Engineered Modular Polyketide Synthases Elucidate How Enoylreductases Collaborate with Downstream Acyl Carrier Proteins and Ketosynthases to Set Stereocenters

Abstract

To engineer modular polyketide synthases that reliably yield stereocomplex products, a detailed understanding of how their enzymatic domains collectively operate is essential. How their enoylreductase (ER) domains function to install either d-or l-oriented alkyl substituents remains largely mysterious. In this study, we investigated the stereoselectivity of ERs and how they work with other domains within tetraketide-heptaketide synthase platforms constructed from components of the pikromycin and rapamycin synthases. We first determined how ER and acyl carrier protein (ACP) domains associate by mutagenically probing the consensus solution predicted by AlphaFold using an in vitro tetraketide synthase platform. Since ERs were found to contact residues conserved in ACPs from diverse module types, we tested whether ERs operate when inserted into modules that only contain ketoreductase and dehydratase domains in several in vivo platforms. ER-inserted modules were observed to generate both unsaturated and saturated intermediates. Increased production of the saturated intermediate was realized when the downstream KS was also swapped for a KS from an ER-containing module. To access products with desired methyl group orientations, it was necessary to consider how the methyl groups interact with the KS of the module that installed them, as well as with the KS of the downstream module. Additionally, we interconverted d- and l-type ERs by swapping a 14-residue loop and propose the molecular basis for how ERs set stereocenters. This research provides new strategies for controlling the orientations of alkyl substituents in complex polyketides and enables access to stereochemical libraries of polyketide drug leads.

Graphical Abstract

INTRODUCTION

Modular polyketide synthases (PKSs) are multienzyme complexes that operate in assembly line fashion to produce bioactive polyketides such as the antibacterial pikromycin, the immunosuppressant rapamycin, and the insecticide spinosyn (Figure 1a).^1–4 Their ability to set stereocenters is of great interest to the synthetic and medicinal chemistry communities, yet harnessing them to reliably produce new molecules and medicines is proving to be a significant challenge.^5–7

Figure 1.

Modular polyketide synthases generate stereocomplex products. (a) Four polypeptides of the pikromycin PKS house 7 modules (updated definition), P1–P7, that biosynthesize the macrolactone narbonolide, a precursor of the macrolide antibiotic pikromycin. The immunosuppressant rapamycin and the insecticide spinosyn are generated by related PKSs. (b) The domains of P5, including P5ER, collaborate to stereoselectively install an α-methyl substituent. (c) How d-type ERs generate d-oriented α-substituents and l-type ERs generate l-oriented α-substituents is a long-standing mystery. (d) The domain composition of γ- and δ-modules shows ERs are located between the structural and catalytic subdomains of KR (KR_s and KR_c). Key: KS^Q, priming KS; KS, ketosynthase; AT, acyltransferase; KR⁰, reductase-inactive but epimerase-active KR; KR_s, KR structural subdomain; ER, enoylreductase; KR_c, KR catalytic subdomain; ACP, acyl carrier protein; TE, thioesterase; unlabeled circle, ACP; unlabeled semicircles, docking domain motifs; squiggle, phosphopantetheinyl arm.

The structure of a polyketide usually parallels the domain structure of its synthase. While this collinearity is suggestive that stereocenters can be controlled by swapping the appropriate ketoreductase (KR) and enoylreductase (ER) domains, recent insights into PKS logic have revealed more sophisticated collaborations between PKS domains than previously recognized (Figure 1b). Ketosynthase (KS) domains can inspect chemistries installed at the α- and β-positions by upstream enzymes, including acyltransferases (ATs), KRs, dehydratases (DHs), and ERs, from the same module (updated definition used),^8–12 as well as chemistries installed at the γ- and δ-positions by the upstream module.^10,13 Domains can also collaborate to control stereochemistry, as the site an acyl carrier protein (ACP) associates with a KR influences how its 18-Å phosphopantetheinyl arm presents the β-keto group to the active site and, thus, the stereoselectivity of the reduction reaction.¹⁴

How ERs stereoselectively reduce α-substituted, trans-α/β-unsaturated polyketide intermediates is a long-standing mystery in PKS enzymology (Figure 1c).^15–17 An l-type ER that yields an l-oriented α-substituent is usually identifiable on a sequence level through the presence of a signature tyrosine 8 residues downstream of a conserved NFRD motif. This tyrosine could serve as the general acid to protonate the re face of the enolate intermediate, yet there is no residue specific to d-type ERs that yield d-oriented α-substituents. In studies of a related enoyl-thioester reductase that protonates the re face of the enolate, stereoselectivity could be reversed by mutating the tyrosine equivalent to the signature tyrosine of l-type ERs to phenylalanine and a nearby residue to a glutamate (PDBs: 4WAS, wild-type enzyme bound to crotonyl-CoA and NADP⁺; 4W99, double mutant with reversed stereoselectivity).¹⁸ The optimal pH for this double mutant is 5.0, as the acidic form of this residue protonates the si face of the enolate.

