Engineering cofactor supply and NADH-dependent d-galacturonic acid reductases for redox-balanced production of l-galactonate in Saccharomyces cerevisiae

d-Galacturonic acid (GalA) is the major constituent of pectin-rich biomass, an abundant and underutilized agricultural byproduct. By one reductive step catalyzed by GalA reductases, GalA is converted to the polyhydroxy acid l-galactonate (GalOA), the first intermediate of the fungal GalA catabolic pathway, which also has interesting properties for potential applications as an additive to nutrients and cosmetics. Previous attempts to establish the production of GalOA or the full GalA catabolic pathway in Saccharomyces cerevisiae proved challenging, presumably due to the inefficient supply of NADPH, the preferred cofactor of GalA reductases. Here, we tested this hypothesis by coupling the reduction of GalA to the oxidation of the sugar alcohol sorbitol that has a higher reduction state compared to glucose and thereby yields the necessary redox cofactors. By choosing a suitable sorbitol dehydrogenase, we designed yeast strains in which the sorbitol metabolism yields a “surplus” of either NADPH or NADH. By biotransformation experiments in controlled bioreactors, we demonstrate a nearly complete conversion of consumed GalA into GalOA and a highly efficient utilization of the co-substrate sorbitol in providing NADPH. Furthermore, we performed structure-guided mutagenesis of GalA reductases to change their cofactor preference from NADPH towards NADH and demonstrated their functionality by the production of GalOA in combination with the NADH-yielding sorbitol metabolism. Moreover, the engineered enzymes enabled a doubling of GalOA yields when glucose was used as a co-substrate. This significantly expands the possibilities for metabolic engineering of GalOA production and valorization of pectin-rich biomass in general.


Results and discussion
Establishing a system for redox-balanced GalA reduction. Previous work by others has shown that, due to its high oxidation state, the reduction of GalA can occur only in the presence of a co-substrate (redoxdonor) such as fructose or glucose, but the molar yield of GalOA per mol co-substrate was low 9 . Presumably, the low yields were due to insufficient redox power provided by hexose metabolism. Therefore, we decided to couple the reduction of GalA to the oxidation of the sugar alcohol sorbitol, which exhibits a higher reduction state compared to glucose. Sorbitol must be oxidized to fructose by a sorbitol dehydrogenase (SDH) to enter glycolysis. Thereby, SDH can provide the necessary reducing equivalents in a stoichiometric manner (Fig. 1).
For co-utilization of sorbitol and GalA, different genetic cassettes, each for overexpression of four genes-a sorbitol transporter, an SDH, a GalA transporter and a GalA reductase-were constructed and integrated into the URA3 locus of the hexose-transporter deficient (hxt 0 ) strain EBY.VW4000 20 . This strain background was chosen to rule out any influence of endogenous hexose transporters on GalA uptake 10 . All cassettes contained the endogenous transporter HXT13 for sorbitol uptake 21 and GatA from A. niger 10 for GalA uptake. As GalA reductases, we chose the well-known enzyme TrGar1 and its orthologue from A. niger (AnGar1), for which a GalA reductase activity had not been demonstrated yet. To provide NADPH or NADH, we selected the YlSdr from Yarrowia lipolytica 22  www.nature.com/scientificreports/ To test the concept, we performed shake flask fermentations under aerobic conditions with these strains in sorbitol-containing media with or without GalA supplementation (Fig. 2).
