Biosynthesizing structurally diverse diols via a general route combining oxidative and reductive formations of OH-groups

Design and verification of the platform pathway for biosynthesis of various diols

The proposed diol biosynthetic platform proposed in this study makes use of the highly active amino-acid synthetic pathways and channels the metabolic flux into diol formation through four reaction steps catalyzed by amino acid hydroxylase, l-amino acid deaminase, α-keto acid decarboxylase, and aldehyde reductase, respectively. In such a reaction cascade, one of the hydroxyl groups of the product diol is first formed through oxidative hydroxylation of an intermediate amino acid that is coupled with the oxidation of α-ketoglutarate (KG) to succinate without the requirement of an extra NAD(P)H, which is highly exergonic (Δ_rG’: −100 ~ −120 kcal/mol). The other hydroxyl group is formed from successive deamination, decarboxylation, and reduction of the resulting intermediates (Fig. 1c). Both the deamination and decarboxylation are thermodynamically highly favorable chemical reactions that can give the necessary push for the thermodynamically less favorable reductive formation of hydroxyl group in the last step.

In addition to the above principal advantages of the general concept, the synthetic route has several specific advantages and attractiveness. First, it stretches from the well-elucidated natural amino-acid biosynthesis pathways. The highly active metabolism of amino acids has been used to biotechnologically produce a number of valuable amino acids such as l-lysine and l-glutamate at a large industrial scale (millions of tons per year). The existing experiences and know-how can be exploited for further pathway engineering aimed at producing corresponding diols. In fact, amino acids with carboxyl and amino groups in their structure have been successfully used as intermediates for the synthesis of other valuable compounds. For example, deamination or decarboxylation of amino acids can generate various appealing carboxylates or diamines such as γ-amino-butyrate (GABA), 1,5-diaminopentane and 1,4-diaminobutane, trans-p-hydroxycinnamate, and urocanate^11,12. Second, the availability of different stereoselective hydroxylases enables the hydroxylation of an amino acid at a certain carbon position can be realized by selecting or designing a proper hydroxylase, offering thereby the possibility to generate structurally diverse diols from the same amino acid (Fig. 1c)^13,14. Last but not least, relatively high purity of target product can be obtained owning to the stringent stereocontrol of hydroxylation positions by hydroxylases, which will greatly reduce the difficulty and cost of the subsequent separation and purification process.

As depicted in Fig. 1b, the first key reaction of the proposed diol synthetic route is the hydroxylation of an amino acid to form a corresponding hydroxyl amino acid in the presence of an amino acid hydroxylase. The hydroxyl amino acid is then converted to the corresponding hydroxyl α-keto acid by an l-amino acid deaminase, followed by decarboxylation and reduction by an α-keto decarboxylase and an aldehyde reductase, respectively, to generate the desired diol.

