An atlas of metallome and metabolome interactions and associations with incident diabetes in the Strong Heart Family Study

Abstract

Background:

Chronic exposure to certain metals plays a role in disease development. Integrating untargeted metabolomics with urinary metallome data may contribute to better understanding the pathophysiology of diseases and complex molecular interactions related to environmental metal exposures. To discover novel associations between urinary metal biomarkers and metabolism networks, we conducted an integrative metallome-metabolome analysis using a panel of urinary metals and untargeted blood metabolomic data from the Strong Heart Family Study (SHFS).

Methods:

The SHFS is a prospective family-based cohort study comprised of American Indian men and women recruited in 2001–2003. This nested case-control analysis of 144 participants of which 50 developed incident diabetes at follow up in 2006–2009, included participants with urinary metal and untargeted metabolomic data. Concentrations of 8 creatinine-adjusted urine metals/metalloids [antimony (Sb), cadmium (Cd), lead (Pb), molybdenum (Mo), selenium (Se), tungsten (W), uranium (U) and zinc (Zn)], and 4 arsenic species [inorganic arsenic (iAs), monomethylarsonate (MMA), dimethylarsinate (DMA), and arsenobetaine (AsB)] were measured. Global metabolomics was performed on plasma samples using high-resolution Orbitrap mass spectrometry. We performed an integrative network analysis using xMWAS and a metabolic pathway analysis using Mummichog.

Results:

8,810 metabolic features and 12 metal species were included in the integrative network analysis. Most metal species were associated with distinct subsets of metabolites, forming single-metal-multiple-metabolite clusters (|r|>0.28, p-value<0.001). DMA (clustering with W), iAs (clustering with U), together with Mo and Se showed modest interactions through associations with common metabolites. Pathway enrichment analysis of associated metabolites (|r|>0.17, p-value< 0.1) showed effects in amino acid metabolism (AsB, Sb, Se and U), fatty acid and lipid metabolism (iAs, Mo, W, Sb, Pb, Cd and Zn). In stratified analyses among participants who went on to develop diabetes, iAs and U clustered together through shared metabolites, and both were associated with the phosphatidylinositol phosphate metabolism pathway; metals were also associated with metabolites in energy metabolism (iAs, MMA, DMA, U, W) and xenobiotic degradation and metabolism (DMA, Pb) pathways.

Conclusion:

In this integrative analysis of multiple metals and untargeted metabolomics, results show common associations with fatty acid, energy and amino acid metabolism pathways. Results for individual metabolite associations differed for different metals, indicating that larger populations will be needed to confirm the metal-metal interactions detected here, such as the strong interaction of uranium and inorganic arsenic. Understanding the biochemical networks underlying metabolic homeostasis and their association with exposure to multiple metals may help identify novel biomarkers, pathways of disease, potential signatures of environmental metal exposure.

The general population is exposed to different levels of metals and metalloids through food, water and air. Certain metals and metalloids, including molybdenum (Mo), selenium (Se), and zinc (Zn) are essential to performing normal biological functions; however, at non-homeostatic levels they may exhibit toxic effects.^1–3 Other metals, including arsenic (As), cadmium (Cd), lead (Pb), and uranium (U) are non-essential and chronic exposure to these metals plays a role in the development of cardiovascular disease, kidney disease, neurocognitive outcomes and some cancers.^4–12 While the interaction between two metals can sometimes lead to synergistic or antagonistic effects,^13–15 data elucidating relevant biologic pathways and mechanisms underlying exposure to multiple metals is sparse. Understanding the biochemical networks underlying metabolic homeostasis and their association with varying metal and metalloid exposures may help identify novel biomarkers and pathways of disease.

The metabolome refers to the complete set of small-molecule metabolites, which are the intermediates and products of metabolism.^16,17 Analytical and computational advancements in metabolomics now enables profiling of thousands of low and high abundance chemicals in biological specimens.^18–20 The number of studies that have examined metabolomics as they relate to exposure to a single metal is steadily growing.^21–25 There are currently no untargeted metabolomic studies, however, that focus on multiple metals exposure in adults.

The emerging field of systems medicine and integrative omics uses informatic techniques to connect and integrate different “omics” and other biomedical data in order to better understand the complex development of disease in a system.^26,27 In this work, we use data integration methods to combine a panel of metals and metalloids measured in urinary samples, hereafter referred to as “the metallome”, with the metabolome, a global untargeted metabolome panel measured in serum samples. By integrating urinary metal biomarkers and untargeted metabolomics, and annotating the resulting network structure or modules of co-regulating metabolites, we characterize the association of both known and unknown chemicals with respect to urinary metal levels and potential biological pathways.

