Linkage Analysis of Urine Arsenic Species Patterns in the Strong Heart Family Study

Abstract

Arsenic toxicokinetics are important for disease risks in exposed populations, but genetic determinants are not fully understood. We examined urine arsenic species patterns measured by HPLC-ICPMS among 2189 Strong Heart Study participants 18 years of age and older with data on ∼400 genome-wide microsatellite markers spaced ∼10 cM and arsenic speciation (683 participants from Arizona, 684 from Oklahoma, and 822 from North and South Dakota). We logit-transformed % arsenic species (% inorganic arsenic, %MMA, and %DMA) and also conducted principal component analyses of the logit % arsenic species. We used inverse-normalized residuals from multivariable-adjusted polygenic heritability analysis for multipoint variance components linkage analysis. We also examined the contribution of polymorphisms in the arsenic metabolism gene AS3MT via conditional linkage analysis. We localized a quantitative trait locus (QTL) on chromosome 10 (LOD 4.12 for %MMA, 4.65 for %DMA, and 4.84 for the first principal component of logit % arsenic species). This peak was partially but not fully explained by measured AS3MT variants. We also localized a QTL for the second principal component of logit % arsenic species on chromosome 5 (LOD 4.21) that was not evident from considering % arsenic species individually. Some other loci were suggestive or significant for 1 geographical area but not overall across all areas, indicating possible locus heterogeneity. This genome-wide linkage scan suggests genetic determinants of arsenic toxicokinetics to be identified by future fine-mapping, and illustrates the utility of principal component analysis as a novel approach that considers % arsenic species jointly.

Inorganic arsenic (iAs) exposure is carcinogenic (IARC, 2004) and increasing evidence also supports its association with cardiovascular disease (Moon et al., 2012; Wu et al., 2014) and diabetes (Maull et al., 2012), although the importance of low-level exposure remains unclear. The iAs species (arsenite, arsenate) are metabolized to form methylated (MMA) and dimethylated (DMA) arsenic species (Thomas, 2009; Vahter, 2002). The enzyme arsenic (III) methyltransferase coded by AS3MT on 10q24.32 is believed to be centrally important in arsenic methylation reactions (Wood et al., 2006), although other enzymes might also be involved (Harari et al., 2013; Ren et al., 2010).

There is growing evidence from human samples that arsenic species differ in their toxicities. Proliferation of T cells in fresh blood collected from human participants recruited at University of New Mexico was more strongly inhibited by exposure to trivalent MMA than exposure to trivalent arsenite (Burchiel et al., 2014). In a study using blood collected from 2 male medical students at the National Autonomous University of Mexico, treating lymphocytes with either trivalent MMA or trivalent DMA induced DNA damage via oxidative-stress, but MMA was effective at lower doses (Soto-Reyes et al., 2005). Human lymphoblastoid cell culture experiments suggested that trivalent MMA had greater impact on mitotic indices than arsenate, arsenite, pentavalent MMA, or trivalent or pentavalent DMA; and trivalent MMA and DMA were able to inhibit tubulin polymerization at lower concentrations than arsenite (Kligerman et al., 2005). If arsenic species are toxicologically distinct, then their biotransformation and elimination may be relevant for human disease risks from arsenic exposure.

