Evaluation over time of the detection capability of a video-tracking system through daily exposure of Danio rerio to sodium hypochlorite, ethanol or bisphenol A

Evaluation Over Time of the Detection Capability of a Video-tracking System Through Daily Exposure of Danio rerio to Sodium Hypochlorite, Ethanol or Bisphenol A. MIGUEL ÂNGELO CAVALEIRO FERNANDES DISSERTAÇÃO DE MESTRADO APRESENTADA AO INSTITUTO DE CIÊNCIAS BIOMÉDICAS ABEL SALAZAR DA UNIVERSIDADE DO PORTO EM TOXICOLOGIA E CONTAMINAÇÃO AMBIENTAIS Miguel Ângelo Cavaleiro Fernandes Evaluation Over Time of the Detection Capability of a Videotracking System Through Daily Exposure of Danio rerio to Sodium Hypochlorite, Ethanol or Bisphenol A. Dissertation for the master degree in Environmental Toxicology and Contamination submitted to the Institute of Biomedical Sciences of Abel Salazar, from University of Porto. Supervisor - Doctor Luis Teles Category - Assistant Professor Affliation – Interdisciplinary Center of Marine and Environmental Research (CIIMAR)/Faculty of Sciences, University of Oporto Acknowledgments I would like to thank to several people who helped me in different levels in the realization of this thesis. Initially I would like to thank to my supervisor, professor Luis Oliva Teles for the opportunity to integrate this project, in a field that I really like and for all its help, knowledge, support and availability provided during the development of this work, over the year. I would also like to thank to all the professors of this master for all the lessons and also for the shared knowledge, especially to professor doctor Vitor Vasconcelos for accepting and allowing the realization of this thesis. To my master colleagues Tiago Afonso, João Amorim and Joana Machado thank you for all the help and support provided over the year. To all my master colleagues, many thank for these two years of shared experiences. I would also like to thank to my friends because they were always present and also for their ability to hear me talk sometimes almost incessantly about this thesis. And finally I would like to thank my parents for the opportunity of studying and following my dreams, for all the advice and help as well as to my family. Abstract Due to the increase of contamination sources worldwide the protection of the natural ecosystems has become a necessity and simultaneously an enormous challenge. It is necessary to develop new, faster and more efficient methods of detecting contamination. In several studies, behavior has proven to be a sensible endpoint which could be used to detect sub-lethal exposures. In a previous work was developed a video-tracking system using zebrafish locomotors behavioral analysis, to detect a sub-lethal concentration (9% 96h LC50) of sodium hypochlorite (SH). The aim of this work was to use this video-tracking system to determine whether the detection capability does not deteriorate after successive exposures of the zebrafish to ethanol, sodium hypochlorite (SH) or bisphenol A (BPA). Three similar video-tracking systems were conceived to record the movement of the zebrafish. In each system four experimental conditions (control, exposure to ethanol, BPA, or SH) were tested at the same time. Fish were exposed once a day for 9 consecutive days to these toxicants for 1h30m, but only the second half hour of each day was used in the analysis. One assay was performed and later was repeated a second time with new fish. In the end the zebrafish locomotor behavior was transformed into XY coordinates and 9 movement descriptors were calculated. A cluster analysis was conducted using Artificial Neural Networks (ANNs), of the type Kohonen to define different behavior categories of the fish submitted to the different experimental conditions, with the information about the movement descriptors. Several correspondence analysis were performed to obtain a measure of the effect caused by the toxicants, that then was analyzed in each day, by linear and orthogonal multiple regression models. The Presence/Absence model analyzed if the behavior of the fish was related with the presence/absence of the respective toxicant in the water. The Moment/Toxicant model allowed analyzing the progress of the behavior response, before and after adding the toxicants in the toxicant experimental units. This model was used to analyze the progress of the behavior response, in the moments before and after for the control experimental units. The Moment/Control model indicated that the behavior of the control fish was not influenced by the practical procedure, which means that the behavior changes detected were only related with the toxicants. The Presence/Absence model indicated that the system was able to successfully detect the three toxicants. With ethanol the detection capability was maintained, but in the case of the SH and BPA a deterioration of the detection capability over the days occurred. The Moment/Toxicant model revealed that all of the toxicants influenced the behavior, but for SH, and BPA a decrease in the amplitude between the Moments Before and After of the behavior effect over the days was detected. This response may be due to the induction of detoxification mechanisms, and biochemical changes that lead to a decreased effect of the toxicants in behavior, or due to the accumulation of adverse effects caused by the repeated exposure to the toxicants. In order to prevent the loss of detection capability some procedures can be adopted such as the regular exchange of fish. In the case of ethanol, the system was resistant to the repeated exposures. Through the ANNs, the correspondence analysis as well as the linear and orthogonal regressions it was possible to use the zebrafish behavior changes induced by the toxicants as a way to detect them, and it was also possible to evaluate the exposure conditions to which the fish were subjected. This study shows that the system was capable of detecting changes in fish behavior exposed to small concentrations, which indicates that it can be an important tool for early warning detections of contamination, it can help understand ecological consequences of exposure and have the potential to be integrated in ecotoxicological studies. Resumo Devido ao aumento das fontes de contaminação um pouco por todo o mundo, a proteção dos ecossistemas naturais tornou-se uma necessidade e ao mesmo tempo um enorme desafio. É necessário desenvolver novos métodos, mais rápidos e eficientes de deteção de contaminação. Em diversos artigos, o comportamento provou ser um parâmetro sensível, e que poderia ser usado para detetar exposições sub-letais. Num trabalho anterior foi desenvolvido um sistema de vídeo-rastreio utilizando a análise do comportamento locomotor do peixe-zebra, para detetar uma concentração sub-letal de hipoclorito de sódio (9% 96h LC50). O objetivo deste trabalho foi utilizar este sistema de vídeo-rastreio para determinar se a capacidade de deteção não se deteriora após exposições sucessivas do peixe-zebra a etanol, hipoclorito de sódio ou bisfenol A. Conceberam-se três sistemas de vídeo rastreio iguais, para registar a movimentação dos peixes-zebra, e testaram-se em cada sistema quatro condições experimentais ao mesmo tempo (controlo, exposição ao etanol, bisfenol A, ou hipoclorito de sódio). Os peixes foram expostos uma vez por dia durante 9 dias consecutivos a estes tóxicos durante 1h30m, mas apenas a segunda meia hora de cada dia foi utilizada na análise. Realizou-se um ensaio, que depois foi repetido uma segunda vez com peixes novos. No final as movimentações dos peixes-zebra foram transformadas em coordenadas XY e 9 componentes de comportamento foram determinados. Foi realizada uma cluster analysis, utilizando uma Rede Neuronal Artificial do tipo Kohonen, para definir diferentes categorias de comportamento dos peixes submetidos às diferentes condições experimentais, com as informações sobre os movimentos descritores. Foram realizadas várias Análises de Correspondência, para se obter uma medida do efeito causado pelas substâncias tóxicas, que, em seguida, foi analisada em cada dia, por modelos de regressão linear múltipla e ortogonal. O modelo Presença/Ausência analisou se o comportamento do peixe estava relacionado com a presença/ausência da respetiva substância tóxica na água. O modelo Momento/Substância tóxica permitiu analisar a evolução da resposta comportamental, antes e depois da adição das substâncias tóxicas nas unidades experimentais substâncias tóxicas. O modelo Momento/Controlo foi utilizado para analisar o progresso da resposta comportamental, nos momentos antes e depois para as unidades experimentais controlo. Este modelo indicou que o comportamento dos peixes controlo não foi influenciado pelo procedimento prático, o que significa que as mudanças de comportamento detetadas estavam apenas relacionadas com as substâncias tóxicas. O modelo Presença/Ausência indicou que o sistema foi capaz de detetar com sucesso as três substâncias tóxicas. Com o etanol a capacidade de deteção manteve-se, mas no caso do hipoclorito de sódio e do bisfenol A, ocorreu uma deterioração da capacidade de deteção ao longo dos dias. O modelo Momento/Tóxico revelou que todas as substâncias tóxicas influenciaram o comportamento, mas para o hipoclorito de sódio, e o BPA, foi detetada uma diminuição na amplitude entre o tempo Antes e Depois do efeito comportamental ao longo dos dias. Esta resposta pode ser devida à indução de mecanismos de desintoxicação e alterações bioquímicas, que levam a um efeito reduzido do tóxico no comportamento, ou devido à acumulação de efeitos adversos causados pela exposição repetida às substâncias tóxicas. De modo a impedir a perda de capacidade de deteção, alguns procedimentos podem ser adotados, tais como a troca periódica dos peixes. No caso do etanol, o sistema mostrou-se resistente às exposições repetidas. Através das Redes Neuronais Artificias, da Análise de Correspondência e da regressão linear múltipla e ortogonal, foi possível usar as mudanças de comportamento do peixe-zebra induzidas pelas substâncias tóxicas como uma forma de detetá-las, e também foi possível avaliar as condições de exposição a que os peixes foram submetidos. Este estudo demostra que o sistema foi capaz de detetar alterações comportamentais nos peixes expostos a pequenas concentrações o que indica que pode ser uma ferramenta importante para deteções de alerta precoce de contaminação, pode ajudar a compreender as consequências ecológicas da exposição e tem o potencial para ser integrado em estudos ecotoxicológicos. i Table of Contents 1. INTRODUCTION ........................................................................................................... 1 1.1. Behavior Analysis ............................................................................. 4 1.2. Zebrafish ... 1.3. Characterization of ANNs .................................................................. 8 1.4. Tolerance .......................................................................................... 9 1.5. Toxicants ........................................................................................ 10 1.5.1. Sodium Hypochlorite ....................................................................... 10 1.5.2. Bisphenol A ..................................................................................... 13 1.5.3. Ethanol ............................................................................................ 16 1.6. Toxicant Selection ........................................................................... 19 2. Objectives ..................................................................................................................... 19 3. MATERIAL AND METHODS ...................................................................................... 21 3.1. Organization of Experimental Material ............................................ 21 3.2. Arenas ............................................................................................. 21 3.3. Recording Areas .............................................................................. 22 3.4. Origin of Test organism .................................................................. 24 3.5. Exposure concentrations ................................................................. 24 3.6. Experimental plan ........................................................................... 26 3.7. Video treatment .............................................................................. 28 3.8. Statistical Analysis ........................................................................... 31 3.8.1. Cluster Analysis ............................................................................... 31 3.8.2. Anova and post-Hoc Test ................................................................ 