Critical Aspects of EPA's IRIS Assessment of Inorganic Arsenic: Interim Report (2013)

4

Hazard Identification

The major goals of the Environmental Protection Agency (EPA) toxicologic review of inorganic arsenic are identification of the disease hazards associated with chronic exposure to arsenic and estimation of the potential disease risks when people are exposed for a portion of their lifetime. For hazard identification, EPA has outlined a process that involves approaches to screen and organize studies, a causality framework for evaluating cancer and noncancer effects, and an evidence framework for determining susceptible life stages and populations. Those approaches have been designed to incorporate recommendations made in other National Research Council reports that are applicable to all Integrated Risk Information System (IRIS) assessments; they are under review by another Research Council committee.

The present committee has focused on EPA’s plans for hazard identification that are specific to inorganic arsenic. EPA has indicated that its assessment of causality will be based primarily on studies in humans under the assumption that any health effects observed in such studies are relevant to humans, regardless of country of origin and regardless of whether the mode of action is understood. Animal and mechanistic data will provide supporting evidence with respect to biologic plausibility. If particular effects are observed only in animal studies, mechanistic data will be used to address questions about the relevance of the data to humans. If mode-of-action data are insufficient for determining relevance, EPA will assume that the effects are relevant to humans.

The sections that follow contain a preliminary survey of the scientific literature on what appear to be the most affected organ systems and focus on the epidemiologic evidence. The strengths and weaknesses of epidemiologic studies of cancer and noncancer effects were discussed in the committee’s workshop by Cantor (2013) and Steinmaus (2013), respectively. The committee also considered mode-of-action information that would inform dose–response analyses, particularly with respect to low to moderate exposure (see Box 5). Consideration was also given to factors that could increase susceptibility to the effects of inorganic arsenic. In approaching its task, the committee did not attempt to conduct systematic reviews of the literature for each organ system. That task lies ahead for EPA. Rather, the committee has attempted to focus on factors that will help to tailor such reviews and to inform decisions about performing dose–response analyses.

The committee notes that among the noncancer effects of inorganic arsenic exposure, particular attention should be paid to the category of diseases that occur with a high prevalence in the US population so that it can be determined whether exposure to inorganic arsenic may be contributing to the underlying burden of disease in the population. That category includes cardiovascular disease, respiratory disease, kidney disease, and diabetes. Adverse effects of early-life exposure should also be considered.

SKIN DISEASES

Arsenic-related skin conditions include two related conditions: preneoplastic or nonneoplastic skin lesions (henceforth referred to as skin lesions) and nonmelanoma skin cancers. The types of skin conditions and their risks are determined by dose, duration, and susceptibility to exposure. The epidemiologic features of the two conditions, especially as related to those determining factors, are described below.

Skin Lesions

Arsenical skin lesions predispose a person to some skin cancers and serve as an indicator of sufficient arsenic exposure for (or increased susceptibility to) the occurrence of other cancer and noncancer diseases (Ghosh et al. 2007). Owing to differences in arsenic metabolism and toxicity among species, observational epidemiologic studies are preferred for determining the association between arsenic exposure and skin lesions.

Skin lesions have a well-established dose–response relationship with arsenic in drinking water. ATSDR (2007) based its chronic minimal risk level of 0.0003 mg/kg-day on skin lesions. Skin lesions were reported to occur at concentrations as low as 10 μg/L in cross-sectional (Ahsan et al. 2006) and prospective cohort (Argos et al. 2011) studies. The association was confirmed at concentrations of 50–100 μg/L (Guha Mazumder 2003; Guo et al. 2006; Argos et al. 2011). However, other studies have found skin lesions to occur only at high concentrations (e.g., Haque et al. 2003).

In a longitudinal study in Bangladesh, the odds of skin lesions were about 70% higher at water concentrations of 50–100 μg/L than at less than 10 μg/L. The odds nearly doubled in those exposed at 100–200 μg/L and nearly tripled in those exposed at 200 μg/L or higher (Argos et al. 2011). The association may be modified by sex (Chen et al. 2006), genetic variations (Ahsan et al. 2007), and diet (Pierce et al. 2011).

The clear dose–response relationship persists when exposure is measured by way of urinary concentration of monomethyl arsenic (MMA), a metabolic intermediate of arsenic. A case–control study in Bangladesh found a dose–response relationship between the percentage of total urinary arsenic that was MMA and skin lesions in 594 cases and 1,041 population-based controls (Ahsan et al. 2007). A small Taiwanese case–control study of 26 skin-lesion patients with sex- and age-matched controls indicated that participants who had a high percentage of MMA (more than 15.5%) had an odds ratio (OR) of 5.5 compared with those who had a low percentage of MMA, and those with low dimethyl arsenic (DMA), less than 72.2%, had an OR of 3.25 compared with those who had high DMA (Yu et al. 2000).

In a Mexican cross-sectional study of 104 people who lived near mining operations and contaminated groundwater (76 exposed at 50 μg/L or more, 28 exposed at 10 μg/L or lower), residents who had skin lesions had urinary MMA concentrations of 7.5 μg/L whereas residents who did not have lesions had 4.8 μg/L (Valenzuela et al. 2005). By using pathway analysis to address the collinear relationships among urinary arsenic metabolites, Kile et al. (2011) have also shown that urinary MMA concentration is associated with an increased risk of skin lesions in a case–control study.

Skin Cancer

Arsenic is an established skin carcinogen, first classified by the International Agency for Research on Cancer (IARC) in 1987 on the basis of observations of patients treated with the arsenical Fowler’s solution. A causal relation between arsenic in drinking water and skin cancer was confirmed in the 2012 IARC publication based on ecologic studies in Taiwan, primarily in the southwest region, where geologic arsenic is endemic (Wu et al. 1989). To date, almost all published studies linking arsenic exposure to skin cancer have found supportive evidence of nonmelanoma skin cancers (basal-cell and squamous-cell carcinoma).

The evidence of skin-cancer risk posed by high arsenic exposure appears to be well supported by cohort, case–control, and ecologic studies, with increasing arsenic exposure associated with increasing skin-cancer risk. A US population-based case–control study of skin cancers also reported evidence of an increase in invasive squamous-cell carcinoma in those who had the highest concentrations of toenail arsenic (Karagas et al. 2001b).

At lower arsenic exposure, some studies have been able to estimate the effect of arsenic exposure on skin-cancer risk although in general the findings from these studies have been less conclusive. The ASHRAM case–control study in Hungary, Romania, and Slovakia found an association between histologically confirmed basal-cell carcinoma of the skin and drinking-water inorganic arsenic concentrations greater than 100 μg/L (Leonardi et al. 2012). The study suggested that at lower concentrations an OR of 1.18 was associated with each 10-μg/L increase in average lifetime water concentration, with the odds of basal-cell carcinoma increasing with increasing concentrations of inorganic arsenic. A case–control study in Slovakia showed statistically significantly higher concentrations of urinary arsenic metabolites in people who had nonmelanoma skin cancer than in controls (Ranft et al. 2003). Arsenic exposure was primarily from residential proximity to a coal-burning power plant. The urinary arsenic concentrations in the study were all less than 50 μg/L but showed an association with nonmelanoma skin-cancer status.

Other studies of lower arsenic exposure have been unable to measure skin-cancer risk at the lower concentrations precisely because of methodologic and sample-size limitations. A Danish ecologic study of low exposure (primarily arsenic at less than 2 μg/L) was conducted with geographic information system (GIS) methods. No association was observed between water arsenic and nonmelanoma skin-cancer incidence (Baastrup et al. 2008). Aside from the extremely low concentrations, the study was limited by lack of histologic specificity in that basal-cell and squamous-cell carcinomas were grouped as one outcome. A small case–control study in Lagunera, Mexico, with 42 cases and 48 controls suggested an increased risk of nonmelanoma skin cancer, which was modified by the presence or absence of human papillomavirus (Rosales-Castillo et al. 2004). Exposure in the study was measured as cumulative exposure as determined by measurement of urinary arsenic and extrapolation to cumulative exposure after assessment of the participant’s residential history.

A more recent US case–control study of squamous-cell carcinoma used sensitive methods for urinary arsenic detection. It reported a dose-related increase in total urinary arsenic after arsenobetaine was subtracted out (median = 4.76 μg/L) and a trend in each urinary fraction (inorganic arsenic, MMA, and DMA), with the association strongest for MMA (Gilbert-Diamond et al. 2013).

A US case–control study of melanoma skin cancer found an increased risk starting at an toenail arsenic concentration of 0.04 μg/g and showed evidence of a dose-related increase in risk above that concentration (Beane Freeman et al. 2004). The risk was particularly strong in those who had a prior skincancer diagnosis. However, the study was limited by the use of colon-cancer cases as controls.

Mode of Action

Few studies have been able to elucidate the molecular or cellular basis of skin lesions or cancer directly in skin tissues in human populations exposed to arsenic. Several in vitro and animal studies have attempted to examine molecular and cellular alterations in response to external exposure to arsenic at various doses. Those studies have identified immune-related, oxidative stress–related, apoptotic, stem-cell, mitochondrial, and genomic alterations that may potentially underlie basal-cell and squamous-cell cancer development in response to arsenic stimuli (Yu et al. 2006; Kitchin and Conolly 2010; Lee et al. 2011; Liao et al. 2011; Tokar et al. 2011a; Zhao et al. 2012; Huang et al. 2013; Lee et al. 2013; Pei et al. 2013). The findings from the experimental studies are difficult to extrapolate directly to understand how arsenic induces skin lesions or tumor development in humans. However, to the extent that the studies involved exposures at relevant dose ranges and are compatible with the epidemiology, the evidence might be worth considering in the IRIS assessment to strengthen or refute results of analyses that are relevant to skin pathology.

Although this section integrates discussions of skin lesions and skin cancers, the committee notes that although skin lesions are clinical markers of susceptibility to arsenic-related skin and internal cancers and other health outcomes, it is not clear whether skin lesions are direct precursors of most skin cancers in arsenic-exposed people. Despite the existing dogma, no published longitudinal studies have documented the transition from skin lesions to skin cancer in humans. Thus, it is important to evaluate modes of action underlying skin lesions and skin cancer carefully to understand how they are similar and different and their potential implications for dose–response analyses for these seemingly related end points.

Key Considerations for the IRIS Assessment

The committee recommends that EPA consider skin studies that have histologic specificity and that allow the separation, for example, of basal-cell carcinoma and squamous-cell carcinoma because current data suggest that they are influenced by arsenic exposure and could have different dose–response relationships.

RESPIRATORY EFFECTS

Lung Cancer

Arsenic exposure via drinking water is an established lung carcinogen in humans. Associations have been consistently observed in highly exposed populations in Taiwan, Japan, Chile, Argentina, and the United States (Guo 2004; IARC 2004, 2012). And ecologic analyses in Chile suggest that in utero or early-life exposure increases the risk of arsenic-related lung-cancer mortality (Smith et al. 2006).

