EXTERNAL SCIENTIFIC REPORT

Transcription

1 EXTERNAL SCIENTIFIC REPORT APPROVED: 24 May 2017 doi: /sp.efsa.2017.en-1251 Biological relevance of the magnitude of effects (considering mortality, sub-lethal and reproductive effects) observed in studies with amphibians and reptiles in view of population level impacts on amphibians and reptiles Manuel E. Ortiz-Santaliestra 1, Joao P. Maia 2, Andrés Egea-Serrano 1, Carsten A. Brühl 3, Isabel Lopes 2 1 Instituto de Investigación en Recursos Cinegéticos (IREC) UCLM-CSIC-JCCM. Ronda de Toledo 12, Ciudad Real, Spain 2 Departamento de Biologia e CESAM, Universidade de Aveiro. Campus de Santiago, Aveiro, Portugal 3 Institut für Umweltwisseschaft, Universität Koblenz-Landau. Fortstrasse 7, Landau, Germany. Abstract Amphibians and reptiles have not been considered in environmental risk assessments of chemicals, which has generated some debate about whether risk posed by some pollutants like pesticides on these animals are covered by surrogates in the groups of fish, mammals and birds. In order to develop a scientifically sound and robust risk assessment scheme it is necessary to have enough information available on the biological relevance of effects observed in laboratory studies in view of population level effects, to identify sensitive life stages and to compare sensitivity of our target study groups with that of their surrogates. With these objectives, a systematic review of toxicological literature on amphibians and reptiles was conducted. The results suggest that, when analysing available information, fish-generated toxicity data seem to be appropriate to cover aquatic amphibians. However, additional research to clarify comparative sensitivity is desirable given the low number of comparisons done on current use pesticides or with the most sensitive amphibian life stages. In the terrestrial environment, the absence of correlation in the toxicity indicators between amphibians or reptiles and birds or mammals indicates that homeothermic vertebrates are not suitable surrogates for toxicity on amphibians or reptiles. When comparing laboratory-generated data with information from other scenarios, an increase in sensitivity in aquatic mesocosm studies was observed when comparisons were conducted on the same species, although indirect effects caused by stressors in the mesocosms other than contaminants (e.g. predators or competitors) possibly contributed to potentiate responses associated with the presence of pollutants in the water. The ratio of toxicity between lab and aquatic mesocosm would point to the fact that assessment factors used for both acute and chronic toxicity in fish would be insufficient to protect amphibians from toxic effects in natural communities. This systematic review shows a limitation of data leading to adequate comparisons for reptiles and terrestrial amphibians, as well as for aquatic amphibians exposed through oral routes. Recommendations are provided about further research needs and possibilities to adapt current risk assessment in order to protect amphibians and reptiles while minimizing additional vertebrate testing. European Food Safety Authority, EFSA Supporting publication 2017:EN-1251

2 Key words: amphibians, reptiles, toxicity, risk coverage, surrogates, risk assessment, pesticides Question number: EFSA-Q Correspondence: 2 EFSA Supporting publication 2017:EN-1251

3 Disclaimer: The present document has been produced and adopted by the bodies identified above as author(s). This task has been carried out exclusively by the author(s) in the context of a contract between the European Food Safety Authority and the author(s), awarded following a tender procedure. The present document is published complying with the transparency principle to which the Authority is subject. It may not be considered as an output adopted by the Authority. The European Food Safety Authority reserves its rights, view and position as regards the issues addressed and the conclusions reached in the present document, Acknowledgements: The authors acknowledge the input provided by Franz Streissl and Jane Richardson (EFSA) during the course of the study. Dick de Zwart kindly provided the information contained in his database to quantify and analyse inter-test variability in aquatic ecotoxicology. Peter Craig and Leo Poshtuma helped contacting Dr. de Zwart. The members of the working group of the EFSA PPR panel elaborating the scientific opinion on amphibians and reptiles provided comments on the design of the data models and the selection of life history model species: Robert Smith (chair), Paulien Adriansee, Annette Aldrich, Cecilia Berg, Philippe Berny, Kyriaki Machera, Silvia Pieper and Christopher Topping. The following authors facilitated the full text of some of their articles, which have been reviewed as part of the present project (in alphabetical order): Giovanni Bernardini, David F. Bradford, Ronald J. Brooks, Rickey Cothran, Pieter De Wijer, Nilda Ester Fink, Douglas Fort, Hemant Ghate, Rosalba Gornati, Paula F. P. Henry, Diane Lynn Larson, Anna Maryańska-Nadachowska, James Masuoka, Claudia Monteiro, Gianluigi Monticelli, Lorin Neuman-Lee, Chan Jin Park, Bruce D. Pauli, Paola M Peltzer, Cristina S. Pérez-Coll, Giorgio Rispoli, Domingos de Jesus Rodrigues, Rosaria Scudiero, Ulrich Sinsch, Sandra Castro Soares, Andrew J. Tindall, Stanley E. Trauth, Henk Vijverberg, Richard J. Wassersug, Rujas Yonle and Thomas Yorio. Margaret Wead and Pamela de Francesco provided bibliographic material from the libraries of Savannah River Ecology Lab and the Federación Bioquímica provincia de Buenos Aires, respectively. The authors declare that they have no competing interests. Suggested citation: UCLM, Universidade de Aveiro, Universität Koblenz-Landau, Manuel E. Ortiz- Santaliestra, Joao P. Maia, Andrés Egea-Serrano, Carsten A. Brühl and Isabel Lopes, Biological relevance of the magnitude of effects (considering mortality, sub-lethal and reproductive effects) observed in studies with amphibians and reptiles in view of population level impacts on amphibians and reptiles. EFSA supporting publication 2017:EN pp. doi: /sp.efsa.2017.en-1251 ISSN: European Food Safety Authority, 2017 Reproduction is authorised provided the source is acknowledged. 3 EFSA Supporting publication 2017:EN-1251

4 Summary Amphibians and reptiles are the two vertebrate taxa with a higher proportion of endangered species, and pesticide-related pollution is recognised as one of the major factors threatening populations of both groups. However, few pesticides have been studied with regards to their effects on amphibians or reptiles, which is partly due to the fact that amphibians and reptiles have not been considered in environmental risk assessments of pesticides. Whether risks of pesticides to amphibians and reptiles are indeed covered by surrogate taxa used in risk assessment or not is currently under debate. In order to develop a scientifically sound and robust risk assessment scheme it is necessary to have enough information on the biological relevance of effects observed in laboratory studies in view of population level effects. It is also necessary to identify the most sensitive life stages and possibilites to extrapolate from endpoints observed in studies with other taxa to avoid animal testing. Information needs to be gathered to examine whether fish, birds and mammals are valid surrogates for amphibians and reptiles. This report presents a systematic review of scientific literature providing toxicity data on amphibians and reptiles, conducted with the purposes to (1) identify the most sensitive amphibian and reptile life stages to chemical pollution, (2) identify the most sensitive endpoints at different levels of biological organization in order to estimate extrapolation factors from laboratory studies to field-collected data, and (3) compare endpoints from amphibians and reptiles with the available information from fish, birds and mammals in order to test the role of these taxa as valid surrogates for amphibians and reptiles. In addition, and because the use of population models to support specific protection goals in risk assessment has gained importance as a measure to identifiy population-level effects and reduce experimental testing, a compilation of information useful to parameterize population models of six selected species is conducted. A comprehensive search of relevant literature on toxicity data related to amphibians and reptiles was conducted, using the appropriate search strings and combinations, in six different source types: multidisciplinary databases of scientific literature (i.e. Web of Science and Scopus), literature included in general amphibian and reptile ecotoxicology compilations, literature compiled in technical reports previously prepared for EFSA, literature sources used for creating amphibian or reptile records in the ecotoxicological database created by De Zwart (see references for details), literature sources used for creating amphibian or reptile records in toxicological online databases (United States Environmental Protection Agency s Ecotoxicology Knowledgebase ECOTOX, and National Library of Medicine s Hazardous Substances Data Bank HSDB), and indexes of herpetological scientific journals of local scope not included in Web of Science or Scopus. After elimination of duplicated or obviosuly irrelevant records (i.e. not relating to toxicity on amphibians or reptiles), an initial screening of the abstract and, when necessay, full text, was conducted in order to verify whether records met the following inclusion criteria: the source relates to amphibians or reptiles, the source describes the exposure to a chemical, the source includes a control or reference for comparison, and the source reports information on an endpoint indicative of the outcome of the exposure. Records passing the initial screening were subjected to a study appraisal on their relevance and reliability. Finally, records passing this study appraisal were considered eligible for data extraction. Data extraction consisted of the retrieval of relevant information, including among other fields: species, age, sex, chemical substance, exposure route and duration, type of recorded endpoint, type of response, exposure concentration, mean effect value of the control and exposed groups and the reported variability measures of these mean effects, and statistical significance of the comparison. As measures of the toxicological effects, we used typical endpoint parameter values (LC/LD/EC/ED/NOEC) when they were included in the papers, or when we were able to estimate them from mean effects after fitting a statistically significant regression model. For the rest of cases, we calculated Hedge s d values as the metric of standardized effect size; d values were relativized as a function of exposure time (d(t)) and a toxicity index was calculated as the regression coefficient of d(t) on the chemical concentration associated to such d(t) value. Effect characterization values (either typical parameters or toxicity indexes) were then analyzed following specific procedures adapted to 4 EFSA Supporting publication 2017:EN-1251

5 each of the three objectives defined above: (1) For identification of sensitive life stages, effect characterization values were compared among life stages for each substance, class, exposure route and type of response, using Marginal Means generated after running Generalized Linear Models, or comparing significant toxicity indexes obtained for each life stage. (2) For estimation of extrapolation factors from laboratory to field, regressions were conducted for each substance, class, life stage, exposure route and type of response to compare effect characterization values from laboratory and field studies; studies in mesocosms simulating natural communities were also included in the comparison as an intermediate step from laboratory to field. Extrapolation factors were derived from significant regression models. (3) For comparison of sensitivity between amphibians or reptiles and their surrogates, we used the LC/LD/EC/ED values from fish, birds and mammals and compared to the data retrieved in our review of literature for each substance, taxon, exposure route and type of response. In addition, Species Sensitivity Distributions (SSD) were modeled for substances for which comparable endpoints were available for a minimum of six different species. The resulting HC 5 were compared, when possible, with those obtained from SSD run with the corresponding surrogate vertebrate taxon. The review was conducted on 3642 full-text records for chemical exposure effects and 556 for life history traits, out of which 1332 and 204, respectively, were finally used for data extraction. The datasets comprised values corresponding to effects of chemicals and 1854 values corresponding to life history traits. Considering the three objectives corresponding to the review of the effects of chemicals on amphibians and reptiles, the main results of the study were: (1) Identification of the most sensitive life stage. Hatchlings (small larvae that have not yet finished internalizing gills, in the case of anurans), resulted as the most sensitive life stage among aquatic amphibians, followed by larvae, whose sensitivity was though not very different from that of hatchlings, and embryos. Some substance-specific exceptions to this general pattern were observed, such as the highest sensitivity of embryos to some cationic substances (e.g. lead, ammonium). (2) Extrapolation from laboratory data to mesocosm and field situations. The intensity of effects observed after aquatic exposure of amphibians in laboratory did not differ from that observed in mesocosms when no species-specific differences were taken into account. However, when data for the same species were compared, time-corrected endpoint values calculated from mesocosm studies were 303 times more sensitive than those calculated from field studies. This extrapolation factor increased to 1429 when only lethal effects were included in the analysis. The additional stress from natural sources like predation or competition to which amphibians are exposed in mesocosms, even considering that compared treatments differed only in the chemical exposure, could explain this finding. The low number of available comparisons with field data does not allow a statistical comparison. (3) Comparison with surrogate taxa. Fish seem to be in general more sensitive than amphibians to waterborne pollutants. This higher sensitivity of fish compared to tested amphibians, and especially the higher sensitivity of the rainbow trout (the species most commonly used in toxicity assessment), indicates fish as potentially useful surrogate species, providing appropriate coverage of the toxicity of pollutants in the water to aquatic amphibians. However, a few cases have been observed in which, because of a lower sensitivity of the rainbow trout, coverage by the standard species is not so clear. The substances involved in these cases do not allow for estimating any particular type of substance or mode of action explaining this finding. Higher sensitivity of fish compared to aquatic amphibians was confirmed by comparison of the SSD-derived HC 5 values. Data from both taxa were correlated, which, although correlation was weak and the number of compared substances was low, shows the potential of fish to act as surrogates for toxicity assessment on amphibian aquatic stages. However, a definitive confirmation on the usage of fish as surrogates for amphibians would require toxicity data generated from the most sensitive amphibian life stages (i.e. hatchlings), which are very scarce. In the terrestrial environment, although birds and mammals are more sensitive than amphibians and reptiles to a higher number of substances than vice versa, currently available data do not allow for 5 EFSA Supporting publication 2017:EN-1251

6 extrapolating toxicity data between groups. Furthermore, the frequency of cases in which amphibians or reptiles are more sensitive than birds or mammals is around 30%. Several groups of substances such as pyrethrins and organochlorine compounds are more toxic to amphibians or reptiles than to homeothermic vertebrates, which means that differences between classes in the way that toxicants are absorbed, metabolised or eliminated may result in entire groups of pollutants causing stronger toxicity on amphibians and reptiles than on birds and mammals. More data should be compiled for other substances on terrestrial toxicity to rule out similar trends showing a comparatively higher sensitivity of amphibians or reptiles, thus leading to a non-protective risk assessment. In general, the outcome of this review reflects a disquieting lack of information on toxicity for reptiles, terrestrial amphibians and, to a lesser extent, aquatic amphibians. Whereas this problem had already been pointed out in previous reviews, the results of the present systematic review come to confirm that much information needs to be generated to extract relevant conclusions on the overall coverage by surrogate vertebrates currently used in environmental risk assessment of the toxicity of chemicals on amphibians and reptiles. 6 EFSA Supporting publication 2017:EN-1251

7 Table of contents Abstract... 1 Summary Introduction Background Objectives Data and Methodologies Literature search about effects of chemicals on amphibians and reptiles Definition and validation of search protocol Inclusion criteria and search strings Quality check of literature search and redefinition of search strings Literature sources Multidisciplinary, international scientific literature databases Thematic literature compilations Specialised databases and EFSA reports Local herpetological journals Review of literature about effects of chemicals on amphibians and reptiles Definition of exclusion criteria Initial screening Study appraisal Data extraction Compilation of relevant information for development of amphibian and reptile landscape or population modelling Selection of model species Initial choices Evaluation of initial choices and final selection Literature search about relevant life history traits for population modelling of selected species Multidisciplinary, international scientific literature databases Local herpetological journals Review of literature about relevant life history traits for population modelling of selected species Data extraction Tools for appraisal of the systematic review Appraisal of systematic review of data Appraisal of extensive literature reviews Analysis of data of effects of chemicals on amphibians and reptiles Processing of data Identification of sensitive amphibian and reptile life stages Comparison of endpoints observed in laboratory with field data Comparison of sensitivity with surrogate taxa Development of species sensitivity distributions Results and Discussion Reviewed literature about effects of chemicals on amphibians and reptiles Identification of sensitive amphibian and reptile life stages Comparison of endpoints observed in laboratory with mesocosm and field data Comparison of sensitivity with surrogate taxa Aquatic environment Terrestrial environment Species sensitivity distributions Species Sensitivity Distribution (SSD) in amphibian species SSDs in surrogate species and comparison with amphibian data EFSA Supporting publication 2017:EN-1251

8 3.6. Life history traits for population modelling of selected species Results of the literature review Summary of information Conclusions Recommendations References Appendix A List of papers used for quality check of the literature searches Appendix B Local herpetological literature sources reviewed Appendix C Initial screening data model Appendix D Study appraisal data model Appendix E Endpoint study data model Appendix F List of amphibian and reptile species in the EU Appendix G Life history trait data model Appendix H Outcome of sensitivity comparisons between life stages Appendix I Outcome of comparisons of sensitivity recorded in laboratory, mesocosm and field studies 133 Appendix J Data used in the regression to compare sensitivity of aquatic amphibians to pollutants from laboratory and mesocosm studies Appendix K Data for comparison of sensitivity between aquatic amphibians and surrogate fish 140 Appendix L Data for comparison of sensitivity between terrestrial amphibians or reptiles and surrogate birds or mammals Appendix M Data used for SSD of amphibians EFSA Supporting publication 2017:EN-1251

9 1. Introduction 1.1. Background Worldwide declines in the populations of amphibians and reptiles rank among the most critical concerns in conservation biology (IUCN 2016). Of the more than 7000 (amphibians) and 9000 (reptiles) known species about 31 and 21%, respectively, are considered endangered or vulnerable to extinction (IUCN 2016). Several natural and anthropogenic factors have been associated with the declines of populations and species of these two groups, pollution being considered a major cause (Sparling 2010, IUCN 2016). The panoply of chemicals being released into the environment is enormous (thousands of chemicals being produced each year in quantities superior to 1 tonne) and pesticides form a specific group of chemicals since they are designed to be biologically active and are released on purpose in the environment. In the EU-28, almost t of active chemicals were sold in 2014 (EUROSTAT 2017). Because it is difficult to generate toxicity data for all existing chemicals, the adverse effects that most of these chemicals may induce to biota are poorly understood, especially with regards to amphibians and reptiles (Hayes et al. 2006, Pauli et al. 2010, Sparling et al. 2010). This lack of information on toxicity of chemicals to amphibians and reptiles is related to the fact that: (i) only recently, standardized protocols were published to run ecotoxicity assays with amphibians, yet, none exists for reptiles (e.g. OECD 2009, ASTM 2013); (ii) additional vertebrate testing to fill the existing knowledge gaps is generally discouraged from the perspective of animal use for experimental purposes; and (iii) these two groups of organisms have not been traditionally considered in environmental risk assessments (ERA) of chemicals that manufacturers must conduct before a substance can be placed on the market (amphibian or reptile toxicity testing is still not specifically required for the authorization of chemicals in Europe according to the Regulation (EC) No 1907/2006). However, in 2013, two pieces of European legislation on pesticide risk assessment for the terrestrial environment were published and mentioned terrestrial amphibians and reptiles: Regulation (EU) 283/2013, setting out the data requirements for risk assessment of active substances, establishes that available and relevant data regarding the potential effects on amphibians and reptiles in the terrestrial environment shall be taken into account in the risk assessment, and the Regulation (EU) 284/2013, which refers to commercial formulations, establishes that, when the risks on amphibians and reptiles in the terrestrial environment cannot be predicted from the available information about active substances, it has to be somehow addressed, according to what each member state considers. Therefore, to date, the risk assessment conducted with other vertebrate taxa has been considered to cover the risks of chemicals to amphibians and reptiles. Namely, in aquatic risk assessment, fish have been used as surrogates for the aquatic life stages of amphibians, while birds and mammals are used as representatives for vertebrates, and therefore as surrogates for terrestrial life stages of amphibians and for reptiles. However, given the poor development of research in amphibian and reptile ecotoxicology it is not possible to accurately state that these three classes of vertebrates (or other taxonomic groups) constitute safe/protective surrogates for ERA of amphibians and reptiles. But, if patterns of toxicity are to be identified, they may help to determine extrapolation factors for future risk assessment schemes of amphibians and reptiles, preventing the need for additional animal experimentation. Some studies have already been carried out aiming at identifying such toxicity patterns. For instance, Kerby et al. (2010) compared the median lethal concentrations of several chemicals on amphibians and eleven invertebrate taxonomic groups. These authors established four types of chemicals (phenols, metals, pesticides, inorganic compounds) and computed the HC 5 (i.e. the median lethal concentration LC 50 for a species more sensitive than 95% of all species included in a species sensitivity distribution) for amphibians and the invertebrate taxa. For metals, pesticides and inorganic compounds, the HC 5 computed for amphibians were overall higher than those computed for the eleven invertebrate groups. However, for the phenol group, 10 of the invertebrate taxa exhibited a significantly higher HC 5 comparatively to amphibians. A trait based analysis of the ecological vulnerability of wildlife to five substances (De Lange et al. 2009) showed that amphibians and reptiles would generally be more vulnerable than the rest of vertebrate groups. Weltje et al. (2013) made a 9 EFSA Supporting publication 2017:EN-1251

