Structural and biosynthetic analysis of the fabrubactins, unusual siderophores from Agrobacterium fabrum strain C58

Abstract

Siderophores are iron-chelating molecules produced by microorganisms and plants to acquire exogenous iron. Siderophore biosynthetic enzymology often produces elaborate and unique molecular through unusual reactions to enable specific recognition by the producing organisms. Herein, we report the structure of two siderophore analogs from Agrobacterium fabrum strain C58, which we named fabrubactin (FBN) A and FBN B. Additionally, we characterized the substrate specificities of the NRPS and PKS components. The structures suggest unique Favorskii-like rearrangements of the molecular backbone that we propose are catalyzed by the flavin-dependent monooxygenase, FbnE. FBN A and B contain a 1,1-dimethyl-3-amino-1,2,3,4-tetrahydro-7,8-dihydroxy-quinolin (Dmaq) moiety previously seen only in the anachelin cyanobacterial siderophores. We provide evidence that Dmaq is derived from L-DOPA and propose a mechanism for the formation of the mature Dmaq moiety. Our bioinformatic analyses suggest that FBN A and B and the anachelins belong to a large and diverse siderophore family widespread throughout the Rhizobium/Agrobacterium group, α-proteobacteria, and cyanobacteria.

INTRODUCTION

Siderophores are high-affinity iron-chelating small molecules made by plants, fungi, and prokaryotes, that are used to acquire exogenous iron. Biosynthesis of siderophores comes at a significant metabolic cost to the producing organism, which is further exacerbated by competitors capturing the ferric-siderophore before it can be absorbed by the producer. As a result, siderophore biosynthesis is an evolutionary battleground of enzymology that leads to a high diversity of elaborate structures^1–4. One such example are the pyoverdines, made by fluorescent pseudomonads. Apart from a characteristic dihydroxyquinoline chromophore present in all of these molecules, pyoverdine biosynthetic enzymology produces remarkable structural diversity by varying the number, identities, and cyclization of amino acids incorporated into these natural products⁵. To date there are more than 60 structurally unique pyoverdines⁶. Thus, exploring siderophore biosynthesis can facilitate the discovery of unusual natural product enzymology and provide insights into how Nature generates natural product diversity.

We have previously reported that Agrobacterium fabrum strain C58, formerly known as A. tumefaciens strain C58^{7, 8}, produces a siderophore assembled by a hybrid nonribosomal peptide synthetase (NRPS)/polyketide synthase (PKS)⁹. This conclusion was based on the associated biosynthetic gene cluster (BGC) encoding two PKS modules, one fatty acyl-CoA ligase module, and seven NRPS modules. This enzymology did not align with the structure of any known siderophore from Agrobacterium spp. or related bacteria. Interestingly, some of the enzymes encoded by this BGC have homologs in the cyanobacterium Anabaena sp. strain PCC7120⁹, and it was subsequently shown that the BGC from this cyanobacterium encodes enzymes that assemble a siderophore of unknown structure¹⁰. More recent bioinformatic analyses of BGCs in cyanobacteria suggest the BGC in PCC7120 is involved in the production of an anachelin analog¹¹.

In our previous study, we were able to purify the siderophore, but did not perform any structural analyses. Efforts to use biochemical characterization of the biosynthetic enzymes to guide structural analysis were thwarted by our inability to obtain soluble NRPS components when heterologously produced in Escherichia coli. Since our initial work, we, and others, discovered that small proteins belonging to the MbtH-like protein (MLP) superfamily dramatically influence the solubility and activity of associated NRPSs^12–14. Genes encoding MLPs are commonly found in association with genes encoding NRPSs and hybrid NRPS/PKSs¹⁵, as is the case with the A. fabrum strain C58 siderophore BGC.

