (SPE) and reversed-phase HPLC-ESI-MS/MS processing. Isotopically labeled internal standards were added to the target analytes to improve analytical precision. In addition, 4-methylumbelliferone glucuronide was used to monitor the efficiency of the deconjugation reactions. Concentrations that were below the LOD were replaced with LOD/√2. This is one of the standard treatments for exposure values below the LOD, but their performance can be quite unstable, especially in scenarios with high correlations or high percentages of values below the LOD, as commonly observed in environmental mixture studies¹⁵. To account for variations in dilution in the urine samples, metabolic concentrations were normalized to creatinine concentrations. The NHANES Laboratory Procedures Manual was used for detailed information on chemical measurement techniques and quality control protocols.

Outcome assessment

Blood samples are preferably taken early in the morning to avoid fluctuations in hormone levels caused by daily activities. The total estradiol (E2) and total testosterone (TT) levels in human serum were measured via isotope-diluted high-performance liquid chromatography‒tandem mass spectrometry (ID‒LC‒MS‒MS/MS). The quantification of the analytes was achieved by incorporating stable isotope-labeled internal standards and external calibrators. The ID‒LC‒MS/MS method utilized a triple quadrupole mass spectrometer with electrospray ionization. Testosterone was ionized in positive ion mode, whereas estradiol was ionized in negative ion mode. The identification of estradiol and testosterone was based on their chromatographic retention times and specific mass‒charge ratio transitions through selected reaction monitoring (SRM). Internal standards were used in the form of stable isotope-labeled testosterone and estradiol, which were marked with 13 C. Sex hormone-binding globulin (SHBG) is a glycoprotein that is produced primarily in the human liver¹⁶. The assay is based on the immunoassay principle involving the specific interaction of sex hormone-binding globulin (SHBG) with its corresponding antibodies, followed by quantification of the chemiluminescent emission, released from the formation of the immune complex. This technique utilizes the high specificity of the antigen‒antibody reaction to detect and quantify SHBG levels within a biological matrix. The resulting chemiluminescent signal, which is directly proportional to the concentration of SHBG, is then accurately measured, providing a sensitive and reliable method for SHBG analysis¹⁷.

The established limits of detection for total testosterone (TT), estradiol (E2), and sex hormone-binding globulin (SHBG) were 0.75 ng/dL, 2.994 pg/mL, and 0.800 nmol/L, respectively. In addition to these measures, we calculated the free androgen index (FAI) and the TT-to-E2 ratio (TT/E2) to estimate the concentration of free (unbound) testosterone in the bloodstream^18,19 The FAI is a proxy for the fraction of free testosterone in the bloodstream²⁰.

The NHANES has implemented quality assurance and quality control (QA/QC) protocols that are fully compliant with the mandates of the Clinical Laboratory Improvement Amendments of 1988. A comprehensive treatment of these QA/QC procedures, including methodological details and operational guidelines, is provided in the NHANES LPM. This comprehensive documentation ensures that the laboratory components of the survey are conducted to the highest standards of accuracy and reliability, thereby contributing to the overall credibility and utility of the health and nutrition data produced by the NHANES for public health research and policy-making.

Statistical analysis

The study population was analyzed via a flowchart, as shown in Fig. 1. Geometric medians (GM), mean values, and selected percentiles were used to show the distribution patterns of serum sex hormone levels and chemical concentrations. Correlation coefficients were calculated to determine individual correlations. To normalize the data prior to conducting subsequent regression modeling, a natural logarithm (ln) transformation was applied due to the skewed distribution of these twenty-five chemicals and sex steroid hormones. This transformation allowed for the application of statistical analyses that assume normally distributed variables.

Fig. 1.

Schematic of study population flow.

Linear regression models were used to assess the relationships between the urinary concentrations of twenty-five chemicals and serum sex hormones in men of reproductive age. Linear regression is a commonly used and effective tool. It is simple, intuitive, easy to interpret, and provides the specific effect of each independent variable on the dependent variable.

