Metabolic engineering of oleaginous yeast Rhodotorula toruloides for overproduction of triacetic acid lactone

Abstract The plant‐sourced polyketide triacetic acid lactone (TAL) has been recognized as a promising platform chemical for the biorefinery industry. However, its practical application was rather limited due to low natural abundance and inefficient cell factories for biosynthesis. Here, we report the metabolic engineering of oleaginous yeast Rhodotorula toruloides for TAL overproduction. We first introduced a 2‐pyrone synthase gene from Gerbera hybrida (GhPS) into R. toruloides and investigated the effects of different carbon sources on TAL production. We then systematically employed a variety of metabolic engineering strategies to increase the flux of acetyl‐CoA by enhancing its biosynthetic pathways and disrupting its competing pathways. We found that overexpression of ATP‐citrate lyase (ACL1) improved TAL production by 45% compared to the GhPS overexpressing strain, and additional overexpression of acetyl‐CoA carboxylase (ACC1) further increased TAL production by 29%. Finally, we characterized the resulting strain I12‐ACL1‐ACC1 using fed‐batch bioreactor fermentation in glucose or oilcane juice medium with acetate supplementation and achieved a titer of 28 or 23 g/L TAL, respectively. This study demonstrates that R. toruloides is a promising host for the production of TAL and other acetyl‐CoA‐derived polyketides from low‐cost carbon sources.

such as Yarrowia lipolytica and Rhodotorula toruloides (also known as Rhodosporidium toruloides) can be used for efficient TAL production due to their potential high flux through the key polyketide precursors, acetyl-CoA, and malonyl-CoA (Abdel-Mawgoud et al., 2018;Park et al., 2018). As a well-known lipid producer, Y. lipolytica was chosen for TAL biosynthesis via heterologous expression of 2-PS (Yu et al., 2018), and the best engineered Y. lipolytica strain achieved a titer of 35.9 g/L TAL in 280 h and a yield of up to 43% of the theoretical yield from glucose (Markham et al., 2018). Rhodotorula toruloides is an oleaginous basidiomycete yeast, which can grow on various sugars and produce a broad range of lipid and nonlipid chemicals (Jagtap & Rao, 2018;. Compared with Y. lipolytica, R. toruloides has a greater substrate range and natively produces triglycerol (TAG) at much higher titers (Jagtap & Rao, 2018;, but was less explored as a result of the unannotated genome sequence (Coradetti et al., 2018;Zhu et al., 2012) and lack of sophisticated genetic tools (Park et al., 2018). Nevertheless, the recent progress in the characterization of constitutive promoters (Nora et al., 2019;Y. Wang et al., 2016), development of CRISPRbased genome editing tools Otoupal et al., 2019;Schultz et al., 2019), RNA interference tool (X. , genome-scale model (Dinh et al., 2019), and functional genomics (Coradetti et al., 2018) enable us to perform metabolic engineering of R. toruloides for production of value-added compounds (Wen et al., 2020), specifically TAL.
In this study, we first expressed codon-optimized 2-PS genes from various organisms in R. toruloides and investigated the production of TAL under different culture conditions. We then created and characterized a broad set of TAL-producing overexpression and knockout gene targets in R. toruloides IFO0880. After combinatorial optimization of various targets, our final strain of I12-ACL1-ACC1 achieved a maximum titer of 28 g/L within 120 h in fed-batch fermentation from glucose with acetate addition. Then, we demonstrated the feasibility of olicane juice, an inexpensive carbon source as the substrate for TAL production, which produced 23 g/L TAL in fed-batch fermentation. This study not only establishes R. toruloides as a novel host organism for TAL biosynthesis but also demonstrates its potential as a biotechnological chassis for the production of high-value chemicals from low-cost substrates.

| Plasmids for 2-PS expression
The codon-optimized 2-PS genes were synthesized with two homologous ends to MfeI and SpeI digested pGI2 (Abbott et al., 2013; backbone, which contains nourseothricin resistance (NAT R ) for yeast, and then assembled with pTEF1 promoter and T35S terminator (pGI2-TEF1-X) for gene expression by Gibson assembly in E. coli (Gibson et al., 2009).

