New tools helped reduce attrition due to poor ADME profiles, and now researchers hope to do the same screening for toxicology.

Part two in a three-part series on advances in ADME-Tox

Patrick McGee Senior Editor

It used to be that the number of drugs failing preclinically due to poor pharmacokinetics hovered at about 40%, but in recent years, that has dropped to about 10%. The reason? Improved in vitro and animal models played a key role, as did more extensive use of tools like mass spectrometry and liquid chromatography/mass spectrometry (LC/MS).

"These allowed for the analysis of trace levels of compounds to be analyzed in biofluids and small animal pharmacological models, so basically you could screen out the pharmacokinetic and pharmacodynamic bad actors in mouse and rat models, which prior to that technology was very difficult to do. So now we are capable of having far more effective in vitro and in vivo screening methods,” says Russell Robins, director of pharmacokinetics, dynamics and metabolism at Pfizer Inc., St. Louis.

click the image to enlarge The results of a computational model developed by William Egan, PhD, used to determine optimal ADME properties by computing polar surface area and hydrophobicity. Because the results formed an ellipse, it came to be called the Egan Egg. (Source: Novartis Institutes for Biomedical Research)

Over the years, in vitro methods have improved and been buttressed by the emergence of newer technologies such as in silico modeling and other modeling tools. But while researchers in drug discovery and development have been able to improve ADME (absorption, distribution, metabolism, and excretion) profiles, toxicology remains a stubborn problem, as evidenced by the fact that failures due to toxicology are in the 30% to 40% range, making it the number one reason for preclinical attrition.

"Predicting efficacy of molecules continues to be a very, very tough thing, but our goal really is to focus on major reasons for these things dying in clinical trials and to try to improve that. Toxicity is a huge reason for things falling out in the clinic, and the more we can do experimentally and computationally to lower that fallout rate, the better off we’ll be,” says John Van Drie, PhD, director of computer-aided drug discovery at the Novartis Institutes for Biomedical Research Inc., Cambridge, Mass.

Automated ADME-Tox
One tool that could help in this effort is LeadStream, a turnkey ADME-Tox system launched earlier this year by Thermo Electron Corp., Waltham, Mass. LeadStream combines three integrated instrumentation modules: the WorkCell, a fully automated, modular platform that conducts a variety of ADME-Tox assays; the Reformatter, which provides online preparation of plates for WorkCell; and an LC/MS system. LeadStream also includes Orchestrator, a piece of software that manages the flow of samples through the laboratory. In designing the system, Thermo Electron sought to automate ADME-Tox processes in the same way that other processes such as high-throughput screening had been automated.

Robins’ labs evaluated LeadStream in collaboration with Thermo Electron because they were interested in seeing how it would work relative to their standard automation and semi-manual methods. Three screens were evaluated: metabolic stability, permeability, and drug-drug interaction. "They mounted those on the LeadStream technology and performed the screens over a statistically relevant set of compounds which we compared to our historical database from our own internal methods, which are less automated. We compared the results statistically and they compared favorably,” Robins says.

Pfizer is still evaluating the system and has yet to decide whether it will purchase one, but Robins says two advantages of the system are high-level vertical integration and process coordination. "It manages sample material flow, sample preparation, sample analysis, and then the out-streaming of that data to a workable database or at any global database that you may have.” Because it is higher throughput, it requires less man hours, so researchers can spend more of their time thinking about what the data means rather than generating the data. Also, because it is not built for one screen, it can be used in a variety of ways.

Cultured Hepatocyte System
Another offering in ADME-Tox, albeit on a smaller scale, is the B-Clear product line from Qualyst Inc., Research Triangle Park, N.C. B-Clear is an in vitro sandwich-cultured hepatocyte system for the assessment of hepatobiliary disposition that offers the potential to study biliary transport of drugs in vitro. Qualyst touts it as the only model of its type available to industry. "Currently, there are no good systems that mimic the integrated transport of drugs by multiple transporters in a cell system. Expressed cell lines or cell-free systems are limited to the study of only a single transporter at a time,” says Gerald Miwa, PhD, vice president of drug metabolism and pharmacokinetics at Millennium Pharmaceuticals, Cambridge, Mass.

