Desktop Drugs - Drug Discovery and Development

in silico predictions establish a beachhead in the war on cancer.

click to enlarge
This figure shows a model of docked hits in the binding site of TRAF6, via electrostatic surface based on APBS calculation. The figure is generated by PyMol. (Source: Department of Experimental Therapeutics, M.D. Anderson Cancer Center)

A hallmark of a good predictive model for drug development is the capacity to grapple with myriad “what-ifs.” Usually, this takes real-world data, and gobs of it. But recently, it was the lack of raw data that sparked an innovation.

Consider the US National Cancer Institute’s NCI-60 panel of human cancer cell lines, a resource that has been screened with over 100,000 chemical entities: “Out of the 60 cell lines that are on the NCI panel, there are none from bladder cancer,” explains Dan Theodorescu, MD, PhD, Paul Mellon Professor of Urologic Oncology and Molecular Physiology, University of Virginia, Charlottesville, Va. “This was a motivation for me to extrapolate from the NCI-60 cell lines to a bladder cancer panel,” an effort requiring a new set of algorithms collectively known as “co-expression extrapolation” or, COXEN.

In brief, COXEN (see diagram, below) uses NCI-60 gene expression data, correlated with in vitro responses to anti-neoplastic agents for all 60 cell lines, and then compares that to expression data from bladder cancer samples. The result is a COXEN score, which is the probability of a therapeutic response, either for a given bladder tumor sample, or for the collection overall. This prognostic ability is invaluable in bladder cancer where there are so few therapeutic options. “At the very least, you want to assign patients to a drug combination that has the highest COXEN score for that particular tumor,” says Theodorescu. The greater value of this approach, however, is the ability to mull over the as-yet-untried: “We could analyze someone’s tumor by COXEN, look at scores for all the 75 FDA-approved [oncology] drugs, and then give that priority list to a medical oncologist who can create an utterly novel combination,” he says.

While this new clinical approach is being validated, COXEN analysis has been expanded to include data from the full battery of chemicals, approved and otherwise, previously thrown at the NCI-60 panel. This has already resulted in 115 hits, and one solid lead for bladder cancer. “C1311 is an analog of the top hit, an agent which has actually been used in two benign conditions, as well as two cancers which are in ongoing Phase 2 trials. It’s especially exciting,” says Theodorescu, “because it’s an orally bioavailable drug.” Furthermore, other drugs with known activity in bladder cancer such as E09 (Eoquin, Spectrum Pharmaceuticals) have also been identified among the top hits, which further supports the ability of this approach to identify effective compounds for bladder cancer.

Door Number Three
“We were walking into a pathway that other pharma companies had been working in for twenty years, and we walked out with a completely different view of it, and the proper target,” marvels Ulrik Nielsen, PhD, vice president of research, Merrimack Pharmaceuticals, Cambridge, Mass. The target is ErbB3, and the therapeutic based on this in silico divination is MM-121.

Using the concepts of systems biology and the methods of a chemical engineer, Merrimack’s team intricately, mathematically described all four receptors in the EGF pathway. The model was then trained with clinical data derived from already approved EGF inhibitors: Erbitux, Tarceva, Tykerb, etc. “It was mostly an academic venture, in the sense that there were so many drugs in the pathway already in clinic—we didn’t think the effort would evolve any new therapeutics,” he says. Yet the model, once matured, then perturbed, was telling them that the optimal target wasn’t on anybody’s R&D radar. “This receptor, erbB3, is sort of the red-haired stepchild of that family,” says Nielsen. “It’s not a tyrosine kinase—it isn’t really over-expressed in cancer—everyone’s been overlooking it. But the model tells us that this is where we should be hitting the pathway.”

This ability to operate free from bias is one of the many advantages of in silico drug design. “Most people still work from the assumption that a cancer target is something that’s over-expressed, yet Herceptin is the only compound that’s really [survived that assumption].”

MM-121, described by Nielsen as “the first systems biology drug” is currently in preclinical testing, and slated for Phase 1 evaluation within the next six months.

New school
With industry pipelines in a state of low flow, and costs skyrocketing, there is greater incentive to explore new venues. “Why not do drug discovery in an academic environment?” asks Shuxing Zhang, PhD, assistant professor in the Department of Experimental Therapeutics, M.D. Anderson Cancer Center, Houston. “M.D. is the number one cancer center in the world. We have very good basic science; we have people in clinical trials, expert clinicians in cancer treatment…” Zhang was retained last year by the department chair to establish a discovery program, with in silico investigations as its core.

