Arsenic Toxicology: Translating between Experimental Models and Human Pathology

Background: Chronic arsenic exposure is a worldwide health problem. How arsenic exposure promotes a variety of diseases is poorly understood, and specific relationships between experimental and human exposures are not established. We propose phenotypic anchoring as a means to unify experimental observations and disease outcomes. Objectives: We examined the use of phenotypic anchors to translate experimental data to human pathology and investigated research needs for which phenotypic anchors need to be developed. Methods: During a workshop, we discussed experimental systems investigating arsenic dose/exposure and phenotypic expression relationships and human disease responses to chronic arsenic exposure and identified knowledge gaps. In a literature review, we identified areas where data exist to support phenotypic anchoring of experimental results to pathologies from specific human exposures. Discussion: Disease outcome is likely dependent on cell-type–specific responses and interaction with individual genetics, other toxicants, and infectious agents. Potential phenotypic anchors include target tissue dosimetry, gene expression and epigenetic profiles, and tissue biomarkers. Conclusions: Translation to human populations requires more extensive profiling of human samples along with high-quality dosimetry. Anchoring results by gene expression and epigenetic profiling has great promise for data unification. Genetic predisposition of individuals affects disease outcome. Interactions with infectious agents, particularly viruses, may explain some species-specific differences between human pathologies and experimental animal pathologies. Invertebrate systems amenable to genetic manipulation offer potential for elaborating impacts of specific biochemical pathways. Anchoring experimental results to specific human exposures will accelerate understanding  of mechanisms of arsenic-induced human disease.

volume 119 | number 10 | October 2011 • Environmental Health Perspectives Review Arsenic exposure via drinking water affects > 140 million people worldwide and causes cancer and broncho pulmonary, cardio vascular, and metabolic diseases and neu ropa thies. Various experimental models have been developed to understand how arsenic exposure causes these diverse disease out comes. Translation of laboratory arsenic toxi cology studies to human health is important but is complicated by inexact dose conver sion between in vitro, murine, and human exposures and speciesspecific metabolic differences. Here, we discuss issues in dose conversion and potential means to translate findings in selected experimental model sys tems to an understanding of human arsenic toxicology. Pheno typic anchoring of results from model systems by tissue dosimetry, gene expression and epigenetic mark profiling, and tissue biomarker identification should promote develop ment of a coherent picture of mechanisms of arsenicinduced human disease. We discuss research needs critical to progress in translation of experimental find ings. We also highlight a humanspecific disease end point and discuss advantages of invertebrate systems to address specific ques tions in a simpler background with fewer confounding factors.

Dose and Exposure Conversion
Data collected in human studies often include exposures but not doses. Urine and toenail arsenic are often used as indicators of body burden but are subject to wide individual varia tion with similar exposures. Dose conver sion between human and murine exposures is a complicated issue. Calculating dose requires careful determination of amounts consumed and is rarely reported. Often, consumption estimates are based on data from published studies. However, water consumption can vary greatly in mice and is markedly different in different strains (Bachmanov et al. 2002). Likewise, human exposure data include an estimate of arseniccontaminated water and/or food consumption. However, body weights are not systematically collected and differ greatly with study population. Hence, calcula tion of human dose with individual precision has not been done. Even with reliable dose estimates, dose conversion between the mouse and human is complicated. An estimate based on body surface area may be reliable for many substances (ReaganShaw et al. 2008), but arsenic metabolism is strikingly different in rodents and humans (Vahter 1999). For these reasons, anchoring results by induced pheno type may be a more useful approach. A simple anchor might be target tissue arsenic levels. Murine tissue dosimetry can be performed readily, although most data currently available are from mice with high arsenic exposures (Devesa et al. 2006;Gentry et al. 2005). Some human data on tissue, blood, and urine arse nic levels have been correlated with exposures in specific popu lations. Thus, this approach is limited in that data available are on a popula tion level, but there are no sys tematic compi lations of these correlations on an individual level. Hence, no direct connection between a specific human exposure and a biological arse nic level is available, and research including these measures is needed. Other approaches to determine exposure equivalence by induced pheno type include anchoring by changes in gene expression, epigenetic marks, or tis sue remodeling biomarker profiles. These approaches are certainly possible within labo ratory models and could readily serve to unify results from experimental systems. However, only very limited data sets are available for human exposures. Thus, there is a great need for research collecting these data from humans who exhibit arsenicinduced disease. These data are critical to translation of experimental results to specific human exposures.
