Integration of mechanistic and pharmacokinetic information to derive oral reference dose and margin‐of‐exposure values for hexavalent chromium

Abstract The current US Environmental Protection Agency (EPA) reference dose (RfD) for oral exposure to chromium, 0.003 mg kg−1 day−1, is based on a no‐observable‐adverse‐effect‐level from a 1958 bioassay of rats exposed to ≤25 ppm hexavalent chromium [Cr(VI)] in drinking water. EPA characterizes the confidence in this RfD as “low.” A more recent cancer bioassay indicates that Cr(VI) in drinking water is carcinogenic to mice at ≥30 ppm. To assess whether the existing RfD is health protective, neoplastic and non‐neoplastic lesions from the 2 year cancer bioassay were modeled in a three‐step process. First, a rodent physiological‐based pharmacokinetic (PBPK) model was used to estimate internal dose metrics relevant to each lesion. Second, benchmark dose modeling was conducted on each lesion using the internal dose metrics. Third, a human PBPK model was used to estimate the daily mg kg−1 dose that would produce the same internal dose metric in both normal and susceptible humans. Mechanistic research into the mode of action for Cr(VI)‐induced intestinal tumors in mice supports a threshold mechanism involving intestinal wounding and chronic regenerative hyperplasia. As such, an RfD was developed using incidence data for the precursor lesion diffuse epithelial hyperplasia. This RfD was compared to RfDs for other non‐cancer endpoints; all RfD values ranged 0.003–0.02 mg kg−1 day−1. The lowest of these values is identical to EPA's existing RfD value. Although the RfD value remains 0.003 mg kg−1 day−1, the confidence is greatly improved due to the use of a 2‐year bioassay, mechanistic data, PBPK models and benchmark dose modeling.


| INTRODUCTION
Chromium exists in drinking water as trivalent chromium [Cr(III)] and hexavalent chromium [Cr (VI)]. Because of the lack of reducing agents in drinking water and relatively neutral pH, the Cr(VI) species predominates in most supplies. The current US Environmental Protection Agency (EPA) reference dose (RfD) for Cr, 0.003 mg kg −1 day −1 , was last updated in 1998 (US EPA, 1998). The RfD is based on a 1-year study in F344 rats exposed to ≤25 ppm Cr(VI) in drinking water (Mackenzie, Byerrum, Decker, Hoppert, & Langham, 1958), where 25 ppm was considered the study no-observable-adverse-effect-level (NOAEL) due to the absence of carcinogenic and non-carcinogenic effects. More recently, however, exposure to 180 ppm Cr(VI) in drinking water increased oral cavity tumors in F344 rats and exposure to ≥30 ppm Cr(VI) increased the incidence of small intestine (SI) tumors in B6C3F1 mice (National Toxicology Program [NTP], 2008). Environmental monitoring indicates that Cr(VI), which is naturally present in water, is detected at an average concentration of~0.001 ppm in US drinking water (Ellis, Johnson, & Bullen, 2002;McNeill, Mclean, Parks, & Edwards, 2012;Oze, Bird, & Fendorf, 2007;US EPA, 2017). Notwithstanding the large disparity between environmental and carcinogenic concentrations of Cr(VI), the observation of tumors in the NTP (2008) cancer bioassay necessitates re-examination of whether the existing RfD value, last updated before the cancer findings, is health protective. To assess this, we use the latest toxicology and mode of action (MOA) data for Cr(VI) to derive RfD values to benchmark against the existing RfD of 0.003 mg kg −1 day −1 .
MOA analysis for cancer outcomes is an important aspect of human health risk assessment (US EPA, 2005). To inform the MOA and risk assessment of oral Cr(VI) exposure, a series of studies were conducted beginning with an overall proposed MOA (Thompson, Haws, Harris, Gatto, & Proctor, 2011), and subsequent 90-day toxicity studies Thompson et al., 2012), transcriptomic analyses Rager et al., 2017), genotoxicity studies (O'Brien et al., 2013;Thompson, Wolf, et al., 2015;Thompson, Young, et al., 2015;Thompson et al., 2017), as well as ex vivo gastric reduction studies and pharmacokinetic modeling Kirman et al., 2012;Kirman et al., 2013;Kirman et al., 2016;Proctor et al., 2012). Other genotoxicity studies were conducted in response to early drafts of the NTP 2-year cancer bioassay (De Flora et al., 2006;De Flora et al., 2008). Based on these studies and others, several scientists and regulatory agencies have concluded that the MOA for the intestinal tumors in mice involves chronic cytotoxicity and regenerative hyperplasia (Becker et al., 2015;Haney, 2015;HealthCanada, 2015;Thompson et al., 2013), and some scientists have proposed RfD values that are protective of both noncancerous and cancerous lesions in the mouse SI (Haney, 2015;HealthCanada, 2015;TCEQ, 2016;Thompson et al., 2014).
We previously developed an RfD for Cr(VI) that focused specifically on intestinal effects . Since that original publication, additional MOA information has been published that further supports threshold mechanisms for the tumors observed in mice and rats. In addition, new data have been published that better characterize the pharmacokinetics of Cr(VI) reduction in humans, as well as quantitatively account for sensitive populations (Kirman et al., 2016).
Herein, we update our risk assessment for the SI and extend our analyses to other endpoints relevant to Cr(VI) exposure. Lesions described in the 2-year cancer bioassay are reviewed for relevance to setting toxicity criteria and are modeled using physiological-based pharmacokinetic (PBPK) internal dose estimates and benchmark dose (BMD) modeling methods. Data-derived extrapolation factors (EFs) are applied to human equivalent doses (HEDs) for derivation of candidate RfD values. The RfD value ultimately selected is designed to be protective against the non-cancer effects of Cr(VI), as well as cancerous effects in the SI. The oral tumors that occurred in F344 rats primarily at 180 ppm Cr(VI) are analyzed using a margin-of-exposure (MOE) analysis. These analyses should be informative for scientists and regulators assessing the health risks associated with oral exposure to Cr(VI).

