Enzymes and pathways in microbial production of 2,3-butanediol and 3-acetoin isomers

Abstract 2,3-Butanediol (BD) and acetoin (AC) are products of the non-oxidative metabolism of microorganisms, presenting industrial importance due to their wide range of applications and high market value. Their optical isomers have particular applications, justifying the efforts on the selective bioproduction. Each microorganism produces different isomer mixtures, as a consequence of having different butanediol dehydrogenase (BDH) enzymes. However, the whole scene of the isomer bioproduction, considering the several enzymes and conditions, has not been completely elucidated. Here we show the BDH classification as R, S or meso by bioinformatics analysis uncovering the details of the isomers production. The BDH was compared to diacetyl reductases (DAR) and the new enoyl reductases (ER). We observed that R-BDH is the most singular BDH, while meso and S-BDHs are similar and may be better distinguished through their stereo-selective triad. DAR and ER showed distinct stereo-triads from those described for BDHs, agreeing with kinetic data from the literature and our phylogenetic analysis. The ER family probably has meso-BDH like activity as already demonstrated for a single sequence from this group. These results are of great relevance, as they organize BD producing enzymes, to our known, never shown before in the literature. This review also brings attention to nontraditional enzymes/pathways that can be involved with BD/AC synthesis, as well as oxygen conditions that may lead to the differential production of their isomers. Together, this information can provide helpful orientation for future studies in the field of BD/AC biological production, thus contributing to achieve their production on an industrial scale.

Each one of the BD isomers has a specific industrial application. The 2S,3S-BD is a building block for the asymmetric synthesis of valuable chiral compounds [6] and it can also be used in the production of printing inks and solvents [7]. On the other hand, the 2R,3S-BD is an antiseptic and humectant for cosmetics [8], besides being the precursor of 2-butanol, applied, for example, in the biofuel and food industries [9,10]. The 2R,3R-BD has multiple applications in agriculture [8], and is used in the asymmetric synthesis of valuable chiral compounds, such as chloro-boronic esters [11].
This levo isomer also shows one of the most remarkable BD features: its anti-freezing property [9], since it presents a freezing point at À60 C [3]. This property was studied and explored even before the 1950s [12]. The variety of BD stereo-specific utilization highlights this molecule's commercial relevance.
The BD cost per ton is estimated to be around 10-50 thousand dollars and BD key derivatives global market reaches approximately 32 million tons and 43 billion dollars per year [13]. Currently, the main feedstocks used for BD production at large scale are butenes from catalytic cracking gases. In brief, a C4 hydrocarbon fraction is obtained after the removal of butadiene and transformed into a mixture of oxides by chlorohydrination and subsequent cyclization. So, BD is produced by the hydrolysis of this mixture and can be separated by vacuum fractionation. The chemical production of BD is expensive and also presents weaknesses regarding safety, global warming and sustainability [14]. Some Chinese companies, such as: Global Biochem, Novepha, and Zhangjiagang Glory Biomaterial can fabricate BD microbiologically from a corn-based feedstock. Acetoin and butanediol isomers. The isomers of acetoin (A) and butanediol (B) with attention to their stereo-centers (in red). The butanediol isomers usually assume the gauched more stable conformation (C), the meso isomer (left) shows closer CH 3 groups while the levo and dextro (right) show CH 3 groups in anti-conformation. The molecular structures were drawn in MolView software (http://molview.org/). Adapted from [5].
However with the increased demand for BD more efficient processes are needed [13].
The microbial production of BD and AC is usually related to the specific natural metabolic outcome of every species that can be enriched in one of the isomers. For example, pathways targeting 2R,3S-BD are better studied in Bacillus sp. [15] and Klebsiella pneumoniae [16]. The Bacillus species includes Bacillus subtilis [17] and Bacillus licheniformis, an extremophile organism [15]. The K. pneumoniae is considered to be the greatest 2R,3S-BD producer due the elevated productivity [16], which even led to studies in order to modify this bacteria into a nonpathogenic strain [18]. The 2R,3R-BD production is largely studied in Paenibacillus polymyxa [19,20], since this organism achieves an extremely high purity of this isomer. For 2S,3S-BD production, Brevibacterium saccharolyticum is the oldest organism studied [21], and recent work shows that Rhodococcus erythropolis [22] and a Serratia strain [23] can also generate this isomer.
