Scholarly article on topic 'Consideration of non-linear, non-threshold and threshold approaches for assessing the carcinogenicity of oral exposure to hexavalent chromium'

Consideration of non-linear, non-threshold and threshold approaches for assessing the carcinogenicity of oral exposure to hexavalent chromium Academic research paper on "Nano-technology"

CC BY-NC-ND
0
0
Share paper
OECD Field of science
Keywords
{Chromium / Hexavalent / Carcinogenicity / Regulatory / Cancer / Toxicity / "Oral exposure"}

Abstract of research paper on Nano-technology, author of scientific article — J. Haney

Abstract A non-linear approach, consistent with available mode of action (MOA) data, is most scientifically defensible for assessing the carcinogenicity of oral exposure to hexavalent chromium (CrVI). Accordingly, the current paper builds upon previous studies (Haney, 2015a, 2015b) to first develop a non-linear, non-threshold approach as well as a non-linear threshold approach for assessing the oral carcinogenicity of CrVI, and then utilizes available MOA analyses and information for selection of the most scientifically-supported approach. More specifically, a non-linear, non-threshold dose–response function was developed that adequately describes the non-linearity predicted for potential human excess risk versus oral dose due to the sub-linear relationship between oral dose and internal dose (added mg Cr/kg target tissue) across environmentally-relevant doses of regulatory interest. Additionally, benchmark dose modeling was used to derive a reference dose (RfD of 0.003 mg/kg-day) with cytotoxicity-induced regenerative hyperplasia as a key precursor event to carcinogenesis in the mouse small intestine. This RfD value shows remarkable agreement with that published previously (0.006 mg/kg-day) based on a more scientifically-sophisticated, physiologically-based pharmacokinetic modeling approach (Thompson et al., 2013b). The RfD approach is the most scientifically-defensible approach based on the weight-of-evidence of available MOA information and analyses conducted for the most scientifically-supported MOA.

Academic research paper on topic "Consideration of non-linear, non-threshold and threshold approaches for assessing the carcinogenicity of oral exposure to hexavalent chromium"

H ||| Contents lists available at ScienceDirect

Regulatory Toxicology and Pharmacology

journal homepage: www.elsevier.com/locate/yrtph

Consideration of non-linear, non-threshold and threshold approaches for assessing the carcinogenicity of oral exposure to hexavalent chromium

J. Haney Jr.

Texas Commission on Environmental Quality (TCEQ), Austin, TX, United States

ARTICLE INFO ABSTRACT

A non-linear approach, consistent with available mode of action (MOA) data, is most scientifically defensible for assessing the carcinogenicity of oral exposure to hexavalent chromium (CrVI). Accordingly, the current paper builds upon previous studies (Haney, 2015a, 2015b) to first develop a non-linear, non-threshold approach as well as a non-linear threshold approach for assessing the oral carcinogenicity of CrVI, and then utilizes available MOA analyses and information for selection of the most scientifically-supported approach. More specifically, a non-linear, non-threshold dose—response function was developed that adequately describes the non-linearity predicted for potential human excess risk versus oral dose due to the sub-linear relationship between oral dose and internal dose (added mg Cr/kg target tissue) across environmentally-relevant doses of regulatory interest. Additionally, benchmark dose modeling was used to derive a reference dose (RfD of 0.003 mg/kg-day) with cytotoxicity-induced regenerative hyperplasia as a key precursor event to carcinogenesis in the mouse small intestine. This RfD value shows remarkable agreement with that published previously (0.006 mg/kg-day) based on a more scientifically-sophisticated, physiologically-based pharmacokinetic modeling approach (Thompson et al., 2013b). The RfD approach is the most scientifically-defensible approach based on the weight-of-evidence of available MOA information and analyses conducted for the most scientifically-supported MOA.

© 2015 The Author. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND

license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

n* Regulatory Toxicology and Pharmacology

Article history: Received 10 July 2015 Received in revised form 19 September 2015 Accepted 15 October 2015 Available online 19 October 2015

Keywords:

Chromium

Hexavalent

Carcinogenicity

Regulatory

Cancer

Toxicity

Oral exposure

1. Introduction

A significant amount of new research has been conducted over the past several years to generate data specifically to better inform the mode of action (MOA) analysis for hexavalent chromium-induced carcinogenesis due to oral exposure and to improve the extrapolation of rodent oral study results to humans (e.g., Thompson et al., 2011a, 2011b, 2012a, 2013a; Kirman et al., 2012, 2013; Proctor et al., 2012; Kopec et al., 2012a, 2012b; O'Brien et al., 2013; Suh et al., 2014; Thompson et al., 2015a, 2015c). Thorough evaluation of these research project data is essential to a better scientific understanding of the carcinogenic MOA operating in relevant rodent studies (e.g., NTP, 2008) and hexavalent chromium (CrVI) toxicokinetics following oral exposure, both of which are of particular importance considering the significant regulatory challenge of extrapolating high oral dose results from laboratory

E-mail address: joseph.haney@tceq.texas.gov.

animal studies to environmentally-relevant human doses that are orders of magnitude lower in a meaningful (not just conservative), toxicologically-predictive manner (e.g., the mouse dose at the lowest water concentration used in NTP, 2008 is about 74,000 times higher than the approximate human dose corresponding to the 35-city geometric mean drinking water concentration reported in EWG, 2010). Consequently, regulatory agencies should duly consider these data to inform key areas of chemical dose—response assessment such as the MOA (e.g., key events), toxicokinetics (e.g., dose-dependent differences in target tissue absorption), and biologically-plausible expectations about potential thresholds and any low-dose risk.

Failure of a chemical assessment's low-dose extrapolation to appropriately consider and incorporate (if scientifically robust and defensible) relevant CrVI research project data on MOA and tox-icokinetics may result in significantly overestimating environmental risk. For example, recent analyses of CrVI toxicokinetic data (Kirman et al., 2012) revealed appreciable dose-dependent differences in target tissue absorption (Haney, 2015a, 2015b). More

http://dx.doi.org/10.1016/j.yrtph.2015.10.011

0273-2300/© 2015 The Author. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

specifically, the dose fraction absorbed (or CrVI absorbed by target tissues per unit dose) progressively decreases with decreasing oral dose, resulting in non-linearity (i.e., sub-linearity) between oral dose and target tissue dose across doses of environmental interest (Fig. 1, reproduced from Fig. 2 in Haney, 2015b). This type of tox-icokinetic information that reveals a non-linear relationship between oral and internal dose should be taken into account in assessing the potential for a non-linear dose—response (USEPA, 2005). Taking this non-linear/sub-linear relationship between oral and internal target tissue CrVI dose into account, Haney (2015b) concluded:

• Decreasing target tissue absorption as doses decrease to lower, more environmentally-relevant doses is inconsistent with linear low-dose extrapolation as the shape of the dose—response curve accounting for this toxicokinetic phenomenon would be nonlinear;

• The magnitude of risk overestimation by a linear low-dose extrapolation approach (e.g., the USEPA draft oral slope factor or SFo) increases significantly as it is used to predict risk at lower, more environmentally-relevant CrVI doses where the dose fractions absorbed by target tissues progressively decrease; and

• A non-linear approach, consistent with available MOA data, is most scientifically defensible for assessing CrVI-induced carcinogenesis.

Accordingly, consistent with results from prior toxicokinetic analyses demonstrating a non-linear/sub-linear relationship between oral dose and internal dose (i.e., target tissue concentration), the current paper builds upon previous studies (Haney, 2015a, 2015b) to:

1) Develop two non-linear approaches (i.e., non-linear, non-threshold and threshold) for assessing the carcinogenicity of oral exposure to CrVI; and

2) Utilize available published MOA analyses and information for selection of the most scientifically-supported approach.

The non-linear, non-threshold approach considered is a novel one for assessing the potential risk of oral exposure to CrVI. The non-linear, threshold approach is represented by the derivation of a reference dose (RfD). While other RfD values have been developed, the RfD derived herein is based on an internal dose metric (unlike the draft RfD in USEPA, 2010), focuses specifically on the duodenum as the most tumorigenically responsive target tissue (unlike

Thompson et al., 2013b), and converts internal dose to external dose based on an independently modeled relationship (Haney, 2015a), thus providing an important point of comparison for previously derived values. Lastly, while the MOA data reviewed herein have been discussed in various studies elsewhere (e.g., Thompson et al., 2013a), this study represents the first review and interpretation of all these data (including newly published studies) to appear in a peer-reviewed scientific journal by staff of a regulatory agency. The regulatory perspective is important because regulatory agencies ultimately determine what low-dose extrapolation approaches are scientifically justified and any impact of new MOA research on health-protective regulations.

2. Materials and methods

Pursuant to prior analyses and justification provided in Haney (2015b), this paper considers two non-linear approaches for assessing the carcinogenicity of oral exposure to CrVI:

• A non-linear, non-threshold low-dose extrapolation approach;

• A non-linear threshold approach.

The non-linear, non-threshold low-dose extrapolation approach is exemplified by the development of a mathematical model (i.e., dose—response function) that adequately describes the non-linearity that would be predicted in excess risk versus oral dose, under the preliminary assumption that excess risk is proportional to target tissue concentration of absorbed CrVI down to zero dose (i.e., under an assumed mutagenic MOA for the sake of comparison), when dose-dependent differences in target tissue absorption are appropriately considered. These differences are discussed in Haney (2015a), which provides a peer-reviewed approach to calculate dose-specific adjustment factors for the draft SFo (USEPA, 2010) based on dose-dependent differences in absorption. These dose-specific adjustment factors that account for the non-linearity in oral dose versus target tissue concentration were used in a second study (Haney, 2015b) to estimate potential excess risk at environmentally-relevant doses (e.g., doses at the federal maximum contaminant level (MCL), 1/3 the MCL, measured drinking water concentrations) and produce an associated dose—response curve (Fig. 2, reproduced from Fig. 3 of Haney, 2015b). The dose—response is non-linear due to the non-linear (i.e., sublinear) toxicokinetics of CrVI absorption by target tissues being taken into account (Fig. 1) (also see Fig. 4 of Haney, 2015a). It is this non-linear/sub-linear dose—response for excess risk which must be

Fig. 1. Dose-dependent changes in mouse target tissue absorption per unit dose and low-dose nonlinearity in absorbed tissue concentration versus dose.

Fig. 2. Potential human excess risk versus lower dose adjusted for dose-dependent differences in target tissue absorption.

adequately described mathematically (down to zero dose) to exemplify the non-linear, non-threshold low-dose extrapolation method.

Although the methods provided in Haney (2015a) can be used to more accurately estimate potential carcinogenic risk at a given oral dose while accounting for dose-dependent changes in target tissue absorption, relatively burdensome dose-specific calculations are required. Therefore, a more practical and user-friendly approach is desirable for calculating excess risk which is non-linear with dose. Such an approach is achieved in this paper through a model (i.e., dose—response function) that makes the calculation of potential excess risks at various doses (accounting for dose-dependent changes in target tissue absorption) much simpler. More specifically, the dose—response function describing the predicted nonlinear relationship between excess risk and oral dose was obtained by modeling the human doses and excess risk estimates from Table 3 of Haney (2015b), reproduced for the current study as Table 1, using USEPA benchmark dose (BMD) software (BMDS version 2.5). An additional dose evaluated in Haney (2015b), the approximate adult human intake at the federal MCL (3E-03 mg/kg-day), was also included in Table 1. Consistent with the definition of excess risk, excess risk was set to zero at 0 mg/kg-day. The development of this model serves as a suitable example of the nonlinear, non-threshold low-dose extrapolation method for the current paper.

The non-linear threshold approach is represented in this paper

by the derivation of an RfD designed to be protective of the potential carcinogenic effects of oral exposure to CrVl. While this approach already appears in the peer-reviewed literature (Thompson et al., 2013b), differences in the analyses result in somewhat different RfD values. For example, the extensive physiologically-based pharmacokinetic (PBPK) modeling used in Thompson et al. (2013b) to derive the mouse internal dose metric for BMD modeling (flux of CrVl into target tissues) and ultimately the human point-of-departure (POD) could not be conducted for the current study. Consequently, an alternative analysis was performed, which can be used for comparison to the Thompson et al. (2013b) RfD (0.006 mg/kg-day).

The RfD for this paper was developed using standard dose—response assessment methodologies (e.g., TCEQ, 2012; USEPA, 1993) and USEPA BMD software (BMDS version 2.5), with diffuse hyperplasia as a key (i.e., necessary but not always sufficient) precursor event in the proposed non-mutagenic MOA for CrVl-induced carcinogenesis (Thompson et al., 2013a) since available data indicate that it precedes tumor formation in the mouse small intestine (e.g., Fig. 1 of Thompson et al., 2011a, Fig. 3 of Thompson et al., 2013b). Based on NTP (2008) data, diffuse hyperplasia only has a strong, well-defined dose—response relationship in the mouse duodenum (see Appendices C and D of NTP, 2008). This is consistent with both significant tissue absorption of CrVl by the duodenum (duodenum tissue concentration > jejunum > ileum; Table 8 of Kirman et al., 2012) and the duodenum as the most

Table 1

Human excess risk estimates adjusted for dose-dependent differences in absorption by target tissues.

Mouse dosea mg/kg-day (at water concentration)

SFo adjustment factorb

Human equivalent dose for modelingc mg/kg-day

Excess risk adjusted for dose-dependent absorption for modeling*1

4.3E-05 (35-city GM of 0.00018 mg/L) 3.1E-03 (highest city of 0.0129 mg/L) 8.0E-03 (l/3 MCL of 0.0333 mg/L) 1.8E-02

2.4E-02 (MCL of 0.1 mg/L) 1.165E+00 (male mouse POD/BMDL10)

2408 34 6.2 4.8 4.3 1

7.1E-06 1.5E-09

5.1E-04 7.9E-06

1.3E-03 1.1E-04

3.0E-03e (at MCL of 0.1 mg/L) 3.1E-04

3.9E-03 4.7E-04

1.9E-01 1.0E-01

a Doses from the federal MCL down to 1/3 MCL from Table 9 of Haney (2015a) or text of Haney (2015b), lower doses based on ratio of water concentration to MCL x dose at

MCL; BMDL10 from USEPA (2010). b From Table 9 or text of Haney (2015a, 2015b).

c Human dose = mouse dose/(70 kg/0.05 kg)025 except for one dose (see last footnote).

d Adjusted risk based on unrounded SFo of 0.525 per mg/kg-day (i.e., 0.1 risk/human dose of 0.1905) x human equivalent dose/SFo adjustment factor. e Corresponds to approximate human intake at the federal MCL: (0.1 mg/L x 2 L/day)/70 kg = 0.003 mg/kg-day.

tumorigenically responsive tissue. Therefore, the duodenum was selected as the critical mouse target tissue for BMD analysis. More specifically, the incidence of diffuse hyperplasia in the duodenum of female mice was used for BMD modeling since:

• Statistical analyses did not reveal differences between male and female mice in hyperplastic or tumorigenic response to CrVI exposure (Thompson et al., 2013b);

• The dose—response for diffuse hyperplasia in female mice is strong and more monotonic than that in male mice (see Tables C4 and D4 of NTP, 2008); and

• Importantly, the water concentrations used in NTP (2008) for female mice correspond to those used in Kirman et al. (2012) to determine added Cr concentrations in mouse target tissues due to CrVI oral exposure, which is a very useful internal dose metric for BMD modeling.

