ToxSci Advance Access originally published online on September 1, 2006
Toxicological Sciences 2006 94(2):368-378; doi:10.1093/toxsci/kfl098
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Hormesis Outperforms Threshold Model in National Cancer Institute Antitumor Drug Screening Database
* Department of Public Health, Environmental Health Sciences Program, Morrill I, N344, University of Massachusetts, Amherst, Massachusetts 01003
Department of Mathematics and Statistics, Lederle Graduate Research Tower, University of Massachusetts, Amherst, Massachusetts 01003
Department of Public Health, Biostatistics and Epidemiology Program, School of Public Health and Health Sciences, Arnold House, University of Massachusetts, Amherst, Massachusetts 01003
Department of Biology, College of the Holy Cross, O'Neil Hall, 107, Worcester, Massachusetts 01610-2395
1 To whom correspondence should be addressed. Fax: 413-545-4692, E-mail: edwardc{at}schoolph.umass.edu.
Received March 31, 2006; accepted August 30, 2006
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Which dose-response model best explains low-dose responses is a critical issue in toxicology, pharmacology, and risk assessment. The present paper utilized the U.S. National Cancer Institute yeast screening database that contains 56,914 dose-response studies representing the replicated effects of 2189 chemically diverse possible antitumor drugs on cell proliferation in 13 different yeast strains. Multiple evaluation methods indicated that the observed data are inconsistent with the threshold model while supporting the hormetic model. Hormetic response patterns were observed approximately four times more often than would be expected by chance alone. The data call for the rejection of the threshold model for low-dose prediction, and they support the hormetic model as the default model for scientific interpretation of low-dose toxicological responses.
Key Words: hormesis; threshold; dose-response; yeast; NCI; U-shaped; J-shaped; bell-shaped; risk assessment; carcinogens; chemotherapeutics; cell proliferation; Saccharomyces.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
The threshold dose-response model has long been recognized as the dominant dose-response model in the biological sciences, including pharmacology and toxicology (Clark, 1926
, 1933, 1937). The threshold model dominates discussion in the leading pharmacological (Hardman and Limbird, 2001) and toxicological textbooks (Eaton and Klaassen, 2001; Hayes, 2001), development of study designs that drive hazard assessment procedures for pharmaceutical and chemical agents, and risk assessment processes used by regulatory and public health agencies worldwide. Despite this fact and a history of broad acceptance in many biological disciplines, the assumption that the threshold model should be used has been recently challenged. An alternative model, the hormesis model, has been proposed based on evidence of its generalizability by biological system, endpoint measured, chemical class tested (Calabrese, 2004, 2005; Calabrese and Baldwin, 2001a,b,c, 2003; Calabrese and Blain, 2005; Calabrese et al., 1999), and high frequency in the toxicological literature (Calabrese and Baldwin, 2001b). Using a priori entry and evaluative criteria, Calabrese and Baldwin (2003) reported that the hormesis model far outperformed the threshold model in a toxicological assessment using approximately 800 dose-response relationships that were broadly representative of commonly employed biological models, endpoints, and chemical agents. The present paper extends these findings by a systematic in-depth analysis of 56,914 dose-response studies in yeast. The analyses demonstrate that the hormesis dose-response model strongly outperforms the threshold model when applied to the extensive and highly standardized National Cancer Institute (NCI) tumor drug screening database using yeast as the test organism. |
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
This study utilized data from the NCI yeast anticancer drug screen, described in detail by Holbeck (2004)
and at the NCI Web site (http://www.dtp.nci.nih.gov/yacds/index.html). Briefly, data from stage 2, which contains the most promising compounds based on preliminary testing, were selected for evaluation; the agents were tested at five concentrations (1.2, 3.7, 11, 33, and 100µM) in 13 yeast strains. The yeast comprises a panel of Saccharomyces cerevisiae strains altered in DNA damage repair or cell cycle control genes, along with the wild-type (wt) strain without such genetic alterations. The NCI Web site contains a description of the genotype of each yeast strain used. The responses reported are derived from the fraction of growth of the yeast strain exposed to the compound relative to the growth of the same yeast strain treated with solvent (i.e., DMSO) control. Yeast cells in the exponential phase of growth were inoculated into synthetic complete medium containing 2% glucose and the test chemical. The starting cell density was 104 cells per well containing 200 µl of medium. (Julian Simon, personal communication). This study, like any analysis of preexisting data, has limitations based on the data that are available. Factors limiting the range of questions that we could analyze were the fact that the NCI database provides the average of two responses and the difference between them but not original optical densities or raw data permitting us to match chemicals to plates. Possible sources of variation were differences among columns or rows in 96-well plates and the lack of randomization of treatments within plates, plates within stacks, and stacks within the incubator. In the main, however, the large size, apparent quality control, and internal consistency of the NCI database minimize risks associated with experimental protocols, and we restricted our analysis to questions where experimental variation can be properly analyzed. Moreover, in each instance where such factors may be relevant to our analysis, we specifically point it out. Whenever a choice of assumptions was possible, we made assumptions so as to ensure that our analysis was conservative, in the sense that the bias, if any, would reduce the chance of observing hormesis.