Herein we report the structure of two siderophore analogs from A. fabrum strain C58 that we refer to as fabrubactin (FBN) A and B and the biochemical characterization of the NRPS and PKS components that produce the siderophore backbones. These siderophores contain a lipid tail at one end and are capped by a 1,1-dimethyl-3-amino-1,2,3,4-tetrahydro-7,8-dihydroxy-quinolin (Dmaq) at the other (Figure 1). This Dmaq moiety is rare in natural products and has previously only been found in three analogs of the cyanobacterial siderophore anachelin^{16, 17}. We provide biochemical evidence that levodopa (L-DOPA) is a precursor for the formation of the Dmaq moiety and is produced by a heme peroxidase. We hypothesize that this heme peroxidase is bifunctional and catalyzes L-DOPA formation and aza-annulation to form the Dmaq moiety. FBN A and B are unusual in that they contain two backbone quaternary carbons each with a free hydroxyl and carboxyl group. We propose these moieties are formed by two Favorskii-like rearrangements catalyzed by a flavin-dependent monooxygenase. Finally, the formation of the two analogs suggests the potential for an unusual NRPS module that first selectively activates β-Alanine (β-Ala), covalently tethering the amino acid to its thiolation domain, and then activates either another β-Ala (FBN A) or captures an activated L-Ala from a neighboring NRPS for amide bond formation (FBN B), generating a dipeptide on the thiolation domain. The structure of the FBNs and the unusual enzymology provide more support for siderophores and their biosyntheses being rich sources of novel chemistry and enzymes. Additionally, bioinformatics suggests the FBNs and anachelins are a large class of Dmaq-containing siderophores found in many bacteria.

Figure 1. Structures of fabrubactin (FBN) analogs A and B.

Unique partial structures 2-(2- aminoethyl) - 2-hydroxy-propanedioic acid (Ahpa; shown in blue), 2-C-carboxy-2,4-dihydroxy-undecanoic acid (Cdua; shown in red), and 1,1-dimethyl-3-amino-1,2,3,4-tetrahydro-7,8-dihydroxy-quinolin (Dmaq; shown in green) are indicated. Only atoms unique to FBN B are numbered on the structure of this analog.

RESULTS AND DISCUSSION

Structure elucidation of FBN A and B.

Growth conditions in Tris-based minimal medium containing the iron chelator dipyridyl were optimized for siderophore production. The siderophores were purified as described in the Experimental Procedures. Fractions containing the siderophore were identified by reactivity with chromeazurol S reagent^{18, 19}. During growth and purification, a parallel culture of an A. fabrum strain C58 disrupted in siderophore production (an fbnN deletion) was used as a negative control. From eight liters of culture, five mg of FBN A and B were obtained for structural studies.

The structures of the two siderophores, FBN A and B (Figure 1), were determined using a combination of MS and NMR. The purified compounds had an m/z of 1008.4334 (M⁺), showing the appropriate M+1, M+2, and M+3 isotopic patterns (Figure S1A). A metabolite with this m/z was not observed in the negative control. The molecular formula of FBN A and B was established as C₄₄H₆₆N₉O₁₆S based on high-resolution electrospray ionization mass spectrometry (HR-ESI-MS) data. The two analogs are identical except for β-Ala in FBN A and Ala in FBN B corresponding to backbone atoms numbered 39–42. Therefore, FBNs will henceforth be used to refer to both analogs unless otherwise indicated. Additionally, ultra-high-resolution Fourier transform ion cyclotron resonance (FT-ICR) MS was also employed to obtain the exact molecular formula information from its isotopic fine patterns, which clearly suggested the presence of one sulfur (Figure S1B, S1C).

Analysis of the ¹³C-¹³C COSY data provided most of the backbone carbons of the structure (i.e., without configurational assignment) (Figure 2, Table S1, and Table S2). Analysis of the 1D NMR, COSY, and HMBC data of FBN A identified three standard amino acids including one Ala and two glycines (Gly), and one β-Ala. Similarly, two Ala and two Gly amino acids were identified in FBN B. HMBC correlations from H-29 (s, δ_H 7.88) to C-25 (δ_C 148.8) and C-27 (δ_C 174.5) suggested the presence of a 2-substituted thiazole (Thz) ring²⁰. The connectivity between C-27 and C-30 and between C-30 and C-31 completed the Ala-Thz fragment (Figure 2, Table S1, and Table S2).

Figure 2. Select 2D NMR analysis for FBN A. 2D NMR correlations to determine FBN A (β-Ala) and FBN B (Ala) structures.

NMR data is provided in Tables S1 and S2.

The HMBC correlations from the proton signals of the two N-methyl groups (H₃-11 and H₃-12) to C-2 and C-10 enabled a linkage of C-2 and C-10 via a nitrogen atom. C-7 and C-8 were assigned as two phenolic carbons based on the chemical shift and were also confirmed by NOESY correlations between OH-8 and H-9 and between H-6 and OH-7. Therefore, one of the remaining unassigned moieties was associated with the Dmaq unit (Figure 2, Table S1, and Table S2)^{16, 17}. The counterion for the Dmaq moiety was trifluoroacetate, used in the HPLC purification. The catecholate in the Dmaq moiety of anachelin is one of the functional groups involved in binding iron¹⁷. The presence of Dmaq in the FBNs provides structural support for their role as siderophores. Mixing a solution of FeCl₃ and siderophore gave a peak at 1043 m/z corresponding to a 1:1 complex as [M+Fe³⁺−3H−H₂O]⁺, determined by ESI-MS (Figure S2).