The overall goal of this study is to use metabolomics to identify potential metabolomic signatures related to exposure to metalloids and metals and to better elucidate metabolic responses related to the development of diabetes. The underlying hypothesis is that chronic exposure to metals from the environment is associated with perturbations at the level of individual metabolites pathways.

Methods

Study Population.

The Strong Heart Family Study (SHFS) is a longitudinal study of cardiovascular disease, its risk factors, and genetic determinants in 13 American-Indian communities from three geographic regions in Arizona, Oklahoma, and North Dakota and South Dakota. Between 2001–2003, 3,665 participants ranging in age from 14 to 93 years were recruited. The visit included the collection of blood and urine, a physical exam, and an interviewer-administered standardized questionnaire. Visits were performed by trained and certified examiners. Due to a tribal request, data from one of the original tribes was not used.

This study includes 145 SHFS participants from a previously conducted nested case-control study.²⁸ In 2001–2003 participants were free of type-2 diabetes. Incident cases of type-2 diabetes were defined as normal fasting glucose at baseline (2001–2003) and a fasting glucose level of ≥7.0 mmol/L or self-reported physician diagnosis at follow-up (2006–2009). Untargeted metabolomic data and urinary metals were measured using biospecimens collected at baseline (2001–2003).

Urine metals.

As part of the SHFS, concentrations of eight urine metals and metalloids [antimony (Sb), cadmium (Cd), lead (Pb), molybdenum (Mo), selenium (Se), tungsten (W), uranium (U) and zinc (Zn)], and four arsenic species [inorganic arsenic (iAs), monomethylarsonate (MMA), dimethylarsinate (DMA), and arsenobetaine (AsB)] were measured from spot urine samples using liquid chromatography-inductively coupled plasma mass spectrometry (ICPMS) at the Trace Element Laboratory of the Karl-Franzens University, Graz, Austria following a previously published protocol.²⁹ The limits of detection (LOD) were 0.015 μg/L for Cd, 0.1 μg/L for Mo, 0.08 μg/L for Pb, 0.006 μg/L for Sb, 2 μg/L for Se, 0.008 μg/L for U, 0.005 μg/L for W, 10 μg/L for Zn, and 0.1 μg/L for all As species. For all metals, we accounted for urine dilution by normalizing their concentrations to urinary creatinine.

Metabolomics.

Using baseline fasting plasma samples, global, untargeted metabolomic assays were performed using high-resolution Orbitrap mass spectrometry conducted at Emory University in the Clinical Biomarkers Laboratory using previously described methods.^18,20,30,31 Briefly, 65 μL plasma samples were treated with acetonitrile, spiked with internal standard mix, and centrifuged at 13,000 g for 10 min at 4°C to remove proteins. Supernatant (100 μL) was removed and loaded into autosampler vials. Mass spectral data were collected with a 10-min gradient on a Thermo LTQ-Velos Orbitrap mass spectrometer (Thermo Fisher, San Diego, CA) to collect data from mass/charge ratio (m/z) 85–2,000 in a positive ionization mode. Three technical replicates were run for each sample using C18 column. Samples were randomized and analyzed in batches of 20. Pooled and calibrated plasma reference samples were included in each batch for quality control. Peak extraction and data alignment were performed using apLCMS.³¹ Feature and sample quality assessment was performed based on coefficient of variation (CV) and Pearson correlation, respectively, based on the technical replicates using xMSanalyzer.³² Data preprocessing included batch-effect evaluation and correction using principal component analysis (PCA) and ComBat.³³ External investigators interested in accessing the SHS data, including metabolomics data can be made available to qualified investigators after a corresponding paper proposal has been approved by the Strong Heart Study Publication and Presentation committee.

Other variables.

Sociodemographic (age, sex, education, BMI), lifestyle (smoking status) and medical information (estimated glomerular filtration rate (eGFR), fasting glucose and diabetes status) were obtained through an interviewer-administered standardized questionnaire and a physical examination. Exams were performed by trained and certified examiners as previously reported.³⁴ Renal function was assessed using the eGFR calculated by the MDRD equation.³⁵

Bioinformatics and Statistical analysis.

