Scholarly article on topic 'Impact of study design and database parameters on NOAEL distributions used for toxicological concern (TTC) values'

Impact of study design and database parameters on NOAEL distributions used for toxicological concern (TTC) values Academic research paper on "Biological sciences"

CC BY-NC-ND
0
0
Share paper
OECD Field of science
Keywords
{TTC / NOAEL / LOAEL / "NOAEL distribution"}

Abstract of research paper on Biological sciences, author of scientific article — Jürg A. Zarn, Emanuel Hänggi, Barbara E. Engeli

Abstract TTC values for chemicals with unknown toxicity but known structure are derived from 5th percentiles of NOAEL distributions from compounds with known toxicity. The impact of chemical structures on TTC values was repeatedly investigated but not the impact of parameters such as study numbers per compound and differences in study design. Recently, study design parameters such as application route with related dose-decrements, dose-spacing and number of animals per group but not exposure duration were found to affect NOAEL distributions. Here, the impact of study design parameters on lowest NOAEL distributions and consequently on TTC values was analyzed in a database on 423 Cramer class III pesticides. Using NOAELs related to lowest LOAELs instead of lowest NOAELs, excluding studies with a dose spacing >8, and standardizing NOAELs to the initial dose animals received shifted the 5th percentile of NOAEL distributions from 0.22 to 0.5mg/kg body weight per day. In contrast, weighting of NOAELs for the study numbers per compound shifts 5th percentiles downwards to lower values by 10–20%. The results show that database and study design parameters influence NOAEL distributions to a minor degree and derived TTC values therefore can be considered reliable in that perspective.

Academic research paper on topic "Impact of study design and database parameters on NOAEL distributions used for toxicological concern (TTC) values"

Contents lists available at ScienceDirect

Regulatory Toxicology and Pharmacology

journal homepage: www.elsevier.com/locate/yrtph

Impact of study design and database parameters on NOAEL distributions ■. used for toxicological concern (TTC) values

Jurg A. Zarn *, Emanuel Hanggi, Barbara E. Engeli

Federal Food Safety and Veterinary Office FSVO, Risk Assessment Division, Stauffacherstrasse 101, 8004 Zurich, Switzerland

CrossMark

ARTICLE INFO

Article history: Received 4 March 2015 Available online 19 May 2015

Keywords: TTC NOAEL LOAEL

NOAEL distribution

ABSTRACT

TTC values for chemicals with unknown toxicity but known structure are derived from 5th percentiles of NOAEL distributions from compounds with known toxicity. The impact of chemical structures on TTC values was repeatedly investigated but not the impact of parameters such as study numbers per compound and differences in study design. Recently, study design parameters such as application route with related dose-decrements, dose-spacing and number of animals per group but not exposure duration were found to affect NOAEL distributions. Here, the impact of study design parameters on lowest NOAEL distributions and consequently on TTC values was analyzed in a database on 423 Cramer class III pesticides. Using NOAELs related to lowest LOAELs instead of lowest NOAELs, excluding studies with a dose spacing >8, and standardizing NOAELs to the initial dose animals received shifted the 5th percentile of NOAEL distributions from 0.22 to 0.5 mg/kg body weight per day. In contrast, weighting of NOAELs for the study numbers per compound shifts 5th percentiles downwards to lower values by 10-20%. The results show that database and study design parameters influence NOAEL distributions to a minor degree and derived TTC values therefore can be considered reliable in that perspective.

© 2015 Elsevier Inc. All rights reserved.

1. Introduction

In the regulatory dietary risk assessment of chemicals (FAO/WHO, 2009), authorities compare the dietary exposure of a chemical in question with the chemical-specific health based guidance value (HBGV) such as Acceptable Daily Intake (ADI), Tolerable Daily Intake (TDI) and others to identify possible health risks. HBGV typically are derived from No Observed Adverse Effect Levels (NOAEL) identified in laboratory animal studies by using safety factors that account for several aspects of the underlying database. There are situations however, where few or no toxicolog-ical studies are available but the chemical structure of a compound is known. In such a situation the Toxicological Threshold of Concern concept (TTC), initially developed by Munro (Munro et al., 1996) based on Cramer's grouping of chemicals (Cramer et al., 1978) offers a scientifically justified, pragmatic assessment tool (EFSA, 2012) which is applied by authorities worldwide (Bruschweiler, 2014). Over the last two decades, using the

Abbreviations: LOAEL, Lowest Observed Adverse Effect Level; NOAEL, No Observed Adverse Effect Level; PPP, Plant Protection Product, Active Ingredient.

* Corresponding author. Fax: +41 58 467 21 99. E-mail addresses: juerg.zarn@blv.admin.ch (J.A. Zarn), emanuel.haenggi@blv. admin.ch (E. Hänggi), barbara.engeli@blv.admin.ch (B.E. Engeli).

http://dx.doi.org/10.1016/j.yrtph.2015.05.015 0273-2300/® 2015 Elsevier Inc. All rights reserved.

toxicological data provided over several decades, the TTC concept was constantly refined. Improvements came from databases on new substances with further toxicological data and reconsiderations of the Cramer classifications (Bhatia et al., 2015; Feigenbaum et al., 2015; Galloway et al., 2013; Kalkhof et al., 2012; Kroes et al., 2004; Laufersweiler et al., 2012; Leeman et al., 2014; Pinalli et al., 2011; Tluczkiewicz et al., 2011; van Ravenzwaay et al., 2011, 2012). For several chemical regulatory fields, adapted TTC approaches were proposed (Kroes et al., 2007; Melching-Kollmuss et al., 2010; Muller et al., 2006). The TTC concept assumes that NOAELs of groups of chemical structures scatter according to defined distributions. Therefore, chemical group specific lowest NOAEL distributions allow to calculate exposure thresholds that should not be exceeded by a compound of unknown toxicity but known structure to warrant safety with a certain probability, e.g. 95%, (EFSA, 2012; Munro et al., 1996).

If regulatory measures for a structurally defined population of chemicals are based on a lowest NOAEL distribution derived from a sample out of this population, it has to be representative for the population in question. In the development of the TTC concept, this point was the subject of intensive investigations on the relation between chemical structure and NOAEL. However, significant overlap of lowest NOAEL distributions for different chemical

groups, such as Cramer class II and III (Munro et al., 1996), indicates that current grouping still does not discriminates satisfactorily. Several publications (Bhatia et al., 2015; Dekant et al., 2010; Kroes et al., 2004, 2007; Muller et al., 2006; Tluczkiewicz et al., 2011) address toxicological endpoint-driven and/or chemical structure driven attempts to divide the world of chemicals into groups represented by satisfactorily discernible lowest NOAEL distributions. Further efforts in this direction might improve the concept. However, considerations on the design and quality of the toxicological studies and the extent of the database per compound as well as of the sample group representing the population of chemicals in question should likewise supplement the aspect of structure-activity relationship in the TTC concept.