So far a large set of canonical amino-acid hydroxylases have been discovered by genome mining or rational engineering^{15,16,17,18,19,20}. A typical canonical amino-acid hydroxylase requires α-KG and molecular oxygen as the two common co-substrates, and one amino acid as the specific substrate^21,22. Interestingly, and very useful for diol production, amino acid hydroxylases of different species can catalyze hydroxylation of the same substrate but at different carbon positions with high stereoselectivity²³. We selected hydroxylases from various species to generate 10 different diols from 6 amino acids in E. coli. A summary of the substrate amino acids and their corresponding diol products is given in Fig. 2a. The reaction schemes of the synthesis of the individual diols through this proposed pathway are shown in Supplementary Figures 1 and 2. The plasmids and primers used in this study are listed in Supplementary Data 1 and Supplementary Data 2, respectively. Specifically, the hydroxylase MFL from Methylobacillus flagellatus KT, which catalyzes the hydroxylation of branched-chain amino acids (BCAAs) at the C-4 position, was expressed in the pET28a plasmid, while L-amino acid deaminase AAD_vul from Proteus vulgaris, α-keto acid decarboxylase KDC from Lactobacillus lactis, and aldehyde reductase YqhD from E. coli W3110 were co-expressed in the plasmid pZA. The two plasmids were co-transformed into E. coli BL21(DE3), yielding strain DL01. DL01 was cultivated in FM-II medium added with different BCAAs as substrate. When 25 mM valine, leucine, norvaline or norleucine were used as a substrate, respectively, 18.5 mg/L MPO, 230.5 mg/L IPDO, 8.7 mg/L 1,3-BDO and 4.2 mg/L 1,3-pentanediol (1,3-PTD) were produced after 48 h of cultivation respectively (Fig. 2b, Supplementary Figure 3 and Supplementary Figure 4–7), demonstrating that this platform pathway functions in vivo. Different from the above BCAAs, isoleucine possesses two C-4 positions (C4¹-OH and C4²-OH). It has been reported that MFL has no catalytic activity on isoleucine¹⁷; but fortunately, selective hydroxylation of the two C-4 atoms can be realized separately by using two different hydroxylases encoded by hilA and hilB from Pantoea ananatis AJ13355¹⁵. Therefore, we constructed strains DL02 and DL03 that separately overexpressed hilA and hilB, in addition to AAD_vul, KDC, and YqhD in E. coli BL21(DE3). Cultivations of DL02 and DL03 in FM-II medium supplemented with 25 mM isoleucine led to the formation of 145.2 mg/L 2-E-1,3-PDO and 12.1 mg/L 2-methy-1,3-butanediol (2-M-1,3-BDO), respectively (Fig. 2c, Supplementary Figure 3, and Supplementary Figure 8 and 9). The significant difference in the production titer of 2-E-1,3-PDO and 2-M-1,3-BDO indicates that HilA has a higher catalytic capability on isoleucine than HilB. This can be attributed to the fact that isoleucine is the natural substrate of HilA, while the natural substrate of HilB is 4¹OH-isoleucine, the product of isoleucine catalyzed by HilA¹⁵.

Hydroxylase GriE from Streptomyces DSM 40835 was reported to selectively catalyze the hydroxylation of amino acids at C-5 position²⁴. Therefore, we constructed the strain DL04 by substituting MFL with GriE. After 48 h of fermentation, DL04 was able to produce 276.4 mg/L 2-methyl-1,4-butanediol (2-M-1,4-BDO), 7.8 mg/L 1,4-BDO, and 11.9 mg/L 1,4-PTD in FM-II medium supplemented with 25 mM leucine, norvaline, and norleucine, respectively (Fig. 2d, Supplementary Figure 3, and Supplementary Figure 10–12). Remarkably, the reported catalytic efficiency of GriE towards the three BCAAs (leucine>>norleucine>norvaline)²⁴ was in accordance with the titers of the corresponding diols (2-M-1,4-BDO>>1,4-PTD>1,4-BDO), indicating that the catalytic efficiency of hydroxylase is a determinant factor for the achievable titers of the diol products.

From the above results, it is confirmed that the hydroxylation of BCAAs by MFL and GriE is under stringent stereocontrol that enables the synthesis of structurally different diols from the same BCAA, as in the cases of leucine, norvaline, and norleucine (Fig. 2b, d). This clearly demonstrates the power and advantages of the biological production of structurally diverse diols using the strategy proposed in this work. To our knowledge, microbial synthesis of the six branched-chain diols MPO, IPDO, 2-M-1,3-BDO, 2-M-1,4-BDO, 2-E-1,3-PDO, and 1,4-PTD has not been reported before. Most of these products belong to important commodity chemicals with a variety of applications. For example, MPO is widely used to produce polyesters, resins, and lubricants²⁵, IPDO serves as an excellent moisturizer in the cosmetic industry, and 1,4-PTD is reported as a valuable building block for the synthesis of chloroquine, one of the most effective medicines for the prevention of malaria²⁶.

