Cultivation of Rhodococcus opacus 557 and Rhodococcus rhodnii 135 was performed with 4-hydroxybenzoate as sole source of carbon and energy (Jadan et al., 2001). Genomic DNA from R. opacus 557 and R. rhodnii 135 was prepared from cells obtained after centrifugation of 50 mL cultures, which were subsequently washed with 50 mM Tris-HCl, pH 7.6 and treated with phenol-chloroform to extract the DNA (Sambrook and Russel, 2001). Escherichia coli DH5α (GIBCO BRL) and clones obtained were grown while shaking at 37°C in lysogeny broth (LB) medium (Sambrook and Russel, 2001) containing ampicillin (100 μg per mL).

Oligonucleotides were designed and synthesized according to the N-terminal and internal sequences of PHBH_Ro and PHBH_Rr (Montersino and van Berkel, 2013). In addition, primers were designed using the sequences of conserved regions of PHBH_Pf (Weijer et al., 1982), and PHBHs from Acinetobacter sp. ADP1 (DiMarco et al., 1993) and Azotobacter chroococcum (Quinn et al., 2001).

The constructs pROPOB1 and pRRPOB1 were obtained by cloning the 870 bp PCR products of primers fw-Rh557 [GAA (CT)AC CCA (AG)GT (CG)GG CAT (ACT)GT] and rev-pobA [CGGT(GC)G G(GC)G G(GC)A C(AGT)A T(AG)T G] with R. opacus 557 or R. rhodnii 135 DNA into the EcoRV site of pBS T-tailed as described elsewhere [pBluescript II SK(+), Stratagene; (Marchuk et al., 1991)]. Inserts obtained from EcoRV digested plasmid DNA were labeled with digoxigenin by using the DIG DNA Labeling and Detection Kit Nonradioactive (Boehringer, Germany) for the detection of fragments on Southern blots of EcoRI-digested R. opacus 557 or R. rhodnii 135 DNA. Respective DNA-fragments were purified from agarose gels, ligated into EcoRI-digested and dephosphorylated pBS. The resulting plasmid was transformed into E. coli DH5α and obtained colonies checked by colony hybridization as described elsewhere (Eulberg et al., 1997). Positive clones pRoPOB1-1 contained a 9.8 kb EcoRI fragment of R. opacus 557 DNA and pRrPOB1-1 a 7.8 kb EcoRI fragment of R. rhodnii 135 DNA, respectively, comprising the complete pobA genes. Subclones containing less flanking DNA regions were obtained by using various restriction endonucleases as shown in Figure S1.

DNA sequencing and sequence analysis was performed with common primers such as T3, T7, M13, or rM13 and respective software as described previously (Gish and States, 1993; Thiel et al., 2005; Felsenstein, 2009).

Rhodococcus opacus 1CP is a model strain for the degradation of aromatic compounds (Eulberg et al., 1997) and encodes a single PHBH-like protein (accession number ANS30736) which is 99% similar to the other Rhodococcus PHBHs reported herein. The corresponding gene pobA was amplified by PCR and cloned into pET16bp as described earlier Riedel et al., 2015. Using the primers pobA-fw (5′-catatgaacacacaggtcgggatc-3′) and pobA-rev (5′-ggtacctcagcccagcggggtgc-3′) allowed introducing NdeI/NotI restriction sites for cloning. The subsequent cultivation and expression was done as described below for the Cupriavidus enzymes.

Ralstonia eutropha (also designated as Cupriavidus necator) JMP134 harbors a number of enzymes involved in degradation or aromatic compounds and amongst those two PHBH-like proteins (accession numbers KX345395 and KX345396 for PHBH_Cn1 and PHBH_Cn2, respectively; Pérez-Pantoja et al., 2008. The PHBH-encoding genes AOR50758 and AOR50759 were codon optimized according to the codon table of Acinetobacter sp. ADP1, synthetically produced, obtained in a pEX-cloning vector and cloned into pET16pb by methods reported earlier Oelschlägel et al., 2015; Riedel et al., 2015. Cloning was performed using E. coli DH5α and LB medium (10 g tryptone, 5 g yeast extract and 10 g NaCl per L) was used with ampicillin (100 μg per ml). For gene expression, the pET construct was transferred to E. coli BL21 (DE3) pLysS and cultivated in LB medium containing ampicillin (100 μg per ml) and chloramphenicol (34 μg per ml). Fernbach flasks (1 L) were used and the cultures were grown at 37°C until an OD600 of 0.2 and subsequently cooled to 20°C. At an OD600 of about 0.5 the gene expression was induced by adding IPTG (0.5 mM) and the cultures were continued at 20°C for 20 h. Afterwards cells were harvested by centrifugation (1 h at 5,000 x g, 4°C) and the pellets were stored at −20°C.

