Structural insights into alcohol dehydrogenases catalyzing asymmetric reductions.

Abstract Alcohol dehydrogenases are a group of oxidoreductases that specifically use NAD(P)+ or NAD(P)H as cofactors for electron acceptance or donation and catalyze interconversion between alcohols and corresponding carbonyl compounds. In addition to their physiological roles in metabolizing alcohols and aldehydes or ketones, alcohol dehydrogenases have received considerable attention with respect to their symmetry-breaking traits in catalyzing asymmetric reactions and have Accordingly, they have become widely applied in fine chemical synthesis, particularly in the production of chiral alcohols and hydroxyl compounds that are key elements in the synthesis of active pharmaceutical ingredients (API) employed in the pharmaceutical industry. The application of structural bioinformatics to the study of functional enzymes and recent scientific breakthroughs in modern molecular biotechnology provide us with an effective alternative to gain an understanding of the molecular mechanisms involved in asymmetric bioreactions and in overcoming the limitations of enzyme availability. In this review, we discuss molecular mechanisms underlying alcohol dehydrogenase-mediated asymmetric reactions, based on protein structure–function relationships from domain structure to functional active sites. The molecular principles of the catalytic machinery involving stereochemical recognition and molecular interaction are also addressed. In addition, the diversity of enzymatic functions and properties, for example, enantioselectivity, substrate specificity, cofactor dependence, metal requirement, and stability in terms of organic solvent tolerance and thermostability, are also discussed and based on a comparative analysis of high-resolution 3 D structures of representative alcohol dehydrogenases.


Introduction
Alcohol dehydrogenases (ADHs) (E.C. 1.1.1.X, where X ranges from 1 to 411) (https://www.brenda-enzymes. org/ecexplorer.php?browser¼1; last accessed 12 November 2018), are a class of oxidoreductases, that catalyze electron transfer between a reductant (electron donor) and an oxidant (electron acceptor). Given their function in catalyzing the reduction of ketones or the oxidation of alcohols via hydride transfer between substrate and NAD(P) þ or NAD(P)H cofactors, ADHs are also referred to as carbonyl reductases, and typically act on hydroxyl or carbonyl compounds that are associated with biologically and pharmacologically active substrates ( Figure 1) [1,2]. Furthermore, stereospecific ADHs possess inherent advantages over chemo-catalysts in terms of their highly chemo-, regio-, and enantioselectivity, resulting in excellent yields and an excess of high enantiomeric products [3,4]. Therefore, stereospecific ADHs have been widely used and attract considerable interest with respect to the production of optically active hydroxyl compounds for active pharmaceutical ingredients (API) and fine chemicals [5][6][7].
Robust ADHs with a suitable utility and availability are necessary for industrial applications. These compounds are ubiquitous in nature and are present in a wide variety of organisms [8,9]. Microorganisms are currently still the main sources of stereoselective ADHs due to the advantage of convenient use, for example, Lactobacillus kefir, Rhodococcus ruber, Rhodotorula sp, Saccharomyces cerevisiae, and Thermus thermophilus. The enzymes discovered and characterized from different sources exhibit distinct physical and enzymatic properties and have accordingly been classified into different superfamilies or families based on their characteristics and protein structure [10].
Despite the diversity of ADHs, the range of their applications has to date remained relatively modest, particularly for industrial applications [11][12][13]. There are disproportionately few examples of the commercialscale application of biocatalysts for the manufacture of fine chemicals, owing to their limitations in terms of activity, substrate spectrum, specificity and sensitivity to typical processing conditions in chemical plants [11,12,14]. However, companies such as Codexis and Merck have developed a series of ADHs for commercial scale applications (Supplementary Table 1). Recent scientific and technical advances in structural biology, protein engineering, and bioinformatics have, nevertheless, opened the door for the development of tailor-made enzymes for industrial purposes and will contribute to the exploration of biodiversity. Besides the protein structure of ADHs, genome mining and a docking model can reveal some novel ADHs suitable for practical applications that can be used to understand this molecular mechanism [15][16][17][18][19]. These advances will facilitate the discovery and optimization of new ADHs customized to overcome the present limitations and thereby expand their industrial applications [20][21][22]. Moreover, modulation of catalytic functions based on a knowledge of their enzyme protein structure can also contribute to advancing our understanding of the molecular mechanisms underlying the ADH-mediated catalysis of asymmetric reactions, as well as interactions between active sites and ligands, including cofactors and their hydroxyl or carbonyl compounds.
Regarding the stereospecificity of ADHs, previously published literature reviews have mainly focused on reaction route and type, reaction system, reaction engineering, and their application [6,7,23]. In this review, we will discuss structural aspects of the molecular mechanisms underlying ADH-mediated asymmetric reactions, including structural profiles and features, and the catalytic machinery involving stereochemical recognition and molecular interactions, in light of recent advances in biotechnology during the past decade. The diversity of enzymatic functions and properties, for example, enantioselectivity, substrate specificity, cofactor dependence, metal requirement, organic solvent tolerance, and thermostability, are also discussed based on comparative analysis of high-resolution 3 D structures of representative ADHs. Obtaining an understanding of the molecular mechanisms underlying the stereospecificity of ADHs will provide a research foundation for further modification of these enzymes as industrially advantageous biocatalysts.

