Enzymatic synthesis of bioactive quinolones and (thio)coumarins by fungal type III polyketide synthases

Abstract

Quinolones are privileged scaffolds for drug discovery that are relatively rare in nature. Here, we characterise two promiscuous fungal polyketide synthases AthePKS and FerePKS, which we had previously found to produce 2‐quinolones in vitro. We challenged the enzymes with several substituted anthranilic acid derivatives, revealing their ability to produce precursors of pharmaceutically relevant quinolones. We also discovered that AthePKS and FerePKS accept other 2‐substituted benzoic acids, leading to the formation of coumarin and thiocoumarin scaffolds. We applied AthePKS in an artificial enzymatic cascade towards an antimicrobial 4‐methoxy‐1‐methyl‐2‐quinolone and demonstrated its in vivo feasibility by successfully expressing the pathway in Escherichia coli. Lastly, we determined the crystal structure of AthePKS, suggesting hotspots for enhancing its catalytic efficiency by enzyme engineering. Our results provide a framework for further engineering of enzymatic routes towards privileged heteroaromatic scaffolds and derivatives thereof.

Quinolones are valuable scaffolds for drug discovery but are rare in nature. Here, we show that two fungal enzymes, AthePKS and FerePKS, can generate 2‐quinolones and two additional heteroaromatic scaffolds. Using AthePKS, we designed an artificial enzymatic cascade towards an antimicrobial quinolone from a simple precursor and implemented it in E. coli. We also determined the AthePKS crystal structure, suggesting hotspots for improving its catalytic efficiency by protein engineering.

Abbreviations

ATP

adenosine triphosphate

CoA

coenzyme A

HPLC

high performance liquid chromatography

LC–MS

liquid chromatography‐coupled mass spectrometry

MS/MS

tandem mass spectrometry

NP

natural product

OMT

O‐methyltransferase

SAH

S‐adenosylhomocysteine

SAM

S‐adenosylmethionine

T3PKS

type III polyketide synthase

Introduction

Natural products (NPs) are an inexhaustible source of biologically active compounds. Structurally and functionally refined through evolution, these molecules have been a major source of inspiration in drug discovery and development for decades [1, 2]. However, the direct use of NPs is often impeded by their poor pharmacokinetics, and their structural complexity renders chemical modification challenging [3, 4]. To bypass these limitations, many researchers have focused on utilising the basic molecular scaffolds of NPs as platforms for drug discovery [5].

One such ‘privileged’ scaffold is 2‐quinolone, also known as carbostyril or 1‐azacoumarin. Found in a group of bioactive alkaloids mainly from Rutaceae plants, it is also at the core of several marketed drugs with different modes of action (Fig. 1). Many synthetic 2‐quinolone derivatives have also been shown to possess promising antimicrobial [6, 7, 8], anti‐Alzheimer's [9], anticancer [10], and antidiabetic [11] activities. Due to the intrinsic photoactive properties of the conjugated 2‐quinolone system, this scaffold is being investigated for use in luminescent materials, cationic metal sensors, and other functional organic materials [12].

Fig. 1.

Examples of bioactive natural products and marketed drugs containing the 2‐quinolone scaffold.

Numerous strategies for the chemical synthesis of 2‐quinolones have been developed [12, 13]. In nature, this scaffold is typically represented by 4‐hydroxy‐1‐methyl‐2‐quinolone, as synthesised by type III polyketide synthases (T3PKSs). The reaction involves decarboxylative condensation of malonyl‐coenzyme A (CoA) with N‐methylanthraniloyl‐CoA to form a diketide intermediate which then cyclises to yield the heteroaromatic quinolone. To date, only a handful of dedicated quinolone synthases have been reported in plants and bacteria [14, 15, 16, 17], and their application in the biosynthesis of 2‐quinolone derivatives remains underexplored.

In a previous study [18], we performed activity profiling of 37 fungal T3PKSs with 12 acyl‐coenzyme A substrates. Unexpectedly, half of the active enzymes converted N‐methylanthraniloyl‐CoA to the heteroaromatic quinolone—an activity not previously observed in fungal T3PKSs. From this set of active enzymes, we selected the two best‐performing T3PKSs for further characterisation. Here, we combined these enzymes with a bacterial CoA ligase to generate a series of substituted quinolone derivatives in one pot. We further demonstrated that both enzymes could accept (thio)salicylic acid derivatives, resulting in the formation of three distinct heteroaromatic scaffolds. Harnessing this promiscuity, we developed a biocatalytic cascade towards an antimicrobial O‐methylated 2‐quinolone with an unknown biosynthetic pathway. Finally, we demonstrated the feasibility of this cascade in a microbial host. Overall, the functional and structural insights generated in this study will guide future efforts towards the sustainable production of privileged heteroaromatic scaffolds.

Results

AthePKS and FerePKS produce N‐, O‐ and S‐containing heterocycles in vitro

First, we set out to test the substrate scope of the two T3PKSs on several anthranilic acid derivatives. Since T3PKSs typically act on CoA‐activated carboxylic acids, we employed the anthranilate‐CoA ligase PqsA from Pseudomonas aeruginosa [19] (UniProt: Q9I4X3) to generate the substrates for the T3PKSs. To streamline the analysis, we opted to perform the assays in a one‐pot continuous cascade with the respective carboxylic acids (1–8) as starter substrates and malonyl‐CoA (9) as the extender unit. We first tested the viability of such an approach with N‐methylanthranilic acid (1) using 1 mol% of each enzyme. Under these conditions, PqsA‐AthePKS and PqsA‐FerePKS converted 12–24% of 1 through N‐methylanthraniloyl‐CoA (1a) to the corresponding quinolone (1b).

We then turned to substituted anthranilic acid derivatives 2–5 (Fig. 2A) and assessed first whether they could be accepted by PqsA. We observed full conversion of 2 and 3 to the respective CoA‐activated intermediates 2a and 3a, whereas 4 and 5 were only partially converted—12% and 36%, respectively (Fig. S1). In the continuous cascade, anthraniloyl‐CoA (2a) was then converted to the 2‐quinolone product by both AthePKS and FerePKS, albeit less efficiently than 1. 5‐Chloro‐anthranilic acid (3) was accepted better than 2 by both enzyme cascades. Despite the low activity of PqsA with substrate 4, we detected trace amounts of the putative 2‐quinolone product with the expected m/z of 178 in reactions with AthePKS. 3,4‐Dimethoxyanthranilic acid (5) was not converted to the 2‐quinolone by either enzyme cascade.

Fig. 2.

Substrate scope analysis of AthePKS and FerePKS. (A) One‐pot continuous cascade with PqsA, Athe/FerePKS, extender unit 9 and carboxylic acids 1–8 (shaded) yields products 1b‐8b. The percentages correspond to % conversion of a given substrate to the final product under the assay conditions by AthePKS and FerePKS, respectively; ‘n.d.’—not detected. (B, C) Analysis of the reaction products of the PqsA‐AthePKS and PsqA‐FerePKS cascades with substrates 6 and 8 showing the formation of coumarin and thiocoumarin products by liquid chromatography‐coupled mass spectrometry (LC–MS). Solid lines represent the extracted ion chromatograms of the expected reaction product; dotted lines represent the authentic standards of 4‐hydroxycoumarin (6b) and 4‐hydroxythiocoumarin (8b). (D) Formation of 3‐methyl‐4‐hydroxy‐2‐quinolone (11) from 2 and methylmalonyl‐Coenzyme A (10) by the enzyme cascades.

