Polyketides are a diverse class of natural products having important clinical properties, including antibiotic, immunosuppressive and anticancer activities. They are biosynthesized by polyketide synthases (PKSs), which are modular, multienzyme complexes that sequentially condense simple carboxylic acid derivatives1. The final reaction in many PKSs involves thioesterase-catalyzed cyclization of linear chain elongation intermediates. As the substrate in PKSs is presented by a tethered acyl carrier protein, introduction of substrate by diffusion is problematic, and no substrate-bound type I PKS domain structure has been reported so far. We describe the chemical synthesis of polyketide-based affinity labels that covalently modify the active site serine of excised pikromycin thioesterase from Streptomyces venezuelae. Crystal structures reported here of the affinity label–pikromycin thioesterase adducts provide important mechanistic insights. These results suggest that affinity labels can be valuable tools for understanding the mechanisms of individual steps within multifunctional PKSs and for directing rational engineering of PKS domains for combinatorial biosynthesis.
Recent advances in the metabolic engineering of PKSs have led to the development of new methods for producing polyketide natural products and analogs1. The pikromycin type I PKS system of S. venezuelae ATCC 15439 has been developed as an important model because it is able to biosynthesize macrolide antibiotics of two different ring sizes, methymycin (Compound 1) and pikromycin (Compound 2) (Scheme 1)2. Pikromycin thioesterase (Pik TE), the terminal catalytic domain of this system, is the source of this natural diversity owing to its unique ability to cyclize both linear hexaketide and linear heptaketide chain elongation intermediates to the 12- and 14-membered macrolactones 10-deoxymethynolide (Compound 3) and narbonolide (Compound 4), respectively. Further processing of 10-deoxymethynolide and narbonolide by the post-PKS tailoring enzymes DesVII (a glycosyltransferase) and PikC (a cytochrome P450 hydroxylase) completes the biosynthesis of methymycin and pikromycin.
Scheme 1: The pikromycin biosynthetic pathway in S. venezuelae ATCC 15439.
Numbers on structures denote atom positions referred to in the text.
Pik TE is well suited as a component of the combinatorial biosynthesis toolbox because it has inherent substrate tolerance. Despite this substrate tolerance, specificity studies of the isolated Pik TE domain suggest that it has some sensitivity toward seemingly innocuous substrate modifications. Reaction of Pik TE with Compound 5, a mimic of the hexaketide chain elongation intermediate, results in exclusive macrolactonization to 10-deoxymethynolide (Compound 3), whereas reaction with the C7-reduced analog Compound 6 results in hydrolysis to acid Compound 7 (Scheme 2)3,4. The PikAIV monomodule, which catalyzes chain extension and macrolactonization by the native TE domain, further demonstrates a modified reactivity profile toward these substrates. Reaction of hexaketide Compound 5 and methylmalonyl coenzyme A (CoA) with PikAIV yields both the nonextended 10-deoxymethynolide (Compound 3) and the extended narbonolide (Compound 4) macrolactones5. Notably, reaction of reduced hexaketide Compound 6 and methylmalonyl CoA with PikAIV gives a mixture of products resulting from hydrolysis (acid Compound 7), extension and lactonization (lactone Compound 8), and extension and macrolactonization (macrolactone Compound 9)5. Thus, Pik TE does not tolerate all minor substrate modifications, and unknown structural and mechanistic details determine whether a Pik TE substrate is cyclized or hydrolyzed.
Scheme 2: Enzymatic reactions of hexaketide substrate analogs with Pik TE and PikAIV.
Numbers on structures denote atom positions referred to in the text. Ac, acetyl.
Pik TE belongs to the 
,
-hydrolase class of serine hydrolases and has the characteristic Ser-His-Asp catalytic triad (Ser148, His268 and Asp176)6. In vivo, the polyketide substrate is bound as a thioester to the phosphopantetheinyl arm of an acyl carrier protein (ACP) and undergoes transesterification to the active site Ser148 of Pik TE. Intramolecular attack by the distal hydroxyl group of the substrate on the acyl-enzyme intermediate affords the macrolactone, with simultaneous release of the product from the enzyme. Pik TE has been the subject of site-directed mutagenesis and additional substrate specificity studies with simple diketide-based substrates6. The crystal structure of Pik TE (ref. 7) shows many similarities to the structure of the 6-deoxyerythronolide B synthase (DEBS) TE (ref. 8), the enzyme responsible for the macrolactonization of the DEBS-derived heptaketide chain elongation intermediate to 6-deoxyerythronolide B (Compound 10, structure not shown). These homologous enzymes share a common protein fold, hydrophobic dimer interface, catalytic triad and open substrate channel7.
