Abstract
These abstracts were presented at the 2017 annual meeting of the UK In Vitro Toxicology Society (IVTS). The meeting was hosted at the Senate House in London, UK on November 23–24, 2017. The main session topics included hepatotoxicity; dermal and barrier toxicity; IVIVE, exposure and non-mammalian; cardiotoxicity; neurotoxicity; and genotoxicity.
The IVTS (www.ivts.org.uk) was founded in 1988 to support scientists active in the study, practice or development of in vitro toxicology.
Profiling potential drug-induced hepatotoxicity through modelling molecular initiating events (MIEs)
Drug-induced liver injury is reported as a major cause of serious adverse effects, and has led to the withdrawal of a significant number of marketed drugs. During drug development, considerable effort is made to identify chemicals with hepatotoxic potential. Current testing strategies involve screening for possible undesirable interactions of drug candidates with multiple relevant subcellular targets. The initial interaction of the chemical with a subcellular target triggering the chain of events, leading to an adverse effect, is referred to as the molecular initiating event (MIE). During this research a number of MIEs relevant for hepatotoxicity were identified and include such targets as hepatic transporters, nuclear receptors and mitochondrial elements. Datasets reflective of each MIE were compiled from various public sources. Several approaches were investigated for the modelling of MIEs including machine-leaning methods (k-nearest neighbours, random forest, decision tree) and expert analysis of chemical structure. A selection of these models were incorporated into a hepatotoxicity profiler, where the overall hepatotoxic potential of a compound is derived from the predictions of individual MIE models. This profiler was evaluated using an in vivo human hepatotoxicity dataset assembled from multiple sources. The talk will touch on the details of data compilation, its modelability and vision for the applicability of the profiler in drug discovery settings.
Physiological and toxicological evaluation of advanced 3D liver models for toxicological profiling in the pharmaceutical industry
There is an urgent need for reliable mechanistically-based predictive cell-based assays and novel translational biomarkers to safely progress drugs into the clinical setting and ensure the better safety of medicines in patient populations. This is especially true in the case of hepatotoxicity. The IMI MIP-DILI project focused on the validation of different 2D and 3D systems, investigations of mechanisms of drug induced hepatotoxicity and the development of more competent cell systems with better predictivity for DILI, including systems useful for long term experiments. In relation to 3D cultures, many advantages were identified:
• 3D cultivation closer to in vivo environment forming tissue like microstructures
• Organ specific functions (e.g. for liver) are maintained over prolonged time
• 3D cultures are maintained for many weeks allowing application in repeated dose long term toxicity assessment
• It is possible to make disease models
• Can be adapted to state of the art “-omics” techniques such as metabolomics, fluxomics and proteomics
• High throughput imaging is possible.
The potential of these test systems to add value to DILI detection was acknowledged within MIP-DILI, indeed a majority of EFPIA members are interested to continue further work with 3D hepatic spheroid models and can see its application in R&D. However, continued validation is required and further in-house experience will need to be generated before such models become part of routine testing.
Transcriptomic profiling of 3D liver microtissue for mechanistic toxicology
Understanding the mechanisms and interactions between chemical exposure and adverse effects on human biology is paramount in being able to make pragmatic decisions on safety. The use of an adverse outcome pathway (AOP) approach based on understanding the underlying mechanistic effects of human disease relies very much on having appropriate and well characterised in vitro models that respond in the same way as the whole organism or target organ of interest. Such models have to accurately represent the “normal” phenotype and have functioning transporter systems, metabolism or signalling pathways. Therefore the establishment of in vitro systems which accurately represent human biology are essential.
Recently more complex organotypic in vitro models that promise a human-like response to chemical toxicity are becoming increasingly popular. Especially for liver several 3D systems have been developed showing a more liver-like phenotype than conventional immortalized cell lines derived from solid tumour or primary 2D liver cells which are known to rapidly de-differentiate thereby losing their hepatic function. To investigate the applicability of 3D liver microtissues (Cyprotex, UK) in human health risk assessment we investigated the transcriptomic profile of untreated microtissues showing the establishment of a more liver-like phenotype over time, enabling elongated exposure for at least 14 days before cells start to lose their hepatic function again. Elongated (repeat) dosing has previously been shown in some cases to be essential for accurate modelling of hepatotoxic responses. Repeat dose exposure to doxorubicin, a substance known for hepatotoxicity, in these 3D microtissues was investigated over the 14 day period of increased hepatic function. Measurement of toxicological endpoints (ATP content, GSH level, etc.) only show a clear toxic reaction starting at day 10, while evaluation of transcriptional changes identify markers of hepatotoxicity at earlier timepoints with this chemotherapeutic drug.
Paracetamol (APAP) overdose intervention – novel biomarker based in silico methods
Paracetamol (acetaminophen, APAP) is one of the most commonly used drugs in the UK and USA but is responsible for 200 deaths per year in the UK alone. Biomarkers currently used to identify APAP-related liver injury, such as alanine aminotransferase (ALT), tend to lack sensitivity. Novel biomarkers (HMGB1, Full and Fragmented K18) are thought to add value. The framework currently used to define an APAP overdose in the clinic is highly dependent upon time since ingestion and initial dose, information which is often unavailable and difficult to predict. Consequently, critically vulnerable patients are often left untreated, and conversely the APAP antidote, N-acetylcysteine (NAC), is often unnecessarily administered. My project uses mathematical and statistical methods to better understand this complex problem and make quantitative predictions of initial dose, time since ingestion and probability of liver injury based on a panel of novel biomarkers in order to better inform clinical implementation of antidote therapy.
