Use of Generalized Weighted Quantile Sum Regressions of Tumor Necrosis Factor Alpha and Kidney Function to Explore Joint Effects of Multiple Metals in Blood

Exposure to heavy metals could lead to adverse health effects by oxidative reactions or inflammation. Some essential elements are known as reactors of anti-inflammatory enzymes or coenzymes. The relationship between tumor necrosis factor alpha (TNF-α) and heavy metal exposures was reported. However, the interaction between toxic metals and essential elements in the inflammatory response remains unclear. This study aimed to explore the association between arsenic (As), cadmium (Cd), lead (Pb), cobalt (Co), copper (Cu), selenium (Se), and zinc (Zn) in blood and TNF-α as well as kidney function. We enrolled 421 workers and measured the levels of these seven metals/metalloids and TNF-α in blood; kidney function was calculated by CKD-EPI equation. We applied weighted quantile sum (WQS) regression and group WQS regression to assess the effects of metal/metalloid mixtures to TNF-α and kidney function. We also approached the relationship between metals/metalloids and TNF-α by generalized additive models (GAM). The relationship of the exposure–response curve between Pb level and TNF-α in serum was found significantly non-linear after adjusting covariates (p < 0.001). Within the multiple-metal model, Pb, As, and Zn were associated with increased TNF-α levels with effects dedicated to the mixture of 50%, 31%, and 15%, respectively. Grouped WQS revealed that the essential metal group showed a significantly negative association with TNF-α and kidney function. The toxic metal group found significantly positive associations with TNF-α, serum creatinine, and WBC but not for eGFR. These results suggested Pb, As, Zn, Se, and mixtures may act on TNF-α even through interactive mechanisms. Our findings offer insights into what primary components of metal mixtures affect inflammation and kidney function during co-exposure to metals; however, the mechanisms still need further research.


Introduction
Heavy metals, for instance, arsenic (As), cadmium (Cd), and lead (Pb), are widely utilized in industry [1]. Based on the International Agency for Research on Cancer, Pb was grouped as a metal that is probably carcinogenic to humans (group 2), and As and

Study Population
The participants for this research were 462 workers enrolled in annual health examinations between 2009 and 2010 in the Kaohsiung Medical University Hospital in Taiwan. We excluded subjects with rheumatoid arthritis, cancer, and chronic kidney disease, those <20 or >80 years old, those who failed to offer specimens, those with eGFR <60 mL/min/1.73 m 2 , and those with unfinished questionnaires or incomplete data for covariates. The flowchart for subjects is displayed in Figure 1. This protocol was approved by the Ethics Committee of Kaohsiung Medical University Hospital (approval number: KMUHIRB-G(I)-20210005; date of approval, 19 February 2021). Each participant provided their informed consent before enrollment. The questionnaire collected information on the following demographic characteristics: gender, age, body mass index (BMI), and lifestyle habits (cigarette smoking, alcohol consumption). Exclusively subjects with whole variables (serum creatinine, white blood counts, seven metals and TNF-α concentrations in serum, and renal function) were included in this study. Finally, 421 participants remained in our database. In our study, the outcome variables were white blood counts, TNF-α levels, serum creatinine levels, estimated glomerular filtration rate (eGFR), and the ratio of TNF-α and white blood cells. All outcome variables were log transformed. The measurement of eGFR was based on the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation, eGFR = 144 × (serum creatinine/0.7) k × (0.993) age [38]. R Statistical Software Version 4.0.3 was used for analysis. GAM and WQS regression used separately the mgcv [39], gWQS [40], and groupWQS [41] packages. renal function. Then, we assessed the relationship between metals and TNF-α levels. Finally, we determined the interrelation between the metal mixtures and TNF-α levels and kidney function.

