EPA/635/R-08/012A DEVELOPMENT OF A RELATIVE POTENCY FACTOR (RPF) APPROACH FOR POLYCYCLIC AROMATIC HYDROCARBON (PAH) MIXTURES DRAFT – DO NOT CITE OR QUOTE September 2009 NOTICE This document is an interagency science consultation draft. This information is distributed solely for the purpose of pre-dissemination peer review under applicable information quality guidelines. It has not been formally disseminated by EPA. It does not represent and should not be construed to represent any Agency determination or policy. It is being circulated for review of its technical accuracy and science policy implications. U.S. Environmental Protection Agency Washington, DC DISCLAIMER This document is a preliminary draft for review purposes only. This information is distributed solely for the purpose of pre-dissemination peer review under applicable information quality guidelines. It has not been formally disseminated by EPA. It does not represent and should not be construed to represent any Agency determination or policy. Mention of trade names or commercial products does not constitute endorsement or recommendation for use. ii DRAFT – DO NOT CITE OR QUOTE EXECUTIVE SUMMARY The U.S. Environmental Protection Agency’s (U.S. EPA’s) Integrated Risk Information System (IRIS) Program is releasing for scientific review a relative potency factor (RPF) approach for polycyclic aromatic hydrocarbon (PAH) mixtures as one approach to assessing cancer risk from exposure to PAH mixtures. The RPF approach is not a reassessment of individual PAH carcinogenicity, but rather provides a cancer risk estimate for PAH mixtures by summing doses of component PAHs after scaling the doses (with RPFs) relative to the potency of an index PAH (i.e., benzo[a]pyrene). The cancer risk is then estimated using the dose- response curve for the index PAH. RPFs for seven individual PAHs were developed in the U.S. EPA (1993) Provisional Guidance for Quantitative Risk Assessment of PAHs (Provisional Guidance) and are utilized extensively within U.S. EPA program offices and other regulatory agencies. The Provisional Guidance, however, does not reflect the most recent research, nor does it consider additional PAHs with carcinogenic potential (such as fjord-region PAHs). The Supplemental Guidance for Conducting Health Risk Assessment of Chemical Mixtures (U.S. EPA, 2000) highlights that approaches based on whole mixtures are preferred to component approaches, such as the RPF approach. Risk assessment approaches based on toxicity evaluations of whole mixtures inherently address specific interactions among PAHs and account for the toxicity of unidentified components of PAH mixtures. They also do not require assumptions regarding the toxicity of individual components (e.g., dose additivity or response additivity). While whole mixture assessment is preferred, there are challenges associated with using these approaches. There are very few toxicity data available for whole PAH mixtures and, in most cases, chemical analyses of the composition of mixtures are limited. In addition, PAH- containing mixtures tend to be very complex; the composition of these mixtures appears to vary across sources releasing these mixtures to the environment and in various environmental media in which they occur. For these reasons, a whole mixtures approach may not always be practicable for risk assessment purposes. This report provides recommendations for development of the RPF approach for PAH mixtures health risk assessment and includes: (1) A rationale for recommending an RPF approach (Section 2); (2) A summary of previous approaches for developing the RPF approach for PAHs (Section 3); (3) An evaluation of the carcinogenic potential of individual PAHs (Section 4); iii DRAFT – DO NOT CITE OR QUOTE (4) Methods for dose response assessment and individual study RPF calculation (Section 5); (5) Selection of PAHs for inclusion in the RPF approach (Section 6); (6) Derivation of RPFs for selected PAHs (Section 7); and (7) Characterization of strengths, weaknesses, and uncertainties associated with the RPF approach to PAH cancer risk assessment (Section 8). The RPF approach involves two key assumptions: (1) similar toxicological action of PAH components in the mixture; and (2) interactions among PAH mixture components do not occur at low levels of exposure typically encountered in the environment (that is, additivity is assumed). Mechanistic studies indicate that the mutagenic and tumor-initiating activity of carcinogenic PAHs requires metabolic activation to reactive intermediates (e.g., dihydrodiol epoxides, quinones, radical cations), which covalently modify DNA targets resulting in mutation, and