Case–control and cohort studies in those and other regions examined risks at lower levels of exposure (reviewed in EFSA 2009; Steinmaus et al. 2013). For example, a hospital-based case–control study in Chile (151 cases and 419 controls) detected increased risks beginning with the category of average water concentrations of 10–29 μg/L which became statistically significant at 30–39 μg/L compared with 10 μg/L or lower (Ferreccio et al. 2000). In a population-based study of 306 lung-cancer cases and 604 controls, the ORs were specifically increased in those who drank water during the peak exposure periods (1958–1970), which was about 40 years or more before the diagnostic period of the cases (2007–2009); however, no test of latency or dose–response effects was performed (Steinmaus et al. 2013).

In a preliminary case–control study in Argentina, there was evidence that the percentage of MMA and polymorphisms in cystathionine beta-synthase modified the risk of arsenic-induced lung cancer (Steinmaus et al. 2010).

In Taiwan, a cohort study (139 lung-cancer cases) found a trend of increasing lung-cancer risk beginning at an arsenic concentration of 100 μg/L; less than 10 μg/L in water was the reference concentration, and there was no observable association with 10-99 μg/L (Chen et al. 2004).

In Bangladesh, a trend of an association between well-water arsenic concentration and risk of malignant lung tumors, in particular squamous-cell tumors, was found in a pathology-based case-control study of smokers (3,223 cases, 1,588 controls); a limitation is that the study used people who had suspicious lung lesions (for example, inflammatory, tubular, or other disease) as controls (Mostafa et al. 2008).

Each of those studies examined trends in estimates of relative risk associated with specific categories of drinking-water arsenic, however, and did not model continuous dose–response relationships. The one study that did examine a linear dose–response relationship was conducted on the basis of GIS exposure models in Denmark and found no association with lung cancer (409 cases) (Baastrup et al. 2008); however, the concentrations of arsenic in drinking water were less than 2 μg/L for most participants. Likewise, in an ecologic analysis in a population of Mormons in Utah (34 lung-cancer deaths) (Lewis et al. 1999), there were no observable associations compared with the US general population. However, such ecologic comparisons at low exposure are not likely to be informative, because of misclassification and confounding. A case–control study in New Hampshire that measured people’s toenail arsenic as a biomarker (223 cases, 238

controls) found an increased OR for small-cell or squamous-cell carcinoma of the lung at higher concentrations of arsenic, especially in those who had a history of lung disease (Heck et al. 2009a). In the Strong Heart Study of 3,932 American Indians 45-74 years old, baseline urinary arsenic concentrations were related to an increased hazard ratio for lung-cancer mortality of 1.56 (95% confidence interaval [CI] 1.02-2.39) over almost 20 years of followup; 78 lung-cancer deaths were observed. There was a linear trend of increasing lung-cancer mortality with increasing urinary arsenic concentrations with no evidence of a threshold (p = 0.04) (Garcia-Esquinas in press). Most recently, a large prospective cohort study in Japan observed a dose-dependent increase in lung-cancer incidence in relation to arsenic intake in food as measured through a food-frequency questionnaire (Sawada et al. 2013). The increase in risk was particularly pronounced in cigarette-smokers.

Mode of Action

The mode of action of arsenic in lung carcinogenesis is not completely understood, but some patterns appear to be emerging. Although arsenic is not directly mutagenic, it has been shown to affect several oncogenic pathways that are relevant to lung cancer, including epigenetic, microRNA, gene expression, and mitochondrial DNA alterations. In particular, three oncogenic pathways have been implicated in lung cancer, including those of the epidermal growth factor receptor (Biscardi et al. 1999; Simeonova and Luster 2002; Andrew et al. 2009a; Li et al. 2011; Stueckle et al. 2012; Sung et all. 2012), phosphoinositide 3-kinase/Akt (Dong 2002; Gao et al. 2004; Zhang et al. 2006; Beezhold et al. 2011; Chen et al. 2012; Z. Wang et al. 2012), and Nrf2/Keap1 (X.J. Wang et al. 2007; Andujar et al. 2010; Zheng et al. 2012).

Key Considerations for the IRIS Assessment

There is evidence of a trend of increasing risk at higher levels of arsenic exposure, but relatively few studies at lower levels of exposure have examined dose–response relationships with lung cancer. It may be possible to model the dose–response relationships from the estimated relative risks associated with categories of exposure. Risks may involve cofactors (including genetic factors) and be modified by timing of exposure (for example, early in life) and may necessitate assessment of potential confounding by cigarette-smoking. Associations could be specific to histologic types of lung cancer. The ability to detect associations at low levels of exposure requires measurement not only of water concentrations of arsenic but of biomarker concentrations that encompass all sources of exposure, including diet, that may be important especially when drinking-water arsenic concentrations are relatively low.

Nonmalignant Respiratory Outcomes

Arsenic’s noncancer pulmonary effects have been less well studied than its lung-cancer effects, but results of a number of human epidemiologic studies suggest deleterious effects of arsenic on a variety of nonmalignant pulmonary outcomes, including respiratory symptoms, airway epithelial damage, impaired pulmonary function, chronic obstructive pulmonary disease (COPD), and tuberculosis (Mazumder et al. 2000, 2005; Milton and Rahman 2002; Milton et al. 2003; Smith et al. 2006, 2011; Parvez et al. 2008; Rahman et al. 2011).

Cross-sectional studies in Bangladesh and India report increased risks of clinical respiratory symptoms in people who have skin lesions and are exposed to arsenic at concentrations greater than 500 μg/L (Mazumder et al. 2000; Milton and Rahman 2002). A prospective evaluation of the participants (11,746) in the Health Effects of Arsenic Exposure Longitudinal Study in Bangladesh revealed that those in the second quintile of water arsenic (7-40 μg/L) had a 27% increase in risk of respiratory symptoms compared with those in the lowest quintile (Parvez et al. 2010). In a subset of the same cohort (950), arsenic exposure was associated with significantly lower forced expiratory volume in 1 second (FEV1) and forced vital capacity (FVC) in people exposed to water arsenic at concentrations above 97 μg/L than in those

exposed at less than 19 μg/L (Parvez et al. 2013). The greatest effect of arsenic was evident in male smokers and people who had skin lesions. A handful of earlier studies in India and Pakistan also observed decreased pulmonary function in those exposed at greater than 250 μg/L (De et al. 2004; von Ehrenstein et al. 2005; Nafees et al. 2011).

A retrospective cohort study in Chile reported that in utero arsenic exposure was associated with a decline in FEV₁ and FVC in adulthood (Dauphine et al. 2011). In utero exposure to arsenic has also been linked with increased risk of acute respiratory infections in infants in a large cohort (1,551) in Bangladesh (Rahman et al. 2011). A cross-sectional study in India reported a 10-fold increase in the prevalence of bronchiectasis, a specific type of COPD, in people who had skin lesions (108) compared with those who did not (150) (Mazumder et al. 2005). An earlier ecologic study in the arsenic-endemic areas of Chile reported an excess of mortality from COPD in people 30–39 years old (especially women) who were exposed to a weighted average arsenic concentration of 570 μg/L during 1955–1969 (Smith et al. 1998). Higher mortality from chronic bronchiectasis was observed only in those who were exposed to arsenic during early childhood or in utero (Smith et al. 2006). The ecologic analysis captured mortality in 1989–2000 in the two cohorts that were exposed to arsenic in early childhood or in utero before (1950–1957) or during (1958–1970) times when the peak exposure reached as high as 1,000 μg/L. Significantly higher risks of death from bronchiectasiswere observed in those who were exposed to arsenic during early childhood (standardized mortality ratio [SMR] = 12.4; 95% CI 3.3–31.7) or in utero and early childhood (SMR = 46.2; 95% CI 21.1–87.7) (Smith et al. 2006). An ecologic study in Taiwan also reported higher mortality from bronchitis in people living in areas with a high prevalence of blackfoot disease than in a reference population or the rest the country during the peak arsenic exposure (median 780 μg/L, 1971-1994) (Tsai et al. 1999).

In addition to COPD and associated outcomes, arsenic exposure has been linked to chronic lung infections, especially pulmonary tuberculosis. In the same population in Chile discussed above, Smith and colleagues reported significantly higher mortality from pulmonary tuberculosis (RR = 2.1; 95% CI 1.7-2.6) resulting from arsenic exposure (Smith et al. 2011). The risk mimicked the exposure history trend in the population (rising risk after chronic exposure and then falling back to normal after cessation of exposure). Although that novel finding is consistent with potential immunosuppression effects of arsenic, it has yet to be replicated in other large population studies.

A recent study in the United States has reported an association between prenatal inorganic arsenic exposure and increased risk of respiratory disease, such as upper respiratory tract infections and colds (Farzan et al. in press).

Mode of Action

As discussed in the committee’s workshop by Lantz (2013), a substantial body of literature from both epidemiologic and animal studies of arsenic suggests impaired immune function, aberrant wound repair, and disrupted matrix and barrier function. A major consideration in elucidating the mode of action should be epidemiologic and animal studies that identify potential pathogenic mechanisms in response to low to moderate arsenic exposures. As indicated above, results of prospective epidemiologic studies suggest that arsenic exposures are linked to chronic loss of lung defenses, such as secretion of CC16 protein from airway cells (Parvez et al. 2008, 2010). Subchronic or in utero exposures of mice to low to moderate concentrations of arsenic in drinking water (10–100 ppb) led to a decrease in immune gene expression and aberration in inflammatory protein expression (Kozul et al. 2009a; Ramsey et al. 2013a) that may make mice more susceptible to airway inflammation (Kozul et al. 2009b). However, other mouse studies report that genes that sustain lung and vascular matrix, wound repair, and barrier function are compromised by arsenic exposure (Lantz et al. 2007, 2009; Hays et al. 2008; Petrick et al. 2009). Those changes in matrix and wound repair observed in mice correlate to changes observed in exposed humans and isolated human cells (Josyula et al. 2006; Lantz et al. 2007; Olsen et al. 2008). Increased matrix metalloproteinase-9 (MMP-9) expression and activity may be important biomarkers of inhibitory effects of arsenic on lung function (Josyula et al. 2006;

Olsen et al. 2008), although they may also be portions of the overall change in inflammatory gene expression. Another important consideration with respect to mode of action is whether in utero or perinatal exposure poses a significant risk of lung dysfunction or disease. Recent animal studies suggest that that may be the case (Lantz et al. 2007, 2009; Hays et al. 2008; Kozul et al. 2009b; Ramsey et al. 2013b), but few epidemiologic studies have examined affects of in utero arsenic exposure on lung disease. The underlying mechanism of chronic inflammation, impaired immune function, and aberrant matrix maintenance is not clear but may involve arsenic-induced oxidative stress (Lantz and Hays 2006). Protection from oxidative stress can reduce arsenic-induced pulmonary inflammatory responses and injury in mouse models (Zheng et al. 2012; Tao et al. 2013).

Key Considerations for the IRIS Assessment

Despite a large number of studies that have demonstrated nonmalignant lung disease after exposure to a wide range of arsenic (40-1,000 μg/L), little information is available to determine a full dose-dependent relationship, especially at low to moderate doses. For the IRIS assessment, the challenge is to integrate information from different studies that focus on different types of nonmalignant respiratory outcomes in populations exposed at different doses and for different periods. Critical synthesis of the human population studies coupled with a coherent integration of mode of actions underlying different nonmalignant respiratory outcomes and phenotypes should be a focus for consideration by EPA.