Overview on various types of ADHs
The structures of more than 2064 ADH proteins from various sources, including eukaryotes, bacteria, and archaea, have been described as a result of X-ray crystallography in PDB (http://www.pdb.org, Query Parameters: Text Search for: dehydrogenase or reductase, and Enzyme Classification is 1: Oxidoreductases and Enzyme Classification is 1.1: Acting on the CH-OH group of donors and Enzyme Classification is 1.1.1: With NAD(þ) or NADP(þ) as acceptor). However, studies on molecular mechanisms involved in ADH-catalyzed asymmetric reactions are still in their infancy.
Supplementary Tables 2 and 3 summarize the structural information and properties of the stereospecific ADHs with known 3 D structures. On the basis of protein chain length, conserved motif, and structural and mechanistic features, ADHs have been classified into three classes that have similar functions but different structures and mechanisms: short-chain dehydrogenase/ reductases (SDR), medium-chain dehydrogenase/reductases (MDR), and long-chain dehydrogenase/reductases (LDR), [24]. SDRs, MDRs, and LDRs all have conserved Rossmann fold structures for cofactor binding but are distinct in terms of their substrate-binding pocket structures and also overall protein structure ( Figure 2). SDRs tend to be small compact proteins containing a single domain, whereas MDRs and LDRs are somewhat larger and have a two-domain structure, namely a catalytic domain and a cofactor-binding domain [25].
SDRs comprise a large superfamily of enzymes that act on a broad spectrum of substrates that possess various chemical structures [26]. From the structural profile, the N-terminal sequences superimpose well and the Rossmann fold responsible for cofactor binding resides within the top of the b-sheets. The C-terminal portion generally functions in substrate binding, and the obvious structural variation in this region results in the diversity of substrate specificities ( Supplementary  Figures 1 and 2) [27]. On the basis of the cofactor-binding motif and the involved active sites, SDRs are classified into the following five families: classical, extended, intermediate, complex, and divergent SDRs [24].
The classical and extended family members are the major types of SDRs. The classical, extended, and intermediate SDRs have the same active site (Tyr-x-x-x-Lys) in the C-terminal region, but have different coenzyme-binding motifs (Thr-Gly-x-x-x-Ala, Thr-Gly-x-x-Glyxx-Gly, and Gly/Ala-x-x-Gly-x-x-Gly/Ala, respectively) in the N-terminal region [24]. However, the coenzymebinding Gly-motif of the divergent family (Gly-x-x-x-x-x-Ser-x-Ala) is differently spaced compared to the corresponding motif in other SDR families, and the active site has a Tyr-x-x-Met-x-x-x-Lys motif instead of Tyr-x-xx-Lys. The divergent family encompasses enoyl reductases from bacteria and plants. The complex family is named after its members, which are components of multifunctional enzyme complexes present in all forms of life, for example, fatty acid synthase. These are NADP(H)-binding proteins with the SDR region fulfilling a beta-ketoacyl reductive function. The members of this group harbor the unique Tyr-x-x-x-Asn motif at the active site rather than the Tyr-x-x-x-Lys motif commonly found in other SDRs, as well as a Gly-Gly-x-Gly-x-x-Gly coenzyme-binding motif.
In contrast to SDRs, MDRs have two functional domains, the catalytic and cofactor-binding domains ( Supplementary Figures 3 and 4). On the basis of differences in sequence patterns, the MDR members are generally classified into several subdivisions, including alcohol dehydrogenases (ADH), cinnamyl alcohol dehydrogenases (CAD), yeast alcohol dehydrogenases (YADH), mitochondrial response proteins (MRF), acetyl-CoA reductases (ACRs), leukotriene B4 dehydrogenases (LTD), polyol dehydrogenases (PDH), quinone oxidoreductase (QOR) and so on [28]. The structure of the ADHs from Clostridium beijerinckii (CBADH, PDB ID: 1PED) and Thermoanaerobacter brockii (TBADH, PDB ID: 1YKF) were the first prokaryote NADP(H)-dependent ADH structures to be reported. The cofactor-binding domain and the catalytic domain in their monomer structures are separated by a deep cleft possessing a single zinc atom in the catalytic site [29]. In addition, an NAD(H)-dependent medium-chain ADH from the hyperthermophilic bacterium Sulfolobus solfataricus (SsADH, PDB ID: 1JVB) has also been identified [30]. Compared with other known ADH structures, the interdomain cleft of SsADH is significantly larger. Moreover, the loop structures of the conserved coenzyme-binding and catalytic domains are more flexible and variable. In terms of subunit structural traits, the oligo structures of ADHs have certain inherent features. In higher plants and mammals, the zinc-containing ADHs, such as horse liver ADH (HLADH), generally occur as dimers, whereas those from bacteria and yeasts are generally tetramers.