Considering the potential modifying effect of age-related steroid hormone levels in men, we performed age-stratified analyses. With age, gonadotropin-releasing hormone secretion decreases, leading to an age-related decline in steroid hormone levels. Large population-based studies have shown that total serum testosterone levels decrease by 0.4% per year and that total testosterone levels decrease by 1.2% per year after the age of 40²¹. We therefore divided the study population into two groups according to age. Those aged 20–39 years composed the young group, and those aged > 39 years composed the middle-aged group.

This study implemented the weighted quantile sum (WQS) regression model to evaluate the combined impact of simultaneous chemical exposure on the levels of sex hormones. This approach investigated the cumulative impact of multiple chemicals, considering their individual contributions to the overall association with sex hormone concentrations. It is commonly used as a multipollutant model to calculate the combined effects of various environmental pollutants^22,23 In summary, WQS provides the flexibility to examine the overall impact of a chemical mixture while accounting for covariates and identifying the individual chemical effects that account for the observed correlations. A WQS index, which represents the chemical mixture, was created by using quartile rankings for each chemical and incorporating the empirically estimated weights for each compound into the composite metric. For this study, a WQS index was created by dividing the chemicals into quartiles and calculating the proportional weights of individual substances that contribute to the total effect of the mixture. Owing to the lack of prior research on the directional relationships between each chemical and sex hormone, we conducted two separate WQS regression analyses. One assumed a positive association between all components of the WQS index and sex hormones, whereas the other assumed a negative correlation.

We used the Bayesian kernel machine regression (BKMR) model to assess the aggregate effects of multiple pollutants. Compared with other models, the BKMR model offers increased flexibility. This model enables the detection of potential interactions and incorporates the exposure‒response patterns of each component within the chemical mixture. This allows for a more nuanced evaluation of the complex relationships among various environmental chemicals and their combined impact on health outcomes²⁴. The study evaluated the cumulative impact by comparing changes in sex steroid hormone levels across all time points against chemicals held constant at the 25th percentile. The univariate exposure‒response relationship for each chemical at its 75th and 25th percentile concentrations was examined meticulously, with all other chemicals held constant at their respective median levels. Finally, we evaluated the remaining markers at the 25th, 50th, and 75th percentile thresholds²⁵. In each participant, we investigated potential interactions within mixtures that may be mediated by variations in sex hormones by comparing the chemical concentrations within the 25th-75th percentile ranges. To determine the significance of individual exposure factors relative to the composite mixture effect²⁶, we calculated conditional posterior inclusion probabilities (PIPs). The BKMR framework can identify nonlinear and nonadditive associations among the chemicals being considered.

Possible confounders, selected on the basis of prior studies, included age, gender, ethnicity, body mass index (BMI), poverty income ratio (PIR), marital status, spending on take-out food, and the 6-month period at the time of the survey. In the NHANES survey, sample weights were applied to mitigate selection bias across age, sex, and racial subgroups. Therefore, we followed previous advice and used an unweighted estimation, as the variables used to calculate sample weights were already included in the adjusted model^27,28 Statistical analyses were performed via Free statistics software and R studio (packages ‘gWQS’ and ‘bkmr’). Differences were considered statistically significant at P < 0.05 (two-sided).

Results

Population characteristics

Table 1 presents the demographic characteristics of the participants from the NHANES involved in this study. The combined sample obtained from the NHANES between 2013 and 2016 was 20,147, from which 1,262 adult male subjects were included in the analytical sample. The average age (± the standard deviation) of these individuals was 49.21 ± 17.38 years. A significant proportion of the participants in this group were identified as non-Hispanic white (38.7%). In the present study, 78% of the sample reported having completed at least a high school education or its equivalent. Most participants were married, and there was an equal split between smokers and nonsmokers. The median PIR was 2.2 (IQR 3.03), and 63.2% of the participants spent little to no money on take-out food. The mean CR (± SD) of the participants was 145.52 ± 84.68 mg/dl, and the mean Alb concentration (± SD) was 54.59 ± 279.6 mg/l.

Table 1.

Descriptive statistics [mean ± SD (range) or n (%)] for demographics in the whole sample (n = 1262).

