Crystal structure and biochemical studies of the trans-acting polyketide enoyl reductase LovC from lovastatin biosynthesis

Abstract

Lovastatin is an important statin prescribed for the treatment and prevention of cardiovascular diseases. Biosynthesis of lovastatin uses an iterative type I polyketide synthase (PKS). LovC is a trans-acting enoyl reductase (ER) that specifically reduces three out of eight possible polyketide intermediates during lovastatin biosynthesis. Such trans-acting ERs have been reported across a variety of other fungal PKS enzymes as a strategy in nature to diversify polyketides. How LovC achieves such specificity is unknown. The 1.9-Å structure of LovC reveals that LovC possesses a medium-chain dehydrogenase/reductase (MDR) fold with a unique monomeric assembly. Two LovC cocrystal structures and enzymological studies help elucidate the molecular basis of LovC specificity, define stereochemistry, and identify active-site residues. Sequence alignment indicates a general applicability to trans-acting ERs of fungal PKSs, as well as their potential application to directing biosynthesis.

Atherosclerosis is the current leading cause of death for adults in the western world (1). Statins inhibit cholesterol biosynthesis and are the most widely prescribed drugs for the prevention and treatment of atherosclerosis (2). Lovastatin (compound 1) is the first statin approved by the Food and Drug Administration and is the direct precursor for manufacture of simvastatin, the second most-prescribed drug worldwide (1). The high impact of statins to human health, including their possible use for cancer and neurodegenerative diseases (3), has prompted vigorous efforts toward synthesis of lovastatin (1). However, the structural complexity of lovastatin has prevented its commercial production by total chemical synthesis (1). In nature, lovastatin is biosynthesized by the fungus Aspergillus terreus (4) using a gene cluster that contains two polyketide synthases (PKSs), LovB and LovF (Fig. 1A) (5). Based on past biochemical studies of lovastatin biosynthesis, our previous bioengineering efforts led to rational generation of lovastatin analogs with enhanced activities and reduced side effects (6). Despite the above advances, many questions about lovastatin biosynthesis remain unanswered. Understanding lovastatin biosynthesis represents an opportunity to leverage our knowledge for future efforts in directing biosynthesis of statin analogs.

Fig. 1.

(A) Proposed lovastatin biosynthetic pathway with 1-3 shown in their acidic forms. (B) Three putative substrates (tetra-, penta- and hepta-ketides) of LovC. (C) In vitro substrates. NADPH (compound 4) and NADH (compound 5) are discussed in the text.

During lovastatin biosynthesis, the iterative type I PKS LovB builds the nonaketide main framework with the aid of the trans-acting enoyl reductase (ER) LovC (4, 7). A second type I iterative PKS, LovF, constructs the 2-methylbutyryl side chain. LovB and LovF are both large multidomain megasynthases that are highly homologous in sequence and domain arrangements (Fig. 1A). When coupled with LovC, LovB iteratively catalyzes more than 30 precisely synchronized reactions to yield a 19-carbon intermediate dihydromonacolin L (DmL) (compound 2) (Fig. 1A). In comparison, LovF only catalyzes one round of Claisen condensation to produce a 2-methylbutyryl intermediate, which is attached to the C8 hydroxyl of monacolin J (compound 3) to produce lovastatin. LovC specifically interacts with only LovB, but not LovF, and accepts only three out of eight possible LovB intermediates as its substrates (Fig. 1A; tetra-, penta-, and heptaketides). Similar trans-acting ERs have also been reported in other fungi such as MokE in Monascus pilosus (8), MlcG in Penicillium citrinum (9), ApdC in Aspergilus nidulans (10–12), cytochalasans (12, 13), and tenellin_ORF3 in Beauveria bassiana (14). These trans-acting ERs are highly homologous (38–80% sequence identity; Fig. S1A) and are observed to interact with megasynthases through specific protein–protein interactions (4). However, how LovC specifically interacts with LovB and why it reduces only at the tetraketide, pentaketide, and heptaketide stages is not well understood.

