Urine Arsenic and Arsenic Metabolites in U.S. Adults and Biomarkers of Inflammation, Oxidative Stress, and Endothelial Dysfunction: A Cross-Sectional Study.

BACKGROUND: Arsenic (As) exposure has been associated with increased risk for cardiovascular disease (CVD) and with biomarkers of potential CVD risk and inflammatory processes. However, few studies have evaluated the effects of As on such biomarkers in U.S. populations, which are typically exposed to low to moderate As concentrations.
OBJECTIVES: We investigated associations between As exposures and biomarkers relevant to inflammation, oxidative stress, and CVD risk in a subset of participants from the New Hampshire Health Study, a population with low to moderate As exposure (n=418).
METHODS: Associations between toenail As, total urine As (uAs), and %uAs metabolites [monomethyl (%uMMAV), dimethyl (%uDMAV), and inorganic (%iAs) species] and plasma biomarkers, including soluble plasma vascular and cellular adhesion molecules (VCAM-1 and ICAM-1, respectively), matrix metalloproteinase-9 (MMP-9), tumor necrosis factor-α, plasminogen activator inhibitor-1 (PAI-1), and urinary oxidative stress marker 15-F2t-isoprostane (15-F2t-IsoP), were evaluated using linear regression models.
RESULTS: Covariate-adjusted estimates of associations with a doubling of urinary As suggested an 8.8% increase in 15-F2t-IsoP (95% CI: 3.2, 14.7), and a doubling of toenail As was associated with a 1.7% increase in VCAM-1 (95% CI: 0.2, 3.2). Additionally, a 5% increase in %uMMA was associated with a 7.9% increase in 15-F2t-IsoP (95% CI: 2.1, 14.1), and a 5% increase in %uDMA was associated with a 2.98% decrease in 15-F2t-IsoP [(95% CI: -6.1, 0.21); p=0.07]. However, in contrast with expectations, a doubling of toenail As was associated with a 2.3% decrease (95% CI: -4.3, -0.3) in MMP-9, and a 5% increase in %uMMA was associated with a 7.7% decrease (95% CI: -12.6, -2.5) in PAI-1.
CONCLUSION: In a cross-sectional study of U.S. adults, we observed some positive associations of uAs and toenail As concentrations with biomarkers potentially relevant to CVD pathogenesis and inflammation, and evidence of a higher capacity to metabolize inorganic As was negatively associated with a marker of oxidative stress. https://doi.org/10.1289/EHP2062.

Introduction

Worldwide, an estimated 200 million individuals are exposed to water arsenic (As) concentrations exceeding the guideline set by the World Health Organization (WHO) (2011). Arsenic is an established human carcinogen that has also been associated with a growing number of noncancer outcomes (NRC 2013). In particular, chronic As exposure has been positively associated with cardiovascular disease (CVD) (Farzan et al. 2015a; James et al. 2015; Moon et al. 2013; reviewed in NRC 2013; Moon et al. 2012). Although the majority of research on As and CVD has been conducted in populations exposed to relatively high As levels, such as in Bangladesh, where concentrations typically exceed the country’s national drinking-water standard of , recent evidence suggests that low to moderate As exposure may be associated with CVD-related morbidity (Gong and O'Bryant 2012; James et al. 2015; Monrad et al. 2017; Moon et al. 2013; Mordukhovich et al. 2009) and mortality (D'Ippoliti et al. 2015; Farzan et al. 2015a; Medrano et al. 2010).

Arsenic-induced CVD may be mediated by increased inflammation and oxidative stress, which adversely affect vascular endothelial function. This inference has been supported by both experimental and epidemiological studies (Burgess et al. 2013; Chen et al. 2007; Engström et al. 2010; Kurzius-Spencer et al. 2016; Lemaire et al. 2015; Ma et al. 2012; Mo et al. 2011; Soucy et al. 2005; Wu et al. 2012, 2014). However, the majority of existing studies utilized high doses of As or evaluated populations exposed to relatively high As concentrations, such as populations in Bangladesh (Chen et al. 2007; Wu et al. 2012). In the United States, most water As concentrations typically fall in the low to moderate range, but higher levels of contamination () have been documented, including in parts of Maine and New Hampshire (Flanagan et al. 2014; USGS 2010). In New Hampshire, 40% of households depend on private wells, and of these wells contain water As concentrations (Karagas et al. 1998).

