Metamorphic enzyme assembly in polyketide diversification

Abstract

The chemical diversity of natural products is fueled by the emergence and ongoing evolution of biosynthetic pathways in secondary metabolism^1-5. However, co-evolution of enzymes as functional assemblies for metabolic diversification is not well understood, especially at the biochemical level. Here, two parallel enzyme assemblies with an extraordinarily high sequence identity form a β-branched cyclopropane in the curacin A (Cur), and a vinyl chloride group in the jamaicamide (Jam) pathways, respectively. The assemblies include a halogenase (Hal), a 3-hydroxy-3-methylglutaryl (HMG) enzyme cassette for β-branching and an enoyl reductase domain (ER). Bioinformatic analysis indicated that the corresponding genes were inserted into modular polyketide synthases (PKSs) via acyltransferase (AT) domain replacement. The Hal from CurA, and the dehydratases (ECH₁s) and decarboxylases (ECH₂s) within the HMG enzyme cassettes and ERs from both Cur and Jam were assessed biochemically to determine the mechanism of cyclopropane and vinyl chloride formation. Unexpectedly, the polyketide β-branching pathway was modified by introduction of a γ-chlorination step on (S)-HMG mediated by Cur Hal, a non-heme Fe^II, α-ketoglutarate (α-KG)-dependent halogenase⁶. In a divergent scheme, Cur ECH₂ was found to catalyze formation of the α,β C=C enoyl thioester, whereas Jam ECH₂ formed a vinyl chloride moiety by selectively generating the corresponding β,γ C=C (enoyl thioester) of the 3-methyl-4-chloroglutaconyl decarboxylation product. A non-conserved Tyr⁸² residue in Cur ECH₂ was crucial to this regiochemical control. CurF ER specifically catalyzed an unprecedented cyclopropanation reaction on the chlorinated product of Cur ECH₂. Thus, the combination of chlorination and polyketide β-branching, coupled with mechanistic diversification of ECH₂ and ER leads to formation of cyclopropane and vinyl chloride substituents. These results reveal a remarkable parallel interplay of evolutionary events in multienzyme systems leading to functional group diversity in secondary metabolites.

The tremendous biosynthetic capability of nature is well exemplified by structurally diverse secondary metabolites that help their hosts, typically microorganisms and plants, to survive and thrive in environmental niches by mediating a broad range of ecological and physiological interactions^1-5. The biosynthesis of secondary metabolites is “diversity-oriented”^1,5,7, targeting the variable environment by producing a vast array of complex chemical structures⁸. This enormous productivity is largely fueled by the rapid evolution of biosynthetic genes and functional alteration of the corresponding enzymes^3,4. As such, the evolutionary history of metabolic gene assemblies informs the origin of their biosynthetic diversity. However, tracing the ancestral forms of multiple genes as a functional collective is elusive, especially when they are dispersed in the genome. Biosynthetic genes from microbial hosts are usually clustered, and are ideal for evolutionary and functional studies⁴. Currently, our understanding of the evolution and function of multienzyme systems in secondary metabolism is largely based on genetic studies, with recent efforts increasingly focused on comparative biochemical analysis⁹.

Modular PKSs originated from fatty acid synthases (FASs) and serve as a paradigm for secondary metabolic systems that evolve to expand chemical diversity¹⁰. These giant biochemical machines are modular assembly-lines that catalyze highly programmed biosynthetic pathways, where final product structures result from variation in chain initiation, extension, termination and tailoring steps^11-13. Moreover, PKSs have a propensity to form hybrids such as with nonribosomal-peptide synthetase (NRPS) modules and recently reported HMG enzyme cassettes^14,15. The metamorphic properties shown in these composite systems drive metabolic diversification, but relatively few details are currently available to describe the mechanistic details.

The curacin and jamaicamide marine cyanobacterial metabolites from Lyngbya majuscula are mixed-polyketide nonribosomal-peptide natural products with potent anticancer and sodium channel blocking activities, respectively^16,17. The parallel components of the Cur and Jam biosynthetic pathways (Fig. 1a) provide an unusual opportunity to investigate the biosynthetic origin of chemical diversity, in the form of cyclopropane ring formation for curacin and vinyl chloride formation for jamaicamide^16,18. Studies on the variant function and selectivity of these highly parallel biosynthetic systems form the subject of this report.

Figure 1. Comparison of enzyme assemblies in the Cur and Jam pathways.

a, Formation of cyclopropane and vinyl chloride functional groups. b, Comparative sequence identities of the enzymes encoded by the two highly similar regions in the Cur and Jam pathways. The aligned DNA sequences are located at the boundaries of these two regions. c, Formation of 3-ACP₃ in the Cur pathway, and hypothesized reactions for 4-ACP₃, 5-ACP₃ and 6-ACP₃. The hypothetic chlorinated intermediates are shown along with the non-chlorinated ones. The β-branching carbon atoms are highlighted in red.