Preliminary results show that initial dose and time since ingestion can be predicted from the biomarker panel with 73.7% and 86.5% accuracy respectively. HMGB1 was found to be the fastest responding biomarker and most significant in predicting liver injury probability. A crucial gap between these in silico results and clinical application is the lack of information and data regarding NAC intervention. Laboratory work, funded by the IVTS fellowship allowed me to attempt filling this data-gap, focusing on the in vitro relationship between cell death, APAP dose and NAC treatment. My aim now is to incorporate information obtained through first-hand experience of standard techniques such as ATP/LDH assays, FACS, qPCR and Western Blotting into the modelling framework, improving clinical relevance and predictivity.
In vitro models of the gas-blood barrier of the lung
The lung is structurally complex in order to protect the delicate peripheral respiratory zone from unwanted inhaled toxins, including noxious gases, airborne particles and microorganisms, whilst achieving effective gas exchange. Nevertheless, there are times when all these agents might access the gas-exchange units and have adverse effects. With respect to inhaled particles, very small micro- and nano-sized particles can relatively easily access the deep lung, and there is strong evidence that such particles in ambient air pollution not only affect the lung (e.g. aggravating asthma, reducing lung growth in children), they also impact the cardiovascular system and have been implicated in diseases such as diabetes. In vitro models of the alveolar unit allow comprehensive (screening) study of putative toxicity and bioreactivity of existing and new (e.g. engineered) particulate materials. A number of “3D” models of the alveolar unit are being used. Studies of in vitro models using primary human lung cells will be described to illustrate the importance of considering all the cellular components of the alveolar unit when assessing particle bioreactivity and translocation at the gas-blood barrier. There have been significant advances in recent years in development of alveolar barrier models to study the behaviour of agents at this interface, including “Lung on a chip”, microfluidics, 3D organoids/spheres and stem cell approaches, the relative benefits of which will be discussed.
Development and application of a fragment-based in silico profiler for the prediction of chemical reactivity and toxicity
The Adverse Outcome Pathway (AOP) paradigm details the existing knowledge that links the initial interaction between a chemical and a biological system, termed the molecular initiating event (MIE), through a series of intermediate events, to an adverse effect1. An important example of a well-defined MIE is the formation of a covalent bond between a biological nucleophile and electrophilic compounds2. This particular MIE has been associated with various toxicological endpoints such as acute aquatic toxicity, skin sensitisation and respiratory sensitisation. This study has investigated the calculated parameters that are required to predict the rate of chemical bond formation (reactivity) of a dataset of α-β-unsaturated carbonyls (Michael acceptors). Reactivity of these compounds towards glutathione was predicted using a combination of a calculated activation energy value (Eact, calculated using Density Functional Theory (DFT) calculation at the B3YLP/6-31G+(d) level of theory), and solvent accessible surface area values (SAS) at the alpha carbon. The analysis produced excellent models for the majority of the dataset. Based on the results, a fragment-based algorithm was developed enabling the reactivity to be predicted for linear Michael acceptors without the need to perform the time-consuming DFT calculations. This presentation will outline the development of the fragment-based algorithm for predicting chemical reactivity and discuss its application for the prediction of toxicity within the AOP approach.
1. G. T. Ankley, R. S. Bennett, R. J. Erickson, D. J. Hoff, M. W. Hornung, R. D. Johnson, D. R. Mount, J. W. Nichols, C. L. Russom, P. K. Schmieder, J. A. Serrrano, J. E. Tietge and D. L. Villeneuve, Environ Toxicol Chem, 2010, 29, 730–741.
2. A. O. Aptula and D. W. Roberts, Chem Res Toxicol, 2006, 19, 1097–1105.
Defining skin xenobiotic metabolism using a combined in vitro/in silico approach
Department of Applied Mathematics, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF UK.
Skin represents an important route of exposure and determining whether such incidental or intentional exposure poses a risk to human health requires consideration of temporal concentration, in addition to assessing the chemical's intrinsic hazard. In order to elicit a toxic response in vivo the chemical must reach its site of action in sufficient concentration, as determined by its absorption, distribution, metabolism and elimination (ADME) profile. Whilst absorption and distribution into and through skin layers have been studied for decades, only more recently has skin metabolism become a subject of intense research, now recognised as playing a key role in both toxification and detoxification processes.
EU directives on animal use for toxicity testing and the lack of human skin for research has prompted an increase in the use of tissue-engineered human skin models. These models are histologically similar to human skin and express metabolising enzymes making them ideal in vitro tools for toxicity testing. In this talk, I will discuss how the use of these models in combination with in silico tools may be used to resolve a significant challenge in predicting toxicity following dermal exposure. I will highlight how in vitro data can be used to drive novel multiscale-mathematical models that predict the kinetics of xenobiotic metabolising enzymes, their transdermal distribution, spatiotemporal metabolite distributions and whole body systemic exposures for a wide range of chemical structures. The ability to predict metabolism in the skin would significantly aid risk assessment and shorten the length of time from discovery to patient benefit.
COSMOlogic GmbH & Co. KG, Imbacher Weg 46, D-51379 Leverkusen, Germany
Biomembrane permeabilities are of general interest in the uptake and distribution of pharmaceutical agents, chemical toxicants and environmental pollutants. The first step of a toxicological adverse outcome pathway is the availability of chemicals at the site of a particular molecular initiating event. Several barriers have to be overcome within biological systems. Therefore, knowledge about substance-specific barrier heights and permeabilities are essential in the extrapolation of in vitro cellular responses to in vivo systems.
We present the new, entirely mechanistic COSMOperm method to predict passive membrane permeabilities. The COSMOperm approach is based on substance-specific free energy profiles ΔG(z) within a biomembrane of interest from COSMO-RS (Conductor-like Screening Model for Realistic Solvation) calculations. These are combined with membrane layer specific diffusion coefficients D(z), for example, in the water phase, the polar head groups and alkyl tails of biochemical phospholipid bilayers. COSMO-RS is based on first-principle quantum chemical structures, and uses physically sound intermolecular interactions (electrostatic, hydrogen bond and van der Waals). For this reason, it is unbiased towards different application scenarios, as cosmetics, chemicals and pharmaceuticals.
