Rational engineering of 2-deoxyribose-5-phosphate aldolases for the biosynthesis of (R)-1,3-butanediol

Carbon–carbon bond formation is one of the most important reactions in biocatalysis and organic chemistry. In nature, aldolases catalyze the reversible stereoselective aldol addition between two carbonyl compounds, making them attractive catalysts for the synthesis of various chemicals. In this work, we identified several 2-deoxyribose-5-phosphate aldolases (DERAs) having acetaldehyde condensation activity, which can be used for the biosynthesis of (R)-1,3-butanediol (1,3BDO) in combination with aldo-keto reductases (AKRs). Enzymatic screening of 20 purified DERAs revealed the presence of significant acetaldehyde condensation activity in 12 of the enzymes, with the highest activities in BH1352 from Bacillus halodurans, TM1559 from Thermotoga maritima, and DeoC from Escherichia coli. The crystal structures of BH1352 and TM1559 at 1.40–2.50 Å resolution are the first full-length DERA structures revealing the presence of the C-terminal Tyr (Tyr224 in BH1352). The results from structure-based site-directed mutagenesis of BH1352 indicated a key role for the catalytic Lys155 and other active-site residues in the 2-deoxyribose-5-phosphate cleavage and acetaldehyde condensation reactions. These experiments also revealed a 2.5-fold increase in acetaldehyde transformation to 1,3BDO (in combination with AKR) in the BH1352 F160Y and F160Y/M173I variants. The replacement of the WT BH1352 by the F160Y or F160Y/M173I variants in E. coli cells expressing the DERA + AKR pathway increased the production of 1,3BDO from glucose five and six times, respectively. Thus, our work provides detailed insights into the molecular mechanisms of substrate selectivity and activity of DERAs and identifies two DERA variants with enhanced activity for in vitro and in vivo 1,3BDO biosynthesis.

