Chemistry-Based Modeling on Phenotype-Based Drug-Induced Liver Injury Annotation: From Public to Proprietary Data.

Drug-induced liver injury (DILI) is an important safety concern and a major reason to remove a drug from the market. Advancements in recent machine learning methods have led to a wide range of in silico models for DILI predictive methods based on molecule chemical structures (fingerprints). Existing publicly available DILI data sets used for model building are based on the interpretation of drug labels or patient case reports, resulting in a typical binary clinical DILI annotation. We developed a novel phenotype-based annotation to process hepatotoxicity information extracted from repeated dose in vivo preclinical toxicology studies using INHAND annotation to provide a more informative and reliable data set for machine learning algorithms. This work resulted in a data set of 430 unique compounds covering diverse liver pathology findings which were utilized to develop multiple DILI prediction models trained on the publicly available data (TG-GATEs) using the compound's fingerprint. We demonstrate that the TG-GATEs compounds DILI labels can be predicted well and how the differences between TG-GATEs and the external test compounds (Johnson & Johnson) impact the model generalization performance.

Leveraging Cell Painting Images to Expand the Applicability Domain and Actively Improve Deep Learning Quantitative Structure-Activity Relationship Models.

The search for chemical hit material is a lengthy and increasingly expensive drug discovery process. To improve it, ligand-based quantitative structure-activity relationship models have been broadly applied to optimize primary and secondary compound properties. Although these models can be deployed as early as the stage of molecule design, they have a limited applicability domain─if the structures of interest differ substantially from the chemical space on which the model was trained, a reliable prediction will not be possible. Image-informed ligand-based models partly solve this shortcoming by focusing on the phenotype of a cell caused by small molecules, rather than on their structure. While this enables chemical diversity expansion, it limits the application to compounds physically available and imaged. Here, we employ an active learning approach to capitalize on both of these methods' strengths and boost the model performance of a mitochondrial toxicity assay (Glu/Gal). Specifically, we used a phenotypic Cell Painting screen to build a chemistry-independent model and adopted the results as the main factor in selecting compounds for experimental testing. With the additional Glu/Gal annotation for selected compounds we were able to dramatically improve the chemistry-informed ligand-based model with respect to the increased recognition of compounds from a 10% broader chemical space.

Test systems in drug discovery for hazard identification and risk assessment of human drug-induced liver injury.

INTRODUCTION: The liver is an important target for drug-induced toxicities. Early detection of hepatotoxic drugs requires use of well-characterized test systems, yet current knowledge, gaps and limitations of tests employed remains an important issue for drug development. Areas Covered: The current state of the science, understanding and application of test systems in use for the detection of drug-induced cytotoxicity, mitochondrial toxicity, cholestasis and inflammation is summarized. The test systems highlighted herein cover mostly in vitro and some in vivo models and endpoint measurements used in the assessment of small molecule toxic liabilities. Opportunities for research efforts in areas necessitating the development of specific tests and improved mechanistic understanding are highlighted. Expert Opinion: Use of in vitro test systems for safety optimization will remain a core activity in drug discovery. Substantial inroads have been made with a number of assays established for human Drug-induced Liver Injury. There nevertheless remain significant gaps with a need for improved in vitro tools and novel tests to address specific mechanisms of human Drug-Induced Liver Injury. Progress in these areas will necessitate not only models fit for application, but also mechanistic understanding of how chemical insult on the liver occurs in order to identify translational and quantifiable readouts for decision-making.

Are zebrafish larvae suitable for assessing the hepatotoxicity potential of drug candidates?