A fully predictive calculation of passive permeation through phospholipid bilayer membranes results in a performance of r2 = 0.89; rmsd = 0.76 log10(cm/s), as compared to gold standard black lipid membrane (BLM) experiments.
Because of its generality, predictivity and sound mechanistic treatment, the COSMOperm approach is a valuable tool for diverse chemical datasets, including the handling of protonated states and ionic structures. In addition, ΔG(z) and D(z) profiles are obtained quickly within minutes, as compared to time-consuming molecular dynamics simulations.
Quantifying both toxicokinetics and toxicodynamics can enable better in vitro to in vivo ecotoxicity extrapolation. This is because toxicokinetics are primarily related to physical and chemical properties whereas toxicodynamics are primarily related to molecular structure. We already have fairly good models to predict toxicokinetic aspects (uptake, elimination) for a wide range of substances depending on substance hydrophobicity. Modelling internal dose is the first step towards better in vitro to in vivo toxicity extrapolation because it allows identification of the concentration of the toxicant in the organism and in the cell which is the biologically relevant dose. The second step is modelling organism level endpoints based on molecular structure and in vitro data. Interestingly, the typical ‘box and arrow’ depiction of toxicodynamic models makes them look similar to AOPs. However one important difference is that in toxicodynamic models the variables are usually representing state variables of the organism, whereas AOPs tend to mostly model cellular state variables. Recognising and addressing this issue of scale will likely lead to better theory and models. Another matter of debate is the trade-off between generality and level of detail needed. I will discuss how a toxicokinetic-toxicodynamic framework can help to develop novel methods to replace animal toxicity testing in ecotoxicology.
Integrated drug effect data using graph databases
ChEMBL is one of the largest and most widely used bioactivity database freely available for the whole community1. In its latest version, ChEMBL contains 15 million data points measured on more than 1.7 million compounds curated from the scientific literature and datasets deposited by companies and not-for-profit organisations. In addition to activity data from early stage drug discovery and approved drugs, ChEMBL also contains useful data on the ADME properties of compounds and on their toxicities.
We are a member of several international projects (eTOX, HeCaToS, EU-ToxRisk, TransQST) which aim to better predict compound toxicity and where we gather and model toxicity data. The goal of HeCaToS is to establish better prediction models for human heart and liver toxicity, by challenging 3D human cardiac and hepatic cell models with relevant doses of known cardio- and hepato- toxicants and then comparing the output with data from heart and liver biopsies taken from patients treated with the same toxicants.2,3
We will describe some of our contributions to the HeCaToS project. These include the analysis of compound effects on protein targets and in higher order assays such as cellular or in vivo assays. We have sought to establish relationships between different types of assays in order to understand the mechanisms leading either to a therapeutic effect (Mode of Action) or to toxicity (Adverse Outcome Pathways). We have also investigated the use of a graph database approach in exploring these relationships. These studies demonstrate that, where the data are available, it is possible to highlight known drug effects.
References
1. Gaulton, A. et al. The ChEMBL database in 2017. Nucleic Acids Res. 45, D945–D954 (2017).
2. Kuepfer, L. et al. A model-based assay design to reproduce in vivo patterns of acute drug-induced toxicity. Arch. Toxicol. (2017). doi:10.1007/s00204-017-2041-7
3. HeCaToS website: http://www.hecatos.eu/home
Using benchmark dose modelling for potency ranking and correlation across endpoints
The benchmark dose (BMD) modelling approach utilises a critical effect size (CES) to calculate a small but measurable effect on the dose response. Confidence intervals are also calculated, which provide a measure of precision for the BMD. Recent advances in this area make it possible to analyse multiple dose responses at the same time with compound used as covariate. For the analysis of continuous in vitro data, the BMD confidence intervals can be plotted in rank order, with the lowest BMD confidence intervals showing that compound to have high potency and vice-versa. This provides a rank order of chemicals, with the example used here being genotoxic potency. When this approach is carried out on continuous in vivo dose response data from a comparable endpoint, then the potency ranks can be compared to each other. This shows whether the in vitro assay estimates potency in the same or similar way to the in vivo assay. The two potency ranks can also be plotted together to see if there is a correlation between these endpoints, and if there is, the in vitro BMD metrics could be used to estimate the in vivo BMD metrics. This approach can assist in dose setting for in vivo studies and when combined with information on mechanism of action can be used to support chemical grouping and read across. The next step is to include information on oral equivalent doses which allow for bioactivity to exposure ratios, which provides the opportunity to incorporate a risk-based strategy.
Developing realistic fish in vitro models to assess impact of contaminants in the aquatic environment
The use of fish in biological and toxicological research is on increase and it is perceived that it will continue to increase in order to meet new regulatory requirements. Inherent limitations in the use of available fish cell lines has necessitated the development and modification of specialized culture techniques from various organs (e.g. liver, gill and gut) to investigate their use in line with the adoption of 3Rs principles. We have developed and established the physiological comparability and metabolic capacity of these cultures grown as 3-dimensional (3D) models. The methodological modifications of these developed models using liver, gill and gut cells in our laboratory offer a reproducible, replacement model for environmental hazard assessments. The presentation will cover some of our recent work in terms of co-culturing the aforementioned organs together providing a ‘virtual fish’ model, mimicking the real in vivo situation. The results will include further characterization (i.e. biochemical, cellular, histological, transcriptomics and proteomics) of these models and biological responses following exposure to different contaminants (metals and organics).
