Engineering a PKS assembly line with a promiscuous loading module-Design of MAT-swapped loading modules for split VEMSFirst, we sought to introduce the polyspecific MAT domain into the VEMS loading module (M0), thereby creating a promiscuous hybrid loading module that allows for priming PK synthesis with various acyl-CoA starter substrates. The VEMS M0 comprises an adenylation (A) domain, responsible for the selection and activation of the starter substrate DHBA, which is subsequently loaded onto the ACP domain (Fig. 2A). Additionally, VEMS M0 features a nonfunctional KR domain38 that is believed to play a structural role39. In the design of hybrid modules, it is inevitable to generate non-cognate domain-domain interactions (DDIs). Depending on the characteristics of the non-cognate interface, this can affect the performance of the engineered PKS. Due to the intricate nature of this issue, two different hybrid designs were evaluated (Fig. 3A): For one hybrid M0 (H1M0) the A-KR section was replaced by the MAT domain derived from mFAS and the VEMS ACP0 was further exchanged with the mFAS ACP, thereby ensuring native DDIs between MAT and ACP, both from mFAS, interacting during transacylation. In this case, a non-cognate interface arises during the translocation step between the mFAS ACP of M0 and the VEMS KS1 of the downstream M1 (Fig. S3). In the second hybrid M0 (H2M0), the VEMS ACP0 was retained providing native interplay between the ACP0 and the KS1 during translocation. In this case, non-cognate DDIs occur during the transacylation between the mFAS MAT and the VEMS ACP0. The constructs were conceived based on a recent design of an mFAS-derived loading module from our lab. In this construct, the mFAS MAT domain was excised and fused to the mFAS ACP via a flexible 12-residue long linker33. Both hybrid modules H1M0 and H2M0 feature a C-terminal DEBS-derived docking domain to enable docking to the split VEMS M1, carrying the matching docking domain N-terminally.Fig. 3: Split VEMS with a promiscuous loading module.A Two designs for a promiscuous loading module, H1M0, and H2M0 were evaluated. mFAS-derived domains are indicated in red. The production of 6-ethyl-4-hydroxy-2-pyrone 3 was analyzed via LC-HRMS (6-ethyl-4-hydroxy-2-pyrone 3 [M + H]+: m/z (calcd.) = 141.0547). Extracted ion chromatograms are shown for both PKSs using either H1M0 or H2M0. B The hybrid loading module H2M0 was challenged with different starter substrates, enabling the production of 2, 3, and 4. Normalized extracted ion chromatograms (EICs) of 4-hydroxy-6-methyl-2-pyrone 2 (triacetic acid lactone, TAL) and its derivatives are depicted (4-hydroxy-6-methyl-2-pyrone 2 [M + H]+: m/z (calcd.) = 127.0390, 6-ethyl-4-hydroxy-2-pyrone 3 [M + H]+: m/z (calcd.) = 141.0547, 4-hydroxy-6-propyl-2-pyrone 4 [M + H]+: m/z (calcd.) = 155.0703, (E)-4-hydroxy-6-propenyl-2-pyrone 5 [M + H]+: m/z (calcd.) = 153.0547, and 4-hydroxy-6-(2-hydroxypropyl)-2-pyrone 6 [M + H]+: m/z (calcd.) = 171.0652). C Mass spectra of compound 2 (found m/z = 127.0398), 3 (found m/z = 141.0550), and 4 (found m/z = 155.0696).Evaluation of hybrid loading modulesBoth hybrid loading modules were produced in E. coli. While H1M0 was attempted to be purified via a single affinity tag, but did not achieve sufficient purity, H2M0 was received in high purity via tandem affinity chromatography, which was further validated via size exclusion chromatography (Fig. S4). To assess the functionality of the hybrid loading modules, they were tested within the context of a complete assembly line (HXM0–M1–M2-TE, Fig. 3A). It was assumed that the MAT domain loads its C-terminal ACP (mFAS ACP for H1M0 and VEMS ACP0 for H2M0) with a starter substrate, which then primes synthesis in split VEMS M1. Next, the starter substrate