Herein, we report the crystal structure of LovC, which represents the structure of a polyketide ER from an iterative reducing Type I PKS. LovC is a 39.5-kDa protein (363 residues) and, surprisingly, is shown to be a unique monomeric member of the medium-chain dehydrogenase/reductase (MDR) superfamily (15). Two cocrystal structures are presented: the LovC–crotonoyl CoA complex and the LovC-NADP⁺ binary complex. To study the trans-acting ERs and their unique substrate specificity, an in vitro assay is essential, and we report the successful development of LovC enzyme assays that may be applied to other trans-acting ERs. We elucidated the structural basis of LovC substrate specificity and found that LovC mutations can alter its substrate specificity. The results can be leveraged with our previous work using trans-acting ERs to biosynthesize new statin analogs such as the compactin precursor (5), and the chemical diversity can potentially be further expanded by using other trans-acting ERs working in conjugation with the megasynthases for directing biosynthesis.

Results and Discussion

Overall Structure: LovC Is a Unique Monomeric ER and MDR Member.

Unlike previously reported ER domains [Protein Data Bank (PDB) ID codes 1GUF and 1ZSY], LovC exists as a monomer as detected from the crystal structure (Fig. 2) and size-exclusion chromatography (Fig. 3). The monomeric state of LovC is highly unique in the MDR superfamily (15), the members of which are functionally dimeric or tetrameric, except for the monomeric mannitol dehydrogenase (16). The MDR superfamily includes the zinc-containing alcohol dehydrogenases (ADHs), leukotriene dehydrogenases, quinone oxidoreductases, the eukaryotic ERs, and membrane-sensing proteins (15). Because none of the MDR proteins of known structures shares significant sequence homology to LovC, we could not solve the LovC structure by molecular replacement. Subsequently, the LovC structure was solved by multiwavelength anomalous dispersion (MAD) technique using selenomethionine-derived protein (Table S1). The LovC structure contains two domains: the catalytic domain (residues 1–134, 283–363) and cofactor-binding domain (Rossmann fold; residues 135–282) (Fig. 2A). The structural comparison between LovC and selected MDR proteins (18–23% sequence identity) shows 1.6–2.4 Å rms deviation (RMSD) between the aligned Cα coordinates (Figs. S1B and S2). All are dimers except LovC. The least conserved region is located between α2 and βA. The principal difference between LovC and these homologs lies in two transition regions between the catalytic and cofactor-binding domains, consisting of α2-αA-loop and α3-loop-α4. LovC has an additional 10 or more residues in each loop (xL1 and xL2; Fig. 2A and Fig. S1B). The extended lengths of these two loops are a unique conserved feature of trans-acting ERs associated with fungal PKSs (Fig. S1A), and, as discussed below, the extended loops may influence the oligomeric state and protein–protein interactions of LovC during lovastatin biosynthesis.

Fig. 2.

(A) Overall fold and quaternary structure of LovC. Two extended loops, xL1 and xL2, are unique to the LovC structure. (B) Comparison of LovC-NADP⁺ (blue) vs. LovC-CO7 (yellow) structures showing the major change at Y296. (C) Comparison of both models (rendered as electrostatic surface) reveals that the protein surface has significant changes. (D) Zoomed view shows that the active site does not change.

Fig. 3.

Gel-filtration interaction study indicative of LovB-LovC complex formation. Inset, SDS/PAGE confirmed interaction: M, marker; 1, purified LovB (335 kDa, doublet resulting from the loss of an N-terminal KS-containing fragment); 2, purified LovC (40 kDa); F9, fraction 9 from the combined run.

The LovC catalytic domain has three β-sheets: the anti-parallel Beta1 (β1-β2), the anti-parallel Beta2 (β5-β4-β6), and the six-stranded Beta3 (β7-β5-β4-β3-β9-β8) (Fig. 2A). The LovC cofactor-binding domain has two copies of the β−α−β−α−β Rossmann fold, in which βA−αB−βB−αC−βC forms a mirror image with βD−αD−βE−βF (Fig. 2A). In dimeric MDR proteins such as the porcine fatty acid synthase (FAS) ER (17) and quinone oxidoreductase (18), the dimer interface includes the backbone hydrogen bonding across βF from each monomer to form a characteristic 12-stranded β-sheet, as well as intermonomer interactions between αF, βE, and βF (Fig. S2). Importantly, computer modeling to create the LovC-LovC homodimer (based on other MDR dimer structures) showed that the presence of the extended xL2 loop (Fig. 2A and Fig. S2) generates significant steric clash with its dimeric partner. The xL2 loop does not exist in dimeric MDR proteins but is conserved in trans-acting ERs associated with iterative fungal PKSs such as MlcG (9), MokE (8), ApdC (10), and tenellin synthase ORF3 (14) (Fig. S1A). Therefore, the steric hindrance caused by the conserved xL2 helps explain why LovC is isolated as a monomeric protein in vitro, and the monomeric state may be a conserved feature shared among the trans-acting ER domains of fungal PKSs. LovC is also the first monomeric MDR structure reported (19).

