Inuencing Trend and Extent of Combined Exposure Levels of Total Arsenic and Inorganic Arsenic on Arsenic Methylation Capacity Among University Students: Findings from Bayesian Analysis Using Kernel Machine Regression

The arsenic (As) methylation capacity is an important determinant of the susceptibility to arsenic-related diseases. Total As (TAs) or inorganic As (iAs) was reported to associate with As methylation capacity individually, however, inuencing trend and extent of their combined exposure levels on methylation capacity remains poorly understood. We measured urinary concentrations of iAs, monomethylarsonic (MMA), and dimethylarsinic (DMA) acids using HPLC-HG-AFS, and calculated the primary (PMI: (MMA+DMA)/TAs) and secondary (SMI: DMA/(MMA+DMA)) methylation capacity indexes in 209 university students in Hefei, China, a non-As endemic area. Subjects were given with a standardized questionnaire to inquire their sociodemographic characteristics. Bayesian kernal machine regression (BKMR) analysis was used to estimate the association of lnTAs and lniAs levels with methylation indices (ln%MMA, ln%DMA, lnPMI, lnSMI). The median concentration of iAs, MMA and DMA was 1.22, 0.92 and 12.17 μg/L, respectively; the proportions of iAs, MMA and DMA were 8.76%, 6.13% and 84.84%, respectively. Females had higher %DMA and lower %MMA, while males had lower %DMA and higher %MMA. The combined levels of lnTAs and lniAs showed monotonic decrease in change of ln%DMA and lnSMI other than ln%MMA, additionally, change in ln%PMI was decreased only when levels of lnTAs and lniAs were larger than their 60th percentiles compared to they were at 50th percentile. With regard to single exposure level, the lnTAs showed positive correlation with ln%DMA, lnPMI, lnSMI when lniAs was set at a specic level; while lniAs showed negative correlation with ln%DMA, lnPMI, lnSMI when lnTAs was set at a specic level; and all the dose-response relationships were nonlinear. Our results suggested that combined levels of TAs and iAs after PMI logarithm The univariate linear regression models analysis was conducted.


Introduction
Inorganic arsenic (iAs) is a geogenic contaminants widely distributed in the environment and commonly identi ed in the groundwater, and the WHO stipulated that the guideline value of iAs concentration levels in drinking water is less than As exists in different chemical forms in nature, the two major iAs species are As and As . In humans, the primary iAs metabolic pathway is methylation, which could take place in most organs of the body, but mainly in the liver (Drobná and Walton et al. 2010; Rahman and Hassler 2014). At present, the metabolism of iAs in the human body has been suggested through two pathways (Sattar and Xie et al. 2016;Mochizuki 2019). However, whatever the metabolic pathway, most of the As and its metabolites are excreted in urine, and a typical urinary As methylation trait contains 10%-30% iAs, 10%-20% MMA, and 60%-80% DMA (Wei and Yu et al. 2016). Therefore, urinary As metabolite concentrations and their proportions are considered as reliable biomarkers of As exposure and As methylation capacity.
Epidemiological studies suggested that high levels of iAs and MMA in urine and low levels of DMA were associated with the incidence of As-related diseases, which included some cancers, heart disease, skin lesion and preschool children's developmental delay and so on ( Consequently, growing interest has been focused on the factors that in uence As methylation capacity. As exposure level was one of important determinants for As methylation capacity, so some previous studies have explored effect of TAs and/or iAs level on As methylation capacity among populations in As endemic area by linear regression analysis (Torres-Sánchez and López-Carrillo et al. 2016; Yang and Chai et al. 2017; Olmos and Astolfo et al. 2021), however, in fact, the relationship between them might not be linear, and and this relationship among populations in a non-As endemic area is not clear. Additionally, iAs, as a part of TAs in urine, correlates signi cantly with TAs, they should not be separated when assessing their in uence on As methylation capacity. Therefore, a more appropriate analysis method should be performed to explore this relationship. Bayesian kernal machine regression (BKMR) analysis, as a new statistical method, could give a more accurate relationship between one variable and an outcome while controlling an other variable at a speci c level which we also wanted to study (Bobb and Valeri et al. 2015). In addition, it could evaluate an nonlinear relationship between exposure and outcome (Bobb and Valeri et al. 2015).
In the present study, we conducted a cross-sectional study to investigate As methylation capacity among some university students in a non-arsenic endemic area, and observe the in uencing trend and extent of combined exposure levels of TAs and iAs on As methylation capacity.

