December 1, 2007 (Vol. 27, No. 21)
Assays, Computer Models, and Databases Improve Prediction of Drug Safety
Speakers at Select Biosciences “Virtual Discovery Europe” conference held in London in late October discussed some of the key questions and challenges of the ADMET field. Pharma is struggling right now with the fact that the drug pipeline paradigm continues to shift. Finding safe drugs faster and earlier runs up against the complication of compound toxicity and/or metabolism.
“The problem with metabolism, particularly in the assessment of toxicity and genotoxicity, is its unpredictability across models and testing strategies,” said Richard Walmsley, Ph.D., CSO at Gentronix (www.gentronix.co.uk).
“Metabolism is a big target, and a moving one that is not consistent across species, or even people, but just because it’s difficult doesn’t mean that you don’t try to hit it. It’s important to tackle these questions to facilitate and speed up the discovery of new drugs,” added Mario Lobell, computational chemist with Bayer Healthcare (www.bayerhealthcare.com).
“Metabolism is complex, and the industry is a long way from the luxury of internationally agreed upon toxicity and genotoxicity tests, either in vivo or in vitro, that both sensitively and specifically predict the toxicity of drugs or their metabolites to humans,” noted Dr. Walmsley. “Despite the progress in worldwide efforts to improve toxicity screening, we still see drugs failing in clinical trials.”
“The in silico models alert us to theoretical hazards, but the downside is that this is only suggestive of a biological risk,” said Dr. Walmsley. “A good in silico tool will be able to tell you whether you are working with a chemistry in which in vitro testing should be carried out during lead optimization.” If the hazard is confirmed in vitro there is the opportunity to select from related chemistries without the problem.
In Vitro Issues
Of course in vitro results are supposed to predict in vivo genotoxicity. Dr. Walmsley, however, cited several recent papers showing that the current in vitro mammalian cell tests too often suggest a hazard that is not confirmed in vivo.
“With an unacceptably low level of specificity, some have asked whether it is even worth doing some of these in vitro genotoxicity studies. Without more accurate in vitro mammalian tests, it seems that animal testing will continue to be a major part of hazard assessment rather than safety assurance.”
The key issue at hand, Dr. Walmsley noted, is the need to predict genotoxicity in humans not just rodents and, by extension, the need to predict human metabolism not just rodent metabolism. “For genotoxicity, we need better predictive tests, since the existing in vitro tests really don’t give us enough information about genotoxicity, general toxicity, or the toxicity of metabolites,” said Dr. Walmsley. “Right now expert panels are looking at methods to improve the accuracy of existing in vitro tests.
“Gentronix meanwhile has developed what appears to be a better in vitro replacement test, GreenScreen® HC.” The company’s GreenScreen line of products are cell-based assays for use in genetic toxicity screening using high-throughput protocols. The assay system is designed for use much earlier in drug discovery than the standard test battery for genotoxicity, providing an accurate preview of these late-stage regulatory tests, reported Dr. Walmsley. One of the protocols for GreenScreen HC that the company recently developed allows the assessment of metabolites generated by S9 liver extracts.
“While in silico models are less expensive and have the potential to guide in vitro testing strategies, at present their productivity is not sufficient to replace in vitro tests,” concluded Dr. Walmsley. “Therefore the focus for the present will remain on in vitro tests, and improvements here should eventually lead to an era where an in vivo test simply confirms safety in the animal.”
“In recent history, computational chemistry and in silico ADMET modeling has been a hot topic in drug development, and our focus has been to expand on in silico research,” said Lobell.
Most of Bayer Healthcare’s in silico ADMET methods have been developed in-house, which has been advantageous to its drug development on multiple levels, reported Lobell. “The consistency of working closely with and collaborating with scientists in-house guarantees consistent results. The scientists are familiar with the methodology and the endpoints we are working with, and in the end, working with in-house data tends to generate consistency.”
