How cis-Acyltransferase Assembly-Line Ketosynthases Gatekeep for Processed Polyketide Intermediates

Abstract

With the redefinition of polyketide synthase (PKS) modules, a new appreciation of their most downstream domain, the ketosynthase (KS), is emerging. In addition to performing its well-established role of generating a carbon–carbon bond between an acyl-CoA building block and a growing polyketide, it may gatekeep against incompletely processed intermediates. Here, we investigate 739 KSs from 92 primarily actinomycete, cis-acyltransferase assembly lines. When KSs were separated into 16 families based on the chemistries at the α- and β-carbons of their polyketide substrates, a comparison of 32 substrate tunnel residues revealed unique sequence fingerprints. Surprisingly, additional fingerprints were detected when the chemistry at the γ-carbon was considered. Representative KSs were modeled bound to their natural polyketide substrates to better understand observed patterns, such as the substitution of a tryptophan by a smaller residue to accommodate an l-α-methyl group or the substitution of four smaller residues by larger ones to make better contact with a primer unit or diketide. Mutagenesis of a conserved glutamine in a KS within a model triketide synthase indicates that the substrate tunnel is sensitive to alteration and that engineering this KS to accept unnatural substrates may require several mutations.

Graphical Abstract

INTRODUCTION

Modular polyketide synthases (PKSs) are multidomain assembly lines that biosynthesize many important medicines, such as the antibiotic erythromycin and the anticancer agent epothilone.¹ Within these molecular machines, it is well established that acyltransferases (ATs) transfer carbon building blocks from α-carboxyacyl-CoA extender units to acyl carrier proteins (ACPs), ketosynthases (KSs) generate carbon–carbon bonds between acyl-ACPs and growing polyketides, and processing enzymes, such as ketoreductases (KRs), dehydratases (DHs), and enoylreductases (ERs), adjust the chemistry of the α- and β-carbons of the extended intermediate. However, the sets of these domains that most closely collaborate, or “modules”, and the mechanisms by which they ensure the correct processing of intermediates are just beginning to be understood.

For the past three decades, most PKS engineers have performed module-swapping using a boundary that places KS at the most upstream position of the module and ACP at the most downstream. This strategy has rarely been successful at producing appreciable quantities of a desired polyketide.² Pairing ACPs and the KSs naturally downstream of them within engineered synthases yielded better results.³ Recently, a phylogenetic study of a group of cis-acyltransferase (cis-AT) PKSs revealed the evolutionary comigration of KSs with upstream processing enzymes.⁴ A phylogenetic study of the related trans-AT assembly lines also showed correlations between clades of KSs and their upstream ACPs.⁵ In agreement with these observations, synthases designed using the updated module boundary downstream of KS have consistently outperformed those designed using the traditional boundary.^6–8

Within the redefined module, KSs work closely with their upstream ACPs and processing enzymes to ensure intermediates are processed correctly.⁹ In addition to recognizing at least ACP through complementary surface chemistry, a KS may possess a substrate tunnel complementary to the chemistry of the fully processed polyketide intermediate. Multiple sequence alignments of KSs from 18 streptomycete cis-AT assembly lines revealed four residues at the dimer interface loop¹⁰ that often covary with polyketide chemistry—most notably, TNGQ for the α-unsubstituted, d-β-hydroxy intermediates generated by B0-type KRs and VMYH for the α-unsubstituted, α/β-unsaturated intermediates generated by B0-type KRs and DHs.^4,11

The inactivation of processing enzymes within a synthase usually results in large decreases in polyketide production and shunt products indicative of a stall in the assembly line.¹² The gatekeeping activities of KSs could explain these phenomena (Figure 1). Studies of processing enzyme knockouts in the monensin PKS showed KR knockouts are among the most severely affected.¹² This could be due to a general exclusion of the β-keto group by KSs that accept more reduced intermediates. The products of ER knockouts contain a new double bond rather than a new hydroxy group, as might be expected from the DH-mediated equilibrium between β-hydroxy and α/β-unsaturated intermediates.^12–14 This could be due to KSs that normally accept a β-methylene most strongly gatekeeping against the intermediates with polar β-hydroxy groups. The authors of the study of the monensin processing enzyme knockouts considered KSs as potential gatekeepers; however, since only weak fingerprints at the dimer interface loop were observed for the KSs that accept α/β-unsaturated substrates, other enzymes were investigated for this role.¹²

Figure 1.

