Review of the three papers in the BELLE Newsletter

Melvin E. Andersen, Ph.D., DABT, CIH

This issue of the BELLE Newsletter contains three contributions investigating potential mechanisms for non-monotonic dose response curves for animal and human carcinogens. The following remarks are not intended as a critical review of the other contributions in this issue of the Newsletter. Instead, I read the three papers together asking what they conveyed in regard to our current state of knowledge about low dose effects and especially in regard to current ideas about the dose response curves for these low doses. Do these contributions show some consistency of outlook; do they suggest common themes that need further evaluation; and, will the development and validation of specific mechanistic models lend enough support to observations of U-shaped dose response curves to alter current approaches to risk assessment.

While each paper provides a somewhat different perspective, all have certain similarities in approach and motivation. The first common characteristic is acceptance of the reality of complex U-shaped dose response curves. The second is a desire to create quantitative structural models to investigate these dose response behaviors based on clearly articulated biological hypotheses. A third is a curiosity about the impact of these behaviors for risk assessment. Despite success in showing that specific mechanistic models can account for these U-shaped curves, none of the papers automatically assume that a particular quantitative model has been proven by the modeling efforts. Instead, the consistency between model predictions and cancer or tumor promotion data indicate that these models may well be correct and encourage new research avenues or quantitative analyses that will bolster or refute these hypotheses. When further studies show these mechanisms to be consistent with a larger body of biological data, such models might eventually find practical applications in risk assessments.

The three papers do differ significantly in the level of biological details included in the models and in the extent to which they lead to specific research efforts for individual compounds. The Downs and Frankowski contribution is a theoretical framework that is more defined by the algebraic and mathematical structures than by a close association of model parameters with specific biological processes. This model includes endogenous and exogenous carcinogens interacting through saturable repair processes. The effective dose to

DNA is coupled to a one-hit model for neoplastic transformation. The authors outline the mathematics of cases that yield the U-shaped curves; they do not emphasize the biological insights obtained from modeling about the characteristics of the system that control these complex curve shapes. This model appears to predict reduced cancer risk at low doses when the exogenous carcinogen (i) reduces the effective dose of the endogenous carcinogen and (ii) is repaired more effectively than the endogenous carcinogen. In essence, there is an interaction between the background and the chemically-induced process such that the exogenous compound interferes with the carcinogenic process responsible for the background tumors. This theme is repeated in the other two papers.

The incidence curves for lung cancer from Radon in residential situations is consistent with a U-shaped dose response curve (Bogen; Figure 2). In his contribution, Bogen describes a mechanistic model for human lung cancer where radon has the potential to cause mutation by altering DNA bases or to kill cells via radiation-induced toxicity. This model is based on a two-stage type cancer model and contains 6 different cell types. Dose response for cell killing and mutation vary with cell type and the relative efficacy of these two processes vary for the different cells. This mechanistic model predicts reduced cancer incidence when cytotoxicity kills cells that constitute spontaneous premalignant clones without producing enough new premalignant cells by mutation to counteract the loss of cells due to toxicity. This paper discusses the role of modeling in hypothesis generation, in scenario testing, and in experimental design. This work shows clear plausibility for the model structure. What new studies are possible to assess whether this plausible model is likely to be correct? The second phase of model verification probably cannot be pursued easily through epidemiological observations. Verification will probably require a combination of epidemiology and specific studies in appropriate animal model systems to provide identification of cell types at risk and cell population kinetics in affected pulmonary tissues. Even in animals, it will be very difficult or impossible to identify and quantify these different cell types that are presumed to contribute to tumorigenesis in this model.

Our work on growth of altered clones after treatment with dioxin was largely motivated by the observations of Pitot et al. (1987) indicating a U-shaped curve for tumor promotion with this compound. We describe clone growth with a stochastic model that predicts growth of initiated cells in the presence of dioxin. A large pool of initiated cells is created by a protocol in which animals are treated with an initiator while the liver is undergoing a wave of cell proliferation. The clone growth simulations predict the growth of intermediate cells, more akin to the P and R cells in Bogen's Figure 1. This more narrow focus simplifies the analysis and the clearer definition of the experimental system also allows design of mechanistic studies in an animal model system.

While it is relatively straightforward to identify and quantify clones, it is impossible to insure that all these cells seen in the different clones have the same genetic changes. We believe that verification of the two-cell tumor promotion model will rely on a variety of data. First, further studies of other promoters should be helpful in demonstrating the reality of U-shaped dose response curves for other liver tumor promoters. From our perspective, the critical questions in completing this generic two-cell promotion/gene induction model are more directed toward the manner in which promoters affect growth characteristics of individual cells and initiate proliferation and subsequent mitoinhibition. These research questions are more accessible to study and interpretation than are initiation studies in which a variety of altered cells of unknown genotype contribute to the population of observable clones.

Our clone growth model proposes that the initiation protocol produces two different cell types. One cell type grows out to observable clones of altered cells in the absence of dioxin. The growth of these cells is inhibited by the mitoinhibitory growth environment caused by dioxin treatment. The second cell type has mutations that reduce sensitivity to the mitoinhibitory growth environment. This second cell type derives a growth advantage in the presence of dioxin and the mitoinhibitory growth environment. If the inhibition of growth of the first type cell occurs at lower concentrations than the growth advantage conferred on the second cell type, fewer clones would be expected compared to the control. This leads to the U-shaped curve in Figure 2; Panel A.

What is the importance of the U-shaped curve for comprehensive evaluations of the effects of these compounds? For dioxin, the two-cell model, if corroborated, would clearly argue for non-linear approaches to carcinogenic dose response analysis for these liver tumors. However, the downward moving portion of the U-shaped response curves does not occur because small doses are healthful. They occur because the effects of the substance on specific cells vary. The radon model proposes low dose cytotoxicity for specific cell populations; the dioxin model proposes low dose mitoinhibitory effects on growth characteristics of these intermediate cells. The U-shaped curves simply are representations of the interaction of multiple contributing biological mechanisms at the cellular level that give rise to a larger macroscopic behavior at the organ level. Low dose cytotoxicity or low dose mitoinhibition have to be considered in relation to effects on other organ systems throughout the body.

All of these three models show that U-shaped dose response can be accounted for by very conventional hypotheses for effects of chemicals on DNA-repair, mutation and cell proliferation rates. We all need to remain open-minded about the presence of complex dose response curves and inquisitive about the mechanisms by which they can be achieved. My first introduction to toxicology was the evaluation of acute toxicity curves for 1,1-dichloroethylene. These curves showed a maximum at several hundred mg/kg and decreased for lower or higher doses (Andersen et al., Environ. Health Perspectives, 977). I spent a considerable effort over the course of two years simply convincing colleagues that these curves were reproducible and deserved serious attention.

In each of the three papers here, the U-shaped behaviors occur because chemical treatment has differential effects on the on-going process producing background lesions and on the neoplastic processes associated with chemically-induced tumor formation. The predicted behaviors from these models actually contradict another tenet of risk assessment ñ additivity to background. All provide evidence that the background processes and chemically-induced processes may interact to decrease the background contribution in the presence of low levels of added compound. Thus, they presume a case that is neither additive nor independent to background. The chemicals actually inhibit background neoplastic processes. As more targeted mechanistic studies are conducted confirming these mechanisms, current risk assessment procedures could be changed to acknowledge that low-dose linearity and additivity to background may not be appropriate assumptions for some classes of carcinogens.