To gain further, structure-based insight into the mechanisms of macrolactonization and hydrolysis by Pik TE and to guide rational enzyme engineering, we sought to develop substrate-based affinity labels. Many functionalities covalently modify serine hydrolases and proteases; however, we noted that the high chemical reactivities of these moieties seemed likely to be incompatible with a densely functionalized polyketide9. Peptidyl -aminoalkylphosphonates, particularly diphenylphosphonates, are well-characterized irreversible inhibitors of serine proteases10. Crystal structures of Cbz-(4-AmPh-Gly)P(OPh)2 bound to bovine trypsin have confirmed the inhibition mechanism of serine proteases by diphenylphosphonates: the active site serine attacks the electrophilic phosphorous to eliminate one phenoxy group, and aging of the complex results in hydrolysis of the second phenoxy group11. The resulting tetrahedral complex mimics the substrate tetrahedral intermediate, with one oxygen atom bound in the oxyanion hole11. We viewed the relative chemical stability of diphenylphosphonates as an asset for their incorporation into polyketide affinity labels that mimic the thioester of an ACP-bound intermediate.
We designed three triketide affinity-label candidates (Compound 11, Compound 12 and Compound 13) to mimic the C1–C6 segment of the pikromycin heptaketide chain elongation intermediate (Scheme 3a). Triketide Compound 11 and reduced triketide Compound 12 lack the C2 methyl group of the heptaketide; however, we did not expect this methyl group to contribute substantially to binding in the enzyme active site. Reduced triketide Compound 12 also mimics the C1–C6 segment of the DEBS heptaketide chain elongation intermediate, and its C3 alcohol is present in the pikromycin hexaketide substrate. Cyclic triketide Compound 13 mimics triketide Compound 11; although we expected Compound 13 to be less reactive than Compound 11 and Compound 12, we anticipated that Compound 11 or Compound 12 might cyclize in aqueous buffer during planned inhibition experiments.
Scheme 3: Structures of compounds used in the present study.
(a) Structures of the three candidate affinity labels. (b) Synthesis of triketide phosphonates Compound 11, Compound 12 and Compound 13. Reagents: a, CS2, KOH, H2O, 86%; b, n-BuLi, EtCOCl, 93%; c, EtCHO, TiCl4, NMP, (–)-sparteine, 98%; d, 2,6-lutidine, TESOTf, 95%; e, (PhO)2P(O)Me, LiHMDS, 90%; f, CSA, 87%; g, Zn(BH4)2, 83%; h, Me2C(OMe)2, p-TsOH, 98%; i, 2.5% aq. Na2CO3, 4%. Bn, benzyl; CSA, camphorsulfonic acid; HMDS, hexamethyldisilazide; NMP, N-methylpyrrolidone; Ph, phenyl; TES, triethylsilyl; Tf, trifluoromethanesulfonyl.
We synthesized triketide Compound 11 in six steps from D-phenylalaninol (Compound 14) (Scheme 3b) and generated thiazolidinethione Compound 16 according to the synthesis of the enantiomer of Compound 1612. The titanium-mediated aldol reaction of Compound 16, followed by protection, direct displacement of the chiral auxiliary and deprotection, afforded triketide Compound 11. We synthesized reduced triketide Compound 12 and cyclic triketide Compound 13 from Compound 11 in one step each. We used 13C NMR analysis of acetonide Compound 20 to confirm the stereochemistry in the directed reduction of Compound 1113.
We assessed the residual activity of Pik TE by monitoring the rate of enzymatic hydrolysis of p-nitrophenylpropionate (PNP-propionate)14 after 8 h of incubation with triketides Compound 11, Compound 12 and Compound 13 (2 mM). Both triketide Compound 11 and reduced triketide Compound 12 substantially reduced the hydrolytic activity of Pik TE, whereas cyclic triketide Compound 13 had only a minimal effect (
10% reduction in 8 h). Reduced triketide Compound 12 completely inactivated Pik TE within 15 min (Supplementary Fig. 1 online), whereas triketide Compound 11 afforded a two-fold reduction in activity. A two-fold loss of enzyme activity was achieved with an eight-fold reduction in the concentration of reduced triketide Compound 12 (250 
M) over the same time period (Supplementary Fig. 1). Covalent modification of Pik TE by reduced triketide Compound 12 was confirmed by LC-MS analysis of the Pik TE adduct fragments obtained by proteolytic digestion with trypsin (Supplementary Fig. 2 online).