The cell pellets were resuspended in 50 mM Tris/sulfate buffer (pH 7.5) while adding 8 units DNaseI (AppliChem—BioChemica, Darmstadt). Cells were broken through ultrasonic treatment (15 times 30 s, 70% power using a HD 2070, MS 72, Bandelin Sonoplus) in an ice-bath. Cell debris was removed by centrifugation (20,000 × g for 20 min, 4°C). After filtration through a cellulose membrane (0.2 μm pore size) to remove remaining particles, the crude extracts were subjected to Ni-chelate chromatography using a 1 ml HisTrap FF column (GE Healthcare) mounted in an ÄKTA fast-performance liquid chromatographer (GE Healthcare). The column was pre-equilibrated with 10 mM Tris/sulfate buffer (pH 7.5). After applying the cell extract, the column was washed with 3 ml loading buffer and then with loading buffer containing 25 mM imidazole until no protein eluted anymore (about 6 ml). Next we started a gradient to achieve 500 mM imidazole in the loading buffer within 6 ml. Target protein eluted during this gradient. Fractions were collected in 1 ml size and checked for standard PHBH activity (see section Enzyme Activity Measurements and Product Analysis). Active fractions were pooled and concentrated and buffer exchanged using Amicon Ultra-15 centrifugal filter devices (30 kDa) in 50 mM Tris/sulfate buffer (pH 7.5) containing 45% glycerol. The enzyme samples were stored at −20°C until further use. Protein concentration was determined by means of a Bradford assay.

Enzyme activity measurements were performed at 30°C in 50 mM Tris/sulfate buffer (pH 7.5), containing 60 μM FAD, 175 μM NAD(P)H (or 0 to 175 μM if varied) and 500 μM 4-hydroxybenzoate (or 0–500 μM if varied). Reactions were started by adding 20–40 nM of enzyme solution. All assays were performed in triplicate and either followed by the decrease in absorption at 340 nm (ε₃₄₀ = 6.22 mM⁻¹ cm⁻¹) or by HPLC analysis of 3,4-dihydroxybenzoate. For HPLC analysis, five samples were taken at 1 min intervals and reactions were stopped adding ice-cold methanol. Before analysis, samples were centrifuged at 17,000 x g for 2 min to remove protein precipitates. HPLC (10 μl sample volume) was performed with a C18 reverse phase column (Knauer) running in a Ultimate3000 (ThermoScientific) UHPLC system. Elution was done isocratically with 0.1% trifluoroacetic acid, containing 30% methanol (flow rate 1 ml per min; 6 min total run time). Authentic standards of 4-hydroxybenzoate, NAD(P)H, NAD(P)⁺ and 3,4-dihydroxybenzoate were used to calibrate the system. Absorption was continuously monitored at 215 nm and spectra of eluting compounds were acquired with a diode array detector.

PHBH protein sequence analyses were performed using the NCBI BlastP-service (Altschul et al., 1990). In total, 70 PHBHs of various bacterial phyla were selected and used for in silico analyses. The protein sequences from P. putida KT2440 (NP_746074; salicylate hydroxylase), C. testosterone TA441 (BAA82878; 3-(3-hydroxyphenyl)propionate hydroxylase), S. chlorophenolicum L-1 (AAF15368; pentachlorophenol monooxygenase), and Acinetobacter sp. ADP1 (AAF04312; salicylate hydroxylase) served as appropriate out-group, as reported earlier (Suemori et al., 2001; Pérez-Pantoja et al., 2008).

The sequence information was used for a phylogenetic analysis allowing functional annotation of PHBH genes. Several algorithms (Fitch-Margoliash, maximum parsimony, maximum likelihood, and neighbor joining) were applied to obtain reliable sequence alignments and representative distance trees. The following software tools were used: Clustal-X (ver. 1.8) (Higgins and Sharp, 1988; Thompson et al., 1997), GeneDoc (ver. 2.6.003), the PHYLIP 3.66 package (PROTDIST and FITCH) (Felsenstein, 2005), and MEGA5 (Tamura et al., 2011). Bootstraps of 1,000 replicates were calculated from the corresponding alignment by means of the PHYLIP 3.66 package (SEQBOOT, PROTDIST, FITCH, and CONSENSE) (Felsenstein, 2005).