| Data selection
Several risk assessments of Cr(VI) have been conducted in the past few years; however, many of these were completed before the publication of directed research aimed at understanding the pharmacokinetics of Cr(VI) and the MOA for gastrointestinal tumors in rodents (NJDEP, 2009;OEHHA, 2011;US EPA, 2010). The current work therefore focuses on using new MOA and pharmacokinetic data to improve the quantitative risk assessment of Cr(VI). However, a formal systematic review such as those described by the Institute of Medicine (IOM, 2011) or the NTP's Office of Health Assessment and Translation (OHAT, 2015) is beyond the scope of this study. Instead, we expand upon the hazard identification recently conducted by US EPA (2010), which resulted in the quantitative dose-response analysis of the cancer and non-cancer endpoints listed in Table 1. Reproductive and developmental effects were also considered and are discussed in  , changes to the internal dose metric and the change in diffuse epithelial hyperplasia incidence (see Section 2.2.3) warrants re-examination.

| Handling of diffuse epithelial hyperplasia data
The hyperplasia data analyzed herein were taken from tables C4 and D4 in NTP (2008), and are summarized in Table 2. These tables provide the incidence of diffuse epithelial hyperplasia as a function of the  (Stout, Herbert, Kissling, et al., 2009)  Hyperplasia data from the jejunum were omitted for doseresponse modeling, for reasons described previously . In brief, hyperplasia incidence in the NTP (2008) bioassay was assessed microscopically via a single 5 μm biopsy taken at the approximate midpoint of each intestinal segment. The mouse duodenum and ileum are each~9 cm long, whereas the jejunum is~19 cm long-implying that the biopsy taken in the mid-jejunum may underestimate hyperplasia in the proximal (duodenal) end of the jejunum, where the chromium level was likely higher than at the midpoint and distal (ileal) end of the jejunum based on chromium levels measured in each segment . Thus, the relationship between hyperplasia and dose is less certain in the jejunum.