Study participants
We randomly recruited volunteers including postgraduates and undergraduates from a medical university in Anhui province, China; in addition, the volunteers must meet the following conditions strictly to ensure that diet and water were major sources of As exposure: eating only in the university canteen for the last two weeks, drinking only municipal water for the last two weeks and not taking medicine or herbs which may contain As for the last two weeks. In total, we enrolled 244 volunteers in the present study and 209 valid participants completed the entire study who completed the questionnaires validly and provided effective urine samples. The study was approved by the ethics committee of Anhui Medical University; in addition, oral and written consent was obtained from all subjects before this study begins.

Urine Samples collection and pretreatment
Morning midcourse urine samples from the participants after getting up and not having breakfast or doing exercise were collected and placed into 10 mL polypropylene plastic tubes respectively. Then the urine samples were sent for determination in time or stored at -80℃ when they could not be detected in time. The determination must be completed within one week after sampling to prevent the transformation of As speciations in urine. The fresh urine was ltered directly to remove impurities with microporous membrane (0.45µm i.d.), and the frozen urine was balanced at 4℃ for 2 hours before ltration. The urine after ltration could be determined directly.

Standards and reagents
All the glass needed for the experiment should be soaked in HNO 3 solution (9:1) for 24 hours, then rinsed with water repeatedly, and washed with ultrapure water nally. Up-s (ultrapure) grade HNO 3  which were all high-grade pure. The experimental water was ultrapure water (18.2 MΩ.cm).

As speciations determination
The As speciations in urine was determined by high performance liquid chromatography-hydride generation-atomic  Table S1.

Stability of As speciation
The limit of detection (LOD) of As speciations ranged from 0.218 µg/L to 0.738 µg/L. The recoveries of added standard were ranged from 83.88-96.63% (Table S2). The results of precision test and stability test were shown in Table S3 and  Table S4, respectively.

Determination of TAs in food
In order to explain the source of As in the urine among this group of university students, we purchased a batch of food and drinking water from the canteen and detected the TAs in these samples by hydride generation-atomic uorescence spectrometry (HG-AFS). Before determination, the weight of the sample was weighed rstly, and then these samples was digested with the prepared digestive solution (HNO 3 : H 2 O 2 = 3:1) for more than 24 hours. After the sample was completely digested, the acid was heated and driven out with the graphite furnace digestion instrument (LabTech, America). The 5% CH 4 N 2 O 4 S and 5% C 6 H 8 O 6 were used as pre-reduced catalyst, 5% HCl was used as carrier uid, 20 g/L KBH 4 and 3.5 g/L KOH were used as reductants.