Pluripotent stem cells for drug discovery and toxicity assessment
Toxicity testing based upon animal models or transformed cell lines is not always an accurate representation of the response of human tissues and organs to xenobiotic substances. The development of human pluripotent stem cells, which are capable of generating many of the cell types found in the adult body, may be an effective solution to address this problem; therefore, this lecture will attempt to present not only the background of what pluripotent stem cells are and how they are made but also how we can use them to produce versatile new toxicity assays for use in pharmaceutical development.
An update on the Comprehensive in vitro Proarrhythmia Assay (CiPA) and the role of mathematical models in risk assessment
The Comprehensive in vitro Proarrhythmia Assay (CiPA) aims to improve assessment of pro-arrhythmic risk associated with novel pharmaceutical compounds to a point where the existing dedicated clinical trial (the Thorough QT study) could be replaced with a more accurate preclinical assessment. The CiPA proposal includes four main components: in vitro ion channel screening, in silico action potential modelling, in vitro stem-cell derived myocyte assays, and a more lightweight clinical check as part of standard Phase I ECG. In this talk I'll give an update on CiPA's four streams and describe the ongoing validation efforts, as well as a more in-depth discussion of the mathematical modelling aspects that I have been contributing towards.
The use of impedance-based systems to detect the structural and functional cardiotoxicity of cancer therapies
Accurate predication of cardiac liabilities pre-clinically is of crucial importance, with current methodologies proving inadequate, evidenced by drug withdrawals and the association of many cancer therapeutics with occurrence of cardiotoxicity. The adverse effects of cancer therapies on the heart can be broadly classified into structural and functional cardiotoxicities, both of which can be detected using iPSC-derived cardiomyocytes (iPSC-CMs) on the real-time impedance-based xCELLigence cardio system. Due to the rapid data acquisition and ability to record the transient minute movements of iPSC-CMs via changes in impedance, information on overall cell health, morphology and contractility can be monitored. Using a similar system, the xCELLigence real-time cell analyser (RTCA), changes in morphology can be detected on cardiomyocyte cell lines such as AC10 cardiomyocytes (AC10-CMs), showing the potential of this platform for detection of structural cardiotoxicities. Both of these impedance-based systems have the advantage that long-term experiments can be conducted, thus allowing recapitulation of the progressive nature of human toxicity development. This study has utilised these real-time impedance-based xCELLigence systems to demonstrate induction of cardiomyocyte hypertrophy by the tyrosine kinase inhibitor sunitinib and the notoriously cardiotoxic anthracyclines, with changes in contractility also detected with sunitinib addition. The sensitivity of these systems at detecting changes in cellular morphology is evidenced by concurrent anthracycline and angiotensin receptor blocker (ARB) treatment, where reductions in hypertrophy were observed. Recent clinical studies have demonstrated that administration of drugs that act upon the angiotensin system may reduce the cardiotoxicity of anthracyclines, thus the reduction of anthracycline-induced hypertrophy by ARBs shows the translational potential of these systems. The methodologies used offer a model for assessing drug-induced cardiotoxicity that is robust, clinically relevant and animal-free, with the study identifying both structural and functional cardiotoxicities and a novel mechanism for anthracycline-induced cardiotoxicity.
Evaluation of in vitro cellular models for use in functional and structural studies of drug-induced cardiotoxicity
Detrimental effects upon the cardiac system are a major cause of drug attrition. Current in vitro methodologies for assessment of drug-induced cardiotoxicity involve sub-optimal screens of non-cardiac cell lines engineered to express cardiac ion channels, or primary tissue with limited utility for clinical translation. The advent of human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CM), with the ability to synchronously beat (contract) in vitro; opened up opportunities for improved identification of cardiotoxicity. However, these cells are costly, time-limited, and require complex maintenance techniques, posing constraints in terms of widespread use. One resolution is the use of immortalised cardiac cell lines, with capability for continuous growth and clinical translation. However, the limitations of these models for detecting both structural and functional cardiotoxicity is unknown. In this study we assess the predictive value of an immortalised non-contractile human cardiac cell line (AC10) and functionally-responsive hiPSC-CM to determine toxicity of the histone deacetylase inhibitor (HDACi) class of drug, thus highlighting the advantages of each type of cell model in relation to structural and functional cardiotoxicity. This not only supports the value of comprehensive cellular screening models, but offers a predictive tool to assess cardiotoxicity that would allow development of efficacious and safer drugs in this class.
The development and characterisation of 3D neuronal microtissues for safety testing
In order to improve our understanding and prediction of neurotoxicity a 3D neuronal microtissue was developed from iPSC derived neurones and astrocytes using ultralow adhesion plates. Microtissues formed over 3 days (day 0) with clear clustering, self-attachment, with a regular shape and stable ATP content within a single form. A gene expression time series analysis of the neuronal microtissues at day 0, 7, 14, 21, 25 and 28 days was performed using Agilent whole human genome oligo microarray G4851B (n = 4 per time point). A total of five statistical comparisons were performed. Each comparison includes one timepoint (7, 14, 21, 25, or 28) relative to timepoint zero. The enriched KEGG pathways amongst the genes that are up-regulated over time are relatively consistent for timepoints 14 and later. The strongest enrichment can be seen for various processes linked to neuronal signalling (e.g. “Retrograde endocannabinoid signaling” and “Dopaminergic synapse”) together with various processes associated with addiction. The gene profiles suggest the MTs continue to mature over the 28 day period. GFAP (astrocyte marker) and βIII-tubulin (neurone marker) immunofluorescence show an increase in astrocyte number on the microtissue periphery suggesting either astrocyte proliferation and/or migration, potentially linked to the maturation. Various known neurotoxicants were exposed to mature and immature microtissues for 72 or 336 hours and cell health was determined using confocal high content screening. Interestingly differential sensitivity to different neurotoxicants was observed depending upon the maturity of the neuronal microtissue.