Replication Procedure
Each chemical was tested four times at the same five concentrations in each of the 13 yeast strains (Table 1). Ninety-six–well plates were used, with 80 chemicals being tested at the same concentration (1.2, 3.7, 10, 33, or 100µM) on one plate, leaving the 16 peripheral wells for controls. Each concentration for that drug was incubated over the same 12-h period on a different plate such that there were five plates run on the same chemical at the same time. Of the 16 control wells, four were allocated to unexposed controls, eight to solvent controls, and four to controls using cycloheximide. All strains were cycloheximide sensitive, and the assay was deemed invalid if there was growth in the presence of cycloheximide. The order of the plates in the growth chamber with respect to chemical tested was systematic and not randomly allocated. The relative position of the controls on each plate was constant. The variability in response was designed to be maximized by using a different source of chemical on a different day and different yeast cultures in each different test (Julian Simon, personal communication). Slightly greater evaporation from the peripheral wells containing the controls may have caused control values to be several percentage points above a normal background. This potential bias was systematic for all plates and, if present, would introduce a negative bias with respect to discerning possible stimulatory responses (Faessel et al., 1999). To ensure that our analysis was conservative with respect to the detection of hormesis, no correction was made for this factor. Factors other than evaporation from peripheral wells may contribute to variation in data from 96-well plates. An analysis of such factors by Faessel et al. (1999) found that differences associated with positions of plates in stacks and of stacks in an incubator tended to be smaller than differences between the middle and edges of plates. Moreover, the importance of these sources of variation is minimal for our purposes as each plate contained its own set of controls.
|
The response data consisted of a ratio of the optical density (OD) of the response well for the treatment divided by the mean of the OD readings of the eight solvent control wells for each concentration. OD readings were at 600 nm. This process was repeated on the second day, and the ratios from the 2 days were averaged. We refer to the average response as the replication response. Two replication responses were produced for each concentration and for each strain/experiment. Data on the NCI Web site provide the average of the two response values and the difference between the two values, but the original OD values are not available. Data were also not available to match chemicals to plates to take advantage of plate effects. It is known that even if the control and treatment responses have the same mean, then the mean of their ratio will be greater than 100%, and the mean of the ratio approaches 100% as the variance of controls gets smaller (Casella and Berger, 2002
). This creates a slight bias in favor of a hormetic model, but since the design used eight controls for each response, the effect is slight. Assuming a lognormal model, if the coefficient of variation is 10%, then the mean of the ratio of a control and treatment response with the same mean is 100.1%.
Evaluation Strategy
Since the goal of this research is to evaluate whether there is nonrandom biological activity as measured by cell proliferation below the toxicological threshold, it is necessary to evaluate individual dose-response relationships. In the five-concentration protocol of NCI, the dose-response study should ideally have at least one concentration in the toxic (i.e., above threshold) domain, a concentration with a response that approximates the control response (i.e., the so-called no observed effect level [NOEL] or the highest dose that does not differ in a significant manner from the control) and several lower concentrations that would be evaluated for biological activity below this NOEL or threshold-like value. The NCI yeast database is unique and useful for this purpose but not ideal, in that some experiments show toxicity but insufficient doses below the toxicologic threshold. Our a priori entry criterion required at least one measurement at a concentration below that that was used to estimate the toxicological threshold. A response in the toxic zone at the highest concentration was considered desirable but not essential since lack of toxicity at the highest concentration in a screening bioassay may simply reflect an inadequate concentration range.