Interactions Between LovC and the LovB Megasynthase.

LovC, a trans-acting ER, may need to remain monomeric to interact with the LovB megasynthase, which has an inactivated ER⁰ domain (residues 1,851–2,250) attributable to the deletion of key active site residues (Fig. S3) (4). Complex formation between LovC and LovB has been hypothesized previously (4). We show by size-exclusion chromatography that in the absence of an in vitro substrate, LovB and LovC coelute, confirming the formation of LovB-LovC complex (Fig. 3). LovC likely interacts with LovB by forming a heterodimer with LovB-ER⁰. However, no complex formation was detected when ER⁰ was cloned, expressed, purified, and incubated with LovC. Therefore, to form the LovC-LovB complex, LovC may need to interact with not just the ER⁰ but multiple domains of LovB [such as the LovB acyl-carrier protein (ACP), dehydratase (DH), MT (methyl transferase), or KR (ketoreductase) domains]. Nevertheless, Fig. 3 represents direct evidence for the association between a trans-acting ER with its megasynthase partner.

Cofactor-Binding and Enzyme-Activity Assays Show That LovC Is an NADPH-Dependent ER.

To determine the cofactor-binding specificity of LovC, we used fluorometric titration with NADPH (compound 4) or NADH (compound 5) (Fig. S4 A and B) (20), and we found a strong preference to bind NADPH. To characterize LovC activity, we searched for in vitro substrates resembling the three putative LovC substrates (Fig. 1B) by ChemDB (21), and we identified NCI-636688 (compound 6) and NCI-636689 (compound 7) that can be reduced by LovC in the presence of NADPH but not NADH (Fig. S4). To explore substrate specificity, we synthesized (E)-2-butenoyl-acyl-N-acetylcysteamine [(E)-2-butenoyl-NAC] (compound 8), (2E,4E)-hexadienoyl-NAC (compound 9), and (E)-2-octenoyl-NAC (compound 10) to mimic the α,β unsaturated di-, tri, and tetraketide intermediates, respectively, and (E)-2-octenoic acid (compound 11) as a “NAC-less” analog of compound 10 (Fig. 1C). The fact that compound 10 can be reduced but not compound 11 supports the importance of the NAC moiety for substrate binding (Table 1 and Fig. S4). For the NCI compounds 6 and 7, the t-butyl group may serve a similar anchoring/orienting purpose as the NAC moiety. The fast turnover rate of compounds 6 and 7 provides a facile assay for ERs in general. These results show that recombinant LovC is enzymatically active as a standalone protein and support that NADPH is its cofactor.

Table 1.

Summary of fluorometric activity assay kinetic parameters for WT and mutant LovC

Sample	Km (μM)	kcat (s−1)	kcat/Km (s−1 μM−1)
Substrate: NCI-636688
WT	540	0.048	8.9 × 10−5
S51A	1,255	0.12	9.5 × 10−5
K54S	63	0.002	3.1 × 10−5
S51A/K54S	142	0.009	6.5 × 10−5
A93M	10	0.023	3.8 × 10−5
S138M	55	0.0007	1.1 × 10−5
T139V	21	0.0014	6.4 × 10−5
N263A	254	0.037	14.6 × 10−5
N263S	247	0.018	7.1 × 10−5
Substrate: octenoyl-SNAC
WT	452	0.0034	0.75 × 10−5
S51A	290	0.0055	1.89 × 10−5
K54S	1,023	0.0009	0.088 × 10−5
S51A/K54S	446	0.0044	0.98 × 10−5
T68V	466	0.012	2.58 × 10−5
A93M	50	0.0012	2.40 × 10−5
S138M	137	0.0006	0.44 × 10−5
T139V	111	0.0013	1.17 × 10−5
N263A	24	0.0008	3.33 × 10−5
N263S	102	0.0012	1.18 × 10−5

NCI, National Cancer Institute. SNAC, S-N-acetylcysteamine.

Open and Closed Conformations of LovC.