10 systematic comparison of the acute and chronic toxicity data (for eight inorganic and 47 organic compounds) of amphibians and fish, and found that, in general, fish are more sensitive than amphibians. Regarding acute sensitivity, amphibians were 10 to >100-fold more sensitive than fish for only six of the 55 analysed compounds, while for chronic toxicity, amphibians were 10 to >100-fold more sensitive than fish for only three of the 52 tested chemicals. The scarcity of toxicity data available for terrestrial stages of amphibians or reptiles makes the establishment of patterns comparing sensitivity between these animals and their surrogates more complicated than for aquatic stages. Friday and Thompson (2012) analyzed studies retrieving acute oral LD 50 values for amphibians and for a bird or mammal species to conclude that acute oral toxicity was higher to surrogate species than to amphibians for 18 out of the 20 compared substances. Crane et al. (2016) compared single oral LD 50 values between amphibians and birds or mammals for 26 substances and found that amphibian toxicity values were correlated with values for mammals, but not with avian ones. They also observed that six of the compared substances had a lower LD 50 in amphibians than in mammals and five had a lower LD 50 in amphibians than in birds. Weir et al. (2010), based on a survey of scientific literature, reported that reptiles were more sensitive than birds to five out of the 15 chemicals that they analysed. These authors also indicated that exposure of reptiles was relatively high, particularly when considering the dermal route, thus suggesting that caution is needed when using birds as surrogates for reptiles in ERA. Actually, the exposure scenarios traditionally used for fish, birds or mammals, often exhibit low relevancy for amphibians and reptiles, especially in the terrestrial environment where dermal exposure is likely to be a relevant route for amphibians (Brühl et al. 2011, Van Meter et al. 2014) and reptiles (Weir et al. 2014), but this exposure route is not considered in tests with birds or mammals. The available information does not seem to be conclusive enough to determine whether current ERA protocols are adequately protective for amphibians and reptiles. Uncertainties in ERA regarding these groups would be solved if they were incorporated in the ERA procedure, but this would involve a series of problems; first, use of vertebrate species in experimentation is discouraged, and the EU legislation (Directive EC/63/2010) recommends the implementation of replacement protocols in order to minimize the number of animals used for experimental purposes. Second, the scarcity of standard test guidelines for aquatic amphibians, and the complete lack of such guidelines for terrestrial amphibians and reptiles, is a flaw for the execution of tests required for risk assessment. And third, including new testing in ERA protocols would involve an additional expenditure of resources both by industry, which must conduct the tests, and by risk assessors from regulatory agencies who must evaluate the dossiers. In order to develop a scientific sound and robust risk assessment scheme it is necessary to have enough information on the biological relevance of effects observed in laboratory studies in view of population level effects. It is also necessary to identify the most sensitive life stages and possibilities to extrapolate from endpoints observed in studies with other taxa to avoid animal testing. Information is needed to verify whether fish, birds and mammals are in fact good surrogates for amphibians and reptiles. Finally, another way to improve the efficacy of ERA protocols without increasing animal testing is the implementation of population models, which allow for estimation of population-level responses from individual-collected data (Willson et al. 2012). These models are being incorporated to some of the scientific opinions on ERA of Plant Protection Products (PPP) that EFSA is currently developing, and their implementation for use in risk assessment with amphibians has been recommended in a recent paper co-authored by experts of different governmental agencies of the United States and Canada (Johnson et al. in press). Actually, some population models have been already developed for extrapolating pollutant effects on amphibian populations (Salice et al. 2011, Willson and Hopkins, 2013). Population models can be useful to define specific protection goals (SPG) associated with population-level effects while saving the enormous effort required to conduct field studies for monitoring population trends. In order to develop robust population models, their parameterization should be based on extensive knowledge about the model species, as the better the accuracy of the parameterization, the more realistic the model will be EFSA Supporting publication 2017:EN-1251

11 1.2. Objectives Promoting a deeper understanding on the comparative toxicity of chemicals among the different vertebrate classes is necessary to support sound and robust ERA of amphibians and reptiles, and eventually establish extrapolation factors. Thus, following the specifications published by EFSA, the present report presents an extensive literature review on amphibian and reptile ecotoxicological information conducted with the purposes of: (1) identifying the most sensitive amphibian and reptile life stages to chemical pollution, (2) identifying the most sensitive endpoints, for amphibians and reptiles, at different levels of biological organization to estimate extrapolation factors from laboratory studies to field-collected data, (3) comparing endpoints from amphibians and reptiles with the information from fish, birds and mammals in order to test the role of these taxa as valid and protective surrogates for amphibians and reptiles, and (4) selecting species covering the taxonomic diversity of amphibians suitable to be used for landscape modelling in order to define population-level protection goals on the basis of such models, and compile information on life-history traits useful to develop such models EFSA Supporting publication 2017:EN-1251

12 2. Data and Methodologies The systematic review of literature was split into two sections. A first section was focused on the review of studies dealing with effects of chemicals on amphibians and reptiles in order to address the objectives 1, 2 and 3 defined above. The second section aimed at compiling information on life history traits relevant for future development of population models using six pre-defined model species. Whereas the workflow was identical (i.e. search, screening, appraisal and data extraction), important differences in specific parts of the methodology existed, motivated by the intrinsic differences in the type of data to be recorded. For this reason, we present the methodology separated for each of the two study sections Literature search about effects of chemicals on amphibians and reptiles Definition and validation of search protocol Inclusion criteria and search strings Several sources of information were defined and records were searched and selected following a PECO (i.e. Population, Exposure, Comparator, Outcome) inclusion procedure (EFSA 2010). For all information sources, records were initially selected according to the following inclusion criteria: 1. The source relates to amphibians or reptiles (Population criterion). 2. The source describes the exposure to a chemical (Exposure criterion). 3. The source reports information on an endpoint indicative of the outcome of the exposure (Outcome criterion). The Comparator criterion (i.e. the source includes a control or reference group for comparison) was not considered in the inclusion strategy as it is difficult to identify with a title or a title + abstract screening. This criterion was further included in the initial screening of the full texts. Two multidisciplinary, peer-reviewed literature databases were used as the primary source of information. These were the Web of Science (Thomson Reuters) and Scopus (Elsevier), where we retrieved potentially relevant records. In order to fulfil the inclusion criteria defined above, the following search strings and combinations were used: Search string: Amphibians Amphibia* OR Anura OR Acris OR Agalychnis OR Alytes OR Anaxyrus OR Ascaphus OR Atelopus OR Barbarophryne OR Batrachophrynus OR Bombina OR Bufo OR Bufotes OR Ceratophrys OR Chiromantis OR Crinia OR Cryptobatrachus OR Cyclorana OR Dendrobates OR Discoglossus OR Duttaphrynus OR Eleutherodactylus OR Epidalea OR Epipedobates OR Fejervarya OR Gastrophryne OR Gastrotheca OR Glandirana OR Hoplobatrachus OR Hyla OR Hyperolius OR Hypsiboas OR Kassina OR Leiopelma OR Lepidobatrachus OR Leptodactylus OR Limnodynastes OR Limnonectes OR Lithobates OR Litoria OR Mantella OR Mantophryne OR Melanobatrachus OR Melanophryniscus OR Microhyla OR Myobatrachus OR Nectophryne OR Neobatrachus OR Nothophryne OR Nyctibatrachus OR Odontophrynus OR Odorrana OR Oreophryne OR Osteopilus OR Pelobates OR Pelodytes OR Pelophylax OR Phyllobates OR Phyllomedusa OR Physalaemus OR Pleurodema OR Polypedates OR Pseudacris OR Pseudophryne OR Ptychadena OR Pulchrana OR Rana OR Rhacophorus OR Rheobatrachus OR Rhinella OR Rhinoderma OR Scaphiopus OR Scinax OR Sclerophrys OR Silurana OR Spea OR Strongylopus OR Telmatobius OR Xanthophryne OR Xenopus OR Caudata OR Urodela OR Ambystoma OR Amphiuma OR Batrachoseps OR Bolitoglossa OR Calotriton OR Chioglossa OR Cryptobranchus OR Cynops OR Desmognathus OR Dicamptodon OR Ensatina OR Euproctus OR Eurycea OR Hydromantes OR Hynobius OR Ichthyosaura OR Lissotriton OR Lyciasalamandra OR Mertensiella OR Necturus OR Notophthalmus OR Oedipina OR Ommatotriton OR Paramesotriton OR Plethodon OR Pleurodeles OR Proteus OR Pseudoeurycea OR Pseudotriton OR Rhyacotriton OR Salamandra OR Salamandrella OR 12 EFSA Supporting publication 2017:EN-1251

13 Salamandrina OR Siren OR Speleomantes OR Taricha OR Triturus OR Gymnophiona OR Caecilia OR Dermophis OR Geotrypetes OR Gymnophis OR Ichthyophis OR Siphonops OR Typhlonectes Search string: Reptiles Reptil* OR Crocodilia OR Alligator OR Caiman OR Crocodylus OR Gavialis OR Rhynchocephalia OR Sphenodon OR Squamata OR Sauria OR Ophidia OR Amphisbaenida OR Acanthodactylus OR Adelophis OR Agama OR Agkistrodon OR Algyroides OR Amblyodipsas OR Ameiva OR Amphisbaena OR Anatololacerta OR Anguis OR Anisolepis OR Anolis OR Archaeolacerta OR Aspidoscelis OR Atlantolacerta OR Australolacerta OR Balanophis OR Bitis OR Blanus OR Boiga OR Bothrops OR Calliophis OR Cerastes OR Chalcides OR Chamaeleo OR Chironius OR Coelognathus OR Coluber OR Corallus OR Coronella OR Crotalus OR Cryptoblepharus OR Ctenophorus OR Ctenosaura OR Daboia OR Dalmatolacerta OR Darevskia OR Dinarolacerta OR Echinosaura OR Elaphe OR Elapsoidea OR Eryx OR Eumeces OR Gallotia OR Gekko OR Geophis OR Gerrhosaurus OR Helicops OR Hellenolacerta OR Hemidactylus OR Hemorrhois OR Heterodon OR Hierophis OR Holbrookia OR Hoplocephalus OR Hyalosaurus OR Hydrophis OR Hydrosaurus OR Iberolacerta OR Iguana OR Imantodes OR Iranolacerta OR Japalura OR Lacerta OR Lacertoides OR Lamprolepis OR Lampropeltis OR Lamprophis OR Leiocephalus OR Lepidodactylus OR Liolaemus OR Liopeltis OR Lycodon OR Lyriocephalus OR Macropisthodon OR Macroprotodon OR Macrovipera OR Malpolon OR Mesalina OR Micrurus OR Naja OR Natrix OR Nerodia OR Omanosaura OR Ophisops OR Parvilacerta OR Phelsuma OR Phoenicolacerta OR Phyllodactylus OR Pituophis OR Platyceps OR Plectrurus OR Podarcis OR Pogona OR Psammodromus OR Psammophis OR Pseuderemias OR Python OR Rhabdophis OR Rhabdops OR Rhacodactylus OR Rhinechis OR Rhinoplocephalus OR Saurodactylus OR Scelarcis OR Sceloporus OR Scincus OR Sphaerodactylus OR Stellagama OR Stenodactylus OR Tarentola OR Teira OR Thamnophis OR Timon OR Trapelus OR Trioceros OR Trogonophis OR Typhlops OR Typhlosaurus OR Uromastyx OR Varanus OR Vipera OR Xenodon OR Zamensis OR Zootoca OR Testudines OR Caretta OR Chelonia OR Chelydra OR Chrysemys OR Dermochelys OR Emys OR Eretmochelys OR Erymnochelys OR Geochelone OR Geoclemys OR Glyptemys OR Gopherus OR Graptemys OR Lepidochelys OR Macrochelys OR Malaclemys OR Manouria OR Mauremys OR Pelodiscus OR Podocnemis OR Pseudemys OR Terrapene OR Testudo OR Trachemys Search string: Chemicals pollut* OR metal* OR arsenic OR cadmium OR copper OR mercury OR selenium OR agrochemical* OR fertilizer* OR nitrate OR nitrite OR pesticide* OR herbicide OR triazine OR atrazine OR glyphosate OR paraquat OR diquat OR insecticide* OR organochlorine* OR DDT OR DDE OR DDD OR lindane OR endosulfan OR organophosphate OR chlorpyrifos OR malathion OR parathion OR carbamate OR carbaryl OR carbofuran OR neonicotinoid OR imidacloprid OR fipronil OR fungicide* OR dithiocarbamate OR tebuconazole OR difenoconazole OR sulphur OR folpet OR captan OR thiram OR rodenticide* OR brodifacoum OR bromadiolone OR warfarin OR difenacoum OR biocide* OR POP OR PCB OR PBDE OR PFOS OR pharmaceutical OR BPA OR ethynilestradiol OR nanomaterial* OR nanoparticle* OR nanotube* Following the PECO approach, search strings Amphibians and Reptiles would correspond to the Population criterion, search string Chemicals to the Exposure criterion, and the combination of either Amphibians or Reptiles with Chemicals to the Outcome criterion. Searches were run on February 26 th 2016 considering the entire time span of each database ( for Web of Science, for Scopus). For the Web of Science, the keywords were searched within the Topic field (including Title, Abstract, Author Keywords, and Keywords Plus). For Scopus, they were searched within the TITLE-ABS-KEY field (title, abstract and keywords). The search progress and resulting records from these literature sources is indicated in the Table 1. The records resulting from searches identified as #5 in Table 1 were exported to an EndNote file (HERP_REF_1) EFSA Supporting publication 2017:EN-1251

14 Table 1: Number of records retrieved from Web of Science and Scopus after searches using the first versions of the different search strings and combinations. Search Keyword search and terms Records in Web of Science Records in Scopus #1 Amphibians 300,093* 178,085 #2 Reptiles 112,570* 74,235 #3 #1 OR #2 401,989* 241,238 #4 Chemicals 7,893,815* 4,843,674 #5 #3 AND #4 11,263 11,653 *Approximate number of records; Web of Science does not eliminate duplicated records (retrieved from more than one of the databases included in the Web of Science) until navigation to or exportation of the last record Quality check of literature search and redefinition of search strings In order to check the validity of reference search protocol defined in the previous section, we created a list of relevant studies covering the different inclusion criteria (e.g. in terms of taxonomy, chemicals and outcomes), which we would expect to be found through the defined literature search strategy. That list was created after reviewing the papers cited by Sparling et al. (2010) in its reference book about ecotoxicology of amphibians and reptiles. The list included 138 records, which are listed in Appendix A. We confirmed that 131 out of the 138 relevant records had been retrieved from the Web of Science and Scopus searches. We used the seven non-retrieved records to improve the quality of the searches, for which we examined the reasons why these records had not been retrieved. We concluded that the absence of those records was motivated because of either (i) only scientific names had been included in the Amphibians or Reptiles search strings, or (ii) some substances were not referred in the Chemicals search string. Therefore, we decided to redo the searches using improved search strings as follows (grey-shadowed terms are the ones incorporated to the improved search strings): Search string: Amphibians Amphibia* OR Anura OR Acris OR Agalychnis OR Alytes OR Anaxyrus OR Ascaphus OR Atelopus OR Barbarophryne OR Batrachophrynus OR Bombina OR Bufo OR Bufotes OR Ceratophrys OR Chiromantis OR Crinia OR Cryptobatrachus OR Cyclorana OR Dendrobates OR Discoglossus OR Duttaphrynus OR Eleutherodactylus OR Epidalea OR Epipedobates OR Fejervarya OR Gastrophryne OR Gastrotheca OR Glandirana OR Hoplobatrachus OR Hyla OR Hyperolius OR Hypsiboas OR Kassina OR Leiopelma OR Lepidobatrachus OR Leptodactylus OR Limnodynastes OR Limnonectes OR Lithobates OR Litoria OR Mantella OR Mantophryne OR Melanobatrachus OR Melanophryniscus OR Microhyla OR Myobatrachus OR Nectophryne OR Neobatrachus OR Nothophryne OR Nyctibatrachus OR Odontophrynus OR Odorrana OR Oreophryne OR Osteopilus OR Pelobates OR Pelodytes OR Pelophylax OR Phyllobates OR Phyllomedusa OR Physalaemus OR Pleurodema OR Polypedates OR Pseudacris OR Pseudophryne OR Ptychadena OR Pulchrana OR Rana OR Rhacophorus OR Rheobatrachus OR Rhinella OR Rhinoderma OR Scaphiopus OR Scinax OR Sclerophrys OR Silurana OR Spea OR Strongylopus OR Telmatobius OR Xanthophryne OR Xenopus OR Caudata OR Urodela OR Ambystoma OR Amphiuma OR Batrachoseps OR Bolitoglossa OR Calotriton OR Chioglossa OR Cryptobranchus OR Cynops OR Desmognathus OR Dicamptodon OR Ensatina OR Euproctus OR Eurycea OR Hydromantes OR Hynobius OR Ichthyosaura OR Lissotriton OR Lyciasalamandra OR Mertensiella OR Necturus OR Notophthalmus OR Oedipina OR Ommatotriton OR Paramesotriton OR Plethodon OR Pleurodeles OR Proteus OR Pseudoeurycea OR Pseudotriton OR Rhyacotriton OR Salamandra OR Salamandrella OR Salamandrina OR Siren OR Speleomantes OR Taricha OR Triturus OR Gymnophiona OR Caecilia OR Dermophis OR Geotrypetes OR Gymnophis OR Ichthyophis OR Siphonops OR Typhlonectes OR frog OR toad OR salamander OR newt 14 EFSA Supporting publication 2017:EN-1251