| Plasmids for overexpression of metabolic gene targets
The plasmid pRTG2-X (X represents gene expression targets) for gene targets expression was constructed based on a previously developed plasmid pRTN , which contains the E. coli genetic elements of pUC19 (pMB1 origin, ampicillin resistance), the S. cerevisiae genetic elements of pRS426 (2µ origin and URA3 selection marker), the strong R. toruloides p17 or pANT promoter, the target gene, the T35S terminator, and a R. toruloides G418 resistance (G418 R ) cassette from NM9 (Johns et al., 2016;Schultz et al., 2019) using DNA assembler . The multiple gene expression plasmid pRTHyg-X was pieced together from NM8 (Schultz et al., 2019) for hygromycin resistance (Hyg R ), pANT promoter, and Tncbt for gene expression.

| Plasmids for gene target knockout
The previously constructed plasmid pRTH-X (X represents gene knockout targets) was used for gRNA cloning and expression , which contains the E. coli genetic elements (pMB1 origin, ampicillin resistance), the S. cerevisiae genetic elements of pRS426 (2µ origin and URA3), a gRNA expression cassette with the IFO0880 5S ribosomal RNA, tRNA Tyr , N20 (targeting the first 10% open reading frame), the S. cerevisiae SUP4 terminator, and a R. toruloides Hyg R cassette from pZPK-PGPD-HYG-Tnos (Lin et al., 2014). For multiple gene knockout, the plasmid pRT-Cas9-SgRNA-Hyg was

| Yeast transformation
Most of the linear fragments for R. toruloides transformation were generated by PCR amplification of the genes or gRNA expression cassettes, together with the selection markers using the primers ZPK F/R, or gRNA F/R (Supporting Information: Table S1), respectively, whereas the fragments with sizes larger than 8 kb (e.g., ACC1-G418/Hyg) were excised from the plasmids by restriction enzyme digestion. Fragments were then cleaned using DNA Clean & Concentrator-5 Kit (Zymo Research) before transforming to R. toruloides.
Briefly, a single colony was picked and cultured overnight at 30°C

| Culture tube or shake flask fermentation
Three to five single colonies were randomly picked from each plate and cultured for 48h at 30°C in 3 ml YPD liquid medium supplemented with appropriate antibiotics in 14 ml culture tubes (VWR) for initial screening. Samples were collected by centrifugation and diluted 20 times for high-performance liquid chromatography (HPLC) analysis. Then, the strain with a production level close to the average titer of TAL was selected and recultured in YPD for 24 h as the seed culture. Fermentations were inoculated from seed culture to media with alternative carbon sources at an initial OD 600 of 0.2 and grown for an additional 72 h before sample preparation. For acetate spike, filter-sterilized 20× NaAc was added to the media at 12 h. Since a deficient cell growth and tiny amounts of TAL were obtained in YP (1% yeast extract and 2% peptone) medium, the yield was calculated based on the produced TAL over all carbons of sugars and/or acetate.

| Fed-batch fermentation
For fed-batch fermentation in bioreactors, single colonies of R.
toruloides I12-ACL1 and I12-ACL1-ACC1 were used to inoculate shake flask cultures with each containing 50 ml of YPAD medium supplemented with appropriate antibiotics, as described in For the fed-batch fermentation experiments with oilcane, the original oilcane feed solution containing a total sugar of 152 g/L (68.9 g/L glucose, 61.6 g/L fructose, and 21.5 g/L sucrose) was concentrated to about 450 g/L total sugar by evaporation through the boiling at atmospheric pressure. The concentrated oilcane feed solution was further autoclaved at 121°C for 30 min before it was used to provide initial sugars in the medium and to feed sugars during the fed-batch fermentation. The initial medium contained a total oilcane sugar of 50 g/L and all other medium components, as described previously for the glucose fed-batch fermentation. The oilcane feeding started when the initial sugars were depleted, as indicated by a sharp decrease in agitation speed and an increase in pH value. Oilcane was fed to control the residual glucose concentrations within 0-10 g/L. All other fermentation conditions, including acetate feeding, were the same as previously described for the fed-batch fermentation experiments with glucose.

