Polyketides are a class of biologically active microbial and plant-derived metabolites that possess a high degree of structural and functional diversity and include many human therapeutics, among them anti-infective and anti-cancer drugs, growth promoters and anti-parasitic agents1. The macrolide antibiotics, characterized by a glycoside-linked macrolactone, constitute an important class of polyketides, including erythromycin and the natural ketolide anti-infective agent pikromycin. Here we describe new mechanistic details of macrolactone ring formation catalyzed by the pikromycin polyketide synthase thioesterase domain from Streptomyces venezuelae. A pentaketide phosphonate mimic of the final pikromycin linear chain-elongation intermediate was synthesized and shown to be an active site affinity label. The crystal structures of the affinity-labeled enzyme and of a 12-membered-ring macrolactone product complex suggest a mechanism for cyclization in which a hydrophilic barrier in the enzyme and structural restraints of the substrate induce a curled conformation to direct macrolactone ring formation.
The biosynthesis of complex polyketides is carried out by type I modular polyketide synthases (PKSs) through the stepwise condensation of simple carboxylic acid derivatives such as malonyl-CoA and methylmalonyl-CoA1. These modular PKSs comprise several large, multifunctional enzymes that catalyze the initiation, elongation, reduction and macrocyclization steps that ultimately give rise to the characteristic macrolactone scaffold typical of macrolide antibiotics such as erythromycin (Compound 1) and pikromycin (Compound 2; structures not shown). The pikromycin PKS shows the unique ability to efficiently generate two natural macrocyclic products, 10-deoxymethynolide (Compound 3) and narbonolide (Compound 4)2,3 (Scheme 1a). Macrolactonization of both is catalyzed by the terminal thioesterase (TE) domain of PikAIV4, suggesting that this enzyme accommodates substrates of varying lengths.
Scheme 1: Pikromycin biosynthesis and substrate mimics.
(a) Terminal steps of the pikromycin biosynthetic pathway in S. venezuelae ATCC 15439. Premature termination of polyketide biosynthesis after the steps catalyzed by PikAIII results in the generation of the 12-membered-ring macrolactone 10-deoxymethynolide (Compound 3), whereas continued elongation by PikAIV results in the biosynthesis of the 14-membered-ring macrolactone narbonolide (Compound 4). Further processing by the post-PKS tailoring enzymes DesVII, a glycosyltransferase, and PikC, a cytochrome P450 hydroxylase, completes the biosynthesis of the macrolide antibiotics methymycin and pikromycin, respectively3. (b) Hexaketide substrate mimics. (c) Diphenylphosphonate affinity labels. Ac, acetyl; Me, methyl; Ph, phenyl.
An understanding and expansion of the naturally relaxed substrate specificity of the pikromycin TE (Pik TE)4 may lead to the development of a general macrocyclization catalyst for use in producing new macrolide therapeutics. In addition to its natural substrates, Pik TE will hydrolyze a variety of simple diketide thioesters, with a slight preference for the 2-methyl-3-keto diketide5. Reaction with late-stage pikromycin chain elongation intermediates6,7 revealed substrate-dependent differences in enzyme-mediated cyclization as opposed to hydrolysis. Reaction of hexaketide thioester (Compound 5) with Pik TE provided efficient macrolactonization to 10-deoxymethynolide (Compound 3), whereas reaction of its C7-reduced analog (Compound 6) resulted in exclusive hydrolysis to acid Compound 77,8 (Scheme 1b). This suggests that, despite its natural substrate tolerance for chain length variation, Pik TE is sensitive to minor functional group modifications of its natural substrates.
Pik TE is related to 6-deoxyerythronolide B synthase (DEBS) TE9, the enzyme responsible for the macrolactonization of the terminal heptaketide chain-elongation intermediate in the biosynthesis of 6-deoxyerythronolide B (Compound 8, structure not shown), the aglycone precursor of erythromycin. Crystal structures of both enzymes9,10 show dimers of the common 
/
-hydrolase fold with a similar hydrophobic interface. A serine-histidine-aspartate catalytic triad is located at the center of a long open substrate channel that spans the width of the enzyme. In vivo, substrates are presented to PKS type I TE domains as thioesters tethered to the phosphopantetheinyl group of a preceding acyl carrier protein (ACP) domain. The N-terminal side of the substrate channel is presumed to be the entrance site and the C-terminal side, the exit site. Despite important insights from these structures, many questions remained regarding the molecular basis of Pik TE substrate selectivity, owing to the lack of ligands in the hydrophobic substrate channel.