15 Search string: Reptiles Reptil* OR Crocodilia OR Alligator OR Caiman OR Crocodylus OR Gavialis OR Rhynchocephalia OR Sphenodon OR Squamata OR Sauria OR Ophidia OR Amphisbaenida OR Acanthodactylus OR Adelophis OR Agama OR Agkistrodon OR Algyroides OR Amblyodipsas OR Ameiva OR Amphisbaena OR Anatololacerta OR Anguis OR Anisolepis OR Anolis OR Archaeolacerta OR Aspidoscelis OR Atlantolacerta OR Australolacerta OR Balanophis OR Bitis OR Blanus OR Boiga OR Bothrops OR Calliophis OR Cerastes OR Chalcides OR Chamaeleo OR Chironius OR Coelognathus OR Coluber OR Corallus OR Coronella OR Crotalus OR Cryptoblepharus OR Ctenophorus OR Ctenosaura OR Daboia OR Dalmatolacerta OR Darevskia OR Dinarolacerta OR Echinosaura OR Elaphe OR Elapsoidea OR Eryx OR Eumeces OR Gallotia OR Gekko OR Geophis OR Gerrhosaurus OR Helicops OR Hellenolacerta OR Hemidactylus OR Hemorrhois OR Heterodon OR Hierophis OR Holbrookia OR Hoplocephalus OR Hyalosaurus OR Hydrophis OR Hydrosaurus OR Iberolacerta OR Iguana OR Imantodes OR Iranolacerta OR Japalura OR Lacerta OR Lacertoides OR Lamprolepis OR Lampropeltis OR Lamprophis OR Leiocephalus OR Lepidodactylus OR Liolaemus OR Liopeltis OR Lycodon OR Lyriocephalus OR Macropisthodon OR Macroprotodon OR Macrovipera OR Malpolon OR Mesalina OR Micrurus OR Naja OR Natrix OR Nerodia OR Omanosaura OR Ophisops OR Parvilacerta OR Phelsuma OR Phoenicolacerta OR Phyllodactylus OR Pituophis OR Platyceps OR Plectrurus OR Podarcis OR Pogona OR Psammodromus OR Psammophis OR Pseuderemias OR Python OR Rhabdophis OR Rhabdops OR Rhacodactylus OR Rhinechis OR Rhinoplocephalus OR Saurodactylus OR Scelarcis OR Sceloporus OR Scincus OR Sphaerodactylus OR Stellagama OR Stenodactylus OR Tarentola OR Teira OR Thamnophis OR Timon OR Trapelus OR Trioceros OR Trogonophis OR Typhlops OR Typhlosaurus OR Uromastyx OR Varanus OR Vipera OR Xenodon OR Zamensis OR Zootoca OR Testudines OR Caretta OR Chelonia OR Chelydra OR Chrysemys OR Dermochelys OR Emys OR Eretmochelys OR Erymnochelys OR Geochelone OR Geoclemys OR Glyptemys OR Gopherus OR Graptemys OR Lepidochelys OR Macrochelys OR Malaclemys OR Manouria OR Mauremys OR Pelodiscus OR Podocnemis OR Pseudemys OR Terrapene OR Testudo OR Trachemys OR crocodile OR tuatara OR lizard OR skink OR snake OR turtle OR terrapin OR tortoise Search string: Chemicals pollut* OR metal* OR arsenic OR cadmium OR copper OR mercury OR selenium OR aluminium OR Pb OR agrochemical* OR fertilizer* OR nitrate OR nitrite OR pesticide* OR herbicide OR triazine OR atrazine OR glyphosate OR paraquat OR diquat OR insecticide* OR organochlorine* OR DDT OR DDE OR DDD OR lindane OR endosulfan OR organophosphate OR chlorpyrifos OR malathion OR parathion OR carbamate OR carbaryl OR carbofuran OR neonicotinoid OR imidacloprid OR fipronil OR fungicide* OR dithiocarbamate OR tebuconazole OR difenoconazole OR sulphur OR folpet OR captan OR thiram OR rodenticide* OR brodifacoum OR bromadiolone OR warfarin OR difenacoum OR biocide* OR POP OR PCB OR dioxin OR furan OR PBDE OR PFOS OR flame_retardant* OR perfluooroctanoic OR pyrene OR fluoranthene OR pharmaceutical OR BPA OR ethynilestradiol OR nanomaterial* OR nanoparticle* OR nanotube* The improved searches were run on March 5 th 2016, and the 138 records selected as relevant were then retrieved. The search progress and resulting records from these new searches are indicated in Table 2. The retrieved records from both databases after searches #5 in Table 2 were exported to an EndNote file (HERP_REF_1b). Among those records, there were too many duplicated and non-relevant records, so we conducted a title only screening to exclude those duplicated and obviously irrelevant records. The selected records were copied to a new EndNote filed (HERP_REF_2a) EFSA Supporting publication 2017:EN-1251

16 Table 2: Number of records retrieved from Web of Science and Scopus after searches using the second and definitive versions of the different search strings and combinations. Search Keyword search and terms Records in Web of Science Records in Scopus #1 Amphibians 361,988* 197,041 #2 Reptiles 192,499* 114,283 #3 #1 OR #2 538,526* 297,907 #4 Chemicals 8,685,290* 5,337,138 #5 #3 AND #4 15,345 15,021 *Approximate number of records; Web of Science does not eliminate duplicated records (retrieved from more than one of the databases included in the Web of Science) until navigation or exportation of the last record Literature sources In the next sections, we describe the search and initial selection of records performed for each source of information. The process of study selection is summarised in the Figure 1. Figure 1: Flow chart of the study selection process corresponding to references reporting effects of chemical substances on amphibians or reptiles EFSA Supporting publication 2017:EN-1251

17 Multidisciplinary, international scientific literature databases As we have described above, Web of Science and Scopus were used as the primary sources of information. The process of reference search in these databases is described in the previous section Thematic literature compilations The titles of the references cited in the two major sources compiling ecotoxicological literature related to amphibians and reptiles were reviewed: RATL: A Database of Reptile and Amphibian Toxicology Literature (B.D. Pauli, J.A. Perrault & S.L. Money Canadian Wildlife Service, Environment Canada). List of scientific literature on amphibian and reptile ecotoxicology published until Available at Ecotoxicology of Amphibians and Reptiles, 2nd edition (D.W. Sparling, G. Linder, C.A. Bishop & S. Krest SETAC / CRC Press). A reference manual on amphibian and reptile ecotoxicology published in A total of 1068 records among the papers cited in these two compilations were identified, according to their title, as potentially meeting the inclusion criteria, 660 of which were already included in the HERP_REF_2a file (including the 138 relevant records used for search quality check, see section ). The new 408 records were incorporated to a new EndNote file (HERP_REF_2b) Specialised databases and EFSA reports A series of databases totally or partially dealing with amphibian and/or reptile ecotoxicology were analysed to include all the information therein into the systematic review. Because the type of information necessary to address the objectives of the project was not coincident with the goal for which those databases were created, and thus the fields included therein did not always fit our objectives, we used the list of primary sources of information (i.e. literature records) instead of the contents of the databases themselves. The considered databases were: Specialised online databases. Two sources of toxicological data were consulted in order to select the records from which information related to amphibians and reptiles had been originated: United States Environmental Protection Agency s Ecotoxicology Knowledgebase (ECOTOX). This is a source for obtaining chemical toxicity data for aquatic life, terrestrial plants and wildlife. The data sources for this database are derived predominantly from the peerreviewed literature. The database is freely available at We retrieved records by searching order by order or, when too many records resulted, family by family. A total of 1050 records were retrieved, out of which 795 were already included in the HERP_REF_2a or HERP_REF_2b files. The 255 new records were then added to the HERP_REF_2b file. National Library of Medicine s Hazardous Substances Data Bank (HSDB). This database is included within the large Toxicology Data Network (TOXNET) of the US National Library of Medicine and provides all kind of toxicological information about potentially hazardous chemicals. The data sources for this database are derived predominately from the peerreviewed literature. The database is freely available at This database is not adapted to general taxonomical search, so the search of relevant data was conducted genus by genus. That search resulted in 2753 records, 54 out which were duplicated. Out of the 2699 unique records, 1521 were excluded in a title screening for being obviously irrelevant, while EFSA Supporting publication 2017:EN-1251

18 were already included in the HERP_REF_2a or HERP_REF_2b files. The 437 new records were then added to the HERP_REF_2b file. Specific databases used in previous, similar studies with other taxa. Hickey et al. (2012) used a large ecotoxicity database, initially compiled by De Zwart (2002), to quantify and analyse inter-test variability in aquatic ecotoxicology. Whereas the ecotoxicity data have been made available as a supplement of the online publication, the primary information sources are not listed therein. In order to identify the papers used to build this database, we contacted Dr. Dick De Zwart, who kindly provided us with a copy of the original database. The 2838 data records pertaining to amphibians came from a total of 195 references, 187 of which were already included in the HERP_REF_2a or HERP_REF_2b files. The eight new records were then added to the HERP_REF_2b file. Databases built to create the EFSA technical reports on amphibians and reptiles: Friday, S. & Thompson, H. (2009). Exposure of reptiles to plant protection products. Available at Friday, S. & Thompson, H. (2009). Compared toxicity of chemicals to reptiles and other vertebrates. Available at pdf. Friday, S. & Thompson, H. (2012). Toxicity of pesticides to aquatic and terrestrial life stages of amphibians and occurrence, habitat use and exposure of amphibian species in agricultural environments. Available at A total of 1064 records were retrieved from the three reports. We did not list unique records but pooled records from all three reports together. A total of 799 records out of the 1064 selected ones were already included in the HERP_REF_2a or HERP_REF_2b files. The 265 new records were then added to the HERP_REF_2b database Local herpetological journals In order to get information as complete as possible, we searched for information in peer-reviewed herpetological journals that, because of their local scope, are not indexed in the Web of Science or Scopus. The list of local herpetological journals considered in this step, as well as the timespan available for search in each case and the tools used to search their contents is shown in Appendix B. We used the search string Chemicals described in section , adapted to the primary journal language when necessary, to search within the titles of the articles published in those journals. When a search tool was not available, we reviewed all tables of contents by copying them into a text processing software (Microsoft Word) and running the searches keyword by keyword using the search tool of the software. Only three records meeting the inclusion criteria were found in the searched contents of these herpetological journals. Once we confirmed that these records were not yet included in either the HERP_REF_2a or HERP_REF_2b files, we added them to the latter Review of literature about effects of chemicals on amphibians and reptiles Once completed all the searches from the different sources described in the previous section, the 3128 records in the HERP_REF_2a file and the 1376 records in the HERP_REF_2b were pooled together (for a total of 4504) and uploaded to a project created in DistillerSR (Evidence Partners), where the literature review was performed. These records correspond to the RefIDs 1 to 4504 of the Ref_IDs dataset available as electronic supplementary material. The first step consisted of retrieving the full text of the selected references, for which we considered (i) open access sources, (ii) online contents subscribed by the different institutions in the consortium conducting this systematic review, (iii) hardcopies from printed series available in the libraries of the institutions in the consortium 18 EFSA Supporting publication 2017:EN-1251

19 conducting this systematic review, (iv) material facilitated by authors upon request, and (v) personal copies of the authors of the present report. We obtained full text for 3642 records. The majority of cases in which full text was not made available corresponded to congress abstracts of conference papers (probably without enough degree of detailed information), old reports not published on the web, theses or dissertations (likely to be published as journal papers), and papers from local sources. The complete process of literature review involving the 3642 retrieved full texts included three sequential levels, two initial levels of screening (level 1: initial screening; level 2: study appraisal) and a final level of data extraction (level 3: endpoint study) Definition of exclusion criteria During the review of the 3642 references mentioned above, which had been initially selected as matching the inclusion criteria on the basis of their titles, we found several records that were unsuitable to be included in the final review. The potential reasons for excluding those records were defined as exclusion criteria and were grouped in three categories: A. Validation of the record. A records was excluded if: A1. It was not possible to access to the information contained in the paper (e.g. wrong reference, unable to find the full text, unable to understand the language). A2. After reading the paper or a significant part of it, it did not meet the inclusion criteria. A3. The reference did not meet the relevance and reliability criteria described below (see section 2.2.3). B. Identification of the essential information. A record was excluded if: B1. There was a lack of compulsory information to fill mandatory fields of the data models designed for data extraction (see section for data model description). B2. There was a lack of details about the methods leading to ambiguous information or interpretation (e.g. use of a formulation without specification of percentage of active ingredient). C. Data collection. A record was excluded if: C1. There was ambiguity or inconsistence in the description of some of the methodological settings or results (e.g. the result in the text does not match the same result when shown in a table or figure). C2. There was duplicity of data, either within the same article or with some results published elsewhere. It should be noted that, when data were referred in a review or meta-analysis, the primary source was the one included in the data model. In this category, duplicated records that had passed unnoticed in the title-only screening were also included. The application of exclusion criteria A1 and A2 should presumably happen during the level 1 of the review, the application of criterion A3 during the level 2, and the criteria B and C during the level 3 of analysis, being the criteria in category B referred to methodological questions and the criteria in category C referred to questions related to results. Nevertheless, when the criterion C1 referred to a previously unnoticed duplicated reference, it was applied during either level 1 or 2 of the review Initial screening The level 1 of the review consisted of an initial screening of the full text to confirm whether the inclusion criteria were actually met. In addition to the inclusion criteria defined to select records during the literature searches, we added the existence of a control or reference group (i.e. Comparator criterion) to complete the PECO approach explained in section The structure of the data model used in this level 1 (initial screening) is shown in Appendix C. The following situations qualified for exclusion of a record in this level of analysis: 19 EFSA Supporting publication 2017:EN-1251

20 The paper is written in a language that none of the reviewers of the procurement team is able to understand proficiently (i.e. a language other than English, French, German, Italian, Portuguese, Spanish or Catalan). The paper does not relate to amphibians or reptiles. The paper does not describe the exposure to a chemical. The paper does not provide a comparison with a control or reference unit without exposure. The paper does not provide information on an endpoint indicative of the outcome of the exposure. All records were independently reviewed by two people. Records were proposed for exclusion if any of the situations qualifying for exclusion was selected. When the outcome of the two reviews was inclusion, the record passed to the level 2 of analysis. When the outcome of the two reviews was exclusion, the record was excluded from the review. When there was a conflict between the two reviews, a third independent reviewer was assigned and his/her decision was determinant. As part of this initial screening, we also classified the studies in the following types: Type 1: Studies establishing a correspondence between the exposure to a given level of a chemical substance and a response shown by exposed individuals as a consequence of such exposure. Four second-level types are considered in this category: Type 1A: In vitro studies. Type 1B: Experimental studies conducted under artificially established conditions (either in laboratory or mesocosms). Type 1C: Experimental studies conducted in the field, under natural conditions. Type 1D: Field studies identifying and quantifying chemical exposure, and one or several responses as consequence of such exposure. Type 2: Studies providing information on some life history trait of one of the model species. Although this section of the review was not intended to retrieve this type of information, there were some papers reporting exposure to chemicals that also contained potentially relevant information on life history traits of some of the model species. The selection of study types allowed for multiple responses. The references being classified within any of the type 1 categories were eligible for data extraction in this section of the project (i.e. studies dealing with exposure to chemicals) once they also passed the level 2 (study appraisal). Studies being classified within the type 2 category were eligible for data extraction in the section of the project aiming at compiling life history trait information for model species susceptible to be used in population modelling. The results of the initial screening of all references are presented in the initial_screening dataset available as electronic supplementary material Study appraisal The level 2 of the review consisted of an evaluation of relevance and reliability of the selected records, which we created following Ågerstrand et al. (2011). This evaluation included 20 questions with three possible answers: yes, no or not applicable. Each positive answer was assigned a value of +1, each negative answer was assigned a value of -1, and not applicable answers counted as zero. A quality score (QS) was calculated for each record according to the following equation: QS = [(sum of yes ) + (sum of no )] / [(sum of yes )+ (sum of no ) ] The structure of the data model Study appraisal, used in this second level of analysis, is shown in Appendix D. We expected the cases in which the answer was not applicable to be study type dependent, thus we pre-defined for which specific study types should each question be answered 20 EFSA Supporting publication 2017:EN-1251

21 (Table 3), although this was for guidance only and the questions being not applicable were decided on a case-by-case basis by the reviewer. Table 3: Scheme of the questions of the relevance and reliability evaluation expected to be answered according to each study type. O: question should be answered. X: question should not be answered. # Question To be answered for studies type 1A 1B 1C 1D 2 1 Is there a high probability that the species can be exposed to the tested substance in nature? O O O O X 2 Are the appropriate life-stage(s) or tissue(s) studied? O O O O X 3 Are the studied endpoint(s) recorded in a way that can be considered as realistic to happen in the wild? O O O O X 4 Is the test exposure scenario (route and duration of exposure, concentrations) representative for likely exposures in nature? O O O X X 5 Have critical parameters influencing the endpoints (ph, temperature) been considered adequately? O O O O X 6 Are the references reported? O O O O O 7 Is the purpose of the study clearly described? O O O O O 8 Is the test compound clearly identified? O O O O X 9 Is the exposure system (duration, concentration, route) properly described? O O O O X 10 Is the test concentration confirmed by analytical methods? O O O O X 11 Are the study organisms properly identified (when relevant: species, subspecies, sex, age)? O O O O O 12 Are the time-points of observations defined? O O O O X 13 Are controls appropriate for the tested organisms, substances and conditions? O O O O X 14 For field-collected organisms, is there information on the history of contamination? X O O O X 15 For field studies, is the habitat of the studied population described? X X O O O 16 Is the statistical design appropriate to address the purpose of the study? O O O O O 17 Are all novel methods and assumptions in the study adequately referred or justified? O O O O O 18 Is the geographical range of the individuals for which traits are reported defined? X X X X O 19 Is sample size from which traits have been calculated reported? X X X X O 20 Is there enough evidence that the sample of individuals for which trait is reported is not biased? X X X X O For studies dealing with chemical pollution, (type 1 according to what is explained in section 2.2.2), we established a threshold value of 0.6, below which the reviewed studies were considered not to meet the minimum relevance and/or reliability to be included in the review, and were therefore excluded according to the exclusion criterion A3 (see section 2.2.1). In order to calibrate this threshold, all records with a QS between 0.5 and 0.6 were reanalysed by a second reviewer in order to avoid reviewer-related biases during appraisal; the outcome of the second appraisal resulted in the inclusion of a 30.9% of the studies that had been initially excluded for having a QS between 0.5 and 0.6. Once all studies had been appraised (including the second appraisal of those with QS between 0.5 and 0.6), we simulated a review at the level 3 of studies with QS between 0.5 and 0.6, and found that only a 21.8% of them would have had information to fill all mandatory fields at the data 21 EFSA Supporting publication 2017:EN-1251

22 extraction level. The results of the study appraisal of all references are presented in the study_appraisal dataset available as electronic supplementary material Data extraction The level 3 of the review process consisted of the extraction of data from the selected records. Two data models were designed: the Endpoint study data model for collecting data on chemical exposure, and the Life history trait data model for collecting data useful for parameterization of the population models. In this section, we will focus on the former. For description of the latter, see section All studies assigned to at least one of the type-1 study categories, once they were included after initial screening and study appraisal, were eligible for data extraction using the Endpoint study data model (Appendix E). Whereas at levels 1 and 2 a single form was completed per reference, at level 3 most of the references required several instances (copies) of the forms to report different experiments within the same study, needed to report effects at different concentrations or on different responses. The way of reporting toxic effects derived from chemical exposure in the studies under review was slightly different depending on how each particular experiment had been conducted. There were two ways of reporting the effects: Data category 1: Reporting endpoints. Endpoints are defined here as benchmark values obtained from the integration of responses measured at different concentrations (e.g. LC 50, EC 50, NOEC, for which calculation it is necessary to make a regression with the percentage of effect at different exposure concentrations). Data category 2: Reporting responses for each tested level. These are studies in which different replicates of experimental units are exposed to different levels (doses, concentrations), including a control treatment, and a magnitude of effect is recorded at each level. For some of these studies it was possible to calculate an endpoint (e.g. LC 50, EC 50, NOEC) and for others it was not (e.g. if only one concentration was tested or if a statistically significant adjustment between exposure level and effect cannot be achieved). In many cases, the results of the study were shown graphically only, being necessary to process the figures with a software for digital image analysis. In these cases, we used the open access ImageJ 1.47v software (National Institute of Health, to record data Compilation of relevant information for development of amphibian and reptile landscape or population modelling Selection of model species The second section of the systematic review was focused on the retrieval of information that can be used in the future for parameterization of population models with amphibians and reptiles. A necessary previous step to this task was the choice of model species for implementing population models (hereafter model species). With this purpose, we used a trait-based approach focused on the identification of those species within each major taxonomic groups susceptible to be at risk of pesticide exposure and toxicity. An updated list of amphibian and reptile species present in the European Union was created using the database of the IUCN red list ( completed with the catalogue of species included in the Atlas of Amphibians and Reptiles of Europe developed by the European Herpetological Society (Sillero et al. 2014; atlas available at The criteria for listing the species were that they are native to the EU territory, excluding overseas areas (i.e. Macaronesian, Northern African and Trans-Oceanic territories). The list of species can be seen in the Appendix F, which also includes information on distribution, namely with the presence in each of the three zones defined for pesticide risk assessment according to the Regulation 1107/2009, taxonomical classification, IUCN category, inclusion in the Annex IV of the Habitats Directive (92/43/CEE) and presence on arable 22 EFSA Supporting publication 2017:EN-1251

