Predicting chemical hazard across taxa through machine learning.

We applied machine learning methods to predict chemical hazards focusing on fish acute toxicity across taxa. We analyzed the relevance of taxonomy and experimental setup, showing that taking them into account can lead to considerable improvements in the classification performance. We quantified the gain obtained throught the introduction of taxonomic and experimental information, compared to classification based on chemical information alone. We used our approach with standard machine learning models (K-nearest neighbors, random forests and deep neural networks), as well as the recently proposed Read-Across Structure Activity Relationship (RASAR) models, which were very successful in predicting chemical hazards to mammals based on chemical similarity. We were able to obtain accuracies of over 93% on datasets where, due to noise in the data, the maximum achievable accuracy was expected to be below 96%. The best performances were obtained by random forests and RASAR models. We analyzed metrics to compare our results with animal test reproducibility, and despite most of our models "outperform animal test reproducibility" as measured through recently proposed metrics, we showed that the comparison between machine learning performance and animal test reproducibility should be addressed with particular care. While we focused on fish mortality, our approach, provided that the right data is available, is valid for any combination of chemicals, effects and taxa.

Zebrafish early life stages as alternative model to study 'designer drugs': Concordance with mammals in response to opioids.

The number of new psychoactive substances (NPS) on the illicit drug market increases fast, posing a need to urgently understand their toxicity and behavioural effects. However, with currently available rodent models, NPS assessment is limited to a few substances per year. Therefore, zebrafish (Danio rerio) embryos and larvae have been suggested as an alternative model that would require less time and resources to perform an initial assessment and could help to prioritize substances for subsequent evaluation in rodents. To validate this model, more information on the concordance of zebrafish larvae and mammalian responses to specific classes of NPS is needed. Here, we studied toxicity and behavioural effects of opioids in zebrafish early life stages. Synthetic opioids are a class of NPS that are often used in pain medication but also frequently abused, having caused multiple intoxications and fatalities recently. Our data shows that fentanyl derivatives were the most toxic among the tested opioids, with toxicity in the zebrafish embryo toxicity test decreasing in the following order: butyrfentanyl>3-methylfentanyl>fentanyl>tramadol> O-desmethyltramadol>morphine. Similar to rodents, tramadol as well as fentanyl and its derivatives led to hypoactive behaviour in zebrafish larvae, with 3-methylfentanyl being the most potent. Physico-chemical properties-based predictions of chemicals' uptake into zebrafish embryos and larvae correlated well with the effects observed. Further, the biotransformation pattern of butyrfentanyl in zebrafish larvae was reminiscent of that in humans. Comparison of toxicity and behavioural responses to opioids in zebrafish and rodents supports zebrafish as a suitable alternative model for rapidly testing synthetic opioids.

Predicting exposure concentrations of chemicals with a wide range of volatility and hydrophobicity in different multi-well plate set-ups.

Quantification of chemical toxicity in small-scale bioassays is challenging owing to small volumes used and extensive analytical resource needs. Yet, relying on nominal concentrations for effect determination maybe erroneous because loss processes can significantly reduce the actual exposure. Mechanistic models for predicting exposure concentrations based on distribution coefficients exist but require further validation with experimental data. Here we developed a complementary empirical model framework to predict chemical medium concentrations using different well-plate formats (24/48-well), plate covers (plastic lid, or additionally aluminum foil or adhesive foil), exposure volumes, and biological entities (fish, algal cells), focusing on the chemicals' volatility and hydrophobicity as determinants. The type of plate cover and medium volume were identified as important drivers of volatile chemical loss, which could accurately be predicted by the framework. The model focusing on adhesive foil as cover was exemplary cross-validated and extrapolated to other set-ups, specifically 6-well plates with fish cells and 24-well plates with zebrafish embryos. Two case study model applications further demonstrated the utility of the empirical model framework for toxicity predictions. Thus, our approach can significantly improve the applicability of small-scale systems by providing accurate chemical concentrations in exposure media without resource- and time-intensive analytical measurements.

Gut microbial transformation of the dietary mutagen MeIQx may reduce exposure levels without altering intestinal transport.

