|Home | About | Journals | Submit | Contact Us | Français|
Arsenic is a nonmutagenic human carcinogen that induces tumors through unknown mechanisms. A growing body of evidence suggests that its carcinogenicity results from epigenetic changes, particularly in DNA methylation. Changes in gene methylation status, mediated by arsenic, have been proposed activate oncogene expression or silence tumor suppressor genes, leading to long-term changes in activity of genes controlling cell transformation. Mostly descriptive, and often contradictory, studies have demonstrated that arsenic exposure is associated with both hypo- and hyper-methylation at various genetic loci in vivo or in vitro. This ambiguity has made it difficult to assess whether the changes induced by arsenic are causally involved in the transformation process or are simply a reflection of the altered physiology of rapidly dividing cancer cells. Here, we discuss the evidence supporting changes in DNA methylation as a cause of arsenic carcinogenesis and highlight the strengths and limitations of these studies, as well areas where consistencies and inconsistencies exist.
Arsenic is an element of major health concern because substantial epidemiologic evidence links inorganic arsenic exposure to a variety of human cancers [201,202]. Chronic low-dose dietary arsenic exposure is causally linked to cancers of the skin, bladder, liver and lung [1-4], and human exposures during gestation are associated with elevated incidence of lung and bladder cancer decades later during adulthood [4-6]. It is estimated that over 100 million people worldwide are exposed to carcinogenic levels of arsenic , the vast majority owing to the consumption of drinking water taken from arsenic contaminated aquifers. Populations affected by arsenic- contaminated drinking water span the globe with significant exposures identified in Bangladesh, India, Taiwan, China, Mexico, Argentina, Chile, Europe and regions of North America.
The mechanism by which arsenic mediates carcinogenesis remains a subject of debate, with evidence supporting several plausible etiologies, including disruption of signaling cascades , elevated levels of oxidative stress [8,9], chromosomal aberrations  and epigenetic changes [2,11]. Arsenic does not cause point mutations in standard mutagenicity assays and is generally considered to be nongenotoxic [12,13]. Uncertainty regarding possible mechanisms of arsenic carcinogenicity persists because arsenic has generally been found to be noncarcinogenic in laboratory animals when administered as a single agent [14,15], with the possible exception of transplacental exposures in mice . In rodents, arsenic is neither a tumor initiator nor a tumor promoter, though it does act as a co-carcinogen by synergistically enhancing tumorigenicity of other carcinogens such as UV irradiation . Thus, inorganic arsenic is generally not carcin ogenic in animal models when administered as a single agent and studies in humans are complicated by the fact that arsenic exposures in human populations can occur in association with exposures to other potential carcinogens.
The term epigenetics was coined by Nanney  to describe systems of cellular heredity based on processes other than changes in DNA sequence. Epigenetic mechanisms allow cells to rapidly alter long-term transcriptional activity, thereby permitting coordinated changes in gene expression without permanently altering the sequence of DNA. Chromatin structures limit access of the transcriptional apparatus to genes, thereby interfering with transcriptional activity. Epigenetic restructuring of chromatin can add or remove these obstructive complexes to stably repress or activate genes, respectively. Environmental factors that disrupt epigenetic programming cause aberrant changes in gene expression that predispose to a broad array of diseases, including cancer, hypertension, Type 2 diabetes and congenital malformations .
The epigenetic effects of arsenic have been an area of intense research interest since it was reported by Mass and Wang that arsenic causes hypermethylation of the p53 gene . Since their report, changes in DNA methylation have been the most intensively studied epigenetic phenomena resulting from arsenic exposure, while changes in histone modification and microRNA expression have begun to receive increasing attention.
DNA methylation is a fundamental determinant of chromatin structure. In general, the extent of methylation at CpG islands, DNA sequence clusters rich in CG dinucleotides in the vicinity of gene promoters, correlates with the long-term silencing of gene transcription [21,22]. By repressing gene expression, DNA methylation provides an essential level of control over genes regulating cell differentiation, proliferation and organism development, and these changes can be passed on to daughter cells during replication . Global disruption of DNA methylation is lethal to mammals, and even focal disruption at imprinted loci can produce serious developmental abnormalities. Changes in promoter CpG island methylation can significantly change cell behavior. DNA hypermethylation has been implicated in the loss of tumor suppressor gene expression [24-26]. Conversely, it was noted more than 20 years ago that oncogenic transformation goes hand-in-hand with genome-wide hypomethylation [27,28]. Hypomethylation appears to contribute to malignant transformation by predisposing cells to chromosomal defects and rearrangements leading to genetic instability [29,30] and spontaneous mutations [31,32]. These observations suggest that genome hypomethylation precedes and likely predisposes to transformation. Thus, both DNA hypermethylation and hypomethylation appear to independently contribute to cancer development and progression.