Inter-individual variation in arsenic kinetic processes has been studied indirectly by comparing the percentages of iAs, MMA, and DMA species (relative to their sum) in urine (Chiou et al., 1997; Gamble and Liu, 2005; Gomez-Rubio et al., 2011; Gribble et al., 2013a; Hopenhayn-Rich et al., 1996; Navas-Acien et al., 2009; Vahter, 2000). The pattern of urine arsenic species is a possible risk factor for skin lesions (Ahsan et al., 2007; Chen et al., 2009; Valenzuela et al., 2009); bladder, urothelial, lung, and skin cancers (Agusa et al., 2010; Chen et al., 2003; Huang et al., 2008a, b; Steinmaus et al., 2006, 2010; Yu et al., 2000) and cardiovascular diseases (Huang et al., 2007, 2009). Endogenous and environmental factors have been associated with urine arsenic species patterns, including sex, age, pregnancy status, nutrition, body mass, smoking, alcohol, diabetes status, and insulin level (Gamble and Hall, 2012; Gomez-Rubio et al., 2011; Gribble et al., 2013a; Su et al., 2012; Tseng, 2009). Arsenic species patterns are also influenced by genetics (Antonelli et al., 2014; Hernandez and Marcos, 2008). In the Strong Heart Study, a large population-based cohort of American Indians with low-moderate arsenic exposure, heritability estimates for each % arsenic species (%iAs, %MMA, and %DMA) were ≥50% (Tellez-Plaza et al., 2013). The causal role of AS3MT in arsenic methylation biology has been studied in both functional experiments (Wood et al., 2006) and Mendelian randomization analysis of AS3MT variants for urine arsenic species (Pierce et al., 2013). This gene has attracted such interest that there have been several comparative studies of AS3MT genetic variation across populations (Fujihara et al., 2007, 2008, 2009, 2011); in study populations from Bangladesh and Northern Mexico, there was a large linkage disequilibrium region around this gene (Engstrom et al., 2013; Gomez-Rubio et al., 2010).

The predominant approach to genetic epidemiology of arsenic kinetics (Hernandez and Marcos, 2008) has been candidate gene or genome-wide association studies (Engstrom et al., 2011; Gomez-Rubio et al., 2010; Gribble et al., 2013b; Hernandez et al., 2008; Lindberg et al., 2007; Meza et al., 2005; Pierce et al., 2012; Schlawicke Engstrom et al., 2009). Although often informative, association studies can have limited power to detect genomic relationships in the presence of allelic heterogeneity (Ott et al., 2011), or for rare variants unless they have a large effect. Association studies are also often performed using genotyping arrays, which may have limited coverage in populations such as American Indians. Family-based linkage studies can complement association studies to find chromosomal regions that co-segregate across generations with urine arsenic species patterns (Ott et al., 2011). We previously reported a preliminary linkage analysis of urine arsenic species patterns in a pilot sample of 487 participants from the Strong Heart Family Study, an ancillary study to the Strong Heart Study, finding suggestive linkage evidence on chromosomes 5 (%iAs), 10 (%DMA), and 11 (%iAs and %MMA) (Tellez-Plaza et al., 2013).

In this study, we aimed to investigate the co-segregation of loci with urine arsenic species in the full sample of the Strong Heart Family Study cohort. In addition to considering each % arsenic species biomarker separately, we also used principal components as a novel approach to describe overall patterns in the 3 % arsenic species measures.

MATERIALS AND METHODS

Study population

The Strong Heart Family Study is composed of 94 extended families from Arizona, Oklahoma, North Dakota, and South Dakota (North et al., 2003). The rich family relationships included 3975 relative pairs in Arizona; 3248 relative pairs in Oklahoma; and 5571 relative pairs in North and South Dakota. Trained nurses and medical assistants collected anthropometric and interview data following standardized protocols (Lee et al., 1990). This study was approved by relevant institutional and tribal ethics review boards. Informed consent was obtained from all participants. For this study, we included Strong Heart Family Study participants with both genotype and arsenic species data measured as part of an ancillary study to evaluate gene-environment interactions for diabetes incidence. Arsenic species data were not available among persons with baseline diabetes. We also excluded 245 participants with one or more urine arsenic species <0.1 μg/l limit of detection, as it is difficult to evaluate arsenic metabolism if arsenic species in urine are undetectable. We also excluded participants with missing data on smoking history (n = 6), current alcohol consumption (n = 3), body mass index (n = 8), educational attainment (n = 5), or with a sample error (n = 2) for an analysis sample size of n = 2189.