Surprisingly few studies have investigated the ER domains of modular PKSs. Their stereoselectivities were examined using triketide lactone synthases in which the offloading module was engineered to contain the KR, DH, and ER domains from either the fifth module of the erythromycin synthase (E5) or the 14th module of the rapamycin synthase (R14).^16,17 Mutations to the signature tyrosine of E5ER converted it from l-type to d-type; however, introduction of a tyrosine to R14ER did not convert it from d-type to l-type. Studies of how isolated ERs reduce isolated enoyl-ACPs and small molecule analogs showed that polyketides must be bound to an ACP for stereoselective reduction.¹⁵

The ER from the third module of the spinosyn PKS (S3ER) was the first to be structurally elucidated (PDB: 3SLK).¹⁹ Crystal structures of ERs from the curacin and juvenimicin PKSs have also been solved (PDBs: 5DP1, 8G7W).^20,21 As a member of the medium-chain dehydrogenase/reductase (MDR) superfamily, the ~330-residue ER domain consists of a nucleotide-binding subdomain (ER_n) that binds NADPH and a substrate-binding subdomain (ER_s) proposed to bind an enoyl-ACP (Figure 1c).^19,22–24 In δ-modules, which contain KR, DH, and ER domains, ER is located between the KR structural and catalytic subdomains (KR_s and KR_c) and forms relatively weak domain–domain contacts with KR (Figure 1d).^19,21 The vast majority of ERs from modular PKS are monomeric, in contrast to those of iterative type I PKSs.¹⁹ No experimentally determined structures report how ACPs dock with ERs; however, a recent cryo-EM structure reveals the association of the ACP and ER domains from the related human fatty acid synthase (FAS, PDB: 8VF7).²⁵

When engineering PKSs, collaborations between processing enzymes and downstream KSs must be taken into account. For example, DHs and KSs work together to install double bonds.¹¹ These KSs specialize in selecting α/β-unsaturated over β-hydroxy intermediates, which interconvert through a DH-mediated equilibrium. Bioinformatics studies identified a VMYH motif in gatekeeping positions 1–4 of Family K, L, and M KSs from γ-modules, which contain KR and DH and generate α/β-double bonds.^8,10 While no similar motif is observable in Family N, O, and P KSs from ER-containing modules, trends are apparent. Each possesses small residues in positions 1–4. Family N KSs that accept α-unsubstituted intermediates predominantly possess a tryptophan at gatekeeping position 29. Family O KSs that accept intermediates with a d-α-methyl group predominantly possess either a tryptophan or a phenylalanine at this position, whereas Family P KSs that accept intermediates with an l-α-methyl group predominantly possess a phenylalanine at this position. The correlation between the methyl orientation and the presence or absence of the tryptophan at position 29 is high in most KS families that accept intermediates with α-methyl groups but somewhat weaker in Family O and P KSs.¹⁰ KS gatekeeping studies have revealed that some synthases, such as the rapamycin PKS, contain promiscuous KSs that nonspecifically interact with polyketide intermediates with the help of an aromatic residue at gatekeeping position 2.^10,26

We designed diverse PKS platforms to study how ERs collaborate with other PKS domains. First, the ER/ACP interface predicted by AlphaFold 3 for both d- and l-type ERs was investigated through an in vitro platform employing P1-P5-P6-P7, a tetraketide synthase comprised of the first, fifth, sixth, and seventh modules of the pikromycin PKS.^8,13,27,28 Relative changes in activity for 21 synthase variants containing mutations to P5ER surface residues corroborated the proposed ER/ACP interface. Since ERs contact residues conserved in the ACPs of diverse module types, we performed ER-insertion experiments into γ-modules within diverse in vivo tetraketide-heptaketide synthase platforms, including one that generates macrolactones.^13,29,30 In each platform, the insertion of a d- or l-type ER yielded the anticipated product. The ER-inserted γ-modules often generate a mixture of α/β-unsaturated and α/β-saturated intermediates; however, swapping their KSs for those naturally from ER-containing modules greatly increases the ratio of α/β-saturated to -unsaturated intermediates. Insertion of the R2 and R8 ERs into R5 of P1-R5-R6-P7 yielded the anticipated, stereoisomeric tetraketide products. Through swapping a 14-residue loop between the R2 and R8 ERs in these synthases, ER stereoselectivities were almost completely reversed. Our results suggest that d- and l-type ERs equivalently generate enolate intermediates, but due to subtle geometric active site differences, the si face is exposed in d-type ERs for protonation by a conserved lysine, whereas the re face is exposed in l-type ERs for protonation by the signature tyrosine.

RESULTS AND DISCUSSION

AlphaFold Predictions for 50 KR_s+ER+KR_c+ACP Regions Show an ER/ACP Interface

Our lab has corroborated AlphaFold predictions for how ACPs dock to KRs and KSs within a PKS module by mutagenically probing short model synthases.^14,31 As ACPs were found to associate with A- and B-type KRs from β-modules at different sites to present opposite faces of the β-keto group of a polyketide intermediate for reduction, it is possible ACPs differentially associate with ERs to influence the stereoselectivities of d- and l-type ERs. Thus, 25 KR_s+ER+KR_c+ACP regions from δ-modules that yield d-α-methyl groups and 25 KR_s+ER+KR_c+ACP regions from δ-modules that yield l-αmethyl groups were supplied to AlphaFold 3 (Figures S1 and S2). For each KR_s+ER+KR_c+ACP, at least one of the 5 predictions show an ER/ACP association (d-type, Data Files 1–25; l-type, Data Files 26–50; sequences, Data File 51; PKS abbreviations, Table S1); the other predictions primarily show the known KR/ACP association (Figure 2a and Figure S3).¹⁴

Figure 2.

Probing the AlphaFold-predicted ER/ACP interface. (a) Stereodiagram showing the KR_s+ER+KR_c+ACP region of P5 predicted by AlphaFold 3. P5ER residues are shown in sticks, as are all P5ACP residues, including S′ and L′ from the conserved DSL motif. (b) The putative interface between P5ACP and P5ER was probed through mutagenesis of P5ER surface residues within the tetraketide synthase P1-P5-P6-P7. The relative in vitro production of tetraketide 1 from P1-P5-P6-P7 point mutants is in agreement with the predicted interface. Mutations to important residues at the ER/ACP interface increase ER-skipping and thus production of tetraketide 2. In all figures, β-ketolactones are represented as keto tautomers, as they dominate the NMR spectra; however, the enol tautomer is usually observed crystallographically due to contacts mediated by hydrogen bonds. Key: prime, ACP element; L1, KR_s/ER linker; L2, ER/KR_c linker; KR_s, KR structural subdomain; KR_c, KR catalytic subdomain; ER_n, ER nucleotide-binding subdomain; ER_s substrate-binding subdomain.