The mutagen and probable human carcinogen 2-amino-3,8-dimethylimidazo[4,5-f]quinoxaline (MeIQx) is metabolized in the colon to 9-hydroxyl-2,7-dimethyl-7,9,10,11-tetrahydropyrimido[2',1':2,3]imidazo[4,5-f]quinoxaline (MeIQx-M1) by conjugation with microbially generated acrolein. However, whether this microbiota-controlled process alters systemic exposure and hepatotoxicity of MeIQx remains unclear. The physiological relevance of this microbial transformation on the systemic exposure of MeIQx was investigated using an in vitro-in vivo extrapolation approach. To address whether microbial transformation influences intestinal transport of MeIQx, the intestinal uptake of MeIQx and its metabolite MeIQx-M1 was quantified using Ussing chambers mounted with different intestinal segments from male Fischer 344 rats. Up to 0.4% of both MeIQx and MeIQx-M1 were transported from the mucosal side to the serosal side of intestinal tissue within 90 min, suggesting that the intestinal uptake of both compounds is similar. With the uptake rates of both compounds, physiologically based pharmacokinetic (PBPK) modeling of the fate of MeIQx in the human body including microbial transformation of MeIQx was performed. Results indicate for the first time that high levels of microbe-derived acrolein would be required to significantly reduce systemic exposure of MeIQx in humans. Finally, neither MeIQx nor MeIQx-M1 were cytotoxic towards human liver HepaRG cells at dietary or higher concentrations of MeIQx. In summary, these findings suggest that gut microbial transformation of heterocyclic amines has the potential to influence systemic human exposure to some extent, but may require significant gut microbial production of acrolein and that further investigations are needed to understand physiological levels of acrolein and competing biotransformation pathways.

Importance of Toxicokinetics to Assess the Utility of Zebrafish Larvae as Model for Psychoactive Drug Screening Using Meta-Chlorophenylpiperazine (mCPP) as Example.

The number of new psychoactive substances (NPS) increases rapidly, harming society and fuelling the need for alternative testing strategies. These should allow the ever-increasing number of drugs to be tested more effectively for their toxicity and psychoactive effects. One proposed strategy is to complement rodent models with zebrafish (Danio rerio) larvae. Yet, our understanding of the toxicokinetics in this model, owing to the waterborne drug exposure and the distinct physiology of the fish, is incomplete. We here explore the toxicokinetics and behavioral effects of an NPS, meta-chlorophenylpiperazine (mCPP), in zebrafish larvae. Uptake kinetics of mCPP, supported by toxicokinetic modeling, strongly suggested the existence of active transport processes. Internal distribution showed a dominant accumulation in the eye, implying that in zebrafish, like in mammals, melanin could serve as a binding site for basic drugs. We confirmed this by demonstrating significantly lower drug accumulation in two types of hypo-pigmented fish. Comparison of the elimination kinetics between mCPP and previously characterized cocaine demonstrated that drug affinities to melanin in zebrafish vary depending on the structure of the test compound. As expected from mCPP-elicited responses in rodents and humans, zebrafish larvae displayed hypoactive behavior. However, significant differences were seen between zebrafish and rodents with regard to the concentration-dependency of the behavioral response and the comparability of tissue levels, corroborating the need to consider the organism-internal distribution of the chemical to allow appropriate dose modeling while evaluating effects and concordance between zebrafish and mammals. Our results highlight commonalities and differences of mammalian versus the fish model in need of further exploration.

Biotransformation of Benzo[ a]pyrene by Three Rainbow Trout ( Onchorhynchus mykiss) Cell Lines and Extrapolation To Derive a Fish Bioconcentration Factor.

