Arsenic, one carbon metabolism and diabetes-related outcomes in the Strong Heart Family Study

Abstract

Background:

Inorganic arsenic exposure and inter-individual differences in its metabolism have been associated with cardiometabolic risk. A more efficient arsenic metabolism profile (lower MMA%, higher DMA%) has been associated with reduced risk for arsenic-related health outcomes; however, this profile has also been associated with increased risk for diabetes-related outcomes. The mechanism behind these contrasting associations is equivocal; we hypothesized one carbon metabolism (OCM) may play a role.

Methods:

We evaluated the association between OCM-related variables (nutrient intake and genetic variants) and both arsenic metabolism biomarkers (iAs%, MMA% and DMA%) and diabetes-related outcomes (metabolic syndrome, diabetes, HOMA2-IR and waist circumference) in 935 participants free of prevalent diabetes and metabolic syndrome from the Strong Heart Family Study, a family-based prospective cohort comprised of American Indian tribal members aged 14+ years.

Results:

Of the 935 participants free of both diabetes and metabolic syndrome at baseline, 279 (29.8%) developed metabolic syndrome over a median of 5.3 years of follow-up and of the 1,458 participants free of diabetes at baseline, 167 (11.3%) developed diabetes over follow-up. OCM nutrients were not associated with arsenic metabolism, however, higher vitamin B6 was associated with diabetes-related outcomes (higher HOMA2-IR and increased risk for diabetes and metabolic syndrome). A polymorphism in an OCM-related gene, methionine synthase (MTR), was associated with both higher MMA% (β=2.57, 95% CI: 0.22, 4.92) and lower HOMA2-IR (GMR=0.79, 95% CI=0.66, 0.93 per 5 years of follow-up). Adjustment for OCM variables did not affect previously reported associations between arsenic metabolism and diabetes-related outcomes; however, the association between the MTR variant and all diabetes-related outcomes were attenuated or reversed direction after adjustment for arsenic metabolism

Conclusions:

Our findings suggest MMA% may be a partial mediator in the association between OCM and diabetes-related outcomes. Formal mediation analyses with longer follow-up period are needed to confirm this finding. Additional research is needed to determine whether excess B vitamin intake is associated with increased risk for diabetes-related outcomes.

1. INTRODUCTION

One carbon metabolism (OCM) is a network of interrelated biochemical reactions critical to the biosynthesis of purines and thymidylate as well as the generation of methyl groups.¹ The transfer of a methyl group to substrates is essential to multiple biological processes including arsenic metabolism.¹ OCM functioning, and adequate methyl group availability, is dependent on essential nutrients that support this pathway including folate, vitamin B₁₂, vitamin B₆, vitamin B₂, and methionine. Therefore, OCM status in epidemiologic and in vivo studies has been evaluated through the measurement of both circulating levels and dietary intake of these nutrients. Low OCM nutrient status in observational studies has been associated with numerous adverse health effects^2-8 including diabetes-related outcomes.^9-12 However, findings on the efficacy of OCM nutrient supplementation in clinical trials to reduce these outcomes have been mixed. ^13-15 Just two trials have evaluated the effect of OCM supplementation (one evaluating folic acid alone,¹⁶ one evaluating a B vitamin combination (folic acid, B₆ and B₁₂) pill¹⁷) on diabetes: results were null except in one high risk group (hypertensive obese participants) sub-analysis.^{16, 17} Smaller supplementation trials focused on folic acid, on the other hand, have suggested supplementation may improve diabetes indicators and complications among diabetes cases.^18-22 Still, some observational studies have suggested high circulating folate levels may actually have deleterious health effects on diabetes-related outcomes, even in the offspring of mothers with high folate during pregnancy.^23-27 Studies evaluating the effect of intake of folate from foods or other OCM nutrients on diabetes-related outcomes are limited, mostly cross-sectional in design and findings are even more conflicting,^{9, 12, 28-35} highlighting the need for more research on this relationship. Single nucleotide polymorphisms (SNPs) in genes encoding enzymes involved in OCM have also been associated with diabetes-related outcomes^36-44 and provide a useful, hypothetically unbiased alternative assessment of OCM status.

Evidence suggests arsenic metabolism may also be affected by OCM status, as has been recently reviewed.⁴⁵ Methylation reactions involved in the metabolism of inorganic arsenic (arsenate and arsenite (iAs)) into mono- and di-methylated arsenicals (MMA and DMA) require the transfer of a methyl group generated by the conversion of S-adenosylmethionine (SAM) to S-adenosylhomocysteine (SAH) in the OCM pathway.¹ Typically, arsenic metabolism in epidemiologic studies is measured as each relative percentage of iAs (iAs%), MMA (MMA%) and DMA (DMA%) out of the three urinary concentrations summed together. DMA has a shorter circulating half-life and is more rapidly excreted than iAs or MMA,⁴⁶- ⁵¹ thus, a urine profile reflecting higher percentages of DMA is generally considered a more efficient arsenic metabolism profile. Randomized clinical trials conducted in highly arsenic-exposed Bangladeshi populations have successfully showed folic acid supplementation can enhance arsenic metabolism, reflected in increases in DMA% and decreases in MMA%, likely through supporting efficiency of the OCM pathway.^{52, 53} Observational cross-sectional studies also suggest OCM nutrients may enhance arsenic metabolism, including in the Strong Heart Study (SHS), the parent study (comprised of older participants exposed to higher arsenic and recruited 10 years earlier, pre-folic acid fortification) to the cohort this analysis was conducted in, the Strong Heart Family Study (SHFS).^54-56 Further, OCM SNPs have been found to influence arsenic metabolism as well.^57-60 This is highly important as arsenic metabolism, specifically higher MMA%, has been identified as a risk factor for arsenic-related health outcomes, including cardiovascular disease, skin lesions and cancer.^61-68 However, higher MMA% has also been associated with a reduced risk for diabetes-related outcomes. Indeed, in the SHS and SHFS, higher MMA% has been associated with lower risk for diabetes,⁶⁹ metabolic syndrome⁷⁰ and elevated waist circumference, as well as lower body mass index (BMI)⁷¹ and homeostasis model assessment for insulin resistance (HOMA2-IR)⁷². The mechanism underlying the contrasting association between arsenic metabolism and diabetes-related outcomes versus other arsenic-related health outcomes is unclear.

The goal of this study was to characterize the role OCM plays in the association between arsenic metabolism and diabetes-related outcomes in an attempt to better understand the contrasting relationships between arsenic metabolism and these outcomes versus other arsenic-related health outcomes. Previous research suggests OCM is strongly tied to both arsenic metabolism and diabetes-related outcomes, making it an ideal candidate to potentially provide a mechanistic pathway between the two variables. In this study, we first evaluated the associations between OCM variables (SNPs in: PEMT, MTR, MTRR, MTHFR, MTHFD1, SHMT1 and CBS; and nutrient intake: vitamin B₆, vitamin B₂ and folate) and both arsenic metabolism (iAs%, MMA% and DMA%) as well as four diabetes-related outcomes (metabolic syndrome, waist circumference, diabetes, and HOMA2-IR). Diabetes-related outcomes were selected based on previously reported associations between arsenic metabolism and these outcomes in the Strong Heart Study and SHFS.^{69, 72, 73} We then evaluated whether OCM influences previously reported associations between arsenic metabolism and diabetes-related outcomes in this study population. We used data from participants in the Strong Heart Family Study, a prospective family-based cohort study of cardiometabolic disease in American Indian tribal members from Arizona, Oklahoma, and North/South Dakota.

