The key to understanding natural systems, Raman says, is that they grow and adapt. They aren’t static machines set into motion that execute preprogrammed instructions. They change in response to different inputs or environments, much like individual organisms.
“If we want to ultimately engineer biological systems to do things that we know engineered systems can do, we have to embrace the idea that there's an entirely new science that needs to focus specifically on systems that have come through the process of evolution, rather than engineering,” he said. “If you understand how principles of design and biology are manifest, then you should be able to go into the lab and create systems that are adaptive and robust too.”
Complex systems, core components
In its most abstract sense, a system is just a bunch of parts that come together to serve a purpose. So, biological systems can be described at many different levels of complexity. Within multicellular organisms, like animals, the systems are different organs and tissues that work with each other to breathe and think and move. At the protein level, amino acids interact with each other to create the folding and function of a protein.
While working with Ranganathan, Raman started using statistics and mathematical formulas to understand systems of proteins. He noticed that out of hundreds of amino acids that may form a protein, just a small number were responsible for its actual functions. What’s more, that small number of amino acids weren’t the obvious candidates based on what he might have guessed from its three-dimensional structure or usual rules of biochemistry. But there were signals that mathematical formulas could predict the important players, and he was intrigued by the idea that those formulas could apply in other contexts too.
During his postdoctoral work at Washington University in St. Louis, Raman had his chance to test this idea. He worked in the lab of Jeff Gordon, a UChicago alum known as “the Father of the Microbiome” for his groundbreaking studies on the links between gut bacteria and obesity. Gordon’s work renewed interest in the links between the world of microorganisms that inhabit the digestive tract and human health and inspired a cottage industry of scientists sequencing and comparing every gut microbe they could find. But a microbiome, whether in the human gut or anywhere else in the natural world, is an exponentially more complex system of independent organisms that interact with each other, their hosts, and their environment—that is, an ecosystem.
Raman applied the mathematical rules that he developed in simple protein systems to human gut microbiomes and found that the same remarkably small number of components of the system could account for their behavior. For example, he found that out of thousands of species of bacteria in samples from acute malnourished children, just 15 were out of balance compared to healthy control samples. Using that data, they also created a supplement to restore those crucial 15 species that helped the children develop and thrive much better than their cohorts. They saw the same results in animal models as well: The loss of the same 15 species of gut microbes led to malnourishment.
“The same dynamics recapitulated themselves over and over again,” Raman said. “So, there was reason to believe that going through the process of evolution actually reduces the dimensionality of those systems down to core components, and those core components are conserved across contexts.”
Engineering microbiomes from scratch
Now as an assistant professor of pathology in the Duchossois Family Institute at UChicago, Raman and his team are using the mathematical rules that reduce complex systems like gut microbiomes down to their core components, and then using that data to engineer new microbiomes from scratch with a specific function in mind. But instead of going about it like an artisanal chef, selecting ingredients based on their distinct properties, they take a systems approach.
Using the DFI’s bank of more than 2,000 gut bacterial strains, they created 100 different custom microbiomes to see how well they could kill off a common pathogen called Klebsiella pneumoniae. Then, using adaptive machine learning algorithms to analyze the results of which combinations did better or worse, they created new microbiomes to test again, refining the process over and over until they had a 24-member, synthetic microbiome that depresses the pathogen in vitro and in mouse models. While the repetitive work for this proof of concept was carried out by a sore-armed and weary postdoc, Raman believes automation can scale up the process to test thousands of synthetic microbiomes for desired functions at a time.