We previously reported that low to moderate As exposure is associated with increased CVD-related mortality among long-term smokers in the New Hampshire Health Study (NHHS) (Farzan et al. 2015a). The goal of the present project was to evaluate whether As exposure is associated with biomarkers of inflammatory processes, endothelial dysfunction, and oxidative stress that may reflect pathogenic mechanisms relevant to CVD (Mozos et al. 2017; Poredoš and Ježovnik 2015), including plasma matrix metalloproteinase-9 (MMP-9), tumor necrosis (), plasminogen activator inhibitor-1 (PAI-1), soluble vascular and intercellular adhesion molecules (VCAM-1 and ICAM-1, respectively), and urinary (), which to our knowledge has not been evaluated previously in relation to As exposure, in the NHHS. We also evaluated whether the percentages of uAs metabolites are differentially associated with these biomarkers because a reduced capacity to fully metabolize inorganic As (iAs) has been associated with many As-related health outcomes (Steinmaus et al. 2010), including CVD (Chen et al. 2013b). Given the increasing evidence that susceptibility to As toxicity may differ by sex (NRC 2013), we also explored potential differences between men and women.

Methods

Participants

Participants in the present study were controls selected as part of a population-based case–control study of keratinocyte cancer in U.S. adults residing in New Hampshire (Gilbert-Diamond et al. 2013). Controls were randomly selected from population lists provided by a) the New Hampshire Department of Transportation for participants and b) Medicaid and Medicare Services for participants and frequency-matched to keratinocyte cancer case distribution on sex and age. Participants were eligible if they could speak English and had a listed phone number. All participants provided informed consent in accordance with the Committee for the Protection of Human Subjects at Dartmouth College. For the present study, biomarkers were measured in plasma (, MMP-9, PAI-1, ICAM-1, VCAM-1) or urine () samples from a subset of 418 control participants (78% of 535 total controls) from the most recent phase of the NHHS (September 2008 through June 2012) who had available plasma () or urine () samples (or both) for biomarker analysis, along with available measures of either toenail As () or urine As ().

General Characteristics

Sociodemographic details, lifestyle factors, and other general characteristics, including age, education, marital status, tobacco smoke use, monthly alcohol consumption, and self-reported height and weight [used to calculate body mass index (BMI)], were determined by a structured in-person interview, which was typically conducted at the participant’s residence.

Water Arsenic

Tap-water samples were collected from participants’ homes and were tested at the Trace Element Analysis Core at Dartmouth College using inductively coupled plasma mass spectrometry (ICP-MS) with a quadrupole collision cell 7500c Octopole Reaction System ICP mass spectrometer (Agilent) and helium as a collision gas to remove polyatomic interferences, as described previously (Karagas et al. 2001). This method has a detection limit ranging from 0.001 to , with of all samples below the limit of detection. Samples with As concentrations below the limit of detection were set to (limit of ).

Toenail Collection and As Measures

Upon enrollment, participants were mailed instructions and materials for the collection of toenail clippings. Toenails were processed and As concentrations were measured at the Trace Element Analysis Core at Dartmouth College using ICP-MS, as described previously (Davis et al. 2014). This method has a detection limit ranging from 0.02 to , with 9% of samples below the limit of detection. Samples with As concentrations below the limit of detection were set to (limit of ). Toenail As levels were available for all 418 participants.

Urine Collection

Urine-collection kits with instructions and materials necessary to collect first-morning-void urine samples were mailed to participants, who were instructed to refrigerate the urine samples until interviewers collected it later that day. Urine samples were aliquoted and frozen at within 24 h of collection and were shipped on dry ice to the University of Arizona for urine As analyses.

Urine As and %As Metabolites

Urine As was measured using a high-performance liquid chromatography (HPLC) ICP-MS system, as described previously (Gilbert-Diamond et al. 2011). This As speciation method quantifies the concentration of iAs species ( and ) and organic As species (, , and arsenobetaine). This method had detection limits ranging between 0.03 and for individual metabolites, with 0%, 4.8%, and 28.9% of the study population below the detection limit for DMA, MMA, and iAs, respectively. All samples had detectable levels of at least one metabolite after exclusion of arsenobetaine. Total urinary As (uAs) was calculated as the sum of , , , and and was adjusted for specific gravity. The proportion of iAs in urine (%uiAs) was calculated as . The proportion of MMA in urine (%uMMA) and the proportion of DMA in urine (%uDMA) were calculated as follows: [] and [], respectively. Any As metabolite concentrations below the limit of detection were set to (limit of ), before calculating uAs or proportions of metabolites in urine. Urine As measures were available for of the 418 total participants.

Blood Collection

Venous blood samples of were collected in heparinized tubes as described previously (Karagas et al. 1999, 2006) at the time of the interview and were available for analysis for of the 418 total participants. Blood was separated by centrifugation at for 20 min at 4°C, and each component (plasma, red blood cells, buffy coat) was labeled and stored separately at until analysis.

Markers of Inflammation and Endothelial Function

Plasma biomarkers were measured at the DartLab (Geisel School of Medicine at Dartmouth, Hanover, NH). Briefly, ICAM-1, MMP-9, , and VCAM-1 were measured using the Meso Scale Discovery (MSD) multispot assay system with V-PLEX or Ultra-Sensitive kits (Dabitao et al. 2011) according to the manufacturer’s instructions. Standards and spikes were measured in triplicate, and samples were measured once. Plates were read on a Sector Imager 2400. MSD Discovery Workbench analysis software with 4-parameter logistic curve-fitting was used for data analysis. PAI-1 was also measured in plasma at the DartLab, using a MILLIPLEX-MAP® human magnetic bead platform (EMD Millipore) according to the manufacturer’s instructions and as described previously (Farzan et al. 2017).