The ER/ACP interfaces are equivalent for both d- and l-type ERs and highly similar to the ER/ACP interface in the human FAS (Figure S4).²⁵ ER makes contact through its substrate-binding subdomain (ER_s) with residues in the loop between β1 and β2, α1, and α3, whereas ACP makes contact through residues in αI′ and αII′ (′ denotes ACP elements).¹⁹ The association between P5ER and P5ACP is representative of the other ER/ACP interfaces. The central contact is between a hydrophobic pocket formed by the side chains of ER residues F54, V57, L58, L61, M63, and S316 and the leucine of the ACP DSL motif, L52′ (numbering based on the first structured residue in each domain). L61 and M63 also hydrophobically interact with the conserved ACP valine V55′. The side chains of S19, L20, L61, R313, and S316 interact with the conserved ACP glutamate E56′. The backbone carbonyls of G18 and L61 form hydrogen bonds with the side chain amide of the conserved ACP asparagine N59′. The side chains of S19 and E21 make polar interactions with the conserved ACP arginine R60′. The side chains of R313 and Q317 also interact with backbone carbonyls at the C-terminal end of αI′. In 13 of the 50 predictions, ACP residues in a loop downstream of αI′ make contact with KR residues on or near its first β-strand, β1 (e.g., Data Files 24, 31, 47). The phosphopantetheinyl arm that is post-translationally appended to S51′ of P5ACP would be in position to access the ER active site as observed in the human FAS (PDB: 8VF7).²⁵ One of the most significant differences of the ER/ACP interface of the human FAS is that the pocket for the DSL leucine is less hydrophobic; however, the lysine that is in the same position as M63 appears to compensate for this by ionically interacting with the phosphodiester group of the phosphopantetheinyl arm.

Other interactions predicted by AlphaFold 3 are equivalent to those observed in the S3 KR_s+ER+KR_c crystal structure (PDB: 3SLK).¹⁹ NADPH is bound to the nucleotide-binding subdomain (ER_n) in the same orientation observed for NADP⁺. Contact between the KR and ER domains is primarily mediated by the short linkers connecting them. The median length of the N-terminal linker that spans the last structured KR_s residue and the first structured ER residue is 6 amino acids (it is unusually long in P5 at 18 amino acids), and the median length of the C-terminal linker that spans the last structured ER residue and the first structured KR_c residue is 2 amino acids. Residues in αA, β8, and β9 of ER usually form a pocket for the C-terminal end of αF from KR.

Mutagenesis of P1–P5–P6–P7 Corroborates the ER/ACP Interface

To probe the predicted ER/ACP interface, we systematically mutated 17 surface residues of P5ER within the in vitro tetraketide synthase platform P1-P5-P6-P7 [comprised of the 4 polypeptides, P1^N, ^CP1P5^N, ^CP5P6^N, and ^CP6P7, where ^N and ^C indicate the AT+(KR/DH/ER)+ACP and KS regions, respectively] (Figure 2b, Figure S5, and Tables S2 and S3).^13,32In vitro reactions were performed with 2 μM synthase, 2 mM methylmalonyl-CoA, and 1 mM NADPH in 400 mM potassium phosphate buffer (pH 7.5). They were quenched during active turnover after 10 min, and the yields of tetraketides 1 and 2 from P1-P5-P6-P7 and its 21 point mutants were compared through liquid chromatography/mass spectrometry (LC/MS) analysis (Figures S6–S12). Standards for 1 and 2 were prepared through extraction from the culture broths of E. coli K207–3 expressing P1-P5-P6-P7 and P1-R5-P6-P7 (vide infra), and a combination of silica gel and semipreparative C₁₈ chromatographies.³² Calibration curves for 1 and 2 were generated using LC/MS.

The results corroborate the ER/ACP interface predicted by AlphaFold 3. Of the 6 residues that comprise the hydrophobic pocket for L52′ of ACP (F54, V57, L58, L61, M63, and S316), F54, V57, L58, and L58 make the most significant contributions, as their mutation to alanine resulted in the greatest decreases in the production of 1 (83%, 48%, 63%, and 72%, respectively). Differential scanning fluorimetry shows that these mutations do not affect the stability of the P5ER domain (Figure S13). These synthases predominantly produce the ER-skipped product 2. Unmodified P1-P5-P6-P7 also generates a small quantity of 2 (a tenth that of 1), suggesting that P5KS does not gatekeep against the α/β-unsaturated diketide intermediate as well as it does the native α/β-unsaturated pentaketide intermediate. Decreased production of 1 (22%) was also observed from the R313A point mutant. This highly conserved arginine apparently plays a larger role than Q317 in contacting the C-terminal end of α1′, as the Q317A point mutant produced nearly as much 1 as P1-P5-P6-P7. Control point mutants yielded anticipated results: the alanine point mutants of T16 and E228, residues distant from the ER/ACP interface, produced as much tetraketide as P1-P5-P6-P7, whereas the alanine point mutant of R201, an invariant residue that interacts with both the NADPH adenine and ribose phosphate, produced only trace amounts of 1 (but a significant quantity of 2).