Drug-induced liver injury (DILI) is poorly predicted by single-cell-based assays, probably because of the lack of physiological interactions with other cells within the liver. An intact whole liver system such as one present in zebrafish larvae could provide added value in a screening strategy for DILI; however, the possible occurrence of other organ toxicities and the immature larval stage of the zebrafish might complicate accurate and fast analysis. We investigated whether expression analysis of liver-specific fatty acid binding protein 10a (lfabp10a) was an appropriate endpoint for assessing hepatotoxic effects in zebrafish larvae. It was found that expression analysis of lfabp10a was a valid marker, as after treatment with hepatotoxicants, dose-response curves could be obtained and statistically significant abnormal lfabp10 expression levels correlated with hepatocellular histopathological changes in the liver. However, toxicity in other vital organs such as the heart could impact liver outgrowth and thus had to be assessed concurrently. Whether zebrafish larvae were suitable for assessing human relevant drug-induced hepatotoxicity was assessed with hepatotoxicants and non-hepatotoxicants that have been marketed for human use and classified according to their mechanism of toxicity. The zebrafish larva showed promising predictivity towards a number of mechanisms and was capable of distinguishing between hepatotoxic and non-hepatotoxic chemical analogues, thus implying its applicability as a potential screening model for DILI.

Phospholipidosis in rats treated with amiodarone: serum biochemistry and whole genome micro-array analysis supporting the lipid traffic jam hypothesis and the subsequent rise of the biomarker BMP.

To provide mechanistic insight in the induction of phospholipidosis and the appearance of the proposed biomarker di-docosahexaenoyl (C22:6)-bis(monoacylglycerol) phosphate (BMP), rats were treated with 150 mg/kg amiodarone for 12 consecutive days and analyzed at three different time points (day 4, 9, and 12). Biochemical analysis of the serum revealed a significant increase in cholesterol and phospholipids at the three time points. Bio-analysis on the serum and urine detected a time-dependent increase in BMP, as high as 10-fold compared to vehicle-treated animals on day 12. Paralleling these increases, micro-array analysis on the liver of treated rats identified cholesterol biosynthesis and glycerophospholipid metabolism as highly modulated pathways. This modulation indicates that during phospholipidosis-induction interactions take place between the cationic amphiphilic drug and phospholipids at the level of BMP-rich internal membranes of endosomes, impeding cholesterol sorting and leading to an accumulation of internal membranes, converting into multilamellar bodies. This process shows analogy to Niemann-Pick disease type C (NPC). Whereas the NPC-induced lipid traffic jam is situated at the cholesterol sorting proteins NPC1 and NPC2, the amiodarone-induced traffic jam is thought to be located at the BMP level, demonstrating its role in the mechanism of phospholipidosis-induction and its significance for use as a biomarker.

Comparisons between in vitro whole cell imaging and in vivo zebrafish-based approaches for identifying potential human hepatotoxicants earlier in pharmaceutical development.

Drug-induced liver injury (DILI) is a major cause of attrition during both the early and later stages of the drug development and marketing process. Reducing or eliminating drug-induced severe liver injury, especially those that lead to liver transplants or death, would be tremendously beneficial for patients. Therefore, developing new pharmaceuticals that have the highest margins and attributes of hepatic safety would be a great accomplishment. Given the current low productivity of pharmaceutical companies and the high costs of bringing new medicines to market, any early screening assay(s) to identify and eliminate pharmaceuticals with the potential to cause severe liver injury in humans would be of economic value as well. The present review discusses the background, proof-of-concept, and validation studies associated with high-content screening (HCS) by two major pharmaceutical companies (Pfizer Inc and Jansen Pharmaceutical Companies of Johnson & Johnson) for detecting compounds with the potential to cause human DILI. These HCS assays use fluorescent-based markers of cell injury in either human hepatocytes or HepG2 cells. In collaboration with Evotec, an independent contract lab, these two companies also independently evaluated larval zebrafish as an early-stage in vivo screen for hepatotoxicity in independently conducted, blinded assessments. Details about this model species, the need for bioanalysis, and, specifically, the outcome of the phenotypic-based zebrafish screens are presented. Comparing outcomes in zebrafish against both HCS assays suggests an enhanced detection for hepatotoxicants of most DILI concern when used in combination with each other, based on the U.S. Food and Drug Administration DILI classification list.

Zebrafish developmental toxicity assay: A fishy solution to reproductive toxicity screening, or just a red herring?