Accordingly, the incidence of diffuse hyperplasia in the duodenum of female mice from NTP (2008) along with the duodenum tissue concentrations (added mg Cr/kg tissue) reported in Kirman et al. (2012) were used for BMD modeling in the principal analysis. Additionally, the estimated 95% lower confidence limits (95% LCLs) on duodenum tissue concentrations were used for a supporting analysis for comparison of BMD results from the principal analysis with a more conservative, alternate BMD analysis. A benchmark response (BMR) of 10% was used so that the BMD and 95% lower confidence limit on the BMD (BMDL) would be calculated at a BMR that does not extrapolate farther than necessary below the range of the data. That is, so that the study data (e.g., a diffuse hyperplasia incidence of 32% (16/50) at the lowest dose) can provide support to the BMD10/BMDL10 values.

Lastly, while the more detailed data analyses and lengthy discussions found in stand-alone MOA analysis papers are beyond the scope of this paper, published MOA analyses and the underlying data (McCarroll et al., 2010; Thompson et al., 2011b, 2013a) are briefly reviewed to assess the overall weight-of-evidence for the most scientifically-supported MOA (i.e., mutagenic versus non-mutagenic/threshold) for CrVI-induced carcinogenesis due to oral exposure through summary and interpretation of the currently available scientific evidence and its strength. See Section 3.3.3.1 for additional details. This weight-of-evidence is used for selection of the most scientifically-defensible, non-linear approach (i.e., non-threshold versus threshold).

3. Results

3.1. Non-linear, non-threshold low-dose extrapolation approach

3.1.1. BMD modeling

The Weibull model had the lowest Akaike Information Criterion (A1C) and was the only model that provided fit to the dataset

Table 2

Weibull model response function fit to modeled dose—response data.

(Table 1) with a goodness-of-fit p value > 0.1 and scaled residuals <|2| (BMDS Wizard version 1.10, BMDS version 2.5). The equation and parameter values for the Weibull model dose—response function (provided by BMD software) fit to the full dataset are provided in Table 2. However, while the model fit to all doses in Table 1 (i.e., the full dataset) provided acceptable fit per goodness-of-fit criteria, visual inspection revealed a less-than-desirable fit at doses of interest for environmental risk assessment (Fig. 3). Dropping the highest dose substantially improved model fit at all doses. Importantly, this includes doses corresponding to water concentrations < the MCL, which are the doses of most interest for environmental risk assessment. The absolute values of scaled residuals for all doses decreased and the goodness-of-fit p value increased (Table 3), with substantially improved fit compared to the full dataset model confirmed by visual inspection for doses of interest (Fig. 3). Similarly, dropping the two highest doses also provided a better-fitting model compared to modeling the full dataset (see Fig. 3, parameter values provided in Table 2). However, compared to dropping only the highest dose: (1) this results in the loss of more information (an additional data point); (2) the scaled residual for zero dose increases; (3) the scaled residual for the lowest dose is not improved; (4) the scaled residuals for the two middle non-zero doses are only marginally improved; and (5) the scaled residual for the approximate human dose corresponding to the MCL increases over 2-fold, with the average of the absolute values of scaled residuals being slightly higher compared to only dropping the highest dose (Table 3). Moreover, higher doses should only be dropped for modeling one dose at a time until adequate model fit is obtained (USEPA, 2012), and only dropping the highest dose was necessary to accomplish satisfactory model fit by visual inspection, a goodness-of-fit p value > 0.1, and scaled residuals <|2|. For the reasons discussed above, the Weibull model fit dropping only the highest dose is deemed the most appropriate, best-fitting model. While models considered to provide overall poorer fits (using the full dataset or dropping the two highest doses) are also shown in Fig. 3 for comparison, these models are not considered further.

3.1.2. Non-linear, non-threshold dose—response model

Fig. 3 shows the best-fitting model for the doses modeled, which include those of most interest for environmental risk assessment (i.e., doses corresponding to water concentrations < the MCL). The model equation and parameter estimates for this best-fitting model (provided by BMD software) are provided in Table 2 and represent the non-linear, non-threshold low-dose extrapolation approach for this paper. Table 4 provides a comparison of the dose-specific excess risks used for modeling (from Table 1) to the excess risks predicted by the best-fitting model (Weibull model with highest dose in Table 1 dropped). Across doses, the average difference between the excess risks used for modeling and the excess risks predicted is 12.7%. Furthermore, the potential excess risks

Weibull model equation: Y [excess risk] = background + (1 - background)*(1 - 2.72( si°pe*doseP°wer))

Parameters

Values for Model fit with full set of dose—response dataa

Values for better-fitting Model (two highest doses dropped)3

Values for best-fitting model (high dose dropped)a

Example calculation

Dose (mg/kg-day) for user input to estimate potential excess risk

Background

7.08154 1.8147

for user input to estimate potential

excess risk

4.72847E-10

67.3438

2.10096

for user input to estimate

potential excess risk

6.95107E-11

38.3192

2.02543

Solve for Y [excess risk]:

0.003b

6.95107E-11 38.3192 2.02543 3.0E-04

See Table 1 for dose—response (i.e., human dose-excess risk) data; Weibull model with highest dose dropped is best-fitting model.

Corresponds to approximate human intake at the federal MCL.

Fig. 3. Non-linear, non-threshold model fit for potential human excess risk versus lower dose adjusted for dose-dependent differences in target tissue absorption.

Table 3

Goodness-of-fit P values and scaled residuals for Weibull model fit to full set of data and subsets.

Human equivalent dosea mg/kg-day Scaled residuals for model fit with full set of dose—response datab (p value = 0.3405) Scaled residuals for better-fitting model (two highest doses dropped)b (p value = 0.7822) Scaled residuals for best-fitting model (high dose dropped)b (p value = 0.8958)

0 0.000 -0.022 -0.008

7.1E-06c -0.955 0.001 0.001

5.1E-04d 0.144 -0.087 -0.111

1.3E-03 1.078 0.679 0.748

3.0E-03e 0.900 -0.148 0.073

3.9E-03 0.974 — -0.160

1.9E-01 -1.345 — —

Overall averagef: 0.771 0.187 0.183

a From Table 1.

b P values and scaled residual values from USEPA BMD software (BMDS version 2.5) output. c Human dose = mouse dose at 35-city drinking water GM of 0.00018 mg/L (EWG, 2010)/(70 kg/0.05 kg)0 25. d Human dose = mouse dose at highest water concentration of 0.0129 mg/L (EWG, 2010)/(70 kg/0.05 kg)0 25. e Corresponds to approximate human intake at the federal MCL. f Averages obtained using the absolute values of the scaled residuals.

predicted for doses corresponding to the MCL (0.1 mg/L) and typical drinking water concentrations (i.e., the 35-city GM of 0.00018 mg/L and highest city drinking water concentration of 0.0129 mg/L in EWG, 2010) are of particular interest to regulatory agencies and are remarkably within ±3.8% of the excess risks used for modeling (see

italicized values in Table 4). These results are indicative of a relatively practical and user-friendly non-linear, non-threshold model that overall, adequately describes the non-linearity in potential excess risk versus oral dose across the doses of most environmental interest.

Table 4

Comparison of excess human risks used for modeling versus excess risks predicted by the best-fitting model.

Human equivalent dosea mg/kg-day Excess risk modeleda Excess risk predicted by best-fitting modelb Difference (%)

7.1E-06c 1.5E-09 1.5E-09 0

5.1E-04d 7.9E-06 8.2E-06 3.8

1.3E-03 1.1E-04 5.5E-05 50

3.0E-03e 3.1E-04 3.0E-04 3.2

3.9E-03 4.7E-04 5.0E-04 6.4

Overall average: 12.7

a From Table 1.

b Weibull model with highest dose from Table 1 dropped is best-fitting model; equation provided in Table 2. c Human dose = mouse dose at 35-city drinking water GM of 0.00018 mg/L (EWG, 2010)/(70 kg/0.05 kg)0 25. d Human dose = mouse dose at highest water concentration of 0.0129 mg/L (EWG, 2010)/(70 kg/0.05 kg)0 25. e Corresponds to approximate human intake at the federal MCL.

3.2. Non-linear threshold approach

3.2.1. BMD modeling

The Log-Logistic and Dichotomous-Hill models provided adequate fit to the mouse data (Table 5) with a goodness-of-fit p value > 0.1 and scaled residuals <|2| (BMDS Wizard version 1.10, BMDS version 2.5), confirmed by visual inspection. These models provided almost identical fits (Fig. 4) with the lowest A1C value, highest goodness-of-fit p value, the same BMD10 values and very similar BMDL10 values (Table 6). The mouse BMD10 value was 1.83 added mg Cr/kg tissue for both models using mean added mg Cr/kg tissue as the internal dose metric. 1nterestingly, this is the same BMD10 value calculated in Thompson et al. (2013b) for their internal dose metric (BMD10 of 1.8 based on intestinal flux of mg Cr absorbed/kg tissue-day). Conservatively using the 95% LCL of duodenum tissue concentrations as the internal dose metric for modeling diffuse hyperplasia in the duodenum resulted in the very similar BMD10 value of 1.74 added mg Cr/kg tissue for both models. The BMDL10 values showed good agreement with their corresponding BMD10 values, all being within a factor of 1.5. Interestingly, the BMDL10 value for both models using mean added mg Cr/kg tissue as the internal dose metric (1.4 added mg Cr/kg tissue rounded to two significant figures) is the same as the BMDL10 value calculated in Thompson et al. (2013b) for their internal dose metric (BMDL10 of 1.4 based on intestinal fl ux of mg Cr absorbed/kg tissue-day). Furthermore, the BMDL10 values using mean added mg Cr/kg tissue as the internal dose metric were very similar to those using the 95% LCL of tissue concentrations (i.e., within a factor of 1.2). This increases confidence in the principal analysis and use of the average BMDL10 value of 1.39 for mean added mg Cr/kg duodenum tissue (Table 6) as the POD. Thus, a mouse BMDL10 of 1.39 added mg Cr/kg tissue will be used as the POD for diffuse hyperplasia in the duodenum for the derivation of an RfD.

3.2.2. Conversion of internal POD to external POD dose

As an internal dose metric, the mouse POD of 1.39 added mg Cr/ kg duodenum tissue must be converted to a corresponding oral dose (i.e., units of mg/kg-day) for derivation of the RfD. The relationship between duodenum tissue concentration (mean mg Cr/kg tissue) and oral dose (mg/kg-day) was modeled in a previous study (Haney, 2015a) using tissue concentration data reported by Kirman et al. (2012). The associated figure and table (Fig. 1 and Table 6 in Haney, 2015a) are reproduced in the current study as Fig. 5 and Table 7. Fig. 5 shows good fit for the Hill model (goodness-of-fit was evaluated by visual inspection with scaled residuals <| 2| and a goodness-of-fit p value > 0.1). The equation and parameter estimates for this dose—response function (provided by BMD software) were used to calculate the oral dose corresponding to the POD duodenum tissue concentration of 1.39 added mg Cr/kg tissue. 1mportantly, the POD tissue concentration falls between two of the

duodenum tissue concentrations modeled and is similar to one of the modeled concentrations (1.5 mg/kg tissue) where the estimated and observed values show excellent agreement (i.e., the scaled residual is 0.421, well below |2|), which increases confidence in the estimate at the POD. As shown in Table 7, a mouse oral dose of 0.31 mg/kg-day is estimated to correspond to the POD duodenum tissue concentration. Note that the application of an animal-to-human uncertainty factor to this mouse POD ultimately results in a value (0.031 mg/kg-day) that is below the lower end of the range of average human equivalent doses (HED values of 0.05—0.1 mg/kg-day) cited in a recent USEPA CrVl PBPK study (Sasso and Schlosser, 2015), practically identical to the more conservative HED of 0.028 mg/kg-day (pH = 5) based on a similar evaluation (e.g., using the BMDL10 for diffuse epithelial hyperplasia), and is 4.5-fold lower than the HED of 0.14 mg/kg-day (pH = 2.5) based on the similar evaluation (see Table 1 of Sasso and Schlosser, 2015). This mouse oral dose POD (0.31 mg/kg-day) will be used to derive an RfD designed to be protective of both non-carcinogenic and carcinogenic effects (i.e., using diffuse hyperplasia as a key precursor event to tumor formation in the small intestine, protecting against hy-perplasia should preclude tumorigenesis).

3.2.3. RfD derivation

An RfD is calculated by dividing the POD by applicable uncertainty factors to reflect data limitations and to derive a value that is below levels where health effects would be expected to occur (TCEQ, 2012). The three applicable uncertainty factors (UFs) are the:

• Animal-to-human uncertainty factor (UFA);

• Intrahuman uncertainty factor (UFH); and the

• Database uncertainty factor (UFD).

The values selected for these UFs are the same as those selected in USEPA (2010). A full UFA of 10 was used to account for inter-species differences in CrVl toxicokinetics and toxicodynamics. Additionally, a full UFH of 10 was used to account for intrahuman variability in toxicokinetics and toxicodynamics as such differences could lead to variability in susceptibility (e.g., proton pump inhibitor (PPl) use increases lifetime average daily dose by 7—10% depending on the internal dose flux metric used; Thompson et al., 2013b). Lastly, an UFD of 1 was used because the toxicity of ingested CrVl has been extensively studied and the database is robust, including many reproductive/developmental studies that have been conducted in multiple species (e.g., mice, rats, monkeys, rabbits). Importantly, the mouse endpoint used in this study for derivation of the RfD (diffuse hyperplasia in the duodenum in NTP, 2008) is more sensitive than the reproductive/developmental effects that occur at higher doses and was the most sensitive endpoint identified in USEPA (2010).