ER Insertion

If ERs could be generally inserted into γ-modules to produce β-methylene intermediates, a wide range of polyketide derivatives could be accessed. Such an insertion seems feasible since the KR/ER interface is relatively small and the ACP residues that interface with ER (αII′ residues L52′, V55′, E56′, N59′, and R60′) are as conserved in γ-modules as δ-modules. Thus, P5ER was inserted into P3 of P1-P3-P6-P7, a model synthase that produces tetraketide lactone 3 (7.6 ± 0.6 mg L⁻¹)(Figure 3a and Figures S14–S18). Based on the S3 KR_s+ER+KR_c crystal structure (PDB: 3SLK), the KR_s+ER +KR_c+ACP AlphaFold 3 models, and multiple sequence alignments of γ- and δ-modules, the most upstream residue of the ER insert was chosen to be 3 residues downstream of the last β-strand in KR_s and the most downstream residue of the ER insert was chosen to be 6 residues upstream of the first β-strand in KR_c (Figure S19).^19,33 Although P1-P3[P5ER]-P6-P7 produced the expected tetraketide 4 (1.4 ± 0.1 mg L⁻¹) in E. coli K207–3, it produced higher quantities of the ER-skipped tetraketide 3 (4.8 ± 0.3 mg L⁻¹) (Figures S20–S23). We also employed an in vitro platform in which polypeptides and substrates can be incubated at desired concentrations and kinetics can be studied (Figure 3a and Figure S5).³² In vitro, 3 is solely produced by P1-P3-P6-P7 at 63.6 ± 8.4 μM, whereas both 4 and 3 are produced by P1-P3[P5ER]-P6-P7 at 6.6 ± 0.6 μM and 3.6 ± 0.5 μM, respectively. These results show that P3KS accepts diketide intermediates whether or not they have been reduced by P5ER and indicate that the reduced diketide acylates P3KS more slowly than the unreduced diketide. In a previous study, a VMYH motif at gatekeeping positions 1–4 was observed to help the Family K P3KS select α/β-unsubstituted intermediates over β-hydroxy intermediates that interconvert through P3DH¹¹. Due to this intrinsic selectivity of P3KS for α/β-unsaturated intermediates, we swapped it for P5KS, although this is also an imperfect solution: P5KS is a Family P KS that naturally accepts intermediates with l-α-methyl substituents, and it would be accepting an intermediate without an α-substituent; also, P3ACP naturally docks to a serine residue at the key position 275 of P3KS, and it would be docking an asparagine at this position in P5KS.^11,31 Despite these potential pitfalls, P1-P3[P5ER+P5KS]-P6-P7 selectively produced tetraketide 4 (1.2 ± 0.2 mg L⁻¹, in vivo; 5.6 ± 0.6 μM, in vitro), and tetraketide 3 was barely detected from both the in vitro and in vivo platforms (Figure 3a). Thus, transacylation of the α/β-unsaturated diketide intermediate to P5KS is impeded relative to P3KS. We further investigated this activity of P5KS with P1-P3[P5KS]-P6-P7. This synthase only produces a small quantity of 3 in both in vivo and in vitro (0.2 ± 0.1 mg L⁻¹ and 0.7 ± 0.1 μM, respectively). P1-P3[P5KS]-P6-P7 produces a 38-fold lower titer of 3 than P1-P3-P6-P7, reaffirming engineering with the updated over the traditional module boundary (Figure S24).¹³ Unexpectedly, trace quantities of 4 are also produced by E. coli expressing either P1-P3-P6-P7 or P1-P3[P5KS]-P6-P7. We hypothesize that an E. coli enzyme, such as FabI, can reduce the α/β-unsaturated diketide intermediate bound to P3ACP at some rate. Indeed, neither P1-P3-P6-P7 nor P1-P3[P5KS]-P6-P7 yielded detectable levels of 4 in vitro.

Figure 3.

ER insertion coupled with KS swapping. (a) Inserting P5ER into P3 in both the in vitro and in vivoP1-P3-P6-P7 platforms resulted in the production of tetraketides 3 and 4, revealing that P3KS accepts both P5ER-reduced and P5ER-unreduced diketide intermediates. Additionally replacing P3KS with P5KS greatly increased selection for the P5ER-reduced intermediate, while only replacing P3KS with P5KS resulted in the production of a trace quantity of 3. Crystal structures of 3 and 4 confirm their absolute stereochemistries. (b) The generality of coupling ER insertion with KS swapping was also tested in both the in vitro and in vivoP1-P2-P3-P6-P7 platforms. Levels of pentaketides 5 and 6 mirrored those of 3 and 4 from the P1-P3-P6-P7 platforms, affirming that the strategy can be generally applied. A crystal structure of 6 confirms its absolute stereochemistry. Schematics shown in both panels are for the in vitro platform. The P1-P3-P6-P7 in vivo platform differs in that the first two polypeptides are fused. Key: ER^Y, l-type ER; tr., trace quantities; n.d., not detected.

To better characterize tetraketides 3 and 4, we aimed to boost their production in E. coli K207–3. A two-plasmid expression system was employed in combination with a T7 promoter attenuation strategy (Figure S25).³² Through this, the titer of 3 from P1-P3-P6-P7 increased 6.5-fold (46.3 ± 1.6 mg L⁻¹), and the titer of 4 from P1-P3[P5ER+P5KS]-P6-P7 increased 4.4-fold (5.4 ± 0.2 mg L⁻¹). Semipreparative quantities of each product could be purified, and the absolute crystal structures of 3 and 4 were determined (Figure 3a).