Metabolomics data were pre-filtered to retain only features with non-zero values in >50% in all samples and > 80% in diabetes stratified groups. Urinary metals and metabolite peak intensity were log-transformed and standardized to unit variance and zero mean (z-scores) prior to analysis. To integrate metals and metabolomic data, we used the xMWAS package in R.²⁷ xMWAS uses multivariate statistical methods such as partial least squares regression (PLS) and sparse PLS for data integration and network analysis techniques to identify communities of tightly connected nodes and to evaluate the centrality (importance) of nodes in the integrative networks. Community detection is performed using the multilevel community detection method. Community detection reveals topological modules comprised of functionally related biomolecules.^36,37 The analysis also identifies nodes that undergo network changes, which is determined based on the eigenvector centrality measure,³⁸ which a measure of the influence of a node in a network. A high eigenvector centrality means that a node is connected to many nodes who themselves have high scores. Multivariate methods such as PLS are suited for cases when the variables are highly collinear and the number of variables is much greater than the number of samples (n<<p).³⁹ To characterize the metabolites and metabolic pathways associated with urinary metals, we used xMSannotator³² for metabolite annotation and Mummichog⁴⁰ for pathway analysis. Mummichog maps all possible metabolite matches to the network and looks for local enrichment, which reflects the true activity because the false matches will distribute randomly, determined by a permutation test (shown as pathway P value). The basic assumption is that approximate annotation at individual compound level can inform meaningful predictions at functional levels as defined by metabolite sets or pathways ⁴⁰. For annotation of metabolites, putative annotations were performed by xMSannotator using 10 ppm tolerance from theoretical m/z. Confidence scores for annotation were assigned by xMSannotator derived from a multistage clustering algorithm. Conservatively we only those annotations with medium to high confidence scores. After database matching to HMDB¹⁶, the matched metabolic features were selected based on the consistency of isotopic patterns determined by correlation of intensity across samples and similarity in retention times. Selected features were assigned with confidence scores. Those with high (i.e. 3) confidence score satisfies all criteria of 1) presence of primary adduct; 2) N, O, P, S/C ratio checks; 3) Hydrogen/carbon ratio check and 4) abundance ratio checks for isotopes, multimers, and multiply charged adducts with respect to the singly charged adducts and ions according to heuristic rules. A confidence score of 2 indicates the presence of other metabolites from the same pathway within the same network module (defined by intensity profiles across samples and also have similar retention times), and the presence of primary adduct. A subset of the metabolites was confirmed using authentic standards from our in-house library, noted as “Confirmed”. To account for potential confounding factors, including smoking status, BMI, age, sex, eGFR, and study center/site, we included co-variables with metals as independent predictors of metabolites.

To examine the network assemblage of metals and metabolites we used PLS regression where P < 0.001. For pathway analysis, we used a less stringent p-value threshold of P < 0.1 to identify an appropriate number of metabolites for our pathway analysis. Correlation efficiencies were determined by p-value threshold and available sample sizes. For all pathway analyses, only enriched pathways with ≥4 metabolites were kept for evaluation and interpretation. In sensitivity analyses, we re-ran analyses further adjusting for baseline fasting glucose levels and used hierarchical cluster analysis to assess potential residual confounding by site.

Results

Table 1 presents baseline characteristics (2001–2003) of the study participants for the overall sample and stratified by incident diabetes status at the end of follow-up. The average age of participants was 33 years. 32% of participants were male and 57% of participants had a history of smoking. Notably, the median (interquartile range) BMI for all participants was 32 (26–36), with the majority of participants being overweight or obese. Participants who went on to develop diabetes at follow-up had higher urinary arsenic levels (iAs, MMA and DMA) than participants who did not develop diabetes (Wilcoxon rank sum test: W=1725.5, p-value=0.0069). There were no significant differences in characteristics between participants included in our study and the larger SHFS population (Supplemental Table 1). Our untargeted high-resolution metabolomic analysis detected 9,382 metabolic features and after filtering, 8,810 metabolic features were used for integration.

Table 1.

