https://orcid.org/0000-0003-2704-8912
Figures
Abstract
Inorganic arsenic is highly toxic and carcinogenic to humans. Exposed individuals vary in their ability to metabolize arsenic, and variability in arsenic metabolism efficiency (AME) is associated with risks of arsenic-related toxicities. Inherited genetic variation in the 10q24.32 region, near the arsenic methyltransferase (AS3MT) gene, is associated with urine-based measures of AME in multiple arsenic-exposed populations. To identify potential causal variants in this region, we applied fine mapping approaches to targeted sequencing data generated for exposed individuals from Bangladeshi, American Indian, and European American populations (n = 2,357, 557, and 648 respectively). We identified three independent association signals for Bangladeshis, two for American Indians, and one for European Americans. The size of the confidence sets for each signal varied from 4 to 85 variants. There was one signal shared across all three populations, represented by the same SNP in American Indians and European Americans (rs191177668) and in strong linkage disequilibrium (LD) with a lead SNP in Bangladesh (rs145537350). Beyond this shared signal, differences in LD patterns, minor allele frequency (MAF) (e.g., rs12573221 ~13% in Bangladesh ~0.2% among American Indians), and/or heterogeneity in effect sizes across populations likely contributed to the apparent population specificity of the additional identified signals. One of our potential causal variants influences AS3MT expression and nearby DNA methylation in numerous GTEx tissue types (with rs4919690 as a likely causal variant). Several SNPs in our confidence sets overlap transcription factor binding sites and cis-regulatory elements (from ENCODE). Taken together, our analyses reveal multiple potential causal variants in the 10q24.32 region influencing AME, including a variant shared across populations, and elucidate potential biological mechanisms underlying the impact of genetic variation on AME.
Author summary
Inorganic arsenic is highly toxic, and exposure to arsenic increases risk for multiple diseases, including cancer. Individuals differ in their ability to metabolize and excrete arsenic, in part due to inherited genetic variation in and around the AS3MT gene, and these differences impact arsenic toxicity risk. To identify candidate causal variants in the AS3MT region, we applied fine-mapping methods to targeted sequencing data from The Health Effects of Arsenic Longitudinal Study (HEALS), the Strong Heart Study (SHS), and the New Hampshire Skin Cancer Study (NHSCS) (Bangladesh, American Indian, and European American populations). We detected 3 independent association signals in HEALS, 2 in SHS, and 1 in NHSCS; and we identified a set of candidate causal variants for each of these signals. One of the identified signals represents a potential causal variant that impacts arsenic metabolism across all three populations. Using omics-QTL co-localization analyses, we show that some of the variants identified act through regulation AS3MT in multiple tissue types. Overall, this work increases our understanding of variation in the AS3MT region and its role in arsenic metabolism across populations.
Citation: Chernoff MB, Delgado D, Tong L, Chen L, Oliva M, Tamayo LI, et al. (2023) Sequencing-based fine-mapping and in silico functional characterization of the 10q24.32 arsenic metabolism efficiency locus across multiple arsenic-exposed populations. PLoS Genet 19(1): e1010588. https://doi.org/10.1371/journal.pgen.1010588
Editor: Scott M. Williams, Case Western Reserve University School of Medicine, UNITED STATES
Received: May 27, 2022; Accepted: December 20, 2022; Published: January 20, 2023
Copyright: © 2023 Chernoff et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All summary statistics generated with this study are included within the manuscript and its supporting information files. Individual-level data requests for all the data underlying results presented in the study can be requested by contacting IPPH@uchicago.edu. Requests will then be routed to the three individual studies, as there are different mechanisms for data access for each study. Normalized expression matrices, summary statistics for eQTLs and mQTLs, and covariates used for QTL mapping are available at the GTEx Portal (https://gtexportal.org/home/datasets). DNAm normalized data is available at GEO (GSE213478); access to the DNAm raw data is provided through the AnVIL platform (https://anvil.terra.bio/#workspaces/anvil-datastorage/AnVIL_GTEx_V9_hg38). All GTEx protected data are available via dbGaP (phs000424.v9 and phs000424.v8.p2). GTEx whole genome sequencing data can be requested through dbGaP (https://gtexportal.org/home/protectedDataAccess).
Funding: Targeted DNA sequencing across all cohorts was supported by National Institutes of Health grant R01 ES023834 (to B.L.P.) The Health Effects of Arsenic Longitudinal Study was supported by National Institutes of Health grants R35 ES028379 (to B.L.P.), R21 ES024834 (to B.L.P.), P42ES010349 (to J.G.), R01 CA107431 (to H.A.), P30 ES027792 (to H.A.), R24 ES028532 (to H.A.), and R24 TW009555 (to H.A.). The New Hampshire Skin Cancer Study was supported by U.S. National Institutes of General Medicine grant P20GM104416 (to M.R.K.) and by National Institutes of Health grants P42ES007373 (to C.Y. Chen) and R01CA057494 (to M.R.K.). The Strong Heart Study has been funded in whole or in part with federal funds from the National Heart, Lung, and Blood Institute, National Institute of Health, Department of Health and Human Services, under contract numbers 75N92019D00027, 75N92019D00028, 75N92019D00029, & 75N92019D00030. The study was previously supported by research grants: R01HL109315, R01HL109301, R01HL109284, R01HL109282, and R01HL109319 and R01HL090863, and by cooperative agreements: U01HL41642, U01HL41652, U01HL41654, U01HL65520, and U01HL65521 and from the National Institute of Environmental Health Sciences (P42ES033719, R01ES032638 and past grant R01ES021367). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This work was also supported by the National Institute of Environmental Health Sciences under the award number R35ES028379-03S1 (B.L.P.), the National Institute of General Medicine under award number T73M007281, The National Institute of Environmental Health Sciences award number 5F30ES031858-02 (M.B.C.), Susan G. Komen Research Training Grant under award number GTDR16376189, and the National Institute of Aging under award number T32AG51146-5. The Genotype-Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by NCI, NHGRI, NHLBI, NIDA, NIMH, and NINDS. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Arsenic-contaminated groundwater is a global public health issue, impacting >220 million individuals worldwide, with >85% of highly exposed individuals living in South Asia based on an exposure level of ≥10μg/L in drinking water [1,2]. The International Agency for Research on Cancer (IARC) classifies inorganic arsenic (iAs) as a “Group 1” human carcinogen [3] with chronic exposure increasing the risk of bladder [4], kidney, lung [5], liver, and skin cancers [3,6]. Exposure to iAs is also associated with increased risk for diabetes [7], as well as cardiovascular, cerebrovascular, and neurologic diseases [8–11]. A hallmark of chronic iAs exposure is the appearance of skin lesions, arsenical hyperkeratosis, typically on the hands and feet of exposed individuals [12,13]. The most common source of iAs exposure is contaminated drinking water [14] with ~2.1 million individuals in the U.S. [15] and 35 to 77 million in Bangladesh [16,17] exposed to iAs above 10ug/L, the maximum contaminant level set by the U.S. Environmental Protection Agency (EPA) [18] and World Health Organization (WHO) [1,19]. Other sources of arsenic exposure include the consumption of contaminated seafood and rice [20–22].
iAs metabolism in humans is composed of a series of reduction and methylation reactions occurring primarily in the liver with some metabolism potentially occurring in other tissues such as the kidney [23,24]. iAs in the form of arsenite (iAsIII) or arsenate (AsV) enters the body. Based on the Challenger model of metabolism [14,23,25], iAsV can be reduced to iAsIII, which can be methylated in a reaction catalyzed by arsenic (+3 oxidation state) methyltransferase (AS3MT) producing monomethyarsonic acid (MMAV) [14], which can be reduced to monomethylarsonous acid (MMAIII). A second methylation step, also catalyzed by AS3MT, produces dimethylarsinic acid (DMAV), which can be reduced to dimethylarsinous acid (DMAIII), the end-product of iAs metabolism in humans.
Individuals’ arsenic metabolism efficiency (AME) is often represented as the percentage of each arsenic species in urine relative to all species (iAs%, MMA%, and DMA%) [7,14], with higher DMA% indicating more efficient metabolism. While age, sex, and environmental factors contribute to inter-individual variation in AME [14,26], inherited genetic variation also plays an important role. Variation in the AS3MT/10q24.32 region has shown clear association with AME in multiple populations, including Bangladesh and American Indian communities in the US, with multiple independent association signals identified [14,27–29]. Prior studies have found that higher levels of urinary MMA% and/or lower levels of urinary DMA% are associated with increased risk for cancer, cardiovascular disease [30] and arsenic-induced skin lesions [14]. AME-associated SNPs have also been shown to impact the risk of arsenic-induced skin lesions, reflecting increased arsenic toxicity among those with lower AME [31].
The causal variants underlying the observed associations between AS3MT/10q24.32 variation and AME remain unknown. Previous studies have had limited SNP density, small sample sizes, and have focused on single populations. In this work, we generate targeted sequencing data for multiple arsenic-exposed populations to identify candidate causal variants underlying the association between 10q24.32 variants and AME. We perform in-silico functional annotation to assess potential functional impact and further prioritize potential causal variants. Finally, we conduct co-localization analyses to examine variants’ potential effects on AS3MT expression and/or nearby DNA methylation. This work enhances our understanding of the genetic mechanism underlying variation in AME and susceptibility to iAs toxicity.
Results
Arsenic exposure and AME
Total urinary arsenic varied substantially across cohorts, with higher concentrations in HEALS compared to SHS and NHSCS (Fig 1). This difference was also reflected in measures of arsenic in participants’ drinking water (S1 Fig). DMA% was highest in NHSCS and lowest in HEALS (Table 1 and S2 Fig). MMA% was highest in SHS and lowest in NHSCS while iAs% was highest in HEALS and lowest in NHSCS (Table 1).