Plasma samples were randomized across plates. Samples were diluted 1:1,000 for soluble ICAM-1 and VCAM-1, 1:10 for MMP-9, 1:400 for PAI-1, and were left undiluted for . Assays were performed in six separate runs. Approximately 10% of sample wells on each plate were reserved for quality-control samples. Replicates of four different nonstudy plasma samples were included on each plate to account for inter- and intraplate variability; coefficients of variation for these five markers ranged from 7–23% and 5–21%, respectively. All assays were within the range of acceptable variability according to the manufacturer’s standards and are considered generally acceptable in epidemiological biomarker studies (Tworoger and Hankinson 2006). Three individuals had VCAM-1 and ICAM-1 levels that were below detection and were set to (limit of ). A portion of the samples tested for PAI-1 could not be reliably assessed () because their values fell above the range of the standard curve; therefore, these samples were excluded, resulting in a sample size of for PAI-1 analyses.

was measured in urine using a competitive enzyme-linked immunoassay kit (Oxford Biomedical Research) according to the manufacturer’s instructions. Before analysis, urine samples were incubated at 37°C for 2 h with to remove glucuronic acid conjugates on for a more accurate assessment of total . Duplicate samples in a 96-well plate were diluted 1:4 with buffer and were incubated with horseradish peroxidase conjugate at room temperature for 2 h, then washed 3 times. concentrations were determined by adding 3,3',5,5'-tetramethylbenzidine (TMB) substrate and quantifying colorimetric readings at against a standard curve. Readings from duplicates were averaged. Approximately 10% of wells were reserved for quality-control samples and reagent blanks. Replicates of a nonstudy composite urine sample were included on each plate to account for inter- and intraplate variability, which were 7.1% and 6.4%, respectively. concentrations were analyzed in a total of 12 separate plates, completed over four separate runs.

Specific Gravity

Specific gravity (SG) was measured using an Atago Pocket Refractometer. Urine analytes (uAs and urinary ) were adjusted for SG using the following equation: (Duty et al. 2005). In models that simultaneously evaluated uAs and urinary , these analytes were not adjusted for SG; instead, SG was included in the model as a covariate.

Statistical Analyses

Summary statistics were calculated for each variable [ for continuous variables and n (%) for categorical variables] in the whole study sample and separately by sex. To stabilize variances for parametric model assumptions and to reduce the influence of extreme values, transformations were applied to variables with skewed distributions. A natural log transformation [ln(X)] was applied to the following variables to normalize their distributions: uAs, toenail As, BMI, VCAM-1, ICAM-1, MMP-9, , PAI-1, and . Arsenic metabolite variables %uiAs, %uMMA, and %uDMA were left untransformed to facilitate the interpretation of results. Level of educational attainment, marital status, and age were categorized as binary variables (at least college educated vs. less than college educated, married vs. not married, and vs. , respectively). Cigarette smoking status was categorized into three levels: never smoker (reference group), former smoker, and current smoker. Alcohol consumption was categorized into four levels: never drinker (reference group), former drinker, current light drinker () and current heavy drinker (). Fish consumption information was derived from a semiquantitative, self-administered food-frequency questionnaire conducted at the time of the interview and was defined as consuming some type of fish product (tuna, dark fish, other fish, breaded fish, or shrimp) . Batch assignment, to control for batch-to-batch variability, was included as a dummy variable by coding dummy variables for the number of lab batches for the respective marker (). The number of batches differs for urine versus plasma markers.

Associations between ln(uAs) or ln(toenail As) and ln-transformed biomarkers were evaluated using linear regression models. Covariates considered for inclusion in regression models based on a priori considerations and data availability were sex, education, age, cigarette smoking status, alcohol consumption history, ln(BMI), batch assignment, water intake, and fish consumption. Potential confounders that did not alter coefficients by (considering all other potential confounders) were removed from the final models. Our final models were adjusted for sex, age, alcohol consumption, smoking status, and batch assignment, with additional adjustment for arsenobetaine when uAs was the exposure of interest. In models exploring , additional adjustment was included for SG. We further assessed the relationships between %uiAs, %uMMA, or %uDMA and all biomarkers. These models were adjusted for the same set of covariates but also included adjustment for total uAs. In sensitivity analyses, we investigated the potential impact of including water intake as a covariate in our models, as well as excluding individuals with high water As (). We also examined the impact of excluding individuals with out-of-reference range SG values. Exploratory analyses of interactions between As measures and sex were evaluated using cross-product terms in linear regression models. For all analyses, we performed complete case analyses, where observations with missingness in a covariate were excluded. A p-value of 0.05 (2-sided) was used as a cut-off to evaluate statistical significance.