Participant characteristics at phase IV (2001–2003) stratified by incident diabetes status at phase V (2006–2009)

	OVERALL N=145	INCIDENT DIABETES N=50	NO DIABETES N=95
Age (years)	33 (22, 45)	35 (25, 45)	31 (20, 44)
Male	46 (32)	16 (32)	30 (32)
BMI (kg/m2)	32 (26, 36)	35 (32, 39)	30 (25, 34)
Smoking
Never	62 (43)	19 (38)	43 (45)
Ever	83 (57)	31 (62)	52 (55)
Education
<High School	56 (39)	18 (36)	38 (40)
≥High School	89 (61)	32 (64)	57 (60)
Center
Arizona	23 (15.9)	9 (18.0)	14 (14.7)
Oklahoma	77 (53.1)	16 (32.0)	29 (30.5)
North Dakota and South Dakota	45 (31.0)	25 (50.0)	52 (54.7)
eGFR (ml/min/1.73m2)	99 (85, 111)	96 (80, 110)	97 (87, 100)
iAs (μg/g creat)	0.44 (0.22, 1.00)	0.75 (0.29, 1.26)	0.38 (0.20, 0.91)
MMA (μg/g creat)	0.68 (0.41, 1.33)	1.11 (0.41, 1.55)	0.61 (0.42, 1.08)
DMA (μg/g creat)	4.24 (2.50, 7.72)	6.13 (3.29, 10.57)	3.68 (2.4, 6.63)
AsB (μg/g creat)	0.36 (0.23, 0.98)	0.48 (0.26, 1.26)	0.32 (0.19, 0.89)
Sb (μg/g creat)	0.10 (0.05, 0.14)	0.11 (0.04, 0.16)	0.09 (0.05, 0.14)
Cd (μg/g creat)	0.52 (0.16, 0.92)	0.64 (0.24, 1.24)	0.39 (0.15, 0.82)
Pb (μg/g creat)	0.99 (0.25, 1.92)	0.8 (0.32, 1.96)	1.06 (0.21, 1.78)
Mo (μg/g creat)	32.90 (22.51, 50.96)	35.92 (22.17, 55.53)	32.88 (24.51, 47.59)
Se (μg/g creat)	45.76 (32.60, 59.01)	46.66 (33.36, 61.03)	44.54 (32.28, 56.09)
U (μg/g creat)	0.03 (0.01, 0.05)	0.03 (0.01, 0.06)	0.03 (0.02, 0.05)
W (μg/g creat)	0.08 (0.06, 0.17)	0.1 (0.06, 0.2)	0.08 (0.06, 0.16)
Zn (μg/g creat)	467.46 (330.50, 621.09)	467.46 (326.88, 615.4)	464.83 (331.09, 612.52)

N (%) or median (interquartile range) shown.

Integrative network and pathway analysis in all samples

Figure 1 shows results from the integrative network analysis. When using a threshold of P < 0.001 for the overall model (corresponding |r| > 0.34) and adjusting for sex, smoking, age, BMI, eGFR, education, and study site/center, Cd, Pb and Zn were associated with distinct subsets of metabolites, forming single-metal-multiple-metabolite clusters. Virtually all metabolites in the integrated network analysis were positively associated with metals. Several metals showed interactions through associations with common metabolites. DMA (clustering with W and MMA), iAs (clustering with U), together with Mo and Se formed a large cluster-circle linking W-DMA-MMA-Mo-Se-U-iAs together through shared metabolites. Arsenobetaine and antimony were also associated through a shared and unidentified metabolite (m/z 347.1256, retention time 12 seconds). Smoking status was not associated with any of the metabolites in the model. In sensitivity analyses, further adjusting for baseline serum glucose levels, clusters were consistent (Supplementary Figure 1). Supplementary Table 2 shows pairwise correlations between metals and metabolites and annotated metabolites. In sensitivity analyses, there was no evidence of potential residual confounding by site (Supplementary Figure 2).

Figure 1. Integrated Network of the Metallome and Metabolome at baseline.

Each community is represented by a different color. Network assembled using PLS regression with P < 0.001 (|r| > 0.28).

In our pathway analysis, to determine whether metabolites that are associated with metals are also associated with biological pathways, we used a less stringent P value threshold between metals and metabolites of P < 0.1 (corresponding |r| > 0.14) in order to obtain sufficient number of metabolites for enrichment analysis. For ease of visual inspection, Figure 2 and Supplementary Table 3 shows results of pathway analysis. Metals were primarily associated with lipid regulation (iAs, DMA, Pb, Mo, Zn), including glycosphingolipid and fatty acid metabolism, amino acid metabolism (AsB, DMA), including alanine, aspartate and pyrimidine, and energy metabolism (Mo, Pb, W, Zn), including carnitine shuttle and urea cycle, pathways.