Sequence logos were constructed as follows: the PHBH_Pf protein sequence was used as input query for a BlastP (NCBI) (Altschul et al., 1990) search using the non-redundant protein sequences database. Only sequences with an E-value smaller than 1e⁻¹⁰⁰ were selected. After filtering the output sequences for duplicates, crystal structure sequences and cloned protein variants using Sequence Dereplicator and Database Curator (SDDC, ver. 2.0) (Ibrahim et al., 2017), the sequences of the protein segment involved in pyridine nucleotide coenzyme binding were selected and aligned using Clustal Omega (Sievers et al., 2011). The top 200 protein segment sequences were used to generate a sequence logo using the WebLogo server (ver. 2.8.2, Crooks et al., 2004). This process was repeated with the PHBH_Ro protein sequence as query input.

The phylogenetic analysis and its outcome is of major relevance for the identification of the pyridine nucleotide coenzyme binding sites. The above described methods were validated by the herein described protein energy profiling, which allows for drawing sequence—structure relations (Heinke et al., 2015).

Obtaining energy profiles from protein structures is realized by means of a coarse-grained residue-level pair potential function. Based on the theoretical assumptions elucidated in Wertz and Scheraga (1978), Eisenberg and McLachlan (1986), and Dressel et al. (2007), this energy model approximates the hydrophobic effect by utilizing buried and exposed preferences for each of the 20 canonical amino acids. Given a set of globular protein structures, one can determine the frequencies for each amino acid of being exposed on the outside or buried inside the protein by using the DSSP program (Kabsch and Sander, 1983) as proposed by Ofran and Rost (2003) or by determining residue orientation and local spatial residue packing (Dressel et al., 2007; Heinke and Labudde, 2012). The energy potential (E_i) is calculated using the following equations: (1)ei=-ln(fbur,ifexp,i), (2)eij=ei+ej, (3)Ei=∑j∈Protein,j≠ig(i,j)[eij].

Given a residue at sequence index i, the single-residue potential e_i is computed using the amino acid-specific buried-exposed frequency ratio (Equation 1). As shown in Equation (2), the pair potential e_ij between two residues at indices i and j corresponds to the sum of single-residue potentials in this model. Finally, by iterating over all residues that are in contact with residue i, the potential E_i is derived (Equation 3). A contact between two residues (i and j) is assumed, if the Cβ - Cβ atom distance is < 8 Å (in case of Gly, Cα atom coordinates are used as spatial reference points instead).

The sequence of residue energy potentials (E₁,…,E_i,…,E_n) corresponds to the protein's energy profile (Dressel et al., 2007; Heinke and Labudde, 2012, 2013; Heinke et al., 2015). In addition, an algorithm for aligning two energy profiles has been adapted from Mrozek et al. (2007) which, besides detecting similarities and differences of residue energy potentials, can also give a distance scoring function (referred to as dScore) as a measure of global energy profile similarity of two energy profiles P₁ and P₂ (Heinke and Labudde, 2013; Heinke et al., 2015): (4)dScore(P1,P2)=-log(xr-x¯PxOpt-x¯P),

where (5)xOpt=δ(|P1|+|P2|)2.

The dScore corresponds to the normalized energy profile alignment raw score x_rwith respect to the average score x_P obtained from random energy profiles and the highest possible dScore x_opt of two profiles with lengths |P₁| and |P₂|. Here, δ acts as an alignment parameter with δ > 0. The negative logarithm leads to a distance-like formulation, with two identical energy profiles yielding a dScore of 0.

Two PHBH structures PDB ID: 1d7l (Ortiz-Maldonado et al., 1999) and PDB ID: 1bgj (Eppink et al., 1998a) were retrieved from the Protein Data Bank (Rose et al., 2011) and used as modeling templates for automated comparative modeling using Modeler (ver. 9.14) (Eswar et al., 2006).