is elongated twice with MalCoA in M1 and M2-TE. The resulting triketide intermediate bound to the ACP2 of the final module M2-TE is then released under pyrone-ring formation. To prove the functionality of the loading module, we decided to use propionyl-CoA (PrpCoA), which is well-accepted by the MAT domain33. Since the VEMS assembly lines may be primed by M1 AT-mediated loading of the extender substrate MalCoA and subsequent M1 KS-mediated decarboxylation, the use of acetyl-CoA (AcCoA) as a starter substrate could lead to ambiguous results. In contrast, successful incorporation of PrpCoA results in the formation of 6-ethyl-4-hydroxy-2-pyrone (3), and can be distinguished from decarboxylative priming with MalCoA leading to triacetic acid lactone (TAL, 2). As analyzed by liquid chromatography-high resolution mass spectrometry (LC-HRMS), with PrpCoA as the priming substrate, expected compound 3 could be detected using H2M0 (Fig. 3A), preserving the native ACP-KS interface of chain translocation. We interpret the lack of activity in H1M0 containing PKS assembly lines as a result of the translocation reaction across the non-cognate interface. The condensation reaction (comprising chain translocation and elongation reaction) has been shown before by others and us to be rate-limiting in polyketide synthesis40,41,42,43,44,45, such that any interference in the interaction between ACP and downstream KS can hinder substrate turnover. We link the lack of activity in H1M0-containing PKS assembly lines to the translocation reaction which involves the non-cognate mFAS ACP:VEMS KS1 interface. Studies have shown that the interference in the translocation reaction can throttle turnover43,44,45,46, and, vice versa, that engineering strategies preserving the cognate interaction can maintain the efficiency of engineered PKSs39,47,48.Exploiting the promiscuity of the MAT domain to produce 2-pyrone derivativesWith a functional MAT-swapped loading module in hand, we sought to exploit the reported promiscuity of the MAT domain33 to load the VEMS-based platform with a pool of starter substrates (Fig. 3B). A total of five starter substrates (acetyl-CoA, propionyl-CoA, butyryl-CoA, crotonyl-CoA, and 4-hydroxybutyryl-CoA), known to be accepted by the MAT domain133, were tested. To investigate the tolerance of the downstream modules, these starter substrates were selected to possess varying chain lengths and oxidation patterns at the α and β positions. LC-HRMS analysis confirmed the production of 2-pyrone compounds (2–4, Fig. 3C) from three out of five tested starter substrates.MAT swaps in elongation modules of split VEMSSubsequently, we sought to replace the MalCoA-specific transferases (AT1 and AT2) of VEMS elongation modules (M1 and M2-TE) with the promiscuous MAT domain. Previous studies have demonstrated that hybrid PKSs created by domain swapping frequently suffer from decreased catalytic activity and insolubility due to misfolding of the protein subunits49,50. Due to the lack of original structural data for VEMS, we utilized computational modeling (ColabFold51) to predict the structures of relevant VEMS subunits. The linker domain (LD) is of particular importance for AT/MAT-swaps, as it serves as a spacer between the KS and transferase (AT or MAT) domain. The LD is comprised of two distinct parts: LD1, which is the part sandwiched between KS and transferase, and LD2, which wraps back onto the KS surface and connects the transferase domain to the downstream ACP domain (Figs. 4A, S5, and S6). The essential question for the AT/MAT-swap is whether the transferase should be swapped, including or excluding the LD or whether it should be cut within the LD fold. In order to increase the chance of creating a functional hybrid construct, a broad range of