Notwithstanding, the separation of the isomers mixtures can be difficult due to their similar physical-chemical properties, such as the boiling point that ranges between 177-182 C [3,5]. This difficulty is the key factor leading to interest in genetic engineering strategies and improvement of culture conditions to produce isolated isomers of higher purity [16,24]. Therefore, it is essential to gather knowledge encompassing most of the microorganisms, pathways and enzymes involved in the formation of each isomer for successful production planning. The present review aims to draw a general scenario for enzymes and pathways used for the microbial production of BD and AC isomers.

Microbial and enzymatic production of BD and AC isomers
The microbial production of BD and AC ( Figure 2) was first reported in the early 1900s, and even before wartime, pilot scale fermentations have been performed [25,26]. Biotechnological routes were of great importance during World War II, when the demand for 1,3butadiene, one of BD derivatives, increased [3,13,27]. At this time, it was hypothesized that different organisms could possess different butanediol dehydrogenase (BDH) enzymes displaying distinct stereo-specificities [28]. Even after this, it was still not completely understood that the metabolic basis behind the mixtures of BD and AC isomers were observed in fermentations. At this time, only one route of AC production was known and it could just explain 3R-AC, which justifies this maze! In the early 1960s, it was hypothesized the existence of an AC racemase, which would be responsible for the formation of S-AC from R-AC [29].
Almost 30 years after the racemase hypothesis, in the late 1980s, the second pathway for AC formation was discovered. This route started with the spontaneous decarboxylation of acetolactate (CAS number 918-  forming diacetyl (CAS number 431-03-8), followed by its enzymatic reduction and formation of a second AC isomer (3S-AC) [30]. Currently, it is known that this pathway is capable of generating both isomers, which will be discussed later. Besides that, the evidence that supported AC racemase existence has never been found, causing the hypothesis to be disregarded [16,23,31].
Nowadays, it is clear that the ratio of isomers is dependent on the microorganism employed, as a consequence of their different enzymatic repertoire. The literature also shows not only the enzymes/pathways, but also the organisms NADH/NAD þ balance is crucial for the determination of which isomer will be produced. This means that other fermentative routes and products play a role in BD/AC metabolism [24,[32][33][34].
The AC2 route can in theory produce both isomers, and the literature hypothesizes that the R/S isomerism may be related to the stereo specificity of diacetyl reductase (DAR) activity, where S-DAR or R-DAR could generate 3S-AC or 3R-AC, respectively [22]. At this point it is necessary to raise two points: i. "DAR'' is frequently used as a synonym for BDH, despite the fact that DAR enzymatic activity cannot generate BD [22,36]. The KEGG database has three orthology entries for "diacetyl reductase" search term (Supplementary Figure  ii. Despite the hypothesis aforementioned where S-DAR or R-DAR could generate 3S-AC or 3R-AC [22], the literature usually associates the AC2 route only to the production of 3S-AC [19,30], probably because 3S-AC is the most frequent product of this pathway among microorganisms. In agreement with this, it was reported in P. polymyxa ATCC 12321 the presence of a DAR gene with 3S-AC forming-activity which was functionally verified [19,37]. The absence of 2S,3S-BD in this organism suggests that all 3S-AC forming-enzymatic activity is due to the S-DAR, and that a S-BDH should not exist in this organism. Our searches only found literature evidence of the S-DAR activity (directly searching or inside the KEGG and Brenda entries), remaining doubtful if a R-DAR really exists.
The BDH enzymes are known to catalyze the production of BD, but these enzymes can also produce AC Figure 3. Overview of the pathways involved in 2,3-butanediol production. The AC1 route is in orange, AC2R in cyan, and AC2S in blue. BD: 2,3-butanediol, BDH: butanediol dehydrogenase, DAR: diacetyl reductase. "?" indicates the lack of evidence for DAR participation in R-AC generation, although some papers assume it. Enzyme names or acronyms are in italics.
using diacetyl as a substrate [30]. These enzymes are classified as R, S or meso-BDHs, the R-BDH converts diacetyl in 3R-AC and belongs to the medium-chain dehydrogenases/reductases (MDR) family [19,23,35,37]. The S-BDH and meso-BDH are both capable of producing 3S-AC from diacetyl, appearing to have similar stereo specificity, and belong to the short-chain dehydrogenases/reductases (SDR) family [16,23,38]. Curiously, one manuscript reported that different meso-BDHs would present S or R-AC-forming activity, with no impact in the conversion of AC to BD [27].