Variables	N	%	Median	IQR
Age (years)			49	29
20–39	418	33.1
40–59	408	32.3
≥ 60	436	34.5
Race
Mexican American	189	15
Other Hispanic	134	10.6
Non-Hispanic White	488	38.7
Non-Hispanic Black	259	20.5
Non-Hispanic Asian	143	11.3
Other Race - Including Multi-Racial	49	3.9
Six–month time period when surveyed
November 1 through April 30	618	49
May 1 through October 31	644	51
Education
Less than 9th grade	110	8.7
9-11th grade (Includes 12th grade with no diploma)	169	13.4
High school graduate/GED or equivalent	304	24.1
Some college or AA degree	354	28.1
College graduate or above	325	25.8
Marital
Married	742	58.8
Widowed	41	3.2
Divorced	111	8.8
Separated	28	2.2
Never married	234	18.5
Living with partner	106	8.4
Smoking
Yes	651	51.6
No	611	48.4
PIR			2.2	3.03
< 1.3	386	30.6
1.3–3.5	456	36.1
> 3.5	420	33.3
Carryout/delivered food (dollar/30 days)			0	32.25
Never	797	63.2
1–50	196	15.5
> 50	269	21.3
	Mean ± SD
CR (mg.dl)	145.52 ± 84.68	131.5	29
Alb (mg.l)	54.59 ± 279.6	9	13.38

Abbreviation: SD = standard deviation; IQR=interquartile range.

Discussion

The linear regression results indicated that when considering the impact of individual substances on health outcomes, most environmental endocrine disruptors were associated with decreased steroid hormone levels, except for BPA and MPB, which exhibited a significant positive correlation with SHBG. When exposure was analyzed as a mixture using the WQS model, the results suggested that mixed exposure was associated with reductions in E2 and the FAI, while showing a positive correlation with SHBG levels.

Previous epidemiology studies have shown that certain phthalates are linked to decreased testosterone levels^7,29,30 The present model revealed an association between phthalate exposure and decreased testosterone levels. Previous animal experiments reported similar findings, indicating that phthalate exposure can alter sex hormone levels³¹. One possible explanation for this is that phthalate metabolites inhibit steroid-producing enzyme activity in Leydig cells, leading to a reduction in testosterone synthesis. The enzymes involved in the testosterone biosynthesis pathway are P450scc, 3β-HSD, and P450c17. Phthalates may also impact the expression of insulin-like factor 3 (Insl3) in testicular interstitial cells, which can affect normal testicular development. However, previous population studies have shown a significant positive association between phthalate exposure and serum estradiol levels. This finding is inconsistent with what we observed in a linear regression model, and a study using a mixture containing MEP, MEHP, MBP, MiBP, MiNP, and MBzP on gene expression in the sex steroid hormone synthesis and degradation pathway. This study suggested that the mixture reduced the expression of five enzymes in the pathway that converts cholesterol to estradiol but that the effect was dose dependent; BPA disrupted endocrine metabolism in humans, primarily by mimicking estrogen hormone activity, inducing DNA methylation changes³², and affecting other pathways. Studies have also shown that exposure to EDCs reduces the expression of Steroidogenic Acute Regulatory protein (StAR) in the testes of offspring mice. The StAR protein is a crucial mitochondrial transporter that plays a vital role in steroid hormone synthesis³³. Polychlorinated biphenyls (PCBs) have a phenolic structure and can competitively bind to estrogen, androgen, and progesterone receptors in the body. This can affect the normal development of reproductive organs and the homeostatic regulation of sex hormone levels.

Linear regression is a research method commonly used in environmental epidemiology, providing direct insights into the impact of individual substances on health outcomes. This method assumes that the relationships between all independent variables are linear. When considering mixed environmental exposures and their complex nonlinear interactions, although interaction terms can be included, this increases the complexity of the model and may lead to issues of multicollinearity, potentially resulting in unstable coefficients. While the marginal effects are straightforward and easy to interpret, they may not accurately reflect the intricate relationships between variables in the real world. Consequently, in linear regression models, marginal effects may underestimate or overestimate the actual impact of independent variables on the dependent variable, and may also overlook significant interactions. In our study, negative correlations were observed between MEHHP, MEHP and MEOHP and E2 when the effects of individual chemicals were estimated using linear regression models. In the WQS model, MECPP, BP3, MHNCH and ETP were given higher weights, and MEHHP, MEHP and MEOHP were given lower weights when the effects of mixed exposures on E2 were investigated. As MECCP, MEHHP and MEOHP are highly correlated, the WQS model assesses the overall impact of individual exposed substances on the results by assigning them weights. In contrast, the impact of Substance A may arise mainly through an indirect correlation with Substance B. Therefore, the correlation between different exposures and effects may be due to the masking effect of another chemical exposure. Although the linear regression results are significant for MEHHP, MEHP and MEOHP; they may be influenced by the correlation with MECCP.