The formation of carbon-carbon bonds via aldol condensation of two carbonyl compounds is indispensable in biological systems and organic chemistry (1)(2)(3). Aldol condensation reactions generate a new ␤-hydroxy carbonyl compound, which is a valuable precursor in the construction of complex organic molecules caused by the formation of up to two new stereogenic centers (4). Using aldehydes as donor substrates in aldol reactions is particularly of interest because this provides the opportunity for sequential aldol condensation reactions to synthesize more complex molecules (5,6). In biological systems, aldolase enzymes catalyze the reversible and stereoselective aldol addition of a nucleophilic donor onto an electrophilic aldehyde acceptor (7). The formation of a new C-C bond is accompanied by the generation of a new stereocenter, making aldolases attractive tools in the synthesis of chiral compounds and bioactive molecules. Therefore, aldolases have emerged as a promising alternative in the biocatalytic synthesis of rare sugars and sugar derivatives, such as statins, iminocyclitols, epothilones, and sialic acids (8 -10).
2-Deoxyribose-5-phosphate aldolases (DERA, E.C. 4.1.2.4) 2 are found in all kingdoms of life and represent the major aldolase group. One of the best-characterized DERAs is the Escherichia coli DeoC, which belongs to the class I (metal-independent) aldolases (9,10). The E. coli deoC is part of the deo operon (deoABCD) involved in the utilization of extracellular deoxyribonucleotides as energy sources (11). It transforms the D-2deoxyribose-5-phosphate (DRP) intermediate into D-glyceraldehyde-3-phosphate and acetaldehyde, which enter glycolysis and the Krebs cycle, respectively (12). The DERA reaction is reversible, because it also catalyzes the aldol condensation between acetaldehyde (the donor molecule) and D-glyceraldehyde-3-phosphate (the acceptor molecule) producing DRP (Scheme 1) (13). This class of aldolases is unique in that it can catalyze the aldol condensation of two aldehydes and does not require a ketone substrate, whereas other aldolases use ketones as aldol donors and aldehydes as acceptors (10,14). It activates the donor molecule (acetaldehyde) via the catalytic Lys residue, forming a covalent Schiff base intermediate (enamine) followed by the carboligation between the acceptor (D-glyceraldehyde-3-phosphate or second acetaldehyde) and the Schiff base (14, 15). The crystal structure of the E. coli DERA (DeoC) adopts the ubiquitous triosephosphate isomerase (TIM) barrel (␣/␤) 8 fold with the catalytic Lys 167 (the Schiff base-forming residue) located on strand ␤6 (16). A proton relay system composed of Asp 102 , Lys 201 , and a water molecule is involved in shuffling a proton between C2 of the acetaldehyde imine and enamine and subsequent C3 hydroxyl protonation. In addition, several biochemical studies suggested that the C-terminal Tyr 259 of the E. coli DeoC is crucial for enzyme activity (16 -18). However, all published crystal structures of DERA show the absence of electron density for the last eight C-terminal residues including Tyr (16, 19 -22). Recently, using a combination of NMR spectroscopy and molecular dynamics simulations, it has been shown that the C-terminal Tyr 259 of the E. coli DeoC enters the active site in catalytically relevant closed states and is required for efficiency of the proton abstraction step of the DERA catalytic reaction (18).
Recently, we established a novel pathway to produce (R)-1,3butanediol (1,3BDO) from acetaldehyde using DERA as the key enzyme (32,33). The non-natural diol 1,3BDO is used as a building block for the production of synthetic polymers, pheromones, fragrances, insecticides, and antibiotics (34 -39). Presently, 1,3BDO has been produced mainly from petroleumbased feedstocks using chemical processes, which require harsh reaction conditions and release toxic intermediates and by-products (38). Therefore, the development of biocatalytic processes for the production of 1,3BDO from renewable feedstocks is of increasing importance (39,40). The recently proposed artificial biosynthetic approach for 1,3BDO production from glucose is based on a reversed fatty acid ␤-oxidation pathway, which includes four heterologous enzymes and requires three NADPH and one CoA molecules per molecule of 1,3BDO produced (41,42). In contrast, the proposed DERA-based pathway for 1,3BDO production involves three heterologous enzymes: pyruvate decarboxylase (PDC, producing acetalde-hyde from pyruvate), DERA (catalyzing aldol condensation of two acetaldehyde molecules to 3-hydroxybutanal), and aldoketo reductase (AKR), which reduces 3-hydroxybutanal (3HB) to 1,3BDO (Scheme 2). The heterologous expression of this pathway in E. coli resulted in the production of 0.3 g of 1,3BDO/ liter from glucose (11.2 mg/g of glucose) (33). Using a systems metabolic engineering approach, the 1,3BDO titer was increased to 2.4 g/liter and yield was increased to 56 mg/g of glucose, further highlighting the potential of aldolases for synthesis of valuable products. This study also suggested that the rate-limiting step of the proposed 1,3BDO pathway is the DERA-catalyzed aldol condensation of acetaldehyde to 3HB (33).
Although our recent studies revealed great potential of DERAs for biocatalytic conversion of acetaldehyde to 1,3BDO (33), this activity (acetaldehyde condensation) has not been examined in depth. The scarcity of data on DERAs limits our efforts on increasing the acetaldehyde condensation activity of these enzymes, which represents the rate-limiting step in the biocatalytic synthesis of 1,3BDO and potentially statin drugs. In this work, after screening 20 purified microbial DERAs, we identified BH1352 from the alkaliphilic bacterium Bacillus halodurans, as well as TM1559 from Thermotoga maritima and E. coli DeoC as the most active aldolases in the DERA-AKRcoupled production of 1,3BDO from acetaldehyde. The crystal structures of these enzymes were determined including the first full-length DERA structure (BH1352) and revealed the catalytic residues and substrate-binding sites. Using structure-based site-directed mutagenesis, we identified the BH1352 residues critical for acetaldehyde condensation and designed several DERA variants with higher activity in the production of 1,3BDO both in vitro (from acetaldehyde) and in vivo (from glucose). We demonstrated that E. coli cells expressing the DERA-AKR pathway with engineered DERA variants produced 5-6 times more 1,3BDO from glucose compared with cells with the WT BH1352.

Phylogenetic analysis of DERA sequences
To provide insight into the phylogenetic diversity of DERAs, 2,553 sequences of putative DERAs were extracted from the Kyoto Encyclopedia of Genes and Genomes Orthology database using the identifier K01619 for the E. coli DERA (DeoC), which is the best-characterized DERA enzyme (23). Initially, this pool of putative DERA proteins included more than 2,500 sequences (2,281 from bacteria, 120 from archaea, and 152 from eukaryotes), but it was reduced to 1,974 proteins after removing redundant sequences. This phylogenetic analysis revealed the presence of five major clusters of DERA proteins including one bacterial domain, one Firmicutes (Bacilli and Clostridia), one mostly Proteobacteria, and two mixed clusters SCHEME 1. Reversible retro-aldol reaction catalyzed by DERA. SCHEME 2. DERA-based pathway for 1,3BDO production from pyruvate.

Engineering DERAs for the biosynthesis of 1,3BDO
( Fig. 1A). To screen DERAs for the bioconversion of acetaldehyde to 1,3BDO, we selected 20 DERA proteins from different phylogenetic groups, which were found to be soluble when expressed in E. coli ( Fig. S1 and Table S1). Based on the phylogenetic analysis, 17 selected DERAs belong to the five large clusters (clusters 1-5), whereas the remaining three proteins were from nonclassified sequences.

