Deciphering the transcriptional regulatory network of Yarrowia lipolytica using machine learning

Abstract

The transcriptional regulatory network (TRN) in Yarrowia lipolytica coordinates its cellular processes, including the response to various stimuli. The TRN has been difficult to study due to its complex nature. In industrial-size fermenters, environments are often not homogenous, resulting in Yarrowia experiencing fluctuating conditions during a fermentation. Compared with homogenous laboratory conditions, these fluctuations result in altered cellular states and behaviors due to the action of the TRN. Here, a machine learning approach was deployed to modularize the transcriptome to enable meaningful description of its changing composition. To provide a sufficiently broad dataset, a wide range of relevant fermentation conditions (nutrient limitations, growth rates, pH values, oxygen availability and CO2 stresses) were run and samples obtained for RNA-Seq generation. We thus significantly increased the number of publicly available transcriptomic dataset on Y. lipolytica W29. In total, 23 independently modulated gene sets (termed iModulons) were identified of which 9 could be linked to corresponding regulons in S. cerevisiae. Strong responses were found in relation to oxygen limitation and elevated CO2 concentrations represented by (i) altered ribosomal protein synthesis, (ii) cell cycle disturbances, (iii) respiratory gene expression, and (iv) redox homeostasis. These results provide a fine-grained systems-level understanding of the Y. lipolytica TRN in response to industrially meaningful stresses, providing engineering targets to design more robust production strains. Moreover, this study provides a guide to perform similar work with poorly characterized single-cellular eukaryotic organisms. Highlights- A large screening setup significantly expands the public RNA-Seq library on Y. lipolytica. - This work provides a systems-level understanding of the TRN in response to industrial stresses. - iModulon analysis can help accelerate bioprocess development. - Results can guide similar work on poorly characterized single-cellular eukaryotic organisms.