- Email: [email protected]
Relevance of non-guideline studies for risk assessment: The coverage model based on most frequent targets in repeated dose toxicity studies
Toxicology Letters 218 (2013) 293–298
Contents lists available at SciVerse ScienceDirect
Toxicology Letters journal homepage: www.elsevier.com/locate/toxlet
Relevance of non-guideline studies for risk assessment: The coverage model based on most frequent targets in repeated dose toxicity studies M. Batke a,∗ , T. Aldenberg b , S. Escher a , I. Mangelsdorf a a b
Fraunhofer ITEM, Nikolai-Fuchs-Str. 1, 30625 Hannover, Germany RIVM-Laboratory for Ecological Risk Assessment, P.O. Box 1, Anthonie van Leeuwenhoeklaan 9, 3720 BA Bilthoven, The Netherlands
h i g h l i g h t s
Coverage approach: evaluation of scope of examination of old repeated dose toxicity studies. Statistical model calculates probability that LOEL or the next higher dose (LOEL + 1) is determined. Evaluation based on Fraunhofer RepDose DB. Statistical model can be transferred to other datasets.
a r t i c l e
i n f o
Article history: Received 31 July 2012 Received in revised form 31 August 2012 Accepted 3 September 2012 Available online 10 September 2012 Keywords: Extrapolation factors Repeated dose toxicity Integrated testing strategy Study quality OECD guideline
a b s t r a c t A common challenge for human risk assessment is the quality of the available animal studies. Nonguideline studies are often limited for different aspects of study design and documentation. Within this publication the relevance of a limited scope of examination is discussed. Preliminary analyses of the RepDose database have shown that liver, body weight, kidney and clinical symptoms are frequently affected in oral repeated dose toxicity in rats and mice (Bitsch et al., 2006), while many other targets are seldom affected. As most of these targets are investigated frequently also in non-guideline studies, it is likely that they provide a reliable NOEL, although the full spectrum of endpoints has not been covered. Based on RepDose data we investigate the relevance of individual targets for determining the LOEL and the consequences for risk assessment. The resulting coverage model for subchronic oral rat studies includes up to six targets and an additional assessment factor for LOEL extrapolation. It can be applied to assess the reliability of non-guideline studies with respect to the scope of examination. Furthermore the application of the coverage model to other databases will increase and/or specify the chemical domain and reveal respective targets as well as effects. © 2012 Elsevier Ireland Ltd. All rights reserved.
1. Introduction In the evaluation of studies on existing chemicals, the risk assessor often faces the situation that repeated dose toxicity studies are available that are rather old, i.e. performed before the implementation of guidelines, or were not conducted according to guidelines. For these studies, less than three dose levels may be available or a low number of animals/dose group. In addition the scope of investigations may be reduced with regard to the number of organs investigated by histopathology. Further, organ weights,
∗ Corresponding author at: Institut für Toxikologie und experimentelle Medizin, Abteilung Chemikalienbewertung, Nikolai-Fuchs-Str. 1, 30625 Hannover, Germany. Tel.: +49 0511 5350 344; fax: +49 0511 5350 335. E-mail address: [email protected] (M. Batke). 0378-4274/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.toxlet.2012.09.002
clinical chemical examinations, hematology or urine analysis may not be available. In the REACH program, the quality of a study is determined by the Klimisch Code (Klimisch et al., 1997): Depending on the scope of examinations and quality of documentation studies are “reliable” (Klimisch Code 1), “reliable with restriction” (Klimisch Code 2) or “not reliable” (Klimisch Code 3). Many of the old, non-guideline studies can be assigned to reliability 2 or even 3. In the case of Klimisch Code 3, the study is usually considered as not sufficient for evaluation and new animal testing is necessary. However, with the aim of saving experimental animals, it is desirable to use existing studies in risk assessment as far as possible. At Fraunhofer ITEM a database on repeated dose toxicity studies (RepDose) has been developed, that contains currently more than 1100 oral repeated dose toxicity studies, mainly on existing chemicals. These studies include subacute, subchronic and chronic studies with rats and mice with reliability 1 and 2, and
294
M. Batke et al. / Toxicology Letters 218 (2013) 293–298
Table 1 Datasets selected from RepDose for further analyses. Study duration
Study type
Subacute Subchronic Subchronic
Comprehensive Comprehensive Old
2.2. Statistics The probability that a study detects the LOEL, or the LOEL + 1, when only a limited number of targets has been investigated in a study, can be calculated by methods of categorical data analysis, in particular loglinear modelling (Agresti, 2002; Bishop et al., 1975; Christensen, 1997; Fienberg, 1980; Powers and Xie, 1999). We aimed to identify the most suitable and generally usable model, avoiding overfitting. Models that overfit will predict badly in new situations: they have limited generalizability. We have found the independent organ model to be generally acceptable to calculate the probability that the LOEL or the LOEL + 1 were detected. The model selection is documented in supplementary material. The multinomial linear model without interaction terms can be interpreted as the independent organ model. For example, at the LOEL, the fraction of studies with particular organ combinations affected can be modelled by multinomial probability parameters ϕ which are proportional to the exponential function of a linear expression of the target organs. Suppose L, K, C, and B are binary values (0 or 1) of the target organs: liver, kidney, clinical chemistry, and body weight. A value of 0 means not affected, or not reported to be affected, while a 1 means affected. Then probabilities ϕ are modelled as:
No. of studies Rats
Mice
52 88 56
≤10 45 ≤10
also some studies with reliability 3. Preliminary analyses of the overall database have shown that liver, body weight, kidney and clinical symptoms are frequently affected in oral repeated dose toxicity in rats and mice (Bitsch et al., 2006), while many other organs are seldom affected. In inhalation studies the respiratory tract and especially the nose and lung are major targets in addition. As most of these targets are investigated frequently also in nonguideline studies, it is likely that non-guideline studies provide a reliable NOEL, although the full spectrum of endpoints has not been covered in the respective study. Based on RepDose data, in this publication we investigate the relevance of individual targets for determining the LOEL and the consequences for risk assessment. A method is developed to calculate the probability that a certain set of targets provides a reliable reference value. A model consisting of up to six targets is proposed for subchronic oral rat studies.
∝ exp(ˇ1 · L + ˇ2 · K + ˇ3 · C + ˇ4 · B). The exponential function factorises into terms to be multiplied:
∝ exp(ˇ1 · L + ˇ2 · K + ˇ3 · C + ˇ4 · B) = exp(ˇ1 · L) · exp(ˇ2 · K) · exp(ˇ3 · C) · exp(ˇ4 · B) = L · K · C · B
Hence, the probability of a study with a certain organ pattern is the product of the probabilities of studies with each individual organ affected. The ˇ-coefficients are estimated by the method of maximum likelihood. Given the data in Table 3 the coverage at the LOEL of a study with liver, kidney and heart examined would be:
2. Materials and methods
Coverage = 1 − (1 − 0.318) × (1 − 0.239) × (1 − 0.057) = 0.51 2.1. Analyses with RepDose RepDose contains effects for different dose levels and general information on the study design as well as a reliability score. The reliability score assesses the study quality with respect to general design and scope of examination. Based on the reliability and the publication date two types of studies were discriminated: studies with “guideline-conform”, OECD-guideline-like design and scope as well as “non-guideline” studies performed before the implementation of OECD guidelines in 1981. Additionally, the following criteria were applied to select the datasets from the RepDose DB:
-
2.3. Simulation of coverage uncertainty Coverage uncertainty was simulated through Bayesian Monte-Carlo sampling of the linear multinomial models, cf. (Albert, 2009). The details are explicated in supplementary material. To sketch the procedure, note that the individual target fractions in Table 3 are estimated from observed fractions at the LOEL, and at LOEL + 1. Thus, the fraction 31.8% of studies showing Liver to be affected at the LOEL, as employed in the above three-organ coverage equation, is estimated from the ratio of studies with the organ affected to the total number of studies: 28/88 = 31.8%, However, this is a so-called point estimate, without any uncertainty addressed. Through Bayesian simulation of the model likelihoods, the uncertainty around these point estimates can be evaluated. This yields the uncertainty of the ˇcoefficients, as well as the estimated fractions of studies with a particular organ affected, and therefore of the individual organ and total coverage. We found that in each organ model, the median of the simulated coverage distributions was very close to the point estimates.
Oral studies (gavage studies, application via drinking water or food). Rats and mice. Subacute and subchronic studies. LOEL available. In addition NOEL, and LOEL + 1 (dose above LOEL) available (only guidelineconformstudies). If a NOEL was not available, also studies with only LOEL and LOEL+1 were included. Conditions were: Only one organ affected at the LOEL or mild effects at the LOEL
3. Results The datasets shown in Table 1 were derived. Datasets with 10 or less studies were not considered for further evaluation. Most analyses were performed with the guideline-conformdataset on rat subchronic studies. Studies were analysed for the targets at their LOEL and at the LOEL + 1. Targets for effects are classical organs (histopathology or organ weight) in toxicological studies, i.e. liver, kidney, brain, etc. In addition, body weight, clinical symptoms, clinical chemistry and haematology are termed targets in our analysis. In the RepDose database no distinction is made between LOELs and LOAELs. Any effect that is statistically different from the control is termed LOEL. Thus, effects that may be considered as non-adverse are also covered in our analyses. The references for the selected studies are given in supplementary material.