Screening of purified DERAs for biosynthesis of 1,3BDO from acetaldehyde
In our previous work, we identified several aldo-keto reductases (AKRs) with significant activity in reducing 3-hydroxybutanal to 1,3BDO (Scheme 2) (32). From these proteins, PA1127 from Pseudomonas aeruginosa was found to exhibit negligible activity against acetaldehyde, making it suitable for coupling of

Engineering DERAs for the biosynthesis of 1,3BDO
the DERA-catalyzed condensation of acetaldehyde (to 3-hydroxybutanal) with the AKR-catalyzed reduction of 3-hydroxybutanal to 1,3BDO (Scheme 2). Using PA1127, we established a coupled enzyme system (DERA ϩ PA1127) (Fig. S2) and screened 20 purified DERAs for transformation of acetaldehyde to 1,3BDO. These screens revealed significant production of 1,3BDO in the presence of 12 DERAs with the highest activity observed in TM1559 from T. maritima (DERA group 1), E. coli DeoC (DERA group 4), and BH1352 from B. halodurans (DERA group 2) (Fig. 1). Because BH1352 was found to support the highest production of 1,3BDO by E. coli cells expressing different DERAs (including TM1559 and DeoC) (33), this protein was selected for detailed structural and biochemical studies of the transformation of acetaldehyde to 1,3BDO.
Because B. halodurans is an alkaliphilic bacterium (grows well at pH Ͼ9.0), we determined the optimal pH range for BH1352 using the retro-aldol DRP cleavage reaction coupled with glyceraldehyde-3-phosphate dehydrogenase and triosephosphate isomerase (16,24,43,44). These assays revealed a broad pH range with the maximal activity of BH1352 at pH 7.2-9.5 ( Fig. S3), whereas the previously reported DERA enzymes from other bacteria showed the highest activity at pH 6.0 -7.5 (24,28). At optimal pH, V max of BH1352 with DRP as substrate was calculated to be 52-67 mol/min/mg protein, which is lower than that for the DERA from Lactobacillus brevis (102 mol/min/mg protein) but higher than other DERAs (0.25-1.00 mol/min/mg protein) (24,28). The L. brevis DERA has also been reported to exhibit high resistance to aldehydes, but this enzyme showed low activity in the AKR-based screen for 1,3BDO synthesis ( Fig. 1B) (28). Steady-state kinetic parameters of BH1352 and its variants were also determined using the DRP cleavage reaction (Table 1). These experiments revealed that BH1352 exhibits typical Michaelis-Menten kinetics with the apparent K m ϭ 0.22 mM, which is close to that for E. coli DeoC and more than 10 times lower than that for the L. brevis DERA (3.3 mM) (24).

Crystal structures of DERAs: overall fold and active site
The crystal structures of BH1352 (PDB codes 6D33 and 6MSW) and E. coli DeoC (PDB code 1KTN) were determined to 2.50 and 1.40 Å resolution, respectively, using the sittingdrop vapor diffusion method (Table 2), whereas the unpublished crystal structure of TM1559 is available from Protein Data Bank (PDB codes 3R12 and 3R13, Joint Center for Struc-tural Genomics). In contrast to E. coli DeoC, the structures of BH1352 and TM1559 revealed the presence of electron density for the C-terminal Tyr residues (Tyr 224 and Tyr 246 , respectively) making them the first full-length DERA structures. In BH1352, Tyr 224 is located on a flexible strand, whereas the TM1559 Tyr 246 is located on the C-terminal ␣-helix ( Fig. 2 and Fig. S4). Analysis of crystal contacts of BH1352 using the quaternary prediction PDBePISA server (http://www.ebi.ac.uk/ pdbe/pisa/) 3 predicted a dimeric state ( Fig. 2A). This was supported by the result of size-exclusion chromatography, suggesting that this protein exists as a dimer in solution (observed molecular mass, 51.7 kDa; predicted mass of monomer molecule, 24.2 kDa) (Fig. S5). It is similar to the dimeric state of hyperthermophilic and L. brevis DERAs but is different from the E. coli DeoC, which was found to exist in a monomerdimer equilibrium (Fig. S6) (21,24). Based on the BH1352 structure, the interfaces between monomers in each adjacent dimer involve 13 hydrogen bonds and buries 1,288 Å 2 surface area. The dimerization interface of BH1352 is composed mainly of hydrophobic interactions including the BH1352 loops containing Pro 16 , Phe 66 , Pro 67 , Leu 68 , Ile 97 , and Phe 160 (Table S2), although the weak dimerization interface of DeoC (573 Å 2 ) is comprised of a single hydrogen bond and two salt bridges in between ␣3 and ␣4 helices of each protomer ( Fig. S6) (21,24,45). On the other hand, the TM1559 structure exhibits stronger dimerization interactions with the interface 1,464 Å 2 between TM1559 protomers including 14 hydrogen bonds and two salt bridges along with many hydrophobic contacts, which is in line with higher structural stability of this protein (Table S2) (24).
Based on the BH1352 structure, its active site is located inside of the ␤-barrel, near its C-terminal side (Fig. 2B). The active site entrance is formed by the several loops connecting ␤-strands (␤1, ␤6, and ␤7) with ␣-helices (␣1, ␣6, and ␣7), containing highly or semiconserved residues including Thr 12 , Leu 14 , Lys 15 , Phe 66 , Ile 128 , Phe 160 , Ser 18ϩ 6, and Ser 209 . The side chains of these residues create a narrow channel providing access of substrates to the catalytic Lys 155 , located on the ␤6 strand (Fig. 3A). In the best-characterized DERA from E. coli, the catalytic Lys 167 is in close proximity to the side chains of conserved Lys 137 and Lys 201 , and the three Lys residues form salt bridges