Statistical analysis
In our present study, As V in all urine samples could not be detected, so iAs represented merely As III . Concentrations of iAs, MMA, DMA and TAs were used as urine As speciation indicators. The percentages of iAs (%iAs), MMA (%MMA) and DMA (%DMA) were de ned as iAs/TAs×100%, MMA/TAs×100% and DMA/TAs×100%, respectively. Two methylation indices, PMI ((MMA + DMA)/TAs) and SMI (DMA/(MMA + DMA)) were also calculated to assess As methylation capacity (Ren and Xu et al. 2019).
The study participants' demographic characteristics were presented as the Mean ± standard deviation (SD) or number (frequency, %), and the information on the daily food matching and weekly exercise was presented as number (frequency, %). The concentration distributions of iAs, MMA, DMA and TAs were described using selected percentiles.
The linear regression models were performed to compare the differences of %iAs, %MMA, %DMA, TAs, PMI and SMI between categories of demographic variables after the %iAs, %MMA, %DMA, TAs, PMI and SMI were natural logarithm transformed.
Bayesian kernal machine regression (BKMR) analysis was used to estimate the association of lnTAs and lniAs levels with methylation indices (ln%MMA, ln%DMA, lnPMI, lnSMI). BKMR could not only study the linear or nonlinear effects of single factors on methylation indices, but also analyze the possible combined effects among multiple factors (Bobb and Valeri et al. 2015). The exposure-response function does not demand a priori speci cation and is modeled neatly, which are its main features. The BKMR model is indicated by the equation: Yi = h{TAsi, iAsi} +βq Zi + ei (Liang and Han et al. 2020). The function h{} in the equation is an exposure-response function, which can accommodate nonlinearity and/or interaction between different exposure levels (lnTAs, lniAs) in mixture; and Z = Z1, Z2 …… Zq was q potential confounders including gender, age, grade, daily food matching, fruit frequency in the last week, time being spent on strenuous physical activity in the last week; and the Gaussian kernel function is applied to simulation researches and real-life scenarios. Models were run up to 10,000 iterations. Firstly, the cumulative effect of combined exposure level to lnTAs and lniAs on urinary As methylation indices of university students was evaluated. Secondly, when lnTAs or lniAs was xed at the 25th, 50th, 75th percentile, the effect of another factor level on the As methylation indices were calculated. Finally, we investigated univariate exposure-response function and 95% con dence intervals for TAs and iAs with another one xed at the median.
All data were statistical analyzed using SPSS for Windows (version 22.0; SPSS UK Ltd., Surrey, UK) and R (version 4.0.5, package "bkmr" and "ggplot2"), and a two-sided p-value < 0.05 was considered statistically signi cant for all the tests unless otherwise indicated.