Several mechanisms are theorized to underlie epigenetic changes in DNA methylation by arsenic. Central to each is an effect of arsenic on the activity of DNA methyltransferase (DNMT) enzymes. DNMTs catalyze the transfer of a methyl group from S-adenosylmethionine (SAM) onto the C5′ position of cytosine at CpG dinucleotides to produce 5-methylcytosine . Inhibition of DNMT activity by arsenic appears to cause whole-genome and localized gene specific demethylation. By contrast, reports that arsenic causes localized DNA hypermethylation are more challenging to explain mechanistically.
The cytotoxicity and metabolism of arsenic is a function of its oxidation state and methylation status . In animals, absorbed pentavalent arsenate (As+5) is rapidly reduced to trivalent arsenite (As+3) primarily in the blood and liver, and is subsequently distributed to tissues throughout the body [34,35]. The bulk of circulating arsenic undergoes biotransformation in hepatocytes where arsenite undergoes a series of sequential oxidative-methylation and reduction steps yielding several methylation products (Figure 1) . This methylation of inorganic arsenic to dimethylarsenic acid (DMA) facilitates excretion , but in the process consumes both SAM and glutathione (GSH) .
The reduction of arsenate (As+5) to arsenite (As+3) is a nonspecific reaction carried out by multiple mammalian enzymes , including purine nucleoside phosphorylase , glyceraldehyde-3-phosphate dehydrogenase  and ornithine carbamoyl transferase . These enzymes appear to facilitate arsenate reduction by converting arsenate into arsenylated metabolites (arsenate esters or anhydrides) that are much more readily reduced to arsenite by GSH than is inorganic arsenate . Regardless of the particular enzymatic pathway, GSH is consumed in support of arsenate reduction. Once reduced, arsenite methyltransferase (AS3MT) catalyzes the addition of a methyl group to arsenite in what appears to be two sequential chemical reactions: methylation of trivalent arsenicals using SAM as the methyl donor and oxidation to pentavalent arsenate. AS3MT is capable of using several different reductants to support the methylation of As+3, including dithiothreitol, glutaredoxine and thioredoxin, the latter of which may be the most effective [43,44]. Following oxidative methylation, monomethylar-sonic acid (MMA+5) is again reduced prior to the addition of another methyl group. This reduction is reportedly carried out by GSH S-transferase Ω (GSTO) [45,46] and by AS3MT . In humans, 10–20% of arsenic is excreted as MMA+5 and 60–80% as DMA+5 . Thus, the methylation of inorganic arsenic utilizes multiple equivalents of GSH and SAM per molecule of arsenic.
Consumption of SAM during the metabolism of arsenic has been proposed as an important consequence of exposure. In the presence of excessive arsenic, this theory suggests that SAM, which would otherwise be used to carry out normal cellular methylation reactions including DNA methylation, is utilized by AS3MT to methylate arsenic leading to depletion of SAM and accumulation of S-adenosylhomocysteine (SAH) (Figure 2) [49-51]. Hence, SAM depletion caused by extensive arsenic metabolism may impose cofactor limitations on the activities DNMTs and presumably other cellular methyltransferases including histone methyltransferases. Furthermore, high-level SAH accumulation negatively regulates SAM-dependent methyltransferase activity by feedback inhibition, further contributing to DNMT inhibition [52,53]. It is notable that arsenic is primarily metabolized in the liver, thus, the inhibition of DNMT activity due to AS3MT-mediated arsenic metabolism is limited primarily to the liver, and possibly other AS3MT-expressing cell types. For this reason, the effect of arsenic on cysteine transsulfuration may be more important for most cell types.
Arsenic exposure in the presence of folate nutritional deficiency may facilitate genomic hypomethylation. Since methylation of inorganic arsenic uses SAM as a methyl donor, diets deficient in folic acid, methionine, choline and vita-min B12 can limit the synthesis and reutilization of SAM. SAH is recycled back to SAM in a multi step process that proceeds via the folic acid-dependent enzyme 5-methyltetra hydrofolatehomocysteine (HCY) methyltransferase (MTR) and betaine HCY methyl transferase (BHMT), and the cycle is completed when methionine is conjugated with ATP to regenerate SAM by methionine adenosyl transferase (MAT). MTR-mediated SAM regeneration is the primary pathway for recycling SAM and requires the presence of tetrahydrofolate (from folic acid) and cobalamin (vitamin B12) as enzyme cofactors. These cofactors are obtained from dietary sources and therefore SAM synthesis is susceptible to dietary deficiencies of these essential nutrients as well as methionine [54,55]. Dietary deficiencies that limit the supply of SAM have been linked to epi genetic changes in gene expression and predispose to the transgenerational occurrence of disease .
It has been proposed that changes in cellular redox status can elicit developmental events, such as cell differentiation, by influencing epigenetic programming [57,58]. This theory proposes that GSH synthesis is bio chemically linked to epigenetic processes through the transsulfuration pathway (Figure 2). HCY is a major source of cysteine for GSH biosynthesis. When increased GSH production is required to offset increases in oxidative stress, induction of the enzyme cystathione β-synthase (CBS) shunts HCY from the methionine cycle into the transsulfuration pathway (Figure 2). In response to arsenic, the expression of enzymes responsible for GSH biosynthesis are rapidly induced [59,60] including those of the transsulfuration pathway . Thus, during periods of high oxidative stress HCY is shunted into the transsulfuration pathway, which limits the availability of recycled SAM necessary for chromatin methylation, resulting in genome-wide DNA hypomethylation [52,61,62].