3.1. Number of target organs Table 2 shows the number of organs affected in subacute or subchronic studies with rats or mice, at the LOEL, the dose level above the LOEL (LOEL + 1) or for all dose levels investigated. At the LOEL on average more than one target organ is affected in all studies types. With increasing dose levels the average number of targets increases from 1.6 to 4.2 for subacute rat studies and 1.8
Table 2 Number of targets affected for different study types. Species
Rat Rat Rat Mouse a b
Study duration
Subacute Subchronic Subchronic Subchronic
Characterization
Comprehensive Comprehensive Old Comprehensive
n
52 88 56 45
Geometric means and the respective geometric standard deviation in brackets. This figure refers only to 41 of the 56 old oral rat studies which have a LOEL + 1 dose.
No. of targetsa LOEL
LOEL+1
Overall study
1.6 (1.8) 1.8 (1.8) 1.8 (2.0) 1.7 (1.7)
3.7 (2.0) 3.4 (2.0) 3.0 (2.0)b 2.3 (1.9)
4.2 (2.0) 6.5 (1.7) 3.6 (2.3) 4.2 (1.9)
M. Batke et al. / Toxicology Letters 218 (2013) 293–298 Table 3 Fractions of targets affected in old (n = 56) and comprehensive subchronic oral rat studies (n = 88). Fractions are given at the LOEL (old and comprehensive studies) as well as at the LOEL + 1 and for LOELs determined by a single target (comprehensive studies). Target
Clinical chemistry Liver Haematology Kidney Body weight Clinical symptoms Testes Urine analysis Forestomach Heart Spleen Brain Stomach Intestine Adrenal gland Thyroid gland Thymus Lymph node Bladder Endocrine system Nose Pancreas Vascular system Salivary gland Pituitary gland Nervous system Ovary Bone Gallbladder/bileduct Lung Uterus Prostate Seminal vesicle Skeletal muscle Adipose tissue Eye
Fraction (% of studies) Old
Comprehensive
LOEL
LOEL
LOEL + 1
LOEL single target
17.9 44.6 16.1 37.5 37.5 17.9 10.7 17.9
38.6 31.8 29.5 23.9 21.6 18.2 8.0 6.8 6.8a 5.7 4.5 2.3 2.3 2.3 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1
56.8 55.7 46.6 40.9 44.3 42.0 17.0 13.6 9.1a 13.6 15.9 8.0 5.7 3.4 8.0 4.5 5.7 3.4 2.3 2.3 2.3 2.3 1.1 1.1
9.1 5.7 11.4 4.5 2.3 1.1
3.6 7.1 5.4 1.8 5.4 5.4
1.8 1.8 1.8 1.8
1.1 3.4a 1.1 2.3
1.1 1.1
295
at the LOEL + 1 are provided additionally. These data are used to estimate whether the scope of a non-guideline study is sufficient to detect a reliable LOEL or the dose above the LOEL (LOEL + 1). In a subset of studies the LOEL is determined by only one target. The fractions of these “single” targets at the LOEL provide additional information on their relevance (Table 3). In non-guideline studies, most frequently liver, body weight and kidney are affected, followed by clinical chemistry, haematology and clinical symptoms. Although in guideline-conform studies the scope of examination is broader, the same targets show up as frequent targets, although with a somewhat different fraction. Most frequently, i.e. in 38.6% of all studies clinical chemical parameters are changed at the LOEL of a study, followed by liver, haematology, kidney, body weight and clinical symptoms. At the dose level above the LOEL (LOEL + 1) again the same targets appear, the fraction of the targets being affected is considerably increased/about doubled. On the other side, there are some targets like the pancreas, the salivary gland or the nose, that are seldom affected at the LOEL of a study in non-guideline studies, at the LOEL and LOEL + 1 in guideline-conform studies. Out of the 88 guideline-conform oral subchronic rat studies 40 studies (45%) have only a single target at the LOEL. In addition to five of the frequent targets (namely haematology, clinical chemistry, liver, kidney, body weight) forestomach and spleen occur with higher fraction (>2%) alone at the LOEL. In those studies, where the LOEL is triggered only by the forestomach, the route of administration is always gavage. Thus, not only the same organs are occurring frequently at the LOEL, furthermore they are also the targets occurring most frequently as sole targets at the LOEL.
1.1
2.3 2.3 2.3 2.3 1.1 1.1 1.1 1.1 1.1 1.1
LOEL + 1: Dose level above LOEL; LOEL single: proportion of the target affected at the LOEL without any other target. Dark grey: Key targets, light grey: other relevant targets. a Only occurring in gavage studies.
to 6.5 for subchronic rat studies, reflecting more toxic effects at higher dose levels and a significantly increased number of effects overtime (p < 0.001, Mann–Whitney–Wilcoxon-Test). In contrast to the number of targets at the LOEL, the number of overall targets at the study is significantly lower for non-guideline compared to guideline-conform subchronic rat studies. Furthermore the number of targets in subchronic rat studies is significantly higher than in subchronic mouse studies. An interesting finding is that already at the LOEL independent from the study type and duration frequently more than one target is affected. In these cases one of the targets could substitute the other for identifying the LOEL. Further, based on our previous analyses we hypothesize that there will be some major target organs. Therefore we analysed the fraction of the target organs at the LOEL. 3.2. Target organs at the LOEL of guideline-conform and non-guideline studies Table 3 shows the fraction of different targets affected at the LOEL in non-guideline and in guideline-conform rat subchronic oral toxicity studies. For guideline-conform studies the target fractions
3.3. Coverage model So far, we described the distinct fractions of targets affected. The fractions observed for guideline-conform studies can be transformed to probabilities as the targets are within the scope of examination of guideline-conform studies. The statistical method developed to calculate the probability to detect a reference value (LOEL or LOEL + 1) refers not only to single targets but to combinations of targets (Section 2.2). The probability to predict a reference value can also be interpreted as the coverage of the targets included in the model. The following most frequent targets were included in our analysis: Liver, kidney, clinical chemistry, body weight, clinical symptoms, haematology, spleen, testes. The probabilities for eight out of 255 possible coverage models for combined targets and their attributed uncertainties are shown in Table 4. The LOEL is predicted with a probability of 87% if seven (model 7) or eight targets (model 8) are combined. Model 5 with six targets performed nearly as well with 86% coverage, indicating that an inclusion of more than six targets does not increase the probability to detect the LOEL significantly. At the LOEL + 1 the same model improves to 98% probability. However, already coverage model 1 with the four targets liver, kidney, clinical chemistry, and body weight achieved 94% probability for LOEL + 1 prediction. Based on the derived probabilities, coverage above 90 or 95% of the LOEL + 1 is possible with different target combinations. The coverage of the LOEL is not higher than 87% with one of the seven or eight target models. An extrapolation factor LOEL + 1 to LOEL is then required corresponding, e.g. to the dose spacing observed in toxicological studies. Therefore we investigated the dose spacing of the 88 guideline-conform rat studies. Fig. 1 shows the log-normally distributed function (proven by Chi-Square, p 0.05, data not shown) of the LOEL + 1/LOEL ratios. Statistical evaluation gives a geometric mean of 2.6 with a geometric standard deviation of 1.5 and a 95th percentile of 5. The LOEL of a non-guideline study, which meets the
296
M. Batke et al. / Toxicology Letters 218 (2013) 293–298
Table 4 Selected target combinations and their coverage at the LOEL and at the LOEL + 1. Model no.
1 2 3 4 5 6 7 8
Targets
LOEL
Liver
Kidney
Clinical chemistry
Body weight
Clinical symptoms
X X X X X X X X
X X X X X X X X
X
X X X X X X X X
X X X X X X X
X X X X X
Haema tology
X X X X X
Spleen
Testes
X X
X X
Coverage 75% 67% 80% 77% 86% 86% 87% 87%
LOEL + 1 Uncertainty
Coverage
Uncertainty
67%, 82% 58%, 75% 73%, 86% 69%, 83% 80%, 90% 81%, 91% 81%, 91% 82%, 92%
94% 92% 96% 96% 98% 98% 98% 99%
91%, 96% 88%, 95% 94%, 98% 93%, 97% 97%, 99% 97%, 99% 97%, 99% 98%, 99%
X: target included in model; coverage: probability to detect the reference value if the respective targets are included in the statistical model (independent model/linear model). The point estimate of the coverage calculated by the best-fitting linear model is practically equal to the 50% percentile of the MCMC simulations for the uncertainty estimation; uncertainty: 2.5 and 97.5% percentile of the posterior coverage distribution determined by MCMC simulation.