The cocrystal structures of LovC-isomerized 2′-phosphate crotonoyl-CoA (referred to as LovC-CO7; detailed in SI Text) and LovC-NADP⁺ are highly similar, sharing an RMSD of 0.45 Å for C_α atoms (Fig. 2B). Interestingly, upon binding with LovC, the 3′-phosphate of crotonyl-CoA is isomerized to 2′-phosphate (Fig. S5A), which provides a ligand-binding mode that is remarkably similar to the cofactor, with near perfect overlap for the 2′-phosphoadenosine moiety of both the bound CO7 ligand and NADP⁺. There are many previous reports of acid/base catalyzed 3′ to 2′-phosphate isomerization of an acyl-CoA (22–25), and it is likely that charged residues near the cofactor ribose binding site may facilitate the isomerization of crotonyl-CoA (Fig. S5).

The major conformational difference between the two cocrystal structures lies at the xL2 loop (residues 290–301), where the LovC-NADP⁺ structure adopts an open conformation with Y296 pointing toward the solvent, whereas the LovC-CO7 structure adopts a closed conformation with Y296 pointing toward the active site (Fig. 2B). The conformational change also significantly alters the surface potential between the two structures around the xL2 region (Fig. 2C) but does not alter the interior portion of the putative substrate-binding pocket (Fig. 2D). The open-closed conformations serve as an indicator for the flexible nature of xL2, so that it can be adaptive for docking with the upstream ACP for subsequent substrate transfer. Indeed, a docking simulation between a homology model of LovB ACP and LovC identified the LovB ACP-LovC docking site near this region (Fig. S6 C–E). The docking simulation also indicates that multiple residues are involved for ACP-LovC interactions. Supporting this prediction, single mutations of xL2, including W292A, Y296A and R298A, do not change the LovB-LovC interactions, as detected by size exclusion chromatography (Fig. S7E). Both xL2 and other regions predicted for LovB ACP-LovC docking are conserved among trans-acting ERs from fungal PKSs (Fig. S1A).

NADP⁺ Binding.

The LovC-NADP⁺ cocrystal structure identifies the cofactor-binding cleft in between the catalytic and Rossmann fold domains (Fig. 2A). Structure and sequence analysis of NAD(P)-binding enzymes has identified a “fingerprint” region (detailed in SI Results and Discussion) (26). For LovC, this region contains the motif GXXTXXA (Fig. S1B), as well as the βB to αC loop, both are important for NADPH binding and are highly conserved for trans-acting ERs. There is substantial interaction between LovC, the adenosine-phosphate, and nicotinamide-ribose moieties (Fig. S5B). The 2′-phosphate of NADP⁺ is situated in an electropositive pocket, whereas the adenine is sandwiched between the highly conserved residues Y215 and I239. Two residues, K54 and N263, form a bridge directly above the nicotinamide ribose, with the side chains of each forming two hydrogen bonds with the ribose hydroxyls. K54 is highly conserved in the trans-acting ERs, and the LovC-NADP⁺ structure showed that it is important for cofactor binding. The highly conserved nature of the NADP-interacting residues indicates that the cofactor-binding motif should be conserved among fungal trans-acting ERs.

NADPH Stereochemistry.

In previous studies of the FAS ER domains, a transfer of the NADPH pro-4R hydride can result in either syn- (in animal and E. coli) or anti- (in yeast) addition to produce [3R, 2S] or [3R, 2R] reduced fatty acids (27–29). It is interesting to note that the cryptic stereochemistry of fatty acid ER and PKS ER can be opposite in the same fungal organism (30, 31). Because the si-face of the cofactor NADPH is blocked by highly conserved active site residues (Fig. S5 B and C), only the pro-R hydride of NADPH can be transferred to the substrate. This conclusion is consistent with a recent stereochemical study of crotonyl-CoA carboxylase/reductase (32), as well as previous analyses of nicotinamide-binding proteins (33), that when NADPH adopt the anti-conformation, the re face of the cofactor is exposed and the pro-R hydride is transferred. Thus, the LovC structure provides a structural basis for the stereospecificity of the trans-acting ERs.

Active Site and Substrate-Binding Pocket.