We then challenged the enzymes with other 2‐substituted benzoic acids that we hypothesised would lead to heterocycle formation. Although CoA activation by PqsA was somewhat lower for substrates 6–8 (74%, 45% and 79% respectively, Fig. S1), they were accepted by both AthePKS and FerePKS with conversions comparable to those of anthranilic acid derivatives 1–3. This resulted in the formation of 4‐hydroxy(thio)coumarins (Fig. 2B,C). An overview of products 1b‐8b is presented in Figs S2–S4. Lastly, we assessed whether AthePKS and FerePKS could incorporate methylmalonyl‐CoA (10) as an extender unit, as observed in nature for several bacterial and plant T3PKSs [20]. To test this, we incubated substrates 2 and 10 with both enzyme cascades to yield 11, a C‐methylated quinolone with reported antimicrobial activity against several species of Mycobacterium [21] (Fig. 2D).

To our knowledge, this is the first report of the enzymatic formation of the 2‐quinolones 3b‐4b, coumarin 7b and thiocoumarin 8b. The 6‐chlorinated 2‐quinolone scaffold of 3b resembles the common pharmacophore of HIV‐1 non‐nucleoside reverse transcriptase inhibitors [22], which is also found in the anticancer drug Olutasidenib (Fig. 1). The 8‐hydroxylation of 4b is a common motif in several clinical trial drugs such as Abediterol and Navafenterol; it is also common in furoquinoline alkaloids from plants, where it is often methylated. 4‐Hydroxycoumarin 6b is the direct precursor of Warfarin and related anticoagulant compounds, while its fluorinated analogue 7b is the monomer of a potent homodimeric inhibitor of human NAD(P)H quinone oxidoreductase‐1 (NQO1), an enzyme overexpressed in several types of tumour cells [23]. Thiocoumarins, including 8b, possess anticoagulant properties similar to coumarins and have been used in commercial rodenticides [24].

Thus, even without optimisation, the one‐pot cascade with PqsA and a fungal T3PKS yielded a variety of pharmaceutically relevant heterocycles with moderate conversions. We then sought to further extend the cascade towards a more complex bioactive product to demonstrate its applicability.

Screening of methyltransferases for the 4‐OH‐methylation of quinolones

While researching the diversity of natural 2‐quinolones from Rutaceae plants, we encountered 4‐methoxy‐1‐methyl‐2‐quinolone (1c), a compound with reported antimicrobial [25, 26], antimalarial [27] and anti‐HIV [28] activity. Furthermore, compound 1c appears to be the precursor of bioactive Rutaceae alkaloids, although their biosynthesis is not experimentally elucidated (Fig. 3A). It has not escaped our notice that 1c is just one enzymatic step away from the AthePKS product 1b—specifically, it is methylated at the 4‐hydroxy position (Fig. 3B). We therefore hypothesised that AthePKS could be applied in a biocatalytic cascade to produce 1c.

Fig. 3.

Screening of O‐methyltransferases (OMTs) for the production of O‐methylated heterocycles 1c, 2c, 6c and 8c. (A) Chemical structures of 4‐methoxy‐1‐methyl‐2‐quinolone (1c) and 2‐quinolones hypothesised to be derived from it. (B) 4‐OH‐methylation of 1b to yield 1c (methoxy group shaded). (C) Screening of five OMTs for the methylation of 1b in cell‐free extracts. Solid lines represent the extracted ion chromatograms in low‐resolution LC–MS of the expected reaction product; dotted line represents the chemically synthesised reference of 1c. (D–G) Assays with purified DnrK, CmOMT2 and MesMOMT for the methylation of quinolones 1b and 2b, coumarin 6b and thiocoumarin 8b (methoxy groups shaded). The reactions were incubated with 0.5 mm substrate, 2 mm S‐adenosylmethionine (SAM) and 5 μm enzyme for 24 h. The methoxy groups are shaded. The data are represented as mean ± SD; n = 3.

Since the biosynthesis of 1c is not elucidated and, to our knowledge, no heterologous O‐methyltransferase (OMT) was ever reported to catalyse the methylation of 1b, there was no obvious candidate enzyme for this biosynthetic step. Hence, we searched the literature for OMTs that are active on substrates that are structurally similar to 1b and identified three candidates: the bacterial enzyme DnrK from the biosynthetic pathway of daunorubicin [29], the plant enzyme CmOMT2 that methylates hydroxycoumarins in other positions [30], and the fungal enzyme CpdC from the biosynthetic pathway of calipyridone A [31] (Fig. S5). We also noticed that the enzymes MesMOMT and GloKOMT from our previous study on the methylation of catechols [32] share the same structural fold as the three aforementioned OMTs and included them in the screening (Fig. S6).

We expressed the five genes in E. coli BL21(DE3) and tested the activity of the OMTs with 1b and S‐adenosylmethionine (SAM) as methyl donor in cell‐free lysates. With three out of five enzymes, we detected a product with an m/z value that corresponds to mono‐methylated 1b and co‐elutes with a chemically synthesised reference of 1c (Fig. 3C). Under the expression conditions, DnrK yielded the most 1c. Interestingly, MesMOMT, which methylated caffeic acid and other catechols in our previous study [32], was also active on 1b.

We expressed and purified the three active OMTs and tested them with 1b at 1 mol% enzyme to substrate. DnrK was still the best‐performing enzyme and under the same experimental conditions, it was also active with the other heterocycles 2b, 6b, and 8b, although the conversion rates were lower than for 1b (Fig. 3D–G, Fig. S7). This led us to select it for the cascade applications. Interestingly, the addition of 5 mm MgCl₂ to the reaction mixture increased the product concentration by 35% (Fig. 3D–G). To our knowledge, the effect of magnesium or other metal ions on DnrK or its homologues has not been previously examined. We therefore assessed the activity of DnrK in the presence of various metal salts (Fig. S8). Among the divalent cations tested, CaCl₂ and MnCl₂ led to an improvement in the conversion of 1b comparable to that with MgCl₂, while CoCl₂, CuSO₄, FeSO₄, NiSO₄ and ZnCl₂ inhibited the reaction. In contrast, the addition of monovalent salts (NaCl, LiCl) or the trivalent FeCl₃ had no notable effect relative to controls containing either ethylenediaminetetraacetic acid (EDTA) or no added metal. Since there is no evidence for a bound metal ion in the DnrK crystal structures (PDB ID 1TW2, 1TW3, 5EEG, 5EEH and 5JR3) [29, 33, 34], further investigation is necessary to determine if this positive effect of certain divalent cations is mediated by direct interaction of the ions with the enzyme active site or if there are indirect effects at play.