Transplacental exposures. In utero exposures to environmental toxicants can have a pro found effect on development of chronic adult diseases. Endocrine disruptors are para digms of develop mental toxicants and are linked to diseases as diverse as prostate cancer (Ho et al. 2006) and obesity (Grun and Blumberg 2009). Consequences of in utero arsenic expo sure in humans are difficult to determine in most cases because exposure is not limited to the in utero period but continues into post natal life. However, a unique situation with a defined period (1958)(1959)(1960)(1961)(1962)(1963)(1964)(1965)(1966)(1967)(1968)(1969)(1970)(1971) of arsenic expo sure occurred in Antofagasta, Chile (Borgono et al. 1977). This unfortunate incident pro vides a cohort with a defined period of expo sure. Increased incidence of a variety of disease conditions associated with the arsenic expo sure was reported shortly after the switch back to lowarsenic water (Borgono et al. 1977). These conditions included increased incidence of broncho pulmonary and cardio vascular dis eases, both now clearly linked to chronic arse nic exposure (Argos et al. 2010). Longterm followup studies of this cohort revealed high mortality from lung cancer and bronchiectasis in the population exposed in utero and during early childhood decades after high exposure ended (Smith et al. 2006). Additionally, the incidence of myo cardial infarction in infants whose mothers were exposed during this period (Rosenberg 1974) indicates that in utero arsenic exposure could induce cardio vascular disease.
In contrast to the striking results from the Antofagasta population, infant mortality-but not spontaneous abortion-showed dose cor relation in a Bangladeshi population (Rahman et al. 2010). This difference may be due to a difference in exposure levels. The water arsenic level in Antofagasta during the high exposure period was approximately 800 μg/L and uni form in the popu la tion because there was a single source of water, whereas the Bangladeshi population experienced variable exposures due to multiple sources: 268-2,019 μg/L (median, 390 μg/L) for infant mortality and 249-1,253 μg/L (median, 382 μg/L) for spon taneous abortion study populations. Taken as a whole, in utero exposure to high levels of arsenic in drinking water appears to be neces sary for obvious adverse effects early in post natal life. It is likely that lower exposures have a more subtle effect, perhaps contributing to chronic adult diseases.
High arsenic exposure in utero affects gene expression in leukocytes from human cord blood (Fry et al. 2007). Gene ontology analysis of altered mRNA expression in arsenicexposed samples revealed that immune, inflammatory, and stress response categories were affected. Network analyses identified JUNB, interleukin (IL) 8, IL1β, and hypoxiainducible factor1α, which are involved in cell cycle regulation, stress response, inflammation, and response to hypoxia, respectively. In addition, nuclear factorκB was integrated into the sub networks and also found to be activated in the cord blood of arsenicexposed infants.
Animal studies indicate that in utero arsenic exposures induce both cancer and athero sclerosis. In utero arsenic exposure (42.5 or 85 ppm) induced tumors in C3H mice (Waalkes et al. 2003) and established the first reproducible laboratory animal model of carcino genesis by inorganic arsenic (iAs) alone. Recent work from this group shows that wholelife exposure at lower levels (6-24 ppm) results in higher tumor incidence (Tokar et al. 2011). Combined with in vitro studies showing enhanced proliferation of stem cells, these results led to the hypothe sis that cancer induced by in utero arsenic exposure is a consequence of arsenicinduced increase in the stem cell popu la tion of target tissues (Tokar et al. 2010).
Both in utero (Srivastava et al. 2007) and post weaning ) expo sures to arsenic in drinking water accelerate and exacerbate athero genesis in the apo lipo protein Eknockout (ApoE -/-) mouse model for athero genesis. These studies showed that athero genesis was induced by arsenic expo sure alone, without the highfat diet normally used to induce athero sclerosis in this model. The in utero arsenic exposure (49 ppm) used in the ApoE -/experiments produces arsenic levels in livers of the pregnant dams (States JC, unpublished data) similar to those observed in livers of people exposed to high levels of arsenic (200-600 ppb) in drinking water in West Bengal (Guha Mazumder 2001). Data from gene expression analyses show induction of immune, inflammatory, and stress response pathways in livers of 10weekold ApoE -/mice exposed to arsenic in utero (States JC, unpub lished data). These pathways were among the top pathways activated in human cord blood lymphocytes discussed above. Hence, the data suggest that these responses are induced in multiple tissues and may be a common basis from which disease processes emerge. Thus, correlation exists in pheno typic anchors (tissue arsenic levels, altered gene expression) between the higher arsenic exposures in Chile and South Asia and these mouse exposures.
Arsenic-induced tissue remodeling. Adverse health effects of chronic arsenic ingestion on the lung include chronic obstructive pulmo nary disease, chronic bronchitis, and bron chiectasis. In separate studies in West Bengal and Bangladesh, chronic arsenic exposure reduced lung function (De et al. 2004;Parvez et al. 2008;von Ehrenstein et al. 2005) and increased respiratory disease symptoms (i.e., cough, chest sounds, shortness of breath) and chronic bronchitis Milton and Rahman 2002). More than 63% of subjects with mean arsenic exposure of 216 ± 211 ppb (compared with 11 ± 20 ppb in controls) displayed increased respiratory complications (Islam et al. 2007). Clearly, highlevel arsenic exposure (200-1,000 ppb) causes adverse respiratory effects. However, effects of lower exposures are not known.