Arsenic is a well-characterized inducer of oxidative stress. Under physiologic conditions arsenite spontaneously reacts with cellular sulfhydryls, including cysteine residues of proteins, nonprotein dithiols and GSH. In the liver arsenic reduction and oxidative-methylation reactions consumes multiple equivalents of GSH, contributing to oxidation of intracellular GSH . Furthermore, arsenic stimulates the generation of reactive oxygen species during its metabolism through the activation of NADH oxidase . Thus, arsenic rapidly depletes intracellular GSH , thereby increasing the oxidation state of exposed cells. Recovery from an arsenic insult is greatly diminished in cells possessing a mutation that impairs GSH synthesis, underscoring the importance of GSH in the protection of arsenic-induced oxidative stress .
To date, a causal relationship between activation of the transsulfuration pathway by arsenic and DNA hypomethylation has not been rigorously tested. However, a report by Coppin et al. suggested that long-term arsenic exposure shunts HCY to the transsulfuration pathway resulting in DNA hypomethylation . In their investigation, the authors treated RPWE-1 cells with 5 μM arsenite for 16 weeks. RPWE-1 cells are a malignantly transformed prostate epithelial cell line with a low capacity to methylate arsenic. Hence, in these cells, significantly reduced SAM depletion is not attributable to AS3MT-mediated oxidative methylation. Throughout the 16-week treatment, arsenite-treated cells displayed an adaptive increase in the expression of enzymes responsible for transsulfuration GSH synthesis. Over the course of the 16 week treatment, an adaptive increase in intracellular GSH concentration (up to fivefold) was accompanied by consistently decreased SAM levels (~20-30%). Concomitantly, DNA methylation was reportedly decreased up to 90% 4 weeks after the initiation of arsenic treatment. It may be important that DNA methylation levels were not reported either at the end of the study nor periodically throughout the 16-week treatment. Surprisingly, the study did not show a reciprocal correlation between SAM levels and GSH synthesis, which would have supported the observed change in DNA methylation status. In addition, the approach used for the DNA demethylation assay did not target specific genes and was not linked to changes in expression of particular genes. Despite these limitations, the evidence from this work suggests that transsulfuration may contribute to arsenic-mediated depletion of SAM and promote DNA hypomethylation. It would be interesting to know if blocking the utilization of HCY for GSH synthesis could block DNA hypomethylation.
In mammals, DNA methylation is catalyzed by three methyltransferase enzymes: DNMT1, DNMT3A and DNMT3B; each with differing activities for maintenance and de novo DNA methylation. DNMT1 is the methyltransferase predominantly responsible for copying methylation patterns during DNA replication and therefore is primarily responsible for propagating epigenetic DNA methylation to daughter cells [65,66]. By contrast, the DNMT3 family appears to establish de novo methylation patterns during embryogenesis. This division of labor is not entirely dichotomous; however, since DNMT1 activity is also required for de novo methylation at non-CpG cytosines  and perhaps to a limited extent within CpG-rich sites [68,69].
Arsenic appears to inhibit DNMT activity by either reducing expression or inactivating the enzyme. Zhao et al. have demonstrated that chronic exposure of cells to low-dose arsenite for 18 weeks inhibited DNA methyltransferase activity in cell lysates by 40%  Interestingly, DNMT1 mRNA expression in the same cells was increased nearly 50%. The loss of DNMT activity in the lysates could result from either a reduction of the protein level or irreversible inactivation of the enzyme. Because the relative amounts of the different DNMT proteins present in the lysates were not evaluated, it remains to be determined whether arsenic-mediated loss of methyltransferase activity reflects changes in DNMT protein levels or enzyme inactivation. However, the loss of activity is not attributable to SAM restriction since enzyme activity was quantified in vitro using standardized concentrations of SAM.
Follow-up work published by the same group using AS3MT-deficent RWPE-1 cells found that, even in the absence of significant arsenic methylation, 30 weeks of continuous arsenic exposure caused genomic hypomethylation . In this study, Benbrahim-Tallaa et al. reported that 7 weeks of arsenic treatment decreased DNMT activity by as much as 60%; this level of activity persisted throughout the remainder of the 30 week treatment . Although the relative amounts of DNMTs present in cell lysates were not evaluated, RT-PCR analysis demonstrated that DNMT mRNA expression levels did not change. This model is of particular interest because the lack of AS3MT in RWPE-1 cells indicates that arsenic methyltransferase-mediated SAM depletion is not responsible for DNMT inhibition, suggesting that arsenic-mediated DNA hypomethylation may be a consequence of a loss of enzymatically active DNMT rather than competition for cofactor(s) with arsenic methylation.