Results

General demographic characteristics of the study participants are shown in Table 1. More than half of the study participants were , with a mean age of 64 y. Approximately 39% of participants were women (as a result of frequency-matching in the parent case–control study), 62% were at least college educated, and 80% were currently married. Arsenic levels in water, shown in Table 2, ranged from 0.001 to , and approximately 5% of our study sample had water As levels exceeding the maximum contaminant level of . Total urinary As levels (excluding arsenobetaine) among participants ranged from 0.825 to , and the percentage of uAs metabolites ranged from 0.7% to 36.5%, 1.0% to 26.4% and 42.3% to 97.7% for %uiAs, %uMMA and %uDMA, respectively. Three individuals had levels of the plasma biomarkers VCAM-1 and ICAM-1 below the detection limit (Table 2). (see Table S1 for study participant characteristics shown separately by sex). The average BMI was higher in men than in women, and alcohol consumption was also generally higher among men. Men were also more likely to be currently married. Although total As exposures did not differ significantly by sex, women on average had a lower %uiAs and %uMMA, and a higher %uDMA, than men. On average, women also had higher plasma ICAM-1 () than men () and somewhat higher, albeit not significant, levels of MMP-9 ( versus among men). Although raw urine concentrations were significantly higher in men () than in women (), SG-adjusted measures did not significantly differ by sex (men, versus women, ).

Table 1

Demographic characteristics of study participants ().

Characteristic	Mean±SD or n (%)	Minimum	25th percentile	Median	75th percentile	Maximum
Age (y)	64±8	26	60	66	70	75
Age ≥65y	235 (56.2)
Men	254 (60.8)
Women	164 (39.2)
BMI (kg/m^2)	28.5±5.5	16.1	24.4	27.4	31.8	49.1
High school education or less	160 (38.3)
College educated or higher	258 (61.7)
Currently married	334 (79.9)
Alcohol consumption status
Never drinker	64 (15.3)
Former drinker	52 (12.7)
Current drinker <median (<336g/mo)	151 (36.1)
Current drinker ≥median (≥336g/mo)	150 (35.9)
Smoking status
Never smoker	157 (37.6)
Former smoker	203 (48.6)
Current smoker	58 (13.9)
Fish consumption
Consumed fish ≥1 time per month	361 (88.9)
Consumed fish <1 time per month	45 (11.1)
Missing fish consumption	12
Urine specific gravity	1.02±0.01	1.00	1.01	1.02	1.02	1.04

Note: Data are complete for all variables unless otherwise indicated. All measured variables were above the limit of detection (LOD) unless otherwise indicated. BMI, body mass index.

Table 2

Descriptive information on As exposure and outcome biomarkers among study participants ().

Arsenic exposure	Mean±SD or n (%)	Minimum	25th percentile	Median	75th percentile	Maximum
Water As (μg/L)	2.6±9.4	0.0	0.1	0.3	0.9	110.5
Water As below LOD	17 (4.1)
Water arsenic <10μg/L MCL	394 (94.7)
Water arsenic ≥10μg/L MCL	22 (5.3)
Water arsenic ≥50μg/L	4 (1.0)
Missing water As	2 (0.5)
Toenail As (μg/g)	5.3±31.2	0.0	0.0	0.1	2.0	507.2
Total uAs, excluding arsenobetaine (μg/L)a	7.1±8.4	0.8	3.1	4.9	8.2	103
Missing total uAs	15 (3.3)
Urinary arsenobetaine (μg/L)	30.2±87.6	0.0	1.3	5.7	26.2	1276.1
Urinary arsenobetaine below LOD	5 (1.2)
Missing urinary arsenobetaine	15 (3.3)
%uiAsb	8.3±5.7	0.7	4.6	7.3	10.3	38.6
%uMMAb	10.6±4.6	1.0	7.4	10.2	13.1	29.2
%uDMAb	81.1±8.7	38.3	76.7	82.0	87.1	97.7
Missing %As metabolites	15 (3.3)
Outcome biomarkers
Plasma VCAM-1 (ng/mL)	224.4±96.9	1.4	166.3	201.9	260.4	1084.9
Plasma VCAM-1 below LOD	3 (0.8)
Missing plasma VCAM-1	22 (5.2)
Plasma ICAM-1 (ng/mL)	336.5±152.5	0.9	234.8	311.3	402.3	1245.0
Plasma ICAM-1 below LOD	3 (0.8)
Missing ICAM-1	22 (5.2)
Plasma MMP-9 (pg/mL)	1090±916	148	530	800	1407	7216
Missing plasma MMP-9	22 (5.2)
Plasma TNF-α (pg/mL)	148±167	22	98	130	166	2995
Missing plasma TNF-α	22 (5.2)
Plasma PAI-1 (ng/mL)	38.7±17.5	14.8	26.6	36.3	46.5	141.1
Missing plasma PAI-1	51 (12.2)
Urine 15-F_2t-IsoP (ng/mL)	4.5±3.6	0.3	2.5	3.5	5.5	33.8

