PNNL

The Science 

Through photosynthesis, cyanobacteria harvest energy from light to fix carbon dioxide into glucose during the daytime. This glucose is used to create valuable chemical products to fuel their activities at night. Modeling this process is complex, as it involves multiple pathways that adapt to energy-light limitations across daily cycle changes based on the Earth’s rotation. A substantial challenge lies with making phenotypic predictions of protein regulators in cyanobacteria’s metabolic state due to the vast possibilities that emerge from multiple regulatory pathways. To address this, a multi-disciplinary team developed a data-driven modeling workflow using a physics-informed machine learning (PIML) algorithm. They used this workflow to study day-to-night gene expression dynamics of a cyanobacterium called Synechococcus elongatus. They revealed the cyanobacterium’s molecular mechanistic responses to illumination, which unlocks new potential for redox proteome experiments. This technique bridges the gap between multi-omics data and predictive physical models. With this, researchers can screen for protein regulators that participate in overlapping enzymatic reactions within multiple metabolism regulatory pathways amid light disturbances. 

The Impact 

The team’s PIML-omics method combines a machine learning algorithm with physics-informed dynamic modeling to specify molecular interactions and their phenotypic responses to environmental perturbations, such as changes in light. The package doesn’t require prior knowledge of the genes and transcription factor interactions. It helps generate testable regulatory models using a novel approach based on multiomics-based sequence-structure-function predictions. The model has applications for bioengineering the design of desired phenotypes for a variety of important bioproducts by providing fundamental insights into the intricate canals of regulatory pathways in cyanobacteria’s response to environmental stimuli. 

Summary 

A multi-disciplinary team of researchers created a data-driven modeling workflow using a PIML algorithm to interpret gene expression dynamics from the cyanobacteria Synechococcus elongatus as it responds to light perturbations over diel cycles. They classified curated expression datasets into phenotypes relevant to light-dependent metabolic reactions and trained a non-linear mathematical model with a curated transcriptome dataset. This helped determine the dynamics of light-responsive elements within carbon metabolism pathways over circadian time. Additionally, they used their framework to experimentally characterize phenotypic changes in protein abundance and cysteine redox status within cyanobacteria under constant illumination and after two hours of sustained darkness. Their results showed that while changes in protein abundance are minimal, significant shifts in cysteine redox status occur. This affects critical regulatory proteins and enzymes in overlapped metabolic pathways. The team’s workflow offers a mechanistic understanding of how molecular changes relate to specific phenotypic traits. Researchers can use this information to predict multiple phenotypes from complex regulatory pathways under environmental stimuli. They can also use this information to bioengineer desired phenotypes in organisms that are used to develop a range of valuable bioproducts. 

Contacts 

Margaret S, Cheung, Environmental Molecular Sciences Laboratory, Pacific Northwest National Laboratory, margaret.cheung@pnnl.gov  

Song Feng, Pacific Northwest National Laboratory, song.feng@pnnl.gov  

Funding 

This project was supported by the Northwest Biopreparedness Research Virtual Environment project (NW-BRaVE), funded by the Department of Energy, Office of Science, Biological and Environmental Research program FWP 81832. Computing capabilities, including the use of the Tahoma computing cluster, was provided by the Environmental Molecular Sciences Laboratory, a Department of Energy Office of Science user facility located at Pacific Northwest National Laboratory. Prior support to NW-BRaVE was provided by the Predictive Phenomics Initiative under the Laboratory Directed Research and Development program at Pacific Northwest National Laboratory.