CARDIOVASCULAR DISEASE

In 2012, a systematic review and meta-analysis of epidemiologic reports of arsenic-related cardiovascular disease concluded that evidence from different countries consistently shows an association between high chronic arsenic exposure and cardiovascular disease (Moon et al. 2012). (See Chapter 3 for comments on the usefulness of this systematic review for EPA’s IRIS assessment.) Until recently, the evaluation of the evidence on low to moderate exposure in drinking water (less than 100 μg/L) has been limited by the low quality of the available studies that support the role of arsenic as a cardiovascular-disease risk factor (Moon et al. 2012). However, a number of epidemiologic studies of arsenic exposures in large populations report dose–response relationships that are useful to evaluate if there is an increased risk of cardiovascular disease at low to moderate arsenic concentrations (Sohel et al. 2009; Wade et al. 2009; Y. Chen et al. 2011a) (see Table 1). The joint evaluation of the dose–response relationship across those studies can provide useful information for performing the IRIS assessment. Cardiovascular disease may be the most important noncancer disease risk posed by environmental arsenic exposures given the high burden of cardiovascular disease worldwide.

Although the current studies suggest that cardiovascular disease risk is increased by low to moderate arsenic concentrations in drinking water and possibly food, there is a need to confirm the relationship. Inclusion of the most recently published data should provide an opportunity for establishing better dose–response relationships in the low-dose region. Ample human studies are available to establish dose–response relationships. However, results of a number of animal studies support the identification of cardiovascular disease risk and provide valuable information on mode of action that reduces the uncertainty regarding causality (see below). The focus should be on human studies that investigated coronary arterial disease, myocardial infarction, cardiovascular disease, and overall cardiovascular-disease mortality because risk of these disease outcomes appears to be increased by low to moderate exposure. Peripheral arterial disease (such as blackfoot disease) can be excluded from the focus because few studies have investigated the association with low to moderate arsenic exposure and the association with high exposure seems to be limited to populations that have poor nutrition (Tseng et al. 2005). Cerebrovascular disease can be included, but there may be few conclusive epidemiologic studies that had adequate exposure and

TABLE 1 Large Cohort Studies of Overall Cardiovascular Disease and Arsenic Exposure Measured at the Individual Level and Reported in Two or More Arsenic Categories in Populations with Arsenic in Drinking Water at Less than 150 μg/L

This table is provided as an example that can be useful also for other end points (such as ischemic heart disease).

Abbreviations: BMI, body-mass index; CI, confidence interval; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; LDL-chol, low-density lipoprotein cholesterol; NR, not reported; SES, socioeconomic status.

^aData on exposures greater than 150 μg/L are not presented.

^bSubset of residents exposed since 1990.

outcome assessment at low to moderate exposure. Even in studies that had high exposures, the relationship of arsenic to cerebrovascular disease appears inconclusive. Finally, hypertension, although shown to have some relation to arsenic exposure (Abhyankar et al. 2012), is less of a priority inasmuch as there are few studies at low to moderate exposure, most of the evidence available is cross-sectional, and it is not a clinical cardiovascular end point.

Mode of Action

As with arsenic-induced cancers, there may be several plausible modes of action for arsenic-related cardiovascular conditions. If determination of the mode of action is needed, EPA should consider evidence and reviews that suggest that chronic inflammation and reactive oxygen species (ROS) formation are central to the pathogenesis of arsenic-induced cardiovascular disease. Chronic vascular inflammation is suggested by a mechanistic epidemiologic study (Wu et al. 2012) within a large prospective study of cardiovascular-disease mortality associated with arsenic (Y. Chen et al. 2011a). Chronic inflammation may be a derivative of a chronic increase in ROS, a known risk factor for cardiovascular disease. Animal studies have been effective in demonstrating that vascular remodeling occurs in response to low to moderate arsenic exposures (10-100 ppb) (reviewed by States et al. 2011) and that ROS generation, especially from NADPH oxidase activation, is essential in promoting vascular pathogenesis (Straub et al. 2008). Atherosclerosis is a primary mechanism in the etiology of arsenic-related ischemic heart disease, and arsenic-induced atherogenesis has been demonstrated in a genetic mouse model of atherosclerosis after low to moderate exposure to arsenic, as discussed by Lantz (2013) in the committee’s workshop (also see Lemaire et al. 2011). In addition, atherogenic potential in mice may be enhanced by in utero arsenic exposures (Srivastava et al. 2007). Alternatively, other modes of action—such as interactions with protein thiols to change protein function, disruption of cardiac and vascular matrix after increased expression of matrix-degrading proteases (such as MMP-9) (Soucy et al. 2005; Hays et al. 2008; States et al. 2009; Wu et al. 2012), or interference with transcriptional regulation of metabolic and inflammatory genes (Padovani et al. 2010; States et al. 2011)—cannot be ruled out. However, those changes in gene regulation may act through or derive from upstream activation of ROS signaling.

Key Considerations for the IRIS Assessment

Ample epidemiologic reports find a causal association between arsenic exposure of humans and increased risk of cardiovascular disease and mortality. A review of current literature, including recent meta-analyses, provides support for identification of arsenic as a hazard for cardiovascular disease. The recent literature includes large prospective studies that provide an opportunity to establish dose–response relationships for arsenic-induced cardiovascular disease. In evaluating the causal relationship between arsenic exposure and cardiovascular diseases, however, EPA should address potential uncertainties resulting from differences between the study population and the general population. Animal data and data that reveal potential modes of action demonstrate sensitivity of the cardiovascular system to arsenic exposures relevant to humans and support identification of modes of action. That can be useful for supporting causality in the epidemiologic studies, help to identify sensitive populations, and inform the dose–response analysis.

BLADDER EFFECTS

Arsenic is a known bladder carcinogen (IARC 2004, 2012). Assessment by IARC included ecologic studies of highly exposed populations in Taiwan, Chile, and Argentina and indicated higher mortality from bladder cancers in exposed than in nonexposed populations. In addition, ecologic studies in Taiwan reported increasing SMRs with higher categories of exposure on the basis of village median water concentrations of arsenic, and associations with well-water arsenic concentrations were supported by evi-

dence from cohort and case–control studies in Taiwan (IARC 2004, 2012). More recently, a case–control study in Chile found evidence of a dose-related increase in bladder-cancer incidence (Steinmaus et al. 2013). Those studies were important for establishing a causal relationship between arsenic exposure and bladder cancer, and, as described later in this report, those from Region II in Chile signal the importance of exposures early in life. Those studies, however, did not estimate the shape of the dose–response curve and cannot inform estimates of the dose–response relationship at lower exposures.

There is supportive evidence that drinking-water arsenic could be affecting bladder-cancer occurrence in the United States; this evidence should be carefully evaluated by EPA, particularly with respect to smoking, which is a known risk factor for bladder cancer (see Chapter 5). The studies include an ecologic study of the proportion of private-well users that found correlations in New England where high arsenic concentrations in bedrock wells are known to occur (Ayotte et al. 2006). Case–control and cohort studies that detected effects at lower exposure tended to have poor statistical precision and mostly examined dose–response relationships by using categories of exposure. For example, in a categorical analysis in a nested case–control study of male smokers (Michaud et al. 2004), the OR was over 2 in the highest tertile of toenail arsenic in long-term smokers (older than 45 years old), but this was not statistically significant. Likewise, a case–control study in New Hampshire found about a 2-fold risk of transitional-cell carcinoma of the bladder in the highest exposure category in smokers but with wide confidence intervals (Karagas et al. 2004). However, an analysis of the New Hampshire data that treated exposure as a continuous variable observed a dose–response relationship with increasing exposure in smokers, and a case–control study in Nevada reported more than a 2-fold OR of bladder cancer in smokers in the highest category of estimated cumulative intake of arsenic via drinking water, but it was evident only for an exposure period 40 years before diagnosis (Steinmaus et al. 2003).

In an earlier US National Bladder Cancer Study, a linear trend in the ORs for bladder cancer was observed in smokers on the basis of index of arsenic exposure (drinking-water concentrations multiplied by number of years consumed), a trend present for the exposure window of 10–19 years before diagnosis (Bates et al. 1995). Fifty years or more of well-water consumption in a region of Argentina that had high contamination was associated with bladder cancer in a case–control study of smokers (Bates et al. 2004). Estimated drinking-water arsenic exposure (limited to up to 40 years before diagnosis), however, was unrelated. A cohort study in Taiwan detected risks at arsenic concentration below 50 μg/L but was based on only a few cases (two men and five women exposed at less than10 μg/L and one man and three women at 10–49 μg/L) (Huang et al. 2008); thus, a dose–response analysis could not be performed. No excess risks were observed in association with the time-weighted average arsenic exposure from drinking water, even among smokers, in a case–control study in Michigan (Meliker et al. 2010). No association was also found in an ecologic analysis of a cohort study of bladder-cancer mortality in Mormons (a nonsmoking population) in Utah (Lewis et al. 1999) or in an analysis of cancer incidence in an Australian cohort (Hinwood et al. 1999). It is conceivable that lack of associations in studies at low exposure could be due to weak statistical power to detect the more modest effects that might be expected, especially in the cohort studies that had few bladder-cancer cases. Studies based on bladder-cancer mortality are inherently biased toward the null because bladder cancer is typically survivable.

Mode of Action

An emerging body of literature suggests that susceptibility to arsenic exposure may involve genetic factors, and studies have helped to inform understanding of the mechanism of action. Findings include interactions between arsenic exposure and polymorphisms in genes involved in DNA repair, cell cycle, xenobiotic and arsenic metabolism, and metal transport (e.g., Hsu et al. 2008; Andrew et al. 2009b; Karagas et al. 2012; Lesseur et al. 2012), and differences in risk depending on percentages or ratios of urinary metabolites (IARC 2012). These studies are discussed in greater detail in Chapter 5 (section on Genetics and Arsenic Metabolism and Toxicity).

There is a vast literature on mechanistic studies related to arsenic tumorigenesis. Experimental designs range from detailed probing of specific molecular pathways to broad studies of genetic differences

associated with carcinogenesis or cancer risk, including gene-expression studies and population-based studies of genomic changes. Potential modes of action have been proposed, such as cytotoxicity and regenerative repair, formation of reactive oxygen species and resulting oxidative stress and DNA damage, impaired DNA repair, dysregulation of signaling pathways that control the fate of cells (cycle progression, proliferation, differentiation, and apoptosis), and perturbations of gene structure. Epidemiologic and experimental studies have shown genetic differences and epigenetic changes in various processes. A mode-of-actin analysis will need to evaluate the results of mechanistic studies. It will need to consider that, just as bladder cancer is not a single disease, there may be multiple means by which arsenic can cause bladder cancer. Which mode of action applies may depend on many factors, including the populations or species in question (in the case of experimental studies), individual genetic differences, nutritional differences, magnitude of arsenic exposure, and the period of exposure (e.g., in utero, adult).

