Assessment of carcinogenic hazard of chemical mixtures through analysis of binary chemical interaction data.

Assessment of the potential health hazard of environmental complex chemical mixtures is one of the most difficult and challenging problems in toxicology. In this article, we describe the development of an innovative computerized system for ranking and predicting potential cancer hazard of chemical mixtures. We take into consideration both the additive risk of individual carcinogens present and the projected overall interaction effect of the mixture based on analyzing and integrating the possible interaction effects of all binary pairs of individual constituents of the mixture. Using this system, it can be predicted that a number of mixtures of polycyclic aromatic hydrocarbons should have a carcinogenic risk lower than that calculated by the simple additivity model, whereas the reverse is true for a number of other mixtures. The system can be very useful in hazard ranking and priority setting in dealing with mixture problems such as cleanup of hazardous waste.


Introduction
Assessment of the potential health hazard of environmental complex chemical mixtures is a difficult and challenging toxicological problem and a subject of major current concern to both the scientific and regulatory communities (1,2). This is particularly true when dealing with cancer as the toxicological end point. Besides the usual problems associated with risk assessment of individual chemicals, three major obstacles associated with mixtures are a) the impossibility of testing myriads of possible combinations of chemicals; b) the lack of a universally accepted index for quantitative measurement of cancer risk of chemical carcinogens; and c) the uncertainty of the possible outcomes of interactions among the various constituents of the mixture.
This article was presented at the IV European ISSX Meeting on Toxicological Evaluation of Chemical Interactions: Relevance of Social, Environmental and Occupational Factors held 3-6 July 1992 in Bologna, Italy.
This project was funded in part with Federal funds from the U.S. Environmental Protection Agency under Contract No. 68-D90007. The scientific views expressed and conclusions reached in this paper are solely those of the authors and do not necessarily reflect or represent the views and policies of the U.S. EPA or of SAIC.
Address correspondence to Yin-tak Woo Despite this uncertainty, there is clear realization that the toxicologic evaluation of complex chemical mixtures is becoming increasingly important for the hazard assessment of environmental sites (3). The epitome of this troublesome uncertainty are the problems of assessing the relative hazard that hazardous wastes disposal sites represent (4) and assessing the priority for cleanup and decontamination. Beyond this acute example, the problem also surfaces in connection with hazards represented by industrial effluents, pollutants in water and air, industrial products consisting of mixtures of chemicals, as well as complex human and veterinary medicinal preparations.
During the past six years, the U.S. Environmental Protection Agency (EPA) and the National Cancer Institute (NCI) have been involved in systematically compiling databases on combined effects of binary mixtures of carcinogens (5), carcinogens and inhibitors (6), and carcinogens and promoters (7). These databases provide crucial information needed for mixture assessment. In this paper, we describe an innovative and pragmatic approach for integrating the information extracted from these databases. We also describe the mathematical, statistical and toxicological considerations required to develop a computerized system, the Integral Search System (ISS). This system is capable of ranking and predicting the potential cancer hazard of complex chemical mixtures.

Conceptual Prnciples Used in Development o ISS
The principles and assumptions used in the development of ISS are described in the User's Manual that accompanies the software (8). Essentially, ISS consists of two major components: a component to calculate the "inherent (cancer) hazard" of the mixture based on an additivity model, and a "hazard modification" component to modify the "inherent hazard" by a "weighting ratio," calculated by analyzing and integrating the possible interaction effects of all the binary pairs of individual constituents of the mixture. Inherent Cancer Hazard Component. The calculation of the "inherent hazard" using the basic additivity model is based on the assumptions that the concentrations of various chemical carcinogens present in the environmental mixtures are usually relatively low with respect to the dose-response curve, and the effects of constituents of a chemical mixture do not interact with each other. Under these conditions, the individual effects of various constituents present in the mixture can be combined in an arithmetically additive manner to provide an estimate of the overall effect. Owing to the lack of systematic method to correct for interaction effects, this basic additivity model has been virtu-ally the only method used for risk assessment of chemical mixtures (9). Ideally, the additive effect should be calculated by taking the concentration of each chemical constituent into consideration (i.e., overall effect = S concentrationi Y potencyi). However, realistically, for virtually all environmental mixtures, information on concentrations of constituents is either incomplete or nonexistent. Thus, for ISS, the overall effect is calculated by summing the potencies of individual constituents (i.e., overall effect effect= I potencyi).