Airway remodeling is a hallmark of many respiratory diseases, including emphysema, asthma, idiopathic pulmonary fibrosis, and bronchiectasis (Jeffery 2001;Niimi et al. 2005;Reynolds et al. 2005). Persistent struc tural changes in tissue develop through a pro cess of injury and dys regulated repair, leading to chronic inflammation and altered extra cellular matrix deposition in the airway wall, eventually obstructing airflow. Chronic lung disease pheno types in populations with high arsenic exposure suggest that extra cellular matrix, aberrant cell motility, and wound repair are arsenic targets. Data support this hypothesis because changes in expression and organization of extra cellular matrix genes and in expression of mediators and enzymes that control matrix remodeling have been observed consistently in a wide range of model systems.
Expression of a large number of extra cellular matrix genes was altered in adult male C57Bl/6 mice exposed to either 10 or 50 ppb arsenic in their drinking water for up to 8 weeks (Lantz and Hays 2006). These altera tions included suppression of several collagen, elastin, and fibronectin isoforms. In addi tion, mRNA for matrix metalloproteinase9 (MMP9), a matrix degradation enzyme, was induced. Disorganization and expansion of elastin and collagen after 8week 50 ppb arse nic exposure were observed around pulmonary airways and blood vessels. Arsenicinduced changes in adult animals also occurred in the extra vascular matrix of small cardiac arteries (Hays et al. 2008).
Matrix is also critical for cell migration, wound repair, and remodeling after injury. Pathway analysis using gene and protein expression data from multiple model systems suggests that wound repair and cell motil ity are two of the more probable processes affected by arsenic exposure (Lantz and Hays 2006;Lantz et al. 2007Lantz et al. , 2009Petrick et al. 2009). Arsenic increased time to close a scratch wound in confluent human airway epithelial cells. This increased closure time (reduced wound repair) was associated with increased expression and activity of MMP9. Arsenic, even in the absence of the wounding, induced significant production of MMP9, and inhibition of MMP9 partially restored repair. Inhibition of repair also occurs in an animal model. Animals exposed to arsenic had less capacity to repair naphthaleneinduced airway injury (Lantz RC, unpublished data).
volume 119 | number 10 | October 2011 • Environmental Health Perspectives During fetal and early postnatal lung develop ment, extra cellular matrix gene expres sion is necessary for proper development of lung and blood vessels. In the highly exposed Antofagasta population, in utero and early postnatal exposure (~ 800 ppb) increased risk of chronic obstructive pulmonary dis ease and bronchiectasis (Smith et al. 2006). After in utero and early post natal exposure in mice (≤ 100 ppb), lung collagen type 1 α2 (Col1a2), Col3a1, and elastin mRNA expres sion increased and exhibited both develop mental time and exposure dependence ). Changes in matrix protein expres sion may result from arsenic inter action with normal develop mental processes. However, wholelung collagen and elastin levels were not significantly altered. Increased mRNA expres sion could be a compensatory response. For example, arsenicinduced increases in MMP9 during early post natal periods, as seen in a mouse model, would degrade matrix, requir ing increased mRNA expression to maintain appropriate protein levels.
Although wholelung levels of matrix pro teins were unchanged, regional decreases in total collagen in adventitia around airways were seen in 28dayold mice exposed to arse nic during development ). Localized decreases in collagen were associ ated with increased levels of smooth muscle around airways and alterations in pulmonary function. Understanding mechanisms for localized arsenic effects requires research.
Of critical importance is whether changes seen in model systems replicate events in human populations. Levels of MMP9 and its inhibitor, TIMP1 (tissue inhibitor of metallo proteinase1), determined in popula tions with low exposures to arsenic (< 20 ppb) through drinking water showed that the MMP9:TIMP1 ratio in induced sputum was positively associated with total urinary arsenic (Josyula et al. 2006). Although the under lying mecha nism is different from that in model systems (increased ratios were due predomi nantly to TIMP1 decreases in humans), the under lying effect-increased degradation of matrix-is the same.
Thus, ingested arsenic alters matrix and matrixassociated proteins in a number of model systems and in humans. Evaluation of arsenicinduced pheno typic alterations, includ ing lung function, lower respiratory infections (predicted from model systems), and changes in mediators affecting matrix deposition, is needed, especially in children. Changes in matrix deposition may be a source of useful tissue biomarkers for pheno typic anchoring.