LOEL + 1 in the 6 main target model, can thus be extrapolated with a factor of 2.6–5 to yield a reliable LOEL. 4. Discussion The aim of this publication was to develop a strategy (coverage approach) to assess the reliability of the LOEL of a non-guideline repeated dose toxicity study with respect to the scope of examination. Based on the RepDose DB it is shown that a limited set of six targets consisting of liver, kidney, clinical chemistry, body weight, clinical symptoms and hematology is frequently affected at the LOEL of a guideline-conform subchronic rat study. Their examination within a study gives a probability of 86% to detect the LOEL (see model 5, Table 4). For the next higher dose level the overall probability that these targets are sufficient for proper detection is as high as 98%. If additional targets are considered, the probability for the LOEL increases marginally and for the LOEL + 1 it stays the same. The coverage model including six targets is thus sufficient based on our dataset. The LOEL of studies with scope of examinations limited to these six targets can be considered to be equal to the LOEL + 1 of a similar guideline-conform study with a probability of 98%. Our analyses demonstrates that the LOEL + 1 is higher by a factor of 2.6 (geometric mean) to 5 (95th percentile) than the LOEL. This gives an indication for the value of an assessment factor to be used in risk assessment. Assessment factors of up to 10 for the LOEL to NOEL extrapolation as indicated in the REACH guidance R8 are supported by our analyses of dose spacing (ECHA, 2010). The
Fig. 1. Distribution of LOEL + 1/LOEL-ratios for all 88 comprehensive subchronic rat studies.
extrapolation from LOEL + 1 to LOEL in the context of the coverage model accounts for the quality of the respective study. Thus our data also support the use of assessment factors > 1 for data quality as proposed by the REACH Guidance Document R8 (ECHA, 2010). Additional aspects of the data quality are discussed in more detail in the integrated testing strategy (see below, Tluczkiewicz et al., 2012). Further the probability of identifying the LOEL increases, if more than one study of limited quality but different scope of examination is available. It may be doubtful that in non-guideline studies the six targets of model 5 are analysed. The fractions of targets in non-guideline studies (rat oral subchronic) as shown in Table 3, however, indicate that the frequent targets observed in guideline-conform repeated dose toxicity studies (rat oral subchronic) are frequently observed as well. This observation gives confidence in the applicability of the coverage approach. These results indicate that some non-guideline Klimisch code 2–3 studies, with reduced scope compared to guideline-conform studies, do still provide reliable information. Their consideration in risk assessment may prevent additional animal studies. It has been shown that the number of targets is higher in rat studies than in mouse studies. This supports the use of rats as first choice for a species to be investigated in toxicological studies, i.e. detecting effects reliably. The sensitivity of rats and mice is, however not principally different, as has been show in our most recent analyses on interspecies differences (Escher et al., 2012). Our findings are supported by other publications addressing the importance of targets (organs). Travlos et al. (1996) identified histopathological lesions in the liver in 20 of 60 and the kidney in 25 of 61 rat subchronic toxicity studies. However they did not restrict their analyses to the LOEL as we did. Nevertheless also their results support the high abundance of liver and kidney effects. Further they showed that also clinical chemical parameters are increased after 13 weeks in a great proportion of studies. As indicators of liver toxicity alanine aminotransferase (32%), sorbitol dehydrogenase (29%) and total bile acids (43%) are affected. Kidney effects were attributed to urea nitrogen (26%), creatinine (18%), albumin (46%) and total protein (51%). Similarly (Heywood, 1981, 1983) identified body weight (82%), food intake (64%), histopathology in general (62%), liver weight (56%), kidney weight (38%), as relevant parameters in 50 subacute, subchronic, and chronic studies. As Travlos et al., they did not distinguish different dose levels. Furthermore they included chronic studies in their analysis, thus the occurrence of tumors may impair their evaluation. Gold et al. (1991) evaluated the cancer potency database and identified that 80% of the carcinogens were positive in at least one of the major eight targets: liver, lung, mammary gland, stomach, vascular system, kidney, hematopoietic system, and urinary bladder. A recent analysis of Martin et al. (2009) analysed 310 chemicals with chronic rat or
M. Batke et al. / Toxicology Letters 218 (2013) 293–298
mouse studies revealing body weight, clinical signs, organ weights and haematology as well as histopathological changes in liver, kidney, thyroid, lung, spleen, and testis as most relevant targets. Over 90% of the chemicals had histopathological findings in these organs. Apart from the chronic studies included in these evaluations, it has to be stated that the chemical domain of Heywood is mostly restricted to pharmaceuticals and the ToxRef DB analysed by Martin and colleagues only contains pesticides. Nevertheless, it can be stated that there is strong support for the observation that depending on study type and chemicals tested a limited number of indicative targets can be identified. Other targets which are rarely affected or affected at higher doses may help to understand toxicological mechanisms but are secondary for the determination of NOEL/LOEL values. The observation of frequent targets is not conclusive as such but has to be extended to the effects occurring at these targets. For the different targets, we combined all effects types referring to organs, i.e. gross pathology, organ weights, histopathology. These are, however three distinct types of observations, that may be present or absent in a toxicological study. Similarly, clinical chemical examinations cover a wide variety of different investigations and it would desirable to identify the individual parameters that are most frequently affected. Furthermore an analysis of effects would need to discuss the adversity and relevance of the single effects as the current analysis is based on LOELs rather than LOAELs. The analysis of effects would be much more specific but even more depending on the chemicals included in the analysis. From the analysis of the single effects also a conclusion on the level of examination needed to detect a reliable LOAEL of the single target would be possible. These are evaluations foreseen with a broader data basis and would be the “coverage within coverage”. However some implications of the frequently observed targets shall be discussed here to show their impact. Body weight is a sensitive parameter for health effects. This has been supported by our analysis. However, as single parameter even at the LOEL + 1 it detects only 44.3% of the LOELs. The importance of body weight on health and healthy development is also obvious in reproduction. Undernutrition during gestation may result in intra-uterine growth retardation and in the following to metabolic disorders like adiposity and diabetes (Holemans et al., 2003; Howie et al., 2011). On the other side, body weight may not be relevant for determining the LOAEL of a study. Body weight changes are related to food intake. If the incorporation of the test substance into the diet reduces the palatability of the diet, this rather has to be considered as a non-adverse effect. Furthermore poor general conditions, e.g. due to local irritation in gastrointestinal or respiratory tract, of the animals may result in reduced food consumption and thus a reduced body weight. In rodents food consumption is directly related to water consumption, thus secondary to reduction of either food or water consumption the other decreases simultaneously. There are different secondary effects due to reduced food and water consumption. Depending on the species and strain used histopathological changes in liver or kidney, changed parameters in hematology or clinical chemistry are described (Kale et al., 2009; Matsuzawa and Sakazume, 1994; Schwartz et al., 1973). Thus the observation of a changed body weight and possible secondary effects has to be interpreted carefully and needs further evaluation within this concept of the coverage approach. Clinical symptoms are another important parameter. They show up at the LOEL in 18.2% of the studies and at the LOEL + 1 in 42% of the cases. Most clinical signs observed during physical examination of individual animals are determined without the aid of instruments. Therefore careful observation of the animals during treatment is of high importance. As specified by OECD (2002), clinical signs may be very sensitive indicators of toxicity. As an example
297
incoordination, muscle twitching, tremor, or diarrhea may indicate acetylcholinesterase inhibition without any morphological changes being evident in nervous tissue. On the other hand, the adversity of some clinical symptoms such as salivation is questionable. Somewhat surprising was the fact that the testes showed up frequently in our analysis. Here, a further in depth analysis is necessary on the specific effects. Increased relative testes weights may be a consequence of decreased body weight, as it is known that weights of testes are not concomitantly reduced with body weight, such as liver or kidney (Feron et al., 1973). As brain that also does not react to body weight, is not one of the major target organs, also other effects may play a role. Histopathology of the testes is a sensitive endpoint in toxicological studies (Mangelsdorf et al., 2003). On the other side, in our (limited) analyses, testes did not show up as single target organ (Table 3), what shows that toxicity to the testes obviously most often is associated with other toxicological effects. Finally, we included spleen in our list of relevant target organs. Although spleen is not frequently affected and does not increase the fraction of detection significantly compared to the 6 relevant organs, it occurs as single organ in 2.3% of the studies. We suspect that this is due to methemoglobinemia. The results of our investigation allow also some more general conclusions: Usually not just one target is affected, but rather more, due to the complex regulation of physiological processes. This becomes evident also from recent studies on toxicity pathways (Martin et al., 2009). Within the data presented here we also started to analyze the interdependencies of targets like clinical chemistry and liver as well as liver and kidney, both at the LOEL (data not shown). A interaction between liver and kidney was shown were as clinical chemistry and liver did not show interaction at the LOEL. This question needs to be followed up by more extensive modeling using a broader database. Results shall be considered in future coverage models. Besides this, there are further limitations in our analysis. First of all, the number of chemicals is limited. However preliminary analyses with other datasets (mouse subchronic, rat subacute) confirm the findings with our guideline-conform dataset of rat subchronic studies (data not shown). However, the scope of examination is more comprehensive for longer study durations as defined by common guidelines like OECD test guidelines (OECD, 1998, 2008). Moreover our data show that the longer a study takes, the more targets are affected overall, but the LOEL is determined by about two organs independent of the study duration. The impact of time and or scope of examination on a study outcome as well as the predictivity of short-term studies has been discussed earlier giving some evidence that the scope of examination is the crucial point rather than the study duration (Batke et al., 2011a; Betton et al., 1994; Kalberlah, 1998). Another limitation refers to the types of structures involved. The RepDose database predominantly covers existing chemicals. More complex molecules may act more specifically at single targets, e.g. due to receptor interactions. Consequently some of the targets not observed beyond the frequent targets or the single targets may be of importance to detect these other modes of toxicity. Therefore it would be interesting to validate and extend the coverage model with other databases on repeated dose toxicity. The application of the coverage model to other databases can either identify targets for more global or more homogeneous sets of chemicals depending on the chemical domain of the respective database. Furthermore these extensions and applications of the coverage model complete the current results and can in addition identify relevant in vivo effects for repeated dose toxicity. In a next step the targets and effects identified by the coverage model specify and prioritize the application and development of alternative in vitro methods.