Note: Data are complete for all variables unless otherwise indicated. All measured variables were above the LOD unless otherwise indicated. , ; As, arsenic; BMI, body mass index; ICAM-1, intercellular adhesion molecule 1; LOD, limit of detection; MMP-9, matrix metalloproteinase 9; PAI-1, plasminogen activator inhibitor-1; , tumor necrosis ; uAs, urine arsenic; %uDMA, proportion of dimethyl arsenical species in urine; %uiAs, proportion of inorganic arsenical species in urine; %uMMA, proportion of monomethyl arsenical species in urine; VCAM-1, vascular cell adhesion molecule 1.

No values for total uAs were below the LOD because total uAs is calculated as a sum of As metabolite measures, and any As metabolites below the LOD were imputed to before variable calculation.

Any As metabolites below the LOD were imputed to before calculation of metabolite percentages.

Associations between As Measures, %As Metabolites, and Biomarkers of Potential CVD Risk

Associations between As measures and each biomarker are shown in Table 3, presented as the percent difference in each marker per doubling of toenail or urinary As. Toenail As was positively associated with VCAM-1, such that a doubling of toenail As was associated with a 1.7% increase in plasma VCAM-1 concentration {[95% confidence interval (CI): 0.20, 3.22]; }. Positive associations were observed between toenail As and both ICAM-1 [1.46% increase (95% CI: , 3.11); ] and [1.65% increase (95% CI: 0.03, 3.29); ). Additionally, a doubling in uAs was associated with an 8.8% [(95% CI: 3.15, 14.69); ] higher concentration in urine; uAs was not related to any of the other biomarkers. In contrast with the other findings, a doubling of toenail As was associated with a 2.29% [(95% CI: , ); ] decrease in plasma MMP-9. Exploratory analyses of associations between total urine or toenail As measures and biomarkers did not appear to differ significantly by sex ( for all cross-product interaction terms) (see Tables S2 and S3).

Table 3

Percent difference (95% CI) in outcome biomarkers for a doubling of arsenic exposure biomarkers.

Biomarker	Toenail As			Urinary As
	n	Percent difference (95% CI)	p-Value	n	Percent difference (95% CI)	p-Value
VCAM-1	396	1.70 (0.20, 3.22)	0.03	347	−4.61 (−10.52, 1.68)	0.15
ICAM-1	396	1.46 (−0.16, 3.11)	0.08	347	−4.25 (−10.60, 2.56)	0.21
MMP-9	396	−2.29 (−4.25, −0.29)	0.03	347	3.25 (−5.07, 12.29)	0.45
TNF-α	396	0.23 (−1.02, 1,50)	0.72	347	−3.16 (−8.11, 2.07)	0.23
PAI-1	367	0.85 (−0.43, 2.14)	0.19	322	−1.19 (−6.31, 4.22)	0.66
15-F_2t-IsoPa	418	1.65 (0.03, 3.29)	0.05	403	8.77 (3.15, 14.69)	<0.01

Note: Percent difference was calculated using the following formula: , where is the arsenic coefficient from linear regression models in which both the arsenic exposure variable and the biomarker were natural log–transformed. Models were adjusted for sex, age, alcohol consumption, cigarette smoking status, and analytic batch. Urine arsenic analyses were additionally adjusted for ln(specific gravity–adjusted arsenobetaine). , ; As, arsenic; CI, confidence interval; ICAM-1, intercellular adhesion molecule 1; MMP-9, matrix metalloproteinase 9; PAI-1, plasminogen activator inhibitor-1; uAs, urine arsenic; , tumor necrosis ; VCAM-1, vascular cell adhesion molecule 1.

For models, uAs was not adjusted for specific gravity. Instead, specific gravity was included as a covariate in models.

We further assessed associations between %As metabolites and each biomarker, which are shown in Table 4 as the percent difference in each marker for a 5% increase in As metabolite. A 5% increase in %uMMA was associated with a 7.92% increase [(95% CI: 2.10, 14.07); ] in , whereas a 5% increase in %uDMA was associated with a 2.98% decrease [(95% CI: , 0.21); ] in . %As metabolites were not related to any of the other biomarkers, with the exception of PAI-1, which in contrast to other findings was negatively associated with %uMMA [ (95% CI: , ); ). We performed exploratory analyses of associations with arsenic metabolite percentages by sex, but the estimates were imprecise, and there were no significant differences in associations between men and women for any of the metabolites or outcomes (see Table S4).

Table 4

Percent difference (95% CI) in outcome biomarkers for a 5% increase in arsenic metabolites.