The BD production is, in its turn, dependent on which AC isomer is available, and also the set of enzymes of each particular microorganism. Besides the BDHs ability to convert diacetyl in AC irreversibly, as explained above, they produce BD from AC ( Figure 3). The R-BDH produces the R chiral center, so it can reversibly convert 3R-AC to 2R,3R-BD; and 3S-AC to 2R,3S-BD [19,23,35,37]. On the other hand, S-BDH and meso-BDH produce the S chiral center, reversibly converting 3S-AC to 2S,3S-BD, and 3R-AC to 2R,3S-BD. These BDHs differ about the AC isomer affinity, as meso-BDH prefers to use 3R-AC while S-BDH shows a preference for 3S-AC, according to [23] and functional information provided by the BRENDA database.

Differences between R, S and meso-BDHs
The R-BDH belongs to the MDR family [23], while S and meso-BDHs belong to the SDR family [38]. Using CDD web tool [39] and PFAM database [40], we observed that R-BDHs presented two conserved domains: alcohol dehydrogenase GroES-like domain (ADH_N, PFAM ID:  PF08240) and Zinc-binding dehydrogenase (ADH_zinc_N, PFAM ID: PF00107) while meso/S-BDHs showed only the SDR conserved domain (adh_short, PFAM ID: PF00106) (Figure 4(A)). This agrees with the analysis of Yu et al. [19], where discrepancies between SDR and MDR BDHs structures were observed. Besides the conserved domains, the authors described different organization of the coenzyme-binding region.
The multiple alignment used sequences for R-, S-, and meso-BDHs with experimental evidence [16,19,38]. The comparison of the P. polymyxa R-BDH (Uniprot ID: E3E9Q2), the B. saccharolyticum S-BDH (Uniprot ID: Q9ZNN8), and the K. pneumoniae meso-BDH (Uniprot ID: Q48436) shows that while S and meso-BDHs are similar (49.6% of global identity), the R-BDH is more different from the other two, presenting only 12% identity when compared to meso-BDH and 12.8% to S-BDH (Supplementary Figure A.2).

Differences between S and meso-BDHs
The most common and studied meso and S-BDHs shows the classical SDR domain (adh_short, PFAM ID: PF00106), as the meso-BDH from K. pneumoniae and the S-BDH from B. saccharolyticum [22] (Figure 4(A)). However, when analyzing the conserved domain of all sequences already named as meso and S-BDH included in this study, we identified that some of them presented another domain, the Enoyl-(acyl carrier protein) reductase (ER) domain (adh_short_C2 -PFAM ID: PF13561). A batch CD-search showed that some of the single conserved domain present in every sequence showed the SDR domain as a specific hit and the ER as nonspecific, while others have the opposite report ( Figure 4(B)). The presence of the ER domain suggests the possibility of a new BDH group and also that the S and meso-BDH domain of some sequences can be different, despite being similar. The catalytic site previously described [38] (Asn-Ser-Tyr-Lys) was identified in all sequences (Supplementary Figure A.3) reinforcing their similarity. Although the literature states in the SDR domain presence, we could not find reports of the ER for meso and S-BDHs. Amongst the ER containing BDHs, only one sequence (mlike_Smar -S. marcescens H30) was known to have confirmed meso-BDH-like activity [41] suggesting the possibility that other sequences with the ER domain were also active.