While Zhao et al. demonstrated that DNMT expression was increased by arsenic and Benbrahim-Tallaa reported that the levels remained unchanged; Cui et al. reported that exposure of HepG2 cells to arsenite for 48 h decreased DNMT1 mRNA expression . Importantly, expression changes were not entirely dose dependent; rather, DNMT1 expression was strongly repressed by low concentrations of arsenite (2 and 5 μM) and unaffected by 10 μM, while DNMT enzymatic activity in nuclear extracts was lost across all arsenic concentrations. In our laboratory, we have observed that treatment of human keratinocyte-derived HaCaT cells with arsenic concentrations up to 5 μM for 72 h significantly inhibited the expression of DNMT1 and DNMT3A (DNMT3B was undetected); enzyme activity and protein expression levels were not assayed .
Taken together, these data paint an uncertain picture for the effect of arsenic on DNMT expression. Although it appears that DNMT activity is reduced by arsenic, the mechanism responsible for this inhibition remains to be determined. The influence that cell type and treatment protocol might have on DNMT expression may be a major contributor to the uncertainty. The ability of arsenic to affect methyltransferase activity might extend beyond simple cofactor inhibition or gene repression. Direct inactivation or degradation of arsenic-adducted proteins is a possibility since DNMTs are sulfhydryl-rich proteins; DNMT1, 3a and 3b contain 41, 28 and 25 cysteines, respectively. A systematic investigation of arsenic effects on DNMT activity, expression and stability is necessary to fill these knowledge gaps. Regardless of the outcome, it would still be a challenge to reconcile DNMT inhibition with studies reporting hypermethylation of CpG islands.
If arsenic acts as a sink for GSH and SAM, then sustained arsenic exposure would be expected to deplete SAM to an extent that other methyltransferases, including DNMT, would be inhibited leading to loss of DNA methylation. While plausible, studies claiming to substantiate this model have deficiencies and are contradicted by other studies linking arsenic exposure to DNA hypermethylation [72-76]. Thus, the effect that arsenic has on DNA methylation is questionable since it is difficult to reconcile opposing results found in various reports. The diversity of model systems, treatment protocols and end points used by different research groups to assay DNA methylation contribute to the uncertainty, making it necessary to evaluate these reports in the context of their methodologies (Table 1).
Mass and Wang were the first to propose that arsenic altered the epigenetic status of chromatin. After treating A459 human adenocarcinoma cells with 2 μM arsenite for 2 weeks, these investigators observed that arsenic increased the methylation state of DNA . This finding was based on assays that evaluated both gene-specific methylation and global DNA methylation. These investigators focused on the p53 gene and demonstrated a sevenfold increase in promoter methylation. Likewise, SssI methyl incorporation assays  demonstrated a significant increase in global DNA methylation following arsenic treatment. The authors did not examine whether these epigenetic changes were associated with changes in p53 gene transcription, and did not extend the methylation studies to genes other than p53. The authors speculated that hypermethylation is a result of elevated DNA methyltransferase stimulated by an elevated SAM level that occurs in response to increased demand .
Shortly after Mass and Wang published their findings, the laboratory of Michael Waalkes published the first of a series of reports linking arsenic treatment to loss of DNA methylation. In this first report, TRL 1215 rat liver epithelial cells cultured in up to 0.5 μM arsenite for 18 weeks became tumorigenic when inoculated into nude mice, whereas similarly passaged control cells were nontumorigenic . The critical tumorigenic change apparently occurred at a point between 8 weeks of exposure, when injected cells were incapable of forming tumors, and week 18, when injected cells were tumorigenic. Other markers indicative of cell transformation that could have been used to help delineate the timeframe of transformation (e.g., growth in soft agar) during this intervening 10 week period of culture were not evaluated. Genome methylation studies demonstrated that prolonged arsenic exposure for 12, 18 and 22 weeks caused substantial DNA demethylation; however, the status of DNA methylation at the apparently nontumorigenic 8 week time point was not reported. Hence, it is not clear in this model if methylation changes are early events predisposing to subsequent transformation or if hypomethylation temporally coincides with transformation.
Both hypomethylation and hypermethylation have been associated with cell differentiation and cancer cell transformation . When these changes occur concurrently, as they appear to do following chronic arsenic exposure, methylation changes at sites regulating expression of individual genes may be more relevant to carcinogenesis than the global trend in DNA methylation. The localized nature of these changes strongly implies that these changes are mediated by mechanisms that are targeted in nature and not the result of nonspecific mechanisms. It has been argued that arsenic-mediated global DNA hypomethylation underlies increased oncogene expression and contributes to malignant transformation. At the same time, several studies have demonstrated hypermethylation of tumor suppressor genes. The possibility that both events occur simultaneously was demonstrated by treating TRL 1215 cells with arsenic for 18 weeks. Tumorigenicity of these cells, when injected into nude mice, was correlated with DNA hypomethylation and elevated c-Myc expression . These transformed cells also had increased expression of cell-cycle genes, including PCNA and cyclin D1 [78-80]. Interestingly, transformation was also associated with repression of genes known to be silenced by promoter CpG methylation, including p21, metallothionein-1 and several inflammatory and tumor suppressor genes. Treating these arsenic-transformed cells with the DNA methyltransferase inhibitor 5-aza-deoxycytidine for 72 h restored p21 and metallothionein-1 expression to levels at or above those in nontransformed control cells , suggesting that during arsenic-mediated malignant transformation promoter hypermethylation occurs in the presence of widespread global genome hypomethylation. To date, this possibility has not been directly investigated, though in the context of cell differentiation and carcinogenesis these findings are not surprising.
