5

Susceptibility Factors

Several factors are known to influence susceptibility to the adverse effects of arsenic. This chapter addresses susceptibility factors discussed in the committee’s workshop by Beck (2013) and Hansen (2013)—life stage, genetics, nutrition, and pre-existing disease—and the additional factors of sex, smoking, alcohol consumption, and exposure to mixtures. The sections on these factors reflect a preliminary survey of the literature that the US Environmental Protection Agency (EPA) should consider more comprehensively and systematically as it conducts its toxicologic assessment of inorganic arsenic.

LIFE STAGES

Life stages refers to different ages and developmental stages throughout life, including prenatal development (pregnancy) and aging. There is increasing evidence that arsenic exposure early in life affects fetal and child health and development and health later in life. However, there seems to be a lack of data on the influence of aging.

Early-life development is a critical window of susceptibility for multiple toxic agents. The main focus has been on developmental neurotoxicity—recently reviewed by Bellinger (2013)—but there is increasing evidence that other organs and functions may be particularly susceptible during development, such as the immune system (Dietert 2011) and the lungs (Ramsey et al. 2013a). Thus, it is essential to evaluate whether early-life exposure to arsenic may affect the risk of the numerous arsenic-related health effects observed in adults and to consider this question in the health risk assessment.

Concerning prenatal arsenic exposure, it seems clear that all metabolites of inorganic arsenic easily cross the placenta to the fetus. Strong correlations between arsenic in maternal blood and in cord blood have been found in women exposed to arsenic-contaminated drinking water (Concha et al. 1998a; Jin et al. 2006; Hall et al. 2007). 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 survival (see section “Pregnancy Outcomes” in Chapter 4). In particular, the developing immune system and central nervous system seem to be susceptible to arsenic exposure, and adverse effects seem to appear even after relatively low arsenic exposure (see sections “Neurotoxicity” and “Immune Effects” in Chapter 4).

Besides the prenatal and perinatal stages, early childhood may be a period of susceptibility to inorganic arsenic exposure. A few epidemiologic studies have indicated that continued exposure after birth may impair children’s growth. A cross-sectional study in China suggested that increased water arsenic concentrations were inversely associated with body weight of children (720) 8–12 years old (S.X.Wang et al. 2007). In a prospective cohort study in Bangladesh, measures of 2,372 infants were related to concentrations of arsenic metabolites in the urine of mothers in early and late pregnancy and of the children at the age of 18 months (Saha et al. 2012). The study adjusted for age (within each age group), sex, maternal body-mass index, socioeconomic status, and birth weight or length. Observed inverse associations of maternal urinary arsenic with children's weight and length at the age of 3–24 months were markedly attenuated after adjustment for relevant covariates. However, the associations of urinary arsenic at the age of 18 months with weight and length at the age of 18–24 months were more robust, particularly in girls. The



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 51
5 Susceptibility Factors Several factors are known to influence susceptibility to the adverse effects of arsenic. This chapter addresses susceptibility factors discussed in the committee’s workshop by Beck (2013) and Hansen (2013)—life stage, genetics, nutrition, and pre-existing disease—and the additional factors of sex, smok- ing, alcohol consumption, and exposure to mixtures. The sections on these factors reflect a preliminary survey of the literature that the US Environmental Protection Agency (EPA) should consider more com- prehensively and systematically as it conducts its toxicologic assessment of inorganic arsenic. LIFE STAGES Life stages refers to different ages and developmental stages throughout life, including prenatal de- velopment (pregnancy) and aging. There is increasing evidence that arsenic exposure early in life affects fetal and child health and development and health later in life. However, there seems to be a lack of data on the influence of aging. Early-life development is a critical window of susceptibility for multiple toxic agents. The main fo- cus has been on developmental neurotoxicity—recently reviewed by Bellinger (2013)—but there is in- creasing evidence that other organs and functions may be particularly susceptible during development, such as the immune system (Dietert 2011) and the lungs (Ramsey et al. 2013a). Thus, it is essential to evaluate whether early-life exposure to arsenic may affect the risk of the numerous arsenic-related health effects observed in adults and to consider this question in the health risk assessment. Concerning prenatal arsenic exposure, it seems clear that all metabolites of inorganic arsenic easily cross the placenta to the fetus. Strong correlations between arsenic in maternal blood and in cord blood have been found in women exposed to arsenic-contaminated drinking water (Concha et al. 1998a; Jin et al. 2006; Hall et al. 2007). Epidemiologic studies, including prospective cohort studies, have provided evidence that arsenic exposure through drinking water during pregnancy may cause dose-dependent im- pairment of fetal and infant growth and survival (see section “Pregnancy Outcomes” in Chapter 4). In par- ticular, the developing immune system and central nervous system seem to be susceptible to arsenic expo- sure, and adverse effects seem to appear even after relatively low arsenic exposure (see sections “Neurotoxicity” and “Immune Effects” in Chapter 4). Besides the prenatal and perinatal stages, early childhood may be a period of susceptibility to inor- ganic arsenic exposure. A few epidemiologic studies have indicated that continued exposure after birth may impair children’s growth. A cross-sectional study in China suggested that increased water arsenic concentrations were inversely associated with body weight of children (720) 8–12 years old (S.X.Wang et al. 2007). In a prospective cohort study in Bangladesh, measures of 2,372 infants were related to concen- trations of arsenic metabolites in the urine of mothers in early and late pregnancy and of the children at the age of 18 months (Saha et al. 2012). The study adjusted for age (within each age group), sex, maternal body-mass index, socioeconomic status, and birth weight or length. Observed inverse associations of ma- ternal urinary arsenic with children's weight and length at the age of 3–24 months were markedly attenu- ated after adjustment for relevant covariates. However, the associations of urinary arsenic at the age of 18 months with weight and length at the age of 18–24 months were more robust, particularly in girls. The 51

OCR for page 51
52 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic effects seemed to appear after low arsenic exposure. Compared with girls in the first quintile of urinary arsenic (less than 16 µg/L, adjusted for specific gravity), those in the fourth quintile (26–46 µg/L) were almost 300 g lighter and 0.7 cm shorter. In a followup of 1,505 mother–infant pairs on whom there were data on concentrations of arsenic, cadmium, and lead in maternal and child urine, the effects of the com- bined exposure to these metals on children’s weights and heights up to the age of 5 years were evaluated (Gardner et al. 2013). The investigators adjusted for family socioeconomic status, maternal tobacco chew- ing during pregnancy, cooking with indoor fires, maternal education, season of birth, parity, and exposure to cadmium and lead. In the longitudinal analysis, the multivariable-adjusted attributable difference in children’s weight at the age of 5 years was -0.33 kg (95% CI -0.60 to -0.06) for high arsenic exposure (95th percentile or higher) compared with the lowest exposure (5th percentile or lower). The correspond- ing multivariable-adjusted attributable difference in height was -0.50 cm (95% CI -1.20 to 0.21). As in the earlier study, the associations were apparent primarily in girls Those findings are supported by experimental studies in which mice were exposed to very low con- centrations of arsenic (10 µg/L in drinking water) prenatally via the dams’ drinking water or postnatally (Kozul-Horvath et al. 2012). Birth outcomes, including litter weight and number of pups, were unaffect- ed, but exposure in utero and postnatally resulted in impaired growth of the offspring. The growth deficits resolved afer cessation of exposure in male mice but not in female mice up to the age of 6 weeks. Adverse effects of arsenic on child health, especially increased risk of infectious diseases and signs of immunosuppression, have been reported in several cross-sectional epidemiologic studies (see section “Immune Effects” in Chapter 4). Whether the underlying effect was initiated before birth or during child- hood is not clear. Similarly, impaired cognitive function has been reported in several epidemiologic stud- ies and indicates that children, especially of preschool age, are susceptible to arsenic-induced neurotoxici- ty (see section “Neurotoxicity” in Chapter 4). The increasing evidence that early-life exposure to inorganic arsenic increases the risk of adverse health effects later in life was discussed in the committee’s workshop by Waalkes (2013). Important evi- dence comes from the studies in northern Chile, showing considerably increased risk of cancer and non- cancer effects in young adults who were born in Antofagasta during or shortly before the period (1958– 1970) of high arsenic concentrations in drinking water (Smith et al. 2006; Yuan et al. 2007; Dauphine et al. 2011; Smith et al. 2012). In particular, the risk of death from chronic bronchiectasis was higher in those who had been exposed in utero and during early childhood and higher in those who had been ex- posed in utero compared with the rest of Chile (Smith et al. 2006) (see section “Respiratory Effects” above). The increased susceptibility to respiratory effects of early-life exposure has support from experi- mental studies that indicated impairment of lung development caused by in utero exposure to arsenic (Ramsey et al. 2013a,b). Recent epidemiologic studies of arsenic-related developmental immunotoxicity have also indicated long-term consequences for susceptibility, for example, to infections (Moore et al. 2009; Ahmed et al. 2012). The human evidence of later-life effects of low-dose early-life arsenic exposure is supported by a se- ries of experimental studies in mice that were given drinking water that contained arsenic at 50 µg/L (mainly arsenate) during gestation (or before mating) and thereafter until weaning. That exposure regimen resulted in marked dysregulation of the hypothalamic–pituitary–adrenal axis in male mice at the age of 35 days (Goggin et al. 2012), interactions with the glucocorticoid receptor and associated signaling pathway (Martinez et al. 2008; Martinez-Finley et al. 2011), and impaired learning and memory (Martinez-Finley et al. 2009). Perhaps the most striking animal studies are those of Waalkes and co-workers, who demonstrated that arsenic exposure in utero, with no additional postnatal exposure, leads to the development of liver, lung, ovary, and adrenal tumors in rats (Waalkes et al. 2003). They have gone on to show that prenatal arsenic exposure increases the carcinogenicity of other agents in the offspring when they reach adulthood (Tokar et al. 2011a). A few recent studies have indicated that maternal arsenic exposure during pregnancy may influence epigenetic markers, specifically those related to DNA methylation, in DNA isolated from the blood cells of newborn children. Two studies in Bangladesh of 113 and 101 mother–newborn pairs reported positive