Assembling a biocatalytic cascade towards an antimicrobial 4‐methoxy‐1‐methyl‐2‐quinolone

With the necessary enzymes at hand, we proceeded to assemble a biocatalytic cascade towards 1c from 1 (Fig. 4A). The initial conditions for the cascade were adapted from the PqsA reaction [19] with the addition of AthePKS and DnrK at the same molar concentration as PqsA, and the corresponding cofactors. The three‐step one‐pot continuous cascade yielded 17 ± 0.75 μm 1c from 500 μm substrate (3.4% conversion) after 24 h (Fig. 4B). Of note, we observed a minor peak corresponding to 1a in the control reaction with no enzymes, suggesting that the CoA ligation step (but not the following steps in the cascade) can occur nonenzymatically at high concentrations of the substrate and cofactors (ATP, Mg²⁺ and CoA) (Fig. 4B and Fig. S9). We then sought to optimise several parameters to increase the titre of the final product. We found that buffering the reaction mix with Tris/HCl instead of HEPES/NaOH at a lower pH is beneficial for the production of 1c (Fig. 4C). Next, we optimised the starting concentration of CoA in the cascade. During the T3PKS reaction, CoA is released from 1a and 9 as a byproduct and can in principle be recycled by the CoA ligase. Therefore, we investigated the minimal CoA requirements and found that adding as little as 15.6 μm CoA does not drastically affect the final titre of 1c, and a 1 : 8 CoA‐to‐substrate ratio afforded the highest titre of to 1c (Fig. 4D). We also found that increasing the concentration of Mg²⁺ above 5 mm in the reaction mix drives the CoA ligation to completion but does not improve the titres of 1b and 1c, indicating that DnrK and AthePKS are rate‐limiting in the cascade (Fig. 4E). Consequently, we evaluated the effect of altering the molar ratio of the enzymes in the cascade. As expected, increasing the concentrations of AthePKS and DnrK relative to PqsA enhanced the product titre. A molar ratio of 1 : 4 : 6 (PqsA : AthePKS : DnrK) gave the highest titre of 1c (Fig. 4F). Following this optimisation, the concentration of 1c in the cascade increased from 17 ± 0.75 to 177 ± 2 μm, corresponding to a 35% conversion of the starting material 1.

Fig. 4.

Assembling a one‐pot biocatalytic cascade for the production of 1c. (A) The artificial pathway towards 1c from 1 featuring three enzymes—PqsA, AthePKS and DnrK. (B) Analysis of the reaction products of the cascade under initial conditions and ‘no enzyme’ control by high‐performance liquid chromatography (HPLC); detection at λ = 315 nm. Initial conditions included 100 mm HEPES/NaOH pH 8, 500 μm 1, 5 mm MgCl₂, 2 mm ATP, 250 μm Coenzyme A (CoA), 500 μm 9, 2 mm S‐adenosylmethionine (SAM), 1 mol% PqsA, 1 mol% AthePKS, 1 mol% DnrK relative to 1. The cascade was further optimised by comparing the concentration of 1c and intermediates at (C) different buffer pH (asterisk indicates the starting conditions); (D) varying molar ratios of the substrate to CoA; (E) varying concentrations of MgCl₂ in the reaction mix; (F) varying molar ratios of the enzymes in the reaction mix. (G) Cascade‐mediated production of 4‐methoxylated heterocycles 1c, 2c, 6c and 8c. The data are represented as mean ± SD; n = 3.

We noticed that the final product 1c constitutes 73 ± 5% of the total cascade intermediates after 24 h, despite reaching only 35% of its theoretical maximum concentration. To assess whether a side/reverse reaction was taking place, we incubated 1b and 1c in the reaction mix with or without the enzymes for up to 72 h (Fig. S10). The final product 1c was relatively stable in the reaction mix, with 90% of it still detectable after 72 h at 37 °C. However, the concentration of the quinolone intermediate 1b decreased by 20% after 24 h and by 40% after 72 h. The disappearance of 1b was accompanied by the accumulation of two new peaks with m/z 363 and m/z 453 (positive ion mode), which we had also detected in the cascade reactions (Fig. S11). The tandem mass spectrometry (MS/MS) fragmentation and UV–visible absorbance spectra of the side product with m/z 363 were similar to those of zanthobisquinolone—a dimer of 1b previously isolated from Zanthoxylum simulans [35]. Given its increased abundance in the ‘no enzyme’ control, its formation is likely nonenzymatic. For the other side product, tandem mass spectrometry did not allow us to speculate on its identity, yet its formation was time‐dependent and increased in the mix containing enzymes. Given the better stability of 1c compared to 1b, we believe that the formation of both side products could be suppressed by increasing the efficiency of the methylation step.

Lastly, we evaluated the performance of the optimised cascade for the production of other 4‐methoxylated heterocycles 2c, 6c and 8c with 2, 6 and 8 as the starting substrates. We observed the formation of the expected products with all substrates, although 4‐methoxythiocoumarin 8c was only present in trace amounts (Fig. 4G).

In vivo implementation of the artificial pathway

The use of expensive cofactors (ATP, CoA and SAM) was one of the cost‐ and scale‐limiting factors in the implementation of the cascade towards 1c. We hypothesised that this could be mitigated by implementing the pathway in a microbial cell factory, enabling in situ cofactor production and recycling. Furthermore, the starting substrate 1 can in principle be derived from anthranilic acid, a common primary metabolite in microbes. Therefore, we endeavoured to implement the cascade in a microbial host as a proof of concept for the production of 1c. We selected E. coli due to its ease of genetic manipulation and the fact that the antimicrobial activity of 1c reportedly does not extend to Gram‐negative bacteria [25].

To express the pathway in E. coli, we cloned PqsA and AthePKS into pETDuet‐1 and DnrK into pRSFDuet‐1. We then cotransformed the resulting plasmids into E. coli BL21(DE3), generating strain s1. After the initial fermentation with fed substrate 1 (final concentration 200 μm), we observed a small peak of the quinolone intermediate 1b, but only trace amounts of the final product 1c (Fig. 5A). We then explored different plasmid configurations by cloning PqsA and AthePKS into the high copy number plasmid pRSFDuet‐1 and DnrK into the medium copy number plasmid pCDFDuet‐1, generating strain s2. We also generated a plasmid combination inspired by the enzyme ratio that worked best in the cascade setting: PqsA in pCDFDuet‐1, and DnrK and AthePKS in pRSFDuet‐1, generating strain s3. Counterintuitively, s3 provided only a marginal increase in the titre of 1c compared to s1. However, in s2, the titre of 1c increased to quantifiable levels and averaged at 2 μm.

Fig. 5.

Implementing the biosynthetic pathway towards 1c in E. coli. No exogenous cofactors (adenosine triphosphate (ATP), coenzyme A (CoA) or S‐adenosylmethionine (SAM)) were added to the fermentation cultures. (A) Extracted ion chromatograms of the pathway intermediates across engineered E. coli strains; the peak for 1c is shaded. s1: PqsA and AthePKS in pETDuet‐1, DnrK in pRSFDuet‐1; s2: PqsA and AthePKS in pRSFDuet‐1, DnrK in pCDFDuet‐1; s3: PqsA in pCDFDuet‐1, DnrK and AthePKS in pRSFDuet‐1. (B) Production of 1c after 48 h of fermentation in E. coli BL21(DE3) (s2) and MG1655(DE3) (s4) from 200 μm substrate 1. (C) Time course of 1c accumulation in E. coli MG1655(DE3) (s4). (D) Influence of substrate concentration on the titre of 1c accumulated over 48 h in s4. In ‘100 + 100’ samples, 100 μm of substrate 1 were added to the culture once every 24 h. (E) Effect of additives (10 mm MgCl₂ and/or 2 mm l‐methionine; 20 μg·mL⁻¹ cerulenin) on the concentration of 1c and pathway intermediates 1 and 1b accumulated over 48 h in s4. The s4 data underlying panels (B–D) were collected in one large experiment, executed in duplicate on the same day. The same s4 sample incubated with 200 μm substrate 1 for 48 h is shown as a reference in these panels. Data are represented as mean ± SD; n = 2.