Arsenic-induced vascular disease in adult animal models. Arsenic exposure is strongly associated with increased cardio vascular disease risk ). High expo sures cause occlusive arteriosclerosis, such as blackfoot disease seen in Taiwan (Tseng 2008) and coronary occlusion in infants in Chile (Rosenberg 1974). Many studies have found increased cardio vascular disease risk with more modest exposures (10-100 ppb). In the United States, mortalities from vas cular diseases were increased in counties where arsenic levels were > 20 ppb relative to those with < 10 ppb (Engel and Smith 1994). Diseases associated with these lower exposures include coronary artery and isch emic heart disease, carotid athero sclerosis, micro circulatory defects, and prolonged QT intervals (Medrano et al. 2010;Tseng 2008;Wang et al. 2009). Arsenic may increase associated vascular disease risk factors, such as systolic hypertension Tseng 2008) and diabetes (NavasAcien et al. 2009). Increased systolic hypertension ) is consistent with direct stimu latory effects of arsenic on vascular smooth muscle (Soucy et al. 2004) and decreased vaso relaxation (Srivastava et al. 2007). Nutritional ), metabolic NavasAcien et al. 2005), and genetic susceptibilities ) to cardio vascular pathologies caused by arsenic implicate enhanced oxidant signaling as a primary mode of action. This appears as endothelial cell dysfunction and metabolic dysregulation from loss of nitric oxide or gain of oxidant signaling ).
Mice may be as sensitive as or more sen sitive than humans to vascular patholo gies caused by low to moderate arsenic exposures. Angiogenesis, tumor angio genesis, and liver sinusoidal vessel remodeling occur in C57BL/6 mice exposed for 2-5 weeks to 1-10 ppb arsenic (Soucy et al. 2003(Soucy et al. , 2005Straub et al. 2008). Mouse models reproduce the athero genic effects of arsenic after in utero (Srivastava et al. 2007) or adult arsenic expo sures (Bunderson et al. 2004;Srivastava et al. 2009). In the mouse heart, arsenic caused peri vascular fibrosis (Hays et al. 2008) and increased expression of matrix remodeling pro teins (e.g., Serpine1 and MMP9) (Soucy et al. 2005). At higher exposures, progressive loss of myo cardial micro vessels (Soucy et al. 2005) and cardio myopathy (Li et al. 2002) occurred. In the developing chicken heart, arsenic affects epithelial to mesenchymal transitions necessary for valves to develop (Lencinas et al. 2010). Arsenic causes liver steatosis, fibrosis, and por tal hypertension in humans (Mazumder 2005) that may predispose individuals to risk of sys temic athero sclerosis and metabolic disease (Targher et al. 2010). Arsenic causes mouse liver sinusoidal endothelial cell (LSEC) capillar ization and peri portal vessel hyperplasia (Straub et al. 2007(Straub et al. , 2008 that resemble similar pathol ogy seen in infants who died from in utero or peri natal arsenic exposures (Rosenberg 1974). As in humans, studies in rabbit models (Pi et al. 2003) and mouse models (Bunderson et al. 2004;Straub et al. 2008) implicated nitric oxide loss and increased oxidant signaling in promoting endothelial cell dysfunction and pathogenic pheno typic change. Thus, animal models recapitulate pathogenic end points that are relevant to arsenicinduced human cardio vascular diseases, and these end points provide pheno typic anchors for systematic investigation of pathogenic mechanisms.
Pheno typic anchors of vessel remodeling and vessel celltomatrix interactions involved in remodeling reveal critical signaling path ways under lying the etiology of arsenicrelated vascular diseases. Matrix inter actions are critical for maintaining vessel integrity, wall cell pheno type, and functional signaling. In a model of epithelial to mesenchymal tran sition in heart valve development, transcrip tomic analysis revealed 382 genes that were responsive to 25 ppb arsenic (Lencinas et al. 2010). Pathway analysis identified clusters of responsive genes involved in cyto skeletal regulation, matrix deposition, and cell adhe sion, as well as in stabilizing an endothelial cell pheno type (Lencinas et al. 2010). The clus ter of cyto skeletalregulating genes included GTPases (Rac1 and similar members of the RhoA GTPase family) known to be activated by arsenic in vascular dys function (Qian et al. 2005;Smith et al. 2001;Straub et al. 2007Straub et al. , 2008 and inflammation (Lemarie et al. 2008). In vivo, arsenic exposure results in membrane localization of Rac1 in capillarized LSEC (Straub et al. 2007(Straub et al. , 2008. In ex vivo studies, arsenicinduced LSEC capillarization was pre vented by inhibiting Rac1 activity (Straub et al. 2007(Straub et al. , 2008. Rac1 also is highly expressed in skin tumors induced by arsenic plus phorbol ester in Tg.AC mice (Waalkes et al. 2008).