that tumor promotion and progression phases may involve parent compound binding to the Ah receptor (AhR) and subsequent alterations of gene expression or a cell proliferation response to metabolite cytotoxicity (see Section 2.4, Similarities in Carcinogenic Mode of Action for PAHs, and Figure 2-3, Overview of the Proposed Key Events in the Mode of Action for PAH Carcinogenicity). As such, there is evidence that an assumption of similar toxicological action is reasonable; however, the carcinogenic process for PAHs is likely to be related to some unique combination of multiple molecular events resulting from formation of several reactive species. The second assumption of no interactions at low levels of exposure is reasonable, but evidence of toxicological interactions among PAHs at higher dose levels has been observed (see Section 2.7, Additivity of PAHs in Combined Exposures). Several approaches have been used previously for the determination of RPFs for PAHs (see Section 3). In the published literature, RPF values were proposed in at least one analysis for a total of 27 PAHs (see Table 3-1). Because these approaches generally relied on similar bioassay data and modeling methods, the resulting RPF values are generally comparable for most PAHs across studies. The RPF approach provided in the current report makes use of more recent data on genotoxicity and tumorigenicity of PAHs. There is a large PAH database on carcinogenicity in animal bioassays, genotoxicity in various test systems, and bioactivation to tumorigenic and/or genotoxic metabolic intermediates. The RPF analysis presented here includes only unsubstituted PAHs with three or more fused aromatic rings containing only carbon and hydrogen atoms, because these are the most widely studied members of the PAH chemical class. The study types that were considered most useful for RPF derivation were rodent carcinogenicity bioassays (all routes) in which one or more PAHs was tested at the same time as benzo[a]pyrene. In addition, in vivo and in vitro data for iv DRAFT – DO NOT CITE OR QUOTE cancer-related endpoints in which one or more PAHs and benzo[a]pyrene were tested simultaneously were obtained, including studies on the formation of DNA adducts, mutagenicity, chromosomal aberrations, sister chromatid exchange frequency, aneuploidy, DNA damage/repair/recombination, unscheduled DNA synthesis, and cell transformation. Although it would be possible to calculate RPFs from studies where a PAH and benzo[a]pyrene were tested by the same laboratory using the same test system but at different times, this approach was not considered because it could introduce differences in the dose-response information that are unrelated to the chemical (e.g., variability associated with laboratory environment conditions, animal handling, food supply, etc.). Thus, studies in which benzo[a]pyrene was not tested simultaneously with another PAH were not considered in the RPF calculations. Studies of AhR binding/activation were not considered for use in deriving RPFs because there is evidence indicating that highly mutagenic fjord-region PAHs are potent carcinogens despite exhibiting lower AhR affinity (reviewed by Bostrom et al., 2002). Likewise, some PAHs that strongly activate the AhR, such as benzo[k]fluoranthene (Machala et al., 2001), are only weakly carcinogenic. In addition, some studies have demonstrated the formation of DNA adducts in the liver of AhR knock-out mice following intraperitoneal or oral exposure to benzo[a]pyrene (Sagredo et al., 2006; Uno et al., 2006; Kondraganti et al., 2003), indicating that Ah responsiveness is not a prerequisite to genotoxicity. These findings suggest that there may be alternative (i.e., non-AhR mediated) mechanisms of benzo[a]pyrene activation in the mouse liver, and that AhR affinity would not be a good predictor of carcinogenic potency. Several study types were excluded from the database because they did not provide carcinogenicity or cancer-related endpoint information for individual PAHs. These include biomarker studies measuring DNA adducts in humans, studies of PAH metabolism, and studies of PAH mixtures. Although these studies contain important information on human exposure to PAH mixtures and the mode of action for PAH toxicity, they generally do not contain dose- response information that would be useful for calculation of RPF estimates. A database of primary literature relevant to the RPF approach for PAHs was developed by performing a comprehensive review of the scientific literature dating from the 1950s through 2009 on the carcinogenicity and genotoxicity of PAHs. The search identified over 900 individual publications