OCR for page 51
Susceptibility Factors 53 associations between maternal arsenic exposure and global methylation (methyl-incorporation or LINE-1 assays) in DNA isolated from cord blood (Kile et al. 2012; Pilsner et al. 2012), whereas no significant difference in in cord blood LINE-1 methylation was observed between arsenic-exposed (55) and - unexposed (16) newborns (as assessed by arsenic in hair and nails) in a cross-sectional study in Thailand (Intarasunanont et al. 2012). The study by Pilsner et al. (2012), which also included the Alu and LUMA assays, indicated that the effects on DNA methylation were largely sex-specific, with arsenic-related in- creased methylation in newborn boys but decreased methylation in girls. Evaluation of methylation in cord-blood DNA of 134 infants in a prospective birth cohort in New Hampshire using the Illumina Infini- um Methylation 450K array found evidence of differential patterns of DNA methylation in relation to fairly low maternal urinary arsenic concentrations in late pregnancy (median 4.1 µg/L, interquartile range 1.8–6.6 µg/L, maximum value about 300 µg/L) (Koestler et al. 2013). Among the 100 loci that had the strongest association with arsenic, those in CpG promoter islands (44) showed mostly (75%) higher methylation in the highest-exposed group (>6.6 µg/L) than in the lowest-exposed group. However, cg08884395 (associated with estrogen receptor 1, ESR1) and cg27514608 (peroxisome proliferator– activated receptor-γ coactivator 1-α [PPARGC1A]) showed inverse associations between arsenic expo- sure and methylation and had statistically significant trends. Although changes in DNA methylation have been observed and associated with arsenic exposure, their biologic effects and meaning at a functional level in the cell and as related to health effects are unknown. Key Considerations for the IRIS Assessment Collectively, the highly suggestive evidence that early-life exposure to arsenic, even at low concen- trations, increases the risk of adverse health effects and impaired development in infancy and childhood and later in life leads the committee to suggest that the timing of exposure, particularly early-life expo- sure, be considered in evaluating epidemiologic studies for dose–response assessment. The increasing body of data on mechanisms, including hormone interactions and epigenetic alterations, in support of those effects warrants consideration. GENETICS OF ARSENIC METABOLISM AND TOXICITY Genetic factors probably confer susceptibility or resistance to inorganic arsenic exposure. Genetic variants may have biologic effects of their own that act synergistically with inorganic arsenic or may have normal function in the absence of inorganic arsenic. Genetics may interact with inorganic arsenic ion mul- tiple ways. Three possible interactions are as follows:  Genetic variants may increase or decrease the absorbed dose of inorganic arsenic or alter the ex- cretion of inorganic arsenic or its metabolites. The committee knows of no examples of this type of sus- ceptibility.  Genetic variants may alter the metabolism of inorganic arsenic independently of excretion or ab- sorption. Most of the arsenic literature deals with this type of susceptibility.  Genetic variants may increase or decrease the organ or cellular toxicity of inorganic arsenic. There is a small body of literature on this type of susceptibility. As described below, there are many examples in the literature regarding genetic susceptibility fac- tors, but their role in risk assessment is still unclear; the prevalence of these factors will vary in different populations that have different underlying ethnicities. Furthermore, the function of the genetic variants will probably vary with dose and may not be relevant in some exposure scenarios. Finally, most of the published reports are of case–control studies that provide few or no data on timing of exposure. An often overlooked underlying aspect of genetic susceptibility to environmental factors is the timing of exposure.

OCR for page 51
54 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic Regardless of the type of environmental factor (for example, chemical, social, or nutritional), the timing of exposure probably plays a critical role in whether a genetic factor will interact with it. Critical Developmental Windows and Gene–Environment Interactions The field of gene–environment interactions is surprisingly underdeveloped. Most environmental health studies of gene–environment interactions have used a small-scale candidate-gene approach and have not planned for replication in independent populations. Underlying the problem is an often neglected aspect of genetic epidemiology studies: gene expression changes from one life stage to another. Genetic variants that produce gene–environment interactions might do so only when the exposure corresponds to a critical developmental window during which the gene is highly expressed (Wright and Christiani 2010). Most gene–environment interaction studies are case–control studies and cannot address critical develop- mental windows. Genomewide Approach vs Candidate-Gene Approach Most human studies of genetic variation and arsenic have focused on a few candidate single nucleo- tide polymorphisms (SNPs) that may modify arsenic toxicity or predict its metabolism. Although such an approach has strengths, including biologic plausibility and clear a priori hypotheses, there are limitations in selecting only a few SNPs for a study (Rebbeck et al. 2004). The potential functional effect of many polymorphisms is sometimes undefined or even controversial. In addition, it is unlikely that one SNP of one gene can account fully for the complexity of function of an entire biologic pathway. Finally, some genes or SNPs that have important roles in arsenic pathogenesis may not have been identified yet, and a candidate approach would not be able to identify such genes. Thus, bias is a concern in the selection of SNPs in a candidate approach (Wright and Christiani 2010). An alternative approach is to use genomewide scans, which capitalize on advances in high- throughput technology and on the data from the completed HapMap Project (Hirschhorn 2005; Hirschorn and Daly 2005). Genomewide scans allow screening of the genome in an unbiased manner with respect to genetic risks factors (Kronenberg 2008). The greatest strength of such an approach is that it allows new biologic relationships to be discovered. Its primary weakness is that well-known biologic associations are ignored. When genetic associations, or gene–environment interactions, are simply ranked, a gene that may have greater biologic plausibility a priori is considered equal to all other genes—even genes that may not be expressed in the target tissue (Wright and Christiani 2010). To date, genomewide association stud- ies have been conducted to identify genetic risk factors for diseases as varied as age-related macular de- generation (Haines et al. 2006), cardiac diseases (Cupples et al. 2007; Vasan et al. 2007), diabetes (Meigs et al. 2007; Wellcome Trust Case Control Consortium 2007), amyotrophic lateral sclerosis (van Es et al. 2007), rheumatoid arthritis (Plenge et al. 2007; Thomson et al. 2007; Wellcome Trust Case Control Con- sortium 2007), and cancer (Broderick et al. 2007; Murabito et al. 2007; Spinola et al. 2007; Zanke et al. 2007). Whole-genome sequencing is an approach that has the advantage of discovering rare variants that may modify arsenic toxicity. Overall, the field of genomics has undergone a paradigm shift from a hy- pothesis-driven approach to a hypothesis-generating approach. Because genomic approaches generate multiple comparisons on a grand scale, tests of statistical significance need to take into account that genes are not necessarily expressed independently of each other and still control for overall experimentwide error. One procedure is a correction for multiple comparisons, such as Bonferroni correction or control of false discovery rates. Such methods are purely statistical, however, and ignore biology. Another approach is to conduct the research in multiple populations sequentially. Researchers generate hypotheses in one population and then test them in separate populations as a second-line screening and validation or replica- tion method (Rodriguez-Murillo and Greenberg 2008). At least one whole-genome interaction study of inorganic arsenic has been conducted (Pierce et al. 2012). The region proximal to the arsenic 3+ methyl- transferase enzyme (AS3MT) gene was identified as containing variants associated with the percentage of

