Nucleotide and amino acid sequences of putative AMDases were acquired from the NCBI¹ and IMG database (Integrated Microbial Genomes)². Sequences were compared to others deposited in the NCBI database using BLAST alignment tools³ (Altschul et al., 1990). Structural information on the enzymes was retrieved from the RCSB-PDB database ⁴.

Sequence data were processed using BioEdit⁵, Clone Manager Suite 9 (SciCentral software, licensed), Staden Package containing Pregap⁶ and Gap4 and SeaView⁷ 4.5.1/4.3.4 (Gouy et al., 2010). Maximum-likelihood phylogenetic trees based on amino acid sequences were constructed using SeaView⁷ 4.5.1 and 4.3.4 (Gouy et al., 2010)].

The representative amino acid sequence of the B. bronchiseptica KU1201 AMDase (IMG Locus Tag Ga0077226_10648), which included all known AMDase-specific motifs (Okrasa et al., 2008), was used for a protein blast search against the IMG and NCBI non-redundant database. Resulting hits with an e-value ≤ 1e-44 and an identity score ≥ 43% were analyzed and evaluated manually. In total, 63 candidate AMDases were identified, including 58 new enzymes.

Amino acid sequences of AMDases (Supplementary Table S1) were aligned using T-coffee 8.99 in sensitive mode. Subsequently, hmmbuild of the HMMER package 3.1b2 was used to construct a profile HMM (Hidden Markov Model) from the multiple alignment file. The HMM logo was visualized using the skylign webtool ⁸. The new domain will be retrievable from the Pfam database collection⁹.

For strains and plasmids used in this study see Tables 1–3. Cloning of amdA (IMG locus tag Ga0099533_11097; 708 bp) and amdV (IMG locus tag Ga0099534_11097; 711 bp) into the expression vector pET21a (Novagen/Merck, Darmstadt, Germany) was accomplished after amplification from genomic DNA using gene specific primer pairs with underlined restriction sites (amdA_for: 5′-AGGGGATCCATGACGAAACCCCATCT-GGG-3′/amdA_rev: 5′-ATGAAGCTTGGCGAAAAGACGGCCGTGGC-3′ and amdV_NdeI_for: 5′-CATATGGTGACACAGCAGCCCCATC-3′/amdV_XhoI_rev: 5′-CTCGAGGGCGA-ACAGCAGTCCC-3′). The gene sequence of amdP (744 bp) was obtained from IMG (IMG locus tag SL003B_3169) and synthesized with codon usage adapted to E. coli (MWG Eurofins, Germany; sequence deposited in Supplementary Information). The constructs were verified by sequencing (MWG Eurofins, Germany) and were subsequently transformed into E. coli BL21. Cultures were grown aerobically in LB medium containing 100 μg/ml ampicillin at 37°C until they reached an OD₆₀₀ of 0.5. After induction with 0.1 mM IPTG, the corresponding proteins were expressed at 22°C for 20 h harboring a C-terminal sixfold histidine tag. The enzymes AmdA, AmdP, and AmdV were purified from crude cell extract with nickel-ion affinity chromatography using Ni-NTA agarose (Qiagen, Hilden, Germany) and checked for the correct sizes by SDS-PAGE. The enzymes were dialyzed over night at 4°C against 10 mM EDTA (pH 7.5) in a SERVAPOR tube (SERVA, Heidelberg, Germany; Ø 16 mm, MWCO 12000-14000).

Soil samples originated from the Botanical Garden in Duisburg, Germany, (latitude: N51°29′15.8″, longitude: E6°46′3.5″) and were incubated aerobically at 30°C in 100 mL flasks for 3 days with a selective enrichment medium [1 L contained 5.0 g NaNO₃ × 3 H₂O; 1.01 g K₂HPO₄ × 3 H₂O; 0.25 g MgSO₄ × 7 H₂O; 0.015 g CaCl₂ × 7 H₂O; 0.14 g FeSO₄ × 7 H₂O; 0.02 g EDTA. All components were prepared as 100-fold concentrated solutions and autoclaved separately. The medium was supplemented with 10 mL of a trace element solution [100-fold containing 15 mg Co(NO₃)₂ × 6 H₂O; 61.8 mg H₃BO₄; 198 mg MnCl₂ × 4 H₂O; 57.5 mg ZnSO₄ × 7 H₂O; 47.5 mg NiCl₂ × H₂O; 2.5 mg CuSO₄ × 5 H₂O; 2.4 mg Na₂MoO₄; 4 mg CaCl₂ × 6 H₂O) and 10 mM phenylmalonic acid (Sigma–Aldrich)]. For streak dilution, agar plates were prepared with the same medium and incubated at the same temperature to get single colonies, which were then transferred onto indicator medium agar plates [500 mL contained 0.5 g KH₂PO₄; 0.5 g yeast extract; 1 g NaCl; 1 g (NH₄)SO₄; pH was set to pH 6.0 with KOH/HCl. After autoclaving and cooling to 60°C, sterile filtered 0.025 g kresol red (solved in 70% ethanol) and 1.8 g phenylmalonic acid (solved in H₂O, pH set to 6 with KOH) was added]. The low amount of yeast extract enabled a fast growth of cell mass. When yeast extract was depleted, phenylmalonic acid was the sole carbon source and its decarboxylation caused a pH increase. Colonies which showed a color shift after 2–3 days provided initial indication of using phenylmalonic acid as a carbon source and were further investigated.