2. METHODS

2.1. Study population

The SHFS recruited 2,919 participants in 1998-1999 (n=428) and 2001-2003 (n=2,491) with follow-up visits conducted in 2006-2009. For this study, we did not include baseline visits conducted in 1998-1999 as OCM nutritional information was only conducted during visits in 2001-2003. We included participants with sufficient urine for arsenic analyses (n=1,719) and who were free of diabetes at baseline (n=1,661). We used urinary arsenic and OCM nutrient data measured from visits conducted in 2001-2003, and diabetes-related outcome (metabolic syndrome, waist circumference, HOMA2-IR, diabetes) data from follow-up visits in 2006-2009. We excluded participants missing information on outcome data (n=91), OCM genetic variant data (n=21) or OCM nutrient status (n=82). We also excluded participants with missing data on education, smoking, alcohol intake, BMI, and estimated glomerular filtration rate (eGFR) (n=9), resulting in 1,458 participants available for analyses evaluating HOMA2-IR, waist circumference or incident diabetes. For incident metabolic syndrome analyses, we further excluded participants with prevalent metabolic syndrome (n=523), resulting in 935 participants available for metabolic syndrome analyses (Figure 1). All participants provided informed consent and study protocols were approved by multiple institutional review boards, participating communities and The Indian Health Service.

Figure 1.

Study Flow Diagram

2.2. Data Collection

Baseline and follow-up visits included bio-specimen collection, physical exam, food frequency questionnaire and an interview-administered questionnaire (age, sex, education, smoking history, alcohol use, medical history).⁷⁴ Exam measures (waist circumference, blood pressure, height, weight) and collection of urine and fasting blood samples were performed by centrally trained nurses following a standardized protocol.⁷⁵

2.3. Urine arsenic determinations

Morning spot urine samples were collected in polypropylene tubes, frozen within 1 to 2 hours of collection, shipped buried in dry ice and stored at −70°C in the Penn Medical Laboratory, MedStar Research Institute, Washington, DC for up to 18 years. For arsenic analyses, urine samples were thawed and up to 1.0 mL from each urine sample was transferred to a small vial, transported on dry ice to the Trace Element Laboratory at Graz University, Austria and stored at −80°C until analyses.

Total urine arsenic concentrations were measured using inductively coupled plasma-mass spectrometry (ICPMS), as has been described in detail previously.⁷⁶ The urine concentrations of arsenite, arsenate, MMA, DMA and arsenobetaine were measured using high performance liquid chromatography/- ICPMS (HPLC/ICPMS). Limits of detection were 0.1 μg/L for all five species. Arsenic species concentrations below the limit of detection (<5% for all species) were imputed as the limit of detection divided by √2. Arsenobetaine levels were low (median=0.52, IQR=0.33-1.13 μg/L), confirming infrequent seafood consumption.

2.4. OCM Nutrient Collection

Dietary intake of OCM-related micronutrients was measured during the baseline visit through estimated daily averages of dietary intake of vitamins B₆ and B₂ and folate in the past-year. These variables, as well as total caloric intake, were measured through an interviewer-administered Block 119-item food frequency questionnaire (FFQ). The Block questionnaire is one of the most widely used questionnaires with demonstrated reliability and validity.⁷⁷ To enhance accuracy of the questionnaire in this cohort, additional questions relating to foods commonly consumed by American Indians were added.⁷⁷

2.5. OCM SNP Selection and Genotyping

DNA was extracted from blood specimens obtained at the baseline visit using organic solvents and was genotyped according to Illumina protocol⁷⁸ using the Illumina Cardio-Metabo DNA Analysis BeadChip (MetaboChip), which contains 196,725 markers. These markers were selected based on a large-scale meta-analysis for cardiometabolic traits such as coronary artery disease and type 2 diabetes. Markers included in this analysis have been genotyped previously using strict QC methods⁷⁹ and were selected based on their role in OCM. Single nucleotide polymorphisms (SNPs) were only included if minor allele frequencies were >2%. For genes with multiple SNPs available, SNPs that had been previously associated with arsenic metabolism or diabetes-related outcomes (or SNPs in perfect LD with these SNPs) were selected. SNPs in the following genes were included: rs4646371 in phosphatidylethanolamine N-methyltransferase (PEMT), rs3818239 and rs17751556 in methylenetetrahydrofolate dehydrogenase (MTHFD1), rs12952556 and rs2273027 in serine hydroxymethyltransferase 1 (SHMT1), rs1801131, rs1801133 and rs2274976 in methylenetetrahydrofolate reductase (MTHFR), rs10495387 in methionine synthase (MTR), rs3776464 and rs16879334 in methionine synthase reductase (MTRR) and rs12482221 in cystathionine β-synthase (CBS).

2.6. Diabetes-Related Variable Collection and Definitions

Incident type 2 diabetes was defined as fasting plasma glucose ≥ 126 mg/dL, self-reported physician diagnosis or self-reported use of insulin or oral diabetes treatment. Baseline and follow-up HOMA2-IR values were calculated with the computed solved model for HOMA2-IR⁸⁰ using fasting glucose and insulin values. Metabolic syndrome was defined according to the National Cholesterol Education Program ATP III guidelines, the most accepted guidelines for the US.^{81, 82} To be considered as a case of metabolic syndrome, participants had to have at least three of the following criteria: elevated waist circumference (≥40 inches in men and ≥35 inches in women); triglycerides ≥150 mg/dL (or on medication); fasting glucose ≥100 mg/dL (or on medication); HDL cholesterol <40 mg/dL for men and <50 mg/dL for women (or on medication); systolic blood pressure ≥130 mmHg or diastolic blood pressure ≥85 mmHg (or on medication). Metabolic syndrome status was measured at baseline and follow-up visits. Waist circumference was measured at the umbilicus while the participant was in a supine position.⁷⁵ Other measurements for metabolic syndrome status were measured as follows. For systolic and diastolic blood pressure, three measurements were taken on the right arm in seated position after 5 minutes of rest with an appropriately sized cuff using a Baum mercury sphygmomanometer; the average of the last two measurements was used for analyses.⁷⁵ Blood samples collected after 12-hour fasting were used to measure triglycerides, glucose, and cholesterol using enzymatic methods.⁷⁵ High-density lipoprotein cholesterol levels were measured by precipitation with heparin and manganese chloride.⁷⁵

2.7. Sociodemographic and Diabetes-Related Risk Factor Data

The standardized questionnaire included sociodemographic data (age, sex, education, income), smoking history, alcohol use, physical activity, and medical history.⁷⁴ Physical exam measures were performed by centrally trained nurses following a standardized protocol and included measurements of height, weight and body fat.⁷⁵ Estimated glomerular filtration rate (eGFR) was measured from age, sex and recalibrated plasma creatinine using the Chronic Kidney Disease – Epidemiology Collaboration formula.⁸³

2.8. Statistical Analyses

Statistical analyses were conducted in R software (version 3.5.1; R Project for Statistical Computing). Arsenic metabolism was computed by dividing iAs, MMA and DMA over the sum of the three × 100 (iAs%, MMA% and DMA%). Arsenic exposure, calculated as the sum of inorganic and methylated arsenic species (ΣAs) in urine and creatinine-corrected to account for urine dilution, was right-skewed and therefore log-transformed in analyses. OCM nutrients (folate, vitamin B₆ and vitamin B₂) were adjusted for total caloric intake using a residual analysis approach. This approach has been suggested over including caloric intake as a variable in regression models, as the nutrient and total caloric intake variables are measured through the same questionnaire, and therefore, their errors are correlated.⁸⁴ For the residual analysis approach, we regressed each log-transformed vitamin intake on log-transformed total caloric intake. We then added the mean log-transformed value of each nutrient to the nutrient residuals to create calorie-corrected nutrient variables. OCM SNPs were included in analyses using co-dominant inheritance models. For the SNPs MTHFD1 (rs3818239), MTHFR (rs2274976), MTHFD1 (rs17751556) and MTR (rs10495387), the homozygous variant genotype was <10 and therefore grouped with the heterozygous genotype in analyses, resulting in just one comparison group to the wild-type genotype.

