Methods for predictable and accelerated engineering of metabolism in eukaryotes

Wealth increases and a growing world population necessitates sustainable methods for producing biomolecules within broad fields like fuels, chemicals, foods, feeds, and pharmaceuticals. The engineering of microorganisms to produce biomolecules from sustainable feedstock can be part of the solution. The development of organisms for bio-production can be considered as an iterative process with four major steps called Design, Build, Test, and Learn (DBTL). This process can be accelerated by the development of new and improved methodologies. I have worked with all four steps in the DBTL cycle and have successfully contributed to developing new tools for regulating the production of selected metabolites. In the work presented in paper no. one (chapter no. 2 in this thesis), 8 hexameric sequences were identified in the untranslated part of the translation initiation site that leads to variable production of a green fluorescent protein (GFP) in yeast. It was shown that these hexameric sequences can be used for predictive and context-independent tuning of protein expression from yeast to CHO cells. Thus, the method has the potential to become a unified tool across a large group of organisms. The carotenoid production was effectively regulated by combining three of the eight hexamers and inserting them in front of the genes crtE and erg9 which are known to code for key enzymes in the carotenoid synthesis pathway.

In paper no. 2 (chapter 3) the aim was to improve genotype-to-phenotype predictions of tryptophan synthesis rate in yeast as a tool for metabolic engineering. Genome scale modeling and literature was used in the identification of five gene targets for engineering, CRISPR/Cas9 facilitated genome editing for the construction of a cell library of combinatorial edits of promoters, high throughput methods for genotype and phenotype characterization, and state of the art machine learning (ML) models for genotype to phenotype prediction. The created library has at least a 5 fold variation in the tryptophan synthesis rate calculated as the ratio between the highest and lowest average rate for individual strains. It was possible to fit different ML models closely to the observed library strains, and these models could also be used to predict efficient strains not seen in the library. Thus, a strain was identified with a tryptophan synthesis rate that was 17 %
larger than the highest rate found in the library and 106 % larger than that of an engineered, high producing platform strain. In future studies, further increases in the tryptophan synthesis rate can be expected by combining relevant edits of new genetic targets with the five presented in this study. The developed methods have the potential to become parts of more standardized workflows for leading metabolic engineering efforts in the routine development of cell factories.