| Benchmark dose modeling
Dose-response modeling for adverse effects was conducted using the US EPA's Benchmark Dose Software v.2.6, using the suite  The term "mouse SI total flux" represents the sum of the three sectional flux values (duodenum + jejunum + ileum), so that total risk to the entire SI is estimated. For example, for SI endpoints characterized in terms of mouse SI sectional flux, we use the PBPK model to determine the mouse pyloric flux value that occurs when the mouse SI total flux (sum of duodenum, jejunum and ileum fluxes) is equal to the POD value. In this way, a POD corresponding to a 10% response in the SI tissue as a whole will be distributed between the sections based upon the gradient of subtissue doses (e.g., 9% response in duodenum, 0.9% response in jejunum and 0.1% response in the ileum as a hypothetical distribution). This approach assumes that, for a given value of pyloric flux, the dose of Cr(VI) delivered to the SIs and to systemic tissues, as well as their associated risks to the tissue, as a whole, are equivalent for all species.
Pyloric flux in humans was used to estimate human equivalent lifetime average daily doses that correspond to the mouse internal POD values by considering variation in toxicokinetic processes for Cr(VI) as a function of age using the following five age groups: (1) neonate (0-3 months); (2) infant/child (0.25-6 years); (3) youth (6-18 years); (4) adult (18-60 years); and (5) elderly (60-75 years), as described by Thompson et al. (2014). Details on the application of the human PBPK model for the chromium risk assessment are summarized in Appendix A.

| Toxicity value derivations
RfD values were derived as follows. The rodent POD was first adjusted by an interspecies EF composed of the toxicodynamic factor (EF AD ).
Per US EPA, PBPK models obviate the need for toxicokinetic factor (EF AK ), because the HED is computed directly via the PBPK model. RfD calculation is as follows: where: RfD is mg kg −1 day −1 ; POD is expressed in terms of internal dose in rodents; HED is mg kg −1 day −1 ; EF AD = EF for interspecies toxicodynamic variation (unitless); EF HK = EF for intraspecies pharmacokinetic variation (unitless); EF HD = EF for intraspecies toxicodynamic variation (unitless).
Because the oral cavity tumors in F344 rats were significantly elevated only at 180 ppm, and mechanistic data support thresholds in oral tissue response to Cr(VI) (see Section 3.1.1), an MOE analysis was conducted for these tumors. The MOE is defined as the ratio of the BMDL 10 in an animal study to the estimated human exposure.
Human exposures can be either mean human exposures or high exposures (e.g., 95th percentile). MOE values ≥30 000 or ≥100 000 are considered by many to indicate low concern for human health (Barlow, Renwick, Kleiner, et al., 2006).
3.1 | Dose-response analysis informed by mechanistic considerations 3.1.1 | Portal-of-entry effects