Sociodemographic, arsenic, vitamin intake and physiological variables were compared in participants with and without metabolic syndrome and diabetes at follow-up, using chi-square and Kruskal Wallis tests for categorical and continuous variables, respectively. Spearman correlations, as well as percentages of participants above the recommended daily intake (RDA) and upper limit (UL), for nutrient variables were evaluated. The associations between OCM status (genetic- and nutrient-related) and arsenic metabolism were evaluated in participants free of prevalent metabolic syndrome and diabetes (n=935) to remove the potential for the effect metabolic state may have on arsenic metabolism. In OCM nutrient analyses, the mean difference and 95% confidence interval (95% CI) of each arsenic species percentage (iAs%, MMA%, DMA%) per interquartile range (IQR) increase in each calorie-corrected nutrient variable (vitamin B6, vitamin B2 and folate) was estimated using mixed linear regression models to account for family-relatedness. In OCM SNP analyses, mixed linear regression models were used to estimate the mean difference (95% CI) of each arsenic species percentage comparing the heterozygous and homozygous variant genotypes to the wild type genotype (reference) for each OCM-related SNP. Models for OCM nutrients and genetic variants association with arsenic metabolism were adjusted for log-transformed ΣAs, age (continuous), sex, region (Arizona, Oklahoma, North/South Dakota), education (<12 years, 12+ years (or above/below appropriate level of schooling if aged <18 years)), BMI (<25, 25-<30, ≥30 kg/m2), smoking status (never, former, current), alcohol use (never, former, current), and eGFR (continuous). In models evaluating waist circumference, BMI was not included as an adjustment for consistency with previous literature.⁸⁵

The association of OCM status (genetic- and nutrient-related) was also evaluated with each diabetes-related outcome. For diabetes, HOMA2-IR and waist circumference outcomes, analyses were conducted in participants free of baseline diabetes (n=1458). In metabolic syndrome analyses, analyses were conducted in participants free of baseline diabetes and baseline metabolic syndrome (n=935). For diabetes and metabolic syndrome, our dichotomous outcomes, we used modified Poisson regression with robust variance⁸⁶ using generalized estimating equations with an independence working correlation structure to account for family clustering. Results were reported as relative risk (RR) and 95% confidence intervals (95% CI) of incident diabetes and metabolic syndrome per IQR and quartile increase in calorie-corrected OCM nutrients as well as comparing the heterozygous and homozygous variant genotypes to the wild-type genotype (reference) for each OCM SNP. For our continuous outcomes, HOMA2-IR and waist circumference, we conducted multi-level models (MLM) in which both HOMA2-IR (log-transformed) and waist circumference values at baseline and at follow-up were treated as the outcome and the linear predictor included the interaction of our OCM variable and time since baseline (in years). Specifically, the time variable included two values for each participant: time=0 and time=follow-up duration (in years). In this way, we were able to estimate the GMR of HOMA2-IR and mean difference of waist circumference and 95% CIs by OCM status variables at follow-up (considering 5 years of follow-up, i.e. time=5). Again, we used mixed effects linear regression models to account for family clustering.

Models evaluating the association between OCM SNPs and diabetes-related outcomes were adjusted for age, sex and four of ten population stratification principal components (which have been used previously^{87, 88} as they describe a significant amount of variation in the population). Models evaluating the association between OCM nutrients and diabetes-related outcomes were adjusted for age, sex, center, education, BMI, kidney function, smoking status and drinking status.

In previous analyses in this cohort, arsenic metabolism has been associated with risk for diabetes, metabolic syndrome and elevated waist circumference, as well as higher HOMA2-IR.^{73, 89} We re-ran these analyses in this study before and after adjustment for OCM variables to evaluate whether OCM status might be confounding these associations. We also re-ran final models evaluating the association between OCM nutrients and SNPs with diabetes-related outcomes after adjusting for arsenic metabolism (i.e., MMA%) and ΣAs, to evaluate whether arsenic metabolism may act as a partial mediator in these associations. Finally, based on our findings, we conducted an exploratory formal mediation analysis, evaluating the direct, and indirect through MMA%, effects of the MTR SNP on HOMA2-IR at follow-up.

Multiple sensitivity analyses were conducted to better understand the association between vitamin B6 and diabetes-related outcomes. First, we ran Spearman correlations between our calorie-corrected B6 variable and dietary food groups (processed meat, red meat, fried chicken, cereals, vegetables and fruits) to determine if there were any high correlations that might potentially be confounding the associations between vitamin B₆ and diabetes-related outcomes; we then used additional adjustments for these food groups separately in the final model. We also tested for potential confounding by niacin by adjusting for it in vitamin B₆ and diabetes-related outcome models. Further, we ran models using raw vitamin values with separate adjustment for total caloric intake instead of calorie-corrected vitamins. Finally, we stratified analyses by sex.

3. RESULTS

3.1. Participant Characteristics

Of the 1,458 participants free of diabetes at baseline, 167 (11.3%) developed diabetes over a median of 5.3 years of follow-up, and of the 935 participants free of both diabetes and metabolic syndrome at baseline, 279 (29.8%) developed metabolic syndrome over follow-up. Participants who developed diabetes or metabolic syndrome at follow-up were older, had higher DMA%, BMI, HOMA2-IR and waist circumference, and lower MMA% (p≤0.05) compared with participants that did not develop diabetes or metabolic syndrome (Table 1). Participants who developed diabetes were also exposed to higher levels of ΣAs than those that did not develop diabetes. Participants who developed diabetes or metabolic syndrome did not differ by vitamin intake or other sociodemographic variables. All calorie-corrected nutrients were significantly correlated with each other (p = <0.001) (Supplemental Table 1). The majority of participants were below the RDA for vitamin B₂ and folate intake, but above the recommended dietary allowance (RDA) for vitamin B₆ intake.

Table 1.