33 3.8.3. Correspondence Analysis ................................................................ 33 3.8.4. Saturated Orthogonal Multiple Linear Regression Analysis .............. 35 3.8.4.1. Presence/Absence Model ...................................................... 35 3.8.4.2. Moment/Control Model......................................................... 36 1 1. Introduction Water is a vital resource for all organisms, including humans. With the increase of human population, the pressure on the water resources has also increased due to industrial development, agriculture, and domestic uses thousands of chemical substances are continuously reaching and contaminating many water resources all over the world (Houtman, 2010). Degradation of water quality poses serious ecological problems and to protect not only human health but also natural ecosystems its necessary to determine the toxicity of chemical substances, and detect quickly situations of environmental contamination (Storey et al., 2011). Risk assessment evaluates the likelihood of adverse effects occurring in ecosystems and relates the disturbance with the magnitude of the impact (Wright and Welbourn, 2002). It has 4 steps: the first is the hazard identification, which aims to determine whether exposure to a chemical substance can cause adverse effects and characterize the strength of the evidence that can have this effect, the second is the dose-response evaluation which aims to determine the relation between the dose of exposure to a contaminant and the following effect, the third is the exposure assessment that analyzes the magnitude and duration of the exposure to the agent and the fourth is risk characterization that summarizes the information of the 3 previous steps and analyzes the relation between the dose and the probability of occurring the adverse effect. Risk characterization puts the assessment of risk in a form that is useful for the competent authorities responsible for the decisions (Wright and Welbourn, 2002). Risk management occurs after risk assessment, and the objective is to take action and minimize the risk and the costs (Wright and Welbourn, 2002). Toxicology is the study of adverse effects of chemical substances on organisms (Chapman, 2002). Toxicity tests are used to determine the concentrations of a substance and the exposure time required to produce critical effects such as mortality, alterations in growth or reproduction (Wright and Welbourn, 2002). These tests can help to understand the mode of action and the physiological or other type of effects of the chemical substances on the organisms (Chapman, 2002).The majority of such tests are conducted under controlled conditions (temperature, water quality, pH) in the laboratory 2 (Chapman, 2002). Institutions such as the Organization for Economic Cooperation and Development, the International Standardization Organization, the United States Environmental Protection Agency and the American Society for Testing and Materials) described several standard toxicity tests (Befyaeva et al., 2010). In this standard tests the exposure of organisms to the test solution can be semi-static regime where the frequency of medium renewal normally depend on the stability of the test substance, or flow-through regime, which continually dispenses and dilutes a stock solution of the test substance (Wright and Welbourn, 2002). Toxicity tests can be acute or chronic. Acute tests are accomplished for relative short periods of time normally between 48h to 96h, and the acute toxicity testing is usually determined by the concentration that is lethal to 50% of the test organisms i.e. the median lethal concentration (LC50) (Wright and Welbourn, 2002). These tests are more used because they are simple to execute and produce fast results (Magalhaes Dde et al., 2007). Chronic tests are executed for longer periods of time, and the chronic toxicity is determined by the lowest concentration that caused a statistically significant effect observed (LOEC) in the organisms and also by the highest concentration that has no statistically significant effect observed in the organisms (NOEC) (Wright and Welbourn, 2002). Chronic tests are designed to detect mostly sub-lethal effects on growth and reproduction (Wright and Welbourn, 2002). However these types of studies are expensive and not practical, because they require a lot of work and time. The toxicological tests can also be executed on the field, in this case they are called ecotoxicological tests and they allow a better comprehension of the effects of the chemical substances on the organisms because they are performed under natural exposure conditions (Chapman, 2002), however ecotoxicological test may also be developed in laboratory. Ecology studies the interactions between organisms and their environment. Ecotoxicology comprises the toxicology and ecology and is the study of the effects of toxic substances on live organisms, populations and communities inside defined ecosystems. The objective of ecotoxicology is to be capable of predicting the effects of toxic substances on natural communities under natural exposure conditions (Chapman, 2002). I exposure to sub-lethal concentrations (low concentrations) instead of acute toxicity to contaminants 3 (Houtman, 2010) the exposure of organisms and rapidly detect situations of contamination. Water monitoring is normally based on chemical analysis (direct identification of substances). These methods have some disadvantages including the discontinuity of sampling (e.g. 3 times per year) that may fail to identify intermittent discharges to the environment. In some cases the interval time between sampling and the results is also a disadvantage. The fact of not all chemical substances are included in the chemical analysis and because of that the detection is not always fast enough to prevent the occurrence of effects in organisms is another disadvantage (Gonzalez et al., 2009). The monitoring of water can also be achieved by physical-chemical parameters and biological methods that use for example physiologically, biochemical or behavior alterations of organisms and also changes in populations dynamics or in community structures to detect contamination (Gonzalez et al., 2009). The physical-chemical analysis can have lack of sensitivity but biomonitoring that uses the organisms to assess changes in the environment is typically sensitive to many chemical substances (Gonzalez et al., 2009). Biomonitoring is more relevant because it uses the mixture of chemical substances existing in the environment and can allow the organisms to integrate over time the potential toxic effects of different chemical substances throughout the life cycle also indicating the overall effects on aquatic ecosystems (Gerhardt, 1995). Bioindicators are organisms used to monitor environment and ecosystems (Gonzalez et al., 2009). Behavior is becoming increasingly important for ecotoxicology being used in several studies including in biomonitoring in order to early detect water contamination (Magalhaes Dde et al., 2007). Behaviors reactions are vastly integrative responses, because behavior is related with biochemical and physiological processes (Brewer et al., 2001). Sub-lethal exposures to toxicants can trigger behavioral responses that allow quantitative measures of mechanisms modifications, which may have the potential to provide knowledge about individual and population effects of environmental contamination (Brewer et al., 2001). The increasing integration of behavior analysis in toxicological and ecotoxicological protocols could help to better understand the impact and effects of chemical exposure in organisms. With the constant development of technology and statistical analysis, behavior could be valuable also for biomonitoring the 4 environment to avoid situations of pollution by detecting toxic substances, increasing the accuracy of risk assessment. 1.1. Behavior analysis Behavioral toxicology in recent years has received an increased attention essentially because of the automation of the techniques to obtain and treat data (Bae and Park, 2014). Behavior is a selective response to internal and external stimulus (Gerhardt, 2007). It is an extremely structured order of actions and reactions designed to allow the best conditions possible in relation to fitness of the organism in the environment (Tierney, 2011). Because higher concentrations are easier to test and analyze, a large number of toxicological assays use them to achieve results quicker (Magalhaes Dde et al., 2007). In most cases environmental contamination for the majority of contaminants only occur in natural aquatic systems at low concentrations that despite not be sufficient to cause mortality may lead to ecological functions losses (Houtman, 2010). These situations happen through behavioral alterations that may affect for example, predation and olfactory capacity (Scott et al., 2003). Behavioral changes induced in organisms by exposure to toxicants are usually subtle and may be detected at lower concentrations than those that cause permanent or irreparable damage with more serious consequences for the organisms, and therefore may be detected before the permanent effects (Scott and Sloman, 2004). The majority of toxicological studies that use sub-lethal concentrations usually evaluate only the effects on chronic developmental or reproductive endpoints (Scott and Sloman, 2004) because they are typically easier to relate with the health of the organisms, although these may be more expensive and time consuming (Melvin and Wilson, 2013). However substantial technological improvements in computers, image analysis and video automation have made it easier and affordable to obtain interpret and apply behavioral endpoints for quantifying behavior in toxicity evaluation (Bae and Park, 2014). 5 Nowadays, one of the most used tools to motorize the activity of organisms is video-tracking. Through the use of video cameras to perform the recordings, the signal from the camera can be converted into a numerical video file and sent to a computer. Then it can be processed in real time by the software or it can be stored and examined later to avoid errors such as shadows (Delcourt et al., 2013). With video-tracking, not only locomotor activity (behavior responses to toxicants) common to several animal taxa can be analyze, but also more complex behaviors (e.g. social interactions, predation, feeding and mating), which might allow a greater understanding of the ecological impacts of environmental contamination (Scott and Sloman, 2004). Alterations in locomotor activity may affect the performance of different behavioral tasks such as the ability to prey (with consequences to growth and longevity), or foraging from predators (Little et al., 1990). Software like LocoScan and EthoVision have been used in numerous studies (Blaser and Gerlai, 2006; Gerlai et al., 2006; Egan et al., 2009) to analyze different effects, on behavior of innumerous substances (e.g. pesticides, personal care products, pharmaceuticals, drugs, ethanol). These systems allow analyzing overall activity through endpoints like distance travelled, acceleration and angular velocity (Brewer et al., 2001; Magalhaes Dde et al., 2007; Liu et al., 2011b), but also other types of behaviors for example social interactions, feeding, or predators avoidance of many different fish species. A meta-analysis of the literature done by Melvin and Scott (2013) comparing studies that assess behavioral responses with acute lethality, developmental and reproductive procedures revealed that behavioral studies are in general more sensible and less time consuming. In fact concentrations ranging from 0.1 to 5.0 percent of the LC50, were the lowest behaviorally effective toxicant concentrations that caused changes in fish swimming behavior (Little and Finger, 1990). In studies with multiple observations, changes in behavior occurred commonly 75% earlier than the onset of mortality (Little and Finger, 1990). Behavioral toxicology as the advantage to be non-invasive, so the implementation of this type of analyses can be a powerful addiction in toxicological investigation in order to obtain more accurate consequences of 6 exposure, especially to environmental low-level exposures (Little and Finger, 1990). With the purpose of monitoring water quality and protect aquatic ecosystems, biological early warning systems (BEWSs) have been developed to offer a rapid warning of contamination incidences (Gonzalez et al., 2009). These systems are capable of detecting different responses of organisms to disturbance, because organisms can detect a wide range of pollutants. The Fish Toximeter, the ToxProtect, and the Aqua-Tox-Control are commercially available systems that use behavioral