We then cotransformed the most promising plasmid combination into E. coli MG1655(DE3), a well‐characterised K‐12 derivative widely used in fermentation studies due to its stable genome and absence of mobile genetic elements such as F plasmids and lambda phage [36]. The obtained strain s4 afforded a product titre of 3 μm after 48 h of fermentation and was used for all following experiments (Fig. 5B). A time course fermentation experiment suggested that incubation beyond 48 h provides little increase in the titre of 1c (Fig. 5C). Lastly, we explored varying substrate concentrations (Fig. 5D) and supplementing with several additives aimed at increasing precursor supply for the T3PKS and OMT reactions (Fig. 5E). The addition of 10 mm MgCl₂ and/or 2 mm l‐methionine, the precursor of the methyl donor SAM, did not improve the titre of 1c compared to the ‘no additives’ control. The addition of SAM was not tested due to its poor uptake by E. coli cells [37]. However, the addition of cerulenin, which inhibits fatty acid synthesis and increases the intracellular pool of malonyl‐CoA [38], increased the titre of 1c to ~ 25 μm (12% conversion). This indicates that the T3PKS is the bottleneck of the in vivo pathway as it has to compete with fatty acid synthase for malonyl‐CoA [39].

Overall, we successfully engineered a biosynthetic pathway to produce 1c in E. coli for the first time. Through copy number optimisation and the addition of cerulenin, we increased the product titre from undetectable levels to ~ 25 μm (12% conversion). Improving the availability of malonyl‐CoA was essential to enhance the production of 1c, which could in the future be addressed with one of the established genetically encoded strategies to arrest fatty acid synthesis [40, 41] or by implementing the recently developed artificial noncarboxylative malonyl‐CoA pathway [42]. However, even in the presence of cerulenin, most of the substrate was not converted to 1b, indicating that the PKS cannot achieve complete turnover. To assess if this is related to its catalytic activity, we determined the apparent steady‐state kinetic parameters of AthePKS (Fig. S12, Table S1). The relatively high apparent K _m for 1a (42.62 μm) indicates that the enzyme may operate suboptimally at the substrate concentrations in vivo. Furthermore, its apparent k _cat (0.058 min⁻¹) is lower than the reported values for plant‐derived quinolone synthases from Rheum palmatum (k _cat = 1.5 min⁻¹) [17], Aegle marmelos (k _cat = 3.8 min⁻¹) [14] and Citrus microcarpa (k _cat = 15.7 min⁻¹) [15]. Although direct comparison of the apparent kinetic parameters is complicated by differences in experimental conditions across these studies, it might be wise to also explore enzymes of plant origin for cascade applications.

Structural characterisation of AthePKS

Finally, we determined the crystal structure of apo‐AthePKS at 1.9 Å and compared it to known T3PKS structures from fungi and plants (Table S3). A canonical homodimer with conserved αβαβα thiolase fold was present in the asymmetric unit, which agrees with the expectation that T3PKS form homodimers in solution (Fig. 6A). The electron density at the catalytic cysteine (C155) is consistent with its oxidation into sulfinic acid—a phenomenon observed in most T3PKS structures determined to date and thought to reflect the intrinsic redox potential and reactivity of this residue [43]. Structurally, AthePKS shares most similarity with ORAS from Neurospora crassa (PDB: 3EUQ, Z‐score 57.9, RMSD 1.5 Å) [44], followed by CsyB from Aspergillus oryzae (PDB: 3WXZ, Z‐score 54.7, RMSD 2.1 Å) [45]—the only fungal T3PKSs crystallised to date. Like in ORAS, the active site cavity of AthePKS is connected to a long tunnel lined by hydrophobic residues (Fig. 6B). The presence of this tunnel suggests that fatty acyl substrates are likely the native substrates of AthePKS, and its activity with 1a is promiscuous rather than physiological. Furthermore, structural alignment yielded hits to the amoebal Steely1 [46] and chalcone synthases from ancient vascular plants [43, 47, 48], as well as several divergent T3PKSs from flowering plants (Table S3). The observed structural similarity, despite broad taxonomic divergence, suggests that the T3PKS architecture has been remarkably conserved through evolution, while allowing functional adaptation across diverse biosynthetic contexts.

Fig. 6.

Structural characterisation of AthePKS. (A) Cartoon representation of the AthePKS dimer (PDB: 9HRB). (B) Comparison of the hydrophobic tunnel connected to the active site in ORAS (carbon atoms—grey; PDB: 3EUT) and AthePKS (carbon atoms—pink). The catalytic triad residues and eicosanoic acid from 3EUT are shown as sticks (other atoms: nitrogen—blue, oxygen—red, sulphur—yellow). (C) Superimposition of substrate 1a from the structure of AmQNS (white; PDB: 7CCT) with the structure of AthePKS. (D) Superimposition of the catalytic triad residues in AmQNS (carbon atoms—grey) and AthePKS (carbon atoms—pink). (E) Active site architecture of AthePKS compared to AmQNS. Catalytic triad and several consensus residues (L214, V271, S338, P375; residue numbering from AmQNS) are hidden for visibility. The structures were visualised in pymol version 3.1.1.

Next, we compared the active site of AthePKS to that of the quinolone synthase AmQNS from Aegle marmelos [49] (PDB: 7CCT), which is complexed with substrate 1a (Fig. 6C). We observed an almost ideal superimposition of the catalytic triad residues (C164, H303 and N336 in AmQNS, Fig. 6D). Position 265, which is occupied by a ‘gatekeeper’ phenylalanine in most T3PKSs and is critical for substrate recognition [50], is a valine in AmQNS and a threonine in AthePKS (T283, Fig. 6E). A smaller residue in this position is associated with the broadening of the active site entrance, which facilitates access for the bulky substrate 1a in quinolone synthases and related enzymes acridone synthases [14, 15, 51]. S132 and A133 in AmQNS are substituted with A116 and T117 in AthePKS. These positions were previously shown to control starter substrate specificity and ultimately the preferred cyclisation mechanism of AmQNS and acridone synthase from Ruta graveolens [14, 51]. In addition, bulky substitutions I254M and G256F in AthePKS are limiting the volume of its substrate‐binding pocket compared to AmQNS and related enzymes (Fig. S13). Manipulation of the highlighted residues might modulate the affinity and catalytic efficiency of AthePKS towards substrate 1a for improved production of 1c. However, it would also be beneficial to perform further cocrystallisation or soaking experiments with AthePKS and its potential substrates in order to obtain ligand‐bound crystals for detailed mechanistic studies.