| Analytical methods
Samples were prepared by diluting in methanol to the linear range, vortex mixing, and centrifuging at 16,000 g for 5 min to remove cells.
After being filtered by a 0.2 μm filter, the supernatant was injected into the HPLC for TAL, sugars, and acetate analyses.

| Rhodotorula toruloides can serve as a TAL producer
High lipid production in oleaginous organisms like R. toruloides suggests a great potential for these organisms to synthesize alternative acetyl-CoA-derived products, specifically type III polyketide, TAL (Markham et al., 2018;Park et al., 2018;Wen et al., 2020).
It was reported that R. toruloides could grow normally under harsh conditions  or in cultures with nonnative products, including fatty alcohols (D. Liu et al., 2020), fatty acid ethyl esters (Y. Zhang et al., 2021), and limonene (S. Liu et al., 2021). The product tolerance assay also showed that R. toruloides possessed a similar growth profile in YPD and YPD with 5 g/L TAL, but longer lag and log phases in YPD with 7 g/L TAL supplementation (Supporting Information: Figure S1). Therefore, R. toruloides can be potentially engineered to produce high titers of TAL without significantly detrimental growth effects.
As the 2-PS gene tested for TAL biosynthesis in yeast and bacteria was mainly from GhPS (Uniprot ID:

| TAL production using various substrates
To evaluate the effects of substrates on TAL production, we chose the commonly used carbon sources, including xylose (X), glycerol (G), sucrose (S), complete synthetic medium (SC), and glucose/xylose mixed sugar (DX) (Figure 1a). The results showed that YPD was a more preferred medium than SC, with a 4-fold higher TAL titer; compared to glucose, glycerol and glucose/xylose produced 5%-10% higher TAL, while xylose and sucrose decreased TAL production. We also observed deficient cell growth and residual sugars in YPX, YPS, and SC media (Supporting Information: Table S3), indicating a positive correlation between cell growth and TAL titer.
It has been demonstrated that acetate feeding was beneficial to acetyl-CoA supply and TAL biosynthesis in S. cerevisiae and Y. We, therefore, supplemented 0.5%, 1%, and 2% NaAc to YPD with the 12 h R. toruloides cell culture and observed significant improvements, 67%-80% higher TAL production under these spiking conditions, representing a similar titer to that of YP2D (3.6 g/L TAL from YP-4% glucose at 72 h) and~30% of theoretical maximum yield (0.47 g/g) calculated from both glucose (2%) and acetate (0.5%) in culture tube (Figure 1b). To explore the potential role of acetate during TAL biosynthesis, we provided 2% NaAc as an alternative carbon source.
However, the TAL produced from YP-NaAc was only 42% of that produced from YPD (Figure 1b). The residual amounts of NaAc were also measured for the above-mentioned NaAc media, and only YPD + 0.5%NaAc showed depletion of acetate after fermentation, while a portion of acetate left in YPD + 1%NaAc, YPD + 2% NaAc, and YP-NaAc (Supporting Information: Table S3). The acetate consumption indicated that acetate may not only act as a substrate for TAL production but also be associated with the redox and regulatory mechanism, which has been elaborated in Y. lipolytica (Markham et al., 2018).