parameters. Some commercially available systems (Multispecies Freshwater Biomonitor, Biological Early Warning System, bbe Fish and Daphnia Toximeter), use more than one specie because different species have different sensitivity and reaction time to the same contaminants (Bae and Park, 2014). Some matters still need to be resolved such as the problems related with data treatment (large quantity of data), advanced processing, the selection of appropriate bioindicators for each environment and the intrinsic variation of behaviors between organisms of the same specie (Bae and Park, 2014). However the development of BEWSs capable of detecting various types of pollutants with great sensitivity, with fast and reliable detection of adverse situations (faster than chemical detection) and with minimal cost of maintenance that are easy to use, could be a future worldwide solution, especially when integrating BEWSs with chemical monitoring. Chemical detection may be necessary to identify the toxicant because even though BEWS can detect a reaction to one substance in a toxicity test, in a natural environment where a mixture of toxicants is present most of the time, to identify the toxicants (although identification is not the purpose of the BEWS). For example after an alarm of the BEWs, the substance that caused the biological effect can be identified by water chemical analysis (Gonzalez et al., 2009). In fact in Europe, specifically in Rivers Elbe, Meuse, Rhine there are BEWSs working in programs of biomonitoring (Gonzalez et al., 2009). In the river Rhine, the results of BEWSs, are always complemented and confirmed with physicochemical water analysis (Diehl et al., 2006). The majority of organisms are sensitive to more toxic substances than some conventional analytical methods that are part of aquatic monitoring 7 (Gonzalez et al., 2009). Several organisms have been used in BEWSs including bacteria, algae, invertebrates and fish (Gonzalez et al., 2009). Fish have an important role in the trophic chain and significant commercial value, being therefore important to protect these organisms from contamination (Viarengo et al., 2007). In this sense fish have been used for the monitoring of drinking water, wastewater effluents, surface water and aquaculture (Bae and Park, 2014). 1.2. Zebrafish Different types of fish species are used in behavioral testing to toxicant exposures or other types of stimulus. Fish have some specific advantages, such as direct contact with the aquatic environment (body surface), ecological important behaviors that are easily observed and quantified in controlled environment, the well documented life cycle that some species have (Scott and Sloman, 2004) and because the early stages of its life cycle are extremely sensitive to contaminants. One of the species most used is the zebrafish (Danio rerio). The zebrafish is a small (up to 4.5 cm), tropical freshwater teleost fish belonging to the family Cyprinidae of the order Cypriniformes that it is native of the Himalayan region (South Asia) (Befyaeva et al., 2010). A great number of Zebrafish can be easily maintained in the laboratory with a relatively low cost, a female can produce up to 200 transparent eggs every other day (Blaser and Gerlai, 2006; Gerlai et al., 2006), it has a rapid reproductive cycle (Befyaeva et al., 2010), its genome has been completely sequenced and several genes of high mammalian homology also have been discovered (Tierney, 2011). Zebrafish has been identified as an excellent model for pharmacology, toxicology and pharmacogenomics studies (Gerlai et al., 2006). Exposure to novelty evokes robust anxiety responses in zebrafish, as with rodents (Blaser and Gerlai, 2006). In the last few years, zebrafish have been used in several paradigms adapted from rodents with video-tracking systems such as the Novel tank test (exposure to a novel arena where vertical behavior is analyzed), open field test (exposure to a novel arena where horizontal behavior is analyzed), the light-dark box as a measurement of scototaxis (dark/light preference), shoaling (measures the 8 effects of anxiety on social behavior) after pre-treatment with for example anxiolytic substances (ethanol and fluoxetine), or potential anxiogenic substances (caffeine) (Egan et al., 2009; Maximino et al., 2011). The effect of predators in behavior has also been studied (Gerlai et al., 2000; Gerlai et al., 2009). Zebrafish behavioral analysis through toxicant induced modifications can be an important process to study the function of the brain (Blaser and Gerlai, 2006). Consequently pecies in toxicological studies including studies that help to uncover the mechanisms of action of certain substances, and their effects on the behavior, but also in biomonitoring studies as an early warning signal. Several statistical analyses can be used in the behavior evaluation, to enable a better understanding of the behavior and to increase the detection sensitivity. 1.3. Characterization of ANNs In the last few years several computational analyses were developed to handle data of behavior tracking. Some techniques include ANNs (Artificial Neural Networks) such as Multi-layer Perceptron (Kwak et al. 2002) and Self-organizing map (SOM) (Park et al., 2005; Liu et al., 2011b). The ANNs have been applied in several areas, because they have numerous characteristics that make them interesting and attractive for prediction (Teles et al., 2006). In contrast to traditional model-based methods the ANNs are nonparametric data-driven self-adaptive methods and because of that incorporate few apriori assumptions (Zhang et al., 1998). They are able to learn from samples and respond to subtle undetected functional relations between the data (Zhang et al., 1998). ANNs can also generalize and can frequently predict an occurrence even with noisy information (Zhang et al., 1998). They are capable of approximate nonlinear multivariate functions to any desired accuracy (Zhang et al., 1998). These characteristics make them ideal to decipher problems that are too complex for conventional statistical methods (Zhang et al., 1998). 9 In several studies, ANNs are used to identify (Kwak et al., 2002) or classify (Park et al., 2005) standard movements previous established by the authors or to set themselves these movements patterns based on variables that describe the fish movement (Liu et al., 2011b). However, the potential of the ANNs has not been fully explored. Only in one previous work (for publication), the ANNs together with the correspondence analysis (SOM-CA) were directly and successfully applied in the detection of toxic substances in the water through the zebrafish behavior. So the ANNs have the potential to evaluate the different behavioral responses that may occur when organisms are exposed to toxicants along time. 1.4. Tolerance The development of tolerance is a response that may occur when organisms are exposed continuously to toxicants, and this response may affect the detection capability of video-tracking systems. presence of different types of contaminants in the environment, especially in aquatic ecosystems (Wirgin and Waldman, 2004). Normally they exist at low concentrations with the exception of punctual discharges that introduce or greatly increase the concentration of some pollutants. These compounds may cause lethal or sub-lethal effects that can impair behavior, morphological or biochemical processes depending on the mode of action of the contaminant. However organisms have the capacity in some cases to tolerate toxicants in contaminated situations and for example fish populations frequently survive and prosper in highly polluted sites (Wirgin and Waldman, 2004). This tolerance may be due to adaptation, this is genetically based resistance. Normally, occurs in relatively long time scales (several generations) at the population level, and represents the plasticity of the genotype (Meyer and Di Giulio, 2003). In a population, where normally exists variability, and in the constant presence of contaminants, the individuals more resistant are selected, resulting in more resistant generations (Ownby et al., 2002). If this selective sappear, although not instantly (Wirgin and Waldman, 2004). Adaptation may cause reduction of 10 fitness, for example greater sensitivity to other chemical substances and abiotic factors (e.g. salinity), but it can also increase tolerance to other substances (Wirgin and Waldman, 2004). Embryos and larvae Killifish (Fundulus heteroclitus) of adults collected from New Bedford Harbor (USA) that is contaminated with polychlorinated biphenyls (PCBs) were more resistant than embryos and larvae of fish from reference sites after the exposure to PCBs, being the concentrations necessary to produce sub-lethal and lethal effects of two orders of magnitude higher than the ones producing effects in reference site embryos (Nacci et al., 1999). Ownby et al (2002) also proved that wild populations of Fundulus heteroclitus (collected form the Elizabeth River), had inherited tolerance to polycyclic aromatic hydrocarbon (PAH). Alteration of the biological functions can also occur at the individual level, when an organism pre-exposed to a particular contaminant become less sensible to his effects when another exposer occurs (acclimation) (Klerks, 1999). This process does not include alterations in the pear in remediated environments (Wirgin and Waldman, 2004). The tolerance to the toxicants probably arises from, the induction of mechanism, such as metallothioneins (proteins), that have the capacity to bind to heavy metals, and act as defense mechanism to protect the organisms (Perez-Coll et al., 1999). Other mechanisms include reduced uptake, storage of toxicant in isolated structures, biotransformation of the toxicant into inactive metabolite, or elimination from the cell by excretion or secretion (Wright and Welbourn, 2002). These mechanisms probably require more energy which prevents energy storage, and may cause consequences on the organism (Holmstrup et al., 2011). 1.5. Toxicants To test the detection capability of the video-tracking system over the days, three toxicants belonging to different chemical groups were selected. The toxicants selected were sodium hypochlorite, ethanol and bisphenol A. 1.5.1. Sodium hypochlorite (SH) 17 metabolizes the acetaldehyde in to acetate (Swift, 2003). In the liver it can also be metabolized by the CYP2E1 and by the enzyme catalase located in the peroxisomes of hepatocytes (Swift, 2003). Acute effects in humans are muscular incoordination, visual impairment, decreased reaction time, behavior changes and severe intoxication can lead to vomiting, nausea and hypothermia and even eventually coma, hypertension, and death (Strohm and Sweet, 2005). Chronic consumption can lead to several types of liver damage (cirrhosis and alcoholic hepatitis), cancer, cardiac problems and during pregnancy can lead to congenital malformations (fetal alcohol syndrome) such as mental deficiency (Strohm and Sweet, 2005). Ethanol is a substance that acts in the brain. Short-term alcohol exposure, normally enhance the action of GABA (Gamma-AminoButyric Acid) and glycine, (inhibitory neurotransmitters) by increasing the function (inhibitory effects) of their receptors GABAA (GABA receptor) and GlyR (glycine receptor) (Mihic et al., 1997). This explains why alcohol in some circumstances decreases anxiety. On the other hand ethanol inhibit the action of glutamate (excitatory neurotransmitter) by inhibiting is receptors, NMDA (N-Methyl-D-aspartate), and -amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid), and KARs (kainate receptors) although ethanol is considered a weak agonist of NMDA even at high concentrations (Dildy-Mayfield and Harris, 1995; Wright et al., 1996). Ethanol also affects the 5-HT3 (5-hydroxytryptamine3) receptor, where the neurotransmitter serotonin (modulator of physiological functions, including perception, aggressiveness, anxiety, sexual behavior) binds, and the nicotinic acetylcholine receptors (nAChRs), where the neurotransmitter acetylcholine binds (Cardoso et al., 1999; Davies et al., 2006). Chronic exposure to ethanol has the contrary effect, which is decreased inhibitory neurotransmission and increased excitatory neurotransmission, in an attempt to reach equilibrium (Faingold et al., 1998). This process, may explain the development of tolerance, for example animals previously exposed to ethanol appear resistant to its effect when compared with controls. Studies with zebrafish reported that ethanol can increase the activity for example of AChE (Rico et al., 2007), and chronic exposure can lead to different significant gene expression levels in brain, suggesting an adaptive response (Pan et al., 2011). Although ethanol is not a threat from an environmental perspective, many studies have shown that ethanol can affect many aspects of zebrafish behavior. 