Key Considerations for the IRIS Assessment

There are a number of considerations in assessing the potential relationship between arsenic exposure and bladder cancer, especially at levels of exposure relevant to the US population. Evaluation of dose–response relationships may necessitate conversion of categories of exposures to fit continuous dose–response curves as was done in the European Food Safety Authority Assessment (EFSA 2009). Complications in evaluating lower exposures include diet’s possible importance as a contributor to exposure and the need to incorporate considerations regarding smoking and other cofactors. It is crucial to assess exposure on an individual level and to include a biomarker and relevant cofactors where possible. A limitation of early ecologic studies of highly exposed populations that relied on mortality data is that bladder cancer is not typically fatal. In addition, timing and duration of exposure can influence the magnitude of the effects of arsenic on bladder-cancer incidence. Associations were observed with more recent exposure in some case–control studies and exposure in the distant past in others. In summary, if possible, the IRIS assessment should focus on the studies of both recent and past exposure that examined bladder-cancer incidence (rather than mortality) and that examined susceptible groups of the population (on the basis of cofactor exposures or genetics) and dose–response relationships.

RENAL EFFECTS

Adverse effects on the kidneys have been found in different arsenic-exposed populations around the world, including the United States. Renal cancer is relatively rare, but chronic renal disease is common. Although reports vary in study design, have different extents of exposure, and use different outcome measures, they appear to concur that arsenic exposure causes renal disease in humans. Some examples are discussed below, but they are not meant to limit EPA’s evaluation of the literature.

Renal Cancer

Arsenic has been shown to cause cancer of tissues in the urinary system other than the bladder; for those, there may be different modes of action for cancer development. Broadly speaking, cancer of the urinary system is either renal-cell carcinoma or cancer of the urothelium (which occurs in the renal pelvis and ureters as well as the bladder). In International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes commonly used in epidemiologic studies of cancers associated with arsenic exposure, urothelial cancer is coded as a subtype of renal cancer rather than bladder cancer. As noted in the committee’s workshop by Cohen (2013), most cancers of the urinary system occur in urothelial tissues, especially the bladder.

Increased mortality from cancer of the kidney, including the renal pelvis and ureters (ICD-9, code 189), has been reported in studies in Argentina (Hopenhayn-Rich et al. 1998), Chile (Yuan et al. 2010),

and China (Tsai et al. 1998; Yang et al. 2004). In Chile, mortality increased by about 10 years after high exposure began, peaked around 25 years later, and then declined; the decline occurred later in women (Yuan et al. 2010). The incidence of urothelial or transitional carcinoma increased with arsenic exposure (Ferreccio et al. 2013). In the committee’s workshop, the study was described as having excellent exposure assessment (Cantor 2013). Yang et al. (2004) calculated cancer as a moving average over the study years of 1971–2000 and found that renal, renal pelvis, and ureteral cancer mortality decreased as the time since the end of exposure increased. They noted that smoking rates remained constant during that time, and this strengthened the case for a causal role of arsenic in these cancers. Chen et al. (2010) examined the incidence of urinary cancer in northeast Taiwan, where arsenic concentration in well water ranged from less than 10 μg/L to more than 300 μg/L. Urothelial-cancer incidence—adjusted for age, sex, education levels, cigarette-smoking, and alcohol use—increased with well-water arsenic concentration. (During the committee’s workshop, it was pointed out that concentrations were based on the well used by the subjects at the time of enrollment in the study.) Although relatively few cancers were observed, the numbers of both total urinary system cancers (45 cases) and urothelial carcinoma (36 cases) were significant and showed positive dose–response relationships.

Only a few papers have reported on cancer of the kidney excluding urothelium (ICD-9 code 189.0). Ferreccio et al. (2013) did not find renal-cell cancer incidence to be increased in Chile, and Chen et al. (2010) reported that renal-cell cancer was not increased in northeastern Taiwan. In southwestern Taiwan, renal-cancer mortality increased 6-fold (in males) to 16-fold (in females) at the highest arsenic concentrations (over 600 ppb) (Chen et al. 1992). Huang et al. (2011) conducted a case–control study of the incidence of pathologically verified renal-cell cancer in the Taipei area of Taiwan, where arsenic concentrations in drinking water range from undetectable to 4 μg/L (average 0.7 μg/L). They reported that the OR for renal-cell cancer was increased (age- and sex-adjusted OR = 2.31) in the top one-third of urinary arsenic (greater than 20.95 μg/g of creatinine) compared with the referent group (greater than 12.3 μg/g of creatinine). In a followup study, the group that had the highest urinary arsenic also had significantly higher urinary 8-hydroxydeoxyguanosine concentrations (also corrected for creatinine), a biomarker of DNA damage that has previously been associated with risk of renal cancer; this highlights a potential mechanism of cancer formation (Huang et al. 2012).

Mode of Action

Potential modes of action for renal cancer include those described in Chapter 6: oxidative stress (production of reactive oxygen species and diminished antioxidant reserves), DNA damage, cell proliferation and accumulation of mutations, and dysregulation of cell-cycle control. Recent studies by Tokar et al. (2012a) specific to the kidney found that an increase in renal-cell carcinoma resulted from in utero exposure to inorganic arsenic followed by DMA after weaning. In a followup study (Tokar et al. 2012b), rat kidney stem cells exposed to arsenic for 10 weeks displayed characteristics of cancer cells—a suggestion that early-life effects may be key to the development of renal cancer.

Renal Disease

Significant higher mortality from renal disease (ICD-9 codes 580–589) has been found in regions that have high arsenic concentrations, such as Taiwan (Tsai et al. 1999) and Chile (Smith et al. 2012), than in regions that have low arsenic. In Taiwan, SMRs for renal diseases decreased from 1971, when use of low-arsenic water became common, to 2000 (Chiu and Yang (2005); this finding strengthens the association with arsenic. In Antofagasta, Chile, chronic renal diseases (chronic renal failure, unspecified renal failure, chronic glomerulonephritis, and sclerosis) were increased in those who were exposed in childhood or in utero and in childhood (Smith et al. 2012).

Renal disease may not be the cause of death, so studies of renal-disease incidence or changes in biomarkers of impaired renal function may reflect the true impact of arsenic exposure on the kidneys better.

The incidence of renal disease (chronic glomerulonephritis, nephrotic syndrome, nephritis and nephropathy, and chronic or unspecified renal failure) was increased in southwestern Taiwan (Wang et al. 2003). In a cross-sectional study of a population exposed to moderate arsenic concentrations in central Taiwan (geometric mean urinary arsenic 70.6 μg/L), urinary arsenic was associated with an estimated glomerular filtration rate (eGFR) of less than 90 mL/min/1.73 m² but not with an eGFR of less than 60 mL/min/1.73 m², the standard definition of reduced eGFR (J.W. Chen et al. 2011). The association of arsenic with measures of glomerular filtration rate has not been evaluated in populations exposed to arsenic at low to moderate concentrations. However, a study in southeastern Michigan, a region that has a population-weighted mean water arsenic concentration of only 11 μg/L reported increased SMRs for renal disease over expected values in both males and females in association with water arsenic (Meliker et al. 2007). The appearance of protein or albumin in the urine can indicate increased permeability of the glomerular filtration barrier, impaired reabsorption of filtered proteins, or release of tubular proteins into the urine as tubular cells die. Increased rates of proteinuria or albuminuria have been reported after low to moderate arsenic exposures in the Strong Heart Study in the United States (Zheng et al. 2013) and after high exposures in Bangladesh (Y. Chen et al. 2011b). Moreover, proteinuria improved with the change to a water supply that was lower in arsenic. Impaired tubule function, indicated by excretion of β-2-microglobulin and N-acetyl-beta-glucosaminidase (NAG), has been found in populations exposed to arsenic at moderate to high concentrations (J.P. Wang et al. 2009; J.W. Chen et al. 2011); alterations in NAG activity in a Korean population that apparently had lower exposure (geometric mean urinary arsenic 8 μg/g of creatinine) have also been reported (Eom et al. 2011). Experimentation in rats showed that NAG excretion was increased a month after exposure to arsenic in drinking water at 30 mg/L (J.P. Wang et al. 2009).

Arsenic is associated with diabetes mellitus in populations exposed to arsenic (see section “Diabetes” below). Diabetes mellitus results in nephropathy, and this complicates the interpretation of the association between arsenic and renal disease. Several studies explicitly considered diabetes and found that the association between arsenic and impaired renal function remained after control for diabetes or in stratified analysis in participants who had and did not have diabetes. In a study in Bangladesh, the association was maintained after adjustment for hypertension and HbA1c; this supports the idea that proteinuria was independent of diabetes (Y. Chen et al. 2011b). Similarly, the association between arsenic and albuminuria was independent of diabetes in people who had low to moderate arsenic exposures in the United States (Zheng et al. 2013). J.P. Wang et al. (2009) reported that increases in urinary NAG were more pronounced in diabetic people than in nondiabetic people in a high-exposure area. Results of experimental studies support that interaction. Kidney weight (absolute and relative to body weight), oxidation of renal proteins, renal malondialdehyde, and measures of glomerular filtration (serum creatinine and urea nitrogen) were more severely affected by arsenic in rats that had been rendered diabetic with alloxan treatment (Patel and Kaila 2010). EPA should consider whether arsenic and diabetes have interactive effects on renal function.

Mode of Action

There have been few mechanistic studies of renal function. Processes described in Chapter 6 on modes of action might apply to kidneys. Studies have provided evidence of mitochondrial toxicity (Peraza et al. 2003) and increased mesangial proliferation secondary to hexokinaseII expression in the mesangial cells of the glomeruli (Pysher et al. 2007). The investigators noted that mesangial cells play an important role in the control of glomerular filtration and are affected by diabetes and that their proliferation decreases GFR. Differences in mechanisms of arsenic elimination could contribute to differences between species and strains in its effects on the kidney. For instance, lack of upregulation of the MRP1 transporter renders BALB/c mice more sensitive to the nephrotoxic effects of arsenic than C57Nl/6J mice (Kimura et al. 2005). Oxidative stress, inflammation, and other mechanisms of cell damage may also apply to kidney cells.

Key Considerations for the IRIS Assessment

There is mounting evidence that exposure to arsenic causes cancer of intrarenal urothelial tissues and renal-cell carcinoma in humans and mice. In humans, the incidence of renal cancer (urothelial and renal-cell carcinoma), renal-disease mortality, and proteinuria decreased after the arsenic concentration in water was reduced in Taiwan and Chile, and this supports a causal role. Results of studies in mice suggest that prenatal and early-life exposure confer susceptibility to renal-cell cancer. Renal-cell cancer is infrequent (about 3%) and has a low death rate (lower than 10%) in the United States (Pili and Rodriguez 2008). Thus, it is unlikely to be detected by mortality-based studies. Chronic kidney disease is more common (about 13% and an additional 0.16% with end-stage disease) (C.Y. Hsu 2011). Evidence is available from US populations that arsenic at low to moderate concentrations in drinking water is associated with proteinuria. In the case of other chronic renal diseases, there is only limited cross-sectional evidence at high exposure. Experimental studies support a role of arsenic exposure in kidney damage. In evaluating the literature on arsenic effects, EPA should consider chronic kidney disease and whether arsenic and diabetes have interactive effects on kidney function.

