Experiments, Analyses and Decisions: Hormesis in Ecotoxicology

A. John Bailer, Ph.D.

Center for Environmental Toxicology and Statistics

Department of Mathematics & Statistics

Miami University

Oxford, Ohio 45056-1641

Tel: 513-529-3538

Fax: 513-529-1493

Email: ajbailer@miavx.acs.muohio.edu


We have been challenged by Peter Chapman to consider the implication of hormesis for both ecotoxicology and ecological risk assessment. A consideration of his proposal led to the following observations and comments. I echo his comments related to the need for experimental designs and statistical tools that accommodate and address hormesis. I expand on his comments related to the basic question of how adverse responses should be defined for hormetic concentration-response patterns for both individual species and populations. Finally, I reflect on the notion of the implication of incorrectly detecting or not detecting hormesis in an ecosystem.

Peter Chapman is to be congratulated on an outstanding summary of the implications of hormesis for ecotoxicology and ecological risk assessment. Discussing hormesis effects in contexts beyond the single-chemical testing controlled experiment is welcome. In particular, the notion of hormetic factors being related to abiotic factors such as habitat modification and more general biotic factors such as the introduction of new species in an ecosystem were insightful. I would like to touch on three topics in this reaction piece. First, what does hormesis require from experimenters in terms of designs and from analysts in terms of quantitative tools? Second, how much is too much for a hormetic hazard? Finally, so what? What if hormesis is present and you ignore it? What happens if hormesis is absent but you believe it is present?

The acknowledgement of the potential presence of hormesis requires reflection by both the experimenter and the biostatistician or data analyst. As noted by Chapman (Table 2, 2000), the possibility of hormesis would necessitate a wider range or spacing of exposures including more exposure concentrations at lower concentration levels. In addition, it is not uncommon for increased variability in responses to be observed along with increasing mean responses. Thus, the concentration range in which hormesis is expected to occur might reflect the most variable region of the concentration-response gradient. This is not an intractable problem. Standard strategies of analysis would suggest over-sampling conditions that are expected to be more variable. In other words, there is no reason to test an equal number of organisms at each and every concentration condition. You might have done previous range finding experiments which suggest that large responses are expected at the highest concentration conditions. If this was an inhibitory response, say number of young produced, then the concentrations associated with minimal number of young will likely exhibit the least variability in response. Thus, investing the same number of organisms at these extreme concentrations may not be as fruitful as reallocating some of these resources for use in the lower exposure conditions. Now this corresponds to a departure from the fairly standard recommendation of equal replication at each exposure condition; however, this is consistent with Chapman's call for a new paradigm for assessment. Chapman also issues the call for statistical techniques that are not confining in analyses of concentration response patterns. He notes regression-based methods that either incorporate effective concentrations as explicit parameters in the models (Van Ewijk and Hoekstra, 1993) or explicitly define potency estimators to reflect inhibition relative to control, zero exposure, response patterns (Bailer and Oris, 1997; Bailer et al. 2000). I believe that both of these strategies reflect improvements over techniques that require restrictive assumptions about concentration response patterns or that attempt to define no effect levels based upon some statistical testing criterion. The primary focus of these methods has been on single stressor testing in a single populations of organisms. Research is required to consider how these methods would be applied in multi-contaminant contexts in which mixtures of stressors are simultaneously encountered. It may be possible to consider multiple predictor generalizations of these models where each predictor corresponds to a different stressor or to interactions of stressors. One possible mixture problem that has been examined using regression-type estimators is setting exposure limits to effluent. In these studies, dilutions of an effluent are tested to explore their impact on some biological response. Since an effluent may contain a rich mixture of different stressors along with certain essential elements, this allows for one natural application of these current regression models to mixtures. While this may help with evaluating a mixtures effect on a single population of organisms, the question of how to best evaluate and define regulatory limits for a stressor in a community or an ecosystem remains open. While some sort of community-level exposure experiments (e.g. pond-level, mesocosm experiments) might be of appeal, it is not clear that you would ever to be able to have sufficient replication of conditions to detect hormetic responses. Alternatively, mechanistic or population modeling to integrate the different effects of the stressors on community structure would have to be investigated.