A two-stage approach was used to assess possible below–toxic threshold biological activity. The first involved estimation of the toxic threshold, while the second assessed the distribution of responses at concentrations lower than the estimated threshold concentration.
Threshold Estimation Strategies
Two strategies were used to estimate the toxicological threshold: the Benchmark Dose (BMD) and the NOEL. Since the results were similar, we only present the BMD results here. The Benchmark Dose 10 (BMD(10)) is the dose at which the response is estimated to have decreased 10% below control value (Crump, 1984). The BMD(10) was selected since 10% bounded the variability of the most variable yeast strain (i.e., response SD ranged from 3.0 to 7.5% for the 13 strains). Use of a BMD based on less than 10% (e.g., 2.5–7.5%) would yield progressively higher estimates of the frequency of hormesis for all parameters estimated (see "Results" section, Fig. 3 and "Supplementary Data" section), suggesting that the current approach (i.e., BMD(10) method) would lead to an overall underestimation of hormesis frequency. An identical analysis using the BMD(2.5) for all parameters reported for the BMD(10) analysis is given within the "Supplementary Data" section. Our method of estimating a BMD(10) is explained below (Fig. 1). Since our goal is to classify toxicity, we did not calculate the lower bound of a confidence interval for that dose. The BMD(10) approximates the control but probably entails a low degree of toxicity. It corresponds to a dose that is slightly higher than the toxicological threshold. This suggests that a dose immediately below and very close to the BMD(10) may itself be within the toxic zone (i.e., a slightly higher concentration than the actual toxicological threshold). This would become less likely with increasing distance between the BMD(10) and the concentration below the BMD(10). For example, for agents with a BMD(10) near 3.7µM, the 1.2µM dose would be close to the toxicity threshold. In contrast, for agents with a BMD(10) approaching 100µM, the 1.2µM dose would be nearly two orders of magnitude below the toxicity threshold.
|
|
A BMD(10) was calculated for each of the 28,457 (2189 chemicals and 13 strains) dose-response experiments using the average of two replications as response. The BMD(10) was estimated through the following procedure.
- The largest concentration with an average response below 90% is identified. Let this concentration be Cbelow, and let the associated response be Rbelow.
- If the average response at the next smallest concentration is at least 90%, then let this concentration be Cabove, and let the associated response be Rabove. The BMD(10) is estimated by linear interpolation on the log concentration scale:
- If the average response at the next lowest concentration below Cbelow is less than 90%, then let this concentration be Cbelow with response Rbelow and return to step 2.
In contrast to a linear or nonlinear regression approach to calculating BMD, the procedure described above is "local" in the sense that the BMD(10) is only calculated using the responses at concentrations that are adjacent to the BMD(10). Further, when this approach is used, the two concentration-response pairs that surround the BMD(10) are chosen using only concentrations above the BMD(10) and one concentration below the BMD(10). Responses at concentrations used to estimate below-threshold responses were not used in the estimation of the BMD.
Assessing the Distribution of Responses below the Toxic Threshold
We used two approaches to assess evidence of stimulated biological activity at concentrations below the estimated threshold of toxic response. The first is a pattern analysis that counts how often both replicates in each experiment were above and below (or equal to) 100% and compares those counts to expected values, assuming a threshold model. The second approach compares the frequency of responses at various levels above and below 100%.
Pattern analysis.
This approach categorized each concentration-response replication for all chemical-strain combinations and analyzed the patterns of responses at each concentration that were above and below (or equal to) 100%. In this approach, each chemical-strain repetition can express one of two different responses: H (response above 100%) and L (response less than or equal to 100%). A simple "fair coin" model was posited for the responses below the BMD(10) where each single replication would have a 50% chance of being above or below (or equal to) 100% and there is statistical independence across responses. This model assumes that the responses at concentrations below the BMD(10) have a median of 100%. It therefore describes a threshold model with minimal distributional assumptions. The fidelity between the observed data and this hypothesized model was tested.