Ketosynthase gatekeeping within the pikromycin PKS. (a) Pikromycin PKS operates in a hand-over-hand manner, with KSs and ACPs passing the growing polyketide through the assembly line until it is cyclized into the pikromycin precursor, narbonolide. Unlabeled circles, ACPs; TE, thioesterase; KS^Q, priming KS; KR⁰, reductase-inactive but epimerase-active KR. (b) KS of the third module (yellow) optimally catalyzes transacylation after the upstream AT, KR, and DH generate a fully processed intermediate. Its substrate tunnel does not bind the malonyl extender unit, the β-ketoacyl intermediate, or the β-hydroxyacyl intermediate as well as it does the α/β-unsaturated intermediate. Because of the DH-mediated equilibrium, the β-hydroxyacyl intermediate may be presented more often than the α/β-unsaturated intermediate to the KS.

We looked beyond the four-residue stretch at the dimer interface loop and beyond the previously analyzed PKSs to more completely characterize KSs that accept intermediates of differing chemistries. Based on the chemistries at the α- and β-carbons anticipated for their substrates, 739 KSs from 92 PKSs were divided into 16 families, which were subdivided into 115 groups based on the chemistry at the γ-carbon. Residues at 32 active site positions were compared through sequence logos generated for each KS Family. Patterns emerged from this analysis, and the roles of several residues were investigated through molecular modeling. A highly conserved glutamine in the substrate binding tunnel proved sensitive to mutation within an engineered triketide synthase. The insights gained for how sets of gatekeeping residues collaborate within the KS active sites will be informative in the rational engineering of these highly sophisticated chemical factories.

RESULTS

Selection of KSs and Active Site Residues.

KSs from 129 characterized PKSs were grouped based on the chemistries at the α-, β-, and γ-carbons of the substrates they accept, as predicted from biosynthetic models of each PKS and sequence analysis of KRs and ERs (Supporting Information and Source Data 1).^1,15 A cladogram of these KSs was generated, revealing significant clustering of KSs that accept less common substrates (nomenclature for KSs explained in Experimental Procedures and Table S1) (Figure 2). KSs collaborating with upstream nonribosomal peptide synthetase (NRPS) domains are distinct, as has been noted in both cis- and trans-AT assembly lines (Figure S1).^5,10 One clade most commonly accepts aminoacyl-extended intermediates modified through cyclization and oxidation. Another clade contains mostly KSs from modules that load aromatic starter units onto the synthase. Interestingly, this clade includes not only KSs from the first modules of ansamycin synthases that accept 3-amino-5-hydroxybenzoyl starter units but also KSs from the fourth modules of naphthalenic ansamycin synthases thought to gatekeep for the oxidation/cyclization that helps generate the second ring of the naphthalene moiety.¹⁶ KSs of cyanobacterial and myxobacterial PKSs form a separate clade from those of actinomycete PKSs, and distinct sets of residues are located in their active sites [PDBs 4MZ0, 2HG4, and 6C9U] (Figure S2).^17–20 After these KSs were removed, 739 KSs, primarily from actinomycete PKSs, were sorted into 115 groups based on the chemistries at the α-, β-, and γ-carbons of their substrates and 16 families (A–P) based on the chemistries at the α- and β-carbons of their substrates (Figures 3 and S3–S6).

Figure 2.

Choice of KSs for gatekeeping analysis. A cladogram of KSs from 129 assembly lines shows phylogenetic relationships. Among the 233 KSs excluded from this study are those from cyanobacteria and myxobacteria, those collaborating with NRPS enzymes (e.g., EpoP_1z), those implicated in selecting for cyclizations (e.g., RifA_4b), and those accepting aromatic starter units. KS nomenclature is described in the Experimental Procedures section.

Figure 3.

KSs sorted by the substrate they accept. Based on the chemistries at the α-, β-, and γ-carbons of their substrates, 739 KSs were placed into 115 groups, indicated in the upper right of each box. The chemistries at the α- and β-carbons of their substrates determine their family (A–P). The number of KSs in each group is reported in the lower right of each box. The identities of the KSs are provided in Figure S4.

Residues were selected for analysis based on their proximity to where the α-, β-, and γ-carbons of substrates are located during the transacylation reaction (Figure 4).^21,22 Positions 1–6 (residues 154–159 in EryKS3, PDB 6C9U) are in the dimer interface loop—the first four corresponding to the residues that comprise the TNGQ and VMYH motifs.^4,10 Positions 7–17 (171–181 in EryKS3) are also at the dimer interface in the form of two helices that help construct the substrate binding tunnel. Positions 18–26 (231–233 and 237–242 in EryKS3) are comprised of a loop that runs parallel to the dimer interface loop and a helix. Positions 27–30 (264–267 in EryKS3) form a loop adjacent to Positions 18–20. Position 31 (380 in EryKS3) is immediately downstream of the KSNIGHT motif, and Position 32 (445 in EryKS3) is immediately upstream of the SGTNAH motif.

Figure 4.