Accordingly, the RfD is calculated as follows:

Table 5

Added chromium duodenum concentrations in B6C3F1 mice.

Drinking water concentration3 (mg SDD/L) Duodenum tissue concentrationa (mean added mg Cr/kg tissue) ±SDa 95% UCLb (added mg Cr/kg tissue) 95% LCLc (added mg Cr/kg tissue) Number of animalsd (n) Diffuse hyperplasia"1 (# animals)

0 0 — — — 50 0

14 7.2 0.8 7.8 6.6 50 16

60 33.5 5.0 37.2 29.8 50 35

170 42.4 12.4 51.5 33.3 50 31

520 60.9 14.1 71.3 50.5 50 42

a From Table 8 of Kirman et al. (2012) with zero shown for added Cr at zero dose.

b 95%UCL = mean + (1.645 x SE) where SE = SD/n0 5 and n = 5. c 95%LCL = mean - (1.645 x SE) where SE = SD/n0 5 and n = 5.

d From Table D4 of NTP (2008).

Fig. 4. Diffuse hyperplasia incidence versus duodenum tissue concentration.

Table 6

BMD modeling results for diffuse hyperplasia in the duodenum of B6C3F1 mice.

BMD model3 Duodenum tissue concentration dose metricb AIC value Goodness-of-fit P value Scaled residual near BMD dose BMD10 (added mg Cr/kg tissue) BMDL10 (added mg Cr/kg tissue)

Log-logistic Mean added mg Cr/kg tissue 239.644 0.4658 0 1.83 1.41

Dichotomous-hill Mean added mg Cr/kg tissue 239.644 0.4658 0 1.83 1.37

Average: 1.83 1.39

Log-logistic 95% LCL added mg Cr/kg tissue 240.839 0.4379 0 1.74 1.21

Dichotomous-hill 95% LCL added mg Cr/kg tissue 240.839 0.4379 0 1.74 1.30

Average: 1.74 1.26

a USEPA BMDS version 2.5. b From Table 5.

RfD = POD/(UFa x UFH x UFD) = 0.31 mg/kg-day/ (10 x 10 x 1) = 0.0031 mg/kg-day

3.3. Weight-of-evidence for the most scientifically-supported carcinogenic MOA

3.3.1. Data relevant to the proposed cytotoxicity-induced regenerative hyperplasia MOA

Over the past few years, many new studies have been conducted to investigate and improve the scientific understanding of the MOA

for CrVl-induced carcinogenicity due to oral exposure. Some of these studies have already assimilated and evaluated relevant study data for MOA analyses under regulatory agency guidance (e.g., Thompson et al., 2011b, 2013a), while the results of other studies have yet to be included in a formal MOA analysis (e.g., Thompson et al., 2015a, 2015b, 2015c). Applying relevant guidance (e.g., USEPA, 2005) in a recent MOA analysis, Thompson et al. (2013a) present evidence in support of a non-mutagenic, cytotoxic MOA for CrVl carcinogenicity with the following key events:

• Absorption of CrVl from the intestinal lumen.

Fig. 5. Mouse duodenum tissue concentration versus daily dose.

Table 7

Duodenum best-fitting model tissue concentration prediction.

Hill model (non-constant variance) equation: Y [tissue conc. in mg Cr/kg at dose] = intercept + v*dosen/(kn + dosen)

Parameters

lnputs

Oral dose (mg/kg-day)

Intercept

Y [tissue conc. in mg Cr/kg at dose]

0.31 (oral POD)

62.397

1.39 (BMDL10)

• Toxicity to intestinal villi.

• Compensatory crypt regenerative hyperplasia.

• Clonal expansion of spontaneous mutations within the crypt stem cells due to chronic proliferation, resulting in late onset tumorigenesis.

Although reduction of CrVl to Crlll occurs in the gastrointestinal (Gl) tract (stomach, intestinal lumen) and is recognized as a detoxification process (e.g., De Flora et al., 1997), it is generally acknowledged that reduction of CrVl in the Gl tract prior to absorption and cellular absorption by target tissues are competing rates (e.g., Proctor et al., 2012; Kirman et al., 2013). Consequently, while these competing processes have recently been studied to better understand them for purposes of PBPK modeling (Proctor et al., 2012; Kirman et al., 2013), the occurrence of CrVl absorption from the intestinal lumen is not a topic of significant scientific debate. Accordingly, since absorption of CrVl from the intestinal lumen is known to occur and is the first event assumed (inherently or explicitly stated) in both the proposed non-mutagenic and mutagenic MOAs for CrVl-induced carcinogenesis (Thompson et al., 2013a; McCarroll et al., 2010), scientific evidence demonstrating this key event in the carcinogenicity study need not be presented here (e.g., see Appendix J of NTP, 2008 and Table 8 of Kirman et al., 2012).

3.3.1.1. Villous cytotoxicity. In regard to the duodenum as the tissue where most tumorigenesis occurred, NTP (2008) indicated that the

incidences of diffuse epithelial hyperplasia were significantly increased in the duodenum of all exposed groups of male and female mice (see Tables 13, C4, and D4 of NTP, 2008). By contrast, the incidence of diffuse epithelial hyperplasia in the jejunum was significantly increased only in the highest female mouse exposure group (516 mg sodium dichromate dehydrate (SDD)/L). The duodenal villi of exposed mice were described as short, broad, blunt, and lined by densely packed, tall columnar epithelial cells compared to those of the controls (see Plates 19 to 22 of NTP, 2008). Frequently, the epithelial cells and cell nuclei were piled up in multiple layers along the long axis of the villi and intestinal crypts were elongated and generally appeared to contain increased numbers of epithelial cells with increased numbers of mitotic figures. Study authors concluded these lesions to be consistent with regenerative hyperplasia secondary to previous epithelial cell injury (NTP, 2008). That is, the study authors themselves cite these villous effects as part of the evidence of cytotoxicity-induced regenerative hyperplasia having occurred in the mouse tissues where tumorigenesis occurred.

ln regard to early suggestive evidence of cytotoxicity, cyto-plasmic vacuolization occurred in the mouse duodenal villi at >170 and >60 mg SDD/L on days 8 and 91, respectively (see Table 4 of Thompson et al., 2013a). Villous atrophy and crypt hyperplasia, on the other hand, both occurred in the duodenum at 520 and >170 mg SDD/L on days 8 and 91, respectively. Generally, these effects did not occur in the jejunum on day 8 and occurred to a lesser extent on day 91 (e.g., lower incidences of crypt hyperplasia in the jejunum at these doses compared to the duodenum). Both crypt area and the number of enterocytes per crypt in the duodenum were significantly increased at >170 mg SDD/L on day 91, but crypt hyperplasia was unable to compensate for the apparent villous damage (e.g., atrophy, blunting, and cytoplasmic vacuoli-zation at >170 mg SDD/L on day 91) and maintain normal, healthy duodenal villi (see Fig. 4(A), (D), and (F) of Thompson et al., 2013a). As opposed to villous toxicity, cytotoxicity to crypt cells was not evident (e.g., decreases in crypt area and the number of enterocytes per crypt, not increases, would be expected as a result of crypt enterocyte cytotoxicity; lack of treatment-related effects on crypt mitotic and apoptotic indices). Thus, data indicate that CrVl-

induced cytotoxicity in the intestine occurs in the villi, and primarily in the duodenum (consistent with the incidence of mouse tumors).

In regard to genotoxicity, karyorrhectic nuclei (KN) and micro-nuclei (MN) were assessed in both the crypt and duodenal villi at doses corresponding to water concentrations up to 520 mg SDD/L for up to 90 days. In vivo assays are useful for further investigation of genotoxicity produced in in vitro systems, and the mammalian in vivo MN test is especially relevant for assessing genotoxicity because although they may vary among species, factors of in vivo metabolism, pharmacokinetics, and DNA repair processes are active and contribute to the responses (OECD, 2014). There were no treatment-related effects in duodenal crypts. Bycontrast, in duodenal villi, KN (clustered at the villi tips) and MN were significantly increased at > 170 and 520 mg SDD/Lon day 8 and at >60 and >170 mg SDD/Lon day 91, respectively (see Tables 5 and 6 of Thompson et al., 2013a). These duodenal villi effect level data for cytogenetic damage correspond to those discussed above for cytotoxicity and indicate that the MN present in villi did not originate in the crypt. Data from a more recent 7-day exposure study (including a positive control group) also showed no treatment-related effects on MN (MN, KN, or g-H2AX immunostaining) in duodenal crypts (Thompson et al., 2015c). Accordingly, available data indicate that both CrVl-induced cytotoxicity and genotoxicity occur in the duodenal villi, not the crypt. This is consistent with villous toxicity-induced regenerative hyperplasia in the crypt, absent any direct cytotoxicity or genotoxicity in the crypt itself.

Lastly, while there are data to suggest that absorbed CrVl may have induced intestinal oxidative stress throughout the 2-year mouse NTP (2008) study and contributed to cytotoxicity in the intestinal villi, these data are not essential for the demonstration of the proposed key events or for the limited purpose of this paper and are therefore not considered further here (see Thompson et al., 2013a).

3.3.1.2. Crypt hyperplasia. Cytotoxicity-induced regenerative proliferation may be viewed as a necessary, but not always sufficient, event for tumor formation for chemicals with a non-genotoxic/ cytotoxic carcinogenic MOA (Butterworth et al., 1995). Accordingly, the occurrence of hyperplasia in target tissues such as the mouse duodenum is relevant as long-term oral exposure to CrVl may facilitate tumorigenesis through the induction of prolonged regenerative cellular proliferation, increasing replication errors and causing the accumulation of spontaneous mutations within stem cells that reside in the base of the crypts (O'Brien et al., 2013). As mentioned in the previous section, crypt hyperplasia occurred in the duodenum at 520 and >170 mg SDD/L on days 8 and 91, respectively. This is consistent with the number of enter-ocytes per crypt being statistically significantly increased and the duodenal crypt area being increased = 45% following 7-day exposure to 520 mg SDD/L (Thompson et al., 2015c; O'Brien et al., 2013), and the crypt area being similarly increased following 7-day exposure to 170 mg SDD/L with both crypt area and the number of enterocytes per crypt in the duodenum being statistically significantly increased at >170 mg SDD/L on day 91 (O'Brien et al., 2013; see Fig. 4(D) and (F) of Thompson et al., 2013a). In the jejunum, crypt hyperplasia occurred at >170 mg SDD/L on day 91 (see Table 4 of Thompson et al., 2013a). The increased incidence of crypt hyperplasia in the duodenum at 170 mg SDD/L (90%) and 520 mg SDD/L (90%) on day 91 is reasonably consistent with the increased incidence of diffuse hy-perplasia (74—88%) at those water concentrations in the 2-year NTP study (e.g., see Table D-4 of NTP, 2008). While Thompson et al. (2013a) did not report an increased incidence of crypt hyperplasia in the duodenum at 60 mg SDD/L on day 91, there was a statistically significant increase in the number of enterocytes per

crypt following 7-day exposure to 60 mg SDD/L as well as a relatively comparable increase following 90-day exposure (although it did not achieve statistical significance), and crypt area was also increased (z45%) at 60 mg SDD/L on day 91 (see Table 1 and Fig. 1(F) of Thompson et al., 2015c and Fig. 2(D) and (F) of O'Brien et al., 2013). Moreover, the recent Cullen and Ward (2015) study does report an increased incidence of crypt hyperplasia for the CrVl research project study at 60 mg SDD/L on day 91 (30%) as well as for the 2-year NTP study (40%) at the same water concentration where diffuse hyperplasia was reported by NTP (2008). lncreased crypt hyperplasia and diffuse hyperplasia in the duodenum at >60 mg SDD/L and in the jejunum at >170 mg SDD/L appear consistent with the water concentrations associated with tumorigenesis (e.g., see Table D2 of NTP, 2008).

As a marker for crypt cell proliferation, the increases in Ki67 expression at 60 mg SDD/L on day 8 and at >170 mg SDD/L on days 8 and 91 add to the already overwhelming weight-of-evidence that CrVl induces crypt hyperplasia at tumorigenic doses (see Fig. 4(H) of Thompson et al., 2013a). Lastly, while it may be hypothesized that CrVl-induced hyperplasia occurs subsequent to a mutation (e.g., McCarroll et al., 2010), not only is there a demonstrated lack of genotoxicity (e.g., increased MN, Kras mutation frequency, g-H2AX immunostaining) in the crypts of exposed mice, but the strong dose—response for diffuse hyperplasia without a commensurate dose—response for focal hyperplasia is inconsistent with this hypothesis (e.g., see Table D4 of NTP, 2008). By contrast, the available data are consistent with regenerative proliferation secondary to cytotoxicity, which was also the conclusion of the NTP (2008) study authors. Additional discussion of data and information relevant to this key event in the proposed MOA (e.g., increased Myc expression and signaling, duodenal crypt mitotic index (Ml) and apoptotic index (Al)) may be found in Thompson et al. (2013a) and related papers (e.g., Kopec et al., 2012a, 2012b).

3.3.1.3. Mutagenesis in crypt cells. The detection of gene mutations in target tissues (that initiate the carcinogenic process) following in vivo chemical exposure through an environmentally-relevant route is the strongest evidence for a mutagenic MOA. O'Brien et al. (2013) measured Kras codon 12 GAT mutation frequency in the duodenal epithelium of mice exposed to 0.3—520 mg SDD/L drinking water for 90 days to assess the potential for CrVl-induced mutation in the small intestine. Kras was selected because:

• lt is often mutated early in human intestinal tumors;

• Codon 12 GGT to GAT mutation is commonly reported in human duodenal tumors and accounts for approximately 12.6% of colon tumors;

• lt has been shown to contribute to small intestine tumorigenesis in mice; and

• Results from several studies suggest Kras codon 12 GAT mutation frequency to be a functional reporter of tumor initiation and/or progression that is amplified during carcinogenesis with mutational loading, perhaps even when the mutational specificity of the mutagen is other than the G to A mutation (Thompson et al., 2013a).

lf mutation preceded and induced crypt proliferation (e.g., McCarroll et al., 2010), then the increase in crypt cells could also increase the likelihood of detecting Kras mutations. However, there were no treatment-related effects on Kras codon 12 GAT mutation frequency, even at carcinogenic CrVl doses that increased crypt proliferation with 7- or 90-day exposure (see Fig. 6(A) of Thompson et al., 2013a). The absence of these effects does not support mutation being the inducer of crypt proliferation or an early, initiating key event in the carcinogenic MOA.