domain boundaries for AT/MAT-swaps was tested. We categorized boundaries into three groups depending on how the LD was treated, i.e., whether it was retained from the acceptor module (VEMS M1 or VEMS M2-TE) or derived from the donor module (mFAS) (Figs. S5 and S6). In H1 constructs, the boundaries were positioned close to the core fold of the AT domain while preserving the LD (LD1- and LD2-part) of the accepting VEMS. Conversely, in H2 constructs, the LD was derived from the donating mFAS. Lastly, in the H3 group, the boundaries were set so that the LD1-part originated from the acceptor module, while the LD2-part was derived from the donor module (Fig. 4B, C).Fig. 4: Boundaries of AT/MAT-swapped VEMS constructs M0-M1*.A Structural model of the VEMS M1 KS-AT didomain fold (monomeric) predicted with ColabFold51. KS, LD1, AT, and LD2 domains are depicted in blue, orange, green, and gray cartoons, respectively. The LD is marked by a gray box. B Zoom into the LD. For a better orientation, LD1-building secondary structure elements are numbered. MAT-swaps of the H1, H2, and H3 group are depicted. The residue ranges in which the junction sites were chosen differ for each hybrid group and are highlighted in red. C Shortened Multiple Sequence alignment (MSA) of the KS-AT didomain sequences of mFAS and all elongation modules from DEBS and VEMS. The full MSA is provided in Fig. S6. Swap junctions upstream of the AT domain are termed usXX (XX is the residue position according to the first junction site us1, indicated in the MSA) and downstream of the AT domain dsXX in accordance with a previous study on AT swaps49. Secondary structure elements of the predicted structure of the VEMS M1 KS-AT didomain are indicated. The range in which the respective swap junctions (indicated in red) of the hybrid groups were chosen is framed with dashed lines.AT/MAT-swapped elongation module VEMS M2-TEFor VEMS M2 TE, a total of 24 hybrid constructs M2*-TE were designed (Fig. S5), of which 6 constructs belong to the H1-group, 8 to the H2-group, and 10 to the H3-group, and subjected to test expression in E. coli (Tables S1–S4). Sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) analysis revealed that certain constructs express, albeit at insufficient quantities (<0.1 mg/L E. coli culture) to allow protein isolation and analysis (Figs. S7 and S8). The difficulty in obtaining functional hybrid constructs in sufficient quantities underscores the intricate nature of the KS-AT didomain fold.AT/MAT-swapped elongation module VEMS M1Similarly, as for VEMS M2-TE, different boundaries for the AT/MAT-swap in the first elongation module M1 were tested. M1 engineering was performed in the context of the full-length polypeptide VemG (M0-M1) because we assumed the highest protein stability under native-like conditions. The boundaries of the best expressible hybrid construct should then be applied to the engineered M1 of split VEMS (M1). Four constructs were tested for expression in E. coli (2 of H1, 1 of H2, and 1 of H3 group, Figs. 4B and S6), of which one construct of the H1 group yielded enough protein for further analysis (Fig. S9). To confirm functionality, the AT/MAT-swapped M0–M1* was assembled with M2-TE to VEMS. LC-HRMS analysis confirmed the formation of venemycin 1 when using the substrates DHBA and MalCoA, indicating the functionality of the hybrid module (Fig. S10). These boundaries were then utilized to create a stand-alone AT/MAT-swapped M1* (Fig. S11) that is compatible with split VEMS. The AT/MAT-swapped hybrid module M1* was also active when used as a stand-alone module within the split VEMS testbed, as confirmed by-product formation (Fig. S10).Since the polyspecific transferase offers promiscuous substrate loading, the hybrid assembly line was challenged