Comparison of above/below-control values in the subtoxic zone of the dose-response.
The threshold dose-response model predicts that responses below the toxicological threshold should randomly vary on either side (i.e., above or below) of control group values (100% response). The hormetic model predicts that there should be a nonrandom stimulatory response (i.e., responses greater than 100%) below the toxic threshold. In order to test which model best accounts for the observed data, above-control (> 100% response) to below-control (
100% response) ratios were detailed for all yeast strains in the various BMD(10) classifications. The nonrandom distribution predicted by the hormesis model would be reflected in a greater frequency of responses above than below the control and in the magnitude of the deviation from the control.
Comparisons were made to responses above 100, 105, 110, 115, and 120% and then to below-the-appropriate-control group response using the formula:
 |
100%, > 105%/ 95.24%, > 110%/ 90.91%, > 115%/ 86.96%, and > 120%/ 83.33%. This methodology was used to take into account the possibility of an unrestricted stimulatory response while the maximum inhibitory response was fixed at zero. |
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Figure 2 describes the concentration-response relationships of the 13 yeast strains to the 2189 chemical agents tested. While there was little change on average from the control response at the lowest concentration (1.2µM), indications of average toxicity start to become evident at 3.7µM, progressing in dose-dependent fashion over the next three concentrations (11, 33, and 100µM). Table 2 shows the number of chemicals with BMD(10) values within each of six BMD(10) classification ranges for each of the 13 strains. The more toxic chemicals are included in the low BMD(10) range (e.g., < 1.2µM), while the chemicals with the lowest toxic potential comprise the highest BMD(10) categories. The data indicate that the wild type and SPY50780 yeast strains had the lowest number (139/2189 and 143/2189) of concentration-responses with BMD(10)s < 1.2µM, indicating that they were the least susceptible strains, a perspective that is supported by plotting of the overall data in Figure 2. In contrast, strains carrying rad50, rad50EPP+, rad18, rad52, and sgs1 were the most susceptible (Fig. 2, Table 2).
|
|
Table 3 is a summary of the below-BMD(10) mean responses and SD for each of the 13 yeast strains. The numbers of chemical-concentration relationships satisfying a priori entry criteria using the BMD(10) methodology are shown in Table 2. Of 28,457 concentration-responses to the 2189 chemicals, 16.7% (4763) gave no evidence of toxicity, having BMD(10) values
100µM. Assessments were performed on this subgroup of responses under the assumption that at higher concentrations, a toxic response would have occurred. When the BMD(10) is less than 3.7µM, there is no concentration that can be assessed for biological activity below the BMD(10), thereby not satisfying our entry criteria. There were 7558 such responses (3798 with BMD(10) < 1.2µM and 3760 with 1.2 BMD(10) < 3.7), accounting for 26.6% of the total responses. Therefore, 73.4% of the total dose-responses were evaluated. Similar findings were observed with the NOEL methodology (data not shown).
|
BMD(10) Response Evaluation
We averaged responses at concentrations when the concentration was below the BMD(10) for each strain. This resulted in averages of between 196 and 572 responses, with the mean and SD (Table 3). The below-BMD(10) mean values (Table 3) are generally consistent across each of the 13 yeast strains within a specific BMD(10) classification as well as across BMD(10) classifications. However, the mean values are modestly lower in the 3.7
BMD(10) 11µM group than in the other groups (p < 0.001), which do not differ significantly from each other. These trends are consistent with median values as well. Consequently, all 13 strains with each of the four BMD(10) classifications had average responses significantly greater than the control (p < 0.001 for each of the four columns in Table 3). These findings are consistent with a nonrandom distribution of responses in the direction of the hormetic dose-response. Findings with the NOEL methodology were similar (data not shown) except that the responses were usually several percentage points higher per strain than for the BMD(10) methodology. Similarly, if a smaller BMD(2.5–7.5) cutoff point were used instead of the BMD(10), the mean responses become progressively higher as the BMD value decreases (Fig. 3 and supplementary data, Table S1).