Seventy PHBH sequences (including 15 sequences of biochemically characterized PHBHs and 55 randomly selected PHBH sequences from various bacteria) were used for automated comparative modeling (average sequence identity of ~50%). For each PHBH sequence, five comparative models were generated from which the model with the best corresponding DOPE score (Eswar et al., 2006) was selected for energy profile calculation. In the first step of energy profile analyses, energy profile distance trees were generated. As shown recently (Heinke and Labudde, 2013; Heinke et al., 2015) such distance trees can indicate functional and structural relations and, in case of PHBHs, can support the proposed molecular evolution. To obtain such distance trees, pairwise energy profile alignments were computed as elucidated and, for each energy profile alignment, the corresponding dScore was derived, leading to an energy profile distance matrix. By utilizing the un-weighted pair group method arithmetic mean (Sokal and Michener, 1958) and neighbor joining (Saitou and Nei, 1987) with the derived distance matrix as input, distance trees were generated.

The Rate4Site tool (ver. 2.01) (Mayrose et al., 2004) was used for determining conserved amino acids in PHBH proteins specific for NADPH and NADH, respectively. Multiple sequence alignments were made from selections containing only sequences of pseudomonads and rhodococci, which were used as input to calculate evolutionary rates for all amino acids applying default settings of Rate4site. The obtained values for conservation were scaled to b-factors ranging between 0 and 100. These b-factors were used to color the image of the crystal structure of PHBH_Pf as example of a NADPH-specific protein. In a similar way, the image of the model structure of PHBH_Ro as an example of a NADH-preferring protein, was colored. The program Pymol (ver. 1.4) (Schreudinger, 2011) was used to create structure images.

The three-dimensional structure of the PHBH_Pf monomer with the FAD cofactor in the out conformation (PDB ID: 1pdh) was used to access the mode of NADPH binding. Docking was performed using HADDOCK (ver. 2.0) (de Vries et al., 2010). The solvated docking was carried out with the recommended parameters of HADDOCK. A distance restraint of 9.0 Å was set between C4N of NADPH and C4a of the flavin cofactor. For rigid-body energy minimization, 2,000 structures were generated, and the 200 lowest energy solutions were used for subsequent semi-flexible simulated annealing and water refinement. Resulting structures were sorted according to intermolecular energy and clustered using a 6.5 Å cut-off criterion. Subsequent cluster analysis was performed within a 2.0 Å cut-off criterion. The structure with the lowest score was selected for generating an image showing the NADPH binding mode of PHBH_Pf.

PHBH sequences determined in this study are available from the GenBank/EMBL/DDBJ nucleotide sequence databases under accession numbers KF234626 for R. opacus 557 and KF234627 for R. rhodnii 135.

Most biochemically characterized PHBHs with known amino acid sequence are strictly dependent on NADPH (Table 1). However, PHBH from R. opacus 557 (PHBH_Ro) and PHBH from R. rhodnii 135 (PHBH_Rr) show a clear preference for NADH (Jadan et al., 2001, 2004). This prompted us to determine the amino acid sequences of PHBH_Ro and PHBH_Rr (see Methods). Genomic R. opacus 557 DNA contained a 1,179-bp open reading frame coding for a PHBH polypeptide of 392 amino acids. The amino acid sequence predicted from the open reading frame corresponded with the experimentally determined N-terminal sequence of the protein (MNTQVGIVGGGPAGLM) and with the N-terminal sequence (TDHFRQYPFAWFGILAEAPP) of an internal 25 kDa tryptic fragment. Genomic R. rhodnii 135 DNA contained a 1,191-bp open reading frame coding for a PHBH polypeptide of 396 amino acids.

In this paper, amino acid residues are numbered according to the sequence of PHBH_Pf (CAA48483) to facilitate reference to the 3D-structure. The amino acid sequences of PHBH_Ro (accession number KF234626) and PHBH_Rr (accession number KF234627) both share 46.7% identical positions with PHBH_Pf (Figure 2). Their helix H2 regions, proposed to be involved in determining the pyridine nucleotide coenzyme specificity (Eppink et al., 1999), deviate in amino acid sequence from that of NADPH-specific PHBHs (Figure 2). The latter enzymes typically contain the fingerprint sequence 32-ERxxx(D/E)YVLxR, while the NADH-preferring Rhodococcus PHBHs contain the sequence 32-E(S/C)RTREEVEGT.