for a target list of 74 PAHs (see Table 2-1) that have been identified in environmental media or for which toxicological data are available. Review of these publications resulted in the identification of more than 600 papers that included carcinogenicity or cancer- related endpoint data on at least one PAH and benzo[a]pyrene tested at the same time. References in the PAH database were sorted into the following major categories: cancer bioassays, in vivo studies of cancer-related endpoints, and in vitro studies of cancer-related endpoints. These categories were further sorted by route (for bioassays) or by endpoint (for v DRAFT – DO NOT CITE OR QUOTE cancer-related endpoints). Each study was reviewed, and critical study details were extracted into tables for each individual endpoint (see Section 4). The tables also include an initial determination of whether the data from each study meet selection criteria for use in the RPF analysis. Studies with data on selected PAHs and benzo[a]pyrene were considered for RPF determination, even if a particular PAH has not been classified by U.S. EPA or International Agency for Research on Cancer (IARC) as a carcinogen. Studies were included in the analysis if the following selection criteria were met: • Benzo[a]pyrene was tested simultaneously with another PAH; • A statistically increased incidence of tumors was observed with benzo[a]pyrene administration, compared with control incidence; • Benzo[a]pyrene produced a statistically significant change in a cancer-related endpoint finding; • Quantitative results were presented; • The carcinogenic response observed in either the benzo[a]pyrene- or other PAH-treated animals at the lowest dose level was not saturated (i.e., tumor incidence at the lowest dose was <90%), with the exception of tumor multiplicity findings; and • There were no study quality concerns or potential confounding factors that precluded use (e.g., no concurrent control, different vehicles, strains, etc. were used for the tested PAH and benzo[a]pyrene; use of cocarcinogenic vehicle; PAHs of questionable purity; unexplained mortality in treated or control animals). If the above criteria were met, studies were selected for use in the analysis regardless of whether positive or negative results were reported. Studies with positive findings were used for calculation of RPFs. Studies with negative findings were used in a weight of evidence evaluation of potential carcinogenicity (discussed in Section 6.1). Dose-response data were extracted from studies with positive findings that met selection criteria. For studies that reported results graphically, individual data points were extracted using digitizing software. In all, over 300 data sets were extracted, reflecting dose-response data from at least one study for 50 of the 74 PAHs included in the analysis. All of the extracted data are presented in Appendix C of this report. Statistical analyses were performed on tumor bioassay data to determine whether the tumor incidence or multiplicity observed at a particular dose represented a statistically significant increase over controls. If statistical analyses were not described in the original report, incidence data were analyzed using Fisher’s Exact test and the Cochran-Armitage trend test. Positive vi DRAFT – DO NOT CITE OR QUOTE findings were indicated by a significant (p < 0.05) difference for at least one dose group by comparison to control (in Fisher’s Exact or an equivalent test) or a significant dose-response trend (Cochran-Armitage or equivalent) for multi-dose studies. For tumor bioassay data reported as tumor count, a t-test was conducted (when variance data were available) to determine whether the count was significantly different from control (p < 0.05). The results of the statistical analyses are shown with the dose-response data in Appendix C. Statistical analyses of the cancer-related endpoint data were not conducted; the study author’s conclusions as to response (positive or negative) was used. Section 5 describes both the methods used for dose-response assessment and RPF calculation in detail. The general equation for estimating an RPF was the ratio of the slope of the dose-response curve for the subject PAH to the slope of the dose-response curve for benzo[a]pyrene. For bioassay data, tumor incidences were modeled using the multistage model within the U.S. EPA Benchmark Dose (BMD) Software (Version 1.3.2). For cancer-related endpoint data in quantal form, this model was also used; for continuous data (either tumor multiplicity or cancer-related endpoint data), the simplest continuous model (linear) within the software was applied. Whenever the data