Comparing the substrate tunnels of KS families. (a) A multiple sequence alignment shows where the 32 substrate tunnel residues are located within KSs. (b, c) Stereodiagrams of the representative KS from Family I (Eco7_2b_I7) show the residues in the 32 positions that help bind a fully processed polyketide intermediate (gray). Residues in cyan are contributed by the other monomer. (d) Sequence logos reveal which residues are conserved in Families A–P.

Distinct Family Fingerprints.

Short Starter Unit Acceptors (Family A).

KSs in Groups A1 and A2 accept acetyl and propionyl substrates, respectively. They most notably possess a methionine at Position 22 instead of smaller residues more common in other families. This is most often preceded by an unusual glycine at Position 21. Acetyl and propionyl acceptors possess ATxQ and AMxQ motifs, respectively, at Positions 1–4 of the dimer interface loop. Leucine and threonine are unusually dominant at Positions 10 and 14, respectively. Isovaleryl acceptors (A4) possess similar fingerprints to acetyl and propionyl acceptors (A1 and A2), while the lone member of the methoxyacetyl acceptors (A3) diverges.

β-Ketoacyl Acceptors (Families B–D).

The dimer interface loops of α-unsubstituted, β-ketoacyl acceptors (B) and d-α-methyl, β-ketoacyl acceptors (C) contain xxxH or xxxQ, respectively, in which Positions 1–3 are small residues. TNGQH is common in Positions 1–5 of l-α-methyl, β-ketoacyl acceptors (D). Family D diverges from B and C at a few other positions: phenylalanine is most common at Position 27 (versus threonine), leucine at Position 29 (versus tryptophan), and isoleucine at Position 32 (versus valine/alanine). That Families B and C more commonly possess valine and alanine at Position 32 than isoleucine is unusual among KSs. Families B–D often possess a negatively charged residue at Position 21 and a serine at Position 30. Glutamine at Position 31 is more commonly replaced in Family C (by glycine, serine, or alanine) than other families and is most commonly replaced by a histidine in the diketide acceptors of Group D1.

β-Hydroxyacyl Acceptors (Families E–J).

The most apparent differences between these families are in the dimer interface loops. α-Unsubstituted-d-β-hydroxy acceptors (F) and d-α-methyl-d-β-hydroxy acceptors (I) contain TNGQ or VNGQ motifs, respectively, while l-α-methyl-d-β-hydroxy acceptors (J) contain a TSGQ motif. In Family J, a phenylalanine is most common at Position 29 rather than a tryptophan. The residues in the dimer interface loops of unsubstituted-l-β-hydroxy acceptors (E) and d-α-methyl-d-β-hydroxy acceptors (G) are typically small and unconserved; however, l-α-methyl-l-β-hydroxy acceptors (H) show a strong ASYQ motif. Family H differs from others at several locations: serine is most common at Position 13, proline at Position 15, alanine at Position 18, glycine at Position 25, methionine at Position 27, and leucine at Position 29.

α/β-Enoyl Acceptors (Families K–M).

The VMYH motif is dominant in the dimer interface loops of α-unsubstituted, trans-α/β-enoyl acceptors (K); α-methyl, trans-α/β-enoyl acceptors (L); and cis-α/β-enoyl acceptors (M). Families K and L are very similar, the main differences being that threonine is most common at Position 22 in Family K and that phenylalanine often replaces tryptophan at Position 29 in Family L. cis-α/β-Enoyl acceptors (M) often contain distinctive residues: isoleucine is almost as frequent as aromatic residues at Position 8 and large hydrophobes are often present at Position 11 rather than threonine. The loop and helix that usually span from Position 6 to Position 15 are absent in some KSs; most of these belong to Families K–M and accept substrates containing several conjugated double bonds (Figure S7).

β-Methylene Acceptors (Families N–P).

The α-unsubstituted, β-methylene acceptors (N); the d-α-methyl, β-methylene acceptors (O); and the l-α-methyl, β-methylene acceptors (P) have very similar fingerprints. However, Position 29 in Family P is most commonly a phenylalanine instead of a tryptophan.

Distinct Fingerprints in Closely Related Groups.

KSs that accept diketides were compared with those accepting longer intermediates with the same chemistries at their α- and β-carbons (Figure 5a). One of the biggest differences was the presence of a larger residue at Position 10 in the diketide acceptors. Most often it is a leucine but can also be an isoleucine, methionine, or a threonine. In contrast, the most conserved residue at Position 10 in most groups accepting longer substrates is a glycine. A larger residue at Position 22 is also generally observed in diketide-accepting KSs, as with the starter unit acceptors of Family A. The SAG motif in Positions 13–15 of Families K and L is replaced by TLS and TQG in the diketide-accepting Groups K1 and L1, respectively; the VMYH motif in the dimer interface loop is also usually replaced with VMYD. In some diketide-accepting groups, such as D1, the overall fingerprint diverges markedly.

Figure 5.