298
M. Batke et al. / Toxicology Letters 218 (2013) 293–298
Thus, our analyses have to be considered as a first step in a process of getting better insight into outcomes of repeated dose toxicity studies and their evaluation. More detailed analyses in the database RepDose are on the way. The coverage concept can be considered as a transparent and systematic tool to rate the scope of non-guideline studies and thus supports the expert based weight of evidence applied to human risk assessment. The current status of the coverage model is implemented in an integrated testing strategy for the endpoint repeated dose toxicity (ITS RepDose) developed within the FP6 EU-project OSIRIS (osiris.simpple.com). The concept of the ITS RepDose was presented at EPAA Annual Conference 2011 (Batke et al., 2011b) and will be published soon in more detail (Vermeire et al., 2012; Tluczkiewicz et al., 2012). In addition to the evaluation of the study scope, the RepDose ITS also address remaining questions on the quality of study parameters. The documentation of the study is addressed in the tool “knock out criteria”, which assess essential study parameters such as substance identity, route of exposure and tested species. The reliability of the study design is weighted in a tool termed QUANTOS (quality assessment of non-guideline toxicity studies), which address criteria such as the statistical power of the study, the study duration, the administered dose groups and the documentation of statistical methods and controls. Conflict of interest The authors declare that there are no conflicts of interest. Acknowledgement The work was financially supported by the EU-Projekt OSIRIS, contract no. GOCE-CT-2007-037017. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.toxlet. 2012.09.002. References Agresti, A., 2002. Categorical Data Analysis, second ed. Wiley-Interscience, New York. Albert, J., 2009. Bayesian Computation with R, second ed. Springer. Batke, M., Escher, S., Hoffmann-Doerr, S., Melber, C., Messinger, H., Mangelsdorf, I., 2011a. Evaluation of time extrapolation factors based on the database RepDose. Toxicology Letters 205, 122–129. Batke, M., Escher, S.E., Tluczkiewicz, I., Aldenberg, T., Kroese, D.E., Buist, H.E., Mangelsdorf, I., 2011b. ITS RepDose: a functional integrated testing strategy for the endpoint repeated dose toxicity. In: EPAA Annual Conference, Brussels. Betton, G., Cockburn, A., Harpur, E., Hopkins, J., Illing, P., Lumley, C., Connors, T., 1994. A critical review of the optimum duration of chronic rodent testing for the determination of non-tumourigenic toxic potential: a report by the BTS Working Party on Duration of Toxicity Testing. Human & Experimental Toxicology 13, 221–232. Bishop, Y.M.M., Fienberg, S.E., Holland, P.W., 1975. Discrete Multivariate Analysis. Theory and Practice. MIT Press.
Bitsch, A., Jacobi, S., Melber, C., Wahnschaffe, U., Simetska, N., Mangelsdorf, I., 2006. REPDOSE: a database on repeated dose toxicity studies of commercial chemicals—a multifunctional tool. Regulatory Toxicology and Pharmacology: RTP 46, 202–210. Christensen, R., 1997. Log-Linear Models and Logistic Regression. Springer. ECHA, 2010. Guidance on information requirements and chemical safety assessment. Chapter R.8: characterisation of dose [concentration]-response for human health. In: ECHA (Ed.), vol. ECHA-2010-G-19-EN. ECHA. Escher, S.E., Batke, M., Hoffmann-Doerr, S., Messinger, H., Mangelsdorf, I., 2012. Interspecies extrapolation based on experimental data of the RepDose database – a probabilistic approach. Submitted. Feron, V.J., de Groot, A.P., Spanjers, M.T., Til, H.P., 1973. An evaluation of the criterion “organ weight” under conditions of growth retardation. Food and Cosmetics Toxicology 11, 85–94. Fienberg, S., 1980. The Analysis of Cross-Classified Categorial Data, second ed. MIT Press. Gold, L.S., Slone, T.H., Manley, N.B., Bernstein, L., 1991. Target organs in chronic bioassays of 533 chemical carcinogens. Environmental Health Perspectives 93, 233–246. Heywood, R., 1981. Target organ toxicity. Toxicology Letters 8, 349–358. Heywood, R., 1983. Target organ toxicity II. Toxicology Letters 18, 83–88. Holemans, K., Aerts, L., Van Assche, F.A., 2003. Fetal growth restriction and consequences for the offspring in animal models. Journal of the Society for Gynecologic Investigation 10, 392–399. Howie, G.J., Sloboda, D.M., Vickers, M.H., 2011. Maternal undernutrition during critical windows of development results in differential and sex-specific effects on postnatal adiposity and related metabolic profiles in adult rat offspring. The British Journal of Nutrition, 1–10. Kalberlah, F.S.K., 1998. Examination of the data as the basis for the quantification of extrapolation factors. Final report of the research project No. 116 06 113 of the Federal Environmental Agency. vol. Fb 797. Bundesanstalt für Arbeitsschutz und Arbeitsmedizin, Dortmund/Berlin, p. 208. Kale, V.P., Joshi, G.S., Gohil, P.B., Jain, M.R., 2009. Effect of fasting duration on clinical pathology results in Wistar rats. Veterinary Clinical Pathology 38, 361–366. Klimisch, H.J., Andreae, M., Tillmann, U., 1997. A systematic approach for evaluating the quality of experimental toxicological and ecotoxicological data. Regulatory Toxicology and Pharmacology: RTP 25, 1–5. Mangelsdorf, I., Buschmann, J., Orthen, B., 2003. Some aspects relating to the evaluation of the effects of chemicals on male fertility. Regulatory Toxicology and Pharmacology: RTP 37, 356–369. Martin, M.T., Judson, R.S., Reif, D.M., Kavlock, R.J., Dix, D.J., 2009. Profiling chemicals based on chronic toxicity results from the US EPA ToxRef Database. Environmental Health Perspectives 117, 392–399. Matsuzawa, T., Sakazume, M., 1994. Effects of fasting on hematology and clinicalchemistry values in the rat and dog. Comparative Haematology International 4, 152–156. OECD, 1998. Test No. 408: Repeated Dose 90-Day Oral Toxicity Study in Rodents. OECD Publishing. OECD, 2002. Guidance Notes for Analysis and Evaluation of Repeat-Dose Toxicity Studies. OECD Publishing. OECD, 2008. Test No. 407: Repeated Dose 28-day Oral Toxicity Study in Rodents. OECD Publishing. Powers, D.A., Xie, Y., 1999. Statistical Methods for Categorical Data Analysis. Academic Press, p. 295. Schwartz, E., Tornaben, J.A., Boxill, G.C., 1973. The effects of food restriction on hematology, clinical chemistry and pathology in the albino rat. Toxicology and Applied Pharmacology 25, 515–524. Tluczkiewicz, I., Batke, M., Kroese, D., Buist, H., Aldenberg, T., Paune, E., Grimm, H., Kühne, R., Schüürmann, G., Mangelsdorf, I., Escher, S. E., 2012. The OSIRIS Weight of Evidence approach: ITS for the endpoint repeated-dose toxicity (RepDose ITS). Submitted. Travlos, G.S., Morris, R.W., Elwell, M.R., Duke, A., Rosenblum, S., Thompson, M.B., 1996. Frequency and relationships of clinical chemistry and liver and kidney histopathology findings in 13-week toxicity studies in rats. Toxicology 107, 17–29. Vermeire, T., Aldenberg, T., Buist, H. E., Escher, S. E., Mangelsdorf, I., Pauné, E., Rorije, E., Kroese, E. D., 2012. OSIRIS, a quest for proof op principle for integrated testing strategies of chemicals for four human health endpoints. Submitted.
