Prediction of Metabolic States from Phosphorylation Changes in Saccharomyces Cerevisiae

While classically signal transduction and metabolism have been studied independently, integrating these two layers should provide better insight into their  interdependencies, and in particular between their key regulators, kinases and metabolic enzymes, respectively [1]. Large-scale studies have associated and quantified changes in proteins and metabolites under steady-state conditions for dozens of perturbations [2, 3]. Nevertheless, it is still a challenge to systematically and accurately identify kinases, or phosphatases, regulating functionally relevant phosphorylation sites (p-sites) in metabolic enzymes.

In this study, we analyzed our ability to predict kinase activities and metabolic concentrations under different genetic perturbations using yeast as a model system. A similar approach to [4] was used to infer the kinases activity in each condition using their substrates fold-changes. Reassuringly, conditions where the kinases were knocked-out or inhibited the activity was predicted to be significantly reduced. Interestingly, within each condition significant changes in the activity of other proteins were observed. For example, HOG1 perturbation is associated with loss of activity of PBS2, and both proteins are known to interact on the same signalling pathway [7]. We then correlated the kinases activity scores with the metabolite concentration changes to identify possible direct or indirect regulation of metabolic enzymes by phosphorylation. To further study this association we built linear regression models for each metabolite using the kinases activity scores as input features. Notably, the majority of the conditions showed a positive and significant correlation between the predicted and the measured values. To better assess the predictability of these models we used independent dynamic phosphoproteomics and metabolomics data-sets also acquired in yeast [5, 6]. Kinase activity scores were calculated as before for each time-point. The predictive models were trained with the previous steady-state data-sets and tested to predict the metabolites changes in the dynamic data-sets in each time point. The best predictive power was achieved on the later stages, possibly because they were trained with steady-state conditions. Rapamycin treated condition had particular good performance most likely associated to the fact that TOR inhibition is present in the steady-state data-sets.

In summary, we found that it is possible in yeast to robustly predict the activity changes of kinases and phosphatases under different perturbations, and to predict from these metabolite concentration changes.

1. Gonçalves et al., “Bridging the layers: towards integration of signal transduction, regulation and metabolism into mathematical models.,” Mol Biosyst, 2013
2. Schulz et al., “Large-scale functional analysis of the roles of phosphorylation in yeast metabolic pathways.,” Sci Signal, 2014
3. Bodenmiller et al., “Phosphoproteomic Analysis Reveals Interconnected System-Wide Responses to Perturbations of Kinases and Phosphatases in Yeast,” Sci Signal, 2010
4. Casado et al., “Kinase-substrate enrichment analysis provides insights into the heterogeneity of signaling pathway activation in leukemia cells.,” Sci Signal, 2013
5. Oliveira et al., “Dynamic phosphoproteomics reveals TORC1-dependent regulation of yeast nucleotide and amino acid biosynthesis.,” Sci Signal, 2015
6. Oliveira et al., “Inferring causal metabolic signals that regulate the dynamic TORC1-dependent transcriptome.,” Molecular Systems Biology, 2015
7. Martin et al., “Differential genetic interactions of yeast stress response MAPK pathways.,” Molecular Systems Biology, 2015.