His₁₀-tagged forms of two putative PHBHs originating from C. necator JMP134 were successfully produced by recombinant expression in E. coli BL21 (DE3) and purified by nickel-chelate chromatography (see section Materials and Methods). HPLC experiments confirmed that both isoforms produce 3,4-dihydroxybenzoate as sole product from 4-hydroxybenzoate (Figure S2). Activity measurements with either NADH or NADPH established that PHBH_Cn2 is strictly dependent on NADPH whereas PHBH_Cn1 can utilize both coenzymes to perform aromatic hydroxylation. Determination of the apparent kinetic parameters k_CAT and K_M (Table 2) through monitoring NAD(P)H consumption as well as 3,4-dihydroxybenzoate production revealed that PHBH_Cn1 has a slight preference for NADH and that the NADPH-specific PHBH_Cn2 is about four times more active than PHBH_Cn1. These experiments also revealed that both enzymes suffer to some extent from uncoupling of substrate hydroxylation resulting in hydrogen peroxide as by-product, thus yielding aromatic product/NAD⁺ ratios of 0.73 and 0.81 for PHBH_Cn1 and PHBH_Cn2, respectively.

We also determined the pyridine nucleotide coenzyme specificity of the His₁₀-tagged form of a putative PHBH from R. opacus-1CP (see section Materials and Methods). Kinetic analysis of this enzyme (PHBH_Ro1CP) established a clear preference for NADH (Table 2).

The amino acid sequences of the PHBHs from C. necator JMP134 and R. opacus 1CP are in agreement with the experimentally determined coenzyme specificities. PHBH_Cn2 contains the NADPH-preferring sequence motif 32-EQRSPEYVLGR, while PHBH_Ro1CP contains the NADH-preferring sequence motif 32-ESRTREEVEGT. The NAD(P)H-dependent PHBH_Cn1 contains the sequence 32-EDCTQAHVEAR.

Bacteria capable of degrading various aromatic compounds convert the consecutive degradation products into 4-hydroxybenzoate, which then can be funneled into the protocatechuate pathway. Thus, the PHBH enzyme necessary for this route can be expected to be common among microorganisms capable of degrading these aromatic compounds.

Using the amino acid sequence of the NADPH-specific PHBH_Pf as query sequence for a BlastP search, we identified many putative PHBHs among bacterial phyla with an aerobic lifestyle. Most of them are present in proteobacteria, while roughly 10% is present in Actinobacteria. In the other domains of life, PHBH is rarely present. In Archaea, a few putative PHBHs are found, while in Eukarya a small number of hypothetical PHBHs are identified in basidiomycetes such as Ceratitis capitata, XP_004528594; Trichosporon oleaginosus IBC246, KLT40385; and Trichosporon asahii var. asahii CBS 2479, EJT53028. Some of them are similar to PHBH-like proteins of proteobacteria, while others show a high similarity to PHBH-like proteins encoded by Streptomyces species (cf. Figure S3a).

By limiting the BlastP output to an E-value smaller than 1e⁻¹⁰⁰, 6135 sequences were retrieved. From these sequences, 1423 had an unique sequence for the loop-helix H2 region. Taking the first 200 sequences of this group for construction of a sequence motif showed that the previously found motif 32-ERxxx(D/E)YVLxR for NADPH specificity is more accurately described by 32-ERx(S/T)x(D/E)YVL(G/S)R (Figure 3A). Similarly, by using the NADH-preferring PHBH_Ro protein sequence as BlastP query, we found 6,337 sequences with an E-value smaller than 1e⁻¹⁰⁰, having 1564 unique loop-helix H2 regions. Taking the first 200 sequences of this group for construction of a sequence motif showed that the NADH-preferring PHBH motif is represented by 32-ExR(S/T)Rxx(I/V)ExT (Figure 3B). After filtering duplicates from the combined total number of 12472 PHBH sequences, 6,482 sequences were unique. Thus, a large overlap exists between the two groups. In the dataset obtained using the PHBH_Pf sequence, 145 sequences were not present in the dataset obtained with the PHBH_Ro sequence. Vice versa, 347 sequences were not found in the PHBH_Pf dataset. The distribution of the sequences not present in each dataset (Figure S3b) shows that most of the sequences only present in the Ro-dataset are found among the first 2,000 sequences, while those for the sequences only present in the Pf-dataset are located in the last 3,000 sequences of each group.