Among the three ADH superfamilies, the LDRs are a relatively heterogeneous group of proteins. Different LDRs do not have entire-chain homology but have been shown to display molecular similarities with respect to sequence segments. Similar to MDRs, LDRs also display a somewhat larger two-domain structure comprising the catalytic and cofactor-binding domains. For example, polyol-specific long-chain dehydrogenases/ reductases (PSLDRs), such as mannitol 2-dehydrogenase from Pseudomonas fluorescens (PfM2DH, PDB ID: 1M2W) [31]. The crystal structures of PfM2DH revealed that this protein comprises two domains, the N-terminal a/b domain containing a nucleotide-binding region and the C-terminal a-helical catalytic domain, which are connected by a tetra-peptide linker of sequence Thr-Asp-Asp-Val [31].

Catalytic mode and conformational change
The high-resolution 3 D structure of ADHs suggests that the catalytic machinery is formed by the conserved key sites and associated ligands in the catalytic domain. In SDR structures, the proton relay system comprises different functional molecules and groups, involving Tyr-OH, 2 0 -OH of the nicotinamide ribose, a Lys residue, and a water molecule positioned by Asn or a homologous Ser [32]. In the inherent catalytic mechanism of SDRs, the involved Tyr plays the role of a general base for proton transfer, Lys lowers the pKa value of the hydroxyl group of Tyr, and Ser is involved in the binding of substrate/product and/or the hydroxyl group of Tyr. The Asn residue in the extended catalytic tetrad functions in maintaining the configuration of active sites for proton relay [33,34]. In the carbonyl reductase from Sporobolomyces salmonicolor (SSCR, PDB ID: 1ZZE; 1Y1P), the hydrogen bond of the Asn207 side chain with the substrate halogen atom and the XH/p interaction of the substrate phenyl group with the side chains of Ser222/Thr223 resulted in the formation of the highly reactive conformation of a-halogenated acetophenones in the active site of the enzyme [35]. In ketoreductases (KREDs, reduction of 3-thiacyclopentanone to 3-thiacyclopentanol) from Lactobacillus kefir, the most commonly used enzymes in industrial pharmaceutical synthesis, Ser143 and Tyr156 stabilize the alkoxide formed on hydride transfer from NADPH. The alcohol is then liberated through a proton relay involving Tyr156, the cofactor ribofunarose, Lys160, a backbone carbonyl, and water ( Figure 3) [36]. In addition to the known roles of the conserved sites in SDRs, study of the (R)-specific ADH from L. brevis (LB-RADH, PDB ID: 1NXQ and 1ZK4, reduction of prochiral ketones to the corresponding secondary alcohols) has extended the above concept and led to the proposal of an "extended proton relay system." Conformational flexibility close to the Ser residue of the catalytic tetrad has been indicated to give more space for the accommodation of an additional water molecule between Tyr and the nicotinamide nucleoside moiety of the cofactor, which would be a further component for proton relay [37].
In MDR, the holo-structures of the ADH from Clostridium beijerinckii (CBADH, PDB ID: 1PED, reduction of ketones to the corresponding secondary alcohols) and TBADH (PDB ID: 1YKF, reduction of ketones to the corresponding secondary alcohols) suggest that Asp150 plays an important role in the determination enzyme stereospecificity. Regarding the direction of hydride transfer, the oxygen atom of Asp150 properly locates towards the 4-position of the nicotinamide group of NADP(H) to form a hydrogen bond with both pro-S and pro-R hydrogen atoms [33].
For LDRs, the strictly conserved Lys fulfills as a proton acceptor. This is different from the SDR superfamily which utilizes a catalytic Tyr as a general base for proton transfer. For example, Lys-295 of PfM2DH is poised to act as the general base in the reaction. Asn-191 and Asn-300 are positioned to direct the O2 proton toward Lys-295 and stabilize a negative charge on the substrate oxygen in the transition state [31]. Corresponding to the principle of proton relay, ADHmediated reduction proceeds through the following steps via an ordered "bi-bi" mechanism. Initially, the free form of enzyme and the cofactor forms a complex, which then binds to the carbonyl substrate, and hydride is transferred from the cofactor (NAD(P)H) to the carbonyl substrate. Thereafter, the hydroxyl product formed and the oxidized cofactor are sequentially released from the enzyme [38].
In proteins, conformational change is generally related to ligand binding and subsequent molecular interactions [39]. For SDRs, conformational change in terms of structural adaptability was first reported for LB-RADH [37]. Structural comparisons of apo-LB-RADH-G37D under different crystallization conditions (PDB ID: 1ZK0, 1ZK1, 1ZK2, 1ZK3, 1ZK4, 1ZJY, and 1ZJZ) has indicated that conformational change clearly occurs in the cofactor-binding region encompassing residues 35-46 and the substrate-binding region of the loop structure ( Figure 4). In addition, conformation differences between the apo-and the ternary complex structures of LB-RADH-G37D further indicates the key flexible region around Ser141, with the neighboring Ser142 belonging to the conserved catalytic tetrad of SDRs. In detail, the carbonyl group of Ser141 in the backbone is oriented away from the bound cofactor and the water molecule in the ternary complex, whereas in the apo-structure, the same group is oriented toward the cofactor-binding site. Therefore, Ser141 displays diverse conformations in both the side chain and the main chain, and the conformational change within the main chain is assumed to accommodate the cofactor and water molecule. Within the structure of the complex formed between LB-RADH and phenylethanol, the side chain conformations of Tyr189, Glu144, and Met205 are also modified for ligand binding and molecular interactions [37].