The liver is clearly a target organ for arsenic toxicity in humans and is an important site for the metabolism of dietary arsenic. To investigate the relationship between global DNA methylation and gene-specific changes during arsenic-mediated transformation, Chen et al. administered drinking water containing moderately-high arsenic concentrations (45 ppm) to adult 129/SvJ mice for 48 weeks. At the end of the treatment, global hepatic DNA methylation was decreased as determined by both HPLC measurement of 5-methyldeoxycytosine content and the SssI methyl incorporation assay . Gene-expression analysis revealed deregulation of a number of genes, including elevated expression of cyclin D1 and estrogen receptor (ER)-α. Both genes were evaluated for changes in promoter CpG methylation by sequencing bisulfite-modified DNA. Of 13 CpG sites in the ER-α promoter, eight demonstrated a significant decrease in methylation status following chronic arsenite treatment. In comparison, the cyclin D1 promoter was unmethylated and remained so following arsenic treatment. Thus, it does not appear that increases in cyclin D1 expression related to arsenic exposure depend on promoter methylation; however, cyclin D1 effects attributed to arsenic may have a basis in the increased expression of ER-α due to epigenetic reprogramming .
A major difficulty in the analysis of the epigenetic outcome of arsenic exposure is that the effects appear to be influenced by the particular model system used. This difficulty can be illustrated by the studies of Cui et al. where the same methods and genes were used to evaluate DNA methylation changes at different arsenic concentrations and in different model systems [71,74]. In one investigation, human-derived HepG2 and Huh-7 liver cancers cell lines were exposed to 2 and 10 μM arsenite for 72 h. In these cells, 2 μM arsenic caused a loss of DNA methylation within the promoters of p16INK4a (CDKN2A), RASSF1A, E-cadherin (CDH1) and GSTP1, and the extent of demethylation was similar to that obtained by treating cells with 5-azadeoxycytidine . Similar results were found in T24 (human bladder carcinoma), HCT116 (colon carcinoma) and HL-60 (human promyelocytic leukemia) cell lines, suggesting that these changes are not unique to a particular cell line. By contrast, 10 μM arsenic had little effect on promoter methylation in these cells, a result that was attributed to cytotoxicity. In a separate report, the same authors administered drinking water containing 100 ppm arsenate to A/J mice for 18 months. Treatment doubled the occurrence of poorly differentiated lung adenocarcinomas from 0.6 to 1.2 tumors/mouse. By contrast to the hypomethylation observed in HepG2 and Huh-7 cells, tumor-containing lung tissue from arsenic-treated mice had dose-dependent hypermethylation of the p16INK4a and RASSF1A promoters . The extent of DNA methylation of normal lung tissue in arsenic-treated animals did not change relative to unexposed controls.
Arsenic is not generally considered to be a complete carcinogen [15,83]; rather, it is a cocarcinogen requiring an accompanying second agent to induce tumors in adult animals [84,85]. For this reason, most carcinogenesis studies in rodent species have employed some initiating event. Xie et al. evaluated the capacity of arsenic to achieve epigenetic changes as a cocarcinogen in Tg.AC mice in conjunction with phorbol 12-myristate 13-acetate (TPA) skin-painting . Tg.AC mice are unique because they harbor multiple copies of a transgene consisting of the v-Ha-ras structural gene fused to a ζ-globin hemoglobin promoter. This mouse line is a validated dermal model for testing skin carcinogenesis in as little as 6 months, compared with the traditional 2-year National Toxicology Program bioassay . Xie administered organic and inorganic arsenicals to Tg.AC mice in drinking water for 17 weeks. A total of 4 weeks after the initiation of arsenic treatment, the tumor promoter TPA was applied to the skin for a 2 week period. At the end of the 17 week study period, global DNA hypomethylation was associated with all forms of arsenic tested (As+3, As+5, MMA and DMA). Interestingly, despite the fact that arsenic is a well documented skin carcinogen and TPA is an inducer of skin papillomas in Tg.AC mice, the authors did not report any dermal response in conjunction with TPA treatments. Presumably, the 17 week study design may have been too short to permit papilloma detection [88,89]. Microarray-based whole-liver transcriptome analysis of treated mice revealed that each form of arsenic produced dissimilar but overlapping patterns of gene expression. Unfortunately, the authors did not investigate whether the observed differences in gene expression were associated with differences in promoter-region DNA methylation.