According to the threshold dose-response model, the distribution of responses below the estimated threshold (e.g., BMD(10)) should approach a 1:1 ratio for above- and below-control values. This was assessed for each BMD(10) classification group for responses > 100/ 100%, > 105/ 95.24%, > 110/ 90.91%, > 115/ 86.96%, and > 120/ 83.33%. Alternatively, one could use a different model and assume the equivalency of a symmetrical response (e.g., > 120%/ 80% rather than > 120%/ 83.3%), but this paper used the prior and more conservative approach. This approach was selected in order to make hormesis more difficult to detect. Table 4, A–D, indicates the distribution of responses below the BMD(10) for the chemicals in the various BMD(10) ranges. For example, the distribution of responses in the 33 BMD(10) < 100µM classification range (Table 4, C) for the 13 yeast strains is nonrandomly distributed in the direction of a hormetic response regardless of the degree of variability in the data. The findings are inconsistent with the threshold model, which predicts a ratio closely approximating 1:1. A comparison of the respective BMD(10) classification groups reveals that in the large variation comparisons (i.e., > 110%/ 90.91%), the proportion of stimulatory responses exceeds those on the "below" side by threefold to over 10-fold. The comparisons are generally similar among the 11 BMD(10) < 33µM, 33 BMD(10) < 100µM, and BMD(10) 100 classifications. While the 3.7 BMD(10) < 11µM classification (Table 4, A) also shows an excess of above-control values, the magnitude of the above/below differential is notably less. The most likely explanation for the reduced response in the 3.7 BMD(10) < 11µM classification is that responses at the 1.2µM concentration in these experiments may have displayed toxicity for some of the chemicals since 1.2µM is very close to the BMD(10) value. Figure 4 provides a simplifying summary of the information in Table 4. The comparisons in Table 4 were also used to compute weighted average estimates of overall ratios of above-control responses to below-control responses. For experiments in the 3.7 BMD(10) < 11µM classification range, above-control responses were seen 1.96 times as often as below-control responses. For the 11 BMD(10) < 33µM, 33 BMD(10) < 100µM, and BMD(10) 100µM classifications, above-control responses were seen 4.64, 4.22, and 6.94 times as often as below-control responses, respectively. A weighted average calculated across all four BMD(10) classifications revealed that the above-control responses predicted by the hormesis model are 4.39 times as frequent as below-control responses. A similar assessment was performed using the BMD(2.5) (see supplementary data—Table S2A–D and Figure S1) with findings consistent with the BMD(10) analysis but even more supportive of the hormetic model.
|
|
Pattern Analysis
Patterns of response below the BMD(10) were compared for six log-spaced ranges of BMD(10)'s. Figure 5 presents these comparisons, based on observed responses that were not used in the calculation of the BMD(10), along with the expected counts under the fair coin threshold model. For instance, for BMD(10)s in the range 3.7
BMD < 6.3, the responses at 3.7µM and 11µM were used to calculate the BMD(10)s, and the figure summarizes the pattern of three responses for the replicates at 1.2µM. For the 11 BMD < 33µM range, there were five possible patterns for the four replicates at 1.2µM: 4H 0L, 3H 1L, 2H 2L, 1H 3L, and 0H 4L. Under the fair coin threshold model, we would expect the fractions of responses that fit those patterns to be 1/16, 4/16, 6/16, 4/16, and 1/16, respectively. Similar procedures were used with BMD(10) values ranging from 33 to 100µM.
|
Strikingly, for each BMD(10) category, the observed responses markedly skew toward patterns that have more "H" (> 100%) responses, and the skewness increases as the BMD(10) increases and toxicity decreases (p < 0.0001). Considering the all-H patterns (left most pattern in each panel), the observed patterns are 1.3, 1.7, 5.3, 6.8, 20.7, and 25.1 times more frequent than the expected counts as the BMD(10) increases from the lowest range (3.7–6.3µM) up to the highest (57–100µM), where 57µM is halfway between 33 and 100µM on the log scale. Further, there is a strong general pattern with the BMD(10) ranges with counts tending to decrease monotonically as the number of Ls in the pattern increases. An assessment using the BMD(2.5) reveals similar findings to the BMD(10) (Fig. 5) but even more supportive of the hormetic model (supplementary data, Figure S2).