For general activity tests, the three different AMDases were assayed using purified recombinant protein. Unless otherwise indicated, the enzymes were added to a substrate solution [10 mM phenylmalonic acid (Sigma–Aldrich) dissolved in 50 mM sodium phosphate buffer, pH 7.0]. The reaction was stopped by the addition of 2 M HCl and by putting the reaction tubes on ice. The decrease of substrate concentration was measured by isocratic, reverse-phase HPLC (LaChrom Elite^®, VWR International GmbH, Germany) with RP-18 columns (LiChroCART 125-3, Puropsher RP-18E und LiChroCART 125-4, Purospher RP-18, VWR International GmbH, Germany). The elution solvent was composed of 80% sodium phosphate buffer (50 mM, pH 7) and 20% methanol with a flow rate of 0.5 ml/min. Analysis of measurements was performed by EZChrom Elite (version 3.2.1, licensed, VWR International GmbH, Germany). Retention times of phenylmalonic acid and phenylacetic acid were 1.5–2.5 min resp. 4.5–6.5 min, dependent on column diameter at a wavelength of 214 nm.

All samples were measured in triplicate with three biological replicates. For determination of the optimal temperature, samples were incubated between 4 and 50°C for 45 min. Temperature stabilities were assayed at 22 and 4°C for up to 30 h and at -20°C for up to 70 days. The impact of pH conditions on enzyme activity was measured using phenylmalonic acid substrate solutions prepared in 50 mM citric acid buffer (pH 4.0, 5.0 and 6.0), 50 mM sodium phosphate buffer (pH 6.0 and 7.0) or 50 mM potassium phosphate buffer (pH 6.0, 6.5, and 7.0) and 50 mM Tris buffer (pH 7.0, 8.0, and 9.0). The pH dependent enzyme activity of AmdA was measured using 20 mM phenyl malonic acid substrate solution solved in 100 mM citric acid buffer (pH 4.3, 5.2, 5.4, and 5.8), 200 mM potassium phosphate buffer (pH 5.8, 6.5, and 7.3) and 30 mM Tris buffer (pH 7.0 and 8.0).

The influence of salts and possible cofactors on AmdP, AmdV, and AmdA was assayed with the following compounds in 1 mM final concentration: sodium phosphate, pyridoxal phosphate, NAD, ATP, thiamine diphosphate, zinc chloride, sodium chloride, and magnesium chloride. The reaction using 10 mM phenylmalonic acid was stopped with 2 M HCl after 45 min.

To determine the solvent stability of the AMDases, ethanol, methanol, isopropanol and DMSO were added to the reaction mixture (10 mM phenylmalonic acid in 50 mM Tris-buffer, pH 7.0) to a final volume of 10% (v/v). The reaction was stopped after 45 min and the activity compared to solvent-free controls.

The selectivity of the three different AMDases in the synthesis of optically pure 2-arylpropionic acids was analyzed by using purified recombinant protein. The enzymes were added to a substrate solution [10 mM 1a or 1b dissolved in 50 mM tris buffer, pH 7.5]. 1a or 1b (see Figure 1A) were prepared as described in (Gaßmeyer et al., 2016). The reaction was stopped by the addition of 2 M HCl and 2-arylpropionic acids were extracted twice with methyl tert-butyl ether (500 μL). Methanol (100 μL) and trimethylsilyl-diazomethane (25 μL) were added and the mixture incubated at room temperature for 30 min. Acetic acid (5 μL) was added, the solvents evaporated to dryness and the residue was re-dissolved in ethyl acetate (200 μL). Conversion and optical purity were then analyzed by chiral gas chromatography on a Shimadzu GC Plus 2010 device (Gaßmeyer et al., 2016); using a FS-Hydrodex-β-6TBDM column (Macherey-Nagel), with a method operating isothermic at 160°C (for naproxen) and 170°C (for flurbiprofen) with an injection split of 1/20. The elution order was for both 2-arylpropionic acids (S) and (R) with baseline separation and a peak difference of 0.5 min. ee-values were determined from the relative peak areas according to standard procedures. ee stands for enantiomeric excess and represents the excess of one enantiomer in a racemic mixture. Errors of ee-value determination were extrapolated from the area difference obtained from the respective racemic authentic standard. From this, the precision of ee-values was found to be ±0.3%.