Baseline Participant Characteristics of American Indians in the Strong Heart Family Study by Incident Metabolic Syndrome and Diabetes Status at Follow-up, 1998-2009

	No MetS at Follow-up	MetS at Follow-up	P- Value	No Diabetes at Follow-up	Diabetes at at Follow-up	P- Value
Total, n (%)	656(70.2)	279(29.8)		1295 (88.8)	163 (11.2)
Age (years), median (IQR)	28.6 (19.4, 41)	33.7 (24.9, 44.1)	<0.001	34.3 (22.4-45.8)	39.2 (29.6-50.7)	<0.001
Sex, n (%)
Female	285(43.4)	114(40.9)	0.51	535(41.3)	64 (39.3)	0.68
Male	371(56.6)	165(59.1)		760(58.7)	99 (60.7)
Education, n (%)
<12 years	171(26.1)	70(25.1)	0.82	313(24.2)	45 (27.6)	0.39
12+ years	485(73.9)	209(74.9)		982(75.8)	118 (72.4)
Smoking Status, n (%)
Never	282(43)	120(43)	0.94	525(40.5)	64 (39.3)	0.87
Ever	119(18.1)	53(19)		264(20.4)	36 (22.1)
Current	255(38.9)	106(38)		506(39.1)	63 (38.7)
Alcohol Intake, n (%)
Never	85 (13)	27(9.7)	0.21	140(10.8)	19 (11.7)	0.11
Ever	137(20.9)	69(24.7)		320(24.7)	52 (31.9)
Current	434(66.2)	183(65.6)		835(64.5)	92 (56.4)
eGFR mL/min/1.73 m2, n(%)
< 60	2 (0.3)	2 (0.7)	0.737	9 (0.7)	1 (0.6)	1.00
≥ 60	654(99.7)	277(99.3)		1286 (99.3)	162 (99.4)
Caloric Intake (kcal), median (IQR)	2228 (1537, 3470)	2271 (1426, 3252)	0.32	2222 (1516-3345)	2294 (1561-3309)	0.92
Vitamin B6 (mg)
Intake, median (IQR)	1.7 (1.1, 2.8)	1.7 (1.1, 2.5)	0.40	1.6 (1.1-2.6)	1.9 (1.1-2.7)	0.26
Supplement Use, n (%)	149(22.7)	77(27.6)	0.13	314(24.2)	39 (23.9)	1.00
Vitamin B2 (mg)
Intake, median (IQR)	1.7 (1.1, 2.6)	1.6 (1.1, 2.5)	0.85	1.7 (1.1-2.7)	1.9 (1.2-2.7)	0.45
Supplement Use, n (%)	149(22.7)	77(27.6)	0.13	314(24.2)	39 (23.9)	1.00
Folate (μg)
Intake, median (IQR)	356(226, 550)	355(222, 514)	0.49	355(232-548)	409 (248-602)	0.25
Supplement Use, n (%)	160(24.4)	85(30.5)	0.06	340(26.3)	42.0 (25.8)	0.97
ΣAs (μg/L), median (IQR)	4.4 (2.9, 7.1)	4.7 (3.1, 7.5)	0.21	4.5 (3.0-7.3)	5.3 (3.2-9.4)	0.006
iAs%, median (IQR)	10.9 (7.5, 15.5)	9.9 (7.3, 13.5)	0.04	10.3 (7.0-14.2)	9.0 (6.7-13.0)	0.04
MMA%, median (IQR)	15.9 (12.6, 19.9)	14(10.9, 17.8)	<0.001	14.8 (11.3-18.5)	13.1 (10.3-16.8)	<0.001
DMA%, median (IQR)	72 (65.5, 78.5)	74.6 (68.9, 80.8)	<0.001	74.2 (67.3-80.4)	78.0 (70.7-82.1)	0.001
BMI (kg/m2), n (%)
Normal	276(42.1)	46(16.5)	<0.001	333(25.7)	9 (5.5)	<0.001
Overweight	205(31.2)	92(33)		385(29.7)	28 (17.2)
Obese	175(26.7)	141(50.5)		577(44.6)	126 (77.3)
HOMA2-IR, median (IQR)	1.1 (0.8, 1.6)	1.5 (1-2.2)	<0.001	1.3 (0.9-2.1)	2.3 (1.5-3.7)	<0.001
Waist Circumference (cm), median (IQR)	90.0 (80.0, 99.0)	99.0 (91.0, 111)	<0.001	97 (87-109)	111 (100-123)	<0.001

Abbreviations: ΣAs (iAs+MMA+DMA); BMI (body mass index); dimethylarsinic acid (DMA); eGFR (estimated glomerular filtration rate); inorganic arsenic (iAs); IQR (interquartile range); metabolic syndrome (MetS); monomethylarsonic acid (MMA)

3.2. Association between OCM Variables and Arsenic Metabolism

In mixed linear regression models, calorie-corrected OCM vitamins (B6, B2 and folate), were not associated with any arsenic metabolism variables (iAs%, MMA%, DMA%) in adjusted models (Table 2). In mixed linear regression models evaluating the association between OCM SNPs and arsenic metabolism, carriers of one or two A alleles in the MTR rs10495387 polymorphism was associated with 2.57% (95% CI: 0.22, 4.92) higher MMA% and non-significantly associated with 4.08% (95% CI -8.32, 0.17) lower DMA% (Table 3). No other OCM SNPs were associated with arsenic metabolism.

Table 2.

Mean Difference in Arsenic Metabolism by IQR Increase (95% CI) in One Carbon Metabolism-Related Nutrient Intake (N=935)

Variable	Model 1	Model 2	Model 3
iAs%
Vitamin B6	0.06(−0.50, 0.61)	0.05 (−0.48, 0.59)	0.16 (−0.36, 0.68)
Vitamin B2	0.23(−0.32, 0.78)	0.07 (−0.48, 0.61)	0.15 (−0.41, 0.70)
Folate	−0.52 (−0.98, −0.06)	−0.21 (−0.65, 0.24)	−0.12 (−0.56, 0.33)
MMA%
Vitamin B6	−0.03 (−0.52, 0.46)	−0.09 (−0.56, 0.38)	−0.08 (−0.53, 0.37)
Vitamin B2	0.47(−0.01, 0.94)	0.26 (−0.20, 0.72)	0.26 (−0.19, 0.70)
Folate	−0.43 (−0.85, 0.00)	−0.15 (−0.56, 0.25)	−0.14 (−0.52, 0.24)
DMA%
Vitamin B6	−0.03 (−0.92, 0.87)	0.04 (−0.80, 0.88)	−0.08 (−0.89, 0.73)
Vitamin B2	−0.70 (−1.55, 0.16)	−0.33 (−1.16, 0.51)	−0.40 (−1.23, 0.42)
Folate	0.95(0.19, 1.70)	0.36 (−0.35, 1.07)	0.26 (−0.43, 0.94)

Model 1: adjusted for log creatine-adjusted urinary total arsenic

Model 2: Model 1 adjustments plus age, sex, center

Model 3: Model 2 adjustments plus education, kidney function, smoking status, drinking status, BMI

Abbreviations: DMA (dimethylarsinic acid); iAs (inorganic arsenic); monomethylarsonic acid (MMA)

Table 3.

Mean Difference (95% CI) in Arsenic Metabolism by One Carbon Metabolism-Related Genetic Polymorphisms (N=935)