There is no consensus on a universally accepted index for the potency of carcinogens. In ISS, for carcinogens with sound dose-response data, the slope factor* (ql*) has been used as an index of potency. For carcinogens with no qj* values and for chemicals strongly suspected to be carcinogenic, structure-activity relationship (SAR)t analysis may be used to provide a rough estimate of their potency. The use of SAR analysis allows inclusion of virtually any type of compound, thereby giving a more realistic estimate of the inherent hazard. Besides SAR, other alternative indexes or scales (e.g., TD50, RfD*) may also be used when sufficient data are available.
To allow conversion between the ql* values and concern level scales, a correspondence table (Table 1) has been developed based on analysis of 134 carcinogens with ql* values to reach a reasonably even distribution among narrative concern level scales. The conversion enables calculation of the arithmetic sum of the potencies of all carcinogenic constituents to give the inherent hazard of the mixture. Since interaction often modifies hazard by orders of magnitude, the inherent hazard must be converted to a linearized scale (exponent index) before application of the weighting ratio (\[R).
Hazard Modification Component.
One of the unique and powerful features of The Slope Factor Iqli is the slope of the straight line from the upper bound risk at zero dose producing an upper bound risk of 1%. It expresses the cancer risk (proportion of population affected ) per unit of dose (in mg/kg/day or mmole/kg/day) 110). Ql can be calculated using commercially available software, such as Global 86. The q* values of a number of carcinogens are available in on-line database such as U.S. EPA's IRIS database ( 11). tSAR analysis (12) has been successfully used at the US EPA 113,14) to screen for potential cancer hazard of new or untested chemicals. Typically, a concern level rating ranging from low to high is developed for a chemical, based on SAR analysis, by a panel of experts Ithe Structure Activity Team or SAT). An expert system, based on formalized knowledge rules developed by SAR experts, to give concern level rating is being developed for the U.S. EPA. aSlope factor (ql*) expressed as cancer risk (proportion of population affected) per mmole/kg/day. bA linear scale of hazard indicators which parallels the ranking sequence of the exponents of the slope factors. CA narrative representation of hazard concern levels of chemicals based on SAR consideration.
ISS is its capability to project a realistic estimate of the overall interaction effect of the mixture based on available interaction data and chemical class assignments. There are four standard response categories (SRCs) for interactions § among chemicals that may cause deviation from the basic additivity model: synergism (syn), promotion (pro), antagonism (ant), and inhibition (inh). The former two represent hazard-amplifying interactions while the latter two represent hazard-reducing interactions. Depending on the relative balance of the totality of hazard-amplifying and hazard-reducing interaction effects, the inherent hazard level may be modified upward or downward. Two important and reasonable assumptions have been made. First, the overall interaction effect can be calculated by integrating the individual interaction effects of all the binary pairs of various constituents of the mixture. Second, in the absence of interaction data on any given binary pair of chemicals, the interaction of the binary pair can be approximated by the predominant interaction effect associated with the binary pair of chemical classes to which each chemical belongs. The hazard modification component consists of five segments: (a) a computer program to generate all possible binary pairs of chemical constituents of mixture, (b) a master database integrating the three U.S. EPA/NCI databases on binary interaction of carcinogen with carcinogen/promoter/inhibitor, (c) search capabilities to locate interaction "hits" (HA) for each binary pair of chemicals, (d) mathematical formalism to calculate the adjusted "inferred" values (HB) for each pair of tRfD, the Reference Dose method has been used by the US EPA for quantitative risk assessment of noncancer end points ( 11). There have been some recent proposals that the RfD method be used for quantitative risk assessment of chemical carcinogens which have been clearly shown to act via nongenotoxic mechanisms. chemicals for which the interaction cannot be located during the search but can be inferred by association to a pair of structural or functional classes to which