Introduction
Terrequinone A (Compound 1) represents a unique member of a family of bisindolylbenzoquinones commonly known as asterriquinones (Fig. 1). Since the initial discovery of cochliodinol (Compound 2)1, an antifungal agent2 from various Chaetomium species3, several new asterriquinones, including asterriquinone CT5 (Compound 3) and asterriquinone B1 (Compound 4), have been isolated from various species of Aspergillus terreus4, 5, Humicola and Botryotrichum6. All share a common dihydroxybenzoquinone core and vary mainly in the pattern of prenylation on the indole substituents. Like terrequinone A, most isolated asterriquinones are cytotoxic compounds that have been shown to intercalate genomic DNA, thus predisposing tumor cells to apoptosis7. Structure-activity studies on the asterriquinones have demonstrated that the hydroxyls of the symmetric dihydroxybenzoquinone core are important for antitumorigenic activity8. However, terrequinone A is the only asterriquinone asymmetrical in its quinone core, bearing an isopentenyl moiety in place of one of the hydroxyls.
Figure 1: Structures of representative asterriquinones.
Compound 1, terrequinone A; Compound 2, cochliodinol; Compound 3, asterriquinone CT5; Compound 4, asterriquinone B1.
Full size image (39 KB)Although terrequinone A was originally isolated from Aspergillus terreus9, a gene cluster for this molecule was recently discovered in Aspergillus nidulans through a genetic screen involving LaeA, a nuclear methyltransferase that acts as a global regulator of natural product gene expression in various Aspergillus species10, 11. By monitoring genetic loci that are downregulated in LaeA deletion strains or upregulated in overexpression strains, five contiguous genes, tdiA–tdiE, were found to be involved in secondary metabolite production. This study constituted the first identification of a biosynthetic pathway for this class of fungal toxins. Bioinformatic analysis of this cluster predicts that tdiA encodes a single-module nonribosomal peptide synthetase (NRPS), tdiB an indole prenyltransferase, tdiC an oxidoreductase, tdiD a pyridoxal-5'-phosphate (PLP; Compound 5)-dependent aminotransferase and tdiE a gene of unknown function with weak homology to S-adenosyl-L-methionine (SAM; Compound 6)–dependent methyltransferases11. Given the predicted functions of the proteins in the cluster, a route to terrequinone A was proposed (Scheme 1).
Scheme 1: Overall reaction pathway to terrequinone A.
Compound 9, IPA; Compound 14, didemethylasterriquinone D; Compound 1, terrequinone A.
Full size image (42 KB)Unique in this pathway is the presence of a single-module NRPS, TdiA. NRPSs are multifunctional enzymes consisting of semiautonomous domains that synthesize a myriad of secondary metabolites12. Using an assembly line logic comprising multiple modules, these enzymes use a thiotemplated mechanism13 to activate, tether and modify amino acid monomers, sequentially elongating the peptide chain and finally releasing the complete peptide. Absent from the TdiA scaffold is the traditional condensation (C) domain, which is responsible for formation of an amide bond between amino acids loaded on sequential modules14, 15. Rather, TdiA contains three individual domains responsible for adenylation, thiolation and thioester cleavage. TdiA's adenylation (A) domain is responsible for recognition of substrate and its subsequent activation as an acyl-O-AMP16; the substrate is then loaded onto the phosphopantetheine arm of the cognate thiolation (T) domain to form an acylthioester17, 18. The single-module TdiA ends with a thioesterase (TE) domain. Such TE domains traditionally catalyze release of the final product by either hydrolysis to the free acid or cyclization to an amide or ester12, 19. Notably, there are no amide, ester or free carboxylic acid functionalities present in terrequinone A, implying a new mechanism of release from this NRPS.
Another notable feature of this pathway involves the prenyltransferase TdiB. The prenylated indole group is a relatively widespread motif throughout many families of natural products, all of which are thought to be derived biosynthetically from L-tryptophan (Compound 7)20. The site of attachment of the prenyl group varies widely, with prenylation observed at all sites around the indole ring. Furthermore, the prenyl group may be tethered through its C1 position or its C3 position, representing nucleophilic attack of the indole on either end of the allyl cation-like transition state of the activated prenyl group21. Biosynthetically, prenylated indoles are formed by prenyltransferases (otherwise known as dimethylallyl transferases). These enzymes are known to use dimethylallyl diphosphate (DMAPP; Compound 8) as the prenyl source and often, although not always, contain an (N/D)DXXD motif as the signature for diphosphate binding. They may be divided into two functional classifications: so-called 'regular' prenyltransferases catalyze attacks at the C1 position of DMAPP, whereas 'reverse' prenyltransferases facilitate attachment of the dimethylallyl group through its C3 (ref. 22). Thus, the role of the prenyltransferase is to bind and activate DMAPP and to enable directed capture of the allyl cation at either the C1 or C3 position of DMAPP. Given the presence of two prenyl groups with different connectivities in the terrequinone A scaffold, we find it intriguing that the Tdi cluster contains only a single gene encoding a prenyltransferase, tdiB.