Oral mucosa
Exposure to Cr(VI) was associated with a relatively late onset of tumors in the oral cavity of male and female F344 rats (NTP, 2008) (Table 1). To date, no non-neoplastic or pre-neoplastic histopathological lesions that might serve as precursor events have been identified in the oral tissue of rats or mice exposed to up to 180 ppm Cr(VI) for 7 days, 13 weeks or 2 years (NTP, 2007(NTP, , 2008; Thompson et al., 2012). Toxicogenomic analyses indicate minimal, if any, gene expression changes in the oral mucosa of F344 rats or B6C3F1 mice exposed to ≤180 ppm Cr(VI) for 7 or 90 days (Thompson et al., 2016). Exposure to 180 ppm Cr(VI) for 28 days did not increase mutant frequency in oral tissue of Big Blue® F344 rats (Thompson, Young, et al., 2015). Taken together, these data indicate that Cr(VI) elicits minimal, if any, direct cellular responses in the oral mucosa of rats or mice.
In 2008, De Flora and colleagues proposed that the oral tumors in rats might have been the result of local irritation and oxidation by Cr(VI) at the highest dose-possibly combined with mechanical stimulation by water bottle cannulae (De Flora et al., 2008). Interestingly, we previously observed dose-dependent decreases in the reduced/ oxidized glutathione (GSH/GSSG) ratio (i.e., increased oxidation) in oral samples in F344 rats but not mice .
However, given the lack of gene expression changes in the oral mucosa (Thompson et al., 2016), the change in GSH/GSSG ratio may not have occurred in the oral mucosa tissue per se, but rather in the saliva or microbiota present in the oral cavity. We have also shown that high levels of Cr(VI) employed in the 2 year bioassay generally impaired the health of rodents, as indicated by reduced water intake, reduced bodyweight and iron deficiency . Taken together, these data indicate that the oral tumors in rats, which were significantly elevated only at 180 ppm Cr(VI), are unlikely to be initiated by direct contact. Moreover, the significant reduction in bodyweight gain suggests that rats exposed to 180 ppm exceeded a maximum tolerated dose, which, according to US EPA (2005) guidance, confounds the relevance of these tumors.
Notably, 350 ppm was determined to be too toxic to use in the 2 year bioassay based on adverse effects observed in the 13 week study (NTP, 2008). Similarly, male mice only received ≤90 ppm Cr(VI) due to toxicity observed at 180 ppm in the 13 week study (NTP, 2008). Although general toxicity from achieving the maximum tolerated dose is not associated with oral tumors in F344 rats per se, rats do appear to have a proclivity toward oral cavity tumor development (NTP, 2008;Stout et al., 2009). Nevertheless, a number of mutagenic and non-mutagenic chemicals induce squamous carcinomas in the oral cavity of rats (Greaves, 2012).
Oral tumors were significantly elevated (14% in males, 22% in females) in the highest treatment group (Table 1) for risk to human health (Barlow et al., 2006).
Tumors of the small intestine Table 2 lists the incidence of intestinal adenomas, carcinomas, and adenomas and carcinomas combined, as well as diffuse epithelial Section 2.2) indicated that the incidence of intestinal tumors did not differ between male and female mice in any intestinal segment.
Neither the main effect (χ 2 (6) = 6.79, P = 0.34) nor the interaction (χ 2 (12) = 14.09, P = 0.30) of sex was significant, indicating that these data could be modeled together. As a demonstration of how well the intestinal flux estimates predict response in the mouse SI, Figure 3 shows a dose-response for the combined incidence of adenomas or carcinomas in each intestinal section of male and female mice as a function of flux. The BMD 10 and BMDL 10 values were 12.8 and 10.3 mg Cr(VI) l −1 day −1 (Table 3). Because intestinal tumors are thought to progress from adenomas to carcinomas (Brix, Hardisty, & McConnell, 2010;Greaves, 2012;McConnell, Solleveld, Swenberg, & Boorman, 1986), adenomas and carcinomas were also modeled separately. As expected, the BMDL 10 for carcinomas (28.1 mg Cr(VI) l −1 day −1 ) was higher than adenomas (13.2 mg Cr(VI) l −1 day −1 ) ( Table 3).

Diffuse epithelial hyperplasia of the small intestine
Exposure to high levels of Cr(VI) induces diffuse epithelial hyperplasia in the duodenum and, to a lesser extent, the jejunum in mice (Table 2). As with tumors, the incidence of diffuse epithelial hyperplasia was highest in the proximal SI (duodenum) and lowest in the distal intestine (ileum). Diffuse epithelial hyperplasia is a non-neoplastic lesion that, under chronic wounding, can lead to increased stem cell proliferation, which can promote transformation and carcinogenesis (Cohen & Ellwein, 1990;Cohen, Gordon, Singh, Arce, & Nyska, 2010;Tomasetti & Vogelstein, 2015). Because Cr(VI) flux through intestinal sections well describes the tumor response in the SI, and hyperplasia is a critical preceding event to tumor formation, intestinal flux was used to model hyperplasia incidence from the NTP (2008) 2 year bioassay. As described above, statistical analyses were conducted to determine whether male and female diffuse epithelial hyperplasia could be modeled together. The overall incidence of hyperplasia did not differ significantly between female and male animals in any of the intestinal segments.
As discussed in Methods, Section 2.2, the jejunum data were not modeled, due to uncertainties introduced by the experimental protocol for diagnosing hyperplasia. Including these data would shift the doseresponse curve rightward, thereby resulting in a less conservative BMDL (see Thompson et al., 2014). Figure 4(A) shows the best fitting model of the combined male and female duodenum and ileum data.
The P value for global fit (0.067) was below the US EPA's preferred target minimum of 0.1, but was higher than the minimum acceptable for the multistage model. Removing the three highest dose groups increased the P value to 0.1 (rounded from 0.098) ( Figure 4B). Examination of Figure 4(A) and the scaled residuals indicates that data points at doses of 3.0 and 7.6 flux units were penalizing the P value.
Removing the 7.6 dose group increased the P value to 0.2 (rounded from 0.199). All three modeling approaches (using all data, dropping three highest dose groups, or omitting one potential outlier group at an SI flux of 7.6 mg Cr(VI)l −1 SI day −1 ) resulted in BMD 10 and BMDL 10 values of 2.1 and 1.7 mg Cr(VI) l −1 day −1 (Table 3) Thompson, Wolf, et al., 2015).