The putative LovC substrate-binding pocket can be identified from the LovC-CO7 structure, in which the four-carbon crotonoyl group extends into a hydrophobic pocket that is adjacent to the position of the nicotinamide ring of LovC-NADP⁺ (Fig. 2D and Fig. S5A). The binding pocket is defined by residues from α1, α3, and the loops between α1−β4 and β5−β6. At the entrance of the pocket, the K54-N263 electropositive bridge could serve as a docking point for the phosphate in PPT group of the ACP-tethered polyketide intermediate (Fig. 4A). Because K54 is highly conserved (Fig. S1A), the ACP docking point and the PPT docking entrance may also be conserved among trans-acting ERs in fungal PKSs.

Fig. 4.

(A) Docking results for the α,β unsaturated heptaketide intermediate attached to PPT. (B) Stereoview of the polyketide-binding pocket. (C) General hydride transfer mechanism proposed for LovC.

To dissect the possible substrate-binding modes, we conducted docking simulations of the putative LovC substrates (α,β unsaturated tetra-, penta-, and heptaketide) to the LovC active site using Gold (34). The results for all three substrates are highly consistent, placing the polyketide moiety in the pocket defined by N49, S51, A93, A135, S138, T139, L142, G282, P283, I285, and F286, whereas the PPT group loops toward the outer surface of the pocket to dock the PPT-phosphate over K54 (Fig. 4 A and B). The putative substrate-binding pocket is large enough to accommodate all putative LovC substrates, and places the Cβ of each within hydride-transfer distance to the C4 of NADP⁺. The docking simulation also identifies a possible oxyanion hole to accommodate the negative charge developed on C=O after hydride transfer. This can be either the side chains of N49, S51, T68, T139, K54, N263, or the NH of G282 (Fig. 4 A and B). These residues are conserved among trans-acting ERs, implying a conserved catalytic mechanism.

Interestingly, MDR sequence or structure alignment does not reveal a conserved catalytic motif among MDR proteins. The active site Tyr proposed for MDR homologs such as PDB ID codes 1GUF (35), 1QOR (36), and 2OBY (37) corresponds to F60 in LovC (Fig. S6A). Therefore, based on docking result and sequence conservation in trans-acting ERs associated with fungal PKS, we proposed that S51, K54, T68, and N263 may serve as the oxyanion hole.

Mutations Identified Key Catalytic and Substrate-Binding Residues.

To determine the roles of substrate pocket residues, we generated two sets of mutants: S51A, K54S, S51A/K54S, T68V, T139V, N263A, and N263S to identity catalytic site(s), and A93M and S138M to decipher substrate-binding residues (Table 1). The mutant proteins were expressed, purified, and subjected to the assays of cofactor binding, fluorometric activity, and in vitro reconstitution to screen for the production of DmL. All mutants displayed an increase in fluorescence when titrated with NADPH, indicating that they are all competent in cofactor binding.

The steady-state kinetic data obtained from the fluorometric activity assay using NCI-636688 or (E)-2-octenoyl-NAC as substrates are summarized in Table 1. With NCI-636688, the k_cat of K54S and T139V are 24- and 35-fold less than WT, respectively, whereas the k_cat of S51A is 2.5-fold greater than WT. Although less drastic, the data from (E)-2-octenoyl-NAC show a similar trend in activity for the mutants compared with WT, supporting that the observed kinetic constants are substrate independent and mutation dependent.

The increased k_cat of S51A came as a surprise because this is the polar residue nearest the position of the thioester carbonyl in the docked substrates (and, therefore, a likely candidate for stabilization of the oxyanion following hydride transfer) (Fig. 4B). The rate increase may be attributable to enhancement of a step other than the catalytic step, such as product release. For the pocket mutants N263A, and N263S, A93M and S138M, we observed a change of relative substrate specificity (Table 1). The result supports that altering the substrate pocket topology by mutagenesis can change substrate specificity (detailed in SI Text) and is a proof of principle for future endeavor to apply synthetic substrate mimics to biosynthesize new analogs. Finally, the in vitro reconstitution of DmL biosynthesis by combining LovB, LovC, and substrates demonstrates that all mutants are enzymatically active and capable of interacting productively with LovB (Fig. S6B).