Urine arsenic measures

Total urine arsenic levels and arsenic species were measured by the Trace Element Laboratory of the Institute of Chemistry—Analytical Chemistry, University of Graz in Graz, Austria, using high performance liquid chromatography inductively-coupled plasma mass spectrometry (HPLC-ICPMS) (Scheer et al., 2012). Samples had been occasionally thawed and re-frozen in the years between sample collection and chemical analysis, precluding accurate measurement of reduction state of arsenicals in these samples (Scheer et al., 2012). The inter-batch variability was monitored by replicate measurements of 3 urine reference materials with certified arsenic levels of 20.3 µg/l (NIST 2669 I), 50.2 µg/l (NIST 2669 II), or 119 µg/l (NIES 18); the coefficients of variation (CV) ranged from 3.8 to 14.4%, with an overall mean CV of 7.9% (n = 46). Percent arsenic species (%iAs, %MMA, and %DMA) were defined as the contributions of inorganic, methylated, and dimethylated arsenical species in urine to the sum of those species. For example, %iAs = iAs/(iAs + MMA + DMA).

Genetic measures

Methods used for microsatellite marker genotyping have been described previously (Tellez-Plaza et al., 2013). In brief, microsatellite markers from the ABI PRISM Linkage Mapping Set-MD 10 version 2.5 (Applied Biosystems, now Thermo Fisher Scientific, Grand Island, New York) set were genotyped in amplified white blood cell DNA using the ABI PRISM 377 Genetic Analyzer. These ∼400 markers were spaced roughly 10 cM apart with density ranging from 2.4 to 24.1 cM. Genotypes were called using the Genotyper software (Thermo Fisher Scientific). SNPs in AS3MT were typed as part of the Cardio-Metabo DNA Analysis BeadChip following the manufacturer’s protocol (Illumina, San Diego, California). In brief, genomic DNA (200 ng) was whole-genome amplified, enzymatically fragmented, and hybridized to the chips containing locus-specific oligomers covalently linked to beads embedded in the chips. The hybridization was followed by a single-base extension with fluorescently labeled nucleotides. All steps were performed on a Tecan Freedom EVO 150 cm liquid handler (Tecan, Männedorf, Switzerland) with Illumina GTS Robot Control software (Illumina). Fluorescence intensities were detected using an iScan (Illumina) and were consequently analyzed using the Illumina GenomeStudio software. Assay accuracy was assessed using sample-dependent and independent controls. Potential misclassification of pedigree relationships was inspected using PREST (McPeek and Sun, 2000; Sun et al., 2002) and Simwalk2/Preswalk (Sobel et al., 2002) software. Mendelian errors and unlikely double recombinants were blanked, with an overall blanking rate <1%. SNP variants were aligned to the UCSC Genome Browser Human Genome Version 18 assembly.

Statistical analyses

A variance components decomposition method implemented in Sequential Oligogenic Linkage Analysis Routines (SOLAR) software was used to estimate heritability and detect linkage to chromosomal locations that affect the variation in arsenic species (Almasy and Blangero, 1998). Analyses were conducted separately for each logit-transformed measure of arsenic species patterns (%iAs, %MMA, and %DMA) and for each study center (Arizona, Oklahoma, and North and South Dakota). We also considered principal components of the logit-arsenic species as another way of describing inter-individual phenotypic differences in urine arsenic species patterns. The first 2 principal components were of substantive interest, while the third principal component was a negligible (eigenvalue 0.01) artifact of the logit-transformation of the 3 dependent % arsenic species variables. Models adjusted for age, age squared, sex, linear, and quadratic interactions between age and sex, study center (Arizona, Oklahoma, and North and South Dakota), obesity (body mass index ≥30 kg/m²), high school completion (education ≥ 12 years), smoking status (ever/never), drinking status (current drinking yes/no), and higher than median iAs exposure (measured by the sum of inorganic and methylated arsenic species in urine; not corrected for urine creatinine). Sensitivity analysis indicates that adjustment for urine creatinine prior to dichotomization of the urine sum of species would have little impact on our final results (data not shown). Residuals from preliminary heritability analyses controlling for covariates were inverse-normalized and carried forward for linkage analyses.

We used the LOD score (negative log of the P value) thresholds set by Lander and Kruglyak (1995) for suggestive (≥1.9) or significant (≥3.3) evidence of linkage. We also stratified each logit-transformed % arsenic species linkage analyses by study center for secondary analyses to examine possible heterogeneity across study centers.