Similarly, Thompson et al. (2013a) suggest that if mutation precedes proliferation, then given crypt hyperplasia as early as day 8 and the high incidence by day 90, alterations in adenomatous polyposis coli (Apc) might be occurring early on (inactivating mutations in Apc results in uncontrolled cell proliferation and early onset of mouse small/large intestine adenomas through constitutive activation of Wnt/b-catenin signaling pathway). Kopec et al. (2012a, 2012b) collected toxicogenomic data in mice and rats exposed to CrVl up to 520 mg SDD/L for up to 90 days. However, there was no indication of changes in Apc expression that might result from genetic or epigenetic silencing of the Apc gene, no increase in b-catenin (Ctnnbl) expression level or indication of Ctnnbl activation (by transcription factor analysis) that would be expected by loss of Apc, nor was there functional enrichment of Wnt/b-catenin signaling pathways. The absence of these changes in the small intestine of CrVl-exposed mice suggests that crypt hy-perplasia and tumorigenesis are not likely the result of early genetic or epigenetic changes related to Apc.

Data from more recent studies also support that DNA damage is not an early, initiating key event in CrVl-induced tumorigenesis (Thompson et al., 2015a, 2015b, 2015c). For example, Thompson et al. (2015a) used X-ray fluorescence (spectro)microscopy (m-XRF) to assess Cr content in the villi and crypt regions of the duodenum from mice exposed to CrVl in drinking water (520 mg SDD/ L) for 13 weeks. Additionally, g-H2AX immunostaining was used to assess DNA damage. As observed previously, CrVl exposure induced villous blunting and crypt hyperplasia in the duodenum. g-H2AX immunostaining was elevated in villi, but not in the crypt compartment, consistent with m-XRF maps revealing mean Cr levels >30 times higher in duodenal villi than in crypts (mean Cr levels in crypt regions were reported to be only slightly above background). Despite the Cr and elevated g-H2AX immunoreac-tivity in villi, no aberrant foci indicative of transformation were evident (see Table S4 of Thompson et al., 2015a). These in vivo study findings provide additional support for a non-mutagenic MOA involving chronic wounding of intestinal villi and crypt cell hy-perplasia as opposed to an MOA involving direct Cr-DNA interaction in intestinal stem cells as an initiating key event for crypt proliferation and tumorigenesis.

ln a drinking water study to assess crypt health along the entire length of the mouse duodenum, mice were exposed up to 520 mg SDD/L drinking water for 7 days (Thompson et al., 2015c). Crypt enterocytes in "Swiss roll" sections were scored as normal, mitotic, apoptotic, karyorrhectic, or as having MN (an oral gavage of 50 mg/ kg cyclophosphamide served as a positive control for MN induction). Exposure to 180 and 520 mg SDD/L significantly increased the number of crypt enterocytes, whereas MN and g-H2AX immuno-staining were not increased in the crypts of dosed mice. Treatment with the cyclophosphamide, on the other hand, was reported to significantly increase crypt MN and qualitatively increase g-H2AX immunostaining. While synchrotron-based X-ray fluorescence (XRF) microscopy revealed strong Cr fluorescence in duodenal villi, negligible Cr fluorescence was present in the crypt compartment. The data from this in vivo study do not support that CrVl adversely affects (i.e., damages DNA in) the crypt compartment where intestinal stem cells reside as an early, initiating key event for crypt proliferation and tumorigenesis. Rather, as negative genotoxicity data in target tissue, these data add to the weight-of-evidence that the MOA for CrVl-induced tumorigenesis in the mouse small intestine involves compensatory crypt enterocyte hyperplasia induced by chronic villous toxicity.

Additional data and relevant information (e.g., weight-of-evidence support from genomic fingerprinting for mutagenic versus non-mutagenic carcinogens, lack of increased MN or g-H2AX immunostaining in the crypt) may be found in Thompson

et al. (2013a) and related papers (e.g., Thompson et al., 2012b, 2015a, 2015c).

3.3.2. Data relevant to the proposed mutagenic MOA

McCarroll et al. (2010) conclude that the weight-of-evidence supports the plausibility that orally administered CrVl may act via a mutagenic MOA for carcinogenicity. However, as these study authors did not have the benefit of being able to review and weigh the substantial data generated through the CrVl MOA research project, this conclusion was based primarily on studies designed for hazard identification as opposed to MOA analysis, and on data from non-target tissues as well as in vitro systems. Regardless, McCarroll et al. propose the following MOA key events:

• lnteraction of absorbed CrVl with DNA

• Mutagenesis

• Cell proliferation

• Tumorigenesis

The ability of absorbed CrVl reduced intracellularly to interact with macromolecules such as DNA is not a topic of significant scientific debate. However, the potential role of CrVl-induced muta-genicity in the MOA for tumorigenesis/carcinogenesis in the small intestine of mice orally exposed to CrVl is the subject of scientific debate. Therefore, since the downstream events (i.e., cell proliferation, tumorigenesis) in the proposed mutagenic MOA depend entirely upon the ability of CrVl to induce mutations that initiate carcinogenesis in target tissues, evidence relevant to this topic is discussed here.

3.3.2.1. CrVl-induced mutagenicity and genotoxicity

3.3.2.1.1. In vitro. CrVl compounds are positive in the majority of in vitro mammalian cell line tests, with their genotoxicity being related to solubility/bioavailability to the targets (ATSDR, 2012). Furthermore, genetic toxicology data show that CrVl can act as an in vitro mutagen, causing point mutations in Salmonella typhimu-rium and Escherichia coli and a concentration-response in mutant colonies at nonactivated concentrations down to noncytotoxic levels in these bacteria (see Table 1 of McCarroll et al., 2010). Mutation formation (i.e., increased reversions) in these bacteria is mitigated by the addition of S9, consistent with the extracellular reduction of CrVl to Crlll by constituents of the microsomal S9 mix and findings that Crlll itself does not induce revertants (Thompson et al., 2013a). McCarroll et al. also report CrVl to be mutagenic in Saccharomyces cerevisae, mammalian cell lines (e.g., Chinese hamster ovary (CHO)), and mouse lymphoma cells, and clastogenic in cultured CHO cells, mouse mammary FM3A carcinoma cells, and human lymphocytes. S9 activation is generally not required to detect the mutagenic effects seen in the in vitro studies. Thus, at least in vitro, data indicate that CrVl has the capacity to be a direct-acting mutagen (McCarroll et al., 2010).

ln regard to other genotoxicity demonstrated in vitro, CrVl has been shown to cause DNA damage/repair in bacteria, DNA-protein crosslinks, DNA strand breaks and adduct formation in mammalian and/or human cells, sister chromatid exchange in CHO cells, and unscheduled DNA synthesis in human lymphocytes and fibroblasts, with DNA adducts being the most abundant form of genetic lesions in mammalian cells. Based on this information, McCarroll et al. conclude that CrVl is generally positive in vitro for multiple geno-toxic endpoints. However, interpreting the relevance of in vitro results to the in vivo conditions of interest presents challenges and appears particularly problematic for CrVl (e.g., non-target tissue results versus actual target tissues, in vitro conditions versus in vivo drinking water exposure ad libitum where extracellular reduction of CrVl occurs and competes with absorption along the Gl tract,

prokaryotic cell results versus eukaryotic cells where intracellular reduction of CrVl can result in cytoplasmic trapping outside the nucleus, roles of oxidative stress, oxidative DNA damage, and epigenetic effects such as DNA hypermethylation in producing genotoxicity in vitro as opposed to direct DNA reactivity). Consequently, the examination of in vivo data and their relevance to the exposure of interest is important.

3.3.2.1.2. In vivo. McCarroll et al. (2010) rely on two assays to demonstrate that CrVl has the ability to induce mutations in vivo. ln the somatic cell mouse spot test, statistically significant positive results (i.e., increased number of coat spots due to loss of the dominant allele for coat color pigmentation through mutation or another genetic event) were obtained in female mice dosed by intraperitoneal (i.p.) injection with 10 mg/kg-day potassium chro-mate on gestational days 8—10, but not 20 mg/kg-day potassium chromate (see Table 2 of Knudsen, 1980). ln the transgenic Muta™ Mouse assay with the lacZ gene as the mutational target, 40 mg/kg potassium chromate via i.p. injection increased mutational frequency in the bone marrow and liver (ltoh and Shimada, 1998). However, results from Knudsen (1980) are equivocal (i.e., effects at 10 but not 20 mg/kg-day), and regardless, McCarroll et al. acknowledge that data generated via the i.p. exposure route are not relevant to the oral route of exposure. This is especially true given that bolus i.p. doses initially bypass the liver and also bypass CrVl reduction in the Gl tract as an important protective detoxification mechanism, especially at doses within the reductive capacity of the Gl tract (e.g., De Flora et al., 1997), and that the mouse tissues shown to be mutagenically responsive in these studies (i.e., mela-noblasts, bone marrow, liver) are not the ones shown to be tumorigenically responsive following oral exposure in NTP (2008) (e.g., the small intestine/duodenum). Thus, these studies only appear to show that environmentally- and physiologically-irrelevant exposure to CrVl is capable of inducing mutation in genes that do not initiate tumorigenesis in tissues that are not relevant to the target tissues of concern for potential carcinogen-esis. For example, although mutation occurred in the skin and liver of mice exposed to doses of 2.7—5.4 mg CrVl/kg and 14 mg CrVl/kg, respectively, a range essentially covering most of the tumorigenic doses in the NTP (2008) study (e.g., 1.4—8.7 mg CrVl/kg-day in female mice), CrVl did not induce tumors in these tissues (e.g., see Table D1 of NTP, 2008). lt stands to reason that if mutagenicity (or other genotoxicity) occurs in tissues that do not subsequently develop tumors (e.g., liver, skin), then that mutagenicity is not a key event (or a predictor of a key event) that initiates tumorigenesis for the chemical in question so does little to elucidate key events in how the chemical initiates cancer in target tissues, much less being indicative of a mutagenic MOA. This applies to the high NTP study doses and even more so to environmentally-relevant doses, which are orders of magnitude lower (e.g., the mouse dose at the lowest water concentration used in NTP, 2008 is over 70,000 times higher than the approximate human dose corresponding to the 35-city geometric mean drinking water concentration reported in EWG, 2010), considering the lack of mutagenicity/genotoxicity in mouse duodenal crypt target tissue (e.g., Thompson et al., 2013a, 2015c) at doses corresponding to drinking water concentrations that induced tumorigenesis in NTP (2008) but still orders of magnitude higher than those applicable to humans.

McCarroll et al. also summarize other in vivo gentoxicity data (see Table 2 of McCarroll et al., 2010). Although the weight-of-evidence for mutagenic potential and its relevance to the MOA cannot be determined simply by the number of positive versus negative genotoxicity study results (i.e., there are many important considerations), rat data are mostly positive and show chromosomal aberrations, DNA-protein crosslinks, and DNA single strand breaks in bone marrow, liver, and liver and brain, respectively.

These effects were generally observed following oral gavage exposure durations that varied from a single dose up to daily doses for a year, with the exception of the liver DNA-protein crosslinks that were found after three weeks of exposure via drinking water (Coogan et al., 1991). The mouse data reviewed are also mostly positive and show DNA strand breaks, MN induction and chromosomal aberrations, and DNA fragmentation in leukocytes, bone marrow, and liver and brain, respectively. DNA strand breaks, fragmentation, and chromosomal aberrations were observed following oral gavage of single doses. While MN were induced in the bone marrow of MS/Ae and CD-1 mice by i.p. injection of 20—80 mg K2CrO4/kg, oral doses higher than the species-specific LD5o values (i.e., doses as high as 160—320 mg K2CrO4/kg) did not induce MN when given by oral gavage in the same study (Shindo et al., 1989). ln regard to environmentally-relevant exposure, McCarroll et al. point out that the three 3-month mouse drinking water studies cited (including NTP, 2007) were negative for MN in bone marrow. Additionally, MN were not induced in the bone marrow, peripheral blood, and/or livers of BDF1 mice and pregnant Swiss mice (or their fetuses) exposed via drinking water up to 500 mg CrVl/L (5000 times the federal MCL) for up to 210 days or by intragastric administration of 17.7 mg CrVl/kg, whereas a single i.p. injection of 17.7 mg CrVl/kg bypassing the environmentally- and physiologically-relevant oral exposure route was capable of producing positive results (De Flora et al., 2006).

As mentioned previously, McCarroll et al. acknowledge that data generated via the i.p. exposure route are not relevant to the assessment of oral exposure, although the study authors deem the data to support that CrVl is positive for genotoxicity in vivo and provide qualitative information for hazard identification. However, these i.p. data only support positive in vivo genotoxicity through an exposure route that is environmentally- and physiologically-irrelevant and contrived to bypass the liver and the CrVl reduction that occurs in the Gl tract. ln addition to environmentally-irrelevant dosing, these data and the bolus oral dose data only show genotoxicity in tissues that are not relevant to the target tissues of concern for potential carcinogenesis (NTP, 2008). For example, although DNA fragmentation occurred in the liver and brain of mice exposed at doses (1.9,19, and 95 mg/kg presumably as Na2Cr2O7 in Bagchi et al., 2002; which would correspond to 0.75, 7.5, and 38 mg CrVl/kg) apparently encompassing the range of tumorigenic doses in the NTP (2008) study (1.4—8.7 mg CrVl/kg-day in female mice), Bagchi et al. attribute this to oxidative damage rather than direct DNA interaction, and moreover, CrVl did not induce tumors in these tissues (see Table D1 of NTP, 2008). Similarly, while chromosomal aberrations (i.e., primarily breaks, but some rearrangements) occurred in bone marrow at a gavage dose (20 mg CrO3/kg or 10.4 mg CrVl/kg in Sarkar et al., 1993) similar to the tumorigenic dose for the highest female mouse exposure group in the NTP study (8.7 mg CrVl/kg-day) and DNA strand breaks occurred in the leukocytes of mice at a gavage dose range (0.59—9.5 mg K2Cr2O7 or 0.21—3.4 mg CrVl/kg in Danadevi et al., 2001) similar to that in the NTP study, the NTP study identified the small intestine (the duodenum in particular) as the target tissue for CrVl-induced tumorigenesis. Furthermore, DNA damage did not occur in lymphocytes at much higher i.p. doses (3.3—27 mg CrVl/ kg) or higher oral doses (z 6—9 mg CrVl/kg-day) in rats exposed for 3—6 weeks via drinking water (Coogan et al., 1991), or in the leukocytes of human volunteers ingesting 5 mg CrVl as a single bolus or 4 mg CrVl/day via drinking water for 15 days (2 L/day of 2 mg CrVl/L drinking water for 15 days; Kuykendall et al., 1996). Again, if genotoxicity (or mutagenicity) occurs in tissues that do not subsequently develop tumors (e.g., liver, brain), then it stands to reason that the genotoxicity is not a key event (or a predictor of a key event) that initiates tumorigenesis for the chemical in question so

does little to elucidate key events in how the chemical initiates cancer in target tissues, much less being indicative of a mutagenic MOA. This applies to the high NTP study doses, and even more so to orders of magnitude lower environmentally doses, considering that doses corresponding to drinking water concentrations that induced tumorigenesis in NTP (2008) did not induce mutagenicity/geno-toxicity in mouse duodenal crypt target tissue (e.g., Thompson et al., 2013a, 2015c).