To investigate the generality of coupling ER insertion with KS swapping, pentaketide synthases P1-P2-P3-P6-P7, P1-P2-P3[P5ER]-P6-P7, P1-P2-P3[P5ER+P5KS]-P6-P7, and P1-P2-P3[P5KS]-P6-P7 were assayed both in vivo and in vitro (Figure 3b and Figure S5). P1-P2-P3-P6-P7 yields 222 ± 24 mg L⁻¹ of 5in vivo and 140 ± 8 μM in vitro (Figures S26–S28). ¹³P1-P2-P3[P5ER]-P6-P7 generates the anticipated, reduced tetraketide 6 (51 ± 4 mg L⁻¹) in E. coli K207–3 (structure confirmed by X-ray diffraction); however, it produced a greater amount of the unreduced tetraketide 5 (169 ± 9 mg L⁻¹)(Figures S29–S32). In vitro, P1-P2-P3[P5ER]-P6-P7 yields comparable levels of the two pentaketides (15.7 ± 1.9 μM of 6 vs 16.4 ± 2 μM of 5) (Figure 3b). In vivo, P1-P2-P3[P5ER+P5KS]-P6-P7 produces significantly more 6 (12 ± 1 mg L⁻¹) than 5 (1.6 ± 0.4 mg L⁻¹), and in vitro, 6 was exclusively produced (2 ± 0.2 μM). The reduced tetraketide 6 could not be detected either in vivo or in vitro from P1-P2-P3[P5KS]-P6-P7, indicating that α/β-unsaturated triketide intermediate is not a substrate for the E. coli enzyme that reduces the α/β-unsaturated diketide intermediate of P1-P3[P5KS]-P6-P7.

Installing d- and l-Oriented Methyl Substituents

While coupling ER insertion and KS swapping worked to generate reduced, α-unsubstituted intermediates both in vivo and in vitro, we sought to engineer modules to deliver d- and l-oriented α-methyl substituents. Subsequent experiments were conducted only in vivo, such that quantities of products sufficient for NMR and/or crystallographic analysis could be obtained. To avoid the potential pitfalls associated with KS swapping, we also sought to utilize promiscuous KSs, such as those from the rapamycin PKS. A recent engineering study in our lab indicated the general compatibility of components from the pikromycin and rapamycin PKSs.¹³ Rapamycin PKS components include 2 γ-modules that generate α-methyl, α/β-unsaturated intermediates (R5 and R11), 2 δ-modules that generate d-α-methyl intermediates (R2 and R14), and 1 δ-module that generates an l-α-methyl intermediate (R8) (Figure 4a and Figure S33). The KSs of R1–R12 are highly similar to one another but distinct from those of R13 and R14. How they gatekeep against incompletely processed intermediates is unclear. We first pursued P1-X-P6-P7 as a platform to study engineered modules at the X position (Figure 4b and Figures S34 and S35). While P1-R5-P6-P7 and P1-R8-P6-P7 yielded their respective, expected tetraketide products, 2 and 1, P1-R2-P6-P7 only produced trace quantities of its expected tetraketide product, 7 (Figure S34). Within P1-R2-P6-P7, P6KS likely gatekeeps against the d-γ-methyl substituent presented to it, as suggested by a previous study.¹³ Thus, we pursued P1-X-R6-P7 (R6 generates a d-β-hydroxy intermediate in the biosynthesis of rapamycin) as a platform, with the expectation that R6KS would accept both d- and l-oriented γ-methyl substituents (Figure 4c).¹³P1-R5-R6-P7 produced the expected tetraketide 8 (3.1 ± 0.4 mg L⁻¹), and each of P1-R2-R6-P7, P1-R8-R6-P7, and P1-R14-R6-P7 yielded products with the mass expected for tetraketide stereoisomers 9 and 10 (12 ± 1 mg L⁻¹, 31 ± 0.6 mg L⁻¹, and 5.9 ± 0.3 mg L⁻¹, respectively) (Figures S36–S47). While the stereoisomers could not be separated using a C₁₈ column, they could be using an OD-HR chiral column (Figure S48). Following silica gel chromatography and semipreparative HPLC, 5.9 mg of the P1-R2-R6-P7 product and 8.6 mg of the P1-R8-R6-P7 product were obtained. Their ¹H NMR spectra are essentially equivalent except for a doublet that appears at 1.01 ppm for the P1-R2-R6-P7 product and 1.06 ppm for the P1-R8-R6-P7 product (Figure 4e). The absolute crystal structures for both tetraketide stereoisomers were obtained (Figure 4c). The crystal structure of 10, also known as drakolide, was previously reported.³⁴ Together, the chiral chromatography and crystal structures confirmed that P1-R2-R6-P7 and P1-R14-R6-P7 produce stereoisomer 9, in which the ε-methyl substituent is d-oriented, at >99% diastereomeric excess (de) and that P1-R8-R6-P7 produces stereoisomer 10, in which the ε-methyl substituent is l-oriented, at >99% de. Interestingly, small quantities of 8 were produced by P1-R2-R6-P7 and P1-R8-R6-P7, but not by P1-R14-R6-P7. This likely results from R14KS gatekeeping against α/β-unsaturated intermediates more stringently than R2KS and R8KS.

Figure 4.