We examined the contribution of AS3MT variants to a linkage signal anticipated on chromosome 10 by including 2 SNPs rs17878846 and rs10509760 from that gene region in a conditional linkage analysis. We summarized the linkage disequilibrium between these SNPs and other 10q24 locus variants measured on the Cardio-Metabo DNA Analysis BeadChip through pair-wise r². There were 24 persons in our sample without genotypes available for rs17878846 and rs10509760 who were excluded from the conditional linkage analysis.

RESULTS

Arsenic Metabolism Biomarkers and Participant Characteristics

The demographic characteristics of the study participants are summarized in Table 1. Mean age in the sample was 34.3 (SD 14.8) years, 59% were female, and mean body mass index was 31.5 (SD 7.8) kg/m².

In this sample, median (interquartile range) urine arsenic species patterns were 10 (7, 14) %iAs, 14 (11, 18) %MMA, and 75 (68, 81) %DMA (Table 1). These patterns varied by sex, age, study center, body mass index, education, smoking, alcohol consumption, and iAs exposure. The arsenic species pattern measure showing the most variability across groups was %DMA. The sample Spearman correlation of total urine arsenic with %iAs was −0.16, with %MMA was −0.23, and with %DMA was +0.22. The heritability of these urine arsenic species patterns, conditional on demographics and high versus low arsenic, varied slightly by study center, ranging from 0.33 for %iAs in Oklahoma to 0.57 for %DMA in Arizona (Supplementary Data).

The first principal component appeared mainly to distinguish more versus less DMA in urine relative to the other species, and explained ∼80% of the phenotypic variability in urine arsenic species patterns. The second principal component explained 19% of the phenotypic variability and appeared to distinguish %iAs versus %MMA after accounting for differences in %DMA (Supplementary Data). The first principal component score had a negative Spearman correlation with total urine arsenic (−0.22), was positively correlated with logit %iAs (0.87) and logit %MMA (0.80) and negatively correlated with logit %DMA (−0.998). The second principal component score had a weak negative Spearman correlation with total urine arsenic (−0.08), and it was negatively correlated with logit %iAs (−0.42), positively correlated with logit %MMA (+0.56), and essentially uncorrelated with logit %DMA (−0.02, P value = 0.48).

Linkage Disequilibrium for AS3MT Index SNPs

Considering a pair-wise r² between SNPs of ≥0.8 to be indicative of linkage disequilibrium, the linkage disequilibrium block for rs17878846 ranges at least from 104 584 497 to 104 783 894. The region in linkage disequilibrium with rs10509760 ranges at least from 104 624 464 to 104 793 052. As a point of comparison, according to the UCSC Genome Browser hg18 assembly, the AS3MT gene extends 104 619 200 to 104 651 645, so each of these 2 index SNPs is tagging a broader region extending a bit beyond the AS3MT gene boundaries in our sample. Pair-wise SNP r² for the linkage disequilibrium blocks are presented in Supplementary Data and Supplementary Data.

Linkage Analysis of Arsenic Metabolism Biomarkers

In our linkage analyses (n = 2189), we localized a quantitative trait locus (QTL) on chromosome 10 near the AS3MT gene (Figs. 1–3), with LOD scores exceeding 4 for both logit- %MMA and %DMA phenotypes, as well as the first principal component of logit % arsenic species (Fig. 4). After conditioning for rs17878846 and rs10509760 at AS3MT (n = 2165), the conditional LOD score was attenuated by approximately half (Fig. 5). The %DMA LOD score decreased from 4.61 at 124 cM to 2.49 at 116 cM and the first principal component LOD score decreased from 4.84 at 121 cM to 2.73 at 116 cM.

There was a suggestive linkage peak on chromosome 6 (Fig. 1 and Supplementary Data) for %iAs largely driven by a signal in Arizona (LOD 3.06, 1-LOD support interval from 0 to 35 cM). A small overall population peak in this region was also visible for %DMA (LOD 1.01) and to a lesser extent %MMA (LOD 0.37); however, those peaks did not meet the threshold for suggestive evidence of linkage (Figs. 2 and 3). There was no visible peak on chromosome 6 for either of the principal components (Fig. 4).