In regulatory risk assessment concepts it is assumed that toxicological threshold doses in tested species exist for non-genotoxic compounds and that they can be approximated by animal studies. A series of laboratory animal studies is conducted allowing for potentially changing sensitivities during lifetime including the pre-natal phase because for a given compound the exact toxicological mechanism of action and the potency often is unknown. If several species are tested for a comprehensive range of toxicological endpoints in a scientifically sound way, the lowest NOAEL achieved is considered an appropriate approximation for the overall lowest toxicological threshold in the tested species. Hence, it seems self-evident that for a given compound the proximity of the NOAEL to the lowest toxicological threshold depends on the design of the individual toxicological studies and the extent/magnitude of the database available for the compound of interest. For example, the number of species investigated, the number of studies performed, the animals used per dose group and the tested dose intervals (dose spacing) are crucial to achieve closest approximation of the lowest toxicological threshold by the NOAEL. If regulatory decisions are derived from distributions of lowest NOAELs, e.g. the use of the TTC concept, and because the quality of the database for each individual NOAEL entering the lowest NOAEL distribution has an impact on the distribution, study design and database-related aspects should be considered. If a compound was only investigated in one study with few animals per dose group and few endpoints investigated, the respective NOAEL has more uncertainty regarding its proximity to the toxicological threshold compared to a compound investigated in several species in a range of studies and therefore with high statistical power. In principle, a study design with large dose spacing favors low NOAELs whereas few animals/endpoints favor high NOAELs relative to the toxicological threshold.

Recent findings in retrospective analyses of pesticide evaluations have shown that study design parameters such as number of animals, dose spacing and the dose decrement during growth of experimental animals observed in many studies significantly influence NOAEL distributions (Zarn et al., 2010, 2011, 2013). Study duration however proved not to have a significant effect on NOAEL.

Based on these experiences, the present work focuses on the influence of such parameters on lowest NOAEL distributions of Cramer class III pesticides and quantifies possible consequences for TTC values derived thereof. It is not the goal of this publication to propose additional or modified TTC values.

2. Methods and database

Publicly available data from PPP evaluations of toxicity studies from the European Food Safety Authority (EFSA) (EFSA, 2014), the Joint FAO/WHO Meeting on Pesticide Residues (JMPR) (WHO, 2011) and the US EPA (US EPA, 2011) were entered into a Microsoft® Office Access 2007 database (Zarn et al., 2011). For a

given study, these data included the chemical class of the compound, the number of animals per group and sex, the study duration, the application route, all dose levels and the NOAEL and the LOAEL as reported by the evaluating authority. Rat, mouse, rabbit and dog studies were included if the application route was oral either by food, drinking water, gavage or capsule administration. Moreover, information was also included on whether the application aimed at constant doses or on constant compound concentrations in food and drinking water and whether constant doses or constant amounts in gavage and capsule applications were applied. Only studies providing both an experimental NOAEL and an experimental LOAEL were included. If several studies were found to result in the same lowest NOAEL, the study with the lowest dose spacing was chosen. If several studies with equal dose spacing were found to result in the same lowest NOAEL the study with the shortest study duration was chosen. All the compounds in the database were grouped to Cramer classes by Toxtree 2.6.6 (downloaded from http://toxtree.sourceforge.net/index.html) and only Cramer class III compounds were further used (three compounds were excluded because they were Cramer class II or I). This reference database PSMTOXDB included 3924 informative studies from 423 compounds. Further details on the database and analyses are described elsewhere (Zarn et al., 2011, 2013). 80 (19%, identified by CAS number) of the compounds in PSMTOXDB are also included in the Tluczkiewicz database (Tluczkiewicz et al., 2011), 61 (14%) in the Munro database (Munro et al., 1996) and an additional 44 (10%) in both. For data processing and data analyses, Microsoft® Office Access 2007, Microsoft® Office Excel 2007, IBM SPSS Statistics 22 for Windows and @risk Professional 6.3.1 were used.

To investigate the influence of several database- (e.g. number of studies per compound) and study-related features (e.g. dose spacing) on NOAEL distributions, several data subsets were generated and compared with each other. The data subsets are described below and a short description of the subsets and their denomination is given in Table 1.

Table 1

Reference database (PSMTOXDB) and data-subset extracts of lowest NOAELs used.

Subset Weighted (w) Selection criteria

NO Organophosphates and carbamates (O)

N Nw Compounds including O

Ns Nsw Standardized NOAEL

N8 N8w Dose spacing 68

N8s N8sw Dose spacing 68, standardized NOAEL

N\O N\Ow Compounds without O

N\Os N\Osw Standardized NOAEL

N8\O N8\Ow Dose spacing 68

N8\Os N8\Osw Dose spacing 68, standardized NOAEL

N-L\O N-L\Ow NOAEL related to lowest LOAEL, without O

N-L\Os N-L\Osw Standardized NOAEL

N-L8\O N-L8\Ow NOAEL related to lowest LOAEL, without O, dose spacing 68

N-L8\Os N-L8\Osw Standardized NOAEL

N\Oallspec Compounds without O, studies in all 4 species

N4\O Dose spacing 64

N8\O_lowweight Dose spacing 68, low-weight half of distribution

N8 \O_highweight Dose spacing 6 8, high-weight half of distribution

NM Compounds including O, Munro's criteria*

NM\O Compounds without O, Munro's criteria*

NMun(cl3) Munro Cramer class 3 compounds

NMun(cl3)\O Munro Cramer class 3 compounds without O

* Criteria described in Munro et al. (1996).

2.1. Exclusion of organophosphates/carbamates

As a basis for further data subsets, a subset was created by excluding organophosphates and carbamates. This was done to analyze the effect of the highly toxic acetylcholine esterase inhibitors on lowest NOAEL distributions and to make our analyses comparable with other published analyses.

Lowest NOAEL distributions where organophosphates and car-bamates were excluded are designated with "\O".

2.2. Restriction of dose spacing

Recently, in analyses on the effect of study duration on NOAELs, the dose spacing (ratio LOAEL/NOAEL) turned out to affect NOAEL distributions (Zarn et al., 2010, 2011). It was shown before that the dose spacing in chronic studies is disproportionately high compared to all other study types. To investigate this influence within the context of the current work, a subset excluding all studies with a dose spacing >8 was created, resulting in 3145 remaining studies for 418 compounds. Chronic studies are especially prone to high dose spacing ratios. A dose spacing cut-off value of 8 was chosen, because it excludes a significant fraction of the studies with highest dose spacing but still maintains a high amount of chronic studies in the database. To find such an optimal cut-off value, the following procedure was applied: the fraction of chronic studies remaining in the database were recorded for each cut-off value between 4 (median dose spacing for the whole PSMTOXDB) and 12 (only few studies have a dose-spacing >12), The fraction of chronic studies in the three subsets N\O (no restriction with regard to dose spacing), N8\O (only studies with dose spacing 68) and N4\O (only studies with dose spacing 64) are 62%, 55% and 36% (Table 2), respectively. The cut-off value of 8 was as the best compromise between exclusion of studies with high dose spacing and maintenance of a reasonable fraction of chronic studies. Although 20% of studies with high dose spacing were hereby excluded the fraction of chronic studies (7%) lost is considered acceptable.

Lowest NOAEL distributions with dose spacing restrictions are designated with an "8" in the nomenclature.

Table 2

Percentages of study types and species providing the lowest NOAEL in the respective data subset. Due to rounding, sums of percentages not always add up to 100% (e.g. studies (only 99%) giving the lowest NOAELs in the N distribution).