In line with previous observations 21 , the (NAD + -dependent) Sor2 conferred robust growth and sorbitol consumption, both of which were markedly delayed by the addition of GalA. This is likely due to the weak acid toxicity and/or inhibition of sorbitol uptake, as Hxt13 is closely related to the yeast galactose permease Gal2 that was previously reported to be competitively inhibited by GalA 23 . With Sor2, only low concentrations of GalOA (up to 0.68 g L −1 ) were produced. The behavior of strains expressing the NADP + -dependent YlSdr was the opposite. Without GalA, only marginal growth and a concomitantly slow sorbitol consumption were measured. This can be explained by the accumulation of cytosolic NADPH that, in contrast to NADH, cannot be re-oxidized, leading to a growth arrest 24 . As expected, the addition of GalA stimulated growth and sorbitol consumption by acting as a redox sink for the accumulating NADPH (see Fig. 1). This is reflected by the substantially higher titer of GalOA in fermentations with YlSdr (up to 2.54 g L −1 ). For both GalA reductases, TrGar1 and AnGar1, resulting titers and molar yields per mol consumed co-substrate (Y P/S ) were similar. Despite slower growth and slower sorbitol consumption, these yields were about 12-fold higher with NADPH-than NADH-dependent SDH. Altogether, Figure 1. Redox-balanced system for GalOA production. The cofactors necessary for the reduction of d-galacturonate by GalA reductases (TrGar1, AnGar1 or AnGaaA) to GalOA can be derived from the oxidation of sorbitol by SDH. By choosing the suitable SDH, either NADH (Sor2) or NADPH (YlSdr) can be accumulated. Fructose produced by SDH subsequently enters glycolysis. www.nature.com/scientificreports/ these results show that the supply of suitable reducing equivalents is the most critical engineering target for GalOA production. Moreover, we demonstrate, for the first time, the activity of AnGar1 as a GalA reductase. Growth, sorbitol and GalA consumption, as well as GalOA formation of engineered S. cerevisiae were also studied in aerobic batch-operated, fully controlled stirred-tank bioreactors (e.g., pH, DO, power input). Figure 3 shows online and offline monitored state variables of the standardized batch process with the best performing strain of shake flask experiments (S. cerevisiae SiHY001 expressing YlSdr and AnGar1). By using pre-cultures grown in the same medium as for batch processes, the lag phase observed in shake flask experiments could be substantially reduced to a few hours. The highest metabolic activity was measured within the first 24 h of the batch process, reaching a CO 2 concentration of 0.25% (v/v) in the exhaust gas and a maximum growth rate of 0.12 h −1 . Finally, 0.45 mol GalOA was formed per mol sorbitol consumed (Y P/S ), indicating that more than 50% of sorbitol was used for CO 2 and biomass formation. Growth and sorbitol consumption rates started to decrease at a process time of 24 h, despite the high remaining sorbitol concentration in the reactor (5.7 g L −1 /31.3 mM). The decrease in the sorbitol consumption rate may be caused by the rather low affinity (K M = 20.4 mM) of Hxt13 for sorbitol 21 and/or accumulation of the possibly toxic product GalOA. Even without further growth of the S. cerevisiae cells beyond 72 h, GalOA was still formed until a final concentration of 4.4 g L −1 after a process time of 191 h. During the stationary phase (72-191 h), the yield coefficient for GalOA on sorbitol (Y P/S, 72-191 h ) increased to 0.79 mol mol −1 , which is comparable to the results observed in shake flasks (Fig. 2) where the growth of the yeast cells was reduced considerably compared to that in the stirred-tank bioreactor. The substrate-specific product selectivity (Y P/E ) of 0.95 mol GalOA per mol GalA (0.93 g GalOA g −1 GalA) indicates a nearly by-product free biotransformation and exceeds selectivities of filamentous fungi under similar conditions in batch processes (0.6-0.9 g GalOA g −1 GalA 15 ). The carbon balance of the batch process could be closed (100% considering an error range of ± 10%) ( Supplementary Fig. S3).