1,3-PDO is probably the most prominent diol whose bioproduction has attracted great attention for decades, e.g., microbial production of 1,3-PDO from biodiesel production-derived glycerol or glucose. However, there are no natural microorganisms that can produce 1,3-PDO directly from sugar. In addition to the well-known synthetic pathway of 1,3-PDO from glucose over glycerol as industrially used by DuPont and Tate&Lyle, further de novo pathways have been developed to synthesize 1,3-PDO, e.g., by expanding the homoserine synthesis pathway using glucose as substrate^9,27 or by developing an aldolase-catalyzed one-carbon assimilation pathway using glucose and methanol as co-substrates⁸. In this study, we sought to produce 1,3-PDO by extending the threonine catabolism in E. coli. This pathway shares a similar structure with the proposed general route except that the deaminase AAD_vul was replaced by the endogenous threonine deaminase IlvA, because IlvA can also remove the –OH group at C-3 position in accompany with the deamination, and the hydroxylase MFL, which is inactive towards threonine, was replaced by one of another two hydroxylases, namely AVI from Agrobacterium vitis and BPE from Bordetella petrii, which possess hydroxylase activities towards threonine¹⁷.

Since the conversion of threonine to 4-hydroxy-threonine catalyzed by threonine hydroxylase and the conversion of 4-hydroxy-α-ketobutyric acid to 1,3-PDO catalyzed by α-keto acid decarboxylase and aldehyde reductase have been demonstrated by the previous studies^8,17,28,29, we carried out an in vitro examination of the catalytic activity of threonine deaminase on 4-hydroxy-threonine and confirmed that 4-hydroxy-threonine produced from the hydroxylation of threonine is further converted by threonine deaminase (Supplementary Figure 14). Next, to verify this pathway in vivo, these two hydroxylases AVI and BPE were assembled into the platform pathway in alternative combination with two α-keto acid decarboxylases, namely KDC from L. lactis and PDC from Zymomonas mobilis, which were employed to produce 1,3-PDO in the previous reports^8,9. The obtained four recombinant strains were cultivated in FM-II medium supplemented with 25 mM threonine. Surprisingly, none of the four strains produced 1,3-PDO. A possible interpretation of this pitfall is that the deaminase IlvA directly competes with the hydroxylase AVI or BPE in the use of threonine as substrate, but the K_cat of IlvA is much higher (~16,000-fold) than that of AVI or BPE^17,30, resulting in the observation that the majority of threonine was quickly deaminized by IlvA without being first hydroxylated by AVI or BPE. To overcome the mismatched catalytic efficiencies of the enzymes, we overexpressed the threonine hydroxylase (AVI or BPE), as well as the α-keto acid decarboxylase (PDC or KDC) and the aldehyde reductase YqhD in E. coli BL21(DE3) so that 4-hydroxy-threonine should be first formed by hydroxylation before the endogenous IlvA fulfills the task of deamination, followed by enhanced decarboxylation and reduction to form 1,3-PDO. As the result, the recombinant strains DL06 and DL08 (both harboring KDC) yielded 6.7 mg/L and 29.1 mg/L 1,3-PDO in shake flask culture, respectively (Fig. 2e, Supplementary Figure 3 and 13). In contrast, another two strains DL05 and DL07 harboring PDC still did not produce 1,3-PDO due to unknown reasons, though both KDC and PDC have been reported to catalyze the conversion of 4-hydroxy-2-ketobutyrate to 3-hydroxy-propionaldehyde^8,9. One possible interpretation is that the activity of PDC is inhibited by certain metabolites produced by the promiscuous catalysis of hydroxylase in DL05 and DL07. Compared with other 1,3-PDO synthetic pathways, the hydroxylation, deamination, and decarboxylation in the proposed pathway are all thermodynamically more favorable reactions (Supplementary Figure 15), which can give the necessary push for the thermodynamically less-favorable reductive formation of hydroxyl group in the last step.