Discussion

In summary, we have developed a one‐pot biocatalytic platform for the synthesis of three different heteroaromatic scaffolds. Using a series of anthranilic and salicylic acid derivatives, we demonstrated the production of several bioactive molecules which had not been produced enzymatically before. In the case of thiocoumarin 8b, our two‐step biocatalytic strategy offers an advantage compared to the conventional 4‐step chemical synthesis [52]. The crystal structure of AthePKS determined in this study may facilitate the engineering of the enzyme to further increase its catalytic efficiency with a particular class of substrates.

Moreover, using AthePKS, we developed an artificial enzymatic cascade towards an antimicrobial Rutaceae alkaloid 1c, achieving its biosynthesis for the first time. To bridge the gap in its cryptic biosynthetic pathway, we searched the literature for candidate O‐methyltransferases and identified an enzyme that, apart from the target intermediate, was active on all other heteroaromatic scaffolds. We also demonstrated the production of unnatural 4‐methoxylated (thio)coumarins through changing the starter substrate in the cascade.

While this work provides a framework for the cascade biosynthesis of 1c both in vitro and in a microbial host, further optimisation is necessary to render these strategies suitable for large‐scale production. The performance of the cascade is currently superior in vitro, and we see several options to improve it further. First, increasing the efficiency of O‐methylation, either through protein engineering or homologue screening, could outrun the formation of side products from the quinolone intermediate. Similarly, the PKS step might be improved by engineering AthePKS or by employing plant quinolone synthases instead, although their performance in a cascade setting would need to be evaluated first. Second, the components of the cascade could be utilised more efficiently. Currently, the first step of the cascade recycles the CoA released during malonyl‐CoA condensation by the T3PKS. This recycling aspect could be enhanced by introducing a malonate‐CoA ligase [53], which would enable in situ generation of malonyl‐CoA—the most costly component of the cascade—and thereby support scale‐up. Additionally, incorporating an SAH degradation/SAM regeneration system would improve both the efficiency of the methylation step and overall cofactor economy in the cascade [54, 55]. Future experimental efforts will focus on implementing these strategies to enable efficient enzymatic synthesis of 1c and related pharmaceutically valuable 2‐quinolones and (thio)coumarins.

Materials and methods

General

Commercial standards of compounds 1b, 2b, 3b, 6b, 7b, 8b were purchased from Sigma‐Aldrich (St. Louis, MO, USA), TCI Europe NV (Haven, Belgium) or BLD Pharm (Shanghai, China). Nuclear Magnetic Resonance (NMR) spectra were recorded on a Jeol 400‐YH spectrometer (400 MHz for ¹H nucleus, 101 MHz for ¹³C nucleus). Deuterated solvents (DMSO‐d ₆ ) were purchased from Sigma‐Aldrich. The chemical shifts are given in parts per million (ppm, δ) and reported relative to the solvent peak at 2.50 ppm (¹H) or 39.52 ppm (¹³C). All spectra were measured at room temperature. ¹H NMR data are reported as: chemical shift (ppm), multiplicity (s = singlet, d = doublet, t = triplet, q = quartet, m = multiplet, dd = doublet of doublets, dt = doublet of triplets, dq = doublet of quartets, br = broad), coupling constants (J) given in Hertz (Hz) and the integration. ¹³C NMR spectra were conducted with proton decoupling and the chemical shifts are reported. NMR data were processed on MestReNova.

Construction of plasmids

The list of all plasmids used in this study is provided in Table S4. Plasmids for the expression of PqsA, AthePKS, FerePKS, GloKOMT and MesMOMT were constructed in our previous studies [32, 56]. The genes coding for DnrK, CmOMT2 and CpdC were codon‐optimised for expression in E. coli and ordered as linear DNA fragments from Twist Bioscience (San Francisco, CA, USA). The genes coding for DnrK, CmOMT2 and CpdC with N‐terminal hexahistidine tag were cloned by restriction and ligation (NcoI/HindIII) into pRSFDuet‐1 for expression under the T7 promoter. For pathway applications, different combinations of the pathway genes were cloned between pETDuet‐1, pRSFDuet‐1 and pCDFDuet‐1 using restriction and ligation or Gibson assembly. All constructs were verified by Sanger sequencing (Macrogen Europe, Amsterdam, The Netherlands). Amino acid sequences of all proteins expressed in this study can be found in the Supporting Information.

Protein expression, purification and crystallisation

Expression

The expression and purification procedures were the same for PqsA, AthePKS, FerePKS, DnrK, CmOMT2 and MesMOMT. Plasmids harbouring the target genes were transformed into chemically competent E. coli BL21(DE3) and maintained on selective LB agar containing the appropriate antibiotic. A starter culture was inoculated from a single colony (5 mL, LB with appropriate antibiotic) and incubated at 37 °C (180 r.p.m., overnight). The main culture was inoculated from the starter culture (1 : 100) into 400 mL of TB autoinduction medium with appropriate antibiotic (20 g·L⁻¹ tryptone, 24 g·L⁻¹ yeast extract, 4 mL·L⁻¹ glycerol, 17 mm KH₂PO₄, 72 mm K₂HPO₄, 0.2% w/v lactose and 0.05% w/v glucose) and incubated at 37 °C (180 r.p.m., 2 h), after which the temperature was lowered to 25 °C (180 r.p.m., overnight).

Purification

The protocol for protein purification was adapted from [32, 56]. The cells were harvested by centrifugation (15 min, 3428  g ), and the pellet was resuspended in 5 volumes of lysis buffer (buffer A including one EDTA‐free protease inhibitor tablet (Roche, Basel, Switzerland); buffer A: 50 mm Tris/HCl pH 7.5, 500 mm NaCl, 20 mm imidazole). The cell suspension was lysed by sonication (50% duty cycle, 7 cycles of 35 s ON/60 s OFF) and cleared by centrifugation for 60 min at 17 000  g . The supernatant was loaded onto the Ni‐NTA affinity matrix (Qiagen, Hilden, Germany) equilibrated with buffer A by gravity flow. The column was washed with 20 column volumes of wash buffer (50 mm Tris/HCl pH 7.5, 500 mm NaCl, 30 mm imidazole) and eluted stepwise with one column volume of buffers B1 to B6 (buffers B1–B6: 50 mm Tris/HCl pH 7.4, 500 mm NaCl, 50 mm imidazole/100 mm imidazole/150 mm imidazole/200 mm imidazole/250 mm imidazole or 500 mm imidazole, respectively). The eluates of each step were collected in separate fractions and analysed by SDS/PAGE. Fractions containing target proteins with low background were pooled and transferred into the storage buffer (10 mm HEPES/NaOH pH 7.5, 50 mm NaCl, 2 mm dithiothreitol, 5% (v/v) glycerol) by 3 cycles of concentration (Amicon® Ultra Centrifugal Filter; 3 kDa molecular weight cutoff) and dilution (1 : 30). For protein crystallisation experiments, AthePKS was additionally subjected to size exclusion chromatography on a Superdex 75 pg column (Cytiva, Wilmington, DE, USA) and eluted into the same storage buffer. The protein concentrations were determined by absorbance at 280 nm (NanoDrop, ThermoFisher Scientific, Waltham, MA, USA) before the purified enzymes were aliquoted and flash‐frozen with liquid nitrogen for storage at −70 °C. SDS/PAGE analysis of the purified proteins is shown in Fig. S14.