In an accompanying article11, we described the design and synthesis of polyketide-based diphenylphosphonate affinity labels Compound 9 and Compound 10 (Scheme 1c) and demonstrated their ability to modify the catalytic serine of Pik TE11. These compounds, which mimic the C1–C6 segment of the pikromycin heptaketide intermediate, provided important new insights into the Pik TE catalytic mechanism. However, two key questions remain. How does Pik TE induce the ends of a natural linear substrate to come together and form, exclusively, a macrolactone product? Why does reduction of a substrate carbonyl to a hydroxyl lead to the linear product in preference to the macrocycle? To answer these questions, we reasoned that affinity labels of extended chain length could be useful probes of the mechanistic details involved in substrate selectivity, thereby providing insight into the basis for discrimination between the competing macrolactonization and hydrolytic reactions catalyzed by Pik TE.
We designed a pentaketide affinity label candidate (Compound 11) to mimic the C1–C9 segment of the pikromycin heptaketide intermediate (Scheme 1a,c). The C9-keto group of the heptaketide chain-elongation intermediate is functionally equivalent to the C7-ketone and C7-alcohol in hexaketide substrates Compound 5 and Compound 6, respectively (Scheme 1b). Therefore, we deemed that incorporation of the primary alcohol in pentaketide Compound 11 at the C9 position was important for revealing substrate-enzyme interactions that determine whether a substrate is cyclized or hydrolyzed.
Pentaketide Compound 11 was synthesized in seven steps from alcohol Compound 12 in a manner analogous to the synthesis of triketide phosphonate Compound 911 (Supplementary Scheme 1 online). The key steps were (i) a titanium-mediated aldol reaction and (ii) direct displacement of the thiazolidinethione chiral auxiliary to install the diphenylphosphonate group.
We examined the ability of pentaketide Compound 11 to inactivate Pik TE by incubating the enzyme with excess Compound 11 and following the time-dependent loss of activity. We assessed residual activity by measuring the rate of enzymatic hydrolysis of p-nitrophenylpropionate, which has been used previously to investigate the hydrolytic activity of thioesterases12,13. Pentaketide Compound 11 reduced the hydrolytic activity of Pik TE at an initial rate of 0.85% min-1 
0.08% min-1 (mean s.d.). A two-fold reduction in activity occurred within 50 min and nearly complete inactivation (
5% residual activity) after 6 h (Supplementary Fig. 1 online). The inactivation rate of pentaketide Compound 11 is intermediate between the rates observed under similar experimental conditions with triketide Compound 9, which gave a 50% reduction in activity after 8 h, and with reduced triketide Compound 10, which yielded complete inactivation after several minutes11.
The crystal structure of Pik TE modified by pentaketide Compound 11 was solved to a resolution of 1.95 Å (Fig. 1 and Supplementary Table 1 online). Initial electron density maps clearly showed modification of the catalytic Ser148 as well as an elongated density with the distal end curled back toward Ser148 (Supplementary Fig. 2 online). The proximal (P1–C3) end of the pentaketide adduct is essentially unchanged with respect to the shorter phosphonate adducts11. Accordingly, neither phenyl group is present, the phosphonate mimics the tetrahedral transition state through hydrogen bonds with His268 and with the NH group of Gly149 in the oxyanion hole, and a potential water of hydrolysis and a DMSO molecule are displaced.
Figure 1: Affinity label adduct and macrolactone product within the Pik TE active site.
Red, O; orange, P; blue, N; green, S; red spheres, water. (a) Refined 2Fo – Fc electron density contoured at 1 
around the modified serine-phosphopentaketide. (b) Stereodiagram of modified serine-phosphopentaketide detailing protein-substrate interactions. Phosphopentaketide Compound 11 main chain atoms are numbered beginning with the proximal phosphorus atom (P1). (c) Refined 2Fo – Fc electron density contoured at 1 around the 10-deoxymethynolide and DMSO molecules trapped within the active site of Pik TE. (d) Stereodiagram of 10-deoxymethynolide (10-DML) detailing product and DMSO interactions.
The serine-phosphopentaketide adduct mimics atoms C1–C9 of the heptaketide tetrahedral intermediate after reaction of Pik TE with the ACP heptaketide thioester. The acyl chain of the serine-phosphopentaketide is initially directed toward the exit site of the substrate channel; however, before exiting the channel, the chain curls back toward the active site, terminating with a hydrogen bond between the distal C9-hydroxyl group and the side chain hydroxyl of Thr77 (Fig. 1b). Density for the full length of pentaketide Compound 11 was observed in all maps examined. However, the distal end is less well ordered than the proximal end, presumably as a result of the lack of specific contacts with the protein as observed previously with the triketide mimics.