Figure 6 plots the fraction of H responses for each strain at the lowest concentration (1.2µM) as a function of the log distance below the BMD(10). As the distance of the 1.2µM concentration below the BMD(10) increases, the frequency of having both replicated responses at this first concentration (1.2µM) being greater than 100% increases to almost 60%, while only 25% would have been expected by chance assuming a threshold model. The figure shows that as the 1.2µM concentration reaches a value of 1/4th of the BMD(10), the probability of responses consistent with the hormesis model become far more common than chance for all strains; for the more highly toxic agents in the lowest BMD(10) category, the response at 1.2µM is often below control values. A similar assessment was performed using the BMD(2.5). It revealed similar findings to the BMD(10) which were even more supportive of the hormetic model (supplementary data, Figure S3).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
The data indicate that responses to concentrations below the toxicological threshold for each of the 13 yeast strains tested with many hundreds of chemically diverse agents are nonrandomly distributed with respect to the control. A variety of complementary methodological evaluations (Tables 3 and 4, Figs. 4–6
) support the same interpretation. These findings indicate that the threshold dose-response model inadequately accounts for biological activity below the threshold. However, the results are consistent with predictions of the hormesis dose-response model. The conclusions based on this large database in yeast are similar to those made in earlier reports of Calabrese and Baldwin (2001b, 2003) using toxicological data representative of a broad range of biological models, endpoints, and chemical agents. The evidence of hormesis in yeast, along with the previous studies, is biologically significant and has potentially important implications for essentially all drug and chemical hazard assessment studies and risk assessments worldwide.
The similar overall response patterns below the toxic zone in all 13 yeast strains for the large number of chemicals in this extensively evaluated public database is a novel finding. Previous publications (see Holbeck, 2004, for a review) with this database have focused on the nature of the above-threshold (rather than below-threshold) responses and their underlying toxicological mechanisms since the goal of the NCI has been principally oriented toward identifying possible antitumor drugs rather than assessing the nature of the dose-response in the low-dose zone.
The pattern analysis assessment indicated that the total number of concentration-responses below the BMD(10) are skewed strongly in the direction predicted by the hormesis model (Fig. 5). These findings are consistent with the earlier reports of Calabrese and Baldwin (2001b, 2003) indicating a similar relationship for data derived from the toxicological literature.
The quantitative nature of the dose-response in the hormetic zone in the present study is also consistent with findings reported for hormesis with other biological models, endpoints, and chemical agents. That is, the hormesis response is usually modest, with the maximum response typically being only 30–60% greater than the controls (Calabrese and Blain, 2005). For example, the data for the wild-type yeast strain in the 11 BMD(10) < 33µM and 33 BMD(10) < 100µM classifications indicate that the proportion of responses exceeding 120% was 26.3 and 16.5%, respectively. In the case of strain SP47080, the respective proportion of responses >120% were 26.2 and 23.7%, respectively.