Study Population
The participants for this research were 462 workers enrolled in annual health examinations between 2009 and 2010 in the Kaohsiung Medical University Hospital in Taiwan. We excluded subjects with rheumatoid arthritis, cancer, and chronic kidney disease, those <20 or >80 years old, those who failed to offer specimens, those with eGFR <60 mL/min/1.73 m 2 , and those with unfinished questionnaires or incomplete data for covariates. The flowchart for subjects is displayed in Figure 1. This protocol was approved by the Ethics Committee of Kaohsiung Medical University Hospital (approval number: KMUHIRB-G(I)-20210005; date of approval, 19 February 2021). Each participant provided their informed consent before enrollment. The questionnaire collected information on the following demographic characteristics: gender, age, body mass index (BMI), and lifestyle habits (cigarette smoking, alcohol consumption). Exclusively subjects with whole variables (serum creatinine, white blood counts, seven metals and TNF-α concentrations in serum, and renal function) were included in this study. Finally, 421 participants remained in our database. In our study, the outcome variables were white blood counts, TNF-α levels, serum creatinine levels, estimated glomerular filtration rate (eGFR), and the ratio of TNF-α and white blood cells. All outcome variables were log transformed. The measurement of eGFR was based on the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation, eGFR = 144 × (serum creatinine/0.7) k × (0.993) age [38]. R Statistical Software Version 4.0.3 was used for analysis. GAM and WQS regression used separately the mgcv [39], gWQS [40], and groupWQS [41] packages.

Biomarkers of Multiple Metals and Inflammation
We collected blood samples at annual health examinations and stored them at 4 • C. We used dilution solution, blending solution with 0.2% ammonia solution, 0.1% Triton X-100, and 0.3% HNO3. All blood samples were analyzed via inductively coupled plasma mass spectrometry (ICPMS) of Thermo Scientific XSERIES 2 in the Kaohsiung Medical University Hospital. We tested the following metal and metalloid concentrations in blood: As, Cd, Pb, Se, Co, Cu, and Zn. The procedure for analyses was customized from the method of Ebba Barany in 2007 [42]. We used two standard reference materials (SEMs), ClinChek ® RECIPE (Munich, Germany) control Level I No. 8880 and Level II No. 8881, to ensure quality control and quality accuracy during analysis. With the SEMs, coefficients of variation were less than 3% for the measurements at high levels (Level II No. 8881) and were less than 5% for the measurements at low levels (Level I No. 8880). The serum from each subject was separated from coagulated blood samples. Serum concentrations of TNF-α were measured by the protocol of Human TNF-alpha Quantikine ELISA Kit measurement (R&D Systems).
The white blood count and serum creatinine levels were analyzed in the clinical biochemical laboratory in the Kaohsiung Medical University Hospital. All laboratories in the University Hospital were qualified by the Taiwan Accreditation Foundation, proving their quality assurance and quality control (QA/QC).

Statistical Analysis
Characteristics of participants were presented as number (%) or mean ± standard deviation (SD). Pearson correlations were followed to demonstrate the association between metals and metalloids levels and TNF-α. Multiple linear regression models were used to examine the association between serum TNF-α level and the seven elements. Model 2 was adjusted covariates for gender, age, BMI, alcohol consumption, and smoking status. The equation of the multivariable linear regression model was: We implemented GAM to survey the dose-response assessment between metals/metalloids and the ratio of TNF-α and white blood cells, adjusted for gender, age, BMI, alcohol consumption, and smoking status. In GAM analysis, the value of the effective degree of freedom (EDF) was greater than 1, which means a complicated relationship between response and predictor. We used smooth functions to predict and plot the relationships between metals/metalloids and the ratio of TNF-α and white blood cells in multivariable GAM.
To identify the association of seven elements and explore the weights of the individual elements, we applied WQS regression. The typical WQS model is as follows: The WQS index represents the weight of each element component. We utilized gWQS regression to certify the association of the mixture with TNF-α, WBC, serum creatinine, the ratio of TNF-α and white blood cells, and eGFR. Estimating the weighted index of the components (e.g., scored into deciles or quartiles) in bootstrap samples of the random training subset and assessing the p-value of the weighted index by a holdout dataset in a generalized linear model are crucial parts of gWQS regression. To reduce collinearity and dimensionality, the weights of components are controlled to be 0 to 1. We can identify the prime contribution to the associations between mixtures of levels of As, Cd, Pb, Co, Cu Se, Zn, and TNF-α in serum.
We also employed the grouped WQS regression to identify the health effects of different mixtures. As, Cd, and Pb are known nephrotoxicants [43]. In this model, all elements were divided into toxic metals (As, Cd, Pb) and essential metals (Co, Cu, Se, Zn). We summed up the results using odds ratios and 95% confidence intervals for each element or mixture and evaluated the weights of metal/metalloid levels in the WQS model.

Results
The demographic characteristics of the study are shown in Table 1 age was 39.8 years old. The means for the blood elements, TNF-α levels, and eGFR are also presented in Table 1. Pearson correlation coefficients, which measured the relationship between each element and TNF-α levels, were from −0.05 to 0.68, as shown in Figure 2. The highest correlation coefficient is 0.68 (p < 0.001), between Pb and TNF-α levels.