In this study, we have expressed tdiA, tdiB, tdiC, tdiD and tdiE in E. coli, overproduced and purified the encoded proteins and reconstituted the full terrequinone A biosynthetic pathway in vitro. The biosynthetic route to terrequinone A shows unusual logic, and its component enzymes show many unique functionalities.
Results
Cloning and expression of Tdi proteins
We amplified the five genes from the terrequinone A biosynthetic cluster from cDNA derived from Aspergillus nidulans strains A4 or FGSC A26 and cloned the products into C-terminal His6-tagged vectors. After sequencing the five genes, we determined that although bioinformatics correctly predicted the sequences of tdiB and tdiC, there were differences in the introns predicted for tdiA, tdiD and tdiE (Supplementary Fig. 1a online). Additionally, we determined that the reported genome sequence spanning tdiD contains an extra base, C464. Expression in E. coli at 30 °C or 15 °C with isopropyl-
-D-thiogalactoside (IPTG) induction yielded 1.6 mg l-
1 TdiA, 2.3 mg l-
1 TdiB, 7.5 mg l-
1 TdiC, 6.3 mg l-
1 TdiD and 16.8 mg l-
1 TdiE. We purified all proteins to homogeneity using nickel-affinity chromatography and, in the case of TdiA, gel filtration chromatography (Supplementary Fig. 1b).
Characterization of TdiD
The first step in the proposed biosynthetic pathway for terrequinone A is formation of indole pyruvic acid (IPA; Compound 9) from L-tryptophan using the aminotransferase TdiD. PLP-dependent aminotransferases typically contain a covalent PLP cofactor bound as a Schiff base to a conserved lysine in the protein active site. Upon binding of an amine-containing substrate, transimination occurs, yielding an intermediate that ultimately undergoes hydrolysis to pyridoxamine-5'-phosphate (PMP; Compound 10), and a keto functionality is installed on the substrate23 (Fig. 2a). Although purified TdiD lacked a bound cofactor, titration experiments demonstrated that it was, in fact, able to bind PLP, although weakly. Whereas free PLP has a 
max at 389 nm, the enzyme-bound PLP showed a max at 416 nm. TdiD showed linear binding of PLP, as demonstrated by an increase in absorbance at 416 nm, past 10 equivalents (data not shown).
Figure 2: Formation of IPA from L-tryptophan by TdiD.
(a) First half of the reaction cycle for aminotransferases (AMT). (b) Time course, evolving as denoted by the arrows, for conversion of PLP (389 nm) to PMP (323 nm) during the conversion of L-tryptophan to IPA, with 30 s between consecutive spectra.
Full size image (60 KB)With the ability of TdiD to bind PLP confirmed, we sought to test whether L-tryptophan could be converted to IPA. Addition of L-tryptophan to reactions containing TdiD and PLP yielded a decrease in absorbance at 389 nm and a corresponding increase in absorbance at 323 nm, indicating conversion of PLP to PMP (Fig. 2b). PMP formation was dependent on TdiD (Supplementary Fig. 2a online) and L-tryptophan (Supplementary Fig. 2b). This conversion is indicative of aminotransferase activity and thus IPA formation. To verify this, we analyzed reactions by HPLC. Over time, L-tryptophan was consumed, and a new peak that coeluted with authentic IPA evolved (Supplementary Fig. 3 online).
In PLP-dependent aminotransferases, there is generally a second substrate that is responsible for regenerating PLP after PMP formation. This substrate is typically a keto acid that acts as an amine acceptor. We tested three typical 
-keto acids (pyruvate (Compound 11), -ketoglutarate (Compound 12), and -keto-
-(methylthio)butyrate (Compound 13)) for their ability to regenerate PLP, as assayed by either an increase in the rate of IPA formation or a decrease in the amount of PMP formed. Although none of these compounds acted as a substrate for TdiD, this did not preclude determination of Michaelis-Menten kinetic parameters for the first half-reaction. Owing to the poor affinity of TdiD toward its PLP and PMP cofactor, we measured multiple turnovers due to diffusion of used PMP out of the TdiD active site and diffusion of new PLP into the active site. By measuring the conversion of free PLP to PMP, we determined that TdiD had a KM of 198 
26 
M and a kcat of 0.19 0.02 min-
1 for L-tryptophan.