Contents lists available at SciVerse ScienceDirect
Toxicology Letters journal homepage: www.elsevier.com/locate/toxlet
Relevance of non-guideline studies for risk assessment: The coverage model based on most frequent targets in repeated dose toxicity studies M. Batke a,∗ , T. Aldenberg b , S. Escher a , I. Mangelsdorf a a b
Fraunhofer ITEM, Nikolai-Fuchs-Str. 1, 30625 Hannover, Germany RIVM-Laboratory for Ecological Risk Assessment, P.O. Box 1, Anthonie van Leeuwenhoeklaan 9, 3720 BA Bilthoven, The Netherlands
h i g h l i g h t s
Coverage approach: evaluation of scope of examination of old repeated dose toxicity studies. Statistical model calculates probability that LOEL or the next higher dose (LOEL + 1) is determined. Evaluation based on Fraunhofer RepDose DB. Statistical model can be transferred to other datasets.
a r t i c l e
i n f o
Article history: Received 31 July 2012 Received in revised form 31 August 2012 Accepted 3 September 2012 Available online 10 September 2012 Keywords: Extrapolation factors Repeated dose toxicity Integrated testing strategy Study quality OECD guideline
a b s t r a c t A common challenge for human risk assessment is the quality of the available animal studies. Nonguideline studies are often limited for different aspects of study design and documentation. Within this publication the relevance of a limited scope of examination is discussed. Preliminary analyses of the RepDose database have shown that liver, body weight, kidney and clinical symptoms are frequently affected in oral repeated dose toxicity in rats and mice (Bitsch et al., 2006), while many other targets are seldom affected. As most of these targets are investigated frequently also in non-guideline studies, it is likely that they provide a reliable NOEL, although the full spectrum of endpoints has not been covered. Based on RepDose data we investigate the relevance of individual targets for determining the LOEL and the consequences for risk assessment. The resulting coverage model for subchronic oral rat studies includes up to six targets and an additional assessment factor for LOEL extrapolation. It can be applied to assess the reliability of non-guideline studies with respect to the scope of examination. Furthermore the application of the coverage model to other databases will increase and/or specify the chemical domain and reveal respective targets as well as effects. © 2012 Elsevier Ireland Ltd. All rights reserved.
1. Introduction In the evaluation of studies on existing chemicals, the risk assessor often faces the situation that repeated dose toxicity studies are available that are rather old, i.e. performed before the implementation of guidelines, or were not conducted according to guidelines. For these studies, less than three dose levels may be available or a low number of animals/dose group. In addition the scope of investigations may be reduced with regard to the number of organs investigated by histopathology. Further, organ weights,
∗ Corresponding author at: Institut für Toxikologie und experimentelle Medizin, Abteilung Chemikalienbewertung, Nikolai-Fuchs-Str. 1, 30625 Hannover, Germany. Tel.: +49 0511 5350 344; fax: +49 0511 5350 335. E-mail address: [email protected] (M. Batke). 0378-4274/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.toxlet.2012.09.002
clinical chemical examinations, hematology or urine analysis may not be available. In the REACH program, the quality of a study is determined by the Klimisch Code (Klimisch et al., 1997): Depending on the scope of examinations and quality of documentation studies are “reliable” (Klimisch Code 1), “reliable with restriction” (Klimisch Code 2) or “not reliable” (Klimisch Code 3). Many of the old, non-guideline studies can be assigned to reliability 2 or even 3. In the case of Klimisch Code 3, the study is usually considered as not sufficient for evaluation and new animal testing is necessary. However, with the aim of saving experimental animals, it is desirable to use existing studies in risk assessment as far as possible. At Fraunhofer ITEM a database on repeated dose toxicity studies (RepDose) has been developed, that contains currently more than 1100 oral repeated dose toxicity studies, mainly on existing chemicals. These studies include subacute, subchronic and chronic studies with rats and mice with reliability 1 and 2, and
294
M. Batke et al. / Toxicology Letters 218 (2013) 293–298
Table 1 Datasets selected from RepDose for further analyses. Study duration
Study type
Subacute Subchronic Subchronic
Comprehensive Comprehensive Old
2.2. Statistics The probability that a study detects the LOEL, or the LOEL + 1, when only a limited number of targets has been investigated in a study, can be calculated by methods of categorical data analysis, in particular loglinear modelling (Agresti, 2002; Bishop et al., 1975; Christensen, 1997; Fienberg, 1980; Powers and Xie, 1999). We aimed to identify the most suitable and generally usable model, avoiding overfitting. Models that overfit will predict badly in new situations: they have limited generalizability. We have found the independent organ model to be generally acceptable to calculate the probability that the LOEL or the LOEL + 1 were detected. The model selection is documented in supplementary material. The multinomial linear model without interaction terms can be interpreted as the independent organ model. For example, at the LOEL, the fraction of studies with particular organ combinations affected can be modelled by multinomial probability parameters ϕ which are proportional to the exponential function of a linear expression of the target organs. Suppose L, K, C, and B are binary values (0 or 1) of the target organs: liver, kidney, clinical chemistry, and body weight. A value of 0 means not affected, or not reported to be affected, while a 1 means affected. Then probabilities ϕ are modelled as:
No. of studies Rats
Mice
52 88 56
≤10 45 ≤10
also some studies with reliability 3. Preliminary analyses of the overall database have shown that liver, body weight, kidney and clinical symptoms are frequently affected in oral repeated dose toxicity in rats and mice (Bitsch et al., 2006), while many other organs are seldom affected. In inhalation studies the respiratory tract and especially the nose and lung are major targets in addition. As most of these targets are investigated frequently also in nonguideline studies, it is likely that non-guideline studies provide a reliable NOEL, although the full spectrum of endpoints has not been covered in the respective study. Based on RepDose data, in this publication we investigate the relevance of individual targets for determining the LOEL and the consequences for risk assessment. A method is developed to calculate the probability that a certain set of targets provides a reliable reference value. A model consisting of up to six targets is proposed for subchronic oral rat studies.
∝ exp(ˇ1 · L + ˇ2 · K + ˇ3 · C + ˇ4 · B). The exponential function factorises into terms to be multiplied:
∝ exp(ˇ1 · L + ˇ2 · K + ˇ3 · C + ˇ4 · B) = exp(ˇ1 · L) · exp(ˇ2 · K) · exp(ˇ3 · C) · exp(ˇ4 · B) = L · K · C · B
Hence, the probability of a study with a certain organ pattern is the product of the probabilities of studies with each individual organ affected. The ˇ-coefficients are estimated by the method of maximum likelihood. Given the data in Table 3 the coverage at the LOEL of a study with liver, kidney and heart examined would be:
2. Materials and methods
Coverage = 1 − (1 − 0.318) × (1 − 0.239) × (1 − 0.057) = 0.51 2.1. Analyses with RepDose RepDose contains effects for different dose levels and general information on the study design as well as a reliability score. The reliability score assesses the study quality with respect to general design and scope of examination. Based on the reliability and the publication date two types of studies were discriminated: studies with “guideline-conform”, OECD-guideline-like design and scope as well as “non-guideline” studies performed before the implementation of OECD guidelines in 1981. Additionally, the following criteria were applied to select the datasets from the RepDose DB:
-
2.3. Simulation of coverage uncertainty Coverage uncertainty was simulated through Bayesian Monte-Carlo sampling of the linear multinomial models, cf. (Albert, 2009). The details are explicated in supplementary material. To sketch the procedure, note that the individual target fractions in Table 3 are estimated from observed fractions at the LOEL, and at LOEL + 1. Thus, the fraction 31.8% of studies showing Liver to be affected at the LOEL, as employed in the above three-organ coverage equation, is estimated from the ratio of studies with the organ affected to the total number of studies: 28/88 = 31.8%, However, this is a so-called point estimate, without any uncertainty addressed. Through Bayesian simulation of the model likelihoods, the uncertainty around these point estimates can be evaluated. This yields the uncertainty of the ˇcoefficients, as well as the estimated fractions of studies with a particular organ affected, and therefore of the individual organ and total coverage. We found that in each organ model, the median of the simulated coverage distributions was very close to the point estimates.
Oral studies (gavage studies, application via drinking water or food). Rats and mice. Subacute and subchronic studies. LOEL available. In addition NOEL, and LOEL + 1 (dose above LOEL) available (only guidelineconformstudies). If a NOEL was not available, also studies with only LOEL and LOEL+1 were included. Conditions were: Only one organ affected at the LOEL or mild effects at the LOEL
3. Results The datasets shown in Table 1 were derived. Datasets with 10 or less studies were not considered for further evaluation. Most analyses were performed with the guideline-conformdataset on rat subchronic studies. Studies were analysed for the targets at their LOEL and at the LOEL + 1. Targets for effects are classical organs (histopathology or organ weight) in toxicological studies, i.e. liver, kidney, brain, etc. In addition, body weight, clinical symptoms, clinical chemistry and haematology are termed targets in our analysis. In the RepDose database no distinction is made between LOELs and LOAELs. Any effect that is statistically different from the control is termed LOEL. Thus, effects that may be considered as non-adverse are also covered in our analyses. The references for the selected studies are given in supplementary material.