Engineering DERAs for the biosynthesis of 1,3BDO
with the side chain oxygens of conserved Asp 102 (16,21). During the DERA-catalyzed synthesis of 2-deoxyribose-5-phosphate (DRP) (Scheme 1), the uncharged nucleophilic Lys 167 of E. coli DeoC attacks the acetaldehyde carbonyl forming a carbinolamine and then a Schiff base, which subsequently tau-tomerizes to an enamine group and attacks glyceraldehyde 3-phosphate (16,43). Finally, hydrolysis of the aldol condensation intermediate produces the free enzyme and DRP. Previous biochemical studies with the E. coli DeoC also suggested an important role for the highly conserved C-terminal

Engineering DERAs for the biosynthesis of 1,3BDO
Tyr 259 , because its replacement by Phe (Y259F) resulted in a ϳ100-fold reduction of the DRP cleavage activity (16,18). Interestingly, the deletion of Tyr 259 (⌬Y259) significantly increased the DeoC activity in the condensation reaction between acetaldehyde and chloroacetaldehyde (27). Using a combination of NMR and molecular dynamics simulations, it has been shown that the DeoC C-terminal tail is intrinsically disordered with the equilibrium between open and catalytically relevant closed states, where Tyr 259 is inserted into the active site close to the catalytic Lys 167 (ϳ 6 Å) (18). Remarkably, the structures of both BH1352 and TM1559 revealed the presence of electron density for the C-terminal Tyr 224 and Tyr 246 , respectively. In TM1559, Tyr 246 is positioned on the C-terminal ␣-helix with the side chain exposed to solvent (Fig. S4), whereas Tyr 224 in BH1352 is located on the flexible C-terminal tail with the side chain stabilized through interactions with the active site of the other BH1352 dimer (Fig. 4). In the BH1352 active site, the Tyr 224 side chain showed two orientations, with the hydroxyl group pointing toward the catalytic Lys 155 (2.7 Å) or toward the ␤1-␣2 loop backbone (near conserved Leu 14 and Lys 15 , 2.9 -3.0 Å) (Fig. 4A). In the second orientation, the aromatic ring of Tyr 224 is part of the hydrophobic wall of the active site (with conserved Leu 14 , Val 63 , Phe 66 , Ile 128 , and Phe 160 ) (Fig.   4B). Thus, in line with a recent work on molecular dynamics simulation with DeoC (18), the BH1352 crystal structure provides the structural indication that the C-terminal Tyr residue of DERAs might be involved directly in the substrate binding and/or catalytic mechanism of this enzyme.