Engineering GalA reductases for NADH-specificity. To identify the amino acid residues responsible for NADPH binding, we generated the structural models of TrGar1 and AnGar1. The crystal structure of the NADPH-dependent aldehyde reductase AKR1A1 from Sus scrofa (PDB ID 1HQT) was the template; this enzyme shares 37% sequence identity with TrGar1 and AnGar1. The homology models of TrGar1 and AnGar1 were constructed in Molecular Operating Environment (MOE; Chemical Computing Group, https ://www. www.nature.com/scientificreports/ chemc omp.com/). Given the high sequence homology between TrGar1 and AnGar1 (63% identity and 81% similarity), their structural models are quite similar (Fig. 4A). In the active site, the phosphoryl group of NADPH interacts electrostatically with the two positively charged residues Lys (254 in TrGar1 or 261 in AnGar1) and Arg (261 in TrGar1 or 267 in AnGar1) (Fig. 4B). In MOE, we performed mutation scanning of these residues and examined the protein stability as well as NADH versus NADPH ligand affinity for the various substitutions ( Supplementary Fig. S4). According to the protein stability results, the lysine residue could be mutated to Leu, Met, Phe, Trp, or Tyr; among these, the mutant to Met also has a better affinity for NADH than for NADPH, which is predictable given that Met is similar in size to Lys, but without the positive charge ( Supplementary Fig. S4). A similar analysis for the arginine residue indicates that, while Ile, Leu, Thr, and Val are stable alternatives for its mutagenesis, the substitution with Leu offers better protein stability and ligand affinity with NADH ( Supplementary Fig. S4). Furthermore, both K261M and R267L preserve the space occupied by Lys and Arg, respectively, in the NADP binding site well. Therefore, we hypothesized that the amino acid substitutions K254M and R260L in TrGar1 or  www.nature.com/scientificreports/ the corresponding mutations in AnGar1 (K261M and R267L, respectively), alone or in combination, will result in the desired change of cofactor-specificity (Fig. 4C). The ability of GalA reductase variants to utilize different cofactors was tested using the sorbitol co-fermentation system as described above, with the exception that enzymes were expressed from multicopy (2µ) plasmids. AnGaaA, a phylogenetically non-related GalA reductase, which naturally accepts NADPH and, albeit to a lesser extent, NADH 18 was included for comparison.
As expected, the two wildtype GalA reductases could only produce a considerable amount of GalOA when in combination with YlSdr, i.e., with NADPH as a cofactor (Fig. 5). The lower titers observed with AnGaaA in comparison to TrGar1 correlate with specific activity differences of these enzymes in yeast, as previously reported 10 . Wildtype TrGar1 and AnGar1 produce comparable amount of GalOA. The mutated variants of TrGar1 showed a decreased GalOA production with NADPH, as expected, but no considerable product formation in the NADH background. In contrast, both single amino acid substitutions K261M and R267L conferred AnGar1 the ability to accept NADH in addition to NADPH. The double mutant K261M/R267L apparently confers NADHspecificity to AnGar1, since high amounts of GalOA are only produced in combination with Sor2. Importantly, the mutated AnGar1 variants are superior to AnGaaA with NADH, demonstrating the feasibility of the enzyme engineering approach.
To investigate the molecular basis of these in vivo observations, we performed in vitro enzyme assays of AnGar1 variants. For this, we selected the wildtype enzyme, the single mutant producing the higher GalOA amount in the NADH background (R267L) and the double mutant K261M/R267L. Consistent with in vivo results, the single mutation only partly and the double mutation almost fully abolished the AnGar1 activity with NADPH as the cofactor (Fig. 6A). When the assay was performed with NADH, all three variants showed comparable activities, which were by an order of magnitude lower compared to those with NADPH of the wildtype enzyme (Fig. 6A). This demonstrates that (also wildtype) AnGar1 accepts NADH, albeit with a substantially lower preference compared to NADPH. However, when NADP was added to competitively inhibit NADH binding, only the mutants were able to efficiently reduce GalA using NADH (Fig. 6B). Whereas the double mutant was fully insensitive to NADP at both NADH concentrations tested, the single mutant showed a considerable sensitivity at the lower (160 µM) but not at the higher (800 µM) NADH concentration. This is consistent with its ability to use NADPH as a cofactor, in contrast to the double mutant (Figs. 5, 6A). Together, these data demonstrate that the change of the cofactor preference is due to a dramatically decreased affinity towards NADP(H), which is in accordance with the expectations based on the structure model. In the cellular context, the wildtype enzyme cannot use NADH supplied by Sor2 due to the presence of NADP (which accumulates in the absence of the enzyme that can re-reduce it, i.e. in the absence of YlSdr). Conversely, the mutated variants can bind NADH, since they are not sensitive to the presence of NADP. This notion also explains why AnGar1 mutants produce more GalOA in vivo in the Sor2 background compared to AnGaaA, which also accepts NADH, but with a by an order of magnitude lower affinity (reflected by the K m value) compared to NADPH 18 .