Two highly similar enzyme assemblies

The two parallel, highly conserved Cur and Jam enzyme assemblies are incorporated into the early PKS modules, and are predicted to catalyze β-branching reactions in the growing chain elongation intermediate^16,18. These unusual embedded domains and discrete enzymes span from CurA to CurF and from JamE to JamJ, and are grouped into three subsets (Fig. 1a): (1) Hals embedded in CurA and JamE; (2) HMG enzyme cassettes containing a tandem acyl carrier protein (ACP) tridomain (ACP₃), including ACP_I, ACP_II and ACP_III embedded in CurA and JamE, discrete CurB and JamF ACP_IVs, CurC and JamG KSs, CurD and JamH HMG-CoA synthase-like enzymes (HCSs), CurE and JamI ECH₁s, ECH₂s embedded in CurF and JamJ; and (3) ERs embedded in CurF and JamJ (Fig. 1a). Comparative analysis of these Cur and Jam enzymes revealed that the sequence identities of the Hals, ACP₃s, ACP_IVs, KSs, HCSs and ECH₁s are extraordinarily high (∼90%), whereas the ECH₂s and ERs are substantially lower (∼60% identity) (Fig. 1b).

Cur and Jam Hals were predicted to be α-KG-dependent non-haem halogenases (less than 20% sequence identity to characterized homologs)^19-21, that catalyze halogenation of unactivated carbon atoms^20-24 through a non-haem Fe^IV=O intermediate^25,26. HMG enzyme cassettes have been demonstrated to catalyze polyketide on-assembly-line β-branching to generate a pendant methyl or ethyl group from a polyketide β-carbonyl^14,15,27. Cur and Jam ERs show ∼50% sequence identity to other ERs in Cur and Jam PKS modules, and belong to the acyl-CoA reductase family that catalyzes NADPH-dependent reduction of α,β C=C (enoyl thioester) in acyl-CoAs or acyl-ACPs²⁸. These two ERs are located upstream of CurF and JamJ KS, an unusual location as ERs typically reside between AT and ACP domains in PKS modules.

AT replacement-mediated PKS hybridization

Bioinformatic analyses of Cur and Jam pathway sequences suggested that the parallel AT-Hal-ACP_I-ACP_II-ACP_III-ACP_IV-KS-HCS-ECH₁-ECH₂-ER-KS-AT gene assembly (Fig. 1b) might have been introduced into the polyketide pathway by AT domain replacement. Based on the DNA and amino acid alignments of CurA—CurF and JamE—JamJ, we found that the highly similar regions, extend from the N-termini of the ATs in CurA and JamE, through the C-terminal “post-AT linkers”²⁹ of the ATs in CurF and JamJ (Fig. 1b, and Supplementary Fig. 1). Recent bioinformatic studies indicate that these highly similar sequences could promote AT domain replacement by homologous recombination^30,31. Thus, a “di-AT domain replacement” might have occurred in Cur or Jam pathways through insertion of the above gene assembly into a pre-existing cluster, which could serve as an efficient strategy for PKS pathway expansion or contraction. This hypothesis is supported by phylogenetic analysis for the KS, AT and dehydratase (DH) domains of the sequenced pathways from L. majuscula (Supplementary Fig. 2).

HMG β-branching with ER saturation

HMG β-branching includes a series of modifications on the β-carbonyl group of polyketide intermediates typically tethered to the tandem ACPs. As shown for curacin A (Fig. 1c), the AT domain loads a malonyl group onto CurB ACP_IV, and the KS catalyzes subsequent decarboxylation to acetyl-ACP_IV. HCS then catalyzes condensation of C-2 from acetyl-ACP_IV and acetoacetyl-ACP₃, to form (S)-HMG-ACP₃ (1-ACP₃). As we have shown previously, ECH₁ catalyzes dehydration of 1-ACP₃ to 3-methylglutaconyl-ACP₃ (2-ACP₃), followed by ECH₂ mediated decarboxylation to generate 3-methylcrotonyl-ACP₃ (3-ACP₃)¹⁴, a presumed precursor for (1R, 2S)-2-methylcyclopropane-1-carboxyl-ACP₃ (5-ACP₃) (Fig. 1c).

This initial study raised two important questions regarding the role of the Cur and Jam HMG enzyme cassettes and ERs in formation of cyclopropane and vinyl chloride moieties: (1) How is the CurF ER involved in cyclopropyl ring formation based on its predicted function as an enoyl reductase, and is Cur ER involved in reduction of 3-ACP₃ to 4-ACP₃ (Fig. 1c)? (2) How is the unusual β,γ C=C of the pendant vinyl chloride group formed in the Jam pathway? As previously proposed, is 3-methyl-3-butenoyl-ACP₃ (6-ACP₃) generated from 3-ACP₃ isomerization⁶, or by differential regiochemical control of double bond formation during ECH₂ decarboxylation³² (Fig. 1c)?