Probing the active site of BH1352 using site-directed mutagenesis
Because the catalytic residues of E. coli DeoC are conserved in BH1352 and TM1559, the same catalytic mechanism can be applied to aldol condensation of two acetaldehyde molecules catalyzed by these enzymes. In the BH1352 active site, the side chain of conserved Asp 92 (Lys 179 in TM1559) forms salt bridges with the conserved Lys 126 (2.7 Å from Asp 92 ), Lys 155 (3.1 Å), and Lys 184 (3.0 Å). We propose that the catalytic Lys 155 forms a Schiff base with the acetaldehyde carbonyl, whereas Asp 92 and Lys 184 are part of the BH1352 proton relay system involved in imine deprotonation to form an enamine (Fig. S7). This is consistent with the results of alanine replacement mutagenesis of BH1352 with the respective mutant proteins (D92A, K126A, K155A, and K184A) showing very low or no catalytic activity both in the DRP cleavage and in acetaldehyde condensation reactions (Fig. 5). This is also supported by the crystal structure of TM1559 in complex with citrate and glycerol (PDB code 3R12), indicating that its active site includes Asp 117 , Lys 150 , Lys 179 (catalytic), and Lys 208 (Fig. S8). Another crystal structure of TM1559 (PDB code 3R13) also revealed the presence of additional electron density in the active site representing an unknown ligand covalently bound to the catalytic Lys 179 (likely representing one of the reaction intermediates).
To identify other BH1352 residues involved in substrate binding, we modeled DRP into the BH1352 active site using the structure of the DeoC-DRP complex of DeoC from E. coli (Fig.  3B) (16,43). The produced model of DRP binding in the BH1352 active site predicts that the side chain of Thr 12 appears to be involved in substrate coordination via hydrogen bonding with the ␤-hydroxyl group of DRP, as well as with Lys 184 and a water molecule (Wat 26 ; Fig. 3B). This is consistent with the results of site-directed mutagenesis, which revealed that Ala replacement of Thr 12 resulted in a catalytic impairment in the DRP cleavage reaction (Fig. 5A). However, the T12A mutant protein exhibited acetaldehyde condensation activity comparable with that of the WT BH1352, suggesting that this residue is not critical for acetaldehyde condensation (Fig. 5B).
The DRP-binding model also suggested that the DeoC Lys 172 (interacting with the DRP phosphate and ␥-hydroxyl groups) is replaced by Phe 160 in BH1352 (Figs. 6 and 7). Moreover, it proposes that the highly conserved BH1352 Arg 190 and DeoC Arg 207 (located close to the conserved phosphate-binding Glyrich loop) might also be involved in the coordination of the DRP phosphate through the bound water molecule (like Lys 172 in DeoC). In addition, BH1352 Lys 15 appears to interact with the phosphate and ␥-hydroxyl groups of DRP. Based on the DERA sequence alignment, BH1352 Lys 15 is conserved in DERAs from Bacilli (group 2), whereas the proteobacterial DERAs (group 4) contain an Asn residue at this position (Asn 21 in DeoC) (Figs. 1  and 7). Ala replacement mutagenesis of Lys 15 rendered BH1352 completely inactive in both retro-aldol and acetaldehyde con-

Engineering DERAs for the biosynthesis of 1,3BDO
densation reactions (Fig. 5), indicating that this residue plays an important role in catalytic activity of this enzyme.
Another notable feature of the BH1352 and TM1559 structures is the presence of a cluster of hydrophobic residues near the catalytic Lys (Lys 155 in BH1352), including four residues conserved in all DERAs (Leu 14 , Val 63 , Phe 66 , and Ile 128 in BH1352 and Leu 40 , Val 88 , Phe 91 , Ile 152 , and Phe 184 in TM1559) (Fig. 6C, Fig. S8, and Table S3). In the BH1352 structure, the side chains of Leu 14 , Phe 66 , and Ile 128 are oriented toward the ␣-carbon of aldol products (Fig. 6), suggesting that these residues provide hydrophobic contacts for ligand binding and that they might be essential for enzyme activity. This was supported by the results of alanine replacement mutagenesis of these residues, which produced mutant proteins with a greatly reduced activity in both reactions (the L14A protein was found to be insoluble) (Fig.  5). Another hydrophobic cluster comprising of three valine residues (Val 154 , Val 177 , and Val 183 ), Ile 170 , and Met 173 is located between the two ␤-strands (␤6 and ␤7) and the ␣6 helix (in BH1352) (Fig. 6). It was previously reported that this cluster may contribute to the sequential aldol condensation, as revealed by the DeoC mutations F200I and M185V (equivalent to Val 183 and Met 173 in BH1352), resulting in enhanced condensation of acetaldehyde and chloroacetaldehyde (27). Also, DeoC Phe 200 is replaced by Val in DERAs from T. maritima and Pyrobaculum aerophilum, both of which show higher sequential aldol condensation of acetaldehyde (24). These results suggest that reducing the size of hydrophobic side chains in this cluster might contribute to higher aldol condensation activity.
Recently, it has been shown that the C-terminal Tyr 259 of the E. coli DeoC is required for the efficient proton abstraction step in the DRP cleavage reaction (18), whereas the previous work with the truncated DeoC ⌬Y259 protein (Tyr 259 deleted) demonstrated an enhanced activity in acetaldehyde condensation with chloroacetaldehyde (27). Because the BH1352 structure suggested that the C-terminal Tyr 224 might directly contribute to substrate binding or activity of this enzyme (Fig. 4), sitedirected mutagenesis was also used to ascertain the role of this residue. We designed and purified four Tyr 224 mutant proteins including Y224A, Y224F, ⌬Y224 (Tyr224 deleted), and ⌬S223/ Y224 (Ser 223 and Tyr 224 deleted) and tested their catalytic activities in the DRP cleavage and acetaldehyde condensation (1,3BDO production) reactions. Interestingly, although the acetaldehyde condensation reactions of these mutant proteins were not affected, their retro-aldol activity was greatly reduced (especially in Y224F), indicating that Tyr 224 is essential for DRP cleavage but not for acetaldehyde condensation (Fig. 5). Thus, the crystal structures of BH1352 and other DERAs from different phylogenetic groups revealed significant differences in substrate coordination and catalysis of DRP cleavage and acetaldehyde condensation.