The Rac1 signaling program mediates arsenicinduced generation of reactive oxygen species that are second messengers for its patho genic effects. Rac1 is an essential subunit of Nox2type NAPDH oxidase, and this oxidase is required for arsenicstimulated largevessel endothelial and LSEC oxidant production (Smith et al. 2001;Straub et al. 2008). Arsenic does not capillarize LSEC in mice lacking this oxidase (Straub et al. 2008). This finding was the first in vivo demon stra tion of a role for NADPH oxidase in arsenic action and the first demon stration that the activation of the oxi dase promotes LSEC capillarization. In a recent study Ghatak et al. (2010) confirmed that NADPH oxidase activity is central to arsenic induced liver fibrosis. Further, chronic activa tion of Rac1 and Nox2type NADPH oxidases are longitudinal risk factors for vascular disease and hypertension (Lee and Griendling 2008). Gainoffunction poly morphisms in oxidase subunit genes are associated with cardio vascular disease in general (San et al. 2008) and with arsenicinduced disease ).
There is a significant knowledge gap in understanding how pheno typic change in indi vidual cell types relates to pathogenic vascu lar remodeling and function. Arsenicinduced LSEC capillarization limits the removal of lipoproteins, lipids, and waste proteins from the circulation and alters normal liver lipid metabo lism (Straub et al. 2008). In addition, zonal distribution of hepatocyte lipid depo sition changes from being exclusively within hepato cytes surrounding the central veins (zone 3) to spreading into hepato cytes sur rounding the portal veins (zone 1). These effects may translate to both liver and sys temic vascular diseases (Targher et al. 2010). Arsenicinduced change in liver cell pheno type and under lying cell matrix appears to alter basic liver structure, function, and metabo lism. However, full investigation of the LSEC responses is hindered by the over whelming mass of hepato cytes masking these responses. Preliminary evaluation of total mouse liver mRNA, microRNA, and proteome responses to lower level arsenic exposures revealed mod est changes (Straub et al. 2009). This modest effect is expected because there is little observ able arsenicinduced change in the hepatocytes. Examination of primary LSEC exposed to arsenic ex vivo, however, demonstrated much greater responses that supported the patho genic in vivo effects (e.g., decreased expression of the scavenger receptor stabilin2). The chal lenges are to determine whether LSECspecific or vascularcell-specific changes can provide markers for arsenicinduced patho genesis and whether preventing arsenic effects in LSECs or vascular cells prevents systemic pathogen esis. Similarly, there is a need to understand how arsenicinduced change in micro vascular pheno type affects organ function, such as in the liver, or systemic metabolic changes that promote cardio vascular and metabolic diseases.
Epigenetic effects of arsenic exposure. The disruption of normal epi genetic control can participate in the etiology of complex human diseases, including psychiatric disorders, cardio vascular disease, diabetes, and cancer. In cancer, pathologic disruption of the nor mal epi genetic state of a cell can be caused by diverse mediators and mechanisms, includ ing environ mental agents, stresses, and cues. Accumulating evidence indicates that arsenic is an environmental toxicant that can mediate epi genetic changes (Reichard et al. 2007;Ren et al. 2011b). Thus, epigenetic control mecha nisms are a nexus of gene-environment inter actions that link cellular responses to arsenic exposure. DNA and histone modification enzymes and the cellular pathways that input signals to them represent potential targets for disruption leading to an altered epi genetic state and phenotype.
Recent work links arsenic exposure to epigenetic state disruption and progression of the diseased state. During carcinogenesis, arsenic exposure induces global DNA hypo methyla tion with hypomethylation frequently found in repetitive elements, although DNA demethyla tion of some gene regula tory regions also occurs (Chen et al. 2004;Jensen et al. 2009a;Reichard et al. 2007). The functional consequences of this DNA hypo methyla tion remain unclear but may involve inappropriate gene activation or altered chro matin structures. Because arsenicals inhibit activity of DNA methyltransferases DNMT1 and DNMT3a (Reichard et al. 2007), this effect may contribute to overall decreased lev els of DNA methylation. However, it may be only one of multiple factors contribut ing to arsenicalinduced epigenetic change, because arsenicals also mediate a coincident DNA hypermethylation of CpG island gene promoters, as well as changes in histone post translational modifications.
Aberrant DNA hypermethylation of CpG island gene promoters is functionally linked to inappropriate transcriptional silencing, and disease progression. This epigenetic lesion has been found in multiple human cell models of arsenicalinduced malignant transforma tion (Cui et al. 2006;Jensen et al. 2009a). In one example, both arsenite and mono methyl arsonous acid (MMA III ) induced malignant transformation of an immortalized uro thelial cell line model of human bladder cancer (UROtsa) (Bredfeldt et al. 2006;Sens et al. 2004). In this model, arsenite and MMA III each induced hundreds of DNA methylation changes across the genome, with a striking overlap in genes targeted by these similar but chemically distinct arsenicals. These results suggest that different forms of arsenic may act similarly in their ability to perturb the epi genetic landscape. For example, in the UROtsa model, both MMA III and arsenite induced DNA hypermethylationassociated gene silenc ing of DBC1 (deleted in bladder cancer 1) and G0S2 (G0/G1 switch regulatory protein 2) (Jensen et al. 2008(Jensen et al. , 2009a. Interestingly, both of these genes display tumor suppressor function and become aberrantly methylated and transcriptionally silenced in clinical blad der cancer (Chang et al. 2010;Habuchi et al. 1998;Hoque et al. 2006;Izumi et al. 2005;Kusakabe et al. 2010;Welch et al. 2009), suggesting that in vitro models of arsenical induced malignant transformation may accu rately reflect epi genetic events that occur in clinical disease. The human relevance of these in vitro studies is further suggested by recent human populationbased studies that found a connection between arsenic exposure and epi genetic dysfunction in bladder cancer (Marsit et al. 2006).