Distributions of total urinary arsenic (μg/g of creatinine) are shown for three arsenic-exposed populations: The Health Effect of Arsenic Longitudinal Study (HEALS, in red), the New Hampshire Skin Cancer Study (NHSCS, in green), and the Strong Heart Study (SHS, in blue). The inset Fig shows SHS and NHSCS only.
Association analyses
Conditional association analysis of TOPMed-imputed data identified association signals for urinary DMA% in the 10q24.32 region for all three cohorts (Fig 2). We identified three independent signals in HEALS (lead SNPs rs145537350, rs12573221, and rs4919687), two in SHS (rs191177668 and rs4919688), and one in NHSCS (rs191177668). The per-allele association estimates for these lead SNPs varied in magnitude from ~2% to ~12% (in DMA% units) (Table 2). These results are similar to those observed for the non-imputed data (S3 Fig). Analysis of MMA% and iAs% across all three cohorts produced results generally consistent with those observed for DMA%, with similar association signals detected for all three arsenic species (S4, S5, S6 and S7 Figs). Alleles associated with increased DMA% tended to be associated with decreased MMA% and iAs%.
P-values are from linear models adjusted for age, sex, and population structure (in HEALS and SHS). The top panel for each population shows the primary association (adjusted for lead SNPs from all non-primary signals, if present). Additional panels show p-values for secondary (HEALS and SHS) and tertiary (HEALS) association signals (adjusted for lead SNPs from all other signals). Three signals were identified for HEALS, two for SHS, and one for NHSCS. LD estimates are based on several 1,000 Genomes populations (SAS for HEALS, MXL/PUR/CLM/PEL for SHS, and EUR for NHSCS).
SHS and NHSCS share a lead SNP (rs191177668) and this SNP is in LD (r2 = 0.5–1.0 depending on reference population) with a HEALS lead SNP, rs145537350 (Fig 3), suggesting a shared causal variant across cohorts. The shared lead SNP in SHS and NHSCS, rs191177668, has a p-value of 8x10-8 and beta estimate of -0.062 in HEALS (isolated primary signal). Controlling for the primary signal, this SNP has p-values of 0.6 and 0.2 in the HEALS secondary and tertiary analyses. We also observe two signals in HEALS and one in SHS that appear distinct (or population-specific), with minimal LD among the lead SNPs (Fig 3). Some lead SNPs differ substantially in MAF across populations (Table 3). For example, SHS lead SNP rs145537350 is common in SHS (MAF = 0.14) but has much lower frequency in HEALS (0.007) and NHSCS (0.005). We further examined MAF differences for the lead SNPs across SHS centers (S1 Table). Here, we find generally consistent MAF across SHS centers, with the exception of rs4919688 which is more common in the Arizona center (MAF = 0.482) compared with the Dakotas and Oklahoma centers (0.252, 0.221 respectively).
All LD estimates are based on 1,000 Genomes data obtained from LDLink. We observe strong LD (r2 = 0.495–1.0) between rs145537350 (HEALS) and rs191177668 (NHSCS and SHS), but weak LD among all other variants.
We examined the impact of BMI as a covariate in each cohort. While BMI is associated with DMA% in each cohort, the association for the lead SNPs identified in each cohort are largely unchanged. In the cohorts with available measurements of arsenic in drinking water (HEALS and NHSCS) water arsenic exposure was inversely associated with DMA% (S8 Fig), as reported previously [14,32]. However, there was no detectable interaction (P<0.05) between any of the lead SNPs and exposure level measured in drinking water.
Fine-mapping
The results of ancestry-specific fine-mapping of individual-level data performed using SuSiE [33,34]] were consistent with those observed in our association analyses, with three distinct 95% confidence sets identified in HEALS and two in SHS (Fig 4). We find overlapping SNPs in the confidence sets for HEALS set 1 and SHS set 1 (S2, S3, and S4 Tables). For SHS, the confidence sets identified through the analysis of summary statistics were somewhat different than those obtained from analyses of individual-level data, with SNPs in summary statistic based SHS confidence set 1 overlapping with SNPs in HEALS sets 1 and 3 (S2–S5 Tables). However, in both the summary statistic and individual-level data analyses, no overlap was observed between HEALS confidence set 2 and any SHS set or between SHS set 2 and any HEALS set. SuSiE did not identify a confidence set for NH, potentially due to the weaker associations observed in this group.
95% confidence sets for HEALS (A) and SHS (B) are each highlighted in different colors and mapped onto the overall DMA% association results.
We examined the shared signal through meta-analysis of the isolated HEALS and SHS association signal results using MANTRA. The 95% confidence set produced contained 8 variants including the lead primary SNPs in HEALS (rs145537350) and SHS (rs191177668) which had posterior inclusion probabilities of 0.488 and 0.251, respectively (S3 Table). This confidence set also included the lead tertiary HEALS SNP (rs4919687), a variant included in HEALS set 3 and in the SHS summary statistic-based set 1.
In-silico functional annotation
We aligned credible set variants with markers of open chromatin (Dnase I), cis-Credible Regulatory Elements (CREs), H3K27Ac, H3K4me3, and transcription factor (TF) binding sites (specifically TFs in cells and tissue types related to the kidney, liver, and heart). Three variants in HEALS confidence set 1 (Table 4 and S9 Fig), one variant in HEALS set 3 (Table 4 and Fig 5), and 12 of the 50 variants in the HEALS set 2 overlapped at least one of the examined features. Examining the 85 variant SHS confidence set that contains the shared lead variant in SHS and NHSCS (rs191177668) as well as the lead variant in HEALS (rs145537350), we find that 24 of the variants overlap functional features (S6 Table).
Candidate causal SNPs from HEALS confidence set 3 (lead SNP rs4919687) overlap with genomic features including candidate cis-regulatory elements and transcription factor binding sites. Highlighted panels show details of candidate causal SNPs overlapping these features. Note that AS3MT is located downstream of BORCS7 and overlaps BORCS7-AS3MT.
Cis-eQTL identification and co-localization analysis
AS3MT is expressed in most human tissue types, and this expression is highly variable ranging from 411.5 TPM in the adrenal gland to 0.92 in whole blood (S10 Fig). We identified cis-eQTLs for AS3MT in 45 tissue types (among 47 analyzed), with 27 tissue types having multiple AS3MT cis-eQTLs (as many as four). Following identification of these QTLs, we wanted to determine if any eQTLs shared a common causal variant with our AME association signals thereby representing a regulatory mechanism by which our identified SNPs impact AME. Cis-eQTLs in 22 tissue types had a lead eSNP in high LD (r2>0.7) with rs4919687, a lead DMA% SNP in HEALS found in confidence set 3. Among these, rs4919690 (in confidence set 3) was the lead eSNP in 14 tissue types. Eight tissue types had a lead eSNP in high LD with rs12573221 (lead SNP in HEALS confidence set 2) and 5 tissues had a lead eSNP in high LD with rs145537350 (lead SNP in HEALS set 1) and rs191177668 (shared lead SNP in SHS and NHSCS). Examining the individual lead SNPs associated with urinary DMA% in the context of our eQTL analysis, we found that HEALS lead SNP rs4919687 had p < 5x10-8 in forty tissue types and the shared lead SNP in SHS and NHSCS, rs191177668, had p-value < 5x10-8 in eight tissue types (based on primary eQTL analysis). The second lead SNP in SHS (rs4919688) was not identified in any of our QTL analyses.
Following the identification of eQTLs in high LD with our AME-associated SNPs, we performed a co-localization analysis to determine whether the same causal variants impact AME and AS3MT expression, providing a regulatory mechanism by which our identified SNPs influence AME. Under the assumption that 50% of DMA% SNPs are eQTLs (p12 = 5x10-6), we found evidence for co-localization between AS3MT cis-eQTLs and the DMA% signal represented by rs4919687 in 21 tissue types (PP of common causal variant (CCV) >80%) (Table 5). The rs4919687 allele associated with decreased AME (lower DMA%) was associated with lower AS3MT expression across all tissue types in which co-localization was observed (Fig 6B), consistent with a mechanism in which lower AS3MT mRNA levels result in lower protein levels and lower enzymatic activity; thereby decreasing AME.
We observe evidence of co-localization in 21 GTEx tissue types, with four example tissue types shown in Panel A (aorta, adipose-visceral omentum, lung, and nerve-tibial). The low efficiency (low DMA%) allele at rs4919687 is consistently associated with lower AS3MT expression in all tissues for which co-localization was observed.
Varying the prior probability (p12) of the percentage of DMA% SNPs that are also cis-eQTLs from 50% to 5% decreased the number of tissue types in which co-localization was observed (PP of CCV>80%) between AS3MT cis-eQTLs and rs4919687 from 21 to 4 (Table 5 and Fig 6A) and resulted in no co-localization between AS3MT cis-eQTLs and other DMA% signals (other than HEALS set 3).
Considering the additional lead signals identified in our population-specific analyses, we observed co-localization between DMA% signal rs12573221 (HEALS set 2) and an AS3MT eQTL in aortic artery. Even under the most liberal priors, we observed no evidence of co-localization for DMA% signals represented by rs145537350 (HEALS set 1) or rs191177668 (shared SNP in SHS and NHSCS).
Beyond AS3MT, we examined 27 additional genes within 500kb of AS3MT. HEALS DMA% lead SNPs rs4919687 (HEALS set 3) was in high LD with eQTLs for multiple genes (in at least one tissue type), including CYP17A1OS, AL356608.1, CYP17A1, BORCS7, NT5C2, and WBP1L. Among these, a BORCS7 eQTL (represented by rs11191421 and rs4919690) present in multiple tissues showed strong LD with HEALS lead SNP rs4919687 (set 3) in 43 tissue types. We also observed two genes, NFKB2 and RPARP-AS1, with cis-eQTLs whose lead SNPs were in high LD with a HEALS DMA% lead SNP rs12573221 (set 2) in tibial nerve and transverse colon tissues respectively. Finally, we observed one gene, SUFU, with a cis-eQTL in high LD with a DMA% association signal from SHS (rs191177668, association-based set 1) in the tibial artery.