First, we sought to test whether CurF ER can saturate 3-ACP₃, the previously established product of Cur ECH₂¹⁴. Thus, the embedded domain was excised and cloned as an N-terminal GST-tagged fusion protein. We also overexpressed and purified the CurA ACP₃ tridomain and each excised single domain (ACP_I, ACP_II and ACP_III) as apo proteins (Supplementary Fig. 3). 1-ACPs were generated as previously described¹², and substrate loading was examined by HPLC (Supplementary Fig. 4). The ACP_I, ACP_II and ACP_III have nearly identical amino acid sequences, and each was efficiently loaded with the HMG substrate. Thus, for convenience we chose excised CurA ACP_II, as well as ACP₃, for subsequent enzyme assays.

Fourier transform ion cyclotron resonance mass spectrometry (FTICR-MS) and infrared multiphoton dissociation (IRMPD) methods were applied to detect mass changes to the HMG substrate covalently linked to the ACP phosphopantetheine (PPant) arm³³. Cur ER function was assessed by coupling it with the Cur ECH₁ and ECH₂ reactions. As reported, Cur ECH₁ catalyzed the reversible dehydration of 1-ACP_II to generate 2-ACP_II, and Cur ECH₂ catalyzed decarboxylation of 2-ACP_II to generate 3-ACP_II, corresponding to 18- and 62-Dalton mass losses from 1-ACP_II, respectively (Fig. 2b and 2c). Cur ECH₁ shows substrate preference for (S)-HMG-ACP_II over (R)-HMG-ACP_II (Supplementary Fig. 5a, 5b and 5c), which is consistent with our previous results using the CoA-linked substrates¹⁴. With Cur ER and NADPH, a 2-Dalton mass addition was observed for 3-ACP_II (Fig. 2d and Supplementary Fig. 6a), corresponding to saturation of the α,β enoyl thioester to generate 4-ACP_II.

Figure 2. Halogenation and cyclopropanation in the Cur pathway.

a-h, Partial FTICR mass spectra (12+ charge state of ACP_II) for Cur ECH₁, ECH₂ and ER reactions excluding (a-d) or including (e-h) the Cur Hal chlorination step. 1-ACP_II was incubated with Cur Hal for 2 h to generate the γ-Cl-1-ACP_II substrate. Reactions were incubated at 30°C for 2 h for the 1-ACP_II substrate and 30 min for the γ-Cl-1-ACP_II substrate. Asterisks denote unidentified species. i, GC-MS analysis of the enzyme products after butylamine cleavage, and comparison with authentic standards. For optimal sensitivity, the chromatograms were recorded at selective ion mode (SIM) by monitoring 55, 57, 83, 115, 155 and 157 atomic mass unit (amu). Retention times of the products were confirmed by coinjection with the authentic standards.

To confirm the structure of the Cur ER reaction product tethered to ACP₃, we cleaved it from the PPant arm with butylamine to generate the corresponding butylamide derivative. Gas chromatography (GC)-MS analysis³⁴ was performed and the readily separable isomers were compared and correlated with authentic standards by mass spectra and coinjection. We used 1-ACP_II and 1-ACP₃ as substrates for Cur ECH₁, ECH₂ and ER reactions, and their products were confirmed as 4 (Fig. 2i, upper trace). However, the relatively poor efficiency of Cur ER-catalyzed reduction of 3-ACP_II (see below), suggested that it was unlikely to be the natural substrate. Moreover, the timing and function of Cur Hal remained to be established, and a key involvement in the β-branching scheme was hypothesized.

Halogenation and cyclopropane ring formation

Bioinformatic analysis and presence of the vinyl chloride in jamaicamide suggest that the Jam and Cur Hal domains might be α-KG-dependent non-haem halogenases^19-21. In the Jam pathway, chlorination evidently occurs on the pendant carbon generated by β-branching. Thus for Cur Hal, we reasoned that cyclopropane ring formation likely involves transient halogenation as in coronatine biosynthesis, where the chloride serves as a leaving group²⁰. However, the timing of the chlorination step remained to be established, and the identity of the Cur pathway cyclopropane ring-forming enzyme was not evident from examination of the Cur gene cluster.