The WQS model, on the other hand, is designed to address multiple highly correlated exposure variables by assigning weights to calculate a composite index that assesses the impact of each EDC on the overall indicator. This composite index represents the joint effect of multiple variables on the outcome. In a WQS model, the weights reflect the relative importance of each variable in the combined effect. Although WQS is essentially a regression model, its marginal effects are not the direct effects of individual variables but rather the contribution of each variable in the composite index. The assignment of weights in WQS, which in our analyses reveals the combined effects of different mixtures of variables, has a greater impact on the outcome variable when the variable with a higher weight changes, and vice versa. When variables are highly correlated, linear regression may face the problem of model instability, making an accurate estimation of the marginal effects of each variable difficult³⁴. The WQS, however, mitigates the problem of covariance to a certain extent by assigning weights and concentrating the combined effects of multiple correlated variables into a single index for analysis.

However, the model has obvious limitations of not being able to assess the combined effects of chemicals with different modes of action simultaneously, and it assumes that the effects of the exposed substances are linear and that the effects of each exposed substance on health outcomes can be weighted and combined. If the true exposure‒outcome relationship is nonlinear or has complex interactions, then the WQS model may not fully capture these complexities. For example, the linear regression model shows that MCOP is significantly associated with lower SHBG concentrations. Therefore, its weight in the WQS model is positively and negligibly correlated with SHBG.

The BKMR model is particularly adept at handling nonlinear relationships and interactions among variables. It can reveal the effects of each variable under various combinations of conditions, referred to as conditional effects. In the BKMR model, conditional effects are typically measured through changes in the predicted values of the model. BKMR can estimate the marginal effects of each variable, representing the influence of a variable on the outcome while holding other variables constant. Furthermore, it visualizes the complex relationships between variables through exposure-response cross-sections, allowing researchers to observe how different levels of exposure impact health outcomes. The BKMR methodology works by calculating the total effect (multipollutant models), individual effects (single-pollutant models), relative importance (PIP), trends in nonlinear dose‒effect curves (CR curves), and interactions (double CR curves), providing a comprehensive solution to the covariance problem²⁵.

In some circumstances, variables that are highly weighted in the WQS model may have significant marginal and interaction effects in the BKMR model, which introduces PIP to help identify the importance of exposure to outcomes. This may indicate that these variables were identified as important exposure factors under both models. In our study, the effect of BP3 exposure on E2 was weighted higher in the WQS model (weight equal to 0.13), and BP3 also had a high PIP in the BKMR model. The distribution of the weights in the WQS model may approximately represent the direction and strength of the effects of certain exposure variables on the outcome variables in the BKMR model. By comparing the weights in the WQS model and the BKMR PIP, more comprehensive information on mixture effects can be obtained. The weights in the WQS model can be viewed as linear approximations of the effects of the variables on health outcomes, whereas the BKMR provides a nonlinear, interactive view of these effects.

BKMR identifies which direction leads to the largest change in the target variable by calculating the direction of the gradient of the target variable in the space of the input variables, which can be regarded as the “direction of the maximum gradient.” If the mixture effect is primarily linear, the WQS weights can be approximated to represent the direction of the “maximum gradient” in the BKMR model. This means that, under the assumption of linearity, WQS and BKMR may yield similar results in identifying significant components.

In our study on E2, the WQS weights were more consistent with the PIP of the BKMR, which may indicate that the mixture effect is largely linear, that the WQS effectively approximates the results of the BKMR, and that the direction of the WQS weighting may approximately represent the direction of the maximum slope in the BKMR model. An inherent limitation of the Bayesian kernel machine regression (BKMR) model is the implementation of the accounting method. The method of interpolating exposure‒response relationships by setting other chemicals at predetermined levels prevents the model from accurately assessing the effects of common exposure scenarios that include a range of high and low chemical concentrations. This limitation prevents a comprehensive assessment of the complex interactions and combined effects of multiple chemical exposures. These shortcomings can be remedied to some extent by the introduction of linear regression models or WQS models, particularly where linear effects dominate or where combined exposure effects are needed. The use of these models in combination can provide a more comprehensive and accurate assessment of the effects of multiple chemical exposures.