Recognition beyond the β-carbon. (a) Sequence logos of KSs that accept diketides and those that accept longer substrates with the same chemistries at their α- and β-carbons are compared. Larger residues at Positions 2, 10, 14, and 22 can make van der Waals interactions with diketide substrates. (b) Sequence logos of KSs that accept substrates with the same chemistries at their α-, β-, and γ-carbons but opposite stereochemical orientation of their γ-methyl groups are compared.

Groups that accept intermediates with γ-methyl substituents of opposite stereochemical orientation but with the same α- and β-chemistries were also compared within each family (Figure 5b). Differences in Family F include Position 2 (tyrosine vs. asparagine), Position 5 (glycine vs. aspartate), Position 10 (leucine vs. alanine), and Position 22 (methionine vs. alanine). Differences in Family G include Position 4 (phenylalanine vs. glutamine), Position 10 (glycine vs. leucine), and Position 29 (tryptophan vs. phenylalanine). Differences in Family H include Position 10 (leucine vs. glycine) and Position 27 (phenylalanine vs. methionine). Differences in Family K include Position 11 (asparagine vs. threonine). Differences in Family P include Position 27 (threonine vs. phenylalanine).

Modeling of Representative KSs Bound to Intermediates.

To learn more about the roles of potential gatekeeping residues, a representative KS from each family was modeled bound to its substrate (Figure 6). Ery(KS3+AT4) (PDB 2QO3) was used as a template to generate homology models; the region downstream of KS until the LPTYxFxxxxxW motif was included because of its structural interaction with KS.¹⁸ The dimer interface loop of EryKS5 (Family P, PDB 2HG4) is in a different conformation from that of EryKS3 (Family G).¹⁹ This conformation is present in most KSs from trans-AT assembly lines, including two double-bond accepting trans-AT KSs (PDBs 4OPE and 4TKT) containing the VMYx motif observed in Families K–M.¹⁰ Since the conserved tyrosine in Position 3 of l-α-methyl-l-β-hydroxy acceptors (H) can interact with the β-hydroxy group if similarly positioned to the VMYx tyrosine, the dimer interface loop of Family H was modeled in the same conformation as the VMYx motif. Thus, the conformation of the dimer interface loop observed in EryKS3 was used for Families A–G and I–J, and the conformation observed in EryKS5 was used for Families H and K–P. The alanine-to-tryptophan (Position 2) point mutant of EryKS3 known to greatly accelerate the condensation of acyl N-acetylcysteamine thioesters with methylmalonyl-ACP was also modeled.²³ The dimeric KSs were energy-minimized with ClusPro 2.0,²⁴ and substrates bound to the reactive cysteines of these KSs were positioned using parameters from structurally determined acyl-KSs (see Experimental Procedures and Supporting Information).^21,25–27

Figure 6.

Models of KSs bound to their substrates. A representative KS from each family was homology modeled and energy minimized before positioning its natural substrate (up to eight carbons) within its substrate tunnel. Residues in the stereodiagrams are colored as in Figure 4. Key hydrogen bonds are indicated with yellow dashes (PDBs provided in the Supporting Information).

Mutation of a Key Glutamine within a Triketide Lactone Synthase.

The models indicate that the side chains of the residues in Position 31 often interact with the β-substituents of substrates. This interaction could help β-hydroxyacyl acceptors select β-hydroxyacyl intermediates over β-keto intermediates. Since almost half of the KSs from the d-α-methyl-β-ketoacyl-accepting Family C contain a glycine at Position 31 and most of the KSs from the β-ketoacyl-accepting Group D1 contain a histidine, these mutations were made to PikKS6 (G) within the engineered triketide lactone synthase Pik167 (Table S2, Figures S7, S8, and S9, Source Data 2).⁸ The activities of these synthases were first observed in vivo within Escherichia coli K207–3 (an engineered strain that converts supplied propionate to methylmalonyl-CoA and contains the PKS-activating phosphopantetheinyl transferase Sfp),²⁸ with the production of triketide lactone being monitored over 1 week. The glycine and histidine point mutants produced ~5-fold less triketide lactone. The in vitro production of pyrone from methylmalonyl-CoA was also monitored over 1 week. The first polypeptide of this synthase as well as the three versions of the second polypeptide were purified from E. coli K207–3 so that the unmutated Pik167 and the two Pik167 point mutants could be incubated with methylmalonyl-CoA in the absence of NADPH. The glycine and histidine mutants produced slightly less pyrone than unmutated Pik167 (1.25- and 2-fold, respectively).