the chemicals belong, and (e) mathematical formalism to calculate the weighting ratio. Segment A. The number (N) of combinations (binary, ternary, etc.) that can occur between the number of individual chemicals present in a mixture is given by the factorial formula: where k is the multiplicity of the combination. For binary mixtures (k = 2), it can be shown that the number of binary pairings increases rapidly with the number of chemicals (e.g., N= 45 for 10 chemicals; N= 4950 for 100 chemicals). Owing to the speed of computers, it is feasible to establish binary pairing between any number n of identified compounds in a mixture. ISS contains a program to generate such a set of binary pairs to search for interaction "hits" in Segment C. Segment B. The ISS master database, integrating and standardizing the three U.S. EPA/NCI databases on binary inter- §The interaction effects involving chemical carcinogens fall into three types: (A) When both chemicals are established carcinogens, the interaction effects may be: I/) additivity, when the combined effect is additive, (ii) synergism, when the combined effect more than additive, and liii) antagonism, when the combined effect is less than additive. 1B) Inhibition, when exposure is to an established carcinogen at an effective dose, together with or preceded or followed by exposure to a noncarcinogenic chemical agent that is an inhibitor of the carcinogenic response. IC) Promotion/cocarcinogenesis, when exposure is to an established carcinogen at subcarcinogenic ("subthreshold") dosels) or to a very weak carcinogen, and this exposure is followed by lin the case of promoter) or simultaneous with lin the case of cocarcinogen) exposure to a noncarcinogenic chemical agent that significantly enhances the carcinogenic response.
Environmental Health Perspectives action of carcinogen with carcinogen/promoter/ inhibitor, contains carcinogenesis binary combination effects about 1000 chemicals of 60 structural/functional classes. It is structured to allow searching by specifying binary pairs of individual chemical names, as well as binary pairs of chemical class terms. Segment C. The ISS search program is designed to search for matching pairs between the set of binary pairs generated in Segment A and those present in the master database to locate interaction "hits" of the four SRCs that may cause deviation from the basic additivity model. For each binary pair of individual chemicals, a "name pair hit" is registered if interaction data on this specific pair is located in the master database. The number of name pair hits of the mixture is then tallied to give the total number (HA) for each of the four SRCs.
Segment D. For each binary pair with no name pair hit, ISS continues to search for interaction hits associated with the binary pair of structural/functional classes (class pair hits) to which the chemicals belong. Since class pair hits only represent possible interaction based on class association, they are given less weight than name pair hits. For each SRC, the raw score of the total number of class pair hits is tallied and then statistically adjusted to give the class pair inferred value (HB). The statistical adjustment of raw score of class pair hits involves consideration of the number of documented interactions between each class pair in the ISS database, the distribution of the type of SRC effect within a class interaction, and the representativeness of classes in the ISS database (8). The equation derived for these adjustments is: [2] HB ef[f(oct)2y  For each SRC, the total interaction effect of the mixture with respect to that particular SRC is the sum of HA and HB: Hsyn = HA syn + HB syn' [3] Segment E. The extent of hazard modification due to interaction is dependent on relative balance between hazard-amplifying and hazard-reducing interactions. This can be best represented by the general form:

WR Hazard-Amplifying Interaction Effects
Hazard-Reducing Interaction Effects where the p, q, r, and s are hazard-modification effectiveness coefficients which reflect the effectiveness of the four types of combination effects to modify the inherent hazard level of chemicals. These coefficients are empirical and should reflect the user's perspective of the entire combination effects literature as well as conceptual biases. It is the present view of the authors thatp = 0.3, q = 0.7, r = 0.3 and s = 0.6 are reasonable values. However, it is only through extended testing of this ranking system and additional experimental data becoming available in the future that an increasingly accurate set of values for these coefficients will be reached.