Histiocytic cellular infiltration of mesenteric lymph nodes
Histiocytic cellular infiltration was observed in the duodenum and mesenteric lymph nodes of mice and rats in the NTP 13-week and 2- year bioassays (NTP, 2007(NTP, , 2008. In the 13-week study, histiocytic infiltration was noted in the mouse duodenum at lower doses than in the mesenteric lymph nodes. As discussed in NTP (2007) (Gopinath, Prentice, & Lewis, 1987). It is thought that mesenteric lymph nodes act as a "storage depot" for macrophages that are unable to degrade ingested cellular contents (Gopinath et al., 1987). The 13- week and 2-year NTP Cr(VI) study reports noted the similarity in histology between the histiocytes in the intestine and mesenteric lymph nodes (NTP, 2007(NTP, , 2008 (Table 3). Incidence data are adapted from NTP (2008). BMD, benchmark dose; BMDL, benchmark dose (with corresponding 95% lower confidence limit); SI, small intestine [Colour figure can be viewed at wileyonlinelibrary.com] that histiocytic infiltration of mesenteric lymph nodes is a consequence of effects in the duodenum, we consider this an adaptive response to the presence of foreign material (i.e., chromium) and, therefore, it was not considered relevant for RfD derivation. It is reasonable to conclude that protection against intestinal injury (e.g., diffuse epithelial hyperplasia) will mitigate histiocytic infiltration into mesenteric lymph nodes.

| Systemic effects
As mentioned previously, no adverse effects were observed in the 2- year bioassay on Cr(III) (NTP, 2010). Therefore, adverse systemic effects of Cr(VI) are potentially due to (1) direct effects of Cr(VI) in the blood, or (2) secondary effects such as changes in blood redox or iron homeostasis. Exposure to Cr(VI) was shown previously to alter serum GSH/GSSG levels and ratios Thompson et al., 2012), as well as induce iron depletion . Because the Cr(VI) PBPK model (Kirman et al., 2017) can estimate the amount of Cr(VI) entering (i.e., fluxing) into the portal circulation from the gastrointestinal tract, this dose metric was used for effects of Cr(VI) manifested beyond the intestinal mucosa, whether by direct or indirect mechanisms.

Liver
The incidence of chronic liver inflammation was significantly elevated in female rats (  (2008), the relevance of the lesions was considered by the NTP study authors to be unknown.
As with the histiocytic infiltration into the duodenum and mesenteric lymph nodes, these infiltrates may be present to scavenge chromium.
Indeed, the NTP study authors described infiltration into the liver as possible evidence of "phagocytosis of some insoluble chemical precipitate." Despite the questionable relevance of these liver effects, BMD modeling was conducted for comparison to adverse effects in the SI ( Figure 6A,B; Table 4).

Pancreas
US EPA (2010) also modeled cytoplasmic alteration of the acinus pancreas of female mice, even though it was not included in the summary of non-neoplastic lesions in NTP (2008). According to the NTP (2008) study authors, cytoplasmic alteration was "characterized by depletion of cytoplasmic zymogen granules from the pancreatic acinar epithelial cells." Loss of zymogen (degranulation) is said to represent a physiological feature rather than a pathological process (Gopinath et al., 1987). Such lesions are observed in rats treated with diuretics (likely related to dehydration), as well as those in conditions of food deprivation (Gopinath et al., 1987). As noted by the NTP (2008) study authors, the significance of these lesions is unknown. Water consumption rates in the two highest male and female dose groups were less than controls throughout the study.
In the second year of the study, the average water consumption was reduced by 15% and 35% in the two highest male dose groups and by 25% and 32% in the two highest female dose groups (B) Temporal concordance (7 days of exposure): As evidenced by crypt hyperplasia in female mice in the absence of neoplastic lesions. Bars represent incidence in hematoxylin and eosin-stained sections . Note: The short bars indicate empirical observations with 0% incidence. Blue line represents mean ± SD for counted cells in a second study (Thompson, Wolf, et al., 2015). (C) Temporal concordance (90 days of exposure): As evidenced by crypt hyperplasia in female mice in the absence of neoplastic lesions. Bars represent incidence in hematoxylin and eosinstained sections . Note: The short bars indicate empirical observations with 0% incidence. Blue line represents mean ± SD for counted cells (O'Brien et al., 2013). DEH, diffuse epithelial hyperplasia; SI, small intestine (NTP, 2008). It is therefore conceivable that the effects in the pancreas might be due to indirect mechanisms, such as reduced water intake due to poor palatability. Despite the questionable significance of pancreatic alterations, these lesions were modeled in both male and female mice using portal flux. The BMD plot for female mice is shown in Figure 6(C), and the BMDL 10 values for males and females are listed in Table 4.