Examination of the reaction orders for IPDO biosynthesis

The results of 1,3-PDO synthesis with different combinations of enzymes presented above indicate that the reaction order (e.g., first hydroxylation or deamination) of the proposed diol biosynthesis pathway may affect the biosynthetic efficiency. In the following, we took IPDO as a further example to address this question in more detail.

Since the reduction of aldehyde by aldehyde reductase always follows the decarboxylation of α-keto acid by α-keto acid decarboxylase, we considered these two reactions as one single step. In this way, a total of six reaction orders (routes) are thus possible, as listed in Fig. 3a. Routes 1 and 2 (R1 and R2) have both hydroxylations as the first reaction step, but differ from each other in the order of deamination and decarboxylation/reduction. To examine whether R1 or R2 or both work, we overexpressed AAD_vul, KDC, and YqhD in the plasmid pZA to yield pZA-aky and transformed it into E. coli BL21(DE3). The obtained strain DL01-R12 was cultivated in FM-II medium with the addition of 5 mM 4OH-Leu, the hydroxylation product of leucine. This strain produced 374.1 mg/L IPDO after 48 h of cultivation (Fig. 3b). Further investigation by enzyme activity assay, however, showed that in the absence of the deaminase AAD_vul, KDC, and YqhD had no catalytic activity on 4OH-Leu (Supplementary Figure 16a, b) and thereby ruled out the possibility of Route 2.

In a similar way, we examined Route 3 and Route 4, which have deamination as the first reaction step and differ from each other in the order of hydroxylation and decarboxylation/reduction. We overexpressed only KDC and YqhD in the plasmid pZA to yield pZA-ky and transformed it together with pET-mfl into E. coli BL21(DE3). The resulted strain DL01-R34 was able to produce 19.2 mg/L IPDO (Fig. 3b) in FM-II medium supplemented with 5 mM 2-ketoisocaproate (KIC), the deamination product of leucine catalyzed by AAD_vul. Moreover, the feasibility of Route 3 was further verified by the fact that MFL showed activity on KIC (see below in Fig. 5c). The possibility of Route 4 was eliminated due to the missing hydroxylation activity of MFL towards isopentanol (IPT, the decarboxylation/reduction product of KIC), as can be observed that no IPDO was produced by the strain DL01-R4 that carried pET-mfl in E. coli BL21(DE3) in FM-II medium supplied with 5 mM isopentanol (Fig. 3b). Finally, we ruled out the possibility of Routes 5 and 6, which have in common the decarboxylation/reduction as the potential first step of the reaction cascade, because both KDC and YqhD had no activity on leucine (Supplementary Figure 16a, c).

Based on the above results, we concluded that there are two pathways working simultaneously for IPDO production from leucine (Fig. 3c), i.e., the reaction steps were performed either in the order of hydroxylation, deamination, decarboxylation, and reduction from leucine (Route 1) or in the order of deamination (leucine to KIC), hydroxylation, decarboxylation, and reduction (Route 3). If glucose is used as the sole substrate, Route 3 consists only of the hydroxylation, decarboxylation, and reduction steps, while Route 1 seems to contain a futile cycle between transamination and deamination from a pathway stoichiometric point of view (Fig. 3c). Surprisingly, Route 1 is more effective than Route 3 as demonstrated by the significantly higher IPDO titer when fed with the same concentration of precursors, i.e., leucine for Route 1 and KIC for Route 3. The reason behind this seems to be due to the lower catalytic capability of MFL on KIC than on leucine (this was experimentally confirmed, see results later), since the latter is regarded as the natural substrate of MFL. As a result, the metabolic flux channeling into Route 1 is expected to be much higher than that into Route 3 under physiological conditions. This implies that effective hydroxylation is the key step in the proposed pathway of IPDO production.