3.1. Number of target organs Table 2 shows the number of organs affected in subacute or subchronic studies with rats or mice, at the LOEL, the dose level above the LOEL (LOEL + 1) or for all dose levels investigated. At the LOEL on average more than one target organ is affected in all studies types. With increasing dose levels the average number of targets increases from 1.6 to 4.2 for subacute rat studies and 1.8
Table 2 Number of targets affected for different study types. Species
Rat Rat Rat Mouse a b
Study duration
Subacute Subchronic Subchronic Subchronic
Characterization
Comprehensive Comprehensive Old Comprehensive
n
52 88 56 45
Geometric means and the respective geometric standard deviation in brackets. This figure refers only to 41 of the 56 old oral rat studies which have a LOEL + 1 dose.
No. of targetsa LOEL
LOEL+1
Overall study
1.6 (1.8) 1.8 (1.8) 1.8 (2.0) 1.7 (1.7)
3.7 (2.0) 3.4 (2.0) 3.0 (2.0)b 2.3 (1.9)
4.2 (2.0) 6.5 (1.7) 3.6 (2.3) 4.2 (1.9)
M. Batke et al. / Toxicology Letters 218 (2013) 293–298 Table 3 Fractions of targets affected in old (n = 56) and comprehensive subchronic oral rat studies (n = 88). Fractions are given at the LOEL (old and comprehensive studies) as well as at the LOEL + 1 and for LOELs determined by a single target (comprehensive studies). Target
Clinical chemistry Liver Haematology Kidney Body weight Clinical symptoms Testes Urine analysis Forestomach Heart Spleen Brain Stomach Intestine Adrenal gland Thyroid gland Thymus Lymph node Bladder Endocrine system Nose Pancreas Vascular system Salivary gland Pituitary gland Nervous system Ovary Bone Gallbladder/bileduct Lung Uterus Prostate Seminal vesicle Skeletal muscle Adipose tissue Eye
Fraction (% of studies) Old
Comprehensive
LOEL
LOEL
LOEL + 1
LOEL single target
17.9 44.6 16.1 37.5 37.5 17.9 10.7 17.9
38.6 31.8 29.5 23.9 21.6 18.2 8.0 6.8 6.8a 5.7 4.5 2.3 2.3 2.3 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1
56.8 55.7 46.6 40.9 44.3 42.0 17.0 13.6 9.1a 13.6 15.9 8.0 5.7 3.4 8.0 4.5 5.7 3.4 2.3 2.3 2.3 2.3 1.1 1.1
9.1 5.7 11.4 4.5 2.3 1.1
3.6 7.1 5.4 1.8 5.4 5.4
1.8 1.8 1.8 1.8
1.1 3.4a 1.1 2.3
1.1 1.1
295
at the LOEL + 1 are provided additionally. These data are used to estimate whether the scope of a non-guideline study is sufficient to detect a reliable LOEL or the dose above the LOEL (LOEL + 1). In a subset of studies the LOEL is determined by only one target. The fractions of these “single” targets at the LOEL provide additional information on their relevance (Table 3). In non-guideline studies, most frequently liver, body weight and kidney are affected, followed by clinical chemistry, haematology and clinical symptoms. Although in guideline-conform studies the scope of examination is broader, the same targets show up as frequent targets, although with a somewhat different fraction. Most frequently, i.e. in 38.6% of all studies clinical chemical parameters are changed at the LOEL of a study, followed by liver, haematology, kidney, body weight and clinical symptoms. At the dose level above the LOEL (LOEL + 1) again the same targets appear, the fraction of the targets being affected is considerably increased/about doubled. On the other side, there are some targets like the pancreas, the salivary gland or the nose, that are seldom affected at the LOEL of a study in non-guideline studies, at the LOEL and LOEL + 1 in guideline-conform studies. Out of the 88 guideline-conform oral subchronic rat studies 40 studies (45%) have only a single target at the LOEL. In addition to five of the frequent targets (namely haematology, clinical chemistry, liver, kidney, body weight) forestomach and spleen occur with higher fraction (>2%) alone at the LOEL. In those studies, where the LOEL is triggered only by the forestomach, the route of administration is always gavage. Thus, not only the same organs are occurring frequently at the LOEL, furthermore they are also the targets occurring most frequently as sole targets at the LOEL.
1.1
2.3 2.3 2.3 2.3 1.1 1.1 1.1 1.1 1.1 1.1
LOEL + 1: Dose level above LOEL; LOEL single: proportion of the target affected at the LOEL without any other target. Dark grey: Key targets, light grey: other relevant targets. a Only occurring in gavage studies.
to 6.5 for subchronic rat studies, reflecting more toxic effects at higher dose levels and a significantly increased number of effects overtime (p < 0.001, Mann–Whitney–Wilcoxon-Test). In contrast to the number of targets at the LOEL, the number of overall targets at the study is significantly lower for non-guideline compared to guideline-conform subchronic rat studies. Furthermore the number of targets in subchronic rat studies is significantly higher than in subchronic mouse studies. An interesting finding is that already at the LOEL independent from the study type and duration frequently more than one target is affected. In these cases one of the targets could substitute the other for identifying the LOEL. Further, based on our previous analyses we hypothesize that there will be some major target organs. Therefore we analysed the fraction of the target organs at the LOEL. 3.2. Target organs at the LOEL of guideline-conform and non-guideline studies Table 3 shows the fraction of different targets affected at the LOEL in non-guideline and in guideline-conform rat subchronic oral toxicity studies. For guideline-conform studies the target fractions
3.3. Coverage model So far, we described the distinct fractions of targets affected. The fractions observed for guideline-conform studies can be transformed to probabilities as the targets are within the scope of examination of guideline-conform studies. The statistical method developed to calculate the probability to detect a reference value (LOEL or LOEL + 1) refers not only to single targets but to combinations of targets (Section 2.2). The probability to predict a reference value can also be interpreted as the coverage of the targets included in the model. The following most frequent targets were included in our analysis: Liver, kidney, clinical chemistry, body weight, clinical symptoms, haematology, spleen, testes. The probabilities for eight out of 255 possible coverage models for combined targets and their attributed uncertainties are shown in Table 4. The LOEL is predicted with a probability of 87% if seven (model 7) or eight targets (model 8) are combined. Model 5 with six targets performed nearly as well with 86% coverage, indicating that an inclusion of more than six targets does not increase the probability to detect the LOEL significantly. At the LOEL + 1 the same model improves to 98% probability. However, already coverage model 1 with the four targets liver, kidney, clinical chemistry, and body weight achieved 94% probability for LOEL + 1 prediction. Based on the derived probabilities, coverage above 90 or 95% of the LOEL + 1 is possible with different target combinations. The coverage of the LOEL is not higher than 87% with one of the seven or eight target models. An extrapolation factor LOEL + 1 to LOEL is then required corresponding, e.g. to the dose spacing observed in toxicological studies. Therefore we investigated the dose spacing of the 88 guideline-conform rat studies. Fig. 1 shows the log-normally distributed function (proven by Chi-Square, p 0.05, data not shown) of the LOEL + 1/LOEL ratios. Statistical evaluation gives a geometric mean of 2.6 with a geometric standard deviation of 1.5 and a 95th percentile of 5. The LOEL of a non-guideline study, which meets the
296
M. Batke et al. / Toxicology Letters 218 (2013) 293–298
Table 4 Selected target combinations and their coverage at the LOEL and at the LOEL + 1. Model no.
1 2 3 4 5 6 7 8
Targets
LOEL
Liver
Kidney
Clinical chemistry
Body weight
Clinical symptoms
X X X X X X X X
X X X X X X X X
X
X X X X X X X X
X X X X X X X
X X X X X
Haema tology
X X X X X
Spleen
Testes
X X
X X
Coverage 75% 67% 80% 77% 86% 86% 87% 87%
LOEL + 1 Uncertainty
Coverage
Uncertainty
67%, 82% 58%, 75% 73%, 86% 69%, 83% 80%, 90% 81%, 91% 81%, 91% 82%, 92%
94% 92% 96% 96% 98% 98% 98% 99%
91%, 96% 88%, 95% 94%, 98% 93%, 97% 97%, 99% 97%, 99% 97%, 99% 98%, 99%
X: target included in model; coverage: probability to detect the reference value if the respective targets are included in the statistical model (independent model/linear model). The point estimate of the coverage calculated by the best-fitting linear model is practically equal to the 50% percentile of the MCMC simulations for the uncertainty estimation; uncertainty: 2.5 and 97.5% percentile of the posterior coverage distribution determined by MCMC simulation.