3.3.3. Carcinogenic MOA guidance and weight-of-evidence assessment

3.3.3.1. MOA guidance considerations. In dose—response assessment, the approach for extrapolation below the observed data (i.e., threshold versus non-threshold) is based on the understanding of the chemical's MOA at each tumor site (USEPA, 2005). There is no default carcinogenic MOA, even for chemicals demonstrating mutagenic activity when data on other possible carcinogenic MOAs are lacking (USEPA, 2007). Furthermore, simply demonstrating plausibility or even a positive weight-of-evidence for mutagenic activity/potential is insufficient to conclude that the MOA is in fact mutagenic (TCEQ, 2012). Such a low scientific standard of proof for demonstration of a mutagenic MOA is contrary to USEPA guidance (USEPA, 2007) that states [emphasis added]:

The determination that a chemical carcinogen is capable of producing mutation is not sufficient to conclude that it causes specific tumors by a mutagenic MOA or that mutation is the only key event in the pathway to tumor induction. For a chemical to act by a mutagenic MOA, either the chemical or its direct metabolite is the agent inducing the mutations that initiate cancer.

This sets a reasonably scientifically-rigorous standard for demonstration of a mutagenic MOA. Most specifically, to demonstrate a mutagenic MOA for a specific tumor the weight-of-evidence of scientific information must sufficiently support that "either the chemical or its direct metabolite is the agent inducing THE mutations that initiate cancer [emphasis added]" at the tumor site. At the most basic level, this requires:

(1) Linking the chemical or a metabolite to mutations; and

(2) Linking those mutations induced by the chemical to the initiation of cancer in target tissues (TCEQ, 2012).

3.3.3.2. Weight-of-evidence assessment

3.3.3.2.1. Mutagenic MOA. For the present assessment of the weight-of-evidence for the most scientifically-supported MOA, it is noted that:

• McCarroll et al. (2010) acknowledge that data best suited to support many aspects of the MOA framework were not available at that time;

• Since that time, the CrVl MOA research project has generated a great deal of data relevant to this very issue;

• The demonstration of mutagenicity/genotoxicity in non-target tissues, especially through environmentally- and physiologically-irrelevant exposure routes (i.p.), is not tantamount to demonstrating CrVl-induced mutagenicity in target cells as the key initiating event for carcinogenesis in the target tissues of animals exposed through drinking water in vivo; and

• Such a demonstration is particularly irrelevant to the possible MOA and potential for tumorigenesis in target tissues at environmentally-relevant drinking water doses (e.g., mean of 0.001 mg/L based on 18,085 samples (see Table S5 of Thompson et al., 2015a) x 2 L/day x 1/70 kg = 2.9E-05 mg CrVl/kg-day) that are orders of magnitude lower than even experimental

drinking water doses not associated with target tissue geno-toxicity (e.g., O'Brien et al., 2013; Thompson et al., 2015a, 2015c).

ln regard to (1) above, the demonstration of CrVl-induced mutations (or even genotoxicity), some high dose mouse oral gavage and i.p. studies have produced genotoxicity in non-target tissues (see Table 2 of McCarroll et al., 2010). However, CrVl-induced mutations have only been demonstrated through an environmentally-and physiologically-irrelevant exposure route (i.p.) and only in nontarget tissues (Knudsen, 1980; Itoh and Shimada, 1998). By contrast and more importantly, entirely relevant drinking water studies have not found CrVl-induced mutagenicity (or genotoxicity) in the target tissue duodenal crypts (e.g., O'Brien et al., 2013; Thompson et al., 2015a, 2015c). This is not particularly surprising given that the conditions under which some chemicals may be shown to cause genotoxicity (e.g., carcinogenic study- or environmentally-irrelevant high doses or exposure routes, non-target cells or tissues, in vitro) may not necessarily be predictive of mutagenic effects in the carcinogenic study laboratory animal target tissues or the target tissues of humans exposed to environmentally-relevant doses (TCEQ, 2012). Thus, the relevance of the positive results, obtained in non-target tissues under unrealistic conditions contrived to produce positive results (e.g., environmentally- and physiologically-irrelevant i.p. exposure), to the MOA in target tissues under realistic exposure conditions (i.e., environmental exposure through drinking water) is questionable at best. Based on currently available evidence, sound scientific judgment dictates that the negative mutagenicity/genotoxicity results obtained in target tissues under conditions that are actually relevant to the MOA operating in the NTP (2008) mouse study (i.e., at oral doses relevant to the study and through the relevant exposure route) should weigh more heavily in the weight-of-evidence for the most scientifically well-supported MOA. For example, the following mutagenicity/genotoxicity findings weigh heavily against data previously construed to support that CrVl is inducing tumorigen-esis in the mouse small intestine through mutagenicity as the initiating, key event:

• The absence of increases in Kras mutation frequency effects;

• The absence of effects on Apc expression; and

• The absence of cytogenetic damage (e.g., elevated g-H2AX immunoreactivity, MN) in the duodenal crypts of mice exposed in vivo to tumorigenic doses (e.g., Thompson et al., 2015a, 2015c).

Rather, these data provide support to a non-mutagenic MOA weight-of-evidence and are consistent with the absence of early tumors and metastases in the 2-year NTP (2008) study, as well as the absence of neoplastic lesions in the 90-day drinking water studies (NTP, 2007; Thompson et al., 2011a, 2012a).

ln regard to (2) above, linking CrVl-induced mutations (or even genotoxicity) to the initiation of cancer in relevant target tissues (e.g., mouse duodenum), there is scientific evidence to the contrary (i.e., CrVl did not induce genotoxicity much less mutagenicity in duodenal crypt target tissue). As alluded to above, relevant drinking water studies designed to address this very question did not find mutagenicity (or genotoxicity) in the target tissue duodenal crypts (e.g., O'Brien et al., 2013; Thompson et al., 2015a, 2015c). Additionally, although mutagenicity and genotoxicity have been induced in some tissues under contrived conditions (e.g., liver, skin, brain), CrVl did not induce tumors in those tissues in the carcinogenicity study (NTP, 2008), which is also antithetical to the demonstration of such a link. So the tissues where CrVl has been shown to induce mutagenicity or genotoxicity in some studies did not develop tumors in the NTP study, and CrVl-induced

mutagenicity and genotoxicity were not found in the target tissue (duodenal crypts) where tumors subsequently developed in NTP (2008) in CrVl MOA studies designed specifically to address this issue. These facts are directly contradictory to the assertion by McCarroll et al. that DNA damage in mouse leukocytes and rat livers (Danadevi et al., 2001; Coogan et al., 1991) show a causal relationship between DNA damage and tumor induction (in different tissues no less) for the NTP (2008) study.

ln summary, current evidence cannot be reasonably construed to support a weight-of-evidence that CrVl induces mutations that are THE initiating event in CrVl-induced carcinogenesis of the small intestine, which is the reasonably scientifically-rigorous standard for demonstration of a mutagenic MOA (USEPA, 2007; TCEQ, 2012). Based on the current database, the lack of relevant data demonstrating even a prima facie case for a link between mutations (or even genotoxicity) induced by CrVl and the initiation of tumori-genesis in the mouse small intestine (e.g., duodenum) renders additional discussion of a mutagenic MOA and the limitations of the McCarroll et al. (2010) analysis unnecessary (e.g., the dose—response for DNA strand breaks in mouse leukocytes is not particularly relevant for showing dose—response and time concordance between mutagenicity and tumorigenesis in the small intestine, particularly given that at the time it was known that a 9-month mouse dose of 4.7 mg CrVl/kg-day from drinking water (a clearly tumorigenic dose in NTP, 2008) was reported not to cause target tissue genotoxicity in De Flora et al., 2008). lnterested readers may refer to Thompson et al. (2013b) for a discussion of their perspectives on the limitations of the McCarroll et al. (2010) study, which complement the data-based scientific arguments made herein. Based on the objective evaluation of currently available information, it is concluded that the hypothesis that mutation is the initiating key event in CrVl-induced intestinal carcinogenesis is inconsistent with the weight-of-evidence (e.g., MOA data collected from target tissue following relevant exposure, late tumor onset in the NTP study, etc.).

3.3.3.2.2. Non-mutagenic MOA. Studies recently conducted for the CrVl MOA research project provide critical target tissue-specific data and a basis for MOA analysis that is much more relevant and scientifically robust than that available previously. This new MOA information addresses the major data gaps cited by McCarroll et al. (i.e., target tissue studies for temporal/dose—response concordance, cell proliferation/hyperplasia studies, reductive capacity studies). For example, McCarroll et al. (2010) indicate that in vivo mutation and duodenal hyperplasia studies with adequate dose selection, sampling and harvest times, and consideration of site concordance and exposure conditions would help inform the overall analysis, and such studies are now available (e.g., O'Brien et al., 2013; Thompson et al., 2013a, 2015a, 2015c). These studies show that to continue to rely on a hodgepodge of disjointed data, much of limited relevance, to support theoretical plausibility as a foundation for establishing a mutagenic MOA in the face of more directly relevant and deterministic MOA study data (e.g., O'Brien et al., 2013; Thompson et al., 2015a, 2015c) would be against the scientific weight-of-evidence based on currently available information. For example, while McCarroll et al. (2010) cite the lack of target tissue studies as a database weakness, study authors still conclude that evidence of in vitro and in vivo mutagenicity in the absence of cytotoxicity rules out cytotoxicity as a likely MOA. However, mutation was only induced by CrVl in vivo through an environmentally- and physiologically-irrelevant exposure route (i.p.) and only in non-target tissues (Knudsen, 1980; ltoh and Shimada, 1998), and the study authors acknowledge that data generated via the i.p. exposure route are not relevant to the assessment of oral exposure. Additionally, evidence that a chemical is capable of producing mutation or other DNA damage under

in vitro conditions (e.g., even at sub-cytotoxic concentrations in bacterial strains) represents only the lowest hierarchal evidence to be considered along with all other relevant evidence (USEPA, 2007; TCEQ, 2012). On the other hand, target tissue data are highest on the scientific evidence hierarchy for evaluating the likely MOA (especially when collected following relevant exposure), and contrary to ruling out cytotoxicity as a MOA, provide support that:

• Mutagenicity/genotoxicity is not induced by CrVl in the target tissue (duodenal crypts) of mice exposed to tumorigenic oral doses through the relevant exposure route (drinking water);

• Indeed, cytotoxicity (and resultant compensatory hyperplasia) occurs in target tissue early during exposure to tumorigenic doses through drinking water (i.e., hundreds of days before the first tumors); and

• The likely carcinogenic MOA is compensatory crypt enterocyte hyperplasia induced by chronic villous toxicity.

Data supporting a carcinogenic MOA weight-of-evidence of compensatory crypt enterocyte hyperplasia induced by chronic villous toxicity are discussed in Section 3.3.1 and in greater detail in Thompson et al. (2013a). That study also contains additional discussions (e.g., dose—response concordance, temporal association, plausibility, human relevance) based on reasonable scientific interpretations of the data, including a discussion of the toxicological similarities between CrVl and captan. For example, captan also causes duodenal tumors and hyperplasia in mice (but not rats) with a long time to tumor (NCI, 1977). Although USEPA had hypothesized that captan has a mutagenic MOA and calculated an SFo (Q1*) based on intestinal tumors in mice, following third-party review and reevaluation of relevant (e.g., MOA) data, the USEPA Office of Pesticide Programs ultimately determined that captan acts through a non-mutagenic threshold MOA that requires prolonged irritation of the duodenal villi as the initial key event and is not likely to be a human carcinogen at doses not causing cytotoxicity and regenerative cell hyperplasia (Gordon, 2007; USEPA, 2004). Such high human doses were considered unlikely (as with CrVl), and based on the MOA weight-of-evidence the genotoxicity of captan does not contribute significantly to human carcinogenic potential at environmentally-relevant doses (Gordon, 2007). While the additional discussions in Thompson et al. (2013a) need not be repeated here, new studies relevant to the weight-of-evidence have subsequently been published.