To test the performance of yeast strains expressing mutated AnGar1 variants in a standardized batch processes, biotransformations were again studied in stirred-tank bioreactors. Here, the genes coding for the GalA reductase variants were integrated genomically as single copies for better stability 25 . As shown before in shake flasks, both singly (R267L) and doubly mutated (K261M/R267L) AnGar1 reductases could reduce GalA to GalOA. The S. cerevisiae strain with the singly substituted GalA reductase reached a final product concentration It should be noted that the comparison of the results achieved in shake flasks and stirred-tank bioreactors is somewhat biased due to different expression levels, as multicopy plasmids were used in the shake flask experiments and genomically integrated single copy constructs in the stirred-tank bioreactor. Moreover, the aeration conditions and, consequently, the respiratory re-oxidation of NADH likely vary between the bioreactor and shake flask fermentations. The biotransformation of GalA was strongly growth-dependent. GalOA formation stagnated when the stationary phase was reached at approximately 72 h in both batch processes (Fig. 7A,C). The lower molar yields of GalOA produced per mol sorbitol (Y P/S ) with NADH-dependent GalA reductase variants compared to the NADPH-dependent reductase may be caused by other NADH-consuming metabolic reactions in yeast. The difference in final product titers in batch processes making use of NADH-dependent GalA reductases might result from the reduced capability of the AnGar1 double mutant to utilize NADH and NADPH at the same time.
The estimated final product selectivity (Y P/E ) of the biotransformation with the AnGar1 double mutant (0.86 mol GalOA per mol GalA) is not significantly lower compared to the single mutant (Y P/E = 0.92 mol mol −1 ). The discrepancy, however, might come from the low final product concentrations of biotransformations with the double mutant being harder to distinguish analytically from the background noise in the UV-chromatogram of fermentation samples. The estimated integral carbon balance data in both batch processes could be closed (100%, considering an error range of ± 10%) (Supplementary Figures S5, S6).
Co-fermentation of GalA and glucose using engineered GalA reductases. Sorbitol and GalA cooccur in fruits 26 , but some major sources of pectin, such as sugar beet pulp, do not contain a large amount of the sugar alcohol. Instead, this feedstock contains glucose, galactose, and arabinose as major carbohydrates 19 . All these substrates are funneled into glycolysis, which produces NADH in the GAPDH reaction. In S. cerevisiae, the predominant fraction of NADH is re-oxidized by the production of ethanol even under aerobic conditions in the presence of glucose due to the Crabtree effect 27,28 . Thus, glycolysis is redox-neutral, and "superfluous" NADH, resulting from the production of biomass, is re-oxidized through the production of glycerol 29 . To minimize the re-oxidation of NADH, we decided to test the performance of our engineered GalA reductase variants in the strain JWY019, which lacks the main alcohol dehydrogenase ADH1 and the glycerol-phosphate-dehydrogenase genes GPD1 and GPD2 30 . In this strain background, the engineered enzymes indeed achieve higher GalOA titers (Fig. 8A), and molar yields per mol consumed glucose (Fig. 8B).
With the K261M/R267L mutant, the yields were more than doubled compared to those observed with the wildtype enzyme. However, they are still below those obtained on sorbitol. This is obviously due to the fact that, despite the deletion of ADH1, GPD1 and GPD2, NADH is still re-oxidized in JWY019 through the residual formation of ethanol (Fig. 8C) and glycerol (Fig. 8D). Strikingly, the production of glycerol is inversely correlated with the increased production of GalOA, supporting the conclusion that GalA acts as a redox sink for the glycolytic NADH. Our process optimization experiments (data not shown) and published results from others 9 show that the production of GalOA under anaerobic conditions is not possible in S. cerevisiae. This observation can likely be attributed to the increased energy demand for transport of GalA, which requires proton symport 31 . Furthermore, it is possible that the export of GalOA occurs in an energy dependent manner, further compromising PK2-1C cells. The cells transformed with the empty vector (EV) were used as a negative control. In (A), the assays were performed with NADPH or NADH alone. The specific activity (mili Units per mg protein, mU mg −1 ) is shown. The Y axis is divided in two segments to better visualize the lower activities. In (B), the assays were performed with NADH and NADP (oxidized form) as a competitive inhibitor at indicated concentrations. Shown are relative activities, calculated as percent of the activity measured at the respective NADH concentration in the absence of NADP. Error bars represent standard deviation of technical triplicates. n.d., not detectable.

