Cyclization of OPA Intermediate

We next explored the cyclization step of OPA to give 1 in the xsb BGC. Upon closer examination of the gene component, we notice that xsbB features an A domain that might adenylate non‐amino acid substrates, due to the lack of a conserved aspartate residue in motif A4 that typically interacts with the α‐NH₂ of amino acid substrates [18] (Figure S2). Also, a 3‐oxoacyl‐ACP synthase encoded by xsbD has a high sequence identity (83.6%) with XpzS [16] that catalyzes amide‐ and ester‐bond formation between the carrier‐protein‐bound aminoacyl with phenazines. Therefore, there are three conceivable pathways to achieve OPA cyclization (Figure 3a): 1) the A domain of XsbB adenylates OPA for loading onto the T domain, and subsequently, XsbD catalyzes intramolecular amide‐bond formation; 2) XsbC thioesterifies the carboxylate of enolpyruvate with a CoASH, followed by amide‐bond formation under XsbD catalysis; or 3) the cyclization occurs in a non‐thiotemplated fashion, during which XsbC activates the carboxylate of enolpyruvate through ATP‐dependent adenylation, followed by direct nucleophilic attack of the NH₂ on the carboxylate without forming a CoA‐bound thioester intermediate, as observed during the biosyntheses of coumermycin A1, [19] dapdiamide, [20] and yatakemycin. [21] To confirm the involvement of genes in OPA cyclization, we individually deleted xsbB, xsbC, and xsbD. While the P _BAD xsbA ΔxsbB and P _BAD xsbA ΔxsbD mutants did not affect the product profile (Figure 3b, traces iii and v), 1 was completely absent in the P _BAD xsbA ΔxsbC mutant (Figure 3b, trace iv), revealing xsbC being solely responsible for the formation of 1 from OPA. We then verified the function of XsbC by in vivo experiments. E. coli BL21(DE3) harboring xsbA, xpzC, and xsbC produced 1, while no production of 1 was observed in E. coli expressing xsbA and xpzC (Figure 3c, traces iii and iv).

To support our hypothesis that XsbC belonging to the phenylacetate‐CoA ligase family mediates OPA cyclization via an ATP‐dependent adenylating manner (Figure 3a), we pursued a maximum‐likelihood‐based phylogenetic analysis [22] to infer the relationships between XsbC and enzymes in the ANL (acyl‐CoA synthetases, NRPS adenylation domains, and luciferase enzymes) superfamily. The resultant tree (Figure S3) showed that XsbC is distantly related to NatL2 and BomJ, both of which are acyl‐AMP ligases involved in intermolecular ester‐bond formation.[ 23 , 24 ] Although a multi‐sequence alignment indicated a low sequence identity (19.8%) of XsbC with NatL2, a closer inspection of the alignment (Figure S4) and homology modeling (Figures S5–8) revealed that XsbC harbors diagnostic binding sites and conserved motifs that are observed in NatL2. [23] The Zn²⁺‐binding tetrad Cys263‐His269‐Cys321‐Cys323 is a unique feature in distinguishing NatL2 from canonical Mg²⁺‐binding CoA ligases and is well conserved in XsbC (Figures S4 and S6). Most importantly, a C‐terminal extension with Lys429 in NatL2 is also present in XsbC. Lys429 from the other monomer forms a salt bridge with the AMP (Figure S7), and thus C‐terminal extension crossing to the other monomer is proposed to lock the functional homodimer in the “close” conformation, which prevents a CoA from entering the binding pocket.

Taking all these data into account, we propose a three‐gene cassette consisting of an anthranilate synthase component I homolog (e.g. SgcD) or ADIC synthase (e.g. XpzC), an FMN‐dependent and [Fe−S]‐containing dehydrogenase (e.g. SgcG and XsbA), and an acyl‐AMP ligase (e.g. XsbC) being responsible for benzoxazolinate biosynthesis.