Based on observations that gene imprinting is critical for fetal development, it has been proposed that gestational disruption of normal epigenetic programming by arsenic predisposes to congenital defects and cancer later in life. Epidemiologic evidence links in utero and early-life human exposures to arsenic with increased risk of cancer mortality during adulthood [4,6]. Likewise, gestational arsenic exposure has been associated with developmental and long-term disease consequences, including pulmonary and cardiovascular diseases, fetal loss and birth defects [4-6,90-92]. Since arsenic readily crosses the placenta and concentrates in the fetus [93-95], maternal arsenic exposure might disrupt epigenetic gene imprinting during susceptible periods of fetal development. Theoretically, this disruption of normal epigenetic programming might lead to aberrant changes in gene expression, which in turn influences fetal development and long-term risk of disease [56,96].
Chemical exposures during sensitive periods of gestation can increase the risk of tumors in the offspring. In mice, gestational exposure to arsenic reportedly increases the incidence of malignant tumors of the liver and lungs in mature offspring [16,97,98]. This transplacental carcinogenesis was investigated by continuously exposing pregnant C3H mice to drinking water containing 0, 42.5 or 85 ppm arsenite during gestational days 8 through to 18. The incidence is strongly affected by gender such that male offspring had a greater incidence of hepatocellular carcinoma at 74 weeks after birth, whereas female offspring had an increased occurrence of pulmonary carcinoma at 90 weeks after birth. Although a promising model for arsenic tumorigenesis, a recent report from a collaborating laboratory using the same mouse strain and a similar exposure regimen reported very different outcomes . This subsequent study found no increase in liver tumors in either male or female mice 52 weeks after in utero exposure. More difficult to reconcile with the original study is the observation that exposure of offspring to drinking water containing 85 ppm of arsenic in addition to the in utero exposure actually protected against liver tumors. Since the 52-week study design of latter work was significantly shorter than the original study, the possibility remains that if the animals had been maintained for the full treatment periods the outcomes may have been more similar.
Based on initial observations that gestational exposure of pregnant C3H mice induced tumors in adult offspring, follow-up studies have attempted to investigate associated epigenetic changes. The long-term effect of gestational arsenic exposure on DNA methylation was assessed in livers of male mice obtained 36 weeks after birth . As with previous studies published by this research team, dams were exposed to 85 ppm arsenic in drinking water during gestational days 8 through to 18. Global and gene-specific DNA methylation changes were determined in normal-appearing liver tissue obtained from 5 control mice and 5 arsenic-exposed mice with hepatocellular carcinoma. In utero arsenic exposure was not associated with any change in global DNA methylation or with changes in cyclin D1 promoter methylation. By contrast, the promoter region of the ER-α demonstrated significantly reduced methylation at 12 of 13 CpG sites. These changes were consistent with those reported by Chen et al. in adult 129/SvJ mice exposed to drinking water containing 45 ppm of arsenic for 48 weeks  with the exception that Chen et al. reported that arsenic produced global DNA demethylation and marked liver histological changes, including steatosis and hypertrophy, but without detectable tumors. It is possible that the differences in the extent of global demethylation as well as differences in carcinogenesis may be attributed to the timing of exposure (in utero exposure vs exposure in mature animals).
A subsequent report addressed the more immediate effect of gestational exposure by assessing DNA methylation status in newborn pups. Following a 10 day gestational exposure to drinking water containing 85 ppm arsenic, livers were obtained from newborn pups shortly after birth . As with adult mice, global methylation status did not predict localized CpG methylation status: there was no change in global DNA methylation, whereas randomly targeted CpG-rich regions became demethylated. These data, and those of the previous study  suggest thatin utero arsenic exposure causes targeted gene-specific changes rather than global genomic demethylation.
As with the preceding animal studies, attempts to investigate the epigenetic effects of arsenic in humans also have had limitations. Subject recruitment, exposure assessment, tissue sample procurement and normalization for confounding factors are just a few of the complications unique to human studies. Evaluated individually, the existing reports of epigenetic changes mediated by arsenic in exposed human populations highlight the need for further investigation.
Chanda et al. investigated the association between DNA methylation changes and arsenic exposure in human cancers . Peripheral whole-blood samples were collected from 158 residents of West Bengal, India, who were chronically exposed to arsenic-contaminated drinking water. Subjects were stratified on the basis of estimated drinking water exposure over the course of at least 6 months. The methylation status of two tumor suppressor genes was investigated using a study design having two phases. The first phase demonstrated that chronic arsenic exposure was associated with the hypermethylation of the p53 promoter in 96 subjects. The second phase demonstrated hypermethylation of the p16INK4a promoter in a different set of 62 subjects. These results appear to be in line with the cell culture studies demonstrating hypermethylation of p53 , and hypermethylation of p16INK4a in arsenic-induced lung adenomas ; however, they contrast with the finding that arsenic induced hypomethylation of p16INK4a and RASSF1A in human liver tumor cells resulting in increased expression of these genes . Recently, Majumdar et al. published a follow-up investigation to the 2006 Chanda study . This latter study stratified subjects on the basis of estimated drinking water exposure and determined DNA methylation status in peripheral blood cells. The results were partially consistent with the gene-specific methylation changes of the original study in that the second highest exposure level was associated with global hypermethylation, but the highest levels of arsenic exposure demonstrated a trend towards hypomethylation. To explain this discrepancy, the authors proposed that lowdose arsenic might induce DNMT3A leading to subsequent DNA hypermethylation, whereas the high-dose exposures cause SAM depletion and DNMT inhibition. Since DNMT activity and expression were not investigated, the extent to which changes in DNMT activity actually contribute to changes in DNA methylation remains speculative.