| Single gene target engineering to improve TAL production
It is generally recognized that malonyl-CoA is the limiting precursor for polyketide synthase (Xu et al., 2011;Zha et al., 2009). Therefore, we overexpressed the endogenous acetyl-CoA-carboxylase (ACC1) to test whether the conversion of acetyl-CoA to malonyl-CoA would facilitate TAL synthesis ( Figure 2). As shown in Figure 3a, overexpression of ACC1 did not markedly enhance TAL production with only 6% improvement in YP2D at 120 h compared to that of the starting strain I12. To further drive the condensation of acetyl-CoA and malonyl-CoA, we introduced a second-round genome integration of GhPS gene and achieved 2.2 g/L TAL in YPD at 72 h (Supporting Information: Figure S3) and 4.8 g/L TAL in YP2D at 120 h, respectively,~11% higher than that of I12 (Figure 3a). Based on these results, we deduced that the limiting precursor for TAL overproduction was acetyl-CoA instead of malonyl-CoA. Therefore, we sought to increase the flux of acetyl-CoA by enhancing its biosynthetic pathways and disrupting its competing pathways.
(1) Enhancing acetyl-CoA biosynthetic pathways: We explored three distinct metabolic engineering strategies and characterized the roles of associated gene targets in TAL production ( Figure 3).
First, we investigated the pyruvate dehydrogenase (PDH) complex pathway and overexpressed its subunits, E1 and E3 (LPD1) in I12 strain. Fermentation showed that both E1 and E3 overexpression improved TAL production by~16%, which reached~5.3 g/L at 120 h (Figure 3a and Supporting Information: Table S4). It is known that the PDH complex is located in the mitochondrial matrix in eukaryotes, and its compartmentalization is mediated via mitochondrial targeting sequence (MTS). However, the improved TAL production indicated that the tested subunits of E1 and E3 may not contain a fully featured MTS, or there may be a leaky expression of these two subunits in the cytoplasm, which was similar to that of the overexpression of PDH complex in S. cerevisiae (Lian et al., 2014) and Y. lipoytica (Markham et al., 2018). As we failed to construct the mutants of other subunits of PDH, overexpression of the complete PDH or a cytoplastic PDH complex (i.e., E. coli cytoPDH) (Cardenas & Da Silva, 2016;Kozak et al., 2014) can be a potential strategy to increase the acetyl-CoA level.
Second, we evaluated the PDH bypass pathway, which converts pyruvate to acetyl-CoA through a three-step reaction sequentially catalyzed by pyruvate decarboxylase (PDC), acetaldehyde dehydrogenase (ALD), and acetyl-CoA synthetase (ACS).
However, no TAL improvement was achieved by ACS1 overexpression, which is similar to the study in S. cerevisiae where ACS1 overexpression did not improve n-butanol production because of low activity or posttranslational deactivation (Lian et al., 2014).
Third, we explored the citrate route, a pathway that generates cytosolic acetyl-CoA from citrate and was reported to be present only in oleaginous yeasts (Pomraning et al., 2019;Vorapreeda et al., 2012;Zhu et al., 2012). The pathway gene ACL1, encoding ATP-citrate lyase, has been overexpressed to increase lipid production in Y. lipolytica (Blazeck et al., 2014;G.-Y. Wang et al., 2015). A multiomic analysis of R. toruloides also revealed that ACL1 was expressed at an extremely high level during the lipogenesis stage (Zhu et al., 2012). Therefore, the endogenous ACL1 was overexpressed in I12 strain (Figure 2), and the TAL production was dramatically improved by 45%, to 6.6 g/L in YP2D in a test tube at 120 h (Figure 3a and Supporting   Information: Table S4), which was~35% of the theoretical yield.
In addition, we overexpressed metabolic targets that could indirectly increase the metabolic flux of acetyl-CoA, including AMPD1 (encoding AMP deaminase) (X.-K. Zhang et al., 2019), ME1 (encoding malic enzyme), PEX10 (encoding peroxisomal matrix protein), and YLACL1 (ACL1 from Y. lipolytica and sequence included in Supporting Information: Table S2) (Blazeck et al., 2014) (Figure 2). The results showed that PEX10 and YLACL1 overexpression increased TAL titer by 12% and 11%, respectively, whereas AMPD1 decreased TAL titer and ME1 had no effect on TAL production (Figure 3a). This suggests that upregulation of β-oxidation by enhancing peroxisome biogenesis through PEX10 overexpression can be an alternative way to recycle acetyl-CoA for TAL production in R. toruloides.
(2) Disrupting acetyl-CoA competing pathways: Removing acetyl-CoA consuming pathways was demonstrated as an effective way to increase the availability of acetyl-CoA. In yeast, the glyoxylate shunt allows acetyl-CoA to be converted into a C4 carbon without carbon loss (Dolan & Welch, 2018). Therefore, we performed the inhibition of two key reactions of the glyoxylate cycle, namely peroxisomal citrate synthase, encoded by CIT2, and cytosolic malate synthase, encoded by MLS1 (Chen et al., 2013), by a previously developed CRISPR/Cas9 method (Schultz et al., 2019). As shown in Figure 3b, compared with I12-Cas9 strain, the deletion of CIT2 and MLS1 improved TAL production by 14% and 20%, respectively.
In addition, we investigated the effects of disrupting other gene targets, including two acyltransferases (encoded by DGA1/LRO1), pyruvate carboxylase (encoded by PYC1), serine esterase or patatindomain-containing protein (encoded by NTE1), and mitochondrial NAD + transporter (encoded by YIA6). Among them, DGA1 and LRO1 are involved in TAG formation in Y. lipolytica (Athenstaedt, 2011), and PYC1, NTE1, and YIA6 were reported to improve TAL production in S.
cerevisiae (Cardenas & Da Silva, 2014). The fermentation showed that the deletion of DGA1 and LRO1 improved TAL titer by 11% and 19%, respectively, while the deletion of NTE1, YIA6, and PYC1 had a marginal effect on TAL production (Figure 3b), which is inconsistent with the observation in S. cerevisiae.