Briefly, Thompson et al. (2015a) demonstrated that g-H2AX immunostaining was elevated in duodenal villi but not in the crypt compartment of mice exposed to 180 mg CrVl/L drinking water for 13 weeks, consistent with mean Cr levels z37 times higher in duodenal villi than in crypts. Thompson et al. (2015c) showed that exposure to 21 and 180 mg CrVl/L significantly increased the number of crypt enterocytes, whereas crypt MN and g-H2AX im-munostaining were not increased (see Section 3.3.1.3). The data from these in vivo studies do not support that CrVl damages DNA in the crypt compartment where intestinal stem cells reside as an early initiating key event for crypt proliferation and tumorigenesis. To the contrary, these data add to the weight-of-evidence that already supports a non-mutagenic MOA as opposed to an MOA involving direct Cr-DNA interaction in intestinal stem cells as an initiating key event for crypt proliferation and tumorigenesis. Similarly, a recent study investigating the mutagenic potential of CrVl in the oral mucosa of Big Blue® transgenic F344 rats found that 28-day exposure to 180 mg CrVl/L did not significantly affect mutant frequency (Thompson et al., 2015b), adding to the weight-of-evidence that CrVl-induced carcinogenicity is unlikely to be due to a mutagenic MOA, especially at environmentally-relevant doses. Additionally, a recent report contains results of a pathology peer

review by two independent board-certified veterinary pathologists of histologically-prepared duodenal specimen slides from the NTP and CrVl MOA research project studies wherein B6C3F1 mice and F344/N rats were exposed to SDD via drinking water (Cullen and Ward, 2015). The primary purpose of the pathology review was to determine: (1) the degree to which histopathological duodenal findings differed between these studies; (2) the degree to which findings in mice and rats were qualitatively different or similar; and (3) whether findings likely represent a single pathological process or independent responses. Some of the significant findings include:

• Histopathological findings in the mice and rats of the NTP (2007) and CrVl MOA research project (SRI, 2011a, 2011b) 90-day studies were qualitatively similar, although the degree to which animals were affected varied between study, species, sex, and exposure duration prior to sacrifice;

• Each of the five major findings (i.e., atrophy/blunting of villi, enterocyte vacuolization, single cell necrosis, histiocytic cellular infiltrates, crypt epithelial hyperplasia) was identified to some degree in both mice and rats;

• Generally, the prevalence and severity of findings tended to be greater in mice, males, and the CrVl MOA research project animals, and with increased dose;

• The major reasons for the differences in results between the NTP and CrVl MOA research project 90-day mouse and rat studies are greater prevalence and severity of effects in the CrVl research project animals (e.g., perhaps due to different sources for strains) and differences in diagnostic interpretation (e.g., perhaps due to specific focus on, and scrutiny of, non-neoplastic morphological changes in Cullen and Ward, 2015); and

• Several of the findings (i.e., atrophy/blunting of villi, enterocyte vacuolization, single cell necrosis, crypt epithelial hyperplasia), however, were qualitatively similar in mice and rats of the different studies and appeared pathologically inter-related, portraying a process in which regenerative crypt epithelial hy-perplasia occurred as a sequela to CrVl-induced villous epithelial cell damage.

exposure groups (60 and 170 mg SDD/L), the prevalences of his-tiocytic cellular infiltrates, atrophy/blunting, and enterocyte vacuolization were higher in the 90-day study compared to the 2-year NTP study, which had a higher prevalence of villous single cell necrosis. On the other hand, epithelial hyperplasia of the duodenal crypt was higher in the 2-year study across all water concentration exposure groups. These data show duodenal villous toxicity and hyperplasia at >60 mg SDD/L for 90 days. Furthermore, for both the 90-day and 2-year exposure durations, these effects generally become progressively much more prevalent at higher water concentrations (170 and 520 mg SDD/L), where duodenal tumors were also much more prevalent. See Cullen and Ward (2015) for all available data as well as a more full discussion of study findings. Lastly, Bourdon-Lacombe et al. (2015) conclude that gene expression profiling (i.e., Thompson et al., 2012b) successfully supports a non-genotoxic MOA for CrVl-induced small intestine tumorigenesis, stating, "heat maps using genes related to various MOAs associated with carcinogenicity (e.g., cell cycle progression, proliferation, oxidative stress and regeneration) clearly show that chromium clusters with other non-genotoxic carcinogens and not with chemicals that are genotoxic."

Key considerations in the weight-of-evidence for the carcinogenic MOA (e.g., dose-response and temporal concordance) are given in Table 9. Based upon review of the current database, including newly published studies, the current paper concurs with Thompson et al. (2013a) and Health Canada (2015) that the weight-of-evidence supports cytotoxicity-induced regenerative hyperpla-sia (i.e., a threshold MOA) as the most scientifically well-supported MOA. A summary of dose—response information relevant to the MOA is provided in Table 10 (similar to Table 1 in Thompson et al., 2013b). Based on available MOA analyses and data, compensatory crypt enterocyte hyperplasia induced by chronic villous toxicity should be considered as required but not always sufficient to induce the late onset intestinal tumorigenesis observed in NTP (2008), and the non-linear threshold (i.e., RfD) approach should be adopted for assessing the potential intestinal carcinogenicity of oral exposure to CrVl.

Table 8 contains mouse histopathological prevalence data of particular interest for the current paper (from Tables 2 and 10 of Cullen and Ward, 2015 and Table 2 of Thompson et al., 2011a). The prevalences of histopathological findings due to 7- and 90-day mouse exposures in the CrVl MOA research project (SRl, 2011a) are presented along with those from the 2-year NTP (2008) study. Generally, across dose groups the prevalences of the various his-topathological findings for the villi and crypts in the CrVl research project 90-day study were within 30% of those for the 2-year NTP (2008) study. Overall agreement between 90-day and 2-year results appears particularly good for the high water concentration exposure group (520 mg SDD/L). For the lower water concentration

4. Discussion and conclusions

Conservative default procedures are intended to provide regulatory scientists with a health-protective approach in the absence of sufficiently robust scientific data that justify an alternative, more chemical-specific and toxicologically-predictive approach. While this is an important function, these default procedures also allow regulatory scientists to avoid decisive scientific judgments based on the scientific weight-of-evidence and conducting dose—response assessments accordingly (e.g., even with a robust scientific weight-of-evidence for a non-mutagenic MOA, the lack of study data in a given area can be deemed uncertainty and used almost indefinitely

Table 8

Prevalence of non-neoplastic findings in the duodenum of female mice exposed to CrVl for 7 days, 90 days, and 2 years. Diagnosis 0 mg SDD/L 60 mg SDD/L 170 mg SDD/L 520 mg SDD/L

90 Daysa 2 Yearsb 90 Daysa 2 Yearsb 7 Daysa 90 Daysa 2 Yearsb 7 Daysa 90 Daysa 2 Yearsb

Villus:

Histiocytic cellular infiltrates 0/10 0/9 1/10 0/10 — 10/10 9/10 — 10/10 10/11

Atrophy/blunting 0/10 0/9 4/10 1/10 0/5 10/10 3/10 3/5 10/10 10/11

Enterocyte vacuolization 0/10 0/9 3/10 1/10 3/5 8/10 5/10 5/5 7/10 7/11

Single cell necrosis 0/10 0/9 0/10 2/10 — 0/10 3/10 — 6/10 7/11

Crypt:

Epithelial hyperplasia 0/10 0/9 3/10 4/10 0/5 6/10 8/10 3/5 7/10 10/11

a CrVl research project (e.g., SRl, 2011a) study slide review results from Table 2 of Thompson et al. (2011a) and Table 2 of Cullen and Ward (2015). b NTP (2008) study slide review results from Table 10 of Cullen and Ward (2015).

Table 9

Key considerations in the carcinogenic MOA weight-of-evidence.

Evidence for non-mutagenic MOA [based on target tissue data following in vivo drinking water exposure]a

Scientific relevance and weightb

Evidence for mutagenic MOA [based on non-target tissue data following various exposure scenarios/conditions]c

Mutagenic potential pertinent to NTP study tumors

(1) No increased Kras mutation frequency in target tissue (i.e., duodenum) due to 90-day exposure to 0.3 —520 mg SDD/L drinking water;

(2) No DNA damage - negative results for MNd, KN, AI, MI, and g-H2AX immunostaining in duodenal crypts due to drinking water exposure for 7 and/or 90 days; and

(3) Additionally, no Apc involvement or increased Wnt/b-catenin signaling due to 90-day drinking water exposure.

CrVl-induced mutagenicity as the initiating event

(1) Same negative mutagenicity and genotoxicity evidence in target tissue as above following 7- and/or 90-day exposure; plus

(2) Signs of duodenal villous toxicity and initial signs of hyperplasia (e.g., larger crypt area) begin as early as day 8 at 170 mg SDD/L, the drinking water concentration where significant villous toxicity (e.g., 100% prevalence of atrophy/blunting) and duodenal crypt hyperplasia (i.e., prevalence, significantly increased enterocytes/crypt and crypt area) is also found later at day 91;

(3) Only one tumor location in each species (portal of entry) despite the presence of Cr in multiple tissues; and

(4) Tumors of the mouse small intestine did not occur in the 90-day NTP study or early in the 2-year study (>451 days _, >625 days ?) and were not associated with lethality or metastases.

Dose—response concordance

(1) Increased duodenum tissue concentrations of Cr at >14 mg SDD/L for 90 days;

(2) Redox changes (GSH/GSSG) in the duodenum at >60 mg SDD/L for 7 days and >14 mg SDD/L for 90 days, with gene expression changes indicative of oxidative stress at >60 mg SDD/L for 90 days;

(3) Signs of duodenal villous toxicity and initial signs of hyperplasia (e.g., larger crypt area) beginning at

170 mg SDD/L for 7 days, and 60 mg SDD/L for 90 days (e.g., 40% prevalence of villous atrophy/blunting, 30% prevalence of crypt hyperplasia);

(4) Increased villous toxicity and significant crypt hyperplasia at 520 mg SDD/L for 7 days, with increased villous toxicity (e.g., atrophy/blunting) and duodenal crypt hyperplasia (i.e., prevalence, significantly increased enterocytes/crypt and crypt area) beginning at 170 mg SDD/L for 90 days;

(5) Similar but increased villous toxicity and duodenal crypt hyperplasia prevalence with further increased enterocytes/crypt and crypt area at 520 mg SDD/L for 90 days;

(6) Qualitatively similar results for villous toxicity and crypt hyperplasia for the 2-year NTP study.

++ Target tissue -+ In vivo +/-

+ Relevant Study Species for SI Tumors +/-+ Relevant Exposure Route -/+ + Relevant Dose(s) +/-++ Drinking Water Exposure -/+ ^ High hierarchy of evidence data from target tissue. ^ Data collected following in vivo exposure of the most relevant study species/strain via the most relevant route, exposure scenario, and dosing regimen (e.g., drinking water ad libitum and NTP study drinking water concentrations).

^ All target tissue data negative for mutagenicity and genotoxicity.

(same +/- as above)

^ High hierarchy of evidence data from target tissue. ^ Data collected following in vivo exposure of the most relevant study species/strain by the most relevant route, exposure scenario, and dosing regimen (e.g., drinking water ad libitum and NTP study drinking water concentrations). ^ Data show target tissue (i.e., duodenal crypt) hyperplasia in the absence of mutagenicity (and genotoxicity), but in the presence of significant villous toxicity (e.g., prevalent atrophy/ blunting), in addition to characteristics not indicative of a mutagenic MOA.

++ Target tissue -+ In vivo +

+ Relevant Study Species for SI Tumors +/-+ Relevant Exposure Route and Dose(s) + ++ Relevant Exposure Scenario -/+ ^ High hierarchy of evidence data from target tissue. ^ Data collected following in vivo exposure of the most relevant study species/strain by the most relevant route, exposure scenario, and dosing regimen (e.g., drinking water ad libitum and NTP study drinking water concentrations).

^ Data are indicative of: (1) Increased tissue concentrations of Cr and redox changes at lower concentrations than those inducing villous toxicity and crypt hyperplasia at the same time point; (2) Significant crypt hyperplasia (60% prevalence) at 520 mg SDD/L on day 8 in the presence of significant villous toxicity (e.g., 60% prevalence of atrophy) and signs of villous toxicity and initial signs of hyperplasia that began at the next lowest dose of 170 mg SDD/L on day 8; (3) Significant crypt hyperplasia (30% prevalence) in the presence of villous toxicity (e.g., 40% prevalence of atrophy/

(1) Positive mutagenicity results in vivo in mouse skin, bone marrow, and liver (transgenic Muta™ mouse) by an environmentally- and physiologically-irrelevant exposure route (i.p.);

(2) Generally positive genotoxicity results in vivo in non-target tissues of rats and mice exposed via routes/ scenarios either entirely or largely irrelevant to the NTP study (e.g., i.p. and gavage as opposed to drinking water exposure that produced negative results in several 3-month studies), except for one rat drinking water study qualitatively positive for DNA-protein crosslinks by elec-trophoresis (negative by the well-established alkaline elution method) in the liver but negative in lymphocytes (Coogan et al., 1991); and

(3) In vitro, generally positive results for mutagenicity/genotoxicity in non-target tissue cells and bacteria.

(1) Same non-target tissue data as above; plus

(2) DNA adducts on pSP189 plasmids transfected into human fibroblasts immortalized with the SV virus and then transfected into E. Coli MBL50, which increased mutation frequency; but

(3) Primarily, DNA strand breaks in the leukocytes of mice exposed via oral gavage (Danadevi et al., 2001), and secondarily DNA-protein crosslinks as evaluated qualitatively by elec-trophoresis (but negative by the well-established alkaline elution method) in the liver (but not in the lymphocytes) of rats exposed via drinking water (Coogan et al., 1991), neither of which are mutations or in target tissue.

To demonstrate dose—response concordance between the key mutational event initiating the carcinogenic process in the mouse small intestine and subsequent events in the hypothesized mutagenic MOA, the McCarroll et al. (2010) analysis hinges upon:

(1) Primarily, DNA strand breaks in Swiss mouse leukocytes (a nonmutation endpoint, not in a tissue susceptible to CrVl-induced carcinogenesis) due to single oral gavage at a cited dose of 0.6 mg/kge (Danadevi et al., 2001); and

(2) Secondarily, DNA-protein crosslinks in the rat liver (another nonmutation endpoint not in target tissue) as assessed qualitatively by electrophoresis (negative by the well-established alkaline elution method) at =6—9 mg CrVl/kg-day

Table 9 (continued)

Evidence for non-mutagenic MOA [based on target tissue Scientific relevance and weightb Evidence for mutagenic MOA [based on

data following in vivo drinking water exposure]a non-target tissue data following various

exposure scenarios/conditions]c

(7) Mutagenicity/genotoxicity is not induced by these drinking water concentrations - no evidence of increased Kras mutations, crypt cytogenetic damage (i.e., negative results for MNd, KN, Al, Ml, and g-H2AX immunostaining), Apc involvement or increased Wnt/b-catenin signaling at tumorigenic doses via drinking water for 7 and/or 90 days; and

(8) Late onset tumorigenesis is induced at the same drinking water concentrations (60—520 mg SDD/L) as these non-mutagenic events.