OCR for page 51
Susceptibility Factors 55 monomethyl arsenic (MMA) and dimethyl arsenic (DMA) metabolites and is associated with the risk of a disease outcome (skin lesions). Replication in an independent population is still needed, however. Recently, a linkage analysis was conducted in the Strong Family Heart Study, a family-based study of mostly American Indian families in Arizona and Oklahoma (Tellez-Plaza et al. 2013). Urinary arsenic and its metabolites were considered as quantitative traits. Linkage studies differ from association studies (typically of case–control design) in studying family members and estimated heritability (the percentage of variance in urinary arsenic metabolites due to genetics). Regions in the genome that have suggestive logarithm of the odds of disease (LOD) scores (over 1.9) were described on chromosomes 5, 9, and 11. AS3MT is found on chromosome 10, and a locus that had an LOD score of 1.8 was found proximal to this site. Like genomewide-association studies, linkage studies identify genomic regions of interest, not specific genes. Further study, typically deep sequencing and the identification of functional variants, is needed before any genes can be considered as susceptibility factors. Finally, in any study of genetic con- tributions to arsenic metabolism, estimates of heritability are probably influenced by the degree of expo- sure in the underlying populations (that is, in the absence of arsenic exposure, a genetic trait cannot exhib- it heritability), and this influence may explain the modest LOD score for the region proximal to AS3MT. Studies of Genetic Susceptibility to Arsenic Arsenic Metabolism: Role of Methylation The metabolism of inorganic arsenic is critical for its toxicity and has been studied extensively in humans and animals. Two processes are involved: reduction and oxidation reactions that interconvert ar- senate and arsenite and methylation reactions, which convert arsenite to MMA and DMA (Tam et al. 1979; Buchet et al. 1981a,b; Marcus and Rispin 1988). Inorganic arsenic is methylated, mainly in the liv- er, to form monomethylated and dimethylated metabolites (Vahter and Marafante 1987; Hopenhayn-Rich et al. 1993). There is also evidence of genetically determined differences in arsenic methylation (Vahter et al. 1995b; Engström et al. 2013; Harari et al. 2013; Schlebusch et al. 2013). The biomethylation process involves both the arsenic 3+ methyl transferase gene and and the glutathione S-transferase (GST) enzyme systems (particularly the omega class) (Styblo et al. 1995). Further upstream, several genes related to one- carbon metabolism (such as folate metabolism genes, which provide methyl groups as substrate for me- thyl transferase enzymes) are also involved. Once arsenic is absorbed into the blood, it undergoes reduc- tion to arsenite and then methylation to yield MMA and DMA, which, in their pentavalent forms, are more readily excreted in urine. Studies have shown that animal species and individuals vary markedly with respect to arsenic methylation efficiency and that efficient methylation of inorganic arsenic is associ- ated with a high rate of excretion (Vahter 2002). Gamble et al. (2005) reported associations between one- carbon donor status and urinary methylated arsenic species in Bangladeshi adults. The percentage of DMA in urine was positively associated with plasma folate, although the percentages of inorganic arsenic and MMA were negatively associated with folate. In a followup study, folate supplementation led to en- hanced arsenic methylation and excretion and lower blood arsenic concentrations (Gamble et al. 2007). Glutathione S-transferase Glutathione could be a necessary catalyst for arsenic methylation through its role in the reduction of arsenate to arsenite. The GST-omega isozyme family is critical in this reaction although purine nucleoside phosphorylase can also reduce arsenate. There are eight classes of GST in mammalian cells, each of which may have subclasses of isozymes. These enzymes catalyze the nucleophilic attack on electrophilic chemicals, including xenobiotics, by glutathione. Various studies have addressed polymorphisms in GST isozymes and either arsenic speciation or toxicity. Chung et al. (2013) found an association between GST null variant (GSTM1 MspI) with increased risk of urothelial cancer. A GST-omega nonsynonymous SNP (Ala140Asp) modified the association between hair arsenic and metabolic syndrome (Wang et al. 2007).

OCR for page 51
56 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic Lesseur at al. (2012) assessed polymorphisms in five GST isozymes (P, M, O, T, and Z). A coding SNP in GST P1 (Ile105Val) modified the association of bladder cancer with toenail arsenic concentrations. Main effects were also noted in the risk of bladder cancer and GSTO2 SNP Asn142Asp and GSTZ1 SNP Glu32Lys. In a Vietnamese population, GST P1 Ile105Val was associated with lower MMA; this sug- gested a protective effect of the SNP (Agusa et al. 2012). Such a finding would be counterintuitive given the study by Lesseur et al. (2012) of this SNP with respect to toenail arsenic and bladder cancer; however, Lesseur et al. did not report arsenic speciation. L.I. Hsu et al. (2011) reported that the GSTT1 null variant modified the association between cumulative arsenic estimated by water concentrations and bladder can- cer. Polymorphisms in other GST genes (GSTO1, GSTO2, GSTP1, and GSTM1) did not modify that as- sociation. Similarly, Hsieh et al. (2011) did not find an association or evidence of effect modification be- tween arsenic exposure and GSTO1 and O2 polymorphisms and carotid atherosclerosis but did find evidence of effect modification between a haplotype of purine nucleoside phosphoralase SNPs and well- water arsenic. Finally, Chung et al. (2011) reported that subjects who had the GSTO2 Ala140Asp SNP had lower proportions of MMA in urine than subjects who were homozygous for the Ala allele; this sug- gests a protective effect of the SNP. That SNP was also assocated with a reduced risk of urothelial carci- noma. Collectively, those studies suggest that the genetic variability of the GST enzymes may influence susceptibility to arsenic toxicity, but the ability to use this information in the IRIS assessment is not clear. Arsenic 3+ Methyl Transferase The enzyme AS3MT catalyzes the methylation of arsenite by using S-adenosyl methionine as the methyl donor. The AS3MT gene has been the target of several candidate-gene studies of genetic suscepti- bility to arsenic. A nonsynonymous SNP M287T and an intronic SNP in the AS3MT gene were both as- sociated with reduce methylated arsenic species in urine in American Indians (Gomez-Rubio et al. 2012). Lower levels of methylation of arsenic also predicted higher risk of metabolic syndrome in women in that study. Beebe-Dimmer et al. (2012) did not find an association between AS3MT SNPs and bladder cancer, but they found evidence of an interaction between arsenic in drinking water and the Met287Thr SNP with an odds ratio of 1.17 per 1-µg/L increase in water arsenic concentration restricted to people who had at least 1 copy of the Thr allele. The previously described study by Lesseur et al. (2012) did not find evi- dence that interactions between AS3MT SNPs and arsenic exposure affected bladder-cancer risk. Agusa et al. (2011) found relationships between urinary arsenic methylation and two noncoding SNPs. Engström et al. (2011) genotyped a suite of SNPs in the AS3MT gene in Bangladesh and Argentinian Andean popu- lations that had high water arsenic exposures. They found a haplotype (series of SNPs on the same allele) that was associated with lower urinary percentage of MMA and one that was associated with higher uri- nary percentage of MMA in both populations. The haplotype with the higher percentage of MMA was the major allele in Bangladesh (52% allele frequency), whereas the haplotype associated with the lower per- centage of MMA was the major allele in Argentina (70% allele frequency). In addition, four polymor- phisms in DNA methylation enzymes DNMT1a and DNMT3b were associated with a higher urinary per- centage MMA in the pregnant women in Bangladesh. Genomic Screens At least two larger-scale genomic screens have been conducted in populations characterized for ar- senic exposure. These studies differ from the previously described linkage study in that they were not conducted in related people and used SNPs rather than short tandem-repeat markers to tag genomic re- gions of interest. Karagas et al. (2012) screened 10,000 nonsynonymous SNPs in a subset of subjects in a case–control study of bladder cancer conducted in New Hampshire (832 cases and 1,191 controls). The top hits were then validated in the larger case–control study. Variants in the FSIP1 gene (for a fibrous sheath interacting protein) and the SLC39A2 gene (for solute carrier protein 39A2, a metal transporter) modified the association between arsenic in drinking water and bladder cancer.

OCR for page 51
Susceptibility Factors 57 In a genomewide genotyping analysis conducted in 1,313 Bangladeshi subjects, a significant associ- ation between arsenic metabolism and toxicity phenotypes and five SNPs proximal to the AS3MT gene was found (Pierce et al. 2012). An expression array conducted in a subset of 950 subjects demonstrated increased AS3MT expression associated with the SNPs, and this suggests that these are cis-quantitative trait loci. Finally, one of the SNPs modified the association between urinary arsenic and premalignant skin disorders in a case–control study in Bangladesh (Pierce et al. 2012). Key Considerations for the IRIS Assessment A number of studies have demonstrated that genetic factors can modify the response to inorganic ar- senic. Most have been candidate-gene studies that focused on arsenic-metabolism genes. A small number have also addressed whether these genes modify the relationship between inorganic arsenic exposure and health effects but only for cancer, metabolic syndrome, cardiovascular disease, and skin lesions. There has been one genomewide scan (of 300,000 SNPs) in a population characterized for inorganic arsenic ex- posure, and it validated previous research on AS3MT gene variants as potential risk factors (Pierce et al. 2012). A family-based linkage study provided moderate evidence that the same region modifies arsenic metabolism (Tellez-Plaza et al. 2013). Although results have not been validated in independent popula- tions and deep sequencing of the region has not been performed, this information, when considered with the results of candidate-gene studies of AS3MT, suggests that this gene is a modifier of inorganic arsenic metabolism. Nonetheless, further work is needed to characterize genetic susceptibility to arsenic. If the methods for incorporating genetics into risk assessment are sufficient at this stage, AS3MT would be a logical choice for incorporation. Candidate-gene studies have identified genetic variants that modify arse- nic toxicity in base-repair genes (Applebaum et al. 2007; Breton et al. 2007; Ebert et al. 2011). These studies were specific to cancer end points. Overall, genetic risk factors for inorganic arsenic toxicity are still not fully understood, and for most genes the genetic risk probably varies too greatly with inorganic arsenic dose, age at exposure, and ethnicity of the underlying population to allow reasonable estimates of increased or decreased risk. SEX DIFFERENCES IN ARSENIC METABOLISM AND HEALTH EFFECTS There is evidence of marked sex differences in the metabolism and toxicity of arsenic. It is essential to evaluate such differences in the IRIS assessment to protect the most susceptible people in the popula- tion. Evaluation of sex differences may also provide information on mechanisms and modes of action of arsenic. It is well documented that the biotransformation of arsenic differs by sex. Other factors that may in- fluence arsenic metabolism, to be considered in evaluating the sex-dependent metabolism, include magni- tude of exposure, age, pregnancy, liver diseases, and smoking (Vahter 2002, 2009). In general, women excrete a greater percentage of DMA and a lower percentage of MMA in urine than do men after similar exposures (Hopenhayn-Rich et al. 1996; Hsueh et al. 2003;Lindberg et al. 2007, 2008a; Tellez-Plaza et al. 2013). That is probably related to women’s having more efficient one-carbon metabolism than men, espe- cially a potential of remethylating homocysteine via the choline–betaine pathway (Zeisel 2011). In addi- tion, arsenic methylation efficiency increases during pregnancy (Concha et al. 1998a; Gardner et al. 2011). Results of a number of studies have indicated that men are more affected by arsenic-related skin ef- fects, including skin cancer, than women (Watanabe et al. 2001; Kadono et al. 2002; Chen et al. 2003; Rahman et al. 2006; Ahsan et al. 2007; Lindberg et al. 2008b; Leonardi et al. 2012). That does not seem to be the case for all arsenic-related cancers, however, and should be studied in more detail. As discussed in the committee’s workshop by Cantor (2013), the indicated differences might be influenced by the background prevalence of the end point under study, and epidemiologic studies have rarely evaluated the differences in relative or absolute risk of arsenic-related health effects. Recent data indicate that girls may