Under the assumption that 50% of DMA% SNPs are cis-eQTLs (p12 = 5x10-6), we identified 25 tissue types in which a cis-eQTL for BORCS7 co-localized with the DMA% signal represented by rs4919687. Thus, both AS3MT and BORCS7 eQTLs co-localize with this DMA% signal. We also observed co-localization between this DMA% signal and tissue-specific eQTLs in CYP17A1OS (thyroid) and CYP17A1 (frontal cortex of the brain) (PP>80%) (S7 Table). Similar to AS3MT results, reducing p12 (to a 5% probability of co-localization), resulted in much lower probabilities of co-localization for BORCS7 (S7 Table).
Cis-mQTL identification and co-localization analysis
Among the nine tissue types analyzed, we identified mQTLs in strong LD with rs4919687 (HEALS set 3) in five tissue types: transverse colon, kidney cortex, lung, ovary, and testis. We also identified mQTLs in high LD with rs12573221 (HEALS set 2) in five tissue types: breast (mammary), lung, skeletal muscle, prostate, and testis. Finally, we identified mQTLs in high LD with rs145537350 (HEALS set 1) in four tissue types: breast (mammary), kidney cortex, lung, and whole blood.
Under the assumption that 50% of DMA% signals are mQTLs (p12 = 5x10-6), we observed co-localization between DMA% signal rs4919687 and mQTLs in two tissue types: colon-transverse (one CpG) and ovary (two CpGs) (S8 Table). For colon, the associated CpG was in the gene body of CYP17A1 while for ovary, both associated CpGs were in the gene body of AS3MT. We did not observe any co-localizations between mQTLs and the DMA% association signals represented by either rs12573221 (HEALS set 2) or rs145537350 (HEALS set 1).
We observed co-localizations between AS3MT cis-eQTL (rs4919690) and cg08650961 cis-mQTL (rs4919690) under the assumption that 50% of mQTLs are also eQTLs (p12 = 5x10-4) and the assumption that 25% of mQTLs are also eQTLs (p12 = 2.5x10-4).
Effect modification analysis
Using known associations between individual characteristics and DMA%, we performed a series of effect modifications analyses to understand the interactions between these variables and our identified variants. We found no interaction between sex and the genetic effects on DMA% (S9 Table); the effects of our identified variants were similar in males and females. We also examined the interaction between smoking and DMA% and observed evidence of interaction between current smoking and the lead SNP in NHSCS (p = 0.04), but not in HEALS or SHS.
Sensitivity analysis
Arsenobetaine, a form of organic arsenic found in seafood (see methods) was correlated with DMA% in all three cohorts (correlation of 0.18, 0.12, and 022 in HEALS, SHS, and NHSCS respectively); however, no clear differences were seen between our initial analysis results and those from a model that included adjustment for arsenobetaine. To further confirm our results, we performed a sensitivity analysis, excluding individuals with DMA% >85 (76 in HEALS, 224 in NHSCS, and 169 in SHS) to avoid the inclusion of DMA measures strongly influenced by exposure to organic arsenic. Here, we found no significant change in results with our lead SNPs still appearing among the top SNPs in the association analyses of all three populations.
Discussion
The relationship between AS3MT genotype, gene expression in the 10q24.32 region, and arsenic metabolism are well-established; however, questions remain regarding the precise causal variants driving these associations, the potential differences in associations across ancestry groups, as well as the specific genes in the region influenced by 10q24.32 genotypes across different tissue types. In this project, we applied fine-mapping approaches to sequencing-based genotype data in the 10q24.32 AME-associated region to identify candidate causal variants across three cohorts exposed to varying levels of arsenic in their drinking water. Fine-mapping analyses revealed that there are likely multiple causal variants in the 10q24.32 region impacting AME (represented by DMA%), with at least one causal variant likely shared across populations. In silico functional annotation and QTL co-localization further revealed that several of our candidate causal variants overlap regulatory features and impact expression of AS3MT and local DNA methylation.
Under the assumption of shared biological mechanisms and shared causal variants across populations, cross-population association analyses can narrow the list of potential causal variants in a region by identifying SNPs showing consistent evidence of association across all examined populations [35–37]. However, this was not the case in our study, as meta-analysis of shared association signals in HEALS and SHS produced a confidence set of 8 variants, a set larger than the corresponding confidence sets based on HEALS alone (4 and 7 variants). This is likely due to the broader signal observed in SHS, resulting in a confidence set with many more SNPs than those observed for HEALS (the result of more extensive LD among nearby variants in SHS). This extensive LD in SHS makes discriminating between potential causal variants more challenging in the meta-analysis context.
A novel finding of our analysis is the identification of associations that appear to be population-specific in both HEALS and SHS, which require further examination. Failure to replicate genetic associations across populations has been observed in many studies and has driven the increased emphasis on diversity in GWAS [38–43]. One explanation for a lack of replication is differences in allele frequencies, which can reduce power in populations with low MAF [34]. For example, one of the HEALS confidence sets (lead SNP rs12573221, confidence set 2) has a MAF range of 6.1–16.5% in HEALS and 0.1–4.4% in SHS; thus, the SNPs are likely too rare in SHS to be examined at the present sample sizes. Similarly, signal 2 (lead SNP rs4919688, confidence set 2) has a MAF of 26.8% in SHS and 2.6% in HEALS, which may explain the lack of replication of this SHS signal in HEALS. Our study used different sample sizes for our three populations. Larger sample sizes, as seen in HEALS, increase the study power and decrease the standard error of effect estimates, and thus improve our ability to identify association signals [44]. It is possible that we simply lacked the power to identify all association signals across all populations.
Beyond MAF, differences in LD likely contribute to the differences in observed signals across populations. HEALS signal three is represented by rs4919687 which has a MAF >0.1 in all populations. However, while this variant represents an independent candiate causal variant and confidence set (set 3) in HEALS, it co-occurs with SNPs from HEALS confidence set 1 in SHS confidence set 1 (derived from summary statics). Furthermore, we find the HEALS tertiary lead SNP (in HEALS confidence set 3) among the top 30 SNPs of the SHS primary association signal. Thus, a single confidence set in SHS (set 1) may capture two confidence sets in HEALS. Differences in LD patterns across the populations is a possible explanation for these observations [35,38,45], causing two distinct signals (in HEALS) to be indistinguishable in SHS.
In addition to these factors, differences in subject recruitment and inclusion among the studies used for this work may also have contributed to the differences in association signals. NHSCS included individuals both with and without skin cancer. Arsenic exposure and AME are risk factors for skin cancer, so this selection could contribute to biases in observed associations [30,46]. Chronic arsenic exposure is also associated with increased health risks, so the long-standing exposure in HEALS may also distinguish it from the other populations and make direct comparison challenging [46]. It is possible that the observed population-specific signals reflect true differences in genetic effects across populations. Gene-gene or gene-environment interactions could result in different effect size estimates across populations [38,40]. Differences in exposure level across populations may also impact the observed associations [40,41,43]. For instance, attenuation of SNP effects in populations with low exposure may result in low power to detect association [40]. We examined the gene-environment interaction between our lead SNPs and arsenic exposure in HEALS and while arsenic exposure did have an independent effect on DMA%, there was no evidence of effect modification between any of our identified variants and arsenic exposure (S10 Table).
We sought to understand the regulatory mechanisms by which the causal variants impact gene function using co-localization analyses focused on AS3MT and surrounding genes and DNA methylation features. AS3MT is expressed in most tissues at detectable levels (S10 Fig), allowing us to examine co-localization across a wide range of tissues. Expression is highest in the adrenal gland, potentially due to co-regulation with nearby CYP17A1 which plays a role in steroid hormone formation or for protection against arsenic which can disrupt endocrine function [47,48]. While co-localization was not detected for adrenal tissue, it was detected in many tissue types with low expression (e.g. subcutaneous adipose). This suggests that our results were not driven by the variability in expression across tissues and that low expression levels did not prevent QTL detection.
The liver is the major site of arsenic metabolism, but we observed only suggestive evidence of co-localization in this tissue type (Table 5 and S11 Fig) We detected AS3MT eQTLs in liver (rs4919690, p = 7.15x10-8), but our relatively small sample size may have limited our statistical power to robustly identify co-localization across all sets of priors analyzed.
Despite the ancestry mismatch between our Bangladeshi participants and the GTEx donors, we found compelling evidence of co-localization between our DMA% association signal (represented by rs4919687, HEALS set 3) and a multi-tissue cis-eQTL for AS3MT in 21 tissue types. The minor allele at SNP rs4919687 was associated with decreased DMA% and decreased AS3MT expression in tissues in which co-localization was observed, supporting the hypothesis that decreased expression results in lower amounts of the enzyme and ultimately lower AME. Our co-localization results suggest that HEALS SNP rs4919690 may be the causal variant underlying this signal (Posterior Inclusion Probability, PIP = 0.09), as it is the lead eSNP for the co-localizing AS3MT eQTL in 14 tissue types, and it is the lead eSNP for co-localizing BORCS7 eQTL in 9 tissue types. The repeated appearance of this SNP as a lead SNP in GTEx QTL analyses (genotyping based on whole-genome sequencing) suggests causality. Some co-localization was observed between AS3MT eQTLs and a second association signal in HEALS, suggesting that this variant may also impact arsenic metabolism by regulating AS3MT expression across tissues.
