Model-driven evaluation of microbial physiology: insights from protein allocation

• Ano de defesa: 2024
• Autor(a) principal: Ferreira, Maurício Alexander de Moura
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa; Microbiologia Agrícola
• Palavras-chave em Português: Micro-organismos - Fisiologia; Sistemas biológicos; Engenharia metabólica; Aprendizado do computador; Microbiologia Agrícola
• Link de acesso: https://locus.ufv.br/handle/123456789/32578; https://doi.org/10.47328/ufvbbt.2024.425

The optimal allocation of proteins to cellular functions is crucial for cell survival and growth. However, the strategies employed by the cell are still elusive, as there are many supposedly conflicting objectives to be considered, such as minimizing the expenditure of resources, while at the same time affording to produce certain enzymes in excess, despite the lower demand for enzyme resources to maintain a certain amount of metabolic flux. Further, certain phenotypes, such as the overflow metabolism, are triggered by changes in resource distribution. In order to tackle these problems, the thesis focuses on the usage of protein-constrained metabolic models in combination with machine learning and integration with multi-omics data. Based on these approaches, here it is predicted the occurrence of overflow metabolism in the form of respiro-fermentative metabolism in the yeast Kluyveromyces marxianus. By integrating the metabolic model of K. marxianus with transcriptomics data, new insights on the genes, enzymes and metabolites involved in ethanol stress were obtained. Next, it is presented a new approach for studying enzyme usage redistribution, PARROT, which minimizes the distance between enzyme usage of an initial growth condition and a changing growth condition, based on the principle of minimal adjustment. The PARROT approach was able to predict enzyme usage in alternative growth conditions with higher accuracy than previous methods. While this approach is useful for studying resource redistribution, it is still not able to predict in vivo protein concentrations, given that the predicted usage is limited to a given metabolic flux and catalytic efficiency. To solve this problem, an approach that combines machine learning with metabolic modelling was developed, termed CAMEL. This approach could accurately predict in vivo concentrations, including for strains that were metabolically engineered. Finally, resource redistribution was evaluated on the context of enzyme promiscuity, from which a network of reactions termed “underground metabolism” can arise. To this end, the approach named CORAL was developed to integrate enzyme promiscuity constraints into metabolic models. It was found that these promiscuous enzymes are important for maintaining growth and providing robustness to disturbances in metabolism. The results obtained in this thesis are relevant to systems metabolic engineering endeavours, providing tools and knowledge to design microbial strains more suitable for industrial applications. Keywords: Systems biology; Metabolic engineering; Microbial physiology; Machine learning.