Results
In our present study, 209 university students completed the questionnaires validly and provided effective urine samples, which included 116 (55.50%) males and 93 (44.50%) females. The mean students' age was 19.94 ± 2.37 years, their mean BMI was 21.42 ± 3.01 kg/m 2 , and most participants (61.24%) were normal weight. There were 118 (56.46%) nonapproaching graduation students and 91 (43.54%) approaching graduation students; 96 (45.93%) students came from rural areas and 113 (54.07%) students came from town (Table 1). The distribution of iAs, MMA, DMA and TAs in urine were shown in Table 2. The detection rates of iAs (As ), MMA and DMA in urine among 209 university students were 100%, 96.65% and 72.73%, respectively. The concentration of DMA was the highest among all the measured As speciations and its median concentration was 12.17 µg/L, the median concentration of iAs was 1.22 µg/L, and the median concentration of MMA was 1.18 µg/L ( Table 2). The proportion distribution of %iAs, %MMA, %DMA in urine was shown in Table 3. The %DMA was the highest, followed by %iAs and %MMA. Abbreviations: TAs, total urinary arsenic; iAs, inorganic arsenic (It mainly refers to the As III , since the As V has not been detected in this population); MMA, monomethylarsonic acid; DMA, dimethylarsinic acid. The association of demographic features, eating habits with As speciations in urine among 209 university students were shown in Table 4. The lnTAs concentration in urine of students over 20 years old was 0.31 µg/L lower than that of students under 20 years old (p < 0.01). Similar to the age, the urine lnTAs concentration of approaching graduation students was 0.34 µg/L lower than that of non-approaching graduation students (p < 0.01). Compared with the males, ln%MMA in urine of females was lower while ln%DMA and lnSMI was higher, and the difference was statistically signi cant. Dietary habits were only correlated with the urine lnTAs. The lnTAs level of students with balanced diet was higher than that of students with unbalanced diet, but the difference was only statistically signi cant when compared to those with less meat and more vegetables (p < 0.01). There was also a signi cant correlation between fruit consumption and urinary lnTAs level, so were water consumption. The more frequent fruit consumption, the lower lnTAs concentration, and the more water consumption, the lower lnSMI.  The univariate linear regression models analysis was conducted.
*, P value < 0.05; **, P value < 0.01; Ref, reference group; TAs, total urinary arsenic; iAs, inorganic arsenic (It mainly refers to the As III , since the As V has not been detected in this population); MMA, monomethylarsonic acid; DMA, dimethylarsinic acid; PMI, the Primary Methylation Index; SMI, the Secondary Methylation Index; ln, the natural logarithm transformed. The univariate linear regression models analysis was conducted.
*, P value < 0.05; **, P value < 0.01; Ref, reference group; TAs, total urinary arsenic; iAs, inorganic arsenic (It mainly refers to the As III , since the As V has not been detected in this population); MMA, monomethylarsonic acid; DMA, dimethylarsinic acid; PMI, the Primary Methylation Index; SMI, the Secondary Methylation Index; ln, the natural logarithm transformed.
The estimated joint effect of lnTAs and lniAs on As methylation indices (ln%MMA, ln%DMA, lnPMI, lnSMI) are shown in Fig. 1-4. We rstly displayed some numerical summaries of their overall effect, which was identi ed as the change in As methylation indices (ln%MMA, ln%DMA, lnPMI, lnSMI) associated with a simultaneous change in lnTAs and lniAs from a particular percentile as compared to when lnTAs and lniAs were at their median values (50th percentile) (Fig. 1A, Fig. 2A, Fig. 3A, Fig. 4A). As can be seen from Fig. 1A, combined exposure was not associated with ln%MMA. When both lnTAs and lniAs levels are above the 60th percentile, the lnTAs and lniAs were signi cantly negatively correlated with ln%DMA; on the contrary, when both lnTAs and lniAs levels are below the 55th percentile, a signi cantly positive correlation was observed when compared to the 50th percentile, and the similar result was also found about lnSMI.
While compared with the 50th percentile, the lnTAs and lniAs were positively correlated with lnPMI only at 35th, 40th and 55th percentile. The variation in the association of lnTAs or lniAs when it increased from 25th to 75th percentile accompanied by another factor was set at 25th, 50th or 75th percentile with As methylation indices (ln%MMA, ln%DMA, lnPMI, lnSMI), respectively was shown in Fig. 1B, Fig. 2B, Fig. 3B and Fig. 4B. LnTAs displayed a signi cantly positive association with ln%DMA and lnPMI levels when lniAs was set at the 25th, 50th, and 75th percentiles. And lniAs displayed a signi cantly negative association with ln%DMA, lnPMI, and lnSMI levels when lnTAs is set at the 25th, 50th, and 75th percentiles, and displayed a signi cantly positive association with ln%MMA. Finally, we also examined the potential nonlinear exposure-response relationship when another factor was held at the corresponding median concentration by BKMR analysis. We found that lnTAs was positively correlated with ln%DMA, lnPMI and lnSMI.
Considering the study participants are university students and environment As exposure level is not very serious in Anhui province (Zhong and Zhang et al. 2019), we thought that their urinary As are mainly source from food and drinking water. Therefore, we measured the TAs content in the food and drinking water from the canteen. The results showed that the TAs content in cereals was the highest among all the our purchased foods, with an average of 90.340 µg/kg, while TAs in fruits, vegetables and meats were similar to be lower. The mean concentration of TAs in drinking water was 0.284 µg/L (Table 5).

Key ndings
In summary, these students exposed to As generally and widely; the levels of iAs, MMA, DMA and TAs had gender difference. With respect to As methylation capacity, gender and exposure level of TAs and iAs were important determinants. Females seem to have stronger As methylation capacity than males; combined levels of TAs and iAs played an important role in reducing As methylation capacity, expecially iAs; and the reduction only occured when TAs and iAs were up to a certain combined level. . Both the percentages of MMA and iAs were lower in our study showed that university students in our study had stronger As methylation capacity.