OCR for page 51
58 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic be at higher risk for developmental effects after early-life arsenic exposure (Hamadani et al. 2011; Saha et al. 2012; Gardner et al. 2013) and boys after prenatal exposure (Kippler et al. 2012). There may be several mechanisms of the observed sex difference in arsenic toxicity. It may be me- diated at least partly by a difference in arsenic metabolism, inasmuch as the reduced form of the monomethylated metabolite, MMA(III), is highly toxic. Efficient methylation to DMA increases the ex- cretion of arsenic from the body (Vahter 2002). Arsenic-related skin lesions, for example, are influenced by methylation efficiency in a sex-dependent manner (Lindberg et al. 2008b). There may also be sex dif- ferences in toxicodynamics. For example, when pregnant mice were exposed to arsenic in drinking water at 42 or 85 mg/L on gestation days 8–18, female offspring developed ovarian and lung tumors and uterine and oviduct hyperplasia whereas male offspring had a highly increased incidence of liver and adrenal tu- mors later in life (Waalkes et al. 2003). Liver tumors were induced in female offspring only when the in utero arsenic exposure was combined with skin application of a tumor-promoting phorbol ester, 12-O- tetradecanoylphorbol-13-acetate (Waalkes et al. 2004a; Liu et al. 2006; Tokar et al. 2010b). Gene- expression analysis of the liver of male mice with hepatocellular carcinoma induced by exposure to arse- nic in utero showed overexpression of ER-α and cyclin D1 and a feminized expression pattern of several cytochrome P450 genes (Waalkes et al. 2004b). In addition, hepatic DNA from male offspring showed a significant reduction in methylation in GC-rich regions (Xie et al. 2007). Interaction of arsenic with ER-α and estrogen-associated functions has previously been reported (Lopez et al. 1990: Chattopadhyay et al. 1999; Stoica et al. 2000; Chen et al. 2002; Du et al. 2012; Treas et al. 2012). Mice that were exposed postnatally showed fewer sex differences in cancer development; arsenic increased lung adenocarcinoma and hepatocellular carcinoma in both sexes, but gallbladder tumors occurred only in males (Tokar et al. 2011b). In studies of mice exposed to more environmentally relevant doses of arsenic (0.05, 0.5, 5.0, or 50 mg/L in drinking water for 4 months), monthly assessment of locomotor activity showed that female mice in all treatment groups exhibited hyperactivity at every test. In contrast, male mice exhibited hyperactivity in the group exposed at 0.5 mg/L and hypoactivity in the highest-dose group after 4 months of exposure (Bardullas et al. 2009). Key Considerations for the IRIS Assessment There is clear evidence of sex differences in the metabolism of inorganic arsenic. Because arsenic metabolism is a recognized susceptibility factor, it seems likely that the toxicity of arsenic could also dif- fer between men and women. However, only a few arsenic-related health outcomes have been evaluated by sex. It is essential to evaluate sex differences in the IRIS assessment to provide protection for the most susceptible people in the population. NUTRITIONAL DEFICIENCIES As discussed in the committee’s workshop by Beck (2013), there is considerable evidence from stud- ies in animal models and in humans that nutritional factors may influence arsenic metabolism and toxicity. For example, studies of animals fed diets deficient in methionine, choline, or proteins have demonstrated decreased methylation and increased tissue retention of arsenic (Vahter and Marafante 1987). In West Ben- gal, India, poor nutritional status (low body weight) was associated with an increased risk of skin lesions (Mazumder et al. 1998). Similarly, in Bangladesh, lower body-mass index has been associated with an in- creased risk of arsenicosis skin lesions (Milton et al. 2004; Ahsan et al. 2006). In addition, using a validated dietary food-frequency survey in the large prospective Bangladesh Health Effects of Arsenic Longitudinal Study, Heck et al. (2009b) found that higher intakes of protein, methionine, and cysteine were associated with 10-15% greater urinary arsenic excretion after controlling for numerous covariates. Those are but a few examples of the extensive literature identifying poor nutrition as a susceptibility factor in arsenic toxicity.

OCR for page 51
Susceptibility Factors 59 Essentially, the most important dietary factors aside from protein and amino acid intake appear to fall into two categories—vitamins and nutrients that are involved in one-carbon metabolism and the syn- thesis of SAM, the universal methyl donor; and selenium, an element whose antagonism of arsenic toxici- ty has long been known (Levander 1977; Zheng et al. 2005). Nutritional Effects on One-Carbon Metabolism Nutritional effects on one-carbon metabolism on arsenic metabolism and toxicity were discussed in the committee’s workshop by Gamble (2013). The synthesis of SAM, which is required for each of the two methylation steps of arsenic metabolism, is influenced by many nutrients that directly or indirectly contribute methyl groups to the one-carbon metabolic pathway (such as folate, choline, and betaine) or serve as cofactors for enzymes in the pathway (such as vitamins B2, B6, and B12) (reviewed in Hall and Gamble 2012). In particular, deficiencies of folate and the B vitamins appear to exacerbate the associa- tions between inorganic arsenic exposure and its health effects in humans. For example, in a population- based cross-sectional study in Bangladesh, the association between arsenic exposure and hypertension was stronger in participants who had lower than average dietary intakes of folate and other B vitamins (Y. Chen et al. 2007b). A nested case–control study of incident skin-lesion cases and matched controls (Pils- ner et al. 2009) in Bangladesh and a case–control study in India (Mitra et al. 2004) reported that folate deficiency is a risk factor for arsenic-induced skin lesions. That folate deficiency might modify the risks of arsenic-induced health outcomes can be explained by the observation that people who are folate- deficient are poor methylators of arsenic and have more of the more toxic inorganic arsenic and MMA metabolites in blood and urine than those who are folate-replete (Gamble et al. 2005). Moreover, folic acid supplementation of folate-deficient study participants facilitated the synthesis and urinary elimina- tion of DMA (and total arsenic) and lowered blood MMA and total arsenic concentrations (Gamble et al. 2007). Another study of the influence of nutritional status on arsenic metabolism, carried out in pregnant women in Bangladesh, found only a marginal effect of plasma folate (after adjustment for the degree of arsenic exposure), which influences the methylation markedly (Li et al. 2008). That effect was due at least partly to the strong effect of pregnancy on arsenic methylation, starting very early in pregnancy and pos- sibly related to the betaine-mediated induction of one-carbon metabolism in early pregnancy (Gardner et al. 2011). Influence of Selenium Status on Arsenic Toxicity Selenium and arsenic antagonize each other’s toxicity, probably because they facilitate the excretion of one another in bile (Levander 1977; Zeng et al. 2005). A selenium–arsenic–glutathione conjugate has been identified in rabbits and mice (Gailer et al. 2002; Burns et al. 2008) but has not been validated in humans. Epidemiologic studies have described inverse relationships between blood or plasma concentra- tions of selenium and the percentage of MMA in blood or urine or both (Basu et al. 2011; Pilsner et al. 2011). A case–control study in Bangladesh also reported inverse relationships between blood selenium concentrations and urinary arsenic and the risk of arsenic-induced skin lesions (Y. Chen et al. 2007a). Key Considerations for the IRIS Assessment It is clear that folate and selenium nutritional status have important effects on arsenic metabolism and toxicity. Therefore, the IRIS assessment should consider the nutritional status of study populations when examining dose–response relationships reported in the epidemiologic literature. Although a nutri- tional deficiency should not exclude an epidemiologic study from inclusion in a systematic review, nutri- tional factors could be qualitatively noted when the weight of evidence is considered. However, concerns about deficiencies affecting arsenic toxicity in the United States should be muted because of the nutrition-