DISCUSSION

KSs may select for properly processed intermediates before performing their more recognized role of extending those intermediates. Because their substrate tunnels are comprised of more than 30 residues at the dimer interface, different sets of amino acids can selectively bind diverse polyketide substrates, as recently characterized in the first module of pyrrolic polyketide assembly lines.²⁹ Since complementarity between a substrate and an enzyme positions reactive groups for catalysis, for the transacylation reaction to optimally transfer a polyketide intermediate from the phosphopantetheinyl arm of ACP to the reactive cysteine of a KS, the chemistries of the acyl-ACP, especially those close to its reactive thioester, should be complementary to features in the substrate tunnel. Gatekeeping would then be explained by incompletely processed intermediates being significantly less complementary to those same features.

Recently, a ketosynthase from an anthroquinone-synthesizing PKS (a type II synthase in which each enzyme is encoded on a separate polypeptide, in contrast to type I modular PKSs) was crystallographically observed bound to a hexaketide intermediate that contains a ketone at every other carbon (PDB 6SMP).³⁰ Most of these are enolized and form hydrogen bonds with one another rather than the ketosynthase. This hydrophilic side of the polyketide faces the dimer interface, while the hydrophobic side interacts with a phenylalanine in an equivalent location to the nearly invariant phenylalanine at Position 26 in the KSs of modular PKSs. In the models of the 16 representative substrate-bound KSs, the substrates are oriented similarly to the hexaketide with their hydroxy and keto groups often forming intramolecular hydrogen bonds.

Families B–D bind substrates with β-keto groups. In d-α-methyl, β-ketoacyl acceptors (C), glycine and serine often substitute for the glutamine usually at Position 31, and in l-α-methyl, β-keto diketide acceptors (D1), a histidine usually substitutes for this glutamine. The substitutions indicate that bound β-keto groups generally face Position 31 and that the hydrophobic side of the polyketide faces the nearly invariant phenylalanine at Position 26. Within the substrate tunnel, the polar keto group needs to coordinate with a hydrogen bond donor. About half of the substrates in Families B–D possess a δ-hydroxy group that could help satisfy this requirement. The others may be more reliant on a hydrogen bond donor from the residue at Position 31, including from the glutamine that usually hydrogen bonds with the backbone carbonyl in Position 30. In the model of EryKS4, the representative KS for Family D, the glutamine in Position 31 does not hydrogen bond with the backbone, freeing its NH₂ to interact with the β-keto group (Figure 6). A TNGQ motif is present in the dimer interface loop of several KSs from Family D, similar to Families F, I, and J. The asparagine NH₂ may also donate a hydrogen bond to the β-keto group. Mutation of this asparagine to alanine in the erythromycin PKS results in a tetraketide shunt product.³¹ With the β-keto group of substrates facing Position 31, KSs can select for the orientation of α-substituents. In Family C the valine that usually substitutes for the isoleucine at Position 32 may permit the binding of d-α-methyl groups. In Family D, KSs select l-over d-α-methyl, β-keto intermediates, the two species being interconverted by an epimerizing C2-type KR.³¹ Since these KSs possess isoleucine at Position 32 and a leucine or phenylalanine at Position 29 rather than the usual tryptophan, a d-α-methyl group would clash with the isoleucine, whereas an l-α-methyl group would fit in the space afforded by the tryptophan to leucine/phenylalanine substitution.

Families F, I, and J bind substrates with d-β-hydroxy groups. The side chain carbonyl of the glutamine at Position 31 could serve as a hydrogen bond acceptor for these hydroxy groups. The dimer interface loops of these families contain a T(N/S)GQ motif, and the asparagine or the serine in Position 2 may also coordinate the d-β-hydroxy group. These interactions would equivalently orient the substrates of these families. A d-α-methyl group would then fit in the KSs of Family I between the tryptophan and isoleucine most commonly in Positions 29 and 32. As some of the largest differences between Families F and I are the dominant leucine and isoleucine residues at Positions 10 and 32 in Family I, these residues may help create a hydrophobic environment for the d-α-methyl group. An l-α-methyl group would fit in the KSs of Family J in the space afforded by a smaller residue that substitutes for the tryptophan usually at Position 29.

Families E, G, and H bind substrates with l-β-hydroxy groups. The α-unsubstituted, l-β-hydroxyacyl acceptors (E) and the d-α-methyl, l-β-hydroxyacyl acceptors (G) have very similar sequence fingerprints, indicating that they bind their substrates equivalently. If the conserved glutamine is coordinated with the backbone carbonyl of Position 30 as in the models, sterics do not permit the l-β-hydroxy group of substrates to coordinate with the glutamine side chain carbonyl. Instead, substrates can bind in orientations similar to those containing d-β-hydroxy groups such that the d-α-methyl groups of the substrates accepted by Family G would project toward a valine most commonly in Position 32. Since KSs from Families E and G make few specific interactions with α- and β-substituents, they may be more tolerant to unnatural substrates. Indeed, the Family G members EryKS2, EryKS3, EryKS6, and PikKS6 accepted the broadest range of diketides in a panel of engineered triketide synthases.² l-α-methyl, l-β-hydroxyacyl acceptors (H) are markedly different from Families E and G and may bind their substrates in an uncommon orientation. Sterics permit the l-β-hydroxy group to coordinate both the side chain carbonyl of the glutamine at Position 31 and the hydroxyl group of a conserved tyrosine at Position 3. This substrate orientation places the l-α-methyl group between the side chains of residues in Positions 2, 3, and 32, which are most commonly serine, tyrosine, and isoleucine. The γ-carbon would be located in the space afforded by a smaller residue substituting for the tryptophan usually in Position 29.