For MDRs, conformational changes and the related mechanisms have been reported based on the structure of HLADH, a typical NAD(H)-dependent MDR. Upon cofactor binding, the conformation of the interdomain crevice between the cofactor-binding domain and the substrate-binding domain shifts from an "open" to "closed" format [40]. In addition, the conformational change of MDRs has been found to involve movement of the catalytic domain. With recent progress in bioand computational chemistry, newly developed modeling and simulation methods, especially quantum mechanical/molecular mechanics (QM/MM) calculations and molecular dynamics (MD) simulations, provide powerful tools to understand the dynamic behavior of enzymes interacting with ligands and predict the corresponding molecular mechanism [41][42][43][44]. Zinc-dependent carbonyl reductase from Candida parapsilosis (CPCR2, PDB ID: 4C4O), belongs to the MDR family, shows two different conformers of Glu66 and two positions of the catalytic zinc ion. The dependence of barriers for the hydride transfer for these two states in the reduction of carbonyl substrate were analyzed using QM/MM steered molecular dynamics simulations. The results of calculations show that the catalytic state (Zn cat ÀGlu out ) has a $20 kcal/mol lower reaction barrier in comparison to the resting state (Zn rest ÀGlu in ). This indicates that the coupled movement of zinc ion and Glu influences not only the ligand exchange but also the catalytic process of MDRs [43].
In LDRs, conformational changes have been reported based on the structure of PfM2DH (reversible oxidation of D-mannitol to D-fructose, D-arabinitol to D-xylulose, and D-sorbitol to L-sorbose). In converting mannitol to fructose, Lys-295 of PfM2DH is positioned to act as the proton acceptor. Hydrogens on the amide nitrogens of Asn-191 and Asn-300 would stabilize the partial negative charge on the O2 in the transition state. In functioning as hydrogen bond donors to the two lone pairs of the substrate oxygen, the asparagines also direct the O2 proton toward Lys-295. Carbonyls of Val-229 and Asn-300 accept hydrogen bonds from Lys-295 and direct the lone pair of electrons on lysine to accept a hydrogen bond from the substrate O2. The side-chain conformation of Asn-300 is therefore critical for catalysis because it functions in orienting both enzyme and substrate groups. The Asn-300 side chain is additionally oriented by a hydrogen bonding interaction with the mannitol O5 [30].

Enantioselectivity
For substrates of a certain category, stereospecific ADHs are generally referred to as (R)-or (S)-specific enzymes, according to the enzymatic selectivity for the chiral center in the substrate/product compound, where the stereoisomer of a chiral molecule is defined following the Cahn-Ingold-Prelog (CIP) rule. In this manner, the stereochemical reaction pattern depends on the attack from the pro-R-or pro-S-hydride of NAD(P)H to the si or re face of the sp 2 -hybridized carbon atom in the C¼O moiety ( Figure 5). Stereoselective ADHs can accordingly be divided into two types, following Prelog's rule or anti-Prelog, in terms of the stereochemical outcomes [45]. For SDRs, the cofactor generally binds in an extended conformation to facilitate 4-pro-S hydride transfer, whereas MDRs catalyze the transfer of the 4-pro-R hydride [24].
The enantioselectivity of ADHs is essentially based on the steric conformation of the substrate-binding pocket and the chemical structure of the substrate/ product. In general, the size difference of the two groups flanking the prochiral center of the carbonyl substrate has a significant influence on the enantioselectivity of enzymes [46]. The models for the carbonyl reductase from Sporobolomyces salmonicolor (SSCR, PDB ID: 1ZZE; 1Y1P) with different carbonyl compounds enable us to elucidate the space orientation of the substrates. The nicotinamide ring of the cofactor is positioned toward the re-face of ethyl phenylglyoxylate, but to the si-face of ethyl 3,3-dimethyl-2-oxobutyrate, producing ethyl (S)-2-hydroxy-2-phenylacetate and ethyl (R)-3,3-dimethyl-2-hydroxybutyrate, respectively (Supplementary Figure 5) [47].