The meso and S-BDHs show different Km values reported for both AC and BD isomers (Table 1) (BRENDA enzymes database, [16,23]. Using Serratia sp. T241 as an example, it is possible to see that those values suggest that S-BDH prefers 2S,3S-BD to 2R,3S-BD, on the contrary of meso-BDH, pointing out that there are differences between the enzyme kinetics [23]. The other point that should be highlighted is the lower Km values for some of the pairs S-BDH/2S,3S-BD when compared to the meso-BDH/2R,3S-BD. This fact also suggests that the enzyme specificity could influence the amount of the isomers produced on each organism. Unfortunately, to the best of our knowledge, there is no comparable data about the diacetyl Km for any of these two enzymes in BRENDA, in the literature. The meso and S-BDHs catalytic site from the sequences analyzed were considered conserved (Figure 4(A) and Supplementary Figure A.3) but Table 1 shows dissimilar Km values between them, suggesting that amino acid differences could be important for the determination of the stereo selectivity. Previous work [38,42] has discussed the role of a non-catalytic amino acid triad, here named stereo-triad, in the maintenance of activity and stereo selectivity in meso-and S-BDHs. They contrasted kinetic parameters between wild-type meso-BDH and mutants with changes in this triad (Gln140Ile, Asn146Phe and Trp190His), observing if this stereo-triad differentiates meso and S-BDHs due to its participation on the substrate recognition and its binding. In general, sequences with previous annotations of meso-BDHs from Uniprot and KEGG databases contained two of the classical triads described in the literature [38,42], Gln-Asn-Trp or Gln-Asn-Met. On the other hand, those previously annotated as S-BDHs revealed other stereo-triads, the Ile-Phe-Trp and Ile-Tyr-Trp. One mutation reported in the literature showed a stereo selectivity reversion from meso-to S-BDH when the triad first residue changed from Gln to Ile [42], supporting the relevance of this amino acid at this position in the S-BDHs triad. The investigation of the stereo-triad seems relevant to determine the enzyme stereo selectivity. However, there is not a large amount of stereospecific kinetic data evaluated for S and meso-BDHs. Several manuscripts are not aware of the differences between meso and S-BDHs and probably annotates the enzymes based on similarity search.

Enzymes grouping and classification
Selected meso-BDHs, S-BDHs and DARs (Supplementary  Table A.1) were aligned (Supplementary Figure A.3) and a phylogenetic tree was constructed ( Figure 5). The sequence-based grouping revealed the family organization independently of the annotations previously attributed to the sequences. We observed four groups: (1) meso-BDH: the sequences were grouped in a single clade supported by a bootstrap of 100%, including meso-BDHs with experimental evidence such as the enzymes from K. pneumoniae [16] and Bacillus licheniformis [43]; (2) S-BDH: These enzymes were organized in five clades, all with high bootstrap support. The main S-BDH branch included the sequence from R. erythropolis that has experimental evidence [22]. Another branch, with only two sequences, contained the second representative with experimental evidence, B. saccharolyticum [21]; (3) ER: All enzymes were grouped in a clade supported by a bootstrap of 100%, including the enzyme with BDH-like activity from Serratia marcescens H30 [41]; and (4) Unknown group: This group contains one DAR (P. polymyxa) with experimental evidence [19] plus three other sequences with different conserved domain and stereo-triad.
Overall, the literature/databases current annotations of the sequences without experimental evidence included in the tree were partially or completely wrong, based on the grouping presented in Figure 5. As an example we could cite dar_Rter (Raoultella terrigena sequence named "DAR" in Uniprot -Q04520), Sm_Lque (Lonsdalea quercina sequence named "DAR" in Uniprot -A0A1H4FIC3), Sm0_Ctur (Cronobacter turicensis sequence named "meso-butanediol dehydrogenase/ (S,S)-butanediol dehydrogenase/diacetyl reductase" in KEGG -CTU_29850), and Sm0_Eclo (Enterobacter cloacae sequence named "meso-butanediol dehydrogenase/(S,S)-butanediol dehydrogenase/diacetyl reductase" in KEGG -EcWSU1_01150), which were actually grouped as meso-BDH. They showed a very small distance among them, the same conserved domain and stereo-triad, strongly pointing to the annotation mistakes. Besides that, it was possible to observe that the enzymes are usually unspecifically annotated as mesoand S-BDH simultaneously ("Sm" or "Sm0") in the literature/databases, but they clearly grouped and showed stereo-triads of only one of them.
The meso-BDH clade ( Figure 5) is composed mainly of sequences with the short-chain dehydrogenase domain (SDR) and two stereo-triads (Gln-Asn-Trp and Gln-Asn-Met) already described [38,42]. Two of the sequences display the enoyl-reductase domain, but while one of them (m_Lsac) has the Gln-Asn-Met typical meso-BDH stereo-triad, the other (S_Wsol) has the Gln-Asn-Leu triad. The S_Wsol sequence presents a different residue in the last position of the triad, and despite Leu not seeming to be conserved in this position, it can use a hydrophobic interaction for AC stereo-selection, one of the two possible chemical interactions described [42]. The final answer on S_Wsol activity can only be made by experimental validation.