An important clue about the timing of chlorination at the β-branching carbon came from previous precursor-incorporation studies in curacin A biosynthesis. NMR data on curacin A labeling by [²H₃,2-¹³C]acetate indicated that the β-branching carbon that forms cyclopropane was labeled by only one deuterium atom (α-isotope chemical shift at C20, 0.295)¹⁸, which was previously interpreted as an anomalous result. However, these data are consistent with chlorination occurring on the HMG β-branching intermediate before ECH₂ catalyzed decarboxylation. Otherwise, the pendant carbon atom would be labeled by either one or two deuterium atoms in a 2:1 ratio (Supplementary Fig. 7).

To identify the function of Cur Hal, we constructed both Hal and the tetradomain Hal-ACP₃ as N-terminal His-tagged proteins (Supplementary Fig. 3). Cur Hal eluted as a dimer from an analytical size-exclusion column. Following His-tag removal by thrombin cleavage, metal content of the protein was analyzed by inductively coupled plasma (ICP)-MS. After reconstitution with a mixture of metal ions and α-KG, more than 90% of Hal was bound to Fe²⁺ (Supplementary Methods), which indicated that it functions as an Fe^II-dependent enzyme. Thus, anaerobic purification coupled with α-KG/Fe^II-reconstitution was performed, as was previously reported to retain optimal activities of α-KG-, O₂- and Fe^II-dependent halogenases^20-22.

Seven acyl-ACP substrates bearing the target pendant β-branching carbon were tested to establish the substrate identity for Cur Hal, including malonyl-ACP_IV, acetyl-ACP_IV, 1-ACP_II, 2-ACP_II, 3-ACP_II, 4-ACP_II and 6-ACP_II (Fig. 1c). Consistent with the [²H₃,2-¹³C]acetate precursor incorporation experiment noted above, we observed formation of the mono-chlorinated species exclusively on 1-ACP_II to generate γ-Cl-1-ACP_II. The chlorinated product was confirmed by FTICR-MS and IRMPD analysis (Fig. 2e and Supplementary Fig. 6a), and corroborated by GC-MS results (see below, Fig. 3h). As expected, Cur Hal showed the same selectivity for (S)-HMG-ACP_II (1-ACP_II) as Cur ECH₁ (Supplementary Fig. 5a, 5b and 5c). In the absence of α-KG or O₂, no chlorinated product was detected with Cur Hal in the presence of HMG substrate (Supplementary Fig. 5d). Notably, the chlorination on the carboxylated γ-carbon of HMG is unusual for α-KG-dependent non-haem halogenases, which have been previously limited to catalyzing modification of unactivated carbons atoms^20-24.

Figure 3. Comparison of ECH₂s and ERs in Cur and Jam pathways.

a-f, Partial FTICR mass spectra (12+ charge state of ACP_II) for Cur and Jam ECH₁, ECH₂ and ER reactions with the γ-Cl-1-ACP_II substrate. The reactions were incubated at 30°C for 30 min. g, Comparison of catalytic efficiencies for cyclopropanation and saturation by Cur and Jam ERs. The product yields in the time-course studies were measured by IRMPD-based quantification. 3-ACP_II was used as substrate for Cur ER saturation, and γ-Cl-3-ACP_II was used as substrate for Cur ER cyclopropanation and Jam ER saturation. Assays were performed in triplicate, and standard deviation error bars are shown. h, GC-MS analysis to identify the structures of Cur and Jam ECH₂ products. The chromatograms were recorded at SIM by monitoring 57, 117, 154 and 189 amu. The retention times of products were confirmed by coinjection with the authentic standards.

Next, we sought to investigate how chlorination of 1-ACP_II affects efficiency of the downstream reaction sequence with the HMG cassette enzymes. 1-ACP_II was converted to γ-Cl-1-ACP_II by Cur Hal (Fig. 2e), and reacted sequentially with Cur ECH₁, ECH₂ and ER. ECH₁ dehydrated γ-Cl-1-ACP_II and the γ-Cl-2-ACP_II product was decarboxylated by ECH₂ to generate γ-Cl-3-ACP_II (Fig. 2f and 2g). The ECH₁/ECH₂-coupled dehydration and decarboxylation with γ-Cl-1-ACP_II was shown to be ∼4-fold faster compared to 1-ACP_II (Supplementary Fig. 8), which might be due either to the electron-withdrawing effect of the γ-chlorine atom to stabilize the negative charge on the intermediate of ECH₂ decarboxylation or to a more effective binding position of the chlorinated versus non-chlorinated substrate in the enzyme active sites. Unexpectedly, when Cur ER was added with Cur ECH₁ and ECH₂ in the presence of γ-Cl-1-ACP_II, no saturation product was obtained. Instead, we observed a 34-Dalton mass reduction from γ-Cl-3-ACP_II (Fig. 2h and Supplementary Fig. 6), demonstrating the elimination of chlorine in the product. This result suggests that the product could be 3-ACP_II, 5-ACP_II or 6-ACP_II (Fig. 1c).