Benzobactin BGC in Diverse Bacteria

To expand the biosynthetic repertoire of benzoxazolinate‐containing natural products, we carried out a survey of all bacterial genomes available in the NCBI database using the protein sequence of XsbC as a query. Besides C‐1027 [6] and ashimides [9] homologous BGCs found in Actinobacteria as we expected, the Blast result revealed that the orphan xvb BGC responsible for benzobactin biosynthesis in entomopathogenic bacteria Xenorhabdus and Photorhabdus [10] abounds in Pseudomonas, and is prevalent across two phyla of bacteria (Figure 2b), the Proteobacteria (e.g. Pseudomonas, Vibrio, Shewanella, and Brenneria) and Firmicutes (e.g. Bacillus, Paenibacillus, Halobacillus, and Clostridium). Such extensive similarities to the xvb BGC suggest that benzoxazolinate‐containing natural products are more widespread in diverse bacteria—from ocean to land—than we thought, and that Pseudomonas being the most abundant genus among all these strains might be prolific producers for these rare natural products.

We selected different Pseudomonas strains for homologously expressing the target BGC by a promoter‐exchange strategy. [25] A BGC, designated as pbz, from Pseudomonas chlororaphis subsp. piscium DSM 21509 was successfully activated as indicated by the significant distinction of HPLC profiles between induced and non‐induced mutants (Figure 4a, traces i and ii). In addition to 2 produced in a high amount, the P. chlororaphis P _BAD pbzA mutant yielded five more benzobactin derivatives with masses ranging from 305 to 1012 (Table S5) as determined by untargeted analysis, and therefore we attempted to isolate them for structure elucidation. The compound with a mass of 305, designated as benzobactin B (3) confirmed by NMR spectroscopy (Table S6 and Figure S1), has one 2‐hydroxymethylserine residue less than 2. Compounds with larger masses (e.g. 609, 628, and 1012) were not isolated due to their low production levels and instability. Among them, the structure of a compound with a mass of 609, designated as benzobactin C (4), was tentatively assigned to be a dimer of 3 by tandem MS/MS and isotope labeling experiments (Figures S9 and S10a). Based on the sum formula (Table S5), we proposed that benzobactins‐628 and 1012 are a dimer and tetramer of 3, respectively, while their exact chemical structures cannot be formulated at the current stage.

Open in a separate window

Figure 4

HPLC analysis of P. chlororaphis subsp. piscium DSM 21509 mutants. a) HPLC‐UV analysis of (i) the constructed promoter exchange mutant P. chlororaphis P _BAD pbzA and (iii–vii) deletion mutants. Deletions were carried out in the P _BAD pbzA mutant. Mutants expressing the pbz BGC were induced by l‐arabinose. 1 (dark blue) and 2 (red), 3 (green), and 4 (orange), as well as the as‐yet‐uncharacterized benzobactins‐628 (light blue) and 1021 (purple), are highlighted in the traces. (i) The P _BAD pbzA and (vi) P _BAD pbzA ΔpbzG mutants shared an identical metabolite profile, indicating that PbzG (homologous to anthranilate synthase component II) might not be involved in benzobactin biosyntheses, or that its function could be complemented by other genes in the genome. b) HPLC‐MS analysis of (i–iii) the ΔpbzF, ΔphzE, and ΔpbzF ΔphzE mutants for benzobactin production. Deletions were carried out in the P. chlororaphis P _BAD pbzA mutant. Shown are the EICs of 2 (red) and 3 (green) in deletion mutants. c) Production of benzobactins in the ΔpbzF and ΔphzE mutants. The production of individual compounds at each deletion mutant is relative to its production in the P _BAD pbzA mutant (in percentage, %). nd=not detected.

Synergistic Effect of ADIC and Anthranilate Synthases

The pbz BGC contains a typical three‐gene benzoxazolinate cassette pbzCEF. An ADIC synthase (PhzE) in a phenazine biosynthetic operon that is 2.6 megabase pairs away from the pbz BGC is also present in the genome (Figure 2b). Again, consistent with the speculation that the ADIC precursor could be synthesized by PhzE and PbzF, both of which are functionally complementary, the P _BAD pbzA ΔphzE and P _BAD pbzA ΔpbzF mutants still yielded 2 and 3 (Figure 4b, traces i and ii), while the products completely lost in the P _BAD pbzA ΔpbzF ΔphzE mutant (Figure 4b, trace iii). Surprisingly, the production titers of 2 and 3 decreased ≈30‐fold in the P _BAD pbzA ΔphzE mutant and ≈900‐fold in the P _BAD pbzA ΔpbzF mutant, compared to that of the P _BAD pbzA mutant (Figure 4c). This suggests that PbzF and PhzE are not only functionally complementary but also have a synergistic positive effect on production level of the downstream benzobactin products, presumably because the ADIC metabolic flux is dramatically increased by both enzymes. The coexistence of pbzF and phzE also occurs in the genomes of X. vietnamensis DSM 22392 and Vibrio nigripulchritudo SOn1 (Figure 2b). These observations perhaps are unsurprising, because phenazines, which often have a high expression level being the pigmentation phenotype of wild‐type strains, are much more widely distributed in bacteria (particularly in Pseudomonas)[ 26 , 27 ] than benzoxazolinate derivatives.