Structure-based insights into molecular interactions between stereospecific enzymes and ligands have further enabled us to identify the functional sites directing the enantioselectivity of ADHs. Those residues with a bulky side chain, such as Tyr, Trp, and Phe, generally play an important role in determining the enantioselectivity of enzymes [37,48]. The 1-(4-hydroxyphenyl)-ethanol dehydrogenase (HPED, PDB ID: 4URE, 2EWM) from strain EbN1, which catalyzes the dehydrogenation of 1-(4-hydroxyphenyl)-ethanol to 4-hydroxyacetophenone and also catalyzes the reverse reaction, is (R)-specific, the (S)-1-phenylethanol dehydrogenase (PED, PDB ID: 2EWM)) from strain EbN1 is (S)-specific [42]. The docking models revealed amino acid side chains (HPED, Phe 187; Ped, Tyr 93) restrict the substrate binding pocket and constitute stereospecificity (Supplementary Figure 6) [49]. For PED, almost all of the recognized product enantiomers have an (S)-configuration; however, the compounds with reversed Cahn-Ingold-Prelog (CIP) priority (e.g. 2-chloro-1-phenylethanol) have an (R)-configuration. Compared with the anti-Prelog complexes, docking modeling studies indicate that the Prelog complexes have shorter distances between the benzylic carbon atoms of the substrates and the NADH nicotinamide, and stronger H-bonds between the NADH nicotinamide and Tyr154. Furthermore, the anti-Prelog models' alternate substrate conformations positioned the phenyl rings in approximately the same region as that of the Prelog models, but forced the alkyl/ester side chains into an alternative binding pocket, taking them in close proximity to the NADH nicotinamide ring, which led to loosened binding owing to steric interference. The limited space of the alternative binding pocket may also explain why anti-Prelog complex models are possible for substrates with relatively small (methyl) side chains, but not for those with bulkier side chains [50].
In addition to residue size having a steric effect on the formation of substrate-binding pockets, the charge characteristic of the sites interacting with ligands will also have an influence on the enantioselectivity of ADHs. The (6R)-2,2,6-trimethyl-1,4-cyclohexanedione (levodione) reductase (LVR) (PDB ID: 1IY8) from Corynebacterium aquaticum M-13 catalyzes the reversible reaction between levodione and (4R)-hydroxy-(6R)-2,2,6-trimethylcyclohexanone. Its complex structural model illustrates the stereospecifically catalytic mode of hydride transfer from cofactor to substrate to yield a product in the (R)-configuration with high selectivity [51]. Furthermore, mutational changes indicate that the negative charge of Glu103 interacts with the positively charged re-face of levodione and thus affects the enantioselectivity (Supplementary Figure 7) [51].

Substrate specificity
With respect to the practical application of stereoselective ADHs for the synthesis of a series of chiral products, it would be desirable to use an enzyme with a broad substrate spectrum [52]. On the basis of the shape and conformation of pockets in the catalytic domain, a given stereospecific ADH is generally specific to certain types of substrates, such as aromatic ketones, alkyl ketones, or ketoesters [53]. Thus, a knowledge of the substrate specificity of enzymes would be very valuable and instructive in guiding the assignment of appropriate candidates for defined substrates.
For substrate accommodation, the hydrophobic channel would be useful for the recognition of substrates with a linear structure that facilitates the approach of a substrate to the active center. As key sites involved in the reactions catalyzed by SDRs, the catalytic triad comprising Tyr, Ser, and Lys plays an important role in the constitution of the proton relay system. The Lys residue within the catalytic triad generally interacts with Tyr via hydrogen bonding. Other conserved sites, including specific Thr, Asn, Ser, and Ala residues, together with water molecules, are involved in the formation of a hydrophilic pocket, which apparently forms a proton bridge between the Lys residue and the bulk solvent. The structure of LB-RADH and its complex modeling reveals the docking conformation of acetophenone (substrate)/phenylethanol (product) and the involved sites. The substrate-binding hydrophobic pocket is built by the nicotinamide moiety of a cofactor and the hydrophobic region at the interface of aG helix and the loop between bE strand and aF helix. The methyl group of the ligand is buried into a pocket comprised of Ile143, Glu144, Leu152, Tyr155, Gly188, and Tyr189. The larger phenyl ring is not fully embedded in a pocket but interacts with the aromatic group of Tyr189 and the hydrophobic side chains of Ala93, Leu152, Val195, Leu198, and Met205 ( Figure 6) [37,54]. The alcohol dehydrogenase/carbonyl reductase from Candida parapsilosis (CpRCR, PDB ID: 3WLE) as a robust,  highly stereoselective biocatalyst following Prelog's rule for ketone reduction. However, CpRCR likewise exhibits a rather narrow substrate scope. Engineering a set of CpRCR variants enabled stereospecific synthesis of a broad range of chiral alcohols [55].