with the non-native extender substrate MMalCoA, which can normally not be processed by split VEMS (Fig. 5). When using an extender substrate mixture of MalCoA and MMalCoA, the derivative 5-methyl-venemycin 7 was only produced by the assembly lines M0–M1* VEMS and M1* split VEMS, carrying the AT/MAT-swap, demonstrating the ability of the hybrid assembly line to incorporate non-native extender substrates. Of note, providing MalCoA and MMalCoA as extender substrates, M1* split VEMS also produced venemycin 1, as a result of loading both malonyl and methylmalonyl by the promiscuous MAT. The derivative 3,5-dimethyl-venemycin 8, which would result from the incorporation of MMalCoA by both elongation modules, could not be observed due to the substrate-specific AT of M2-TE.Fig. 5: M1* in VEMS and split VEMS.A M1* can be operated in the context of VEMS and split VEMS. The resulting assembly lines are able to produce the native VEMS product venemycin 1. Additionally, the MAT domain allows the incorporation of MMalCoA which is not accepted by the native AT1. This results in the production of a methylated venemycin derivative 7 which is not accessible by utilizing VEMS. B Normalized EICs of venemycin and its derivatives (venemycin 1 [M + H]+: m/z (calcd.) = 221.0445, 5-methyl-venemycin 7 [M + H]+: m/z (calcd.) = 235.0601, and 3,5-dimethyl-venemycin 8 [M + H]+: m/z (calcd.) = 249.0758. C Mass spectra of compound 1 (found m/z = 221.0442) and 7 (found m/z = 235.0610).Promiscuous assembly line with substrate-dependent starting pointsThe mFAS-derived MAT domain possesses a dual functionality, meaning that it can prime but also continue fatty acid synthesis by loading the appropriate starter (AcCoA) or extender substrate (MalCoA). This property is not found in modular PKSs, in which transferases are more diversified and either load the starter for priming synthesis or the extender substrate for growing the polyketide intermediate. In this regard, the AT/MAT-swapped M1* offers the chance to harness M1* as an alternative starting position for polyketide synthesis. As a first test, we supplied propionyl-CoA to the hybrid M1* assembly line, which cannot be processed by split VEMS but is accepted by the MAT domain33. Since AcCoA may arise in situ during synthesis by decarboxylation of MalCoA, we chose to work with propionyl-CoA. Confirmed by LC-HRMS, the hybrid split VEMS assembly line produced the respective 2-pyrone 3 (Fig. 6A). Based on these data, we conclude that the production involves priming by M1*, followed by an elongation step by M1* before handing over the intermediate to M2-TE. We propose that the propionyl-loaded ACP transfers propionyl back to the intramodular KS domain, instead of translocating it to the downstream module, which is a mechanism observed in some modular PKSs in the context of “module stuttering”52. Subsequently, M2-TE catalyzes a second elongation step and the release of compound 3.Fig. 6: Promiscuous assembly line with substrate-dependent starting points.A The dual functionality of the MAT domain can be used to start the PK synthesis at M1* and allows the utilization of propionyl-CoA as a starter substrate, resulting in the production of compound 3, which would not be accessible via split VEMS. EICs of 6-ethyl-4-hydroxy-2-pyrone 3 [M + H]+: m/z (calcd.) = 141.0547. B Utilization of the MAT domain in the elongation module M1* allows starting the synthesis at two different positions of the assembly line, depending on the nature of the used starter substrate. EICs of the compounds are provided in Figs. S12 and S13.The dual functionality of the MAT domain turns split VEMS into an assembly line with two priming points that can be utilized depending on the chemical structure of the starter substrate: DHBA-derivatives are