PREGNANCY OUTCOMES

Fetal Exposure to Arsenic

Early-life development is recognized as a potentially critical window of vulnerability to effects of toxic agents. It is essential to evaluate the potential adverse effects of fetal and postnatal exposure to inorganic arsenic. Experimental and human studies have shown that both inorganic arsenic and its methylated metabolites readily pass the placenta (Concha et al. 1998a; Jin et al. 2006; Hall et al. 2007). An important feature is the increase in the efficiency of maternal arsenic methylation during pregnancy (Concha et al. 1998a; Li et al. 2008; Gardner et al. 2011), which results in lower exposure of the fetus to inorganic arsenic and MMA, the most toxic arsenic metabolite, with advancing gestation. The maternal urinary fraction of MMA decreases markedly in the middle of the first trimester (Gardner et al. 2011), whereas the fraction of inorganic arsenic decreases slowly throughout gestation. Those findings are supported by studies in mice that show mainly DMA in blood and tissues of the newborn in spite of high maternal exposure to inorganic arsenic (10–85 mg/L in drinking water) during gestation (Devesa et al. 2006; Jin et al. 2006). Thus, the timing of urine sampling during pregnancy may have implications for the evaluation of arsenic methylation efficiency.

Birth Weight

Arsenic is embryotoxic and teratogenic in experimental animals that are given high doses (Wang et al. 2006; Hill et al. 2008). Because the kinetics and toxicity of arsenic vary among animal species, the extrapolation of data to humans is difficult.

With respect to human studies, a number of small, mainly cross-sectional studies have had inconclusive results concerning birth weight in relation to arsenic exposure, probably mainly because of the small samples. Of the larger studies, two ecologic studies in northeastern Taiwan and northern Chile found arsenic-related decreases in birth weight whereas a study in Inner Mongolia found an opposite association. In Taiwan, the authors compared birth weights in an area (four townships with 18 villages) that had well-water arsenic concentrations of 0.2–3,600 μg/L (30% of wells had over 50 μg/L) with those in matched townships (based on urbanization level) that had arsenic in public water supplies of less than 0.9 μg/L; the birth weights in the latter were an average of 29.0 g higher (95% CI 13.6–44.6 g), after adjustment for maternal age, education, marital status, and infant sex (Yang et al. 2003). In the Chilean study, the authors compared birth weights in the town of Antofagasta, which had arsenic in the public water supply at an average of 42 μg/L (33–53 μg/L) with those in Valparaiso, which had arsenic at less than 1 μg/L (Hopen-

hayn et al. 2003b). The multivariable-adjusted model showed that birth weights in the latter were an average of 57 g higher (95% CI -9 to 123 g).

A study in Inner Mongolia compared mean birth weight (9,890; from prenatal-care records) in four groups of “subvillages”, categorized according to the average of available well-water arsenic concentrations: less than 20, 21–50, 51–100, and over 100 μg/L (Myers et al. 2010). However, wells in a village were measured for arsenic only if the well in one of five randomly selected households exceeded the local standard of 50 μg/L, and this might have influenced the finding that birth weights in the village group that had the highest mean arsenic concentration averaged 50 g higher than those of the group that had the lowest mean arsenic concentration. In addition, well-water screening for arsenic took place in 1991–1997 whereas the pregnancies studied occurred from December 1996 to December 1999.

A recent population-based prospective study involving 1,578 mother–infant pairs in rural Bangladesh and measurements of arsenic in maternal urine collected in early and late gestation (median 80 μg/L, 10th–90th percentiles 26–400 μg/L) found a significant inverse association between size at birth and urinary arsenic concentration (Rahman et al. 2009). The dose-dependent difference in birth size was obvious mainly at maternal urinary arsenic concentrations below 100 μg/L; each increase of 1 μg/L was associated with a 1.68-g reduction in birth weight. Arsenic exposure was also associated with smaller head and chest circumferences in a similar manner. In the same cohort of pregnant women, a longitudinal evaluation of fetal growth characteristics in early and late gestation, as measured with ultrasonography, supported a weak association between maternal urinary arsenic concentrations and fetal size, mainly in boys (Kippler et al. 2012).

Fetal and Infant Loss

A few studies have probed the associations between prenatal arsenic exposure and fetal and infant loss. The associations with fetal loss are less convincing, particularly at low-dose exposure, but there seem to be fairly consistent results concerning infant mortality. Several of the few available human studies are, however, ecologic in design. Evaluation of trends in pregnancy outcomes in Antofagasta, in northern Chile, indicated an increased rate of late fetal loss (overall 3%) during the period that had the highest water arsenic concentrations (about 800 μg/L during 1958-1970) relative to mortality in Valparaiso, which had essentially no arsenic in the drinking water (Hopenhayn-Rich et al. 2000). The rate ratio was 1.7 (95% CI 1.5-1.9) for stillbirth, 1.53 (95% CI 1.4-1.7) for neonatal mortality, and 1.26 (95% CI 1.2-1.3) for postneonatal mortality after adjustment for location and calendar time. In an ecologic study in Bangladesh, outcome data on 30,984 pregnancies in 600 villages, grouped geographically in 16 centers, were related to the average water arsenic concentrations in the centers (each based on seven to 14 wells) (Cherry et al. 2008). After adjustment for socioeconomic and health factors, the OR for stillbirth (overall 3.4%) was 1.80 (95% CI 1.14-2.86) at 50–80 μg/L compared with that at less than 10 μg/L. Arsenic-related increases in risk of spontaneous abortion, stillbirth, neonatal death, and preterm birth have also been reported by cross-sectional studies in Bangladesh (192 and 533 women) and West Bengal (202 women), in which retrospective data about drinking-water sources and outcomes of previous pregnancies were collected (Ahmad et al. 2001; Milton et al. 2005; von Ehrenstein et al. 2006). The risk ratios were 2–3 for both spontaneous abortions (two studies) and stillbirths (all three studies) in high-exposure groups; but even in the largest of the studies, the number of cases was small. In the largest Bangladeshi study (Milton et al. 2005), the OR for neonatal death was 1.8 (95% CI 0.9-3.5) for water concentrations above 50 μg/L (70 cases) compared with concentrations below 50 μg/L (16 cases).

A few population-based studies in Bangladesh with individual exposure data have been conducted. Kwok et al. (2006) reported pregnancy-outcome data on 2,000 women in three areas that had known high arsenic concentrations in drinking water. A weak but statistically significant association between arsenic concentrations in drinking water and birth defects (OR 1.005 for all defects combined, 95% CI 1.001–1.010) was found, but no other adverse effects on pregnancy outcomes. Another study collected fetal and infant mortality data on 29,134 pregnancies in Matlab, Bangladesh (Rahman et al. 2007). Data on indi-

vidual arsenic exposure were based on interviews about history of drinking-water sources, carried out in a parallel study, which also measured arsenic concentrations in all functioning tube wells in Matlab. Women who were using water that had arsenic at 277-408 μg/L (fourth quintile) had a significant increase in the relative risk (RR) of fetal loss of 1.14 (95% CI 1.01–1.30) and of infant death of 1.29 (95% CI 1.08–1.53) (Rahman et al. 2007). There was a significant dose–response relationship between arsenic exposure and risk of infant death. In a later population-based, prospective cohort study of 2,924 pregnant women in the same area, the OR of spontaneous abortion was 1.4 (95% CI 0.96-2.2) in women who had urinary arsenic concentrations in the fifth quintile (249-1,253 μg/L, median 382 μg/L) compared with women in the first quintile (less than 33 μg/L) (Rahman et al. 2010). However, both the second and third quintiles showed similar ORs, so the dose–response relationship was not convincing. Infant mortality increased more clearly with increasing maternal urinary arsenic concentrations; a hazard ratio of 5.0 (95% CI 1.4–18) was calculated in the fifth quintile of maternal urinary arsenic concentrations (268–2,019 μg/L, median 390 μg/L) compared with the first quintile (less than 38 μg/L). Most of the infant deaths occurred in the first week after birth. Because the passage of arsenic into breast milk is limited, breastfeeding protects the infants from arsenic exposure (Concha et al. 1998b; Fangstrom et al. 2008).

Mode of Action and Susceptibility Factors

Fetal growth is influenced by multiple factors, including genetic predisposition, maternal nutrition, and environmental exposures. The mechanisms by which arsenic might affect birth size are not well understood but may involve oxidative stress or perturbation of oxidative defense that leads to placental insufficiency (Ahmed et al. 2011). Arsenic-associated changes in gene expression in cord blood related to stress, inflammation, and apoptosis have been reported (Fry et al. 2007). And experimental studies have shown that arsenic, even at low doses, may act as an endocrine-disrupting chemical (Bodwell et al. 2004; Davey et al. 2007, 2008) and influence the insulin growth-factor system, glucose homeostasis, and cellular growth. Recently, arsenic has also been associated with altered DNA methylation in cord blood and with histone modifications (Cronican et al. 2013). The immunosuppressive effect of arsenic (see section “Immune Effects” below) may contribute to infant mortality.

Key Considerations for the IRIS Assessment

It seems clear that all metabolites of inorganic arsenic easily cross the placenta to the fetus. A number of epidemiologic studies, including prospective cohort studies, have provided evidence that arsenic exposure through drinking water during pregnancy may cause dose-dependent impairment of fetal and infant growth and infant survival. Data on the effects of arsenic on fetal loss are inconclusive. For some outcomes, particularly birth size, the effects of arsenic seem to start at low-level exposure. There appear to be available data that can be used for dose–response characterization concerning birth size and infant growth and possibly infant mortality.

NEUROTOXICITY

Arsenic has traditionally been categorized as a peripheral neurotoxicant; it produces a clinical picture of severe polyneuropathy after severe poisoning. Occupational exposure, as can occur in copper smelters, has been associated with peripheral neuropathies and related poor motor function (Sińczuk-Walczak et al. 2010). However, recent animal and human studies suggest that arsenic neurotoxicity, even after environmental exposures, can include the central nervous system.

Human Population Studies

Studies of Children

Epidemiologic reports on neurotoxicity are largely consistent with the experimental studies described above. Low to moderate concentrations of arsenic in drinking water have been associated with neurocognitive deficits in children (see Table 2). Tsai et al. (2003) conducted an ecologic study of adolescents and found lower neurodevelopmental test scores, after adjustment for socioeconomic factors, in Taiwanese children who lived in areas that had high concentrations of arsenic in drinking water. Calderón et al. (2001) found that urinary arsenic concentrations were inversely associated with verbal IQ in 80 Mexican children (6–8 years old) who lived near a smelter even after adjustment for blood lead concentrations. Mean concentrations of urinary arsenic per gram of creatinine were 62.91 μg/L-g (27.54–186.21 μg/L-g) in the exposed population and 40.28 μg/L-g (18.20–70.79 μg/L-g) in the nonexposed population. Urinary arsenic also predicted subscales of verbal comprehension and long-term memory formation. Data from a small pilot study conducted in the United States show an inverse association between hair arsenic and IQ in adolescents (11–13 years old) (Wright et al. 2006) and suggest that arsenic toxicity may occur at exposures lower than those found in Bangladesh. The median concentration of arsenic in hair was 18 ppb.