Many of the arsenicmediated epi genetic genesilencing events linked to gene promoter DNA hypermethylation were also accompanied by changes in the histone code in these same regions, specifically hypo acetylation of histones H3 and H4 (e.g., Jensen et al. 2009a). The temporal order of and mechanisms involved in this multi faceted epi genetic reprogramming are not clear. The epi genetic state change may result from a new epigenetic program being enacted by arsenicaldriven alterations in cell signaling inputs. Alternatively, arsenicals may act on multiple epigenetic modifier enzymes to shortcircuit the epigenetic program. Indeed, changes in both histone phosphorylation and histone methylation that appear independent of DNA methylation changes occur after arsen ical exposure (Jensen et al. 2009b;Zhou et al. 2008). Taken together, these results indicate that arsenicals likely disrupt multiple epigenetic pathways.
Epidemiological studies of Chilean popu lations show an arsenicrelated increase in lung and bladder cancer mortality, as well as a long latency between the time of major arsenic exposure and increased disease rates (e.g., Marshall et al. 2007). The long latency suggests that arsenicals may damage the epig enomic integrity of progenitor or stem cell populations and that the expanded popula tions arising from these progenitors retain the epigenetic changes. This type of epigenetic initiation event is consistent with the first step in the recently proposed epigenetic progeni tor theory of carcino genesis (Feinberg et al. 2006). Specifically, we predict that arsenicals induce changes in the epigenetic terrain of progenitor cells that are faithfully inherited from cell generations, even after removal of the initiating toxicant. Thus, arsenicals may act as epi mutagens-agents capable of altering the epi genome of cell populations, resulting in changes in gene expression and pheno typic shifts. This longterm epigenetic damage may remain silent until other critical events occur (e.g., loss of p53, immortalization), at which time the arsenicalinduced epigenetic changes may be pheno typically "unmasked" and help drive evolution of the malignant phenotype (e.g., suppression of tumor suppressor genes). The precise mechanisms responsible for arse nic's disruption of a cell's epigenetic state are being elucidated and will be critical for a full understanding of arsenical action. Research profiling epigenetic changes in human tissues is needed to validate the epigenetic changes observed in vitro.
Cutaneous effects of arsenic and human papilloma viruses. In humans, skin is the most sensitive target organ for chronic arsenic expo sure (Yoshida et al. 2004). Even at lowlevel exposures, arsenic increases risks for pigmen tation changes (melanosis), hyper keratosis, Bowen's disease, and non melanoma skin can cer (NMSC) (Agency for Toxic Substances and Disease Registry 2007; Chen et al. 2009). Although chronic arsenic exposure is causally volume 119 | number 10 | October 2011 • Environmental Health Perspectives linked with skin disease, cutaneous arsenico sis is solely a human disorder for reasons that remain unknown (Rossman et al. 2002). However, humanspecific hyper keratosis may be linked to enhanced viral infection and immune suppression observed in laboratory studies.
One possible explanation for the human specificity of the effect of arsenic on skin is an inter action with a viral skin pathogen. Arsenic exposures inhibit immune function, at least in part by inhibiting immune surveillance of dendritic cells and CD4 cell activation (Lantz et al. 1994;Liao et al. 2009). By compromising immune function, arsenic impairs the immune response to viruses. This effect has been demon strated for influenza A, for which arse nic exposure elevates viral titers and increases morbidity (Kozul et al. 2009;Yu et al. 2006). Similarly, it has been known for more than a century that arsenic exposure can reactivate latent herpes infections (Au and Kwong 2005;Lanska 2004). Likewise, human papillomavirus (HPV), a humanspecific pathogen, shares sev eral clinical features with arsenicosis and may contribute to arsenical skin disease. Cutaneous HPV establishes infection by evading detec tion by skin dendritic cells (Langerhans cells). Therefore, it is reasonable that immune inhibi tion by arsenic could unmask pre existing infec tions or impair the immunologic response to new exposures (Frazer et al. 1999).