Assembly of the last partial structures, Ahpa and 2-C-carboxy-2,4-dihydroxy-undecanoic acid (Cdua) was accomplished by ¹H-¹H COSY and HMBC. Since both C-34 (δ_C 78.0) and C-44 (δ_C 77.6) were sp3 quaternary carbons, one hydroxy and one carboxylic acid group was attached to each to satisfy the molecular formula (Figure 2, Table S1, and Table S2). The presence of two carboxylic acid groups were further confirmed by ESI-MS² and FT-MS² data (Figure S3A), showing two significant peaks [M−COOH]⁺ and [M−2*COOH]⁺ at 964.4371 m/z and 920.4474 m/z, respectively.

Attempts to confirm the connectivity of the aforementioned partial structures using FT-MS² experiments were conducted concurrently with the NMR-based structure elucidation. Accurate mass measurements and fine structure isotopic patterns resulted in the assignment of exact molecular formulas of fragments (Figure S3B)^{21, 22}. The fragment ion at m/z 548.2283 could be assigned the elemental composition C₂₄H₃₄N₇O₆S⁺, corresponding to Dmaq−Gly1−Ala−Gly2−Thz-Ala. Detection of the ion at m/z 764.3029 (C₃₂H₄₆N₉O₁₁S⁺), resulted in the sequence assignment of a β-Ala−Ahpa after the Ala-Thz residue. The neutral loss between m/z 764.3029 and parent ion was produced through cleavage of the amide bond of the Cdua residue (Figure S3B). Observation of HMBC correlations from NH-42 and H₂-45 to C-43 suggested the connectivity between Cdua and β-Ala, thus completing the structure of the FBNs without configurational assignment.

Substrate specificity of the initiation module, FbnG.

The unusual structures of the FBNs raised questions about how these siderophores are biosynthesized. We previously identified a large portion of the FBN BGC⁹. Here we extended the BGC to additional ORFs to account for the biosynthesis of necessary precursors, transcriptional regulation, FBN transport across the inner and outer membranes, and the FBN biosynthetic enzymes (Table S3). At the core of FBN biosynthesis is an NRPS/PKS hybrid megasynthase (Figure 3A) and understanding the substrate-specificities of these enzymes would provide insights into how these siderophores are assembled. Each of the adenylation (A) domain and acyltransferase (AT) domain-containing modules were overproduced and purified in E. coli as either N- or C-terminal histidine-tag fusions and assayed for substrate specificity. The likely initiating module was predicted to be FbnG based on it containing an N-terminal acyl-CoA-ligase domain (AL). The purified, histidine-tagged version of this protein was assayed for fatty acid recognition using standard ATP/PP_i exchange assays. FbnG was found to preferentially activate decanoate over octanoate, nonanoate, and 3-hydroxydecanoate (Figure S4A). This result was unexpected since the FBN structures suggest that a hydroxylated fatty acid would be the initial substrate. The BGC codes for a flavin-dependent monooxygenase (FMO, FbnE) that may catalyze the hydroxylation of the thioesterified decanoate; however, as discussed below, we hypothesize that FbnE catalyzes the C2 hydroxylation of thioesterified intermediates on the PKS T domains during Favorskii-like rearrangement reactions. Currently, it is not clear if FbnE is involved in modification of decanoate while tethered to FbnG or whether FbnG has a different substrate specificity in vivo.

Figure 3. Biosynthetic scheme of the NRPS and PKS components.

A) Hybrid polyketide synthase/nonribosomal peptide synthetase encoded by the fbn BGC. Substrates incorporated by each module are displayed as the numbered chemical moiety that is added to the preceding structure (left to right) and thioesterified to the protein. B) Proposed mechanism of putative iterative NRPS FbnD leading to FBN A formation. C) Proposed mechanism of capture of L-Ala thioesterified to the T domain of either FbnO or FbnP leading to FBN B formation. Domains are abbreviated as follows: AL, acyl-CoA ligase; T, thiolation KS, ketosynthase; AT, acyltransferase; KR, ketoreductase; FMO, flavin monooxygenase; C, condensation; A. adenylation; Cy, cyclase; OX, oxidase; E, epimerase, Rd, reductase; MLP, MbtH-like protein. R is conditional upon the presence of the β-carbon (shown in parentheses as 0 or 1): if 1, R = H (FBN A), if 0, R = CH₃ (FBN B).