OCR for page 51
60 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic al status of the population with regard to folate and selenium. The United States mandated the fortifica- tion of foods with folic acid in 1998, thereby decreasing the prevalence of folate deficiency (and rates of adverse birth outcomes associated with deficiency during pregnancy). However, some segments of the population rely on maize rather than wheat and thus do not benefit from supplemented foods. People ex- posed to arsenic in Bangladesh and much of South Asia have a high prevalence of folate deficiency, which exacerbates arsenic toxicity (Gamble et al. 2005, 2007; Hall and Gamble 2012). The selenium con- tent of foods varies with the selenium content of soils in which they are grown. In the United States, soil selenium concentrations vary widely (USGS 2012). However, comprehensive reviews of blood and serum selenium concentrations in the United States (e.g., Combs 2001) have led to the conclusion that “the risk of selenium deficiency in the general US population is negligible” (Laclaustra et al. 2009). PRE-EXISTING DISEASE, SMOKING, AND ALCOHOL CONSUMPTION Pre-existing Disease Risk assessment is increasingly concerned about the vulnerability that stems from interaction be- tween a chemical’s mode of action and disease processes that can occur in segments of the population. With respect to inorganic arsenic, the evidence summarized in Chapter 4 that arsenic increases the risk for several major diseases—including cardiovascular, respiratory, and renal disease and diabetes—suggests that this source of vulnerability (pre-existing disease) may be particularly pertinent, especially given the high rates of these diseases in the United States. Because this is an emerging field of vulnerability as- sessment, several underlying principles and implications for quantitative risk assessment are outlined be- low in relation to considerations for inorganic arsenic. Chemical interactions with background disease processes may be of several types as shown in Figure 3:  The disease alters chemical action by altering toxicokinetics so as to change internal dose materi- ally.  The disease alters chemical action by affecting host defense mechanisms, as can occur when a path- ologic condition is associated with chronic inflammation and oxidative stress.  The chemical increases the likelihood of disease by damaging systems that are also affected by the disease process. In the first two interactions, the presence of the disease process in an exposed person can be ex- pected to shift the dose–response curve because the chemical is more effective in a diseased person. In the third, the disease is more likely to occur in an exposed person because of the chemical’s contribution to the pathologic process. Those interactions may occur in many exposed people, but at low dose they are most pertinent in those who are furthest along the disease spectrum and thus make up a group that is highly vulnerable both to the disease and to chemical toxicity. The interactions are most likely to be important on a population level if the vulnerable group constitutes a sizable fraction of the population. For example, if the disease is rare, few people are at heightened risk; this might be important for these few, but the population risk may not be greatly affected. In contrast, if the disease is common, many people will have it, and many others will have risk factors and subclinical biomarkers indicative of it that can in time lead to it. That implies a vulnerability distribution in which healthy members of the population are resistant to the chemical and those on the threshold of disease are most vulnerable to a chemical-induced shift to clinical disease. From a dose–response perspective, such a distribution may cause effects to be seen at exposures well below the threshold in a typical person because sensitive members of the population will respond at lower doses and

OCR for page 51
Disease process  Loss of function,  tissue damage  Altered disease  inflammation  risk or severity  Altered host defense Preclinical effect Clinical disease Altered  toxicokinetics Chemical exposure  Toxicokinetics Toxicodynamics Upstream effect Toxic effect Altered  Altered effective  internal dose  dose  Altered chemical  Altered chemical  potency potency FIGURE 3 Potential interaction between chemical exposure and disease process. Interaction could lead to altered chemical potency as disease process affects host toxicokinetics or defense mechanisms (downward arrows) creating a vulnerability to chemical effect. Interaction could also affect disease risk especially if the chemical and disease have similar upstream pathways and clinical outcomes (upward arrows). In this case, chemical exposure creates an additional risk fac- tor for disease to occur. 61

OCR for page 51
Susceptibility Factors 65 are available for determining whether exposure to environmental tobacco smoke interacts with arsenic with respect to skin, lung, or other end points. There are numerous potential mechanisms for the syner- gism, including inhibition of DNA repair by arsenic, which increases smoking-induced genetic damage. Most of the abovementioned studies of interactions between arsenic and smoking in Bangladesh and in occupationally exposed people involved men. There appears to be much less information on women. It was reported, however, that tobacco-chewing, which is prevalent in women in Bangladesh, was associat- ed with a considerably higher risk of skin lesions in women than that in women who did not use tobacco (Lindberg et al. 2010). It was particularly obvious in women who had the lowest efficiency of arsenic methylation. Because smoking is still relatively common, the finding of synergism with arsenic for at least four disease outcomes will be an important consideration. Indeed, of all susceptibility factors evaluated, smok- ing is probably the one on which the findings at both high and low to moderate arsenic exposure were most consistent. However, several uncertainties and data gaps with respect to this interaction could be acknowledged in attempting a quantitative assessment. The data gaps include lack of smoking–arsenic interaction studies for other end points, limitations of the information on the effects of secondhand tobac- co smoke, the incomplete description of interaction dose–response relationships, and the lack of mecha- nistic understanding. Regarding interaction dose–response relationships, the interaction between two agents will depend on the relative size of the doses: the nature of the interaction may change as the ratio of exposures varies (e.g., see Chou and Talalay 1984). For arsenic, the interaction matrix for skin lesions suggests that the best chance to see synergism is at higher doses of arsenic (Melkonian et al. 2011), and this might suggest that the interaction is less likely to affect the dose–response relationship at low arsenic doses. However, that is uncertain given the limitatins of the statistical power in the epidemiologic data to find low-dose arsenic effects and interactions. For cardiovascular disease, for instance, there is evidence of effect modification at low to moderate concentrations of arsenic (Y. Chen et al. 2011a; Moon et al. 2013). Furthermore, the literature has generally evaluated smoking status (current, former, or never) without defining the amount of smoking needed to begin to see an arsenic interaction. A plausible quantitative approach is a sensitivity analysis in which the smoking-interaction syner- gism size effect reported for skin lesions, lung cancer, bladder cancer, and cardiovascular disease is ap- plied to the dose–response relationship for these end points to determine the degree to which it would change the potency calculation. If that degree is found to be influential, consideration can be given to providing a separate potency estimate for smokers. A separate estimate could be useful in standard-setting and particularly in risk–benefit analyses when the burden of disease related to arsenic exposure is as- sessed in the general population in light of the proportion of people who smoke. Public education regard- ing radon includes the cancer risk in nonsmokers vs smokers because of the synergistic interaction be- tween these agents. If the data are sufficient, perhaps a similar approach could be used for public education concerning arsenic. Potential Interaction with Alcohol Consumption Although it is likely that many people have dual exposure to arsenic and ethanol, the potential for an interaction has not been extensively studied (Bao and Shi 2010). Results of a study of rats indicate that ethanol increases arsenic retention in the liver and kidneys and increases arsenic-induced hepatotoxicity (Flora et al. 1997). The epidemiologic data indicate higher concentrations of arsenic in urine in conjunc- tion with alcohol ingestion, and this suggests a toxicokinetic interaction, higher arsenic concentrations in alcoholic drinks, or increased water intake after alcohol intake (Tseng et al. 2005). Ethanol may augment arsenic-induced oxidative stress and induction of angiogenic factors that would promote tumor growth (Klei and Barchowsky 2008; L. Wang et al. 2012). There are additional mechanistic avenues through which an interaction could occur and potentially affect the outcome of epidemiologic studies (Bao and Shi 2010). That may prove to be a subject of productive research but has not been sufficiently developed to understand the nature and dose–response relationship of coexposure interaction on key end points.

OCR for page 51
66 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic Key Considerations for the IRIS Assessment Interaction with some disease processes or chemical coexposures, particularly smoking, may in- crease vulnerability to the effects of arsenic. Evidence that arsenic alters disease incidence or interacts with upstream disease biomarkers shows which end points are most likely to involve disease-related vul- nerability. Evaluation of the size and nature of vulnerable population groups will help to determine whether available epidemiologic studies adequately capture these groups. That consideration will also show how the response at doses below the range of observation might be affected with respect to the fea- sibility of defining a population threshold or dose-dependent transition. Quantitative approaches may in- volve separate analysis of vulnerable groups if such group-specific data are available or adjustment of the overall population response to account for specific chemical interactions or vulnerability factors. MIXTURES AND COEXPOSURES Coexposures or mixtures that include inorganic arsenic complicate the risk evaluation of arsenic in at least two ways: arsenic may interact with other agents that potentially modify the effect of arsenic, and there may be a combination effect of arsenic with other similarly acting chemicals (such as other metals). Arsenic may interact with other agents because it perturbs pathways and end points shared by the agents. The effect can be a “zero interaction” called additivity (for example, anticholinesterase pesticides have a cumulative effect on key enzymes [Timchalk et al. 2005]), synergism (greater than additive, such as arsenic or radon plus smoking), or antagonism (less than additive, such as the antagonistic relationship between arsenic and selenium). Those interactions are typically dose-dependent for each agent, so as- sessing how one chemical alters the response to another can be complicated. Thus, the nature of the inter- action may depend on the dose ratio and require testing multiple dose combinations of the interacting chemicals to explore fully the types of interaction possible (Jonker et al. 2005); or interactions that may occur at high doses of one or both chemicals may not occur at lower doses (an interaction threshold; Yang and Dennison 2007). The main concern at low doses is that an exposure to a chemical that is well tolerat- ed in the population (for example, below its threshold dose in most or all people) can become an im- portant risk contributor through additive or synergistic interaction. Zero interaction, or additivity, is typi- cally seen as shifting the dose–response curve to the left (more potent) according to simple dose additivity, whereas a synergistic interaction may have a different dose–response curve. Dose additive or synergistic interactions are likely to be highly variable in the population because exposure to the agents is not uniform and leads to a range of responses and vulnerabilities. That added variability may be important to capture in a risk assessment for chemicals with which prominent interactions are likely to occur at dos- es experienced in the population or with which coexposure to multiple similarly acting chemicals is like- ly. Additional uncertainty factors are possible, but background interaction with similarly acting chemicals can be a reason that thresholds can be difficult to define at the population level and can lead to the poten- tial for linear-appearing low-dose slopes and related modeling (NRC 2009). Inorganic arsenic has varied modes of action and numerous end points, so the potential for chemi- cal–chemical interaction or combination effects is large (Hong et al. 2004; Hore et al. 2007; Islam et al. 2011; Naujokas et al. 2013). Two types of interaction that have been identified beyond those previously discussed with selenium and tobacco smoke are interactions with other metals and with polycyclic aro- matic hydrocarbons. Coexposure and Potential Interaction with Other Metals Like arsenic, a number of metals are pro-oxidants and share underlying effects on proteins, signaling pathways, and genetic instability; thus, numerous arsenic–metal interactions are plausible (Hartwig 2013).