For the determination of AMDase-specific searching criteria an amino acid sequence alignment of known AMDases was established initially (Figure 1B). We then verified, that the catalytically crucial amino acids described in the postulated reaction mechanism (Okrasa et al., 2008; Lind and Himo, 2014) are conserved within all representatives. For the alignment only enzymes were chosen which (i) have a proven AMDase activity of the order of Bordetella’s prototype and (ii) are able to decarboxylate phenylmalonic acid and more complex prochiral substrates. Five enzymes met these requirements (Figure 1B): Four of them are the characterized AMDases mentioned above and additionally one enzyme from Variovorax sp. HH01. The strain is an isolate of an enrichment culture derived from soil with phenylmalonic acid as sole carbon and energy source (see Material and Methods) and the respective enzyme was characterized in the course of this work. Based on the published AMDases and our Variovorax enzymes, we were able to establish a total of 12 conserved motifs that could be used to identify AMDase sequence homologs (Table 4). The conserved amino acids described in Okrasa et al. (2008) and listed in Table 4 include all AMDase-specific protein structural motifs known today, like the hydrophobic pocket, the dioxyanion hole and the aromatic binding pocket and additionally two sequence motifs surrounding the residues, which determine the product’s stereoisomery. These conserved sequence motifs were used to screen data bases for further enzymes with AMDase activity. In these analyses only proteins, which met all 12 criteria with identical amino acids or amino acid residues with similar physicochemical properties were chosen. With the established search algorithm, 58 further ORFs which most likely encoded enzymes with AMDase activity were identified from the IMG and NCBI databases. The amino acid sequence alignment of these enzymes was used to create a Hidden Markov Model (HMM) shown in Figure 1C and enabled the deduction of six conserved sequence motifs (Table 5) which seemed specific for the prototype AMDase from B. bronchiseptica and include all known catalytically active residues. The HMM was implemented into the search algorithm and used for a reliable prediction of AMDase-active enzymes. AMDase are members of the protein superfamily aspartate/glutamate racemases, including aspartate racemases (EC 5.1.1.13), glutamate racemases (EC 5.1.1.3), hydantoin racemases (EC 5.1.99.5) and maleate cis-trans isomerases (5.2.1.1). AMDases of B. bronchiseptica’s prototype with their conserved sequence patterns are clearly defined within the superfamily, as can be seen in Figure 2. Characteristic catalytic elements are the dioxyanion hole, the hydrophobic pocket, the large binding pocket for aromatic residues and a thiolate hole for the accommodation of the deprotonated Cys residue. The search algorithm and deduced sequence patterns allow a well-defined distinction between “true” AMDases and other members of the aspartate/glutamate superfamily for the very first time.

To date, 63 ORFs encoding putative AMDases were identified (status January 2016). All of the 25 amd-carrying bacterial genera belong exclusively to the class of Alpha-, Beta- and Gamma-proteobacteria (Figure 3). Alpha-proteobacteria include 25 amd genes, distributed over 16 genera, Beta-proteobacteria contain 34 AMDase genes distributed over six genera and within Gamma-proteobacteria four ORFs encoding AMDases were found, spread over three genera (Figures 4A,B). In strains with complete genome sequences, AMDase genes are encoded on the chromosomes. A singularity represents the strain Nitratireductor pacificus pht-3B which is the only isolate that encodes two different amd genes. N. pacificus pht-3B is part of a pyrene-degrading consortium of enriched sediment from the Pacific Ocean (Nie et al., 2012).

It is intriguing that for most of the genera in which AMDases were identified; only one or two different strains with the amd genes could be detected. However, within the genus Variovorax 23 strains harbored an amd gene and within the genus Alcaligenes seven strains that carried amd genes were detected (Figure 5B). This observation may indicate that the amd gene was originally evolved within the genus Variovorax or a closely related group of bacteria. The findings also suggest that the amd gene is not part of an essential core genome of any of these bacteria but rather part of the accessory pan-genome.