Interestingly, among the actinobacterial sequences presently available, most comprise the NADH-preferring fingerprint. However, Mycobacteria have a mixed type motif, often the first or both arginine(s) of the NADH-fingerprint are present but the remaining part is lacking. In addition, many mycobacterial sequences have parts of the NADPH-preferring fingerprint, especially, x(D/E)YVL(G/S)R. Among the Streptomyces sequences many have the NADH-preferring fingerprint, but some also have a mixed type like Mycobacteria. However, these mixed-type fingerprints do not have larger parts of the NADPH-fingerprint and are more similar to the sequence of Cupriavidus PHBH_Cn1 and thus might accept both NADH and NADPH. Among rhodococci, the NADH-fingerprint is highly conserved and only a few examples of a mixed type were identified, for example among plant pathogens as Rhodococcus fascians, which shows a similar sequence to some Mycobacteria. Hereafter, we focused on bacterial PHBHs from which 70 sequences (including the 15 biochemically characterized PHBHs and 55 randomly chosen candidates of various bacteria) were chosen for further analysis of the pyridine nucleotide coenzyme specificity.

The 70 selected PHBH amino acid sequences and 4 distinct proteins (as out-group as reported elsewhere; Pérez-Pantoja et al., 2008) were used to generate an extended multiple sequence alignment (Figure S1). All sequences in the alignment (except ZP_01743892) harbor the three consensus sequences of flavoprotein hydroxylases involved in FAD binding (Eppink et al., 1997). Furthermore, residues in direct contact with the aromatic substrate are strongly conserved. These residues include Tyr201 and Pro293, which interact with the phenolic moiety, and Ser212, Arg214, and Tyr222, involved in binding the carboxylic group of 4-hydroxybenzoate (Schreuder et al., 1989). With exception of Ser212 (97% Ser, 3% Thr), these residues are 100% conserved.

As already indicated by the pairwise similarity data, the distance tree of bacterial PHBHs (Figure 4) does not reflect the taxonomic relationships, in contrast to what one might expect for a chromosomally encoded enzyme. While some branches in the distance tree represent sequences of only relatively closely related strains, such as various Burkholderia strains or various Rhodococcus strains, other branches represent relatively closely related PHBHs from taxonomically distant bacteria (e.g., from the phyla of proteobacteria Acinetobacter sp. ADP1, C. necator JMP134, Polaromonas sp. JS666, Mesorhizobium loti MAFF303099, and Rhodospeudomonas palustris CGA009).

Interestingly, the distance tree clearly reflects the pyridine nucleotide coenzyme preference shown in Table 1. All NADPH-specific PHBHs are located on one side of the tree and on the opposite side the NADH-preferring enzymes are clustered. In between these types we mostly find PHBHs for which a pyridine nucleotide coenzyme preference is not proven yet. However, this preference can be predicted from the phylogenetic tree, and we conclude that representatives closer to the NADH-assigned PHBHs can use both coenzymes, with a preference for NADH. We experimentally confirmed this conclusion by determining the pyridine nucleotide specificity of PHBH_Cn1, a newly produced representative of this group (Table 2). In the other part of the tree closer to the NADPH-assigned enzymes, PHBHs may also use both pyridine nucleotides but tend to be stricter or even exclusively dependent on NADPH. The out-group of the distance tree includes NAD(P)H-dependent enzymes (Pérez-Pantoja et al., 2008) and intersects the NAD(P)H using putative PHBHs close to the NADH-preferring PHBH type. From an evolutionary point of view this makes sense since a PHBH-predecessor protein might have used both nicotinamide cofactors or even had a preference for NADH. However, more questions on the PHBH evolution need to be answered, e.g., has the pyridine nucleotide coenzyme preference happened by chance or by adaptation, and why does it seem to be stable among certain bacteria, especially among Actinobacteria? Most Actinobacteria show a NADH-preferring fingerprint or a slightly altered one with the exception of Mycobacteria. This might be related to lifestyle and environment of those bacteria, which needs further investigations.

