Directed biosynthesis of fluorinated polyketides

Modification of polyketides with fluorine offers a promising approach to develop new pharmaceuticals. While synthetic chemical methods for site-specific incorporation of fluorine in complex molecules have improved in recent years, approaches for the direct biosynthetic fluorination of natural compounds are still rare. Herein, we present a broadly applicable approach for site-specific, biocatalytic derivatization of polyketides with fluorine. Specifically, we exchanged the native acyltransferase domain (AT) of a polyketide synthase (PKS), which acts as the gatekeeper for selection of extender units, with an evolutionarily related but substrate tolerant domain from metazoan type I fatty acid synthase (FAS). The resulting PKS/FAS hybrid can utilize fluoromalonyl coenzyme A and fluoromethylmalonyl coenzyme A for polyketide chain extension, introducing fluorine or fluoro-methyl disubstitutions in polyketide scaffolds. Addition of a fluorine atom is often a decisive factor toward developing superior properties in next-generation antibiotics, including the macrolide solithromycin. We demonstrate the feasibility of our approach in the semisynthesis of a fluorinated derivative of the macrolide antibiotic YC-17.

enzymes that catalyze addition of F atoms in secondary metabolism 6 . Thus, new methods that enable fluorine derivatization are urgently needed to bridge the gap between the inherent bioactivity of a natural compound and its development as a human therapeutic agent.
Polyketide natural products comprise over 10,000 molecules with a wide range of bioactivities and are among the most prominent classes of approved clinical agents 7,8 . In nature, polyketides are assembled mainly from simple monomeric acetate and propionate units by polyketide synthases (PKSs). In the type I cis-AT subclass, PKSs occur as multi-functional protein mega-complexes comprising a series of catalytic domains organized in modules on one or few polypeptide chains. Typically, one PKS module requires a minimum of three domains for a two-carbon extension of a growing polyketide intermediate: an acyltransferase (AT) domain that selects an acyl-coenzyme A (CoA) extender unit and transfers the acyl moiety to the acyl carrier protein (ACP) domain, and a ketosynthase (KS) domain that accepts a growing chain from the ACP of the previous module and catalyzes a decarboxylative Claisen condensation to extend the polyketide chain. A canonical PKS module may further contain up to three additional domains, a ketoreductase (KR), a dehydratase (DH) and an enoylreductase (ER) that tailor the βketo functionality prior to the next round of chain extension. The final module in the biosynthetic assembly line typically contains a thioesterase (TE) domain located at the C-terminus and is responsible for polyketide chain release as a linear chain or a macrocyclic product. Engineering of modular polyketide biosynthesis for the directed assembly of new-to-nature polyketides is a highly aspired aim 9 , and offers an alternative or complementary approach to organic synthesis.
Changing substrate specificity of a single enzymatic domain of a specific PKS module by protein engineering enables, for example, the regioselective modification of the product during biosynthesis.
Enzymes that catalyze direct fluorination of polyketides remain unknown. Previous efforts have demonstrated that engineered PKS assembly lines at even-numbered carbon centers enable loading of non-canonical extender substrates during the chain elongation process (Fig. 1a) 10 . Indeed, also fluoromalonyl-CoA was incorporated into a growing triketide chain followed by spontaneous off-loading lactonization 11,12 . However, the application of this concept to the formation of a complete macrolide structure had not been demonstrated. In canonical modular PKSs, extender subunits are selected by the AT domains, which act as the "gatekeepers" of polyketide biosynthesis and typically ensure the introduction of a defined acyl-CoA with high substrate specificity ( Supplementary Fig. S1). We have recently demonstrated that the promiscuous AT domain from metazoan fatty acid synthase (FAS), termed malonyl-/acetyl transferase (MAT), is able to transfer various acyl-CoA moieties with high efficiencies 13 , different to AT domains from PKSs 14 . We hypothesized that this domain may also transfer fluorinated extender substrates into a PKS module, since FASs and PKSs are structurally and biochemically related (Fig 1b) [15][16][17] . To test the feasibility of this approach, we chose to work with module 6 of DEBS including its C-terminal TE domain (DEBS M6+TE) (Fig. 1c) 18 .