In contrast to the compact structure of SDRs, the structure of MDRs generally consists of two distinct domains, a large and small domain. Of these, the small domain is the substrate-binding one, and the low sequence and structural homologies of this domain can result in diverse substrate specificity [10]. In the only Zn-dependent MDR from archaea characterized to date (SsADH), the channel structure is involved in substrate entrance and a stereospecific reaction occurs upon contact between the substrate and the catalytic zinc ion. On the basis of the resolved protein structure of SsADH, it is assumed that the substrate can reach the active center through the hydrophobic channel, which induces a conformational change in the loop structure of the enzyme's catalytic domain [30]. Depending upon the distance to the metal ion, the active residues can be divided into three parts, namely, the inner (Ser40 and Trp95), central (Phe49, Ile120, Leu295, Trp117, and Leu272), and outer (Asn51, Leu52, and Leu286 from another subunit) parts (Figure 7) [30]. ADH from Thermoethanolicus brockii (TbSADH) is an attractive industrial biocatalyst, due to its high thermostability, but it also fails in such a ketone reduction. The variants I86V/W110L/L294Q and I86N/C295N are excellent catalysts for this transformation of N-Boc-3-pyrrolidinone (>99% e.e. at conversions of 98%) and 3-ketothiolane (99% e.e.), respectively [56]. TeSADH has been further engineered by generating several mutants expanding its substrate specificity. For example, W110A TeSADH is able not only to reduce phenyl-ring-containing ketones with high enantioselectivity, but it can also racemize the corresponding enantiopure alcohols [57,58]. W110A/I86A TeSADH also expanded substrate specificity to accommodate ketones bearing two sterically demanding groups [59]. Although the Trp95 residue has a significant influence on substrate specificity, the enzyme is highly active toward compounds with broad side chains, including flurbiprofenal, naproxenal, ketoprofenal, and fenoprofenal. Trp95 binds ligands in two distinct modes: one in which Leu272 and Leu295 form the hydrophobic pocket for binding naproxenal and flurbiprofenal, and the other whereby Trp117 and Phe49 connect ketoprofenal and fenoprofenal with more flexibility via p-p-interactions [30,60].
Similar to MDRs, the LDRs also have a somewhat larger two-domain structure comprising a catalytic domain and a cofactor-binding domain. The active site of PfM2DH is positioned at the interface of the two domains. Arg-373 is the only residue from the C-terminal domain that interacts directly with the bound NADþ. Important secondary structural elements that contribute residues to the active site are a1 and loop regions at the C termini of b3 and b7, b10 from the N-terminal domain, and a10 from C-terminal domain. A conserved lysine (Lys-295), located on a10, is assumed to act as the general acid/base in the reaction [31,61].
In addition, even for defined types of substrate, certain enzymes will perform dissimilar activities that correspond to a series of substrate derivatives with substituted groups. In this regard, the stereospecific carbonyl reductase from C. parapsilosis has been reported to show a catalytic preference for a-substituted acetophenone derivatives, by forming a hydrogen bond between the active sites of the enzyme and the electron-withdrawing groups at the a-position of substrates. This is consistent with the general rule that the strength of a hydrogen bond is in the order: hydroxyl > bromine > hydrogen [62]. pH can also influence the equilibrium of oxidation/reduction reactions. In reverting the reaction during ketone reduction, a hydride ion is initially transferred from NADH to the carbonyl group's carbon atom, resulting in an intermediary alkoxy (alcoholate) anion with concomitant proton transfer from the Tyr residue to the alcoholate. The protons that are used in the active site tyrosine phenolate's re-protonation are retrieved from the solvent via a proton relay system that is composed of the Asn residues and conserved Lys together with the ribose hydroxyl groups of the NADH cofactor [63,64].

Cofactor dependence
The activity of SDRs, MDRs, and LDRs is dependent on pyridine nucleotides, and in this regard, these enzymes have a structurally similar dinucleotide-binding Rossmann fold [65]. The strand topology of the Rossmann fold forms a classic cofactor-binding domain with a Gly-rich structural motif, which is critical for structural integrity and accommodation of the pyrophosphate portion of the nucleotide cofactor [23]. ADHs depend on cofactors acting as electron acceptors or donors. The cofactor requirements of ADHs are separated into two classes, that is, NAD(H) and NADP(H) dependence. In general, the presence of a positively charged residue in the cofactor-binding domain determines whether the enzyme requires a pyridine nucleotide cofactor with or without a phosphate group. Additionally, the positively charged residue is assumed to play an important role in the enzyme affinity for NADPH, rather than NADH, by forming a salt bridge with the phosphate moiety at the 2 0 -position of AMP.
The conserved active sites and functional motifs for SDRs have been identified by analysis of structure-function relationships, which have revealed the role of these elements in cofactor binding [63]. In the case of binding the adenine ring of a cofactor, a hydrophobic pocket is formed by certain conserved residues, including Leu, Val, and Ala (Leu39, Val63, and Ala 90 in PED; Leu34, Val111, and Ala89 in the L-2,3-butanediol dehydrogenase (L-BDH, PDB ID: 3A28)), An Asp residue (Asp38 in PED; Asp33 in L-BDH) forms hydrogen bonds with the two hydroxyl groups of the adenine ribose, which is conserved in SDRs to selectively bind NADH (Supplementary Figure 8) [16,33,34]. In the case of binding the nicotinamide ring, conserved Val, Ile, and Thr residues are involved in polar interactions with the nicotinamide moiety and are located in a hydrophobic wall for binding the nicotinamide in the re-face position. Furthermore, Ile and Leu form a conserved sequence of hydrophobic residues [16,33,34]. In addition, in the case of L-BDH, Gln12, Ile14, and Thr189 have molecular interactions with the pyrophosphate moiety of the cofactor by forming hydrogen bonds, and Asn88, Tyr154, and Lys158 form hydrogen bonds with the ribose moiety of cofactor (Supplementary Figure 8) [16,33,34].