activated by M0 (ref. 39) and forwarded to M1, while acyl-CoAs can be incorporated by MAT of M1*. Furthermore, the promiscuity of MAT in M1* enables the modulation of position 5 of the 2-pyrone scaffold when non-native extender substrates are used (Fig. 6B). We demonstrated the versatility of this assembly line by combining 5 different starter (DHBA, 3-hydroxybenzoic acid, acetyl-CoA, propionyl-CoA, and butyryl-CoA) and two extender substrates (MalCoA and MMalCoA) leading to 10 products with a 2-pyrone scaffold generated by one engineered PKS.AT/MAT-swaps in mono-modular PIKS systemTo further interrogate the utility of the AT/MAT-swap strategy, we engineered a late-stage elongation module of the well-characterized PIKS, which is responsible for the production of the precursors of the antibiotics methymycin and pikromycin53,54. We chose the recently engineered PIKS module 5 (PIKS M5) that is C-terminally elongated with the PIKS M6 TE domain giving an overall domain structure of KS-AT-KR-ACP-TE (termed PIKS M5-TE)37. Domain boundaries for the AT/MAT-swap were chosen in accordance with the H1 boundaries previously established by Rittner and Joppe et al. 25 (Fig. S14), which also performed best in the comprehensive hybrid design screening of VEMS. PIKS M5-TE and the AT/MAT-swapped PIKS M5*-TE were produced in E. coli, with PIKS M5*-TE yielding about 50% of the amount of the non-swapped PIKS M5-TE (Fig. S15).First, the activities of PIKS M5-TE and PIKS M5*-TE were tested chemoenzymatically by the in vitro product formation assay established by Sherman and coworkers55 (Fig. 7A). Activated PIKS pentaketide was employed as the elaborate starter substrate mimic and MMalCoA was added as the native extender substrate of PIKS M5. The NADPH consumption of KR5 was monitored fluorometrically to evaluate the reaction progress (Fig. 7C). As confirmed by LC-HRMS, the hybrid module PIKS M5*-TE was able to produce the methymycin precursor 10-deoxymethynolide (10-dml, 14), although at a reduced speed of 0.18 min-1 compared to PIKS M5-TE (1.6 min-1). Next, we challenged the hybrid PIKS M5*-TE with the non-native extender substrates MalCoA and fluoromethylmalonyl-CoA (FMMalCoA). The use of di-substituted FMMalCoA was first described in the context of a FAS/PKS hybrid of DEBS M625. FMMalCoA gives access to motifs found in solithromycin. Additionally, the incorporation of the F/Me-substitution demonstrates the ability of direct disubstitution of macrolactone scaffolds. The formation of the expected demethylated and fluorinated 10-dml derivatives 2-demethyl-10-dml 16 and 2-fluoro-10-dml 17 was confirmed by LC-HRMS (see Fig. 7B for normalized EICs and S16 for mass spectra). NADPH monitoring revealed the highest turnover using the extender substrate MalCoA (0.48 min-1), which is the preferred substrate of the MAT domain. Yet, 2-demethyl-3-oxo-10-dml 15 was identified as the dominant product when incubated with MalCoA due to impaired acceptance of the demethylated intermediate by the KR5 domain such that the TE releases the non-reduced macrolactone 1556. Unexpectedly, the formation of 2-fluoro-10-dml 17 was also observed with PIKS M5-TE, indicating that the native AT domain is “leaky” for the di-substituted FMMalCoA substrate.Fig. 7: Analysis of AT/MAT-swapped PIKS M5-TE in chemoenzymatic synthesis.A Reaction scheme for the PIKS M5-TE-mediated elongation of PIKS pentaketide with native substrate MMalCoA or non-native substrates MalCoA and FMMalCoA to produce WT product 14 and derivatives thereof (15–17). B EICs of 10-dml and its derivatives detected by LC-HRMS (14 [M + H-H2O]+: m/z (calcd.) = 279.1955, 15 [M + H-H2O]+: m/z (calcd.) = 263.1642, 16 [M + H-H2O]+: m/z (calcd.) = 265.1798, 17 [M + H-H2O]+: m/z (calcd.) = 297.1860). For both constructs, the