Most individuals are exposed to dermal HPV during their lifetimes (Pfister 2003). In fact, many individuals have anti bodies against HPV, thereby demonstrating prior exposure (Masini et al. 2003). Such exposures may be of little consequence for individuals with nor mal immune function; however, individu als with impaired immune function are at significantly increased risk of HPV infection and NMSC. Patients with epidermo dysplasia verruci formis have an immune defect that pre vents recognition of HPV, resulting in severe skin infection and a 90% increase in NMSC risk (Pfister 2003). Likewise, immuno suppressive therapy increases the risk of skin warts and pre malignant actinic keratoses 2fold and risk of squamous cell carcinoma 150fold (Shamanin et al. 1996;Stockfleth et al. 2004). Thus, arsenicinduced immune suppression may increase HPV infectivity.
Only a handful of studies have investigated the occurrence of HPV infection in dermal arsenicosis. Ninety NMSC patients recruited from an arsenicendemic region of Mexico were evaluated for the serological presence of HPV16-reactive anti bodies (RosalesCastillo et al. 2004). The odds ratios for NMSC in patients with a positive history for high arse nic exposure or the presence of antibodies against HPV were 4.53 and 9.04, respectively. This risk increased to 16.5 when both high level arsenic exposure and HPV were present (RosalesCastillo et al. 2004). Although it has not been systematically investigated, several case studies have directly detected HPV infec tion in arsenical skin lesions. HPV types 16 and 41 have been detected in squamous cell carcinomas taken from arsenicexposed patients (Grimmel et al. 1988;Neumann et al. 1987), and HPV23 was identified in multiple hyper keratotic papules from a single patient (Gerdsen et al. 2000). Somewhat in contrast with these findings, Ratnam et al. (1992) detected HPV in only 2 of 33 arsenical keratoses isolated from four patients. The dif ferences among these findings are not surpris ing given the small study size, the > 100 types of HPV, and the technical challenge associated with broadly detecting cutaneous HPV types (Dang et al. 2006;Vasiljevic et al. 2007).
In addition to arsenic's effect on immune function, arsenic may promote integration of HPV DNA into the genome of keratino cytes, the process underlying HPVmediated neoplasia (Jones and Wells 2006). Damage to episomal HPV DNA, such as that caused by oxidative stress, is a critical step triggering genomic integration of the virus and expres sion of genes that promote keratino cyte prolif eration and inhibit differentiation (Jones and Wells 2006). By promoting integration, arse nic may enhance the tumori genicity of HPV (Germolec et al. 1997;Milner 1969;Rossman 1998). Together, HPV and lowconcentration arsenic may target epidermal stem cells to pro mote keratinocyte proliferation and inhibit normal differentiation (Egawa 2003;Liu et al. 2010). Although the effect of arsenic on HPVinfected cells is unknown, preliminary data suggest that arsenic increases cell division and delays differentiation of HPVinfected keratino cytes in organo typic skin cultures, leading to delayed differentiation, increased suprabasal cell division, and suprabasal skin thickening (Reichard JF, unpublished data). Clearly, more research on arsenic enhance ment of viral infections in both animals and humans is needed.
Metabolism, genetics, and model systems. Human dose dependence for any arseniclinked pheno typic outcome depends on multiple criti cal factors, such as intra cellular chemical trans formation, tissue distribution, reactivity, and efflux (Thomas 2007). Each can be affected by individual genetic variability, so depar tures from the "norm" in dose responsiveness and outcome often occur. Use of genetically manipulable models can undoubtedly enhance our understanding of these processes and their importance to toxicity mechanisms.
The methylated derivatives mono methyl arsonic acid (MMA V ) and dimethyl arsinic acid (DMA V ) were believed to be detoxi fied metabolites (Vahter 1999). However, detection of the methylated As III (+3 oxida tion state) species in urine (Le et al. 2000) altered perceptions because MMA III is signifi cantly more toxic than either iAs or the other metabo lites (Petrick et al. 2000;Styblo et al. 2000). Some 10-20% of urinary metabolite in humans is MMA [much higher than for most mammals (Vahter 2002)]; the expecta tion that a portion of this is MMA III might account for higher human susceptibility to pathologic outcomes compared with rodents. Studies of arsenicexposed populations link urinary MMA levels and individual suscepti bility to a range of arsenicrelated pathologies (Smith and Steinmaus 2009). The genetic con tribution to this association is important, with data suggesting that several pathways might contribute to differential MMA levels (e.g., uptake, onecarbon metabo lism, speciation, efflux). The Sadenosylmethionine-dependent enzyme arsenic (+3 oxidation state) methyl transferase (AS3MT) is capable of transform ing iAs to produce MMA and DMA species of both +3 and +5 oxidation states (Li et al. 2005;Thomas et al. 2004). Certain intronic and extra genic AS3MT polymorphisms (along with more extended local haplo types) are associ ated with higher DMA:MMA ratios (Gomez Rubio et al. 2010;Schlawicke et al. 2009), whereas the exon 9 poly morphism M287T is associated with higher urinary levels of MMA (Hernandez and Marcos 2008). Recently, this M287T allele was associated with both elevated damage to DNA (SampayoReyes et al. 2010) and enhanced premalignant skin lesions (Valenzuela et al. 2009), suggesting a mechanistic connection to higher MMA levels. More detailed study of the catalytic proper ties of AS3MT alleles and their response to input from other intersecting pathways (e.g., onecarbon metabolism, redox environment, feedback inhibition) is required.