No prior evidence suggests a role for BORCS7 in arsenic metabolism; however, we observe co-localization between a DMA% association signal and cis-eQTLs for BORCS7 in multiple tissues (as observed previously [49]). Expression levels of AS3MT and BORCS7 are correlated in nearly all tissues in which co-localization is observed, suggesting co-regulation by a common causal variant (and potentially other mechanisms) or two causal variants in very strong LD. The correlated expression of the two genes has been noted previously [50]. A mouse strain carrying a human BORCS7/AS3MT was created to study arsenic metabolism as the AS3MT promoter abuts the 3’ UTR of BORCS7 [51]. However, this study did not describe any specific role for BORCS7 in arsenic metabolism. It is likely that the SNPs in this region are pleiotropic, influencing both AS3MT and the expression of the surrounding genes. Furthermore, co-regulation of these genes has been previously reported in multiple tissue types [31,49,50,52].
The co-localization of mQTLs with DMA% association signals further increases support for rs4919690 as a causal variant, as mQTLs for 10q24.32 CpGs co-localize with eQTLs (represented by lead SNP rs4919690) for AS3MT in lung tissue. The DMA% decreasing allele at rs4919690 (A) is associated with increased methylation of cg08650961, a CpG located in the body of the CNNM2 gene outside of a CpG island, and decreased AS3MT expression in lung tissue.
The population mismatch between our studies of AME (Bangladeshi and American Indian populations) and GTEx donors (primarily European ancestry) likely decreased our power to detect co-localization in both our eQTL and mQTL analysis. Furthermore, the smaller sample size of SHS likely increased the standard error of association estimates in this population, and when this is combined with the population mismatch between SHS and GTEx we likely had a decreased ability to detect co-localization in this population Our mQTL analysis sample sizes were also small, some n<100 samples, and our analyses were restricted to 9 tissue types with available DNAm data. This limited our power for mQTL detection and our ability to detect co-localization. Additionally, in the populations examined for this study, we are not able to fully assess the contribution of organic sources of arsenic to variability in our AME phenotype, which could potentially bias the associations observed.
In this study, we do not assess the association of AME-related SNPs with arsenic toxicity risks. However, we have previously shown that the association signals represented by rs12573221 and rs4919687 show clear associations with arsenic-induced skin lesion risk in HEALS [53]. Similarly, the novel signal for SHS and NH that we report here (rs191177668), which is in strong LD with and can serve as a proxy for the novel signal in HEALS (rs145537350), is also associated with skin lesion risk in the dataset previously described (OR = 1.64, CI = 1.2, 2.24) [53].
This study builds on previous work in the 10q24.32 region [29,31,49,54,55] in several ways. Previous studies have found signs of positive selection near the AS3MT gene associated with efficient arsenic metabolism [56]. They found selection signals in multiple populations including several from South America and East Asia and found that several the SNPs associated with the protective haplotype were also associated with MMA%. None of the identified SNPs showed a significant association with DMA%, though in the context of our study one SNP did appear in lD with the secondary lead SNP in SHS.
We leverage data from multiple arsenic-exposed cohorts with diverse ancestry in a single study, allowing us to consider both shared and population-specific effects of inherited genetic variation on AME. Additionally, our targeted sequencing data enabled us to identify a novel, independent association between 10q24.32 variation and DMA% in HEALS that we were unable to detect in our previous array-based work [28]. We further increased our sample size for cis-eQTL analyses using the latest data from GTEx and incorporated both expression and methylation data into our examination of the SNPs’ mechanism of action. Finally, we provide evidence that there are likely multiple causal SNPs within the 10q24.32 region associated with AME. Together, this allowed us to provide evidence regarding the potential causal variants and mechanisms underlying the established association between the 10q24.32 region and AME. Future studies can build on these findings, potentially establishing cellular or animal models with perturbations of potential causal sites in order to directly assess the impact of specific alleles on arsenic metabolism.
Methods
Ethics statement
The study protocol for the Health Effects of Arsenic Longitudinal Study (HEALS) was approved by the Institutional Review Boards of The University of Chicago, Columbia University, and the Bangladesh Medical Research Council. Details of the study were explained, and verbal informed consent was obtained from all participants. For the Strong Heart Study (SHS) the goals and procedures of the study were explained, and a signed consent form was obtained from each participant [57]. The study protocol was approved by the Indian Health Service Institutional Review Boards, and the participating communities [7,58]. The NHSCS was approved by the Committee for the Protection of Human Subjects of Dartmouth College. At enrollment, participants underwent a written, informed consent process [59].
Study populations
This project leveraged data from three studies of arsenic-exposed individuals: The Health Effects of Arsenic Longitudinal Study (HEALS), the Strong Heart Study (SHS), and the New Hampshire Skin Cancer Study (NHSCS) (Table 1).
HEALS [17] is a prospective cohort study of a population from Araihazar, Bangladesh exposed to iAs through contaminated well water. 11,746 adults aged 18–75 years were recruited at baseline (1999–2001). All provided verbal informed consent. Water samples were collected from all 5,966 wells in the study area and tested for arsenic. In-person interviews, clinical evaluations, and blood and spot urine sample collection occurred in participants’ homes by trained physicians using structured protocols. Follow-up in-person interviews were conducted biennially. The study protocol was approved by the Institutional Review Boards of The University of Chicago, Columbia University, and the Bangladesh Medical Research Council. Urinary arsenic species in baseline samples were previously measured for 4,794 participants by the Columbia University Trace Metals Core Laboratory. For this study, we included 2,426 individuals with both arsenic metabolite data and DNA available for targeted sequencing.
SHS [11,57,60] was established in 1989 to examine cardiovascular disease and risk factors among American Indian men and women. It includes participants from 12 tribes in the American Southwest, Northern Plains, and Southern Plains. Arsenic species were measured in 3,973 participants using procedures that have been described previously [61,62]. As some SHS participants are relatives, we restricted DNA for sequencing to 997 unrelated individuals with arsenic metabolite data and DNA available. We excluded 123 participants due to low coverage. An additional 6 were excluded due to missing arsenic species and/or covariate data. Therefore, our final dataset includes targeted sequencing for 868 individuals with arsenic species data.
NHSCS [63] is a population-based case-control study of squamous cell carcinoma of the skin. A total of 510 incident cases of histologically-confirmed, invasive squamous cell carcinoma were recruited and 483 NH residents, frequency matched on age and sex, were recruited as controls. Urine and home drinking water samples were collected and used to measure total water and urinary arsenic concentration urinary arsenic metabolites. Among the 993 participants, 288 lacked sufficient DNA and were excluded. This resulted in targeted sequencing for 706 participants with existing data on urinary arsenic metabolites.
Measurement of urinary arsenic metabolites
Separation of arsenic species in urine samples from all three populations was performed using high-performance liquid chromatography [8,17,63]. This procedure was followed by IPC-MS to quantify arsenic species in urine samples. For this analysis, iAsIII and iAsV were summed to obtain total iAs, and each arsenic species (iAs, MMA, and DMA) is expressed as a percentage of the sum of each of these three species (iAs+MMA+DMA). Arsenocholine (AsC) and arsenobetaine (AsB) are nontoxic forms of (organic) arsenic and were excluded from our primary analyses. Details regarding the measurement protocols and the limit of detection for each metabolite used by each study have been described previously and can be found in S1 Text (Supplemental Methods) [17,54,58,63].
The genotype-tissue expression project (GTEx)
SNP genotype and gene expression (RNA-seq) data from GTEx v8 [64,65] were used in this project, including data from 838 donors and 49 tissue types. The protocols for sequencing, processing, and quality control have been previously described [66,67].
Targeted DNA sequencing
Protocols for the collection and extraction of DNA in HEALS, SHS, and NHSCS have been previously described [17,57,63]. Targeted sequencing for the 10q24.32 region was performed in all three cohorts using an Illumina TruSeq Custom Amplicon (TSCA) Kit designed in Illumina’s DesignStudio. Variant selection for targeted sequencing focused on variants in the 99% confidence set for each of the two AME association signals previously reported in HEALS [28] (>200 variants). In addition, variants in moderate LD (r2 > 0.4 based on the 1000 Genomes (1KG) BEB population) with the lead variant of either association signal were also selected (42 and 208 variants for the primary and secondary signals respectively). An additional set of 527 variants was included to capture/tag all common variants in the 10q24.32 region. Coding regions of genes in the 10q24.32 region (AS3MT, BORCS7, CNNM2, CYP17A1, and NT5C2) were also selected to capture protein-coding variants. In total, 1,114 “targets” were selected (with targets corresponding to specific variants or exons of interest). The final TSCA design included 781 small regions across 1.5 Mb of the 10q24.32 region, each 400–450 bp in length. These amplicons covered 858 of our “targets” (with 256 targets being undesignable due to proximity to repetitive DNA sequences).
Read alignment and genetic variant calling
Targeting sequencing data from all three cohorts (HEALS, SHS, and NH) were processed at the University of Chicago Bioinformatics Core using the Genome Analysis Toolkit (GATK) [68] Best Practices Workflow for germline short variant discovery [69]. For variant calling, raw paired-end reads were mapped to the hg19 reference using the Novoalign software. Variants were called for each sample using GATK HaplotypeCaller in GVCF mode to produce intermediate GVCFs which were then consolidated to a single GVCF file with the GenomicsDBImport tool. We used this consolidated file to perform joint variant calling across all samples by cohort, using GATK GenotypeGVCFs, which provided a set of raw SNPs and indels in VCF format. Biallelic SNPs with any of the following properties were excluded: QualbyDepth (QD) <2.0, FisherStrand (FS) >60.0, RMSMappingQuality (MQ) <40.0, StrandOddsRatio (SOR) >3.0, MappingQualityRankSumTest (MQRankSum) <-12.5, or ReadPosRankSum <-8.0. Additionally, indels with one or more of the following properties were excluded: QD <2.0, FS >200.0, ReadPosRankSum <-20, Inbreeding Coefficient <-0.8, or SOR >10.0. Finally, samples with low coverage, indicated by Depth of Coverage (DP) <30 were removed.