Variable	n	iAs%	MMA%	DMA%
CBS (rs12482221)
AA	365	0.00(Reference)	0.00(Reference)	0.00(Reference)
AG	409	0.15(−0.7, 0.99)	−0.18 (−0.89, 0.52)	0.04(−1.28, 1.36)
GG	161	−0.42 (−1.62, 0.79)	−0.29 (−1.23, 0.66)	0.70(−1.12, 2.52)
SHMT1 (rs12952556)
AA	637	0.00(Reference)	0.00(Reference)	0.00(Reference)
AG	269	0.42(−0.51, 1.36)	0.03(−0.67, 0.72)	−0.45 (−1.75, 0.84)
GG	29	0.80(−1.63, 3.23)	1.54(−0.64, 3.73)	−2.35 (−6.56, 1.87)
SHMT1 (rs2273027)
AA	286	0.00(Reference)	0.00(Reference)	0.00(Reference)
AG	463	0.51(−0.42, 1.45)	−0.21 (−1.04, 0.62)	−0.31 (−1.82, 1.21)
GG	186	0.00(−1.18, 1.19)	−0.33 (−1.31, 0.64)	0.33(−1.50, 2.15)
PEMT (rs4646371)
GG	360	0.00(Reference)	0.00(Reference)	0.00(Reference)
GA	452	−0.44 (−1.35, 0.47)	0.18(−0.56, 0.92)	0.26(−1.13, 1.65)
AA	123	−0.47 (−1.89, 0.94)	0.08(−0.99, 1.15)	0.39(−1.74, 2.52)
MTRR (rs3776464)
AA	513	0.00(Reference)	0.00(Reference)	0.00(Reference)
AT	358	0.13(−0.79, 1.06)	0.25(−0.43, 0.93)	−0.38 (−1.75, 0.98)
TT	64	0.01(−1.72, 1.75)	0.25(−1.72, 1.75)	−0.26 (−3.09, 2.57)
MTRR (rs16879334)
CC	586	0.00(Reference)	0.00(Reference)	0.00(Reference)
CG	306	0.13(−0.79, 1.06)	0.28(−0.41, 0.96)	−0.41 (−1.78, 0.95)
GG	43	0.01(−1.72, 1.75)	0.26(−1.28, 1.80)	−0.27 (−3.10, 2.55)
MTHFR (1801131)
CC	586	0.00(Reference)	0.00(Reference)	0.00(Reference)
CT	306	0.07(−0.85, 0.98)	−0.28 (−1.02, 0.46)	0.21(−1.17, 1.60)
TT	43	−1 (−2.84, 0.85)	−0.97 (−2.46, 0.52)	1.97(−0.46, 4.39)
MTHFR (1801133)
AA	473	0.00(Reference)	0.00(Reference)	0.00(Reference)
AC	394	0.07(−0.73, 0.87)	0.42(−0.29, 1.13)	−0.49 (−1.79, 0.81)
CC	68	1.18(−0.62, 2.98)	1.02(−0.42, 2.46)	−2.20 (−4.82, 0.43)
MTHFR (2274976)
GG	790	0.00(Reference)	0.00(Reference)	0.00(Reference)
GA/AA	145	−0.89 (−2.03, 0.25)	−0.75 (−1.68, 0.18)	1.64(−0.10, 3.39)
MTHFD1 (rs3818239)
AA	838	0.00(Reference)	0.00(Reference)	0.00(Reference)
AG/GG	77	−1.08 (−2.32, 0.15)	−0.40 (−1.42, 0.62)	1.48(−0.38, 3.34)
MTHFD1 (rs17751556)
AA	870	0.00(Reference)	0.00(Reference)	0.00(Reference)
AG	97	−0.32 (−2.00, 1.35)	0.65(−0.76, 2.06)	−0.32 (−3.05, 2.40)
MTR (rs10495387)
CC	892	0.00(Reference)	0.00(Reference)	0.00(Reference)
CA/AA	43	1.51(−0.64, 3.66)	2.57(0.22, 4.92)	−4.08 (−8.32, 0.17)

Models adjusted for age, sex, ΣAs, 4 genetic PCs

Abbreviations: DMA (dimethylarsinic acid); iAs (inorganic arsenic); monomethylarsonic acid (MMA)

3.3. Association between OCM Variables and Diabetes-Related Outcomes

In mixed modified Poisson regression models, IQR increases in calorie-corrected vitamin B6 were associated with greater risk for incident metabolic syndrome (RR=1.15; 95% CI=1.02, 1.29) and incident diabetes (RR=1.28; 95% CI=1.10, 1.48) (Table 4). Vitamin B₆ was also borderline associated with higher HOMA2-IR in mixed multi-level models (GMR=1.04; 95% CI=1.00, 1.08 per 5-years of follow-up). Results were similar categorizing vitamin B₆ into quartiles. Results were consistent in sensitivity analyses stratifying by sex; using non-calorie-corrected vitamin B₆ values and adjusting separately for caloric intake in models; adjusting for niacin intake; and further adjusting final models separately for intake of fried chicken, processed meat, red meat, vegetables and fruit (data not shown). Calorie corrected vitamin B₂ was also associated with increased risk for incident diabetes (RR per IQR increase in vitamin B₂=1.21; 95% CI=1.03, 1.41). Folate was not significantly associated with any diabetes-related outcome.

Table 4.

Association between One Carbon Metabolism-Related Nutrient Intake and Diabetes-related Outcomes

Variable	Relative Risk for Metabolic Syndrome (n=935)	Mean Difference in Waist Circumference (N=1458)	Relative Risk for Diabetes (n=1458)	GMR for HOMA2-IR (n=1458)
Vitamin B6 (mg)a
<1.2	1 (Reference)	(Reference)	1 (Reference)	1 (Reference)
1.2 – <1.8	1.17 (0.89 1.53)	1.18 (−1.21, 3.57)	1.28 (0.82 2.00)	1.07 (0.98, 1.17)
1.8 – <2.7	1.08 (0.81 1.43)	1.33 (−1.08, 3.74)	1.00 (0.62 1.63)	0.98 (0.89, 1.07)
≥2.7	1.40 (1.06 1.85)	0.47 (−1.94, 2.88)	1.72 (1.13 2.61)	1.07 (0.98, 1.18)
IQR	1.15 (1.02 1.29)	0.03 (−1.01, 1.07)	1.28 (1.10 1.48)	1.04 (1.00, 1.08)
Vitamin B2 (mg)a
<1.2	1 (Reference)	(Reference)	1 (Reference)	1 (Reference)
1.2 – <1.8	1.08 (0.80 1.44)	0.13 (−2.26, 2.51)	1.23 (0.80 1.90)	0.97 (0.88, 1.06)
1.8 – <2.8	1.26 (0.96 1.65)	0.16 (−2.25, 2.57)	1.17 (0.75 1.82)	1.00 (0.91, 1.09)
≥2.8	0.93 (0.69 1.27)	−1.83 (−4.24, 0.58)	1.53 (1.00 2.34)	1.02 (0.93, 1.12)
IQR	1.01 (0.89 1.14)	−0.95 (−2.01, 0.12)	1.21 (1.03 1.41)	1.01 (0.97, 1.05)
Folate (μg)a
<234	1 (Reference)	(Reference)	1 (Reference)	1 (Reference)
234 - <355	0.83 (0.63 1.10)	1.48 (−0.89, 3.86)	1.08 (0.71 1.63)	0.95 (0.87, 1.04)
355 - <560	1.03 (0.78 1.35)	2.11 (−0.27, 4.50)	1.35 (0.88 2.05)	1.00 (0.91, 1.09)
≥560	0.97 (0.74 1.26)	−0.70 (−3.13, 1.73)	1.27 (0.83 1.94)	1.05 (0.96, 1.15)
IQR	1.02 (0.90 1.15)	−0.93 (−1.96, 0.11)	1.15 (0.97 1.36)	1.01 (0.97, 1.05)

^aVitamins are calorie-corrected

Models adjusted for age, gender, center, education, BMI, kidney function, smoking status, drinking status.

Abbreviations: GMR (geometric mean ratio);

HOMA2-IR (homeostasis model assessment for insulin resistance)

Three OCM SNPs in CBS, PEMT and MTR genes had significant associations with diabetes-related outcomes. Each addition of the variant G allele in the CBS SNP was associated with an allele-dependent decrease in waist circumference per 5 years of follow-up (βAG= −0.51, 95% CI=−2.36, 1.33; βGG= −0.3.28; 95% CI=−5.82, −0.74) (Table 5). CBS was not associated with any other diabetes-related outcome. For PEMT, each addition of the variant A allele was associated with an allele-dependent increase in both metabolic syndrome (RRGA=1.12, 95% CI= 0.88, 1.43; RRAA=1.37; 95% CI=1.02, 1.85) and waist circumference (β_GA= 1.92, 95% CI=−0.13, 3.71; β_AA= 2.51, 95% CI=−0.14, 5.17 per 5 years of follow-up). A similar trend was observed for diabetes; however, the association was not significant. For MTR, one or two copies of the variant A allele was associated with lower HOMA2-IR (GMR=0.79, 95% CI=0.66, 0.93 per 5 years of follow-up). The trend was consistent for metabolic syndrome and diabetes, but not significant.