Data subset Species Study

Rat Dog Mouse Rabbit c sc sa mg dv

NO 63 31 2 5 39 17 8 17 19

N 52 37 5 6 59 16 4 11 9

Ns 39 45 6 10 44 19 5 15 17

N8 48 36 6 10 52 18 5 11 14

N8s 38 41 6 15 40 18 7 13 22

N\O 52 36 6 6 62 16 3 11 9

N\Os 38 44 7 11 46 18 5 14 16

N8\O 46 38 7 9 55 18 4 10 12

N8\Os 35 43 6 15 43 18 6 12 22

N-L\O 46 37 5 11 57 16 3 10 14

N-L\Os 35 42 7 16 45 20 4 9 21

N-L8\O 44 38 5 12 49 19 5 11 16

N-L8\Os 35 42 7 17 41 21 6 9 23

N\Oallspec 53 35 5 6 64 18 2 9 7

N4\O 45 29 8 17 36 20 8 11 25

N8\O_lowweight 50 31 8 11 50 18 6 11 15

N8 \O_highweight 43 45 5 8 60 18 3 9 10

NM 87 - 13 - 35 54 1 10 -

NM\O 86 - 14 - 36 53 - 11 -

c, Chronic; sc, subchronic; sa, subacute; mg, multi-generation; dv, developmental.

2.3. NOAELs related to lowest LOAELs

As an alternative to investigate and to minimize the influence of dose spacing on the NOAEL distributions, not the lowest NOAEL itself for a given compound was identified but the NOAEL related to the lowest LOAEL. This was done because the lowest NOAEL for a given compound derives not necessarily from the study with the lowest LOAEL but the latter was the closest approximation of the toxicological threshold for a substance of interest. If this approach is combined with restrictions of the database for dose spacing, NOAEL distributions might result that more accurately approximate toxicological thresholds than simply selecting for lowest NOAELs.

NOAEL distributions based on lowest LOAELs are indicated by "N-L" in the nomenclature.

2.4. Standardization method for NOAEL and LOAEL from different

study types

For each study, if applicable, besides the lowest NOAEL and the lowest LOAEL as reported, also a standardized NOAEL and a standardized LOAEL was calculated and added to the databases. The standardization was done for NOAELs of studies with designs applying the compound either by providing feed with constant compound concentrations or by capsule or gavage without regular dose adjustment accounting for decreasing doses as body weights increase. To make NOAELs and LOAELs from different study designs and durations better comparable, the NOAELs and LOAELs are standardized to the initial dose animals received by multiplying NOAELs and LOAELs with the species and study specific dose decrement factors (Table 3). For the present work, dose decrement factors derived from previous projects were used (Zarn et al., 2010, 2011, 2013).

Lowest NOAEL distributions based on selection of lowest standardized NOAEL are designated by an "s" in their name.

2.5. Weighting procedure for NOAELs of dataset quality

The toxicity studies required by authorities and submitted by applicants are very different regarding design and scope between rat, dog, mouse and rabbit. In rats, fetal developmental, multi-generation, subchronic and chronic studies are required and for dogs and mice only subchronic and chronic studies (up to 2 years). In rabbits, only fetal developmental studies are required. Additionally, voluntarily the companies also provide occasionally subacute studies in rat, dog and mouse. In the PSMTOXDB, the overall median number of studies of any design per compound is 9 (interquartile range (IQR) 5). The median number of studies per compound and species is 5 for rats, 2 for dogs, 2 for mice and 1 for rabbits.

To account for the number of informative studies per compound and the number of species tested, a dataset quality/sensitivity weighting system was developed. To weight approximately the

Table 3

Dose standardization factors for studies with a design not providing constant doses. This are studies providing feed with constant compound concentrations or by capsule or gavage without regular dose adjustment accounting for decreasing doses as body weights increase. Multiplying the reported NOAEL in a study using non-constant doses by the factors below yields the dose animals were exposed to at the beginning of the respective study. For example, multiplying a chronic rat feeding study NOAEL by 2.9 gives the corresponding initial dose.

Subacute Subchronic Chronic

Mouse 1.2 1.3 1.6

Rat 1.6 2.1 2.9

Dog 1.1 1.2 1.4

toxicological sensitivity (severely derogated by the different data requirements for the four species) of the four species considered in this paper (mouse, rat, rabbit, dog), in the dataset N\Oallspecies (only compounds with at least one informative study in all four species) the frequencies of the species giving the lowest NOAEL per compound were identified. It was found that the mouse had the lowest frequency for giving the lowest NOAEL, followed by the rabbit. Rats and dogs nearly had the same frequency of being the most sensitive species. All species frequencies were standardized to the species with the lowest frequency (mouse; sensitivity set as 1). This resulted in relative species sensitivities of 10 (rat), 7 (dog) and 1 (mouse and rabbit). For example, a rat study tenfold more often gave the study with the lowest NOAEL compared to the mouse. For each compound, these relative species sensitivity values were then multiplied by the number of the informative studies per species (1 or 2 or 3 (if p3 studies available)) and then summated over the 4 species. The maximum possible weight value for a compound therefore was 57 (3*10 + 3*7 + 3*1 + 3*1). Since the number of studies and endpoints covered per species are very different, the species sensitivity factor as developed above certainly is affected by the number of studies and the study types performed. Therefore it is important to note that this approach of weighting probably overestimates the sensitivity of rats (large database per compound) in relation to dogs, mice and especially rabbits (rather small database per compound). However, it should be noted that dogs nearly as often as the rats give the lowest NOAEL for given compounds although fewer studies with lower statistical power and less endpoints covered are done in this species.

The resulting distribution of the "sensitivity" factors over all compounds was grouped into three weight categories. Weight 1 was allocated to compounds with "sensitivity" factors <31, weight 3 to compounds with "sensitivity" factors p31 and 648 and weight 12 for compounds with "sensitivity" factors >48. The slight increase in multiples (1, 3, and 12) of weight value factors should favor in a somewhat balanced way lowest NOAELs of compounds with an extensive database. The boundaries for the three categories were allocated based on the distribution of "sensitivity" factors with the aim to group approximately 10% of the compounds in the lowest, 70% in the medium and 20% in highest weight class.

In practice, the lowest NOAEL of a compound with a "sensitivity" factor >48 (therefore weight 12) was used 12-fold and the lowest NOAEL of a compound with a weighting factor <31 (therefore weight 1) was used only once in generating a lowest NOAEL distribution.

It should be noted, although favoring well documented lowest NOAELs in principle seems scientifically justified, there is considerable arbitrariness in the factors used.

Lowest NOAEL distributions based on weighted compounds are designated by the letter "w" in the name.

2.6. Data analyses

To analyze the influence of dose spacing, dose decrements in certain study types and data base differences between compounds on the NOAEL and LOAEL distributions, 5th percentiles of the cumulative empirical distributions and of appropriate distribution models were calculated. For each empirical NOAEL and LOAEL distribution the ratio of the 97.5th and the 2.5th percentiles was calculated. This ratio is an estimate of the range (tightness of distribution) of NOAELs (and LOAELs) for 95% of compounds in a given data subset.

All NOAEL distributions were tested for the most appropriate distribution model using @risk Professional 6.3.1. The best fit selection was done by using Akaike Information Criterion (AIC). In Table 4 besides the empirical 5th percentiles of the lowest

NOAEL distributions also the 5th percentiles of the best fitted model is given. On the 5th percentile level of the best-fitting models there was good agreement between the distribution models and the empirical NOAEL distribution. Unless otherwise stated, throughout the whole document 5th percentiles of distributions are referring to the empirical distribution.