The greatly diminished, but not abolished, activity of K54S and T139V is of particular interest. A structurally equivalent lysine to K54 in SDR enzymes is crucial for cofactor binding (38) and catalysis (39), whereas the side chain hydroxyl of T139 lies within 3.2 Å from the C4 of the nicotinamide (the site of hydride transfer) (detailed in SI Text). As a control, we solved the 1.7-Å crystal structure of LovC-K54S bound with NADP⁺ and generated the S51A/K54S mutant. The LovC K54S structure confirms no significant conformational change from that of the wild-type (RMSD C_α of 0.21 Å; detailed in SI Results and Discussion and Fig. S7 A–D). Furthermore, the turnover rate and binding affinity of S51A/K54S displayed additive effect of the single mutants, S51A and K54S. Therefore, the low K54S activity is attributed directly to the loss of functionality provided by the K54 side chain. The above result agrees with the recent mutational study by the Leadlay group on the modular type I PKS ER domain, which also concludes that no single residue serves as oxyanion hole or Brønsted acid (40). Rather, a combination of multiple side-chains, waters, or the backbone NH of G282 may all contribute to enzyme catalysis, a unique feature observed for PKS ER domains.

Proposed Mechanism for the Observed LovC Substrate Specificity.

Based on the above analyses, we propose the following mechanism to interpret the observed LovC substrate specificity for the putative α,β unsaturated tetra-, penta-, and heptaketide intermediates: For an ER, the cofactor NADPH is proposed to bind before substrate binding (41). LovC then gains proximity to the LovB megasynthase through protein–protein interaction (Fig. 3 and Figs. S6 and S8), which may occur by forming a heterodimer with the LovB ER⁰ domain and extra help from interactions with other LovB domains, such as ACP. The LovB-ACP is then docked to the electropositive surface of LovC near K54, and the polyketide substrate is delivered from ACP to the LovC substrate pocket. The pro-R hydride of NADPH is then transferred to C3 of the polyketide alkene group (Fig. 4C), and the resulting enolate oxide is stabilized by the oxyanion hole (candidate residues include K54 and amide NH of G282). The enolate collapses and proton transfer from a Brønsted acid (likely water) to C2 completes enoyl reduction. Substrate docking indicates that because of the pocket size limitation, LovC cannot accommodate polyketides larger than the heptaketide intermediate (Figs. 1B and 4 A and B). On the other hand, although the smaller di- and triketide intermediates do fit in the active site, the shorter di- and triketide may adopt multiple, nonproductive orientations, as demonstrated by the docking result of these shorter ketides, the double bonds of which restrict their flexibility and prevent a proper orientation in the LovC active site, thus discouraging productive binding for enoyl reduction. Furthermore, chain elongation by KS may compete for the shorter ketides. In comparison, the longer tetra- and pentaketide intermediates (the first two putative substrates for LovC) can be extended to the back of substrate pocket, promoting more significant hydrophobic contact to anchor the polyketide, thus resulting in the productive-binding mode. In addition, the hexaketide intermediate is strongly driven to undergo Diels–Alder cyclization rather than enoyl reduction. In conclusion, LovC structural analyses help explain substrate specificity, provide insight into the stereochemistry of enoyl reduction, and identify candidate active site residues important for catalysis.

Biological Significance.