Permanent fish cell lines constitute a promising complement or substitute for fish in the environmental risk assessment of chemicals. We demonstrate the potential of a set of cell lines originating from rainbow trout ( Oncorhynchus mykiss) to aid in the prediction of chemical bioaccumulation in fish, using benzo[ a]pyrene (BaP) as a model chemical. We selected three cell lines from different tissues to more fully account for whole-body biotransformation in vivo: the RTL-W1 cell line, representing the liver as major site of biotransformation, and the RTgill-W1 (gill) and RTgutGC (intestine) cell lines, as important environment-organism interfaces, which likely influence chemical uptake. All three cell lines were found to effectively biotransform BaP. However, rates of in vitro clearance differed, with the RTL-W1 cell line being most efficient, followed by RTgutGC. Co-exposures with α-naphthoflavone as potent inhibitor of biotransformation, assessment of CYP1A catalytic activity, and the progression of cellular toxicity upon prolonged BaP exposure revealed that BaP is handled differently in the RTgill-W1 compared to the other two cell lines. Application of the cell-line-derived in vitro clearance rates into a physiology-based toxicokinetic model predicted a BaP bioconcentration factor (BCF) of 909-1057 compared to 920 reported for rainbow trout in vivo.

A validated algorithm for selecting non-toxic chemical concentrations.

The maximal chemical concentration that causes an acceptably small or no effect in an organism or isolated cells is an often-sought-after value in toxicology. Existing approaches to derive this value have raised several concerns; thus, it is often chosen case-by-case based on personal experience. To overcome this ambiguity, we propose an approach for choosing the non-toxic concentration (NtC) of a chemical in a rational, tractable way. We developed an algorithm that identifies the highest chemical concentration that causes no more than 10% effect (= EC10) including the modeled 95% confidence intervals and considering each of the measured biological replicates; and whose toxicity is not significantly different from no effect. The developed algorithm was validated in two steps: by comparing its results with measured and modeled data for 91 dose-response experiments with fish cell lines and/or zebrafish embryos; and by measuring actual effects caused by NtCs in a separate set of experiments using a fish cell line and zebrafish embryos. The algorithm provided an NtC that is more protective than NOEC (no-observed-effect-concentration), NEC (modeled no-effect concentration), EC10 and BMD (benchmark dose). Despite focusing on small-scale bioassays here, this study indicates that the NtC algorithm could be used in various systems. Its application to the survival of zebrafish embryos and to metabolic activity in cell lines showed that NtCs can be applied to different effect measurements, time points, and levels of biological organization. The algorithm is available as Matlab and R source code, and as a free, user-friendly online application.

From the Cover: Zebrafish Larvae Are Insensitive to Stimulation by Cocaine: Importance of Exposure Route and Toxicokinetics.

Zebrafish (Danio rerio) larvae have been suggested as vertebrate model to complement or even replace mammals for rapidly assessing behavioral effects of psychoactive drugs. Yet, divergent responses have been reported in mammals and fish despite the conservation of many drug targets. Cocaine, eg, acts as stimulant in mammals but no such response has been documented for zebrafish larvae. We hypothesized that differences in exposure routes (inhalation or injection in mammals vs waterborne in fish) may be a reason for differences in behavioral responses. We characterized cocaine toxicokinetics by liquid chromatography-mass spectrometry and found its rapid uptake into larvae. We used Matrix-assisted laser desorption ionization-mass spectrometry imaging for the first time to characterize internal distribution of cocaine in zebrafish larvae. Surprisingly, eyes accumulated the highest amount of cocaine and retained most of it even after 48 h depuration. We attribute this to trapping by pigment melanin, a thus far little explored mechanism that may also be relevant for other basic drugs. Cocaine also reached the brain but with levels similar to those in trunk indicating simple passive diffusion as means of distribution which was supported by toxicokinetic models. Although brain levels covered those known to cause hyperactivity in mammals, only hypoactivity (decreased locomotion) was recorded in zebrafish larvae. Our results therefore point to cocaine's anesthetic properties as the dominant mechanism of interaction in the fish: upon entry through the fish skin and gills, it first acts on peripheral nerves rapidly overriding any potential stimulatory response in the brain.

Tralopyril bioconcentration and effects on the gill proteome of the Mediterranean mussel Mytilus galloprovincialis.