18 Gerlai et al. (2000) exposed zebrafish for 1h in an acute treatment to 0.25% v/v, or 0.5% v/v (considered intermediate doses), and the fish presented increases in general activity and aggression, and diminished fear (distance to the image) in relation to an predator image and shoaling. This type of behavior is linked to the anxiolytic properties of ethanol. At higher concentration (1% v/v), decreased the activity, however the authors argued that this response, is probably due to sedative effects of ethanol. Egan et al. (2009) demonstrated that zebrafish exposed to ethanol (3% v/v) acute treatment (5 minutes), presented decreased erratic movements and increased exploration, and ethanol (3% v/v) chronic treatment (7 days), caused increased exploration, average velocity and total distance travelled, all considerate anxiolytic effects. Larvae zebrafish (6 days post fertilization) exposed to ethanol demonstrated hyperactivity at lower concentrations (0.5 2% v/v) and hypoactivity at higher concentration (4% v/v) (de Esch et al., 2012). Gerlai et al. (2006) also showed that zebrafish exposed chronically, for two week to ethanol (0.25% v/v) had developed a significant tolerance to this substance. Fish exposed to the highest (1.00% v/v) acute ethanol concentration (for 1h) that were exposed earlier to the chronic ethanol treatment, presented decreased distance from the predator, an anxiolytic effect that was not attenuated by the chronic exposure. All zebrafish exposed only to acute ethanol concentrations (0.25% v/v, 0.5% v/v, 1% v/v) presented an almost linear concentration response, and the higher ethanol concentration increased the total path length swum by the fish. One of the explanations for this particular result is that maybe the strain (long-fin wild type) was more resistant (than the one used in 2000), and thus a higher concentration was necessary to cause hypoactivity (Gerlai et al. 2006). Some behavior variation has been reported between zebrafish species exposed to similar ethanol concentrations, as the previously mentioned (Dlugos and Rabin, 2003). Gender differences were also detected in chronic ethanol (0.5% v/v) exposed wild-type zebrafish (10 weeks), because females had present this response (Dlugos et al., 2011). Tran and Gerlai (2013) also demonstrated in a 1h time-course experience (behavior monitoring during all period), that zebrafish exposed acutely to ethanol (1.00% v/v) exhibit an inverted U shaped trajectory in distance travelled. These authors argued that initially, the temporal trajectory increased due to an elevation of ethanol levels in the brain resulting in stimulation, and then the levels reached the maximal blood/brain ethanol levels which lead to a depression in the trajectory of the distance 19 travelled. However zebrafish in the same situation that had been previously exposed to chronic concentration (0.5% v/v) demonstrated a blunted response, suggesting tolerance. 1.6. Toxicant Selection BPA and SH are capable of causing behavioral changes in zebrafish, (Magalhaes Dde et al., 2007; Nimkerdphol and Nakagawa, 2008; Saili et al., 2012; Wang et al., 2013) and in this sense are good objects of study. Furthermore, because they are widely used, they are constantly released in the environment, and from this point of view are ecologically relevant. They also have other advantages, including all the three toxicants have known values of 96h LC50 for zebrafish and a high solubility in water, even BPA (moderately soluble) that has the lowest solubility in water 300 mg/L (Shareef et al., 2006). Ethanol has a lower toxicity than the others two toxicants, but it causes different effects on behavior. Several studies such as Dlugos and Rabin, (2003), Gerlai et al., (2006), Gerlai et al., (2009), Dlugos et al., (2011), Tran and Gerlai, (2013), Tran et al., (2014), use ethanol in chronic exposures with zebrafish, and proved the development of behavior tolerance, which makes this toxicant ideal to serve as positive control , to verify if all the experimental procedure and the subsequent statistical analysis are suited to detect this type of behavior, that eventuality may occur. 2. Objectives After the development in previous works (for publication) of a videotracking system with zebrafish, which proved to be very accurate and sensitive, the main objective of this work was to determine whether the detection capability of this system does not deteriorate after successive exposures of the zebrafish to the toxicants ethanol, SH and BPA. Other aims included: 20 Use the zebrafish behavior changes induced by environmental disturbances (toxicants), as a way of detecting them, through the time series of the Kohonen ANNs. Use the ability of the Kohonen ANNs, the correspondence analysis and the saturated orthogonal multiple linear regression analysis as a method of diagnosis of environmental conditions to which the test organisms were exposed. 21 3. Material and Methods 3.1. Organization of Experimental Material Initially, a Tetra aquarium was assembled with a capacity of 50 liters (64x34x29 cm; length x width x height), equipped with a Trixie aquarium heater (50 Watts; 30-60 L), and one Trixie water pump filter (7 Watts; 40-60 L), capable of chemical cleaning, through activated carbon for chlorine, and biological cleaning by sponge. Four Aqua Cyan 15 aquariums (34x17x24 cm; length x width x height), with a maximum capacity of fifteen liters equipped with aquarium heaters Aquapor 25 and water pumps filters Rena H20 Max 35 were also assembled. Each of these four aquariums was divided into three equal parts, through the use of fine a fin net, in order to isolate simultaneously and separately twelve fish (three per aquarium). 3.2. Arenas To assemble three arenas sets (the tanks where the fish were filmed) with the dimensions of 35x20x15 cm (length x width x height), glass plates 2mm thick were glued together, and they were internally divided by glass plates in four equal divisions (each division represents a recording arena) in order to isolate one fish per arena. The bases of the arenas were also glass plates 2mm thick. The external glazing of the recording arenas was frosted glass that allowed greater contrast of footage, however to the internal glass plates (that divided the arenas) white plastic boards were glued to eliminate visual contact and completely isolate the fish from each other. 22 Figure 2: Scheme of the arenas. A-Control, B-SH, C-Ethanol, DBPA. 1Aqua Pro internal filter M200 water pump, 2Trixie aquarium heater (25 W; 15-30 L), 3Water bath. The horizontal plane was used because the activity and movement of zebrafish is higher in this plane that in the vertical plane (Vogl et al., 1999). The depth of water in the arenas was reduced to a minimum value (10 cm 1.5 L) to further reduce the vertical movement, but this reduction did not affect the overall level of fish activity. With this video-tracking program, the fish were kept individually in each organisms in the same arena, when they were close to each other, due to overlapping. This problem is also an obstacle to other screening programs, because it can lead to fish incorrect identifications (Delcourt et al., 2009). 3.3. Recording Areas The arenas were placed in three recording areas (figure 2), each one with one water bath, (48x17x36 cm; length x width x height), that had a transparent base, an Trixie aquarium heater (25 Watts; 15-30 L), and a Aqua Pro internal filter M200 water pump (5 Watts; up to 45 L). The walls were covered by silver foil to reflect the light to the arenas, and about 10 liters of water which were always changed before each trial. Underneath the water bath were placed twelve LED cylindrical lamps of 60 Watts (four for each area) to uniformly illuminate the arenas. The basis of the arenas was made of glass and the base of the water baths was transparent, because in this case the light can pass through those surfaces and better highlight the fish silhouette which improves the footage quality. The LED lamps have been selected because unlike incandescent lamps, they do not heat significantly and therefore do not increase the temperature of the water bath and consequently of the arenas. Through the aquarium resistances previously mentioned and with the help of four Digital Internal/External Thermometers (Model RT 801 Version 12) with an accuracy of ±0.5ºC, the temperatures of the aquariums and of the arenas were 23 set and maintained to 28±0.5ºC, the ideal for the species in question. The culture medium used in the arenas consisted of tap water aged at least 48h, in a 50 litter aquarium, with no fish and water pumps with chemical and biological cleaning. Figure 3: Scheme of one recording area. 1Polyurethane isolation, 2Led lamps, 3Water bath, 4540L IR camera, 5Polystyrene, 6Expanded cork. To minimize possible disturbance such as vibrations originating from the ground, but also sounds that could disturb the experience, specifically noises, the walls (sides and back) of the recording area were covered with expanded cork boards, and above the cameras, expanded cork boards and polystyrene were also placed in supports (these last ones facing inwards to reflect the light). On the surface of the table where the recordings took place, polyurethane isolation was also applied to minimize possible vibrations originating from the ground (figure 3). When recordings occurred, expanded cork boards were put in the front, to isolate completely the entire system and the arenas. Furthermore, in the expanded cork boards were fixed white cardboard to reflect the light and thereby obtain an enhanced contrast of the fish in the arena, which facilitates image processing. Three 540L IR cameras (model CACO0008) flow electronics with super high color resolution were used in the recordings (one for each recording area). The data were stored in one Intel® Pentium® Dual CPU computer (2:00 [email protected] 24 GHz, 1.87 GB RAM) system with a Microsoft Windows XP Home Edition version 2002 Service Pack 3, through the DSS1000 program version 4.7.0041 of 2004. 3.4. Origin of Test organism In this study, about 40 wild-type adult zebrafish (Danio rerio) with about 3 months of age and 2.5 to 3.0 cm length from the same batch (ORNI-EX, Lda, Arcozelo, Vila Nova de Gaia, Portugal) were used. The fish were acclimated for two month prior to any test in one aquarium tetra of 50 liters, because of the risk of mortality due to a possible disease, habituation to the new environment or simply due to the intrinsic differences between fish even being from the same batch. However in this case mortality was not observed. The fish were fed once a day with TetraMin Bio Active Formula, 1h after the end of each trial, and kept in a 12h of light/12h of dark photoperiod. The fish were used exclusively in these trials. 