The above work represents the structural and functional characterization of a trans-acting ER from an iterative fungal PKS. The LovC structures help identify the possible ACP docking site, as well as residues important for substrate binding and enzyme catalysis. Looking beyond lovastatin biosynthesis, when we submitted the LovC sequence to the BLAST server, in addition to the trans-acting ER domains that are well studied to associate with fungal PKSs, we also found eight sequence homologs (32–59% identity) in Aspergillus clavatus, six homologs (36–58% identity) in Magnaporthe grisea, twelve homologs (29–46% identity) in Aspergillus niger, and five homologs (32–36% identity) in Gibberella zeae. Note that the fungal fatty acid ER domain has a completely different fold (42). Furthermore, an extensive sequence comparison of LovC with other MDR family members confirms that no known alcohol dehydrogenase or quinone reductase share > 20% sequence identity with LovC. Namely, the trans-acting ER domains form a subfamily that is distinct from other MDR family members, as analyzed in details by a recent paper on the MDR family (15). Therefore, these LovC homologs, with 30–60% of sequence identity to LovC, most likely also serve as trans-acting ER domains that are used for polyketide biosynthesis. If this is true, these fungi should have not only possess the trans-acting ER domains, but also the corresponding PKS megasynthases whose ERs are inactivated. As presented in the SI Text, the inactivated ER⁰ domain has a shortened linker region between β3 and β4, resulting in the absence α1 which contains important active site residues of MDR proteins (including K54 of LovC). This can be used as a diagnostic property when we search for homologs of the LovB-ER⁰ using the BLAST server. As a positive control, the BLAST search identifies known ER⁰ domains from the monacolin synthase in Monascus pilosus and the compactin synthetase in Penicillium citrinum, both synthases have well-characterized trans-acting ER domains. Significantly, the BLAST search also reveals the existence of ER⁰ domains in PKS or PKS/nonribosomal peptide synthetase megasynthases whose functions are not known yet, such as the six PKSs with ER⁰ in Aspergillus clavatus (25–30% identity), five in Magnaporthe grisea (27–33% identity), two in Aspergillus niger (21–23% identity), and one in Gibberella zeae (21% identity). In all ER⁰ homologs, the key helix–loop region between β3 and β4 is shortened, similar to the LovB-ER⁰ domain. The above analyses support that these megasynthases have inactivated ER⁰ domains, as well as corresponding trans-acting ER domains that can produce partially reduced polyketides such as brefeldin, cytochalasin, and equisetin from Aspergillus clavatus, Ace1 from Magnaporthe grisea, and zearalenone from Gibberella zeae. In conclusion, the significance of the LovC crystal structure lies in its uniqueness as a trans-acting, monomeric ER domain, and its general applicability to other trans-acting ER domains that react with corresponding megasynthases with inactivated ER domains. In the future, this work will facilitate the elucidation of biosynthetic pathways in these ER-inactivated megasynthases, as well as an in vitro assay to characterize other trans-acting ER domains.

Methods

Protein Expression and Purification.

Recombinant MatB, LovC, and LovB were expressed and purified as detailed in SI Methods. Selenomethionine-substituted LovC protein was produced in E. coli BL21(DE3) using metabolic inhibition of the methionine pathway in M9 minimal medium detailed in SI Methods.

Site-Directed Mutagenesis.

Site-directed mutagenesis was achieved using the QuikChange II kit (Stratagene). Mutagenic oligonucleotides are listed in SI Methods.

Cofactor-Binding Assay.

An Hitachi 4500 fluorescence spectrometer (50-nm slit; 700V) was used to record an emission scan (350–600 nm) with excitation wavelength set to 340 nm (slit width 5.0 nm). NADH or NADPH (0.1–50 μM) was added to 0.1 M potassium phosphate buffer (pH 7.0; 100 μL) with or without 50 μM LovC. The fluorescence titration plots (Fig. S3A) were generated from the λ_max emission at 455 nm.

Fluorometric Activity Assay.

Potential in vitro substrates of LovC were obtained from the Developmental Therapeutics Program/National Cancer Institute (DTP/NCI) small molecule repository (43) or synthesized as NACs. The synthesis and assay conditions are detailed in SI Methods.

Reconstitution of DmL Biosynthesis.

LovB (25 μM) was incubated with 25 μM MatB, 25 μM LovC (WT or mutant), 100 mM malonate, 5 mM CoA, 20 mM ATP, 2 mM NADPH, 2 mM S-(5′-adenosyl)-l-methionine chloride (SAM), and 7 mM MgCl₂ in buffer [100 mM NaH₂PO₄ (pH 7.4), 10% (vol/vol) glycerol, 2 mM DTT] at ∼25 °C overnight. Processing of reactions and product detection is detailed in SI Methods.

Gel-Filtration Interaction Studies.

Samples of purified LovB (10 μM; 100 μL) and LovC (10 μM; 100 μL) were incubated for 20 min and then run on a Superdex 200 10/300 GL column (equilibrated with Buffer C), and the 8.9-mL peak fraction representing the LovB-LovC complex was concentrated to 250 μL before SDS/PAGE.

Crystallization, Data Collection, Phasing, Model Building, and Refinement.

Cocrystals of LovC with crotonoyl CoA (LovC-CO7), SeMet LovC with crotonoyl CoA (SeMet LovC-CO7), and LovC WT or mutant K54S with cofactor NADP⁺ (LovC-NADP, LovC K54S-NADP) were grown at 25 °C using the sitting drop vapor diffusion method. The condition of crystallization and data collection is detailed in SI Methods. Multiwavelength anomalous dispersion (MAD) methods were used to solve the structure of the SeMet LovC-CO7. Details of phasing, model building, and refinement are provided in SI Methods. Detailed statistics are listed in Table S1.