Characterization of TdiA
Transformation of free L-tryptophan into IPA by TdiD suggested to us that perhaps IPA would act as a substrate for activation and loading by the A domain of the TdiA NRPS. Although amino acids are the typical substrates for NRPS adenylation domains, recently -keto acids have been demonstrated to be competent substrates of A domains in the cereulide biosynthetic system24. As in the cereulide system, the conserved aspartate side chain that normally charge-pairs with the NH3+ of the traditional amino acid substrate has been substituted for a hydrophobic residue, alanine, in TdiA (Fig. 3a). To address the substrate specificity of TdiA, we performed an ATP-PPi exchange assay on both IPA and L-tryptophan. Although we did not detect any reversible aminoacyl-O-AMP formation in the presence of L-tryptophan, TdiA successfully activated IPA with a kcat of 103 4 min-
1 and a KM of 1.2 0.3 M (Fig. 3b). The extremely low KM of TdiA for IPA is probably advantageous, owing to the instability of IPA in aqueous conditions. Such an affinity may allow immediate transfer of IPA from TdiD to TdiA.
Figure 3: Characterization of TdiA.
(a) Domain organization for TdiA. Inset table shows the ten amino acids conferring substrate specificity based on residue positions within PheA46 (shown in blue) for TdiA, four L-tryptophan–activating A domains and four -keto acid–activating A domains. Numbers in parentheses refer to the module number in multimodular NRPSs. (b) Michaelis-Menten plot for activation of IPA by TdiA. (c) HPLC traces (280 nm) showing products resulting from incubation of IPA with TdiA. Peaks corresponding to IPA appear at Rt 34.8 min and didemethylasterriquinone D at Rt 40.4 min. (d) Mass spectrum (ES-
) of didemethylasterriquinone D product resulting from TdiA incubation.
With the substrate specificity of TdiA determined, we assessed the full turnover of IPA to the putative bisindolebenzoquinone product, didemethylasterriquinone D (Compound 14). After post-translation priming with coenzyme A (CoA) and the phosphopantetheinyltransferase Sfp25, which were necessary to convert TdiA from the inactive apo to the active holo form (Supplementary Fig. 4 online), TdiA was able to convert IPA into a new product (Fig. 3c) that had a mass of 369.1 [M-H]
(Fig. 3d). This mass is consistent with didemethylasterriquinone D, and we definitively confirmed the structure to be that of didemethylasterriquinone D by NMR (see Methods). We measured the rate of formation of product Compound 14 from IPA to be 0.46 0.01 min-
1.
We attempted to form a diaminobenzoquinone by incubating TdiA with IPA imine generated in situ from coincubation of L-tryptophan and the amino acid oxidase RebO26. Under these conditions, only the previously observed dihydroxybenzoquinone core Compound 14 formed (data not shown), probably arising from hydrolysis of the IPA imine to the ketone before A domain activation. We presume the inability to form the diaminobenzoquinone stems from the selectivity of the A domain for the -keto acid rather than the -imino acid.
The symmetric connectivity of the two IPA molecules incorporated into the final product is thought to arise by head-to-tail dual Claisen condensations facilitated by the TE domain found at the C terminus of TdiA (Scheme 2). Indeed, when the active site Ser774 contained within the conserved GXSXGG motif was mutated to alanine, TdiA was no longer capable of generating didemethylasterriquinone D from IPA (Supplementary Fig. 5 online). To our knowledge, this is the first instance in which a TE domain has been found to catalyze carbon-carbon bond formation.
Reconstitution of terrequinone A biosynthesis
In order to investigate the final stages of terrequinone A biosynthesis, we probed the ability of TdiB, TdiC and TdiE to turn over didemethylasterriquinone D. Given the homology of TdiC to zinc- and NADH (Compound 15)-dependent quinone oxidoreductases, we included zinc sulfate and NADH in the reaction buffer. To enable TdiB-mediated prenyltransferase activity, we also included DMAPP, magnesium chloride and TCEP, and to facilitate the postulated methyltransferase activity of TdiE, we included S-adenosylmethionine (SAM). Under these conditions, didemethylasterriquinone D was cleanly converted to terrequinone A (Fig. 4a), as confirmed by NMR and mass spectroscopy (see Methods).
Figure 4: Characterization of TdiB/TdiC/TdiE trienzyme system.
HPLC traces (280 nm) showing products resulting from incubation of didemethylasterriquinone D with TdiB (5 M), TdiC (5 M) and/or TdiE (5 M) in the presence of DMAPP (5 mM), magnesium chloride (5 mM), zinc sulfate (200 M), NADH (5 mM), SAM (250 M), TCEP (2.5 mM) and BSA (1 mg ml-
1) in 75 mM Tris (pH 8.0) and 10% (vol/vol) DMSO. (a) Traces resulting from different TdiB/TdiC/TdiE enzyme combinations. (b) Traces resulting from omission of various cofactors. Peaks corresponding to starting material Compound 14 appear at Rt 11.4 min, O-isopentenyldidemethylasterriquinone D at Rt 11.7 min, ochrindole D at Rt 13.5 min and terrequinone A at Rt 14.3 min.