Table 5.

Association between One Carbon Metabolism-Related Genetic Polymorphisms and Diabetes-Related Outcomes

Variable	n	Relative Risk for Metabolic Syndrome (N=935)	n	Mean Difference in Waist Circumference (N=1458)	Relative Risk for Diabetes (N=1458)	GMR for HOMA2-IR (N=1458)
Total	935		1458
CBS (rs12482221)
AA	365	1.00(Reference)	558	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
AG	409	1.02(0.81, 1.29)	658	−0.51 (−2.36, 1.33)	1.10 (0.81, 1.51)	0.99 (0.91, 1.07)
GG	161	1.24(0.93, 1.66)	242	−3.28 (−5.82, −0.74)	1.03 (0.67, 1.57)	0.99 (0.89, 1.11)
SHMT1 (rs12952556)
AA	637	1.00(Reference)	1002	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
AG	269	1.02(0.81, 1.28)	410	0.97(−0.91, 2.86)	1.03 (0.75, 1.43)	1.03 (0.95, 1.12)
GG	29	1.07(0.59, 1.93)	46	2.08(−2.72, 6.87)	0.64 (0.22, 1.82)	1.01 (0.82, 1.24)
SHMT1 (rs2273027)
AA	286	1.00(Reference)	454	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
AG	463	0.89(0.71, 1.11)	732	0.24(−1.68, 2.16)	1.24 (0.87, 1.78)	1.05 (0.97, 1.14)
GG	186	1.06(0.79, 1.42)	272	1.19(−1.33, 3.71)	0.93 (0.57, 1.51)	1.08 (0.96, 1.20)
PEMT (rs4646371)
GG	360	1.00(Reference)	584	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
GA	452	1.12(0.88, 1.43)	681	1.92(0.13, 3.71)	1.07 (0.80, 1.44)	1.00 (0.92, 1.08)
AA	123	1.37(1.02, 1.85)	193	2.51(−0.14, 5.17)	1.21 (0.80, 1.84)	1.01 (0.90, 1.14)
MTRR (rs3776464)
AA	513	1.00(Reference)	775	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
AT	358	0.96(0.78, 1.19)	579	−0.09 (−1.91, 1.74)	1.11 (0.79, 1.55)	1.01 (0.93, 1.09)
TT	64	0.86(0.54, 1.36)	104	−0.43 (−3.89, 3.04)	1.13 (0.61, 2.09)	0.90 (0.77, 1.04)
MTRR (rs16879334)
CC	514	1.00(Reference)	776	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
CG	357	0.97(0.79, 1.19)	578	−0.22 (−2.05, 1.60)	1.11 (0.79, 1.56)	1.01 (0.93, 1.09)
GG	64	0.86(0.54, 1.36)	104	−0.50 (−3.96, 2.96)	1.13 (0.61, 2.09)	0.90 (0.77, 1.04)
MTHFR (rs1801131)
CC	586	1.00(Reference)	943	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
CT	306	0.99(0.80, 1.24)	449	0.73(−1.14, 2.59)	0.80 (0.57, 1.13)	1.01 (0.93, 1.09)
TT	43	0.71(0.37, 1.37)	66	2.03(−2.16, 6.22)	0.72 (0.28, 1.80)	1.12 (0.93, 1.34)
MTHFR (rs1801133)
AA	473	1.00(Reference)	748	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
AC	394	1.08(0.89, 1.33)	599	−1.03 (−2.79, 0.73)	0.94 (0.69, 1.28)	0.95 (0.88, 1.03)
CC	68	0.87(0.56, 1.34)	111	−2.34 (−5.61, 0.93)	0.83 (0.47, 1.46)	0.98 (0.85, 1.13)
MTHFR (rs2274976)
GG	790	1.00(Reference)	1235	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
GA/AA	145	1.07(0.82, 1.40)	223	1.21(−1.11, 3.52)	0.91 (0.60, 1.37)	1.08 (0.98, 1.20)
MTHFD1 (rs3818239)
AA	838	1.00(Reference)	1311	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
AG/GG	77	1.00(0.73,1.37)	147	0.95(−1.85, 3.75)	0.94 (0.50, 1.76)	1.03 (0.91, 1.16)
MTHFD1 (rs17751556)
AA	870	1.00(Reference)	1362	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
AG/GG	97	0.68(0.40, 1.17)	96	−1.90 (−5.33, 1.53)	0.91 (0.48, 1.70)	0.94 (0.81, 1.09)
MTR (rs10495387)
CC	892	1.00(Reference)	1390	0.00(Reference)	1.00 (Reference)	1.00 (Reference)
CA/AA	43	0.84(0.54, 1.32)	68	−0.03 (−4.01, 3.94)	0.67 (0.30, 1.50)	0.79 (0.66, 0.93)

Models adjusted for age, sex and 4 population stratification PCs

Abbreviations: GMR (geometric mean ratio); HOMA2-IR (homeostasis model assessment for insulin resistance

Of the three OCM nutrients and twelve OCM SNPs, MTR was the only variable associated with both a diabetes-related outcome (HOMA2-IR) and an arsenic metabolism variable (MMA%). Therefore, we ran additional models evaluating the effect the MTR SNP had on diabetes-related outcomes after adjustment for MMA% and ΣAs. Adjustment for arsenic variables attenuated associations, with the association for HOMA2-IR almost losing significance (GMR=0.83, 95% CI=0.71, 0.99). Further, the very slight negative association between MTR and waist circumference before MMA% adjustment reversed to a slightly positive association after adjustment (β=1.35, 95% CI=−2.51, 5.21) (Table 6). Based on these findings, we also used a mediation model to evaluate the direct, and indirect through MMA%, effects of MTR on HOMA2-IR at follow-up. Of the total −0.184 (95% CI: −0.35, 0.02; P-value: 0.092) reduction in log HOMA2-IR at follow-up, the direct effect of MTR accounted for −0.138 (95% CI −0.341, 0.06; P-value: 0.180) and the indirect effect of MTR through MMA% accounted for −0.046 (95% CI −0.095, 0.00; P-value: 0.036). Finally, we attempted to understand the potential role of OCM as a confounder in the relationship between the previously reported associations with arsenic metabolism (lower MMA% and higher DMA%) and increased risk for diabetes-related outcomes by adjusting separately for OCM nutrients and SNPs. Associations remained consistent before and after adjustments (data not shown).

Table 6.