Structure-based engineering of BH1352 for enhanced production of 1,3BDO
The crystal structures of BH1352 and TM1559 revealed that their substrate-binding pockets also include the side chain of a  red boxes in A and B), each displayed in detail on the right.

Engineering DERAs for the biosynthesis of 1,3BDO
semiconserved Phe (Phe 160 in BH1352 and Phe 184 in TM1559) ( Fig. 6 and Fig. S9). This residue is conserved in most DERAs from clusters 1 (mixed group) and 2 (Firmicutes), but it is replaced by a Lys residue in Proteobacterial DERAs (cluster 4) including E. coli DeoC (Fig. 7). In the Lactobacillus brevis DERA (LbDERA), the replacement of the homologous Phe 163 by Tyr has been shown to result in enhanced sequential condensation of acetaldehyde and chloroacetaldehyde, probably by promoting substrate access (30). We found that the BH1352 Phe 160 was not essential both for the retro-aldol (DRP cleavage) and acetaldehyde condensation reactions, because the F160A mutation had no significant effect on both reactions (Fig. 5). However, the DRP cleavage activity of BH1352 was negatively affected when Phe 160 was mutated to Glu, Gln, Lys, Met, Trp, or His and slightly stimulated by mutation to Tyr (ϳ23%) (Fig.  5A). Interestingly, the acetaldehyde condensation via BH1352 increased almost three times in the F160Y protein and was also increased in the F160E (72%) and F160H (44%) proteins ( Fig. 5B  and Fig. S10). In contrast, the replacement of Phe 160 with Lys, Gln, Met, or Trp had no significant effect on this activity. These results suggest that similar to LbDERA (30), the substitution of Phe 160 by Tyr in BH1352 enhances acetaldehyde binding and/or condensation but has no effect on DRP cleavage (Fig.  S9). Based on the BH1352 crystal structure, the hydroxyl group of Tyr 160 (in F160Y) might interact with the main chain amide of conserved Lys 15 (3.3 Å) located on the ␤1-␣1 loop (Leu 13 -Thr 19 ) near the absolutely conserved Leu 14 (Fig. S9C). Our mutagenesis studies demonstrated that Lys 15 is critical for catalytic activity of BH1352, whereas Leu 14 is part of the hydrophobic cluster near the catalytic Lys 155 (L14A mutant protein was found to be insoluble) (Figs. 5 and 6). We propose that the hydroxyl group of Tyr 160 provides a stabilizing effect on the conformation of both Leu 14 and Lys 15 in the BH1352 active site, resulting in increased acetaldehyde condensation activity of this enzyme.
We also mutated the semiconserved residues Ile 170 and Met 173 of BH1352, located near the catalytic Lys 155 (Fig. 6D), to examine whether the reduction or increase of their hydrophobic side chains will affect the catalytic activity of BH1352 and improve acetaldehyde condensation. Our coupled DERA-AKR assays (1,3BDO production) revealed a 40 -50% increase in the production of 1,3BDO by the purified mutant proteins I170V, M173I, M173L, and M173V compared with the WT BH1352, whereas I170A showed reduced activity (Fig. 5). In contrast, the retro-aldol (DRP cleavage) activity of BH1352 was not signifi-

Engineering DERAs for the biosynthesis of 1,3BDO
cantly affected by these mutations. Interestingly, the replacement of Met 173 by a polar residue (Thr) had a strong negative effect on both BH1352 activities, indicating that retaining hydrophobicity at this position is critical for catalytic activity of this enzyme. We also designed the BH1352 double mutant protein F160Y/M173I, which showed 1,3BDO formation activity comparable with that of F160Y, suggesting that these two mutations are not synergistic in acetaldehyde condensation (Fig. 5). Thus, both BH1352 F160Y and F160Y/M173I proteins exhibit enhanced activity in acetaldehyde condensation reaction and can be used for in vitro and in vivo production of 1,3BDO.