Applying Adverse Outcome pathways (AOPs) to support Integrated Approaches to testing and Assessment (IATA) for identification of potential developmental neuro-toxicants
Recent societal concerns have been raised linking the rise in children's developmental learning disabilities to chemical exposures. However, there is a lack of information concerning the developmental neurotoxicity (DNT) hazard posed by industrial and environmental chemicals due to a complex current testing that is entirely based on animal studies. New testing approach, based on a battery of in vitro DNT assays anchored to common key events identified in the existing DNT Adverse Outcome Pathways (AOPs) will be discussed. AOP-informed Integrated Approaches to Testing and Assessment (IATA) will be proposed for screening and prioritization of chemicals with DNT potential. For generation of new data IATA framework should be based on a set of non-testing and in vitro test methods that can be used in a flexible combination (fit-for-purpose). Such IATA would facilitate an application of mechanistic knowledge into DNT evaluation, increasing scientific confidence in decision making process, delivering data that could contribute to hazard identification and characterization and possibly safety assessment of chemicals, speeding up evaluation of thousands of compounds present in industrial, agricultural and consumer products that lack safety data on DNT potential.
The AOP concept relies on understanding causal relationship between the Molecular Initiating Event (MIE), in which a chemical interacts with a biological target, resulting in a sequential series of measurable key events (KEs), which are triggered at different biological levels (cellular, tissue, organ) ultimately resulting in adverse outcome (AO) manifesting in an individual organism and/or a population. DNT AOPs hold great potential to impact the manner in which in vitro DNT data can be interpreted since the causative links between MIEs, KEs and AO are based on empirical, mechanistic data and biologically relevant knowledge, providing more certainty for regulatory use.
Moreover, AOPs provide a strong biological/pathophysiological rationale to compound classification, which is usually based on chemical structures correlated to apical endpoints from animal experiments. It is an important tool that facilitates generation of the data needed for formation of chemical biological categories: chemicals can be grouped according to their MIEs, and common KEs. AOP-based biological chemical grouping has the potential to add a value for DNT testing due to the complex nature of the underlying biology that is currently inadequately captured by chemical category formation (structure or reactivity). Furthermore, read-across and toxicity classification models can be vastly improved when large amounts of in vitro data are available from high-throughput testing. However, currently the limited number of the developed DNT AOPs has hampered both judgement of the predictive ability, as well as regulatory use of high-throughput in vitro DNT data.
The concept that underlies the AOP framework can also guide more effective selection of existing in vitro DNT data and can advise on the most relevant in vitro tests to be included in Integrated Approaches to Testing and Assessment (IATA) for generation of new data reflecting appropriate coverage of MIEs and KEs.
A novel in silico predictive model of thyroid hormone metabolism in the brain: Towards a fully parametrized materno-foetal PBPK model
Department of Applied Mathematics, Liverpool John Moores University
Thyroid hormones (TH) are essential for the control of metabolism and neurodevelopment. Altered thyroid levels during critical periods of development could result in adverse outcomes in the developing human foetus.
Exogenous compounds can exert thyroid effects through a variety of mechanisms leading to thyroid dysfunction. Globally this dysfunction can contribute to childhood neurological impairments in humans. Life-long exposure to a vast mixture of chemicals in low doses, which can be cumulative for persistent, and the large physiological range of TH in humans results in a large variation of measurements between individuals. This makes studies of human populations very difficult.
My poster details a novel in silico method of modelling TH homeostasis to establish an understanding of under what circumstances chemicals can perturb this homeostasis sufficiently to produce an adverse effect in human neurodevelopment (in utero and infant).
The model is formulated using an extension of Graph Theory called Petri Nets. It utilises a hybrid of non-deterministic and stochastic methods to model the concurrent processes involved. Ordinary differential equations describing the rate of change of concentrations of THs and related enzymes can be translated from the Petri Net.
The objective of my PhD is to model foetal-maternal thyroid homeostasis using physiologically based pharmacokinetics (PBPK) to develop a multi-compartmental model that will cover key developmental stages of the human foetus when critical neurodevelopmental effects of thyroid hormones are occurring in the hypothalamic-pituitary-thyroid (HPT) axis. This can hopefully be of benefit in a wide range of scenarios including Adverse Outcome Pathways (AOPs).
Whole genome sequencing of human tumours has revealed distinct patterns of mutation that hint at the causative origins of cancer. This can be tested by examining the mutational signatures induced in experimental systems by putative cancer-causing agents. We have generated such signatures in mutagen-exposed mouse embryo fibroblasts (MEFs) and in human induced pluripotent stem (hiPS) cells. The results reveal that each mutagen induces a characteristic mutation signature that, in some cases, matches a signature found in human tumours.
In an analysis of somatic mutations in cancers for which tobacco smoking confers an elevated risk, it was found that smoking is associated with increased mutation burdens of multiple different mutational signatures, which contribute to different extents in different tissues. One of these signatures, mainly found in tissues directly exposed to tobacco smoke, is attributable to misreplication of DNA damage caused by tobacco carcinogens. Others likely reflect indirect activation of DNA editing by APOBEC cytidine deaminases and of an endogenous clock-like mutational process. The results are consistent with the proposition that smoking increases cancer risk by increasing the somatic mutation load although direct evidence for this mechanism is lacking in some cancer types.
Application of the ToxTracker reporter assay in a mode of action approach for genetic toxicology assessment
Toxys B.V., Leiden, the Netherlands
ToxTracker is a mammalian stem cell-based reporter assay that detects activation of specific cellular signalling pathways upon exposure to unknown compounds (Hendriks et al, Tox Sci 2016). ToxTracker contains six different GFP-tagged reporters that allows discrimination between induction of DNA damage, oxidative stress and protein damage in a single test.