The experimental design to determine the final product in the presence of both Hal and HMG-cassette enzymes was streamlined by a one-pot reaction using Cur Hal-ACP₃, ECH₁, ECH₂ and ER. Cur (apo) Hal-ACP₃ was loaded from 1-CoA, desalted, and mixed with Cur ECH₁, ECH₂ and ER in an anaerobic environment. The reaction was initiated by exposing the mixture to air. To confirm product structure, the acyl groups linked to Hal-ACP₃ were cleaved with butylamine and the butylamide derivatives were compared with the authentic standards by GC-MS. Direct correlation was confirmed by the mass spectra and coinjection with authentic standards to a single species identified as the cis-2-methylcyclopropane-1-carboxyl compound (Fig. 2i, lower trace), demonstrating the formation of 5-ACP by an unprecedented ER-catalyzed cyclopropanation reaction, presumably via an intramolecular nucleophilic substitution. The internal nucleophile, presumably the resonance-stabilized α-carbanion, is believed to be generated by ER-catalyzed hydride transfer from NADPH to the β-carbon of enoyl thioester^35,36.

Functional differentiation of ERs

Due to the similarity between Cur and Jam ERs (∼65% sequence identity; higher than those between Cur/Jam ER with other PKS ERs in the Cur and Jam pathways), we sought to test whether Jam ER can catalyze the same cyclopropanation when presented with γ-Cl-3-ACP_II. Likewise, Jam ER was prepared (Supplementary Fig. 3) and assayed in the same way as the Cur ER. Unexpectedly, only the saturated product, γ-Cl-4-ACP_II, was observed (Fig. 3c and Supplementary Fig. 6b), indicating that its activity is typical of a canonical PKS ER.

The distinct functions of Cur and Jam ERs motivated us to compare the catalytic efficiencies of Cur ER cyclopropanation vs. Jam ER saturation of the chlorinated substrate, and the efficiencies of cyclopropanation of the chlorinated substrate vs. saturation of the non-chlorinated substrate by Cur ER. This was accomplished using time-course studies by measuring product yields under uniform reaction conditions. It was not possible to measure enzyme kinetic parameters (k_cat and K_M) due to the tendency of ER to aggregate and the solubility limits of ACP-tethered substrates. Thus, γ-Cl-3-ACP_II was employed to assess cyclopropanation by Cur ER, compared to reduction by Jam ER, and 3-ACP_II was used to compare reductive efficiency of the Cur and Jam ERs. IRMPD-based MS analysis (e.g. peak abundance of PPant ejection products (PEPs)³³) provided a convenient method to quantify the yields of ER saturation products that correspond to a 2-Dalton mass change (Supplementary Fig. 9 and 10; see Supplementary Methods). Similarly, Cur ER-catalyzed cyclopropanation was quantified by preparing 4-ACP_II, as an internal standard for 5-ACP_II.

We found that Jam ER saturation and Cur ER cyclopropanation of γ-Cl-3-ACP_II are faster by ∼400-fold and ∼50-fold, respectively, than is Cur ER saturation of 3-ACP_II under identical experimental conditions (Fig. 3g). For 3-ACP_II, Jam ER saturation is ∼240-fold faster than is Cur ER saturation (Supplementary Fig. 11). This comparison confirmed that Jam ER has retained canonical function as an α,β enoyl reductase, in contrast to the Cur ER as a cyclopropanase. Given the proposed hydride-transfer step for both Cur and Jam ERs, their mechanisms are likely differentiated after formation of the α-carbanion intermediate, which functions as an intramolecular nucleophile (Cur ER) or is protonated (Jam ER) (Fig. 5b).

Figure 5. Impact of enzyme assembly evolution on β-branching chemical diversity.

a, Proposed ancestral forms of the enzyme assemblies in Cur and Jam pathways. b, The functional diversification of ERs. c, Differential regiochemical control by ECH₂s.

Regiochemical control by ECH₂s

Ascertaining the role of chlorination in cyclopropane ring formation during curacin biosynthesis strongly suggested that a similar chlorination event occurs in the Jam pathway. Given the extraordinarily high similarity between Cur and Jam Hals (92% sequence identity), we surmised that the two pathways diverge after the halogenation step, resulting in differential catalytic processes leading to the vinyl chloride moiety in jamaicamides. In the two high similar enzyme assemblies, the Cur and Jam ECH₂ domains have lowest sequence identities (59%) (Fig. 1b), and likely function as a key branch-point determinant. Accordingly, Jam ECH₂ was prepared in the same way as the Cur ECH₂ (Supplementary Fig. 3). We also expressed and purified Jam ECH₁ to generate substrate for Jam ECH₂. The function of these Jam enzymes were subsequently investigated with Cur substrates starting from 1-ACP or γ-Cl-1-ACP_II to establish what controls introduction of the β,γ vinyl chloride group.