Terrequinone A formation was absolutely dependent on the presence of DMAPP and NADH (Fig. 4b); however, when we substituted NADPH (Compound 16) for NADH, we observed an identical product profile (data not shown). We included both divalent zinc and magnesium, respectively, in the original reaction mixture to allow for activity of the oxidoreductase TdiC and the prenyltransferase TdiB. In the presence of magnesium, zinc was not required for turnover, but we observed a reduction in product formation when we omitted both salts from the reaction mixture. This result suggests that magnesium can substitute for zinc in the TdiC active site and that zinc can substitute for magnesium to a certain extent. The weak magnesium dependence presumably derives from TdiB activity, in line with observations of other reported prenyltransferases in which mild rate enhancements up to threefold have been observed in the presence of millimolar concentrations of magnesium or calcium ions27, 28, 29. The use of magnesium chloride and calcium chloride at 5 mM concentrations gave identical product profiles (data not shown); hence, we were able to use these salts interchangeably in assays of the TdiB/TdiC/TdiE enzyme system. Notably, despite the homology of TdiE to SAM-dependent methyltransferases, TdiE is not a methyltransferase, as SAM was not required for terrequinone A formation by the TdiB/TdiC/TdiE system (Fig. 4b). We were not entirely surprised by this, for two reasons: TdiE lacks the typical SAM binding motifs30; and no orphan methyl groups are found in the final natural product scaffold.
The greatest yield of terrequinone A required the presence of all three enzymes: TdiB, TdiC and TdiE (Fig. 4a). None of these in isolation catalyzed turnover of didemethylasterriquinone D. Similarly, the combination of TdiE with either of the other two enzymes gave no turnover at all. TdiB and TdiC in combination gave some terrequinone A, but product formation was vastly reduced relative to the three-enzyme system.
Over the time course of the TdiB/TdiC/TdiE reaction, we observed an intermediate (at retention time (Rt) 13.5 min) forming and then disappearing from the assay mixture, with apparent conversion to terrequinone A (Fig. 5a). This species proved to be isolable and was identified as compound Compound 17 by NMR and mass spectroscopy. This compound corresponds to ochrindole D, which had previously been isolated from the sclerotia of Aspergillus ochraceus31, further demonstrating that these prenylated bisindolylquinone natural products are widespread within Aspergilli. Ochrindole D (Compound 17) could in turn be converted to terrequinone A by the action of TdiB alone; TdiC and TdiE were not required for this second prenylation reaction (Fig. 5b).
Figure 5: Turnover of intermediates on route to terrequinone A.
(a) HPLC traces (280 nm) showing a time course for incubation with TdiB, TdiC and TdiE. (b) Products resulting from incubation of ochrindole D (Compound 17) (Rt 14 min) with TdiB, TdiC and/or TdiE. (c) Time course for incubation with TdiB and TdiC. (d) Products resulting from incubation of O-isopentenyldidemethylasterriquinone D (Compound 18) (Rt 12.2 min) with TdiB, TdiC and/or TdiE. In a and c, peaks corresponding to starting material Compound 14 appear at Rt 11.4 min, Compound 18 at Rt 11.7 min, Compound 17 at Rt 13.5 min and Compound 1 at Rt 14.3 min. In b and d, peaks corresponding to Compound 17 appear at Rt 14.0 min, Compound 18 at Rt 12.2 min and Compound 1 at Rt 15.0 min. In d, Compound 17 impurity (Rt 14 min) is converted to Compound 1 (Rt 15 min) in presence of TdiB (indicated by asterisks).
Full size image (74 KB)In the absence of TdiE, bisindolylquinone 14 was converted to a mixture of products by TdiB and TdiC (Fig. 5c). The major two peaks corresponded to terrequinone A (Rt 14.3 min) and O-isopentenyldidemethylasterriquinone D (Compound 18) (Rt 11.7 min), as identified by NMR and mass spectroscopy. After isolating Compound 18, we demonstrated that this compound was not a substrate for any combination of TdiB, TdiC and TdiE (Fig. 5d). Compound Compound 18 therefore represents a dead-end shunt product that cannot be converted to terrequinone A.
Overall, the TdiB/TdiC/TdiE enzyme system robustly converts didemethylasterriquinone D to terrequinone A, a transformation involving net double prenylation at two distinct sites of the parent scaffold. This activity is dependent on the presence of DMAPP and NAD(P)H and is boosted by divalent metal ions such as magnesium, calcium and/or zinc. In the absence of TdiE, product is formed, but significant amounts of a shunt metabolite (Compound 8) also form by an off-pathway irreversible O-prenylation event (Scheme 3).
Scheme 3: Conversion of didemethylasterriquinone D (14) to terrequinone A (1) by TdiB, TdiC and TdiE.