Bacterial strains with amd genes derived from six isolation sources (Figure 4C): soil, plants (plant associated or endophytes), wastewater, oceans, hosts, and air. For 16% of the strains no distinct isolation source was mentioned. The majority (32%) is composed of strains that have been isolated from soil. All strains and gene locus tags are listed in Supplementary Table S1 (Supplementary Information).

Comparison of amd flanking sequences enabled a more precise classification and characterization of the AMDases (Figure 5). One of the most widely spread ORFs was a gene encoding a mandelate racemase/muconate lactonizing enzyme (MR/MLE, e.g., IMG gene ID of Variovorax sp. HH01: 2639906160) which was present in 61% of the available sequences. Mandelate racemases (EC 5.1.2.2) and muconate cycloisomerases (EC 5.5.1.1) belong to the enolase superfamily and are involved in aromatic metabolism (Hasson et al., 1998). Moreover, genes commonly occurring were encoding tripartite tricarboxylate transporters (TTT). These sodium ion-dependent transporters translocate various substances, including citrate and its derivatives and are frequently found in Beta-proteobacteria (Antoine et al., 2003). Further genes cumulating in the sequence surrounding amd are ORFs coding for TRAP transporters and ABC transporters (Schneider and Hunke, 1998). TRAP transporters are tripartite ATP-independent periplasmic transporters which are common in prokaryotes (Mulligan et al., 2011). Like TTTs and ABC transporters, they consist of cytoplasmic membrane proteins and a periplasmic solute-binding protein.

Comparison of amd neighboring sequences combined with the amino acid sequence similarities allowed a first classification of all known AMDases into eight enzyme clusters (Figure 5B). Cluster TTT-I to TTT-V comprise isolates with genes encoding a TTT next to the amd genes. With exception of Azoarcus sp. all Beta-proteobacteria belong to the clusters TTT-I to TTT-III. Cluster TTT-IV and TTT-V are composed of Alpha-proteobacteria. By now there is only one cluster with amd genes flanked by ORFs encoding ABC transporters. This cluster includes the marine isolates Cucumibacter marinus DSM 18995 and N. pacificus pht-3B as well as the isolates of Devosia sp. There are two clusters of AMDases with ORFs encoding TRAP transporters nearby. Thereby the cluster TRAP-I is the only enzyme cluster with representatives of all three proteobacterial classes carrying AMDase coding genes.

For further investigations the enzymes AmdA, AmdV, and AmdP were chosen. AmdA and AmdV originated from Variovorax sp. HH01 resp. Variovorax sp. HH02. The Variovorax strains are isolates from one soil sample and showed the highest activity on indicator medium plates. Furthermore, special interest lies in them as no other enzymes from this genus have been characterized so far. AmdP has been found by data base search as a candidate to prove the applicability and reliability of the searching criteria. The corresponding gene originated from Polymorphum gilvum SL003B-26A1, a strain which was originally isolated from a crude oil-polluted saline soil and was able to grow at temperatures up to 50°C (Nie et al., 2012). This makes AmdP a promising candidate for putative industrial applications. The enzymatic properties of the newly identified AmdA, AmdP, and AmdV were assayed using recombinant purified protein. A turnover of 10 mM of the substrate phenylmalonic acid was measured by HPLC. Concerning the temperature optimum, all three enzymes had their optimal activity in a comparable mesophilic range between 30 and 37°C (Figure 6A). AmdV showed maximal activity at 30°C and possessed 44 ± 3% activity at 10°C. Above 30°C, however, activity rapidly decreased with 6 ± 2% remaining at 50°C. The highest activity of AmdA from Variovorax sp. HH01 could be detected at 34°C. At 10°C, the activity decreased to 33% and above 40°C, activity decreased rapidly until almost no enzyme activity could be measure above 50°C. The highest conversion of the substrate by AmdP could be measured at 37°C. Below this temperature, the activity of this enzyme was lower compared to AmdA and AmdV, at 10°C even 2.6-fold lower than the activity of AmdV. Above 37°C, AmdP showed a better activity compared to the other two AMDases with 4.8-fold more substrate conversion at 45°C than AmdV.

The optimal pH conditions of the enzymes were assayed using phenylmalonic acid substrate solutions. All three enzymes are pH-dependent and showed the highest activity at a neutral to slightly acidic pH, mostly with Tris and sodium phosphate buffer (Figure 6B). AmdA showed its highest activity at pH 6.0, and AmdV as well as AmdP at a pH of 7.0 (Tris buffer). Below pH 4.0, no enzymatic activity could be detected, whereas at a pH of 8.0 and 9.0, 85% residual activity could be measured for AmdA and at pH 9.0, more than 65% activity remained in case of AmdV and AmdP. The preference for neutral or slightly basic conditions agrees well with that of the well-characterized AMDase from B. bronchiseptica (Miyamoto and Ohta, 1992a).