The present findings illustrate the importance of study design in the assessment of hormesis. A comparison of the lowest concentration (1.2µM) to the BMD(10) revealed that the chance of a stimulatory response becomes greater as the difference between 1.2µM and the BMD(10) increases. Approximately 60% of the time, both replicate responses at 1.2µM exceeded control values for BMD(10) values between 50 and 100µM, compared to only 16% for BMD(10) values between 1.2 and 3.7µM. These findings are consistent with the observations of Calabrese and Baldwin (2001b) that there is an optimal range of hormetic responses starting at about 1/3–1/4 of the estimated toxic threshold. It is likely that the low response at the concentration below the BMD(10) in the 3.7 BMD(10) < 11µM classification (Table 3) was due to there being a substantial proportion of such responses within the toxicity zone. The likelihood of a hormetic response increases as the distance from the BMD(10) increases, at least up to the limits presented in the present database. Since the semilog concentration spacing covered only a 100-fold concentration range, usually including toxicity at the high concentrations, it was not possible to explore the concentration-response range at which a return to control values would be expected. This would have required several concentrations lower than 1.2µM. Based on the hormesis database (Calabrese and Blain, 2005), about 80% of the hormetic responses are within 100-fold of the dose of the toxic threshold. Regardless of the genetic differences among the 13 yeast strains, the overall response to the 2189 chemicals was similar. These findings suggest that the hormetic response is a general one, unrelated to a specific cell cycle regulatory mechanism or DNA repair pathway. Similar quantitative features of the hormetic dose-response occur in models representing broad phylogenetic diversity and various cancer and noncancer-related endpoints (Calabrese and Baldwin, 2001b, 2003). The findings with yeast are consistent with those seen with NCI cancer drug screening data for 70 human tumor cell lines and up to 55,000 chemicals, involving over 3.3 million dose-responses (Calabrese, Staudenmayer, and Stanek, in preparation) (http://dtp.nci.nih.gov). Our present findings are drawn from the U.S. NCI database for screening of potential antitumor agents, and the consistently observed stimulation of proliferation in the below-threshold zone may have significant implications for the design of new antitumor drugs, drug testing, and the management of patients in clinical settings (Calabrese et al., 2006).
The current findings are particularly important because they demonstrate the inadequacy of the traditional threshold dose-response model in predicting below-threshold responses. They also indicate that the hormetic model is consistent with these subtoxic responses. The findings suggest that the hormetic responses are more fundamental than threshold responses and support recent arguments that the hormesis model should be considered as the default dose-response model for scientific interpretation of toxicological responses (Calabrese, 2004). Several features of the study design and/or methodological evaluation (e.g., peripheral placement of controls on the 96-well plate, use of a nonsymmetrical model to assess above/below 100% responses, use of the BMD(10) instead of the NOEL or BMD's with lower cutoff points [2.5, 5.0, and 7.5]) favored conservative estimates and may have caused an underestimation of the frequency of hormesis. Thus, the inadequacies of the threshold model are probably greater than presented, while the predictive capacity of the hormesis model in the below-threshold zone exceeds that reported. The findings argue for a paradigm shift in our understanding of the dose-response relationship, the central pillar of pharmacology and toxicology.
|
SUPPLEMENTARY DATA |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Supplementary data are available online at http://toxsci.oxfordjournals.org/.
|
NOTES |
|---|
Disclaimer: The U.S. Government is authorized to reproduce and distribute for governmental purposes notwithstanding any copyright notation thereon. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsement, either expressed or implied, of the Air Force Office of Scientific Research or the U.S. Government.
|
ACKNOWLEDGMENTS |
|---|
Effort sponsored by the Air Force Office of Scientific Research, Air Force Material Command, United States Air Force (USAF). The assistance of Dr Julian Simon, Fred C. Hutchinson Cancer Research Center, is gratefully acknowledged.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Calabrese EJ. (2004) Hormesis: From marginalization to mainstream. A case for hormesis as the default dose-response model in risk assessment. Toxicol. Appl. Pharmacol. 197:125–136.[CrossRef][Web of Science][Medline]
Calabrese EJ. (2005) Paradigm lost, paradigm found: The re-emergence of hormesis as a fundamental dose response model in the toxicological sciences. Environ. Pollut. 138:379–412.[CrossRef][Medline]
Calabrese EJ and Baldwin LA. (2001a) Hormesis: U-shaped dose-responses and their centrality in toxicology. Trends Pharmacol. Sci. 22:285–291.[CrossRef][Medline]
In Calabrese EJ and Baldwin LA (Eds.). Introduction: Scientific foundations of hormesis. Crit. Rev. Toxicol. (2001b) 31:351.[CrossRef][Web of Science][Medline]
Calabrese EJ and Baldwin LA. (2001c) The frequency of U-shaped responses in the toxicological literature. Toxicol. Sci. 62:330–338.
Calabrese EJ and Baldwin LA. (2003) The hormetic dose response model is more common than the threshold model in toxicology. Toxicol. Sci. 71:246–250.