Mass spectra of known intermediates are shown (Compound 22, hydrodidemethylasterriquinone D; Compound 18, O-isopentenyldidemethylasterriquinone D; Compound 17, ochrindole D; Compound 1, terrequinone A).
Full size image (96 KB)Discussion
Terrequinone A joins a growing family of bisindole alkaloids isolated from a variety of marine and terrestrial sources with a wide spectrum of pharmacological activities. This class of molecules includes the indolocarbazoles rebeccamycin (Compound 19)32, 33, 34 and staurosporine (Compound 20)35, 36, 37, the related bacterial pigment violacein (Compound 21)38, diketopiperazine-containing fellutarines39, 40, piperazine-containing hamacanthins39, 40, 41, 42 and dragmacidins39, 40, amide-containing coscinamides39, imidizole- or oxozole-containing topsentins39, 41, 42 and nortopsentins39, imidazolinone-containing rhopaladins39, 40 and pyrimidine-containing hyrtinadines43. Of these, only the gene clusters and biochemical characterization of rebeccamycin, staurosporine32, 33, 34, 36, 37 and violacein38 have previously been reported. In all three cases, the bisindole alkaloid is assembled by enzymatic oxidation of L-tryptophan to its imine, followed by oxidative dimerization to a core that is elaborated by even further oxidation.
The five-gene terrequinone A locus was recently identified based on the fact that it is controlled by the A. nidulans global secondary metabolite regulator LaeA10, 11. Here we present in vitro characterization of the five fungal proteins (TdiA-TdiE) purified from E. coli. Among items of note are (i) the head-to-tail dimerization of tethered -ketoacyl-NRPS intermediates on TdiA to set up the tetrasubstituted benzoquinone nucleus and (ii) bisprenylation by a single enzyme, TdiB. Terrequinone A, and presumably all the related asterriquinones, represent a new biosynthetic logic for nonoxidative dimerization of L-tryptophan-derived monomers to produce bisindole alkaloids.
The terrequinone A pathway was predicted to start with TdiD acting as an L-tryptophan transaminase. Indeed, although TdiD purified as the apoprotein from E. coli, it was PLP dependent in its activity in converting L-tryptophan to Compound 9. Because the PLP was weakly bound, multiple turnovers occurred without the addition of an -keto acid cosubstrate and allowed determination of the kinetic parameters.
The use of a PLP-dependent transaminase as the first oxidative step in the Tdi pathway is a meaningful divergence point for the flux of L-tryptophan to Compound 1 relative to the flux of L-tryptophan to rebeccamycin, staurosporine and violacein (Fig. 6), which use a flavoprotein oxidase26, 32, 38. In the latter case, the flavoproteins RebO, StaO and VioA yield the IPA imine as the initial oxidation product. Isomerization to the eneamine tautomer slows hydrolysis, and one of the -nitrogens of the two starting L-tryptophan molecules is incorporated into the dimerized scaffold32. By contrast, both amino nitrogens are removed by TdiD, leaving the keto and enol forms of IPA for subsequent reaction to terrequinone A; the -amino groups of L-tryptophan are not present in the quinone product. In fact, we have demonstrated that the IPA imine is an incompetent substrate for TdiA, and in situ generation of the IPA imine using RebO simply yielded Compound 14 when coincubated with TdiA, presumably owing to stalling of biosynthesis until hydrolysis to the ketone occurs for further processing to the bisindole alkaloid.
Figure 6: Biochemically characterized natural products derived from L-tryptophan dimerization.
Compound 19, staurosporine; Compound 20, rebeccamycin; Compound 21, violacein; Compound 1, terrequinone A. Atoms derived from individual L-tryptophan monomers are depicted in blue and red.
Full size image (34 KB)The dimerization strategy also differs markedly between the Tdi pathway and that used in the biosynthesis of rebeccamycin, staurosporine and violacein. In the latter biosynthetic manifolds, the enzyme-mediated dimerizations occur oxidatively in solution, requiring hemoprotein oxidases as catalysts to produce five-membered pyrrole ring–containing dimers (Fig. 6)32, 35, 38, 44. Oxidative homodimerization of the IPA imine proceeds at the free-acid level without any participation of NRPS catalytic logic or machinery. By contrast, the terrequinone A dimerization is nonoxidative, with no hemoproteins required. Formation of the six-membered dihydroxyquinone framework requires the NRPS module TdiA, acting on covalently tethered IPA-thioesters (Scheme 2).