We solved the crystal structures of the Pik TE adducts with triketide Compound 11 and reduced triketide Compound 12 to resolutions of 2.1 Å and 1.85 Å, respectively (Fig. 1 and Supplementary Table 1 online). We used crystallization conditions similar to those previously reported7, with the addition of 5% DMSO to facilitate introduction of the affinity label to the crystals. We also solved the structure of unmodified Pik TE crystallized under identical conditions to control for DMSO-induced changes to the active site (Fig. 1e and Supplementary Table 1). Initial electron density maps clearly indicate that both affinity labels covalently modified the active site Ser148 (Fig. 1a,c) in both the A and B monomers, albeit with lower occupancy in the B site. As expected, no phenoxy groups were observed in the electron density for either modified enzyme. The affinity labels are located within the substrate channel and are pointed toward the exit 'C side' of the channel (Fig. 1f,g)7. No DMSO-induced changes to the protein structure were detected.
Figure 1: Structure of affinity-labeled Pik TE (O, red; P, orange; N, blue; water, red spheres).
(a,b) 2Fo – Fc density (1
) for Pik TE modified with reduced triketide Compound 12 (a) and triketide Compound 11 (b). (c,d) Stereodiagram of reduced triketide Compound 12 (c) and triketide Compound 11 (d) with Pik TE, showing selected residues and the affinity label (C, yellow). (e) Fo – Fc density (3) for unmodified Pik TE fit with DMSO. DMSO and water of hydrolysis (blue) are displaced by an affinity label (yellow). (f,g) Pik TE surface with view of substrate channel. (f) Triketide Compound 11 viewed from the exit channel. (g) Perpendicular cutaway view of triketide Compound 11 in relation to the catalytic triad.
Although the two affinity labels differ at C3 (Scheme 3), with either a carbonyl or an alcohol at this position, they interact identically with the Pik TE active site (Fig. 1b,d). The phosphonate structures mimic the tetrahedral covalent intermediate of transesterification. In both structures, the phosphorus atom mimics the C1 atom of the heptaketide substrate, and the pro-R oxygen nearest the active site histidine substitutes for the sulfur atom of the thiolate leaving group. The pro-R oxygen displaces a potential water of hydrolysis observed in the unmodified Pik TE structure (Fig. 1e). The active site histidine is hydrogen bonded to the sulfur mimic in perfect position to donate a proton to the leaving group.
One important feature of esterases, amidases and peptidases is the 'oxyanion hole', which stabilizes the oxyanion intermediate through interaction with two backbone hydrogen bond donors. Typically, in ,-hydrolases, one donor is from the NH group immediately following the active site residue (Gly149 in Pik TE), and the second is from the loop connecting strand 3 and helix 3 (residues 75–83). In unmodified Pik TE, this loop is not close enough to the active site to serve as a hydrogen bond donor and does not move on addition of either affinity label. Though the pro-S phosphonate oxygen interacts with the Gly149 NH group as predicted7, no second hydrogen bond donor from the protein is in direct contact with the phosphonate. Rather, a water molecule bridges between the phosphonate and the backbone NH and side chain OH groups of Thr77 (Fig. 1b,d). The structures of affinity-labeled Pik TE suggest that the need to accommodate both ends of the polyketide during cyclization of the acyl-serine intermediate may preclude the formation of a classic oxyanion hole with two hydrogen bond donors.
The crystal structure of unmodified Pik TE has strong positive electron density in the oxyanion hole (Fig. 1e). This peak was best modeled as a DMSO molecule that forms a hydrogen bond with the Gly149 NH in the oxyanion hole. Formation of the phosphonate adduct evidently displaced DMSO from the oxyanion hole, as this density is not present in either structure of affinity-labeled Pik TE.
The affinity labels extend from the active site toward the exit channel but make only indirect, water-mediated contacts with the enzyme (Fig. 1b,d,f,g). This is not surprising, as Pik TE is able to accommodate several substrates for both cyclization and hydrolysis reactions3,4,6. The natural hexa- and heptaketide substrates differ at the end proximal to the thioester, and therefore tolerance in substrate recognition in this region is to be expected.
The overall protein architecture is unchanged with respect to the unmodified structure. Three reported Pik TE structures determined from crystals grown in various conditions, and at different buffer pHs (pH 7.6, 8.0 and 8.4), show that the diameter of the substrate channel increases with pH (ref. 7). Although the crystals we used were grown at pH 7.6, the crystallization conditions were otherwise similar to those used for the reported pH 8.0 structure (LiSO4 precipitant with added MgCl2). The structures reported here (both the affinity-labeled and unmodified enzyme) most closely resemble that solved at pH 8.0 with LiSO4 (ref. 7). Therefore, it seems that the variation observed in channel architecture, though perhaps important biologically, may be more a result of the crystal form than of the solution pH.
