Mathematical Models Lend Penn Vet Professor Insights Into Diabetes
For a trove of examples of the rich outcomes of interdisciplinary work, look no further than Darko Stefanovski’s research portfolio.
Stefanovski, an assistant professor of biostatistics in the University of Pennsylvania School of Veterinary Medicine, originally trained as a computer scientist, but has made insights into conditions like obesity and diabetes that have long eluded biologists. Wielding mathematical models like divining rods, he uncovers the hidden forces at play in biological systems.
“Although math has been used in the biological sciences, especially statistics when it comes to subjects like genetics and bioinformatics, mathematical modeling is not as widespread,” says Stefanovski. “But if you think back about Einstein and his theories, he was doing something similar. In his case, he didn’t do the experiements, he came up with these concepts, and only later did the measurements prove the concepts were true. Mathematical models are doing the same thing in biology: uncovering the dynamics at work.”
Originally from Macedonia, Stefanovski began pursuing a computer science degree there before immigrating to the United States, where he felt he would receive stronger training.
In 1999, he earned an undergraduate degree in computer and information science from Shepherd University in Shepherdstown, West Virginia. After graduating, he learned through a friend who worked at the National Institutes of Health about an opportunity to apply mathematical modeling at the National Institute of Diabetes and Digestive and Kidney Disorders.
There, he supplied his knowledge in computer programming to the work of researchers including Anne Sumner and Richard Bergman. They were using mathematical models to investigate how glucose was broken down, using data from the classic glucose tolerance test. The model, known as MINMOD, short for Minimal Modeling, helps calculate insulin sensitivity and pancreatic function from the reults of repeated glucose tolerance tests, a method that can reveal early signs of diabetes.
“What you're trying to do is take this very complex data that has been collected and summarize it,” says Stefanovski. “And through the process of summarizing it, you’re building a mathematical model.”
While working at the NIH, Stefanovski was invited to Penn to work with Raymond Boston, a professor of applied biomathematics at Penn Vet’s New Bolton Center, who has since retired but remains an important collaborator, to refine and update MINMOD. Along the way he earned his master’s in bioengineering in 2003. The updated software, MINMOD Millennium, remains in use by diabetes researchers, pharmaceutical companies and physicians around the world to evaluate insulin sensitivity.
“The advantage of the program is that it makes mathematical models accessible and user-friendly to clinicans,” Stefanovski says.
Besides evaluating insulin sensitivity, the model revealed another force acting in the metabolic system, a pathway through which glucose breaks down independent of insulin. This little-understood process is known as glucose effectiveness, and later became a focus for Stefanovski.
Though math and computer science were his research methods of choice, Stefanovski realized that he needed to know more biology to be able to effectively contribute new insights that may one day improve people’s lives.
“How do we solve diabetes was always my concern,” he says. “The answer was not in better math, but in learning more about the system.”
For that reason, following his master’s degree, Stefanovski decided he needed to learn more about “the system,” that is, the human body. He earned his doctoral degree in physiology and biophysics at the University of Southern Calfiornia, working under Bergman, the original mastermind behind MINMOD.
Specifically, he continued to zero in on the underlying mechanism of glucose effectiveness, studying two enzymes he suspected to be involved in breaking down sugar, glucokinase and glucose regulatory protein. Toward the end of his degree, he also turned his attention to lactate, which can be a product of glucose metabolism. As Stefanovski continued in a faculty position at USC, he developed a model of lactate kinetics that could reveal glucokinase activity in the liver, providing important information about a person’s metabolic state, as glucokinase acts as a glucose “sensor” that can trigger shifts in metabolism. Reduced activity of the enzyme is associated with the onset of diabetes.
The model also identified a potential mechanism for glucose effectiveness. Stefanovski was among those who collaborated with Francis Collins, now director of the NIH, to find risk factors for Type 2 diabetes. Among the genetic variations that increased susceptibility were mutations in glucokinase.
Stefanovski’s work in this area continues at Penn. He is using molecular dynamics, computer simulations of molecular biophysical interactions, to speed the process of designing a small molecule compound that can target glucokinase, regulating glucose levels and thus potentially serve as a therapy for diabetics.
Stefanovski’s expertise has occasionally taken him in unexpected directions. For example, in 2011, he was called upon to work with Bergman and other colleagues to develop a better model to measure body fat than the widely used body mass index, or BMI. After a thorough evaluation of existing data on which measurements correlated most closely with the Dexa scans, the gold standard for evaluating body fat percentage, he and colleagues published a new measure, using only hip measurement and height. The result, referred to as the body adiposity index, provides a reliable estimate of body fat percentage across ethnic groups, something that the standard BMI calculations, which require weight and height, often fail to do.
At Penn Vet, which Stefanovski rejoined in 2014, he has been called upon by a diverse set of researchers to employ his mathematical modeling expertise to assist them in their studies. He continues a collaboration with his predecessor, Raymond Boston, on the biological modeling software WinSAAM.
"Dr. Boston has a prominent presence in my career," says Stefanovski. "Needless to say, I would not have been here if it wasn’t for him.”
Stefanovski has assisted Penn Vet’s Amy Durham, an assistant professor, on a project involving feline cancer and Louise Southwood, also an assistant professor, on her research on equine intestinal disease. He has also collaborated with professors in the Perelman School of Medicine to lend biostatistical and modeling assistance.
Together with Penn Vet’s Mary Robinson, director of the Equine Pharmacology Laboratory, he’s engaging in a project that brings his research on diabetes and glucose metabolism squarely into the veterinary realm, exemplifying the concept of One Health, or the connection between human, animal and environmental health.
Laminitis is a condition universally dreaded by horse owners. In it, tissues in the horses’ feet become inflamed, which in some cases can lead to an inability to walk. It has been shown that horses can develop a condition very similar to metabolic syndrome in people, whereby they become obese, levels of inflammatory molecules rise and the body produces more insulin than normal to try to combat decreasing insulin sensitivity in the beta cells of the pancreas. In horses, these increasing levels of insulin seem to contribute to laminitis.
“Horses have terrible problems with insulin because high levels of insulin are associated with laminitis,” Stefanovski says.
To prevent insulin levels from spiraling out of control, Stefanovski and Robinson are examining whether triggering the glucose effectiveness pathway could suppress the rise in insulin.
“We’re hoping that activating this mechanism might help horses,” he says.
Looking ahead, Stefanovski looks forward to more collaborations in the vet school, medical school and beyond, hopefully leading to tangible outcomes for people and animals.
“It’s funny to think that I got into the field thinking, with my computer science background, I might go work for Pixar or some place like that,” he says. “But I’ve since realized that finding ways to visualize data through these models actually has many more beneficial effects than just satisfying the visual senses.”