Crystallisation

The crystallisation of AthePKS was performed using the sitting‐drop vapour diffusion method at a temperature of 18 °C. A 1 : 1 μL mixture of reservoir solution and freshly prepared protein was employed. Initial crystallisation screening utilised commercial kits, specifically JCSG⁺ and Morpheus from Molecular Dimensions Co. Subsequently, optimisations were performed to obtain higher quality crystals. Tetragonal AthePKS crystals were observed at a protein concentration of 10 mg·mL⁻¹ in a crystallisation buffer composed of 0.1 m Tris (pH 8.5), 0.03 m MgCl₂, 0.03 m CaCl₂, 10% (w/v) PEG8000 and 20% (w/v) PEG200. Crystals were harvested using nylon loops 3–4 days prior to diffraction. The crystals were briefly immersed in cryoprotectant solutions, prepared by supplementing the reservoir solutions with 20–30% (v/v) glycerol, before being flash‐cooled in liquid nitrogen.

Diffraction data collection, structure determination and refinement

Diffraction data were collected at beamline P11 of the PETRA III facility (DESY, Hamburg, Germany). The collected data were auto‐processed using xdsapp [57], scaling and merging statistics were subsequently handled with aimless in the ccp4 suite [58]. The structure of AthePKS was determined by molecular replacement. The monomer of type III polyketide synthase from Neurospora crassa (PDB ID: 3E1H) [59] was utilised as the search model, employing the program molrep [60]. Further refinement of the structure was carried out using refmac5 [61] and manual adjustments made in coot [62]. The data collection and refinement statistics are presented in Table S2. The structure of AthePKS was deposited in the PDB under accession code 9HRB. Figures were created in pymol version 3.1.1. Structural comparison was performed with dali [63] and sequence alignments were performed with clustal omega [64].

Chemical synthesis of reference compounds

The general procedure for the synthesis of methoxylated heterocycles 1c, 2c, 6c and 8c was adapted from [65] (Scheme 1). Briefly, 1.5 mL (3 mmol) of trimethylsilyldiazomethane (2 m in hexane, CAS 18107‐18‐1, Sigma‐Aldrich) was added to a solution of the starting material (1b, 2b, 6b or 8b; 1 mmol) in diethyl ether (3.5 mL) at 0 °C under N₂. The reaction mixture was warmed to RT, stirred for 30 min (controlled by TLC) and then concentrated and purified by MPLC (4 g, 0–100% EtOAc/Pentane). ¹H and 13C NMR spectra of all chemically synthesised compounds can be found in Figs S15–S18.

Scheme 1.

Chemical synthesis of methoxylated heterocycles.

4‐methoxy‐2H‐chromen‐2‐one was synthesised by following general procedure to obtain 6с (117 mg, 66%) as an off‐white solid. ¹ H NMR (400 MHz, DMSO‐d ₆) δ 7.77 (dd, J = 8.0, 1.6 Hz, 1H), 7.64 (ddd, J = 8.7, 7.2, 1.6 Hz, 1H), 7.41–7.32 (m, 2H), 5.89 (s, 1H), 4.00 (s, 3H) ppm. ¹³ C NMR (101 MHz, DMSO‐d ₆ ) δ 165.8, 161.6, 152.7, 132.7, 124.2, 122.8, 116.4, 115.1, 90.2, 57.0 ppm.

4‐methoxy‐1‐methylquinolin‐2(1H)‐one was synthesised by following the general procedure to obtain 1с (136 mg, 72%) as an off‐white solid. ¹ H NMR (400 MHz, DMSO‐d ₆ ) δ 7.80 (dd, J = 8.0, 1.6 Hz, 1H), 7.58 (ddd, J = 8.6, 7.1, 1.6 Hz, 1H), 7.41 (dd, J = 8.6, 1.0 Hz, 1H), 7.20 (ddd, J = 8.0, 7.1, 1.0 Hz, 1H), 5.98 (s, 1H), 3.89 (s, 3H), 3.50 (s, 3H) ppm. ¹³ C NMR (101 MHz, DMSO‐d ₆) δ 162.3, 161.8, 139.4, 131.3, 122.5, 121.4, 115.5, 114.5, 96.2, 56.1, 28.5 ppm.

4‐methoxyquinolin‐2(1H)‐one was synthesised by following general procedure to obtain 2с (129 mg, 74%) as an off‐white solid. ¹ H NMR (400 MHz, DMSO‐d ₆ ) δ 11.35 (s, 1H), 7.76 (dd, J = 8.0, 1.5 Hz, 1H), 7.50 (ddd, J = 8.6, 7.2, 1.5 Hz, 1H), 7.28 (dd, J = 8.3, 1.1 Hz, 1H), 7.15 (ddd, J = 8.3, 7.2, 1.1 Hz, 1H), 5.88 (s, 1H), 3.92 (s, 3H) ppm. ¹³ C NMR (101 MHz, DMSO‐d ₆) δ 163.3, 163.2, 138.6, 131.0, 122.2, 121.4, 115.2, 114.5, 96.7, 56.1 ppm.

4‐methoxy‐2H‐thiochromen‐2‐one was synthesised by following general procedure to obtain 8с (70 mg, 36%) as an off‐white solid. ¹ H NMR (400 MHz, DMSO‐d ₆ ) δ 8.08 (dd, J = 8.1, 1.1 Hz, 1H), 7.64–7.57 (m, 2H), 7.51–7.44 (m, 1H), 6.19 (s, 1H), 3.99 (s, 3H) ppm. ¹³ C NMR (101 MHz, DMSO‐d ₆ ) δ 182.9, 166.7, 135.4, 131.1, 126.7, 125.8, 125.6, 123.0, 101.6, 57.0 ppm.

In vitro assays

PqsA substrate scope analysis

To probe the substrate scope of PqsA, reactions were set up in triplicate with carboxylic acids 1–8 and a ‘no substrate’ control and contained 2 mm ATP, 2 mm MgCl₂, 0.5 mm substrate, 0.2 mm CoA and 5 μm PqsA in 100 mm Tris/HCl pH 7.5; reaction volume: 50 μL. Additional negative control reactions were set up under identical conditions lacking ATP. The reactions were incubated at 37 °C for 2 h without shaking and analysed by quantifying the residual –SH groups of coenzyme A in the reaction mixture using 5,5′‐dithiobis‐(2‐nitrobenzoic acid) (DTNB, Ellman's reagent) as described previously [66].

T3PKS substrate scope analysis

The substrate scope of AthePKS and FerePKS was assessed in a one‐pot continuous cascade with PqsA. The reaction mix consisted of 0.5 mm substrate 1–8, 20 mm MgCl₂, 2 mm ATP, 0.25 mm CoA, 0.5 mm malonyl‐CoA (9) or methylmalonyl‐CoA (10) in 100 mm Tris/HCl pH 7.5; reaction volume: 30 μL. 5 μm PqsA and 5 μm Athe/FerePKS were added simultaneously. For each substrate, negative control reactions were set up under identical conditions lacking PqsA and the T3PKS. The reactions were incubated at 37 °C for 24 h without shaking, after which an equal volume of acidic methanol (MeOH +0.1% formic acid) was added to precipitate the proteins. The reactions were centrifuged at 17 000  g for 10 min and analysed by LC–MS and/or HPLC for quantification.