3.5. Exposure concentrations The concentrations of the toxicants used must be ecological relevant (in the range of those found in natural environments), to obtain more realistic consequences of exposure, but at the same time the concentrations should be sub-lethal to zebrafish, because the objective is to see if the normal behavior could be impaired, without compromising the fish physical integrity. To fulfill these conditions, the toxicant concentrations used were 9% of the 96h LC50 for the three toxicants. The selection of the toxicant concentrations was based in a previous experience, were the established concentration was 0.5 mg/L of SH. This concentration was extremely low because it represented 1% of (48 mg/L)) or approximately, 9% of the 96h LC (5.5 mg/L) (Magalhaes Dde et al., 2007; Pitanga, 2011), and did not cause permanent damage or death to the zebrafish. In the case of BPA, the concentration (0.891 mg/L) used is also ecological relevant, because although in the environment normally much lower concentrations are detected, higher concentrations have 25 already been detected in sewage sludge, for example 1.363 mg/kg according to Fromme et al., (2012). Even taking into account that higher concentrations than 100 µg/L are capable of causing endocrine effects (Lahnsteiner et al., 2005; Mandich et al., 2007), normally long exposures are required (days or weeks) to produce this effects, but in the present work the daily exposure time was much shorter (1h30m) and only for 9 days. The detection capability and reliability of the system was tested because these concentrations were very low, perhaps to an incipient sub-lethal level. Table 1: From a Panreac proanalysis SH solution with a purity of 7% (w/w) and a density of 1.15 kg/L, it was prepared the required concentration. For ethanol it was used ethanol absolute (Fisher Chemical), Analytical reagent grade with a purity of 99.5%. The density of ethanol was considered to be 0.790 kg/L = 790 mg/ml at room temperature, (roughly 20 ºC). For BPA to achieve the 9% concentration required (0.891 mg/L), 30 mg of 2,2-bis(4-hydroxyphenyl)propane for synthesis (BPA) were added to a 1L volumetric flask filled with water and mixed in an agitator for 24h before each trial at 25.0°C. The solubility of BPA is 300±5 mg/L at 25.0±0.5°C (Shareef et al., 2006) and to allow a clear dissolution of BPA in the water, a factor 10 (ten times less) was use to prepare the stock solution (30 mg/L). The volumetric flask was wrapped in silver foil, to prevent photodegradation. For all toxicants the dilution water was dechlorinated tap water filtered through activated carbon, (Staples et al., 2011). To the respective arenas was added 9.6 µL of SH, or 46 ml of the stock solution prepared for BPA, or 2.51 mL of ethanol. To maintain the same depth of water in all the arenas, so that a possible difference in this parameter would not influence the behavior of the fish, which ultimately could had led to analysis errors, dechlorinated water at 28ºC was added to the arenas with 50 ml beakers. To the control and SH arenas was added 46 ml of dechlorinated water, and to the ethanol arenas was added 44.5 ml. All the toxicants were purchased from VWR International. Concentration used (9% of the 96h LC50) Sodium Hypochlorite 0.500 mg/L 5.5 mg/L (Pitanga, 2011), Bisphenol A 0.891 mg/L 9.9 mg/L (Hartmann, 2012) Ethanol 1278 mg/L 14200 mg/L (Martins et al., 2007) Toxicants 96h LC50 26 3.6. Experimental plan At the beginning of each trial, the fish were transferred individually to the arenas where the acclimation was of 10 minutes, due to the transfer and the possible stress given to the fish. Then there were 1h30m of exposure time without toxicant. The last hour of exposure was recorded (recording time) and designated by m moment (first recording time) was the time before the addiction of the solutions, and this experimental plan was used in order for each fish to be used as control of himself. In the analysis it was only use the first half an hour of the recording time. After the first recording time, each arena was contaminated with the respective toxicant solution, and the controls with water (to maintain the same depth in every arena). Then the second exposure time was also of 1h30m. The last hour of exposure time was recorded (recording time), designated by moment . This moment (second recording time) was the time after the addiction of the solutions (figure 4). In the analysis, only the first half an hour of the recording times was used. Using the behaviors of the same organisms, with and without toxicant, the possible variability between individuals was reduced, which probably improve the quality of the statistical analysis. In the end of the trials, the fish were kept individually in four Ciano Aqua 15 aquariums that were divided in three equal spaces. Each aquarium had three fish that were exposed to the same experimental condition, in other words, one aquarium with three fish that were the controls, the other one with three fish exposed to BPA, the other with three fish exposed to ethanol, and the last one with three fish exposed to SH. Four nets were used to transfer the fish, one for each experimental condition. The water used in the arenas was discarded after each trial, and the arenas were only filled up in the next day before a new trial. Figure 4: Experimental plan of each trial and toxicant. 33 analysis of the error found in the Training and Testing series. All the other parameters were maintained as initially, in other words as default. 3.8.2. Anova and post-Hoc Test First the ANOVA (analysis of variance) was performed to see if there were statistically significant differences between the groups of behavior categories defined by cluster analysis in relation to the averages of all movement descriptors. Then the post-Hoc test (Scheffe's), and the homogeneous subset, were performed to analyze the average values of the movement descriptors of each category and compare it. These tests were executed in the SPSS (Software version 21) program because the Statistica program homogeneous subset. 3.8.3. Correspondence Analysis To, obtain a quantifiable measure of the behavior effect caused by each toxicant, several correspondence analysis were performed using different partitions of the data in terms of day, assay and toxicant. Of all correspondence analysis tested, the partitions of data with all the toxicants, by day and assay had the best results in terms of significance level and conclusions consistency. In the correspondence analysis the 12 behavior categories defined by the cluster analyses were used as row variable. As column variable the conditions W, BPA, Et and SH were used. These conditions classify the experimental units. W had all experimental units in the Moment Before (without toxicant) and also the controls at the Moment After (after adding the toxicants) in others words this category had all experimental conditions without toxicant. The BPA, Et and SH conditions represent the experimental units with toxicant at the Moment After (after adding the respective toxicants, BPA, ethanol and SH). The case selection conditions were day=i (i= 1, 2, 3, 5 e 9) and assay=k (k=1 e 2). The supplementary column points were the frequencies profile of the different 34 behavior categories in each experimental unit organized by replica (a, b and c), toxicants (Control, BPA, Et and SH), Moment (Before and After), day (1, 2, 3, 5 and 9) and assay (1 and 2). was possible to see that all types of experimental conditions (W, BPA, Et and SH) were clearly discriminated spatially in the three dimensions defined by the different correspondence analysis, as represented in figure 8. In the multidimensional space defined by each correspondence analysis (by day and assay), three vectors (Toxicant vectors) that represent the effects caused by the toxicant on the fish behavior were determined. These vectors (vectors BPA, Et and SH) were determined by the difference between point W and the respective toxicant point. The points W, BPA, Et and SH represent the group of experimental units that belong to each of these types of experimental units. These points were the midpoints of each group of experimental conditions defined by the Column variable in each Correspondence Analysis. The Euclidean distance of each point (experimental unit) to the respective Toxicant vector (BPA, Et and SH) was calculated through the scalar projection of the vector formed by each point with the corresponding Toxicant vector. The objective of these Distances BPA, SH and Ethanol was to have a measure of the position of each point within the gradient formed by the respective Toxicant vector. These Distances measure the degree of modification of each experimental unit in the direction of the three Toxicant vectors. These Distances were standardized to allow comparisons between the different days and assays. In the calculation of the Distances to avoid subjective selections of the dimensions for Figure 8: Correspondence analysis, plot of row column coordinates (experimental conditions W, SH, Et, BPA): dimension 2x3, day 1, assay 1. 35 the representation of the effect caused by the toxicant and to simplify the comparison of the different measures, all the three dimensions created by the correspondence analysis were used and the standard procedures were always the same. 3.8.4. Saturated Orthogonal Multiple Linear Regression Analysis The degree of alteration (Distances BPA, Ethanol and SH) observed in each experimental unit over time (days) was analyzed by different models of saturated orthogonal multiple linear regression analysis (Box et al.,1978). These analysis were made by toxicant (BPA, Et, SH), and by assay. The assays were individually analyzed (assay 1 and assay 2) but also in ensemble. When the factors associated with the assay were not significant the two assays were tested in an ensemble analysis, without discriminating them because the differences were not significant. Thus, significant effects observed in one of the assays are highlighted if they are also significant in the ensemble analysis. However a significant effect in the ensemble analysis cant in the individual analyzed was more relevant than if it were only significant in the individual analyzed. 3.8.4.1. Presence/Absence Model This regression model was used to test if the changes in the behavior profiles of the fish measured by the Distances Ethanol, BPA and SH were significantly correlated statistically with the presence or the absence of the respective toxicant in the water. The independent variables were with and without toxicant in the water (ToxW), Day and Assay. The dependent variable was the Distance to the respective Toxicant vector and they were analyzed separately. The independent variables, were combined in a full factorial design, and were coded in orthogonal polynomial coefficients. The independent variable ToxW has the value +1 in the Moment After in the toxicant experimental units, and the value -1 in the remaining experimental units and times (Moment Before in the toxicant 36 experimental units and the two moments in the control experimental units). The independent variable Day, with 5 levels, one for each day analyzed, were decomposed in linear, quadratic and cubic terms, with a regressor for each one (table 2). The independent variable Assay takes the value -1 for assay 1 and +1 for assay 2. The subsequent 16 regressors were Intercept (b0), Assay, AssayD1, AssayD2, AssayD3, AssayToxW, AssayToxWD1, AssayToxWD2, AssayToxWD3, D1, D2, D3, ToxW, ToxWD1, ToxWD2, ToxWD3. 3.8.4.2. Moment/Control Model This regression model was used to analyze the progress of the behavior response, in the moment Before and After for the control experimental units. The independent variables were the Moment, Day and Assay, the dependent variable was the distance to the respective Toxicant vector (Et, SH or BPA), in the control experimental units. These distances were analyzed separately in this model. The independent variables, were combined in a full factorial design, and were coded in orthogonal polynomial coefficients. The variable Moment has the value -1 in the experimental units before adding the toxicant and the value +1 after adding the toxicant. The two other independent variables Day and Assay were encoded in the same way as in the previous model (Presence/Absence model). The subsequent 16 regressors were Intercept (b0), Assay, AssayD1, AssayD2, AssayD3, Assay Moment, Assay MomentD1, Assay MomentD2, Assay MomentWD3, D1, D2, D3, Moment, MomentD1, MomentD2, MomentD3. Table 2: Encoding of the independent variable Day. 37 3.8.4.3. Moment/Toxicant Model This regression model was used to analyze the progress of the behavior response, before and after adding each toxicant, in the experimental units with toxicant. The independent variables were the Moment, Day and Assay. The dependent variable was the distance to the respective Toxicant vector (Et, SH or BPA). Each of the toxicants was analyzed separately in this model. The distances to the Toxicant vector Et were used for the ethanol experimental units, the distances to the Toxicant vector BPA were used for the BPA experimental units, and the distances to the Toxicant vector SH were used for the SH experimental units. The independent variables, were combined in a full factorial design, and were coded in orthogonal polynomial coefficients. The three independent variables (Moment, Day and Assay) were encoded in the same way as in the previous model (Moment/Control). The resulting 16 regressors were equal to the regressors of the previous model (Moment/Control). 38 4. Results 4.1. Custer analysis The cluster analysis was used in order to define behavior categories using the behavior of the fish exposed to the test conditions. In table 3 the error in the validation series (final error) of the cluster analysis is presented. In this analysis, the error was only of 0.064550, which represent only 6%. The weight of each movement descriptor for each category is presented in table 4. However, to facilitate the analysis and the interpretation of these results it was necessary to use another type of statistical analysis. The same data used in the cluster analysis, was introduced in the SPSS (SPSS 21 Software) program to execute other types of tests, such as the Post Hoc test. Table 3: Error of each time series of the ANN. Table 4: Weight of each movement descriptor in the 12 behavior categories. 