The 2 principal components showed very distinct linkage signals (Fig. 4). The first principal component’s main signal on chromosome 10 was not observed for the second principal component. In contrast, the second principal component’s largest linkage signal was localized to chromosome 5 (LOD 4.21, 1-LOD support interval 111 to 120 cM), in a region not showing suggestive linkage signal for any individual % arsenic species biomarker nor for the first principal component (Supplementary Data). Similarly, there appeared to be a signal on chromosome 14 for the second principal component (LOD 3.20, 1-LOD support interval 46 to 67 cM) that was stronger than in the % arsenic species analyses.

Center-stratified analyses revealed potential locus heterogeneity, with some loci appearing suggestive or significant at 1 center but not at the other 2, and not in the combined analysis. In particular, Arizona showed a significant linkage signal (LOD 3.74, 1-LOD support interval from 39 to 57 cM) on chromosome 11 for %MMA but this was absent in the other 2 study centers (Fig. 2).

DISCUSSION

This study identified possible genetic determinants of arsenic toxicokinetics and illustrated a novel approach explicitly considering the inter-relationships among the % arsenic species. The strongest evidence for QTL in our study was with the first principal component on chromosome 10 (LOD 4.71), a component that reflected how %DMA levels were inversely related to %iAs and %MMA levels. Smaller peaks on chromosome 10 were observed for %iAs (LOD 2.21), %MMA (LOD 4.12), and %DMA (LOD 4.61). The second principal component, which distinguished the balance of %iAs and %MMA independent of the level of %DMA, localized a QTL to chromosome 5 (LOD 4.27). This QTL did not pass the suggestive threshold for the % arsenic species when modeled separately.

Principal component scores in the context of arsenic species have several advantages compared with the conventional separate analysis of % arsenic species measures. First, it simplifies the interpretation by consolidating 3 inter-related measures into 2 independent measures. Second, the principal components may be reflecting biological processes more directly. The first component might be reflecting methylation broadly or even the second methylation step specifically, as these would lead to higher DMA relative to the other species. The evidence for a QTL near the AS3MT gene seen with the first principal component was stronger than the evidence for a QTL seen with any % arsenic species biomarker, consistent with the principal component score having less measurement error as an indicator of the underlying biological processes. The second component could be reflecting a transport process that would only influence the inter-relationship between iAs and MMA in urine independent of the DMA, or another unknown biological process.

Despite the value of these principal component scores for characterizing inter-individual variability in urine arsenic species patterns, there are limitations to this approach. The orientation of the principal component axes depends on the sample, so principal component scores from different study populations are not directly comparable. Also, the value of an individual’s score depends on their compatriots in the study sample. Nevertheless, loci identified by studies using principal component scores can be compared across studies.

This study extends previous arsenic species pattern research in the Strong Heart Family Study (Tellez-Plaza et al., 2013). We confirmed the QTL found in chromosome 10 in our preliminary analysis, which at the time was only suggestive (Tellez-Plaza et al., 2013). Although we failed to replicate earlier findings of suggestive linkage of %iAs on chromosomes 9 and 11 in the overall analyses, on chromosome 11 we found significant evidence for a QTL for %MMA in Arizona. For chromosome 5, our second principal component QTL was in a different part of the chromosome than a suggestive peak for %iAs in the preliminary analysis (1-LOD support interval 29–57 cM). The discrepant findings for chromosomes 5 and 9 might be due to our restriction to diabetes-free participants if the preliminary findings, which included persons with diabetes, were partially informed by genes expressed differently under the conditions of diabetes. We were unable to test for locus-diabetes interactions as arsenic was not measured in participants with diabetes at baseline due to the study design. However, this interaction is possible, as co-morbid disease status can sometimes affect linkage signals (Colilla et al., 2003), diabetes may have epigenetic consequences (Reddy and Natarajan, 2013), and even the earliest arsenic genetic epidemiology research suggested that the genetic determinants of urine arsenic species may be context (ie, age) dependent (Chung et al., 2002).