In an extensive validation study using 250 reference compounds, ToxTracker classified the genotoxic carcinogens as genotoxic with a sensitivity of 94%. The non-genotoxic carcinogens and non-carcinogens were classified as non-genotoxic by ToxTracker with a specificity of 95%. Interestingly, various compounds that give misleading positive results in the conventional in vitro genotoxicity assays did not activate the DNA damage reporters but did induce high levels of oxidative stress or protein damage in ToxTracker.
Next we investigated if ToxTracker could provide insight into the mode of action of genotoxic compounds. By assessing the differential induction of the two DNA damage reporters, ToxTracker was able to discriminate between a mutagenic and clastogenic mechanism of genotoxicity. Furthermore, we found that the assay could discriminate between a clastogenic and aneugenic mode by the selective induction of the Rtkn-GFP DNA strand break reporter. Furthermore, induction of the Rtkn-GFP reporter was significantly slower (>12 h) for the mitotic spindle poisons compared to clastogenic compounds (8 h).
Finally, by staining for phosphorylation of histone H3 and including a DNA stain for polyploidy in the reporter cell lines, ToxTracker can identify an aneugenic MOA by inhibition of cell cycle kinases.
In vitro testing for DNA damage (genotoxicity) is a fundamental part of tiered safety assessment, specifically to identify genotoxic hazards that may need to be followed up with in vivo tests. However, it is hoped that the development of more sophisticated in vitro approaches may, one day, replace some traditional in vivo approaches altogether.
In order to develop more sophisticated in vitro genotoxicity testing approaches, we have been considering the type of dosing used (acute v sub-chronic) and the impact of this dosing regime on the outcomes achieved. We have shown that repeat dose testing can produce considerably different results compared to the traditional acute dosing.
For example, in TK6 cells, a more chronic dosing regime (5 day, daily exposures) reduced the genotoxicity observed when cells are exposed to genotoxic methylating agents (MMS, MNU). In these studies, the acute doses were fractionated so that the cumulative dose was equal, but the fractionated dose was far less mutagenic (Chapman et al., 2015).
In further experiments using carcinogens which are not DNA reactive (non-genotoxic carcinogens), we have shown the opposite effect. That is, that fractionated doses “draw out” the genotoxicity which is otherwise not observed with acute dosing at the same cumulative levels. For example, Nickel Chloride is not genotoxic in TK6 cells during acute (24 hour) exposures. However, 5 day daily exposures to 1/5th of the same doses shows marked genotoxicity.
A further extension of the repeat-dose exposure systems is the development of passive-dosing systems which produce low level continual exposures (particularly to hydrophobic compounds). We have developed some passive dosing approaches for genotoxicity in lymphoblastoid cells (MCL5) exposed to the hydrophobic genotoxin B[a]P.
Hence, developing more sophisticated exposure scenarios that more accurately reflect human exposure situations (low dose and chronic) may produce more biologically meaningful data. These repeat-dose testing scenarios may also lead to the development of in vitro approaches to detect non-genotoxic carcinogens. Ultimately these developments may reduce the use of animals in testing.
Chapman K, Doak SH, Jenkins GJS (2015). Acute dosing and p53-deficiency promote cellular sensitivity to DNA methylating agents. Tox Sci 144: 357–365.
Toxicants that inhibit Apoptosome formation identified using split luciferase assay for Apaf-1 oligomerization
Pharmacology and Therapeutics, School of Medicine, NUI Galway
Here, we report an in vitro split luciferase based assay that can be used to identify modulators of apoptosome formation, a key step in programmed cell death. In this method, an N-terminal fragment of luciferase is fused to one protein and a C-terminal fragment is fused to a second protein. Luciferase activity is reconstituted if the two proteins interact and bring the fragments into close proximity with each other. In the current study, we demonstrate for the first time that split luciferase/Apaf-1 reagents show dATP and cytochrome c dependent luciferase activity that was blocked by a known inhibitor of apoptosome formation (NS3694). This assay was then used to screen a library of toxicants and identify six new inhibitors of Apaf-1. As apoptosome formation is required for spermatogenesis in insects and mammals, this assay could be used to complement existing in vitro tests for reproductive toxicants.
School of Science and Technology, Nottingham Trent University, Nottingham, UK
Organophosphates are widely spread and overused not only in agriculture, but also can be used as solvents, fire retardants and other categories. This high exposure to OPs and associated increased rate of pesticide poisoning is a major public health concern. The OPs selected for this study were chlorpyrifos (CPF), and its metabolite chlorpyifos oxon (CPFO); these were chosen because they are widely used, have been studied for toxicity, and can induce all of the known types of OP depending on the dose, age of the patient, and the duration of exposure. A key interest to the current work is to investigate the ability of sub-lethal concentrations (1, 3, 10 μM) of these OPs to disrupt the differentiation of mouse N2a neuroblastoma cells, and to analyze the underlying changes in the levels of cytoskeletal gene expression and protein synthesis. To understand the possible effects of OPs on cytoskeletal gene expression, oligonucleotide probes were generated to study the associated changes in expression for genes encoding βIII tubulin, the microtubule associated protein tau (MAPT) and neurofilament heavy chain (NFH). These probes were optimized for qPCR analysis and the levels of the corresponding proteins were detected by Western blot analysis. The results indicated that exposure of differentiating N2a cells to sub-lethal concentrations of OPs was associated with impaired cell differentiation and disruption of cytoskeletal proteins, together with reduced expression of genes encoding βIII tubulin, MAPT and NFH.