peaks of 15–17 were normalized to 10-dml 14. C Turnover rates measured by NADPH consumption of M5-TE- and M5*-TE-mediated reactions using different extender substrates.AT/MAT-swaps in bi-modular DEBS systemNext, we applied the AT/MAT-swap to DEBS3, the ultimate bimodular polypeptide of the 6-deoxyerythronolide B-producing DEBS assembly line. DEBS3 comprises two covalently connected elongation modules, module 5 (M5) and module 6 (M6), each possessing the KS-AT-KR-ACP module composition and a C-terminal TE domain facilitating product release. Overall, we designed three bimodular PKS/FAS hybrid proteins by systematically replacing the DEBS AT domains (AT5 and AT6) with the MAT domain using H1 boundaries (Fig. S14). In hybrid M5*–M6-TE, DEBS AT5 was replaced with the MAT domain, in hybrid M5–M6*-TE DEBS AT6 was replaced with the MAT domain, while in M5*–M6*-TE both AT domains were replaced with the MAT domain.While hybrid M5*–M6-TE and M5–M6*-TE would allow the modification of position 4 or 2 of the C14-macrolactone product, respectively, hybrid M5*–M6*-TE simultaneously targets both positions. Recombinant production in E. coli yielded the hybrid M5–M6*-TE in high purity and similar yields to the WT DEBS3 (11 mg/L and 13 mg/L of purified protein, respectively, Fig. S17). In contrary, M5*-M6-TE was obtained in lower yield (0.4 mg per liter) compared to DEBS3, and the double hybrid M5*–M6*-TE could not be isolated at all.As before, the activated PIKS pentaketide was employed as a starter substrate, and MMalCoA was added as the native extender substrate of DEBS AT5 and AT6 (Fig. 8A)33. The reduction of NADPH by KR5 and KR6 was monitored fluorometrically to evaluate the reaction progress (Fig. 8B). DEBS3 as well as both hybrids revealed to be active and produced the C14-macrolactone 3-hydroxy-narbonolide 18 by two elongations. Of note, both hybrids, as well as native DEBS3 used as a control, produced 10-dml 14 as a byproduct. Two pathways could account for the detected 12-membered macrolactone byproduct: Production of 10-dml involves either omission of M6 after elongation by M5, followed by direct release through the TE domain of M6, resembling the natural PIKS pathway that accounts for 10-dml synthesis, or direct loading of KS6 of module 6 with the PIKS pentaketide followed by elongation and processing to 10-dml, a mechanism which was previously exploited by others and us19,25,57.Fig. 8: Analysis of AT/MAT-swapped DEBS3-TE in chemoenzymatic synthesis.Enzyme activity check. A Reaction scheme for the consecutive elongation of PIKS pentaketide with native substrate MMalCoA mediated by hybrids M5*–M6-TE and M5–M6*-TE to produce WT product 18. 10-dml 14 is formed as a byproduct, which is explained by either skipping of M6 and direct translocation onto the TE domain or by direct loading of PIKS pentaketide onto the KS of M6. B Both hybrids demonstrate high turnover rates monitored by consumption of NADPH.Subsequently, the functional hybrid proteins were examined for their ability to produce derivatives of 14-membered macrolactone by providing 1:1 mixtures of extender substrates (MMalCoA/MalCoA or MMalCoA/FMMalCoA) and NADPH (Fig. 9A, B).Fig. 9: Production of 14-membered macrolactones derivatized at C-2 or C-4.A Reaction scheme for the M5*–M6-TE- or M5–M6*-TE-mediated conversion of PIKS pentaketide with MMalCoA/MalCoA substrate mixture to produce macrolactone derivatives 19 and 21. B Reaction scheme for the M5*–M6-TE- or M5–M6*-TE-mediated conversion of PIKS pentaketide with MMalCoA/FMMalCoA substrate mixture to produce macrolactone derivatives 20 and 22. We note that the chemical identity of compounds 19–22 was determined by HPLC-HRMS, which does not give conclusive evidence about the regioselectivity of polyketide modification. C, D EICs of