Screening of the OMTs using cell‐free extracts

To screen for the desired quinolone 4‐O‐methyltransferase activity, the candidate genes were expressed from plasmids c173, c192 and c379‐c381 in E. coli as described in ‘Protein expression, purification and crystallisation’ (culture volume: 10 mL). About 1 mL of each culture was centrifuged at 4500  g for 10 min, and the pellets were incubated at −70 °C for 45 min. Afterwards, the cells were thawed and resuspended in 200 μL of lysis buffer 25 mm Tris/HCl pH 7.5, 1 mg·mL⁻¹ lysozyme and 1 : 10 000 benzonase (Novagen, Merck, St. Louis, MO, USA) and incubated at 30 °C without shaking for 1 h. Following the incubation, 600 μL of 25 mm Tris/HCl pH 7.5 were added, and the insoluble cell fraction was removed by centrifugation at 17 000  g for 15 min. About 20 μL of each cell‐free lysate was then used for OMT assays together with 200 μm substrate 1b, 5 mm MgCl₂ and 1 mm SAM in 25 mm Tris/HCl pH 7.5; final volume: 100 μL. The reactions were incubated at 30 °C for 24 h, quenched with an equal volume of acidic methanol to precipitate the proteins, centrifuged at 17 000  g for 10 min and analysed by LC–MS.

In vitro OMT assays

Substrate scope of the purified DnrK, CmOMT2 and MesMOMT was assessed in triplicate by incubating 0.5 mm substrate 1b, 2b, 6b or 8b, 2 mm SAM and 5 μm enzyme in 100 mm HEPES‐NaOH pH 8; reaction volume: 30 μL. To assess the influence of Mg²⁺ and other metal ions on the activity of the OMTs, an identical set of reactions was prepared with the addition of 5 mm respective metal salt, EDTA or ultrapure water for the ‘no metal’ control. For each substrate, negative control reactions were set up under identical conditions lacking the OMT. The reactions were incubated at 37 °C for 24 h, quenched with an equal volume of acidic methanol to precipitate the proteins, centrifuged at 17 000  g for 10 min and analysed by LC–MS.

Steady‐state kinetics

The protocol for determining the apparent steady‐state kinetic parameters was adapted from [56]. The analysis was performed using 500 nm AthePKS and 6.25 to 200 μm substrate 1a in a reaction mix consisting of 50 mm Tris/HCl pH 7.5 and 400 μm malonyl‐CoA at 37 °C. For the estimation of the initial velocity using linear regression, the reactions were quenched after 5, 10 and 15 min with an equal volume of acetonitrile with 0.1% formic acid, centrifuged and stored at 4 °C before HPLC analysis (see Analysis and quantification of target compounds). The peak areas were integrated and converted to concentrations in μm based on the calibration curve with authentic standard of 1b. The apparent initial velocities of all experiments performed in triplicate were plotted against substrate concentrations using graphpad prism 8, and the apparent K _m and k _cat constants were determined by nonlinear regression with the Michaelis–Menten equation. A full report of the regression statistics is given in Table S1.

One‐pot three‐step biocatalytic cascade

The initial conditions for the three‐step cascade included 0.5 mm substrate 1, 5 mm MgCl₂, 2 mm ATP, 0.25 mm CoA, 0.5 mm malonyl‐CoA and 2 mm SAM in 100 mm HEPES‐NaOH pH 8; reaction volume: 30 μL. PqsA, AthePKS and DnrK (1 mol% or 5 μm each) were added simultaneously, and the reactions were incubated at 37 °C for 24 h. For pH and buffer optimisation, HEPES‐NaOH pH 8 was substituted for HEPES‐NaOH (pH 7 and pH 7.5) or Tris/HCl (pH 7, pH 7.5, pH 8). To assess the effect of varying Mg²⁺ concentration, MgCl₂ was added at concentrations of 5, 10, 20 and 40 mm. To assess the effect of lowering the starting concentration of CoA, reactions were set up with CoA to substrate ratios of 1 : 1, 1 : 2, 1 : 4, 1 : 8, 1 : 16 and 1 : 32. At each stage of the optimisation, negative control reactions were set up under identical conditions lacking the cascade enzymes. For time course experiments, the reactions were incubated for up to 72 h and samples were taken every 24 h. The optimised conditions included 0.5 mm substrate 1, 5 mm MgCl₂, 2 mm ATP, 62.5 μm CoA, 0.5 mm malonyl‐CoA and 2 mm SAM in 100 mm Tris/HCl pH 7.5. Cascade reactions with substrates 2, 6 and 8 were set up under the optimised conditions and incubated for 24 h. All reactions were quenched with an equal volume of acidic methanol to precipitate the proteins, centrifuged at 17 000  g for 10 min and analysed by LC–MS.

Fermentation

Strains harbouring the quinolone pathway genes were generated by transforming the appropriate combinations of plasmids into chemically competent [67] E. coli BL21(DE3) or electrocompetent [68] E. coli MG1655(DE3). The list of all strains used in the fermentation experiments is provided in Table S5. Starter cultures were prepared from two individual colonies of the final strains in 5 mL LB supplemented with appropriate antibiotics. The cultures were incubated overnight at 37 °C with agitation, and 100 μL thereof were used to inoculate the main cultures (10 mL TB autoinduction with antibiotics; sterile glass tubes). The substrate was added after 3 h of growth at 37 °C, 200 r.p.m. (OD₆₀₀~0.6). These cultures were incubated at 30 °C, 200 r.p.m. for 48 h. Samples of 667 μL were taken after 48 h and extracted with the double volume (1334 μL) of acidic methanol (MeOH + 0.1% formic acid). The extracts were dried, resuspended in 25 μL of acidic methanol and analysed by LC–MS. For the time course experiment, the cultures were incubated for 120 h, and samples of 667 μL were taken every 24 h. When required, 10 mm MgCl₂, 2 mm l‐methionine or 20 μg·mL⁻¹ cerulenin were added to the cultures at the time of substrate addition.

Analysis and quantification of target compounds

The detection of in vitro reaction and fermentation products was performed as described in [56] using a low‐resolution Waters Acquity Arc UHPLC–MS system equipped with a 2998 PDA detector and a QDa single‐quadrupole mass detector. The samples were separated over a Waters XBridge BEH C18 3.5 μm 2.1 × 50 mm column at 40 °C with a concentration gradient (solvent A: water + 0.1% formic acid and solvent B: acetonitrile + 0.1% formic acid) at a flow rate of 0.5 mL·min⁻¹ (2 μL injections). The following gradient was used: 5% B for 2 min, 5–90% B over 3 min; 90% B for 2 min; 5% B for 3 min. For E. coli fermentation extracts, a longer gradient was employed: 5% B for 2 min, 2–50% B over 8 min, 50–90% B over 5 min, 90% B for 2 min, 5% B for 3 min. MS analysis was carried out in both positive and negative ion modes with the following parameters: probe temperature of 600 °C; capillary voltage of 1.0 kV; cone voltage of 15 V; scan range 100–1250 m/z. The acquired data were analysed using the proprietary software masslynx.