Jack Birch Unit of Molecular Carcinogenesis, Department of Biology, University of York, YO10 5DD, UK
Led by research into breast cancer, bladder cancer is being classified into subtypes of disease based on transcriptomic and mutational profiles which could define susceptibility to different therapies. Meta-analysis of current classifications by different researchers suggests all the approaches used identify two main groups of muscle-invasive bladder cancer; frequently dubbed “luminal” for the differentiated tumours and “basal” for the less-differentiated. A CYP1A1-activated pro-drug (ICT2700) was previously described for bladder cancer and here we propose to refine patient selection based on predicted CYP-metabolic function.
In an in vitro model of normal human urothelium, CYP-metabolism was inextricably linked to differentiation by the expression of the essential reductases; P450 Oxidoreductase (POR) and cytochrome b5 (CYB5A). This association of CYP metabolic potential with differentiation extended into tumours where CYP1A1, POR and CYB5A transcripts were all significantly enriched in the luminal tumours of the Cancer Genome Atlas consortium. We examined a separate cohort of MIBC tumours histologically (using KRT5/6 and GATA3 to provide the basal/luminal classification) and found a small group of POR over-expressing tumours (11%) that were all classified as luminal. In vitro studies of basal (T24 a.k.a. “EJ138”) and luminal (RT4 and RT112) bladder cancer cell lines showed ethoxyresorufin O-deethylation activity was only detectable in luminal cell lines and linked to POR expression. Furthermore the IC50 for ICT2700 assessed by MTT assay was 0.18μM (±0.11 SD) in luminal RT112 cells and 3.77μM (±1.9 SD) in basal T24 cells supporting the concept that targeting a sub-group of luminal MIBC patients will enhance efficacy.
We are currently extending studies of CYP-metabolism and ICT2700-cytotoxicity into additional representative cancer cell lines and normal tissue models before planning xenograft studies. However, a reliable CYP1A1 antibody may be required for histological classification of MIBC patients and to-date all those tested have proved non-specific.
Centre for Molecular Informatics, Department of Chemistry, University of Cambridge
Extrapolation of in vitro to in vivo toxicity is challenging, especially when in vivo derived exposure measures, such as dose and Cmax, are not available. In this work we employed rule-based methods with modifications to improve the biological relevance of the associations, taking into account physicochemical properties of compounds. Two lowest effective levels (LEL), 15mg/kg/day and 500mg/kg/day were extracted from ToxRefDB for hepatotoxicity in rats for 673 compounds. These endpoints were combined with 361 bioactivity measurements from ToxCast, as well as 29 calculated physicochemical properties.
We found that the most important bioactivity classes were consistent at both toxicity levels, namely activities against Cytochrome P, immunological responses and nuclear receptor activities, which align with known mechanisms of hepatotoxicity. While nuclear receptor activity is not currently tested in the four commercial in vitro models for hepatotoxicity, their associations with hepatotoxicity was strengthened in terms of accuracy to around 15% when proper physicochemical conditions were met. Interestingly, we find that the most frequent physicochemical properties used in rules, namely number of rotatable bonds and number of rings, are linked to bioavailability parameters such as permeability and plasma protein binding, respectively.
Overall, we suggest that the coverage of hepatotoxicity in vitro models can be improved by incorporating a broad range of bioactivity classes. We also propose to consider physiochemical properties of compounds when interpreting outcomes from in vitro models, which at least to some extent seem to be able to substitute the use of in vivo exposure measures at primary stages of toxicity screening.
Computational Chemistry & Biology, Global Research & Development, Merck KGaA, Darmstadt, Germany
The results of biological assays, such as metabolic stability assays, are often associated with high variability. Despite this disadvantage, it is crucial in drug discovery and other disciplines to be able to rely on such assay results. A novel approach based on in silico predictions and confidence scoring (CS) triggering re-testing is suggested to increase confidence in experimental data. After successful validation of the in silico models and confidence scoring, 138 experiments with high confidence (CS >0.7) were repeated. In this study 15.2% of these experiments were not in concordance with the initial experimental results when repeated. While correctly in silico predicted experiments (n = 73) had a low rate of changed experimental outcomes (4.1%), incorrectly in silico predicted experiments (n = 65) had a higher rate of changed experimental outcomes (27.7%). This study suggests that in silico predictions have the potential to identify experimental results which are potentially inaccurate or variable. This concept may be used to identify experiments that benefit from re-testing. Ultimately such a re-testing strategy can lead to an overall more accurate data pool which has a positive effect on decision-making. As a further benefit, the increase in accuracy of the underlying in vitro data allows a refinement of in silico models and thereby more accurate predictions. This may even be interpreted as a virtuous circle of data quality.
The kidney is the most important organ involved in the elimination of chemicals and their metabolites as such are a common site for toxicity. Chemical induced nephrotoxicity is one of the leading causes of acute kidney injury (AKI) worldwide. Here we show the characterisation and evaluation of three kidney in vitro models. Renal proximal tubular epithelial cells (RPTEC) grown in 2D, RPTEC spheroids grown in ultralow adhesion plates (ULA) and kidney multi-cellular microtissues (MTs) consisting of RPTEC, renal fibroblasts (RF) and endothelial cells (EC). The spheroids and MTs display uniform size, shape, longevity and more accurately reflect the complex in vivo microenvironment compared with traditional 2D cell cultures. Known nephrotoxicants were exposed to these tissues for either 72h or over repeat dose exposures of 9 and 14 days. Following exposure confocal high content screening (HCS) analysis of mitochondrial dysfunction, glutathione content, and reactive oxygen species was determined combined with a measure of cellular ATP. The lowest minimum effective concentration (MEC) was selected for each compound. The aminoglycoside gentamicin responds (239-385 μM) in all three models but only following a 9-day exposure with comparable sensitivity. Diclofenac, the most commonly implicated NSAID reported to cause nephrotoxicity, exhibited glutathione depletion at 3-fold (9 days) and 6-fold (14 days) greater sensitivity over 2D RPTEC cultures, responding within 5x human Cmax levels. Camptothecin and cyclosporine A respond in all three models within 1x human Cmax levels (0.01-0.02 μM and 0.1-6 μM respectively). The chemotherapy agent cisplatin was shown to only respond in 3D models within 5x human Cmax (13.7-19.5μM). Further endpoints will be evaluated using HCS to identify other potential early markers of nephrotoxicity such as lysosomal lesions as would be expected to occur following gentamicin exposure. We show the use of in vitro kidney MTs with confocal HCS, following repeat dosing, allows for improved in vitro to in vivo extrapolation of potential chemical induced kidney liabilities.