Unusual Serine Hydroxymethyltransferase

The bimodular NRPS, PbzD, appears to be responsible for loading and elongating 1 (Figure 5a), since the A₁ domain is non‐amino‐acid substrate specific, as indicated by its motif A₄ (Figure S2). This is supported by the P _BAD pbzA ΔpbzD mutant that abolished production of benzobactins but accumulated that of 1 (Figure 4a, trace v). It is proposed that the incorporation of a 2‐hydroxymethylserine into the benzoxazolinate moiety is performed by the second module of PbzD with the assistance of PbzB which is a putative serine hydroxymethyltransferase. Pyridoxal‐5′‐phosphate‐dependent (PLP) serine hydroxymethyltransferases were reported to catalyze the transfer of a hydroxymethyl group from 5,10‐methylenetetrahydrofolate (mTHF) to glycine or alanine to afford l‐serine or α‐methylsereine, as found during the biosyntheses of ashimides [9] and JBIR‐34/35 (ref. [28]). PbzB has high sequence similarity with AsmD [9] and FmoH, [28] and contains a conserved PLP‐binding loop (Figure S11) in which the lysine residue forms a Schiff base with PLP. [29] Therefore, we assumed that PbzB could catalyze mono‐hydroxymethylation of serine or/and bis‐hydroxymethylation of glycine to afford 2‐hydroxymethylserine.

Open in a separate window

Figure 5

Proposed biosynthesis of benzobactins in P. chlororaphis subsp. piscium DSM 21509. a) Proposed biosynthetic pathways of 2 and 3. b) Proposed biosynthetic pathway of 4. Domain involved in the corresponding reaction step is highlighted in green. The color codes enzymes are corresponding to the encoded genes in Figure 2. A, adenylation; T, thiolation; C, condensation; and TE, thioester domains.

To verify the origin of building blocks in vivo, we cultivated the P _BAD pbzA mutant in ¹⁵N media supplemented with l‐serine and glycine. Exemplified by compound 3, while no inverse mass shifts were obtained in the ¹⁵N medium supplemented with l‐serine, we observed—1 Da mass shift in the ¹⁵N medium supplemented with glycine, which resulted from the incorporation of a glycine residue (Figure S10b). Phylogenetic analysis revealed that the PbzD‐A₂ domain unexpectedly falls into the clade of A domains with cysteine specificity and is separate from those with glycine or serine specificity (Figure S12). This suggests that PbzD‐A₂ domain might be evolved for larger substrates and adenylate non‐glycine substrates for loading onto PbzD‐T₂ domain. Thus, we propose that PbzB catalyzes the conversion of glycine into 2‐hydroxymethylserine prior to being subjected to the PbzD‐A₂ domain activation (Figure 5a).