Substrate specificities of the adenylation domains.

All the NRPS modules were insoluble when heterologously overproduced in E. coli unless they were co-overproduced with the MLP FbnK; thus, we conclude every A domain is MLP-dependent. Each of the purified modules was assayed for amino acid activation. According to the solved FBN structures, the amino acids that must be accounted for in the biosynthesis are: two glycine residues, one L-cysteine (L-Cys) residue, two (FBN A) or three L-Ala (FBN B) residues, one (FBN B) or potentially two (FBN A) β-Ala residues, and likely an L-DOPA residue (Figure 1). Interestingly, the megasynthase only contains seven NRPS modules for eight amino acids, with one of the remaining modules (FbnQ) lacking an A domain (Figure 3A).

FbnH was hypothesized to be the terminal module since it contains a C-terminal reductase (Rd) domain to reductively release the product from the megasynthase. Based on the tetrahydroquinolone termini of the FBNs, FbnH was predicted to activate L-DOPA. FbnH was assayed with all 20 proteinogenic amino acids and L-DOPA, with L-DOPA being the only amino acid that stimulated significant ATP/PP_i exchange (Figure S4B); thus, FbnH is the terminal module of the megasynthase.

FbnO is a two-module NRPS subunit. Screening the A domain of the first module for amino acid activation using the standard 20 proteinogenic amino acids detected the highest activation with L-Ala, but also activation of L-Cys, Gly, and L-Ser (Figure S4C). To determine the preferred substrate, the specificity constant (V_max/K_m) was determined for each amino acid (Table S4). These data support L-Ala as the preferred substrate for the first module of FbnO. The second module of FbnO contains both a cyclization (Cy) domain and an oxidase (Ox) domain. This domain organization suggested the second module of FbnO would be involved in forming the thiazole rings of the FBNs. Consistent with this hypothesis, the A domain of this module was specific for L-Cys (Figure S4D).

FbnP has two modules, each containing an A domain. We were unable to generate a construct that enabled soluble production of the N-terminal module to be produced, but we could overproduce and purify both the full-length FbnP and the C-terminal module. The C-terminal A domain was specific for L-Ala, while the full-length FbnP activated both L-Ala and Gly (Figure S4E, S4F). The presence of a C-terminal epimerase (E) domain suggests the stereochemistry of this residue in the FBNs will be D-Ala. These data are consistent with FbnP functioning immediately after FbnO to form the central region of the FBNs.

FbnD was the only A domain-containing module remaining. Of the 20 proteinogenic amino acids tested, FbnD was found to weakly activate only Gly. The BGC codes for a homolog of aspartate α-decarboxylase (FbnA) that converts L-Asp to β-Ala, and the FBNs contain β-Ala moieties. Based on these observations, it is reasonable to propose that FbnD activates β-Ala. ATP/PP_i assays using Gly, L-Ala, D-Ala, β-Ala, and β-Ala-Gly determined the A domain was specific for β-Ala activation (Figure S4G).

Based on the combined ATP/PP_i assay results and our structural analysis, it was reasonable to place FbnQ, which lacks an A domain, as the module between FbnP and the terminal module, FbnH. The structures of the FBNs suggest the penultimate NRPS module would introduce Gly. Since the N-terminal module of FbnP is the only one specific for Gly, we propose that it loads Gly onto the T domain of FbnQ in trans (Figure 3A).

FbnD is an unusual NRPS.