The S-BDHs group contains five clades ( Figure 5) where the major clade contains not only more sequences but the R. erythropolis S-BDH (S_Rery) with experimental activity [22]. The Bifidobacterium actinocoloniiforme (S_Bact) sequence is the closest one to the B. saccharolyticum S-BDH (S_Bsac), with experimental evidence [21], but the S_Bact change in the catalytic site suggests an unknown effect on its activity. The S_Bact has an Asp instead of Asn in the first catalytic tetrad position despite the SDR and stereo-triads are pretty normal. The S-BDH organization in five clades suggests they are more diverse than meso-BDHs and probably have more than one evolutionary origin. The S-BDHs sequences mainly have the SDR domain but also the ER. The stereo-triad was determinant in the group identification and only two variations were found (Ile-Tyr-Trp and Ile-Phe-Trp), both already described [38,42]. All six ER S-BDHs contain one of the two stereotriads and probably have the S-selectivity, but one exception was the SDR S-BDH (Sm_Rcal) that showed the stereo-triad Ile-Leu-Trp (Supplementary Figure A.3), not described yet. This triad has a Leu in the second position, a hydrophobic amino acid like the Phe already known for this position, suggesting despite not being canonical it could present the S-selectivity. The Sm_Rcal clade also contains four of the ER S-BDHs (Sm0_Xaut, Sm0_Relo, Sm0_Cmas, and Sm_Msp), most probably being the more divergent clade inside the S-BDH group. This clade S-stereo selectivity is doubtful and needs careful experimental evaluation.
The ER group was identified as a single clade where all sequences displayed the ER domain, instead of the SDR one, and almost all of them have the stereo-triad Val-Asp-Thr ( Figure 5), two atypical features for the Thr-Phe-Ala (TFA) in yellow; triads without known stereo-activity are shown in white. The active site residues (Asn-Ser-Tyr-Lys) were the same to almost all sequences, except for S_Bact, Sm1_Mext, Sm1_Pphe, Sm1_Tpro, Sm1_Smar, mlike_Smar, Sm1_Yruc and Sm1_Smel that presented Asp-Ser-Tyr-Lys and hSm1_Asp (Ser-Ser-Tyr-Lys). Sequences with experimental evidence are marked with " Ã " and the recently updated entries in Uniprot now named as diacetyl reductase are marked with "þ". The sequence names were composed of two blocks, the literature/database stereo selectivity annotation, and the species name initial letters. The first name block was composed of S (S-BDH) or m (meso-BDH) or Sm (S/meso-BDH) or Sm0 (S/meso-BDH, KEGG ID: K003366) or Sm1 (S/meso-BDH, KEGG ID: K18009). The prefixes "p" (putative), "h" (hypothetical) or "like" were added to respect the original annotation. The species name was shortened to the first genus letter and the first three species letters; the exceptions were organisms without genus determination, where "sp" was used. Only bootstrap values above 70% were shown.
BDHs. We also noticed another atypical characteristic: half of the ER enzymes present Asp-Ser-Tyr-Lys catalytic tetrad instead of Asn-Ser-Tyr-Lys, and one more sequence (hSm1_Asp) has Ser-Ser-Tyr-Lys (Supplementary Figure A.3). These atypical features do not prevent activity as S. marcescens H30 sequence (mlike_Smar) have meso-BDH-like activity [41] and produces 2R,3S-BD satisfactorily, supporting the active tetrad Asp-Ser-Tyr-Lys. This enzyme shows the same stereospecific profile as a meso-BDHs, meaning it produces/uses S-AC, 2S,3S-BD, and 2R,3S-BD but not 2R,3R-BD and 3R-AC. It was surprising, as none of its triad residues (Val-Asp-Thr) could be found between the most conserved and frequent residues of the classical meso-BDH stereo-triads [42]. Notwithstanding, they are physico-chemically different from the traditional meso-BDHs stereo-triad residues. The canonical Gln (polar uncharged), Asn (polar uncharged) and Trp/Met (aromatic ring/hydrophobic) differ from Val (hydrophobic), Asp (negatively charged), and Thr (polar uncharged). Moreover, it is important to underline one experimental catalytic difference between mlike_Smar and meso-BDHs, as only the BDHs can use S-AC in pH 5. The presence of a sequence with meso-BDH-like activity in this clade suggests this short-chain reductase family potentially owns BDH-forming activity. However, as a consequence of the differences listed above (different clades, conserved domain, catalytic tetrad, stereo-triad and catalytic properties) we believe they should not be classified as meso-BDHs and suggest naming this family as Enoyl-Reductase (ER).