OCR for page 51
Susceptibility Factors 67 Candidates for such interaction with arsenic include lead, mercury, cadmium, chromium, nickel, and co- balt (Yao 2008; Valko et al. 2005). For example, cadmium and arsenic appear to have cumulative effects on renal-tubule leakage, as evidenced in humans exposed at moderate environmental doses (Huang et al. 2009). Those effects appear to be mediated by oxidative stress inasmuch as oxidative biomarkers in- creased in a pattern that appears additive among these agents. Interactions involving arsenic and other metals have not been well researched. Coexposure and Potential Interaction with Polycyclic Aromatic Hydrocarbons Another possible interaction is between inorganic arsenic and carcinogenic polycyclic aromatic hy- drocarbons (PAHs). PAH-related DNA adducts and mutagenesis are increased by coexposure to arsenic, and this interaction appears synergistic in at least some experimental systems (Maier et al. 2002; Fischer et al. 2005). The most plausible mechanism is that arsenic-induced inhibition of DNA repair allows the PAH adducts to be longer-lived and thus to build up in host DNA. Several studies that have tested that hypothesis, however, have failed to find impaired removal of PAH–DNA adducts in the presence of arse- nic (Maier et al. 2002; Chiang and Tsou 2009). Other mechanisms are possible, including arsenic-induced impairment of the p53 response to DNA damage (Chen et al. 2005; Shen et al. 2008). The fact that arsenic also potentiates ultraviolet-induced genetic damage (Wiencke et al. 1997; Chen et al. 2005) demonstrates the potential for interactive effects beyond PAHs. Although those interactions remain to be explained, they raise the possibility that at least some of the arsenic-induced tumors seen in human studies are co- mutagenic effects. That theoretically points to an alternative cancer-potency estimate based on the dose– response relationship in which arsenic potentiates the toxicity–genotoxicity–carcinogenicity of other agents. However, that may be somewhat speculative given limitations in quantitative dose–response data and in the mechanistic information needed to explore such possible potentiation fully. Therefore, this is a subject worthy of mention by EPA as an additional mechanistic consideration that may help to explain some of the end points associated with arsenic exposure. For interactions in which there is arsenic coexposure with other carcinogens, the standard risk- assessment assumption of dose additivity is reasonable. Key Considerations for the IRIS Assessment In its evaluation of inorganic arsenic, EPA should consider coexposure to other metals and PAHs because these are the most documented. In accumulating the evidence on mixtures that include arsenic, it would be helpful to take note of which chemicals arsenic tends to occur with in the environment. Coexpo- sure to some metals (such as lead and cadmium) is particularly likely because of their similar sources, persistence, and retention in topsoil (ATSDR 2007; Wang and Fowler 2008). Arsenic in drinking water sometimes occurs with manganese, uranium, radon, and other elements (Ayotte et al. 2011). (Arsenic, cadmium, chromium, lead, and mercury are among the most common metals on the basis of site frequen- cy as tabulated by the Agency for Toxic Substances and Disease Registry (ATSDR 2009). Arsenic also occurs with other environmental chemicals, including PAHs (Mori et al. 2011). Three cases of possible interactions could guide EPA’s assessment of coexposures in the context of their affecting the interpretation of dose–response relationships in key epidemiologic studies. Case 1. The key epidemiologic study does not have subjects who are more vulnerable as a result of coexposure to interacting metals or PAHs. For example, some Superfund sites might have hot spots of metal contamination involving arsenic and other metals. That could represent a special exposure scenario that is not captured by statistical sampling of a general population. By not counting or identifying such vulnerable people, an epidemiologic study might not capture the risk in this subgroup, and that could lead

OCR for page 51
68 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic to potential underestimation of risks to those who are subject to the coexposure. The concern is partially mitigated by recommendations made later in this report (see Chapter 7) to calculate potency values for multiple end points, not just the most sensitive one. The risk of each end point associated with arsenic could then be considered cumulatively with similarly acting metals that co-occur in that exposure groups. Case 2. The key epidemiologic study has arsenic as a covariate (it is positively correlated) with oth- er important metals and PAHs, but they are not measured. In such a study, the health effects might be completely (and incorrectly) attributed to arsenic without considering the contribution of other chemicals. That could lead to overestimation of the effects of arsenic. Case 3. The key epidemiologic study has interacting metals or PAHs distributed by chance across the arsenic exposure groups; this would increase the unexplained variability in each group. The additional variability would tend to obscure or weaken the effect of arsenic. In considering those scenarios, it would be helpful for EPA to assess, to the extent possible, how in- teracting metals or PAHs might co-occur in the epidemiologic study populations in comparison with the target population of the risk assessment. That could lead to a better understanding of potential underesti- mation or overestimation of risk and a description of how coexposures might contribute to overall poten- cy on the one hand and uncertainty on the other and of whether cumulative risk posed by metals, PAHs, or other chemicals is important to consider. SUMMARY As described above, populations may be susceptible to the effects of arsenic because of their sex, life stage, nutritional status, metabolism-related polymorphisms, disease status, smoking status, and expo- sure to other chemicals that could interact with arsenic to produce a cumulative or synergistic effect. Con- sideration of susceptible subpopulations and life stages in deriving risk-based values is a complex task. Developing a potency adjustment for such populations is feasible if the appropriate dose–response data are available for a given subgroup in comparison with the general population. With respect to inorganic arsenic, it is plausible that those in pre-existing or high-risk categories for chronic diseases (such as cardi- ovascular disease, diabetes, and renal disease) could be more vulnerable to arsenic toxicity inasmuch as arsenic has been associated with increased risks of those conditions in at least some cases or has been shown to act on upstream biomarkers of disease risk (e.g., Wu et al. 2012; Kunrath et al. 2013). Thus, the available literature should be examined to assess whether such information is available in existing epide- miologic studies or systematic reviews. Windows of susceptibility appear to occur in connection with in utero or early-life exposure with respect to ensuing cancer, cardiovascular, and respiratory risk. Nutrition- al factors (such as low folate intake), genetic deficiency in arsenic methylation, and tobacco smoke might also predispose some people to arsenic toxicity. The committee supports EPA’s plans to use systematic reviews to evaluate susceptibility factors. For many of the factors considered in this chapter, it is unlikely that adequate data will be available to incorporate the considerations quantitatively in estimating risks. The factors that appear to have the strongest evidence are in utero and early-life exposures, sex differences, and smoking. If adequate data are found, consideration could be given to determining a separate potency calculation for a vulnerable subgroup that would be carried through the IRIS process to the extent possible and to determining wheth- er the vulnerability is appropriately represented in the study population relative to the target (US) popula- tion. In the absence of adequate data, the size of the relevant subgroup (if large, likely to contribute to population risk) and whether the mechanism for vulnerability (if known) is likely to be operable at low dose could be evaluated. At high dose, both vulnerable and less vulnerable people can be affected; but at low dose, only the most vulnerable are likely to be affected. If that is a small percentage of the population, there might not be enough people to permit recording a statistically significant effect at low dose. The larger the fraction of the population that has a particular vulnerability, the greater the chance that the pop- ulation will continue to have a significant response at low dose and thereby extend the dose range of ob-

OCR for page 51
Susceptibility Factors 69 served effects. That is the apparent explanation of the linear-appearing population response to particulate matter and cardiovascular mortality in spite of the likelihood that a threshold for the effect occurs at the individual level (NRC 2009). As the dose is lowered, fewer people are responsive, so statistical signifi- cance is harder to achieve, but this does not mean that the risk at low dose is zero. In cases in which a siz- able subpopulation is vulnerable to a particular effect, it is reasonable to extend the dose–response rela- tionship below the range of observation by a modest extrapolation.