To get more insight into the evolutionary relationship of the pyridine nucleotide coenzyme specificity, we extracted energy potentials of residues located in the PHBH coenzyme fingerprint motifs from energy profile datasets. Pairwise alignments of these sub-energy profiles have been computed and used for deriving dScores which, similar to the strategies elucidated in the Materials and Methods section, have been processed by un-weighted pair group method arithmetic mean clustering (Figures S4, S5) or neighbor joining hierarchical clustering (Figures S6, S7). A multiple sequence alignment-like representation of these energy potentials (Figures S8, S9) illustrates a strong relationship between residue composition, pyridine nucleotide coenzyme specificity, and energetic properties. First, it becomes clear that conserved residues in these motifs yield a conservation of their energetic state, with most energy potentials being relatively low. It can be proposed that these energetically conserved residues serve as fold stabilizing elements in these motifs as well as in the intra-molecular environment of helix H2. Compared to NADPH-specific and NAD(P)H-dependent PHBHs, NADH-preferring PHBHs yield a high-energetic, unstable environment (Figure S10), which is energetically determined by the presence of two conserved Glu-residues and variable positions which are predominantly occupied by destabilizing residues, such as Asp, Glu and Arg (Zhou and Zhou, 2004). In contrast to these findings, residues in the coenzyme fingerprint motif of NADPH-specific and NAD(P)H-dependent PHBHs yield comparatively low energy potentials and thus are partly stabilizing the binding moiety. It can be concluded that this deviation in molecular stability can contribute to the pyridine nucleotide coenzyme specificity and is an important driver of PHBH evolution.

We used the Rate4site program (Materials and Methods section) to assess the evolutionary rate of NADPH-specific and NADH-preferring PHBHs. Figure S11 shows that the NADH-enzymes have more regions (colored red) susceptible to mutation compared to the NADPH-enzymes. Indeed, also the loop region with the coenzyme-binding motif is a little more mutation sensitive in the NADH-preferring enzymes, indicative of a strong selection favoring specific amino acids in the NADPH-specific enzymes.

Studies from PHBH variants generated using site-directed mutagenesis support the idea that Tyr38 and Arg42 of helix H2 confer the specificity of PHBH_Pf for NADPH (Eppink et al., 1998b, 1999; Huang et al., 2008). Based on these findings and the fact that the nicotinamide ring of NADPH binds at the re-side of the flavin (Manstein et al., 1986), we docked the NADPH in the enzyme-substrate complex of PHBH_Pf with the isoalloxazine moiety of the FAD cofactor oriented in the out conformation. As can be seen from Figure 5, the docking predicts that His162 and Arg269 interact with the pyrophosphate moiety of NADPH (Eppink et al., 1998a; Wang et al., 2002) and that Arg33, Tyr38 and Arg42 of the NADPH-specific fingerprint sequence 32-ERx(S/T)x(D/E)YVL(G/S)R are involved in orienting the adenosine 2′-phosphate part of NADPH (Eppink et al., 1999).

At present, crystal structures of 28 different group A flavoprotein monooxygenases are available in the Protein Data Bank, and for most of these enzymes, the preference for the nicotinamide cofactor is known. Structural alignment of the subfamily members showed similar folds for the FAD and substrate binding domains, which is indicative for a conserved interdomain binding mode of the NAD(P)H coenzyme (Treiber and Schulz, 2008). We aligned the loop segments of these enzymes, putatively involved in NAD(P)H binding, based on the structural position of the adenine moiety of the FAD cofactor and the N- and C-termini of these loops. The alignment obtained from the loop segment sequences (Figure 6A) suggests that the proteins can indeed be grouped in NADPH- or NADH-dependent enzymes and the associated distance tree shows this feature as two separate clusters (Figure 6B). The NAD(P)H-dependent enzymes are located in both the NADH- and NADPH-cluster. Based on type of cluster, the “putative monooxygenase” from P. luminescens (PDB ID: 4hb9) is likely NADH-dependent, whereas the “putative monooxygenase” from P. aeruginosa (PDB ID: 2x3n) is likely NADPH-dependent.

Within the whole subfamily, there is no clear consensus motif present for NADH- or NADPH-dependency. However, most NADPH-dependent enzymes have an arginine at position 44 (PHBH_pf numbering), which is capable of H-bond formation, in contrast to the corresponding residue in the NADH-group. The NADH-group has instead a 'GxG' motif near the end of the loop, with x being mostly a hydrophobic residue.

The loop segments do not show a clear consensus structure (Figure 7). Those from PHBH enzymes contain a small helix, but none of the others structures have this feature. A few structures are missing some amino acid residues in the loop segment, due to low electron density in the diffraction dataset, which indicates that here the loop is flexible. This flexibility might change upon NAD(P)H binding, which could be essential to allow for the isoalloxazine moiety movement of FAD (i.e., “in/out” conformation).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.