The field of arsenic genetic epidemiology has heretofore emphasized the role of common genetic variants in explaining population-level variability in arsenic kinetics and susceptibility. The heterogeneity across study centers, and residual heritability unexplained by the AS3MT SNPs included in the conditional linkage model, may suggest an important role for private or population-specific mutations as well. This would have implications for regulatory policies designed to protect the public from arsenic health effects, as “uncertainty factors” for non-cancer risk assessments (Alexeeff et al., 2002; Gaylor and Kodell, 2002) that consider only common modifiers such as common AS3MT SNPs, sex, body mass index, or smoking status would underestimate the extent of inter-individual variability, and could result in regulatory standards inadequate to protect sensitive subpopulations. Quantifying the frequency of rare mutations conferring an impact on arsenic kinetics and susceptibility, and their relative importance for arsenic kinetics and for clinical outcomes, is an important and novel direction for future arsenic susceptibility genetic research. Exploring the role of rare variants in arsenic susceptibility is consistent with the broader push in modern genetic epidemiology to understand the role of low frequency and rare genetic variation in explaining phenotypic variance (Agarwala et al., 2013; Cooper et al., 2010; Kiezun et al., 2012; Ng et al., 2010; Sobreira et al., 2010). The family design of the Strong Heart Family Study may be an asset in further exploring the role of rare genetic variation in explaining inter-individual variability in arsenic kinetics (Cheung et al., 2013; Saad and Wijsman, 2014). Lastly, we did not detect any suggestive peaks on chromosome 12, where we recently identified SNPs in the liver transporter SLCO1B1 explaining a substantial proportion of the phenotypic variance in %MMA and %DMA in a small (n = 157) convenience sample (Gribble et al., 2013b). The discrepancy in findings for chromosomes 11 and 12 may be due to the complementary strengths of linkage and association designs to detect different kinds of genetic signals (Ott et al., 2011). It may also be in part due to a spurious finding in our earlier study related to the small sample size or to locus heterogeneity across study center. Apparent locus heterogeneity across study centers might also be due to differential power to detect linkage signals across study centers if measurement error is different across study centers. We did estimate higher heritabilities for %MMA and %DMA (residuals) in Arizona than the other study centers (Supplementary Data) which could be consistent with less phenotypic measurement error in that study center (Almasy and Blangero, 2010); or it could reflect a true difference in heritability across study centers, for example if there is an exposure-by-gene interaction observable in the study center with highest arsenic exposure levels. An arsenic exposure-by-locus interaction is plausible as arsenic exposure has epigenetic effects across the genome (Bailey and Fry, 2014) and many genes may have differential expression by arsenic exposure (Chavan et al., 2011; Hou et al., 2014; Jutooru et al., 2010; Klei and Barchowsky, 2008; Tchounwou et al., 2003).

Localization of these broad linkage peaks is challenging due to the presence of multiple potentially relevant genes in these regions, and limited understanding of the biology of arsenic absorption, distribution, metabolism, and elimination in humans. The chromosome 6 linkage signal, suggestive for %iAs but not the other % arsenic species, is near several methyltransferase genes including the euchromatic histone-lysine N-methyltransferase EHMT2 (Kim et al., 2013), threonylcarbamoyladenosine tRNA methylthiotransferase (CDKAL1) (Okada et al., 2012), and cap-specific mRNA (nucleoside-2’-O-)-methyltransferase 1 (CMTR1) (Smietanski et al., 2014). However, it is also possible that the linkage peak is not due to a methyltransferase but to some other part of the arsenic kinetic machinery, known to include reduction enzymes and transporters as well (Hernandez and Marcos, 2008). Transporters near the linkage peak include HFE, which has been previously associated with kinetics of metals including iron (Hanson et al., 2001), lead (Hopkins et al., 2008), and manganese (Claus Henn et al., 2011); and ABCB2, which is important for antigen processing (Lankat-Buttgereit and Tampe, 2002) and may be relevant to the emerging research on arsenic immunotoxicity (Ahmed et al., 2011; Farzan et al., 2013). It is thought that arsenic metabolism may be related to creatinine pathways (Gamble and Hall, 2012; Thomas, 2009), and there is some toxicological evidence for an arsenic association with renal elimination of uric acid (Jauge and Del-Razo, 1985) although the epidemiological evidence is scarce (Kuo et al., 2015). We cannot exclude the possibility that some of the chromosome 10 evidence for linkage is explained by rare variants within AS3MT not tagged by the haplotype index SNPs. The principal components’ linkage analyses suggested that there may be distinct genetic determinants for %DMA versus other species, than for the balance of %iAs and %MMA after accounting for %DMA. The chromosome 5 region associated with principal component 2 includes many genes possibly relevant for arsenic kinetics, including solute carrier anion transporter family 4, member C1 (SLCO4C1) (Toyohara et al., 2009), and solute carrier anion transporter family 6, member A1 (SLCO6A1) (Lee et al., 2004) which both are in the same transporter super-family as the SLCO1B1 gene preliminarily associated with % arsenic species patterns in a sample from the Strong Heart Family Study (Gribble et al., 2013b; Roth et al., 2012). The chromosome 14 peak for principal component 2 is near potential candidate genes including tRNA methyltransferase 5 (TRMT5) (Christian et al., 2013), valosin-containing protein lysine (K) methyltransferase (VCPKMT) (Kernstock et al., 2012), and the transporter solute carrier family 38, member 6 (SLC38A6) (Sundberg et al., 2008). Linkage fine-mapping studies are needed to further evaluate the signals from this study.

