Deciphering Virocell Metabolic Dynamics and Ecosystem Outputs Using a Novel Integration of Machine Learning and Metabolomics
Sumudu Rajakaruna1* (email@example.com), Marion Urvoy2, Roya AminiTabrizi1, Cristina Howard-Varona2, Matthew B. Sullivan2, and Malak M. Tfaily1,3
1University of Arizona, 2The Ohio State University, and 3Pacific Northwest National Laboratory
The overarching goal of this project is to establish ecological paradigms for how viruses alter soil microbiomes and nutrient cycles by developing foundational (eco)systems biology approaches for soil viruses. Here, the team used metabolomics to investigate phage-specific metabolic reprogramming in virus-infected cells (virocells) to build critically needed model systems and in silico resources and tools that can be extended to new soil model phage-host systems. Researchers used an already established marine phage-host model system as the base to (1) characterize metabolic dynamics of virocell infection, (2) assess the output of virocell metabolic reprogramming on ecosystem function, and finally (3) develop the analytics that will be directly transferable to soil systems.
Microorganisms, including bacteria and viruses, play a vital role in biogeochemical cycles and global ecosystem function. At present, virus contributions are largely assessed through community-scale geochemical measurements or through evolutionary inferences such as identifying horizontal gene transfer. However, viruses also metabolically reprogram their bacterial host cells towards virion synthesis during infection, which effectively makes the infected cells (virocells) ecologically, metabolically, and physiologically different from uninfected bacteria. Here the team investigated the endo- and exometabolomes of an ecologically important marine heterotroph (Cellulophaga baltica strain #18), independently infected by three viruses with different morphologies, genomes and, therefore, infection strategies, to understand phage-specific virocell metabolic reprogramming and ecosystem outputs in a highly resolved infection time course. Through numerous ordination and multivariate statistical analyses, researchers show not only that a virocell’s metabolite dynamics are different to that of an uninfected cell, but also that such metabolite dynamics differ temporally and between each virocell in ways that qualitatively associate with virus infection efficiencies, likely impacting ecosystem function. To assess this impact most comprehensively, researchers are optimizing metabolite annotation—a major limitation in metabolomics—via testing and implementing combined machine learning and deep learning algorithms to obtain probabilistic annotations for unknown metabolites. These efforts have enabled a 24% increase in total annotations so far, which ultimately would significantly improve the biological interpretation of metabolomics results. Given that soil microbes can be infected by viruses at any given time, collectively these findings suggest that viruses can play an important role in regulating present and future carbon cycling.
This research is based upon work supported by the U.S. Department of Energy Office of Science, Office of Biological and Environmental Research under Award Number DE-SC0020173 and DE-SC0023307 to the Ohio State University.