Genotype quality control and imputation
Quality control (QC) of called SNPs was performed in PLINK v1.9 [70]. Beginning with 9,858 variants on chromosome 10 across all cohorts, QC included filtering by genotyping rate, sample missingness, and minor allele frequency (details in Supplemental Methods in S1 Text) and resulted in 455 10q24.32 variants in 2,357 samples in HEALS, 449 variants in 558 individuals in SHS, and 437 variants in 648 individuals in NHSCS. We imputed missing genotypes in the 10q24.32 region for each cohort using the TOPMed Imputation Server and the TOPMed reference panel [71], resulting in 36,468 10q24.32 variants in HEALS, 21,594 in SHS, and 28,401 in NHSCS.
Association analysis
Data were analyzed using PLINK v1.9 [70] for SHS and NHSCS and using Genome-wide Complex Train Analysis (GCTA) [72] for HEALS to allow adjustment for cryptic relatedness using a genetic relationship matrix in a linear mixed model [28,31,54,72,73]. To generate this kinship matrix, we used genome-wide SNP data available for 2,434 HEALS participants. All regression models were adjusted for age and sex. Both sex and age are associated with AME; therefore, they were included as covariates in order remove variation in the outcome of interest and increase the power of our analysis. For SHS, we further included five SNP-based principal components (PCs) as covariates to account for population structure within the dataset (PCs provided by SHS). PCs were not available for NHSCS due to lack of genome-wide SNP data. While the populations differ by weight, we do not perform association analyses across cohorts and therefore did not include BMI as a covariate in the models.
Our initial association analyses identified the SNP with the smallest P-value for each population (i.e., lead SNP). We then conditioned on the lead SNP(s) and ran a second association analysis for the region. We repeated this process of conditioning on identified lead SNPs until no additional, independent, signals were identified (P<5x10-6). To enable downstream co-localization analyses, we isolated each identified association signal by including the lead SNPs for all other identified signals as covariates. Arsenobetaine is an organic form of arsenic that is highly stable and the major arsenic species found in most seafood [22]. Arsenobetaine is a marker of seafood intake, a diet that can also contain arsenosugars and arsenolipids, which were not measured but which have the potential to contribute to DMA% [22]. We therefore examined the relationship between arsenobetaine and DMA% and adjusted for arsenobetaine in multivariate models.
Fine-mapping to identify candidate causal variants
We applied the Sum of Single Effects (SuSiE) [33] method (in R) to each cohort individually to identify confidence sets of candidate causal variants in each population. SuSiE uses an iterative Bayesian stepwise selection approach to produce credible sets of variables that are designed to be as small as possible while capturing the causal variant. SuSiE can take summary statistics as input or analyze individual-level data. Both options were used in this analysis to assess consistency. Under the assumption that causal variants are shared across populations, we can take advantage of population-specific patterns of linkage disequilibrium (LD) to narrow down the number of potential causal variants [45]. Therefore, we used MANTRA to meta-analyze the isolated association signals across cohorts [74,75].
In-silico annotation of credible sets
Confidence set variants were examined using the UCSC Genome Browser [76] and ENCODE data [77]. We identified overlapping regulatory features (DNase hypersensitivity, histone marks, transcription factor binding sites, etc.) for all variants in each set.
Identifying relevant eQTLs
We identified cis-eQTLs using expression data from 47 tissue types (cell lines excluded) from GTEx v8, with sample sizes ranging from 73 (Kidney Cortex) to 706 (Skeletal Muscle). We mapped cis-eQTLs for AS3MT and all genes within 500kb [64] using a series of linear regressions implemented in FastQTL [78] (previously described by GTEx and downloaded from https://github.com/francois-a/fastqtl). Regressions were adjusted for all covariates provided by GTEx for each tissue, including five genotyping PCs, sequencing platform and protocol, sex, and PEER factors (Probabilistic Estimation of Expression Residuals) which account for non-genetic factors that contribute to variation in gene expression [64,79].
To identify the eQTL(s) for a given gene, FastQTL finds the variant most strongly associated with gene expression (based on P-value). Once a lead SNP is identified, conditional analyses were used to identify additional independent cis-eQTLs and isolate each cis-eQTL signal (by adjusting for the lead SNPs for a given cis-eQTLs).
To identify eQTLs that may share a causal variant with a DMA% association signal, we identified cis-eQTL lead eSNPs in strong LD (r2 >0.7) with a DMA% lead SNP. Due to the varied ancestries of our cohorts (and GTEx), we considered LD estimated from multiple 1KG groups (European (EUR), Bengali (BEB), and Admixed Americans (AMR)).
Identifying relevant mQTLs
We used post-QC, inverse-normalized DNA methylation data generated for GTEx samples and post-QC genotype data from GTEx v8 [64,66,67,80]. Our analysis included a total of 856 samples from 367 donors [67]. The number of samples for each tissue type ranged from 42 to 190, and a total of 754,054 CpGs (measured using Illumina EPIC arrays) were examined across tissues [80]. The methods for mQTL identification have been described previously by Oliva et al [80]), and are the same as those used for our eQTL analysis and implemented in FastQTL [78] (https://github.com/broadinstitute/gtex-pipeline/tree/master/qtl). Multiple testing correction was performed [81] and conditional mQTL analysis was used to identify independent mQTLs using the same procedure as described for the eQTL analysis. We focused on mQTLs within the 10q24.32 AME-associated region (CpGs within 500 kb of AS3MT) and identified CpGs with mQTLs in high LD (r2>0.7) with at least one of our DMA% association signals.
Co-localization of DMA% and QTL signals
To determine if the observed DMA% association signals and the identified cis-eQTLs and mQTLs (from GTEx) share common causal variant(s), we conducted a Bayesian test for co-localization [81] using the coloc R package applied to all overlapping SNPs between the sets of summary statistics [82–84]. We applied coloc to our isolated DMA% association signals paired with the tissue type-specific, isolated cis-QTL association signals for all pairs for which the lead SNPs showed high LD (r2 >0.7) in a relevant 1KG population. The ancestry mismatch among GTEx (primarily European), HEALS (Bangladeshi), and SHS (American Indian) led us to examine LD estimates based on EUR, SAS, and AMR populations (for HEALS and SHS respectively) using data from the 1KG Project implemented in LDlink [85,86].
For co-localization, we specified a prior probability for a SNP being causal only for DMA% (p1), only for gene expression or methylation (p2), and for both DMA% and expression/methylation (p12). We set the overall probability of being causal for DMA% (p1 + p12) as 10−5 and the overall probability of being an eQTL as 10−3 (p2 + p12), which is based on the average number of eGenes identified in GTEx and is typical in studies of complex traits and eQTLs. We then conducted a series of analyses in which the value of p12 was varied to correspond to a 5%, 10%, 25%, and 50% probability that a causal variant for DMA% is also an eQTL. This corresponded to four values of p12: 5×10−7, 2×10−6, 2.5×10−6, and 5×10−6. To assess co-localization between our tissue-specific, isolated mQTL signals and eQTL signals, we set the overall probability of being associated with an eQTL (p1 + p12) as 10−3 and the overall probability of being an mQTL as 10−3 (p2 + p12). We conducted a series of analyses in which the value of p12 was varied to correspond to a 5%, 10%, 25%, and 50% probability that a causal variant for an eQTL is also an mQTL (p12: 5×10−5, 1×10−4, 2.5×10−4, and 5×10−4).
Co-localization analyses produce a posterior probability (PP) of co-localization (H4), the PP that different causal variants underlie the two signals (H3), the PP there is an identifiable causal variant for DMA% but no detectable QTL signal (H1, and vice versa, H2), and the PP of no identifiable causal variant or QTLs (H0).
Supporting information
S1 Text.
a. Supplemental Methods: i. The Genotype-Tissue Expression Project. ii. Measurement of Urinary Arsenic Metabolites. iii. Quality Control for GTEx Whole Genome Sequencing Data. iv. Genotype Quality Control and Imputation. v. GTEx Expression Data Quantification. vi. Read Alignment and Genetic Variant Calling. vii. MANTRA meta-analysis.
https://doi.org/10.1371/journal.pgen.1010588.s001
(PDF)
S1 Fig. Arsenic Exposure Across Arsenic-Exposed Populations.
a. Measures of total water arsenic (μg/L) measured for individuals in three arsenic-exposed populations. The Health Effect of Arsenic Longitudinal Study (HEALS, in red), the New Hampshire Skin Cancer Study (NHSCS, in blue). Measurements in HEALS are based off arsenic levels in wells frequented by the individual participants; in NHSCS measurements were taken from participants’ homes.
https://doi.org/10.1371/journal.pgen.1010588.s002
(PDF)
S2 Fig. Distribution of urinary DMA percent across cohorts.
a. DMA% distribution in three arsenic-exposed populations. The Health Effect of Arsenic Longitudinal Study (HEALS, in red), the New Hampshire Skin Cancer Study (NHSCS, in green), and the Strong Heart Study (SHS, in blue).
https://doi.org/10.1371/journal.pgen.1010588.s003
(PDF)
S3 Fig. Non-imputed DMA% Conditional Association Results.
a. Results of a pre-imputation genetic association study of arsenic metabolism efficiency (DMA%) in the 10q24.32 region in three arsenic-exposed populations. P-values were generated with linear models adjusted for age and sex as well as kinship (HEALS) and population structure (SHS). The SNP with the strongest association is labeled in each panel. The top panel for each population shows the overall association results, the next shows p-values from models adjusted for the initial lead SNP, and the bottom panel shows the result of models adjusted for both previously identified variants. Three variants were identified in the Health Effect of Arsenic Longitudinal Study (HEALS), four in the Strong Heart Study (SHS), and one in New Hampshire Skin Cancer Study (NHSCS).
https://doi.org/10.1371/journal.pgen.1010588.s004
(PDF)
S4 Fig. HEALS MMA% Conditional Association Results.