Biomarker		%uiAs		%uMMA		%uDMA
	n	Percent difference (95% CI)	p-Value	Percent difference (95% CI)	p-Value	Percent difference (95% CI)	p-Value
VCAM-1	347	2.30 (−2.88, 7.75)	0.39	3.76 (−2.79, 10.75)	0.27	−2.13 (−5.54, 1.40)	0.23
ICAM-1	347	3.30 (−2.30, 9.23)	0.25	1.99 (−4.92, 9.41)	0.58	−2.08 (−5.74, 1.73)	0.28
MMP-9	347	−1.70 (−8.19, 5.26)	0.62	−5.86 (−13.58, 2.56)	0.17	2.62 (−2.06, 7.52)	0.28
TNF-α	347	−0.15 (−4.33, 4.22)	0.95	−2.20 (−7.31, 3.20)	0.42	0.73 (−2.17, 3.72)	0.62
PAI-1	322	1.44 (−2.88, 5.95)	0.52	−7.67 (−12.56, −2.51)	<0.01	1.68 (−1.30, 4.74)	0.27
15-F_2t-IsoP	403	0.97 (−3.66, 5.83)	0.69	7.92 (2.10, 14.07)	<0.01	−2.98 (−6.07, 0.21)	0.07

Note: The percent difference was calculated using the following formula: , where is the arsenic coefficient from linear regression models in which only the outcome biomarker variable was natural log–transformed. Models were adjusted for sex, age, smoking status, alcohol status, analytic batch, ln[specific gravity (SG)–adjusted arsenobetaine], and ln(SG-adjusted uAs). In models evaluating urinary arsenic, SG was included as a covariate in models and urinary measures (uAs, urinary arsenobetaine, and ) were not adjusted for SG. , ; CI, confidence interval; uAs, total urine arsenic; %uDMA, proportion of dimethyl arsenical species in urine; %uiAs, proportion of inorganic arsenical species in urine; %uMMA, proportion of monomethyl arsenical species in urine.

Finally, we performed a number of additional sensitivity analyses. First, we estimated associations between arsenic exposure and , VCAM-1, and ICAM-1 after excluding five participants with water As (see Table S5). Associations were similar to those in the primary models. We also repeated the analyses after excluding six participants with urine SG values outside of the normal reference range, but none of the point estimates differed from the primary analysis by (data not shown). We also tested the potential impact of including water intake as a covariate in our analyses of urine As and and found that our estimates were not substantially altered by this additional adjustment (see Table S6).

Discussion

Arsenic is thought to contribute to the development of cardiovascular disease by increasing the production of reactive oxygen species, which may mediate inflammatory responses, alterations in gene expression, and impaired nitric oxide signaling (Solenkova et al. 2014; States et al. 2009; Wu et al. 2014). These factors might also contribute to endothelial dysfunction and altered vascular tone, ultimately increasing the risk for hypertension and atherosclerosis (Solenkova et al. 2014; States et al. 2009; Wu et al. 2014). In this cross-sectional study of U.S. men and women, we evaluated associations between As exposure and a urinary indicator of oxidative stress () and circulating markers of endothelial dysfunction (VCAM-1, ICAM-1, PAI-1) and inflammation (MMP-9, ). Toenail As, an indicator of As exposure in the previous 6–12 mo (NRC 2013), was associated with significantly higher plasma VCAM-1 and urine concentrations and with a nonsignificant increase in plasma ICAM-1 (). However, toenail As was also associated with significantly lower plasma MMP. Urine As, which represents more recent As exposure, largely from the previous two days (NRC 2013), was associated with higher urinary concentrations. %MMA in urine was associated with higher , suggesting that As methylation capacity is reduced in association with oxidative stress. However, %MMA was also associated with significantly lower plasma PAI-1.