Application of engineered BH1352 variants for in vivo production of 1,3BDO in E. coli
Recently, we demonstrated the production of 1,3BDO by engineered E. coli cells expressing a heterologous, aldolasebased pathway containing the WT BH1352, PA1127, and PDC from Zymomonas mobilis (33). Because this work suggested that 1,3BDO production is limited by the activity of BH1352, we designed two novel E. coli strains expressing either the BH1352 F160Y (BDO-1) or F160Y/M173I (BDO-2) variants and compared 1,3BDO production with the original strain expressing the WT BH1352 (BDO-0). In these experiments, we used the E. coli strain LMSE51C with several nonessential genes deleted including pyruvate-formate lyase (pflB), lactate dehydrogenase (ldhA), acetolactate synthase (ilvB), and aldehyde/alcohol dehydrogenase (adhE) (with the goal of increasing the carbon flux to 1,3BDO and reducing byproduct formation) (Fig. S11) (33). The three E. coli strains (BDO-0, BDO-1, and BDO-2) were grown in a bioreactor with pH control (maintained at 7.0) using a fed-batch aerobic fermentation performed by the addition of extra glucose (3%) after the induction of 1,3BDO pathway expression by isopropyl ␤-D-thiogalactopyranoside addition (1 mM). After the cultivation, the BDO-0 strain (WT BH1352) produced 0.2 g/liter of 1,3BDO with a yield of 4 mg/g of glucose (Fig. 8). Under the same experimental conditions, BDO-1 (BH1352 F160Y) produced 0.9 g/liter of 1,3BDO with a yield of 18 mg/g of glucose representing a 4.5-fold increase both in the 1,3BDO titer and yield (Fig. 8). Even higher 1,3BDO production was observed in the BDO-2 strain, which produced up to 1.1 g/liter of 1,3BDO with a yield of 28 mg/g of glucose (a 5.5-and 7-fold increase, respectively) (Fig. 8). Different titers of 1,3BDO produced by the BDO-1 and BDO-2 strains might be due to slightly different expression levels of the BH1352 F160A and F160A/M173I proteins in E. coli cells. Thus, using structurebased protein engineering, we have identified two BH1352 mutant proteins supporting enhanced biosynthesis of 1,3BDO both in vitro (from acetaldehyde) and in vivo (from glucose).

Conclusions
Using a combination of purified DERAs and an aldo-keto reductase (PA1127), we have identified three microbial DERAs with high activity in the transformation of acetaldehyde to 1,3BDO. The crystal structure and site-directed mutagenesis of BH1352 provided insights into the molecular mechanisms of substrate selectivity and acetaldehyde condensation activity of DERAs. By targeting hydrophobic residues near the catalytic Lys 155 of BH1352, we generated two variants of this enzyme (F160Y and F160Y/M173I) with enhanced activity in acetaldehyde condensation and 1,3BDO production. E. coli cells expressing these BH1352 variants as part of the DERA ϩ AKR pathway produced 5-6 times more 1,3BDO from glucose compared with cells with the WT BH1352. The designed BH1352 variants can be used as a starting material for future protein engineering efforts aimed at improving the activity of DERAs and their performance in the biotechnological production of 1,3BDO and other chemicals.

Phylogenetic and sequence analyses
The phylogenetic tree was generated by retrieving 2,553 sequences from UniProt using Kyoto Encyclopedia of Genes and Genomes Orthology identifier K01619, which represents DERAs (EC 4.1.2.4) involved in the pentose phosphate pathway. The original data set was reduced to 1,974 sequences by removing redundant sequences and increasing gap-free sites using CD-HIT and MaxAlign using MAFFT online alignment (https://mafft.cbrc.jp/alignment/server/) 3 (47)(48)(49). The tree was built using FastTree 2.1.5 and visualized by Interactive Tree of Life (http://itol.embl.de/) 3 (50,51). The DERA sequence alignment and phylogenetic analysis were conducted as described in our previous study (32). Structural images of BH1352 were prepared using PyMOL Molecular Graphics System, version 1.8 (Schrödinger, LLC).  TM1559 (center row). The secondary structure elements derived from the structures of BH1352 and E. coli DeoC are shown above and below the alignment, respectively. Residues conserved in all proteins are shown in white type on a red background. The columns with red residues indicate the presence of more than 70% of biochemically similar residues. The catalytic residues are indicated by cyan boxes with red residue numbers, whereas the columns with black boxes and residue numbers indicate the substrate entrance residues. The residues of the hydrophobic amino acid clusters are labeled with black circles.

Engineering DERAs for the biosynthesis of 1,3BDO Gene cloning, protein purification, and mutagenesis
The DERA genes studied in this work (Table S1) were cloned, overexpressed in E. coli, and affinity-purified (Fig. S1) as described previously (52). Site-directed mutagenesis of BH1352 was performed using the Phusion high-fidelity DNA polymerase (New England BioLabs) accordingly to the manufacturer's protocol.

Protein crystallization and structure determination
Purified BH1352 was crystallized at room temperature using the sitting-drop vapor-diffusion method using protein concentration of 10 mg/ml and reservoir solution of 0.1 M Tris-HCl (pH 8.5), 0.2 M magnesium chloride, 25% (w/v) PEG 3350, and 10 mM acetaldehyde. The crystal was cryoprotected in the same buffer supplemented with 2% PEG 200 and flash-frozen in liquid nitrogen. Diffraction data for the BH1352 apoenzyme crystal were collected at 100-K at a Rigaku home source Micromax-007 with R-AXIS IVϩϩ detector. Diffraction data were processed using HKL3000 (53). The structure was solved by molecular replacement using Phenix phaser and the structure of a putative aldolase (PDB code 3NGJ) (54). Model building and refinement were performed using Phenix.refine and COOT (54,55). TLS parameterization was utilized, and B-factors were refined as isotropic. Structure geometry and validation were performed using the Phenix MolProbity tools. The data collection and refinement statistics are summarized in Table 2.