With regards to the cofactor specificity of SDRs, three active sites residing in the cofactor-binding region have been proposed to determine the preference toward NADH or NADPH. Specifically, two basic residues, located in the conserved Gly-x-x-x-Gly-x-Gly motif between helix aB and strand bA and the loop between strand bB and helix aC, respectively, determine the NADPH dependence, whereas an Asp residue in the bB/ aC loop confers preferential binding for NADH. The mechanisms underlying NADPH dependence have been shown to be dependent on two basic residues, generally Arg and Lys (Arg44 and Lys48 in SSCR), which interact with the phosphate moiety of adenosine monophosphate (AMP) of the cofactor via a salt bridge [10]. LB-RADH is an exception to the "two basic sidechains" rule, in that it is strictly NADPH-dependent but has only one basic residue (Arg38) in the bB/aC loop. Therefore, this unexpected finding indicates that a single basic residue would be sufficient for an SDR enzyme to exhibit NADPH dependence [37].
The binding of a cofactor generally leads to a structural change in the mobile region, involving active site molecular interactions. Structural comparisons between apo-SSCR (PDB ID: 1ZZE) and the SSCR/NADPH complex (PDB ID: 1Y1P) have revealed that the Ile91-Tyr101 residues comprising the mobile region form the hydrogen bond network and van der Waals interactions with the cofactor by moving to the crevice between the ribose of NMN and the adenine ring of AMP. The residues between Pro216-Ser222 constitute another mobile region, in which the side-chain of Ser222 shifts in an opposite direction to form a hydrogen bond with a single phosphate group in NMN [10].
With regards to active sites in MDRs for cofactor binding, the Gly-rich sequence motif, Gly-x-Gly-x-x-Gly, is primarily associated with NAD(H) dependence, similar to that of SDRs. The G198D variant of the NAD(P)Hdependent ADH from Thermoanerobacter ethanolicus (TeSADH) showed a significant switch of cofactor dependence from NADPH towards NADH compared to the wild type enzyme [66]. In addition, Asp residues, such as Asp223 in HLADH, generally form hydrogen bonds with the adenine ribose moiety. Two basic residues, Arg-Arg or His-Arg, are generally involved in the formation of the positively charged pocket used for binding the cofactor pyrophosphate moiety. Another basic residue, Lys, located between Asp and the pyrophosphate moiety of the cofactor, is assumed to provide a positive charge and play a role in pH dependence for NAD þ binding [30]. For the NADP(H) dependence of MDRs, the residue Tyr is considered the most important site in terms of determining the cofactor preference for NADP(H) rather than for NAD(H) (CBADH, Tyr218). This residue is positioned toward the adenine ring and hydrogen bonded to a ribose phosphate oxygen of NADP(H) via its hydroxyl group (Supplementary Figure 9). However, for NAD(H)dependent MDRs, Tyr is generally replaced by Phe, which interacts with the adenine moiety of NADP(H) [29].
For LDRs, cofactor interactions are mainly determined by the N-terminal domain, which in PfM2DH has an His-x-Gly-x-Gly-x-x-x-Arg fingerprint in Helix a1. Furthermore, PfM2DH has no conserved tyrosine and is not dependent on Zn 2þ or other metal cofactors. These features indicate that the structure/function relationships of PSLDRs may differ from those of the other LDRs [31,61].

Metal requirements
Among the stereoselective ADHs, MDRs are a class of enzymes requiring zinc ions for catalytic activity and structural stability [25,29]. From the perspective of structure-function relationships, zinc ions that are firmly bound in the subunit of MDRs generally serve as both catalytic and structural zinc [25]. The conserved active sites for binding catalytic zinc are Cys, His, and Asp located in the catalytic domain, the binding to which involves one sulfur atom, one nitrogen atom, and one oxygen atom, respectively, whereas four cysteines containing four sulfur atoms are required for binding structural zinc [25].
The structures of CBADH provide insights into the molecular interactions between zinc and the surrounding active sites, particularly the catalytic zinc. The conserved residues Cys37, His59, and Asp150 in CBADH interact with the catalytic zinc in the active center. As an exception to the typical MDRs, there is no structural zinc in CBADH, whereas instead, the structural element implicated in stabilizing the enzyme is formed by an ion-pair between Asp89 and Lys111 and a hydrogen bond between Glu94 and Asn104 [29]. However, most MDRs generally have additional structural zinc bound to four Cys residues, which maintains the structural stability of the oligomer. In SsADH, however, one Cys residue within the four-Cys motif is replaced by Glu, which functions to enhance the rigidity of the lobe structure by forming a hydrogen-bond network, and thereby contributes to increased enzyme thermostability [29].