LOEL + 1 in the 6 main target model, can thus be extrapolated with a factor of 2.6–5 to yield a reliable LOEL. 4. Discussion The aim of this publication was to develop a strategy (coverage approach) to assess the reliability of the LOEL of a non-guideline repeated dose toxicity study with respect to the scope of examination. Based on the RepDose DB it is shown that a limited set of six targets consisting of liver, kidney, clinical chemistry, body weight, clinical symptoms and hematology is frequently affected at the LOEL of a guideline-conform subchronic rat study. Their examination within a study gives a probability of 86% to detect the LOEL (see model 5, Table 4). For the next higher dose level the overall probability that these targets are sufficient for proper detection is as high as 98%. If additional targets are considered, the probability for the LOEL increases marginally and for the LOEL + 1 it stays the same. The coverage model including six targets is thus sufficient based on our dataset. The LOEL of studies with scope of examinations limited to these six targets can be considered to be equal to the LOEL + 1 of a similar guideline-conform study with a probability of 98%. Our analyses demonstrates that the LOEL + 1 is higher by a factor of 2.6 (geometric mean) to 5 (95th percentile) than the LOEL. This gives an indication for the value of an assessment factor to be used in risk assessment. Assessment factors of up to 10 for the LOEL to NOEL extrapolation as indicated in the REACH guidance R8 are supported by our analyses of dose spacing (ECHA, 2010). The
Fig. 1. Distribution of LOEL + 1/LOEL-ratios for all 88 comprehensive subchronic rat studies.
extrapolation from LOEL + 1 to LOEL in the context of the coverage model accounts for the quality of the respective study. Thus our data also support the use of assessment factors > 1 for data quality as proposed by the REACH Guidance Document R8 (ECHA, 2010). Additional aspects of the data quality are discussed in more detail in the integrated testing strategy (see below, Tluczkiewicz et al., 2012). Further the probability of identifying the LOEL increases, if more than one study of limited quality but different scope of examination is available. It may be doubtful that in non-guideline studies the six targets of model 5 are analysed. The fractions of targets in non-guideline studies (rat oral subchronic) as shown in Table 3, however, indicate that the frequent targets observed in guideline-conform repeated dose toxicity studies (rat oral subchronic) are frequently observed as well. This observation gives confidence in the applicability of the coverage approach. These results indicate that some non-guideline Klimisch code 2–3 studies, with reduced scope compared to guideline-conform studies, do still provide reliable information. Their consideration in risk assessment may prevent additional animal studies. It has been shown that the number of targets is higher in rat studies than in mouse studies. This supports the use of rats as first choice for a species to be investigated in toxicological studies, i.e. detecting effects reliably. The sensitivity of rats and mice is, however not principally different, as has been show in our most recent analyses on interspecies differences (Escher et al., 2012). Our findings are supported by other publications addressing the importance of targets (organs). Travlos et al. (1996) identified histopathological lesions in the liver in 20 of 60 and the kidney in 25 of 61 rat subchronic toxicity studies. However they did not restrict their analyses to the LOEL as we did. Nevertheless also their results support the high abundance of liver and kidney effects. Further they showed that also clinical chemical parameters are increased after 13 weeks in a great proportion of studies. As indicators of liver toxicity alanine aminotransferase (32%), sorbitol dehydrogenase (29%) and total bile acids (43%) are affected. Kidney effects were attributed to urea nitrogen (26%), creatinine (18%), albumin (46%) and total protein (51%). Similarly (Heywood, 1981, 1983) identified body weight (82%), food intake (64%), histopathology in general (62%), liver weight (56%), kidney weight (38%), as relevant parameters in 50 subacute, subchronic, and chronic studies. As Travlos et al., they did not distinguish different dose levels. Furthermore they included chronic studies in their analysis, thus the occurrence of tumors may impair their evaluation. Gold et al. (1991) evaluated the cancer potency database and identified that 80% of the carcinogens were positive in at least one of the major eight targets: liver, lung, mammary gland, stomach, vascular system, kidney, hematopoietic system, and urinary bladder. A recent analysis of Martin et al. (2009) analysed 310 chemicals with chronic rat or
M. Batke et al. / Toxicology Letters 218 (2013) 293–298
mouse studies revealing body weight, clinical signs, organ weights and haematology as well as histopathological changes in liver, kidney, thyroid, lung, spleen, and testis as most relevant targets. Over 90% of the chemicals had histopathological findings in these organs. Apart from the chronic studies included in these evaluations, it has to be stated that the chemical domain of Heywood is mostly restricted to pharmaceuticals and the ToxRef DB analysed by Martin and colleagues only contains pesticides. Nevertheless, it can be stated that there is strong support for the observation that depending on study type and chemicals tested a limited number of indicative targets can be identified. Other targets which are rarely affected or affected at higher doses may help to understand toxicological mechanisms but are secondary for the determination of NOEL/LOEL values. The observation of frequent targets is not conclusive as such but has to be extended to the effects occurring at these targets. For the different targets, we combined all effects types referring to organs, i.e. gross pathology, organ weights, histopathology. These are, however three distinct types of observations, that may be present or absent in a toxicological study. Similarly, clinical chemical examinations cover a wide variety of different investigations and it would desirable to identify the individual parameters that are most frequently affected. Furthermore an analysis of effects would need to discuss the adversity and relevance of the single effects as the current analysis is based on LOELs rather than LOAELs. The analysis of effects would be much more specific but even more depending on the chemicals included in the analysis. From the analysis of the single effects also a conclusion on the level of examination needed to detect a reliable LOAEL of the single target would be possible. These are evaluations foreseen with a broader data basis and would be the “coverage within coverage”. However some implications of the frequently observed targets shall be discussed here to show their impact. Body weight is a sensitive parameter for health effects. This has been supported by our analysis. However, as single parameter even at the LOEL + 1 it detects only 44.3% of the LOELs. The importance of body weight on health and healthy development is also obvious in reproduction. Undernutrition during gestation may result in intra-uterine growth retardation and in the following to metabolic disorders like adiposity and diabetes (Holemans et al., 2003; Howie et al., 2011). On the other side, body weight may not be relevant for determining the LOAEL of a study. Body weight changes are related to food intake. If the incorporation of the test substance into the diet reduces the palatability of the diet, this rather has to be considered as a non-adverse effect. Furthermore poor general conditions, e.g. due to local irritation in gastrointestinal or respiratory tract, of the animals may result in reduced food consumption and thus a reduced body weight. In rodents food consumption is directly related to water consumption, thus secondary to reduction of either food or water consumption the other decreases simultaneously. There are different secondary effects due to reduced food and water consumption. Depending on the species and strain used histopathological changes in liver or kidney, changed parameters in hematology or clinical chemistry are described (Kale et al., 2009; Matsuzawa and Sakazume, 1994; Schwartz et al., 1973). Thus the observation of a changed body weight and possible secondary effects has to be interpreted carefully and needs further evaluation within this concept of the coverage approach. Clinical symptoms are another important parameter. They show up at the LOEL in 18.2% of the studies and at the LOEL + 1 in 42% of the cases. Most clinical signs observed during physical examination of individual animals are determined without the aid of instruments. Therefore careful observation of the animals during treatment is of high importance. As specified by OECD (2002), clinical signs may be very sensitive indicators of toxicity. As an example
297
incoordination, muscle twitching, tremor, or diarrhea may indicate acetylcholinesterase inhibition without any morphological changes being evident in nervous tissue. On the other hand, the adversity of some clinical symptoms such as salivation is questionable. Somewhat surprising was the fact that the testes showed up frequently in our analysis. Here, a further in depth analysis is necessary on the specific effects. Increased relative testes weights may be a consequence of decreased body weight, as it is known that weights of testes are not concomitantly reduced with body weight, such as liver or kidney (Feron et al., 1973). As brain that also does not react to body weight, is not one of the major target organs, also other effects may play a role. Histopathology of the testes is a sensitive endpoint in toxicological studies (Mangelsdorf et al., 2003). On the other side, in our (limited) analyses, testes did not show up as single target organ (Table 3), what shows that toxicity to the testes obviously most often is associated with other toxicological effects. Finally, we included spleen in our list of relevant target organs. Although spleen is not frequently affected and does not increase the fraction of detection significantly compared to the 6 relevant organs, it occurs as single organ in 2.3% of the studies. We suspect that this is due to methemoglobinemia. The results of our investigation allow also some more general conclusions: Usually not just one target is affected, but rather more, due to the complex regulation of physiological processes. This becomes evident also from recent studies on toxicity pathways (Martin et al., 2009). Within the data presented here we also started to analyze the interdependencies of targets like clinical chemistry and liver as well as liver and kidney, both at the LOEL (data not shown). A interaction between liver and kidney was shown were as clinical chemistry and liver did not show interaction at the LOEL. This question needs to be followed up by more extensive modeling using a broader database. Results shall be considered in future coverage models. Besides this, there are further limitations in our analysis. First of all, the number of chemicals is limited. However preliminary analyses with other datasets (mouse subchronic, rat subacute) confirm the findings with our guideline-conform dataset of rat subchronic studies (data not shown). However, the scope of examination is more comprehensive for longer study durations as defined by common guidelines like OECD test guidelines (OECD, 1998, 2008). Moreover our data show that the longer a study takes, the more targets are affected overall, but the LOEL is determined by about two organs independent of the study duration. The impact of time and or scope of examination on a study outcome as well as the predictivity of short-term studies has been discussed earlier giving some evidence that the scope of examination is the crucial point rather than the study duration (Batke et al., 2011a; Betton et al., 1994; Kalberlah, 1998). Another limitation refers to the types of structures involved. The RepDose database predominantly covers existing chemicals. More complex molecules may act more specifically at single targets, e.g. due to receptor interactions. Consequently some of the targets not observed beyond the frequent targets or the single targets may be of importance to detect these other modes of toxicity. Therefore it would be interesting to validate and extend the coverage model with other databases on repeated dose toxicity. The application of the coverage model to other databases can either identify targets for more global or more homogeneous sets of chemicals depending on the chemical domain of the respective database. Furthermore these extensions and applications of the coverage model complete the current results and can in addition identify relevant in vivo effects for repeated dose toxicity. In a next step the targets and effects identified by the coverage model specify and prioritize the application and development of alternative in vitro methods.