This study has numerous strengths but also several limitations. Information on arsenic in drinking water for the study participants was not available, although arsenic in drinking water has been historically >10 µg/l in the communities from Arizona and in some of the communities from North and South Dakota. In the communities from Oklahoma arsenic levels in drinking water is <10 µg/l, and food is suspected as the main source of arsenic. The interpretation of % arsenic species as a biomarker of iAs metabolism and elimination is strengthened by the result that arsenobetaine was a small proportion of the total urine arsenic, consistent with little seafood intake in the population. There was a large sample size with rich family relationships allowing stable estimation of the co-segregation of genetic loci with arsenic species phenotypes. This study was restricted to participants without diabetes at baseline as it was nested within an incident diabetes risk factor study; future work should characterize these genetic relationships among persons with diabetes. Findings from this population may or may not generalize to other populations with different distributions of variables, such as body mass index, predictive of urine arsenic species patterns.

In conclusion, this study identified major loci likely involved in arsenic kinetic processes, located on chromosomes 5, 6, and 10. These loci, together with a few loci that might be relevant for Arizona communities, should be followed up in fine-mapping studies. The evidence for linkage differed by study region, and was only partly explained by common variation in AS3MT, indicating that rare variation may be an important part of inter-individual variability in arsenic susceptibility, or perhaps that additional genes on chromosome 10 are important for arsenic processing. The novel approach of considering the inter-relationship of the % arsenic species through principal components analysis provided additional insights by detecting a strong linkage signal on chromosome 5, and may be a fruitful approach for future arsenic susceptibility genetics research. Arsenic exposure has been associated with incident cardiovascular disease (Moon et al., 2013), cancer mortality (Garcia-Esquinas et al., 2013), and diabetes prevalence (Gribble et al., 2012) in the Strong Heart Study. Further evaluating the role of genetic determinants of arsenic dose and toxicity and of gene-environment interactions may contribute to the prevention and control of arsenic-related disease (Flanders, 2006; Rothman and Greenland, 2005).

FUNDING

National Heart Lung and Blood Institute (R01HL090863 and by SHS grants HL41642, HL41652, HL41654 and HL65521) and the National Institute of Environmental Health Sciences (R01ES021367); National Institute for Environmental Health Sciences (T32ES013678-07 to M.O.G.). M. Tellez-Plaza was supported by the Strategic Action for Research in Health Sciences (CP12/03080) from the Spanish Ministry of Economy and Competitiveness and the European Funds for Regional Development (FEDER). This investigation was conducted in facilities constructed with support from the National Center for Research Resources, National Institutes of Health (C06 RR013556 and C06 RR017515).

ACKNOWLEDGMENTS

The authors declare no conflicts of interest.

REFERENCES