The analysis of the described model addresses three important issues in mixture exposure. First, it examines the aggregate impact of concomitant exposure to multiple chemicals on health outcomes. Second, it considers the significance of the effects of chemicals on health status. Finally, it explores the interactions between different chemicals.

However, our study has several limitations. First, our data are unweighted and representative only of U.S. adult males. The literature suggests that unweighted estimators perform better than weighted estimators when the covariates used to calculate sample weights are included in regression models^27,35. Our study aimed to find a comprehensive analytical solution to address the effects of mixed exposures on health outcomes by comparing different models. Therefore, unweighted analyses were not conducted. Additionally, we used raw data from the NHANES database and substituted values below the limit of detection (LOD) with a standardized figure to ensure an unbiased estimation of effects instead of using multiple interpolation. Notably, slight differences in the categories of EDCs were detected between the 2013–2014 and 2015–2016 cycles. As a result, some EDCs were detected only in the 2015–2016 cycle. This may lead to conflicting results and the exclusion of all highly relevant chemicals in the same model. Because the NHANES database contains information on the occupations and industries of the respondents, the limited coverage of these data, the paucity of referable information, and our assumption that the study participants were not at risk for specific occupational exposures were limitations of our study. In conclusion, our study did not allow for causal inferences about the effects of mixed environmental endocrine disruptor exposures on male steroid hormones. Therefore, additional large-scale prospective cohort studies and empirical investigations are imperative to substantiate the current evidence base within the medical field.

Conclusions

Linear regression, WQS regression and BKMR analytical frameworks were used to assess the associations between EDCs and male steroid hormones. The results from all three models collectively suggest that EDCs can impact male steroid hormone levels. MECPP, MEHHP, MEOHP, MEHP, BP3, and BPA are associated with steroid hormones in different directions, potentially reducing male fertility through various biological pathways. These results also highlight the complexity of the effects of EDCs on male steroid hormones. Our study focuses on the utility of using multiple methods to assess the health effects of exposure to complex chemical mixtures. A combination of different research methods, multifaceted interpretations of results, and complementary validation of different research methods can provide more reliable results for mixed chemical exposure studies. Further prospective studies on the effects of mixed endocrine disruptors on male fertility, as well as mechanistic studies, which are important for male fertility protection, could be conducted in the future based on the present study.

Table S3 summarizes the averages and distributions of the 25 chemical exposures. The limits of detection (LODs) ranged from 0.1 to 1.7 ng/mL, depending on the metabolite. Metabolite concentrations that fell below the limit of detection were imputed with the value of LOD/√2. To increase the normality of the data, a natural logarithm (ln) transformation was applied to the parameters, as all phthalate metabolite concentrations had skewed distributions. The transformed parameters were then used as continuous variables in the statistical analyses.

Relationships between chemical exposure levels and male sex steroid hormones: a linear regression model was employed

Table 2 shows the correlation between single concentrations of environmental endocrine disruptor metabolites and sex hormones in a representative sample of adult males in the United States. The data indicate that exposure to these chemicals is associated with lower levels of E2, TT, and the FAI but is positively correlated with SHBG and the ratio of TT to E2. Negative associations were found between BP3, BPA, MECPP, MEHP, MEOHP and E2; between MECPP, MEHHP, MEOHP and TT; between MCOP and SHBG; and between BPA, TRS, MPB, PPB, MBP, MEHHP, MEHP, MiBP, MEOHP and the FAI. The concentrations of BPA and MPB were positively associated with SHBG. Positive associations were found between BP3, PPB, and MEHP and TT/E2.

Table 2.

The linear regression showed that that the chemical exposures were in general negatively associated with E2 and TT, and FAI, while positively associated with SHBG and TT/E2.