The presence of the unit number in both the numerator and the denominator of the WR provides an equation that will yield a working and realistic WR value even in limit circumstances when the hazard amplifying/reducing ratio would become zero or infinity either as a result of actual data or because of partial absence of data. Furthermore, this equation will yield WR = 1, and will leave the inherent hazard invariant when the hazard amplifying/ reducing ratio is 1 because of complete balance of hazard amplifying/reducing effects, or 0, because of total absence of data on combination effects.
Substituting for Heffect terms, the equation for WR is: 1+p(HA syn + HB syn )+ q(HA pro +HB pro) [6] Where Combining the outputs from the inherent hazard component and the hazard modification component, the relative hazard ranking value of any complex mixtures with known constituents can be computed by multiplying the inherent hazard (in unit of exponent index) of the mixture by the weighting ratio. The numerical value thus obtained can be used for ranking purposes as well as for conversion back to concern level term using the conversion table in Table 1.

System Overiew
The ISS system is a menu-driven, userfriendly software system that can be used efficiently by personnel with minimal computer training. It is designed to be used on any IBM-PC compatible (preferably AT) computer with at least 512 kBytes memory and a hard disk with 10 mBytes of free disk space. The software was written in Clipper (Nantucket Software) using dBase III plus database files. Over 14,000 lines of source codes were used to write this program. Figure 1 presents the system logic flowchart which entails steps for the evaluation of the inherent hazard of the mixture, steps for binary pairing by chemical names and chemical class term assignments (CTA), Volume 102, Supplement 9, November 1994 and steps for search for hits in the ISS mas-Once the list of chemical names of the ter database and for computation of a compounds in a mixture is established and weighting ratio for adjusting the inherent the slope factors or concern level terms are hazard.
introduced, the program computes the inherent hazard. The compounds are then phenol; pyrene and urethan paired into all of the possible binary combinations of names. The ISS uses these name pairs to search for matching pairs present in the database. If a specific name pair is found in the database, that specific interaction (or interactions) is counted as a name hit and the search for that name pair is terminated.
Since not all possible chemical name pairs can be found in the ISS database, it is most probable that no interaction hit will be registered for many of the name pairs. Each of these chemical name pairs is then converted into its chemical class pair, so that a class interaction search can be carried out. If any interaction exists between the two classes in a pair, the extent of interaction is calculated and that inferred value is regarded as a class pair hit; the search for that class pair is then terminated.
As a safeguard against the possibility that some chemicals may have been missed (in a search by chemical class search criterion), because of improper or incomplete class term assignment (CTA), the system provides a listing of those chemical classes that did not intersect with any binary pairing of classes in the three databases ("no hit" classes). The chemicals in the no hit classes are identified and rerouted for "Criteria Review and Adequacy Control of Chemical Structural Class Assignment" to verify the correctness of the CTA, and the search may be repeated using the new CTA (if any).

Application to Sample Mixtures
To illustrate the use and the reasonable operation of the ISS system, four sample mixtures have been tested using the system. Table 2 summarizes the results of the ISSgenerated hazard modification weighting ratios (WR) for these mixtures. Mixture 1, a polynuclear aromatic hydrocarbon (PAH) mixture containing two potent carcinogens (benzo[a]pyrene and 7,12-dimethylbenz[a]anthracene) and three weak or inactive compounds, has a WR of 0.63 indicating that the combined effect is expected to be less than that calculated by using an additivity model. This is consistent with data review that the predominant interaction among PAHs is antagonism (15) and the experimental findings (16) that several PAH mixtures tend to have lower carcinogenic potential than that expected by adding the carcinogenic potentials of individual PAHs. Not all PAH mixtures are expected to interact identically. This can in fact be projected by the ISS system. A slight change in the chemical composition of mixture 1 (replace 3,9 Environmental  overall cancer hazard ql* becomes 4.06 x 106, which is in the high-concern range. This example illustrates the importance of taking interaction effects into account in assessing the potential cancer hazard of mixtures.