allowed, benchmark response (BMR) values of 10% for quantal data and 1 standard deviation from the control value for continuous data were used to calculate the slope by linear extrapolation to the origin for consistency across data sets. Alternative BMR values were used in select instances, as described in Section 5.3. For data sets that included only a single dose, or those for which no model fit was achieved with the selected models, a point estimate RPF was calculated. The RPFs calculated from individual studies for each PAH were used in a weight of evidence evaluation to assess the potential carcinogenicity of each compound (see Section 6) and in the derivation of a final RPF for each compound (Section 7). The selection of PAHs to be included in the RPF approach began with an evaluation of whether the available data were adequate to assess the potential carcinogenicity of each compound. At least one RPF value was calculated for each of 50 PAHs. For 16 of these compounds, only a single RPF value derived from an in vitro cancer-related endpoint (primarily mutagenicity assays) was available (see Table 6-1). Due to the limited data available for these 16 compounds, no further evaluation of these PAHs was conducted, and they were not selected for inclusion in the RPF approach. For the remaining 34 PAHs, a weight of evidence evaluation (see Figure 6-1) was conducted to assess the evidence that each PAH could induce a carcinogenic response. This evaluation did not constitute a formal weight of evidence evaluation of carcinogenic potential; rather, an expedited approach was developed using the data collected to determine whether the available information for each PAH was adequate to draw a conclusion regarding carcinogenic potential. When the data were considered adequate for a given PAH, it was selected for vii DRAFT – DO NOT CITE OR QUOTE inclusion in the RPF approach; if the data were not considered adequate to assess potential carcinogenicity, the PAH was excluded. In vivo tumor bioassays that included benzo[a]pyrene were given the greatest weight in assessing the potential carcinogenicity of a given PAH; data from other bioassays and cancer-related endpoint studies were used to supplement the weight of evidence when the bioassay data that included benzo[a]pyrene were conflicting or negative. Structural alerts for PAH carcinogenicity or mutagenicity (as defined in Section 2.5 as the presence of a classic bay region or fjord region in a PAH containing at least four benzene rings) were noted in the evaluation for each PAH, but were not used explicitly in the weight of evidence evaluation. The weight of evidence evaluation (Section 6) indicated that the available data were adequate to determine that 23 of the 34 PAHs were potentially carcinogenic, that three PAHs (anthracene, phenanthrene, and pyrene) exhibited little or no carcinogenic potential, and that data were inadequate to evaluate the carcinogenic potential for eight PAHs. The eight PAHs with inadequate data were excluded from the RPF approach. For the three PAHs for which there were sufficient data to conclude that the PAH had minimal carcinogenic potential (i.e., robust negative tumor bioassay data and cancer-related endpoint data), a final RPF of 0 was recommended. While there is little quantitative difference between selecting a final RPF of 0 for a given PAH and excluding that PAH from the RPF approach, this is an important distinction for uncertainty analysis. There is substantial uncertainty in the risk associated with PAHs that are excluded from the RPF approach due to inadequate data; these compounds could be of low or high potency. However, for PAHs with an RPF of 0, there is evidence to suggest that these compounds are of little or no carcinogenic potential, and the uncertainty associated with the cancer risk for these compounds is markedly reduced. For each of the remaining 23 compounds, a final nonzero RPF was derived. A number of options were considered for deriving an RPF from among the numerous values calculated for each individual PAH. These options included: prioritizing bioassay RPFs from different exposure routes based on relevance to environmentally-relevant routes; prioritizing bioassay RPFs based on target organs considered relevant to human susceptibility to PAH carcinogenesis; prioritizing RPFs based on quality of the underlying study; prioritizing cancer-related endpoints by their correlation with bioassay potency (i.e., ability to predict bioassay potency); and aggregating RPFs across all bioassays, all cancer-related endpoints, or across all endpoints. In the end, it was concluded that the available data did not provide a