My second set of observations relate to the general question of how should one set a regulatory exposure limit for a stressor that exhibits a hormetic concentration-response pattern. This is a non-trivial question even in the trivial context of a single chemical stressor exposure in a laboratory. If we consider inhibition to reflect a toxic response (e.g. reduced reproductive output), then should we consider toxicity as being defined as relative to a baseline of responses in a zero exposure condition? Alternatively, inhibition relative to a baseline of maximally induced response may make sense. Bailer and Oris (2000) considered the estimation of inhibition relative to different baselines; however, they did not address the question of hormesis being positive, neutral or adverse as we are challenged to consider by Chapman (2000). If the hormetic response reflects a positive change, then inhibition relative to a baseline of maximal response seems justified. If the hormetic response is neutral or adverse, then defining inhibition concentration relative to the responses in unexposed populations would appear justified. Paralleling the concerns raised in the previous paragraph, the definition of impact is clearer for a single stressor (or dilution of a stressor mixture) effect on a single response in a single population of organisms. It is not uncommon for stressors to modify more than one response. For example, high effluent concentrations may not only inhibit reproduction in daphnids but may also lead to mortality before reproduction might occur. In this case, simultaneous modeling or hierarchical modeling of the responses might be in order. A nice example of this hierarchical modeling was given by Wang and Smith (2000) in which they implemented a mixture model where the observed number of offspring was the product of two variables: the true number of offspring; and a dichotomous mortality indicator. A logistic model was used to model mortality while a Poisson model was used to model the number of offspring conditional on survival. As an alternative to hierarchically modeling responses, some multivariate statistical models might be employed when multiple responses are being considered. It is not clear how a potency estimate would be defined in the context of multivariate response patterns. What if only some of the responses measured exhibited hormetic response patterns? How would you evaluate the organisms experience in this circumstance? The complication of considering multivariate response patterns for multiple species living in community is an order of magnitude more difficult still.

Finally, so what? What if hormesis is present and you ignore it? What happens if hormesis is absent but you believe it is present? These are the classic health screening questions reflected in false negative and false positive errors. Figure 1 in Chapman's essay, suggests that ignoring hormesis when it is present may lead to selecting an exposure limit for an essential metal or metalloid that leads to a deficiency. In a seemingly paradoxical way, a lower exposure limit may be more hazardous than a higher exposure level. Alternatively, defining a hazard level assuming a positive hormetic effect is present could lead to exposure limits that are not sufficiently protective.

Ultimately, better understanding of mechanism may support the use of hormetic concentration-response patterns in ecotoxicology. This will require a tremendous effort to appropriately integrate this information into analyses. Chapman's essay provided us with both a challenge and a call for such work.


Bailer, A.J., and Oris, J.T. (1997) Estimating inhibition concentrations for different response scales using generalized linear models. Environmental Toxicology and Chemistry 16: 1554-1559

Bailer, A.J. and Oris, J.T. (2000) Defining the baseline for inhibition concentration calculations for hormetic hazards. Journal of Applied Toxicology 20: 121-125.

Bailer, A.J., Hughes, M.R., Denton, D. and Oris, J.T. (2000) An empirical comparison of effective concentration estimators for evaluating aquatic toxicity test responses. Environmental Toxicology and Chemistry 19: 141-150.

Bailer, A.J., Elmore, R.T., Shumate, B.J. and Oris, J.T. (2000) A simulation study of characteristics of statistical estimators of inhibition concentrations. Environmental Toxicology and Chemistry 19: 3068-3073.

Chapman, P. (2001). The implications of hormesis to ecotoxicology and ecological risk assessment (ERA). BELLE Newsletter, 10(1):this issue.

Van Ewijk, P.H. and Hoekstra, J.A. (1993) Calculation of the EC50 and its confidence interval when subtoxic stimulus is present. Ecotoxicology and Environmental Safety 25: 25-32.

Wang, S.C.D. and Smith, E.P. (2000) Adjusting for mortality effects in chronic toxicity testing: mixture model approach. Environmental Toxicology and Chemistry 19: 204-209.