Discussion
In our study, we found an association between mixtures of metals/metalloids, TNF-α, and renal function. We utilized different statistical strategies to examine the effects of serum metals and their mixtures. Based on the viewpoint of each element, the multivariable linear regression examined single metal/metalloids and adjusted covariates. This study also found that serum Pb, Se, and Co levels expressed a non-linear relationship with TNF-α and eGFR. Our main finding was the significant relationship between the weights of all types of mixtures and TNF-α and eGFR among workers in metal industries. In addition, the grouped WQS regressions provided quantifiable insight into the individual weights of the mixtures between essential metals and toxic metals. In grouped WQS regressions, the toxic metals significantly increased the level of TNF-α compared with the essential metals.
Whether in the occupational area or general environment, an individual may possibly be exposed to multiple metals everywhere simultaneously [44,45]. Hengstler et al. showed the interaction between the odds of a high level of DNA single-strand breaks (DNA-SSB) and co-exposure to metals, including lead, cobalt, and cadmium. [46]. In addition, early studies reported interactions between essential metals and toxic metals [47]. There is some support in the literature for the current findings, especially regarding the exposure to multiple heavy metals [48,49]. Karri et al. reported significant differences in the expression of protein information between heavy metal mixtures and single metals and demonstrated that the health risks of metal mixture exposure are a greater threat than those of single-metal exposure [50]. Among toxic and essential metals, Turksoy et al. found that the levels of Se and Zn affected the concentrations of TNF-α, IL-6, and IL-10 [51]. Even excessive levels of essential elements may induce adverse health effects. We observed this relationship between Se and Co with TNF-α in the GAM model. Intriguingly, we only observed a significant association between age, TNF-α level, and the ratio of TNF-α level and WBC in the essential metal mixture models.
We observed a relationship between co-exposure to the toxic metal mixture (Pb, As, and Cd) as measured in serum and lower eGFR. Sanders et al. found associations between metals (Hg, As, Cd, and Pb) in urine with eGFR, but there was no significant difference between blood metal mixtures in the general population by WQS regression [52]. Research showed high levels of blood Pb and Cd were associated with renal function (adjusted odds ratios: 2.369, 95%: 1.868-3.004) in a general population of adults from the National Health and Nutrition Examine Survey (NHANES) dataset [53]. Our findings on the toxic mixture model are similar to those identified by Luo et al., who reported Pb was the major association with eGFR compared to the other metals in the mixture model [54]. The protective mechanisms of Zn modifying the relationship between metallothionein and kidney function in vivo have been reported [55], but we did not find similar associations between either the essential metal mixture or any of the other metal mixtures. It is known that the kidney is an organ susceptible to the toxicity of metals due to its ability to reabsorb and condense metals. The degree of harm relies upon the nature, portion, course, and length of openness which might be increased when openness happens across various metals [56]. Tsaih et al. found that longitudinal decreases in renal capacity among middle-aged and older people seem to rely upon both long-term lead exposure and circulating lead, with an impact that is generally articulated among those who were diagnosed with metabolic syndrome, i.e., subjects who were probably a susceptible population [57]. In the three mixture models, we observed a significant association between serum creatinine and gender as well as eGFR and age.
These findings for single metals were in line with previous studies. The levels of TNF-α were affected by occupational exposure of workers to Pb and Cd (median serum TNF-α: 56 pg/mL) versus workers who were not exposed (median serum TNF-α: 22.4 pg/mL) [58,59]. The associations between biomarkers and TNF-α in the study were consistent with other studies which have examined TNF-α levels following exposure to metal/metalloids [10,11]. As exposure increased significant inflammation (serum TNF-α: 5.21 pg/mL) in workers compared to the control group, but Se was associated with lower TNF-α levels in workers [10]. We observed an association between combined exposure to Pb, As, and Zn, as measured in serum, and the level of TNF-α. In our linear regression of single metals, serum As, Cd, Pb, Co, Cu, and Zn levels were significantly associated with TNF-α, but there were no associations with Se. These findings emphasize the possibility of unique health impacts from multiple metal/metalloid co-exposures that would otherwise go undetected in single-metal investigations as well as the efficacy of WQS regression in detecting complicated relationships. We further used the ratio of TNF-α and WBC to model the inflammatory response by WQS regression, in which the component of Se was higher than Zn compared to the outcome of TNF-α level only. The level of Se has the property of alleviating Pb harm through diminishing oxidative pressure via the nucleotide-binding domain, leucine-rich-containing family, and pyrin-domain-containing-3 (NLRP3) inflammasome, according to an animal study [60]. In a cross-sectional study, the authors recommend a decrement in the synthesis of TNF-α; however, our results show the inverse [61]. Future research should examine the contribution of multiple metal/metalloid exposure to immune cell counts (lymphocytes, monocytes, neutrophils) and other health effects among different populations.
A strength of our study was that the WQS regressions and grouped WQS regressions were used to examine the effect of metal/metalloid mixtures on inflammation and renal function in workers. In addition, seven metal/metalloids and TNF-α levels in serum were surveyed to comparatively and comprehensively express internal exposure. We acknowledge additional study limitations. We cannot determine causal relationships between the levels of serum for multiple elements and inflammatory biomarkers, serum creatinine, and renal function in this cross-sectional study. The lack of information on seafood consumption and supplements could lead to a gap in our results [62]; however, compared to occupational exposure, the effect of seafood was limited. On the other hand, we used metals in the blood as biomarkers that included every source of metals, including food, seafood, and water. In addition to these limitations, serum or urine cotinine may be a better biomarker for assessing smoking exposure, rather than only a questionnaire [63]. Another limitation was that we did not collect urine samples from workers; therefore, we could not investigate the species forms of As in urine, such as dimethylarsinic acid, monomethylarsonic acid, arsenocholine, and arsenobetaine.