298
M. Batke et al. / Toxicology Letters 218 (2013) 293–298
Thus, our analyses have to be considered as a first step in a process of getting better insight into outcomes of repeated dose toxicity studies and their evaluation. More detailed analyses in the database RepDose are on the way. The coverage concept can be considered as a transparent and systematic tool to rate the scope of non-guideline studies and thus supports the expert based weight of evidence applied to human risk assessment. The current status of the coverage model is implemented in an integrated testing strategy for the endpoint repeated dose toxicity (ITS RepDose) developed within the FP6 EU-project OSIRIS (osiris.simpple.com). The concept of the ITS RepDose was presented at EPAA Annual Conference 2011 (Batke et al., 2011b) and will be published soon in more detail (Vermeire et al., 2012; Tluczkiewicz et al., 2012). In addition to the evaluation of the study scope, the RepDose ITS also address remaining questions on the quality of study parameters. The documentation of the study is addressed in the tool “knock out criteria”, which assess essential study parameters such as substance identity, route of exposure and tested species. The reliability of the study design is weighted in a tool termed QUANTOS (quality assessment of non-guideline toxicity studies), which address criteria such as the statistical power of the study, the study duration, the administered dose groups and the documentation of statistical methods and controls. Conflict of interest The authors declare that there are no conflicts of interest. Acknowledgement The work was financially supported by the EU-Projekt OSIRIS, contract no. GOCE-CT-2007-037017. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.toxlet. 2012.09.002. References Agresti, A., 2002. Categorical Data Analysis, second ed. Wiley-Interscience, New York. Albert, J., 2009. Bayesian Computation with R, second ed. Springer. Batke, M., Escher, S., Hoffmann-Doerr, S., Melber, C., Messinger, H., Mangelsdorf, I., 2011a. Evaluation of time extrapolation factors based on the database RepDose. Toxicology Letters 205, 122–129. Batke, M., Escher, S.E., Tluczkiewicz, I., Aldenberg, T., Kroese, D.E., Buist, H.E., Mangelsdorf, I., 2011b. ITS RepDose: a functional integrated testing strategy for the endpoint repeated dose toxicity. In: EPAA Annual Conference, Brussels. Betton, G., Cockburn, A., Harpur, E., Hopkins, J., Illing, P., Lumley, C., Connors, T., 1994. A critical review of the optimum duration of chronic rodent testing for the determination of non-tumourigenic toxic potential: a report by the BTS Working Party on Duration of Toxicity Testing. Human & Experimental Toxicology 13, 221–232. Bishop, Y.M.M., Fienberg, S.E., Holland, P.W., 1975. Discrete Multivariate Analysis. Theory and Practice. MIT Press.
Bitsch, A., Jacobi, S., Melber, C., Wahnschaffe, U., Simetska, N., Mangelsdorf, I., 2006. REPDOSE: a database on repeated dose toxicity studies of commercial chemicals—a multifunctional tool. Regulatory Toxicology and Pharmacology: RTP 46, 202–210. Christensen, R., 1997. Log-Linear Models and Logistic Regression. Springer. ECHA, 2010. Guidance on information requirements and chemical safety assessment. Chapter R.8: characterisation of dose [concentration]-response for human health. In: ECHA (Ed.), vol. ECHA-2010-G-19-EN. ECHA. Escher, S.E., Batke, M., Hoffmann-Doerr, S., Messinger, H., Mangelsdorf, I., 2012. Interspecies extrapolation based on experimental data of the RepDose database – a probabilistic approach. Submitted. Feron, V.J., de Groot, A.P., Spanjers, M.T., Til, H.P., 1973. An evaluation of the criterion “organ weight” under conditions of growth retardation. Food and Cosmetics Toxicology 11, 85–94. Fienberg, S., 1980. The Analysis of Cross-Classified Categorial Data, second ed. MIT Press. Gold, L.S., Slone, T.H., Manley, N.B., Bernstein, L., 1991. Target organs in chronic bioassays of 533 chemical carcinogens. Environmental Health Perspectives 93, 233–246. Heywood, R., 1981. Target organ toxicity. Toxicology Letters 8, 349–358. Heywood, R., 1983. Target organ toxicity II. Toxicology Letters 18, 83–88. Holemans, K., Aerts, L., Van Assche, F.A., 2003. Fetal growth restriction and consequences for the offspring in animal models. Journal of the Society for Gynecologic Investigation 10, 392–399. Howie, G.J., Sloboda, D.M., Vickers, M.H., 2011. Maternal undernutrition during critical windows of development results in differential and sex-specific effects on postnatal adiposity and related metabolic profiles in adult rat offspring. The British Journal of Nutrition, 1–10. Kalberlah, F.S.K., 1998. Examination of the data as the basis for the quantification of extrapolation factors. Final report of the research project No. 116 06 113 of the Federal Environmental Agency. vol. Fb 797. Bundesanstalt für Arbeitsschutz und Arbeitsmedizin, Dortmund/Berlin, p. 208. Kale, V.P., Joshi, G.S., Gohil, P.B., Jain, M.R., 2009. Effect of fasting duration on clinical pathology results in Wistar rats. Veterinary Clinical Pathology 38, 361–366. Klimisch, H.J., Andreae, M., Tillmann, U., 1997. A systematic approach for evaluating the quality of experimental toxicological and ecotoxicological data. Regulatory Toxicology and Pharmacology: RTP 25, 1–5. Mangelsdorf, I., Buschmann, J., Orthen, B., 2003. Some aspects relating to the evaluation of the effects of chemicals on male fertility. Regulatory Toxicology and Pharmacology: RTP 37, 356–369. Martin, M.T., Judson, R.S., Reif, D.M., Kavlock, R.J., Dix, D.J., 2009. Profiling chemicals based on chronic toxicity results from the US EPA ToxRef Database. Environmental Health Perspectives 117, 392–399. Matsuzawa, T., Sakazume, M., 1994. Effects of fasting on hematology and clinicalchemistry values in the rat and dog. Comparative Haematology International 4, 152–156. OECD, 1998. Test No. 408: Repeated Dose 90-Day Oral Toxicity Study in Rodents. OECD Publishing. OECD, 2002. Guidance Notes for Analysis and Evaluation of Repeat-Dose Toxicity Studies. OECD Publishing. OECD, 2008. Test No. 407: Repeated Dose 28-day Oral Toxicity Study in Rodents. OECD Publishing. Powers, D.A., Xie, Y., 1999. Statistical Methods for Categorical Data Analysis. Academic Press, p. 295. Schwartz, E., Tornaben, J.A., Boxill, G.C., 1973. The effects of food restriction on hematology, clinical chemistry and pathology in the albino rat. Toxicology and Applied Pharmacology 25, 515–524. Tluczkiewicz, I., Batke, M., Kroese, D., Buist, H., Aldenberg, T., Paune, E., Grimm, H., Kühne, R., Schüürmann, G., Mangelsdorf, I., Escher, S. E., 2012. The OSIRIS Weight of Evidence approach: ITS for the endpoint repeated-dose toxicity (RepDose ITS). Submitted. Travlos, G.S., Morris, R.W., Elwell, M.R., Duke, A., Rosenblum, S., Thompson, M.B., 1996. Frequency and relationships of clinical chemistry and liver and kidney histopathology findings in 13-week toxicity studies in rats. Toxicology 107, 17–29. Vermeire, T., Aldenberg, T., Buist, H. E., Escher, S. E., Mangelsdorf, I., Pauné, E., Rorije, E., Kroese, E. D., 2012. OSIRIS, a quest for proof op principle for integrated testing strategies of chemicals for four human health endpoints. Submitted.