Calabrese EJ, Baldwin LA, Holland C. (1999) Hormesis: A highly generalizable and reproducible phenomenon with important implications for risk assessment. Risk Anal. 19:261–281.[Web of Science][Medline]
Calabrese EJ and Blain R. (2005) The occurrence of hormetic dose responses in the toxicological literature, the hormesis database: An overview. Toxicol. Appl. Pharmacol. 202:289–301.[CrossRef][Web of Science][Medline]
Calabrese EJ, Staudenmayer J, Stanek EJ. (2006) Drug development and hormesis. Changing conceptual understanding of the dose response creates new challenges and opportunities for more effective drugs. Curr. Opin. Drug Discov. Devel. 9:117–123.[Web of Science][Medline]
Casella G and Berger RL. (2002) Statistical Inference 2nd ed (Duxbury, Pacific Grove, CA) pp. 190.
Clark AJ. (1926) Applied Pharmacology 1st ed (J & A Churchill, London).
Clark AJ. (1933) Mode of Action of Drugs(Arnold, London).
Clark AJ. (1937) Handbook of Experimental Pharmacology(Verlig Von Julius Springer, Berlin).
Crump KS. (1984) A new method for determining allowable daily intakes. Fundam. Appl. Toxicol. 4:854–871.[CrossRef][Web of Science][Medline]
Eaton DL and Klaassen CD. (2001) Principles of Toxicology. In Klaassen CD (Ed.). Casarett and Doull's Toxicology: The Basic Science of Poisons 6th ed (McGraw-Hill, New York) pp. 11–34.
Faessel HM, Levasseur LM, Slocum HK, Greco WR. (1999) Parabolic growth patterns in 96-well plate cell growth experiments. In Vitro Cell. Dev. Biol. 35:270–278.
In Hardman JG, Limbird LE, Gilman AGEds (Eds.). Goodman & Gilman's The Pharmacological Basis of Therapeutics (2001) 10th ed (McGraw-Hill, New York).
In Hayes AW (Ed.). Principles and Methods of Toxicology (2001) 4th ed (Taylor and Francis, Philadelphia, PA).
Holbeck SL. (2004) Update on NCI in vitro drug screen utilities. Eur. J. Cancer 40:785–793.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  E. J. Calabrese, E. J. Stanek III, M. A. Nascarella, and G. R. Hoffmann Hormesis Predicts Low-Dose Responses Better Than Threshold Models International Journal of Toxicology, September 1, 2008; 27(5): 369 - 378. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Hoffmann and W. Stempsey The hormesis concept and risk assessment: are there unique ethical and policy considerations? Human and Experimental Toxicology, August 1, 2008; 27(8): 613 - 620. [Abstract] [PDF]
|
|
|
|
|
|
K. Elliott A case for deliberation in response to hormesis research Human and Experimental Toxicology, July 1, 2008; 27(7): 529 - 538. [Abstract] [PDF]
|
|
|
|
 |
|
 O.-N. Bae, K.-M. Lim, J.-Y. Han, B.-I. Jung, J.-Y. Lee, J.-Y. Noh, S.-M. Chung, M.-Y. Lee, J.-Y. Lee, and J.-H. Chung U-shaped Dose Response in Vasomotor Tone: A Mixed Result of Heterogenic Response of Multiple Cells to Xenobiotics Toxicol. Sci., May 1, 2008; 103(1): 181 - 190. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. J Calabrese Biomedical implications of hormesis: Part I Human and Experimental Toxicology, February 1, 2008; 27(2): 121 - 121. [PDF]
|
|
|
|
|
|
K. S. Crump Limitations in the National Cancer Institute Antitumor Drug Screening Database for Evaluating Hormesis Toxicol. Sci., August 1, 2007; 98(2): 599 - 601. [Full Text] [PDF]
|
|
|
|
|
|
E. J. Calabrese, J. W. Staudenmayer, E. J. Stanek III, and G. R. Hoffmann Hormesis and High Throughput Studies: Crump's Analysis Lacks Credibility Toxicol. Sci., August 1, 2007; 98(2): 602 - 603. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||