Conclusions
This study demonstrated that metal/metalloid mixtures in serum were associated with TNF-α levels and eGFR by GAM. Heavy metals are still one of the crucial environmental hazards associated with inflammation and kidney health. The effects of metal co-exposure are likely modified by the concentration and duration of exposure, genetic predisposition, and other individual factors. Further epidemiological studies in a framework with more inflammatory cytokines and biomarkers of oxidative stress and controlled environmental and individual factors would increase the clarity regarding the mechanisms of the metal/metalloid mixtures in occupational exposure.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/ijerph19127399/s1, Figure S1: Forest plot of single-metal effects for TNF-α after adjusting covariates. Figure S2: Forest plot of single-metal effects for TNF-α after adjusting covariates and WBC. Figure S3: Forest plot of single-metal effects for WBC after adjusting covariates. Figure S4: Forest plot of single-metal effects for serum creatinine after adjusting covariates. Figure S5: Forest plot of single-metal effects for eGFR after adjusting covariates. Figure S6: Forest plot of single-metal effects for the ratio of TNF-α and WBC after adjusting covariates. Figure S7: Covariate plots after adjustments for the mean predicted value of the toxic metals mixture index in serum and measures for TNF-α (a), WBC (b), and serum creatinine (c), TNF-α/WBC (d), and eGFR (e). Figure S8: Covariate plots after adjustments for the mean predicted value of the essential metals mixture index in serum and measures for TNF-α (a), WBC (b), and serum creatinine (c), TNF-α/WBC (d), and eGFR (e). Figure S9: Covariate plots after adjustments for the mean predicted value of the seven metals mixture index in serum and measures for TNF-α (a), WBC (b), serum creatinine (c), TNF-α/WBC (d), and eGFR (e); Table S1: Associations of single serum metal levels with TNF-α using regression models. Table S2: Associations of single serum metal levels with WBC using regression models. Table S3: Associations of single serum metal levels with serum creatinine using regression models. Table S4: Associations of single serum metal levels with estimated glomerular filtration rate (eGFR) using regression models. Table S5: Associations of single serum metal levels with the ratio of TNF-α and WBC using regression models. Table S6. Results of toxic metals mixture by WQS regression analyses with covariates. Table S7 Results of essential metals mixture by WQS regression analyses with covariates. Table S8. Results of all metals mixture by WQS regression analyses with covariates.

Institutional Review Board Statement:
The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board at Kaohsiung Medical University Hospital (KMUHIRB-G(I)-20210005; date of approval, 19 February 2021). Approval was waived for individual consent forms due to de-identification in the Taiwan Biobank of data and specimens.
Informed Consent Statement: Individual explanation with consent form was conducted for each subject.

Data Availability Statement:
The data could be applied to use via an application proceeding at Kaohsiung Medical University.