PNNL

Abstract

Despite the wealth of available transcriptomics datasets, it is still difficult to gain mechanistic insights of causal relationships without an understanding of the chemical or biological content underpinning the correlated observable traits. The main challenge of mechanistic modeling for predicting phenotypes lies in the recording of the vast possibilities that emerge from multiple regulatory pathways of which the combination of environmental and genomic cues further complicates data interpretation. To meet the technical challenge of predictive capabilities, we established a workflow of a data-driven modeling from a well-established physics-informed machine learning algorithm to interpret the movement of gene expressions over diel cycles in response to light perturbation from the publicly available transcriptomic datasets of Synechococcus elongatus. To tease apart the dependent variables attributed to carbon fixation from the independent variables such as circadian time, we presented a curated transcriptome dataset of light-dependent reactions in photosynthesis as a basis of training a mathematical model to relate observable expression of central carbon metabolism genes over circadian time. In addition, by grouping the curated expression datasets into modulated ones that recapitulates the essential phenotypes, this data-driven workflow reveals hierarchical regulatory dynamics connecting the ones known responsive to light variation to multiple pathways in the central carbon mechanism. We used quantitative global and redox proteomic techniques to characterize changes in abundance and cysteine redox status of S. elongatus proteins collected from illuminated before and a two-hour transition from light to dark. While protein abundance changes were minimal, we detected significant redox status shifts for cysteine residues including sigma factors, enzymes involved in both the oxidative pentose phosphate and the Calvin-Benson cycle. Such observation indicates the separation of time scales in the regulatory dynamics that couples circadian cycle and diverse metabolic pathways after the stimuli with light by the distinction of fast redox or slow biochemical reactions through the hierarchy of an interacting networks. The workflow we developed enabled a mechanistic understanding of relating molecular changes to certain phenotypic traits through multiple regulatory pathways under environmental stimuli.