Families K–M bind intermediates with α/β-double bonds. The conserved VMYH motif in the dimer interface loop may be in the same conformation observed in structurally characterized KSs from trans-AT assembly lines (PDBs 4OPE and 4TKT).¹⁰ Together, the distal features of the methionine and tyrosine residues can make van der Waals interactions with planar α/β-double bonds, while sterically preventing β-ketoacyl and β-hydroxyacyl substrates from forming hydrogen bonds with the conserved glutamine at Position 31. Space for the α-methyl groups is provided in almost half of the KSs in Family L through a phenylalanine that replaces the tryptophan usually in Position 29. Of the KSs analyzed here, 85% with a 14-residue deletion between Positions 6 and 15 belong to Families K–M, apparently to accommodate long, conjugated substrates (Figure S7).

Families N–P bind substrates with β-methylenes. Their substrate tunnels are coated with methylenes and methyl groups from residues such as alanine, valine, isoleucine, leucine, and methionine in Positions 2, 10, 22, and 32. Their dimer interface loops consist of unconserved small residues and could assume conformations in which the side chains are closer to bound substrates, as crystallographically observed in EryKS5.¹⁹ Intermediates containing a β-keto group would not bind well due to the general absence of hydrogen bond donors, and intermediates containing a d-β-hydroxy group would not bind well since no hydrogen bond donor is in Position 2 as in Families F, I, and J. Intermediates containing an α/β-double bond would not make shape-complementary interactions with the side chains of the small residues in the dimer interface loop. As with the α-methyl groups of substrates from other families, d-oriented methyl groups would face Position 32, which is usually an isoleucine in Family O, and l-oriented methyl groups would face Position 29, which is usually a phenylalanine in Family P that substitutes for a tryptophan.

Although the most significant fingerprints can be seen in KSs that gatekeep for chemistries at the α- and β-carbons, chemistries at other carbons may influence substrate tolerance as well. A set of larger, more hydrophobic, residues is usually present in KSs that accept short substrates such as starter units (Family A) and diketides. These are most commonly in Positions 2, 10, 14, and 22 and likely interact with substrates through van der Waals interactions. An alanine-to-tryptophan substitution at Position 2 of EryKS3 has been shown to greatly increase the rate that small acyl N-acetylcysteamine thioesters condense with methylmalonyl-ACP.¹⁹ Based on the model of this point mutant, the tryptophan is positioned to make significant van der Waals interactions with the short substrates. Sequence fingerprints can differ between KSs that accept substrates with opposite stereochemistry at the γ-carbon. As in KSs that accept diketide substrates, a leucine usually replaces a smaller residue at Position 10 in Groups F5, G6, H6, I7, and J5. As gatekeeping beyond the β-carbon could hamper the engineering of new assembly lines, such residues may need to be mutated to increase tolerance to unnatural substrates.

Polyketide assembly-line engineers should take note of gatekeeping features that affect KS substrate tolerance. The first KSs of synthases that accept acetyl or propionyl units will not be suitable in the downstream modules of an engineered synthase. The same is true for the second KSs that accept relatively short diketide substrates. Scrutiny over whether a KS naturally binds features downstream of the β-carbon should be applied. The analysis here indicates that residues at some positions (e.g., Position 10) can make substantially different interactions with substrates containing opposite stereochemistry at the γ-carbon. Engineers can learn from the highly similar KSs in synthases such as the mycolactone and rapamycin PKSs that are tolerant of diverse substrates.^32,33 With the exception of Position 29 not being a tryptophan in KSs accepting l-α-methyl groups, the substrate tunnel residues of these KSs do not vary much with respect to the substrates they accept. Such KSs may rely on less specific van der Waals interactions between substrates and a large hydrophobic residue in Position 2 in the dimer interface loop (tryptophan in mycolactone KSs, phenyl-alanine/tyrosine in rapamycin KSs; Figure S9). This may be a general strategy for accepting substrates with challenging substituent patterns. For example, the four KSs from the rapamycin, tautomycin, and tautomycetin synthases use this less specific strategy to accept the Group F5 substrate, in contrast to the 14 KSs that employ a TNGQ motif in their dimer interface loops to recognize the apparently more accessible Group F6 substrate that differs in the orientation of its γ-methyl substituent. Mutagenesis has been employed to identify residues that increase the substrate tolerance of KSs within engineered synthases. A combination of mutations at Positions 22 and 29 in EryKS6 (Family G) increases its tolerance for enantiomeric diketide intermediates up to 2.5-fold.³⁴