The only diacetyl reductase (DAR) with experimental evidence [19] in the tree (dar_Ppol from P. polymyxa) showed the SDR domain and a unique stereo-triad among the sequences analyzed (Thr-Phe-Ala). This DAR is grouped with three other sequences in a clade in Figure 5 marked with a question mark because it is very doubtful if the enzymes in this clade can be considered putative DARs. They display different conserved domains and non-catalytic triads, besides that the dar_Ppol sequence shows less than 30% of identity with the other three enzymes. The grouping of these four sequences may only mean that among all sequences in the tree, they are more closely related. Consequently, the tree topology clearly shows that the dar_Ppol does not group with the BDHs or with ERs, suggesting that DAR is really another enzyme family.
We noticed that eleven of the sequences from the meso-BDH and S-BDH group clade were recently updated in the Uniprot database and are now annotated as DAR. There are four sequences from the meso-BDH clade (m_Kter, m_Lsak, S_Wsol, and Sm_Lque) and seven from the S-BDH group (Sm_Krob, S_Bact, S_Mwis, pS_Bcoa, S_Phau, Sm_Tsac, and Sm_Cmyx). We doublechecked the KEGG orthology group to which they belong. It was observed that all eleven BDH sequences fit the K03366 group, consistent with traditional meso and S-BDHs, on the other hand, P. polymyxa DAR, the only DAR with experimental evidence, belongs to the K00059 group. This information is also in agreement with the phylogenetic tree, where these enzymes did not group with P. polymyxa DAR but with BDHs.
These results underline that there are differences between S and meso-BDHs sequences, both globally and at the stereo-triad, supporting the discrepancy in the substrate preference and kinetic parameter value. It is the first time that a DAR stereo-triad is addressed in the literature, as well as its comparison with BDHs. This result reinforces the differences between DAR, BDHs and ER enzymes and the need for more research around its kinetic characteristics and the presence in microorganisms of biotechnological interest for the AC/ BD production.

Unusual enzymes/pathways involved in AC/BD isomer determination
In spite of the classical AC/BD pathways and enzymes, there are still unusual contributions for the production of these compounds. The alcohol dehydrogenase activity participation on BD synthesis was identified in the specific circumstance when an organism produced only one of the AC isomers, have only one type of BDH but was capable of producing two BD isomers [24]. For example, in B. licheniformis the production of 2R,3S-BD and 2R,3R-BD was observed, even though only a meso-BDH was reported and no R-BDH was functionally or genomically detected. The production of the 2R,3R-BD in this organism was explained by the presence of the glycerol dehydrogenase (GDH; E.C. 1.1.1.6), that acts as a R-BDH using 3R-AC to produce 2R,3R-BD [15]. Its contribution was noticed not only in B. licheniformis but also in other organisms as K. pneumoniae [16] and a Serratia strain [23]. The GDH that acts as a R-BDH seems capable of converting R-AC in 2R,3R-BD and S-AC in 2R,3S-BD, as well to catalyze the reverse reactions, but not to convert S-AC in 2S,3S-BD, neither the inverse reaction. Based on multiple sequence alignment, it was observed that GDH enzymes showing BDH activity apparently display considerable identity among different species [44]. Similar results were observed in 2016 [15], suggesting homology between R-BDHs and GDHs with R-BDH activity.
Another pathway contributing to the BD production is known as the "BD cycle" and involves the formation of acetylacetoin (CAS number 7338-73-0), followed by its conversion into acetylbutanediol (CAS number 993-71-5) and BD ( Figure 6) [34,45]. The BD isomerism related to the BD cycle is not yet known, despite acetylbutanediol has stereoisomers. This lack of information may be related to the absence of the enzymes involved in the BD cycle in databases, as we could not find acetylacetoin reductase (AAR) neither acetylacetoin synthase (AAS) in KEGG, Metacyc, NCBI and BRENDA databases, which also means that the BD cycle pathway does not appear in metabolic pathway databases (as KEGG and Metacyc), as well that these enzymes do not have an Enzyme Commission (EC) number.
Most papers do not even show the BD cycle pathway, as a study of Paenibacillus brasilensis metabolism [46], although it seems to be of great importance in Bacillus species, as it was already studied in B. subtilis [47], B. licheniformis [15], B. pumilus [48], B. cereus [34] and B. stearotermophilus [49]. Besides that, a study from 2002 [50] used a cell-free extract of various bacteria to determine the production of acetylacetoin and acetylbutanediol and attributed the formation of these products to AAS and AAR, respectively. The compounds could be detected in most of the organisms, suggesting that there are enzymatic activities involved in their synthesis, despite the absence of AAS and AAR enzymes and BD cycle in databases.