Temporal concordance

• Decreased GSH/GSSG ratio

• Nrf2 activation/oxidative stress

• Significantly increased Cr content in duodenal villi

• Signs of duodenal villous toxicity (e.g., cytoplasmic vacuolization, atrophy) and initial signs of hyperplasia (e.g., larger crypt area) beginning at 170 mg SDD/L

• Absence of significantly increased Cr content in duodenal crypts

• Absence of aberrant nuclei (e.g., MNd, KN) or g-H2AX immunostaining in duodenal crypts

• No Apc or Wnt/b-catenin changes

• Transcript changes consistent with non-mutagenic MOA

• lncreased signs of villous toxicity and incidence of crypt hyperplasia at 520 mg SDD/L

Day 91

• Decreased GSH/GSSG ratio

• Nrf2 activation

• Significantly increased Cr content in duodenal villi

• Increased g-H2AX immunostaining in duodenal villi in the absence of aberrant villous foci indicative of transformation

• Diffuse hyperplasia at >62.5 mg SDD/L

• Significant duodenal villous toxicity (e.g., atrophy/ blunting) and crypt hyperplasia beginning at 60 mg SDD/L

• Absence of significantly increased Cr content in duodenal crypts

• Absence of aberrant nuclei or g-H2AX immunostaining in duodenal crypts

• No change in Kras mutation

• No Apc or Wnt/b-catenin changes

• lncreased and significant villous toxicity and duodenal crypt hyperplasia (i.e., prevalence, significantly increased enterocytes/crypt and crypt area) at 170 and 520 mg SDD/L

2-Year (NTP Study)

• Qualitatively similar results for villous toxicity and crypt hyperplasia

• Adenomas (>451 days _, >693days ?)

• Carcinomas (>729 days _, >625 days?)

blunting) beginning at 60 mg SDD/L on day 91 that is progressively more prevalent at 170 and 520 mg SDD/L in the presence of increased and significant villous toxicity and in the absence of crypt mutagenicity/ genotoxicity; and (4) Qualitatively similar results for villous toxicity and crypt hyperplasia in the 2-year study.

^ Duodenal tumors occur at these same doses.

++ Target tissue -+ In vivo +

+ Relevant Study Species for SI Tumors +/-+ Relevant Exposure Route and Dose(s) + ++ Relevant Exposure Scenario -/+ ^ High hierarchy of evidence data from target tissue. ^ Data collected following in vivo exposure by the most relevant route, exposure scenario, and dosing regimen (e.g., drinking water ad libitum and NTP study drinking water concentrations).

^ Data are indicative of: (1) Signs of villous toxicity and initial signs of crypt hyperplasia on day 8 at 170 mg SDD/L, with increased villous toxicity and significant crypt hyperplasia on day 91 at the same dose; (2) Significant villous toxicity and crypt hyperplasia on day 8 at 520 mg SDD/L, with further increased villous toxicity and significant hyperplasia (e.g., prevalence, crypt area) on day 91 at the same dose (in the absence of crypt mutagenicity/genotoxicity); and (3) For the 2-year study, qualitatively similar results for villous toxicity and crypt hyperplasia.

^ Temporally, these effects occurring as early as day 8 significantly precede the late onset duodenal tumors (e.g., adenomas at >451 days).

for 3 weeks via drinking water (Coogan et al., 1991).f

To demonstrate temporal concordance between key events in the hypothesized mutagenic MOA for the carcinogenic process in the mouse small intestine, the McCarroll et al. analysis again hinges upon:

• Primarily, DNA strand breaks in Swiss mouse leukocytes at day 1 due to single oral gavage at a dose of 0.6 mg/ kge (Danadevi et al., 2001);

• Secondarily, after 3 week exposure to = 6—9 mg CrVl/kg-day via drinking water (=day 21), DNA-protein crosslinks in the rat liver (another non-mutation endpoint not in a tissue susceptible to CrVl-induced carcinogenesis) evaluated qualitatively by electrophoresis (but negative by the well-established alkaline elution method) (Coogan et al., 1991)f; plus

• Cell proliferation (hyperplasia) at day >90.

a Based on Tables 4, 7, 8, and 9 and Fig. 4 of Thompson et al. (2013b); Thompson et al. (2015a, 2015c), O'Brien et al. (2013), and Cullen and Ward (2015). b While this column provides examples of important considerations for the weight-of-evidence, a simplistic consideration of these factors alone (e.g., the number of"+" on each side) would represent a significant oversimplification of the scientific judgment necessary for a weight-of-evidence determination as a more holistic consideration of the data is necessary (e.g., cohesiveness of the data supporting a given MOA, interpretation of data obtained under different experimental conditions providing differing results and its contextual meaning for the MOA and overall weight-of-evidence); ">" = greater weight-of-evidence; "+" = attribute present; "-" = attribute absent; "+/-" or "-/+" = data having mixed attributes;"++" = attribute considered of particular importance; Sl = small intestine; relevant = same as the NTP (2008) mouse study that found the tumors of the small intestine for which the underlying MOA is at issue. c Based on Tables 1 and 2, Figs. 3 and 4, and text of McCarroll et al. (2010).

d Negative MN results from studies conducted by NTP (as cited in Table 2 of McCarroll et al., 2010) add to the weight-of-evidence.

e This dose cited by McCarroll et al. (2010) in units of mg Na2Cr2O7/kg should actually be in units of mg K2Cr2O7/kg (Danadevi et al., 2001); the corresponding Cr dose is 0.21 mg CrVl/kg.

f By contrast, although not noted by McCarroll et al. (2010) in the discussion of dose—response concordance, DNA damage (i.e., DNA-protein crosslinks) did not occur in lymphocytes in Coogan et al. (1991) after 3 and 6 weeks of drinking water exposure to =6—9 mg CrVl/kg-day, doses that are considerably higher than the single oral gavage doses in Dandevi et al. (0.21—3.4 mg CrVl/kg), and DNA damage did not occur in lymphocytes even at much higher i.p. doses (3.3—27 mg CrVl/kg in Coogan et al., 1991).

as a justification for the continued withholding of scientific judgment and use of conservative defaults). Such undue reliance on default procedures reduces scientific credibility, discourages important new scientific research, and can reduce regulatory chemical dose—response assessments to data collection and

modeling exercises that do not account for important chemical-specific information or accurately reflect risk.

Regulatory dose—response assessment should be guided by data-informed, chemical-specific approaches (e.g., low-dose extrapolation, PBPK), in lieu of undue reliance on default

Table 10

Summary of dose—response data relevant to the MOA.

Response3 Drinking water concentration mg SDD/L

0.3 (0.1 mg CrVl/L) 4 (1.4 mg CrVl/L) 14 (5 mg CrVl/L) 60 (20 mg CrVl/L) 170 (60 mg CrVl/L) 520 (180 mg CrVl/L)

Cr in duodenum (villi) X ✓ ✓ ✓ ✓

Oxidative changes X ✓ ✓ * ✓* ✓*

Gene expression changes X ✓* ✓* ✓* ✓*

Villus toxicity b X X ✓ ✓ * ✓*

Crypt hyperplasia b X X ✓ ✓ ✓ *

K-ras mutations X X * *

Crypt MN X X * * X

Crypt DNA damage (g-H2AX) NA 7 NA 7 NA 7

a ✓ = presence of response due to 90-day exposure, with "*" denoting that 7-day exposure also induced the effect; 7 = absence of response; NA = not assessed. b From Table 8.

approaches, when they are sufficiently scientifically robust and defensible. Failure to do so may result in significantly overestimating environmental risk and diverting governmental and public attention away from more important public health issues (e.g., obesity, cardiovascular disease, tobacco use), thereby leading to misprioritization and misallocation of efforts and the limited resources available. For example, the Coachella Valley Water District indicates that although no drinking water wells exceed the federal MCL, treating water from the 30 wells that are above California's new CrVl standard of 10 ppb will increase water bills by $30—50 per month just to comply with the standard, when their customers' per capita income of approximately $22,000 per year is already less than half the state average (CVWD, 2013,2015a, 2015b; CalDOF, 2015). ln the present case, there are sufficiently robust scientific data relevant to the carcinogenic MOA for CrVl to justify a more data-informed, chemical-specific approach.

Accordingly, the current paper built upon the previous study (Haney, 2015b) to develop both a non-linear, non-threshold approach and a non-linear threshold approach for assessing the oral carcinogenicity of CrVl. For the non-linear, non-threshold approach, a dose—response function was developed (based on the Weibull model) that adequately describes the non-linearity expected in human excess risk versus oral dose due to the sub-linear relationship between oral dose and internal dose (added mg Cr/kg target tissue) across the environmentally-relevant doses of regulatory interest (Figs. 1—3). For the non-linear threshold approach, BMD modeling was used to derive an internal mouse POD (i.e., BMDL10 of 1.39 added mg Cr/kg duodenum tissue) for diffuse hy-perplasia in the duodenum (Fig. 4 and Table 6). Modeling was then used to convert this internal mouse dose to the corresponding mouse oral POD of 0.31 mg/kg-day (Fig. 5 and Table 7). Using applicable UFs (i.e., UFA of 10, UFH of 10, UFD of 1) resulted in an RfD of 0.003 mg/kg-day, which is considered protective of cytotoxicity-induced regenerative hyperplasia as a key precursor event to carcinogenesis in the mouse small intestine and happens to correspond to the approximate human dose at the MCL (i.e., MCL of 0.1 mg/L x 2 L/day/70 kg = 0.0029 mg/kg-day z 0.003 mg/kg-day).

This RfD value (0.003 mg/kg-day) shows remarkable agreement with that published previously (0.006 mg/kg-day) based on a more scientifically-sophisticated approach that utilized extensive PBPK modeling to address important factors such as human variation (e.g., pH-dependent CrVl reduction across age groups, diurnal variation in gastric lumen factors, sensitive subpopulations such as PPl users) (Thompson et al., 2013b). Convergence of these two RfD values derived using appreciably different methods increases confidence in candidate RfD values near or within this range (0.003—0.006 mg/kg-day) that are based on more appropriate dose metrics (i.e., dose metrics more closely related to the toxic effect

such as internal dose to the critical target tissue as opposed to oral dose). For example, Health Canada (2015) has just derived a tolerable daily intake of 0.0044 mg/kg-day based on the weight-of-evidence for a threshold MOA for CrVl-induced carcinogenesis via oral exposure, which was used to calculate a proposed health-based maximum acceptable concentration of 0.1 mg/L for CrVl in drinking water. Comparison of the RfD values suggests that the RfD derived in the current paper, which approximately corresponds to human intake at the MCL, could be somewhat conservative. This is consistent with negative results for villous toxicity, proliferation, and the various other endpoints studied in the MOA research project (e.g., Cr in duodenum, redox changes, gene changes, crypt cytogenetic damage, Kras mutations, pre-neoplastic lesions) at the MCL and the next higher water concentration, which was 14-fold higher (see Table 1 of Thompson et al., 2013b).

The RfD approach is the most scientifically-defensible, nonlinear approach based on the weight-of-evidence of available MOA information for the most scientifically-supported MOA. That is, as the weight-of-evidence supports compensatory crypt enterocyte hyperplasia induced by chronic villous toxicity as the carcinogenic MOA for CrVl-induced carcinogenesis, an RfD should be developed (USEPA, 2005). Health Canada (2015) concurs that the evidence for a mutagenic MOA is weak, and confidence in a cytotoxic MOA is high. Despite the weight-of-evidence for a non-mutagenic MOA, if the MOA were actually mutagenic, estimates of the potential risk associated with an RfD of 0.003 mg/kg-day (excess risk of 1.8E-04 to 3E-04) would be near the upper end of the USEPA acceptable excess risk range and lower than the risk associated with the federal MCL for arsenic (excess risk of 5E-04 at 10 ppb based on the USEPA lRlS drinking water unit risk of 5E-05 per ppb), which may be deemed acceptable (Haney, 2015b). However, after due consideration of the associated strengths, limitations, and other factors in the context of public health protection, either the RfD derived herein or an RfD based on a more scientifically-sophisticated approach (e.g., Thompson et al., 2013b) may be considered more appropriate.

5. Uncertainties

Extrapolation of experimental animal data to estimate potential human cancer risk yields uncertainty. Although the risk estimates contained in the non-linear, non-threshold section assume low-dose linearity of target tissue dose (not oral dose) and risk (note that a sub-linear dose—response results when risk is expressed as a function of oral dose due to the non-linear relationship between tissue dose and oral dose), based on the weight-of-evidence for a cytotoxicity-induced regenerative proliferation (i.e., non-mutagenic) MOA for carcinogenicity, actual risk for low oral doses of CrVl (e.g., <0.006 mg/kg-day) may be as low as zero. There is also

some uncertainty associated with the BMD model used to model the human doses and excess risk estimates in Table 1. That is, USEPA BMD software may not contain all models that would fit the data and the dose—response function from another model might return slightly different excess risk results at a given dose. However, the model selected (Weibull model) provides good fit to the full dataset as well as data subsets (Table 3). Beyond this, the uncertainties associated with the non-linear, non-threshold assessment are the same as those associated with the toxicokinetic (i.e., dose-dependent, dose fraction absorbed) analyses conducted in the Haney (2015a, 2015b) peer-reviewed publications. Readers are referred to those published, open access studies for additional discussion of uncertainties.

Extrapolation of experimental animal data to humans also gives rise to uncertainties for the non-linear threshold (i.e., RfD) approach. ln recognition of these uncertainties, UFs were applied to the POD (i.e., BMDL10) in deriving the RfD. More specifically, standard UFs generally considered to account for the uncertainties associated with a number of steps in extrapolating laboratory animal data to humans were used, including an UF for intrahuman variability in susceptibility (i.e., a full UFH of 10 as well as a full UFA of 10). Consistent with default procedures designed to be sufficiently conservative in recognition of the uncertainties associated with deriving toxicity factors, it is important to note that PBPK models are useful in accounting for interspecies and intraspecies differences in pharmacokinetics, and the RfD value (0.006 mg/kg-day) calculated by Thompson et al. (2013b) based on a more scientifically-sophisticated assessment utilizing mouse and human PBPK modeling that addresses key factors in human variation (e.g., pH-dependent CrVl reduction across age groups, diurnal variation in gastric lumen factors, sensitive subpopulations such as PPl users) suggests that the RfD derived herein (0.003 mg/kg-day) may adequately account for such uncertainties.