Lobell presented his ADMET Traffic Lights system for in silico ADMET characterization of HTS hit structures and its application to kinase inhibitors. “The biggest challenge in the in silico field is making an accurate model,” he said. “And for that, it’s not enough to generate data; you need to be able to access that data and make sense of it.”
The system is one that Lobell invented, developed, and helped implement at Bayer. “The primary purpose of the in silico ADMET Traffic Lights is not that it’s an automatic filter but rather an aid to help MedChem experts analyze and prioritize HTS hit sets with regard to potential ADMET-related liabilities.
“We have built our scoring system on ADMET-relevant parameters, which are well known to medicinal chemists and can be addressed by structural modifications. Also the scoring system is simple. The generated output is easy to interpret due to the intuitive traffic light color scheme and a scoring system in the convenient range from zero to ten,” said Lobell.
Genkyotex (www.genkyotex.com) uses multiple computer models in order to develop drugs that block enzymes that produce oxygen radicals.
“We are currently optimizing three chemical series for hypertension, myocardial ischemia/reperfusion, and lung fibrosis,” said Cedric Merlot, head of knowledge management and computational chemistry. “Initially we plan to partner one program, and we hope that we will take the other two to approval ourselves.”
The company’s seed funds are limited, thus resources cannot be wasted on synthesizing and testing compounds that will obviously not make it in ADME and safety. Early-stage optimization is vital, Merlot added. “We take full advantage of ADME prediction and computational toxicology to make sure that the drugs we are testing in vivo are worth it. We have access to in silico ADME profiles, which allows us to focus on compounds that have a greater chance to be bioavailable. Then, and most importantly, we are using computational toxicology to identify potential liabilities.”
Merlot’s team tries to make the best use of computers for supporting its projects. “We have a partner that helped us a lot. Drug Design Technologies (www.drugdesigntech.com) provided us with a corporate database and efficient prediction tools for early ADME and safety assessment,” said Merlot.
“Historically, in silico models have been used to filter out compounds that were predicted to be bad, sometimes filtering too much and resulting in the loss of valuable leads,” explained Merlot. “What we are doing is using them for asking the right question.”
His group runs a large panel of models on each of their chemical series. Positive results are then followed up with in vitro assays to confirm the prediction. Then, this information is interpreted in the context of a project. “Knowing the therapeutic area or the disease, we can say whether this liability is a safety concern or not,” said Merlot. “It increases our confidence in the chemical series.”
Current prediction models are worth using in drug discovery, concluded Merlot. “Probably no model is better than the others; it is just about finding the one(s) that will answer the question.”
Maximizing Current Knowledge
Aureus Pharma (www.aureus-pharma.com) develops knowledge databases that integrate structured information with search engines. The company’s knowledge management platform investigates the ADME properties of drugs. In conjunction with ADMET computational tools, Aureus has successfully designed in silico models for the early identification of unwanted side activities.
One of the biggest challenges in this field is selecting the right data, said Elodie Dubus, application scientist. “When people are looking to extract data, it’s sometimes difficult to know the differences in parameters.” Aureus employed this structured information to generate data sets that use cheminformatics to predict cytochrome inhibition or hERG blockage and then successfully built in silico models as a result, reported Dubus.
Its fully annotated, structured knowledge database contains all the pertinent biological and chemical information on the metabolic properties of drugs, added Mary Donlan, director of marketing. “We mine the literature for drug target and ADME information, analyzing articles, patents, and FDA reports,” she explained. “Then we bring the knowledge into a relational database. Our focus with regard to ADME is to gather data about drug-drug interactions into a searchable database. When you have such a database, it’s possible to drill down and select consistent data to build a predictive model and also identify potential drug-drug interactions.”
Parameters can be set for searches depending on the development stage, Donlan noted. “If you know that ADME is a consideration, then that can be used at the beginning of the drug discovery process. Establishing safety first saves money and time in the long run.”