Association Between MTR and Diabetes-Related Outcomes Before and After Adjustment for Arsenic Metabolism

	Final Model	Final Model + MMA% & ΣAs adjustment
MTR rs10495387
RR of Incident Diabetes (N=1458)
CC	1	1
CA/AA	0.67 (0.30, 1.50)	0.70(0.31, 1.60)
GMR of HOMA2-IR (N=1458)
CC	1	1
CA/AA	0.79 (0.66, 0.93)	0.83(0.71, 0.99)
Mean Difference in Waist Circumference (N=1458)
CC	Reference	Reference
CA/AA	−0.03 (−4.01, 3.94)	1.35(−2.51, 5.21)
RR of Incident Metabolic Syndrome (N=935)
CC	1	1
CA/AA	0.84 (0.54, 1.32)	0.93(0.58, 1.48)

All models adjusted for age, sex and 4 population stratification PCs

Abbreviations: ΣAs (sum of iAs, MMA and DMA); GMR (geometric mean ratio); HOMA2-IR (homeostasis model assessment for insulin resistance); MMA (monomethylarsonic acid); RR (relative risk)

4. DISCUSSION

In this study of American Indian men and women aged ≥14 years, OCM nutrients were not associated with arsenic metabolism, however, higher vitamin B₆ was consistently associated with three of the four diabetes-related outcomes studied (higher HOMA2-IR and increased risk for diabetes and metabolic syndrome). Adjustment for arsenic metabolism in these models did not affect the associations between vitamin B₆ and diabetes related outcomes. Further, previously reported associations between arsenic metabolism and diabetes-related outcomes were not affected by adjustment for any of our OCM (nutrient or genetic) variables. However, the MTR rs10495387 polymorphism was associated with both lower MMA% and higher HOMA2-IR per 5 years of follow-up. After adjustment for MMA% the association between the MTR SNP and all diabetes-related outcomes were attenuated or reversed direction, suggesting MMA% may be a partial mediator in the association between this OCM SNP and diabetes-related outcomes.

Arsenic metabolism has been identified in epidemiological studies as a risk factor for health outcomes associated with arsenic exposure. Indeed, higher MMA% has consistently been associated with greater risk for skin lesions,^90-93 cancer^{68, 94-98} and cardiovascular disease.^99-101 As a result of these findings, clinical trials have been conducted in highly arsenic-exposed regions of the world in order to enhance arsenic metabolism (i.e., increase DMA% and decrease MMA%). In contrast, lower MMA% has been identified as protective of diabetes-related outcomes, including diabetes, increased BMI, elevated waist circumference, higher HOMA2-IR and metabolic syndrome.^{69, 71-73, 102} The mechanism behind these contrasting associations is unclear, but essential to uncover in order to determine if the link between arsenic metabolism and diabetes-related outcomes is an epidemiological artifact or a true association that should be considered in future arsenic health effects prevention trials. Both arsenic metabolism and diabetes-related outcomes are tightly connected to OCM, providing an intriguing potential pathway to explain this equivocal relationship.

We characterized OCM in our study through two different metrics: OCM SNPs and intake of OCM nutrients. Despite inherent limitations to these variables as proxies to true OCM status that can be obtained through direct measurement of circulating OCM metabolites, they do have a distinct advantage: they are not influenced by a participant’s metabolic state allowing for a better understanding of relationship directions. OCM nutrients including vitamin B₆, vitamin B₂ and folate were selected for this study based on their role in OCM as outlined in Figure 2. OCM facilitates the generation of methyl groups, through the activation of methionine to S-adenosylmethionine (SAM), which provides the methyl group for most methylation reactions in the body.¹ S-adenosylhomocysteine (SAH) is formed as a product of these reactions and inhibits OCM by tightly binding to most methylation transferase enzymes.¹ SAH is only removed from these enzymes and hydrolyzed to homocysteine if existing homocysteine is either remethylated to form methionine or rerouted to the transsulfuration pathway.¹⁰³ Vitamins B₆ and B₂ are co-enzymes in the conversion of tetrahydrofolate (THF) to 5,10-methylene-THF and the subsequent conversion of 5,10-methylene-THF to 5- methyl tetrahydrofolate (THF), respectively. Dietary folate can enter the OCM pathway as 5-methyl tetrahydrofolate (THF), which can then transfer a methyl donor to homocysteine to generate either THF or methionine. Vitamin B₆ also plays a role in the transsulfuration pathway acting as a co-factor in the conversion of homocysteine to cystathionine, which is used in the generation of glutathione.¹⁰³

Figure 2. Role of One Carbon Metabolism Related Nutrients and Genes in One Carbon Metabolism Pathway.

Genes are identified in pink and nutrients are identified in green. a) the PEMT gene encodes the PEMT enzyme which converts phosphatidylethanolamine to phosphatidylcholine which supports the remethylation of homocysteine through the choline dependent pathway; b) the CBS gene encodes the CBS enzyme which uses c) vitamin B₆ as a co-factor to convert homocysteine into cystathionine; d) Vitamin B₆ is also used as a co-factor in the e) SHMT1-catalyzed conversion of tetrahydrofolate to 5,10 methylene tetrahydrofolate; f) the MTHFD1 gene encodes a protein with three enzymatic activities, each involved in the conversion of tetrahydrofolate to 5,10 methylene tetrahydrofolate; g) the MTHFR gene encodes the rate limiting enzyme MTHFR which catalyzes the conversion of 5,10 methylene tetrahydrofolate to 5-methyl tetrahydrofolate using h) vitamin B₂, in the form of flavin adenine dinucleotide, as a co-factor; 5-methyl tetrahydrofolate can also enter the one carbon metabolism pathway through intake of i) dietary folate; j) MTR catalyzes the conversion of homocysteine to methionine using k) B₁₂ as a co-factor and 5-methyl tetrahydrofolate as the methyl donor; MTR is reactivated through reductive remethylation by l) MTRR. Abbreviations: CBS (Cystathionine beta synthase); MTHFD1 (Methylenetetrahydrofolate Dehydrogenase); MTHFR (Methylenetetrahydrofolate reductase); MTR (Methionine synthase); MTRR (Methionine synthase reductase); PEMT (Phosphatidylethanolamine N-Methyltransferase), SHMT1 (Serine hydroxymethyltransferase)

Our null results on the association between OCM nutrients and arsenic metabolism contrasts with findings from randomized supplementation trials conducted in highly exposed regions in Bangladesh which have shown folic acid supplementation can enhance arsenic metabolism (increase DMA% and reduce MMA% in urine).^{52, 53} Further, our group has previously reported an association between higher intake of vitamins B₆ and B₂ and enhanced arsenic metabolism in the Strong Heart Study, the parent study to the SHFS comprised of older American Indian tribal members recruited in 1989-1991.⁵⁶ It is possible the lack of association in this study is in part due to the relatively high intake levels of these OCM nutrients compared to the parent study, which occurred prior to mandatory folic acid fortification. Some evidence suggests increasing OCM nutrient intake only has an effect on arsenic metabolism when OCM status is low.^{54, 104} Indeed, a small US-based study evaluating this relationship also reported null findings.¹⁰⁵ However, two more recent larger US studies, with B vitamin levels similar to our population, found significant associations between B vitamin intake and arsenic metabolism. Howe et al reported reductions in MMA% with higher weighted sum of B vitamins.¹⁰⁶ It is possible our lack of findings is a result of missing key vitamins. Howe’s weighted sum was strongly influenced by both B₁₂ and thiamin, vitamins we did not have data on. A sensitivity analysis evaluating a variable representing higher intakes of all three B vitamins using principal components was also not significantly associated with arsenic metabolism (data not shown). Spencer et al found an association between higher B6 and lower iAs%, but an association between higher B12 and higher iAs% and lower DMA%.¹⁰⁷ Together, these contrasting findings highlight the need for additional prospective analyses to better understand the association between B vitamin intake and arsenic metabolism.