Conclusion
In brief, we have developed an innovative computerized system capable of giving approximate but realistic assessments of the potential cancer hazards of chemical mixtures. The ISS is the first scientific attempt where all known interactions in chemical carcinogenesis are brought to bear on the cancer hazard assessment of complex chemical mixtures. The information generated by ISS is highly useful to both the scientific and regulatory communities for hazard ranking and for stimulating strategic research.
It is important to realize that no conceptual development of any hazard ranking procedure can remedy the considerable uncertainties that may surround the experimental data on which the ranking is based. Great approximations, tenuous extrapolations, imperfect and incomplete study designs, and enormous data gaps are sometimes encountered in the literature on the carcinogenic effects of chemical combinations. Critically lacking are studies on dose-response relationships in chemical interactions relative to carcinogenesis. Among the mass of data that have been analyzed and tabulated in the ISS master database, studies involving dose-response relationship of the interacting partners of chemicals are virtually nonexistent. It is recognized that the validity of ISS cancer hazard assessments can be limited by this dearth of data because the dose-response relationships can be of critical importance in some instances where, depending on experimental conditions and levels of the interacting compounds, the same combination may be hazard amplifying or hazard reducing (19). Furthermore, there is no standardization or accepted method of evaluating or expressing the intensity of the different combination responses. These uncertainties and limitations must be kept in mind when evaluating the carcinogenic effects of chemical combinations.
Aside from its capability to assess the potential cancer hazards of chemical mixtures, ISS is also a unique system in which all scientific data on binary interactions _I 71 I Figure 2. Printout of the ISS screen showing the profile of interaction "hits" and calculation of the weighting ratio (WR) for mixture 3. (See Table 2 for composition of mixture 3.) Figure 3. Printout of the ISS screen showing the application of weighting ratio to modify the sum of slope factors (ql*) calculated by using additivity model to arrive at the final concern level for mixture 4. (See Table 2 for composition of mixture 4.) dimethylbenz[a]anthracene with cyclopenta[cd] pyrene) yields a mixture (mixture 2) with a higher WR of 0.81 reflecting a partial offsetting of the predominant antagonism among PAHs by the known synergistic interaction (17) between the specific pair of benzo[a]pyrene and cyclopenta-[cd]pyrene.
In contrast to the hazard-reducing interaction among PAHs, the WR of mixtures 3 and 4 exceeds 1.0, indicating that the combined effect is expected to be greater than that calculated by using an additivity model. The ISS-generated screen which shows the interaction profile that contributes to a WR of 1.5 for mixture 3 is shown in Figure 2. There are seven name pair hits*-four hazard-enhancing (three promotion and one synergism), two hazard-reducing (two inhibition) and one neutral (i.e., additive)-and a variety of class pair hits. Using the formulae and statistical considerations previously described, these hits yield an overall WR of 1.5.
The ISS-generated WR for mixture 4 is 2.52 mainly because of multiple promotion hits with few or no hazard-reducing hits. The impact of applying this WR to the assessment of potential cancer hazard is depicted in Figure 3. There are no qj* values available on any of the compounds in mixture 4. Using SAR consideration, a concern level of low-moderate has been assigned to benz[a]anthracene and urethan. Croton oil is given a concern level of marginal. Using the correspondence table (Table 1), these concern levels can be converted to estimated q,* values (1 x 100 for low moderate and 1 x 10-2 for marginal) for the purpose of calculation of overall inherent hazard. Using the additivity model, the estimated overall "inherent haz- The actual records of interaction data can be retrieved using the databases (5-7) or search softwares (15,18