Expose	TT		E2		SHBG		FAI		TT/E2
	95CI	P_value	95CI	P_value	95CI	P_value	95CI	P_value	95CI	P_value
BP3	-0.01 (-0.02 ~ 0.01)	0.453	-0.02 (-0.03~-0.01)	0.005	-0.01 (-0.03 ~ 0)	0.124	0 (-0.01 ~ 0.02)	0.568	0.02 (0 ~ 0.04)	0.013
BPA	-0.02 (-0.05 ~ 0.01)	0.214	-0.02 (-0.05 ~ 0)	0.045	0.03 (0 ~ 0.06)	0.032	-0.05 (-0.08~-0.02)	0.002	0.03 (0 ~ 0.06)	0.088
BPF	-0.01 (-0.03 ~ 0.02)	0.567	-0.01 (-0.03 ~ 0.01)	0.233	0 (-0.02 ~ 0.02)	0.976	-0.01 (-0.03 ~ 0.01)	0.547	0.01 (-0.01 ~ 0.03)	0.492
BPS	-0.01 (-0.03 ~ 0.02)	0.465	-0.01 (-0.03 ~ 0.01)	0.162	0 (-0.02 ~ 0.02)	0.777	-0.01 (-0.04 ~ 0.01)	0.302	0.01 (-0.01 ~ 0.03)	0.387
TCC	0.01 (-0.01 ~ 0.03)	0.191	0.01 (0 ~ 0.02)	0.085	0.01 (0 ~ 0.02)	0.188	0 (-0.01 ~ 0.02)	0.804	-0.01 (-0.02 ~ 0.01)	0.326
TCS	-0.01 (-0.03 ~ 0.01)	0.244	-0.01 (-0.02 ~ 0)	0.153	0.01 (0 ~ 0.02)	0.119	-0.02 (-0.04 ~ 0)	0.01	0 (-0.01 ~ 0.02)	0.62
BuP	0.01 (-0.02 ~ 0.04)	0.626	0 (-0.02 ~ 0.02)	0.943	0.01 (-0.01 ~ 0.04)	0.309	-0.01 (-0.03 ~ 0.02)	0.713	0 (-0.02 ~ 0.03)	0.766
EPB	0 (-0.03 ~ 0.02)	0.707	-0.01 (-0.03 ~ 0.01)	0.174	0.01 (-0.01 ~ 0.03)	0.258	-0.02 (-0.04 ~ 0.01)	0.148	0.01 (-0.01 ~ 0.03)	0.414
MPB	0 (-0.02 ~ 0.02)	0.89	-0.01 (-0.02 ~ 0.01)	0.219	0.02 (0 ~ 0.04)	0.012	-0.02 (-0.04 ~ 0)	0.041	0.01 (0 ~ 0.03)	0.145
PPB	0 (-0.02 ~ 0.01)	0.755	-0.01 (-0.02 ~ 0)	0.059	0.01 (0 ~ 0.03)	0.055	-0.01 (-0.03 ~ 0)	0.047	0.01 (0 ~ 0.03)	0.04
2,5-DCP	-0.01 (-0.02 ~ 0.01)	0.372	0 (-0.01 ~ 0.01)	0.898	0 (-0.02 ~ 0.01)	0.495	0 (-0.02 ~ 0.01)	0.74	-0.01 (-0.02 ~ 0.01)	0.29
2,4-DCP	-0.01 (-0.03 ~ 0.02)	0.601	0 (-0.02 ~ 0.02)	0.885	0.01 (-0.01 ~ 0.03)	0.441	-0.01 (-0.04 ~ 0.01)	0.231	-0.01 (-0.03 ~ 0.02)	0.636
