BMDL 10 values for tumor types being tested in studies with two or more dose levels were calculated for compounds. A BMDL 10 value based on studies with three or more dose levels was available for 62 substances. Taking into account that benchmark dose modeling is dependent on a dose—response curve, the dataset based on studies with three or more dose groups plus controls is generally considered more reliable than studies testing two dose groups.
Figure 3. Density plot with histograms for different types of PODs with indication of the respective fifth percentiles by dotted lines in respective colors. Therefore, we compiled a dataset of chemicals for which all three types of POD were available Figure 2 , Table 3.
The fifth percentile in the EDT 10 dataset is slightly but not significantly higher. The NOEL values are the largest dataset comprising all the non-genotoxic compounds. Most of the non-genotoxic carcinogens belong to Cramer class 3. Out of the substances, belong to Cramer class 3, 5 to Cramer class 2, and 18 to Cramer class 1. The exclusion of the potentially bioaccumulating substances and the steroids mostly affects Cramer class 3, which is reduced to 90 compounds.
The distribution of the values is similar to the original Munro data for Cramer classes 3 and 1 as shown in Figure 4. Table 4. Figure 4.
Cumulative frequencies of the original Munro data and non-genotoxic dataset for Cramer classes 1 and 3. Table 5. Like all other substances remaining after exclusion of genotoxicants and those generally excluded from TTC, non-genotoxic compounds are currently assigned to the Cramer classes.
This approach is based on the observation that tumor formation is one of several chronic adverse effects and that the underlying mechanisms show a threshold Clewell et al.
This approach points to the question of whether non-genotoxic carcinogens, in the context of TTC, are best described by their most sensitive tumor or by their non-neoplastic adverse effects and which dose descriptor is the most conservative. The dataset used in this study was subject to a consistency check, which included the exclusion of non-carcinogenic compounds Cefic, and non-human relevant tumor types as well as an allocation of the mode of action non-genotoxic vs.
Furthermore, linear extrapolation to a virtual safe dose with a tumor risk of 1—1,, implicates high uncertainty. This is in line with findings from a previous study, which showed that BMD confidence intervals for tumor responses correlate with NOEL values from sub chronic toxicity studies Braakhuis et al. Since several authors Gaylor, ; Goodman, have argued that the classical two-year bioassay by NTP will result in tumors at the HTD, it is interesting to note that within this dataset tumors do not primarily start to occur at the highest dose tested.
The observation of predominant non-neoplastic lesions supports the hypothesis for the mode of actions of non-genotoxic compounds made by Braakhuis et al. They stated that exposure to low doses of non-genotoxic carcinogens will change some biological processes slightly, whereas the repeated exposure will lead to overt disturbances, e. NOEL values of subchronic studies were additionally compared with those of chronic studies to compensate for possible deficiencies in the long-term study.
Within this comparison, the differences between the NOELs were not higher than the well-established extrapolation factor of 2 for subchronic to chronic exposure EFSA, The data support the hypothesis that there is no significant added value of chronic studies neither with respect to POD Braakhuis et al.
Nevertheless, we used the chronic dataset for threshold derivation simply, as it was larger than that for subchronic effects and covers a greater number of compounds. It has been hypothesized that there may be a minimal e. This discrepancy of the effect is most likely due to the limited statistical power of the animal studies Hardy et al. Both values will feed into the final analysis to derive TTC values. The assignment of Cramer classes to the dataset is a prerequisite for the derivation of the respective TTC values for Cramer classes 3 and 1.
The shortcomings and ongoing improvements of Cramer classifications using the OECD Toolbox or Toxtree have already been well-documented in previous publications Bhatia et al. As we aimed to compare the TTC values to the original values derived by Munro, we decided to use the original Cramer decision tree from the OECD Toolbox for the classification of the data set. Most of the compounds in the dataset were assigned to Cramer class 3.
As commonly occurs, very few or no compounds were assigned to Cramer class 3. As commonly occurs, very few or no compounds were assigned to Cramer class 2 and for this dataset only 18 compounds into Cramer class 1 Patel et al. The determination of a TTC value for Cramer class 1 compounds was, thus, not reasonable, as the fifth percentile would be based on a single substance only. The NOEL values of the 18 non-genotoxic carcinogens are consistently distributed within the Cramer class 1 dataset Figure 4.
This preliminary analysis is an indication that the Cramer class 1 threshold can be applied to non-genotoxic compounds. EFSA calculated a value of 1. This comparison supports the safe application of the Cramer class 3 TTC value to compounds being non-genotoxic carcinogens. A further extension of Cramer class 3 using data from non-genotoxic carcinogenic herbicides, insecticides, fungicides, and other agrochemicals is available with Heusinkveld et al.