One of the challenges confronting the field of combinatorial biosynthesis is the need to understand mechanisms of polyketide biosynthesis. Pik TE (ref. 7), DEBS TE (ref. 8), DEBS ketoreductase 1 (KR1; with the NADPH cofactor bound)15 and the DEBS ketosynthase 5–acyl transferase 5 (KS5–AT5) didomain (with three linker domains)16 are the only type I PKS enzymes for which crystal structures have been reported. Molecular docking of enzyme-bound intermediates and products with Pik TE (ref. 7) and DEBS TE (ref. 8) has been used to propose models for their mechanisms of macrolactonization. Although structurally unrelated bacterial TEs17,18 and homologous human19,20, bovine21 and bacterial7,8,22,23 TEs have been crystallized, attempts to co-crystallize TEs with substrates and substrate analogs have been met with mixed success21,23,24. Given the challenges in obtaining crystal structures of substrates, intermediates and products bound to PKS domains, we were motivated to develop substrate-based affinity labels for structural and mechanistic studies.
Both triketide Compound 11 and reduced triketide Compound 12 irreversibly inactivate Pik TE; however, reduced triketide Compound 12 does so more effectively. As expected, cyclic triketide Compound 13 was less reactive with Pik TE, likely owing to the presence of a single electron-withdrawing phenoxy group and its structural dissimilarity to the natural substrates. The difference in reactivity of triketide Compound 11 and reduced triketide Compound 12 with Pik TE is noteworthy because their identical binding in the crystal structures offers no clues to the reactivity difference. One explanation for this observation is that the C3 alcohol of reduced triketide Compound 12 results in closer structural similarity to the hexaketide substrate of Pik TE (Scheme 1). Reaction of hexaketide Compound 5 and methylmalonyl CoA with PikAIV affords a 4:1 ratio of 10-deoxymethynolide (Compound 3) and narbonolide (Compound 4)5. This raises the possibility that the hexaketide chain elongation intermediate (with a C3 alcohol), not the heptaketide (with a C3 ketone), is the preferred substrate for Pik TE. Synthesis of extended-chain affinity labels that more closely mimic the structures of the hexa- and heptaketide substrates could clarify these differences in reactivity.
The structures of affinity-labeled Pik TE show an unusual architecture in the oxyanion hole, in which only one NH group directly donates a hydrogen bond to the oxyanion and the second hydrogen bond is mediated through water. Another feature of the active site is the lack of specific contacts between the active site and either affinity label. An 'induced fit' to close the large substrate-binding channel around the substrate has been predicted based on different channel sizes in three crystal forms7. However, binding of the affinity labels results in no change in the size of the substrate channel. The need for access by the distal hydroxyl group during macrolactonization, and the need to accommodate substrates that differ at the proximal end, may explain the lack of specific, direct contacts between substrate and enzyme. Another explanation is the minimal need for the TE domain to discriminate among candidate substrates, as all substrates are delivered by an ACP domain that is tethered to the TE domain.
Development of affinity labels for Pik TE represents a considerable advance in understanding the mechanism and specificity of this multifunctional enzyme. Although the short chain lengths of triketide Compound 11 and reduced triketide Compound 12 do not allow for a complete understanding of Pik TE macrolactonization, we gained important mechanistic insights into this process. Most importantly, the affinity-label adducts provide strong evidence that substrates are not recognized by specific hydrogen bonds or by a protein surface of complementary shape. Additionally, the oxyanion of the tetrahedral intermediate is stabilized by only one direct hydrogen bond from the protein. The use of substrate-based affinity labels also opens new avenues for exploration of other regions of the active site and other catalytic domains of PKSs. Our results suggest that substrate-based affinity labels can be used as valuable tools for understanding the mechanisms of PKSs and for future efforts to engineer PKS domains for creating new bioactive compounds.
Methods
Chemistry.
General procedures, detailed experimentals and copies of 1H and 13C NMR spectra for all new compounds synthesized in Scheme 3b are available in Supplementary Methods online.
Purification of Pik TE.
The production and purification of Pik TE are described in Supplementary Methods.
Inactivation of Pik TE.