| Multiple gene target engineering to improve TAL production
To further investigate the effects of multiple gene targets on TAL production in a combinatorial manner, we selected the top targets that improved TAL production more than 12%, that is, ACL1, ALD5, MLS△, LRO1△, PDH-E3, PDH-E1, CIT2△, PDC1, and PEX10 for the second round of metabolic engineering based on I12-ACL1 strain. In addition, we included ACC1 as its overexpression may result in improved malonyl-CoA concentration in an acetyl-CoA enhanced strain, I12-ACL1. We successfully obtained the mutant strains that overexpressed ACL1, ALD5, PDH-E3, PDH-E1, PDC1, PEX10, and ACC1 through random genome integration using HYG selection, and the optimal combination was ACL1-ACC1, which produced 6.9 g/L TAL in YP2D at 120 h, representing a 29% improvement compared with I12-ACL1 strain (Figure 3c and Supporting Information: Table S4). Unfortunately, we failed to obtain the correct mutants with MLS1, LRO1, and CIT2 deletion after transforming linear fragments containing Cas9-SgRNA-Hyg into I12-ACL1 strain.
Although a decent number of colonies were growing on YPD + HYG plates, which meant the Hyg expression cassette was integrated into genome, none of the colonies had the expected genome mutation or indel after sequencing-based genotyping. Compared to the high genome editing (gene knockout) efficiency through transforming SgRNA-Hyg to I12-Cas9 strain, the low editing efficiency in I12-ACL1 may be caused by the coexpression of Cas9 and SgRNA in a single fragment of Cas9-SgRNA-Hyg, which could result in the toxicity or lethality to the host cells. Therefore, a two-step transformation of individual Cas9 and SgRNA could be beneficial for genome editing in I12-ACL1, and future efforts will be made to develop other antibiotic or auxotrophic selection markers in R. toruloides, or introduce recyclable expression platforms, such as a replicable episomal plasmid (Schultz et al., 2021) or a Cre/loxP site-specific recombination system (Díaz et al., 2018).