Reproductive and developmental toxicity
The effects of Cr(VI) on reproductive and developmental toxicity were determined previously by the US EPA to occur at higher exposure doses than effects in the NTP (2008)  for female reproductive toxicity in Swiss albino mice (Murthy, Junaid, & Saxena, 1996). The maximum allowable dose level was based on a NOAEL; however, US EPA (2010) concluded that NOAEL/lowobservable-adverse-effect-level values could not be identified in Murthy et al. (1996) due to inadequate reporting of data. In brief, Murthy et al. (1996) exposed Swiss albino mice to 250, 500 and 750 ppm Cr(VI) in drinking water for 20 days, and reported ovarian effects (decreases in the number of follicles) in mice exposed to 250 ppm. In the same study, mice in another group were exposed to 0.05, 0.5 and 5 ppm Cr(VI) for 90 days. Unlike the 20 day study, no quantitative data were provided; rather, it was reported that, using electron microscopy, there were disintegrated membranes in follicular cells of the 5 ppm group (Murthy et al., 1996). OEHHA of female mice and rats exposed to ≤1.4 ppm for 90 days ; Supporting information Figure S1). Similarly, NTP reported no significant increases in plasma or erythrocyte chromium levels in female mice exposed to 5 ppm Cr(VI) for 6, 13, 182 or 371 days (NTP, 2008). It is therefore unlikely that low Cr(VI) levels in drinking water pose a direct risk to ovarian follicles. As previously mentioned, US EPA (2010) did not derive a POD from Murthy et al. (1996), due in part to inadequate reporting in the study.
The NTP has conducted several studies that inform the potential for reproductive and developmental toxicity effects from Cr(VI) exposure. No significant microscopic lesions were observed in ovaries of F344 rats (mice not examined) exposed to ≤350 ppm in the NTP (2007) 90-day drinking water study, nor were such lesions observed in earlier feed studies in mice and rats (NTP, 1996a(NTP, , 1996b. No effects on ovary weight or reproductive performance were observed in F 0 or F 1 BALB/c mice (NTP, 1997). Testis weight in F344 rats was unaffected by exposure to ≤350 ppm Cr(VI) for 13 weeks (NTP, 2007). Similarly, testis weights in B6C3F1 and BALB/c mice  Table 4. Incidence data are adapted from NTP (2008). BMD, benchmark dose; BMDL, benchmark dose (with corresponding 95% lower confidence limit) [Colour figure can be viewed at wileyonlinelibrary.com] were unaffected by exposure to ≤350 ppm Cr(VI) for 13 weeks; however, testis weight was reduced 11% in am3-C57BL/6 mice, which was attributed to a 36% decrease in bodyweight (NTP, 2007). Earlier feed studies with Cr(VI) also found no effects on testis weight in rats or mice (NTP, 1996a(NTP, , 1996b(NTP, , 1997.  Figure S2).