3. Results and discussion

3.1. PSMTOXDB compared to other databases

Several databases containing partly different substances used lowest NOAEL distributions for the derivation of TTC values (Table 5). Among the best known are the Munro database (Munro et al., 1996) containing among other compounds 437 Cramer class III industrial chemicals, pharmaceuticals, food substances and environmental, agricultural and consumer chemicals. The Tluczkiewicz database (Tluczkiewicz et al., 2011) contains 372 Cramer class III compounds compiled from 4 different existing databases including parts of the Munro database. The Pinalli database (Pinalli et al., 2011) primarily used 113 food contact materials and the authors derived surrogate NOAELs by multiplying available TDIs with a presumed default safety factor. The Kalkhof database (Kalkhof et al., 2012) used subacute and subchronic mouse and rat studies on 724 industrial chemicals. The Feigenbaum database (Feigenbaum et al., 2015) includes the NOAELs that were the basis for the ADIs for 322 pesticides evaluated by the EFSA. The Munro, Tluczkiewicz and the Kalkhof databases excluded dog studies and used default time extrapolation factors to yield a presumed chronic NOAEL from subacute and subchronic NOAELs. The Tluczkiewicz database additionally excluded developmental and reproductive studies. The Pinalli and the Feigenbaum databases used TDIs and ADIs, respectively, and the corresponding NOAELs for their analyses. Information on the species, the study types and the database were not available.

In summary, not only the databases used for lowest NOAEL distribution analyses (partly with certain overlaps of compounds included) by different authors but also the approaches applied in deriving lowest NOAELs were different. Therefore, the focus of the present work was on effects of several study design-related parameters on NOAEL and LOAEL distributions and the derived percentiles of the distributions. The parameters accounted for were the dose spacing in studies, time extrapolation factors used for standardization of NOAELs of different study designs, putative sensitivity differences of different species (related to different data requirements) and the extent of the underlying database. For this purpose, data subsets were extracted from PSMTOXDB considered appropriate to evaluate effects of such study design and database related effects on NOAEL and LOAEL distributions. The results are presented and discussed in the following chapters.

3.2. PSMTOXDB calibrated to Munro database

Organophosphates and carbamates belong to the most toxic pesticides and their intended effect, inhibition of the acetylcholinesterase in insects, additionally is also the critical effect (lowest NOAEL) in the mammalian toxicity profile in the majority of these compounds. In many analyses on NOAEL distributions these easy to identify chemical structures were excluded for certain analyses. To make our analyses comparable to other authors' works, we also created data subsets without organophosphates and carba-mates. The 5th percentile in the NOAEL distribution including the organophosphates and carbamates, N, is 0.06 mg/kg bw per day. Excluding organophosphates and carbamates (N\O), the 5th

Table 4

Percentiles of cumulative distributions of lowest NOAELs and of NOAELs corresponding to lowest LOAELs. Ratios of 97.5th and 2.5th percentiles were calculated to give an estimate of the span of the NOAELs for 95% of the distributions. The distribution models were identified by @risk to best describe the empirical data.

Compounds (number) NOAEL (mg/kg bw per day) LOAEL (mg/kg bw per day)

5th percentile 5th percentile

Distribution Empirical Model Ratio 97.5/2.5 Empirical Ratio 97.5/2.5

NO 59 Levy 0.03 0.02 1'613 0.10 632

N 423 Pearson6 0.06 0.08 2 661 0.40 739

Nw 423 Pearson6 0.04 0.07 1 500 0.22 1 045

Ns 423 Pearson6 0.10 0.13 2 273 0.73 657

Nsw 423 Pearson6 0.08 0.10 1 368 0.36 860

N8 418 Lognormal 0.12 0.16 2 703 0.50 730

N8w 418 Pearson6 0.10 0.15 1 633 0.31 626

N8s 418 Lognormal 0.20 0.25 2'027 0.80 533

N8sw 418 Pearson6 0.14 0.21 1 521 0.50 546

N\O 364 Loglogistic 0.22 0.18 924 1.12 268

N\Ow 364 Pearson6 0.22 0.18 390 1.00 250

N\Os 364 Loglogistic 0.32 0.30 703 1.55 323

N\Osw 364 Pearson6 0.31 0.30 431 1.40 280

N8\O 362 Pearson6 0.40 0.34 631 1.40 286

N8\Ow 362 Pearson6 0.34 0.36 300 1.10 182

N8\Os 362 Lognormal 0.50 0.44 700 1.70 282

N8\Osw 362 Pearson6 0.50 0.52 325 1.50 187

N-L\O 364 Loglogistic 0.28 0.22 1 453 1.00 254

N-L\Ow 364 Pearson6 0.22 0.23 500 1.00 150

N-L\Os 364 Loglogistic 0.42 0.36 650 1.50 240

N-L\Osw 364 Pearson6 0.37 0.36 664 1.40 170

N-L8\O 362 Pearson6 0.45 0.35 796 1.26 283

N-L8\Ow 362 Pearson6 0.40 0.37 324 1.10 142

N-L8\Os 362 Pearson6 0.51 0.54 807 1.70 282

N-L8\Osw 362 Lognormal 0.50 0.68 390 1.50 187

N\Oallspec 228 Loglogistic 0.24 0.19 417 1.15 127

N4\O 343 Lognormal 0.50 0.42 936 1.50 384

N8\O_lowweight 181 Lognormal 0.50 0.33 827 1.50 400

N8 \O_highweight 181 Loglogistic 0.31 0.30 294 1.39 100

NM 407 lognormal 0.05 0.09 3013 0.43 776

NM\O 350 Lognormal 0.22 0.17 804 1.09 310

NMun(cl3) 437 Lognormal 0.20 0.20 24 320

NMun(cl3)\O 410 Lognormal 0.30 0.32 12 768

Table 5

Comparison of different databases.

Database Compounds Cramer class III compounds Species Studies Prioritization of studies Standardization

Munro (Munro et al., 1996) Different 437 (410 without Mice, Developmental, Lowest NOAEL Subchronic NOAELs divided by 3 to give an

ranges of organophosphates rats, reproductive, anticipated chronic NOAEL

application and carbamates) rabbits, hamsters subchronic, chronic

Tluczkiewicz (Tluczkiewicz Different 372 (without Mice, Subacute, NOAEL of Subchronic and subacute NOAELs divided

et al., 2011) ranges of organophosphates); rats subchronic, study with by 2 and 6 to give an anticipated chronic

application compiled from 4 different existing databases chronic longest duration NOAEL

Pinalli (Pinalli et al., 2011) Food 113 Not Not specified, TDI divided by No standardization

contact specified only noted 100 to give

materials "typically 90 day oral studies'' surrogate NOAEL

Kalkhof (Kalkhof et al., 2012) Industrial 724 Mice, Subacute, Lowest NOAEL Subchronic and subacute NOAELs divided

chemicals rats subchronic, by 2 and 6 to give an anticipated chronic NOAEL

Feigenbaum (Feigenbaum et al., Pesticides 322 (279 without Not Not specified NOAEL No standardization

2015) organophosphates and carbamates) specified reported by EFSA as basis for the ADI

Present study Pesticides 423 (364 without Mice, Developmental, Lowest NOAEL Besides the NOAELs as reported in studies,

organophosphates rats, reproductive, and NOAEL also NOAELs multiplied by species and

and carbamates) rabbits, subacute, corresponding study specific factors to give anticipated

dogs subchronic, chronic to lowest LOAEL initial doses. For comparison, also Munro's time extrapolation was used.

percentile increases to 0.22 mg/kg bw per day and the span of NOAEL decreases from 2661 (97.5th/2.5th percentile ratio) to 924.