The literature also presents other reductases apart from the traditional acetoin-forming DAR enzyme (discussed above on topic 4). Gao utilized a carbonyl reductase (E.C. 1.1.1.184) for AC production [51] while Park used an enzyme referred as "short-chain-acyl dehydrogenase" (this activity does not have EC) showing S-BDH-like activity [52]. We observed this sequence has the ER domain and the stereo-triad Leu-Met-Asn (data not shown), not described for the BDH group in Figure 5. This highlights the relevance of the multiple features for the stereo-activity determination.
There is also an artificial synthetic pathway involving the computationally designed enzyme formolase (E.C. not found) [53], a carboligase which converts acetaldehyde into AC. The initial formolase purpose was to provide a more efficient carbon utilization than the natural pathways. This enzyme shows clear action on the conversion of formaldehyde to dihydroxyacetone, contributing to formate assimilation and supporting further improvement in C1 metabolism engineering to use C1 substrates in the green chemistry [54]. However, more recently it was demonstrated that formolase was also able to produce racemic AC from acetaldehyde through whole-cell biocatalyst. One application of this pathway is the production of BD from bioethanol, to facilitate butanone and 2-butanol production in vitro [55]. Although formolase apparently is not stereospecific, since it generates both 3R-and 3S-AC, it is a BD-related pathway worthy to explore.

Impact of oxygen supply on isomer production
Considering the AC production from acetolactate by acetolactate decarboxylase (AC1 route), BD production consumes 1 mol of NADH per 6 carbons entering as substrate, as the BDH reaction (AC to BD) is the only one involving NADH. This pathway is capable of generating two BD isomers: 2R,3R-BD or 2R,3S-BD (BD1 route, Figure 7). On the other hand, when AC is produced from diacetyl (AC2 route) the BD pathway consumes 2 mols of NADH per 6 carbons. The first one is used at the BD formation from AC and the second at the diacetyl conversion to AC. The BD2 route can generate all BD isomers (2R,3R-BD, 2S,3S-BD and 2R,3S-BD (Figure 7) considering all enzymes that exist on an organism.
The NADH production occurs mainly by the pathways that oxidize the carbon source used by the microorganism. The glucose can be considered the universal carbon source as almost all organisms show the ability to perform glycolysis and produce 2 mols of NADH per 6 carbons used as substrate [56]. On the other hand, glycerol assimilation produces 4 mols of NADH per 6 carbons and is frequently cited as an interesting carbon source because of its abundance and low cost [57].
Considering glucose oxidation and BD1 route fermentation, only 1 NAD þ would be restituted, and other simultaneous fermentative pathways or respiration may be needed to balance the NAD þ /NADH concentrations. Notwithstanding, the BD2 route restitutes all the 2 NAD þ used in glycolysis. Considering the glycerol metabolism, none of the BD pathways is able to balance the NAD þ /NADH concentrations, leading to a NADH excess of 2 or 3 mols per 6 carbons used depending on the BD route. The NADH excess remains to be used by the presence of other fermentative and oxidative pathways [24] that can still compete for this cofactor. This means various other fermentations can occur simultaneously to the BD pathway, for example ethanol (2-4 NADH consumed), lactic acid (1 NADH consumed; and 1,3-propanediol (2 NADH consumed per 6 carbons used as substrate (data from Metacyc database). Besides fermentations, some microorganisms can execute the oxidative phosphorylation pathway (respiration), which can restitute a great NAD þ quantity [58,59].
Usually, the oxygen presence allows the NADH use on respiration, generating more energy than if it was used on fermentation. In this situation, the NADH will probably be preferentially consumed by oxidative phosphorylation. Thus, we could say that the BD production can be controlled by the oxygen availability [60] that is a Figure 7. NADH balance involving pathways with relevance to AC/BD production. The numbers in the left indicate the NADH molecules that can be produced or consumed (related to pyruvate and normalized to 6 carbons). Enzymes are in italic and carbon-assimilation routes are circled. Orange indicates the AC1 BD production route, blue the AC2. AC: acetoin; DA: diacetyl; BD: 2,3-butanediol; PD: 1,3-propanediol; ADC: acetoin decarboxylase; DAR: diacetyl reductase; BDH: butanediol dehydrogenase; LDH: lactate dehydrogenase; ADH: alcohol dehydrogenase; PFL: pyruvate formate-lyase; AcDH: acetaldehyde dehydrogenase; GDT: glycerol dehydratase; PDH: 1,3-propanediol dehydrogenase. Intermediate molecules were hidden, but all enzymes involved in each fermentative route were shown. The molecular structures were drawn in MolView software (http://molview.org/) and the pathways were adapted from the Metacyc database.