clear scientific basis for prioritizing RPFs except for a preference for bioassay data over cancer-related endpoints. As a consequence, final RPFs were derived from bioassay data for any PAH that had at least one RPF based on a bioassay. viii DRAFT – DO NOT CITE OR QUOTE For each potentially carcinogenic PAH with bioassay data, the average RPF was calculated from bioassays with positive results. For those PAHs that did not have any estimated RPF based on a bioassay, but for which the weight of evidence evaluation indicated a potential for carcinogenic response (e.g., dibenz[a,c]anthracene), the final RPF was calculated from all cancer-related endpoint studies with positive results. In both cases, nonpositive results were not included in the calculation. The final RPF for each PAH was reported to one significant figure. The range of RPF values was also reported. Presenting the RPFs in this manner provides an average and maximum estimate for each PAH that has data from multiple studies. All tumor bioassay RPFs (across all exposure routes, species, sexes, and including both tumor incidence and tumor multiplicity RPFs) were combined to estimate the mean and range, except as follows. In some cases, two separate RPFs were calculated in the same group of animals. There were two situations in which this occurred: RPFs for different target organs in the same animals, and RPFs based on incidence of tumors and tumor count in the same animals In these instances, the higher value of the two RPFs was included in the average and range, and the lower value was dropped from the combined data. Several options were considered for the determination of a final RPFs (e.g., arithmetic mean, geometric mean, weighted average, maximum, or order of magnitude estimates). The arithmetic mean and range were chosen as a simple approach to describing the calculated RPF values available for each PAH. Other statistical measures (i.e., geometric mean, weighted average) were not considered appropriate due to the limited number of RPF values calculated for most PAHs and the variability in the RPF estimates. Most PAHs (19/26, 73%) had ≤3 calculated RPF values and the range of RPF values was greater than an order of magnitude for several compounds (6/26 PAHs). The variability in RPF estimates is likely due to differences in study design parameters (e.g., route, species/strain, exposure duration, exposure during sensitive time periods, initiation vs. promotion and complete carcinogenesis protocols, tumor incidence vs. multiplicity reporting) and dose-response methods (modeled vs. point estimates). Calculation of a weighted average was not possible because there is no clear biological rationale for choosing among study types or tumor data outcomes. Providing order of magnitude estimates, as has been previously done for estimating RPFs for PAHs, was not considered to be superior to calculating simple means. Including the range in the estimated RPFs was considered to be informative to the user for characterizing uncertainty. Once a final RPF was derived for a given PAH, the resulting value was assigned a relative confidence rating of high, medium, or low confidence. The relative confidence rating characterized the nature of the database upon which the final RPF was based. Confidence rankings were based on the robustness of the database. For final RPFs based on tumor bioassay data, confidence ratings considered both the available tumor bioassays and the size and ix DRAFT – DO NOT CITE OR QUOTE consistency of the cancer-related endpoint database. The most important factors that were considered included the availability of in vivo data and whether multiple exposure routes were represented. Other database characteristics that were considered important included the strength of evidence of genotoxicity data and SAR information, the availability of more than one in vivo study, and whether effects were evident in more than one sex or species. Very low relative confidence was reserved for final RPFs based on cancer-related endpoint data only (e.g., dibenz[a,c]anthracene). An RPF of zero was only applied if the data implied high or medium relative confidence. Table 1 shows the average RPFs based on tumor bioassay data with their associated range and relative confidence ratings, and an overview of the tumor bioassay database (total number of studies, exposure routes tested, species tested, sexes tested) for each PAH. Table 2 shows the average RPF for dibenz[a,c]anthracene, the only RPF based on cancer-related endpoint data, with its associated range, relative confidence rating, and an overview of the database for this compound. x DRAFT – DO NOT CITE OR QUOTE