The distinction between FBN A and B is the presence of a β-Ala or L-Ala linking the Cdua and Ahpa portions of the siderophores (Figure 1). As discussed below, the Ahpa moiety is likely to be derived from β-Ala and malonyl-CoA loaded onto FbnD and FbnN, respectively. The functions of all the other modules are accounted for, so the introduction of the linking β-Ala or L-Ala must come from an unusual mechanism. β-Ala may be introduced by an iterative function of FbnD. Briefly, β-Ala would be activated and tethered onto the T domain of FbnD, forming a β-Ala-S-FbnD intermediate. The A domain of this module would then work a second time to form a β-Ala-AMP intermediate, but since there is only a single T domain, amide bond formation would occur when the amino group of the thioesterfied β-Ala attacks the activated carbonyl of the β-Ala-AMP, thereby forming β-Ala-β-Ala-S-FbnD (Figure 3B). Our proposed mechanism is analogous to streptothricin biosynthesis wherein a β-lysyl polymer is formed by an iterative module²³, with the exception that two distinct A domains are used; one forms the thioesterified β-lysyl unit and a second activates the remaining β-lysyl-AMP intermediates for capture by the growing thioesterified β-lysyl polymer. Here, we propose a single A domain will perform both activities. This is a reasonable hypothesis based on the observations that a single NRPS module can have both the thioesterified amino acid and an amino acyl-AMP intermediate present simultaneously^{24, 25}. However, at this time we cannot eliminate the possibility that a second FbnD module provides the second β-Ala in trans.

The formation of the L-Ala bridge observed in FBN B is more challenging to address. We tested three hypotheses. First, we assessed whether FbnD was able to activate L-Ala after forming a β-Ala-S-FbnD intermediate. This was tested by first forming the β-Ala-S-FbnD intermediate then assaying whether the A domain gained the ability to activate L-Ala. The results from this experiment were negative, with L-Ala activation by β-Ala-S-FbnD not being statistically different than the control reactions (Figure S5). Next, we investigated whether β-Ala-S-FbnD was able to capture L-Ala from either L-Ala-AMP or L-Ala thioesterified to a T domain from either FbnO or FbnP (Figure 3C), the two modules that activate L-Ala. To test this we formed β-Ala-S-FbnD, added [³H]-L-Ala, ATP, and either apo- or holo-FbnO or FbnP, and investigated whether [³H]-L-Ala-β-Ala-S-FbnD was formed by scintillation counting the FbnD band excised from an SDS-PAGE gel. No radiolabel was detected on β-Ala-S-FbnD in either experiment nor in control experiments where FbnO and FbnP were absent. The lack of detectable radiolabel on β-Ala-S-FbnD in the control experiments also eliminated the possibility that L-Ala was activated by FbnD at some low level and then tethered to the β-Ala-S-FbnD. These series of negative results suggest the relationship between FbnD, FbnO, and FbnP will require further investigations to decipher how FBN A and B are assembled.

Substrate specificities of the PKSs FbnF and FbnN.

The PKSs were incubated with [¹⁴C-C2]-malonyl-CoA or [¹⁴C-C2]-(R,S)-methylmalonyl-CoA and subjected to SDS-PAGE followed by phosphorimaging to check for [¹⁴C]-labeling. Both FbnF and FbnN were radiolabeled when incubated with [¹⁴C-C2]-malonyl-CoA but not when they were incubated with [¹⁴C-C2]-(R,S)-methylmalonyl-CoA, indicating that malonyl-CoA is the preferred substrate of both (Figure S6).

Given the gene organization in the BGC, we predict FbnF would function after FbnG and FbnN would function after FbnD (Figure 3A). However, it is unclear how the available enzymology would account for the quaternary carbons, C-34 and C-44, or the lack of hydroxyl groups on C-35 and C-45, which would be expected due to the presence of ketoreductase (KR) domains in FbnN and FbnF. Additionally, assuming canonical NRPS-PKS enzymology, the number of backbone carbons in both Cdua and Ahpa, assembled in part by the PKSs, is one less than expected.

FbnE is a putative flavin-dependent monooxygenase.

Taken together, the aforementioned data suggest the FBNs undergo two Favorskii-like rearrangements to excise two carbons, C-37 and C-54, from the backbone as carboxylic acids. We propose in each case, the rearrangement is catalyzed by the putative flavin-dependent monooxygenase (FMO), FbnE, stimulated by a double oxidation of the α-carbons C-34 and C-44 from the acyl groups introduced by FbnF and FbnN (Figure 4A, 4B). Currently, it is unclear what role the KR domains play since both the solved FBN structures and our proposed mechanism for backbone contraction by Favorskii-like rearrangements are incompatible with established KR function. To reveal the origins of the excised carbons, we attempted ¹³C-acetate feeding studies with C-1, C-2, and both carbons labeled; however, the addition of acetate to the growth media dramatically decreased FBN production and, as a result, the degree of ¹³C incorporation was too low to be interpretable. Repeated attempts to find the appropriate time for ¹³C-acetate addition were unsuccessful, with disruption of FBN production occurring whenever acetate was added to the growth medium.