The identity of the reaction products was further confirmed by high‐resolution and MS/MS analyses using a Shimadzu LC20‐XR system (Shimadzu Benelux, Den Bosch, The Netherlands) coupled to a Q Exactive Plus mass spectrometer (Thermo Fisher Scientific, USA). The samples were separated over a Waters XBridge BEH C18 reversed‐phase column at 50 °C with a concentration gradient (solvent A: water + 0.1% formic acid and solvent B: acetonitrile + 0.1% formic acid) at a flow rate of 0.5 mL·min⁻¹ (2 μL injections). The following gradient was used: 5% B for 2 min, 5–90% B over 3 min; 90% B for 2 min; 5% B for 3 min. MS and MS/MS analyses were performed with electrospray ionisation (ESI) in positive or negative ion mode at a spray voltage of 3.5, and sheath and auxiliary gas flow set at 48 and 11, respectively. The ion transfer tube temperature was 255 °C. Spectra were acquired in data‐dependent mode with a survey scan at m/z 80–1200 at a resolution of 70 000 followed by MS/MS fragmentation of the top 5 precursor ions at a resolution of 17 500. A stepped collision energy of 30–40–55 was used for fragmentation, and fragmented precursor ions were dynamically excluded for 10 s. The acquired data were analysed using mzmine 4.0.3 [69]. Spectral data of all enzymatic products and reference compounds can be found in Figs S19–S30.

Quantification of the reaction products was performed using reversed‐phase HPLC (instrument: Shimadzu LC‐20AD; autosampler: SIL‐20AC, T = 15 °C, 10 μL injection; flow rate: 1.2 mL·min⁻¹; column: NUCLEOSHELL RP18 90 Å, 100 × 4.6 mm, 2.7 μm, T = 40 °C; detector: SPD‐20A, λ = 355 nm (N‐methylanthranilic acid and N‐methylanthraniloyl‐CoA) and λ = 315 nm (4‐hydroxy‐1‐methyl‐1‐quinolone, 4‐methoxy‐1‐methyl‐2‐quinolone and all other T3PKS products); solvent A: water with 0.1% FA, solvent B: ACN with 0.1% FA; gradient: 10% B for 2 min; 10–95% B over 3 min; 95% B for 2 min; 95–10% B over 3 min). The peak areas were integrated and converted to concentrations in μm based on calibration curves with the authentic standards (Figs S31–S33).

Conflict of interest

The authors declare that there are no conflicts of interest.

Author contributions

NS and KH conceived the study. NS performed all biochemical experiments and analysed the data. AO performed the chemical synthesis and compound characterisation under supervision of SS. LZ and MRG determined and refined the crystal structure. NS wrote the manuscript with contributions from LZ, AO, MRG, SS and guidance from KH. All authors read and approved the final version of the manuscript.

Supporting information

Fig. S1. CoA ligation reactions with PqsA and substrates 1–8.

Fig. S2. Chemical structures of substrates 1–8 and enzymatically generated products thereof.

Fig. S3. Extracted ion chromatograms in low‐resolution LC–MS of reaction products of Athe/FerePKS with substrates 1–8 compared to the ‘no enzyme’ control and reference compounds.

Fig. S4. HPLC analysis of reaction products of Athe/FerePKS with substrates 1–8 compared to the ‘no enzyme’ control and reference compounds.

Fig. S5. Native reactions catalysed by (a) DnrK, (b) CmOMT2 and (c) CpdC.

Fig. S6. Scheme of caffeic acid methylation by MesMOMT.

Fig. S7. Extracted ion chromatograms in low‐resolution LC–MS of reaction products of DnrK, CmOMT2 and MesMOMT with substrates 1b, 2b, 6b and 8b compared to the ‘no enzyme’ control and chemically synthesised reference compounds.

Fig. S8. The effect of adding 5 mm metal salts or ethylenediaminetetraacetic acid (EDTA) to the assay of DnrK with substrate 1b.

Fig. S9. Non‐enzymatic formation of 1a in the cascade.

Fig. S10. Incubation of final product 1c (a) and quinolone intermediate 1b (b) in the reaction mix with or without the enzymes for up to 72 h.

Fig. S11. Formation of side products from 1b and 1c.

Fig. S12. Michaelis–Menten kinetics of AthePKS with 1a as substrate.

Fig. S13. Multiple sequence alignment of AthePKS with known quinolone and acridone synthases.

Fig. S14. SDS/PAGE of the purified proteins used for in vitro reactions in this study.

Fig. S15. ¹H and ¹³C NMR spectra of the chemically synthesised compound 6c (NSO‐01).

Fig. S16. ¹H and ¹³C NMR spectra of the chemically synthesised compound 1c (NSO‐02).

Fig. S17. ¹H and ¹³C NMR spectra of the chemically synthesised compound 2c (NSO‐03).

Fig. S18. ¹H and ¹³C NMR spectra of the chemically synthesised compound 8c (NSO‐04).

Fig. S19. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive ion mode) for (a, b) reference compound 1b and (c, d) enzymatically generated 1b.

Fig. S20. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) reference compound 2b and (c, d) enzymatically generated 2b.

Fig. S21. Comparison of the UV absorption and extracted ESI‐HR‐MS spectra (positive mode) for (a, b) reference compound 3b and (c, d) enzymatically generated 3b.

Fig. S22. UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) enzymatically generated 4b.

Fig. S23. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) reference compound 6b and (c, d) enzymatically generated 6b.

Fig. S24. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) reference compound 7b and (c, d) enzymatically generated 7b.

Fig. S25. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) reference compound 8b and (c, d) enzymatically generated 8b.

Fig. S26. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) reference compound 11 and (c, d) enzymatically generated 11.

Fig. S27. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) chemically synthesised 1c (NSO‐02) and (c, d) enzymatically generated 1c.

Fig. S28. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) chemically synthesised 2c (NSO‐03) and (c, d) enzymatically generated 2c.

Fig. S29. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) chemically synthesised 6c (NSO‐01) and (c, d) enzymatically generated 6c.

Fig. S30. Comparison of the UV absorption and ESI‐HR‐MS/MS spectra (positive mode) for (a, b) chemically synthesised 8c (NSO‐04) and (c, d) enzymatically generated 8c.

Fig. S31. Calibration plots used during Athe/FerePKS substrate scope analysis.

Fig. S32. Calibration plots used during OMT substrate scope analysis.

Fig. S33. Calibration plots used during cascade optimization experiments.

Table S1. Diffraction data collection, structure determination, and refinement statistics.

Table S2. Structural alignment of AthePKS with known T3PKSs from PDB with the Dali server.

Table S3. List of strains and plasmids used in this study.

Table S4. Michaelis–Menten kinetic parameters of AthePKS with 1a as substrate.

Table S5. List of E. coli strains used in fermentation experiments in this study.

Data availability statement

The crystal structure of AthePKS was deposited in the PDB under accession number 9HRB. Raw data not included in the main text or Supporting Information are available from the corresponding author upon reasonable request.