For both 1-ACP_II and γ-Cl-1-ACP_II substrates, Jam ECH₁ and ECH₂ catalyzed successive dehydration and decarboxylation steps as expected (Fig. 3d, Supplementary Fig. 12b, 12c, 12f, and 12g). However, when Jam ER was added, only ∼20% of the saturated product was detected for the non-chlorinated substrate (derived from 1-ACP_II, Supplementary Fig. 12i). No mass change was observed for the corresponding chlorinated substrate (derived from γ-Cl-1-ACP_II, Supplementary Fig. 12j), indicating that the Jam ECH₂ product is not a substrate for Jam ER. In addition, the ECH₁s were switched in the Cur and Jam ECH₁/ECH₂/ER coupled reactions with γ-Cl-1-ACP_II substrate, and no change in the product profile was observed, suggesting that both Cur and Jam ECH₁s generate the same product (γ-Cl-2-ACP_II), consistent with their 94% sequence identity. Thus, the Jam ECH₂-catalyzed decarboxylation product of γ-Cl-2-ACP_II was predicted to be γ-Cl-6-ACP_II (β,γ C=C; Fig. 3d) with a vinyl chloride group, instead of γ-Cl-3-ACP_II (α,β C=C; Fig. 3a). UV spectral comparison of Cur and Jam ECH₂ decarboxylation products revealed that their UV absorption patterns are slightly different between 250 and 280 nm (Supplementary Fig. 13b), which reflects isomeric α,β or β,γ C=C (enoyl thioester) functionality in the molecules.

To determine the structures of the decarboxylation products, one-pot reactions using Cur Hal-ACP₃, Cur ECH₁ and Cur or Jam ECH₂s, and GC-MS analysis were performed as described above. For the reaction including Cur ECH₂, the main product contained primarily an α,β C=C in the E configuration, with trace amounts of the β,γ C=C isomer (Fig. 3h, upper trace) quantified to be ∼3% (see below, Fig. 4b). In contrast, reactions using Jam ECH₂ showed high regiochemical control to generate exclusively the β,γ C=C product, with ∼85% in the E configuration and ∼15% of the Z isomer (Fig. 3h, lower trace). The exclusive E configuration of the vinyl chloride C=C in jamaicamide natural products¹⁶ suggests that the small amount of Z double bond product is likely due to utilization of the curacin substrate, which is less sterically hindered than the jamaicamide substrate (Fig. 1a). Notably, Jam ECH₂ decarboxylation had lower regiochemical control using the non-chlorinated substrate, and generated ∼80% β,γ C=C and ∼20% α,β C=C products, which further explains the partial enoyl reduced product observed following Jam ECH₁, ECH₂ and ER reactions with this substrate (Supplementary Fig. 12i). Given the normal function of ER to catalyze only α,β C=C (enoyl thioester) saturation, the selective formation of β,γ C=C product by Jam ECH₂ renders Jam ER superfluous in the biosynthesis of jamaicamides. In general, α,β C=C ECH₂ products are energetically preferred and frequently identified or predicted in other pathways^15,27,37-40, except the pathways of pederin and its structural analogs^41,42, which are predicted to generate β,γ C=C products (Supplementary Fig. 14a).

Figure 4. Loss of Cur ECH₂-mediated regiochemical control by site-directed mutagenesis.

a, The hypervariable region (in magenta) of Cur ECH₂ and the active site chamber modeled with the chlorinated substrate. The S-configuration of the HMG γ-carbon is preferred based on modeling results. b, Activity and regiochemical control of ECH₂ WT and Cur ECH₂ mutants. γ-Cl-1-ACP_II was used as the substrate for all reactions. (Left) HPLC analysis for ECH₁/ECH₂ coupled dehydration and decarboxylation. All reactions were quenched after 10 min incubation at 30°C. (Right) IRMPD-based quantification to measure the percentage of β,γ C=C products. The coupled ECH₁/ECH₂ reactions were incubated for 45 min before treated with Jam ER for 45 min at 30°C. Assays were performed in triplicate, and standard deviation error bars are shown.

Loss of regiochemical control by mutation

To understand the regiochemical control of ECH₂-catalyzed decarboxylation, the previously solved Cur ECH₂ structure³² was modeled with the chlorinated substrate (Fig. 4a). The catalytic efficiencies of the wild type (WT) and mutants of Cur ECH₂ were compared by performing the ECH₁/ECH₂ coupled assay with γ-Cl-1-ACP_II substrate. Moreover, we measured the ratios of the two possible decarboxylation products (α,β and β,γ C=C), γ-Cl-3-ACP_II and γ-Cl-6-ACP_II.