23 lands, which was based on what is described in the IUCN red list sections for each species. For the species not listed by the IUCN, as well as for one of the species listed therein (Natrix natrix), for which the habitat is not described, it was not possible to find an information source good enough to harmonise habitat descriptions. Therefore, we directly looked for papers in Scopus describing the presence of those species in agricultural lands, and we found positive results for Natrix natrix (Meister et al. 2010) and Vipera berus (Leibl and Völkl 2009) in Germany, although for the latter species, reports from other areas indicate that it would avoid agricultural zones (Reading et al. 1996) Initial choices The initial selection criteria considered for the selection of amphibian and reptile model species were that i) they are present on arable lands, ii) they are included in the Annex IV of the Habitats Directive, and iii) they occur in as many of the zones defined for pesticide risk assessment according to the Regulation 1107/2009 as possible. Reptiles For reptiles, there were only two listed species present in all three European risk assessment zones, the sand lizard (Lacerta agilis) and the grass snake (Natrix natrix). While it seemed logical to select these species, the case of the grass snake presented some problems such as the fact that this species is more or less associated with water bodies (and therefore not representative of the ecology of most snakes) and also that data in Annex IV relates only to two subspecies, not to the entire taxon. Therefore, we looked for an alternative model species among those snakes appearing in two of the three zones and found two species: the horned viper (Vipera ammodytes) and the Aesculapian snake (Zamenis longissimus). Because of its wider distribution range, the selection of Zamenis longissimus as model species seemed the best option. It is worth noting the absence of chelonians from this list. Actually, none of the seven species (four terrapins and three tortoises) is identified as present in arable lands. We ran a literature search to retrieve data on the presence of these species on arable lands, which provided records solely referring to terrestrial tortoises (Testudo sp.), indicating that the progression of agriculture is causing fragmentation and extinction of populations (Hailey and Willemsen 2003, Anadón et al. 2007, Perez et al. 2012, Couturier et al. 2014). However, the presence of tortoises, and in particular of Hermann s tortoise (Testudo hermanni) in areas of traditional agriculture not exposed to intense mechanisation has been reported (Bertolero et al. 2011), which could make this species a suitable chelonian model. Amphibians We focused our efforts on looking for an anuran model species since the Working Group of the EFSA PPR Unit elaborating the scientific opinion on amphibians and reptiles is already developing a population model for a caudate species, the crested newt (Triturus cristatus) (Topping et al. in prep). There were seven anuran species meeting all selection criteria. In order to select one, we listed a series of ecological and biological traits that can be relevant in determining the risk that pesticides pose on these species (i.e. age at sexual maturity, clutch size, egg laying site, duration of the breeding season, juvenile diet type, metabolic rate, home range, maximal dispersal distance and maximum migration distance). The information was extracted from Trochet et al. (2014), except in the cases of Bufotes viridis and Pelophylax lessonae, for which we used the AmphibiaWeb ( For each parameter, we identified the worst case in order to make conservative choices and assigned, when possible, a score of 0 (best case) or 1 (worst case), in an analogous way as Wagner et al. (2014) and Mingo et al. (2016) did in their evaluations of pesticide exposure risks for European amphibians and reptiles, respectively. The final score per species was the average value of all assigned trait-specific scores (Table 4) EFSA Supporting publication 2017:EN-1251

24 Table 4: Initial evaluation of candidate anuran model species based on relevant biological and ecological parameters. Param. Worst case Sexual maturity (years) Clutch size Egg laying site Breeding season duration Diet of juveniles Metabolic rate Home range Max dispersal distance (m) Max migration distance Longer Smaller Lentic Explosive Herbivorous Higher Larger Longer Longer Species Descr. Score Descr. Score Descr. Score Descr. Score Descr. Score Descr. Score Descr. Score Descr. Score Descr. Score Bombina bombina Pelobates fuscus Hyla arborea Bufotes viridis Epidalea calamita Rana arvalis Pelophylax lessonae Lentic 1 Prolong. 0 Insectivorous 0 N/A 60 0 N/A Lentic 1 Explosive 1 Herbivorous 1 N/A N/A Lentic 1 Explosive 1 Insectivorous N/A N/A Lentic or lotic 1 Explosive 1 Insectivorous 0 N/A N/A N/A Lentic 1 Explosive 1 Herbivorous Lentic 1 Explosive 1 Herbivorous N/A Permanent 0 Explosive 1 Insectivorous 0 N/A N/A N/A N/A 0.60 Final score 24 EFSA Supporting publication 2017:EN-1251

25 The natterjack toad (Epidalea calamita) obtained the highest score. Looking at each parameter, the only case in which this species would not cover the other candidates was in metabolic rate. For this reason, we proposed to add a second anuran model species, the European tree frog (Hyla arborea), which has a high metabolic rate and, furthermore, shows considerable differences in ecology compared to the natterjack toad, thus providing a gradient in ecological variability Evaluation of initial choices and final selection In order to check the appropriateness of our initial proposal of model species, we looked at two review papers published in the last years by researchers of University of Trier that compiled some information on European amphibians (Wagner et al. 2014) and reptiles (Mingo et al. 2016) with the purpose of estimating pesticide risks on these species. Both studies followed a similar methodology, which involved the estimation of a pesticide risk factor based on parameters relevant for exposure and sensitivity (Table 5). Three parameters were used in each case to obtain an exposure. The exposure index is then corrected by the probability of species to spatially overlap with areas of pesticide application (estimated from literature data in the case of reptiles and with a GIS analysis in the case of amphibians). Table 5: Parameters used for estimating pesticide risk factors on recent review papers on European amphibians and reptiles. EF: exposure factors used to calculate the exposure index. Group Amphibians Reptiles Paper Wagner et al. (2014) Mingo et al. (2016) Risk factor name PRF (pesticide risk factor) ERF (exposure risk factor) Exposure index name HEI (habitat exposure index) ERI (exposure risk index) EF1 EF2 EF3 Habitat exposure risk Regular occurrence within cultivated landscapes Migration behaviour SVL and body mass (surrogates of physiology) Breeding accumulation Clutch size and clutches/year Whereas Mingo et al. (2016) includes a significant proportion of the reptilian European species, including all the widely distributed ones, Wagner et al. (2014) is restricted to amphibian species listed in the Annex II of the Habitats Directive (92/43/CEE), including only 24 species and generally excluding the widely distributed ones. We compared the PRF/ERF obtained by Wagner et al. (2014) and Mingo et al. (2016) with our preliminary selection of species (Table 6). Reptiles In Testudines, the ERF from Mingo et al. (2016) suggests Testudo graeca and Emys orbicularis to be more at risk than T. hermanni. We did not consider E. orbicularis as it is a freshwater species and it seemed more reasonable to focus on a terrestrial species (actually, exposure of freshwater turtles to waterborne pollutants would probably be covered by amphibians). The reason why we did not select T. graeca is because it was not identified by the IUCN as present on arable lands (neither was T. hermanni, but we confirmed its presence on arable lands based on the literature and results in Friday and Thompson (2009)). Although the occurrence in agricultural areas is implicitly within the ERF (and so T. graeca could be a suitable choice), the native distribution area of T. hermanni in the EU (whole Mediterranean basin) is considerably larger than that of T. graeca (Balcan / Greek Peninsulas). Based on this and the fact that the ERF is also high in the case of T. hermanni, we did not alter the initial choice of T. hermanni. For Saurians, we decided to select a lacertid, as this is the most biodiverse family of the group, and we were unsure that choosing a skink (like Chalcides bedriagai) would represent the ecology of the group. Thus, the most plausible alternative to L. agilis according to Mingo et al. (2016) data was Podarcis siculus. The limitation of this species is its restricted distribution (present only in the southern 25 EFSA Supporting publication 2017:EN-1251

26 zone) compared to L. agilis (present in the three zones), so the initial choice of L. agilis was maintained. For Ophidians, it is noteworthy that the preliminary choice, Z. longissimus, was not included in the Mingo et al. (2016) study, probably because they did not find evidence enough on its occurrence on agricultural areas. The first alternative derived from that paper, Coronella austriaca, was not one of the preliminary choices because it was not listed by the IUCN as present on arable lands. However, given that Mingo et al. (2016) listed this species in their study, we assumed that there is evidence about the occurrence of the species in agricultural areas and, also because it is widely distributed (present in the three zones), we modified our initial choice. Table 6: Comparison of preliminary choices of reptile model species with exposure risk factors (ERF) estimated by Mingo et al. (2016) for reptiles and with pesticide risk factors (PRF) estimated by Wagner et al. (2014) for amphibians, with indication on the position of each species in the ranking of most susceptible species. Group Preliminary choice (ERF/PRF) Species with >ERF/PRF Testudines (turtles) Testudo hermanni (0.38; 3 rd ) Testudo graeca (0.56; 1 st ) Emys orbicularis (0.41; 2 nd ) Saurians (lizards, geckos, skinks, iguanas ) Lacerta agilis (0.39; 5 th ) Chalcides bedriagai (0.52; 1 st ) Podarcis siculus (0.51; 2 nd ) Ablepharus kitaibelii (0.47; 3 rd ) Chalcides ocellatus (0.47; 3 rd ) Ophidians (snakes) Zamenis longissimus (not included) Coronella austriaca (0.49; 1 st ) Telescopus fallax (0.26; 2 nd ) Elaphe quatuorlineata (0.21; 3 rd ) Vipera ammodytes (0.21; 3 rd ) Natrix tessellata (0.08; 5 th ) Caudates (newts, salamanders) Anurans (frogs, toads) Triturus cristatus (0.13; 4 th ) Triturus dobrogicus (0.17; 1 st ) Triturus carnifex (0.14; 2 nd ) Triturus karelinii (0.14; 2 nd ) Epidalea calamita (not included) Hyla arborea (not included) Pelobates fuscus (0.40; 1 st ) Rana latastei (0.24; 2 nd ) Bombina bombina ( rd ) Discoglossus galganoi (0.19; 4 th ) Bombina variegata (0.07; 5 th ) Discoglossus sardus (0.03; 6 th ) Alytes muletensis (0.01; 7 th ) Amphibians For Caudates, Triturus cristatus appears in fourth place in the ranking of pesticide risk, although the PRF values are similar for the four Triturus species occupying the top four positions in the ranking. Given the wide distribution of T. cristatus compared to the other species, it was chosen to represent the ecology of this group. For anurans, the comparison of our initially selected model species with data from Wagner et al. (2014) could not be made because they limited their work to species included in Annex II of the Habitats Directive (where E. calamita and H. arborea are not listed). The Habitats Directive establishes that for Annex II species, special areas of conservation (SAC) must be defined. Wagner et al. (2014) estimated the percentage overlap with agricultural lands only by calculating how much of the SACs defined for each species corresponds to agricultural areas. Because SACs do not exist for E. calamita or H. arborea, we could not estimate a value for these species compared to the PRF calculated by Wagner et al. (2014). However, we followed the method for HEI estimation (EF1, EF2 and EF3 criteria, see Table 5) in order to extract some conclusions anyway EFSA Supporting publication 2017:EN-1251

27 EF1: habitat exposure risk: 2 risk points are awarded to species with high habitat exposure risk (i.e. species known to often inhabit agricultural fields or to breed in small, open water bodies in agricultural landscapes, or species for which a positive significant logistic regression of the presence on the proportion of agricultural area is calculated). 1 risk point is awarded to species with intermediate habitat exposure risk (i.e. species that can be found in agricultural areas and also in areas with lower risk of pesticide exposure, or species for which a non-significant logistic regression of the presence on the proportion of agricultural area is calculated). No risk points are awarded to species with low habitat exposure risk (i.e. species that are never found in agricultural areas, or species for which a negative significant logistic regression of the presence on the proportion of agricultural area is calculated). EF2: migration behaviour: 1 risk point is awarded to species with a significant proportion of populations/individuals performing annual breeding migrations. No risk points are awarded to species not commonly performing breeding migrations. EF3: breeding accumulation: 1 risk point is awarded to species that accumulate in specific sites for breeding. No risk points are awarded to species that do not accumulate in specific sites for breeding. In the approach to apply EF1, it was unclear how to define high/intermediate habitat risks, as all amphibian species present in agricultural areas are also commonly present in non-agricultural areas. Therefore, we established the differentiation between high/intermediate habitat risk on the basis of assignation made by Wagner et al (2014) to those species analysed in their paper. Therefore, E. calamita would be at high habitat risk whereas H. arborea would be at intermediate habitat risk. For EF2 and EF3, both species got both risk points (they have relatively long migration distances and aggregate for breeding). Therefore, the resulting HEI values were 4 for E. calamita and 3 for H. arborea. Even without the information on % overlap of the distribution areas with agricultural fields, we confirmed that the species with highest PRF values according to Wagner et al. (2014) were those with HEI values between 3 and 4 (see Table 1 therein). Thus, and considering the limitation in the diversity range covered by Wagner et al. (2014), we kept the intial choices for anuran model species. In summary, the selected model species for which life history information was retrieved were the great crested newt (Triturus cristatus), the natterjack toad (Epidalea calamita), the European treefrog (Hyla arborea), the Hermann s tortoise (Testudo hermanni), the sand lizard (Lacerta agilis), and the smooth snake (Coronella austriaca) Literature search about relevant life history traits for population modelling of selected species Considering the type of information that we aimed at retrieving in this part of the project, the literature searchers were limited to two of the sources used for the review of the effects of chemicals detailed in section 2.1.2: the multidisciplinary, international scientific literature databases and the local herpetological journals. The rest of the sources are focused on chemical exposure, so we did not expect to get information from papers therein that had not been already selected for the analysis of toxic effects EFSA Supporting publication 2017:EN-1251

28 Multidisciplinary, international scientific literature databases The same two multidisciplinary, peer-reviewed literature databases described in section were used as the primary source of information: the Web of Science and Scopus. Potentially relevant records were retrieved using a combination of two search strings: The search string species included current and past scientific names as well as common names for the six model species. The search string traits included keywords related to all life history traits identified as providing relevant information to the landscape-based population models that are expected to be developed. Search string: Species triturus_cristatus OR great_crested_newt OR bufo_calamita OR epidalea_calamita OR natterjack_toad OR hyla_arborea OR european_treefrog OR testudo_hermanni OR hermann s_tortoise OR lacerta_agilis OR sand_lizard OR coronella_austriaca OR smooth_snake Search string: Traits life_history OR survival_rate OR mass OR size OR length OR width OR life_cycle OR development OR phenology OR laying_rate OR laying_frequency OR emergence OR sexual_maturity OR lifespan OR longevity OR home_range OR migrat* OR movement OR displacement OR food_intake OR metabolic_rate Searches were conducted on September 30 th 2016 considering the entire time span of each database ( for Web of Science, for Scopus). For the Web of Science, the key words were searched within the Topic field (including Title, Abstract, Author Keywords, and Keywords Plus). For Scopus, they were searched within the TITLE-ABS-KEY field (title, abstract and keywords). The numbers of records resulting from the searches are shown in Table 7. Table 7: Number of records retrieved from Web of Science and Scopus after searches using the search strings and combinations about life history traits. Search Keyword search and terms Records in Web Records in of Science Scopus #1 Species 7,392 1,988 #2 Traits 15,118,774 13,250,962 #3 #1 AND #2 2, Local herpetological journals The search was performed on contents of journals listed in the Appendix B using similar procedures as described for the review of studies reporting chemical effects. The same search strings and combinations as described in section were used to search within the titles of the articles published in those journals. The keywords in the search string Traits were adapted to the primary language of the journal. When a search tool was not available, we reviewed the tables of contents by copying them in a text processing software (Microsoft Word) and running the searches keyword by keyword using the search tool of the software. A total of 19 relevant records were found in the contents of these local herpetological journals Review of literature about relevant life history traits for population modelling of selected species Following the same procedure as for the studies reporting effects of chemical exposure, the retrieved records were initially subject to a title only screening to remove duplicates and obviously irrelevant 28 EFSA Supporting publication 2017:EN-1251

29 records. The screening resulted in the selection of 765 records that were exported to an EndNote file (LIFE_HISTORY) (Figure 2). These records were uploaded to the project created in DistillerSR for literature review, and correspond to the RefIDs 116, 225, 2085, 4295, 4331, 4396, 4397 and 4505 to 5262 of the Ref_IDs dataset available as electronic supplementary material. The full text retrieval was conducted using the same sources as described for chemical exposure studies (see section 2.2). As it happen with those studies dealing with effects of chemical exposure, the majority of cases in which full text was not made available corresponded to congress abstract of conference papers (without enough degree of detailed information), old reports not published on the web, theses or dissertations (likely to be published as journal papers), and papers from local sources. Figure 2: Flow chart of the study selection process corresponding to references reporting life history traits of the model species. The complete process of literature review involving the 543 retrieved full texts included the same three sequential levels as for studies reporting effect of chemical exposure: the two initial levels of screening (level 1: initial screening; level 2: study appraisal) were conducted using the same data models as for studies on chemical exposure, shown in Appendices C and D. However, for life history trait studies, a threshold of QS=0 was determined for data inclusion. We decided to apply a different threshold because the number of questions of the study appraisal that applied to this type of studies was generally low, and thus the penalty exerted by each negative answer was too high. A QS>0 meant that the number of affirmative answers in the appraisal was higher than the number of negative answers. For data extraction (level 3: life history traits), a new data model adapted to the type of information to be retrieved was created EFSA Supporting publication 2017:EN-1251

30 Data extraction The data model named Life history trait, shown in Appendix G, was designed to collect the information on relevant life history traits of the selected species for population model parameterization. The structure of that data model, which is what ultimately determined the type of information to be collected, was produced in accordance with the Working Group of the EFSA PPR unit elaborating the scientific opinion on amphibians and reptiles Tools for appraisal of the systematic review Besides appraisal of each particular reference to determine its relevance and reliability, we considered how the appraisal of the entire review could be conducted. We focused primarily on the section related to studies reporting effects of chemical exposure, as this was the main part of the project (addressing three of the four objectives) and the most sensitive part to potential deficiencies in methodology. With this purpose, we looked at recommendations published by EFSA for critically appraising systematic reviews (EFSA 2015), and considered how the methodology of our review would fit into such recommendations. In particular, we focused in the critical appraisal tools for assessing the quality of systematic reviews and extensive literature reviews, which apply to the present project, and are detailed in appendices A and D of EFSA (2015), respectively. It must be noticed that the purpose of this section is not to appraise our own work, but to review whether elements for critically appraising it are available Appraisal of systematic review of data According to EFSA (2015), one of the main situations when it is necessary to appraise the methodological quality of a systematic review is when EFSA outsources systematic reviews and needs to appraise the quality of the systematic reviews performed by the external contractors. Table 8 reflects the list of questions proposed in Appendix A of EFSA (2015) for appraisal of systematic reviews and how they apply to the review presented in this report. Table 8: Elements for appraisal the systematic review according to the critical appraisal tool published by EFSA (2015). # 1 Appraisal question Element for appraisal A Review question and eligibility criteria for study selection A1 Was the review question clearly Inclusion criteria were defined according to a PECO approach formulated? based on the objectives of the project (section ) Were the eligibility criteria related to A2 study characteristics appropriate to Eligibility criteria for selecting studies were established a priori answer the review question and clearly based on the review question, and resulted in inclusion criteria defined a priori? A3 If report characteristics were used as eligibility criteria, are they appropriate to meet the review question? If applied, Report characteristics were not used as eligibility criteria were these criteria defined a priori? B Search process B1 Was the extensive literature search performed in an appropriate way? See Table 9 C Study selection process C1 A rapid assessment was conducted by one reviewer (in Were preventative steps taken to EndNote), followed by full-text examination conducted minimise bias and errors in the study independently by two reviewers (in Distiller) combining a more selection process? experienced and a less experienced reviewers within the team C2 Were the results of the study selection Only 12 out of the 1760 references reporting chemical exposure 1 As identified in Appendix A in EFSA (2015) EFSA Supporting publication 2017:EN-1251