Following the same approach as before, these 95 compounds were assigned to Cramer classes all were class 3 , potentially bioaccumulating compounds were excluded, and the respective fifth NOEL percentile was derived.
Steroids were not contained in this dataset. These values are similar to the percentiles derived for the non-genotoxic dataset of this project and, thus, further, substantiate the applicability of Cramer class 3 and the associated TTC value for non-genotoxic carcinogens.
The exclusion of substances exhibiting non-human relevant mechanisms, as we did in the data set, is generally accepted Boobis et al. The exclusion of several structural groups from the application of the TTC is generally agreed Boobis et al. This means that compounds with similar structural properties are not contained in the TTC datasets, e.
Other compounds, such as dibenzodioxin or-diphenyl-derivatives, are known for their excess toxicity and potential to bioaccumulate. For these compound classes, category-specific TTC values, such as that for organophosphates, do not exist at this time, as the value is likely to be too low to be practically applicable.
Nevertheless, some substances from these excluded classes are part of the original Munro dataset, and their NOELs contribute to the respective TTC values. This is due to the fact that only few substances are concerned, and these are scattered over the full range of NOELs.
In the analysis, the exclusion of the potentially bioaccumulating substances from the non-genotoxic data has a significant influence on the TTC value derived from the NOEL values. Furthermore, the high toxicity of these substances has already been widely accepted, as this is the reason for excluding this group from TTC application. It is, however, discussed that bioaccumulating substances are within the scope of the TTC concept Leeman et al. The definition of bioaccumulating potential, as well as the dataset, was different in this publication, especially the latter being the most probable reason for the different outcomes.
Through the application of the method of random leave out, we nicely show in an objective way that the excluded substances are distinct, and the very same reasoning holds true for the steroids. The method of randomly leaving out a certain number of chemicals not only supports the exclusion of substance groups but also indicates the robustness of the fifth percentile.
The similar ranges of about a factor of 2—3 for all ranges of the TTC values increase the confidence of the authors in the TTC values, as this order of magnitude is approximately similar to one dose spacing difference. Existing in silico or in vitro methods to detect non-genotoxic carcinogen mode of actions are insufficiently accurate yet Benigni et al. Consequently, when applying the current TTC concept, structures cannot reliably be identified as non-genotoxic carcinogens.
After searching the identifiers in the Escher dataset using the Dashboard, data was retrieved for all but two substances: diphenylmethane diisocyanate and dipropylene glycol monomethyl ether. The CAS numbers of these were searched within ChemSpider and the resulting names were compared with that in the Escher dataset; where there was a match, the average mass was extracted. This was carried out using two of the original modules, namely the Cramer rules original Patlewicz et al.
Additionally, these substances were also profiled using 3 custom modules developed ad hoc by Nelms et al. These custom profilers in Patlewicz et al. The custom modules allowed for the Kroes workflow to be replicated. The custom profilers developed in Nelms et al.
The Cramer structural class assignments as provided in Carthew et al. Study data were initially identified within ToxValDB that were either subacute, subchronic, chronic, reproductive, developmental, or multigenerational study type. This was carried out by creating a new field i. ToxValDB records were then filtered to remove ambiguous records, i.
New columns were created to capture toxicity values adjusted on the basis of duration of exposure—this allowed for subchronic and subacute values to be used in the analysis. Rodent and rabbit species names were standardized so that these could be readily selected for filtering.
In this case, records where rat, mice, rabbits, and partial names were identified in the set of studies and mapped to a generic common species tag of rodent. ToxValDB data was then merged with the 3 Cramer structural classes and were processed further as follows: 1 for substances with only 1 study, this was retained; 2 for substances with more than 1 study, extreme outliers, i.
Figure 1 describes the workflow to generate the Cramer class datasets. In the original TTC derivations for oral exposures by Munro et al. If the generative distribution for each Cramer structural class were normally distributed, then the ECDF and CDF plots should closely match each other.
CDFs were plotted by computing the mean and standard deviation of the Cramer structural class data and using those as arguments to compute the CDFs with a sample size of 1, The assumptions of normality were further tested using the Shapiro-Wilk test and by plotting a quantile-quantile plot qqplot. The goodness of fit from the qqplot provided another means to visually inspect whether the Cramer structural class data for each class were normally distributed.
Inspection of the ECDFs for the 3 Cramer structural classes allowed the level of separation between classes to be assessed visually. Substances that were assigned to one of the Cramer classes, were also profiled on the basis of their MOA for aquatic toxicity using the Verhaar module Verhaar et al.
An overall outcome was taken of the 3 profiling outcomes as follows: if the profiling outcomes all agreed, that formed the final outcome else the majority outcome was taken. If the schemes disagreed with each other, the most conservative outcome was taken. A Kolmogorov-Smirnov K-S test Conover, was also performed to verify the difference in distributions.