Several large cross-sectional studies have been conducted in areas of endemic arsenic exposure. Wasserman et al. (2004) conducted a cross-sectional cohort study of 201 Bangladeshi 10-year-old children that demonstrated a strong dose-dependent adverse effect of increased drinking-water arsenic on IQ after adjustment for covariates. The mean arsenic concentration was 118 μg/L, and the range was wide, 0.1–790 ppb. Individual urinary arsenic concentrations did not significantly predict IQ (p = 0.09 for full-scale IQ), but the direction of the effect was the same as for arsenic in water. In a followup study conducted in Bangladesh by the same team in children whose water arsenic ranged from 0.1 to 464 μg/L, whole-blood arsenic in 8- to 11-year-old children was significantly negatively related to several Wechsler Intelligence Scale for Children, 4th Edition (WISC-IV) subscale scores, including verbal comprehension. Urinary arsenic (per gram of creatinine) was significantly negatively associated with verbal comprehension scores (Wasserman et al. 2011). In another large study, Von Ehrenstein et al. (2007) assessed 351 children in West Bengal and found inverse associations between urinary arsenic and vocabulary scores, object assembly, and picture completion. Reconstructed cumulative and peak exposures were estimated from drinking-water arsenic concentrations and were not predictive of test scores.

TABLE 2 Urinary Arsenic Concentrations in Studies of Neurotoxicity

^aTable 2 in source paper provides incorrect values of standard deviation, so this table uses the mean and the range.

Abbreviations: Cr, urinary creatinine; IQR, interquartile range.

Early-life exposure (during pregnancy and infancy) was not associated with infant development as assessed at the age of 7 or 18 months by the Bayley Scales of Infant Development in a large cohort study conducted in Bangladesh (Tofail et al. 2009; Hamadani et al. 2010). However, in a followup study of the same children (more than 1,700), inverse associations between maternal urinary arsenic in pregnancy and verbal/full-scale IQ (Wechsler Preschool and Primary Scale of Intelligence) were found at the age of 5 years of age but only in girls (Hamadani et al. 2011). Associations were stronger with the cross-sectional measures of urinary arsenic, again only in girls. Results of a study of mice support that sexually dimorphic finding: female pups were more vulnerable than males to the developmental effects of arsenic exposure with regard to locomotor activity and depletion of dopamine in the nucleus accumbens (Bardullas et al. 2009). Another study in Bangladesh indicated effects on motor function. Parvez et al. (2011) assessed motor function in 8- to 11-year-old children with the Bruininks-Oseretsky Test of Motor Proficiency and found an inverse association of total composite scores with water arsenic and with biomarkers of arsenic exposure. Whether the effects are peripheral or central nervous system effects is not clear.

With respect to behavioral studies, Roy et al. (2011) conducted a cross-sectional study of school-age children (median age 7 years) in Mexico and found that urinary DMA concentrations in boys were associated with higher scores scores on the oppositional, cognitive problems and ADHD subscales of the teacher ratings on the Conners Comprehensive Behavior Rating Scales, although adjustment for cognitive-test results diminished the association. That might suggest that the behavioral findings were driven largely by cognitive deficits.

Studies in Adults

The relationship between arsenic exposure and neurotoxicity in adults is less well studied than in children. A community-based participatory research study conducted in rural elderly adults found that GIS-based groundwater arsenic exposure was significantly related to poorer scores in language, visuospatial skills, and executive functioning. Long-term low-level exposure to arsenic was also associated with poorer scores in global cognition and memory (O’Bryant et al. 2011). No biomarkers were collected. The effects of arsenic in elderly populations is a particular research need.

Animal and in Vitro Experimental Studies

A study by Rodríguez et al. (2002) showed learning and behavioral deficits in rats exposed prenatally to arsenic via maternal drinking water. Increased spontaneous locomotor activity and increased errors in delayed alternation tasks (a test of memory and executive function) were found in arsenic-exposed offspring compared with nonexposed controls. Animals in the exposed groups consumed arsenic at 2.93–4.20 mg/kg per day, far more than any human study has reported except Dekeish et al. (2006) in Japan. (That report described the consequences of ingesting arsenic-contaminated infant formula, which led to a mass poisoning and reports of 130 fatalities. An IQ score of less than 85 was reported in 20.6% of survivors.)

Animal studies have also shown a dose–response relationship between arsenic in the brain and arsenic in drinking water—a demonstration that arsenic crosses the blood–brain barrier to the central nervous system (Rodríguez et al. 2002; Luo et al. 2009; Xi et al. 2009, 2010). In each of those animal studies, doses of arsenic were probably in the range of milligrams per kilogram per day, although not all studies reported water consumption rates. All studies involved concentrations of inorganic arsenic greater than 1,000 μg/L. Nagaraja and Desiraju (1994) found delayed acquisition and extinction of operant behaviors in rats exposed to sodium arsenate at 5 mg/kg per day in drinking water for 3 months. Increased concentrations of arsenic in cerebellar tissues have been associated with impaired performance on the Morris water-maze test (working and spatial memory) in mice (Y. Wang et al. 2009). Those deficits might be due to increased oxidative toxicity and changes in neurotransmission.

Increased neurotoxic oxidative stress (lowered concentrations of reduced glutathione and increased lipid peroxidation) in mouse brains after oral administration of arsenic trioxide was reported by Rao and Avani (2004) and Chaudhuri et al. (1999). An in vitro study demonstrated that neurite outgrowth was suppressed by sodium arsenite possibly because of induced neuronal apoptosis (Aung et al. 2013). Changes in concentrations of neurotransmitters—such as acetylcholine, dopamine, serotonin, and norepinephrine—have been found in the central nervous system after arsenic exposure (Nagaraja and Desiraju 1993, 1994; Chattopadhyay et al. 2002a,b). In studies using brain explants and neuronal cell cultures, neural networking and increased reactive oxygen intermediates were altered after exposure to arsenic (Chattopadhyay et al. 2002a). In rats exposed to arsenic during pregnancy, fetal brain neurons underwent apoptotic changes and neuronal necrosis (Chattopadhyay et al. 2002b). However, not all studies have demonstrated adverse effects of developmental exposure to inorganic arsenic, and null findings have been reported (Gandhi et al. 2012).

There is a growing literature on inorganic-arsenic toxicity that uses doses more commonly encountered in human exposure scenarios. The most commonly used drinking-water concentration is 50 μg/L, which is within the range seen in many epidemiologic studies. Martinez-Finley et al. (2008) focused on perinatal exposure to inorganic arsenic (defined as pregnancy through weaning at the age of 23 days). They reported that inorganic arsenic at 50 μg/L in drinking water was associated with increased depressive behaviors. (This period of exposure is parallel to a prenatal-exposure paradigm in humans, and preweaning mice are developmentally equivalent to a third-trimester human fetus.) Plasma corticosterone concentrations were about two-fold higher in the exposed group compared with controls. Hippocampal corticotropin-releasing-factor receptor binding was increased, and serotonin 5HT1A receptor binding was reduced. Overall, the results suggest that affected offspring might be predisposed to depressive-like behavior because of disrupted regulatory interactions between the hypothalamic–pituitary–adrenal axis and the serotonergic system in the dorsal hippocampal formation. In a followup study by the same group (Martinez-Finley et al. 2009), mice exposed to inorganic arsenic perinatally took longer to acknowledge a novel object and demonstrated deficits in spatial memory on an eight-arm radial-maze task. Arseni- treated animals had lower concentrations of glucocorticoid receptors in the brain coupled with higher plasma corticosterone; this suggests that the lower glucocorticoid receptor concentrations are a primary event that is followed by a compensatory increase in blood concentrations of corticosterone. A third paper by the group (Martinez-Finley et al. 2011) reported glucocorticoid-receptor–mediated transcriptional deficits in the mitogen-activated protein kinase/extracellular signal-related kinase pathway. That pathway plays a critical role in learning and memory formation, and consideration should be given to whether it could be an underlying cause of learning deficits associated with inorganic arsenic.

Key Considerations for the IRIS Assessment

A growing body of literature on both animals and humans suggests that low to moderation concentrations of arsenic are associated with neurologic deficits, particularly in IQ tests in children. Animal studies have demonstrated deficits in executive function, motor function, and spatial memory. Mechanisms appear related to oxidative-stress–induced apoptosis and effects on neurotransmitters. Human studies on arsenic neurotoxicity have been conducted primarily in children. Prenatal exposure did not predict performance on general developmental tests at the age of 18 months in Bangladeshi children but did predict performance at the age of 5 years; this either suggests a latent period for arsenic exposure in utero or suggests that subtle effects of arsenic exposure cannot be detected in the relatively insensitive psychometric tests used in toddlers and can be adequately assessed only at higher ages. Most studies have used IQ tests or other tests of general cognitive function and their subscales. No clear pattern has emerged with respect to nonverbal skills associated with inorganic arsenic exposure; however, deficits in verbal cognitive skills are the most common finding reported in epidemiologic studies (see Table 3).

TABLE 3 Most Sensitive Neurodevelopmental End Points in Human Studies

There is evidence of a dose–response relationship in animal studies, but the dose range used is orders of magnitude higher than what is typically assessed in human studies. A small body of literature that used 50 ppb during perinatal exposures has demonstrated measurable effects, but there has not been an attempt to delineate the dose–response relationship. Epidemiologic studies are not uniform in how they assess exposure. Studies have used a wide variety of exposure metrics, including concentrations in water, urine, hair, and blood. A compounding difficulty is that few studies are prospective and the dose–response relationship may depend on the timing and duration of exposure. One study that showed a clear dose–response relationship between water arsenic and IQ is the one by Wasserman et al. (2004) in Bangladesh. Lower full-scale and performance IQ scores were observed for each increase in the quartile of water arsenic. Exposure was similar to that in the United States in the lowest quartiles (less than 5.5 ppb and 6–50 ppb). Water arsenic concentrations of 10 ppb and 50 μg/L were associated with raw score losses of 3.8 and 6.4 points, respectively, but these are not equivalent to IQ points, because the WISC-III has not been standardized in Bangladesh. Results of studies of behavior and more domain-specific tasks (such as executive function and behavior) suggest possible adverse effects, but this aspect of neurotoxicity has been relatively understudied. Overall, more epidemiologic studies are needed to confirm animal findings. Motor function has also been understudied, as have the effects of arsenic exposure on cognitive function in adults, especially the elderly.

DIABETES

Several epidemiologic studies have evaluated the association between arsenic exposure and the risk of diabetes. In 2011, the National Toxicology Program (NTP) comprehensively evaluated the epidemiologic and experimental evidence on arsenic and diabetes end points (Maull et al. 2012). After considering consistency among populations, the strength and temporality of the associations, and biologic plausibility, the NTP review concluded that existing human data provided “limited to sufficient” support of an association between chronic exposure to arsenic at high concentrations (150 μg/L or more in drinking water) and diabetes. Below 150 μg/L, concentrations that are more relevant for arsenic risk assessment, the NTP review judged that the evidence was “insufficient” to conclude that arsenic was associated with diabetes because of the lack of prospective studies, limitations in exposure and outcome assessments, and lack of adjustment for relevant confounders.

Some studies, however, characterized by better measures of outcome and exposure were supportive of an association, including several cross-sectional studies in Mexico (Coronado-Gonzalez et al. 2007; Del Ra-

zo et al. 2011). For dose–response considerations, the Mexican studies included water arsenic concentrations of less than 10 μg/L to greater than 100 μg/L (geometric mean 24.4 μg/L) (del Razo et al. 2011). Another relevant study was conducted in pregnant women who lived in an area of Oklahoma in which at least 25% of drinking-water sources had arsenic concentrations greater than 10 μg/L (Ettinger et al. 2009). In that study, arsenic measured in blood and hair samples at delivery was associated with impaired glucose tolerance during pregnancy, but the association with hair arsenic was not statistically significant (Ettinger et al. 2009). Evidence from National Health and Nutrition Examination Survey (Navas-Acien et al. 2008, 2009a; Steinmaus et al. 2009), although directly relevant to the US population, was considered limited in the NTP review because of challenges in interpreting urinary arsenic biomarkers in the presence of seafood intake (Navas-Acien et al. 2011).