It seems that the higher metabolic flux of Route 1 is reflected by its more favorable thermodynamic profile compared with that of Route 3 (Fig. 4). As depicted in Fig. 4a, b, Route 1 starts with a stronger exergonic reaction with a higher energy driving force than Route 3. The calculated thermodynamic profile of Route 1 with a gradual decrease of the (Δ_rG) is more harmonized than that of Route 3. It would be interesting to further investigate how the three key constraints of a pathway, namely thermodynamics, enzyme kinetics, and biochemical properties of intermediates (e.g., permeability, toxicity, or stability), shape the reaction order and determine its overall activity. For the regulation and optimization of a reaction cascade, it would be also interesting to know if there exists an ideal thermodynamic profile as indicated by Fig. 4c. To this end, more intracellular and systematic information obtained under physiological conditions is needed. For both the working reaction orders, only one molecular NADPH is required for generating one molecular IPDO from leucine. In addition, they do not require expensive co-enzymes. Because the oxidative pathway of hydroxyl group formation does not lead to the loss of carbon atoms, the theoretical yield of diol is high. The theoretical yield of IPDO from glucose is estimated to be as high as 0.628 mol/mol glucose based on flux balance analysis (FBA) (Supplementary Figure 17). All these features indicate that it should be a robust and effective pathway.

Optimization of the IPDO-producing route through enzyme mining and improved precursor supply

To further optimize and demonstrate the efficiency of Route 1 for IPDO biosynthesis, we screened for hydroxylase and α-keto decarboxylase with improved catalytic efficiency. l-amino acid deaminase was not screened here, as AAD_vul from P. vulgaris showed the highest activity on leucine among all the reported l-amino acid deaminase from different species³¹. To this end, heterologous hydroxylases and α-keto decarboxylases of different origins were selected (Fig. 5a). All of the coding genes were synthesized with optimized codons. The resulting recombinant strains were inoculated and cultivated in FM-II medium with or without the addition of 25 mM leucine. The starting strain DL01 (named here as IP01 for enzyme screening purposes) harboring the plasmids pET-mfl and pZA-aky yielded 230.5 mg/L and 15.3 mg/L IPDO in the presence or absence of supplemented leucine (Fig. 5b). Two additional α-keto decarboxylases, PDC from Z. mobilis and THI3 from Saccharomyces cerevisiae, were selected to compare with KDC. The obtained strain IP02 harboring PDC accumulated 64.8 mg/L and 5.9 mg/L IPDO with or without additional leucine, which represented a reduction of 71.9% and 61.4% compared with those achieved in DL01. Similarly, when KDC was replaced by THI3 in strain IP03, the titer of IPDO was significantly reduced to 21.4 mg/L and 1.4 mg/L, respectively (Fig. 5b). In addition to the hydroxylase MFL, four hydroxylases, namely AVI, IDO, GOX, and BPE from Agrobacterium vitis, Bacillus thuringiensis, Gluconobacter oxydans, and Bordetella petrii, respectively, were also selected for screening. Like MFL, they were all reported to stereospecifically catalyze the hydroxylation of amino acids at C-4 position¹⁷. Among the derived strains, IP04, IP05, and IP06 harboring AVI, IDO, and GOX, respectively, presented varying degrees of decreased production titers of IPDO compared with strain IP01. In contrast, IP07 harboring BPE showed the accumulation of 220.3 mg/L and 13.7 mg/L IPDO, respectively, in the presence and absence of additional leucine, indicating a high catalytic activity of BPE comparable to that of MFL in IPDO production. Taking together the screening results, the strain IP01 and IP07 turned out to be the two best IPDO-producing strains. Given that the K_m value of MFL on leucine is lower than that of BPE¹⁷, the strain IP01 was selected as the best strain for the follow-up optimization study.