consequence of the aeration and agitation rates applied in the BD fermentative process. It is relevant to remember that most BD pathways and carbon sources combinations lead to a theoretical NADH excess (except glucose and AC2 route) stressing that NADH consumption in respiration driven by the oxygen is essential to balance NADþ/ NADH and allow BD production. Consequently, the oxygen supply and AC/BD pathways relation is critical and extensively discussed in the literature [32,59,60].
In general, high oxygen availability leads to a respiration based NADH consumption avoiding the AC to BD reaction and trapping the carbons as AC [61]. Even the AC2 route can be blocked by the same reason causing the reduction or depletion of the AC and BD production in high aeration [58]. On the other hand, all AC and BD routes demand oxygen at some extent to consume the NADH excess. The AC routes are favored by different low oxygen conditions, while the AC1 would happen mostly in very low oxygen supply, the AC2 pathway would need only low. The AC2 spontaneous conversion of acetolactate to diacetyl depends on oxygen as electron acceptor [62], despite the extra NADH (compared to AC1) that should be used on respiration. One problem of the AC2 route needing higher oxygen supply than AC1 is that the enzyme acetolactate synthase is inactivated in this condition [27]. The enzymes expression is also controlled by the oxygen availability. There is evidence on an up-regulation of acetolactate synthase and acetolactate decarboxylase in very low oxygen conditions [27], meaning an increase of the AC1 and the BD1 routes. In agreement with this, it has been already shown that the production of 2R,3R-BD isomer exclusively, or at least with greater purity, can be reached through very low oxygen supply [32,61,63]. The Table 2 summarizes the data around conditions and genetic modifications leading to isomer production improvement.
Together, this information provides some clues about the isomers produced in each oxygen condition. First, it seems clear that a very low oxygen supply favors R-AC and 2R,3R-BD production, while a low oxygen supply inhibits the AC1 route and increases AC2. Depending on the enzymes present on the organism the AC2 routes and following reactions that can generate any of the AC and BD isomers. The oxygen abundance condition has as the most probable end-product the diacetyl, also suggested by the overexpression of a NADH oxidase (mimetizing the respiration high NADH consumption) in Lactobacillus lactis [36].

Conclusions
The BD and AC are chemicals of high industrial interest, with particular applications for each isomer. There is not a unique biological production pathway for all the AC and BD isomers and the pathway for each isomer may also differ among organisms. The presence of the BDH enzyme is quite common, the R-BDH is a mediumchain dehydrogenase/reductase that is very dissimilar to the meso and S-BDH that are short-chain dehydrogenases/reductases (SDR). We have identified remarkable sequence differences between these two SDRs, as well among them and diacetyl reductase (DAR) and observed that, despite possessing the same conserved domain, they organized separately with strong bootstrap support in the constructed phylogenetic tree. The enoyl-reductases (ER) were revealed as related sequences that have reported BDH-like activity. The meso, S-BDHs and ER also have different stereo-triads (Gln-Asn-Trp and Gln-Asn-Met for meso-BDH; Ile-Tyr-Trp and Ile-Phe-Trp for S-BDHs; and Val-Asp-Thr for ER) that are essential for stereospecific substrate recognition, supporting the BDH biochemical reports of a different Km to a same substrate. We highlight a huge misannotation of the SDRs in the databases and literature, even identifying the different ER named, as they were BDHs. We could also show that DAR exists as an enzyme apart and that more studies about its kinetic parameters are needed, as well as its expression by microorganisms. This review either brought attention to non-canonical routes and enzymes, pointing out that there are more enzymes involved in the AC/BD production than the traditional acetolactate reductase and BDH. When it comes to the impact of the oxygen supply on AC/BD pathways, the literature indicates that 2R,3R-BD is the major product in micro-aeration conditions, while diacetyl is the most probable main product under high oxygen availability. This information can guide future works focusing on specific BD isomer production and bioprocess optimization that, in turn, may play a crucial influence in the industrial production of these compounds.

Disclosure statement
No potential conflict of interest was reported by the author(s).