Since the publication of the NTP report (Maull et al. 2012), several studies have been published, including one cross-sectional study (Gribble et al. 2012) and two prospective studies (James et al. 2013; Kim et al. 2013) conducted in the United States. They were conducted in rural communities that had stable populations and low migration rates from Arizona, Colorado, Oklahoma, North Dakota, and South Dakota. Arsenic was measured at the individual level either in urine or in drinking water, diabetes was measured by using standardized outcome definitions, and the associations were adjusted for relevant risk factors, including body-mass index and sociodemographic factors. Table 4 provides the evidence from those studies as an illustration of important features of the studies that should be considered. Some recent studies have also reported that genetic variation in several genes (such as AS3MT, NOTCH2) could increase the risk of type 2 diabetes (Drobna et al 2013; Pan et al. 2013). Overall, recent epidemiologic studies support an association between arsenic at low to moderate concentrations (10–50 μg/L) and diabetes risk. They are directly relevant to evaluating a dose–response relationship for US populations and should be carefully reviewed by the IRIS program in evaluating diabetes as a relevant end point to be included quantitatively for arsenic risk assessment.

Mode of Action

An increasing number of experimental and mechanistic studies have evaluated the potential mode of action of arsenic in causing diabetes. The NTP review of the experimental evidence concluded that although as a whole the experimental literature was inconclusive, recent studies designed to evaluate diabetes-related end points generally supported a link between arsenic and diabetes (Maull et al. 2012). Diabetes could be related to the inhibition of insulin production by pancreatic β cells or the inhibition of basal or insulin-stimulated glucose uptake. Relevant mechanisms by which arsenic could affect β-cell function and insulin sensitivity include oxidative stress, glucose uptake and transport, gluconeogenesis, adipocyte differentiation, calcium signaling, and epigenetic effects (Diaz-Villasenor et al. 2007; Maull et al. 2012). In a genomewide study of peripheral blood lymphocytes in northern Mexico, moderate to high arsenic exposure was associated with hypermethylation and hypomethylation of several diabetes-related genes (Smeester et al. 2011). Additional coexposures, especially to a high-fat diet, could be particularly important for US populations: experimental evidence on rodents suggests that arsenic interacts with a high-fat diet to induce glucose intolerance without changes in plasma insulin concentrations (Paul et al. 2011).

Key Considerations for the IRIS Assessment

Results of recent epidemiologic evidence from studies conducted in Mexican and US populations, including prospective studies conducted in Arizona (Kim et al. 2013) and Colorado (James et al. 2013), support the association between low to moderate arsenic exposure and diabetes risk. The studies provide useful information for the IRIS assessment to use in evaluating the dose–response relationship at arsenic concentrations relevant for US populations. Mechanistic and animal evidence constitutes information that is relevant for evaluating the mode of action in arsenic-related diabetes. The NTP review, systematic reviews of the literature, and the possible conduct of dose–response meta-analyses should guide the inclusion of diabetes as a relevant health end point for quantitative arsenic risk assessment.

TABLE 4 Recent Studies of Diabetes and Arsenic Measured at the Individual Level That Reported Two or More Arsenic Categories^a

^aThe studies were conducted in populations with drinking water concentrations of arsenic less than 100 μg/L in the United States and were published after the NTP (2011) review.

Abbreviations: BMI, body-mass index; CC, case–control; CO, cohort; CS, cross-sectional; C/NC, cases/noncases; FPG, fasting plasma glucose; HbA1c, hemohemoglobin A1C; OGTT, oral glucose tolerance test.

EFFECTS ON LIVER, PROSTATE, AND PANCREAS

Liver

Arsenic is known to have potential carcinogenic and toxic effects on the liver (IARC 2004, 2012). Ecologic epidemiologic studies of populations in Taiwan and Chile have reported increased liver-cancer mortality in relation to arsenic exposure. In studies conducted in southwest Taiwan, Chen et al. (1986) reported an age- and sex-adjusted OR of 2.67 for liver-cancer mortality in those who had used well water that had high arsenic concentrations for 40 years or more compared with a relatively nonexposed area. In another study, they reported a dose–response relationship between arsenic concentrations in artesian well water and mortality associated with a number of cancers, including liver cancer in males, but the trend in females was not significant (Wu et al. 1989). In southwest Taiwan, in an area that had a median arsenic concentration in drinking water of 780 μg/L, Tsai et al. (1999) reported an increase in mortality due to liver cancer compared with local and national reference groups. Rivara et al. (1997) in a study in Antofagasta and Calama-Chuquicamata, Chile, determined that in an area that had high concentrations of arsenic in both the soil and water, some as high as 800 μg/L of drinking water, there was an increase in mortality (RR = 2.2) due to liver cancer in comparison with a region that had low concentrations of arsenic. Smith and co-workers (Smith et al. 1998; Liaw et al. 2008; Smith et al. 2012) also studied that population in Antofagasta, Chile. In their initial study (Smith et al. 1998), they observed increases in mortality associated with bladder and lung cancer but no increase in overall liver-cancer mortality from arsenic. In a later study, they found increased liver-cancer mortality in those who were exposed in utero and when young during peak exposure periods (before a change in water treatment). Specifically, they found higher childhood mortality from liver cancer than expected (RR = 10.6) (Liaw et al. 2008) and increased liver-cancer mortality in young adults (SMR = 2.5) (Smith et al. 2012), albeit on the basis of a small number of cases.

Not all epidemiologic studies have identified increases in liver cancer in relation to arsenic exposure. In a study of cancer mortality associated with arsenic in the drinking water in Cordoba, Argentina, although increases in lung and renal cancer incidence were found to be dose-dependent, mortality from liver cancer was similar in all three arsenic-exposed (low, medium, and high) groups (Hopenhayn-Rich et al. 1998). Guo (2003) used data from the Taiwan National Cancer Registry on over 40,000 patients in 243 townships who had liver cancer to determine the distribution of hepatocellular carcinoma and cholangiocarcinoma. No difference in the distribution of cancer cell types was found between the townships with endemic arsenic intoxication and the other townships. The study also found no association between arsenic and hepatocellular carcinoma. Baastrup et al. (2008) also found no increase in liver-cancer cases associated with arsenic in a Danish cohort study that had low drinking-water concentrations of arsenic (0.05–25.3 μg/L). However, none of those studies examined timing of exposure.

In studies of liver toxicity, Islam et al. (2011) measured arsenic concentrations in the hair and nails of 200 people in Bangladesh (which correlated with drinking-water concentrations) and found that indicators of liver toxicity (serum alkaline phosphatase, aspartate transaminase, and alanine transaminase) were significantly higher in the high-exposure group (over 50 μg/L in drinking water). Similarly, in a study of people in West Bengal, India, Das et al. (2012) found higher concentrations of bilirubin, alanine transaminase, aspartate transaminase, and antinuclear antibodies (an indicator of autoimmune status) in the serum of people who lived in an area that had an average arsenic concentration of 203 μg/L in drinking water than in serum of nonexposed people.

Arsenic and liver cancer have been studied in experimental animals. Waalkes et al. (2003, 2004a,b, 2006) found that male mice from dams exposed to sodium arsenite (0, 42.4, and 85 ppm) in drinking water on gestation days 8–18 had dose-dependent increases in hepatocellular-carcinoma incidence and multiplicity. An increase in hepatocellular carcinomas was also found in male and female mice exposed to arsenic (0, 6, 12, and 24 ppm in drinking water) in utero and over their lifetime (Tokar et al. 2011b).

Key Considerations for the IRIS Assessment

There is some epidemiologic evidence that liver cancer may be associated with arsenic exposure, in particular with exposure early in life, but the total body of work is not completely coherent. There is also evidence of hepatocyte damage, but overall the studies are not conclusive, especially with respect to low to moderate arsenic exposure. Thus, liver disease is unlikely to have high priority for the IRIS assessment.

Prostate

The prostate is among the tissues potentially associated with arsenic toxicity, including cancer (IARC 2004). In a number of studies of prostate and other cancers in Taiwan, increased prostate-cancer mortality was noted with a dose-related gradient of arsenic in drinking water from wells (Chen et al. 1985; Wu et al. 1989; Chen and Wang 1990; Yang et al. 2008). Notably, in a cohort study of mortality from malignant neoplasms in 314 precincts and townships, an age-adjusted increase in mortality from prostate cancer was associated with arsenic concentrations in well water (Chen and Wang 1990). The authors also found that mortality from prostate cancer declined when the contaminated water sources were replaced (Yang et al. 2008); this supports a causal link. Lewis et al. (1999) examined the relationship between arsenic in drinking water and prostate cancer in an ecologic analysis of a cohort of Mormon residents of Utah and found an SMR of 1.45 and support of a dose–response relationship. In an ecologic study in Canada that evaluated areas that had soil concentrations of arsenic greater than 100 mg/kg and water concentrations greater than 10 μg/L, Hinwood et al. (1999) found an increase in prostate cancer (standardized incidence ratio 1.14) that was the only cancer significantly increased after stratification into exposure categories. In American Indians in Arizona and North and South Dakota who participated in the Strong Heart Study, baseline urinary arsenic concentrations (median 9.7 μg/g of creatinine; interquartile range 5.8–15.6) were associated with increased prostate-cancer mortality over almost 20 years of followup (Garcia-Esquinas et al. in press).

However, associations with prostate cancer have not been consistently found in epidemiologic studies. Rivara et al. (1997) examined mortality from prostate cancer in residents of two regions in Chile, one of which was exposed to high concentrations of arsenic in drinking water (some as high as 800 μg/L), and found an RR of 0.9. Baastrup et al. (2008) found no increase in mortality from prostate cancer due to arsenic in a Danish drinking-water cohort study in which exposures ranged from 0.05 to 25.3 μg/L, although most participants were exposed at below 2 μg/L.

Mode of Action

A series of studies that used cell culture have also suggested that arsenic has an ability to transform normal cells into a cancer-like state. Achanzar et al. (2002) incubated human prostate epithelial cells (RWPE-1) in 5 μM arsenic and found a malignant transformation of the cells that produced epithelial-cell tumors when placed in nude mice. Benbrahim-Tallaa et al. (2005) found that that transformation was associated with aberrant genomic DNA methylation and K-ras oncogene activation at the same arsenic concentration. It was associated with Ras signaling activation (Benbrahim-Tallaa et al. 2007). Similarly, Tokar et al. (2010a) found that arsenic at 5 μM transformed the human prostate epithelial/progenitor cell line WPE-stem to a malignant cancer stem-cell–like phenotype. Singh et al. (2011) found that exposure of human prostate epithelial cells (RWPE-1) to arsenic caused nuclear DNA damage and mutations in mitochondrial DNA and resulted in increased cell survival when arsenic concentrations were as low as 8 pM. They also found (Treas et al. 2012) that arsenic in combination with estrogen altered epigenetic regulatory-gene expression, global DNA methylation, and histone modification in those cells.

