Experimental Design and Statistical Analysis of Studies to Demonstrate a Threshold in Genetic Toxicology: A Mini-review

Abstract

A mechanistic understanding of genotoxicity is important for the risk assessment of the exposure of human populations to chemicals. The nature of the dose response relationship at low doses is valuable information in the evaluation of the biological importance of such exposures. A range of mathematical and statistical approaches have been used to try to characterize responses at these low doses. Methods include mathematical models which do or do not include thresholds and statistical methods which try to identify No-observable eŠect levels (NOELs). It is important to appreciate that determination of an NOEL is not evidence for a threshold. There is an increasing appreciation of the potential to identify`pragmatic' thresholds using experimental systems with a range of biomarkers. The accurate characterization and estimation of these doseresponse relationships requires careful experimental design which can improve the accuracy of the estimates of the response while avoiding the introduction of artifactual eŠects. Statistical approaches such as Design of Experiment (DoE) methodology, which builds on the traditional factorial design, can provide e‹cient approaches for the description and estimation of dose-response relationships of both individual and combinations of agents. Estimation approaches such as the benchmark dose methodology and the concept of thresholds of toxicological concern provide practical methods for addressing the threshold problem.Key words: statistics, experimental design, threshold, genotoxicity IntroductionThe objective of this paper is to provide an overview of the statistical and experimental design issues involved in the design and interpretation of studies to identify thresholds associated with exposure to genotoxic agents.It has become an axiom that genotoxic chemicals induce DNA damage at any level of exposure and do not have a threshold in their dose-response relationships. Madle et al. (1), for instance, stated that``...it is generally agreed that there are no thresholds for genotoxic eŠects of chemicals, i.e., that there are no doses without genotoxic eŠects.'' This concept is the basis of risk assessment strategies for chemicals with genotoxic potential. These consider that genotoxic carcinogens do not have a threshold while those that act by non-genotoxic mechanisms may have a threshold (the default assumption in other areas of toxicology). Chemicals with thresholds are regulated via limit values such as ADI (Allowable Daily Intake) or TDIs (Tolerable Daily Intake) while non-thresholded chemicals are regulated using concepts such as ALARP (As Low As Reasonably Practical) or ALARA (As Low As Reasonably Achievable). (For recent reviews, see (2,3)).Genetic damage can be gene mutation, chromosome damage (clastogenicity) and chromosome loss (aneuogenicity) and is detected by batteries of short-term mutagenicity tests. Increasingly, some aneuogens, based upon their mechanism of action (MOA), are considered to have thresholds but that, in the absence of evidence to the contrary, gene mutag...

Cancer Risk at Ultra-low Dose: Lessons Learned from 40,000-animal Cancer Dose-response Studies

“…Thresholds: The concept of thresholds in genetic toxicology has received extensive recent review (5,(22)(23)(24)(25), and a number of reports have appeared presenting evidence for the existence of thresholds for genotoxic carcinogens. One example is a recent report by Fukiushima and colleagues (26), who reviewed the dose-response eŠects of 0.001-100 ppm dietary 2amino-3,8-dimethylimidazo[4,5-f ]quinoxaline (MeIQx) on several biomarkers of cancer risk in a medium term liver bioassay with 1145 male F344 rats.…”

“…Thus a data set forming a straight line by simple linear regression analysis is not synonymous with a dose-response showing true direct proportion``linearity''. Moreover, simple inspection of plotted data sets can mislead because plots of the same data set using diŠerent transforms will have quite diŠerent shapes, and interpretation without appropriate statistical procedures is perilous at best (22). For example a semi-log transform has been put forward as a superior means to establish threshold for the mega-mouse ED01 study (27) and for our trout DBC results (28).…”

“…For example a semi-log transform has been put forward as a superior means to establish threshold for the mega-mouse ED01 study (27) and for our trout DBC results (28). This approach, however, has raised concern on many levels including visual distortion, inappropriate statistical application, and absence of a convincing rationale for its application to cancer (22,(29)(30)(31). We elected to use logarithmic dose progression in our design, and log-log regression in the analysis of our data sets because this approach addresses the question of direct proportion in dose-response without assumption, is capable of analyzing unbounded data sets, avoids data compression, and lacks potential for visual distortion.…”