39 4.2. Anova and post-Hoc Test In table 5 are presented the results of the ANOVA conducted to compare the 9 movement descriptors (dependent variables) between the behavior categories. In the table is indicated that the significance values were lower than 5% (<0.05). These values indicate that the results between the categories were statistically significant in all variables analyzed, and therefore the null hypothesis was rejected. Significant differences between the average values of the behavior categories were detected. Table 5: Analysis of variance performed between behavior categories. Post Hoc tests are a posteriori analysis that in this case, has the purpose to find patterns (relationships) between groups or subgroups of data. Homogeneous subset is the same test with a different data organization that helps to interpret the results in a more intuitive way. The homogeneous subset is done by different 40 statistical tests such as or Tukey's test, and they help the analysis by pairing of means to see if there is a difference. In this case although the two tests were performed, the Scheffe's test was used because it was a more conservative multiple comparisons technique. The results obtained in the Scheffe's test are compiled in tables 6, 7 and 8. The average value of the x coordinate and the average value y were the movement descriptors that had maximum ability to discriminate the behavior categories, because all the values for this two movement descriptors were significantly different. The coordinates of the center of the arena were 88 mm and 50 mm (x,y), which means that values much inferior or superior to these indicates that the fish were close to the walls of the arena. The behavior category 1.2 for the movement descriptor average value of x coordinate presented the lowest value 22.615 mm which indicates that the fish were frequently in the left side of the arena, close to the glass of the arena and the category 3.3 presented the highest value 122.279 mm which indicates that the fish were frequently in the right side of the arena, close to the glass. The behavior category 2.2 for the movement descriptor average value of Y coordinate presented the lowest value 26.826 mm which indicates that the fish were frequently in the underside of the Table 6: Summary of the homogeneous group analysis for the movement descriptors, average value of coordinate x and y, and linear velocity. The numbering indicates the subsets of each behavior category. 41 arena close to the glass. The category 4.3 presented the highest value 71.543 mm which indicates that the fish were frequently on the top side of the arena, close to the glass. The analysis of the movement descriptor linear velocity revealed that the behavior category 4.3 had the highest value (64.081 mm/s), followed by the behavior category 1.3 (63.516 mm/s). The behavior categories 3.1 and 4.1 presented the two lowest linear velocities, 1.228 mm/s and 4.725 mm/s respectively. The linear velocity had 9 subsets in total which reveals a good discriminatory capacity of this movement descriptor. Table 7: Summary of the homogeneous group analysis for the movement descriptors, linear acceleration, angle and angular velocity. The numbering indicates the subsets of each behavior category. The behavior categories relatively to the linear acceleration in the case of the Scheffe's test were in the same group for all categories, which indicates that this movement descriptor was uninformative. For the movement descriptor angle the behavior categories 1.2 and 1.3 presented the highest values, 1.735 degrees and 1.491 degrees respectively which indicate that the fish in this category turned predominantly to the right. The behavior categories 4.2 and 4.3 for the movement descriptor angle presented the two the two lowest values, -4.010 degrees and -2.691 degrees respectively which indicate that the fish in this category turned predominantly to the left. This descriptor movement revealed a good discriminatory capacity because it had 8 subsets in total. For the angular 42 velocity the categories 4.1 and 4.2 presented the two highest values, 372.292 degrees/s and 363.324 degrees/s respectively which indicate that the fish turned at high speed. The categories 1.1 and 2.1 presented the two lowest values for the angular velocity, 105.010 degrees/s and 117.173 degrees/s respectively which indicate that the fish turned at low speed. For the angular acceleration the categories 4.2 and 4.3 presented the two highest values, 1873.655 degrees/s2 and 1784.101 degrees/s2 respectively. The categories 1.1 and 3.1 presented the two lowest values for the angular acceleration, 585.417 degrees/s2 and 283.705 degrees/s2 respectively. This movement descriptor revealed a great discriminatory capacity because it had 10 subsets in total. For the meander the behavior category 4.1 and had highest value, 118.952 degrees/mm followed by the category 3.1 with 45.668 degrees/mm. These values indicate that the fish has a pattern of movements composed by many changes of directions. The categories 1.1 and 2.3 presented the two lowest values for the meander, 1.460 degrees/mm and 2.510 degrees/mm respectively which indicate that the fish has a pattern of movements composed by few changes in direction with more straight trajectories. The meander had 8 subsets in total, which reveals a good discriminatory capacity of Table 8: Summary of the homogeneous group analysis for the movement descriptors, angular acceleration, meander and standard deviation of X/Y coordinates. The numbering indicates the subsets of each behavior category. 49 Figure 12: Representation of the dependent variable Distance SH in the control group, in function of the Assay, Moment and Day (Moment/Control model). Vertical bars denote 0.95 confidence intervals. The dependent variable Distance SH in the control groups (table 11) show any statistically significant trend in relation to the analyzed factors and the interaction between them. The significance levels were always above 0.05 in both assays, separately (data not represented) and together, for all regressors analyzed. This result indicate that there differences between the Moment Before, and the Moment After in the controls (p <0.88402) and also that the factors Assay and Day did not influenced the behavior as it can be seen in figure 12. The interactions between all the factors were also not statistically significant. The regression analysis for dependent variable BPA in the controls is sown in table 13. The distances BPA for the controls are represented in figure 13. 50 Table 13: Regression analysis (Moment/Control) of the dependent variable Distance BPA in the control group, considering the factor Assay. Statistically significant regressors Figure 13: Representation of the dependent variable Distance BPA in the control group, in function of the Assay, Moment, and Day (Moment/Control model). Vertical bars denote 0.95 confidence intervals. In the case of BPA, because the regressor Assay was statistically significant, the factor Assay had to be considerate. The control groups presented values of Distance BPA (table 13), on average, higher in assay 2 than in assay 1 (bAssay=0.3901; sig=0.0002) but this measure was not significantly influenced by any other factor analyzed, and the interaction between them. As in the previous b* Std.Err. (of b*) b Std.Err. (of b) t(52) p-value Intercept -0.3518 0.1200 -2.9309 0.0050 Assay 0.3932 0.1210 0.3901 0.1200 3.2505 0.0020 AssayMoment -0.0746 0.1210 -0.0740 0.1200 -0.6164 0.5403 AssayD1 0.1239 0.1210 0.0870 0.0849 1.0246 0.3103 AssayMomentD1 0.1533 0.1210 0.1075 0.0849 1.2670 0.2108 Moment -0.0564 0.1210 -0.0560 0.1200 -0.4663 0.6430 D1 0.1865 0.1210 0.1308 0.0849 1.5417 0.1292 AssayD1 0.0473 0.1210 0.0332 0.0849 0.3912 0.6972 N=60 F(7,52)=2,3354 p<0,03765 Std.Error of estimate: 0,92967 51 case, differences between the Moment Before and the Moment After in the controls were not detected, and also that the factor Day did not influenced the behavior as it can be seen in figure 13. The interactions between all the factors were also not statistically significant. The regression analysis for dependent variable Ethanol in the controls is demonstrated in table 14. The distances Ethanol for the controls are represented in figure 14. Table 14: Regression analysis (Moment/Control) of the dependent variable Distance Ethanol in the control group, considering the factor Assay. Statistically significant Figure 14: Representation of the dependent variable Distance Ethanol in the control group, in function of the Assay, Moment and Day (Moment/Control model). Vertical bars denote 0.95 confidence intervals. b* Std.Err. (of b*) b Std.Err. (of b) t(55) p-value Intercept -0.5658 0.0994 -5.6937 0.0000 AssayMomentD1 0.2626 0.1218 0.1516 0.0703 2.1569 0.0354 Moment -0.2728 0.1218 -0.2226 0.0994 -2.2400 0.0292 D1 0.0616 0.1218 0.0356 0.0703 0.5062 0.6148 MomentD1 0.1931 0.1218 0.1115 0.0703 1.5861 0.1184 N=60 F(4,55)=3,1104 p<0,02228 Std.Error of estimate: 0,76979 52 In the measure Distance Ethanol, the control fish did not present values with an evolution as neutral as in the previous toxicants. The Before and After effect (Moment) was statistically significant and negative (b Moment = -0.2226, sig = 0.0292), which means that the measure Distance Ethanol decreased between the moment Before and After in this group of fish (table 14). But this difference tends to decrease over the days (bAssayMomentD1=0.1516, sig=0.0354). This tendency is particularly visible in assay 2 being practically zero on the last day (figure 14). 4.6. Moment/Toxicant Model This regression model had the objective of understanding the progress of the behavior response, before and after adding each toxicant, in the toxicant experimental units. In this regression the linear, quadratic and cubic term of the variable was also analyzed, however as in previous analyzes, these factors were gradually eliminated until the regression was significant. As in previous models, when the factor Assay was not statistically significant, the two assays were treated as if they were one. In this model, the dependent variables (Distance BPA, SH and Ethanol), registered only in the toxicant groups, were analyzed based on moment Before and After (Moment), the assay (Assay) and also based in the repetition (Day). With all the toxicants, the respective measures (Distance) of the exposed fish were not influenced by the factor assay throughout their range (table 15, 16 and 17). No differences were detected between the two assays, and the factor In all toxicant groups the factor Moment influenced in a statistically significant way (and in this case positive way), the respective average Distance (Ethanol, b=0.3834, sig=0.0061; BPA, b=0.5179, sig=0.0004; SH, b=0.5510, sig=0.0001), which means that the Distances of the Moment After were higher than the Distances of the Moment Before (tables 15, 16 and 17). 53 Table 15: Regression analysis (Moment/Toxicant) of the dependent variable Distance Ethanol in the ethanol group, considering the two assays as one. Statistically significant Figure 15: Representation of the dependent variable Distance Ethanol in the ethanol group, in function of the Assay, Moment and Day (Moment/Toxicant model). Vertical bars denote 0.95 confidence intervals. For ethanol the differences between the Moment Before and After adding ethanol, were similar in all the days (b MomentD1=-0.0636, sig=0.5062), especially in assay 1 as is shown in figure 15. b* Std.Err. (of b*) b Std.Err. (of b) t(56) p-value Intercept 0.7770 0.1345 5.7750 0.0000 Moment 0.3536 0.1241 0.3834 0.1345 2.8496 0.0061 D1 0.0752 0.1241 0.0576 0.0951 0.6058 0.5471 MomentD1 -0.0830 0.1241 -0.0636 0.0951 -0.6690 0.5062 N=60 F(3,56)=2,9783 p<0,03909 Std.Error of estimate: 1,0422 54 Table 16: Regression analysis (Moment/Toxicant) of the dependent variable Distance BPA in the BPA group, considering the two assays as one. Statistically significant regressors Figure 16: Representation of the dependent variable Distance BPA in the BPA group, in function of the Assay, Moment and Day (Moment/Toxicant model). Vertical bars denote 0.95 confidence intervals. For BPA and SH the effect of the factor Moment tends to decrease with days (BPA, b MomentD1=-0.2101, sig=0.0340; SH, b MomentD1=-0.2368, sig=0.0112) (tables 16 and 17) reaching practically zero in the last days. This tendency is particularly visible in assay 2 for BPA (figure 16) and in assay 1 for SH (figure 17). b* Std.Err. (of b*) b Std.Err. (of b) t(52) p-value Intercept 0.5997 0.1364 4.3963 0.0001 Moment 0.4459 0.1175 0.5179 0.1364 3.7966 0.0004 D1 -0.0052 0.1175 -0.0043 0.0965 -0.0447 0.9645 D2 -0.0322 0.1175 -0.0223 0.0815 -0.2738 0.7853 MomentD1 -0.2558 0.1175 -0.2101 0.0965 -2.1782 0.0340 MomentD2 -0.0173 0.1175 -0.0120 0.0815 -0.1474 0.8834 D3 -0.1251 0.1175 -0.1028 0.0965 -1.0654 0.2916 MomentD3 0.0357 0.1175 0.0294 0.0965 0.3043 0.7621 N=60 F(7,52)=2,9264 p<0,01167 Std.Error of estimate: 1,0567 55 Table 17: Regression analysis (Moment/Toxicant) of the dependent variable Distance SH in the SH group, considering the two assays as one. Statistically significant regressors Figure 17: Representation of the dependent variable Distance SH in the SH group, in function of the Assay, Moment and Day (Moment/Toxicant model). Vertical bars denote 0.95 confidence intervals. b* Std.Err. (of b*) b Std.Err. (of b) t(52) p-value Intercept 0.7338 0.1273 5.7666 0.0000 Moment 0.4806 0.1110 0.5510 0.1273 4.3298 0.0001 D1 0.0364 0.1110 0.0295 0.0900 0.3280 0.7443 MomentD1 -0.2921 0.1110 -0.2368 0.0900 -2.6318 0.0112 D2 0.0239 0.1110 0.0164 0.0760 0.2151 0.8305 MomentD2 0.1160 0.1110 0.0795 0.0760 1.0450 0.3008 D3 -0.1337 0.1110 -0.1084 0.0900 -1.2047 0.2338 MomentD3 0.0982 0.1110 0.0796 0.0900 0.8847 0.3804 F(7,52)=4,1648 p<0,00105 Std.Error of estimate: 0,98571 N=60 56 5. Discussion 5.1. Characterization of the Behavior Categories Using the cluster analysis it was possible to define 12 behavior categories of the fish submitted to the different experimental conditions. The 6% error of the validation series of the cluster analysis (table 3) was minimal and represented the quality of the cluster analysis, the precision quality of the system. The ANOVA and the post-Hoc test allowed realizing that these behavior categories had statistically significant differences, and that all the categories were different from each other. Due to the massive amount of data and especially the number of days used in the analysis, it was difficult to establish associations between the behavior categories with each toxicant throughout the days, but the post-hoc test permitted the characterization of the behavior categories based on the movement descriptors. The categories 3.1 and 4.1 presented the two lowest linear velocities (1.228 mm/s and 4.580 mm/s respectively), the two lowest standard deviation x/y (0.627 mm and 2.238 mm respectively), the two highest meander ((45.668 rad/mm and 118.95 respectively) and in the case of the category 4.1 the highest angular velocity (372.292 rad/s). Curiously because the Angle values were positive the fish in these two categories also turned predominantly to the right. These results indicated that the fish in those categories presented slow movement and probably more stops, with low dispersion in space and many changes of direction that in the case of the category 4.1 were done at high angular velocity. When in states of heighted anxiety the zebrafish specie tends to display erratic movement (many changes of direction in a relatively small space), and more stops (freezing, when stationary) (Egan et al., 2009). Base on the movement descriptors the categories 3.1 and 4.1, fit in this type of behaviors, and several substances as acute caffeine and alarm pheromone had be proven to triggered these behaviors in zebrafish (Egan et al., 2009). In a previous work (for publication) acutely exposed zebrafish to SH (1h30m), also presented this type of behavior (freezing, increased changes of direction, decreased velocity). The categories 1.3, 2.2 and 2.3 had high linear velocities (63.227 mm/s, 59.906 mm/s and 59.697 mm/s respectively), low meander values (3.187 57 rad/mm, 5.742 rad/mm and 2.510 rad/mm respectively) and high standard deviations of x/y (25.616 mm, 17.156 mm and 25.191 mm respectively) and these parameters indicate fast movements and less stops with few changes of direction and high dispersion in space. These types of behaviors are usually related with normal exploratory behavior and activity. The categories described in this section were the most distinguishable and for that reason were highlighted, as stated above it was impossible to establish associations between the behavior categories with each toxicant throughout the days, but this was not one of the objectives of the work. 5.2. Analysis of the Regressions The results obtained by the Presence/Absence model (table 9, 10 and 11) were within the expectations, all the differences between the Distances BPA, Ethanol and SH were statistically significantly and detected the presence of the respective toxicant in water (ethanol, b ToxW=0.7032; BPA, b ToxW=0.6624; SH, b ToxW=0.7551; with sig<0.0000 in all analyzes). In relation to ethanol the differences were similar over the days (table 9). This conclusion was confirmed by the model Moment/Toxicant (table 15), where only the experimental units in which ethanol was added were analyzed, and the Distances between the Moments Before and After were similar over the days. In this work the development of behavior tolerance to ethanol was not observed, which was beneficial for the detection capability of the system that remained unchanged over the days. However, Gerlai et al. (2006) found that zebrafish exposed to ethanol for two weeks to 14% of the 96h LC50, had developed behavior tolerance, and only an acute exposure to 56% of the 96h LC50 after the chronic exposure attenuated the development of tolerance. Several other works also have reported the development of tolerance in zebrafish chronically exposed, but the concentrations that caused this response were always superior to the 9% 96h LC50 used in the present work because they were normally between 14% of the 96h LC50 and 28% (Dlugos and Rabin, 2003; Gerlai et al., 2009; Dlugos et al., 2011; Tran and Gerlai, 2013; Tran et al., 2014). Contrary to the reported in 58 several works and the initial expectations, the zebrafish did not develop tolerance to ethanol in this work, (because the fish reacted always to the presence of ethanol), and the same had happened with wild-type zebrafish in a study performed by Dlugos and Rabin (2003) that have exposed zebrafish for two weeks to 28% of the 96h LC50 of ethanol and did not detect the development of tolerance. In this previous work, another strain of zebrafish (long-fin striped zebrafish) exposed chronically for two weeks, in contrary to the other strain, developed tolerance to ethanol (the average distance between each fish and its nearest neighbor were similar to the pre-exposure values like in the previous study that in this work the wild-type zebrafish strain used was less resistant to the treatment (although they analyze the shoal behavior instead of the locomotor), and like advanced by Dlugos and Rabin (2003) possibly due to differences in both the response of the central nervous system, as well as the ability of the CNS to adapt to ethanol. Maybe in this work the exposure time employed, more a chronic daily exposure (once a day). Probably in the present work the concentration or the neurotransmitters pattern adaptation (Gerlai et al., 2009; Tran et al., 2014), which highlights the sensibility of the system to quickly detect exposure situations. exposure that was sufficient to cause behaviors effects (hyperactivity), typical observed in acute exposure to low concentrations such as 14% of the 96h LC50 of ethanol (anxiolytic, decreased erratic movements and increased exploration) (Gerlai et al., 2000), in each day, which explains the similar behavior response happens with chronic exposures. In the case of SH and BPA and contrary to what happened with ethanol, through the distances analyzed in the Presence/Absence model (tables 9, 10 and 11) the system was always able to detect the presence of the respective toxicant, although these differences diminished over the days which lead to a decrease of the detection capability over the days (BPA, bToxWD1=-0.1851; SH, bToxWD1=- 0.1449; with sig<0.0486 in all analyzes) (tables 9 and 10). This decreased, is caused by the decrease of the differences of the distances between the Moments Before and After demonstrated by the model Moment/Toxicant (BPA, bMomentD1=- 0.2101, sig=0.0340; SH, bMomentD1=-0.2368, sig=0.0112) (tables 15 and 16). The 65 Canistro, D., Melega, S., Ranieri, D., Sapone, A., Paolini, M., Gustavino, B., Monfrinotti, M., Rizzoni, M., 2012. Modulation of cytochrome P450 and induction of DNA damage in Cyprinus carpio exposed in situ to surface water treated with chlorine or alternative disinfectants in different seasons. Mutation Research - Fundamental and Molecular Mechanisms of Mutagenesis 729 (1/2), 81-89. Cardoso, R.A., Brozowski, S.J., Chavez-Noriega, L.E., Harpold, M., Valenzuela, C.F., Harris, R.A., 1999. Effects of ethanol on recombinant human neuronal nicotinic acetylcholine receptors expressed in Xenopus oocytes. Journal of Pharmacology and Experimental Therapeutics 289 (2), 774-780. ecotoxicology. Marine Pollution Bulletin 44 (1), 7-15. assessment of the environmental fate of bisphenol A. Human and Ecological Risk Assessment 8 (5), 1107-1135. Davies, D.L., Asatryan, L., Kuo, S.T., Woodward, J.J., King, B.F., Alkana, R.L., Xiao, C., Ye, J.H., Sun, H., Zhang, L., Hu, X.-Q., Hayrapetyan, V., Lovinger, D. M., Machu, T. K., 2006. Effects of ethanol on adenosine 5'-triphosphate-gated purinergic and 5-hydroxytryptamine receptors. Alcoholism: Clinical And Experimental Research 30 (2), 349-358. de Esch, C., Slieker, R., Wolterbeek, A., Woutersen, R., De Groot, D., van der Linde, H., Willemsen, R., 2012. Locomotor activity assay in zebrafish larvae: Influence of age, strain and ethanol. Neurotoxicology and Teratology 34 (4), 425433. Delcourt, J., Becco, C., Vandewalle, N., Poncin, P., 2009. A video multitracking system for quantification of individual behavior in a large fish shoal: Advantages and limits. Behavior Research Methods 41 (1), 228-235. Delcourt, J., Denoël, M., Ylieff, M., Poncin, P., 2013. Video multitracking of fish behaviour: a synthesis and future perspectives. Fish and Fisheries 14 (2), 186204. 66 Diehl, P., Gerke, T., Jeuken, A., Lowis, J., Steen, R., van Steenwijk, J., Stoks, P., Willemsen, H.-G., 2006. Early Warning Strategies and Practices Along the River Rhine. In: The Rhine (Knepper, T.P. edition), pp. 99-124. Berlin: Springer. Dildy-Mayfield, J.E., Harris, R.A., 1995. Ethanol inhibits kainate responses of glutamate receptors expressed in Xenopus oocytes: Role of calcium and protein kinase C. Journal of Neuroscience 15 (4), 3162-3171. Dlugos, C.A., Brown, S.J., Rabin, R.A., 2011. Gender differences in ethanolinduced behavioral sensitivity in zebrafish. Alcohol 45 (1), 11-18. Dlugos, C.A., Rabin, R.A., 2003. Ethanol effects on three strains of zebrafish: Model system for genetic investigations. Pharmacology Biochemistry and Behavior 74 (2), 471-480. Dorn, P.B., Chou, C.-S., Gentempo, J.J., 1987. Degradation of bisphenol A in natural waters. Chemosphere 16 (7), 1501-1507. Egan, R.J., Bergner, C.L., Hart, P.C., Glasgow, E., Amri, H., Zukowska, Z., Kalueff, A.V., Cachat, J.M., Canavello, P.R., Elegante, M.F., Elkhayat, S. I., Bartels, B. K., Tien, A. K., Tien, D. H., Mohnot, S., Beeson, E., 2009. Understanding behavioral and physiological phenotypes of stress and anxiety in zebrafish. Behavioural Brain Research 205 (1), 38-44. Elia, A.C., Anastasi, V., Dörr, A.J.M., 2006. Hepatic antioxidant enzymes and total glutathione of Cyprinus carpio exposed to three disinfectants, chlorine dioxide, sodium hypochlorite and peracetic acid, for superficial water potabilization. Chemosphere 64 (10), 1633-1641. Emmanuel, E., Perrodin, Y., Blanchard, J.M., Vermande, P., Keck, G., 2004. Toxicological effects of disinfections using sodium hypochlorite on aquatic organisms and its contribution to AOX formation in hospital wastewater. Environment International 30 (7), 891-900. 67 Environmental Protection Agency of United States (USEPA). 2012. Attachment H Total Residual Chlorine Discharge Limit Calculation NPDES Permit No. MA0103284. Boston, Massachusetts. European Commission. 2011. Commission Directive 2011/8/EU of 28 January 2011 amending Directive 2002/72/EC as regards the restriction of use of Bisphenol A in plastic infant feeding bottles. Official Journal of the European Union. Faingold, C.L., N'Gouemo, P., Riaz, A., 1998. Ethanol and neurotransmitter interactions--from molecular to integrative effects. Progress in Neurobiology 55 (5), 509-535. Farabollini, F., Porrini, S., Della Seta, D., Bianchi, F., Dessì-Fulgheri, F., 2002. Effects of Perinatal Exposure to Bisphenol A on Sociosexual Behavior of Female and Male Rats. Environmental Health Perspectives 110 (3): 409 414. Flint, S., Markle, T., Thompson, S., Wallace, E., 2012. Bisphenol A exposure, effects, and policy: A wildlife perspective. Journal of Environmental Management 104, 19-34. Freitas, J.G., Barker, J.F., 2013. Denatured ethanol release into gasoline residuals, Part 1: Source behaviour. Journal of Contaminant Hydrology 148, 67-78. Fromme, H., Küchler, T., Otto, T., Pilz, K., Müller, J., Wenzel, A., 2002. Occurrence of phthalates and bisphenol A and F in the environment. Water Research 36 (6), 1429-1438. Gerhardt, A., 1995. Monitoring behavioural responses to metals in Gammarus pulex (L.) (Crustacea) with impedance conversion. Environmental Science and Pollution Research 3 (2), 15-23. Gerhardt, A., 2007. Aquatic behavioral ecotoxicology - Prospects and limitations. Human and Ecological Risk Assessment 13 (3), 481-491. 