Figure 4. Proposed Favorskii rearrangement mechanisms of FBNs.

Favorskii rearrangements are catalyzed by flavin monooxygenase FbnE when intermediates are thioesterified to FbnF (A) or FbnN (B). R₁ is conditional upon the presence of the β-carbon (shown in parentheses as 0 or 1): if 1, R₁ = H (FBN A); if 0, R₁ = CH₃ (FBN B).

Favorskii-like rearrangements are commonly observed in dinoflagellate natural products, such as the brevetoxins, and structural evidence of similar enzymology is seen in fungal aspyrones^{26, 27}. In bacteria, there are two known FMOs involved in Favorskii-like rearrangements of polyketides, EncM and AmbI from enterocin and ambruticin biosynthesis, respectively^28–30. Additionally, the structure of myxoprincomide from Myxococcus xanthus DK1622 suggested to us the FMO domain associated with MXAN_3779 is likely to catalyze a Favorskii-like rearrangement³¹.

FbnE is a homolog of the MXAN_3779 FMO domain (43% identical), yet it is not homologous to either EncM or AmbI, which in turn are also not homologous to each other. One key difference in the reactions catalyzed by these enzymes is the fate of the excised backbone carbon. In enterocin, it is retained via intra-molecular esterification, while in ambruticin and myxoprincomide, it is lost via decarboxylation. Based on the FBN structures, we propose the reactions catalyzed by FbnE retain the backbone carbon as a carboxylic acid thereby creating the unusual quaternary carbons C-34 and C-44 that are likely to be essential for iron chelation (Figure 4A, 4B).

Formation of L-DOPA.

We established FbnH is selective for L-DOPA (Figure S4B). We hypothesized that L-DOPA is made from L-Tyr by FbnM prior to loading onto the megasynthase. FbnM is weakly homologous to SfmD (28% identical), a member of a recently discovered group of heme-containing peroxidases that catalyze the conversion of 3-methyltyrosine to 3-hydroxy-5-methyltyrosine in the tetrahydroisoquinoline family of natural products³². Additionally, FbnM contains the HXXXC motif found in SfmD thought to be required for heme binding. Initial attempts to overproduce soluble FbnM in E. coli were unsuccessful. We noticed FbnM shows homology to only the C-termini half of SfmD. We hypothesized that the gene immediately upstream of fbnM may code for a protein that complexes with FbnM based on its genomic location and our finding of FbnL/FbnM homolog fusions in WP_096621997.1 (e.g. from Microchaete diplosiphon).

FbnL and FbnM were found to be soluble when the encoding genes were co-expressed and the proteins co-purified (Figure 5A). The UV-Vis spectra of the FbnL/M pair is consistent with other heme-containing enzymes (Figure 5B)³³. Additionally, the presence of both genes on a single plasmid resulted in production of a brown pigment that was not present when the genes were expressed independently (Figure S7). Connor et al. reported dark brown pigmentation when overproducing Orf13, an unrelated heme peroxidase confirmed to catalyze the ortho-hydroxylation of L-Tyr to L-DOPA³⁴. It is the subsequent polymerization of L-DOPA, through a dopaquinone intermediate, that produces the pigment³⁵. Thus, it is reasonable that one role of FbnL/M is to hydroxylate L-Tyr, using H₂O₂ as the oxidant, thereby forming L-DOPA, which is loaded onto the NRPS FbnH.

Figure 5. FbnL/M is a heme peroxidase that converts L-Tyr to L-DOPA.

A) SDS-PAGE/Coomassie Blue staining of co-overproduced and Ni-NTA-purified FbnL/M. Imidazole concentrations used for protein elution are noted above each lane. B) UV-vis spectrum of 10 μg of purified FbnL/M complex in 120 μL of buffer with (solid line) or without (dashed line) the addition of 100 μM DTT. The soret band typical of heme B shifts from 404 nm to 420 nm signifying the ferric to ferrous iron conversion. C) Detection of L-DOPA in culture supernatants. Samples were induced with 100 μM IPTG where indicated. The averages of three independent reactions with standard deviations are shown. D) Three representative HPLC chromatograms of reactions (top to bottom): no hydrogen peroxide added, L-DOPA added rather than L-Tyr, and L-Tyr plus hydrogen peroxide added.