MCNP	0 (-0.04 ~ 0.03)	0.815	-0.01 (-0.03 ~ 0.02)	0.572	-0.02 (-0.04 ~ 0.01)	0.285	0.01 (-0.02 ~ 0.04)	0.508	0.01 (-0.02 ~ 0.04)	0.522
MCOP	-0.01 (-0.04 ~ 0.01)	0.385	-0.01 (-0.02 ~ 0.01)	0.541	-0.03 (-0.05~-0.01)	0.009	0.02 (-0.01 ~ 0.04)	0.18	0 (-0.02 ~ 0.03)	0.861
MECPP	-0.05 (-0.09~-0.01)	0.014	-0.03 (-0.06~-0.01)	0.018	-0.01 (-0.04 ~ 0.02)	0.439	-0.04 (-0.07 ~ 0)	0.057	0.02 (-0.02 ~ 0.05)	0.34
MBP	-0.02 (-0.06 ~ 0.02)	0.438	-0.01 (-0.04 ~ 0.02)	0.419	0.03 (0 ~ 0.06)	0.067	-0.05 (-0.09~-0.01)	0.016	0.01 (-0.03 ~ 0.05)	0.619
MCPP	-0.01 (-0.04 ~ 0.02)	0.438	-0.01 (-0.03 ~ 0.01)	0.181	-0.01 (-0.03 ~ 0.01)	0.464	0 (-0.03 ~ 0.02)	0.85	0.01 (-0.01 ~ 0.04)	0.292
MEP	0 (-0.03 ~ 0.02)	0.681	0 (-0.01 ~ 0.02)	0.86	0.01 (-0.01 ~ 0.03)	0.425	-0.01 (-0.03 ~ 0.01)	0.259	-0.01 (-0.03 ~ 0.01)	0.462
MEHHP	-0.05 (-0.08~-0.01)	0.011	-0.03 (-0.06~-0.01)	0.016	0.01 (-0.02 ~ 0.04)	0.671	-0.05 (-0.09~-0.02)	0.002	0.01 (-0.02 ~ 0.05)	0.465
MHNCH	0 (-0.04 ~ 0.03)	0.872	0.01 (-0.01 ~ 0.04)	0.345	-0.01 (-0.04 ~ 0.01)	0.367	0.01 (-0.02 ~ 0.04)	0.558	-0.02 (-0.05 ~ 0.01)	0.256
MEHP	-0.03 (-0.06 ~ 0.01)	0.16	-0.04 (-0.07~-0.02)	0.001	0.02 (-0.01 ~ 0.06)	0.116	-0.05 (-0.09~-0.02)	0.004	0.05 (0.01 ~ 0.08)	0.01
MiBP	-0.03 (-0.07 ~ 0.01)	0.185	0 (-0.03 ~ 0.03)	0.901	0.02 (-0.02 ~ 0.05)	0.295	-0.04 (-0.08~-0.01)	0.022	-0.02 (-0.05 ~ 0.02)	0.36
MiNP	-0.03 (-0.06 ~ 0.01)	0.114	-0.02 (-0.04 ~ 0.01)	0.195	-0.01 (-0.04 ~ 0.02)	0.599	-0.02 (-0.05 ~ 0.01)	0.218	0 (-0.03 ~ 0.03)	0.85
MEOHP	-0.05 (-0.09~-0.01)	0.009	-0.04 (-0.06~-0.01)	0.008	0 (-0.03 ~ 0.03)	0.912	-0.05 (-0.08~-0.01)	0.008	0.02 (-0.02 ~ 0.05)	0.325
MBzP	0 (-0.03 ~ 0.03)	0.826	0 (-0.02 ~ 0.02)	0.832	0.01 (-0.02 ~ 0.03)	0.487	-0.01 (-0.04 ~ 0.02)	0.43	0 (-0.03 ~ 0.03)	0.951