31 D D1 D2 E E1 E2 F F1 F2 G G1 H H1 process consistent with the eligibility that reached the data extraction level were further excluded by criteria previously defined? not meeting the inclusion criteria Data extraction from the included studies Data models for data extraction were designed a priori to identify information that needed to be extracted Was data extraction carried out Specific fields were included in the data model Endpoint study appropriately and adequately? Was the to identify how data were translated from the paper to the data approach defined a priori? model, which the funding sources were, and if there were potential or declared conflicts of interest Were preventative steps taken to minimise bias and errors in the data extraction process? Data extraction of each paper was carried out by a single reviewer, so a procedure for resolving disagreements was not relevant at this level. Final datasets were reviewed and homogenised by a single reviewer. Review forms included a checkbox to flag instances in which introduced data were doubtful. The occurrence of duplicated records that could have passed unnoticed was conducted on the finally selected records by looking for matches in title and authors, and also on the database including all extracted data by looking for matches in the combination substance*species tested Assessment of the methodological quality of the studies included in the review Was the methodological quality of the individual studies, which were included in the review, appropriately and A study appraisal was conducted and reliability check was done adequately appraised? Was the approach defined a priori? Although study appraisal was conducted initially by a single reviewer per reference, those references excluded because they did not reach the minimum relevance and reliability but were Were steps taken to minimise bias and close to reach the inclusion threshold were assigned to an errors when appraising the alternative reviewer for re-appraisal. If the second appraisal methodological quality of the studies resulted in inclusion, the reference passed to the level 3 included in the review? (strategy to minimize the loose of data). The appraisal was tested by assessing the relevance of the established minimum quality threshold (0.6). All these procedures are explained in section Data analysis and synthesis of results Was the analysis and synthesis of the individual effect estimates properly undertaken? Was the approach defined a priori? If there was the need or opportunity to perform additional analyses (e.g. sensitivity or subgroup analyses, metaregression), were they performed? Was the approach defined a priori? Was risk of publication bias addressed? Was the approach defined a priori? Did the conclusions reflect the results of the review and any limitation in the We assumed the granularity in the data results (data categories 1 and 2, see section 2.2.4) and designed different procedures for data extraction and analysis depending on the type of data. When not directly reported as endpoint values (data category 1), individual effects were determined through the calculation of effect sizes. In these cases (data category 2), a field in the data model was established to identify statistically significant effects Additional statistical analyses were conducted. Subgroups for running these analyses were defined a priori and adapted to the objective of each analysis. Details on statistical analyses are provided in section 2.5 Data analysis and synthesis of results Risk of publication bias was addressed by identifying cases when potential conflicts of interest could occur. Duplicated information was identified (an inclusion criteria was set up for this purpose). To minimize location bias, we extended our searches of information to local sources. Other potential sources of bias were not addressed Data analysis and synthesis of results The conclusion include those aspects of the objective in which data limitation has prevented from reaching relevant 31 EFSA Supporting publication 2017:EN-1251

32 I I1 I2 process? If a protocol was provided, are appropriate justifications given for any described deviations from the protocol? Have any competing interests been identified? conclusions, as well as recommendation for future research based on data needs Additional considerations Deviations from the proposed protocol were minimised, but changes were previously discussed and approved by EFSA No, it is declared in the acknowledgements Appraisal of extensive literature reviews According to EFSA (2015) and as with systematic reviews, one of the main situations when it is necessary to appraise the methodological quality of an extensive literature review is when it is conducted as part of a systematic review. EFSA outsources systematic reviews and needs to appraise the quality of the systematic reviews performed by the external contractors. Table 9 reflects the list of questions proposed in Appendix D of EFSA (2015) for appraisal of extensive literature reviews and how they apply to the review presented in this report. Table 9: Elements for appraisal the extensive literature review according to the critical appraisal tool published by EFSA (2015). # 2 Appraisal question Element for appraisal A Assessing the search strategy A1 Was the review question appropriately The review question was translated into key elements following translated into search concepts? the PECO approach (section ) Because we did not want to restrict the searches, we generated A2 large search strings that resulted in an excess of retrieved Was the search string an optimal records, most of which were excluded in the title only screening combination of the search concepts for (Figure 1). sensitivity and precision? Search strings and combinations were defined to meet the inclusion criteria (section ) A3 A4 A5 A6 A7 A8 A9 Were the appropriate free-text terms (i.e. (terms in the title and abstract) identified for each search concept? Were appropriate controlled terms (subject headings) identified for each search concept and information source used (when applicable)? Was a pilot study carried out (when applicable)? Was the appropriate spelling used? Was the appropriate syntax used? Were the appropriate line numbers used? Was the use of Boolean and proximity operators appropriate? The search string included species common and scientific names, and terms were hierarchized (e.g. pesticide-herbicide, anura-bufo) We identified as controlled terms some keywords that would lead to ambiguity (e.g. Pb was used instead of lead ) Yes, pilot study for checking quality of the search protocol was conducted and used to improve the protocol (section ) We did not conduct a search with intentional spelling errors to check whether appropriate spelling was used. However, we confirm that records corresponding to all keywords listed in the Chemicals search string were retrieved (although we assume that this does not guarantee the absence of spelling errors) We used the truncation symbol * for name roots leading to different declinations of the same word, and checked that it did not lead to errors in the searches Different line numbers were used, as shown in Tables 1, 2 and 7 Boolean operator OR was used to separate all terms within each search strings. Results of each search string were then combined with the Boolean operator AND in order to fulfil the Outcome criterion of the PECO approach 2 As identified in Appendix D in EFSA (2015) EFSA Supporting publication 2017:EN-1251

33 A10 A11 A12 B B1 Were limits appropriately used? Limits were not used in order to retrieve as much information as possible Were search filters (if used to identify study designs) appropriately used? Search filters were unnecessary and therefore were not used Was the search strategy correctly Yes, as explained in the sections describing the procedure used adapted for each database used? for review of each literature source Assessing the information sources searched Assess if the search was extensive enough, i.e. assess if the right (relevant In order to make the searches as extensive as possible, several and reliable) combinations of sources were used, all of which included different types of information sources were records searched 2.5. Analysis of data of effects of chemicals on amphibians and reptiles Processing of data All compiled data relative to characterization of chemical effects on amphibians and reptiles are presented in the endpoint_study dataset available as electronic supplementary material. All records in the endpoint_study dataset (i.e. all instances of all references reaching level 3 of the review) were initially eligible for effect characterization. However, some data records were finally excluded because they were unsuitable to be used in any of the analyses. These data records were: Records reporting the effects of chemical mixtures. Although these records should have been excluded because of the impossibility of filling one of the mandatory fields of the data model (i.e. Substance), they were included in the data model to offer the possibility of running further analysis in the direction of chemical mixture toxicity assessment. Records reporting the effects measured in laboratory studies in which a stressor was added to the experimental enclosure, or in which housing conditions were considered suboptimal. The reason for excluding these records was that it was impossible to elucidate the effect of the stressor or of the stressful conditions on the outcome of the study, and that each stressor or stressful condition would presumably influence the outcome of the chemical exposure in a different way. Therefore, comparisons including these records could have provide erroneous conclusions. This did not apply to mesocosm or field studies, in which exposure conditions simulate or represent natural scenarios, and therefore the presence of stressors is part of the study conditions. Records reporting effects that were measured after a recovery time following the exposure (for continuous exposures). In these cases it is not possible to relate the observed effect to a given exposure time, and influence of post-exposure periods may differ among substances and organisms, which makes these records unsuitable for comparison. For the rest of the records, effect characterization was done in a different way depending on data categories defined in section For data category 1, the endpoint value (LC 50, LD 50, NOEC, etc.) was directly used as a measure of effect characterization. For data category 2, we used different, nonexcluding approaches to characterize the effect depending on the study design and parameter to record: Calculation of effect sizes (applicable to all studies of data category 2): we calculated Hedge s values as the metric of standardized effect size following the methodology described in Egea- Serrano et al. (2012). By comparing mean response values between the control and a given exposed group, this parameter provides a measure of the magnitude of the treatment effect while adjusting for small sample sizes (Rosenberg et al. 2000). To calculate Hedge s value (d) at each concentration, we used the following equation adapted from Gurevitch et al. (1992): d = [(Ȳ e Ȳ c ) / s] * log J 33 EFSA Supporting publication 2017:EN-1251

34 Where Ȳ e and Ȳ c are the mean intensities of the response (e.g. mortality rate, body mass) recorded at the exposed and control groups, respectively, and s is the pooled standard deviation of the control and exposed groups, calculated as: s = {[(N e 1) * σ e 2 + (N c 1) * σ c 2 ] / [N e + N c 2]} where N e and N c are the samples sizes in the exposed and control groups, respectively, and σ e and σ c are the standard deviations of the means Ȳ e and Ȳ c. Standard deviations (σ) were directly used when reported in the papers. When they were not reported, we estimated σ following different methods depending on the type of variability measure reported: When standard errors (SE) were reported, σ was estimated as a function of SE and the sample size (N), as: σ = SE * N When confidence intervals were reported, a normal distribution was assumed to assign a probability value to the confidence interval. The limits of the confidence interval for a probability 1 - α (X -α/2, X α/2 ) can be calculated as: Ȳ ± Z α/2 * σ / N Where Ȳ is the sample mean and Z α/2 the critical probability value leaving outside a proportion of the population equal to α/2. For instance, for a 95% confidence interval (critical value Z α/2 = 1.96), σ was estimated as: σ = {[ N * (X -α/2 -Ȳ) / ] + [ N * (X α/2 -Ȳ) / 1.96 ]} / 2 Although we acknowledge that this estimation of standard deviation may lead to some errors because of the existence of non-normally distributed data, the impact that these errors in probability estimation might have in the final effect size calculation is expected to be small. When ranges were reported, we proceeded the same way as for confidence intervals, assuming that the range was actually indicating a 95% confidence interval. This assumption is consistent with the overall assumption of a type I error probability of 5% (i.e. an alpha value of 0.05). The term log J is a corrector that removes biases because of small sample sizes, and approaches 1 as sample size increases. It is adapted from the correction factor J(m) used by Gurevitch et al. (1992); because of the strong variability across reviewed studies in sample size, we used the log J, which is a logarithmic expression of the J(m) factor. It is calculated as: log J = log ({1 [2 / (N e + N c )]} * 10) Every single response associated with a concentration and an exposure time that had a variability measure estimator resulted in a d value that was used as effect size estimator. We then obtained a time-corrected d value (d(t)) estimator by correcting the d value as a function of the exposure duration as follows: d(t) = d / ln(t) with t being the duration of the exposure in days for continuous exposures. This normalization has the problem of assuming that the outcome of effects evolves in a linear manner over time, which is very unlikely. However, some sort of linearity in the outcome of effects over time can be assumed for short-term exposures. The logarithmic model in time correction will minimize the errors derived from the lack of linearity in outcome of effects over time for longterm exposures EFSA Supporting publication 2017:EN-1251

35 Effect sizes of data retrieved from single dosage studies were considered time-independent and were not corrected by any duration. In these cases, d was directly used as effect estimator. Calculation of LC/LD 50 or LC/LD 10 (applicable to lethal effects). If a minimum of three concentrations or doses were tested in a given experiment, mortality rates were plotted against each corresponding exposure concentration or dose. If dose-dependence was clearly observed from the plotted values, a probit regression model of the mortality on the exposure level was obtained. If the model was significant at the level p<0.05, LC/LD 50 and/or LC/LD 10 were estimated as the exposure level associated with an estimated mortality of 50% or 10% of the exposed individuals, respectively. Benchmark probability levels were only assumed when they were within the range, or less than 5% away, of percentage effects recorded during the study (e.g. LC 50 calculation was considered valid only when at least one of the tested treatments caused a minimum mortality rate of 47.5%). Calculation of EC/ED 50 or EC/ED 10 (applicable to studies reporting sublethal effects). For sublethal effects reported as rates (e.g. percentage of anomalous individuals), we estimated EC 50 /ED 50 /EC 10 /ED 10 following the same procedures as described for LC/LD. For the rest of the sublethal responses, we first obtained the percentage of response compared to the control as a measure of effects. If a minimum of three values were available, these percentages of response were plotted against their corresponding exposure concentrations/doses. If dosedependence was clearly observed from the plotted values, different regression models were run in order to look for the best adjustment. The considered regression models were linear, quadratic, logarithmic, potential and exponential. For the last three, percentage effects and exposure concentrations were transformed before the regression according to the equation: Transformed value = log (1 + original value) The following conditions needed to be met in order to consider a model as valid for endpoint estimation: The model is significant at the level p<0.05. The interception value is between -5 and 5, indicating that the percent effect in conditions of no exposure is not significantly different from zero (with an error margin of ± 5%). Because percent effects are calculated under the assumption that effects in conditions of no exposure is null, any regression for which the intercept value (i.e. the percent effect for a concentration = 0) deviates more than 5% from the null value was considered to not properly fit the relationship between concentration and effect. The benchmark probability level(s) to be considered (i.e. 10% and/or 50%) were within the range, or less than 5% away, of percentage effects calculated from the data recorded during the study. The equation of the regression model showing the best adjustment (highest R 2 ) out of all models meeting these three requirements was selected to estimate the corresponding endpoint value. This was achieved by calculating the concentration or dose associated with a percentage response of 50% and/or 10%. Calculation of no observed effect concentrations (NOEC) or no observed adverse effect level (NOAEL) (applicable to all studies with at least one treatment statistically different and one treatment statistically non-different from the control). NOEC or NOAEL were identified as the highest concentration or dose, respectively, which effect was not significantly different from that associated with the control treatment. NOEC or NOAEL were considered only if there was at least one tested, higher concentration causing an effect significantly different from that recorded in the control, and if there was a logic in the dose-dependence occurrence of statistically significant differences (i.e. all concentrations or doses causing effects nondifferent from controls were lower than all concentrations or doses causing effects significantly different from those recorded in controls) EFSA Supporting publication 2017:EN-1251

36 A database with all valid effect characterization values (i.e. d(t), d, or endpoints either directly taken as category 1 data or estimated from category 2 data ) was used for further data analysis. Data corresponding to endpoint values were time-corrected in an analogous way as described above for the d values: (example for LC 50 ): LC 50 (t) = LC 50 / ln(t) Time-corrected values were used for identification of sensitive life stages and for comparison of endpoints observed in laboratory and field studies, which are explained in the next sections Identification of sensitive amphibian and reptile life stages The database with effect characterization values was split in different subgroups resulting from the combination of the categories of the variables shown in Table 10. Table 10: Variables and categories within each variable used for creating subgroups of the database with effect characterization values. Variable Categories Additional notes Class and life stage Study type Amphibians: -Embryo (until hatching) -Hatchling (until free swimming larval stage) -Larva (until metamorphosis climax) -Metamorphic (until end of metamorphosis) -Juvenile (until attainment of sexual maturity) -Adult Reptiles: -Embryo -Juvenile (until attainment of sexual maturity) -Adult Laboratory Mesocosms Field Given the overall limitation of data, we did not consider splitting classes into smaller groups (orders, families). This approach has the potential problem of comparing data from more and less sensitive species, but the overview of the dataset shows that large percentage of the records comes from a limited number of species. Splitting into smaller groups would have reduced the amount of data. In the comparison of field vs. laboratory studies, however, alternative analysis to the all-species comparisons are conducted, when possible, with data collected from concurrent species (see section 2.5.3) Mesocosm studies were identified as semifield studies in the data model Endpoint study and used as an additional category for comparison of data from laboratory and field studies Substance (not listed, see dataset) The option of pooling substances by chemical classes, uses or modes of action was disregarded because of even if pooling is done on the basis of these criteria, data show that toxicity among substances can be very different. Pooling would have therefore compromised the validity of the comparison Environment and exposure route 3 Aquatic: -Waterborne (mass volume -1 ) -Water overspray (mass surface -1 ) -Oral (food content) (mass mass food -1 ) -Oral (single dose) (mass body mass -1 ) -Oral (repeated dose) (mass body mass -1 day -1 ) -Sediment spiking (mass mass sediment -1 ) -Maternal transfer (mass embryo mass -1 ) -Intraembryonary (mass volume of injected The way of splitting exposure routes had to do not only with the exposure route itself but also with the units in which the exposure levels were reported 3 Units for concentration or dose measurement are indicated in italics after each exposure route EFSA Supporting publication 2017:EN-1251

37 Effect characterization value 4 Type of response fluid -1 ) Terrestrial: -Overspray (mass surface -1 ) -Substrate overspray (mass surface -1 ) -Spiking of substrate (mass mass substrate -1 ) -Spiking of water substrate (mass volume -1 ) -Oral (food content) (mass mass food -1 ) -Oral (single dose) (mass body mass -1 ) -Oral (repeated dose) (mass body mass -1 day -1 ) -Inhalation (mass volume -1 ) -Maternal transfer (mass embryo mass -1 ) Aquatic/Terrestrial: -Intramuscular (mass body mass -1 ) -Intraperitoneal (mass body mass -1 ) -Intraveneous (mass body mass -1 ) -Subcutaneous (mass body mass -1 ) In vitro (mass volume -1 ) d d(t) EC 10 (t) EC 50 (t) ED 10 (t) ED 50 (t) LC 10 (t) LC 50 (t) LD 10 (t) LD 50 (t) LOEC(t) LOAEL(t) NOEC(t) NOAEL(t) Mortality Abnormalities (including gross pathologies and histopathology) Behaviour (including activity) Development Endocrinology and reproduction Genotoxicity Growth Haematology Immunology Metabolism and biochemistry Neurotoxicity Physiology The assignation of each record to each type of response was based on the combined evaluation of answers provided to the fields Type of effect observed/measured, Further description of observed/measured effects, System level categorisation of effect and Target tissue in the Endpoint study data model (see Appendix E) Identification of sensitive life stages was achieved through a comparison of effects among different life stages. Only data collected from laboratory studies were considered in this part of the study. Data were analysed using different protocols depending on the effect characterization value: Time-corrected endpoints were compared among subgroups with different life stages matching all the other splitting categories. A Generalized Linear Model (GzLM) was run in each case with the time-corrected endpoint as dependent variable and the life stage as predictor, considering linear responses after checking for data normality (McCullagh and Nelder 1989). Identification of most sensitive stage was done by comparing marginal means of those GzLM for which the predictor effect was found significant. If more than two life stages were compared, pairwise comparisons were conducted using the Least Significant Difference test. 4 Differences in data analysis procedure apply depending on the effect characterization value EFSA Supporting publication 2017:EN-1251