a. Results of a post-imputation genetic association study of MMA% in the 10q24.32 region in HEALS. P-values were generated with linear models adjusted for age, sex, and kinship. The top panel represents overall association results, the second shows p-values from models adjusted for the initial lead SNP, and the bottom panel shows the result of models adjusted for both previously identified variants. Two lead variants were identified in this analysis: chr10:102842818 and chr10:102888092. We further note that the primary signal likely captures the tertiary signal identified in the analysis of DMA%.
https://doi.org/10.1371/journal.pgen.1010588.s005
(PDF)
S5 Fig. HEALS iAs% Conditional Association Results.
a. Results of a post-imputation genetic association study of iAS% in the 10q24.32 region in HEALS. P-values were generated with linear models adjusted for age, sex, and kinship. The top panel represents overall association results, the second shows p-values from models adjusted for the initial lead SNP, and the bottom panel shows the result of models adjusted for both previously identified variants. Three lead variants were identified in this analysis: chr10:103078084 (rs145537350), chr10:103089387 (rs12573221) and chr10:102853348. We further note that the primary and secondary signals are identical to two signals identified in the analysis of DMA%.
https://doi.org/10.1371/journal.pgen.1010588.s006
(PDF)
S6 Fig. SHS MMA% Conditional Association Results.
a. Results of a post-imputation genetic association study of MMA% in the 10q24.32 region in SHS. P-values were generated with linear models adjusted for age, sex, and population structure. The top panel represents overall association results, the second shows p-values from models adjusted for the initial lead SNP, and the bottom panel shows the result of models adjusted for both previously identified variants. Two lead variants were identified in this analysis: chr10:103078084 (rs145537350) and chr10:102842863 (rs4919688). We further note that the second signal is identical to the secondary signal identified in the analysis of DMA% and that the lead DMA% signal was the second most significant signal in the primary analysis and is in strong LD with the identified lead variant.
https://doi.org/10.1371/journal.pgen.1010588.s007
(PDF)
S7 Fig. SHS iAs% Conditional Association Results.
a. Results of a post-imputation genetic association study of iAS% in the 10q24.32 region in SHS. P-values were generated with linear models adjusted for age, sex, and population structure. The top panel represents overall association results, the second shows p-values from models adjusted for the initial lead SNP, and the bottom panel shows the result of models adjusted for both previously identified variants. Two lead variants were identified in this analysis: chr10:103213304 and chr10:103015726.
https://doi.org/10.1371/journal.pgen.1010588.s008
(PDF)
S8 Fig. Correlation of DMA% with arsenic exposure (based on drinking water arsenic concentration in HEALS (A) and NHSCS (B).
a. In both cohorts, we observe a negative correlation in which higher exposure is associated with lower DMA%.
https://doi.org/10.1371/journal.pgen.1010588.s009
(PDF)
S9 Fig. Genomic annotations of HEALS confidence set 1.
a. Results of in silico-functional analysis reveals overlap with genomic annotations for candidate causal SNPs from HEALS confidence set 1, corresponding to the primary association signal (lead SNP rs145537350). Multiple SNPs in the confidence set overlap genomic features including candidate cis-regulatory elements and transcription factor binding sites. Highlighted panels show details of candidate causal SNPs overlapping these features.
https://doi.org/10.1371/journal.pgen.1010588.s010
(PDF)
S10 Fig. Distribution of AS3MT expression across human tissues.
a. A. The distribution of AS3MT expression across human tissues reveals higher expression in the adrenal gland compared with all other available tissues. B. With the removal of the adrenal gland, we can see some variation in AS3MT expression across tissue types. Fig produced using the GTEx portal.
https://doi.org/10.1371/journal.pgen.1010588.s011
(PDF)
S11 Fig. Co-localization between HEALS DMA% association signal (rs4919687) and AS3MT eQTLs in the Liver.
a. We detect AS3MT eQTLs in the liver and observe evidence of co-localization (Panel A), but this evidence was not consistent across all sets of priors analyzed. The low efficiency (low DMA%) allele at rs4919687 is associated with lower AS3MT expression in the liver, though this pattern is observed more strongly in other tissue types (Panel B).
https://doi.org/10.1371/journal.pgen.1010588.s012
(PDF)
S1 Table. Minor Allele Frequency (MAF) comparison of lead signals across SHS centers.
https://doi.org/10.1371/journal.pgen.1010588.s013
(DOCX)
S2 Table. HEALS Confidence Sets (C.S.) from population-specific fine- mapping analysis.
https://doi.org/10.1371/journal.pgen.1010588.s014
(DOCX)
S3 Table. SHS Confidence Sets (C.S.) from population-specific fine- mapping analysis based on summary association statistics.
https://doi.org/10.1371/journal.pgen.1010588.s015
(DOCX)
S4 Table. SHS Confidence Sets (C.S.) from population-specific fine- mapping analysis based on primary sequencing data.
https://doi.org/10.1371/journal.pgen.1010588.s016
(DOCX)
S5 Table. Results of MANTRA meta-analysis of shared signal in HEALS and SHS.
https://doi.org/10.1371/journal.pgen.1010588.s017
(DOCX)
S6 Table. In-Silico Functional Examination of SHS Primary Sequencing Data Confidence Set 2.
https://doi.org/10.1371/journal.pgen.1010588.s018
(DOCX)
S7 Table. Co-localization of eQTLs for genes in the 10q24.32 region (excluding AS3MT) DMA% association signal (lead SNP rs4919687) identified in Bangladeshi individuals with posterior probability of colocalization > 80% (for p12 = 5x10-6).
https://doi.org/10.1371/journal.pgen.1010588.s019
(DOCX)
S8 Table. Co-localization (PP>80%) of mQTLs in the 10q24.32 region identified in GTEx tissues with the association signal for DMA% (lead SNP rs4919687) identified in a Bangladeshi population across a range of prior probabilities.
https://doi.org/10.1371/journal.pgen.1010588.s020
(DOCX)
S9 Table. Effect modification analysis of the effect of the interaction between DMA% associated variants and sex on DMA% in HEALS.
https://doi.org/10.1371/journal.pgen.1010588.s021
(DOCX)
S10 Table. Gene-Environment analysis of the effect of the interaction between DMA% associated variants and water arsenic concentration on DMA%.
https://doi.org/10.1371/journal.pgen.1010588.s022
(DOCX)
S11 Table. AS3MT eQTL Analysis Detailed Results.
https://doi.org/10.1371/journal.pgen.1010588.s023
(XLSX)
S12 Table. non-AS3MT gene eQTL Analysis Detailed Results.
https://doi.org/10.1371/journal.pgen.1010588.s024
(XLSX)
S1 Data. Data Summary Statistics by Population.
a. HEALS Isolated Primary Association. b. HEALS Isolated Secondary Association. c. HEALS Tertiary Association. d. NHSCS Isolated Primary Association. e. SHS Isolated Primary Association. f. SHS Secondary Association.
https://doi.org/10.1371/journal.pgen.1010588.s026
(ZIP)
Acknowledgments
We thank all the men and women who participated in the Health Effects of Arsenic Longitudinal Study, the New Hampshire Skin Cancer Study, and the Strong Heart Study and all the research staff who contributed to data collection and processing.
Web Resources
The Genotype-Tissue Expression (GTEx) Project: https://gtexportal.org/home/
FastQTL [50] (previously described by GTEx and downloaded from https://github.com/francois-a/fastqtl)
References
- 1. Ravenscroft P, Brammer H, Richards K. Arsenic Pollution: A Global Synthesis. RGS-IBG Book Ser. 2009 Apr 30;1.
- 2. Podgorski J, Berg M. Global threat of arsenic in groundwater. Science. 2020 May 22;368(6493):845–50. pmid:32439786
- 3.
IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. Some drinking-water disinfectants and contaminants, including arsenic. IARC Monogr Eval Carcinog Risks Hum. 2004;84:1–477.
- 4. Di Giovanni P, Di Martino G, Scampoli P, Cedrone F, Meo F, Lucisano G, et al. Arsenic Exposure and Risk of Urothelial Cancer: Systematic Review and Meta-Analysis. Int J Environ Res Public Health. 2020 Jan;17(9):3105. pmid:32365627
- 5. Wei S, Zhang H, Tao S. A review of arsenic exposure and lung cancer. Toxicol Res. 2019 May 1;8(3):319–27. pmid:31160966
- 6. Chen YC, Guo YLL, Su HJJ, Hsueh YM, Smith TJ, Ryan LM, et al. Arsenic methylation and skin cancer risk in southwestern Taiwan. J Occup Environ Med. 2003 Mar;45(3):241–8. pmid:12661181
- 7. Kuo CC, Howard BV, Umans JG, Gribble MO, Best LG, Francesconi KA, et al. Arsenic Exposure, Arsenic Metabolism, and Incident Diabetes in the Strong Heart Study. Diabetes Care. 2015 Apr 1;38(4):620–7. pmid:25583752
- 8. Moon K, Guallar E, Navas-Acien A. Arsenic Exposure and Cardiovascular Disease:An Updated Systematic Review. Curr Atheroscler Rep. 2012 Dec 1;14(6):542–55. pmid:22968315
- 9. Carroll CR, Noonan C, Garroutte EM, Navas-Acien A, Verney SP, Buchwald D. Low-level inorganic arsenic exposure and neuropsychological functioning in American Indian elders. Environ Res. 2017 Jul;156:74–9. pmid:28334644
- 10. Grandjean P, Herz KT. Trace elements as paradigms of developmental neurotoxicants: Lead, methylmercury and arsenic. J Trace Elem Med Biol. 2015 Jul 1;31:130–4. pmid:25175507
- 11. Astrid Suchy-Dicey, Carolyn Noonan, Ekaterina Burduli, Mateen Farrah J., Longstreth W.T., Buchwald Dedra, et al. Urinary Arsenic and Cadmium Associations with Findings from Cranial MRI in American Indians: Data from the Strong Heart Study. Environ Health Perspect. 128(12):127009. pmid:33332184
- 12. Zeng Q, Zhang A. Assessing potential mechanisms of arsenic-induced skin lesions and cancers: Human and in vitro evidence. Environ Pollut. 2020 May 1;260:113919.