Unlike MDRs, the majority of SDRs retain their activities in the absence of a bound metal ion. As an exception, however, it has been reported that LB-RADH requires a rigidly bound magnesium ion for its activity and is completely inactivated by the removal of this ion [55]. Within the tetramer structure of LB-RADH, the two magnesium ions do not directly interact with the catalytic center, but have direct connection with carboxylate groups in the C-terminal region. Therefore, the involved magnesium ions might play a role as structural metal ions that stabilize the tetramer structure.

Organic solvent tolerance
For the reduction of ketones that have low solubility in aqueous buffers, organic solvents are often used in the reaction media, which can impair enzyme catalytic activity. Therefore, ensuring the stability of ADHs under such denaturizing conditions is important for achieving high catalytic function efficiency [67].
In this regard, the enantioselective ADH-'A' has been reported as an extraordinary biocatalyst with exceptionally high tolerance toward organic solvents, particularly acetone (up to 50%, v/v), 2-propanol (up to 80%, v/v), and hexane (up to 99%, v/v) [60]. The 3 D structure of ADH-'A' (PDB ID: 3JV7) illustrates the molecular mechanism underlying the organic solvent resistance of the enzyme, which was reported as the first evidence and insight into enzyme chemostability from the viewpoint of structure-function relationships. The enzyme is a tetramer, each subunit of which contains a single zinc ion. Compared with other ADHs, the surface of the overall structure of ADH-'A' is more hydrophobic and the loop region more compact. Moreover, ADH-'A' possesses a larger number of salt bridges at the interfaces between dimers. These structural features, namely the greater area of hydrophobic surface and higher quantity of molecular interactions such as salt bridges, are assumed to contribute to the enzyme's stability toward organic solvents [68]. TBADH has high tolerance towards organic solvents. In TBADH, there is a crevice from the surface to the active site provides products and substrates access. This opening is lined with the hydrophobic residues Ile49, Leu107, Trp110, Tyr267, Leu294 as well as Cys283 and Met285 from another molecule within the tetrameric assembly. This might explain the enzyme tolerance toward organic solvents [69].
However, some organic solvents are not necessarily detrimental toward ADHs, and indeed can maintain enzyme activity by stabilizing enzyme structure and regenerating cofactors as reducing agents. In asymmetric reduction by TeSADH in pure hexane, the organic solvent made the process more efficient by allowing high concentrations of substrates to be used [70]. It also makes this enzymatic asymmetric reduction accessible to hydrophobic substrates. Therefore, certain ADHs, such as that from the hyperthermophilic archaeon Pyrococcus furiosus (PFADH) [71], Pcal_1311 [72], HvADH2 [73], ChnA and Ebn2 [74], can retain their activity at high concentrations of organic solvent (Supplementary Table 4). Especially in industrial processes, ADHs with organic solvent-tolerance are greater.

Thermostability
An increase in reaction temperature is generally used to accelerate enzyme reaction rates, and thus biocatalysts with greater thermostability would be more useful and practical for chemical synthesis reactions. ADHs from (hyper) thermophilic organisms have recently gained considerable interest due to their high thermal stability and practicality [75]. Some factors contribute to thermal stability and include more hydrophobic contacts in solvent inaccessible areas, increased numbers of salt bridges, more hydrogen bonds, more proline residues, and a higher degree of compactness [69,76]. Some thermostable ADHs have also been reported (Supplementary Table 4), One example of such an enzyme is PFADH derived from the hyperthermophilic archaeon P. furiosus, which catalyzes asymmetric reductions of aryl ketones and keto esters with high resistance to thermal inactivation [71,77].

Concluding remarks
ADHs offer considerable promise for further process improvements owing to their chemo-, regio-, and stereoselectivity and have received much attention with respect to their roles in catalyzing asymmetric reactions for API synthesis. Although the general principles of oxidoreduction reactions catalyzed by ADHs have been clarified to a certain extent, there remain a number of gaps in our understanding regarding the mechanisms underlying stereoselective recognition and the stereochemical interactions of ADHs with chiral building blocks.
Biochemical approaches rely on enzyme activities, yet despite considerable efforts devoted to developing biocatalytic systems and processes for industrial practice, the application range of stereoselective ADHs remains relatively modest. This deficiency can mainly be attributed to the perceived limitations of available enzymes with desired features relating to stereoselectivity, substrate spectrum, cofactor dependence, activity, and stability. Therefore, there remains imperative to enhance the functional applicability of enzymes. A structure-based comprehension of the function of these enzymes would represent a useful research foundation from which to explore further strategies designed to tailor enzymes as industrially advantageous biocatalysts. With recent progress in biochemistry, structural biology and computational chemistry, these will contribute to gain a more comprehensive understanding of the catalytic mechanisms and prediction of corresponding molecular mechanisms. This will provide the basis to generate novel ADHs suitable for practical application. Therefore, integrating the various aspects of structural biology, biochemistry, and protein engineering, will contribute to advancing progress in the development of stereospecific ADHs, which in turn will expand the range of potential applications in the future.