TdiA is a lone three-domain fungal NRPS module set within this five-protein pathway, with each of its three domains of unique interest. To achieve its active form, the A-T-TE tridomain protein generated from E. coli must be posttranslationally primed with phosphopantetheine on the apo T domain. This may be biochemically accomplished in such a heterologous system by the catalytic agency of purified B. subtilis Sfp phosphopantetheinyltransferase25. The A domain is different from most adenylation domains in that it lacks the conserved active site aspartic acid whose -carboxylate side chain normally charge pairs with the -amino group of an amino acid substrate16. Indeed, the A domain is unable to activate L-tryptophan but instead activates the -keto acid IPA as the acyl-O-AMP and then tethers it in a thioester linkage to the terminal thiol of the T domain's phosphopantetheinyl arm. This is the second such example of an adenylation domain that uses an NRPS -keto acid, following our recent report on keto acid activation and tethering in cereulide biosynthesis24. In line with the amino acid code observed in this cereulide A domain, TdiA has an alanine residue in place of the aspartic acid of amino acid–activating A domains (Fig. 3a).
The lack of dimerization activity observed for the S774A mutant of TdiA validates the role of the TE domain in facilitating coupling of the IPA monomers. Transfer of the activated IPA moiety from the T domain to the active site Ser774 of the TE domain would form the IPA-O-TE acyl enzyme intermediate19. This ester is probably sequestered from hydrolysis, enabling the adjacent A-T didomain pair to reload with another IPA acyl thioester. At this point, we propose that the two acyl groups in cis undergo directed head-to-tail condensation, with the C carbanion of one IPA enolate moiety executing nucleophilic attack on the activated carbonyl group of the adjacent tethered IPA acyl group (Scheme 2). It is not yet clear if the direction of condensation is retrograde or anterograde. Based on the assumption that the thioester carbonyl (as the more thermodynamically activated carbonyl) behaves as the electrophile, the upstream IPA moiety would be transferred to form a linear tethered dimer in the TE active site. A second, now intramolecular, attack of an IPA enolate on the ester carbonyl will complete the cyclization and release of the dihydroxyquinone Compound 14 from the TE active site (Scheme 2).
This one NRPS module in TdiA thus activates two keto acids, tethers them sequentially and holds them in cis, one on the T domain as a thioester and one on the TE domain as an oxoester. The TE domain then directs a double Claisen condensation to generate two carbon-carbon bonds, fashion the quinone nucleus and catalyze the sequential disconnection of both covalent attachments to the protein. This represents a new disconnection mechanism for an assembly line chain-terminating TE domain. It is probable that all asterriquinones are generated by such NRPS-mediated dimerization in a deceptively simple three-domain, one-module NRPS protein. To our knowledge, this is the first example of a TE domain that is capable of catalyzing carbon-carbon bond formation.
TdiC, a NADH-dependent oxidoreductase that aligns with known zinc- and NADH-dependent quinone reductases, acts on the nascent bisindolylquinone scaffold, didemethylasterriquinone D. TdiC-mediated hydride addition to the tetrasubstituted quinone would create the hydroquinone oxidation state (hydrodidemethylasterriquinone D; Compound 22) (Scheme 3). This reduction is required to reverse the polarity of the electrophilic quinone to the nucleophilic hydroquinone Compound 22 that acts as a carbanion equivalent for the first prenylation event at C2 of the original quinone nucleus. This occurs by attack on C1 of the allylic cation-like transition state derived from DMAPP in the active site of the prenyltransferase TdiB. Elimination of water would result in net replacement of a hydroxyl with a prenyl at C2 of the quinone ring. This monoprenylated quinone Compound 17 can be detected and isolated in HPLC assays as an intermediate that builds up and then goes on to the final product Compound 1.
In the absence of TdiE, a second monoprenylated species Compound 18 builds up in the reaction mixture. Compound Compound 18 is prenylated instead on the oxygen at C2 of the dihydroxyquinone. This is an irreversible O-alkylation that is off pathway and competes with the C2 carbon prenylation. This competition between on-pathway C2 prenylation and off-pathway O-prenylation is where the protein TdiE acts, shunting the flux of the reaction down the productive pathway by an as yet unknown mechanism. TdiE could be a chaperone or partner protein, presenting the bisindolylhydroquinone Compound 22 to the prenyltransferase TdiB, but that will be the subject of future investigations. We have demonstrated that TdiE does not act as a methyltransferase, to which it has weak homology, because SAM is not a required cofactor in the production of terrequinone A. This is not entirely surprising, given the absence of any orphan methyl groups in the natural product Compound 1. TdiE is reminiscent of VioE in the bisindole alkaloid violacein pathway, in that VioE is required for shuttling a reactive intermediate toward an on-pathway product rather than a dead-end shunt product38.
