Improved production of Taxol® precursors in S. cerevisiae using combinatorial in silico design and metabolic engineering

Background Integrated metabolic engineering approaches that combine system and synthetic biology tools enable the efficient design of microbial cell factories for synthesizing high-value products. In this study, we utilized in silico design algorithms on the yeast genome-scale model to predict genomic modifications that could enhance the production of early-step Taxol® in engineered Saccharomyces cerevisiae cells. Results Using constraint-based reconstruction and analysis (COBRA) methods, we narrowed down the solution set of genomic modification candidates. We screened 17 genomic modifications, including nine gene deletions and eight gene overexpressions, through wet-lab studies to determine their impact on taxadiene production, the first metabolite in the Taxol® biosynthetic pathway. Under different cultivation conditions, most single genomic modifications resulted in increased taxadiene production. The strain named KM32, which contained four overexpressed genes (ILV2, TRR1, ADE13, and ECM31) involved in branched-chain amino acid biosynthesis, the thioredoxin system, de novo purine synthesis, and the pantothenate pathway, respectively, exhibited the best performance. KM32 achieved a 50% increase in taxadiene production, reaching 215 mg/L. Furthermore, KM32 produced the highest reported yields of taxa-4(20),11-dien-5α-ol (T5α-ol) at 43.65 mg/L and taxa-4(20),11-dien-5-α-yl acetate (T5αAc) at 26.2 mg/L among early-step Taxol® metabolites in S. cerevisiae. Conclusions This study highlights the effectiveness of computational and integrated approaches in identifying promising genomic modifications that can enhance the performance of yeast cell factories. By employing in silico design algorithms and wet-lab screening, we successfully improved taxadiene production in engineered S. cerevisiae strains. The best-performing strain, KM32, achieved substantial increases in taxadiene as well as production of T5α-ol and T5αAc. These findings emphasize the importance of using systematic and integrated strategies to develop efficient yeast cell factories, providing potential implications for the industrial production of high-value isoprenoids like Taxol®. Supplementary Information The online version contains supplementary material available at 10.1186/s12934-023-02251-7.


in silico Design and Analyses
Our methodology utilized three design algorithms (OptKnock, OptGene, and OptForce) in conjunction with the latest yeast genome scale model 8.5.0.We conducted simulations under complete synthetic yeast medium conditions, using either glucose or galactose as carbon sources, and targeted overproductions of cytosolic acetyl-CoA or GGPP.
OptKnock identifies reaction knock-outs to enhance target metabolism [1], while OptForce can predict reaction upregulations, downregulations, or knock-outs [2].On the other hand, OptGene suggests gene deletions for optimising target compound production [3].When OptKnock and OptForce suggested reactions, we considered the associated genes for those reactions.
These algorithms can also suggest a set of genomic modifications.However, we adopted a combinatorial approach in this study.We pooled predicted genes (or corresponding genes of reactions) based on specific criteria.Reactions not associated with any genes were excluded from further analysis.For instance, OptKnock suggested deleting reaction r_4362 (a dipeptidase reaction converting Gly-Met to L-methionine and L-glycine in vacuole) and r_4483 (glycine transport from cytosol to vacuole) to improve GGPP production in glucose-containing medium.However, these reactions lack associated genes in the model, so we disregarded such predictions for wet-lab validation.
To streamline the labour and cost involved, we eliminated reactions associated with multiple genes.For example, OptKnock recommended knocking out r_0832 for GGPP overproduction in glucose-containing medium.However, r_0832 involves three genes (YDR148C, YFL018C, and YIL125W) for knockout, so we focused on reactions associated with single genes.Furthermore, we did not consider reaction downregulations suggested by the OptForce framework, as fine-tuning reaction rates might require alternative strategies like enzymatic inhibitions or gene regulations.
To combine modifications from different frameworks, we shortlisted the most promising candidates through further analysis.We determined maximum potential fluxes using flux variability analysis (FVA) algorithm [4] and assessed flux improvements by constructing a metabolite interaction network [5,6].Subsequently, we experimentally evaluated 17 genomic modifications, consisting of nine gene deletions and eight gene overexpressions.The secondlevel strains were engineered based on the experimental findings.
In our simulation, the inorganic components of the complete synthetic medium (CSM) were left unconstrained, while the uptake rates of carbon sources-glucose or galactose-were maintained at a constant level, as these are potential limiting factors in the growth medium.Specifically, we allowed unconstrained uptake of the inorganic compounds listed in Table S6, which include phosphate, sulfate, ammonium, oxygen, sodium, potassium, chloride, copper, manganese, zinc, magnesium, calcium, iron, and hydrogen.

Figure S3 :
Figure S3: Acetyl-CoA-centred sub-metabolite interaction map of the wild-type model

Figure S10 :
Figure S10: de novo purine nucleotide biosynthesis pathway.The orange circles represent the metabolites involved in the reactions, and the genes are shown in purple.The underlined genes show the overexpressed genes.The arrows indicate the direction of the fluxes.

Figure S11 :
Figure S11: Phosphopantothenate biosynthesis pathway.The orange circles represent the metabolites involved in the reactions, and the genes are shown in purple.The underlined genes show the overexpressed genes.The arrows indicate the direction of the fluxes.

Figure S12 :
Figure S12: Dissolved oxygen (DO) concentrations of three-day cultures of KM1-derived strains measured by the BioLector microbioreactor system in the galactose-containing CSM.

Figure S13 :
Figure S13: Dissolved oxygen (DO) concentrations of three-day cultures of EJ1-derived strains measured by the BioLector microbioreactor system in the glucose-containing CSM.

Figure S14 :
Figure S14: Gas chromatography shows the compounds' peaks produced by KM32 and the mass spectrum of taxadiene.The retention time of the taxadiene peak was at 7.27 th minutes.

Figure S15 :
Figure S15: Gas chromatography shows the compounds' peaks produced by KM32 and the mass spectrum of T5α-ol.The retention time of the T5α-ol peak was at 8.20 th minutes.

Table S1 :
The primer list used to produce the donor DNA parts for the gene deletions ** Plus (+) sign shows the overlapping neighbour parts targeted by the red sequences *** If the overlapping sequences (red) show higher affinity than the annealing sequences (black) for the same DNA template, only annealing parts should be used first to amplify the target regions.****For:forward, Rev: reverse, UHA: upstream homology arm, DHA: downstream homology arm, Col: for colony PCR (coupled with corresponding UHA For) TableS2: crRNA sequences used to target the corresponding genes for deletions * The red nucleotides represent the corresponding PAM sequences

Table S3 :
The primer list used to produce the donor DNA parts for the genomic integrations The red sequences show the overlapping fragments with neighbour parts, while the black sequences show the annealing fragments.** Plus (+) sign shows the overlapping neighbour parts targeted by the red sequences *** If the overlapping sequences (red) show higher affinity than the annealing sequences (black) for the same DNA template, only annealing parts should be used first to amplify the *

Table S4 :
The yeast strains used in the study

Table S5 :
[7] gene candidates predicted by the design algorithms.17 of them (in bold) were prioritised by in silico analyses, maximum potential flux and metabolite interaction maps.The rest of the gene candidates were not used in the experimental studies.The gene candidates suggested by OptGene or the genes of the related reactions targeted by OptKnock and/or OptForce.Both the gene codes and their scientific names are shown.**Thesegene were not used in this study as they were previously integrated into our engineered yeast strain LRS6[7].Experimentally tested gene candidates are shown in bold. *

Table S6 :
The components of the yeast nitrogen base used in the study Figure S2: The ACtivE method