The region in which the acyl chain reverses direction occurs at the transition between the hydrophobic environment of the substrate channel and the exterior of the protein (Fig. 2a,b). At this point, two well-ordered water molecules coordinate the side chain of Gln183, the backbone carbonyl of Ala217 and the C5-hydroxyl group of the pentaketide Compound 11 (Fig. 1b). This region of the exit site forms a 'hydrophilic barrier' in which the hydrophilic protein surface and the bulk solvent environment direct the acyl chain back into the substrate channel toward the active site. The hydrophilic barrier is 'soft', as it clearly does not trap reaction products within the substrate channel.
Figure 2: Electrostatic surface representation of Pik TE substrate channel with affinity label and with macrolactone product.
(a) Cut-away view of ligand-free Pik TE showing electrostatic surface of the substrate channel, water network (blue spheres) and DMSO molecule trapped at the active site in the oxyanion hole. Exit side of the channel is labeled at the right. (b) Magnified view of phosphopentaketide Compound 11 in the hydrophobic chamber. (c) Magnified view of 10-deoxymethynolide and DMSO in the substrate channel. Electrostatic surfaces were calculated using APBS21.
The 1.8-Å crystal structure of Pik TE (S148A) in complex with its natural reaction product 10-deoxymethynolide Compound 3 (Supplementary Table 1) shows that the 12-membered-ring macrolactone occupies a hydrophobic cavity near the exit of the substrate channel (Figs. 1c,d and 2c and Supplementary Fig. 2). As with ligand-free Pik TE, a single DMSO molecule occupies the oxyanion hole11. The 10-deoxymethynolide product makes few direct contacts with the protein. The only hydrogen bonds are mediated through water and the bound DMSO molecule (Fig. 1d). The DMSO may have blocked 10-deoxymethynolide from fully penetrating the active site. The 10-deoxymethynolide molecule is approximately 5 Å removed from the catalytic triad, does not reach the most hydrophobic region of the channel, disrupts the hydrophilic-barrier water network that was observed in all other Pik TE structures11, and is rotated with respect the phosphopentaketide (Fig. 2b,c). Thus, the contacts of 10-deoxymethynolide are unlikely to emulate any that would be present in the acyl-serine hexa- or heptaketide intermediate; nevertheless the binding site illustrates the large volume of the substrate channel that can be sampled by ligands, as well as the conformation of 10-deoxymethynolide bound to the enzyme.
The Pik TE structure is essentially unchanged in the apo-, affinity-labeled and product-bound enzymes10,11 and is thus inconsistent with a proposed induced-fit mechanism for substrate recognition10. Without specific substrate contacts or enzyme conformational change, how is the terminal hydroxyl of the natural substrates positioned to attack C1 for cyclization? We note that the natural substrate conformations are restricted by resonance of the conjugated enones at C7–C9 in the hexaketide and C9–C11 in the heptaketide. Resonance is expected to enforce planarity on C6–C10 in the hexaketide and C8–C12 in the heptaketide. To determine whether the enforced planarity is compatible with the apparently fixed active site structure, we modeled two key acyl-serine intermediates.
The acyl-serine heptaketide was modeled directly from the pentaketide Compound 11 adduct by addition of four carbon atoms and replacement of the phosphonate with an ester (Fig. 3a). The acyl-serine hexaketide was modeled by moving the 10-deoxymethynolide molecule into the active site so that the analogous atoms overlaid those of the serine-phosphopentaketide and by replacing the 10-deoxymethynolide ester with serine ester plus hydroxyl (Fig. 3b). In both cases, all atoms fit in the substrate site without steric conflicts. The planarity resulting from the conjugated enone positions the terminal, attacking hydroxyl group in line with the serine ester. In addition, the enone carbonyl oxygen of both acyl intermediates can be anchored through a hydrogen bond to the side chain of Thr77 (Fig. 3). Thus, it appears that the curl of the acyl chain enforced by the hydrophilic barrier, the conformational restraints of the substrate, and a single anchor hydrogen bond are sufficient to place the distal hydroxyl group of the substrate within attacking distance of the acyl-serine ester, favoring macrolactonization over the competing hydrolysis reaction.
Figure 3: Model of acyl-serine hepta- and hexaketides in the active site shown from the top views (left; orientation is similar to that in Fig. 1b) and side views (right).
(a) Acyl-serine heptaketide. (b) Acyl-serine hexaketide. Important substrate-protein hydrogen bonds, as well as the interaction between the attacking, terminal hydroxyl and the C1 carbon, are indicated with dashes. Conjugated enone acyl-chain carbon atoms are colored yellow and non-conjugated carbons green.