38 For the analysis of d and d(t) values, linear regressions were conducted within each subgroup with the d or d(t) value as dependent variable and the log-transformed exposure concentration or dose associated with each d or d(t) value as explanatory variable. When a significant regression was found, the slope (i.e. non-standardized regression coefficient) was recorded as a toxicity index, reflecting how the size of effect changes for every unit of increase in the log-transformed exposure level. Thus, the higher the toxicity index, the higher the sensitivity of the analysed group. When a regression was non-significant, the toxicity index was interpreted as null. Toxicity indexes were compared among subgroups with different life stages matching all the other splitting categories. For each pair of compared data, the life stage with the highest toxicity index was recorded as the most sensitive one (Figure 3). Figure 3: Example showing a comparison of the regressions run on data from hatchling and larval amphibians exposed to arsenic in the water showing effects on growth. Larval data fit to a significant regression model resulting in a toxicity index of (y = 0, x, R2 = 0.736, p = 0.006), whereas hatchling data did not fit to a significant model (y = x, R2 = 0.008, p = 0.801). In this example, larvae were recorded as being more sensitive than hatchlings. For each pair of life stages, the pool of comparisons resulting in a significant difference (i.e. a significant marginal mean comparison in the GzLM or a difference in toxicity indexes estimated from regressions of the effect sizes on exposure levels) was analysed for the general trend. With this purpose, relative frequencies of stages appearing as the most sensitive in each pair were compared with a chi-square test with the null hypothesis of the compared life stages occurring as the most sensitive with the same frequency. Finally, the same type of analysis of frequencies was conducted for each type of response separately, in order to investigate trends in life stage-related sensitivity as a function of the analysed responses EFSA Supporting publication 2017:EN-1251

39 Comparison of endpoints observed in laboratory with field data The comparison of sensitivity shown by individuals exposed in laboratory, mesocosm or field studies was done in a similar way as the comparison of sensitivity among life stages. Subgroups were defined as shown in Table 10 with only two modifications: Amphibian life stages were classified as embryo, larva, juvenile and adult. Life stages hatchling, larva and metamorphic defined in Table 10 were pooled together for this part of the study. This allowed for increasing the number of potential comparisons through reduction of the number of subgroups. Although some differences existed in sensitivity between hatchlings, larvae and metamorphic individuals (see section 3.2), these differences did not justify the maintenance of so many different life stages in an analysis that was not really focused on life stage comparison. Type of responses were initially maintained as in Table 10, but when there were not enough data to run comparisons, all sublethal responses were merged into a single response category (sublethal). This procedure applied to the comparisons involving data from the same species. Statistical analyses were conducted using the same protocol as for the comparison of life stages, but in this case all data were considered and the comparisons were made between subgroups differing in the study type category (i.e. laboratory, mesocosm, field) and matching all the other categories. Additionally, the initial analysis conducted without considering the species from which data had been collected was repeated species-by-species. When data were not enough to run the analysis for any particular species, the tests were repeated pooling all sublethal responses in a single category. In order to estimate assessment factors allowing for extrapolation of laboratory generated data to field scenarios, we used all endpoints (no data from d/d(t) regressions were used here) for which there were concurrent values from laboratory- and mesocosm- or field-generated data matching all categories defining subgroups. Paired (laboratory vs. mesocosm, laboratory vs. field), time-corrected endpoint values were plotted and a regression model with the laboratory values as explanatory variable and the mesocosm or field values as dependent variable was run. The regression was then repeated considering pairs of values that had been generated using the same species. The assessment factors were calculated from the regression results, using the slope of the regression as a quantitative indicator of the relationship between effects shown in laboratory and in mesocosm or field conditions Comparison of sensitivity with surrogate taxa Sensitivity of amphibians or reptiles to pollutants was compared to that of other vertebrate groups that are commonly evaluated in ERA and are therefore used as surrogates for herpetofauna (i.e. fish in the aquatic toxicity assessment, and birds and mammals in the terrestrial toxicity assessment). This analysis was based on the comparison of endpoint values only (i.e. toxicity indexes used in the two previous sections could not be applied in this case as no comparable data existed for surrogates). Information about endpoint values for surrogates was retrieved from three sources: A series of databases shared by EFSA: ecotoxicological properties of active substances of PPP, pesticide endpoints and chemical hazards, chemical hazards to birds and mammals, and comparison of NOEC-EC values. Toxicology Data Network (TOXNET) included in the National Library of Medicine s Hazardous Substances Data Bank (HSDB) of the U.S. National Library of Medicine (available at Environmental Residue-Effects Database (ERED) of the U.S. Army Corps of Engineers (available at For this part of the study, only data retrieved from experiments run with technical grade or high purity substances were considered. For each endpoint value in our database of effect characterization values 39 EFSA Supporting publication 2017:EN-1251

40 for amphibians and reptiles, we searched all possible correspondences from surrogates matching the technical grade purity or active substance, endpoint, exposure route and frequency, and exposure time (except for the single dosage-derived endpoints, for which exposure time was not considered). Open ended (greater than or less than) values were not used in the analyses in order to make comparisons as robust as possible through the use of parametric correlations. Three sets of comparisons were made: aquatic toxicity values, terrestrial toxicity values for amphibians, and terrestrial toxicity values for reptiles. Within each set, two correlational analyses with different purposes were conducted: The first analysis was run with the purpose of comparing sensitivity between amphibians or reptiles and their surrogate taxa. For each comparable pair (i.e. data with concurrent endpoint, substance, exposure route and frequency, and, when relevant, exposure time) the most sensitive species on each side (amphibians/reptiles and surrogates) was selected, taking the most sensitive endpoint value when more than one value was available. For terrestrial amphibians and reptiles, specific comparisons were conducted considering data from birds only, from mammals only and from birds and mammals together, selecting always the lowest endpoint out of the pool of data under consideration. The second analysis was conducted with the aim of evaluating the degree of coverage that surrogate taxa confer to amphibians and reptiles in terms of sensitivity to toxic effects. For this analysis, surrogate data were considered only from species commonly used in environmental risk assessment, including the rainbow trout (Oncorhynchus mykiss) among fish, the mallard (Anas platyrhynchos), northern bobwhite (Colinus virginianus) and the Japanese quail (Coturnix japonica) among birds, and the rat (Rattus sp.), mouse (Mus musculus), guinea pig (Cavia procellus) and rabbit (Oryctolagus cuniculus) among mammals. For amphibians or reptiles, all tested species with a corresponding data on the surrogate side were included. The most sensitive endpoint value of the selected species was used when more than one was available. Each comparison consisted of a linear regression of values retrieved from amphibians or reptiles and their surrogates, with the Pearson s correlation coefficient used as indicator of a significant relationship between series of data and the slope of the significant models interpreted as a quantitative measure of the relative difference in sensitivity between groups. The relative frequency of cases in which one or another group of species was more sensitive (i.e. had a lower endpoint value) was analysed with Wilconxon s rank tests. Because there were almost no endpoint matches involving long-term toxicity values (only three values), only short-term toxicity data were considered. As data were paired, analyses were initially conducted with all possible data pairs related to short-term toxicity instead of splitting them by specific endpoints. An additional analysis was further conducted including 96h-LC 50 values only (for aquatic toxicity assessment) as this specific parameter was the most repeated one in the list of compared endpoint values Development of species sensitivity distributions Species sensitivity distributions (SSD) were created for combinations of substance*endpoints for which data for at least six amphibian or reptilian species were available. The minimum sample size is consistent with the three to six species suggested by Baker et al. (1994), but lower than the 15 to 50 species shown by Newman et al. (2000) as necessary to provide accurate estimates. The estimates of Newman et al. (2000) are based on comparisons of several methods to calculate hazard coefficients from SSD, and are therefore more reliable than the proposed minimum sample size of six. However, we preferred to use this less conservative approach in order to present as much information as possible, as we did not expect a long list of substances having been tested in six or more amphibian or reptilian species. SSD were constructed according to the standard approach, considering a log-normal or log-logistic distribution of the variable (Newman et al. 2000, Grist et al. 2002). Only endpoint values retrieved or calculated from experiments run with technical grade or high purity substances were considered, with 40 EFSA Supporting publication 2017:EN-1251

41 two exceptions: butachlor and glyphosate. For the former, all data useful to create SSD came from studies conducted with emulsifiable compounds with butachlor purity between 50 and 60%. For the latter, the vast majority of data useful to SSD came from experiments using formulations with different purities (commonly between 29% and 48%). Especially for amphibians, concerns on toxicity of glyphosate-based formulations have traditionally focused on co-formulants rather than on the active ingredient (Mann and Bidwell 1999), resulting in a broad use of formulations instead of technical grade glyphosate salts in ecotoxicity tests. Therefore, for these two pesticides (i.e. butachlor and glyphosate), because there were not enough data from high purity materials to calculate SSDs, we modeled SSD using toxicity data of formulated products. Consequently, these SSDs will be useful to investigate the relative sensitivity of a certain amphibian species with regards to other amphibians, but not to compare SSD endpoints (see below) with those from other SSD (e.g. fish SSD) run with the active ingredients themselves. A logistic regression between the endpoint concentration and the cumulative number of species affected was obtained using the software available from the United States Environmental Protection Agency ( (Neter et al. 1990, Posthuma et al. 2002, EPA 2005), which generates a probability table and a plot showing a central tendency and prediction intervals. The 5 th percentile of probability associated with each SSD was recorded as the hazard coefficient protecting 95% of the species in the group (HC 5 or HD 5 ). This standard approach assumes that data are derived from a random sample of species from the statistical population (Poshtuma et al. 2002), which has been considered to be inappropriate (e.g. Maltby et al. 2005) because the species for which toxicity data are published come from various non-random mechanisms (e.g. abundance, common laboratory use or expected sensitivity). O Hagan et al. (2005) proposed to use expert judgement to correct potential bias in the species selection, but this is very complicated in practice for amphibians and reptiles because the proportion of studied species in ecotoxicology is very small (Schiesari et al. 2007) and inferences on their global position in sensitivity distributions across the entire classes are not possible. Although sensitivity is not necessarily related to taxonomy, we can expect that similarities in physiology can account for some kind of common responses to chemical exposure across phylogenetically related individuals. Therefore, for each constructed SSD we provided information on the taxonomical diversity (i.e. number of represented genera and families) as a surrogate for randomness, in terms of sensitivity, of sampled species. HC 5 or HD 5 values resulting from SSD were compared, when possible, with similar values retrieved from surrogate taxa. Because the availability of data for comparison of this parameter was very limited, we used the data extracted for comparison of sensitivity from EFSA databases, TOXNET and ERED (section 2.5.4) to create SSD of surrogates and calculate HC 5 /HD 5 values for comparison with those of amphibians or reptiles. For comparison of HC 5 or HD 5 values from amphibians and surrogates, data were transformed to fit a normal distribution and a t-test for paired samples was conducted. 3. Results and Discussion 3.1. Reviewed literature about effects of chemicals on amphibians and reptiles Figure 4 summarizes the outcome of the review of the 3642 references for which full texts were retrieved EFSA Supporting publication 2017:EN-1251

42 Figure 4: Progress of the different steps of the literature review with indication of exclusion criteria applied to excluded references. The 1332 references finally used for data extraction generated a total of data records, to which another 1622 data records as endpoint values estimated from the information contained in the paper were added. After applying the filters for valid data explained in section and calculating d or d(t) values when possible, the final database with all valid effect characterization values consisted of records (4989 corresponding to endpoint values and corresponding to d or d(t) values) Identification of sensitive amphibian and reptile life stages The compiled data allowed for effectively running 485 comparisons between life stages corresponding to the same taxon, substance, endpoint, exposure environment and route, and type of recorded response. The outcomes of all specific comparisons are shown in the Appendix H. These comparisons included data from 102 different chemical substances, out of which 16 are pesticides currently approved for use in the EU (five herbicides, four fungicides including copper compounds as a single product, and seven insectides), comprising together 116 out of the 485 comparisons. Out of all comparisons, 481 referred to amphibian aquatic stages, which means that the identification of sensitive life stages within reptiles or terrestrial amphibians was not possible due to the scarcity of data. No comparisons were possible involving amphibians exposed through food in the aquatic environment, as neither pre-larval stages nor metamorphic individuals ingest food, and there were virtually no data on juvenile or adult amphibians fed in the aquatic environment suitable for comparison with data from larvae EFSA Supporting publication 2017:EN-1251

43 The only comparison possible for terrestrial amphibians involved LD 50 data after overspraying individuals with glyphosate-based products (Appendix H). This comparison resulted in a higher sensitivity of juveniles as compared to adult individuals (Wald s Χ 2 = 5.543, 1 d.f., p = 0.019), which is consistent with the higher expected chemical absorption because of the higher surface to volume ratio in juvenile individuals. Nevertheless, it is obvious that more data are necessary to come to a conclusion regarding inter-stage sensitivity to pollutants in terrestrial amphibians. The summary results of the 481 comparisons made on aquatic amphibian stages are shown in Table 11. Hatchlings appeared in general as the most sensitive life stage, with the exception of their comparison with metamorphic, which included only three data pairs. When compared to larvae, however, the relative frequency of cases showing increased sensitivity of hatchlings was not significantly different from the frequency of cases showing decreased sensitivity. However, hatchlings and not larvae were more sensitive than embryos in a number of cases significantly higher than the number of cases that embryos were more sensitive. Besides hatchlings, larvae were always more sensitive than the other life stages, although when compared with embryos the percentage of cases in which sensitivity of larvae was higher than that of embryos was not statistically different from the percentage of cases in which embryos resulted more sensitive than larvae. Embryos and adults were in general less sensitive. The comparisons involving metamorphic or juveniles were little relevant because of the low number of comparable series of data. Table 11: Results of the comparison by pairs of life stages made on amphibians exposed to different waterborne substances in the aquatic environment. Each comparison was done on data referred to concurrent substances, effect characterization parameters and response types. Results of the chi-square test for comparison of frequencies of cases in which one or another life stage was identified as more sensitive are shown (for detailed data, see Appendix H). Compared life stages N Comparison results Sig. (Χ 2 test) Embryo vs. hatchling 79 Embryo more sensitive: 7.5% Hatchling more sensitive: 23.8% Embryo vs. larva 208 Embryo more sensitive: 20.1% Larva more sensitive: 20.6% Embryo vs. metamorphic 6 Embryo more sensitive: 16.7% Metamorphic more sensitive: 33.3% Embryo vs. juvenile 4 Embryo more sensitive: 25% Juvenile more sensitive: 0% - Embryo vs. adult 21 Embryo more sensitive: 33.3% Adult more sensitive: 14.3% Hatchling vs. larva 93 Hatchling more sensitive: 20.0% Larva more sensitive: 16.3% Hatchling vs. metamorphic 3 Hatchling more sensitive: 0.0% Metamorphic more sensitive: 33.3% - Hatchling vs. juvenile 1 No differences in sensitivity - Hatchling vs. adult 14 Hatchling more sensitive: 28.6% Adult more sensitive: 0% - Larva vs. metamorphic 12 Larva more sensitive: 25.0% Metamorphic more sensitive: 0% - Larva vs. juvenile 4 Larva more sensitive: 25.0% Juvenile more sensitive: 0% - Larva vs. adult 35 Larva more sensitive: 31.4% Adult more sensitive: 5.7% Metamorphic vs. juvenile 0 - Metamorphic vs. adult 0 - Juvenile vs. adult 1 Juvenile more sensitive: 100% EFSA Supporting publication 2017:EN-1251

44 Several studies point to the newly hatched individual as the most sensitive stage in amphibians (Herkovits and Pérez-Coll 1993, Ortiz-Santaliestra et al. 2006). Hatchling individuals possess external gills, which increase the surface area in contact with the environment. Once gills are internalized (in anuran amphibians), the surface for potential diffusion of pollutants decreases, as surface to volume ratio also decreases with development. Embryos are protected by the egg membrane and the jelly envelope, which is known to prevent absorption of certain contaminants to the inner part of the egg (Marquis et al. 2006, Edginton et al. 2007). Higher sensitivity of embryos was particularly found for some substances, especially lead (Pb 2+ ). The majority of inter-stage comparisons of toxicity of this metal resulted in a higher sensitivity of embryos compared to hatchlings or larvae, in contrast to the general trend (Appendix H). To a lesser extent, also ammonium (NH 4 + ) tended to generate stronger effects on embryos than on other stages. Lead and ammonium share the characteristic of appearing as cations in the water. Cations are known to inhibit the hatching enzyme responsible for egg membrane degradation and egg elongation as the embryo grows (Freda and Dunson 1985). The inhibition of this enzyme causes the embryo to curl inside the egg and ultimately to die because of asphyxia. Furthermore, there seems to be a relationship between the potential for the egg jelly coat to act as a barrier of diffusion and the lipophilicity of the substance, which would explain why jelly coat represents only minor protection against lead or ammonium nitrate toxicity. However, some studies have also shown that high tolerance of embryos is unrelated to the potential protection conferred by the jelly coat (Herkovits and Pérez-Coll 1993, Bridges 2000). Thereby, other factors like the egg membrane, low metabolic profile or minimum surface to volume ratio would also play a role in diminishing the relative sensitivity of amphibian embryos compared to other aquatic stages. The comparison of life stage sensitivity as a function of the analysed responses did not reveal any trend different from the general one (Figure 5), with hatchlings and larvae usually appearing as the most sensitive stages. Embryos appeared to be particularly sensitive when analysing developmental effects, whereas hatchlings seemed to suffer lethal effects at a higher extent than other life stages EFSA Supporting publication 2017:EN-1251

45 Figure 5: Relative frequencies of cases in which a given life stage out of the pair compared was more sensitive than the other, classified per type of measured response. Asterisks (*) indicate cases in which the comparison of relative frequencies of each life stage (excluding indifferent comparisons) were close to significance (0.05 < p < 0.1; no comparisons resulted in significant differences at the level p < 0.05), according to chi-square tests (for detailed data, see Appendix H). In the case of reptiles, three data sets pairing juvenile and adult toxicity data were compared, all of them regarding lead toxicity. The comparison of EC 10 values corresponding to parameters related to metabolism and biochemistry did not result in significant differences in sensitivity between ages (Wald s Χ 2 = 1.343, 1 d.f., p = 0.246). Two regressions of effect size values on exposure concentrations resulted in toxicity indexes showing opposite results: when responses associated with haematological parameters were considered, adults resulted more sensitive than juveniles, whereas juveniles suffered increased effects relative to adults when variables associated with metabolism and biochemistry were taken into account (Figure 6). Nevertheless, the scarcity of data for reptiles prevents from extrapolating these results to get general conclusions with respect to the entire class EFSA Supporting publication 2017:EN-1251

46 Figure 6: Regressions to calculate toxicity indexes for comparison of sensitivity between juvenile and adult reptiles exposed to lead, based on responses reporting effects on a) haematology or b) metabolism and biochemistry. Considering haematological responses juveniles appear as more sensitive than adults since there is a significant regression model that results in a toxicity index for the former and not for the latter. The opposite happens when considering metabolic/biochemical responses. For details on methodology, see section Comparison of endpoints observed in laboratory with mesocosm and field data A total of 52 comparisons between sensitivity shown in laboratory, mesocosm or field studies (see details in Table 12) were run considering the availability of data matching class, substance, endpoint, exposure environment and route and type of recorded response. The outcomes of all specific comparisons are shown in the Appendix I. These comparisons included data from 13 chemical substances, five of which (i.e. chlorpyrifos, glyphosate, malathion, pyraclostrobin and triclopyr) are pesticides currently approved for use in the EU. These five substances comprised 22 out of the 52 comparisons. When data were restricted to make the comparisons on the same species tested in either laboratory, mesocosm or field, the number of runs was 29, including nine chemical substances with glyphosate, malathion and triclopyr as the only pesticides currently approved in the EU. These three pesticides comprised 16 of the 29 comparisons. The limited amount of data for reptiles made it impossible to run any comparison between study types for this group, so this part of the study had to be conducted with amphibians only. As happened with the life stage comparisons, the majority of comparisons (all but one) referred to aquatic amphibians (basically from hatching to metamorphosis) exposed through contaminated water. Relevant conclusions could be extracted only from the comparison between laboratory and mesocosm data (N=42), as the amount of available data for comparing between laboratory and field studies was low (N=8, one of which refers to the terrestrial environment; Table 12). Nevertheless, the few cases in which responses recorded in the field and in the laboratory could be compared, animals tested in laboratory appeared as more sensitive than those tested in the field. The effects caused by pollutants on aquatic amphibians when tested under the semi-field conditions settled in mesocosms resulted in stronger effects compared to what was recorded in laboratory conditions, although the difference was significant only when the compared effects came from the same species. This happened even considering that, under specific mesocosm conditions, the addition of chemicals at low doses may benefit amphibian larvae by removing their predators (e.g. Relyea et al. 2005). When all possible comparisons were run, regardless of whether the lab and mesocosm studies had been conducted with the same species or not, there were no differences in sensitivity between laboratory and mesocosm studies EFSA Supporting publication 2017:EN-1251