The solvent stabilities of the three AMDases were measured in 10% solutions (v/v) of ethanol, methanol, isopropanol, and DMSO diluted in 10 mM Tris buffer pH 7.0. AmdA, AmdV, and AmdP were similarly susceptible to the negative influences of the solvents (Figure 6C). Ethanol and isopropanol decreased the enzyme activities of the AMDases to less than 10% of the controls without solvent. With methanol, the substrate turnover was also strongly reduced to approx. 15% (AmdA 13 ± 5%, AmdV 15 ± 2%, and AmdP 16 ± 2%). DMSO reduced the enzymatic activity of AmdA and AmdV only to approx. half of the control (48 ± 4 and 50 ± 4%, respectively) and AmdP still possessed 82 ± 7% activity in this solvent.

Concerning the influence of salts and possible cofactors, none of the tested effectors sodium phosphate, pyridoxal phosphate, NAD, ATP, thiamine diphosphate, zinc chloride, sodium chloride, and magnesium chloride (all 1 mM in final concentration) showed a beneficial effect on the enzymatic activity of AmdA, AmdV, and AmdP, proving that the enzymes function cofactor-independently (Figure 6D). Magnesium and especially zinc chloride even reduced enzyme activity significantly (MgCl₂: 81 ± 1% residual activity for AmdA and 83 ± 5% for AmdV. ZnCl₂: 74 ± 9% for AmdV, 38 ± 5% for AmdA and 8 ± 4% for AmdP compared to the effector-less control).

The thermal stabilities and optimal storage conditions of the three AMDases were assayed at 22, 4, and -20°C in 50 mM Tris buffer pH 7.0 and 50 mM sodium phosphate buffer pH 7.0, respectively, by measuring the enzymes’ residual activities. In case of all three enzymes, storage in sodium phosphate buffer seemed to be beneficial compared to Tris buffer (Figure 7A). The residual activity of AmdA (Variovorax sp. HH01) after 7 days at 22°C in Tris buffer comprised only 6 ± 4% while in sodium phosphate buffer, 64 ± 10% activity were still measurable after 19 days at the same temperature. The best storage conditions for AmdA were at 4°C in sodium phosphate buffer with 61 ± 12% residual activity after 24 days. AmdP from P. gilvum SL003B-26A1 was stable for more than 7 days with 40 ± 6% residual activity after incubation at 22°C and with 69 ± 7% at 4°C (Figure 7C). Although AmdV (Variovorax sp. HH02) lost most of its activity after storage at 22°C for 7 days (11 ± 5% residual activity), it was highly stable at 4°C and retained 96 ± 5% of its initial activity after 11 days (Figure 7B). As shown in Figure 7D, for a long-term storage of the three enzymes, it is recommendable to choose sodium phosphate buffer (50 mM and pH 7.0) and -20°C as the AMDases keep more than 85% of their initial activity after at least 70 days (AmdA 89 ± 13%; AmdV 85 ± 9%, and AmdP 89 ± 7%).

The selectivity of all three AMDases was analyzed with aryl malonic acid derivates of the arylpropionic acids naproxen and flurbiprofen. All enzymes showed a very high (R)-selectivity for both substrates as the obtained ee-values were greater than 99% indicating a huge excess of the (R)-enantiomer (Figures 1A and 8). In the case of flurbiprofen, spontaneous decarboxylation during the preparation of the aryl malonic starting material limited the ee. In the synthesis of naproxen, all three AMDases showed (R)-preference and a very high enantioselectivity (>99% ee). In the case of flurbiprofen, the malonic acid is of limited stability. During the synthesis of the starting material, a small amount of racemic flurbiprofen (<4%) is formed that cannot be separated from the flurbiprofen malonate. As a consequence of this contamination, the enantiomeric excess of flurbiprofen is somewhat decreased. This effect is known and can be circumvented by the optimization of the reaction conditions. Nevertheless, the selectivity of the three AMDases in the synthesis of flurbiprofen can be assumed to be very high. Terao et al. (2003) showed an ee of only 92% for (R)-flurbiprofen using the AMDase from B. bronchiseptica. This effect is known and can be circumvented by the optimization of the reaction conditions (Gaßmeyer et al., 2016). In the synthesis of naproxen, all three AMDases showed (R)-preference and a very high enantioselectivity (>99% ee).