Effective Animal Models
Instead of focusing on a particular platform technology, Synta Pharmaceuticals’ biomarker approach zeros in on important questions in the biology of human disease. “We let the biology choose the assay, not the other way around,” comments Kevin Foley, Ph.D., director of in vivo pharmacology.
Synta’s scientists try to understand the molecular basis of a disease and develop animal models to test compounds for the specific biology that translates into human disease. “Traditionally, the problem has been the gold-standard model. If you’re developing a drug for arthritis, you use the collagen-induced arthritis model—everyone uses that model no matter what type of drug is being developed. That’s where you run into problems—the gold-standard model reflects some of the biology going on but not all of it. What we try to do is link animal models and human disease to have a better understanding of both ends so you can choose the best development pathway to develop drugs.”
These animal models have proven effective in development of the company’s vascular disrupting agent, STA-9584, which targets tumor vasculature. It binds to the colchicine site of tubulin, selectively targeting endothelial cells of tumors, but not normal tissue. This shuts off blood flow in the entire tumor (not just the center), leading to hypoxia and necrosis.
In preclinical animal models, the agent has shown improved therapeutic index relative to other VDAs such as combretastatin, Dr. Foley says. Necrosis and apoptosis will be used as biomarkers in the clinic to examine what happens in the tumors when treated with STA-9584. “We’re not using animal models to say, ‘yes, this drug is going to be clinically active,’ but we’re using it to say ‘it’s more likely to be clinically active compared to competing agents in this field.’”
Modeling and Simulation
In silico disease models and virtual patients are the basis of Entelos’ PhysioLab® platforms, which are used to identify and validate drug targets and develop biomarkers.
“Identifying a biomarker means identifying a selection of measurable attributes that would provide the desired result,” says Alex Bangs, Ph.D., cofounder and CTO. This is done by running simulations of many patients under the clinical conditions of interest on current standards of care, new therapeutic options, competitor products, etc., and then mining those simulation results to identify the best marker panel. “Having a model that includes more physiology of the biology around the disease represented gives us a rich environment to mine for these markers. Once we have identified the panel of markers, we can then trace them back into the physiology and understand why they are good markers and have confidence in using them to distinguish patients.”
Dr. Bangs emphasizes that the uniqueness of the PhysioLab platform is that it’s a mechanistic model, versus a statistical model that cannot distinguish relationships between high-level clinical markers. “Having the deeper representation of physiology (the mechanisms) means there are more potential places to identify markers in the system that could be predictive. To do the same in a clinical trial, you would have to measure too many things because you may not know in advance which would be most predictive.”
It’s important to simulate whole populations when looking at biomarkers to get a good idea of how patients will respond before the drug goes to market. “We create many virtual patients by varying genetic and environmental conditions to statistically mirror real clinical populations through public studies or previous clinical trials. The data is used to help us create the appropriate population.”
The company currently has PhysioLab platforms for asthma, type I and II diabetes, rheumatoid arthritis, cholesterol, and metabolism, and is developing cardiovascular and skin allergy platforms. Dr. Bangs says the FDA is reviewing mechanistic models in terms of how they can be helpful for questions regarding patient variability and for new therapeutics where there isn’t an existing body of clinical data.