Results and discussion
Initially, we tested whether the polyspecific MAT domain of murine FAS is able to select fluoromalonyl-coenzyme A (F-Mal-CoA) as a substrate, and whether it accepts the ACP domain of the DEBS M6 for substrate processing. F-Mal-CoA (1) was chemically synthesized following the four-step route to the thiophenyl-ester by Saadi and Wennemers 19 with the additional subsequent transacylation of the F-Mal moiety to free CoA (Fig. 2a, Supplementary Fig. S2-S3).
In an enzyme-coupled fluorometric assay with the domains as individual proteins 13 , we observed excellent transfer kinetics of MAT for F-Mal-CoA (K m /k cat = 6.9 × 10 6 M -1 s -1 ) as well as its ability to catalyze transacylation with DEBS ACP6 (Supplementary Fig. S4-S7). The specificity constant of MAT for loading ACP6 with methylmalonyl moieties (K m /k cat = 6.9 × 10 6 M -1 s -1 ) was 2 -3 orders of magnitude higher than DEBS AT6 (Supplementary Table S1), which can be explained by the inherently high transacylation rates of the MAT domain.
We constructed two hybrids of S. erythraea DEBS and murine FAS, which differed in the  With the catalytically competent hybrid H1 in hand, we aimed to produce 12-membered macrolactones from the pikromycin pathway that are diversified at the C-2 position (Fig. 1a,  Supplementary Fig. S11). These are particularly interesting as exhibiting microbial activity against erythromycin-resistant Staphylococcus aureus strains and bind in a unique way to the 50S ribosomal subunit [20][21][22] . Elongation of the pentaketide (9) with MM-CoA, Mal-CoA and F-CoA, respectively, produced 10-deoxymethynolide (10) as well as the respective desmethylated macrocycle (12) and the fluorinated analog (14) 23 . The absence of NADPH led to the C-3 oxidized species (11, 13 and 15) (compounds 10-15 were confirmed by HRMS; Fig. 3a  Intriguingly, when seeking to conduct scale-up to isolate milligram quantities, we faced challenges for the reactions to the fluoro-compounds 14 and 15. Here, very low amounts of products were obtained, and reactions were contaminated with significant levels of side products. When working-up the reaction mixture to target compound 15, we identified compound 16 as the main product, presumably generated from the hexaketide intermediate via hydrolysis, decarboxylation and cyclohexanone formation (Fig. 3d). This side reaction has been reported previously as originating from the narrow substrate specificity of the DEBS TE domain, preventing macrolactonization to the 12-membered ring when a keto-group is present at C-3 24 .
We elected to pause further analysis of the origin for the low yields of compound 14, and reasoned that chemical instability relates to the C-2 HFC group.  4a). Finally, F-MM-CoA proved to be accepted by H1 for elongation of the pentaketide (9) in presence of NADPH to produce 2-fluoro-10-deoxymethynolide (18) (for mechanistic implications, see Supplementary Fig. S14). Conversion of the extender unit was verified by the NADPH consumption assay (turnover rate: 0.14 ± 0.02 min -1 ) as well as HRMS, and reaction scale-up provided full structural analysis by NMR (Fig. 4b).  Chemical synthesis was performed analogously to 1 from the respective Meldrum's acid and the product was converted enzymatically with pentaketide 9 and NADPH to macrolactone 18.

Methods
Methods, additional references and spectra are available in the supplementary information.

General synthesis of fluorinated CoA-Ester
Fluoro-Meldrum's acid was synthesized from Meldrum's acid in three steps using a previously described method utilizing Selectfluor ® . Fluoromethyl-Meldrum's acid was directly received from Methyl-Meldrum's acid with Selectfluor ® . The meldrum's acids were treated with trimethylsilylthiophenol to produce the respective thiophenyl halfesters, and fluorinated CoA-Esters were eventually synthesized by transacylation from thiophenyl halfesters to free coenzyme A (CoASH). The CoA-Esters were purified by precipitation with acetone (-20 °C).

General procedure for the biosyntheses
Small scale biosynthesis of TKLs and macrolactones were carried out with 4-6.1 µM enzyme, 5 mM 2 or 1 mM 9, 200 µM X-CoA and 60 µM NADPH in the assay buffer (400 mM phosphate buffer, 20 % (v/v) glycerol, 1 mM EDTA, 0.8 % DMSO, pH 7.2) or the reaction buffer (250 mM potassium phosphate, 10 % glycerol, pH 7) at 25 °C. The reactions were followed fluorometrically by monitoring the consumption of NADPH. Products were extracted with EtOAc and confirmed by HPLC-MS.

General procedure for the up-scaled syntheses of macrolactones
In order to receive larger amounts of macrolactones, reactions were carried out in 10-50 mL scale with the final concentrations of 5-10 µM H1, 300-600 µM 9, 400-4000 µM X-CoA and 500-1000 µM NADPH in the reaction buffer at 25 °C. After at least 4 h of incubation, products were extracted with EtOAc and purified on a silica column. The macrolactones 15 and 16 were additionally purified by HPLC on a C18 column.