From the Munro database (Munro et al., 1996) (downloaded from http://www.efsa.europa.eu/de/supporting/pub/159e.htm),

the Cramer class III compounds (dataset NMun(cl3) with 437 compounds) and the Cramer class III compounds except organophos-phates and carbamates (dataset NMun(cl3)\O with 410 compounds) were extracted and compared with the datasets NM

(407 compounds) and NM\O (350 compounds). These two latter datasets contain the lowest NOAELs of PSMTOXDB including and excluding organophosphates/carbamates, respectively. NM and NM\O were generated by applying Munro's selection criteria to PSMTOXDB, i.e. mainly excluding dog studies and subacute studies. Furthermore, as done by Munro and colleagues, NOAEL of sub-chronic studies (but not for multi-generation studies) were divided by 3 to extrapolate to a hypothetic chronic NOAEL. In Munro's paper (Munro et al., 1996) it is stated, that a significant proportion of the subchronic NOAELs were lower than chronic NOAEL. It is however not clear from the publication, whether all subchronic NOAELs in the reference database were divided by 3 and then the lowest NOAEL per compound was identified or whether the lowest NOAELs per compound were identified first and then, if a lowest NOAEL for a compound was a subchronic NOAEL, was divided by 3. For the present analysis, it is assumed that Munro and colleagues divided all subchronic NOAELs by 3 and only then performed the selection of lowest NOAELS. Therefore, in PSMTOXDB all subchronic NOAELs were also divided by 3. It is assumed that the pairs of subsets NMun(cl3) and NM on one side and NMun(cl3)\O and NM\O on the other side were generated on different databases (Munro's database and PSMTOXDB, respectively) but by applying similar selection criteria.

The 5th percentile of the empirical NOAEL distributions are 0.05 and 0.22 for NM and NM\O and for NMun(cl3) and NMun(cl3)\O the values are 0.20 and 0.30. The 4-fold difference between NM and NMun(cl3) is explained mainly by the different proportions of organophosphates and carbamates (among the most toxic compounds in PSMTOXDB) in the two datasets. In PSMTOXDB, the proportion of organophosphates and carbamates is approximately 14% and only 6% in the Munro database. It is assumed that the higher the proportion of the highly toxic organophosphates and carba-mates is in a distribution, the lower the 5th percentile would be. This is supported by the fact that the 4-fold difference between the 5th percentiles of NM and NMun(cl3) decreases to a less than twofold difference if organophosphates and carbamates are excluded (NM\O and NMun(cl3)\O).

Although the differences in the 5th percentiles of NM\O and NMun(cl3)\O are less than twofold, the shapes of the cumulative distributions are significantly different (Fig. 1). The NMun(cl3)\O dataset seems to contain more compounds with relatively low tox-icity compared to MN\O. This is also illustrated by the difference in the 97.5th/2.5th percentile ratios. NM\O NOAELs are within a range of less than 3 orders of magnitude whereas NMun(cl3)\O NOAELs cover a range of more than 4 orders of magnitude. Another explanation might be that the Munro database includes compounds with relatively few toxicological studies. As will be discussed later this shifts lowest NOAEL distributions to higher values.

Using the Munro criteria on PSMTOXDB (NM\O) results in a lowest NOAEL distribution virtually indiscernible from the distribution derived by including studies of all exposure durations, of all species and not using exposure duration standardization factors (N\O) (data not shown). The effects of two major parameters on lowest NOAEL distributions in the Munro criteria neutralize each other. First, the exclusion of shorter than subchronic studies and all dog studies shifts the Munro lowest distributions to higher values whereas the application of the exposure duration standardization factor (division of subchronic NOAELs by 3) shifts it to lower values.

3.3. Effect of dose spacing

It is reasonable to assume that the wider the dose spacing in studies is the wider NOAELs and LOAELs scatter around the toxico-logical threshold. Therefore, heterogeneity in dose spacing of

0.01 0.1 1 10 100 1000 NOAEL (mg/kg bw per day)

Fig. 1. Lowest NOAEL distributions selected according the criteria of (Munro et al., 1996). The solid line depicts the distribution of lowest NOAELs from PSMTOXDB (NM\O) selected and transformed according to the criteria of Munro. The short-dashed line depicts the Cramer class III compounds in the Munro database (NMun(cl3)\O). In both distributions, organophosphates and carbamates were excluded. The inserted plot magnifies the area around the 5th percentile of the cumulative distributions.

studies may have an effect on NOAEL and LOAEL distributions and hence on related analyses. Recently, it was demonstrated that the heterogeneity in dose spacing of studies influences NOAEL distributions and thereby influences statistical analyses of the effects of different exposure durations on NOAEL (Zarn et al., 2013). In the present analyses, the dose spacing also significantly affects lowest NOAEL distributions. For the N\O distribution, the 5th percentile is 0.22 mg/kg bw per day. Selecting lowest NOAELs only from studies with dose spacing 68 (N8\O) increases the 5th percentile by 79% to 0.40 mg/kg bw per day and the 5th percentile for N4\O by 123% to 0.50 mg/kg bw per day (Table 4 and Fig. 2).

The shift of lowest NOAEL distributions to higher values as a consequence of applying dose spacing values as cut off criterion is significantly more pronounced than for the corresponding LOAEL distributions (Fig. 3).

The 5th percentiles of LOAELs of the N8\O and N4\O distributions increase only by 25% and 35%, respectively, compared to N\O. This indicates that in a fraction of the studies (predominantly in the chronic studies) the dose levels expected to give a NOAEL (based on findings in preceding shorter-term studies) were chosen conservatively low by the study directors to ensure that a NOAEL is achieved at all. Therefore, lowest NOAEL distributions not accounting for the dose spacing provide overly low lowest NOAEL distributions.

3.4. NOAELs related to lowest LOAELs

As it was found that lowest NOAEL distributions are sensitive to the heterogeneity of dose spacing in the underlying dataset but that corresponding LOAEL distributions are quite insensitive, an alternative approach to investigate and to minimize the influence of dose spacing was evaluated. Instead of selecting the lowest NOAELs per compound, the lowest LOAELs were identified and then the corresponding NOAELs were used for lowest NOAEL distributions. By this selection of lowest NOAEL distributions, the 5th percentile of N-L\O (selection of NOAELs corresponding to lowest LOAELs) is increased by 26% to 0.28 mg/kg bw per day compared to 0.22 mg/kg bw per day in the N\O distribution (direct

0.01 0.1 1 10 100 1000 NOAEL (mg/kg bw per day)

Fig. 2. Influence of dose spacing on lowest NOAEL distributions from PSMTOXDB. In all lowest NOAEL distributions, organophosphates and carbamates were excluded. The solid line depicts the lowest NOAEL distribution without study selection regarding dose spacing (N\O). The short-dashed line depicts the lowest NOAEL distribution of studies with dose spacing 68 (N8\O). The long-dashed line represents the lowest NOAEL distribution of studies with dose spacing 64 (N4\O). The inserted plot magnifies the area around the 5th percentile of the cumulative distributions.

selection of lowest NOAELs). If this procedure is applied to studies preselected for dose spacing 68, the effect is smaller. The 5th percentile of N-L8\O is increased by 12% to 0.45 mg/kg bw per day compared to 0.40 mg/kg bw per day in N8\O. These upward shifts in NOAEL distributions selected on lowest LOAELs were not pronounced enough to attain statistical significance (Mann Whitney-U Test, p 0.05). This analysis shows that NOAEL selection based on lowest LOAELs only partly minimizes the distortive effects of NOAELs of studies with high dose spacing on NOAEL

distributions. However, a combination of both methods, extracting NOAELs corresponding to lowest LOAELs from studies preselected for dose spacing 68, minimizes the effect of dose spacing and warrants that the lowest LOAEL (not necessarily coming from the same study with the lowest NOAEL) was caught.