macrolactones detected by LC-HRMS (18 [M+Na]+: m/z (calcd.) = 337.2304; 19/21 [M+Na]+: m/z (calcd.) = 363.2142, 20/22 [M+Na]+: m/z (calcd.) = 395.2204). For each construct, peaks 19/21 and 20/22 are normalized to 3-hydroxy-narbonolide 18, respectively.While for both substrate mixtures, native DEBS3 produced WT product 3-hydroxy-narbonolide 18 only (with 10-dml 14 observed as a byproduct via module skipping, see Figs. S18A and S19A, pathway (1)), the AT/MAT-swapped hybrid proteins led to more complex product outputs. M5*–M6-TE produced 4-demethyl-3-hydroxy-narbonolide 19 from the MMalCoA/MalCoA mixture (Fig. 9A) as well as 4-fluoro-3-hydroxy-narbonolide 20 from the MMalCoA/FMMalCoA mixture (Fig. 9B). In MMalCoA/MalCoA-containing samples, hardly any WT product 18 was observed in LC-HRMS and demethylated C-4 regioisomer 19 was the main product reflecting the efficiency of the MAT domain for MalCoA over MMalCoA (Fig. 9C for normalized EICs and S20 for mass spectra). In contrast, a mixture of WT product 18 and C-4-fluorinated product 20 was observed in samples containing MMalCoA/FMMalCoA substrate mixture (Fig. 9D for normalized EICs and S20 for mass spectra). The main product was compound 18, which is in line with the diminished efficiency of MAT towards FMMalCoA observed before25. For MMalCoA/MalCoA substrate mixtures, we also identified 12-membered products 10-dml 14, 2-demethyl-3-oxo-10-dml 15 and 2-demethyl-10-dml 16 from LC-HRMS, which was attributed to either skipping of hybrid module M5* and direct loading of the starter substrate to M6 and elongation with MMalCoA (see Fig. S18B, pathway (2)), or elongation with MalCoA in hybrid module M5*, optional KR5 skipping, followed by skipping of module M6 and direct TE-mediated offloading (see Fig. S18B, pathways (3) for 15 and (4) for 16). We note that the chemical identity of compounds was determined by HPLC-HRMS, which does not give conclusive evidence of the regioselectivity of polyketide modification. Given this limitation, NMR analysis of compounds will be necessary to confirm the modification at position 4 without ambiguity.Similarly, hybrid M5–M6*-TE was subjected to mixtures containing MMalCoA/MalCoA or MMalCoA/FMMalCoA in order to prepare C-2 analogs 2-demethyl-3-hydroxy-narbonolide 21 and 2-fluoro-3-hydroxy-narbonolide 22 of WT product 18. Mass peaks attributed to the respective C-2-demethylated or C-2-fluoromethylated compounds 21 and 22 were identified (Fig. 9C for 21 and Fig. 9D for 22). Substantial shifts in retention time of approx. 0.5 min for 3-hydroxy-2-demethyl-narbonolide 21 and 0.3 min for 2-fluoro-3-hydroxy-narbonolide 22, with respect to the C-4 compounds 19 and 20, indicate that we produced different regioisomers. Chang and coworkers30 have recently shown that C-2 and C-4 regioisomers of C14 macrolactones can be distinguished based on their retention time. However, again we note the NMR analysis will need to confirm suggested compound structures. Hybrid M5–M6*-TE produced the demethylated C-2 regioisomer 21 as the main product in mixtures containing MMalCoA/MalCoA. No WT product 18 was observed in LC-HRMS, but 2-demethyl-3-oxo-10 dml 15 and 2-demethyl-10-dml 16 were identified (see Fig. S18D) and explained by direct loading of the starter substrate onto KS6 of the hybrid module M6*, elongation with MalCoA and offloading of the non-reduced (KR-skipping) and the reduced compound (Fig. S18C, pathways (5) and (6)). The absence of 10-dml 14 in LC-HRMS indicated efficient translocation of the intermediate from M5 to M6*, loading of MalCoA in M6* only, and no M6*-skipping. A product mix of C14-membered products 18 and 22 (with 10-dml 14 as a byproduct, see Fig. S19) was obtained for MMalCoA/FMMalCoA assays similar to the product mix obtained by hybrid M5*–M6-TE.