CVD is a complex outcome with multiple etiologic pathways. Arsenic is hypothesized to affect several of these pathways, including oxidative signaling, induction of cytokines, systemic inflammation, and vascular endothelial activation. During inflammation or endothelial injury, leukocytes are recruited to the microenvironment by a combination of chemokines and adhesion molecules. VCAM-1 and ICAM-1 are often expressed at lesion-prone sites, with VCAM-1 expression preceding atherosclerotic plaque development (Mozos et al. 2017; Nakashima et al. 1998; Poredoš and Ježovnik 2015). Monocytes, attracted to the lesion by these adhesion molecules, penetrate the arterial intima, where they take on a tissue macrophage-like phenotype, secreting cytokines and matrix metalloproteinases, such as MMP-9, which degrade the extracellular matrix and thereby promote disease progression (Libby et al. 2002). Elevated levels of VCAM-1 and ICAM-1 have been previously related to a number of risk factors for CVD and metabolic disease, such as obesity, hypertension, diabetes, and carotid intima media thickness (Blankenberg et al. 2003; Bosanská et al. 2010; Hwang et al. 1997; Pankow et al. 2016). Although previous studies have reported a positive association between these markers and an increased risk of CVD morbidity and mortality (Hwang et al. 1997; Ridker et al. 1998; Rohde et al. 1998), a more recent study reported an inverse association between prospectively measured VCAM-1 levels and CVD risk, suggesting that this association may require further investigation (Kunutsor et al. 2017). Furthermore, it is unclear whether relatively small changes in these plasma markers, such as those observed in the present study, may be clinically relevant or related to elevated CVD risk. In this study, toenail As was associated with significantly higher serum VCAM-1 and with a nonsignificant increase in serum ICAM-1 (), which indicate early cardiovascular effects and are consistent with findings for other As-exposed populations. VCAM-1 and ICAM-1 have been previously related to chronic As exposure across multiple populations, including evidence from studies in Bangladesh (Chen et al. 2007; Karim et al. 2013; Wu et al. 2012) as well as from recent findings by our group in a New Hampshire pregnancy cohort that maternal urine As is associated with higher levels of both maternal plasma VCAM-1 and infant cord blood VCAM-1 and ICAM-1 levels (Farzan et al. 2017). However, our findings also suggest a possible inverse relationship between As and MMP-9, which is inconsistent with the proposed mechanism. More work is needed to elucidate whether altered levels of these markers are related to development of disease.

Although previous studies have linked As exposure with altered levels of plasma markers of inflammation and endothelial function, including VCAM-1 and ICAM-1 (Burgess et al. 2013; Chen et al. 2007; Kurzius-Spencer et al. 2016; Wu et al. 2012), few have examined associations at environmentally relevant levels of As exposure (generally ), such as those found in most of New Hampshire and across the United States in private wells. Our results suggest that, similar to what has been reported for populations exposed to relatively high levels of As, such as populations in Bangladesh, relatively low levels of As exposure may influence inflammation and other mechanisms involved in CVD pathogenesis. The idea that lower levels of As exposure could play a role in CVD development has been supported by an increasing number of epidemiological studies from our research group (Farzan et al. 2015a, 2015b) and from others (D'Ippoliti et al. 2015; Gong and O'bryant 2012; James et al. 2015; Medrano et al. 2010; Monrad et al. 2017; Moon et al. 2013; Mordukhovich et al. 2009). Experimental studies have also shown that As can elicit adverse cardiovascular effects at doses much lower than those required to induce cancer (Lemaire et al. 2011, 2015; Soucy et al. 2005; Straub et al. 2007). For example, a recent mouse study demonstrated that As concentrations as low as 10 ppb increase monocyte adhesion to the endothelium via enhanced VCAM-1 binding (Lemaire et al. 2015). Given the potential for widespread low to moderate As exposure from both water and food sources (Davis et al. 2017; Navas-Acien and Nachman 2013), even relatively small As-related contributions to disease risk could have broad implications for public health.

Reactive oxygen species produced by As exposure are known to induce a wide range of effects that are thought to underlie a number of chronic health conditions, including CVD (reviewed in Jomova et al. 2011; Wu et al. 2014). Oxidative stress responses are thought to be linked to endothelial dysfunction, as indicated by their ability to induce VCAM-1 and ICAM-1 expression (Wadham et al. 2004). Experimental and epidemiological studies have reported associations between As exposure and higher levels of oxidative stress biomarkers, such as oxidized low density lipoprotein and oxidative DNA damage biomarker 8-oxo-2′-deoxyguanosine (Engström et al. 2010; Karim et al. 2013). Although As exposure has been shown to induce 8-isoprostane levels in vitro (Han et al. 2005), to our knowledge, the relationship between As exposure and isoprostanes has not been evaluated in human populations. Isoprostanes are prostaglandin-like compounds that are endogenously generated by nonenzymatic free-radical-catalyzed peroxidation of esterified arachidonic acid. Elevated urinary is thought to reflect widespread oxidative stress and systemic levels of lipid peroxidation end products, and it has been positively associated with CVD risk factors, such as smoking and hypertension (Minuz et al. 2002; Morrow et al. 1995); our results suggest that this marker may also increase with As exposure. In our study, we measured the major urinary metabolite , which is thought to be a reliable marker of systemic oxidative stress (Dorjgochoo et al. 2012; Milne et al. 2015). Additionally, is not subject to collection-related oxidation artifact, is not influenced by dietary lipid content, and is stable over time if samples are stored at (Morrow et al. 1999; Prasain et al. 2013). However, the extent to which levels of measured in urine may be influenced by local conditions in the bladder remains unclear. Experimental animal studies have demonstrated that the bladder can produce , but the extent to which levels measured in our subjects may be influenced by conditions in the bladder is not known. We cannot rule out that production of by the bladder epithelium could have influenced our measurements (Tarcan et al. 2000). Further research is needed to establish whether is a reliable marker of As-related oxidative stress and whether alterations in its levels may be related to development of CVD.