Scientific Reports
| (2020) 10:19021 | https://doi.org/10.1038/s41598-020-75926-5 www.nature.com/scientificreports/ the energetic balance of the cell. Although further interventions into the NADH metabolism (e.g. the deletion of all alcohol dehydrogenase genes) are necessary to increase the yield of GalOA, our results demonstrate that NADH-dependent GalA reductases are useful to improve the co-fermentation of GalA with neutral sugars such as glucose or galactose.

Conclusion
Here, we demonstrated that the supply of suitable redox cofactors is a critical engineering target to enable efficient reduction of GalA in S. cerevisiae. This can be achieved by feeding a co-substrate exhibiting a high reduction state, such as sorbitol (or other polyols). We show for the first time that AnGar1 has a GalA reductase activity. Using the sorbitol-based screening system and structure-guided mutagenesis, we developed AnGar1 variants that, to our knowledge, represent the first reported GalA reductases with a higher preference for NADH compared to NADPH. The altered cofactor-specificity enables the coupling of GalA reduction to glycolysis, resulting in higher yields of GalOA when glucose is used as a redox donor. Hence, the engineered AnGar1 should prove valuable for GalA utilization in pectin-rich hydrolysates, which contain neutral sugars such as glucose, galactose, or arabinose, all of which are funneled into glycolysis. Moreover, the NADH-dependent GalA reductases could facilitate the coupling of GalOA production to the oxidation of glycerol, an abundant waste product that could be supplemented to pectin-rich hydrolysates.   Transformation of yeast cells. For yeast cell transformation, 50 mL YPD culture was inoculated with 1 mL of a YPD preculture and agitated at 200 rpm and 30 °C. The optical density was measured at 600 nm (OD 600 ). When OD 600 reached 0.8-1, the culture was pelleted at 3,000 × g for 3 min and washed with 25 mL sterile water. Cells equivalent to 5 OD 600 units were pelleted at 5,000 × g for 1 min and used for one transformation. , ethanol (C) and glycerol (D) were measured in culture supernatants by HPLC analysis. In (B) the molar yields of GalOA (mol per mol consumed glucose) were calculated after 9 days of cultivation. The difference between the wildtype and the double mutant is statistically significant (t-test P < 0.005), whereas the difference between the wildtype and the single mutant is not (P > 0.05). The corresponding glucose consumption and growth curves are shown in Supplementary Fig. S7. Mean values and standard deviations of biological triplicates are shown. The same color code applies to all panels. www.nature.com/scientificreports/ To this end, 240 μL 50% (w/w) polyethylene glycol, 36 μL 1 M lithium acetate, 10 μL ssDNA, and either 250 ng of plasmid-based or 5 µg of linear DNA in 64 μL water were added to the cells. The reaction set-up was mixed thoroughly and incubated at 42 °C for 20 min. Subsequently, the cells were pelleted at 5,000 × g for 30 s, resuspended in 500 μL YPD medium and spread on appropriate plates. Successfully transformed cells were expected to form colonies after 2-4 days of incubation at 30 °C.

Cultivation of yeast cells in shake flasks. Colonies of strains transformed with plasmids for expression
of different d-galacturonic acid reductase variants were scraped off for an overnight preculture in synthetic complete medium lacking uracil (SC-Ura) supplemented with 2% (w/v) maltose. Precultures of non-plasmid strains were started from a single colony in synthetic complete medium with all essential medium compounds supplemented. The main culture was grown in a 300 mL shake flask in 50 mL SC medium 33 lacking uracil and supplemented with 0.5% (w/v) d-galacturonic acid and 1% (w/v) sorbitol or 2% (w/v) glucose, respectively, at 30 °C and shaking at 200 rpm. The medium was buffered with 100 mM potassium phosphate, pH 6.3. The growth was monitored through OD 600 -measurement, and samples were withdrawn for HPLC-analysis.