Several limitations complicate the interpretation of these studies. The minimum cut-off for subject inclusion was a 6-month exposure to contaminated drinking water, but epigenetic changes were not stratified on the basis of exposure time. Since the time-frame necessary to achieve measurable epigenetic changes in blood is unknown, short exposure periods might have falsely limited detection of early changes, leading to an underestimation of epigenetic changes. Furthermore, DNA methylation of the p53 and p16INK4a promoters was assayed in different sets of subjects using different techniques. Thus, it cannot be concluded that hypermethylation of one gene correlates with hypermethylation of the other gene in the same individuals. Identification of signature genes undergoing congruent methylation and demethylation in the same cells and tissues should be a major focus of future investigations.
Marsit et al. examined 351 bladder tumors from arsenic exposed residents of New Hampshire, USA . In this study, subjects were designated as either arsenic exposed or control on the basis of toenail arsenic measurements, and methylation levels were determined for promoter regions of p16INK4a, RASSF1A and PRSS3. The study found no change at the p16INK4a promoter but observed significant increase in methylation at the RASSF1A and PRSS3 promoters. These results are partially consistent with those of Cui et al. mentioned earlier, who found dose-dependent hypermethylation of RASSF1A and p16INK4a in lung adenomas of A/J mice . As with A/J mouse lung tumors, RASSF1A DNA methylation in bladder tumors increased with higher tumor grade.
Diets deficient in folate, methionine, vitamin B12 and choline restrict the availability of substrates necessary for SAM synthesis and may inhibit DNA methylation [54,55] contributing to oncogene activation [102,103]. Since arsenic methylation consumes both SAM and GSH, arsenic exposure in the presence of dietary methyl deficiencies may exacerbate DNA hypomethylation. To test this hypothesis, Okoji et al. fed adult C57BL/6J mice a diet deficient in methionine, choline and folate. At the end of a 21-day pretreatment period, four dosage groups were administered arsenic-containing drinking water for an additional 130 days. Necropsied livers demonstrated a dose-dependent global DNA hypomethylation. Gene-localized DNA methylation was quantified at two sites upstream of the Ha-ras gene: one was a low CpG-density region (segment 474/976) and the other a high CpG-density region (segment 962/1487). In the low CpG region, arsenite treatment plus methyl deficiency decreased the level of methylation at five of 11 restriction sites cleaved by methylation-sensitive restriction enzymes (EcoRII, StuI, AluI, AvaII and XhoI) and increase methylation at one site (HhaI). Notably, although the activities of these six restriction enzymes are sensitive to 5-methyl cytosine, only two are sensitive to cytosine methylation at CpG dinucleotides (XhoI and HhaI), whereas the other four are inhibited by methylation of cytosines outside the CpG dinucleotide (EcoRII, AluI, StuI and AvaII). The majority of methylation changes were detected by enzymes sensitive to methylation of non-CpG dinucleotides within the low density CpG region upstream of the Ha-ras transcription start site. Thus, the majority of methylation changes occurred on non-CpG cytosines. By contrast, high-density CpG island regions upstream of Ha-ras transcription start site remained unmethylated regardless of treatment conditions . Since the authors did not associate the effects of arsenic or methyl deficiency with gene expression, it is not possible to determine whether these changes had a measurable physiological effect on Ha-ras expression. Other studies have noted that there is not a clear-cut correlation between Ha-ras promoter methylation and mRNA expression . In parts of the world where environmental exposure to arsenic is endemic, poor folate nutritional status may promote arsenicotic skin lesions (melanosis, leukomelanosis and keratosis) and cancer through the disruption of normal DNA methylation. To test this hypothesis, Pilsner et al. recruited 294 Bangladeshi men and women and stratified them on the basis of dietary folate status and arsenic exposure. Surprisingly, the results demonstrated that arsenic had no effect on global DNA methylation in the low-folate group but it had a dose-dependent correlation with DNA hypermethylation in the highfolate group . A follow-up study compared arsenic-exposed individuals with and without skin lesions, matched for arsenic exposure, sex and age. Consistent with the previous study, arsenic exposure was associated with a global increase in DNA methylation in peripheral blood samples. Stratification by nutritional folate status revealed that DNA hypermethylation correlated with arsenic exposure only in subjects with adequate folate status, but this association was only applicable to control cases, or rather, subjects without skin lesions. Hence, the authors concluded that arsenic exposure was positively associated with genomic methylation only if folate status is adequate . To explain these counter intuitive findings, the authors proposed a rationale similar to that of Mass and Wang, discussed earlier, wherein adequate folate is permissive for an adaptive increase in genomic methylation, possibly through increased DNMT activity.