In the larger context, more insightful stud ies into the mechanisms and consequences of arsenic uptake, speciation, distribution, reten tion, and efflux in vivo are necessary. Reports on metabolitespecific transport into and out of cells (Drobna et al. 2010;Liu et al. 2006), as well as mouse studies on organspecific distribution, retention, and excretion of spe cific metabolites (Kenyon et al. 2008), have appeared. MMA species can accumulate in cells (perhaps owing to their reactivity), whereas DMA is readily exported. More genetically amenable models are now available for study. Arsenitefed AS3MTknockout mice pro duced low levels of methylated metabolites but accumulated high levels of iAs (up to 20fold higher than wildtype mice) in various tissues Hughes et al. 2010), sup porting methylation as a key pathway for arse nic elimination. Such iAs accumulation led to early death (Yokohira et al. 2010).
Organisms such as Drosophila and yeast are simpler eukaryotes that have genetic advantages and few confounders. These organisms provide experimentally accessible models capable of rapidly generating fresh insight and testable hypotheses. Thus, Drosophila lacks a homolog of AS3MT, but introducing the human AS3MT gene allows both MMA and DMA species to be produced. Arsenitefed trans genic flies show important dosedependent differential effects of these species in vivo com pared with the wildtype, with significantly impaired chromosomal stability at 9 ppm but enhanced viability at an acute exposure of > 60 ppm owing to reduced arsenic accu mulation (Muñiz Ortiz et al. 2011). The data integrate the idea that methylated arsenicals are more damaging to macro molecules yet are more readily eliminated and that iAs dose makes all the difference to pheno typic conse quence. Importantly, the quantitative conse quences of other human AS3MT alleles (e.g., M287T) can be tested readily in this system. The availability of a transcriptomewide RNA interferencebased gene knockdown system in Drosophila should provide novel screens that identify pathways intersected by such metabo lites. Complementary approaches already initi ated in yeast using a gene deletion library have identified novel pathways pertaining to arsenite methylation and histone H4 methylation that are relevant in human cells (Jo et al. 2009;Ren et al. 2011a).

Disease Outcome Dependence on Interaction with Genetics and Other Environmental Factors
Humans exposed to arsenic do not all suc cumb to a single disease. Some develop cancer, whereas others develop cardio vascular disease or neuropathies. The reason for the different responses to similar exposures is unclear. A hint is apparent in the differential response of dif ferent strains of mice to similar in utero arsenic exposures. C3H, CD, or Tg.AC mice develop earlier and more severe cancer (Tokar et al. 2011;Waalkes et al. 2003Waalkes et al. , 2008, whereas ApoE -/mice develop earlier and more severe athero sclerosis (Srivastava et al. 2007). These responses are clearly linked to the dis ease predisposition of the mice, and this dispo sition appears to be aggravated by the arsenic exposure. Thus, arsenic inter action with the genetic background of the organism deter mines the disease outcome in these models. In humans, disease outcome also is likely dependent on inter action with other exposures in addition to individual genetic predisposi tion. Immunosuppression by arsenic exposure may increase susceptibility to infectious agents (Kozul et al. 2009). Thus, increased sensitivity to viral infections could increase onco genesis if the individual is exposed to onco genic viruses such as HPV. Chronic arsenic exposure causes hyper reactivity to lipopoly saccharide (Arteel et al. 2008), suggesting that aggravated inflammatory responses to bacterial infections or even to non pathogenic exposures could be aggravated. Hence, arsenic exposure may prime the system for exaggerated response to a second hit that could be a biological or physical agent, diet, or altered metabolism encoded by individual genetics. Thus, more studies both of human genetics and disease outcome and of structure/function relation ships of polymorphic genes involved in arsenic metabolism are needed.

Conclusions
Chronic arsenic exposure, mostly via contami nated drinking water, causes a multitude of diseases. It is unclear what governs the specific pathology induced in any given individual. However, genetic susceptibility to a particular disease and interaction with other environmen tal factors play major roles in determining dis ease outcome of arsenic exposure. Anchoring of experimental models for arsenic toxicology to specific human exposures is essential to gain ing a mechanistic understanding of how arsenic exposure leads to specific human pathologies. Global gene expression profiling, epigenetic mapping, and markers of tissue remodeling offer promise as pheno typic anchors. Full development of anchors requires extensive research to profile gene expression, to map epi genetic marks, and to identify biomarkers in target, or surrogate, tissues in arsenicexposed populations. Human research that includes dosimetry would have greatest impact.