Inserting ER domains to control methyl substituent orientation. (a) Rapamycin PKS components used in this study. (b) P1-X-P6-P7 was investigated as a platform for ER insertion. The insertion of R5 and R8 from the rapamycin PKS at X yielded expected tetraketides, 2 and 1, but the insertion of R2 and R14 only yielded trace quantities of 7. Since P6KS apparently selects against d-oriented γ-methyl substituents, this platform was not pursued. (c) P1-X-R6-P7 was chosen as a platform for ER insertion, as inserting the rapamycin modules at X yielded expected tetraketides, likely owing to the relaxed gatekeeping activity of R6KS. Crystal structures of 9 and 10 show their absolute stereochemistries. (d) ERs from the rapamycin PKS that were inserted in R5 of P1-R5-R6-P7 stereoselectively reduced the diketide intermediate, with R2ER and R14ER generating a d-oriented methyl group and with R4ER and R8ER generating an l-oriented methyl group. When ERs from the pikromycin and spinosyn PKSs, P5ER and S3ER, were inserted into R5, they were both observed to generate an l-oriented methyl group. This was not anticipated of S3ER, as it does not naturally reduce an α-substituted intermediate. (e) The d-oriented ε-methyl group of 9 yields a doublet at 1.01 ppm in its ¹H NMR spectrum, whereas the l-oriented ε-methyl group of 10 yields a doublet at 1.06 ppm. Key: ER^Y, l-type ER; ER^V, d-type ER.

To investigate whether ER insertion would be as effective in the P1-R5-R6-P7 platform as it is in the P1-P3-P6-P7 and P1-P2-P3-P6-P7 platforms and yield expected methyl group orientations, P1-R5[R2ER]-R6-P7, P1-R5[R4ER]-R6-P7, P1-R5[R8ER]-R6-P7, and P1-R5[R14ER]-R6-P7 were engineered (R4ER does not reduce an intermediate during rapamycin biosynthesis but is predicted by AlphaFold 3 to be equivalent to R8ER) (Figure 4d,e and Figure S49). Gratifyingly, each of the ER-inserted synthases generates its anticipated product, along with a small quantity of 8 (Figures S49–S56). ¹H NMR analysis revealed that both P1-R5[R2ER]-R6-P7 and P1-R5[R14ER]-R6-P7 generated 9 at >99% de and both P1-R5[R4ER]-R6-P7 and P1-R5[R8ER]-R6-P7 generated 10 at >99% de. Given the successful insertion of the rapamycin ERs into R5, we sought to determine whether ERs from other PKSs could be inserted into this rapamycin γ-module. Thus, P1-R5[P5ER]-R6-P7 and P1-R5[S3ER]-R6-P7 (S3 is the third module of the spinosyn PKS) were constructed. Both produced the anticipated stereoisomer 10 with >99% de at comparable levels to the synthases containing rapamycin ER insertions. While the stereoselectivity displayed by P5ER was anticipated, as it natively generates an intermediate with an l-oriented α-methyl group, the stereoselectivity displayed by S3ER was surprising, as this ER natively reduces an α-unsubstituted intermediate. This implies that even ERs that do not generate stereocenters perform stereocontrolled reductions.

While the KSs of R1–R12 are highly similar, their differences could still be significant. Having inserted ERs into R5, we investigated whether the KSs that naturally accompany these ERs are better acceptors of the ER-reduced intermediates than R5KS (Figure 4e and Figure S56). P1-R5[R2ER+R2KS]-R6-P7 and P1-R5[R2ER+R8KS]-R6-P7 were engineered and respectively observed to produce 1.9-fold and 4.9-fold greater titers of 9 than P1-R5[R2ER]-R6-P7. P1-R5[R8ER+R2KS]-R6-P7 and P1-R5[R8ER+R8KS]-R6-P7 were also engineered and respectively observed to produce 8.0-fold and 8.5-fold greater titers of 10 than P1-R5[R8ER]-R6-P7. One possible explanation for these relative increases is that both R2KS and R8KS possess a phenylalanine at gatekeeping position 29, whereas R5KS possesses a tryptophan at this position.

We next sought to test the ER insertion strategy in a more challenging platform where the inserted ER would need to reduce the lower-energy α/β-double bond of a conjugated dienoyl moiety (~4 kcal/mol lower than an enoyl moiety, as estimated using Benson group additivity) (Figure 5a).³⁵ This platform was inspired by P1-P2-P3-P5-P7, a synthase identified in a recent study that generates a compound with the mass of its anticipated pentaketide product, 13 (Figure S57).¹³ P1-P2-P3-R5-P7 was constructed and observed to generate its anticipated product 11 (Figures S58–S61). P1-P2-P3-R2-P7 and P1-P2-P3-R8-P7 were also constructed and observed to generate compounds with the mass expected for their anticipated, stereoisomeric pentaketide products, 12 and 13, respectively (Figures S62–S68). Chiral chromatography and NMR revealed that each synthase only produced one stereoisomer (Figure S69).

Figure 5.

ER insertion into more challenging pentaketide and heptaketide platforms demonstrates that ERs maintain comparable stereocontrol with longer tetraketide and hexaketide substrates. (a) P1-P2-P3-R5-P7 generates diene 11, P1-P2-P3-R2-P7 generates pentaketide stereoisomer 12, and P1-P2-P3-R8-P7 generates pentaketide stereoisomer 13. The latter two synthases also produce 11, as reduction of a dienoyl intermediate is less favorable (~4 kcal/mol) than the reduction of an enoyl intermediate. The insertion of R2ER and R8ER into R5 of P1-P2-P3-R5-P7 yielded 12 and 13, respectively, along with significant quantities of 11. (b) P1-P2-P3-P4-R5-R5-P7 generates linear heptaketide 14, P1-P2-P3-P4-R5-R2-P7 generates macrolactone stereoisomer 15, and P1-P2-P3-P4-R5-R8-P7 generates macrolactone stereoisomer 16. The latter two synthases also generate small quantities of 14. The insertion of R2ER and R8ER into R5 of P1-P2-P3-P4-R5-R5-P7 yielded 15 and 16, respectively, along with small quantities of 14. Key: ER^Y, l-type ER; ER^V, d-type ER.