Pik TE (20 M) was incubated at 30 °C with each of the triketide phosphonates Compound 11, Compound 12 and Compound 13 (2 mM each) in 50 mM HEPES buffer, pH 8.0, supplemented with 5% DMSO. As controls, two additional reactions were prepared and treated similarly. The first contained no enzyme and the second contained Pik TE (20 M) but no phosphonate compound. After 8 h, aliquots (8 l) of each reaction mixture were withdrawn and diluted 25-fold into 50 mM HEPES buffer, pH 8.0. We then assessed the residual activity of the diluted enzyme mixture by measuring the hydrolysis of PNP-propionate at 30 °C. We initiated the reactions by adding 1 mM PNP-propionate and followed hydrolytic activity by monitoring the release of the p-nitrophenolate ion at 400 nm using a Spectromax M5 multimode plate reader (Molecular Devices). Time course studies with reduced triketide Compound 12 were performed essentially as described above, using shorter preincubation times (2–30 min) and a range of affinity-label concentrations (0.25–2 mM).
Crystallization of Pik TE.
Protein was concentrated to >10 mg ml-1 and dialyzed against 10 mM HEPES, pH 7.0, 2 mM DTT before crystallization. Crystals were obtained as previously described7 by vapor diffusion in a hanging-drop tray at 20 °C. The well buffer contained 1.2–1.4 M Li2SO4, 100 mM HEPES, pH 7.6, 80 mM MgCl2, 2 mM DTT and 5% DMSO. Crystals were of space group P21212 (a = 107.7, b = 130.6, c = 56.2 Å). The solution surrounding the crystals was exchanged with buffer equivalent to the well buffer with the addition of 5 mM triketide Compound 11 or reduced triketide Compound 12; crystals with triketide Compound 11 were soaked for 84 h and those with reduced triketide Compound 12 were soaked for 72 h before being frozen in liquid N2.
Data collection and structure solution.
X-ray diffraction data were collected at the General Medicine and Cancer Institutes Collaborative Access Team (GM/CA CAT) beamline (ID-23D) at the Advanced Photon Source (Argonne National Labs). For triketide Compound 11, data to 2.1 Å resolution were processed using the HKL-2000 software package (HKL Research)25. Model phases were calculated using the previously reported structure of Pik TE (ref. 7). The structure was modeled using O26 and Coot27, and it was refined with REFMAC5 (ref. 28) to a final Rwork of 0.197, with an Rfree of 0.231 (Supplementary Table 1). Initial Fo – Fc difference maps for the triketide Compound 11 adduct had strong electron density continuous with Ser148, confirming covalent modification and indicating the positions of atoms P1–C5 of the phosphonate adduct for the A monomer and of atoms P1–C3 for the B monomer. In the final model, occupancy of the phosphonate adduct was modeled at 1.00 for the A monomer and at 0.66 for the B monomer. At this occupancy, the average B factors for the adducts (41.1 Å2) were comparable to the average side chain B factors for the protein (38.2 Å2). For reduced triketide Compound 12, data to 1.85 Å resolution were processed, and the structure was solved by difference Fourier analysis using phases from the refined structure of the triketide Compound 11 adduct of Pik TE. The reduced triketide Compound 12 structure was refined to a final Rwork of 0.190, with an Rfree of 0.238 (Supplementary Table 1). As with triketide Compound 11, lower occupancy was observed for the B monomer than for the A monomer. The final structure was modeled with an occupancy of 0.75 for the adduct attached to the A monomer and an occupancy of 0.50 for the B monomer. At these occupancies, the average B factors for the adducts are 42.9 Å2. Fo – Fc difference density in the initial maps showed positions P1–C4 for the A monomer and P1–C3 for the B monomer. The terminal atoms of reduced triketide Compound 12 appear to be more mobile than those in triketide Compound 11 and may be present in multiple conformations. Stereochemical restraints for both covalent serine phosphonate products were generated using the PRODRG server29. PyMOL was used to prepare all graphics (Fig. 1)30.
Database accession numbers.
Coordinates and structure factors have been deposited in the Protein Data Bank (PDB ID 2H7Y for the triketide adduct and 2H7X for the reduced triketide adduct). Structure of Pik TE is PDB ID 1MNA (ref. 7).
Author contributions
J.W.G. and R.A.F. were responsible for the design and synthesis of the affinity labels; J.D.K. and D.H.S. were responsible for conducting the Pik TE inactivation experiments; and D.L.A., J.D.K. and J.L.S. were responsible for Pik TE production, crystallography and structural analysis.
Note: Supplementary information is available on the Nature Chemical Biology website.