Notably, we observed differential associations between the urinary As metabolites and . Previous studies have reported that individuals with higher %uMMA or a lower ratio of uDMA to uMMA have a higher risk of many As-related health outcomes (Steinmaus et al. 2010), including carotid intima-media thickness and CVD (Chen et al. 2013a, 2013b). Our findings that urine and toenail As levels were associated with higher urine concentrations of the oxidative stress marker and that individuals with higher %uMMA had higher urine , whereas those with higher %uDMA had lower , suggest that an enhanced ability to metabolize iAs to DMA might reduce oxidative stress resulting from As exposure. Previous studies in more highly exposed populations have reported that women appear to be more efficient at metabolizing iAs to DMA than men, and it has been suggested that this may contribute to sex differences in certain As-related health outcomes, such as skin lesions (Ahsan et al. 2006; Jansen et al. 2015; Lindberg et al. 2008). Consistent with previous studies, we observed that on average, women had lower %uiAs and %uMMA and higher %uDMA than men. However, our findings with regard to sex differences in associations between the measured biomarkers and the distribution of urine As metabolites were inconclusive.

Unexpectedly, we found that toenail As was inversely associated with plasma MMP-9, which is known to play a role in several stages of atherosclerosis and has been prospectively linked to CVD mortality (Hansson et al. 2009, 2011). However, the relationship between As exposure and plasma MMP-9 has been somewhat inconsistent across studies. For example, two cross-sectional studies in the United States and Mexico reported positive associations between As exposure and serum MMP-9 (Burgess et al. 2013; Kurzius-Spencer et al. 2016). However, this finding was not confirmed by a prospective study in Bangladesh: In that study, no overall associations between As exposure and MMP-9 were observed, but slight decreases were found in MMP-9 levels at moderate urinary As concentrations, suggesting a possible nonmonotonic relationship (Wu et al. 2012). Although there are a number of potential reasons for these inconsistencies, it is possible that differences in exposure metrics (e.g., water/food As, urine As, and toenail As) or variability in methods of MMP-9 detection could have affected our ability to compare across studies. Given the importance of MMP-9 as a marker of CVD, these conflicting findings warrant further investigation.

Importantly, our study had several weaknesses. First, because our study was a cross-sectional analysis, some of our findings may be subject to reverse causality. For example, there is some evidence that As may adversely affect renal function (Zheng et al. 2015; reviewed in Zheng et al. 2014) and, though not extensively studied, it has been hypothesized that reduced renal function may affect the distribution of As metabolites in urine (Peters et al. 2015; Zheng et al. 2015). Thus, individuals with subclinical or clinical CVD may have increased biomarkers of oxidative stress, endothelial dysfunction, or inflammation in addition to reduced renal function and may therefore have an altered distribution of As metabolites in urine. We also lacked information on participants’ CVD status and other potential health conditions at the time of the interview and were unable to account for this in our models. Further, our sample size was insufficient to evaluate the shape of the dose–response relationships for each exposure and outcome in detail. Future studies evaluating arsenic exposure in relation to these markers in larger study populations would be helpful in exploring potential nonmonotonic relationships. Urinary arsenic was only measured at a single time point; therefore, the potential variability in typical exposure levels may not be taken into account for all of the individuals in our study sample. Urinary arsenic is considered to be a reliable measure of recent arsenic exposure that appears to remain relatively consistent over time in adults (Gamble et al. 2006), but differences in short- versus long-term windows of exposure reflected by urine and toenails may be responsible for some of the inconsistencies in our findings. Finally, because our study population consisted almost entirely of Caucasian older adults residing in New Hampshire, our findings may not be generalizable to other populations in the United States or elsewhere.

Our study also had many strengths. One strength of our study was the measurement of both toenail As and uAs, which allowed us to estimate associations for both long-term and recent As exposure on a suite of biomarkers of potential CVD risk. The use of urine as an As biomarker provided a measure of As exposure from various sources including diet and drinking water. Additionally, we acquired speciated uAs measures and were therefore able to also evaluate associations between the different As metabolites, which may have differing toxicities, in relation to the same biomarkers. Although one previous study has evaluated %uMMA in relation to plasma MMP-9 and did not observe a significant association (Burgess et al. 2013), to our knowledge, associations between As metabolites and other biomarkers potentially relevant to CVD have not yet been studied.

Conclusion

In our cross-sectional study population of U.S. adults, most of whom had low to moderate As exposure, As was positively associated with biomarkers that may be relevant to CVD pathogenesis, including VCAM-1, ICAM-1, and , although an inverse association with MMP-9 was also observed. Furthermore, lower %uMMA and higher %uDMA, which may indicate more efficient As metabolism, was inversely associated with urine , a marker of oxidative stress.

Click here for additional data file.