P1-P2-P3-R2-P7 and P1-P2-P3-R8-P7 also produced significant quantities of 11, with P1-P2-P3-R8-P7 producing as much 11 as it does 13. We learned that P1-P2-P3-P5-P7 also produces 11, at approximately half the level it produces 13. From these results, we hypothesize that ER establishes an equilibrium in which levels of its dienoyl substrate are comparable to those of its reduced product and that P5KS is more selective for the reduced intermediate than R8KS. We next inserted both R2ER and R8ER into R5 of P1-P2-P3-R5-P7 (Figure 5a and Figure S70). Gratifyingly, P1-P2-P3-R5[R2ER]-P7 produced 12, and P1-P2-P3-R5[R8ER]-P7 produced 13 (3.6 mg L⁻¹ vs 3.2 mg L⁻¹)(Figures S71–S73). Both also generated a significant quantity of 11 (1.6 mg L⁻¹ vs 1.4 mg L⁻¹). Swapping R2KS and R8KS for R5KS in these synthases did not significantly change the titers of 12 or 13.

We then attempted ER insertion within a platform that can generate macrolactones, P1-P2-P3-P4-R5-X-P7 (Figure 5b and Figure S74). This platform was chosen based on studies of hybrid synthases in our lab in which P1-P2-P3-P4-R5-R5-P7, P1-P2-P3-P4-R5-R2-P7, and P1-P2-P3-P4-R5-R8-P7 were each observed to generate a compound possessing the mass of its anticipated product. The titers for each synthase were quantified: P1-P2-P3-P4-R5-R5-P7 generates the linear polyketide 14 (0.6 mg L⁻¹), P1-P2-P3-P4-R5-R2-P7 generates macrolactone stereoisomer 15 at 0.1 mg L⁻¹, and P1-P2-P3-P4-R5-R8-P7 generates macrolactone stereoisomer 16 (2.5 mg L⁻¹) (Figures S75–S86). Macrolactone stereoisomers 15 and 16 could be distinguished by chiral chromatography and NMR. Gratifyingly, the ER-inserted modules R5[R2ER] and R5[R8ER] operate as desired, with P1-P2-P3-P4-R5-R5[R2ER]-P7 generating macrolactone stereoisomer 15 (0.2 mg L⁻¹) and P1-P2-P3-P4-R5-R5[R2ER]-P7 generating macrolactone stereoisomer 16 (0.5 mg L⁻¹) (Figure S87). Further installing R2KS and R8KS in these modules led to a range of outcomes, with P1-P2-P3-P4-R5-R5[R2ER+R2KS]-P7 not producing detectable levels of 15 and P1-P2-P3-P4-R5-R5[R8ER+R8KS]-P7 producing 4-fold more 16 (Figure S88).

The molecular basis of ER stereoselectivity

How ERs set stereocenters has long been a mystery. In each of the employed platforms, insertion of R2ER resulted in Doriented methyl substituents while insertion of R8ER resulted in l-oriented methyl substituents. In both sequence and structure, these ERs are highly similar (Figure 6a and Figures S1–S3). The only significant difference is the α1–β4 loop region of the substrate-binding subdomain (residues 59–72: VALGMVDDKRLAGG in R2ER vs IALGTYPGEAVIGA in R8ER). Thus, these 14 residues were swapped between P1-R5[R2ER]-R6-P7 and P1-R5[R8ER]-R6-P7, which produces 9 (4.7 mg L⁻¹) and 10 (4.0 mg L⁻¹), respectively (Figure 6b,c and Figures S89–S97). Incredibly, P1-R5[R2ER¹⁴]-R6-P7 produces 10 at 5.6 mg L⁻¹ and >99% de, and P1-R5[R8ER¹⁴]-R6-P7 produces 9 at 5.8 mg L⁻¹ and >98% de (Figure S97). To determine whether smaller loop swaps could also reverse the stereoselectivities of R2ER and R8ER, synthases with 9-residue swaps (residues 64–72: VDDKRLAGG in R2ER, YPGEAVIGA in R8ER) and 5-residue swaps (residues 64–68: VDDKR in R2ER, YPGEA in R8ER) were constructed. Compared with the 14-residue-swapped synthases, both the 9-residue-swapped synthases, P1-R5[R2ER⁹]-R6-P7 and P1-R5[R8ER⁹]-R6-P7, as well as the 5-residue-swapped synthases, P1-R5[R2ER⁵]-R6-P7 and P1-R5[R8ER⁵]-R6-P7, are less stereoselective and less active. The tyrosine in the R8ER loop is the signature tyrosine of l-type ERs proposed to donate a proton to the re face of the enolate. Its equivalent in the R2ER loop is the signature valine of d-type ERs. However, swapping only these residues does little to reverse ER stereoselectivity. P1-R5[R2ER^V64Y]-R6-P7 still produces 9 (4.6 mg L⁻¹), and P1-R5[R8ER^Y64V]-R6-P7 predominantly produces 10 (10 at 0.9 mg L⁻¹, 9 at 0.2 mg L⁻¹)(Figure S97).^16,17 Apparently, the structure of the α1-β4 loop region is critical to the stereoselectivity and activity for both d- and l-type ERs, and several residues in these 14-residue regions make important contributions.

Figure 6.