68 Gerlai, R., Chatterjee, D., Pereira, T., Sawashima, T., Krishnannair, R., 2009. Acute and chronic alcohol dose: population differences in behavior and neurochemistry of zebrafish. Genes, Brain and Behavior 8 (6), 586-599. Gerlai, R., Lahav, M., Guo, S., Rosenthal, A., 2000. Drinks like a fish: zebra fish (Danio rerio) as a behavior genetic model to study alcohol effects. Pharmacology Biochemistry and Behavior 67 (4), 773-782. Gerlai, R., Lee, V., Blaser, R., 2006. Effects of acute and chronic ethanol exposure on the behavior of adult zebrafish (Danio rerio). Pharmacology Biochemistry and Behavior 85 (4), 752-761. Gonzalez, C., Greenwood, R., Quevauviller, PP., 2009. Rapid Chemical and Biological Techniques for Water Monitoring. Chichester: John Wiley & Sons, Inc. Goulding, A.T., Shelley, L.K., Kennedy, C.J., Ross, P.S., 2013. Reduction in swimming performance in juvenile rainbow trout (Oncorhynchus mykiss) following sublethal exposure to pyrethroid insecticides. Comparative Biochemistry and Physiology, Part C 157 (3), 280-286. Hartmann, C. Bisphenol A: Environmental Fate and Behaviour, Human Health Effects Implications for EU-chemical Policy. Vien: University of Vien. 2012. Holmstrup, M., Sørensen, J.G., Overgaard, J., Bayley, M., Bindesbøl, A.-M., Slotsbo, S., Fisker, K.V., Maraldo, K., Waagner, D., Labouriau, R., Asmund, G., 2011. Body metal concentrations and glycogen reserves in earthworms (Dendrobaena octaedra) from contaminated and uncontaminated forest soil. Environmental Pollution 159 (1), 190-197. Houtman, C.J., 2010. Emerging contaminants in surface waters and their relevance for the production of drinking water in Europe. Journal of Integrative Environmental Sciences 7 (4), 271-295. 69 Jenner, H.A., Taylor, C.J.L., van Donk, M., Khalanski, M., 1997. Chlorination byproducts in chlorinated cooling water of some European coastal power stations. Marine Environmental Research 43 (4), 279-293. Klerks, P.L., 1999. Acclimation to contaminants by the grass shrimp Palaemonetes pugio: Individual contaminants vs. mixtures. Ecotoxicology 8 (4), 277-286. Kuiper, G.G., Lemmen, J.G., Carlsson, B., Corton, J.C., Safe, S.H., van der Saag, P.T., van der Burg, B., Gustafsson, J.A., 1998. Interaction of estrogenic chemicals and phytoestrogens with estrogen receptor beta. Endocrinology 139 (10), 42524263. Kwak, I.S., Chon, T.S., Chung, N.I., Kang, H.M., Kim, Y.S., Kim, J.S., Koh, S.C., Lee, S.K., 2002. Pattern recognition of the movement tracks of medaka (Oryzias latipes) in response to sub-lethal treatments of an insecticide by using artificial neural networks. Environmental Pollution 120 (3), 671-681. Lahnsteiner, F., Berger, B., Kletzl, M., Weismann, T., 2005. Effect of bisphenol A on maturation and quality of semen and eggs in the brown trout, Salmo trutta f. fario. Aquatic Toxicology 75 (3), 213-224. Lee, H.J., Chattopadhyay, S., Gong, E.-Y., Ahn, R.S., Lee, K., 2003. Antiandrogenic Effects of Bisphenol A and Nonylphenol on the Function of Androgen Receptor. Toxicological Sciences 75 (1), 40-46. Lindholst, C., Bjerregaard, P., Wynne, P.M., Marriott, P., Pedersen, S.N., 2003. Metabolism of bisphenol A in zebrafish (Danio rerio) and rainbow trout (Oncorhynchus mykiss) in relation to estrogenic response. Comparative Biochemistry and Physiology, Part C 135 (2), 169-177. Little, E., Archeski, R., Flerov, B., and Kozlovskaya, V., 1990. Behavioral indicators of sublethal toxicity in rainbow trout. Archives of Environmental Contamination and Toxicology 19 (3), 380-385. Little, E., Finger, S.E., 1990. Swimming behavior as an indicator of sublethal toxicity in fish. Environmental Toxicology and Chemistry 9 (1), 13-19. 70 Liu, J., Wang, R., Huang, B., Lin, C., Wang, Y., Pan, X., 2011a. Distribution and bioaccumulation of steroidal and phenolic endocrine disrupting chemicals in wild fish species from Dianchi Lake, China. Environmental Pollution 159 (10), 28152822. Liu, Y., Chon, T.S., Lee, S.H., 2011b. Analysis of behavioral changes of zebrafish (Danio rerio) in response to formaldehyde using Self-organizing map and a hidden Markov model. Ecological Modelling 222 (14), 2191-2201. López-Galindo, C., Garrido, M.C., Casanueva, J.F., Nebot, E. 2010a. Degradation models and ecotoxicity in marine waters of two antifouling compounds: Sodium hypochlorite and an alkylamine surfactant. Science of the Total Environment 408 (8), 1779-85. López-Galindo, C., Nebot, E., Rubio, D., Vargas-Chacoff, L., Mancera, J.M., Casanueva, J.F., Solé, M., 2010b. Biomarker responses in Solea senegalensis exposed to sodium hypochlorite used as antifouling. Chemosphere 78 (7), 885893. López-Galindo, C., Nebot, E., Rubio, D., Vargas-Chacoff, L., Mancera, J.M., Casanueva, J.F., Solé, M., 2010c. Sublethal responses of the common mussel (Mytilus galloprovincialis) exposed to sodium hypochlorite and Mexel®432 used as antifoulants. Ecotoxicology and Environmental Safety 73 (5), 825-834. Magalhaes Dde, P., Armando da Cunha, R., Albuquerque dos Santos, J.A., Buss, D.F., Baptista, D.F., 2007. Behavioral response of zebrafish Danio rerio Hamilton 1822 to sublethal stress by sodium hypochlorite: ecotoxicological assay using an image analysis biomonitoring system. Ecotoxicology 16 (5), 417-422. Mandich, A., Bottero, S., Cevasco, A., Massari, A., Pedemonte, F., Benfenati, E., Maggioni, S., Erratico, C., Viganò, L., 2007. In vivo exposure of carp to graded concentrations of bisphenol A. General and Comparative Endocrinology 153 (1/3), 15-24. 71 Manduzio, H., Monsinjon, T., Galap, C., Leboulenger, F., Rocher, B., 2004. Seasonal variations in antioxidant defences in blue mussels Mytilus edulis collected from a polluted area: Major contributions in gills of an inducible isoform of Cu/Zn-superoxide dismutase and of glutathione S-transferase. Aquatic Toxicology 70 (1), 83-93. Martins, J., Teles, L.O., Vasconcelos, V., 2007. Assays with Daphnia magna and Danio rerio as alert systems in aquatic toxicology. Environment International 33 (3), 414-425. Matsuda, S., Matsuzawa, D., Ishii, D., Tomizawa, H., Sutoh, C., Nakazawa, K., Sajiki, J., Shimizu, E., Amano, K., 2012. Effects of perinatal exposure to low dose of bisphenol A on anxiety like behavior and dopamine metabolites in brain. Progress in Neuro-Psychopharmacology and Biological Psychiatry 39 (2), 273279. Maximino, C., da Silva, A.W.B., Herculano, A.M., Gouveia, A., 2011. Pharmacological analysis of zebrafish (Danio rerio) scototaxis. Progress in NeuroPsychopharmacology and Biological Psychiatry 35 (2), 624-631. Melvin, S.D., Wilson, S.P., 2013. The utility of behavioral studies for aquatic toxicology testing: A meta-analysis. Chemosphere 93 (10), 2217-2223. Meyer, J.N., Di Giulio, R.T., 2003. Heritable adaptation and fitness costs in killfish (Fundulus heteroclitus) inhabiting a polutted estuary. Ecological Appplications 13 (2), 490-503. Mihic, S.J., Ye, Q., Wick, M.J., Koltchine, V.V., Krasowski, M.D., Finn, S.E., Mascia, M.P., Valenzuela, C.F., Hanson, K.K., Greenblatt, E.P., Harris, R. A.. Harrison, N. L., 1997. Sites of alcohol and volatile anaesthetic action on GABA(A) and glycine receptors. Nature 389 (6649), 385-389. Moriyama, K., Tagami, T., Akamizu, T., Usui, T., Saijo, M., Kanamoto, N., Hataya, Y., Shimatsu, A., Kuzuya, H., Nakao, K., 2002. Thyroid hormone action is disrupted by bisphenol A as an antagonist. Journal of Clinical Endocrinology and Metabolism 87 (11), 5185-5190. 72 Nacci, D., Coiro, L., Champlin, D., Jayaraman, S., McKinney, R., Gleason, T.R., Munns Jr, W.R., Specker, J.L., Cooper, K.R., 1999. Adaptations of wild populations of the estuarine fish Fundulus heteroclitus to persistent environmental contaminants. Marine Biology 134 (1), 9-17. Nimkerdphol, K., Nakagawa, M., 2008. Effect of Sodium Hypochlorite on Zebrafish Swimming Behavior Estimated by Fractal Dimension Analysis. Journal of Bioscience and Bioengineering 105 (5), 486-492. Ownby, D.R., Newman, M.C., Mulvey, M., Vogelbein, W.K., Unger, M.A., Arzayus, L.F., 2002. Fish (Fundulus heteroclitus) populations with different exposure histories differ in tolerance of creosote-contaminated sediments. Environmental Toxicology and Chemistry 21 (9), 1897-1902. Pan, Y., Kaiguo, M., Gerlai, R., Razak, Z., Westwood, J.T., 2011. Chronic alcohol exposure induced gene expression changes in the zebrafish brain. Behavioural Brain Research 216 (1) , 66-76. Park, Y.S., Chung, N.I., Choi, K.H., Chon, T.S., Cha, E.Y., Lee, S.K. (2005). Computational characterization of behavioral response of medaka (Oryzias latipes) treated with diazinon. Aquatic Toxicology 71 (3), 215-228. Perez-Coll, C.S., Herkovits, J., Daniel, P., D'Eramo, J.L., Fridman, O., 1999. Metallothionein induction and cadmium uptake in Bufo arenarum embryos following an acclimation protocol. Environmental Pollution 106 (3), 443. Pitanga, F.L. The effect of sodium hypochlorite in different aquatic organisms. Reyhanian, N., Volkova, K., Hallgren, S., Bollner, T., Olsson, P.E., Olsen, H., Hallstrom, I.P., 2011. 17alpha-Ethinyl estradiol affects anxiety and shoaling behavior in adult male zebra fish (Danio rerio). Aquatic Toxicology 105 (1/2), 4148. 73 Rico, E.P., Rosemberg, D.B., Dias, R.D., Bogo, M.R., Bonan, C.D., 2007. Ethanol alters acetylcholinesterase activity and gene expression in zebrafish brain. Toxicology Letters 174 (1/3), 25-30. Saili, K.S., Corvi, M.M., Das, S.R., Przybyla, J., Anderson, K.A., Tanguay, R.L., Weber, D.N., Patel, A.U., 2012. Neurodevelopmental low-dose bisphenol A exposure leads to early life-stage hyperactivity and learning deficits in adult zebrafish. Toxicology 291 (1/3), 83-92. Scott, G.R., Sloman, K.A., 2004. The effects of environmental pollutants on complex fish behaviour: integrating behavioural and physiological indicators of toxicity. Aquatic Toxicology 68 (4), 369-392. Scott, G.R., Sloman, K.A., Wood, C.M., Rouleau, C., 2003. Cadmium disrupts behavioural and physiological responses to alarm substance in juvenile rainbow trout (Oncorhynchus mykiss). Journal of Experimental Biology 206 (11), 17791790. Shareef, A., Angove, M.J., Wells, J.D., Johnson, B.B., 2006. Aqueous solubilities of - - ethynylestradiol, and bisphenol A. Journal of Chemical and Engineering Data 51 (3), 879-881. Staples, C.A., Tilghman Hall, A., Friederich, U., Klecka, G.M., Caspers, N., 2011. Early life-stage and multigeneration toxicity study with bisphenol A and fathead minnows (Pimephales promelas). Ecotoxicology and Environmental Safety 74 (6), 1548-1557. Storey, M.V., van der Gaag, B., Burns, B.P., 2011. Advances in on-line drinking water quality monitoring and early warning systems. Water Research 45 (2), 741747. Strohm, B.H., Sweet, L.I., 2005. Ethanol. In: Encyclopedia of Toxicology (Wexler, P., Anderson, B., Peyster, A.d., Gad, S., Hakkinen P.J., Kamrin, M., Locey, B., Mehendale, H., Pope, C., Shugart, L. edition), pp. 263-266. New York: Elsevier Inc. 74 Swift, R., 2003. Direct measurement of alcohol and its metabolites. Addiction 98, 73-80. Teles, L.O., Vasconcelos, V., Pereira, E., Saker, M., 2006. Time series forecasting of cyanobacteria blooms in the Crestuma Reservoir (Douro River, Portugal) using artificial neural networks. Environmental Management 38 (2), 227-237. Tierney, K.B., 2011. Behavioural assessments of neurotoxic effects and neurodegeneration in zebrafish. Biochimica et Biophysica Acta 1812 (3), 381389. Tran, S., Chatterjee, D., Gerlai, R., 2014. An integrative analysis of ethanol tolerance and withdrawal in zebrafish (Danio rerio). Behavioural Brain Research. In press. Tran, S., Gerlai, R., 2013. Time-course of behavioural changes induced by ethanol in zebrafish (Danio rerio). Behavioural Brain Research 252, 204-213. Viarengo, A., Lowe, D., Bolognesi, C., Fabbri, E., Koehler, A., 2007. The use of biomarkers in biomonitoring: A 2-tier approach assessing the level of pollutantinduced stress syndrome in sentinel organisms. Comparative Biochemistry and Physiology, Part C 146 (3), 281-300. Vogl, C., Spieser, O.H., Grillitsch, B., Scholz, W., Wytek, R., 1999. Qualification of spontaneous undirected locomotor behavior of fish for sublethal toxicity testing. Part I. Variability of measurement parameters under general test conditions. Environmental Toxicology and Chemistry 18 (12), 2736. Wang, X., Dong, Q., Chen, Y., Jiang, H., Xiao, Q., Wang, Y., Li, W., Bai, C., Huang, C., Yang, D., 2013. Bisphenol A affects axonal growth, musculature and motor behavior in developing zebrafish. Aquatic Toxicology 142 143, 104-113. Wirgin, I., Waldman, J.R., 2004. Resistance to contaminants in North American fish populations. Mutation Research - Fundamental and Molecular Mechanisms of Mutagenesis 552 (1/2), 73-100.