We tested this hypothesis using in vivo and in vitro approaches. We employed a colorimetric assay based on an L-DOPA dioxygenase from Mirabilis jalapa (mjDOD)³⁶ to determine whether fbnLM-expressing E. coli were exporting excess L-DOPA. MjDOD converts L-DOPA to betalamic acid, which spontaneously reacts with amino acids to form yellow pigmented betaxanthins. A greater betaxanthin signal was detected when E. coli BL21(DE3) contained the fbnLM overexpression vector compared to the empty vector control (Figure 5C). The L-DOPA concentration also increased when fbnLM expression was induced with IPTG. We also detected L-DOPA formation from L-Tyr by purified FbnL/M using HPLC (Figure 5D). The low level of turnover detected is not unexpected based on the well-established understanding of how heme peroxidases self-inactivate during catalysis with H₂O₂³⁷. The detection of L-DOPA in these experiments confirm the heme peroxidase FbnL/M can convert L-Tyr to L-DOPA and is the likely source of L-DOPA for incorporation into the FBNs by FbnH.

Proposed biosynthetic mechanism of the Dmaq heterocycle.

To form the Dmaq moiety, FbnL/M hydroxylates L-Tyr (1) forming L-DOPA (2). The L-DOPA is loaded onto the NRPS FbnH (3) and condensed with the upstream backbone of FBN (4) donated by FbnQ (Figure 6). To set up the aza-annulation, the FBN backbone must be reductively released from FbnH to form the semi-aldehyde (5) by the reductase domain of FbnH. An amino group is then transferred to the carbonyl carbon by FbnJ, a homolog of glutamate-1-semialdehyde amino transferases (HemL superfamily, COG0001) to form 6. It has been proposed that Dmaq formation in the anachelins occurs similarly to cyclo-DOPA, i.e. through a tyrosinase-catalyzed oxidation of either L-Tyr or L-DOPA to dopaquinone, which enables the aza-annulation³⁸. However, a tyrosinase is not encoded in the genome of A. fabrum C58, thus the ring closure to form FBN must be accomplished by other means. We propose the FbnL/M pair performs a second function by catalyzing the oxidation of the catechol to a quinone (7). In the following non-enzymatic steps, as have been previously proposed³⁸, an intra-molecular Michael addition yields 8, which tautomerizes resulting in the aza-annulation product (9). This is followed by double N-methylation by FbnI (pfam 13649, class I SAM-dependent methyltransferase) to form the mature Dmaq moiety (10), although it is possible this step precedes quinone formation (Figure 6).

Figure 6. Proposed mechanism for the biosynthesis of the Dmaq moiety.

Fabrubactins and anachelins belong to a large siderophore family.

The Dmaq moiety has only been observed in the anachelins and the FBNs. The biosynthetic formation of Dmaq is defined by the presence of FbnH-M homologs, with the occasional absence of an FbnK homolog (the MbtH-like protein). The organisms encoding homologs of this proposed Dmaq biosynthetic “cassette” are widespread in the Rhizobium/Agrobacterium group, α-proteobacteria, and cyanobacteria (Figure 7). To identify BGCs likely to be involved in the production of the Dmaq family of siderophores, we focused our initial attention on finding FbnL and FbnM homologs since these are the most distinct proteins in FBN biosynthesis. We also searched for FbnH-K homologs, NRPS and PKS enzymology, and proteins involved in siderophore transport (e.g. TonB, outer-membrane siderophore transporters). We did not fully analyze partial BGCs from incompletely assembled (meta)genomes; their includion would add at least double the number of BGCs of interest.

Figure 7. BGCs for the production of Dmaq-containing siderophores.

Schematic representations of the BGCs proposed to be involved in the production of Dmaq-containing siderophores. Arrow colors represent genes coding for the following: Green (NPRS or PKS enzymology), Blue (Favorskii-like rearrangement FMOs), Purple (L-DOPA-recognizing NRPS involved in Dmaq formation); Purple with Red asterisk (L-DOPA-recognizing NRPS with additional domains prior to reductase domain); Red (Dmaq-forming enzymes); Orange (ORFs coding for other functions). ^a, representative of BGCs in A. fabrum (strains 12D13, 1D132, ATCC 31749, ARqua1, and Arqua), A. tumefaciens (strains 15955, 15-174, CFBP5877, CFBP5499, LBA4404, F4), Rhizobium species. (NFIX01, NFIX02, UGM030330-04); ^b, representative of BGCs in A. vitas (strains SZ1, AB3, AB4, 15-172, 1D1609), A. tumefaciens strains 15-174 and 1D1609, R. rhizosphaerae MH17). Identifications of the locus tags for the 5’ and 3’ genes for each BGC are listed in Table S5.