Despite an accumulating number of investigations indicating that arsenic affects the epigenetic status of cells and tissues, there is surprisingly little consensus regarding what exactly are the changes that occur. Both increased and decreased DNA methylation has repeatedly been reported at both the genome-wide and gene-specific levels (Table 1). The reasons for this diversity of findings may be attributable in some cases to technical experimental details since arsenic has different physiologic effects depending on dose and duration of exposure. However, it appears that assays of genome-wide methylation status generally demonstrate hypomethylation more often than not (with the exception of in utero exposures), while assays of individual genes can demonstrate either hypermethylation or hypomethylation, depending on the tissue, cell type and gene examined. Future investigations capitalizing on genomic technologies should provide a powerful tool for obtaining a more comprehensive picture of arsenic epigenetic effects. The use of high-throughput deep sequencing and microarray technologies should be useful for identifying signature genes that undergo hypermethylation and hypomethylation as a result of arsenic exposure.
In general, it is difficult to reconcile the incongruent findings reported by the diverse array of model systems discussed here. Even among models sharing similar treatments techniques, findings tend to be divergent and difficult to interpret. It is clear that a simple mechanistic model, such as repression of DNMT activity by SAM depletion, is overly simplistic [49,48,71]. Rather, DNA methylation patterns suggest that arsenic may cause both hyper- and hypomethylation simultaneously, in all likelihood on a gene-by-gene basis with an overall trend towards global demethylation. Although it may seem counterintuitive that hypermethylation can occur at one gene while hypomethylation occurs at another, it is likely that this is in fact the case. The fact that investigators have chosen to study a wide array of different genes as an index of DNA methylation status makes it difficult to compare the effects of arsenic amongst the various model systems. In this regard, it is a rare occurrence when different groups used the same genes to index DNA methylation status. In addition, most authors typically use a single technique to evaluate methylation changes, which contributes to the confusing array of conclusions, since different methods of determining DNA methylation analysis provide different types of information with different biases and artifacts . Awareness of these limitations and the application of appropriate methods for the desired research aim may help to resolve some of these issues. In addition, by capitalizing on current genomic technologies, future investigations will be able to correlate DNA and transcriptional changes so that model systems can be compared globally with sequence-level resolution.
In this review we have focused on arsenic effects on DNA methyltransferases; however, chromatin structure also depends on a complex network of histone modifying enzymes, including histone methyltransferases. If cofactor restriction is indeed a primary mechanism mediating DNMT inhibition, then it stands to reason that a multitude of other methyltransferases may also be inhibited; presumably this would include histone methyltransferases as well. Inhibition of histone methyltransferase by SAM depletion could profoundly affect both gene activation and repression. For example, demethylation of trimethylated lysine-27 in histone H3 (H3K27me3) coincides with a generalized increase in gene activation [109,110] and loss of DNA methylation . Likewise, loss of H3K4me3 is associated with gene repression and DNA imprinting. Thus, a more nuanced view of the role that SAM depletion may play may be necessary to explain the apparently incongruous findings reported to date. Since histones are proteins that constantly turn over, changes in histone methylation patterns will prove to be the next important target of arsenic intoxication .
Recent technological advancements including methylation-sensitive deep sequencing and microarrays now provide useful tools for characterizing epigenetic changes globally with sequence-level sensitivity. In parallel, application of similar techniques to the study of gene expression provides a comprehensive snapshot of gene-expression changes. Together, using a variety of bioinformatic approaches, chromatin methylation changes can be integrated with gene-expression changes to map the inter-relationships between chromatin methylation and gene-expression patterns in arsenic target tissues. Over the next few years next-generation sequencing techniques should help to address questions surrounding genome-wide versus regional changes in DNA methylation. Furthermore, other epigenetic mechanisms influencing gene expression, such as histone modification and microRNA, have scarcely been investigated.
More daunting are the questions surrounding the model system in which to investigate the epigenetic effects of arsenic. Investigators attempting to address these concerns are faced with dilemmas that accompany their choice of exposure assessment, sample procurement and end points to measure. From an environmental standpoint, a major question is how best to model relevant human exposures. In the laboratory, the applicability of animal model systems to human exposures remains uncertain. Although the next several years might bring some consensus to the field, it is currently difficult to perceive how any current model best reflects disease occurrence associated with human arsenic exposure. Certainly, however, if it is established that epigenetic changes consistently observed in an animal model also occurring in target tissues of humans, this would represent a major step forward.
Research in the authors' laboratory is supported by NIEHS grants R01 ES06273, R01 ES10807 and the NIEHS Center for Environmental Genetics grant P30 ES06096.
Financial & competing interests disclosure
The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.
No writing assistance was utilized in the production of this manuscript.
Papers of special note have been highlighted as:
■ of interest
■■ of considerable interest