Demographic characteristics and eating habits
The percentage of As metabolites in urine is commonly used to re ect the As methylation capacity. Previous studies suggested that variability in As methylation capacity was due, in part to variations in gene polymorphisms, as well as gender, age, BMI, smoking, drinking alcohol and eating habits (Tseng 2009;Wnek and Medeiros et al. 2009; Ren and McHale et al. 2011).
In the present study, TAs and %MMA in the urine of male was higher than that of female, while the ln%DMA and lnSMI was lower than that of female. Our ndings were consistent with previous study researched by Huang  is not the case in the present study, and we only found that diet (food matching daily) was associated with TAs. However, our study didn't assess the real nutritional status of the participants since a food frequency questionnaire was not used to nd out all information of those substances capable to modify both metabolism and toxicity of As.

TAs and iAs levels
We explored the correlation between urinary TAs and iAs levels with the indexes of As methylation capacity by BKMR analyses. The results showed that there were positive correlations between TAs with the three As methylation indices (%DMA, PMI, SMI), and negative correlation with %MMA; however, iAs presented an opposite trend. It should be noted that both PMI and SMI were correlated negatively with combined levels of TAs and iAs. These results reminded us that iAs may played a more important role in reducing As methylation. Our ndings differed from a study from Xu et al. (

Strengths and limitations
This study explored in uencing trend and extent of combined exposure levels of total arsenic and inorganic arsenic on arsenic methylation capacity among university students using BKMR analysis for the rst time. Nevertheless, our study still has several limitations. Firstly, spot urine samples could re ect only short-term exposure, and the time of day at sample collection may in uence the observation of %iAs, %MMA and %DMA. In order to minimize this in uencing effect, morning urine samples were collected at the same time every day during the execution stage of study. Secondly, the dilution degree of urine was not corrected by the speci c gravity of urine creatinine, although it doesn't have any impact on the As methylation capability. Thirdly, we didn't have data about complete dietary in university students, only determined a few common TAs contents of some food, it's an important factor for As methylation capacity. Finally, this was a cross-sectional study, which could only explain the correlation between the As methylation capacity and the in uencing factors investigated, and the causal conclusion could not be obtained.

Conclusions
In the present study, we found that these students exposed to As generally and widely; the levels of iAs, MMA, DMA and TAs had gender difference, and females seem to have stronger As methylation capacity than males. The combined exposure levels of TAs and iAs played an important role in As methylation, expecially iAs, and the reduction only occured when TAs and iAs were up to a certain combined level. The impact of TAs and iAs on As methylation capacity should be payed more attention in the future.

Declarations
Joint effect of the total arsenic and inorganic arsenic in urine on ln%MMA by Bayesian Kernel Machine Regression (BKMR) Model was adjusted for gender, age, grade, daily food matching, fruit frequency in the last week and time being spent on strenuous physical activity in the last week. Fig. 1a: overall effect of the lnTAs and lniAs (estimates and 95%CI, gray dashed line at the null ). This plot compared the ln%MMA level when lnTAs and lniAs were at a particular quantile to when they were at the 50th percentile, respectively. Fig. 1b: independent association of lnTAs or lniAs (estimates and 95%CI, gray dashed line at the null). This plot compared the ln%MMA level when lnTAs or lniAs was at the 75th vs. 25th percentile, when another factor was xed at either the 25th, 50th or 75th percentile. Fig. 1c: univariate exposureresponse function and 95% con dence bands for lnTAs or lniAs with another factor being xed at the median.

Figure 2
Joint effect of the total arsenic and inorganic arsenic in urine on ln%DMA by Bayesian Kernel Machine Regression (BKMR) Model was adjusted for gender, age, grade, daily food matching, fruit frequency in the last week and time being spent on strenuous physical activity in the last week. Fig. 2a: overall effect of the lnTAs and lniAs (estimates and 95%CI, gray dashed line at the null ). This plot compared the ln%DMA level when lnTAs and lniAs were at a particular quantile to when they were at the 50th percentile, respectively. Fig. 2b: independent association of lnTAs or lniAs (estimates and 95%CI, gray dashed line at the null). This plot compared the ln%DMA level when lnTAs or lniAs was at the 75th vs. 25th percentile, when another factor was xed at either the 25th, 50th or 75th percentile. Fig. 2c: univariate exposureresponse function and 95% con dence bands for lnTAs or lniAs with another factor being xed at the median. Figure 3