The replacement of the glutamine at Position 31 of PikKS6 (Family G) in the triketide synthase Pik167 indicates that substrate binding tunnels are sensitive to alteration. Even though this glutamine does not appear to form a hydrogen bond with the l-β-hydroxy group of the presented substrate, its mutation to a glycine or histidine changes the shape and chemistry of the tunnel enough to result in ~5-fold decreases in activity. While glycine and histidine are located at Position 31 in several KSs that accept β-keto groups, these residues did not enable PikKS6 to bind a β-keto substrate. They likely collaborate with other residues (e.g., Positions 2 and 32) within their native KSs. Assembly-line engineers may find it preferable to employ a KS that naturally binds a desired substrate rather than engineer one through site-directed mutagenesis.

Microbes often make significant genomic and metabolic investments in modular PKSs.³⁵ As these assembly lines contain tens to hundreds of enzymatic domains, checkpoints are important to ensure that the biologically active final product is being faithfully synthesized. The enzyme mediating the key carbon–carbon bond-forming reaction also takes on the role of gatekeeper. Positioned at the downstream end of each module, KSs can validate that the upstream processing enzymes have performed their reactions. To harness these synthetic factories and generate pharmaceutically relevant compounds, assembly-line engineers must understand the recognition that takes place within their substrate tunnels. This study illuminates these crucial interactions and further enables our collaboration with some of the best organic chemists in the natural world.

EXPERIMENTAL PROCEDURES

KS Sorting and Sequence Logos.

The sequences of 972 KSs (and the downstream region including AT and the LPTYxFxxxxxW motif) were obtained from 129 characterized cis-AT assembly lines (see Supporting Information and Source Data 1). As in a previous study, they were named by the polypeptide in which the KS-containing module starts, the number of the module within that polypeptide, and the α–δ (a–d, or z for all others) architecture of the module (Supporting Information and Table S1).¹⁵ To focus on the gatekeeping logic employed by PKSs primarily from actinomycetes, 233 KSs were excluded from the analysis (Figures 2 and S1–S3). The remaining 739 KSs from 92 assembly lines were sorted into families and groups (Figures 3 and S4). KSs were further identified by their group name and whether they are associated with a docking domain (DD).³⁶ Sequence logos of 32 substrate tunnel residues were generated for each family using a combination of Clustal X 2.0 and the WebLogo server (Figure 4).^37,38

Model Generation.

Sixteen representative KSs were selected based on how well their sequences match the sequence logo of their family. These KSs and their structurally connected, downstream ATs were homology modeled using the structure of Ery(KS3 + AT4) (PDB 2QO3) as a template in ProtMod.³⁹ As each KS+AT didomain is homodimeric, all-atom SCWRL models were generated for each monomer.⁴⁰ To remove steric clashes, energy minimization was performed on the dimers using the protein–protein docking server ClusPro 2.0.²⁴ The CCP4 extension AceDRG was employed to create a restraint dictionary for each ligand (the first eight carbons of chains being used for longer substrates).⁴¹ Parameters from structurally determined acyl-KSs (PDBs 2BUI, 2GFY, 2IX4, 6ROP) helped position substrates on the reactive cysteines with the program Coot.^{21,25–27,42} The hydrogen bond between the thioester and the amide of Position 32 was maintained between 2.7 and 3.1 Å (from oxygen to nitrogen), and the C_α–C_β torsion angle of the reactive cysteine was maintained between 72 and 82°. The tail of each polyketide was positioned similar to the aforementioned acyl-KSs, and substituents were placed next to features of the KSs with complementary shape and chemistry (PDB files of acyl-KS representatives from each family and the EryKS3 point mutant are provided in the Supporting Information).

General Analytical Procedures.

Samples were analyzed by HPLC using a Waters 1525 HPLC system fitted with a Microsorb-MV 300–5 C₁₈ column (4.6 × 250 mm). A solvent system of H₂O and CH₃CN (0.1% v/v formic acid) was used with a flow rate of 1 mL min⁻¹. After equilibration with 5% CH₃CN, a linear gradient from 5 to 100% over 15 min was run, followed by 100% CH₃CN for 3 min. High-resolution mass spectra were obtained on a 6230 TOF LC/MS scanning from m/z 50 to 3200 in positive ionization mode. Size-exclusion and ion-exchange chromatographies were carried out using a Biologic DuoFlow (Bio-Rad) FPLC system.