47 Table 12: Results of the comparison by pairs of study types (laboratory, mesocosm, field) conducted with amphibians. Each comparison was made on data referring to concurrent substances, life stages, effect characterization parameters and response types. In a first set of comparisons, all species were considered together and comparisons were run regardless of whether species in the compared pair of data were concurrent or not. In a second set of comparisons, only data coming from the same species were compared. Results of the chisquare test for comparison of frequencies of cases in which one or another study type was identified as causing more severe effects are shown (for detailed data, see Appendix I). Without considering concurrent species Considering concurrent species Compared study types stages N Comparison results Sig. (Χ 2 test) Lab vs. mesocosm 42 More sensitive in lab: 19.0% More sensitive in mesocosm: 16.7% Lab vs. field 8 More sensitive in lab: 12.5% More sensitive in field: 0.0% - Mesocosm vs. field 2 No differences in sensitivity - Lab vs. mesocosm 24 More sensitive in lab: 0.0% More sensitive in mesocosm: 16.7% - Lab vs. field 4 More sensitive in lab: 25.0% More sensitive in field: 0.0% - Mesocosm vs. field 1 No differences in sensitivity - The subset of data comparing laboratory and mesocosms effects was analysed more in detail to assess how extrapolation factors varied among the different types of recorded responses. In nonlaboratorial studies, it is infrequent to analyse responses different from survival, growth or development. The distribution of frequencies for each of these three responses (or for consolidated sublethal responses in the case of concurrent species data) was similar to what was shown for the consideration of the whole pool of data. When all species were analysed together, the frequencies of endpoint values obtained in the laboratory being more or less sensitive than those obtained in mesocosm were not different (Figure 7a), whereas when only comparisons involving concurrent species were performed, endpoints obtained from mesocosm studies were generally more sensitive than those obtained from laboratory studies (Figure 7b). Figure 7: Relative frequencies of cases in which effect characterization endpoints from laboratory or mesocosms studies were significantly higher than those from the other scenario, classified per type of measured response. Results are divided in data from pairs of values collected independently on the species (a) and data from pairs of values collected from the same species (b). For detailed data, see Appendix I. The degree of uncertainty when extrapolating data from laboratory studies to the field was analysed by exploring the numerical relationship between endpoint values obtained in laboratory and those 47 EFSA Supporting publication 2017:EN-1251

48 obtained in the field or, as an intermediate step, in mesocosms. When time-corrected endpoint values were compared between field and laboratory, no significant correlation was found, either when pairs of values were considered for concurrent species, or for any amphibian species. This does not necessarily mean that effects of chemicals in the field cannot be extrapolated from laboratory studies, but simply that the amount of comparable data is not high enough to estimate a significant assessment factor. When time-corrected endpoint values were compared between laboratory and mesocosms studies, regardless of whether laboratory and mesocosm endpoints were calculated from the same species or not, no significant correlation was found allowing for the estimation of an extrapolation factor of effects from laboratory to mesocosm scenarios (Figure 7). For these comparisons, besides potential differences related to laboratory vs. mesocosm scenarios, variability in sensitivity could also be motivated by differences in sensitivity between the species included in the comparison. In a metanalysis with a high number of data, the effect of interspecific differences in sensitivity could be minimized because of statistical replication, but the amount of data available for running the present comparison was not high enough to rule out potential effects due to the use of different species. When data analysis was conducted with pairs of data considering concurrent species, i.e. eliminating the effect of interspecific differences in sensitivity, a significant correlation was observed (Figure 8), confirming the higher sensitivity of mesocosm-originated values shown by frequency comparisons (Table 12). The slope of the regression of lab- vs. mesocosm- data was (95% confidence interval: ), which means that mesocosm-originated, timecorrected endpoint values were on average 303 times lower (i.e. 303 times more sensitive) than laboratory-originated values. These regression values were obtained from the comparison of endpoints measured in laboratory vs. mesocosm studies after exposure to eight different substances involving data retrieved from up to 65 different species (Appendix J). Figure 8: Plot of paired values collected from laboratory and mesocosm studies corresponding to the same substance, exposure route, endpoint, life stage and response type. Data are divided depending on whether values within each compared pair were collected independently from the species (blue elements) or from the same species (red elements). Central tendency line (and 95% confidence intervals) for the best adjustment of each subgroup are shown, including the results of the linear regression to estimate assessment factors when possible. The arrow indicates a pair of data that outlied the plot. When extrapolation data were calculated for each specific response, lab- and mesocosm-generated time-corrected endpoints related to lethal effects were different for both species-independent and concurrent species data (Figure 9). For species-independent data, the slope of the regression of lethal endpoint time-corrected values was (95% C.I.: ), which means that mesocosm EFSA Supporting publication 2017:EN-1251

49 originated endpoints were on average 1429 times more sensitive than laboratory-originated endpoints. For concurrent species data, the slope was (95% C.I.: ) meaning that mesocosm endpoints for lethal effects were 333 times more sensitive than laboratory endpoints for such effects. No differences in sensitivity of endpoints between laboratory and mesocosm data for sublethal effects were found (all Wilcoxon s Rank Test p 0.068). Figure 9: Plot of paired values collected from laboratory and mesocosm studies corresponding to lethal effects associated with the same substance, exposure route, endpoint and life stage. Data are divided depending on whether values within each compared pair were collected independently of the species (blue elements) or for the same species (green elements). Central tendency line (and 95% confidence intervals) for the best adjustment of each subgroup are shown, including the results of the linear regression to estimate assessment factors when possible. Although some values in the figure look like zero, they are really positive although very low (<0.01) values; in order to display the great variability in values, the scales of the figure are depiected in a way that make these values indistinguishable from zero. In a risk assessment framework, the fact that sensitivity recorded in mesocosm is higher than that recorded in laboratory conditions could be interpreted as opposing to the accepted rule that lower tiers must lead to more conservative assessments. In aquatic risk assessment, mesocosms are one of the high-tier approaches to refine laboratory exposures (EFSA PPR Panel 2013), and therefore it is assumed that risks derivated from assessment in mesocosm must be lower than those derivated from laboratory assessments. Fish and other vertebrates are not included in aquatic mesocosms conducted as part of risk assessment because of the strong influence that they would have on the components of the system that are normally under evaluation (invertebrates or plants). However, the structure of aquatic mesocosms used in risk assessment according to the proposed guidelines (EFSA PPR Panel 2013) is not comparable to the structure of mesocosms commonly used in amphibian ecotoxicological assessment. In the laboratory, environmental stress is reduced since test organisms are provided with enough food to reduce interspecific competition, have constant and optimal environmental conditions (temperature, light), and additional stress sources like predation are excluded. In this scenario exposure is high since natural break down processes are reduced and avoidance of contaminated habitat is not possible. Graney et al. (1994) defined aquatic mesocosms as intermediate-sized systems, such as a dug-out ponds or in situ enclosures, that can be replicated and manipulated to test both structural and functional parameters as representative aquatic ecosystems. According to this definition, the mesocosms designed in the majority of amphibian ecotoxicological studies aim at 49 EFSA Supporting publication 2017:EN-1251

50 representing aquatic ecosystem as reliable as possible. Therefore, in a mesocosm study, exposure to a chemical is more realistic, and generally lower (e.g. through the presence of sediment and plants to which waterborne pollutants can adsorb) than in the laboratory. However, other environmental stressors such as fluctuating conditions, competition and predation are normally present in mesocosm studies and can affect the response of the animals to pollution. The effect magnitude used to compare laboratory and mesocosm studies was attributable to the addition of pollutants only (e.g. when predators were present in the mesocosms the effect characterization value was calculated by comparing responses in the exposed + predator and the control + predator treatment; treatment without predators were never used as controls for treatments with predators). However, the simple presence of stressors other than pollution can induce a stress response in animals that magnifies the effects associated with the exposure to chemicals. Additionally, the effects of pollutants on the ability to cope with environmental stress can indirectly enhance the mortality caused by other stressors, for instance through the reduction of the response of scape from predation (Ortiz-Santaliestra et al. 2010). The possible presence of pathogens in mesocosm could also turn amphibians more sensitive to doses of certain chemicals that affect the immune system and that, in absence of pathogens (the likely scenario in the laboratory), have no visible effects on organisms (Christin et al. 2003, 2004). Another possible explanation for the higher toxicity in mesocosms compared to laboratory is that, because mesocosm experiments are usually run outdoors, the presence of ultraviolet radiation may result in photo-enhanced toxicity of chemicals. For example, a higher toxicity was observed in amphibians exposed to UV and the insecticide carbaryl, comparatively to carbaryl alone (Zaga et al. 1998). In summary, the higher sensitivity to addition of chemicals to the water shown by amphibians tested in mesocosms when compared to laboratory studies would indicate that the interaction of the tested chemicals with other stressors enhances the magnitrude of their impacts, highlighting the importance of taking indirect effects of chemical exposure into account in order to complete an ecologically reliable assessment. When the sensitivity recorded in laboratory studies was compared to that shown in the field, the outcome of the comparison was completely different from when laboratory studies were compared to mesocosms. This difference could be in the study design; whereas in mesocosm animals are openly exposed to the natural stressors referred above (e.g. predation, competition), many field studies are conducted with caged individuals that are exposed to the applied chemical conditions in each treatment, but protected at least from predation. This particular scenario would approach the type of aquatic mesocosms described in aquatic risk assessment guidelines (EFSA PPR Panel 2013), where the factors susceptible to have an influence on the tested organisms (in this case amphibians) is reduced or at least minimized. This design would explain the observed reduced sensitivity in the field when compared to laboratory assays, because of the ameliorated exposure as a consequence of pollutant degradation, dissipation or adsorption to plants or sediments, together with the restriction of sources of indirect effects. Nevertheless, it is important to remind that the laboratory vs. field comparison in the aquatic environment involved seven pairs of comparable data only, which limits the relevance of conlusions in this context. With regards to terrestrial amphibians, the only possible comparison was that between laboratory and field-originated data corresponding to studies overspraying juvenile individuals with pyraclostrobin. Regressions of effect size calculated from mortality caused by the fungicide treatment on the applied rate did not reveal any difference in toxicity trends between laboratory and field-originated data (Figure 10) EFSA Supporting publication 2017:EN-1251

51 Figure 10: Regressions to calculate toxicity indexes from lethal effects shown by terrestrial amphibians after overspraying with pyraclostrobin-based products, for comparison of sensitivity between field and laboratory scenarios. The absence of effectively calculated toxicity indexes (both regressions are non-significant) did not allow to elucidate which scenario led to more severe effects, in spite of the differences observed in the effect size vs. exposure plots Comparison of sensitivity with surrogate taxa Aquatic environment The comparison of data corresponding to the most sensitive amphibian and fish species included 81 values relative to short-term toxicity endpoints (either LC 50, EC 50 or NOEC) (Appendix K). Linear regression adjustment was y = x (R 2 = 0.739, p < 0.001). The slope of (95% C.I.: ) means that each unit of decrease of the fish endpoint value is associated with an average decrease of the amphibian endpoint value of 1.90 units (Figure 11). The frequency of cases in which fish were more sensitive than amphibians was higher than the frequency of cases with amphibians as more sensitive (64.2 vs. 35.8%), although this difference was not statistically significant (Wilcoxon s Rank Test fish-amphibian Z = , p = 0.127). A variant to this comparison was run considering only data for 96h-LC 50 comparisons, including 63 of the 81 pairs of data. The results obtained were similar as when all data were pooled together (y = x; slope 95% C.I.: ; R 2 = 0.741, p < 0.001) (Figure 12). The frequency of cases in which fish were more sensitive was higher, in a close-to-significant manner, than the frequency of cases in which the opposite trend was verified (69.8 vs. 30.2%; Wilcoxon s Rank Test fish-amphibian Z = , p = 0.086) EFSA Supporting publication 2017:EN-1251

52 Figure 11: Plot of paired, log-transformed toxicity endpoint values from aquatic amphibians and fish matching the same substance, exposure route, exposure time and endpoint. Central tendency line and their 95% confidence intervals for the best data correlation are shown. The black line illustrates the hypothetical equality between amphibian and fish sensitivity indicators (for detailed, non-transformed data, see Appendix K). Figure 12: Plot of paired, log-transformed values of 96h-LC50 values from aquatic amphibians and fish. Central tendency line and their 95% confidence intervals for the best data correlation are shown. The black line illustrates the hypothetical equality between amphibian and fish sensitivity indicators (for detailed, non-transformed data, see Appendix K) EFSA Supporting publication 2017:EN-1251

53 The comparison of data involving the rainbow trout as the most commonly used fish species in environmental risk assessment of chemicals included 47 pairs of data. The regression equation was y = x, with a slope (95% C.I.: ) that was in this case lower than that found in the comparison of the most sensitive species on each side (Figure 13). The correlation between series of data was significant but very weak (R 2 = 0.089, p = 0.041) and the comparison of frequencies of cases in which fish or amphibians were more sensitive, even although trout were more sensitive in the majority of cases (57.4%), was not significant (Wilcoxon s Rank Test fish-amphibian Z = , p = 0.882). The variant to this analysis including the 43 data pairs referring to 96h-LC 50 values revealed essentially the same as when all data were analysed together (y = x, slope 95% C.I.: ), although the correlation remained close to the statistical significance (R 2 = 0.070, p = 0.083; Figure 14). Despite the higher frequency of cases in which trout was more sensitive than the corresponding amphibian species (58.1%), the comparison was neither in this case significant (Wilcoxon s Rank Test fish-amphibian Z = , p = 0.875). These results suggest that, although rainbow trout seems to be among the most sensitive fish species, there are exceptions to this pattern that could compromise the coverage that fish toxicity testing, and in particular assays conducted on rainbow trout, supposedly provides to amphibians. Figure 13: Plot of paired, log-transformed values of toxicity endpoints from aquatic amphibians and rainbow trout matching the same substance, exposure route, exposure time and endpoint. Central tendency line and their 95% confidence intervals for the best data correlation are shown. The black line illustrates the hypothetical equality between amphibian and fish sensitivity indicators (for detailed, non-transformed data, see Appendix K) EFSA Supporting publication 2017:EN-1251

54 Figure 14: Plot of paired, log-transformed values of 96h-LC50 values from aquatic amphibians and rainbow trout. Central tendency line and their 95% confidence intervals for the best data correlation are shown. The black line illustrates the hypothetical equality between amphibian and fish sensitivity indicators (for detailed data, see Appendix K). Although our data on comparisons of sensitivity among life stages revealed that hatchlings were the most sensitive life stage (section 3.2), the six endpoint data that were available for comparison with surrogates corresponding to hatchlings resulted less sensitive for the amphibian species than for the selected fish species (Figure 15). In order to understand patterns leading to differential sensitivity between fish and aquatic amphibians, the comparison of each substance s pair of endpoint values was analysed substance by substance. As seen in Figure 15, a clear pattern is not found in the responses shown to each type of substance, which could explain those cases in which amphibians resulted more sensitive than fish. Among PPPs, the substances exerting higher toxicity to amphibians than to fish were the only compared strobilurin fungicide (azoxystrobin), the only compared sulfonylurea (nicosulfuron) and viologen (paraquat) herbicides, the only compared neonicotinoid insecticide (imidacloprid), the only compared microbiocide (triclosan), one of the two compared cyclohexenone herbicides (cycloxydim), one of the five compared carbamate insecticides (thiobencarb), and two of the nine compared organosphosphate insecticides (fenitrothion and parathion, although for the latter higher amphibian toxicity applied only to one of the two compared endpoints). Among chemicals other than PPP active ingredients, the higher toxicity of organochlorine compounds like chloroform, PCBs or thrichloroethylene is remarkable but unexpected, considering that fish were more sensitive than amphibians to the six compared legacy organochlorine insecticides (i.e. aldrin, endosulfan, endrin, dieldrin, heptachlor and toxaphene) EFSA Supporting publication 2017:EN-1251

55 Figure 15: Outcome of each compared pair of endpoint values considering the most sensitive amphibian and fish available data. Red arrows indicate comparisons in which the amphibian data came from individuals exposed as hatchlings (for detailed data, see Appendix K) EFSA Supporting publication 2017:EN-1251

56 Some of the substances to which amphibians resulted more sensitive than fish have been described as potential endocrine disrupting chemicals (EDCs) (e.g. Triclosan, Veldhoen et al. 2006). The high sensitivity of amphibians to hormonal disruption, either through alteration of thyroid hormonal processes involved in development and metamorphosis or of estrogenic hormones involved in maturation and sex determination has been widely reported, and is recognised as one of the main aspects making amphibians different from other vertebrates in terms of toxicological susceptibility. However, other EDCs in the comparison reported to affect sex determination in amphibians, such as the widely studied atrazine (Hayes et al. 2002), showed higher toxicity to fish than to amphibians. It must be noticed that this comparison includes short-term toxicity data only. Short-term toxicity evaluation is not the most appropriate way to detect endocrine disrupting effects, as the moment when the key hormonal process susceptible to become affected happens is very localized in time (e.g. maturation of gonadal tissue, hormonal changes leading to metamorphosis climax), and exposure of individuals to chemicals over a few-day time period will unlikely coincide with the sensitive stage. Comparable long-term toxicity data were available for two substances only (abamectin and novaluron, with NOEC values and 9.6 times lower for fish than for amphibians, respectively). Weltje et al. (2013) included a comparison of NOEC values from fish and amphibians regardless of whether they match in exposure time or not, and did not observe any particular pattern showing that EDCs were differentially more toxic to amphibians than to fish, although the substance to which differential toxicity was greatest to amphibians was perchlorate, a well-known thyroid disrupting chemical. In the comparative analysis of the potential coverage provided by the rainbow trout, it was observed that amphibians were more sensitive than rainbow trouts to four substances that appeared more toxic to fish when the most sensitive fish species was used: carbendazim, diazinon, pyperonyl butoxide and chlorobenzene (Figure 16). These results would reflect that, for any reason, the rainbow trout is particularly tolerant to certain chemicals compared to other species, although the chemicals in the list above do not reflect any particular trend in terms of properties or modes of action. The impact of this specific outcome in the efficacy of coverage was not too high, as the frequency of cases of rainbow trout showing decreased sensitivity (and therefore providing a weak coverage) was low, although a further review of the location of rainbow trout within fish species sensitivity distributions seems appropriate to reaffirm the role of this species as surrogate for aquatic amphibian stages EFSA Supporting publication 2017:EN-1251

57 Figure 16: Outcome of each compared pair of endpoint values considering the most sensitive amphibian and the rainbow trout (for detailed data, see Appendix K) EFSA Supporting publication 2017:EN-1251