Based on our results, the catalytic activities of WT and mutant Cur ECH₂s were significantly increased with the chlorinated substrate, possibly due to γ-Cl stabilization of the carbanion intermediate (Fig. 4b, left panel). However, their relative catalytic activities are similar to our previous results for the non-chlorinated substrate³². Cur ECH₂ Y82F and Jam ECH₂ WT had activities close to Cur ECH₂ WT, indicating that Tyr⁸² is not essential for decarboxylation.

Next, we measured the ratio of α,β and β,γ C=C decarboxylation products to investigate whether the site-directed mutations can elucidate a basis for double bond regiochemical control by Cur ECH₂. Changes in the ratio of α,β and β,γ C=C products were assessed by measuring UV absorbance ratios (A_280nm/A_250nm, Supplementary Fig. 13) for HPLC peaks corresponding to ECH₂ decarboxylation products (Fig. 4b). Measured peak ratios for Cur ECH₂ WT, K86Q, K86A, H240Q and H240A are ∼1.75, for Jam ECH₂ WT the ratio is 2.23, but for Cur ECH₂ Y82F it is 1.85. The intermediate value for Cur ECH₂ Y82F suggests a mixture of α,β and β,γ C=C products. These products can be distinguished directly by using Jam ER as a reagent to selectively reduce α,β C=C (Fig. 3c) followed by IRMPD to quantify product ratios (Supplementary Fig. 9c). The level of β,γ C=C product (γ-Cl-6-ACP_II) for Cur ECH₂ WT, K86Q, K86A, H240Q and H240A was ∼3% of the total product formed, but was ∼30% of total product generated by Cur ECH₂ Y82F (Fig. 4b). Based on the site-directed mutagenesis results, positioning of the Tyr⁸² hydroxyl group seems crucial for regioselectivity (α or γ position, Fig. 4a) of the protonation step following collapse of the presumed enolate intermediate (Fig. 5c). The Tyr⁸² resides in a hypervariable region (Fig. 4a, α2–loop–α3, in magenta) and is a non-conserved residue for ECH₂ enzymes³². Our results suggest that ECH₂ regiochemical control might be easily affected by mutations that occur in this hypervariable region, thus serving as a facile strategy to introduce functional group diversification.

Discussion

The Cur and Jam pathways enable us to witness the remarkable process of evolutionary diversification in secondary metabolism based on comparative biochemical analysis of two parallel β-branching enzyme assemblies (Fig. 5a). DNA duplications and insertions in natural product pathway evolution are readily identified in these systems, but the mutations in enzyme assemblies leading to generation of chemical diversity require direct analysis to elucidate function. Here we show at the biochemical level that subtle changes in amino acid sequences of only two members (e.g. ECH₂ and ER) of the 10 component β-branching enzyme system are ultimately responsible for distinct chemical outcomes.

Both Cur and Jam enzyme assemblies contain Hal domains that were evidently recruited and embedded in a modular PKS to impart new chemical diversity. Recent studies on this class of α-KG dependent non-haem halogenases have been reported as discrete enzymes in secondary metabolite pathways^6,20-23, but this integrated domain represents an unprecedented example of pathway diversification. Cur and Jam are further diversified by the amino acid sequence variation in downstream enzymes to yield different catalytic activities. Specifically, the Cur ER domain was shown to be a cyclopropanase catalyzing nucleophilic displacement of the chlorine atom leading to a highly strained and unusual functional group (Fig. 5b). In contrast, the Jam ER domain was found to retain reductase function for the curacin α,β enoylthioester substrate, but is inactive against the corresponding β,γ enoylthioester isomer. Thus, in addition to the cyclopropanation strategies of Zn²⁺-dependent CmaC^20,43 and the recently reported FAD-dependent dehydrogenase KtzA⁴⁴, where chloride also serves as a leaving group, the NADPH dependent Cur ER-catalyzed cyclopropanation represents a new strategy for generating a thioester enolate and subsequent ring formation. Structural insights to reveal the sequence variations of Cur ER that stabilize the α-carbanion while supporting closure of the highly strained cyclopropane is key to understanding its functional evolution. Moreover, regiochemical control is presumed to be the result of a protonation step accompanying enolate collapse after ECH₂-mediated decarboxylation (Fig. 5c). Thus, further pathway diversification is reflected in select amino acid sequence changes that direct alternative double bond regiochemistry in the jamaicamide products. These parallel yet distinct systems demonstrate the mutability of enzymes within complex metabolic pathways, and reveal their metamorphic properties for creating chemical diversity in biologically active natural products.