Antifouling (AF) systems are used worldwide as one of the most cost-effective ways of protecting submerged structures against heavy biofouling. The emergence of environmentally friendly AF biocides requires knowledge on their environmental fate and toxicity. In this study we measured the bioconcentration of the emerging AF biocide tralopyril (TP) in the Mediterranean mussel Mytilus galloprovincialis and investigated the effects of TP on the mussel gill proteome following acute (2days) and chronic (30days) exposure, as well as after a 10-day depuration period. The experiments were carried out with 1µg/L TP; blank and solvent (5×10(-5)% DMSO) controls were also included. Proteomics analysis was performed by mass spectrometry-based multidimensional protein identification technology (MudPIT). Differentially expressed proteins were identified using a label-free approach based on spectral counts and G-test. Our results show that TP is rapidly accumulated by mussels at concentrations up to 362ng/g dw (whole tissues), reaching steady-state condition within 13days. Ten days of depuration resulted in 80% elimination of accumulated TP from the organism, suggesting that a complete elimination could be reached with longer depuration times. In total, 46 proteins were found to be regulated in the different exposure scenarios. Interestingly, not only TP but also DMSO alone significantly modulated the protein expression in mussel gills following acute and chronic exposure. Both compounds regulated proteins involved in bioenergetics, immune system, active efflux and oxidative stress, often in the opposite way. Alterations of several proteins, notably several cytoskeletal ones, were still observed after the depuration period. These may reflect either the continuing chemical effect due to incomplete elimination or an onset of recovery processes in the mussel gills. Our study shows that exposure of adult mussels to sublethal TP concentration results in the bioconcentration of this biocide in the tissues and modulates the expression of several proteins that may intervene in important metabolic pathways.

Toxicology across scales: Cell population growth in vitro predicts reduced fish growth.

Environmental risk assessment of chemicals is essential but often relies on ethically controversial and expensive methods. We show that tests using cell cultures, combined with modeling of toxicological effects, can replace tests with juvenile fish. Hundreds of thousands of fish at this developmental stage are annually used to assess the influence of chemicals on growth. Juveniles are more sensitive than adult fish, and their growth can affect their chances to survive and reproduce. Thus, to reduce the number of fish used for such tests, we propose a method that can quantitatively predict chemical impact on fish growth based on in vitro data. Our model predicts reduced fish growth in two fish species in excellent agreement with measured in vivo data of two pesticides. This promising step toward alternatives to fish toxicity testing is simple, inexpensive, and fast and only requires in vitro data for model calibration.

Measured and modeled toxicokinetics in cultured fish cells and application to in vitro-in vivo toxicity extrapolation.

Effect concentrations in the toxicity assessment of chemicals with fish and fish cells are generally based on external exposure concentrations. External concentrations as dose metrics, may, however, hamper interpretation and extrapolation of toxicological effects because it is the internal concentration that gives rise to the biological effective dose. Thus, we need to understand the relationship between the external and internal concentrations of chemicals. The objectives of this study were to: (i) elucidate the time-course of the concentration of chemicals with a wide range of physicochemical properties in the compartments of an in vitro test system, (ii) derive a predictive model for toxicokinetics in the in vitro test system, (iii) test the hypothesis that internal effect concentrations in fish (in vivo) and fish cell lines (in vitro) correlate, and (iv) develop a quantitative in vitro to in vivo toxicity extrapolation method for fish acute toxicity. To achieve these goals, time-dependent amounts of organic chemicals were measured in medium, cells (RTgill-W1) and the plastic of exposure wells. Then, the relation between uptake, elimination rate constants, and log KOW was investigated for cells in order to develop a toxicokinetic model. This model was used to predict internal effect concentrations in cells, which were compared with internal effect concentrations in fish gills predicted by a Physiologically Based Toxicokinetic model. Our model could predict concentrations of non-volatile organic chemicals with log KOW between 0.5 and 7 in cells. The correlation of the log ratio of internal effect concentrations in fish gills and the fish gill cell line with the log KOW was significant (r>0.85, p = 0.0008, F-test). This ratio can be predicted from the log KOW of the chemical (77% of variance explained), comprising a promising model to predict lethal effects on fish based on in vitro data.