In our designed pathway, leucine is one of the most important precursors for IPDO production. The results shown above demonstrated that IPDO titer from the strain IP01 was significantly improved if additional leucine was supplemented into the medium containing glucose as substrate, implying that the native leucine supply was not sufficient for efficient production of IPDO from glucose. An enhancement of the leucine supply in IP01 is obviously necessary. For this purpose, leucine operon (leuA^fbrBCD³²) from E. coli W3110 was overexpressed on the p15AS vector under the control of J23119 promoter (http://parts.igem.org/). The resulting strain IP08 showed an increase in the final IPDO titer from 15.3 mg/L (in IP01) to 20.5 mg/L without supplementing leucine (Fig. 5c). This was accompanied with a rise of the leucine concentration in the culture broth from 0.14 mg/L to 0.22 mg/L (Fig. 5d). KIC is the direct precursor of leucine. In a previous study, acetolactate synthase AlsS from B. subtilis was employed to increase the supply of cytoplasmic KIC³³. In view of this, we further overexpressed alsS from B. subtilis as well as ilvCD genes from E. coli to yield strain IP09, thereby enhancing the flux from pyruvate to KIC. Compared with IP08, this strain showed a significant increase in IPDO titer, reaching 55.7 mg/L. However, 7.08 mg/l-valine was also accumulated in the culture broth of IP09, which was 3.04-fold higher than that of IP08 (Fig. 5d). Since KIC is the common precursor of leucine and valine, it is reasonable that overexpressing alsS and ilvCD led to a considerable part of the metabolic flux from KIC being shunted to the synthesis of valine. To address this issue, strain IP10 was constructed by overexpressing tyrosine aminotransferase TyrB from E. coli which promiscuously catalyzes the conversion of KIC to leucine but can’t convert α-ketoisovalerate (KIV) to valine³⁴. After 48 h of cultivation, a final titer of 77.6 mg/L IPDO was achieved by the strain IP10, which was 1.62-fold higher than that of IP09. As expected, the leucine concentration in the broth further raised to 0.35 mg/L, whereas the valine decreased to 5.43 mg/L (Fig. 5d), indicating that overexpressing TyrB indeed improved the metabolic flux into the leucine synthesis pathway.

Directed evolution of MFL for improved IPDO synthesis

As already mentioned above, hydroxylation is the rate-limiting step in the proposed pathway of IPDO production. Previous research indicated that the catalytic activity of MFL on leucine is so poor that it is far from being suitable for technical application¹⁷. Therefore, to search for MFL mutants with higher hydroxylase activity EP-PCR was applied to construct a random MFL mutant library. For screening, we developed an approach based on the following principle: hydroxylation of leucine by MFL is coupled with oxidation of α-KG to succinate, so that the growth restoration of a succinate-auxotrophic strain is strictly dependent on the leucine hydroxylation, thus coupling cell growth to hydroxylase activity (Fig. 6a). A succinate-auxotrophic strain Δ3A (E. coli BL21 ΔsucAΔaceAΔputA (DE3)(pLysS))³⁵ was employed as a selection tool for screening positive mutants. The suitability of this strain for coupling growth rate with MFL activity was evaluated before the screening. Plasmid pET-mfl was transformed into the Δ3A strain and cultivated in M9 medium supplemented with 0.2 g/L leucine, 0.2 g/L α-KG, and 0.1 mM isopropyl β-d-1-thiogalactopyranoside (IPTG). As a control strain, Δ3A carrying pET28a was cultivated in parallel. Both strains grew poorly in M9 medium, although Δ3A-pET-mfl grew slightly better than the control strain after 60 h of cultivation (Supplementary Figure 18). This was mainly due to strains’ poor growth environment and the heavy burden caused by the addition of IPTG in the lag phase. To solve this problem, we enriched the medium and induced the expression of MFL when the cells were in the exponential phase. After adding 4 g/L yeast to the M9 medium both strains exhibited better growth; in particular, Δ3A-pET-mfl presented a notably faster growth after induction by IPTG (Fig. 6b). Hence, the modified M9 medium was used for the directed evolution of MFL.