To verify the conversion of glycine into 2‐hydroxymethylserine by PbzB, we expressed and purified PbzB with a His‐SUMO‐tag from E. coli BL21(DE3) (Figure S13a) and conducted in vitro enzymatic assays (Figure 6a). Incubation of PbzB with glycine and cofactors, PLP and mTHF, in the potassium phosphate buffer at pH 7.5 showed the formation of 2‐hydroxymethylserine and a trace amount of serine (Figure 6b, trace ii). This finding motivated us to examine whether PbzB can also catalyze the conversion of d‐/l‐serine into 2‐hydroxymethylserine. Surprisingly, under the same enzymatic reaction condition, the production of 2‐hydroxymethylserine that was converted from d‐/l‐serine was much higher than that from glycine (Figure 6c). While glycine was the authentic building block in vivo, serine was a more favored in vitro substrate for PbzB than glycine. Two possible explanations for the seemingly conflicting results are that 1) in in vitro, the one‐step conversion of serine to 2‐hydroxymethylserine is faster than the two‐step conversion of glycine to 2‐hydroxymethylserine; 2) in in vivo, the degradation of serine to glycine catalyzed by the serine hydroxymethyl transferase from primary metabolism is faster than the formation of serine to 2‐hydroxymethylserine catalyzed by PbzB. We then carried out a steady‐state kinetic analysis for the formation of 2‐hydroxymethylserine using d‐/l‐serine (Figure 6d and e), which suggested that the affinity of PbzB for l‐serine is 9‐fold higher than that for d‐serine as indicated by the K _m values, and that PbzB is 22‐fold more efficient at catalyzing the formation of 2‐hydroxymethylserine with l‐serine than that with d‐serine based on the k _cat/K _m values.

Open in a separate window

Figure 6

In vitro characterization of 2‐hydroxymethylserine conversions. a) Enzymatic conversion of glycine and d‐/l‐serine to 2‐hydroxymethylserine using the recombinant serine hydroxymethyltransferase PbzB. b) EICs of d‐/l‐serine (black, 106.0492 [M+H]⁺) and 2‐hydroxymethylserine (red, 136.0604 [M+H]⁺) are shown, while the mass of glycine is too small to be detected. c) Quantification of 2‐hydroxymethylserine production with d‐/l‐serine and glycine substrates. Steady‐state kinetics of formation of 2‐hydroxymethylserine from d) l‐serine (K _m=69.52±0.03 μM, k _cat=0.151±0.006 min⁻¹) and e) d‐serine (K _m=602.90±56.67 μM, k _cat=0.058±0.003 min⁻¹).

Bioinformatics and Structure of PbzB

Serine hydroxymethyltransferases involved in the central metabolism have been studied in various organisms. [30] While GlyA from E. coli, one of the best‐studied examples, catalyzes reversible conversion between serine and glycine and thereby affords mTHF as the major source of C1 unit in cells, [31] the newly identified PbzB involved in the benzobactin biosynthesis catalyzes production of 2‐hydroxymethylserine using l‐serine, d‐serine, or glycine. Sequence alignment analysis (Figure S11b) of PbzB and its homologs AsmD, [9] FmoH, [28] and XvbB [10] (PbzB type) in comparison with structurally and biochemically characterized GlyA‐type enzymes revealed a high sequence similarity of over 42%. Almost all residues involved in coordinating N‐pyridoxyl‐glycine‐5‐phosphate (PLG) and mTHF were conserved among the analyzed proteins (Figure S11b). While GlyA‐type enzymes contain a conserved YAEG— ‐RYY motif, where the two tyrosine residues (Y65 and Y55) are crucial for efficient substrate conversion and the formation of cation π‐interaction with the ligand,[ 32 , 33 ] the PbzB‐type enzymes harbor a TAEG— ‐RYH motif (Figure S11b).

