This summary was written for the EPA STAR program, August 2007.

My dissertation work has focused on models for evaluating the effects of combination exposures. Although this problem has been studied for decades, there is still little consensus on the best ways of evaluating whether a particular combination is additive, synergistic, or antagonistic. This is important both in assessing the additivity of an experimental combination exposure-and thus learning about its mechanism-as well as for predicting the joint action of a mixture, a critical function of risk assessment.

Toxicologists define interaction as a departure from a model for additivity, usually called a “null model”. The null model connects the dose-response relationships of chemicals when given individually to the joint effect of a combination of those chemicals. Correct choice of a null model is the key first step in assessing interaction; departure from the model then indicates synergy or antagonism. By contrast, epidemiologists rely on definitions of interaction which proceed from theoretical models of binary exposures and binary outcomes. Although these models are carefully developed to contribute to an understanding of confounding and bias, continuous exposures and outcomes are rarely treated.

The first part of my work has been focused on the toxicologic null model of “concentration addition”, which is used for combinations that work by a similar mechanism and so may be substituted for each other. Concentration addition, which is commonly used by toxicologists, cannot accurately model combinations which include partial agonists (i.e., chemicals which provide a less-than-full response in the system under study). We have developed a simple extension of concentration addition which allows for inclusion of partial agonists in mixtures. Because partial agonists compete with full agonists for receptors, but have a lower response, they can antagonize the effect we would expected from the full agonist given alone. Using a cell culture assay, we have tested combinations of dioxin, polychlorinated biphenyls, and other agonists and antagonists of the aryl hydrocarbon receptor. These experiments have shown that this “generalized concentration addition” model can accurately predict the joint effects of these agents. The new model may provide a more sophisticated tool for assessing the risks of mixtures of similarly-acting agents.

The second part of my dissertation has involved comparing the different conceptions of additivity and interaction in toxicology and epidemiology. Some null models, like that of “independent action”, are used in both fields; however, the toxicologic model of concentration addition has no counterpart in epidemiology. Indeed, concentration-additive joint effects, which would be considered non-interactive by toxicologists, do not meet the epidemiologic criterion for non-interactivity, and may be perceived as synergistic or antagonistic depending on the shapes of the individual dose-response relationships. These two fields have somewhat different emphases; while toxicologists are more likely to be searching for mechanistic detail in carefully controlled laboratory studies, epidemiologists typically look for public health outcomes among highly complex causal webs dominated by human behaviors. Consequently, each field's models have strength and weaknesses. For example, epidemiologists have thought about the problems of confounding and bias, typically ignored by toxicologists, while toxicologists are more attuned to the issues involved in the use of continuous exposure data. By exploring each theoretical framework from the perspective of the other, we can improve our understanding of interactions in both fields.

As the number of compounds in use increases, and the opportunities for combined exposures grow, a sophisticated understanding of how exposures work together to cause an effect becomes more and more important. By contributing to the development of better null models, this project improves both our biological and our public health understanding of joint exposures.