Synthase Construction, Expression, and Purification.

The second expression plasmids of the Pik167 glycine and histidine point mutants were generated using mutagenic primers and SLiCE assembly (Table S1).⁴³ The two plasmids encoding Pik167 and each of its point mutants were cotransformed into E. coli K207–3. Cells were grown in Luria–Bertani Miller Broth with 50 mg L⁻¹ kanamycin at 37 °C until OD₆₀₀ = 0.6. Cells were induced with 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG) and shaken at 15 °C for 16 h. They were then collected (4000g, 25 min), resuspended in lysis buffer [100 mM potassium phosphate, 400 mM NaCl, 5 mM tris(2-carboxyethyl)-phosphine (TCEP), pH 7.5], sonicated, and centrifuged (20 000g, 40 min). The Ni-NTA affinity resin was washed with 20 mL of lysis buffer and incubated with the supernatant for 16 h at 4 °C. After batch binding, the resin was washed twice with 50 mL of lysis buffer and once with 50 mL of lysis buffer containing 5 mM imidazole. The proteins were eluted from the beads by stirring them with 10 mL of lysis buffer containing 250 mM imidazole for 30 min. They were then concentrated using Amicon Ultra centrifugal filters (Merck), and the buffer was exchanged for 400 mM potassium phosphate, 150 mM NaCl, 1 mM TCEP, pH 7.5. Based on SDS-PAGE analysis, the first polypeptide common to each synthase as well as the second polypeptide of the histidine point mutant required further purification. Thus, both were polished through ion-exchange chromatography [50 mM HEPES, 1 mM TCEP, pH 7.5, 0–1 M NaCl over 1 h on a HiTrap Q HP (5 mL)] and size-exclusion chromatography [50 mM HEPES, 150 mM NaCl, 1 mM TCEP, pH 7.5, at 0.5 mL min⁻¹ on a Superdex 200 Increase 10/300 GL]. All proteins were concentrated to 10 mg mL⁻¹, flash frozen, and stored at −80 °C until needed (Figure S7).

In Vivo Production of Triketide Lactone.

A single E. coli K207–3 colony was used to inoculate 50 mL of Luria–Bertani (LB) media containing 100 mg L⁻¹ kanamycin and 100 mg L⁻¹ streptomycin. After 16 h of growth at 37 °C and 250 rpm, 1 L Erlenmeyer flasks containing production media [200 mL total; 5 g L⁻¹ yeast extract, 10 g L⁻¹ casein, 15 g L⁻¹ glycerol, 10 g L⁻¹ NaCl, and 100 mM potassium phosphate buffer (pH 7.6)], 100 mg L⁻¹ kanamycin, and 100 mg L⁻¹ streptomycin were inoculated with 10 mL of culture. When OD₆₀₀ = 1.0, cells were cooled to 19 °C and both 20 mM sodium propionate and 100 μM IPTG were added. Cultures were incubated at 19 °C at 240 rpm for 1 week. Each day, a sample (1.0 mL) was acidified with 50 μL of 6 M HCl, the resulting insoluble material was removed by centrifugation (15 000g, 10 min), and the supernatant was extracted three times with 0.5 mL of ethyl acetate. The organic layers were concentrated in vacuo, and 0.7 mL of methanol was added for HPLC analysis (Figure 7 and Data Source 2).

Figure 7.

Mutation of a conserved tunnel residue. Activities of the engineered triketide synthase Pik167 and its point mutants, Q31G and Q31H (glutamine in Position 31 of PikKS6, marked with asterisk, was mutated to glycine and histidine). (a) Triketide lactone production from E. coli K207–3 transformed with the expression plasmids of each synthase was monitored by LC/MS. Unmutated Pik167 produces ~5-fold more triketide than either mutant. (b) Purified polypeptides of each synthase were incubated with (2RS)-methylmalonyl-CoA, and pyrone production was monitored by LC/MS. Unmutated Pik167 produces 1.25- and 2-fold more pyrone than the glycine and histidine point mutants, respectively. Data was collected from triplicate experiments and averages are shown.

In vitro Production of Pyrone.

Reactions (1 mL) were performed at 25 °C in a microcentrifuge tube containing 400 mM potassium phosphate, 5 mM TCEP, 0.2 mM (2RS)-methylmalonyl-CoA, and 10 μM of each protein at pH 7.5. TCEP (1 mM) was supplied daily to offset its degradation in phosphate buffer. Samples (100 μL) were quenched with 6 M HCl (5 μL), and the resulting insoluble material was removed by centrifugation (15 000g, 10 min). The supernatants were extracted twice with 500 μL of ethyl acetate, concentrated in vacuo, and redissolved in 200 μL of methanol for HPLC analysis (Figure 7 and Source Data 2).