| Scale-up TAL production in bioreactor
To evaluate the possibility of scaling up fermentation, fed-batch fermentation was performed using strains I12-ACL1 and I12-ACL1-ACC1, which showed high TAL titers in culture tube fermentations.
The results showed that the cell OD 600 and TAL production with 5 g/L NaAc addition was higher than that with 10 g/L NaAc (Supporting Information: Figure S4), and I12-ACL1-ACC1 produced 7.1 g/L TAL at 96 h, which was 40% more than that of I12-ACL1 in YP2D + 0.5% NaAc. The yield of TAL production in YP2D + 0.5% NaAc for I12-ACL1-ACC1 reached 35% of the theoretical yield, which was similar to culture tube fermentation of I12-ACL1-ACC1 in YP2D.
However, it took 24 h less fermentation time, indicating a higher oxygen concentration may benefit TAL production. The production of intracellular lipids by I12-ACL1-ACC1 was also characterized in shake flask media of YP2D and YP2D + 0.5% NaAc, and it showed that TAG and phospholipids were the main components of lipids in YP2D + 0.5% NaAc and YP2D, reaching 41.0 mg/g dry cell weight (DCW) and 33.2 mg/g DCW at 96 h (Supporting Information: Figure S5A), respectively. Meanwhile, the titer of TAG, 0.55 g/L was achieved in YP2D + 0.5% NaAc (Supporting Information: Figure S5B), which was only 7% of that of TAL (7.9 g/L), indicating TAL is the main product for I12-ACL1-ACC1.
For bioreactor fermentation, R. toruloides strain I12-ACL1-ACC1 was set up for scale-up using fed-batch cultures in a medium containing yeast extract, peptone, glucose, and other trace metals, and produced 28 g/L of TAL from glucose-based medium at 118 h, representing a high volumetric productivity of 0.24 g/L/h ( Figure 4a,b, and Supporting Information: Table S5). The highest yield of 0.079 g TAL/g carbon source (glucose and acetate) for I12-ACL1-ACC1 was also achieved at 118 h, which was 17% of the theoretical yield. It was found that~40 g/L glycerol was accumulated at 118 h, indicating that glycerol was the main byproduct, and an enhanced glycerol utilization pathway could be used to improve TAL production. Furthermore, to demonstrate the feasibility of fermentation using low-cost feedstocks and to take advantage of the capability of R. toruloides to assimilate a diverse range of substrates, including monosaccharides, oligosaccharides, and organic acids, we performed fed-batch fermentation using oilcane juice, a first-generation feedstock which is comprised of 152 g/L total sugars. As shown in  Table S5), while the productivity was 0.19 g/ L/h before 120 h. Although the yields in glucose-and oilcane juicebased fed-batch fermentation were relatively low, the volumetric productivities in both conditions were much higher than that in culture tube (0.058 g/L/h) or shake flask (0.074 g/L/h), and even higher than that reported in Y. lipolytica (0.12 g/L/h) (Markham et al., 2018). Meanwhile, the fed-batch fermentation provided us with more knowledge about improving TAL production by optimizing fermentation conditions, including feeding rates, DO/pH control (Sun et al., 2021), and in situ product separation (Lee et al., 2016).

| CONCLUSION
In this study, the codon-optimized 2-PS gene (GhPS) was introduced into R. toruloides for TAL biosynthesis, and the resultant strain I12 produced~2 g/L TAL in a culture tube. The dramatic improvement of TAL production by acetate addition suggests that acetate can not only serve as a substrate but also stimulate TAL production. Although the integration loci or genome architecture may affect the specific gene expression profile, it was found here that ACL1, a citrate route enzyme, could be a preferred gene target to accumulate acetyl-CoA flux, and its overexpression improved TAL titer by 45% compared to I12 strain, which was 35% of the theoretical yield. The concurrent expression of ACL1 and ACC1 further improved TAL by 29% in the culture tube, and the scale-up bioreactor fermentation achieved 28 or 23 g/L TAL from glucose or low-cost oilcane juice with acetate spike, respectively. This study demonstrates that R. toruloides represents a promising microbial cell factory for the production of polyketides and other acetyl-CoA-derived chemicals.

DATA AVAILABILITY STATEMENT
The data that supports the findings of this study are available in the Supporting Information: Material of this article.