To gain an insight into the substituted residues involved in substrate binding in PbzB, we set out to structurally analyze its apo‐ and substrate‐bound form. While crystals of the apo‐form diffracted to 2.8 Å, we only obtained poorly diffracting anisotropic crystals for PbzB co‐crystallized with l‐serine and mTHF. The crystal structure of apo‐PbzB (Figure 7a; Table S7; PDB 7QWZ) shared the overall fold of GlyA‐type enzymes with an RMSD value of 2.86 to GlyA (Figure 7b and c). PbzB, same as GlyA‐type enzymes, forms homodimers with two active sites shared between the two monomers (Figure 7a and b). However, the residues coordinating PLG and mTHF were not visible in the PbzB structure, since they were probably too flexible in the absence of a ligand (Figure 7d), and therefore, we modeled PbzB using Alphafold. [34] The simulated structural model had a confidence of over 95% and an RMSD of 0.97 and 0.84 for the crystal structures of apo‐PbzB and GlyA from E. coli, [31] respectively (Figure 7c). Given that GlyA binds PLP catalyzing formation of PLG in presence of glycine, we postulated that in addition to the formation of PLG, PbzB generates N‐pyridoxyl‐serine‐5‐phosphate (PLS) as an external aldimine formed by PLP and l‐serine (Figure 7e). An in‐depth analysis of the active sites revealed that all residues involved in substrate binding within GlyA and PbzB are superimposable (Figure 7f). Even Y55 and Y65 in GlyA are superimposable with T68 and H78 in PbzB, but these alterations might enlarge the binding pocket of PbzB. While Y55 and Y65 in GlyA perfectly coordinate PLG, T68 and H78 in PbzB open the space for the binding of PLS. Particularly, the substitution of Y55 by T68 may lead to a reduction of negative charges in the binding pocket and thus enable coordination of serine bound to PLP (Figure 7g–i). We constructed a site‐directed mutant of PbzB to investigate the key residues and attempt to shrink the capacity of the binding pocket. T68 and H78 were both mutated to Tyr as in GlyA (Figure S13). However, the PbzB T68Y H78Y mutant did not catalyze formation of any detectable amount of serine or 2‐hydroxymethylserine, suggesting that T68 and H78 residues are crucial to maintaining the function of PbzB.

Open in a separate window

Figure 7

Structural and bioinformatical analysis of PbzB and its homologs. a) Overall‐fold of the crystal structure of apo PbzB. The individual monomers of the dimeric complex are shown in grey and orange (PDB 7QWZ). b) Overall‐fold of the crystal structure of GlyA in complex with PLG (yellow) and mTHF (pink). The two monomers of the GlyA dimer are shown in grey and blue (PDB 1DFO). [31] c) Root mean square deviation (RMSD) of PbzB apo structure (PDB 7QWZ), PbzB model, AsmD model, FmoH model, and XvbB model (models obtained by Alphafold) in comparison to the crystal structure of GlyA (PDB 1DFO). [31] d) Overlay of the PbzB Alphafold model (green) and the PbzB apo crystal structure (grey). Undissolved regions of the PbzB crystal structure are highlighted with red in the PbzB Alphafold model. e) Comparison of the PLP‐bound substrates PLG and PLS. f) Overlay of the residues present in the active site of PbzB (Alphafold model, green) and GlyA (crystal structure, PDB 1DFO, grey). g) Detailed view of the GlyA active site coordinating mTHF (pink) and PLG (yellow). h) Modeled view of the PbzB active site coordinating mTHF (pink) and PLS (orange). i) Overlay of the active sites of GlyA and the PbzB Alphafold model. In (g)–(i), the residues with red circles are critical for fitting the amino acid substrate into the active site. In i, possible stereo hinderance between serine and Y65/Y55 in GlyA is indicated.

Consistent with a general observation that central metabolism enzymes have higher k _cat and k _cat/K _m values than secondary metabolism enzymes, [35] the k _cat and k _cat/K _m values of GlyA [36] is 4200‐fold and 360000‐fold larger than those of PbzB for l‐serine as a substrate. However, the K _m value of PbzB is ≈9‐fold lower than GlyA, suggesting that PbzB has a better affinity for l‐serine than GlyA. Although PbzB is much less efficient than GlyA, a low flux of 2‐hydroxymethylserine formation can be maintained by PbzB that satisfies the requirement of benzobactin production without disturbing primary metabolisms. Therefore, one could assume that benzobactin‐producing bacteria remodel an enzyme from the central amino acid metabolism to deliver a building block for the NRPS biosynthetic pathway. Since only 1 and no benzobactins were observed in the P _BAD pbzA ΔpbzB mutant (Figure 4a, trace iv), 2‐hydroxymethylserine furnished by PbzB is crucial for the downstream assembly line.

We acknowledge the help of Viktor Lehmann from the Philipps University of Marburg during the crystallization process of PbzB and the constructive suggestion on in vitro pathway reconstructions of Dr. Henrike Niederholtmeyer from the Max Planck Institute for Terrestrial Microbiology. Yi‐Ming Shi was supported by the Alexander von Humboldt Foundation. This work is supported by the LOEWE Centre for Translational Biodiversity Genomics (LOEWE TBG) and the ERC advanced grant (835108) to Helge B. Bode. Open Access funding enabled and organized by Projekt DEAL.