HPLC analysis. The samples were treated with 5-sulfosalicylic acid to a final concentration of 5% (w/v). The analysis was done with an Ultimate 3000 HPLC system (Thermo Fisher Scientific) equipped with a NucleoGel Sugar 810 H (Macherey and Nagel) column. The column temperature was set to 30 °C, and the eluent (5 mM H 2 SO 4 ) flow rate was 0.4 mL/min under isocratic conditions. The signal was recorded using a refractive index detector (Shodex RI-101, Shoko Scientific Co.).
Protein extraction and enzyme assays. CEN.PK2-1C cells transformed with AnGar1 plasmids (SiHV079, SiHV101 and SiHV102) or with the empty plasmid as a control were grown in 50 ml SC-Ura media containing 2% (w/v) glucose until an OD 600 = 2.0-2.5. Subsequently, cells were harvested by centrifugation, washed and stored at -80 °C until further processing. After thawing on ice, the cells were mechanically disrupted in 10 mM potassium phosphate buffer (pH 7.2) by shaking (10 min at 4 °C) with glass beads (0.45 mm diameter) using a Vibrax cell disruptor (Janke & Kunkel, Staufen, Germany) and the cell debris was subsequently removed by centrifugation (15,000 × g, 5 min, 4 °C). Protein concentration of clear crude extracts was determined by the Bradford method, using bovine serum albumin as a standard. Enzyme assays were performed basically as described previously 18 . In detail, the reaction mixtures contained (in 200 µl) 10 mM potassium phosphate buffer (pH 7.2), 100 mM GalA, 160 or 800 µM NADPH or NADH and NADP as a competitive inhibitor, where indicated. The reaction was started by adding 10 µl of the substrate solution. The oxidation of NAD(P)H during 10 min was recorded by measuring the change of the absorbance at 340 nm. The specific activities (expressed as mili Units, mU per mg protein) were calculated by dividing the slope measured at 340 nm by the reaction time and protein amount in the reaction mixture. www.nature.com/scientificreports/ High performance liquid chromatography (HPLC) analysis was used for metabolite detection in the fermentation broth. Samples were centrifuged for 10 min at 13,000 × g, and the supernatant was filtered before analysis (0.22 μm pore size, Chromafil RC20/15 MS, Macherey-Nagel GmbH & Co.KG, Düren, Germany). An Aminex HPX-87H column (Biorad, Munich, Germany) was used for separation of substrates and products applying HPLC (Agilent 1100, Agilent Technologies Inc., Santa Clara, USA) equipped with an RI detector (Agilent 1200, Agilent Technologies Inc., Santa Clara, USA) and a thermostat (Mistral, Spark Holland, VE Emmen, the Netherlands). Twenty microliters of samples (or standards) was injected, and separation was performed at a constant column temperature of 60 °C, at a flow rate of 0.5 mL min −1 with 5 mM sulfuric acid as the mobile phase.

Batch processes in stirred-tank bioreactors and offline analytics. Seed cultures for inoculation
Modeling of AnGar1 and TnGar1. The homology models of AnGar1 and TrGar1 were generated with the 'Homology Model' function of the program package Molecular Operating Environment (MOE; Chemical Computing Group, https ://www.chemc omp.com/), using as a template the crystal structure of the NADPHdependent aldehyde reductase AKR1A1 from Sus scrofa (PDB ID 1HQT). The amino acid sequence identity and similarity between AKR1A1 and AnGar1 (or TnGar1) are 37% and 59%, respectively. The homology models generated were scored with GB/VI. The mutation residue scan and resulting protein stability and ligand affinity parameters were performed in MOE Protein Designing function with the Forcefields Amber10 and EHT.