Many
A few recent studies claim that high concentrations of Cr(VI) disrupt endocrine function and, therefore, label Cr(VI) as an "endocrine disruptor" (Banu et al., 2017;Stanley et al., 2013). We therefore queried US EPA's Tox21 consortium database to determine whether data were available to support the notion of Cr(VI) as an endocrine disrupter. There was no significant indication of androgen, estrogen, or thyroid receptor activation/binding (see Appendix B). Nevertheless, these data do not preclude the possibility that high concentrations of Cr(VI) disrupt endocrine function indirectly (e.g., oxidative stress, iron perturbation). Based on these considerations, reproductive and developmental toxicity were not considered further for RfD development.

| Reference dose derivation
Based on BMD modeling of non-neoplastic lesions (Tables 3 and 4 Figure 7). The 95th percentile for the combined distribution (normal and hypochlorhydria) also corresponds to a gastric pH of approximately 4.2. Using these data, a value for EF HK was calculated using a ratio of doses:

| Reference dose selection
Candidate RfD values are shown in Figure 8. Effects in the pancreas resulted in the highest candidate RfD values of 0.02 mg kg −1 day −1 (  (Kirman et al., 2017), which contains three reduction pools rather than one in the earlier models Kirman et al., 2013). Data indicate a low capacity/fast reduction pool, a higher capacity/slower reduction pool and a high capacity/slow reduction pool. The estimated capacities for the fast reduction pools in humans is 0.68-2.6 mg l −1 (depending on whether fed or fasted), 6.1 mg l −1 in mice and 7.1 mg l −1 in rats. These data indicate that depletion of the fast pool occurs at lower Cr(VI) doses in humans than in either mice or rats. Nevertheless, these pools, which  are replenished between bouts of exposure, are sufficient for reducing environmental levels of Cr(VI) that are typically ≤0.003 mg l −1 .
The RfD proposed herein is health protective for most individuals.
For example, Cr(VI) reduction is pH-dependent, and therefore, life stage differences in gastric pH are accounted for in the PBPK model.
Life stage differences in water and food intake are also accounted for in the PBPK model. Because there are fast, medium and slow reduction pools that each have different capacities, the simultaneous accounting for life stage differences in intake and gastric pH are factored into the estimation of safe human doses. In addition to life stage, the human PBPK model was used to address human variation by considering Cr(VI) reduction in individuals with high gastric pH due to use of medication such as PPIs, or those with hypochlorhydria. Quantitative differences in dose were used to support the pharmacokinetic human variability EF (i.e., EF HK ).
The RfD proposed herein is identical to the current EPA RfD that is based on a NOAEL from a 1-year bioassay in rats exposed to ≤25 ppm Cr(VI) (Mackenzie et al., 1958). In that study, no adverse effects were observed in rats exposed to up to 25 ppm Cr(VI) in drinking water. EPA characterizes the confidence in their RfD as low because of the "small number of animals tested, the small number of parameters measured, and the lack of toxic effect at the highest dose tested." This "low" confidence in the existing RfD based on Mackenzie et al. (1958) does not mean that the value is not health protective, but rather that the scientific basis of the RfD could be improved. This is reflected by EPA's adjustment of the NOAEL by a 1000-fold uncertainty factor. Our proposed RfD is based on considerably more scientific information, including data from a 2-year bioassay, rodent PBPK models developed using target tissue and gastric reduction data, human PBPK models informed by human pharmacokinetic data, quantitative dose-response modeling and MOA research. The uncertainties associated with potential pharmacodynamic differences across species and individuals are each addressed with default EF AD and EF HD values of 3 each. These EFs are akin to the threefold interspecies and intraspecies uncertainty factors (UF A and UF H ) factors often applied to account for pharmacodynamic uncertainties. As such, only a 10-fold uncertainty is applied in the proposed RfD, as compared to the 1000-fold uncertainty in the current EPA RfD. Although the RfD herein is identical to that listed in EPA's Integrated Risk Information System (IRIS), the scientific basis of our value is greater than the IRIS value and therefore we characterize the confidence in our RfD as "high." In conclusion, we have derived several candidate RfD values for Cr(VI) using sophisticated risk assessment approaches that greatly improve the confidence in the RfD. The low end of these values (i.e., 0.003 mg kg −1 day −1 ) is identical to the existing RfD in IRIS, suggesting that drinking water criteria based on an RfD of 0.003 mg kg −1 day −1 are sufficiently protective of human health. Importantly, the information gained from recent 2-year bioassays and MOA research greatly improve the scientific basis for these toxicity criteria. OEHHA.