Figure 2. Metabolic pathways associated with urinary metals at baseline.

Metabolic pathways at P < 0.1 (|r| > 0.14). Size of circle mapped to the number of significant metabolites in pathway (≥4), transparency of the circle mapped to the p-value.

Although each metal generally had distinct metabolic responses, three metals were associated with the glycosphingolipid pathway (iAs, Pb, Zn). Two metals were associated with the glycerophospholipid pathway (iAs and Mo), the fatty acid metabolism pathway (DMA, Mo) and the cholecalciferol/vitamin D3 metabolism pathway (U, Se). Inorganic arsenic was associated with the glycosphingolipid and glycerophospholipid pathways. DMA was associated with amino acid (alanine and aspartate), nucleotide sugar, and fatty acid metabolism pathways. Pathway analysis results were consistent when further adjusted for baseline serum glucose levels (Supplementary Figure 2).

Integrative network and pathway analysis in samples stratified by incident diabetes status

Figure 3 shows the integrated network results stratified by participants who went on to develop diabetes at follow-up. Among participants who went on to develop diabetes, there was a cluster-circle linking U-iAs-W-AsB together through a number of shared metabolites, whereas Cd, MMA, DMA, Se, Sb, Pb, and Zn formed single-metal-multiple-metabolite clusters. Among participants with normal fasting glucose levels at follow- up there were fewer metabolites associated with U and iAs at the given threshold compared to the diabetic group (15 compared to 72). Instead, the essential metals Mo and Se were associated with more metabolites (Mo: 43 vs 9; Se: 15 vs 1). Overall, it was evident that metal interactions through associations with common metabolites were stronger for iAs-U in participants who went on to develop diabetes, whereas metal interactions were stronger for essential metals in participants who did not go on to develop diabetes. Comparing centrality measures in the incident diabetic group versus those who did not go on to develop diabetes also showed the greatest change for Mo, U, and iAs, indicating a decrease of importance for Mo and increased importance for U and iAs (Supplementary Table 4). In sensitivity analyses, further adjusting for baseline serum glucose levels, clusters were consistent (Supplementary Figure 2) and there was no evidence of potential residual confounding by site (Supplementary Figures 5 and 6).

Figure 3. Integrated network plots of the baseline metallome and metabolome stratified by incident diabetes at follow-up.

Each community is represented by a different color. Network assembled using PLS regression with P < 0.001.

In pathway analysis stratified by incident diabetes status, among participants who did not develop diabetes at follow-up (Figure 4 and Supplementary Table 7 and 8), both essential and non-essential metals were associated with a number of metabolic pathways, primarily including lipid regulation (iAs, MMA, DMA, Cd, Mo, U, W, Zn), amino acid metabolism (MMA, DMA, Se) and coenzyme/vitamin metabolism (MMA, AsB, Mo, U) pathways. Among participants who went on to develop diabetes (Figure 5), only non-essential metals were associated with metabolic pathways, including lipid regulation (Pb, Sb, U, W), amino acid metabolism (AsB, MMA, DMA, Sb, U), and coenzyme/vitamin metabolism (iAs, MMA, DMA, Cd, W) pathways, energy metabolism (iAs, MMA, DMA, U, W), xenobiotic degradation/metabolism (DMA, Pb) and the phosphatidylinositol phosphate metabolism (iAs, U) pathways. For essential metals, such as Mo and Se, there were generally fewer metabolites associated with essential metals compared to non-essential metals (Figure 3), and essential metals were not enriched in any metabolic pathways (determined by Mummichog permutation test).Results were consistent when further adjusted for baseline serum glucose levels (Supplementary Figures 7 and 8).

Figure 4. Baseline metabolic pathways associated with metals among participants without diabetes at follow-up.

Metabolic pathways at P < 0.1. Size of circle mapped to the number of significant metabolites in pathway (≥4), transparency of the circle mapped to the p-value.

Figure 5. Baseline metabolic pathways associated with metals among participants with diabetes at follow-up.

Metabolic pathways at P < 0.1. Size of circle mapped to the number of significant metabolites in pathway (≥4), transparency of the circle mapped to the p-value.