After isolation of intermediate Compound 17, it can be converted to the final doubly prenylated natural product Compound 1 by the action of TdiB and DMAPP alone. Neither TdiC nor TdiE has any detectable role in this second prenylation step. The monoprenyl benzoquinone nucleus of Compound 17 is again electrophilic and is not a suitable partner for reaction with another molecule of DMAPP. Instead, the indole ring, a known nucleophilic site for enzymatic prenylations in other natural products, acts as the reaction partner. In this case, nucleophilic attack occurs via the indole C2, with DMAPP captured this time at C3 rather than C1, yielding terrequinone A via a reverse prenylation reaction. Thus, TdiB can carry out two consecutive prenyl transfers to the bisindolylquinone scaffold of Compound 14: the first prenyl moiety affixes to the hydroquinone nucleus and the second to one of the indole rings directly. Furthermore, the versatility of TdiB is emphasized by the observation that DMAPP can be captured regioselectively at C1 in the first prenylation and then at C3 in the second prenylation. How the single enzyme TdiB can mediate these two distinct prenylations is worthy of future structural and mechanistic study; in particular, the facial selectivity of attack on DMAPP and the concerted or stepwise nature of these processes are of interest.
This is the first case of a prenyltransferase acting iteratively on a natural product scaffold. This is in contrast to the A. fumigatus pathway to fumigaclavine C (Compound 23) in which two prenylations, one with C1 connectivity29 and the other with reverse C3 connectivity45, require two separate prenyltransferases. Although we do not think it surprising that a single prenyltransferase could be responsible for two prenylations in other fully symmetrical asterriquinones with identical prenyl connectivities, the marked asymmetry of Compound 1 emphasizes the versatility of TdiB.
Comparison of the RebD/VioB-mediated oxidative dimerization route for two IPA imines with the NRPS-mediated nonoxidative dimerization of IPA-S-enzyme intermediates by TdiA uncovers distinct ring sizes and nitrogen contents in the coupled products as well as diverse orientations of the monomers. The head-to-head heme protein–mediated coupling of the Reb and Vio pathways probably involves prior one-electron oxidation of each of the IPA eneamine molecules, followed by C radical coupling. One of the two nitrogens subsequently gets eliminated as NH3, while the other acts as a nucleophile in a C-N bond-forming step to give the pyrrole ring. The head-to-tail pathway of the TdiA-mediated coupling gives a six-membered (rather than five-membered) ring system. As proposed above, this probably occurs by tandem enolate-mediated Claisen condensations on the T- and TE domain-tethered IPA monomers. The TdiA-mediated coupling is nonoxidative, in contrast to the oxidative RebD/VioB couplings, and the subsequent TdiC step is reductive. This reflects nature's chemical versatility in its use of distinct enzymes and coenzymes to control fates and redox states in the dimerization of L-tryptophan to different natural product scaffolds.
In summary, the Tdi pathway is an intersection of NRPS and isoprenoid biosynthetic machinery, where TdiA and TdiB are noteworthy catalysts. The bisindolylbenzoquinone framework is constructed by IPA dimerization on a single NRPS module (TdiA), whereas distinct tandem prenylation regiochemistries are catalyzed by TdiB. In characterizing the biosynthetic pathway to terrequinone A, this work sets the stage for future mechanistic work to uncover more details about these enzymes.
Methods
General materials.
Bacterial strains, plasmids, materials and instrumentation are described in Supplementary Methods online.
Cloning and expression of TdiA-TdiE.
All fungal DNA isolation and manipulation, as well as cloning and expression of TdiA-TdiE, are described in Supplementary Methods.
Compound synthesis and characterization.
Preparation of dimethylallyl diphosphate, as well as isolation of Compound 14, Compound 1, Compound 17 and Compound 18 from TdiA, TdiB, TdiC and TdiE incubations, are described in Supplementary Methods.
Characterization of TdiD.
Titration reactions measuring PLP binding by TdiD contained 40 M TdiD, 25 mM HEPES (pH 7.75) and 150 mM NaCl. PLP was titrated in 0.1 equivalents at a time, and the change of absorbance at 416 nm, which is indicative of the PLP-enzyme complex, was monitored by UV-visible spectrophotometry.
Reactions to determine the KM of TdiD toward L-tryptophan contained 25 mM HEPES (pH 7.75), 150 mM NaCl, 250 M PLP, 2 M TdiD and various concentrations of L-tryptophan ranging from 1–500 M. The increase in absorbance at 323 nm, which is indicative of PMP formation, was monitored continuously for 20 min by UV-visible spectrophotometry.
Reactions to determine the kcat of TdiD toward L-tryptophan contained 25 mM HEPES (pH 7.75), 150 mM NaCl, 200 M PLP, 10 M TdiD and various concentrations of L-tryptophan ranging from 50–1,000 M. The decrease in absorbance at 389 nm, which is indicative of PLP consumption, was monitored continuously for 12 min by UV-visible spectrophotometry (
389 = 5,618 M-
1 cm-
1).