- 13.
Arsenic Toxicity: What are the Physiologic Effects of Arsenic Exposure? | Environmental Medicine | ATSDR [Internet]. 2021 [cited 2021 Jun 4]. Available from: https://www.atsdr.cdc.gov/csem/arsenic/physiologic_effects.html.
- 14. Jansen RJ, Argos M, Tong L, Li J, Rakibuz-Zaman M, Islam MT, et al. Determinants and Consequences of Arsenic Metabolism Efficiency among 4,794 Individuals: Demographics, Lifestyle, Genetics, and Toxicity. Cancer Epidemiol Prev Biomark. 2016 Feb 1;25(2):381–90. pmid:26677206
- 15. Ayotte JD, Medalie L, Qi SL, Backer LC, Nolan BT. Estimating the High-Arsenic Domestic-Well Population in the Conterminous United States. Environ Sci Technol. 2017 Nov 7;51(21):12443–54. pmid:29043784
- 16. Smith AH, Lingas EO, Rahman M. Contamination of drinking-water by arsenic in Bangladesh: a public health emergency. Bull World Health Organ. 2000;78(9):1093–103. pmid:11019458
- 17. Ahsan H, Chen Y, Parvez F, Argos M, Hussain AI, Momotaj H, et al. Health Effects of Arsenic Longitudinal Study (HEALS): description of a multidisciplinary epidemiologic investigation. J Expo Sci Environ Epidemiol. 2006 Mar;16(2):191–205. pmid:16160703
- 18.
Arsenic and Clarifications to Compliance and New Source Monitoring Rule: A Quick Reference Guide.: 2.
- 19. Mohammed Abdul KS, Jayasinghe SS, Chandana EPS, Jayasumana C, De Silva PMCS. Arsenic and human health effects: A review. Environ Toxicol Pharmacol. 2015 Nov 1;40(3):828–46.
- 20. Melkonian S, Argos M, Hall MN, Chen Y, Parvez F, Pierce B, et al. Urinary and Dietary Analysis of 18,470 Bangladeshis Reveal a Correlation of Rice Consumption with Arsenic Exposure and Toxicity. PLOS ONE. 2013 Nov 15;8(11):e80691. pmid:24260455
- 21. Barrett JR. Rice versus Drinking Water: Estimating the Primary Source of Arsenic in the U.S. Diet. Environ Health Perspect. 2017 Jun 30;125(6):064004. pmid:28728137
- 22. Taylor V, Goodale B, Raab A, Schwerdtle T, Reimer K, Conklin S, et al. Human exposure to organic arsenic species from seafood. Sci Total Environ. 2017 Feb 15;580:266–82. pmid:28024743
- 23. Drobná Z, Walton FS, Paul DS, Xing W, Thomas DJ, Stýblo M. Metabolism of arsenic in human liver: the role of membrane transporters. Arch Toxicol. 2010 Jan 1;84(1):3–16. pmid:20020104
- 24.
Arsenic Toxicity: What is the Biologic Fate of Arsenic in the Body? | Environmental Medicine | ATSDR [Internet]. 2021 [cited 2022 Aug 23]. Available from: https://www.atsdr.cdc.gov/csem/arsenic/biologic_fate.html.
- 25. Drobna Z, Styblo M, Thomas DJ. An Overview of Arsenic Metabolism and Toxicity. Curr Protoc Toxicol Editor Board Mahin Maines Ed—Chief Al. 2009;42(431):4.31.1–4.31.6. pmid:25419261
- 26. Shen H, Niu Q, Xu M, Rui D, Xu S, Feng G, et al. Factors Affecting Arsenic Methylation in Arsenic-Exposed Humans: A Systematic Review and Meta-Analysis. Int J Environ Res Public Health. 2016 Feb;13(2):205. pmid:26861378
- 27. Agusa T, Fujihara J, Takeshita H, Iwata H. Individual Variations in Inorganic Arsenic Metabolism Associated with AS3MT Genetic Polymorphisms. Int J Mol Sci. 2011 Apr;12(4):2351–82. pmid:21731446
- 28. Pierce BL, Kibriya MG, Tong L, Jasmine F, Argos M, Roy S, et al. Genome-wide association study identifies chromosome 10q24.32 variants associated with arsenic metabolism and toxicity phenotypes in Bangladesh. PLoS Genet. 2012;8(2):e1002522. pmid:22383894
- 29. Balakrishnan P, Vaidya D, Franceschini N, Voruganti VS, Gribble MO, Haack K, et al. Association of Cardiometabolic Genes with Arsenic Metabolism Biomarkers in American Indian Communities: The Strong Heart Family Study (SHFS). Environ Health Perspect. 2017 Jan;125(1):15–22. pmid:27352405
- 30. Kuo Chin-Chi Moon Katherine A., Wang Shu-Li Silbergeld Ellen, Ana Navas-Acien. The Association of Arsenic Metabolism with Cancer, Cardiovascular Disease, and Diabetes: A Systematic Review of the Epidemiological Evidence. Environ Health Perspect. 125(8):087001. pmid:28796632
- 31. Pierce BL, Tong L, Argos M, Gao J, Jasmine F, Roy S, et al. Arsenic metabolism efficiency has a causal role in arsenic toxicity: Mendelian randomization and gene-environment interaction. Int J Epidemiol. 2013 Dec;42(6):1862–72. pmid:24536095
- 32. Ahsan H, Chen Y, Kibriya MG, Slavkovich V, Parvez F, Jasmine F, et al. Arsenic Metabolism, Genetic Susceptibility, and Risk of Premalignant Skin Lesions in Bangladesh. Cancer Epidemiol Prev Biomark. 2007 Jun 1;16(6):1270–8. pmid:17548696
- 33. Wang G, Sarkar A, Carbonetto P, Stephens M. A simple new approach to variable selection in regression, with application to genetic fine-mapping. bioRxiv. 2020 Jun 27;501114.
- 34. Wang G, Sarkar A, Carbonetto P, Stephens M. A simple new approach to variable selection in regression, with application to genetic fine mapping. J R Stat Soc Ser B Stat Methodol. 2020;82(5):1273–300.
- 35. Teo YY, Ong RTH, Sim X, Tai ES, Chia KS. Identifying candidate causal variants via trans-population fine-mapping. Genet Epidemiol. 2010;34(7):653–64. pmid:20839287
- 36. Wu Y, Waite LL, Jackson AU, Sheu WHH, Buyske S, Absher D, et al. Trans-Ethnic Fine-Mapping of Lipid Loci Identifies Population-Specific Signals and Allelic Heterogeneity That Increases the Trait Variance Explained. PLoS Genet [Internet]. 2013 Mar 21 [cited 2021 May 24];9(3). Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3605054/. pmid:23555291
- 37. Saccone NL, Saccone SF, Goate AM, Grucza RA, Hinrichs AL, Rice JP, et al. In search of causal variants: refining disease association signals using cross-population contrasts. BMC Genet. 2008 Aug 29;9:58. pmid:18759969
- 38. Sirugo G, Williams SM, Tishkoff SA. The Missing Diversity in Human Genetic Studies. Cell. 2019 Mar 21;177(1):26–31. pmid:30901543
- 39. Galinsky KJ, Reshef YA, Finucane HK, Loh PR, Zaitlen N, Patterson NJ, et al. Estimating cross-population genetic correlations of causal effect sizes. Genet Epidemiol. 2019;43(2):180–8. pmid:30474154
- 40. Carlson CS, Matise TC, North KE, Haiman CA, Fesinmeyer MD, Buyske S, et al. Generalization and Dilution of Association Results from European GWAS in Populations of Non-European Ancestry: The PAGE Study. PLOS Biol. 2013 Sep 17;11(9):e1001661. pmid:24068893
- 41. Adeyemo A, Rotimi C. Genetic Variants Associated with Complex Human Diseases Show Wide Variation across Multiple Populations. Public Health Genomics. 2009 Dec;13(2):72–9. pmid:19439916
- 42. Drake KA, Torgerson DG, Gignoux CR, Galanter JM, Roth LA, Huntsman S, et al. A genome-wide association study of bronchodilator response in Latinos implicates rare variants. J Allergy Clin Immunol. 2014 Feb 1;133(2):370–378.e15. pmid:23992748
- 43. Tang H. Confronting ethnicity-specific disease risk. Nat Genet. 2006 Jan;38(1):13–5. pmid:16380723
- 44. de Candia TR, Lee SH, Yang J, Browning BL, Gejman PV, Levinson DF, et al. Additive Genetic Variation in Schizophrenia Risk Is Shared by Populations of African and European Descent. Am J Hum Genet. 2013 Sep 5;93(3):463–70. pmid:23954163
- 45. Schaid DJ, Chen W, Larson NB. From genome-wide associations to candidate causal variants by statistical fine-mapping. Nat Rev Genet. 2018 Aug;19(8):491–504. pmid:29844615
- 46. Argos M, Kalra T, Pierce BL, Chen Y, Parvez F, Islam T, et al. A Prospective Study of Arsenic Exposure From Drinking Water and Incidence of Skin Lesions in Bangladesh. Am J Epidemiol. 2011 Jul 15;174(2):185–94. pmid:21576319
- 47. Gomez-Rubio P, Meza-Montenegro MM, Cantu-Soto E, Klimecki WT. Genetic association between intronic variants in AS3MT and arsenic methylation efficiency is focused on a large linkage disequilibrium cluster in chromosome 10. J Appl Toxicol JAT. 2010 Apr;30(3):260–70. pmid:20014157
- 48. Sun HJ, Xiang P, Luo J, Hong H, Lin H, Li HB, et al. Mechanisms of arsenic disruption on gonadal, adrenal and thyroid endocrine systems in humans: A review. Environ Int. 2016 Oct;95:61–8.