The zebrafish embryotoxicity/teratogenicity assay is described as a useful alternative screening model to evaluate the effect of drugs on embryofoetal development. Fertilized eggs were exposed to different concentrations of 15 compounds with teratogenic (8) and non-teratogenic (7) potential until 96h post-fertilization when 28 morphological endpoints and the level of compound uptake was assessed. The majority of drugs testing positive in mammals was also positive in zebrafish (75% sensitivity), while a relative high number of false positives were noted (43% specificity). Compound uptake determination appears useful for clarifying classifications as teratogenic or potential overdose although assay sensitivity could be improved to 71% if the exposure threshold, previously suggested as ∼50ng/larvae, is reconsidered. The zebrafish assay shows some potential, though limited in its current form, as a screening tool for developmental toxicity within Janssen drug development. Further assay refinement with respect to endpoints and body burden threshold is required.

Screening for phospholipidosis induced by central nervous drugs: comparing the predictivity of an in vitro assay to high throughput in silico assays.

Drug-induced phospholipidosis is a side effect for which drug candidates can be screened in the drug discovery phase. The numerous in silico models that have been developed as a first line of screening are based on the characteristic physicochemical properties of phospholipidosis-inducing drugs, e.g. high logP and pK(b) values. However, applying these models on a predominantly high lipophilic, basic CNS chemistry results in a high false positive rate and consequently in a wrong classification of a large number of valuable drug candidates. Here, we tested 33 CNS-compounds (24 in vivo negative and 9 in vivo positive phospholipidosis-inducers) in our in house developed in vitro phospholipidosis screening assay (Mesens et al., 2009) and compared its predictivity with the outcome of three different, well established in silico prediction models. Our in vitro assay demonstrates an increased specificity of 79% over the in silico models (29%). Moreover, by considering the proposed plasma concentration at the efficacious dose we can show a clear correlation between the in vitro and in vivo occurrence of phospholipidosis, improving the specificity of prediction to 96%. Through its high predictive value, the in vitro low throughput assay is thus preferred above high throughput in silico assays, characterized by a high false positive rate.

Evaluation of the embryotoxic potency of compounds in a newly revised high throughput embryonic stem cell test.

The ability of murine-derived embryonic stem cells (D3) to differentiate into cardiomyocytes is the basis of the embryonic stem cell test (EST). With the EST, chemicals and pharmaceuticals can be assessed for their embryotoxic potency early on in the development process. In order to come to a higher throughput EST, a 96-well based method was developed based on low attachment well plates that allow for the formation of embryonic bodies from which the stem cells can differentiate. Twelve test compounds were selected based on their reported in vitro and in vivo embryotoxic potency. In the 96-well based EST, reportedly strong embryotoxic compounds 5-fluorouracil, 6-aminonicotinamide (6AN), methylmercury chloride, and hydroxyurea were correctly ranked with corresponding Relative Embryotoxic Potency values (REP, based on the EC(50) (microM) value of 6AN) of 2.6 +/- 2.9, 1, 2.0 +/- 3.1, and 0.07 +/- 0.05, respectively. Moderately embryotoxic compounds valproic acid, boric acid, methoxyacetic acid, and lithium chloride resulted in a correct ranking with REP values of 0.01 +/- 0.003, 0.001 +/- 0.001, 0.0007 +/- 0.001, and 0.0006 +/- 0.0004, respectively. The included nonembryotoxic compounds Penicillin G, acrylamide, and saccharin did not result in an inhibition of D3 cells to differentiate into cardiomyocytes, other than related to cytotoxicity (REP value of 0.00001). However, diphenhydramine resulted in an inhibitory effect similarly to the strong embryotoxic compound hydroxyurea, with a REP value of 0.40 +/- 0.36. However, further evaluation suggested this was due to direct inhibition of the contractile capacity of the D3 cardiomyocytes, rather than an embryotoxic mechanism. The 96-well based EST is a promising addition to the screening process of newly developed chemicals and pharmaceuticals.