Directed by Zhang, the new Molecular Modeling Service (MMS) at M.D. Anderson is being offered to speed drug discovery by incorporating the techniques of bioinformatics, chemoinformatics, and systems biology. “This is a different concept to try to put them together.” In the broadest sense, the application of these approaches is either structure-based, or ligand-based. For instance, given a protein structure, MMS can run a molecular docking search against a database of five million known compounds, with a query of 50,000 candidates taking no more than two computing days.

More complex investigations will soon be handled by a Web-based portal within MMS called “M.D. Workbench.” This service will allow outside academics the opportunity to use a relational database to discern signaling pathways crucial to tumor survival. Should this process result in a virtual drug candidate, Zhang hopes to soon have a database robust enough to assess a candidate’s ADME/Tox (absorption, distribution, metabolism, excretion and toxicity) profile—all before a single milligram of drug has been manufactured.

For now, the challenge is the data. “One thing I really need,” says Zhang, “is the ADME/tox studies. The data from academics is very limited, so we would really like to have some clean data for real drug compounds from pharma.”

click to enlarge The COXEN algorithm was applied to a combination of data sets: microarray expression data from the NCI-60 panel; bladder cancer cell lines; and tumor tissues from bladder cancer patients to identify active drug leads for human bladder cancer. Source: Dan Theodorescu, PhD)

Connecting the dots
The underlying problem we see in biomedical research is data fragmentation,” says John Quackenbush, PhD, professor of biostatistics and computational biology, Dana Farber Cancer Institute, Boston. “We have all this clinical data stored on a whole host of clinical databases, but a lot of it is completely divorced from research data.” This conflict was directly affecting a number of groups at Farber including the multiple myeloma group, who were being frustrated in their efforts to design new clinical trials based on data from previous investigations.

The solution, simply stated, is something like couples counseling—all parties in the same room with a moderator to facilitate communication. For Quackenbush, a timely $1 million Oracle Commitment Grant helped pay for this new technological intervention.

“If you have two databases that won’t talk to each other, the solution is to create a third database. And once we did that, we realized we could reach out and pull in all the information that’s available in the public domain,” explains Quackenbush. Data from sources like ChemBank, GenBank, or DrugBank can all be drawn into a given query: questions as mundane as ‘how many samples of EGFR over-expressers do I have in the freezer?’ to ‘Does XYZ kinase map to ABC receptor?’ At this level of sophistication, the utility for drug discovery work becomes obvious.

The multiple myeloma project is early in development, and still in-house, but the eventual goal is to make the architecture a public resource, allowing anyone to mirror the system’s elegant profile.

About the Author
Neil Canavan is a freelance journalist of science and medicine based in New York.

This article was published in Drug Discovery & Development magazine: Vol. 11, No. 5, May, 2008, pp. 26-28.

Screening using isogenic human cancer cell lines
Caliper Life Sciences, Inc. (Hopkinton, Mass.) announced in April a partnership with Horizon Discovery Ltd. (Cambridge, U.K.) that expands Caliper Discovery Alliances and Services’ (CDAS) oncology cell line and screening capabilities for testing single drugs and combination therapies. Through this alliance, CDAS now offers unique, genetically-defined, and isogenic human cancer cell lines that allow researchers to better identify and characterize personalized drugs targeted at a specific subset of patients.

Isogenic cell-line pairs, where one cell line contains a genetic alteration or mutation of interest and the other contains the normal gene, represent cell-based models that accurately portray specific human diseases in the context of their matched normal genetic backgrounds. The inclusion of isogenic cell-line pairs in the discovery process enables scientists to better understand the mechanism of action of lead compounds, directly identify patient-relevant compounds from large compound libraries, re-profile existing drugs for new therapeutic indications, assess the efficacy of drug combinations, and identify potential side effects earlier in the discovery process, according to CDAS.

“The alliance with Horizon Discovery adds significant value to existing CDAS offerings and illustrates Caliper’s commitment to providing accurate in vitro and in vivo models for oncology research,” said David Manyak, executive vice president of drug discovery services, Caliper Life Sciences. “The addition of Horizon’s isogenic cell lines to our existing oncology cell proliferation panel, and the ability to correlate results from these isogenic cell lines to efficacy in specific patient populations, further solidifies Caliper’s in vitro-in vivo-human (IIH) bridge. Access to these tools will enable researchers to enhance the success rate and reduce the cost of discovering targeted monotherapy or combination therapies that better treat disease with fewer adverse events.”

The discovery of targeted or personalized medicines that treat patients with a specific genetic makeup is a long and difficult process. Using more predictive cellular models that better reflect the complexity of biological systems will accelerate and rationalize many aspects of the discovery process, including target selection, lead-compound generation, and identification of the most responsive patients for clinical studies. The availability of such models also enables molecular diagnostic findings to be incorporated into existing drug discovery processes, CDAS says.