Under normal circumstances, people want to keep infections away from their computers. But for Gary An, reconstructing nasty infections inside a computer is a research project, not an act of cyber-terrorism. In collaboration with laboratories at the University of Chicago Medicine studying infectious diseases, An is creating computer models that simulate the delicate, complex balance between bacteria and their home in the human gut. By replicating the results of previous experiments within computational space, An hopes to enable in silico experiments that can push scientific discovery beyond the confines of the lab bench.
In two recent papers, An, associate professor of surgery at the University of Chicago Medicine, describes computer models that simulate a common post-surgical infection and a rare but very serious gastrointestinal condition that strikes premature infants. In both cases, the model uses research findings to recreate at least a portion of the microbe-host relationship in the gut, in order to look for clues as to why bacteria that normally live peacefully inside the intestines suddenly decide to attack.
Pseudomonas aeruginosa, a common cause of infection in patients recovering from surgery, is one such tenant turned aggressor. Most people are colonized by P. aeruginosa without ever realizing it, as it finds a quiet home in the lining of the gut alongside thousands of other bacterial species. That’s until the human host grows ill or undergoes major surgery and the comfort of the gut is disrupted, sometimes causing P. aeruginosa to riot. But this breakdown of what John Alverdy, professor of surgery, calls “molecular diplomacy” doesn’t always occur, creating a need for models that can describe how this system works — and fails.
“Part of the point of studying this particular bacterium and its role in infection is that there is a dynamic interplay between the bacterium itself and its environment…which happens to be a human,” An said. “The idea is to use the modeling as an iterative tool to help basic science labs do what they do a bit more efficiently. Using this model, we can gain insight into what they need to work on in the future.”
Surgical fellow John Seal led the effort to create a computer simulation of P. aeruginosa in the gut using agent-based modeling, where the simple actions of individual bacteria or cells sum up to a dynamic, complex system. In this case the “agents” are the bacterial and epithelial cells, which are set in Seal’s model against a backdrop of the gut’s mucosal layer and intestinal lumen. How those agents interact, and what various changes in the environment do to their behavior, were modeled after the results of experiments conducted over many years in Alverdy’s laboratory. As the model grew, the designers frequently tested it back against those experiments, to make sure the results of running the model accurately replicated what has been observed in the lab.
“It’s actually a very rigorous sort of thing, since we’re making it all up,” An said. “There are no laws of physics or chemistry in the model. It will do whatever you make it do. It’s potentially dangerous to get off track easily, therefore you constantly need to have reality checks back to the behavior of the system.”
Now that the system is in place, the science can move into the in silico world. Researchers can challenge the system in ways that would be difficult to do in a lab dish or an animal model, and can receive feedback on how the bacteria will respond. In the 2011 paper, published in Theoretical Biology and Medical Modeling, the team simulated conditions of host stress by adding inflammatory factors or endogenous opioids to the model to observe how they affect the virulence of P. aeruginosa — the likelihood that it will break its truce with the surrounding gut and attack. These simulations can generate new hypotheses to test in the laboratory, and also clue researchers into how best to run those experiments, such as suggesting the best timepoints to sample in a real-world experiment.
In another model, the same principles are applied to a disease called necrotizing enterocolitis, or NEC. Seen most often in babies born prematurely, the disease causes severe inflammation of the intestines that must be treated with severe surgery to avoid mortality. But despite clues gathered from animal experiments about the importance of early diet, feeding tubes, and the makeup of the gut’s bacterial ecosystem, no full model exists to predict which babies are at risk for this disease. An NEC computer model, built by An with fellow Moses Kim and professor of surgery and pediatrics Donald Liu and published in Surgical Infections, could help identify factors that put a premature infant at risk.
“It can’t be every kid that’s born early, can’t be every kid that you feed by tube, can’t be every kid that swallows some bugs. It has to have some particular component associated with that,” An said. “If, for instance, you could find a gene or an expression level in a premature infant that would suggest that their metabolic stress management capability is decreased, then you could identify a sub-population of kids that were at risk beyond the 3 in 1,000 incidence that it currently is at.”
Currently, the models are limited by the information that has already been collected in traditional experiments. But with advanced microbiome sampling techniques giving biologists a flood of data about the bacterial worlds that live inside our body, future challenges will lie in making sense of how literally thousands of agents all interact with each other and human cells to maintain a healthy balance or spiral off into dangerous infections.
An says that his models are “modular,” able to be scaled up as new knowledge comes to light about the ecology of the human gut. Eventually, An said, computers will even be trained to create the models themselves, converting the data and text from scientific publications into better and better simulations. Those models can then be published withi electronic scientific journal articles, allowing other researchers to test it out and tinker with their own hypotheses. That automation will be increasingly crucial as the complexity of the microbiome out-paces the ability of simple computers — or their operators — to comprehend it.
“That’s emerging as probably an important fundamental concept,” Alverdy said. “To really try to understand and effectively fight infection, it’s something that the human brain can’t compute.”
Seal, J., Alverdy, J., Zaborina, O., & An, G. (2011). Agent-based dynamic knowledge representation of Pseudomonas aeruginosa virulence activation in the stressed gut: Towards characterizing host-pathogen interactions in gut-derived sepsis Theoretical Biology and Medical Modelling, 8 (1) DOI: 10.1186/1742-4682-8-33
Kim, M., Christley, S., Alverdy, J., Liu, D., & An, G. (2012). Immature Oxidative Stress Management as a Unifying Principle in the Pathogenesis of Necrotizing Enterocolitis: Insights from an Agent-Based Model Surgical Infections, 13 (1), 18-32 DOI: 10.1089/sur.2011.057