"B-Clear also has the potential for automation in a high-throughput format. The system may be very useful during drug discovery in identifying and screening out compounds that may be cleared too rapidly by biliary excretion or in identifying the transporters responsible for biliary excretion. We have just begun an evaluation of the potential value and technical limitations of the B-Clear system,” says Miwa, who is also on Qualyst’s scientific advisory board.

Hurel Corp., Beverly Hills, Calif., has also introduced a new in vitro tool, a microfluidic biochip with separate but fluidically interconnected "organ” or "tissue” compartments. Each compartment contains a culture of living cells drawn from or engineered to mimic the function or functions of the respective organ or tissue of a living animal. The Hurel system has microfluidic channels between the compartments that permit a culture medium that serves as a blood surrogate to circulate. Drug candidates are added to the medium and are distributed to the cells in the compartments. The effects of the compounds are detected by measuring or monitoring physiological events such as cell death, differentiation, or disturbances in metabolism or signal transduction pathways.

An LC/MS unit is one of three integrated instrumentation modules comprising LeadStream, an automated turnkey ADME-Tox system launched earlier this year. (Source: Thermo Electron Corp.)

Computational Models
While tools are available to generate data, the question is what to do with that data once it has been gathered. Miwa’s lab is trying to develop a method that takes in vitro and other data and combines it to create a physiologically based pharmacokinetic model to predict drug-drug interactions. They have been collaborating with Rene Levy, PhD, professor and chair of pharmaceutics at the University of Washington, Seattle. Levy conducted research on the molecular mechanisms of inhibition interactions based on identification of cytochrome P450 isozymes catalyzing the metabolism of old and new antiepileptic drugs, and he created a database comprising clinical and in vitro data on drug interactions collected from the literature.

"He [Levy] has a database with real drugs,” says Miwa. "We have in vitro enzyme systems and the ability to take those existing drugs and do some in vitro studies. We also have the computational modeling abilities internally to be able to integrate this information and predict what the clinical pharmacokinetic behavior will be in human subjects.” They have used the database to look specifically for cytochrome P450 3A4-specific inhibition studies. As the most abundant human cytochrome P450, 3A4 oxidizes a broad spectrum of drugs using a number of metabolic processes.

They found about a dozen compounds for which information on 3A4 inhibition was available and which also met Levy’s quality criteria. "We did our modeling from only the in vitro data, and then we projected what the clinical results would be. He came back in with the clinical results to say ‘Yes, you’re right,’ or No, you’re wrong.’….Right now, out of eight compounds for which we really have complete data sets, the model is predicting probably very accurately six.” Miwa says that it will take a few more cycles to refine the model.

Jump Starting ADME-Tox Studies Although new tools have lowered the rate of compounds failing preclinically due to poor ADME profiles, the same cannot be said for eliminating compounds due to toxicity. "In vitro toxicology has remained elusive. It’s one thing to do the Ames test or micronucleus or apoptosis or mitochondrial function, all of the monolithic toxicity tests that have been around for a while,” says Russell Robins, director of pharmacokinetics, dynamics and metabolism at Pfizer Inc., St. Louis. "But how is the compound going to impact an organism? How do you translate that into whole-body toxicity? That remains elusive.” John Van Drie, PhD, director of computer-aided drug discovery at the Novartis Institutes for Biomedical Research Inc., Cambridge, Mass., remembers trying to convince a toxicologist to try modeling toxicity in the mid-1990s when he worked at Pharmacia. "He said to me, ‘John, to a toxicologist, all molecules are toxic,’ which is an accurate statement, but it’s not really that helpful….I think the toxicologists weren’t really eager to embrace the potential for in silico modeling until really very recently.” But that has changed, Van Drie adds. One factor is the US Food and Drug Administration (FDA), which has gotten involved in in silico toxicology modeling and has proposed rules on it. "Any time the FDA talks, people listen,” says Van Drie. Toxicology could use a spark along the lines of that provided for ADME when Pfizer’s Christopher Lipinski, PhD, outlined his "rule of five” in a landmark paper in 1997. "Lipinski simplified it, made it really accessible to the masses, and really drew people’s attention to it. There hasn’t been that similar sort of event in toxicity.”