Reactions to determine possible -keto acid substrates for regenerating PLP from PMP after reaction with L-tryptophan contained 25 mM HEPES (pH 7.75), 150 mM NaCl, 250 M PLP, 500 M L-tryptophan, 30 M TdiD and 500 M -keto acid (pyruvate, -ketoglutarate and -keto--(methylthio)butyrate). At various time points ranging between 10 and 180 min, 10 l was quenched into 100 l MeOH and centrifuged at 16,000g to pellet the protein. The supernatant containing the product was analyzed by HPLC on a Phenomenex 250 
4.6 mm C18 5
Luna column using a gradient of 0%–50% acetonitrile over 20 min starting in 0.1% TFA in H2O; absorbance at 280 nm was monitored.
ATP-PPi exchange assay for TdiA A domain substrate specificity.
Reactions (100 l) contained 75 mM Tris (pH 7.5), 10 mM MgCl2, 5 mM DTT, 5 mM ATP, 1 mM sodium [32P]pyrophosphate (0.18 Ci) and 0.1 M TdiA, with substrate concentrations ranging from 0.1–500 M for IPA and 0.05–17 mM for L-tryptophan. Reactions were incubated at 25 °C for 5 min (IPA) or 15 min (L-tryptophan) and were quenched by addition of 500 l 1.6% (wt/vol) activated charcoal, 200 mM tetrasodium pyrophosphate and 3.5% (vol/vol) perchloric acid in water. The charcoal was pelleted by centrifugation and washed twice with 500 l 200 mM tetrasodium pyrophosphate and 3.5% perchloric acid in water. The radioactivity bound to the charcoal was then measured by liquid scintillation counting. Note that enzyme concentrations and reaction times were chosen such that ATP-PPi exchange remained under 10% of equilibrium levels.
Didemethylasterriquinone D formation by TdiA.
Reactions to measure product formation by TdiA contained 50 mM Tris (pH 8.0), 10 mM MgCl2, 200 M CoA, 5 M TdiA and 3 M Sfp. After a 45-min incubation at 25 °C to prime the T domain, the reaction was initiated by addition of 5 mM ATP and 200 M IPA. At desired time points ranging from 1–120 min, 100 l were quenched into 200 l MeOH and then centrifuged at 16,000g to pellet protein. The supernatant containing the product was analyzed by HPLC on a Phenomenex 250 4.6 mm C18 5 Luna column using a gradient of 0%–100% acetonitrile over 50 min starting in 0.1% TFA in H2O; absorbance at 280 nm was monitored.
Reaction of TdiA with imino-IPA.
Reactions to measure possible formation of a diaminobenzoquinone core contained 50 mM Tris (pH 8.0), 10 mM MgCl2, 200 M CoA, 5 M TdiA and 3 M Sfp. After a 45-min incubation at 25 °C to prime the T domain, the reaction was initiated by addition of 5 mM ATP, 500 M Trp, 5 M RebO and 5 mM NH4Cl. After incubation for 5 h at 25 °C, the 100-l reaction was quenched into 200 l MeOH and centrifuged at 16,000g to pellet protein. The supernatant containing the product, combined with a 50-l DMSO wash of the pellet, was analyzed by HPLC on a Higgins Analytical 50 4.6 mm C18 5 CLIPEUS column using a gradient of 0%–100% acetonitrile over 15 min starting in 0.1% TFA in H2O, monitoring absorbance at 280 nm.
Conversion of didemethylasterriquinone D into terrequinone A.
Reactions used to convert didemethylasterriquinone D into terrequinone A contained 100 M didemethylasterriquinone D, 500 M DMAPP, 5 mM MgCl2, 75 mM Tris (pH 8.0), 2.5 mM TCEP, 1 mg ml-
1 BSA, 200 M ZnSO4, 5 mM NADH, 250 M SAM, 2 M TdiB, 2 M TdiC, 2 M TdiE and 10% (vol/vol) DMSO in a 100-l volume. In order to test for necessary cofactors and proteins, reactions were repeated omitting DMAPP, MgCl2, ZnSO4, NADH, SAM, MgCl2 and ZnSO4, TdiB, TdiC, TdiE or combinations of the three enzymes. After 2 h, reactions were quenched with 200 l methanol and centrifuged at 16,000g to pellet protein. The supernatant containing the product combined with a 50-l DMSO wash of the pellet was analyzed by HPLC on a Higgins Analytical 50 4.6 mm C18 5 CLIPEUS column using a gradient of 0%–100% acetonitrile over 15 min starting in 0.1% TFA in H2O; absorbance at 280 nm was monitored.
Accession codes.
The correct cDNA sequences for tdiA–tdiE have been deposited in GenBank (tdiA: EF550581; tdiB: EF550582; tdiC: EF550583; tdiD: EF550584; tdiE: EF550585).
Note: Supplementary information and chemical compound information is available on the Nature Chemical Biology website.
Author contributions
C.J.B. and A.R.H.-J. contributed equally to this work.