Molecular basis of ER stereoselectivity. (a) Superposition of the AlphaFold 3 predictions for the d-type R2ER and the l-type R8ER reveal that their most significant difference is the α1-β4 loop region (residues 59–72, P5ER numbering). This region is boxed in a sequence alignment of the N-terminal residues of the rapamycin ERs. The signature valine of d-type ERs and signature tyrosine of l-type ERs are indicated. (b) Swaps were performed using 14-, 9-, or 5-residue stretches. These stretches were swapped between P1-R5[R2ER]-R6-P7 and P1-R5[R8ER]-R6-P7, which normally produce 9 and 10, respectively. The 14-residue swaps resulted in a nearly complete reversal, with P1-R5[R2ER¹⁴]-R6-P7 and P1-R5[R8ER¹⁴]-R6-P7 producing 10 and 9, respectively. The 9- and 5-residue-swapped synthases are less stereoselective and less active. (c) ¹H NMR spectroscopy was used to quantify the relative production of stereoisomers 9 and 10. (d) The phosphopantetheinyl-bound diketide intermediate was modeled in the R2ER and R8ER active sites. In d-type ERs, the conserved lysine protonates the si face of the enolate intermediate. In l-type ERs, the signature tyrosine protonates the re face and prevents the conserved lysine from accessing the si face by pinning the substrate to the nicotinamide ring. Modeling was enabled by the ternary complex of a related enoyl-thioester reductase (PDB: 4WAS) in which the thioester carbonyl of the substrate is bound to the 2′-OH of the nicotinamide ribose and a highly ordered water (Wat). Key: ER*, ER containing 14-residue swap.

Models for how α-methyl substituted intermediates are bound within the active sites of the d-type R2ER as well as the l-type R8ER and P5ER were constructed to help understand how these enzymes perform stereoselective reduction (Data Files 17, 45, 47, and 52–54) (Figure 6d). The positions of the ACP domains and NADPH molecules are based on the ER/ACP-associated human FAS structure (PDB: 8VF7) and the AlphaFold 3 KR_s+ER+KR_c+ACP predictions.²⁵ A phosphopantetheinyl arm bound to an α-methyl diketide intermediate was then modeled into the active site of the l-type P5ER based on the positions of the phosphopantetheinyl arm in the ER/ACP-associated human FAS structure and the crotonyl-CoA intermediate in an enoyl-thioester reductase ternary complex (PDB: 4WAS).¹⁸ The signature tyrosine is not only in position to protonate the re face of the enolate (O_ε-C_α: 3.8 Å) but also pins the si face to the nicotinamide ring such that it cannot be accessed by the neighboring conserved lysine, K248. Modeling with the l-type R8ER proceeded equivalently. The phosphopantetheinyl arm bound to an α-methyl diketide intermediate could not be modeled into the active site of the d-type R2ER in same conformation as in the l-type ERs due to steric clashes with the conserved lysine. This lysine is positioned relatively closer to the active site since the alkyl portion of its side chain interacts with the signature valine of d-type ER rather than the conserved proline residue immediately downstream of the signature tyrosine of l-type ERs. The intermediate was modeled closer to the conserved arginine of the NFRD motif. In this orientation, the si face of the enolate is accessible to the conserved lysine (N_ε-C_α: 3.9 Å). Thus, enolate formation within all ERs most likely occurs through Michael attack of the NADPH hydride at the β-carbon, with stabilization of the negatively charged enolate provided by ionic interactions with the positively charged nicotinamide ring and hydrogen bonds with the neighboring ribose 2′-OH and a water bound at the N-terminal end of α4, as observed in the enoyl-thioester reductase ternary complex.¹⁸ In l-type ERs the α-carbon of the enolate would acquire a proton on its re face from the signature tyrosine (except those possessing NFHD rather than the NFRD motif that apparently employ this histidine, Data Files 27–28), whereas in d-type ERs the α-carbon of the enolate would acquire a proton on its si face from the conserved lysine. This mechanism is consistent with a study on an enoyl-thioester reductase in which the re face of crotonyl-CoA is protonated by a tyrosine from the wild-type enzyme, positioned equivalently to the tyrosine of l-type ERs (O_ε-C_α: 3.2 Å in PDB 4WAS), and the si face is protonated by a glutamic acid in a Y/F, W/E double mutant, positioned near the conserved ER lysine. The lysine may be conserved in l-type ERs to decrease the pK_a of the signature tyrosine.

CONCLUSIONS

Engineering modular PKSs requires a deeper understanding not only how their domains operate, but also how they collaborate with one another. Here, we study one of their most common yet mysterious domains, ER. We elucidate how it associates with intermediate-bound ACP domains to conduct chemistry that can yield d- or l-oriented substituents. We show that ERs can be inserted into modules and perform the chemistry anticipated of them but that the intermediates generated by those modules are also dependent on the gatekeeping properties of the KS. We learned that ER insertions could be coupled with KS swaps to access desired products. As KSs can be sensitive not only to the methyl substituents installed by the module to which they belong but also to those installed by the upstream module, we also employed modules from the rapamycin synthase with more relaxed KS gatekeeping properties to access desired products. We were also able to interconvert d- and l-type ERs for the first time, which helped us determine the molecular basis for ER stereoselectivity. The chemical logic uncovered in this study not only helps understand how PKSs naturally function but also informs PKS engineers how to rapidly access polyketide derivatives with new substituents and stereochemistries.

Supplementary Material

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacs.6c03764.

Data Files containing coordinates for KR+ER+ACP models; correspondence and requests for materials should be addressed to A.T.K. (ZIP)

Data Source File 1 containing measurements from in vitro and in vivo reactions (XLSX)

Data Source File 2 containing measurements from in vitro and in vivo reactions (XLSX)

Additional methods, figures, tables, and characterization (PDF)

Data Availability Statement

Data supporting the findings of this work are available within the paper and its Supporting Information files. Triplicate measurements are available in Data Source Files 1–2. The X-ray coordinates for the structures reported in this study were deposited at the Cambridge Crystallographic Data Centre (CCDC) under deposition numbers 2471795–2471800 (1, 3, 4, 6, 9, and 10).