Discussion

In this study we integrated the metallome with the metabolome to identify potential metabolic signatures (metabolites and metabolic pathways) associated with metals and further associated with diabetes development. In our overall analysis, we found that while most metals clustered in single metal-multiple metabolite clusters, there was evidence of interaction between iAs and U, which tightly clustered together through shared metabolites. Through stratifying by diabetes status five years later, we found that this tight cluster of iAs and U was present in participants who went on to develop diabetes but was not present in participants who continued to have normal fasting glucose levels. The majority of the metabolites associated with the eight metals examined, fell on different metabolic pathways. The pathway analysis more broadly showed metals affect three biological pathways: lipid regulation, amino acid metabolism, and energy metabolism.

In our overall analysis and diabetes stratified analyses, iAs was consistently associated with metabolites on the glycosphingolipid metabolism pathway. Among participants who did not go on to develop diabetes, U formed a single-metal-multiple metabolite cluster and was associated with the phytanic acid peroxisomal oxidation pathway, a lipid metabolism pathway. Among participants who went on to develop diabetes, however, iAs and U were tightly clustered together through shared metabolites. Among participants who went on to develop diabetes, both iAs and U were associated with the phosphatidylinositol phosphate and glycosphingolipid metabolism pathways. In our overall analysis of the combined groups, however, the majority of the metabolites associated with iAs and U fell on different pathways. Although, iAs and U formed a cluster in the overall combined analysis, the eigenvector values for iAs and U did not show strong centrality. The inconsistency between centrality measures and network visualization was likely because; 1) centrality is scaled to 0 to 1 for every node and thus the importance of iAs and U in the whole network was dampened by the overwhelming effect of BMI as a major metabolic cluster in the combine analysis; 2) the eigenvector centrality also considers the quality of connections. The eigenvector score is only high when the node is connected to many nodes (in this case the metabolites) who themselves have high scores, and this is not the situation for the iAs/U in the combined analysis as they only had a few shared metabolites that were connected to another node. Therefore, the combined analysis reflected the dilution of non-diabetic group in which iAs and U did not form a major metabolic cluster and indicated the importance of stratification in analysis by disease status and of validation by future studies with larger populations.

The imbalance of essential metals and exposure to non-toxic metals may lead to disruption of signaling pathways central to normal metabolic homeostasis. Our study showed that comparing participants who went on to develop diabetes to those who stayed non-diabetic, iAs and U were associated with more metabolites which are enriched in pathways of hexose phosphorylation, phosphatidylinositol phosphate (PIP) metabolism pathway and gluconeogenic amino acid metabolism. Phosphatidylinositol-3,4,5-triphosphate (PIP₃) is a prime mediator of insulin resistance.^41,42 While an in vitro study using differentiated adipocytes treated with subtoxic concentrations of iAs(III) showed little to no effect on the activity of phosphatidylinositol 3-kinase, which synthesizes PIP_3,⁴³ experimental studies have yet to examine the effects of the metal-mixture of iAs and U on the PIP pathway. In epidemiological studies, As has been associated with type 2 diabetes, including evidence from the SHFS.^44,45 The association between U exposure and diabetes, however, is less clear.^46,47 Other non-essential toxic metals, like Pb and Cd, might also increase oxidative stress, disrupt glucose uptake and alter glucose regulation,⁴⁸ whereas the imbalance of essential meals, like Zn, might adversely affect storage and secretion of insulin and eventually lead to diabetes development.⁴⁹

Previous studies have reported how exposure to metals, including As, U and Cd, relate to the metabolome; however, these studies focused on exposure to a single metal without accounting for co-exposure to other metals. In a study of adults from Chihuahua, Mexico, among participants with diabetes urinary As was uniquely associated with 42 urinary metabolites and 17 plasma metabolites related to amino acid, energy and nucleic acid metabolism pathways⁵⁰. These findings are consistent with the results from our study which found that among participants who went on to develop diabetes, DMA and MMA were also associated with metabolites in the amino acid and nucleic acid metabolism. In another study of urinary As levels and urinary metabolomics, 31 metabolites were associated with urinary As, 6 with a known identity (1,2-dithiane-4,5-diol, threonine, phosphoric acid, pyroglutamic acid, (R*,S*)-3,4-dihydroxybutanoic acid, and succinic acid).²⁵ Although the study did not conduct a pathway enrichment analysis, the metabolites associated with both water As and urinary As were related to amino acid and uric acid metabolism. Findings are consistent with the results of our analysis which found DMA and AsB were associated with alanine/aspartate metabolism and pyrimidine metabolism pathways, respectively. In a study examining the metabolome of rats who were exposed to U in drinking water, U exposure was associated with metabolites that were mostly related to the tryptophan and vitamin B3 metabolic pathways.⁵¹ Those results differ from our study which found U was related to the vitamin D3 pathway in our overall model and when stratified by incident diabetes status, it was associated with lipid, energy, nucleotide and phosphatidylinositol phosphate pathways. The rodent study, however, measured metabolites in urine samples whereas our study measured metabolites in human plasma samples. To our knowledge, only one other metabolomic study has examined a panel of multiple urinary metals. In that study among 232 pregnant women, exposure to Cd, Co, Cu, Cs, Mn, Th and Vn during pregnancy was associated with several metabolites that affected amino acid metabolism, however, only amino acid metabolites were examined and the mixture of metals was not studied.⁵²