Reduction of the enone keto group to a hydroxyl seems to eliminate any contribution to catalysis by substrate conformational restraints. Based on the serine-acyl hexaketide and heptaketide models, the active site could accommodate a 180° rotation about the respective bond, displacing the attacking hydroxyl group more than 6 Å, and increasing the likelihood that the serine ester will be cleaved by a competing hydrolysis reaction. This provides a rationale for the observation that the reduced hexaketide forms only the hydrolyzed product upon reaction with Pik TE7,8, and a 3:4 mixture of macrolactone and hydrolyzed, reduced heptaketide upon reaction with the native PikAIV module14. Thus, the difference in reactivity of the natural and reduced substrates may be due to entropy of the substrate, as opposed to enthalpy of protein-substrate interactions.
The Pik TE substrate channel is critical to the mechanism and selectivity of macrolactonization. The channel provides a hydrophobic environment for the similarly hydrophobic polyketide substrates. A spacious chamber, just to the exit side of the catalytic Ser148, is the most hydrophobic region of the channel and is the site of substrate binding (Fig. 2). The simplest and most compelling interpretation of the structures is that a 'hydrophilic barrier' at the distal end of the exit channel induces longer substrates to curl back toward the catalytic residue Ser148 (Fig. 2), enabling macrolactonization. The size, lack of hydrogen bonding groups and hydrophobic surface of this region of the channel are well suited to accommodate acyl chains of varying lengths and multiple conformations, but are suggestive of poor substrate selectivity. In vivo, Pik TE requires minimal specificity for substrate recognition because delivery of the substrate by a tethered ACP domain reduces the need for discrimination among a variety of potential substrates. The only likely competitor for the active site is a loaded but unextended proximal ACP. The long and negatively charged entrance side of the substrate channel should effectively exclude the negatively charged methylmalonyl ACP species from the active site. Additionally, the narrow substrate channel is not consistent with a proposal that glycosylated linear chain elongation intermediates are substrates for Pik TE15.
Our model of substrate positioning for the macrolactonization reaction has implications for efforts to engineer Pik TE to catalyze additional ring formation reactions. The catalytic assistance provided by the restricted substrate conformation and the lack of substrate-positioning hydrogen bonds indicate that a variety of macrolactones may be formed, provided that the substrates have a structural feature to steer the attacking hydroxyl group toward the acyl serine. Additionally, the position of 10-deoxymethynolide at the exit port indicates that the channel has space for even larger substrates and that the exit end of the channel could be engineered to accommodate different substituents.
Many other PKS pathways possess thioesterase domains that form macrolactone rings. A substrate channel punctuated by a hydrophilic barrier is a general means by which essentially linear, hydrophobic substrates can be induced to curl into a conformation suitable to form a macrocycle. Substrate conformational restraints from conjugated double bonds may assist cyclization by positioning the attacking hydroxyl group for macrolactone formation. Polyketides without conjugated systems or other conformational assistance may require additional substrate-enzyme contacts. For example, DEBS TE, the only other PKS TE domain of known structure, may use more hydrogen-bonded contacts, as predicted, in its more polar active site chamber9.
Our studies of the pentaketide Compound 11 affinity label have revealed an unexpected and unique mechanism that precisely positions the end of the polyketide chain for macrolactone formation by Pik TE. The crystal structure of the pentaketide adduct of Pik TE also provides a framework for protein engineering experiments aimed at expanding the repertoire of enzyme-catalyzed macrolactonization reactions. Finally, the work validates further the utility of diphenylphosphonate-based affinity labels to probe the mechanisms of PKS enzymes.
Methods
Chemistry.
General procedures, detailed experimental procedures and copies of 1H and 13C NMR spectra for all new compounds synthesized in Supplementary Scheme 1 are available in the Supplementary Methods online.
Production and purification of Pik TE and Pik TE (S148A).
The alanine mutant of Pik TE residue Ser148 was generated from the plasmid encoding wild-type Pik TE (pCAte2)7 through QuikChange site-directed mutagenesis (Stratagene) following the vendor's protocols. The expression and purification of wild-type Pik TE has been described previously7,11. Identical protocols were used for expression and purification of Pik TE (S148A).
Crystallization of Pik TE and Pik TE (S148A).
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 described11 by hanging-drop vapor diffusion. Crystallization buffers were supplemented with 5% DMSO to allow for addition of either diphenylphosphonate pentaketide (Compound 11) or 10-deoxymethynolide (Compound 3)6. The S148A Pik TE mutant was used for the 10-deoxymethynolide complex. Pik TE crystals were soaked at 20°C with 5 mM diphenylphosphonate pentaketide (Compound 11) for 72 h before harvesting, whereas Pik TE (S148A) crystals were equilibrated at 4 °C before being soaked for 24 h with 1 mM 10-deoxymethynolide.