Enzyme assays
Purified DERAs were initially screened using a DERA-AKR coupled assay with 50 mM acetaldehyde as substrate in the following reaction mixture (0.2 ml): 100 mM triethanolamine buffer (pH 7.5), 10 mM NADPH, DERA (250 g/ml), and AKR (PA1127, 250 g/ml). The production of 1,3BDO was measured using HPLC, following 2 h of incubation at room temper-ature (used in DERA-catalyzed reactions) (22). The reaction samples were filtered through centrifugal filter device (10,000 cut-off, VWR) to remove enzymes and dried to get rid of residual acetaldehyde from the samples using a vacuum concentrator. The dry samples were dissolved in the same volume of double-distilled H 2 O and analyzed using HPLC (Dionex Ultimate 3000, Thermo Scientific) equipped with an Aminex HPX-87H column, equilibrated with 5 mM H 2 SO 4 as an eluent with a flow rate of 0.6 ml/min at 50°C. 1,3BDO was detected using a refractive index detector (Shodex RI-101). Assay conditions were optimized by varying concentrations of DERA, AKR, and NADPH, and the optimal conditions included 100 g/ml each of DERA and AKR and 10 mM NADPH (Fig. S2).
The kinetic parameters of purified DERA were determined using DRP cleavage reaction using a glyceraldehyde-3-phosphate dehydrogenase/triosephosphate isomerase (GDH/TPI)coupled assay. The DERA-catalyzed retro-aldol reaction produces acetaldehyde and D-glyceraldehyde-3-phosphate, which is converted into dihydroxyacetone phosphate by TPI and further reduced by GDH consuming NADH. The detailed assay conditions were as follows: 100 mM triethanolamine buffer, pH 8.5, 0.5 mM NADH, WT or mutant DERA (1 g/ml), TPI (11 units/ml), GDH (1 unit/ml), and DRP (from 4 M to 4 mM) in a 200-l reaction mixture at 30°C. The kinetic parameters were calculated by a nonlinear regression analysis of raw data fit to the sigmoidal function using GraphPad Prism software (version 5.04 for Windows).
For the analysis of DERA resistance against acetaldehyde, a freshly prepared acetaldehyde solution (final concentration, 100 mM) was added to the incubation mixture containing 2 mg/ml of purified BH1352 (WT or mutant proteins). The incubation solution aliquots were taken and diluted for further use in a DRP cleavage assay (1 mM DRP). The activity of DERA samples was analyzed immediately after acetaldehyde addition and then at regular time intervals. The residual DERA activity was calculated by comparison with control samples without acetaldehyde (containing enzymes and buffer).

Strains and plasmids
The strains and plasmids used in this study were adopted from the previous work and are listed in Table S4 (33,34). Expression of pBD1 (pTrC99A harboring BH1352, PA1127, and PDC from Z. mobilis) in LMSE51C was used as the WT control (BDO-0) to demonstrate the in vivo effect of the mutations in BH1352.

E. coli cultivation in mini-bioreactors
For in vivo studies on 1,3BDO production, the designed E. coli strains were cultivated in 500-ml bioreactors (Applikon Biotechnology Inc.) equipped with Rushton impellers and electrodes for pH and dissolved oxygen essentially as described previously (33). The first seed culture was prepared by inoculating 10 ml of LB supplemented with 100 g/ml of ampicillin from a single colony and grown at 37°C. 50 ml of modified M9 medium (supplemented with 100 g/ml of ampicillin and 0.5 mg/ml of thiamine and containing 0.1 M MOPS at pH 7.3) was inoculated with the first seed culture in a 250-ml baffled flask and grown at 37°C and 200 rpm for 16 h. The second seed Engineering DERAs for the biosynthesis of 1,3BDO culture was then used to inoculate 300 ml of modified M9 medium (without MOPS) and supplemented with 100 g/ml of carbenicillin in the bioreactors. The pH was controlled at 7.0 by the addition of 10% NH 4 OH, stirrer speed at 1,500 rpm, temperature at 37°C, and air flow rate at 1.5 vvm. When the culture density reached A 600 nm between 7 and 8, protein expression was induced by the addition of 1 mM isopropyl ␤-D-thiogalactopyranoside. After 30 min, the air flow rate was reduced to 0.37 vvm (25% of the initial vvm) to reduce dissolved oxygen, and 3% glucose was additionally supplemented.