By Pauline Gee, PhD
Vice President, Predictive Biology MDS Pharma Services
Chair, SBS ADMET Special Interest Group

Those of us in the industry, and particularly members of the SBS ADMET Special Interest Group (ADMET SIG), are aware that screening for adsorption, distribution, metabolism, excretion, and toxicology (ADMET) presents challenges that are at least an order of magnitude more complex than those found in screening for hits and even validation of hits at the secondary screening stage. And yet, we must strive to achieve success in this area.

The costs of drug candidates that fail in later stages of development are rapidly overtaking the costs that go into developing a successful drug, so it is easy to see why failing early is the only logical strategy. Nevertheless, still, we are optimists. We ask ourselves, at each step, "what if we have inadvertently screened out a blockbuster?" This is because, at each stage, the exceptions to accepted dogma loom large in our minds.

How do we resolve such a conflict in developing methods to screen out candidates for undesirable ADMET properties or characteristics?

If Only....
An obvious solution is develop high-throughput (HT) ADMET assays or test systems that are predictive. If only you knew at the outset what you know later in development, or after a drug has been on the market for several years, you could have, or would have, or should have chosen this instead of that. How do we increase the probabilities of choosing the best candidate going forward? It is easy to look at the data retrospectively, and see all of the pieces that were consistent with the outcome.

A more rational strategy to maximize the probabilities of success is to devise systems predicated on the mechanisms of actions of a compound in a cell and in the whole body. If one has a good understanding of how the cell, the organ, the animal, and the human are responding to a compound, then one can devise a strategy to detect and measure such responses. This is obvious, but by no means is it a trivial matter to execute.

Retrospectively, one can devise early screening assays to detect and quantify ADMET on compounds that have been well characterized chemically, in traditional screening assays, in animal models, in the clinic, and in the global human population-if the compound is a drug that has been on the market for several years. The data gathered from the entire continuum of development- from discovery to exposure in humans representing myriad genetic variations-provide context and reality calibrations for early screening assays.

Therefore, empirical results from many compounds with diverse chemical properties gathered into a relational, contextual database may prove useful in calibrating "reality" for early ADMET screening strategies. In addition to the routine (and required) results, which may reveal a lot about the mechanisms of action, the maturation of microarray technologies for measuring the transcriptome-and the growing development of detection and quantification of the proteome, metabolome, and SNP genotyping of individuals-provide quantitation to the biochemical pathways recruited by the mechanisms of actions.

Identifying Biomarkers
Along with the maturation of microarray platforms, numerical analyses of these data have also yielded some agreement among biomathematicians for the distillation and production of practical outcomes that identify parameters that have the potential to be biomarkers or sets of biomarkers. Such renderings of millions of data points have produced a couple of hundred biomarkers or endpoints that can be used in HT on a microarray, and fewer than 50 endpoints that can be measured singly or multiplexed in cell-free or cell-based assays. Results from these early screening assays can then be interpreted in the context of datasets available from later development, and most importantly from responses in humans.

How early and how useful it is to screen for ADMET depends on how reliable and predictable the assays are in screening novel and new chemical entities versus known and well-characterized compounds. Furthermore, early ADMET screening strategies necessitate a higher purity in the libraries of compounds to be screened. Hit-finding strategies depend on the specificity of targets to ameliorate compound impurities; however, these impurities may overwhelm any true screening results for ADMET by either masking an adverse effect through quenching interactions or producing an adverse effect that is not present in the true, purified active compound.

When is an Assay Predictive?
How does one define whether an assay is predictive? Are truth tables useful in these evaluations or is it necessary to redefine the metrics when describing the probabilities of sets of biomarker assays? Do we need to rethink how to use existing assays? Do the existing assays need to be re-evaluated or calibrated against a different reference dataset? Can we calibrate in vitro screening assays with in vivo genomics and proteomics data? Within the walls of industry, non-profit, and academic institutions alike, there are treasures in terms of good assays and balanced robust databases. But, isolated as silos, do they represent healthy innovation and competition or would there be even greater value to these treasures if they were shared or complemented with other systems?

Overall, changes are needed to improve our probabilities of success. This was recognized by the Food and Drug Administration with the release of its report calling for industry to use new technologies to improve drug discovery and development (Challenge and Opportunity on the Critical Path to New Medical Products; http://www.fda.gov/oc/initiatives/criticalpath/whitepaper.html).

In a more tangible effort, the National Institutes of Health set up a separate Request for Application (RFA) to develop novel preclinical tools for predictive ADMET as one of its Roadmap Initiatives. Funding is to start in September 2005 under these initiatives and is expected to continue for four years (http://grants1.nih.gov/grants/guide/rfa-files/RFA-RM-04-023.html).

This initiative evolved from the NIH Summit Workshop conclusions that perhaps a re-think of ADMET was overdue and that "high risk", novel, and "high impact" technologies might be well received by both at the scientific and regulatory camps. Consistent with the federal restrictions on human stem cell research, only cell lines that are in the Human Embryonic Stem Cell Registry are eligible model systems for this RFA. One of the hurdles to acquiring high quality, in vitro screening data is not having good model cell lines whose characteristics approach those of in vivo physiology and genetics and are stable over long periods of time. Cell lines developed from human stem cells, transfected with constitutive telomerase activity, might have the potential to narrow this gap, but one must question the value of developing assays only on those available in the Registry.

In the past, the SBS ADMET SIG has occasionally focused on biomathematical rendering of screening data. We have had Stanley Young, formerly of Glaxo Research Institute, provide insights on the reliability and predictability of Ames Test mutagenicity data. More recently, Leminig Shi, from the FDA's National Center for Toxicological Research presented on the utility of the agency's database and Alan Roter from Iconix Pharmaceuticals presented on the depth of DrugMatrix.

When the focus was on predicting hepatotoxicity, John Senior from the FDA presented the different challenges in defining and measuring the many contributions of the liver to drug development and the requirements of HT ADMET.

More recently, with the push to fail compounds earlier and to increase the probability of success of compounds selected to go forward, Jean-Pierre Valentin from AstraZeneca provided insights into how HT ADMET is incorporated into a secondary screening program, while JoAnn Scatina from Wyeth Research presented a strategy that involved development criteria and resources at earlier discovery stages. Coming from the other direction, Christopher Lipinski provided colorful insights into standards that must be met to acquire useful HT ADMET data from the chemical structure-a minimum, if one hopes to perform quantitative structural activity relationships.

Looking Ahead
While we have received positive feedback on all of our presenters and most sessions have been well attended, we would like to know if, perhaps, the membership would prefer something entirely different. For example, would the SBS membership best be served with a forum that allows for more participation and discussion, given that the whole area of HT ADMET is being re-evaluated and looking forward to new ways of thinking?

There is no better place to start than at the meeting of the ADMET Special Interest Group at the San Francisco Regional Meeting. "Exploiting the Druggable Genome: A West Coast Focus" will be held April 21-22, 2005 at the San Mateo Marriott, San Mateo, CA. Please send your thoughts on how the ADMET SIG might better serve you to feedback@sbsonline.org. Everything is open for consideration and discussion, from the format of our meetings to content, focus, and especially to volunteers at every level to put exciting, informative sessions together.