Cardiac microtissue models for the improved prediction of drug induced cardiac hypertrophy
Drugs can exert morphological damage to the myocardium such as hypertrophy. Repeat dose drug exposures are often required for the manifestation of such toxicities, however, current preclinical in vitro models focus primarily on a single cell population in a restrictive two-dimensional format with limited longevity. The myocardial cell population comprises 70% non-myocytes such as endothelial and fibroblasts. Recent data suggests these cells are involved in the maturation of stem cell derived cardiomyocytes in vitro, however their role in drug induced structural cardiotoxicity is yet to be established. Here, we assessed human derived cardiac microtissues (MTs) formed as either monocultures (cardiomyocytes alone), co-cultures (cardiomyocytes with endothelial or fibroblasts) or tri-cultures (cardiomyocyte, endothelial & fibroblasts) using a low adhesion cell suspension method. Utilising time course brightfield imaging of cardiac MT area throughout a repeat dose compound exposure setting, hypertrophy can be detected prior to gross cytotoxicity. MTs were exposed to 10 known hypertrophy inducing cardiotoxins and 4 other structural cardiotoxins (plus 2 negatives) over 14 days. At day 14, fluorescent HCS analysis of calcium homeostasis and mitochondrial dysfunction was combined with a measure of cellular ATP, to determine structural cardiotoxicity. With a 10-fold Cmax cut off applied the mono-MTs correctly predicted cardiac hypertrophy while the tri-MTs failed to indicate hypertrophy. Mono-, co- and tri-MTs were then exposed to a panel of compounds and assessed for increased B-Type Natriuretic Peptide (BNP), a known marker of cardiac hypertrophy. BNP highlighted hypertrophic responses in all MT models, however MT area increase is restricted in MTs comprising cardiac fibroblasts. This study shows how using a single organotypic human derived 3D model per well and automated multiplexed confocal HCS can enhance the in vitro to in vivo translation of the potential for drug induced hypertrophy. However, once cardiac fibroblasts are introduced no hypertrophic responses can be detected with size measurement alone, despite equal general structural cardiotoxicity prediction.
Phenyl acetate hydrolysis and MTT reduction as markers for enzyme activity and stability in human skin tested in vitro
As late as the 1990s, dermal metabolism was not considered to be important for dermal absorption, dermal toxicity or drug efficacy, and analytical methods were not sensitive enough to detect the low levels of turnover or account for losses. Now, both analytical sensitivity and our understanding of the contribution that dermal metabolism makes to disease (e.g. skin sensitization) and drug efficacy have progressed considerably.
Because of the now recognised importance of dermal metabolism, in vitro studies using human skin are conducted to evaluate metabolism. The presence and activity of enzymes should be demonstrated in the tissues used for these studies.
Human skin discs from 3 donors were prepared and tested in vitro to assess esterase and (primarily) mitochondrial reductase activity, using phenyl acetate (PA) and methylthiazolyldiphenyl-tetrazolium bromide (MTT) assays, respectively. Incubations were performed on arrival at the laboratory (i.e. fresh), following overnight storage at 32°C, +4°C and −20°C, and following extended storage for 1, 2 or 4 weeks at −20°C and 80°C. PA was also measured after a 24 h incubation at the end of each storage period to assess any change in activity during an extended in vitro incubation.
For MTT and PA, there were no statistical differences between fresh and stored or pre- and post-incubation for any storage duration. In conclusion, skin may be stored for short periods before use in dermal metabolism studies.
Development of Solute Carrier (SLC) Transporters Kidney Cell Models Using hTERT-immortalized Renal Proximal Tubule Epithelial Cells (RPTEC/TERT1)
The disposition and clearance of drugs by the kidney relies largely on a subset of membrane transport pumps known as solute carrier proteins (SLC). Among the SLC families, OAT1(SLC22A6), OCT2(SLC22A2) and OAT3(SLC22A8) are the most important transporters in kidney tissue, often acting as the limiting factor for solute clearance. As a result, the FDA and EMA recommend these transport pumps as targets for drug-drug interaction studies. Thus, there is a requirement for an in vitro model that has human kidney origination, accurate clinical predictability and consistent data output.
Unfortunately, primary RPTEC cells lose OAT1, OCT2 and OAT3 transporter expression in culture. Transiently expressing these transporters in primary RPTEC cells also shows large variations between production lots. In our study, we have generated transporter cell models using a well characterized hTERT-immortalized Renal Proximal Tubule Epithelial Cells that stably overexpress the OAT1, OCT2 or OAT3 gene. Clones show typical epithelial morphology and correct marker expression. Most importantly, we verified that the overexpressed transporters have normal transport activities using 5-CF (5Carboxyfluorescein) and ASP+(4-(4-(dimethylamino)styryl)-N-methylpyridinium iodide) uptake assays and the uptake can be inhibited by well know inhibitors in a dose dependent manner. Overall, our data has shown that these modified cell lines are very useful tools which provide human kidney tissue related results, improved consistency over time, and have more predictability for clinical trials versus current models for in vitro testing.