3.5. Standardization method for NOAEL and LOAEL from different study types

Recently, it was demonstrated that extending the duration of exposure of rodent toxicity studies beyond 4 weeks does not statistically significantly lower NOAELs and LOAELs if study design parameters are accounted for (Zarn et al., 2011). This suggests that even a 4-week dosing period is capable of approximating the threshold of toxicity as reliably as a chronic study. Although extended exposure duration indeed might increase the severity of effects beyond 4 weeks at the LOAEL identified in a subacute study, the LOAEL itself is not further lowered. In certain study designs the dose is applied either by providing feed at constant compound concentrations or by capsule or gavage without regular dose adjustment accounting for the constant dose decrements related to the body weight increase (Zarn et al., 2013). In such study designs, the NOAEL reported is an average over time within a constantly decreasing dose curve. If the dose in a study was constant over time, the average dose could be seen as the expected value of the random variable dose. However, within a dose curve constantly decreasing due to body weight increase, the average dose over time is not an approximation of the expected value of the dose but rather a mathematical measure without biological justification. Therefore, to account for the toxicological significance of the early exposures and to make NOAELs and LOAELs from different study designs and durations comparable, the NOAELs and LOAELs were standardized to the initial dose the animals received in the respective studies. In principle, this is the opposite of the procedure used by Munro (Munro et al., 1996) and others where NOAELs of subchronic studies (average doses over subchronic phase) were divided by a default factor to extrapolate to an expected chronic NOAEL (average doses over chronic phase).

In Fig. 4 the effect of standardizing NOAELs to the initial dose at the beginning of the respective study is exemplified with the lowest NOAEL distribution and the standardized lowest NOAEL distribution. The standardizing of NOAELs in the lowest NOAEL distribution N8\Os increases the 5th percentile by 24% to 0.50 mg/kg bw per day and the 97.5th/2.5th percentile ratio is hardly changed. This effect of standardizing the reported NOAEL to the initial dose on the 5th percentiles and the 97.5th/2.5th per-centile ratios is consistently seen (Table 4) in lowest NOAEL distributions and the effect was statistically significantly in all cases (Mann Whitney-U Test, p = 0.05).

3.6. Weighting of NOAELs

All TTC analyses are based on NOAEL distributions using the lowest NOAELs identified for the compounds of interest but the comprehensiveness of the databases used in these analyses varies. Therefore, it is difficult to assign differences in the outcome of analyses from different databases to true chemical group-specific different toxicities. As every additional study for a given compound either lowers the currently lowest NOAEL or is above the currently lowest NOAEL and hence does not change it, a growing database for a given compound by tendency lowers its lowest NOAEL. Theoretically, the more extensive a database becomes by including new studies for given compounds, by tendency the more the distribution shifts to lower values. To investigate whether this plausible assumption is verifiable by data, the N8\O distribution was ranked according to the weighting values of the compounds and then

0.01 0.1 1 10 100 1000 LOAEL (mg/kg bw per day)

Fig. 3. Influence of dose spacing on LOAEL distributions from PSMTOXDB. For LOAEL distributions, the LOAELs related to the lowest NOAELs were used. In all LOAEL distributions, organophosphates and carbamates were excluded. The solid line depicts the LOAEL distribution without study selection regarding dose spacing (L\O). The short-dashed line depicts the LOAEL distribution of studies with dose spacing 68 (L8\O). The long-dashed line represents the NOAEL distribution of studies with dose spacing 64 (L4\O). The inserted plot magnifies the area around the 5th percentile of the cumulative distributions.

Fig. 4. Effect of weighting and standardization of NOAELs on lowest NOAEL distributions from PSMTOXDB. In the three lowest NOAEL distributions, organophosphates and carbamates were excluded and only studies with a dose spacing 68 were used. The solid line depicts the lowest NOAEL distribution without any further transformation (N8\O). The short-dashed line represents the distribution of weighted NOAELs (N8\Ow) and the long-dashed line outlines the distribution of the standardized NOAELs (N8\Os). The inserted plot magnifies the area around the 5th percentile of the cumulative distributions.

divided into two equal fractions. N8\O_lowweight is the fraction of compounds with relatively few studies per compound and/or few species tested (low weighting values) and N8\O_highweight is the fraction of compounds with relatively many studies per compound and/or many species tested (high weighting values). In Fig. 5 the two sub-distributions N8\O_lowweight and N8\O_highweight are depicted as the un-weighted NOAELs. N8\O_highweight has a lower 5th percentile (0.31 versus 0.50 mg/kg bw per day) and a lower NOAEL span (97.5th/2.5th percentile ratio of 294 versus 827) than the N8\O_lowweight distribution. The two sub-distributions N8\O_lowweight and N8\O_highweight were statistically significantly different (Mann Whitney-U Test, p = 0.05).

So, the broader the database of toxicity studies for given compounds is, the more the NOAEL distribution shifts to lower values and the tighter the distribution becomes. Because the extent of the database affects lowest NOAEL distributions, all lowest NOAEL subsets generated from PSMTOXDB by applying different selection criteria were additionally transformed by weighting the individual NOAELs with the categorized weighting factors (1, 3 or 12, respectively) derived for the respective compounds. Fig. 4 shows the effect of weighting on the cumulative NOAEL distributions. The 5th percentiles decrease by 15% (from 0.40 to 0.34 mg/kg bw per day) and the range of NOAELs by 53% (97.5th/2.5th percentile ratio decreases from 631 to 300). The values reflecting the shift (5th percentiles) and the shape of the distribution (97.5th/2.5th percentile ratios) for the pair of cumulative lowest NOAEL distributions N8\O and N8\Ow are similar for other pairs of weighted and un-weighted cumulative lowest NOAEL distributions (Table 4). The present method for weighting of the database extent decreases the lower bound (2.5th and 5th percentile) little to moderately and moderately the upper bound (97.5th percentile) of cumulative lowest NOAEL distributions. It is concluded that although the higher NOAELs are over-represented in un-weighted distributions the 5th percentiles shifts downwards only by 1020% after weighting of the NOAELs for their underlying database. Statistically, the small shifts between the pairs of distributions attained statistical significance (Mann Whitney-U Test, p = 0.05) only in the pairs N8s vs. N8sw and N8\Os vs. N8\Osw and

Fig. 5. Effect of weighting of NOAELs on lowest NOAEL distributions from PSMTOXDB. The lowest NOAEL distribution N8\O (organophosphates and carba-mates excluded, only studies with a dose spacing 68) was divided into two equal fractions. In one fraction only NOAELs from compounds with high weighting scores (N8\O_highweight, solid line) and in the other fraction only NOAELs from compounds with low weighting scores (N8\O_lowweight, short-dashed line) were included. The inserted plot magnifies the area around the 5th percentile of the cumulative distributions.

