Modeling microbial complexity

Will Harcombe combines experimental and computational models to decode complex microbial interactions.

By Bernard Cook III

The role of E. coli in causing food-borne illness is well known. Less well known, however, is the way microbes interact with each other. Like humans, they exchange resources and live in communities. Understanding these complex interactions could help scientists predict and even control the properties of a microbial community—yielding new tools for treating infections, promoting human and animal health, and improving soil productivity. 

Will Harcombe, an associate professor of Ecology, Evolution and Behavior and a member of the BioTechnology Institute, studies microbial interaction by looking at the exchange of chemicals among microbial species. “One of the fundamental ways [microbes] can influence each other is just by taking up and releasing chemicals into the environment,” he says. “Will they compete by trying to eat the same resource? Or will they facilitate each other by one secreting a product that helps another?”

Model Behavior

To make sense of microbial interactions, Harcombe starts with a model of how he expects microbes to interact. At its core, a model is a tool that describes how the output of a system depends on its inputs. Harcombe’s models consist of equations that relate the abundance of two microbial species, the output, to defined inputs like available nutrients and spatial distribution. Using the model, he can test what happens to the abundance of two microbial species when placed in an environment with defined nutrients. Importantly, this allows him to observe what the microbes do when they interact in the ways he suspects.

After generating computer-based results, Harcombe replicates this experiment in the lab.  Harcombe places two or more species of microbes together and tracks the resources they exchange. At the same time, he observes population-level characteristics such as growth rates, colony shape and species abundance. By simultaneously monitoring resources and community traits, he can answer questions about the spatial organization, the ratio between species and how available nutrients impact the exchange of resources. 

Sometimes experiments align with the results generated by his model. In other cases, they differ from — or even contradict — the model.

Two-Species Surprise

In one study, Harcombe and his colleagues wanted to investigate the dynamics between two microbial species, each of which depends on the other for survival. One (E. coli) was modified to require a nutrient—methionine—produced by the other (S. enterica). S. enterica, for its part, needs acetate excreted by E. coli.

One might expect these microbial colonies would thrive when cultured together — and when cultured alone, neither would make it. Per his computer model, that’s precisely what happened. When cultured in the lab, he saw the same results.

 Then came a surprise. In the lab, Harcombe introduced a virus that infects one species but not the other. Because each species’ survival depends on the other, Harcombe expected that infection would impact both. 

When he introduced a virus that only infected S. enterica, his prediction was spot on. When he introduced a virus that only infected E. coli, he ended with fewer E. coli as predicted, but surprisingly, he also found more S. enterica relative to pre-infection numbers. What was he missing?

Answering that question experimentally would have required painstaking tests, altering one variable at a time until he understood what was happening. While that approach might eventually provide an answer, it would be costly. Without knowing which explanations were plausible, Harcombe could spend years testing possible mechanisms responsible for the odd result. 

Productive Insights

Enter modeling. If the microbes weren’t behaving the way he expected, there must be something else; different metabolites or different types of interactions to explain the observations. So rather than running experiments to identify the mechanisms at play, Harcombe put those possibilities into the model. Once he found a model approximating his observation, he could confirm his conjecture with additional experimentation.

Harcombe suspected some E. coli developed resistance to the virus, protecting them from extinction. He further postulated that others would burst and release a wealth of nutrients. Together, they might provide an abundance of the acetate S. enterica need to thrive. Incorporating these mechanisms into his model yielded results that closely mirrored the experimental model.

“The biggest surprise to me in this study is that bacteria could get the elements they needed from other species by eating the dead,” says Harcombe.

Back in the lab, Harcombe ran additional experiments to test the model. For the most part, they supported his hypothesis. The only thing missing was a second consequence of E. coli developing a resistance to the virus. Virus-resistant E. coli produced more acetate than their nonresistant counterparts.

Harcombe is pleased with the way the experiment-plus-model approach has boosted his ability to solve microbial mysteries, and how the solutions stand to benefit humanity. Harcombe and colleagues use these models to predict how microbes respond to antibiotics and engineer interactions into microbial communities in soils to render crops more resilient to a changing climate. “Going forward, I’m really optimistic that we will be able to continue to develop our understanding of fundamental processes and apply it in these different areas.”

Mapping Uncertainty

Can gene regulatory networks help scientists predict cell behavior and improve therapeutics for cancer and other diseases?

By Bernard Cook III

It’s one of the big mysteries of biology. Why do genetically identical cells often differ in the way they move, which proteins they produce and how they respond to their surroundings? Casim Sarkar, a professor in Biomedical Engineering and a member of the BioTechnology Institute, is working to solve this enigma by studying gene regulatory networks — the interaction between genes and the proteins they produce.  

Sarkar likens these gene regulatory networks to computer code that governs which genes are turned on, what proteins they produce and in what quantity. As it turns out, a cell’s behavior (for example, how it moves) and the decisions it makes (whether to move) are largely determined by gene regulatory networks and how they interact with the cell’s environment. 

To begin unraveling the link between gene regulatory networks and cell behavior, Sarkar’s lab borrowed the concept of the “epigenetic landscape” from developmental biologists. Initially, the epigenetic landscape was developed as a visual metaphor to demonstrate how a stem cell (a cell that has not yet developed a specific function) becomes one of many cells with a defined function, like a neuron or a muscle cell. According to Sarkar, “The landscape looks like a ski slope where you can take different paths all starting from the same point.” 

Like a ski slope, the epigenetic landscape has defined features like height, steepness and surrounding hills and valleys, which collectively inform what a cell — in this case, the skier — will do next. Picture placing a marble on a Pringle: in some directions, it may move downward and in others it can only go up. Let go, and the marble will most likely move downward. Like this, a cell on or near a crest is likely to move downhill and perform actions that are more likely, while a cell in a valley is met with resistance and likely won’t change its behavior at all (unless, like a skier, it has accumulated sufficient prior momentum to move upwards).

Instead of simply using the landscape to conceptualize cell behavior, Sarkar’s lab is developing a computer-based epigenetic landscape to predict what a cell in a given state might do next — and the likelihood of that outcome. This approach is particularly useful because it allows him to incorporate elements like cell-to-cell variability and DNA modifications, two features of gene regulatory networks that may push two otherwise identical cells to do different things in the same circumstances. In his model, these elements play a role in determining the shape of the landscape, which allows Sarkar and his team to make predictions about cellular behavior by accounting for these confounding elements. Importantly, this approach also allows him to pinpoint facets of the gene regulatory network that drive two identical cells to behave differently. 

Understanding why identical cells take different paths may help improve therapies for cancer and other conditions. In cancer, for example, tumors consist primarily of cells that proliferate rapidly. But some cancers also contain dormant cells that evade standard chemotherapy and often cause relapse. Sarkar’s approach could identify aspects of the gene regulatory network that push some cells to choose dormancy, which in turn, may help researchers keep these cells dormant or identify an intervention that awakens dormant cells and renders them vulnerable to standard chemotherapy. 

Sarkar’s long-term goal is to make concrete predictions about what a cell might do in specific environmental conditions as a result of the underlying gene regulatory network. Solving this mystery could help engineers working with stem cells create tissues with the desired structure and function and improve therapeutic strategies to combat antibiotic resistance.