Data collection and structure determination.
X-ray diffraction data were collected at the GM/CA beamline (ID-23D) at the Advanced Photon Source (APS, Argonne National Laboratory). Data for the pentaketide Compound 11 adduct and the 10-deoxymethynolide complex were processed to resolutions of 1.95 Å and 1.8 Å, respectively (Supplementary Table 1), using the HKL2000 package16. A dataset collected with wild-type Pik TE and 10-deoxymethynolide was collected at slightly lower resolution, with no differences in compound binding. The structures were solved by molecular replacement using previously published structures10,11; however, ordered-solvent structures were determined independently for the two structures. Model building was done with Coot17 and O18 and refinement with REFMAC518. Initial coordinates and crystallographic restraints for 10-deoxymethynolide and serine-pentaketide phosphonate were generated with the PRODRG server19. The 10-deoxymethynolide product is present in both monomers of the Pik TE dimer. We examined datasets for several crystals treated with pentaketide Compound 11, but in all cases found density for the product only in monomer A and not in monomer B. This is a fairly common finding in protein crystallography, even in cases such as the Pik TE crystal, in which the observation cannot be explained by differences in protein conformation or crystal lattice contacts20. Density for only a single conformer of the full-length pentaketide Compound 11 was observed in all maps examined. The atomic sites for the proximal end of the adduct were fully occupied, whereas those for the distal end were either partially occupied or subject to high thermal motion. No electron density was observed in any map for a different ordered position of the adduct.
Inactivation of Pik TE with diphenylphosphonate (Compound 11).
Assays of Pik TE inactivation by pentaketide Compound 11 were carried out as previously described11. Pik TE (20 
M) was incubated at 30 °C with 2 mM pentaketide Compound 11 in 50 mM HEPES buffer, pH 8.0, supplemented with 5% DMSO. Two additional reactions were prepared as controls and treated similarly. The first contained no enzyme and the second contained Pik TE (20 M) but no phosphonate compound. Over 6 h, 8-l aliquots were removed and diluted 25-fold into 50 mM HEPES, pH 8.0. The residual activity of the diluted enzyme mixture was then assessed by measuring the hydrolysis of p-nitrophenylpropionate at 30 °C and normalized to a non-inhibited control. The reactions were initiated by the addition of 1 mM p-nitrophenylpropionate and hydrolytic activity was followed by monitoring the release of the p-nitrophenolate ion at 400 nm using a Spectromax M5 multimode plate reader (Molecular Devices).
Modeling.
The natural heptaketide acyl enzyme was modeled directly from the Ser148 phosphonate pentaketide adduct. No adjustments to the protein were made because the protein structure is identical, within experimental error, in the apo enzyme, the triketide adduct, the reduced triketide adduct, the pentaketide adduct and the product complex. The serine phosphonate was converted to a serine ester; four carbon atoms and oxygen substituents were added to the end of the pentaketide chain, bond lengths were adjusted, and planarity was imposed on atoms affected by the resonant C=C and C=O double bonds. Initial coordinates, bond length and torsion parameters for the hexa- and heptaketide serine esters were generated with the PRODRG server19, and O18 was used for manual modeling of the adducts. Single bonds at the end of the chain were rotated into approximately staggered conformations to avoid collisions with protein atoms. The conformation of the proximal end of the modeled heptaketide acyl enzyme was not changed relative to the pentaketide Compound 11 adduct. The conformation of 10-deoxymethynolide was used as a basis for the natural hexaketide acyl enzyme model. The ester of 10-deoxymethynolide was overlaid with the Ser148 phosphonate of the pentaketide Compound 11 adduct. The bond forming the macrolactone was broken and single bonds at the end of the hexaketide were rotated as for the heptaketide acyl enzyme model.
Database accession numbers.
Coordinates and structure factors have been deposited in the protein data bank under PDB ID 1HFJ for the pentaketide Compound 11 adduct and 1HFK for the 10-deoxymethynolide complex.
Author Contributions
D.L.A., J.D.K. and J.L.S. were responsible for the Pik TE production, crystallography and structural analysis; J.W.G. and R.A.F. were responsible for the design and synthesis of the diphenylphosphonate pentaketide; and J.D.K. and D.H.S. were responsible for conducting the Pik TE inactivation experiments.
Note: Supplementary information is available on the Nature Chemical Biology website.
