What is a living thing? What are cells? How do they work? Pacific Northwest National Laboratory researcher William Cannon and University of California, Riverside’s Mark Alber and Samuel Britton set out to find answers to these questions by studying the chemical reactions involved in cellular metabolism, the process that regulates life in living things.
“We understand tornadoes and hurricanes and how the warm air mixes with the cool air to create winds,” said Cannon. “Inside cells, there are cycles that are essentially spinning like tornadoes. The issue is that the cell, like a tornado, can only exist under certain conditions and the cell needs to be able to control those conditions. We want to understand how the cells regulate the enzymes that are responsible for the reactions and cause metabolism to work.”
The research team investigated two hypotheses from the late 1960s that postulated that regulation in metabolism has two purposes: to control the metabolites, and to maintain energy flow inside the cells. Until now, the hypothesis that cells regulate the enzymes to control common biochemical pathways had not been proven with experiments or computer simulations.
Cannon, who holds a joint appointment at UC Riverside and has a background in chemistry, biophysics, and physics, said, “Computational mathematics is the most exciting work. In experimental work, you can only work with what the instrument can measure—which often isn’t a direct measure of the phenomena you want to understand. You can build models from first principles to really understand how things work.”
Reinforcement learning is a trial-and-error approach
Cannon’s research team used two methods to control the concentrations by regulating the reactions. The first method, called a deterministic method, relies on statistical thermodynamics and control theory. This approach assumes that each time a method is ran, the same results will happen, meaning there are no bumps or barriers in the energy landscape. The second method relies on statistical thermodynamics and control theory but also uses reinforcement learning, a type of machine learning (ML).
While it takes longer to run and reach a solution, reinforcement learning can get around bumps and barriers. In both methods, they changed the usual equation that describes the dynamics of the system, or mass action kinetics, to a maximum entropy formulation of the dynamics.
Using the mathematical and computational models, the scientists found that the cells were able to control the conditions, “so that the interior did not become thick and gooey like molasses,” said Cannon. They found that the cells regulate by controlling the metabolite concentrations, and their ML approach was highly accurate.
No other approach has been shown to rapidly identify points of metabolic regulation
In a study published in the Journal of the Royal Society Interface, Cannon and his team found that the first part of the original hypotheses was true. “The regulation actually causes the reactions to be much further from equilibrium. The nonequilibrium nature of a regulated reaction is due to the regulation; the regulation is not caused by the nonequilibrium nature,” Cannon said.
For the second phase of the research, the team is trying the same ML approach on a system with close to a thirtyfold increase in reactions. They expect the second method will perform even better.
This study was part of a three-year, $2.1 million grant from the U.S. Department of Energy, Office of Science Graduate Student Research award and the Biological and Environmental Research program; the National Institute of Biomedical Imaging and Bioengineering; and the National Science Foundation. Read more about this research here.
Learning how cells regulate and control themselves is essential to advance in important research areas, such as medicine, fundamental science, and synthetic biology, “We can predict how to engineer cells to do things we want them to do. We can predict what will happen if we change the environment that they are in. We can model them, and those models will help us guide experimental engineering efforts, like converting plants into biofuel,” said Cannon.