Key Considerations for the IRIS Assessment

There is some, albeit modest epidemiologic evidence of an association between arsenic and prostate cancer, including two studies in the United States. In vitro studies of arsenic suggest that a change in phenotype may occur in isolated prostate cells exposed to arsenic. Given that evidence and the burden of prostate disease in US men and its associated health-care costs, the committee believes that the IRIS assessment of inorganic arsenic should at least consider the issue of prostate cancer.

Pancreas

Increased incidence of or mortality from pancreatic cancer in relation to arsenic-contaminated drinking water has not been reported in populations of Taiwan, Argentina, Chile, or Bangladesh that were exposed to high concentrations of arsenic (IARC 2004). However, increased pancreatic-cancer incidence was found in an ecologic comparison of children born before and after a 1955 episode of arsenic-contaminated milk powder (used for bottle-fed infants) in Japan (Yorifuji et al. 2010, 2011). Excess mortality from pancreatic cancer was observed in children younger than 5 years old during the contamination episode. The study was ecologic, and exposure occurred only during a limited period early in life, so an evaluation of a dose–response relationship was not possible. Recent studies at low to moderate exposure have also reported an association between arsenic and pancreatic cancer. In a hospital-based case–control study in Spain, the OR for pancreatic cancer when the highest and lowest quartiles of toenail arsenic concentrations were compared (>0.106 vs ≤0.052 μg/g) was 2.02 (95% CI 108–3.78) (Amaral et al. 2012). In American Indians in Arizona and North and South Dakota who participated in the Strong Heart Study, baseline urinary arsenic concentrations (median 9.7 μg/g of creatinine, interquartile range 5.8–15.6) were associated with increased pancreatic-cancer mortality over almost 20 years of followup (Garcia-Esquinas et al. in press). Although limited because of uncertain exposure assessment, GIS analysis in Florida also noted higher pancreatic-cancer incidences in relation to residential proximity to wells that had arsenic concentrations above 10 μg/L (Liu-Mares et al. 2013). Although the mechanism has not been characterized, one hypothesis is that arsenic increases the risk of diabetes (see section “Diabetes: above), which is a risk factor for pancreatic cancer.

Mode of Action

Studies of arsenic as a causative agent of pancreatic cancer are few. Results of studies of arsenic trioxide by Xue et al. (2009) indicated a dose-dependent effect on AR42J cells (rat pancreatic epithelial cell line). At low concentrations, it caused apoptosis; but at high concentrations (8 μM), it caused oncosis as identified by an increase in malignant changes in the cells detected with laser scanning confocal microscopy.

Key Considerations for the IRIS Assessment

Although there is emerging evidence of potential carcinogenic effects of arsenic on the pancreas, it may be difficult for EPA to establish risks on the basis of available evidence.

IMMUNE EFFECTS

In the hazard assessment of inorganic arsenic, it is important to consider effects on the immune system. The immune system is required to protect against infection. Dysregulation of the immune system may influence cancer and cardiovascular-disease risk. Both innate and adaptive immune responses should be considered. Epidemiologic, rodent, and in vitro data all support a relationship between inorganic arse-

nic exposure and immune system effects. The immunologic effects could include sensitization to exposure to viruses and bacteria; such effects have been seen in human populations exposed to arsenic.

It is also important to consider the relation of adverse effects on the immune system to periods of developmental susceptibility, when the thymus plays a substantial role. In early pregnancy, for example, the immune system changes toward Th2 cytokine response, which protects the fetus from being recognized as foreign (Calleja-Agius and Brincat 2008). As a result, Th1-dependent cytokine production is suppressed, and the child is born with Th2-biased immune responses. After birth, the immune system requires maturation of the Th1 cytokine response to achieve effective resistance to diseases (Dietert and Zelikoff 2008). That series of prenatal and postnatal events seems to be more sensitive to toxic insult than is the adult immune system (Dietert 2009). Suppression of theTh1-dependent function increases susceptibility to infections and reduces response to childhood vaccination (Heilmann et al. 2006; Vorderstrasse et al. 2006).

The hazard assessment of inorganic arsenic should include the following body of literature and integrate mode-of-action information where possible.

Epidemiologic Evidence

Results of several cross-sectional studies in both children and adults indicate that arsenic may be immunosuppressive. Results of recent studies indicate that arsenic is immunotoxic early in life when the thymus is an essential organ of the immune system. In a mother–child cohort in rural Bangladesh, thymus size, measured with ultrasonography, was inversely associated with maternal exposure to arsenic through drinking water, measured as the concentration of arsenic metabolites in urine in early and late pregnancy (range 5.5–1150 μg/L; n = 1,556) (Moore et al. 2009), which corresponded to similar concentrations in the drinking water (Vahter et al. 2006). Maternal urinary arsenic was also positively associated with diarrhea and acute respiratory infections in the infants (n = 1,552) (Rahman et al. 2011). The findings suggest that in utero arsenic exposure impaired child thymus development and increased morbidity, probably via immunosuppression. The results of those studies are supported by those of a recent study conducted in the United States in which low prenatal inorganic arsenic exposure was associated with an increased risk of respiratory disease, such as upper respiratory tract infections and colds (Farzan et al. 2013). With respect to mode of action, it is important to consider the relationship of dose and time dependence of exposure to key immunologic events.

In a followup of the Bangladeshi study (Ahmed et al. 2011), immune and inflammatory markers, measured with immunohistochemistry and multiplex cytokine assay, were assessed in 130 women. Significant positive associations between maternal urinary arsenic and placental markers of 8-oxoguanine (8-oxoG) and proinflammatory cytokines—such as interleukin-1beta (IL-1β), tumor necrosis factor-alpha (TNFα), and interferon-gamma (IFNγ)—were found. The 8-oxoG measured in the placenta was also positively associated with the placental proinflammatory cytokines. Urinary arsenic in early gestation was inversely associated with CD3+ T cells in the placenta.

In a study in Thailand, researchers examined the extent to which maternal arsenic exposure affects gene expression in the newborn by assessing cord-blood genomic signaling. The researchers monitored gene-expression profiles in a population of 32 newborns whose mothers experienced varied arsenic exposure during pregnancy. The investigators used genomewide gene-expression analyses to identify genes that were predictive of prenatal arsenic exposure in a later test population. Those genes could be reduced to a set of 11 transcripts that maintained the maximal predictive capacity to classify prenatal arsenic exposure. Analyzing the genes in the context of their protein–protein and protein–DNA interactions demonstrated that prenatal arsenic exposure was associated with genes involved in stress, inflammation, metal exposure, and apoptosis in the newborn (Fry et al. 2007). Many of those genes are key players in the immune response. Exposure in the study was determined on the basis of arsenic in toenail samples. A toenail-water comparison suggested that the cohort was exposed to high concentrations of inorganic arsenic as evidenced by toenail measurements of up to 68.63 μg/g. A dose–response relationship was observed

between arsenic in toenails and the expression-level changes in the identified set of genes. Moreover, the genes shared binding sites for transcription factors that are known to be metal-responsive (such as metal-responsive transcription factor 1).

Experimental Evidence

Researchers examined mRNA and protein-expression changes in the lungs of mice that were chronically exposed to arsenic in food or drinking water to evaluate whether immune modulation contributes arsenic-related disease risk in the lung (Kozul et al. 2009a). Groups of C57BL/6J mice were exposed to arsenic in drinking water or food at 10 or 100 ppb for 5–6 weeks. A whole-genome transcriptome profiling assay of the lungs revealed substantial alterations in the expression of genes involved in innate immune response and in cell adhesion and migration, channels, receptors, differentiation, and proliferation. The investigators further demonstrated that genes for IL-1β, IL-1 receptor, a number of toll-like receptors, and several cytokines and cytokine receptors were also altered (Kozul et al. 2009a).

Following up on that work, the researchers investigated the effects of arsenic exposure on respiratory influenza A (H1N1) virus infection. C57BL/6J mice were exposed to arsenic at 100 μg/L in drinking water for 5 weeks and were then infected with influenza A/PuertoRico/8/34 (H1N1) virus by intranasal inoculation. Increased morbidity was observed in the mice, and higher titers of influenza were found in whole-lung homogenates. Several alterations in immune response were also found. For example, a decrease in dendritic cells in the mediastinal lymph nodes was observed early in infection relative to nonexposed controls (Kozul et al. 2009b).

Acharya et al. (2010) investigated arsenic-induced carcinogenesis and effects on the immune system in a mouse model. Ethylnitrosourea was used to induce tumors in mice, and arsenic was used as a promoter. Effects on the immune system were assessed on the basis of cytokine (TNFα, IFNγ, IL-4, IL-6, IL-10, and IL-12) production of lymphocytes and the specific apoptotic cascade in lymphocytes. Arsenic, especially in combination with ethylnitrosourea, induced marked neoplastic changes. Those findings were supported in histologic studies and by evidence of severe immune suppression that resulted from cytokine modulation and lymphocyte death.

Key Considerations for the IRIS Assessment

Taken together, the evidence of a relationship between exposure to inorganic arsenic and altered immune function warrants consideration of the immune system in the IRIS assessment. The immunologic changes seem to increase the risk of respiratory symptoms in particular. Those effects have been observed after both lower and higher exposures to inorganic arsenic, although few studies are available on low to moderate exposure. It is conceivable that such effects may contribute to impairmet in fetal and infant health and to detrimental health effects in adults.

SUMMARY

To help EPA set priorities among its efforts, the committee has created the following hierarchy of health end points of concern:

•   Tier 1: Evidence of a causal association determined by other agencies and/or in published systematic reviews.

o   Lung, skin, and bladder cancer (Celik et al. 2008; IARC 2012).

o   Ischemic heart disease (Navas-Acien et al. 2005; C.H. Wang et al. 2007; Moon et al. 2012).

o   Skin lesions (ATSDR 2007).

•   Tier 2: Other priority outcomes.

o   Prostate and renal cancer.

o   Diabetes.

o   Nonmalignant respiratory disease.

o   Pregnancy outcomes (infant morbidity).

o   Neurodevelopmental toxicity.

o   Immune effects.

•   Tier 3: Other end points to consider.

o   Liver and pancreatic cancer.

o   Renal disease.

o   Hypertension.

o   Stroke.

o   Pregnancy outcomes (fetal loss, stillbirth, and neonatal mortality).

The first tier consists of end points that have been identified by other agencies or in systematic reviews as having a causal association with inorganic arsenic. The second tier consists of other high-priority outcomes of concern. The last tier consists of end points for which the evidence appears to be less strong. These categorizations will be refined by EPA after it conducts its more comprehensive analysis.

EPA’s draft plans indicate that a causal determination framework will be used to categorizethe evidence on the different end points into five possible categories: causal relationship, likely to be a causal relationship, suggestive of a causal relationship, inadequate to infer a causal relationship, and not likely to be a causal relationship. The committee supports that five-category approach and recommends that strength-of-evidence judgments be characterized with respect to the modified Bradford Hill criteria for causality. In evaluating the studies, it will be important to consider the timing of exposure with respect to life stage, duration of exposure, and the latent period for the health outcome. The assessment of causality will help EPA to set priorities among end points for later analysis of mode of action and dose–response relationships.