We did, however, see strong consistent positive associations between vitamin B₆ and our diabetes-related outcomes, including increased risk for metabolic syndrome and diabetes, as well as higher HOMA2- IR. These findings conflict with numerous observational studies which have found low OCM nutrient status to be associated with adverse health effects^5-7 including cardiometabolic outcomes.^{2-4, 8-11} Low levels of OCM nutrients resulting in impaired OCM functioning, in turn causing elevations in homocysteine, has been identified as a potential mechanism behind these findings.¹⁰⁸ Still, causality has been questioned as folic acid and B-vitamin combination pill supplementation trials have had inconsistent results for reducing cardiovascular-related outcomes,^{13, 14} with mostly no significant benefits reported for diabetes incidence.^{16, 17} Further, some evidence suggests high serum folate may actually increase mortality in diabetes patients.²⁴ Finally, observational associations may also be a result of reverse causality. Animal models have shown diabetes can alter OCM functioning and metabolite levels, including reductions in vitamin B₆.^{109, 110} This could explain the difference in findings, as metabolic state would not necessarily affect intake. Still, it’s unclear why higher B₆ intake would be associated with greater risk for diabetes-related outcomes. We conducted several sensitivity analyses to evaluate whether confounding might explain the association. Our calorie-corrected B₆ vitamin was not associated with caloric intake or other food patterns that have been associated with diabetes related outcomes (e.g, processed meat⁷⁷, cereals/carbohydrates) (Supplemental Figure 1). Models evaluating vitamin B6 and diabetes-related outcomes remained consistent after adjustment for these food patterns as well as stratifying by sex. Although residual confounding cannot be disregarded, it is also possible the association is not spurious. Some studies have suggested mandatory B vitamin fortification, resulting in excess intake, has contributed to the obesity epidemic.^111-113 In support of this hypothesis, The National Academies Scientific Evaluation of Dietary Reference Intakes and its Panel on Folate, Other B Vitamins, and Choline has reported that there is some evidence that the B vitamin niacin, when used to treat patients with hypercholesterolemia, can cause impaired glucose tolerance.¹¹⁴ Finally, the association may be tied to cystathionine: vitamin B₆ is a co-factor in cystathionine production (Figure 2), and cystathionine has been associated with cardiovascular outcomes.¹¹⁵ Additional research in large prospective cohorts is needed to better understand this relationship.

As with OCM nutrients, twelve OCM SNPs in seven genes were selected based on their role in OCM as outlined in Figure 2, as well as their availability in the MetaboChip illumina, a custom genotyping array of over 200,000 SNPs related to metabolic, cardiovascular and anthropometric traits.78 In our study, MTR was the only gene with a SNP significantly associated with arsenic metabolism. One or two of the variant alleles was associated with higher MMA% and non-significantly associated with higher iAs% and lower DMA% compared to the wild-type genotype, among participants free of baseline metabolic syndrome and diabetes. This is consistent with previous studies which have found the variant genotype of a different MTR SNP to be significantly associated with arsenic metabolism in the same direction (higher iAs% and MMA% and lower DMA%).^{57, 116} MTR plays a key role in OCM by catalyzing the remethylation of homocysteine to methionine and variants in the gene have been associated with differences in homocysteine levels.¹¹⁷ The MTR SNP in our study was associated with lower HOMA2-IR over 5 years of follow-up and with non-significant decreases in all three other diabetes-related outcomes. Consistently, a common MTR SNP, A2756G, has also been associated with diabetes-related outcomes, including obesity³⁹ and hypertriglyceridemia.⁴⁰ Because the MTR rs10495387 SNP was associated with both higher MMA% and reduced risk for diabetes-related outcomes, it served as an ideal OCM variable to explore whether OCM influences the association between arsenic metabolism and diabetes-related outcomes. However, adjusting for MTR rs10495387 did not affect any of these previously reported associations. Adjustment for other OCM SNPs as well as OCM nutrients also did not affect any of the previously reported associations, suggesting that in this study population, our measures of OCM were not confounding the relationship between arsenic metabolism and diabetes-related outcomes. In contrast, when we adjusted models evaluating the association between MTR rs10495387 and diabetes-related outcomes for MMA% and ΣAs, associations with diabetes-related outcomes (diabetes, metabolic syndrome and HOMA2IR) were attenuated, providing some evidence MMA% may serve as a partial mediator in the association between OCM and diabetes-related outcomes. This was supported by our exploratory mediation analyses, which showed a significant indirect effect of MTR through MMA% on HOMA2-IR at follow-up. Since a single OCM variable cannot fully represent the pathway functioning, more complex mediation modeling, such as structural equation models that create latent variables for OCM, are needed to confirm, and better understand, these relationships. Further, utilizing Mendelian randomization would provide enhanced understanding of the role OCM SNPs play in these complex intersecting relationships. Finally, we cannot discount the possibility of reverse causality in the observed associations between arsenic metabolism and diabetes-related outcomes. Studies with multiple urine arsenic measurements are needed to understand the effect metabolic state may have on arsenic metabolism.

This study was limited by a reduced sample size for metabolic syndrome analyses due to high prevalence of baseline metabolic syndrome. Our evaluation of dietary OCM intake was limited by lack of measurement of choline, methionine and vitamin B₁₂, all important nutrients in OCM functioning that were not available in the SHFS. Further, despite the strength of FFQ data in its ability to provide OCM nutrient estimates not affected by metabolic state, FFQs are a self-reported dietary assessment tool that is associated with underestimates of intake, particularly energy and proteins.^{118, 119} Assessment of nutritional biomarkers and OCM metabolites as a third measure of OCM function would have enhanced analyses and study interpretations. Finally, multiple urine arsenic measurements would have allowed for better evaluation of direction of associations.

5. CONCLUSIONS

Our study identified vitamin B6 intake as a strong predictor of development of diabetes-related outcomes, a finding that remained consistent in multiple sensitivity analyses. Additional research is needed to better understand these results and determine whether chronic high intake of vitamin B₆ has the potential to disturb metabolic activity. We also identified a SNP in the MTR gene that was associated with both arsenic metabolism and HOMA2-IR. The association between this SNP and diabetes-related outcomes were attenuated after adjustment for arsenic metabolism. Additional research is needed to confirm directions of ssociation; however, these findings add to the evidence that a complex and interconnected relationship exists between arsenic metabolism, OCM and diabetes-related outcomes.

Highlights.

• One carbon metabolism was assessed as B vitamin intake and genetic polymorphisms

• Higher vitamin B₆ intake was associated with increased risk for diabetes outcomes

• A methionine synthase polymorphism was associated with arsenic metabolism

• The same polymorphism was also associated with insulin resistance

• Arsenic metabolism, one carbon metabolism and diabetes appear to be interconnected

Abbreviations:

(ΣAs)

Arsenic exposure measured as urine arsenic levels

(CBS)

Cystathionine β-synthase

(DMA)

Dimethylarsinic acid

(eGFR)

Estimated glomerular filtration rate

(FFQ)

Food frequency questionnaire

(HOMA2-IR)

Homeostasis model assessment for insulin resistance

(iAs)

Inorganic arsenic

(MTR)

Methionine synthase

(MTRR)

Methionine synthase reductase

(MTHFD1)

Methylenetetrahydrofolate dehydrogenase

(MTHFR)

Methylenetetrahydrofolate reductase

(MMA)

Monomethylarsonic acid

(OCM)

One carbon metabolism

(PEMT)

Phosphatidylethanolamine N-methyltransferase

(SAM)

S- adenosylmethionine

(SAH)

S-adenosylhomocysteine

(SHMT1)

Serine hydroxymethyltransferase 1

(SNP)

Single nucleotide polymorphism

(SHS)

Strong Heart Study

(SHFS)

Strong Heart Family Study