An MFL mutant library with a capacity of 5000~6000 was constructed by EP-PCR. The quality of the mutant library was assessed before screening for mutants with higher activities. Sequencing results indicated that the quality of the mutant library was satisfactory and can be used for the subsequent screening of MFL mutants (Supplementary Figure 19). The MFL mutant library was then transformed into the Δ3A strain and seeded on M9 agar plates supplemented with 4 g/L yeast, 0.2 g/L leucine, 0.2 g/L α-KG, and 0.1 mM IPTG. After 48 h of cultivation, 18 large-sized colonies were selected and their hydroxylase activities were determined. The results indicated that two mutants MFL_m1 and MFL_m2 showed improved activities, reaching 1.9-fold and 1.5-fold of that of the wild type, respectively (Fig. 6c). The remaining mutants exhibited comparable or even decreased activities (Supplementary Figure 20). Surprisingly, in addition to improved hydroxylation of leucine, it was observed that promiscuous hydroxylation activities of MFL_m1 and MFL_m2 on KIC were also improved by 1.8-fold and 1.3-fold, respectively (Fig. 6c). Therefore, these two MFL mutants could promote IPDO production by simultaneously improving the performance of both synthetic routes (Route 1 and Route 3) (Fig. 3c) in the host strain.

Sequence alignment analysis showed that the residue His138 in the wild-type MFL was replaced by Gln in MFL_m1, whereas the residue Ser98 was replaced by Gly in MFL_m2. As no crystal structure of MFL has been reported yet, homologous modeling was applied to predict the MFL structure. Homologous modeling generally requires that the sequence identity between the template protein and the target protein is higher than 30%³⁶. After performing BLAST homology search against the NCBInr database, we found that MFL shares a 31.2% identity in amino acid sequence with the best hit MpDO, which is a member of the dioxygenases family (PF10014) (Supplementary Figure 21) and was chosen as the template for homologous modeling of MFL. It is observed that two β-sheets are packed against each other, forming a cup-shaped β-sandwich with a topology characteristic of the double-stranded β-helix fold, a classic structure that is shared by the family members. Residues involved in binding Fe²⁺ and α-KG are strictly conserved between MFL and MpDO (Supplementary Figure 21), suggesting that they may share the same enzymatic mechanism. The substrate-binding site of MFL is formed by sheet 1 and three variable loops (Fig. 6d). It is suggested that the plasticity of the active site is very important for the catalysis of cupin dioxygenases³⁷. The flexible configuration of the three loops leads to differences in the size and accessibility of the active site³⁸. However, it should be noted that the modeled structures of the three loops may not be reliable due to the low sequence identity between MFL and the template MpDO. According to the modeling, position 138 is located on an α-helix at the opposite side of the active site pocket (Fig. 6d, Supplementary Figure 21). The S98G mutation is located in a loop region (Fig. 6d, Supplementary Figure 21). This loop was reported to play a role in adjusting the plasticity of the active site in MpDO through conformation changes³⁷. Whether the S98G mutation has an effect on the conformation change of this loop needs further experimental verification.

To examine the performance of MFL_m1 on IPDO biosynthesis, MFL_wt in strain IP10 was replaced by MFL_m1, yielding strain IP11. Batch fermentation with IP11 was performed in a shake flask. As shown in Fig. 6e, after 24 h of cultivation its OD reached 10.8 and gradually increased to 16.8 at 96 h, which is similar to that of the control strain IP10. IPDO production of IP11 was steadily increased in an almost linear way throughout the whole cultivation period and appeared to level off near the end of the cultivation, reaching a final titer of 0.31 g/L. It is remarkable that the IPDO titers from the strain IP11 were 1.3~2.0 times higher than that from strain IP10 in the cultivation time from 24 h to 96 h of cultivation, which is well correlated with the elevated MFL activity in strain IP11. This again confirms that the MFL-catalyzed step is the primary limiting step in the constructed IPDO synthetic pathway, and highlights the effectiveness of boosting MFL activity for further improvement of IPDO production. However, a high level of extracellular valine was still detected in the fermentation broth of strain IP11. The concentrations of leucine and valine in the culture broth of strain IP11 were maintained at levels comparable to what was observed in the strain IP10 during the whole cultivation time (Fig. 6f).