From this analysis we identified 40 BGCs, in three groups, that we predict to be involved in the production of a Dmaq-containing siderophore (Figure 7, Table S5). The largest group (Group I in Figure 7) includes the FBN BGC, many members of the Rhizobium/Agrobacterium group, and other α-proteobacteria. Based on the similarity of these BGCs, we anticipate all will produce FBN analogs. Furthermore, homology between NRPS components provided clear indications of where insertions or deletions of modules has occurred. Group II includes three cyanobacteria BGCs, one of which is the putative anachelin BGC from Anabaena cylindrica PCC7122¹¹. These do not show the same similarity in NRPS/PKS components shared in Group I and only Nostoc sp. ‘Lobaria pulmonaria’ codes for a homolog of FbnE; thus, these will likely share the Dmaq moiety, but will be structurally distinct from each other. Group III BGCs have the necessary components for generating a Dmaq moiety, but the NRPSs associated with L-DOPA activation and reductive release of the siderophore backbone are split onto two distinct proteins. Interestingly, five of the six members of this group have a significant modification to the NRPS component for Dmaq formation. These five have the reductase domain on a separate polypeptide than the C-A-T domains, and these proteins contain an N-terminal non-heme β-hydroxylase domain (CmlA_N, pfam18456), a central UlaG-like domain (COG2220), and TubC-N-terminal NRPS docking domain (pfam18563). This domain organization suggests the Dmaq moiety itself will be structurally distinct, potentially with a β-hydroxylated L-DOPA incorporated, resulting in a hydroxyl group on the C4 of the Dmaq moiety. The sixth member of this Dmaq siderophore Group III is found in a Planctomcetaceae isolate and while it has a split FbnH homolog, it lacks the additional domains seen in the other five members of the group and likely will form a standard Dmaq moiety.

CONCLUSIONS

We have solved the structures of two siderophores from A. fabrum C58, FBN A and B, and characterized the specificities of the core enzymes of the NRPS/PKS megasynthase previously linked to the production of these molecules⁹. Taken together, our data suggest the involvement of unusual enzymology in the assembly of the FBNs. We identified an unusual NRPS module, FbnD, that first activates and tethers β-Ala to its T domain, and then captures a second molecule of β-Ala from itself (FBN A) or L-Ala from some other source (FBN B) to form the bridge between CduA and Ahpa components of the FBNs. We determined the FBN siderophores contain an unusual C-terminal Dmaq moeity previously only seen in the anachelin siderophores^{16, 17}. We have provided biochemical evidence that a two-protein heme peroxidase (FbnL/M) catalyzes the conversion of L-Tyr to L-DOPA and propose this protein complex also catalyzes the formation of a dopaquinone intermediate the enables aza-annulation to form one of the rings of Dmaq. Due to its high affinity for iron, Dmaq, when synthetically conjugated to a polyethylene glycol, has been shown to bind surface-adhered metal-oxides and display protease-resistant properties³⁹. This application has the potential to attach a desired natural product onto a metal-coated surface by using a combinatorial biosynthesis approach to selectively add Dmaq to the molecule. A better understanding of the Dmaq-producing suite of biosynthetic enzymes is needed to facilitate such combinatorial approaches.

The structures of the FBNs suggest the involvement of a Favorskii-like rearrangement-catalyzing FMO, FbnE, distinct from the other two known bacterial FMOs that carry out similar chemistry^{28, 29}. Favorskii rearrangements in bacterial NRPS- and PKS-made natural products are rare and little is known about the substrate selectivity of the FMOs. A deeper understanding of these enzymes could allow for unprecedented backbone-carbon manipulations of nonribosomal peptides or polyketides. We propose FbnE is responsible for the formation of the two quaternary carbons, C-34 and C-44 (Figure 4), that make the FBNs unique among bacterial polyketide-associated natural products. We have identified related hybrid NRPS/PKS-encoding BGCs in multiple rhizobacterial species that we propose also make FBN analogs containing the Dmaq moiety and quaternary backbone carbons (Figure 7). Finally, we provide bioinformatic evidence that the enzymology for Dmaq formation is found in many bacteria, suggesting there are still many members of this siderophore family yet to be discovered.

METHODS

Details of Experimental Methods are provided in the Supporting Information.