In Silico Modeling
While improved in vitro and other technologies are enhancing research abilities, many see the improvements as incremental and look instead to in silico modeling. Robins, for one, believes virtual screening is the "Holy Grail” of the screening business. "You could actually look at the genome or a proteome and start playing with the chemistry in silico before you commit chemists and the industrial machine to test it out. But, I don’t think we’re there yet,” says Robins.

"For one thing, we’re pretty good at local modeling, which means you still have to make the chemistry, test it, and build the model. We’re not so good at global modeling yet, but virtual screening would be quite a boon, because then almost everything that you did take the time to synthesize and put ahead would at least be a virtual winner. You wouldn’t be spending 99% of your chemistry time on things that don’t advance.” Chyung Cook, PhD, senior research scientist at Baxter Healthcare, Deerfield, Ill, says that while in silico modeling is very promising, it has several shortcomings, including its accuracy. For example, even if the prediction rate for effective compounds is 80% accurate, the 20% eliminated could include good candidates.

"In general, what the vendors provide falls short of what we need,” says Van Drie. "We are devoting a fair bit of effort to either adapting things that have been published in the literature to our own internally developed tools, or doing our own work to develop our own models that we then disseminate to people on drug discovery teams.” William Egan, PhD, who works with Van Drie on ADME-Tox modeling, developed a computational model several years ago to determine optimal ADME properties by computing polar surface area and hydrophobicity. When shown on a graph, the results for optimal properties formed an ellipse which came to be called the Egan Egg.

Egan believes that good in silico tools help with the compound design cycle because they speed up the process. "You can do 10 iterations very quickly if you have an idea that looks decent, and then the chemist can get to work making it so they can test it. One of the most important things about computational models is helping you propose a hypothesis, because that’s really what every molecule is.”

In response to research needs, vendors continue to develop new modeling tools. In July, the BioSciences Group of Fujitsu Computer Systems, Westwood, Mass., unveiled a new technique for generating enhanced predictions for ADME/Tox research. It consists of a docking-based approach combined with off-the-shelf and custom-built technologies and is designed to develop viable and effective predictive models.

The technique consists of a computational workflow designed to build metabolite models, identify and model active sites, and then run the results through a high-throughput docking algorithm. This then docks the database of metabolites predicted against all cytochrome P450 varieties, resulting in a metric for relative distances of docked configurations that help determine how much a compound and its derivative classes interact with differing cytochrome P450 models.

Last year, Bio-Rad Laboratories’ informatics division in Philadelphia drew a great deal of attention with its KnowItAll ADME/Tox Consensus Modeling Environment. The system is a suite of tools for the computer-based prediction of a potential drug’s ADME/Tox profile, including a large collection of predictive models, applications to build and validate predictive models, experimental ADME/Tox data, and other tools. The company has released new databases that contain measured ADME/Tox properties including bioavailability, carcinogenic potency, and toxicity.

Cook says more and more work is being done to create integrated methods combining preclinical in vitro, in vivo, and in silico data to predict human pharmacokinetics and exposure. GastroPlus from Simulations Plus Inc., Lancaster, Calif., does just that. While this tool is not perfect, Cook says, it is a step in the right direction. "It’s coming slowly, but again, human bodies are more complex than simple computer models, so I don’t know how far we can go. It remains to be seen.”