The carcinogenic MOA weight-of-evidence also has associated uncertainty. New scientific data continue to be generated and published and it is possible that future studies appearing in the scientific peer-reviewed literature with data relevant to the MOA for CrVl-induced carcinogenesis due to oral exposure could affect the scientific interpretation and/or weight of prior scientific data. Additionally, while the CrVl MOA research project has been rather extensive and was designed to fill data gaps in the scientific understanding of the carcinogenic MOA, it can always be said that residual scientific uncertainty remains. Readers are referred to the individual MOA studies and analyses for discussions of the associated limitations by the respective study authors (e.g., Thompson et al., 2011a, 2011b, 2012, 2013a; Kirman et al., 2012, 2013; Proctor et al., 2012; Kopec et al., 2012a, 2012b; O'Brien et al., 2013; Thompson et al., 2015a, 2015c; McCarroll et al., 2010). However, the existence of some degree of uncertainty in a given area does not change the weight-of-evidence for the most scientifically-supported MOA based on the information currently available, which supports the non-linear threshold (i.e., RfD) approach as the most scientifically-defensible, non-linear approach. Nevertheless, in regard to uncertainty surrounding the carcinogenic MOA, Haney (2015b) evaluated what the potential excess risk could be at various RfD values in the event that a weight-of-evidence finding for a non-mutagenic MOA were incorrect, accounting for dose-dependent differences in absorption by target tissues. Results indicated that despite the weight-of-evidence for a non-mutagenic MOA, if the MOA was assumed to be mutagenic, estimates of the potential risk associated with an RfD of 0.003 mg/kg-day (excess risk of 1.8E-04 to 3E-04) would be near the upper end of the USEPA acceptable excess risk range and lower than the risk associated with the arsenic federal MCL (excess risk of 5E-04 at 10 ppb).

Acknowledgments

The author would like to thank the staff and management of the Toxicology Division of the TCEQas well as the rest of the agency for their support in developing an important scientific manuscript. The views and conclusions expressed herein may be those of the study author and not necessarily those of the TCEQ.

Conflicts of interest

The author declares that there are no conflicts of interest.

Transparency document

Transparency document related to this article can be found online at http://dx.doi.org/10.1016/j.yrtph.2015.10.011.

References

Agency for Toxic Substances and Disease Registry (ATSDR), 2012. Toxicological Profile for Chromium. Available at: http://www.atsdr.cdc.gov/toxprofiles/tp7. pdf.

Bagchi, D., Balmoori, J., Bagchi, M., et al., 2002. Comparative effects of TCCD, endrin, naphthalene and chromium (Vl) on oxidative stress and tissue damage in the liver and brain tissues of mice. Toxicology 175, 73—82.

Bourdon-Lacombe, J., Moffat, l., Deveau, M., et al., 2015. Technical guide for applications of gene expression profiling in human health risk assessment of environmental chemicals. Regul. Toxicol. Pharmacol. 72, 292—309.

Butterworth, B., Conolly, R., Morgan, K., 1995. A strategy for establishing mode of action of chemical carcinogens as a guide for approaches to risk assessments. Cancer Lett. 93,129—146.

California Department of Finance (CalDOF), 2015. California Personal lncome Data for 2012. Available at: http://www.dof.ca.gov/HTML/FS_DATA/LatestEconData/ FS_lncome.htm (accessed 24.06.15).

Coogan, T., Motz, J., Snyder, C., et al., 1991. Differential DNA-protein crosslinking in lymphocytes and liver following chronic drinking water exposure of rats to potassium chromate. Toxicol. Appl. Pharmacol. 109, 60—72.

Cullen, J., Ward, J., 2015. Pathology Peer Slide Review of Non-neoplastic Findings in the Duodenum of Mice and Rats Treated with SDD in ACC-TS and NTP Studies, July 2, 2015. Available at: http://cr6study.info.

Coachella Valley Water District (CVWD), 2013. Coachella Valley Water District Chromium Comprehensive Annual Financial Report for the Fiscal Year Ended June 30, 2013, December 10, 2013.

Coachella Valley Water District (CVWD), 2015a. Coachella Valley Water District Chromium-6 Update, February 27, 2015. Available at: http://www.cvwd.org/ documentcenter/view/2112.

Coachella Valley Water District (CVWD), 2015b. Coachella Valley Water District Chromium-6 Update, May 19, 2015. Available at: http://www.cvwd.org/ DocumentCenter/View/71.

Danadevi, K., Rozati, R., Saleha Banu, B., et al., 2001. ln vivo genotoxic effects of potassium dichromate in mice leukocytes using comet assay. Food Chem. Toxicol. 39, 859—865.

De Flora, S., Camoirana, A., Bagnasco, M., et al., 1997. Estimates of the chromium(Vl) reducing capacity in human body compartments as a mechanism for attenuating its potential toxicity and carcinogenicity. Carcinogenesis 18, 531—537.

De Flora, S., llcheva, M., Balansky, R., 2006. Oral chromium (Vl) does not affect the frequency of micronuclei in hematopoietic cells of adult mice and of trans-placentally exposed fetuses. Mutat. Res. 610, 38—47.

De Flora, S., D'Agostini, F., Balansky, R., et al., 2008. Lack of genotoxic effects in hematopoietic and gastrointestinal cells of mice receiving chromium (Vl) with drinking water. Mutat. Res. 659, 60—67.

Environmental Working Group (EWG), 2010. Chromium-6 in US Tap Water. Environmental Working Group. December 20, 2010. Available at: http://static.ewg. org/reports/2010/chrome6/chrome6_report_2.pdf.

Gordon, E., 2007. Captan: transition from 'B2' to 'not likely'. How pesticide registrants affected the EPA cancer classification update. J. Appl. Toxicol. 27, 519—526.

Haney, J., 2015a. Use of dose-dependent absorption into target tissues to more accurately predict cancer risk at low oral doses of hexavalent chromium. Regul. Toxicol. Pharmacol. 71, 93—100.

Haney, J., 2015b. lmplications of dose-dependent target tissue absorption for linear and non-linear/threshold approaches in development of a cancer-based oral toxicity factor for hexavalent chromium. Regul. Toxicol. Pharmacol. 72, 194—20 .

Health Canada, 2015. Chromium in Drinking Water (Document for Public Consultation). Prepared by the Federal-Provincial-Territorial Committee on Drinking Water, Health Canada. Available at: http://www.healthycanadians.gc.ca/health-system-systeme-sante/consultations/chromium-chrome/alt/chromium-chrome-eng.pdf.

ltoh, S., Shimada, H., 1998. Bone marrow and liver mutagenesis in lacZ transgenic mice treated with hexavalent chromium. Mutat. Res. 412, 63—67.

Kirman, C., Hays, S., Aylward, L., et al., 2012. Physiologically based pharmacokinetic model for rats and mice orally exposed to chromium. Chem. Biol. lnteract. 200, 45—64.

Kirman, C., Aylward, L., Suh, M., et al., 2013. Physiologically based pharmacokinetic model for humans orally exposed to chromium. Chem. Biol. lnteract. 204, 13—27.

Knudsen, l., 1980. The mammalian spot test and its use for the testing of potential carcinogenicity of welding fume particles and hexavalent chromium. Acta. Pharm. Toxicol. 47, 66—70.

Kopec, A., Kim, S., Forgacs, A., et al., 2012a. Genome-wide gene expression effects in B6C3F1 mouse intestinal epithelia following 7 and 90 days of exposure to hexavalent chromium in drinking water. Toxicol. Appl. Pharmacol. 259,13—26.

Kopec, A., Thompson, C., Kim, S., et al., 2012b. Comparative toxicogenomic analysis of oral Cr(Vl) exposure effects in rat and mouse small intestinal epithelium. Toxicol. Appl. Pharmacol. 262,124—138.

Kuykendall, J., Kerger, B., Jarvi, E., et al., 1996. Measurement of DNA-protein crosslinks in human leukocytes following acute ingestion of chromium in drinking water. Carcinogenesis 17,1971—1977.

McCarroll, N., Keshava, N., Chen, J., et al., 2010. An evaluation of the mode of action framework for mutagenic carcinogens case study ll: chromium(Vl). Environ. Mol. Mutagen 51, 89—111.

NCl, 1977. Bioassay of Captan for Possible Carcinogenicity, CAS No. 133-06-2. National Cancer lnstitute (NCl) Carcinogenesis Technical Report Series 15.

NTP, 2007. National Toxicology Program Technical Report on the Toxicity Studies of Sodium Dichromate Dihydrate (CAS No. 7789-12-0) Administered in Drinking Water to Male and Female F344/N Rats and B6C3F1 Mice and Male BALB/c and am3-C57BL/6 Mice. NTP Toxicity Report 72. NlH Publication No. 07—5964.

NTP, 2008. National Toxicology Program Technical Report on the Toxicology and Carcinogenesis Studies of Sodium Dichromate Dihydrate (CAS No. 7789-12-0) in F344/N Rats and B6C3F1 Mice (Drinking Water Studies). NTP Toxicity Report 546. NlH Publication No. 08—5887.

O'Brien, T., Ding, H., Suh, M., et al., 2013. Assessment of K-Ras mutant frequency and micronucleus incidence in the mouse duodenum following 90-days of exposure to Cr(Vl) in drinking water. Mutat. Res. 754, 15—21.

Organization for Economic Cooperation and Development (OECD), 2014. OECD Guideline for the Testing of Chemicals: Mammalian Erythrocyte Micronucleus Test (TG 474). Organization for Economic Co-operation and Development. OECD Publishing.

Proctor, D., Suh, M., Aylward, L., et al., 2012. Hexavalent chromium reduction kinetics in rodent stomach contents. Chemosphere 89, 487—493.

Sarkar, D., Sharma, A., Talukder, G., 1993. Differential protection of chlorophyllin against chromium and chlordane in mouse bone marrow in vivo. Mutat. Res. Lett. 301, 33—38.

Sasso, A., Schlosser, P., 2015. An evaluation of in vivo models for toxicokinetics of hexavalent chromium in the stomach. Toxicol. Appl. Pharmacol. 287, 293—298.

Shindo, Y., Toyoda, Y., Kawamura, K., et al., 1989. Micronucleus test with potassium chromate (Vl) administered intraperitoneally and orally to mice. Mutat. Res. 223, 403—406.

Southern Research lnstitute (SRl), 2011a. Final Report on the 90-day Repeat Dose Toxicity Study of Sodium Dichromate Dehydrate Administered in Drinking Water to B6C3F1 Mice, April 1,2011. Southern Research lnstitute Study Number

13026.01.01.

Southern Research lnstitute (SRl), 2011b. Final Report on the 90-day Repeat Dose Toxicity Study of Sodium Dichromate Dehydrate Administered in Drinking Water to Fischer Rats, June 21, 2011. Southern Research lnstitute Study Number

13026.01.02.

Suh, M., Thompson, C., Kirman, C., et al., 2014. High concentrations of hexavalent chromium in drinking water alter iron homeostasis in F344 rats and B6C3F1 mice. Food Chem. Toxicol. 65, 381—388.

Texas Commission on Environmental Quality (TCEQ), 2012. Guidelines to Develop Toxicity Factors (RG-442). Texas Commission on Environmental Quality, Office of the Executive Director, Austin, Texas.

Thompson, C., Proctor, D., Haws, L., et al., 2011a. lnvestigation of the mode of action underlying the tumorigenic response induced in B6C3F1 mice exposed orally to hexavalent chromium. Toxicol. Sci. 123, 58—70.

Thompson, C., Haws, L., Harris, M., et al., 2011b. Application of the US EPA mode of action framework for purposes of guiding future research: a case study involving the oral carcinogenicity of hexavalent chromium. Toxicol. Sci. 119, 20—40.

Thompson, C., Proctor, D., Suh, M., et al., 2012a. Comparison of the effects of hex-avalent chromium in the alimentary canal of F344 rats and B6C3F1 mice following exposure in drinking water: implications for carcinogenic modes of action. Toxicol. Sci. 125, 79—90.

Thompson, C., Hixon, J., Proctor, D., et al., 2012b. Assessment of genotoxic potential of Cr(Vl) in the mouse duodenum: an in silico comparison with mutagenic and nonmutagenic carcinogens across tissues. Regul. Toxicol. Pharmacol. 64, 68—76.

Thompson, C., Proctor, D., Suh, M., et al., 2013a. Assessment of the mode of action underlying development of rodent small intestinal tumors following oral exposure to hexavalent and relevance to humans. Crit. Rev. Toxicol. 43, 244—274.

Thompson, C., Kirman, C., Proctor, D., et al., 2013b. A chronic oral reference dose for hexavalent chromium-induced intestinal cancer. J. Appl. Toxicol. 34, 525—536.

Thompson, C., Seiter, J., Chappell, M., et al., 2015a. Synchrotron-based imaging of chromium and gamma-H2AX immunostaining in the duodenum following repeated exposure to Cr(Vl) in drinking water. Toxicol. Sci. 143,16—25.

Thompson, C., Young, R., Suh, M., et al., 2015b. Assessment of the mutagenic potential of Cr(Vl) in the oral mucosa of big blue® transgenic f344 rats. Environ. Mol. Mutagen 56, 621 —628.

Thompson, C., Wolf, J., Elbekai, R., et al., 2015c. Duodenal crypt health following exposure to Cr(Vl): micronucleus scoring, g-H2AX immunostaining, and synchrotron X-ray fluorescence microscopy. Mutat. Res. Genet. Toxicol. Environ. Mutagen 789—790, 61—66.

United States Environmental Protection Agency (USEPA), 1993. Reference Dose (RfD): Description and Use in Health Risk Assessments. US Environmental Protection Agency, Washington, DC. Available at: www.epa.gov/iris/rfd.htm.

United States Environmental Protection Agency (USEPA), 2004. Captan; cancer Reclassification; Amendment of Reregistration Eligibility Decision; Notice of Availability. US Environmental Protection Agency, Washington, DC. Fed. Regist. 69, 68357—60.

United States Environmental Protection Agency (USEPA), 2005. Guidelines for Carcinogen Risk Assessment. EPA/630/P-03/001B. US Environmental Protection Agency, Washington, DC.

United States Environmental Protection Agency (USEPA), 2007. Draft Framework for Determining a Mutagenic Mode of Action for Carcinogenicity: Using EPA's 2005 Cancer Guidelines and Supplemental Guidance for Assessing Susceptibility from Early-life Exposure to Carcinogens. EPA 120/R-07/002-A. US Environmental Protection Agency, Washington, DC.

United States Environmental Protection Agency (USEPA), 2010. Toxicological review of hexavalent chromium. ln: Support of Summary lnformation on the lnte-grated Risk lnformation System (lRlS). EPA/635/R-10/004A. US Environmental Protection Agency, Washington, DC.

United States Environmental Protection Agency (USEPA), 2012. Benchmark Dose Technical Guidance Document. EPA/100/R-12/001. US Environmental Protection Agency, Washington, DC.