OCR for page 51
6 Mode of Action The Environmental Protection Agency (EPA) should evaluate the biochemical and systems-biology in- formation available on arsenic to determine how it can help to explain the pattern of adverse health out- comes and the mode of action that best matches the underlying biology. Similarly, variability and its effect on the low end of the dose–response curve and the distinction between adaptive changes and adverse re- sponses can be addressed in the context of mode of action. Mode-of-action analysis provides a framework for data integration around a common theme for an agent and a specific health outcome (Andersen et al. 2000; Carmichael et al. 2011). The strategic vision for mode-of-action analysis in risk assessment has been laid out in EPA’s cancer-risk guidelines (EPA 2005b) and has been successfully applied to the assessment of chloroform (EPA 2001) and trichloroeth- ylene (EPA 2011). EPA’s draft plans indicate that this mode-of-action framework will be included in its analysis of adverse-outcome pathways. Mode-of-action data can be extended to a number of uses beyond the cancer-risk guidelines, including noncancer health outcomes and sensitive population assessments. Mode-of-action analysis is a systematic approach to understand the relationship between exposure to an agent and biologic outcomes and can affect interpretation of events at the low end of the dose–response curve. It uses the identified key instigating events to drive dose–response analyses. An acknowledged challenge for EPA in using this approach is that many of the existing mode-of-action data have been de- veloped at relatively high exposure to arsenic and the objective is to understand attributable risk at lower exposures. In addition, strong data on actual systemic exposure to the various forms of arsenic after inges- tion are often difficult to obtain. Identifying those data gaps and their potential effect on the ability to ex- trapolate to potential effects at low exposures wil be an important part of the IRIS assessment process. The mode-of-action framework (Boobis et al. 2006, 2008; Carmichael et al. 2011) in conjunction with the human-relevance framework (Meek et al. 2003) provides a transparent method of organizing in- formation for hazard identification and risk assessment that includes exposure information, dose–response information, a clear conclusion, identified data gaps, and potentially susceptible populations. It permits the integration of cancer and noncancer risk assessment (Carmichael et al. 2011) and provides a transpar- ent mechanism to share results with stakeholders. A similar approach that uses a graphical presentation of how epidemiologic and toxicologic data intersect is described by Adami et al. (2011). Similar discussions of new approaches to hazard identification and risk assessment are provided in Science and Decisions: Advancing Risk Assessment (NRC 2009) and Toxicity Testing in the 21st Century: A Vision and a Strate- gy (NRC 2007) and are meant to extend the concepts provided in the “Red Book” (NRC 1983). IMPORTANT ASPECTS OF MODE-OF-ACTION ANALYSIS The key aspect of mode-of-action analysis is that it provides an evidence-based assessment, integra- tion, and synthesis of all the available data on a chemical, its adverse health effects, and its biology. The data integration and synthesis of a mode-of-action analysis incorporate various types of data, including epidemiologic data, standard toxicity-testing data, data from mechanistic biochemical and metabolism studies, data from gene-expression studies, and data on epigenetic changes. As discussed in the commit- tee’s workshop by Clewell (2013), those data can come from in vivo or in vitro studies involving a variety 70

OCR for page 51
Mode of Action 71 of species, systems, and cell types whose relevance to humans could be assessed. A chemical can modu- late numerous upstream events, and determining whether these converge to form one or more modes of action for a specific adverse outcome can be challenging. Therefore, mode-of-action analysis endeavors to provide a plausible hypothesis for an adverse outcome (Andersen et al. 2000; Meek et al. 2003); there can be uncertainties in the sequence of events, in the causal nature of the events, and in whether more than one pathway is involved. Mode-of-action analysis is distinct from a detailed understanding of mechanism of action and places into perspective all relevant scientific data that link key events and pathways to an adverse health out- come that is consistent with the underlying biology of the target tissue. It is that aspect of mode-of-action analysis that places available data into a context that can be used to interpret vulnerability among species and among individuals so that changes in dose–response relationships in going from high to low expo- sures can be understood. Thus, mode-of-action analysis is an essential step in Integrated Risk Information System (IRIS) assessment, and the committee applauds EPA’s efforts to organize the in vitro and in vivo mechanistic data and its plan to use this analysis in interpreting dose–response relationships and intersub- ject variability in the inorganic arsenic assessment. Mode-of-action analysis should be applied to organize and synthesize data and to harmonize cancer and noncancer end points, especially in the case of arsenic, on which there are large data gaps in regions of the dose–response relationship in key areas and low-dose exposures of regulatory concern. Although association studies alone are not sufficient for quantitative risk assessment, they provide a qualitative guide for hazard identification. One should rigorously examine each (epidemiology) study in the context of the modified Bradford Hill criteria to determine appropriateness for risk assessment (Boobis et al. 2006, 2008). Coexposures are particularly important to take into account because differences in cancer tumor spectrum or other health outcomes (such as diabetes, cardiovascular disease, immune effects, and developmental effects) between populations may be due to the presence of different contaminants (such as lead, mercury, cadmium, and silica) or different dietary inadequacies (such as zinc, selenium, folate, and protein). Similarly, toxicology studies are often performed to detect hazards and may not be designed for description of mode of action. Mechanistic studies performed in vitro and in vivo can inform under- standing of the mode of action. An emphasis on hazard characterization is warranted and can be achieved by using a mode-of-action approach in which the necessary key events for the pathogenesis of each health outcome are examined in a dose-dependent and time-dependent manner. Not all proposed modes of action will be supported by adequate data, and there will not always be complete data on the health outcomes under consideration, but the mode-of-action framework should be worked through for the proposed modes of action for each adverse health outcome of concern. It is under- stood that for many of the health outcomes, a detailed mode-of-action analysis will not be possible, be- cause of lack of information. Providing the available data and identifying data gaps and the effect of the lack of information on confidence in the assessment of risk, particularly at low exposures, will be im- portant parts of the evaluation. When approached in this manner, a proposed mode of action provides a unifying hypothesis for interpreting and integrating multiple studies and for explaining differences be- tween the sexes, among populations, among different ages, and among species. Several test cases for can- cer mode of action have been provided, and this framework has been extended to noncancer end points (Sonich Mullin et al. 2001; Meek et al. 2003; Boobis et al. 2006, 2008; Rhomberg et al. 2011). The mode- of-action framework provides a transparent method for examining the weight of scientific evidence sepa- rate from policy considerations but with an eye to informing science policy (Carmichael et al. 2011). In beginning a mode-of-action analysis, the committee recommends that the IRIS program use an approach similar to that followed in the IRIS assessments of chloroform (EPA 2001) and trichloroeth- ylene (EPA 2011). The process needs to begin with the question to be answered about the health effects of exposue to arsenic. Identifying the adverse outcomes of concern and determining the strengths and weaknesses of the available data are important steps in this process. Toxicokinetic data are essential for understanding how exposure corresponds to internal dose and how they vary among individuals. Expo- sure assessment will influence the dose-dependent and time-dependent response to an agent and is im- portant for dose–response analysis and for risk assessment. Understanding exposure data related to each

OCR for page 51
72 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic health outcome throughout the dose range is a particularly important aspect of the process and is an acknowledged difficulty that IRIS assessors will face in light of the existing arsenic database. EPA’s draft plan for the inorganic arsenic assessment describes proposed evidence tables that will be organized by health category and will capture exposure and outcome information in human studies. A modified form is needed for animal or in vitro studies to record such information as the form of arsenic used, the species and strain, and the tissue or cell line. Understanding how the assessment was performed will be greatly facilitated by the use of additional tables that list modes of action for each health outcome (both cancer and noncancer) and include exposure, tissue dose (if possible), and key events. The analysis should further demonstrate how the documented exposure and time course of exposure lead to the key events. The tables should cite literature that was used to construct the analysis and discuss supporting and conflicting evidence. Linkage of adverse health outcomes to the key events in a mode-of-action analy- sis—when coupled with exposure, metabolism, tissue accumulation of parent and metabolites, and rate- limiting steps (such as protein binding, reactive oxygen signaling, cytotoxicity, proliferation, apoptosis, receptor effects, and immune suppression)—provides a strong hazard identification and characterization analysis. This evidence map of the exposure–response relationship, exposure frequency, and duration— when coupled with the time dependence of the sequential progression and time course of the observed health effects—should match the proposed key events predicted by the mode of action. Such a mode-of- action assessment can be particularly helpful in the analysis of potential confounding effects, of the role of simultaneous exposure to other agents or actions, and of individual risks and population susceptibili- ties. Mode-of-action analysis is a transparent method for examining data and interpreting data gaps and for providing an integrated assessment of hazard identification and risk characterization. The linkage of in vitro to in vivo animal to human dose and dose duration response to temporal onset of key events is nec- essary (Andersen et al. 2000; Tsuda et al. 2003; Slikker et al. 2004a,b). The data can then be compared to see whether they are qualitatively and quantitatively similar. That comparison could be facilitated by cre- ating concordance tables in the range of observations for in vitro, in vivo animal, and human data. The tables should take into account the number of cases and the incidence, the actual exposure (on the basis of a common unit, such as micrograms per liter of plasma, micrograms per gram of tissue, current exposure, total cumulative exposure, or tissue concentrations), species or population (and subpopulation) differ- ences, coexposures, and the timing and route of exposure in the analysis. It would be helpful to document intervention studies that perturb the adverse-outcome pathway either upstream or downstream of the pro- posed effect and thus provide evidence of the relative importance to the outcome in question. In addition, it would be useful to document how the observed health outcomes and mode of action attributed to arse- nic could be modulated by other potentially causal agents, such as micronutrient deficiency, other metal excess, host vulnerability factors, life stages, pre-existing disease, diet, sex, genetic background, and smoking. Box 6 lists the major steps of mode-of-action analysis. BOX 6 Steps of Mode-of-Action Analysis 1. Provide problem formulation statement 2. Tabulate adverse outcomes with supporting and conflicting data 3. Provide pharamacokinetic data throughout the exposure and temporal range for each adverse health outcome and its precursors 4. List modes of action for each adverse outcome, linking pharmacokinetic and pharmacodynam- ic information to health outcomes in an exposure and temporal manner 5. Construct a concordance table to provide strengths and weaknesses of each proposed mode of action for each species, population, and subpopulation