- 49. Chernoff M, Tong L, Demanelis K, Vander Griend D, Ahsan H, Pierce BL. Genetic Determinants of Reduced Arsenic Metabolism Efficiency in the 10q24.32 Region Are Associated With Reduced AS3MT Expression in Multiple Human Tissue Types. Toxicol Sci. 2020 Aug 1;176(2):382–95. pmid:32433756
- 50. Li M, Jaffe AE, Straub RE, Tao R, Shin JH, Wang Y, et al. A human-specific AS3MT isoform and BORCS7 are molecular risk factors in the 10q24.32 schizophrenia-associated locus. Nat Med. 2016 Jun;22(6):649–56. pmid:27158905
- 51. Koller BH, Snouwaert JN, Douillet C, Jania LA, El-Masri H, Thomas DJ, et al. Arsenic Metabolism in Mice Carrying a BORCS7/AS3MT Locus Humanized by Syntenic Replacement. Environ Health Perspect. 2020 Aug;128(8):87003. pmid:32779937
- 52. Duarte RRR, Troakes C, Nolan M, Srivastava DP, Murray RM, Bray NJ. Genome-wide significant schizophrenia risk variation on chromosome 10q24 is associated with altered cis-regulation of BORCS7, AS3MT, and NT5C2 in the human brain. Am J Med Genet Part B Neuropsychiatr Genet Off Publ Int Soc Psychiatr Genet. 2016 Sep;171(6):806–14.
- 53. Pierce BL, Tong L, Dean S, Argos M, Jasmine F, Rakibuz-Zaman M, et al. A missense variant in FTCD is associated with arsenic metabolism and toxicity phenotypes in Bangladesh. PLOS Genet. 2019 Mar 20;15(3):e1007984. pmid:30893314
- 54. Delgado DA, Chernoff M, Huang L, Tong L, Chen L, Jasmine F, et al. Rare, Protein-Altering Variants in AS3MT and Arsenic Metabolism Efficiency: A Multi-Population Association Study. Environ Health Perspect. 129(4):047007. pmid:33826413
- 55. Karagas MR, Gossai A, Pierce B, Ahsan H. Drinking Water Arsenic Contamination, Skin Lesions, and Malignancies: A Systematic Review of the Global Evidence. Curr Environ Health Rep. 2015 Mar 1;2(1):52–68. pmid:26231242
- 56. De Loma J, Vicente M, Tirado N, Ascui F, Vahter M, Gardon J, et al. Human adaptation to arsenic in Bolivians living in the Andes. Chemosphere. 2022 Aug 1;301:134764. pmid:35490756
- 57. Lee ET, Welty TK, Fabsitz R, Cowan LD, Le NA, Oopik AJ, et al. The Strong Heart Study. A study of cardiovascular disease in American Indians: design and methods. Am J Epidemiol. 1990 Dec;132(6):1141–55. pmid:2260546
- 58. Navas -Acien Ana, Umans JG, Howard BV, Goessler W, Francesconi KA, Crainiceanu CM, et al. Urine Arsenic Concentrations and Species Excretion Patterns in American Indian Communities Over a 10-year Period: The Strong Heart Study. Environ Health Perspect. 2009 Sep 1;117(9):1428–33. pmid:19750109
- 59. Rees JR, Zens MS, Gui J, Celaya MO, Riddle BL, Karagas MR. Non Melanoma Skin Cancer and Subsequent Cancer Risk. PLOS ONE. 2014 Jun 17;9(6):e99674. pmid:24937304
- 60.
Research Overview [Internet]. [cited 2021 Mar 2]. Available from: https://strongheartstudy.org/Research/Research-Overview.
- 61. Zheng LY, Umans JG, Yeh F, Francesconi KA, Goessler W, Silbergeld EK, et al. The Association of Urine Arsenic with Prevalent and Incident Chronic Kidney Disease: Evidence from the Strong Heart Study. Epidemiol Camb Mass. 2015 Jul;26(4):601–12. pmid:25929811
- 62. Scheer J, Findenig S, Goessler W, Francesconi KA, Howard B, Umans JG, et al. Arsenic species and selected metals in human urine: validation of HPLC/ICPMS and ICPMS procedures for a long-term population-based epidemiological study. Anal Methods. 2012 Feb 2;4(2):406–13. pmid:22685491
- 63. Diane Gilbert-Diamond, Zhigang Li, Perry Ann E., Spencer Steven K., Gandolfi A. Jay, Karagas Margaret R. A Population-based Case–Control Study of Urinary Arsenic Species and Squamous Cell Carcinoma in New Hampshire, USA. Environ Health Perspect. 2013 Oct 1;121(10):1154–60.
- 64.
GTEx Consortium, Laboratory, Data Analysis &Coordinating Center (LDACC)—Analysis Working Group, Statistical Methods groups—Analysis Working Group, Enhancing GTEx (eGTEx) groups, NIH Common Fund, NIH/NCI, et al. Genetic effects on gene expression across human tissues. Nature. 2017 Oct 11;550(7675):204–13.
- 65. Carithers LJ, Ardlie K, Barcus M, Branton PA, Britton A, Buia SA, et al. A Novel Approach to High-Quality Postmortem Tissue Procurement: The GTEx Project. Biopreservation Biobanking. 2015 Oct;13(5):311–9. pmid:26484571
- 66.
GTEx Portal [Internet]. [cited 2021 Apr 28]. Available from: https://www.gtexportal.org/home/.
- 67. Consortium TGte. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science. 2020 Sep 11;369(6509):1318–30. pmid:32913098
- 68. Poplin R, Ruano-Rubio V, DePristo MA, Fennell TJ, Carneiro MO, Auwera GAV der, et al. Scaling accurate genetic variant discovery to tens of thousands of samples. bioRxiv. 2018 Jul 24;201178.
- 69.
Germline short variant discovery (SNPs + Indels) [Internet]. GATK. [cited 2021 May 20]. Available from: https://gatk.broadinstitute.org/hc/en-us/articles/360035535932-Germline-short-variant-discovery-SNPs-Indels-.
- 70. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet. 2007 Sep;81(3):559–75. pmid:17701901
- 71. Das S, Forer L, Schönherr S, Sidore C, Locke AE, Kwong A, et al. Next-generation genotype imputation service and methods. Nat Genet. 2016 Oct;48(10):1284–7. pmid:27571263
- 72. Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet. 2011 Jan 7;88(1):76–82. pmid:21167468
- 73. Yang J, Zaitlen NA, Goddard ME, Visscher PM, Price AL. Advantages and pitfalls in the application of mixed model association methods. Nat Genet. 2014 Feb;46(2):100–6. pmid:24473328
- 74. Morris AP. Transethnic Meta-Analysis of Genomewide Association Studies. Genet Epidemiol. 2011 Dec;35(8):809–22. pmid:22125221
- 75. Li YR, Keating BJ. Trans-ethnic genome-wide association studies: advantages and challenges of mapping in diverse populations. Genome Med [Internet]. 2014 Oct 31 [cited 2021 Jun 7];6. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4254423/. pmid:25473427
- 76. Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, Zahler AM, et al. The human genome browser at UCSC. Genome Res. 2002 Jun;12(6):996–1006. pmid:12045153
- 77. Davis CA, Hitz BC, Sloan CA, Chan ET, Davidson JM, Gabdank I, et al. The Encyclopedia of DNA elements (ENCODE): data portal update. Nucleic Acids Res. 2018 Jan 4;46(Database issue):D794–801. pmid:29126249
- 78. Ongen H, Buil A, Brown AA, Dermitzakis ET, Delaneau O. Fast and efficient QTL mapper for thousands of molecular phenotypes. Bioinformatics. 2016 May 15;32(10):1479–85. pmid:26708335
- 79. Stegle O, Parts L, Piipari M, Winn J, Durbin R. Using probabilistic estimation of expression residuals (PEER) to obtain increased power and interpretability of gene expression analyses. Nat Protoc. 2012 Feb 16;7(3):500–7. pmid:22343431
- 80. Oliva M, Demanelis K, Lu Y, Chernoff M, Jasmine F, Ahsan H, et al. DNA methylation QTL mapping across diverse human tissues provides molecular links between genetic variation and complex traits. Nat Genet. 2022 Dec 12;1–11. pmid:36510025
- 81. Storey JD, Tibshirani R. Statistical significance for genomewide studies. Proc Natl Acad Sci U S A. 2003 Aug 5;100(16):9440–5. pmid:12883005
- 82. Giambartolomei C, Vukcevic D, Schadt EE, Franke L, Hingorani AD, Wallace C, et al. Bayesian Test for Colocalisation between Pairs of Genetic Association Studies Using Summary Statistics. PLOS Genet. 2014 May 15;10(5):e1004383. pmid:24830394
- 83. Wallace C. Statistical Testing of Shared Genetic Control for Potentially Related Traits. Genet Epidemiol. 2013;37(8):802–13. pmid:24227294
- 84. Wallace C, Rotival M, Cooper JD, Rice CM, Yang JHM, McNeill M, et al. Statistical colocalization of monocyte gene expression and genetic risk variants for type 1 diabetes. Hum Mol Genet. 2012 Jun 15;21(12):2815–24. pmid:22403184
- 85. Machiela MJ, Chanock SJ. LDlink: a web-based application for exploring population-specific haplotype structure and linking correlated alleles of possible functional variants. Bioinforma Oxf Engl. 2015 Nov 1;31(21):3555–7. pmid:26139635
- 86. Alexander TA, Machiela MJ. LDpop: an interactive online tool to calculate and visualize geographic LD patterns. BMC Bioinformatics. 2020 Jan 10;21(1):14. pmid:31924160