Stratified analyses explained the modifying effect of age, with associations between the concentrations of EDCs and sex hormones being more pronounced in middle-aged and elderly males than in young males. Specifically, in the young male group, BPA was negatively correlated with TT, E2 and the FAI; BP3 was negatively correlated with E2 and positively correlated with TT/E2; MCNP, MCOP and MCPP were negatively correlated with SHBG; MEP was negatively correlated with the FAI; MHNCH was positively correlated with TT; and MiNP was negatively correlated with SHBG. However, following the application of the Benjamini‒Hochberg correction, none of the p values observed in the young male group were statistically significant.

Among middle-aged and elderly men, BP3 was negatively correlated with E2 and SHBG. BPA was positively correlated with SHBG and negatively correlated with the FAI. TCS was positively correlated with the FAI and SHBG. MPB was negatively correlated with E2 and the FAI but positively correlated with SHBG and TT. PPB was negatively correlated with E2 and the FAI but positively correlated with TT. MECCP was negatively correlated with TT, E2 and the FAI. MBP was positively correlated with SHBG and negatively correlated with the FAI. MEP was positively correlated with SHBG. MEHHP was negatively correlated with TT, E2 and the FAI. MHNCH was negatively correlated with SHBG. MEHP was negatively correlated with E2 and the FAI. MiBP was negatively correlated with the FAI. MEOHP was negatively correlated with TT, E2 and the FAI. MEOHP was negatively correlated with TT, E2 and the FAI. For the above results, the p value was still statistically significant after Benjamini‒Hochberg correction (Table S1).

The WQS regression model was utilized to evaluate the connection between chemical concentrations and male sex steroid hormones

Table 3 illustrates the relationships between combined chemical exposure and sex hormone levels, as analyzed by WQS regression models. The WQS index positive model shows a positive correlation between the WQS index and SHBG. Conversely, the WQS index negative model has an inverse correlation between E2 and the FAI.

Table 3.

Associations between WQS index and sex hormone indicators among men in NHANES 2013–2016.

Models	TT		E2		SHBG		FAI		TT/E2
	P_value	β	P_value	β	P_value	β	P_value	β	P_value	β
Model 1 positive	0.2	-0.032	0.76	-0.007	0.032	0.055	0.874	-0.004	0.118	0.046
Model 2 negative	0.37	-0.023	0.01	-0.052	0.38	-0.02	0.033	-0.056	0.893	-0.004

In the WQS model, substances with higher weights represent substances that contribute more to the overall impact of the environmental mixture. By constructing weighted indicators to quantify the contribution of each component to the total effect, substances with larger weights play a key role in environmental pollution, perhaps because of their greater toxicity, longer half-life, greater bioaccumulation, or other factors. In our surveyed population, TCC (weight = 0.29), EPB (weight = 0.16) and BPA (weight = 0.11) were the primary metabolites that were chiefly responsible for the positive correlations with SHBG. MHNCH (weight = 0.14), MECPP (weight = 0.13), BP3 (weight = 0.13) and ETP (weight = 0.1) were negatively associated with E2. MEOHP (weight = 0.2), MBP (weight = 0.1), TCC (weight = 0.1), PPB (weight = 0.1) and BPA (weight = 0.09) were the primary metabolites implicated in the adverse correlations with the FAI (Table S3).

The BKMR model was utilized to evaluate the connection between chemical concentrations and male sex steroid hormones

Figure 2 shows statistically significant associations among the concentrations of these 25 chemicals.

Fig. 2.

Pairwise Spearman correlations among urinary concentrations of 25 chemicals or metabolites in the population (N = 1262), NHANES, USA, 2013–2016.

The analysis using the BKMR, as shown in Fig. 3, revealed that the combined effect of chemical mixtures at or above the 55th percentile was inversely associated with the levels of E2 and total TT. Chemical mixture concentrations below the 50th percentile were negatively correlated with SHBG, whereas those above the 50th percentile were positively correlated with overall SHBG. When the concentration of the chemical mixture was greater than 50 quantiles, the FAI and TT/E2 were negatively correlated. Figure 3 illustrates the exposure‒response relationships for the chemicals.

Fig. 3.

Overall effect of the mixture on sex steroid hormones. (A) TT, (B) E₂, (C) SHBG, (D) FAI, (E) TT/E₂ estimated using Bayesian kernel machine regression overall effect (BKMR).

The PIPs identified through the BKMR model are detailed in Supplementary Table S2. The BKMR model identified MECPP (PIP 0.21), MEHHP (PIP 0.20), and MEOHP (PIP 0.16) as significant contributors to the associations observed with TT. BP3 (PIP 0.15) and MEHP (PIP 0.11) were shown to be relatively important chemicals for E2. MCOP (PIP 0.42) and MECPP (PIP 0.28) were important contributors to the associations with SHBG.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

LUO wrote the main manuscript text, ZHAO made substantial contributions to the conception or design of the work, DONG acquisition, analysis, or interpretation of data. All authors reviewed the manuscript.

Data availability

The datasets generated and/or analysed during the current study are available in the NHANES repository, https://www.cdc.gov/nchs/nhanes/index.htm.

Declarations

Competing interests

The authors declare no competing interests.