N-L8\Os vs. N-L8\Osw. Although the present method of weighting was powerful enough to demonstrate the relevance of the database available per compound within a given lowest NOAEL distribution, it was weak in revealing the effect if un-weighted and weighted lowest NOAEL distributions were compared. However, the 5th per-centiles and the 97.5th/2.5th percentile ratios consistently decrease in lowest NOAEL distributions if the NOAELs are weighted.

The principle of favoring well documented lowest NOAELs over poorly documented lowest NOAELs scientifically seems justified. However, the implementation of such a system bears certainly considerable arbitrariness as the "sensitivity" factors used directly are related to the database per species available. And the database per species available is related to the pesticide approval data requirements.

4. Summary

In current regulatory risk assessment practice, often a NOAEL is the point of departure for derivation of health-based guidance values such as ADIs and TDIs. The NOAEL of a study is identical with one of the doses pre-assigned during the study planning and is the dose at which no statistically and/or biologically significant adverse effects are observed. For non-genotoxic compounds, it is assumed that species-specific toxicological thresholds per compound exist. For a given study, design parameters such as dose spacing and statistical power determine how far a NOAEL is apart from the toxicological threshold. The number of studies and the number of species tested for a given compound are additional determinants in the approximation of the lowest toxicological threshold. However, since not all species can be tested with utmost sensitivity and statistical power, regulatory toxicology studies always are only attempts to approximate the toxicological threshold for a compound with a reasonable effort. After decades of animal testing of chemicals there is a wealth of data available that can be investigated for general principles regarding structure-activity relationships as was done in the formation of the TTC concept over the last decades. The present work focused on the effects of study

design and database parameters on NOAEL distributions and tried to identify possible areas of improvement in the derivation of tox-icological default values (such as TTC values) from scientifically appropriate NOAEL distributions. For such a retrospective analysis, differently designed study types and designs are very helpful as only variability in parameters permits the identification of the influence of a parameter on an outcome of interest.

Before different parameters were investigated for their influence on lowest NOAEL distributions, the PSMTOXDB was calibrated to the Munro database to estimate where the PSMTOXDB data lie relative to the Munro data. Using the 5th percentile as reference point, the lowest NOAEL distribution (NM\O) of Cramer class III compounds was less than twice as toxic but the NOAEL range was more than tenfold smaller compared to Munro's distribution (NMun(cl3)\O). If the chemically very heterogeneous pesticides are compared to Munro's dataset it seems that "pesticides" in general are significantly but only slightly more toxic than other Cramer class III compounds.

An important study design parameter is the dose spacing of studies. Using only studies with restricted dose spacing (N4\O and N8\O) shifts the lowest NOAEL distribution ca. twofold (5th percentile) upwards. A similar but milder effect is achieved if not the lowest NOAELs are selected but the NOAELs related to the lowest LOAELs.

It was found recently that exposure duration has no significant effect on NOAEL beyond 4 weeks of exposure. Based on this finding that a NOAEL essentially is revealed within the first 4 weeks of a study, the standardization method used by Munro to make NOAELs from different study designs comparable was adapted. Instead of standardizing subchronic NOAELs to chronic NOAELs by division by 3, in all studies using non-constant dosing regimens the doses were standardized to the initial dose animals received. This takes into account that the initial high doses apparently contribute more to the identification of the NOAEL than the decreasing doses observed in many study designs as exposure durations increase. The standardization factors used are species and study duration-specific values (Zarn et al., 2010, 2013). Using this standardization method shifts lowest NOAEL distributions by 10-50% (e.g. N\O and N\Os) upwards.

The extent of the toxicological database available for the compounds influences the NOAEL distributions. The lowest NOAEL distribution N8\O was divided into two fractions. N8\O_lowweight is the fraction of compounds with a weaker underlying database (less studies and/or species per compound) than the fraction N8\O_highweight. It was found that the lowest NOAEL distribution of N8\O_highweight was shifted downwards and that the distribution was tighter. Based on this finding the weighting system was applied to all datasets. There was a consistent downward shift of lowest NOAEL distributions if the lowest NOAELs of more intensely studied compounds were higher weighted than lowest NOAELs of less intensely studied compounds (e.g. N8\O and N8\Ow). However the effect of weighting clearly seen within a lowest NOAEL distribution (N8\O segmented into N8\O_lowweight and N8\O_highweight) was only weakly translated if a lowest NOAEL distribution was transformed to a weighted distribution. The weighting system might be amended by choosing other weighting factors and by including for example weights for study types (not only study number) and statistical power of the study (animal number per dose group).

5. Conclusion

The present analyses on lowest NOAEL distributions from PSMTOXDB demonstrate effects of study design and database-related parameters of varying potency on the distributions.

If only studies were used with restricted dose spacing, a marked upward shift of lowest NOAEL distributions was observed. By trend, a similar but much less pronounced shift is achieved by using NOAELs related to lowest LOAELs instead of directly selecting lowest NOAELs.

A likewise upward shift of lowest NOAEL distributions comes from standardization of NOAELs to the initial dose animals received in the respective study. This standardization to the initial dose seems justified as for a given compound no shift to lower NOAELs is observed upon extension of exposure duration.

A lowest NOAEL distribution downward-shifting effect was observed if the number of studies and species tested per compound were allowed for. However, the performance of the weighting system used is weak and might be improved. Additionally, the weighting system might be severely distorted by several factors, e.g. regarding species sensitivities because rats, dogs, mice and rabbits are tested in quite different study types.

In conclusion, if databases of NOAELs are used for quantitative analyses of chemical structure group-related NOAEL distributions, clear selection criteria and transforming procedures allowing for study design and database parameters might be necessary in case that several studies per compound are available. In the present work, the most prominent effect came from differences in dose spacing between studies. Therefore, in databases including several studies per compound, the most relevant NOAELs are the NOAELs related to the lowest LOAELs from a dose spacing restricted subset. Reasonable dose spacing cut off values should be assigned after statistical evaluations of the database. To account for different study designs and exposure durations, NOAELs might be standardized to initial dose animals received at the beginning of the respective study. To account for differences in the extent of data per compound, a weighting system might be applied to give different weights to toxicologically differently investigated compounds.

In summary, for NOAELs of Cramer class III pesticides in PSMTOXDB the 5th percentile of 0.22 mg/kg bw per day (N\O distribution, no specific selection criteria, no transformations) increases more than twofold to 0.50 mg/kg bw per day in the more specific N-L8\Osw distribution (Fig. 6).

0.01 0.1 1 10 100 1000 NOAEL (mg/kg bw per day)

Fig. 6. Summary effect of selection criteria and transformations on lowest NOAEL distributions (organophosphates and carbamates excluded) from PSMTOXDB. The solid line depicts the lowest NOAEL distribution without specific study selection criteria or transformations (N\O). The short-dashed line represents the distribution of weighted and standardized NOAELs related to lowest LOAELs from studies with a dose spacing 68 (N-L8\Osw). The inserted plot magnifies the area around the 5th percentile of the cumulative distributions.

As the parameters investigated here for their influence on NOAEL distributions are not chemical group-specific but rather related to study design/databases, the findings might be applicable to NOAEL distributions of chemicals other than pesticides and also to other Cramer classes

The results show that database and study design parameters slightly influence NOAEL distributions and that current TTC values are therefore affected only to a minor extent by these parameters and can be considered reliable in that perspective.