58 Terrestrial environment Data for comparing the most sensitive species of amphibians and of reptiles with those of birds and mammals are shown in Appendix L. Terrestrial amphibians The comparison of most sensitive species of amphibians with their corresponding surrogates (Figure 17) revealed a lack of association in toxicity values shown by the absence of significance of the regression models when the most sensitive avian or mammalian species was used as surrogate (y = x, R 2 = 0.030, p = 0.476). When relative frequency of cases in which surrogates were more sensitive than amphibians were compared with the frequency of cases in which surrogates were less sensitive, it also resulted in non-significant differences (Wilcoxon s Rank Test surrogateamphibian: Z = , p = 0.573), in spite of surrogates being more sensitive than amphibians in a higher number of cases (63.2%). Figure 17: Plot of paired values of toxicity endpoints from the most sensitive terrestrial amphibian and the most sensitive bird or mammal species matching the same substance, exposure route, exposure time and endpoint. The black line represents the equality between amphibian and surrogate sensitivity indicators (for detailed data, see Appendix L). The data used in the comparison analysing coverage of terrestrial amphibians by standard species was not very different from those used in the comparison to identify sensitive taxa. Because the number of considered standard species was high (contrarily to when we did the same evaluation with fish data, see section 3.4.1), for the majority of the substances the most sensitive surrogate species was one of the standard ones. Only for five of the 19 compared endpoints the most sensitive surrogate species was none of the ones included in the standard species list defined in section In consequence, there were no essential differences between the two comparisons (Figure 18) EFSA Supporting publication 2017:EN-1251

59 Figure 18: Outcome of each compared pair of endpoint values considering the most sensitive terrestrial amphibian and the most sensitive bird or mammal out of a) all available species, b) the standard species commonly used for toxicity assessment. For detailed data, see Appendix L. When surrogate taxa were considered separately (Figure 19), no significant correlations were found between endpoints from terrestrial amphibians and endpoints from birds (y = x, R 2 = 0.006, p = 0.811) or mammals (y = x, R 2 = 0.029, p = 0.484). In both comparisons, the number of chemicals to which surrogates resulted more sensitive than amphibians was higher than the number of chemicals to which amphibians were more sensitive than surrogates (Figure 20), but this difference was not significant (Wilcoxon s Rank Test bird-amphibian: Z = , p = 0.084; mammal-amphibian: Z = , p = 0.601) EFSA Supporting publication 2017:EN-1251

60 Figure 19: Plots of paired values of toxicity endpoints from the most sensitive terrestrial amphibian and the most sensitive bird (left) or mammal (right) species matching the same substance, exposure route, exposure time and endpoint. The black lines represent the equality between amphibian and surrogate sensitivity indicators (for detailed data, see Appendix L). Figure 20: Outcome of each compared pair of endpoint values considering the most sensitive terrestrial amphibian and the most sensitive bird (top) or mammal (bottom). For detailed data, see Appendix L EFSA Supporting publication 2017:EN-1251

61 Reptiles When endpoints from the most sensitive reptilian species to each tested compound were compared to the corresponding endpoints from the most sensitive bird or mammal species, we found the same result as with terrestrial amphibians, with a lack of correlation between reptilian and surrogate data (Figure 21, y = x, R 2 = , p = 0.936). Birds or mammals were more sensitive than reptiles in a 70.6% of the comparisons, but this frequency was not significantly different than random according to the Wilcoxon s Rank Test (Z = , p = 0.619). With regards to the comparisons relative to the defined standard bird or mammal species, only in four of the 17 cases the most sensitive surrogate was not one of these standard species. In those four cases, surrogates were more sensitive than reptiles regardless of whether the most sensitive species or the most sensitive standard species was selected on the surrogate side (Figure 22). Figure 21: Plot of paired values of toxicity endpoints from the most sensitive reptile and the most sensitive bird or mammal species matching the same substance, exposure route, exposure time and endpoint. The black line represents the equality between amphibian and surrogate sensitivity indicators (for detailed data, see Appendix L) EFSA Supporting publication 2017:EN-1251

62 Figure 22: Outcome of each compared pair of endpoint values considering the most sensitive reptile and the most sensitive bird or mammal out of a) all available species, b) the standard species commonly used for toxicity assessment. For detailed data, see Appendix L. When bird and mammal endpoints were compared separately with those of reptiles (Figure 23), we did not find any significant correlation between reptilian and surrogate endpoints (reptiles vs. birds: y = x, R 2 = 0.058, p = 0.501; reptiles vs. mammals: y = x, R 2 = , p = 0.989). As seen when both surrogate taxa were considered together, the number of substances for which surrogate endpoint was lower than reptilian endpoint was higher than the number of substances for which the opposite was true (Figure 24), but this higher frequency of increased sensitivity of surrogates was not statistically different from random in any case (Wilcoxon s Rank Test bird-reptile: Z = , p = 0.386; mammal-reptile: Z = , p = 0.943) 62 EFSA Supporting publication 2017:EN-1251

63 Figure 23: Plots of paired values of toxicity endpoints from the most sensitive reptile and the most sensitive bird (left) or mammal (right) species matching the same substance, exposure route, exposure time and endpoint. The black lines represent the equality between reptilian and surrogate sensitivity indicators (for detailed data, see Appendix L). Figure 24: Outcome of each compared pair of endpoints considering the most sensitive reptile and the most sensitive bird (top) or mammal (bottom). Detailed data are in Appendix L EFSA Supporting publication 2017:EN-1251

64 Coverage of terrestrial amphibians and reptiles by birds and mammals Pyrethroid insecticides were systematically more toxic to amphibians or reptiles than to birds or mammals. This effect has been previously pointed out by other comparisons of toxicity data (Weir et al. 2010) and is supposed to be related to the slower metabolism of the substances in poikilothermic animals compared to homeothermic ones, leading to a longer retention in the organism of the parent compound, which is more toxic than the metabolites. Likewise, the lower metabolism could lead to a reduced rate of elimination of bioaccumulative compounds, for which amphibians would appear as more sensitive than birds or mammals to the toxicity of organochlorine insecticides. On the other hand, a lower metabolism would be an advantage for substances like organothiophosphate insecticides or rodenticides, which become more toxic when metabolized. The fact that no different patterns have been identified between comparisons of the most sensitive pair and comparisons with the commonly used standard species is not necessarily due to a proper coverage conferred by such standard species, but because of the limited data on terrestrial toxicity available for amphibians or reptiles that can be compared with an endpoint surrogate value. The data limitation is not only a consequence of the low number of substances tested on amphibians and reptiles, but also due to a lack of concurrency of the exposure ways. Whereas most of the few toxicity studies conducted on terrestrial amphiabians or reptiles use percutaneous (by overspray or contact with treated surfaces) or topical egg exposures, the majority of data from surrogates refer to oral exposures, and information on dermal exposures (which could compare for example to overspray) of birds and mammals is almost limited to the dermal toxicity assays conducted with rodents. Furthermore, when dermal exposure data are available for birds or mammals, exposure concentrations are provided as mass per volume of the applied product, whereas overspray tests with amphibians or reptiles usually measure exposure levels as application rates (mass per surface area unit) and egg topic exposure report exposure levels as mass of substance per egg weight unit. The lack of consistency of the units often makes the data incomparable. All comparisons of sensitivity between amphibians or reptiles and the terrestrial vertebrates used as surrogates in risk assessment, regardless of whether surrogates were considered together or as bird and mammals separately, took to the same conclusion: there is a non-significantly higher number of substances to which surrogates are more sensitive than amphibians or reptiles. Correlations between endpoint values recorded in one and another group of species are very weak and never reach statistical significance. This absence of correlations reveals that, with currently available data, it is not possible to ensure that birds and mammals are valid surrogates for testing chemical toxicity on terrestrial amphibians and reptiles, as no predictions on the effects on herpetofauna can be made from the effects measured in homoeothermic vertebrates. This conclusion contrasts with that of Crane et al. (2016), who suggested that birds and mammals would be valid surrogates for assessing acute toxicity on terrestrial amphibians. Our data do agree with those of Crane et al. (2016) in pointing the higher frequency of available cases in which sensitivity of birds and mammals is higher than that of amphibians. These authors based their conclusion on appropriate coverage provided by terrestrial surrogates in two facts: first, once applied the assessment factor used in ERA of pesticides to the amphibian sensitivity endpoint, there was only one case (DDT) for which surrogate endpoint was not protective enough. According to the data that we have retrieved, the toxicity of the four analysed pyrethroids on amphibians would neither be within the protection range provided by the assessment factor, leading to a total of five out of 19 tested substances which toxicity on amphibians would not be covered by that of surrogates (even after application of the assessment factor). The second fact used by Crane et al. (2016) to support their conclusion is that they found a significant correlation between terrestrial amphibian and mammalian endpoint data. This finding contrasts with the results obtained in the present review, where no significant correlation between terrestrial amphibians and surrogates has been detected. The lists of substances included in the comparisons run in the present project and by Crane et al. (2016) are very similar. The reasons why both comparisons reflect some differences in their outcomes could be in the fact that Crane et al. (2016) included open-ended endpoint values (i.e. those expressed as greater than), and thus they had to use a non-parametric correlation (Spearman). We preferred to avoid those values and run our comparison through Pearson s correlations with accurately defined data. The fact that these methodological variations lead to different results is 64 EFSA Supporting publication 2017:EN-1251

65 probably a consequence of the small amount of available data and reflects the limited relevance of conclusions extracted in this context by any of the studies. Finally, it is worth to mention that, whereas these comparisons are conducted mainly with the purpose of evaluating the role of birds and mammals as surrogates of amphibians and reptiles in pesticide risk assessment, the number of currently used pesticides included in these comparisons is very low (two active ingredients in our comparison between terrestrial amphibians and birds or mammals, four in our comparison between reptiles and birds or mammals, and four in the comparison made by Crane et al. 2016) Species sensitivity distributions Species Sensitivity Distribution (SSD) in amphibian species Considering the toxicity data obtained from the literature for amphibian and reptile species, 13 sets (Table 13) met the requirements needed to conduct the SSD analyses (i.e. six or more species and substance with high active ingredient purity, with the exception of data from butachlor and glyphosate, see section 2.5.5). All data used to create SSD are available in the Appendix M. All SSD constructed were relative to amphibians (there were not enough data from reptiles for any active substance to create SSD), twelve of which referred to waterborne exposures in the aquatic environment, and only one referred to LD 50 resulting from dermal exposure to glyphosate through overspray in the terrestrial environment (Figure 25a). Among the aquatic exposure SSD, 11 corresponded to 96h-LC 50 (i.e. ammonium nitrate, butachlor, cadmium, carbaryl, chlorpyrifos, copper, endosulfan, glyphosate, malathion, sodium chloride and zinc, Figure 25b-l) and the other one corresponded to the 96h-NOEC for glyphosate (Figure 25m). Taxonomic diversity of the species represented in each SSD, as well as HC 5 /HD 5 calculated in each case are shown in Table 13. Table 13: Hazard coefficients (HC 5 and HD 5 ) estimated from SSD conducted on amphibians. The taxonomic diversity within each SSD is indicated as number of species (Spp.), genera (Gen.) and families (Fam.) and categories within each variable used for creating subgroups of the database with effect characterization values. For detailed data, see Appendix M. Active substance Endpoint Environment and exposure route Taxonomic diversity in the SSD Spp. Gen. Fam. HC 5 / HD 5 Units Glyphosate LD 50 Terrestrial, overspray mg a.e./dm 2 Ammonium nitrate 96h-LC 50 Aquatic, waterborne mg/l Butachlor 96h-LC 50 Aquatic, waterborne mg/l Cadmium 96h-LC 50 Aquatic, waterborne mg/l Carbaryl 96h-LC 50 Aquatic, waterborne mg/l Chlorpyrifos 96h-LC 50 Aquatic, waterborne mg/l Copper 96h-LC 50 Aquatic, waterborne mg/l Endosulfan 96h-LC 50 Aquatic, waterborne µg/l Glyphosate 96h-LC 50 Aquatic, waterborne mg a.e./l Malathion 96h-LC 50 Aquatic, waterborne mg/l Sodium chloride 96h-LC 50 Aquatic, waterborne mg/l Zinc 96h-LC 50 Aquatic, waterborne mg/l Glyphosate 96h-NOEC Aquatic, waterborne mg a.e./l 65 EFSA Supporting publication 2017:EN-1251

66 a b c Figure 25: Graphical representation of the Species Sensitivity Distributions (SSD) for different substances and exposure scenarios (a-m) constructed for amphibian species with indication of hazard coefficient corresponding to the 5th, 20th and 50th percentiles (i.e. benchmark values covering endpoint-based sensitivity of 95%, 80% and 50% of the species in the SSD). For detailed data, see Appendix M. Continues on the next page EFSA Supporting publication 2017:EN-1251

67 Figure 25 (cont.) d e f (continues on the next page) 67 EFSA Supporting publication 2017:EN-1251

68 Figure 25 (cont.) g h i (continues on the next page) 68 EFSA Supporting publication 2017:EN-1251

69 Figure 25 (cont.) j k l (continues on the next page) 69 EFSA Supporting publication 2017:EN-1251

70 Figure 25 (cont.) m A relevant analysis after running the SSD in amphibians was the evaluation of Xenopus laevis as amphibian model species in toxicity testing. The few standard guidelines for toxicity testing available so far in amphibians use this species as model (OECD 2009, 2015). The main reason for this is that X. laevis is easy to house and breed in captivity given its fully aquatic habits, and has a relatively short generation time compared to other amphibians. However, the absence of a terrestrial phase makes this species unsuitable for terrestrial toxicity testing. In order to investigate the extent at which X. laevis may act as an umbrella species to determine aquatic toxicity of chemicals on amphibians, it is necessary to review how its sensitivity compares to that of other species. Ten out of the 12 amphibian SSDs for aquatic toxicity data included X laevis (Appendix M), and only in one case the endpoint for X. laevis was below the calculated HC 5 (Figure 25m). The most frequent case was that of the endpoint of X. laevis fitting between the HC 25 and the HC 50, with another case in which such endpoint was between the HC 5 and the HC 25, and three cases in which the X. laevis endpoint was higher than the HC 50 of the SSD. Given that some of these values can be interpreted mistakenly because of the low number of species in some of the SSDs, we calculated the percentage of species other than X. laevis in each SSD on the protected side (i.e. with an endpoint value higher than that of X. laevis). Only for two of the 10 SSDs X. laevis protected more than 95% of the other species. The mean (±SD) percentage of other species covered by the X. laevis endpoint across all SSDs was 50.9±32.5 %. These data reflect that X. laevis would not be particularly sensitive to chemicals in the water compared to other amphibian species, and therefore that its use as standard species for toxicity testing would require the consideration of an assessment factor. The quotient between the X. laevis endpoint value and the HC 5 in the different SSDs ranged from 0.47 (for 96h- NOEC of glyphosate, where X. laevis was comparatively most sensitive, Figure 25m) to (for 96h- LC 50 of endosulfan, where it was comparatively least sensitive, Figure 25h), with a mean±sd value of 38.9± SSDs in surrogate species and comparison with amphibian data The only SSD that was conducted for terrestrial individuals on the basis of the available data involves overspray of juvenile or adult amphibians. Because this exposure route is not commonly tested in birds or mammals, we were unable to find an SSD for birds or mammals which outcome could have been comparable to that of our terrestrial SSD. Therefore, the search of surrogate data on HC 5 derived from SSD was focused on fish-generated information. We found in the literature seven HC 5 values from fish that were comparable with the HC 5 resulting from the SSD conducted with amphibian data (Table 14). These values were completed with another 70 EFSA Supporting publication 2017:EN-1251

71 four HC 5 that were calculated using the 96h-LC 50 data for fish available in the databases provided by EFSA (section 2.5.5) and in TOXNET (Figure 26). Table 14: Hazard coefficients (HC 5 ) for fish retrieved or estimated from database-collected data. Active substance Endpoint Environment and exposure route Number of species in SSD HC 5 Units Source EFSA databases (N=1) 96h-LC 50 Aquatic, waterborne mg/l Carbaryl and TOXNET (N=13) 96h-LC 50 Aquatic, waterborne mg/l Willming et al. (2016) Chlorpyrifos 96h-LC 50 Aquatic, waterborne mg/l Willming et al. (2016) Copper 96h-LC 50 Aquatic, waterborne mg/l Willming et al. (2016) Endosulfan 96h-LC 50 Aquatic, waterborne µg/l TOXNET Glyphosate 96h-LC 50 Aquatic, waterborne mg a.e./l TOXNET EFSA databases (N=3) 96h-LC 50 Aquatic, waterborne mg/l Malathion and TOXNET (N=7) 96h-LC 50 Aquatic, waterborne mg/l Willming et al. (2016) Sodium chloride 96h-LC 50 Aquatic, waterborne mg/l Willming et al. (2016) Shell Canada Energy 96h-LC 50 Aquatic, waterborne mg/l Zinc (2013) 96h-LC 50 Aquatic, waterborne mg/l Willming et al. (2016) a Figure 26: Graphical representation of the Species Sensitivity Distributions (SSD) constructed for fish species with indication of hazard coefficient corresponding to the 5th, 20th and 50th percentiles (i.e. benchmark values covering endpoint-based sensitivity of 95%, 80% and 50% of the species in the SSD). Continues on the next page. Figure 26 (cont.) 5 HC 5 considered for comparison out of those retrieved or calculated for the same substance (the most sensitive endpoint is used in each case) EFSA Supporting publication 2017:EN-1251

72 b c d For seven of the eight compared substances, the HC 5 value for fish was lower than that obtained for amphibians (Figure 27). However, no statistically significant differences were found when comparing 72 EFSA Supporting publication 2017:EN-1251

73 log-transformed (to fit normal distribution, confirmed by Kolmogorov-Smirnoff tests) paired HC 5 values (t 7 = 1.86, p = 0.121). Glyphosate was the only tested substance resulting more toxic to amphibian than to fish species. However, we must remind that SSD of amphibians exposed to glyphosate was run considering formulations, which toxicity on amphibians is generally higher than that of the active ingredient (Mann and Bidwell 1999). Therefore, in this case the HC 5 value for amphibians could be overestimating the toxicity of the herbicide when compared to the value in fish. The results of the comparison of HC 5 values were in general congruent with the comparisons of individual endpoint data (section 3.4.1). A significant Pearson correlation was found between fish and amphibian HC 5 values (Figure 27, logtransformed data: R 2 = 0.785, p = 0.021), which, in combination with the general higher sensitivity of fish compared to amphibians, suggests that fish could be appropriate surrogates for toxicity of chemicals on aquatic stages of amphibians. However, it should be noticed that the number of substances suitable for comparison was low (N = 8, only half of which are currently used pesticides if copper compounds are considered), which could limit the relevance of this conclusion. Figure 27: Comparison of HC5 values between aquatic amphibian stages and fish for eight substances. HC5 values were estimated from Species Sensitivity Distributions or, in some of the fish values, retrieved from the literature. The dashed line illustrates the hypothetical equality between each pair of values. Dots above that line correspond to substances for which fish are more sensitive than amphibians, and vice versa Life history traits for population modelling of selected species Results of the literature review Figure 28 summarizes the outcome of the review of the 543 references for which full texts were retrieved, 19 of which were shared with the section in which the effects of chemical exposure were reviewed EFSA Supporting publication 2017:EN-1251

74 Figure 28: Progress of the different steps of the literature review with indication of exclusion criteria applied to excluded references about life history traits Summary of information The 204 references used for data extraction generated 1854 data records. The information retrieved is presented in the life_history_trait dataset available as electronic supplementary material, and distribution of records among the different species and life history traits is summarized in Table 15. In general, much information was available for the six selected species, which supports their choices as model species for future landscape or population models. The information available for the nattaerjack toad, the Hermann s tortoise and the sand lizard was especially abundant in comparison with the other three species. Table 15: Summary of records per species and life history trait recorded during the course of the literature review. Totals indicate how much information was retrived for each species. Stage Trait Triturus cristatus Egg / embryonic Epidalea calamita Hyla arborea Testudo hermanni Lacerta agilis Egg dimensions Hatching traits Coronella austriaca 74 EFSA Supporting publication 2017:EN-1251