Methods

Chemicals

1-CoA and 2-CoA were enzymatically generated using HMG reductase^14,45 and Cur ECH₂¹⁴. 6-CoA and the butylamide derivatives were synthesized as described in Supplementary Methods. (1R,2S)-2-methylcyclopropanecarboxylic acid was a gift from Timothy M. Ramsey (Novartis Institutes for Biomedical Research, Inc.). All other chemicals were from Sigma-Aldrich.

Construction of plasmids and overexpression and purification of proteins

The expression plasmids for Cur ACP_IV, Cur ECH₁ and Cur ECH₂ were constructed in our previous work¹⁴. The expression plasmids for Cur ACP₃, ACP_I, ACP_II and ACP_III were gifts from Christopher T. Walsh (Harvard Medical School). Cur Hal, Hal-ACP₃ and ER genes were amplified from the cosmid pLM54¹⁸. Jam ECH₁, ECH₂ and ER genes were amplified from cosmid pJam3¹⁶. The primers for the plasmid construction are listed in Supplementary Table 1. His-tagged proteins were expressed in E. coli BL21 (DE3) transformed with the corresponding plasmids.

His-tagged Cur and Jam ECH₁s, ECH₂s and ERs, as well as Cur ACP constructs were purified using Ni-nitrilotriacetate (Ni-NTA) HisTrap column followed by desalting or gel-filtration. The purification of Cur Hal and Hal-ACP₃ was performed under inert atmosphere by using an ÄKTA FPLC (GE Healthcare) with its tubing linked to a glove box (Coy Laboratory Products), which is similar to the system previously described^20,21,46. The Hal N-terminal His-tag was removed by thrombin for subsequent metal content analysis. See Supplementary Methods for detailed protocols.

Metal content analysis of Cur Hal

His-tag cleaved Cur Hal was anaerobically reconstituted with 1 mM α-KG and Fe²⁺ (or a metal mixture). The metal content of Cur Hal was measured by ICP-MS (Finnigan).

Cur Hal functional assays

1-ACP_II was served as the substrate for Hal catalytic activity assays. Typically, ∼200 μl of reaction mixture containing 50 μM 1-ACP_II, 5 μM Cur Hal, 50 μM fresh Fe(NH₄)₂(SO₄)₂, and 0.5 mM α-KG in 50 mM Tris-HCl buffer (pH 7.5) was prepared in a glove box. The reaction was initiated by exposing the mixture to air, and incubated at 30°C for 2 h to achieve full conversion to γ-Cl-1-ACP_II. The product was detected by FTICR-MS and IRMPD as described in Supplementary Methods.

Cur and Jam ECH₁/ECH₂ functional assays

1-ACP_II or γ-Cl-1-ACP_II was served as the substrate for ECH₁ and ECH₂ assays. Typically, ∼50 μM 1-ACP_II or γ-Cl-1-ACP_II was incubated with 1 μM ECH₁ or ECH₁/ECH₂ in 50 mM Tris-HCl buffer (pH 7.5) at 30°C. The reactions were examined by reverse-phase HPLC, and the products were detected by FTICR-MS and IRMPD.

Cur and Jam ER functional assays

Typically, ER reactions were performed by incubating ∼50 μM γ-Cl-3-ACP_II or 3-ACP_II with 1 μM ER and 0.5 mM NADPH in 50 mM Tris-HCl buffer (pH 7.5) at 30°C. Alternatively, the ER reaction was coupled with ECH₁/ECH₂ dehydration and decarboxylation. The products were analyzed by FTICR-MS and IRMPD.

One-pot reactions and GC-MS analysis

Each one-pot reaction was performed by incubating ∼50 μM ACP₃ or Hal-ACP₃ loaded with 1-CoA, and ∼10 μM Cur and/or Jam enzymes with the corresponding cofactors in 50 mM Tris-HCl buffer (pH 7.5) at 30°C for 5 min. The reactions were initiated by exposing the reaction mixture to air. The products were cleaved from Cur ACP₃ and Hal-ACP₃ PPant arms by butylamine aminolysis to generate the butylamide derivatives that were subsequently analyzed by GC/MS³⁴ and compared with the authentic standards for structure determination. See Supplementary Methods for detailed protocols.

Analysis of regiochemical control by ECH₂ WT and mutants

The ratios of α,β and β,γ C=C products of the ECH₂ decarboxylation were measured for Cur ECH₂ WT and mutants, and Jam ECH₂ WT by a coupled ECH₁/ECH₂ dehydration and decarboxylation assay. 50 μM γ-Cl-1-ACP_II was incubated with 2 μM Cur ECH₁ and 2 μM Cur or Jam ECH₂ in 50 mM Tris-HCl buffer (pH 7.5) at 30°C for 45 min. The reaction mixtures were treated with 2 μM Jam ER and subjected to IRMPD-based quantification as described Supplementary Methods.