Funding sources statement

The authors declare that this work was not funded by any sources.

Conflict of interest statement

The authors declare that there are no conflicts of interest.

Transparency Document

The Transparency document associated with this article can be found in the online version.

References

Bhatia, S., Schultz, T., Roberts, D., Shen, J., Kromidas, L., Marie, A.A., 2015. Comparison of Cramer classification between Toxtree, the OECD QSAR Toolbox and expert judgment. Regul. Toxicol. Pharmacol. 71, 52-62.

Bruschweiler, B.J., 2014. The TTC approach in practice and its impact on risk assessment and risk management in food safety. A regulatory toxicologist's perspective. Chimia (Aarau.) 68, 710-715.

Cramer, G.M., Ford, R.A., Hall, R.L., 1978. Estimation of toxic hazard - a decision tree approach. Food Cosmet. Toxicol. 16, 255-276.

Dekant, W., Melching-Kollmuss, S., Kalberlah, F., 2010. Toxicity assessment strategies, data requirements, and risk assessment approaches to derive health based guidance values for non-relevant metabolites of plant protection products. Regul. Toxicol. Pharmacol. 56, 135-142.

EFSA, 2012. Scientifi Opinion on Exploring options for providing advice about possible human health risks based on the concept of Threshold of Toxicological Concern (TTC). Available from: <http://www.efsa.europa.eu/en/efsajournal/ pub/2750.htm>.

EFSA, 2014. Rapporteur Member State assessment reports submitted for the EU peer review of active substances used in plant protection products. Available from: <http://dar.efsa.europa.eu/dar-web/provision> (accessed 8.1.15).

FAO/WHO, 2009. Principles and Methods for the Risk Assessment of Chemicals in Food. Available from: <http://www.who.int/foodsafety/chem/principles/en/ index1.html> (accessed 12.6.14).

Feigenbaum, A., Pinalli, R., Giannetto, M., Barlow, S., 2015. Reliability of the TTC approach: learning from inclusion of pesticide active substances in the supporting database. Food Chem. Toxicol. 75, 24-38. doi:http://dx.doi.org/10. 1016/j.fct.2014.10.016. Epub@2014 Nov 10.

Galloway, S.M., Vijayaraj, R.M., McGettigan, K., Gealy, R., Bercu, J., 2013. Potentially mutagenic impurities: analysis of structural classes and carcinogenic potencies

of chemical intermediates in pharmaceutical syntheses supports alternative methods to the default TTC for calculating safe levels of impurities. Regul. Toxicol. Pharmacol. 66, 326-335.

Kalkhof, H., Herzler, M., Stahlmann, R., Gundert-Remy, U., 2012. Threshold of toxicological concern values for non-genotoxic effects in industrial chemicals: re-evaluation of the Cramer classification. Arch. Toxicol. 86,17-25.

Kroes, R., Renwick, A.G., Cheeseman, M., Kleiner, J., Mangelsdorf, I., Piersma, A., Schilter, B., Schlatter, J., van, S.F., Vos, J.G., Wurtzen, G., 2004. Structure-based thresholds of toxicological concern (TTC): guidance for application to substances present at low levels in the diet. Food Chem. Toxicol. 42, 65-83.

Kroes, R., Renwick, A.G., Feron, V., Galli, C.L., Gibney, M., Greim, H., Guy, R.H., Lhuguenot, J.C., van de Sandt, J.J., 2007. Application of the threshold of toxicological concern (TTC) to the safety evaluation of cosmetic ingredients. Food Chem. Toxicol. 45, 2533-2562.

Laufersweiler, M.C., Gadagbui, B., Baskerville-Abraham, I.M., Maier, A., Willis, A., Scialli, A.R., Carr, G.J., Felter, S.P., Blackburn, K., Daston, G., 2012. Correlation of chemical structure with reproductive and developmental toxicity as it relates to the use of the threshold of toxicological concern. Regul. Toxicol. Pharmacol. 62, 160-182.

Leeman, W.R., Krul, L., Houben, G.F., 2014. Reevaluation of the Munro dataset to derive more specific TTC thresholds. Regul. Toxicol. Pharmacol. 69, 273-278.

Melching-Kollmuss, S., Dekant, W., Kalberlah, F., 2010. Application of the ''threshold of toxicological concern'' to derive tolerable concentrations of ''non-relevant metabolites'' formed from plant protection products in ground and drinking water. Regul. Toxicol. Pharmacol. 56,126-134.

Muller, L., Mauthe, R.J., Riley, C.M., Andino, M.M., Antonis, D.D., Beels, C., DeGeorge, J., De Knaep, A.G., Ellison, D., Fagerland, J.A., Frank, R., Fritschel, B., Galloway, S., Harpur, E., Humfrey, C.D., Jacks, A.S., Jagota, N., Mackinnon, J., Mohan, G., Ness, D.K., O'Donovan, M.R., Smith, M.D., Vudathala, G., Yotti, L., 2006. A rationale for determining, testing, and controlling specific impurities in pharmaceuticals that possess potential for genotoxicity. Regul. Toxicol. Pharmacol. 44,198-211.

Munro, I.C., Ford, R.A., Kennepohl, E., Sprenger, J.G., 1996. Correlation of structural class with no-observed-effect levels: a proposal for establishing a threshold of concern. Food Chem. Toxicol. 34, 829-867.

Pinalli, R., Croera, C., Theobald, A., Feigenbaum, A., 2011. Threshold of toxicological concern approach for the risk assessment of substances used for the manufacture of plastic food contact materials. Trends Food Sci. Technol. 22, 523-534.

Tluczkiewicz, I., Buist, H.E., Martin, M.T., Mangelsdorf, I., Escher, S.E., 2011. Improvement of the Cramer classification for oral exposure using the database TTC RepDose - a strategy description. Regul. Toxicol. Pharmacol. 61, 340-350.

US EPA, 2011. Fact Sheets on New Active Ingredients (this database was not further maintained by US EPA, accessed 27.4.11).

van Ravenzwaay, B., Dammann, M., Buesen, R., Schneider, S., 2011. The threshold of toxicological concern for prenatal developmental toxicity. Regul. Toxicol. Pharmacol. 59, 81-90.

van Ravenzwaay, B., Dammann, M., Buesen, R., Flick, B., Schneider, S., 2012. The threshold of toxicological concern for prenatal developmental toxicity in rabbits and a comparison to TTC values in rats. Regul. Toxicol. Pharmacol. 64,1-8.

WHO, 2011. Joint FAO/WHO Meeting on Pesticide Residues (JMPR) publications. Available from: <http://www.who.int/foodsafety/publications/jmpr-monographs/en/> (accessed 20.3.15).

Zarn, J.A., Hanggi, E., Kuchen, A., Schlatter, J.R., 2010. The significance of the subchronic toxicity in the dietary risk assessment of pesticides. Regul. Toxicol. Pharmacol. 58, 72-78.

Zarn, J.A., Engeli, B.E., Schlatter, J.R., 2011. Study parameters influencing NOAEL and LOAEL in toxicity feeding studies for pesticides: exposure duration versus dose decrement, dose spacing, group size and chemical class. Regul. Toxicol. Pharmacol. 61, 243-250.

Zarn, J.A., Engeli, B.E., Schlatter, J.R., 2013. Characterization of the dose decrement in regulatory rat pesticide toxicity feeding studies. Regul. Toxicol. Pharmacol. 67, 215-220.