OCR for page 51
Mode of Action 73 POTENTIAL MODES OF ACTION OF ARSENIC An important aspect of the IRIS assessment of inorganic arsenic will be provision of a proposed mode of action for each observed health outcome, including the associated supporting and contradictory evidence. If after execution of a mode-of-action framework analysis a cohesive mode of action is not ap- parent or it is clear that multiple modes of action may be involved, a mode-of-action summary statement should indicate that while elaborating the data and hypotheses assessed. Determination of the key events in an adverse-outcome pathway is in the context of the exposure and duration of exposure leading to the key intermediary steps in the pathogenesis of the health outcome. The key events must be linked in an exposure and temporal pattern for each of the steps in the pathway (cf Andersen et al. 2000; Tsuda et al. 2003; Slikker et al. 2004a,b). Where possible, data from multiple studies should be integrated to link tis- sue dose and biologically effective dose with the pathogenesis of the adverse outcome of concern. Thus, both dose response and temporal response must be consistent for a given hypothesized mode of action. For each primary health outcome, the precursor events (the rate-limiting or key events) can be determined as a function of time and exposure relative to the health outcome. Defining the key events consists of de- scribing the absorption, metabolism, accumulation, and retention of arsenic or its metabolites in conjunc- tion with any biochemical or functional change during the pathogenetic process that leads to the health outcome of concern (Boobis et al. 2009). For arsenic, defining the exposure at which tissue metabolism and accumulation occur at levels sufficient to trigger the next key event is important and could be in- formed by pharmacokinetic and pharmacodynamic models. The temporal and exposure relationship for the next key event for each health outcome in pathogenesis would permit linkage to the adverse outcome. The time course and dose dependence should be coherent with respect to dose, time, species, age, sex, and population for each adverse health outcome of concern. The IRIS assessment should consider the plausible modes of action for each adverse outcome that might result from oral exposure to arsenic. Several modes of action have been described in the literature were discussed in the committee’s workshop by Cohen (2013) and detailed in Chapter 4 as potentially associated with arsenic action. They include binding to protein sulfhydryl groups, reactive oxygen genera- tion, reactive oxygen species (ROS) signaling and the oxidative stress that ensues, and perturbation of DNA methylation (epigenetic factors) (Kitchin and Wallace 2008; Kitchin and Conolly 2010). Tissue- specific binding to protein sulfhydryl groups could be explored as a mode of action that can lead to inhibi- tion of a specific protein activity (Kitchin and Wallace 2008). Sulfhydryls (such as glutathione), vicinal thiols (such as pyruvate dehydrogenase and some zinc finger proteins), or selenium may be the cellular target of arsenic and its metabolites. An example of how the arsenic mode-of-action analysis could be structured is provision of what is known about the dose-dependent interactions of arsenic with protein sulfydryls, vicinal dithiols, and zinc- containing proteins and how the interactions might lead to biochemical changes (such as ROS genera- tion), functional changes (such as mitochondrial damage), and epigenetic changes as a function of expo- sure. How those effects could culminate in the development of a specific health outcome can then be de- scribed to the extent possible on the basis of the underlying data (see, for example, Snow et al. 2005), especially with respect to their dose dependence. That approach is supported by the temporal associations that are observed in mice exposed to a high concentration of arsenic in drinking water (Kitchin and Conolly 2010). The temporal analysis is useful and when coupled with a dose-dependent and exposure- dependent profile may provide an integrated sequence of events that define arsenic adverse-outcome pathways. Such a framework may highlight the induction of ROS and redox imbalance as an integrative mode of action for lung cancer, bladder cancer, and cardiovascular disease. Arsenic’s action is complex, and many mechanisms of action and several possible modes of action have been assessed in the context of exposure of humans to high doses through drinking water. It is gen- erally accepted that arsenic is not directly mutagenic but rather acts via an indirect mechanism that in- volves secondary mediators (such as oxidant damage, modified proteins, and immune suppression). Other research has provided data to support proliferation as a mode of action of bladder carcinogenesis associat- ed with high-dose exposure to inorganic arsenic. In adult rats given arsenic directly, cytotoxicity is ob-

OCR for page 51
74 Critical Aspects of EPA’s IRIS Assessment of Inorganic Arsenic served and is followed by regenerative repair (Nascimento et al. 2008; Suzuki et al. 2008a,b, 2009a,b, 2010, 2012; Yokohira et al. 2010). Similarly, an increased incidence of bladder neoplasms is observed in adult mice exposed to high doses of arsenic during the period of rapid cell proliferation in utero (Tokar et al. 2010b). Another hypothesized mode of action is epigenetic modulation that may influence arsenic- induced signaling response. The most studied aspects of the epigenome are DNA methylation, covalent posttranslational histone modifications, and microRNAs, and they have been associated with both cancer and noncancer effects. Epigenetic modifications are believed to influence signaling events and phenotypic variation in cells. DNA methylation is the most widely studied epigenetic modulation in humans and oth- er mammals; it can influence gene silencing by directly affecting the affinity of transcription factors for their DNA binding sites. Arsenic exposure has been associated with decreased S-adenosyl methionine concentrations (and increased concentrations of S-adenosyl homocysteine), altered methyltransferase ac- tivity, and changes in DNA methylation patterns in animals and humans (Reichard and Puga 2010). Dose and temporal association of DNA methylation changes with key initiating events would need to be de- rived to support its role in mode of action analyses. Gene-expression and other -omic data can support the determination of mode of action. Several groups have performed gene expression in cell lines, in primary human cells, and in tissues from animals treated with arsenic, and the results need to be organized by health outcome and dose, exposure duration, and time course of response as reviewed by Ankley et al. (2010). Analysis of the quality of the data and the normalization methods need to be considered. Several additional caveats need to be considered in that driver genes may be expressed at low levels and not detected, the relevant change may not be at the gene transcription level, many changes are simply responses and not causal mechanisms, and not all changes are adverse. In addition, the genomic profile should be obtained of the tissue type of origin (specifically, primary cells from this tissue for in vitro studies) for the health outcome of concern. Because genes are not independent variables, one should focus on rate-limiting steps for the proposed mode of action and key events. Integration of -omic data permits linkage among tissues, and the Bayesian network-analysis approaches may extend to comparison of populations (Subramanian et al. 2005; Geneletti et al. 2011; Schadt and Björkegren 2012). Pathway mapping can be a useful place to start in such qualitative analyses (Clewell et al. 2011; Yager et al. 2013), but current datasets need to be critically assessed with regard to normalization procedures and statistical control of batch effects and multiple comparisons. In addition to pathway mapping, one might focus on ligand through receptor to signaling to transcription-factor analysis or on the transcription-factor set (for example, zinc finger proteins with vicinal thiols). It is important to remember that genes are not independent variables and that statistical significance is not equivalent to biologic relevance (Moggs et al. 2004; Mootha et al. 2004). Several papers provide a framework for the use of gene-expression data to inform understanding of key events (Bercu et al. 2010; Gentry et al. 2010); however, they have focused on perturbation of pathways, not on approaches to determining mode of ac- tion or key events (Moggs et al. 2004; Mootha et al. 2004). Gene-expression analyses permit hypothe- ses—such as selenium depletion, zinc finger protein inactivation, or interruption of mitochondrial respira- tion—to be explored to support assessment of mode of action and key events. Modified Bradford Hill criteria for assessing causality have been used to provide a structure for as- sessing causality systematically in toxicologic and epidemiologic studies (Adami et al. 2011). Causality assessment is a key defining principle in mode-of-action analyses and is an integral part of a toxicologic assessment. The application of the modified Bradford Hill criteria in the integration of mechanistic, toxi- cologic, and epidemiologic data is necessary to determine which adverse health outcomes are causally associated with a specific exposure, exposure duration, and timing of exposure. The application of these criteria to combined mechanistic and epidemiologic data has been conceptualized by Adami et al. (2011), and they have been applied to the effects of atrazine on breast cancer by Simpkins et al. (2011). EPA pro- poses the use of adverse-outcome-pathway analysis in a separate construct that is similar to mode-of- action analysis but does not consider that some changes are physiologic or even adaptive and not adverse. The utility of mode-of-action and adverse-outcome-pathway analysis lies in the fact that their dependence on the modified Bradford Hill criteria for causality assessment relies on exposure and temporal depend- ence for coherence of any observed effects; thus, this emphasizes attributable risk.

OCR for page 51
Mode of Action 75 SUMMARY Mode-of-action analyses permit a transparent assessment of the data supporting or refuting the hu- man health effects of exposure to oral inorganic arsenic in the US population. Both cancer and noncancer effects can be integrated through a mode-of-action analysis. Such analysis permits an unbiased use of all the available data to examine the effects of arsenic exposure and exposure duration at different doses. A mode-of-action analysis needs to be compiled for each health outcome that has been ascribed to arsenic exposure to inform the risk assessment. Numerous mechanisms of action have been proposed for arsenic- associated health outcomes (including protein binding, ROS generation, epigenetic effects, and cytotoxi- city), and it is unlikely that a single mode of action is responsible for all the observed adverse health out- comes. Mode-of-action analysis permits the integration of data to advance understanding of the coher- ence, biologic plausibility, and human relevance of findings throughout the exposure–response continuum and provides a transparent means of synthesizing the data. Mode-of-action analysis may be particularly useful in understanding low-dose exposure and dose-dependent transitions that may occur with increasing dose. The exposure (the tissue or biologic effect level) is considered in the context of metabolism, transport, and accumulation in the tissues of interest. Where nonlinearities in the pharmacokinetics or pharmacodynamics are identified, they can be used to understand the time and dose dependences of the key rate-limiting events and of the proposed adverse health outcome. The mechanistic, toxicologic, and epidemiologic data can then be interpreted in a comprehensive mode-of-action framework that facilitates an improved understanding of exposure–response relationships and interhuman variability of response.