Metabolism is highly complex and involves thousands of different connected reactions. Studying the individual parts of metabolism is difficult because perturbing a single enzyme or a single metabolic pathway may impact the function of a large part of the complete network.⁵³ Prediction of enzymatic activity from correlations in human study is even more challenging given the diversity of genetic background and complex confounding variables, and relies on mechanistic studies using controlled experimental models. However, one benefit of using metabolomics as a holistic approach is to provide comprehensive information on the complex network response to sub-toxic exposures. For example, in our study, we found that hydroxyisocaproic acid (an end product of leucine metabolism) was strongly associated with iAs, W and U levels only in the group of participants who went on to develop diabetes, indicating the role of branched chain amino acids (BCAA) in linking As exposure to diabetes. This correlation was in line with recent finding that branched-chain α-keto acid dehydrogenase (BCKDH), core enzyme complex in BCAA metabolism-controlled response to As toxicity in C. elegans.⁵⁴

The Strong Heart Family Study represents a unique resource for exploring associations between environmental metals exposure and untargeted metabolomics in a well-characterized cohort study that has provided extensive information on the health effects of metals in American Indians living in rural and small-town communities in the US.^13,55–58 One limitation of this study is that urinary Pb and Se may not be the ideal biomarkers of exposure for these metals. Urine biomarkers, however, are commonly used to assess exposure and internal dose to these elements as they integrate multiple exposure sources. The metals in our study have relatively short half-lives in urine, and are generally considered reliable biomarkers of recent exposure. The exception is urinary Cd, which is an accepted biomarker of long-term exposure as it is thought to also reflect accumulation in the kidneys. With chronic exposure, however, urine metal biomarkers with relatively short half-lives can also serve as a proxy of long-term exposure. As expected, participants who went on to develop diabetes had higher As levels compared to participants who did not develop diabetes, however, urinary U levels were similar between both groups, indicating that the metabolic effects of metals in the context of dynamic interaction with other metals are complex. While urinary U levels were not different between diabetes stratified groups, our clustering findings for U and As may suggest that the metabolic effects of U among people who go on to develop diabetes maybe potentiated by co-exposure to other metals, such as As. Stratified network analyses, as in the present study, can reveal associations that depend on interactions with other variables and warrant future validation studies. Another limitation of our study is the relatively small sample size. Our results will require further validation in a larger study population. However, this is the first metallome-metabolome integrative omics study to look at the effects of metal mixtures on metabolic pathways. In this study, we focused on how metals influence different metabolomic pathways, however, some of the metabolomic pathways related to metals may more likely indicate dietary co-intake rather than effect, for example lead and caffeine metabolism and arsenobetaine and folate metabolism. This study presents and atlas of metal exposure on human plasma metabolites and provides important preliminary evidence towards the identification of metabolomic signatures of metal-mixtures.

Conclusions

Understanding the biochemical networks underlying metabolic homeostasis and their association with exposure to multiple metals may help identify novel biomarkers, pathways of disease, and potential signatures of environmental metal exposure. In this integrative analysis of multiple metals and untargeted metabolomics in incident type 2 diabetes, the results show common associations with fatty acid, energy, and amino acid metabolism pathways. Results for individual metabolite associations differed for different metals, however, indicating that larger populations will be needed to establish the metal-metal interactions detected here, such as the strong interaction of U and inorganic As.

Highlights.

• Integrated network analysis of urinary metals and blood metabolomics was performed

• Most metal species formed single-metal multiple-metabolite clusters

• Inorganic arsenic and uranium clustered together through shared metabolites

• Energy and amino acid metabolism related to arsenic in those who developed diabetes