Humanized Mouse Tox Model
Current animal models are often poorly predictive of ADMET in man, noted Mike Piper, Ph.D., senior business development manager at CXR Biosciences. “This is driven by profound interspecies differences in the expression levels and functions of proteins involved in ADMET.” It’s also a major reason for development failure in the pharmaceutical and chemical industries.
CXR and TaconicArtemis have developed transADMET mice panels in which key murine ADMET genes are knocked out and their human counterparts inserted. Dr. Piper discussed three such panels—Cytochrome P450, Nuclear Receptor, and Drug Transporter—each representing critical pathways in compound metabolism, disposition, and safety.
For example, Cytochrome P450-dependent monooxygenases are a group of enzymes that account for the Phase I metabolism of the majority of drugs. Together, CYP3A4 and CYP2D6 catalyze the metabolism of over 60% of drugs in clinical use, according to Dr. Piper who noted these enzymes have diverged significantly between species, both in their multiplicity and substrate specificity. This divergence can cause altered drug metabolism profiles between animals and humans, leading to differences in pharmacokinetics, efficacy, and toxicity.
“We have derived and sourced a series of humanized and knockout CYP3A4 and CYP2D6 mouse models,” said Dr. Piper. “Depending on your needs, we can offer transADMET models where CYP450 expression is under control of the human promoter, or models with gut- or liver-specific CYP3A4 expression.”
The situation is much the same for nuclear receptor panels. Nuclear hormone receptors play a major role in regulating the body’s response to chemical exposure. The Pregnane X receptor (PXR) and Constitutive androstane receptor (CAR) have the ability to bind a wide range of exogenous and endogenous ligands and to control the expression of genes highly relevant to compound metabolism such as the cytochrome P450s and drug transporters.
However, sequence variation in PXR and CAR between animals and humans results in differences in the ability of exogenous ligands to interact and activate these transcription factors. Experimental data obtained in traditional animal models or in vitro test systems, may therefore, not reflect the interactions that occur in man, noted Dr. Piper.
Rule-Based Predictive Modeling
Functional genomics has yet to deliver on its promise for predictive ADME/Tox said Quin Wills, M.D., CSO at SimuGen. This is largely because of the difficulty researchers have adequately interpreting results, especially multivariate results, he contended.
“The problem isn’t the data, it’s the models.” Physicists have remarkably accurate models constructed from first principles but biologists don’t. What’s needed, said Dr. Wills, are flexible modeling platforms that can incorporate prior knowledge (learned biology) by using Bayesian statistical techniques and also incorporate rules (e.g., weights and thresholds) created by researchers.
SimuGen offers high-throughput screening analysis software and HT Stream, as well as various services. One of HT Stream’s strengths, said Dr. Wills, is its ability to easily create rule-based models for use when analyzing assay results.
Suppose you are screening compounds and looking for late-stage apoptosis by identifying Caspase overexpression and nuclear morphology. “You could create a rule that says when Caspase9 is expressed 2.5 times normal and when there is 25 percent nuclear shrinkage, call late-stage apoptosis. Then the software takes over and uses that rule and tries to find the lowest dosage at which the rule starts becoming true.”
A common criticism of modeling software is that it’s difficult to use for non-mathematicians. Dr. Wills agrees but he insists that “the HT Stream has been structured in a way that anybody can use it. It’s a very simple workflow.”
“The only restriction is that for every chemical you test, there must be at least six concentrations done in triplicate, so at least 18 data points. If you are measuring the level of expression of a gene, you need 18 data points to allow us to do rigorous automated quality control. Once you start working with less data it’s very difficult to do that automatically.”
Global vs. Local QSAR
Johann Gasteiger, Ph.D., founder of Molecular Networks, presented an overview of various computational approaches to predictive ADME/Tox based on the company’s extensive cheminformatics suite.
In discussing modeling chemical toxicity, he noted that QSAR (quantitative structure activity relationship) efforts in toxicology tend to focus on a class of compounds. However, compounds in the same class often have very different toxic modes of action. For example, he presented data on phenols indicating a diversity of mechanism of action including polar narcotics, uncouplers of oxidative phosphorylation, precursors to soft electrophiles, and soft electrophiles themselves.
For this reason, he said, using global QSAR models is a poor idea. It’s necessary to first classify chemicals according to their toxic mechanism of action and then derive local QSAR models. Doing this greatly enhances the predictability of the models.