For each dataset, the 5th percentile and TTC values for each Cramer class were calculated individually for local and systemic effects. Comparisons of the structural features of the Carthew and Escher datasets relative to the ToxValDB dataset was performed using the ToxPrint chemical fingerprints Yang et al.
Data processing was conducted using the Anaconda distribution of Python 3. Table 1 shows the assignments of the substances into their specific TTC assignment after processing through the Kroes decision tree, the Cramer rules and the custom profilers from Nelms et al.
Similar to other TTC analyses e. Munro et al. The set of substances assigned by Cramer structural classes formed the basis of the remainder of the analysis.
The starting dataset comprised , records for 4, different substances. Subsetting and filtering ToxValDB to identify study records that met certain criteria in terms of study type, units, and tagged by assignment into one of the 3 Cramer class resulted in 1, studies for Cramer I, 77 studies for Cramer II, and 1, studies for Cramer III substances.
The breakdown of chemicals and studies are reflected in Table 2. Table 2. Figure 2. Overlaying CDFs distributions from a normal distribution on the ECDFs suggests that the Cramer structural classes are poorly aligned and not following a normal distribution Supplementary Figure 1. The Wilks-Shapiro test statistic for Cramer structural class I was 0. Given the lack of correspondence between the CDFs and ECDFs and the results of the Wilks-Shapiro test, it was concluded that the Cramer structural class distributions were not normally distributed and that computing the 5th percentile from the CDFs was not appropriate.
The results of Class II contained too few substances to be considered further. It is worth noting that the 5th percentiles and their associated TTC values for Cramer I and III are not particularly different, highlighting the lack of separation in the distributions between the 2 classes see Table 4. In Escher et al. For the Carthew et al. Table 4. This can be partially explained by the chemical makeup of the respective datasets as evidenced by the 2D structural landscape.
The Escher dataset is more broadly represented within the ToxValDB landscape whereas the smaller Carthew dataset is sparsely distributed across the other 2 datasets see Figure 3. Figure 3. A comparison of the chemical overlap between the three datasets was performed on the basis of their chemical identifiers using DTXSID.
A Venn diagram showing the overlaps is shown in Supplementary Figure 3. A comparison of the overlaps between the 3 datasets were undertaken to compare the actual toxicity values reported to evaluate whether the differences in TTC values were more likely to be as result of differences in the toxicity values. There were substances in the ToxValDB dataset that overlapped with the Escher dataset but their reported values appeared in many cases quite different. This is reflected by the low pearson correlation coefficient of 0.
However, the pearson correlation coefficient was low at 0. Although chemical landscape between the datasets is one contributing factor to explaining why the TTC values derived could be different, the differences in the toxicity values appears to play a larger role, indeed for substances that were common between the datasets, there was a low agreement between the toxicity values reported. The lack of separation in the Cramer structural classes and the fact that the data did not fit the expected theoretical distribution prompted further evaluation of the dataset and consideration of another means of subcategorizing the chemicals.
For the Escher et al. Table 5. Reported and rederived TTC values of the Escher et al. The upper curve shows how the incidence of death from cardiovascular disease changes with increased levels of cigarette smoking.
The lower curve shows the mortality rate from lung cancer. How would you asses these dose-response data in humans? Think about how you would interpret this information before you look at the answer.
All Rights Reserved. Date last modified: August 29, Wayne W. PH Module 2 - Exposure Assessment. Toxicology How can we estimate expected health effects from exposure data with limited human-health information? Animal Studies The Draize test is an acute toxicity test that was introduced in by FDA toxicologists to test the effects of cosmetics. Another option is to euthanize animals at fixed intervals of time to examine tissues and body fluids for abnormalities Still other procedures look for toxic effects of agents using cells grown in culture in vitro tests rather than in vivo.
The purpose of these tests is to: Gain information about toxicity that can potentially be extrapolated to humans. Estimate the doses at which effects occur through dose-response curves Identify subtle but potentially important effects by examining tissues and fluids Explore the biological mechanisms by which adverse effects occur Establish regulatory limits for various exposures including emissions Threshold Dose Dose-response curves are frequently sigmoidal, or S-shaped, and for many substances small doses do not appear to be toxic.
The experimental conditions can be carefully defined and controlled diet, temperature, ventilation, age of the animals, etc. By using a specific strain of mice or rats genetic variability is minimized. Large, curated databases on toxic effects of chemicals provide us with the opportunity to set TTC for many hazards and substance classes and thus offer a precautionary second tier for risk assessments if hazard cannot be excluded.
This allows focusing testing efforts better on relevant exposures to chemicals. Keywords: alternative methods; computational toxicology; exposure; risk assessment; toxicity limits.
0コメント