Genome Mining of Isoindolinone-Containing Peptide Natural Proucts

Abstract

Peptide natural products (PNPs) are an important source of bioactive compounds. Recent studies have shown oligopeptides or pseudopeptides can be synthesized by amide bond-forming enzymes such as ATP-grasp enzymes and amide bond synthetases (ABSs). By focusing on ATP-grasp enzymes as part of a conserved biosynthetic enzyme ensemble, genome mining of PNPs was performed on three biosynthetic gene clusters (BGCs) from diverse fungi, including Coccidioides immitis RS, the causative agent of valley fever. We demonstrate the conserved enzymes synthesize a common dipeptide fragment, l-leucine-l-O-isoindolinone-homoserine (l-Leu-l-Isd), which is modified and diversified into three PNPs (1-3) by associated enzymes in the three pathways. Pathway reconstitution and enzymatic assays led to the characterization of six ATP-grasp enzymes and ABSs that catalyze di-, tri- and tetrapeptide formation. From the C. immitis BGC, a flavoenzyme catalyzing the direct oxidation of l-tryptophan to l-oxindolylalanine was discovered. Our work validates ATP-grasp enzymes and ABSs as leads to mine new PNPs, and further showcases the biocatalytic potential of these enzymes in catalyzing amide bond formations.

Graphical Abstract

INTRODUCTION

Peptide natural products (PNPs) are compounds that contain multiple amide bonds connecting amino acid or carboxylic acid building blocks (Figure 1). Many PNPs display potent biological activities as human therapeutics or pesticides.¹ The largest families of PNPs are classified as nonribosomal peptides (NRPs) and ribosomally synthesized and posttranslationally modified peptides (RiPPs).² These compounds are synthesized via nonribosomal peptide synthetases (NRPSs) and RiPPs-associated enzymes, respectively. Using these well-characterized biosynthetic machinery as guides, discovery of NRPs and RiPPs from microbial and plant genomes has afforded new natural products.^3–5 In addition to these two canonical families of PNPs, an additional group of PNPs that is not derived from either of these pathways has garnered interests in recent years. These PNPs are short peptides or pseudopeptides that have useful bioactivities, including antibiotic dapdiamide A,⁶ proteasome inhibitor cystargolide A,⁷ cysteine protease inhibitor E-64,⁸ and histone acetylase inhibitor NK13650 A (Figure 1B).⁹

Figure 1.

Peptide natural products (PNPs). (A) Mechanism of peptide bond formation by ATP-grasp and ABS enzymes; (B) PNPs that are derived from the functions of ATP-grasp and/or ABS enzymes; (C) isoindolinone PNPs discovered in this work.

The formation of peptide or pseudopeptide bonds in these PNPs follow a different biosynthetic logic compared to that of NRPs and RiPPs. Two types of ATP-dependent, amide bond-forming enzymes are utilized in the biosynthetic pathways of compounds shown in Figure 1B: ATP-grasp enzymes and amide bond synthetases (ABSs, Figure 1A).^10,11 Both of these types are standalone enzymes that are significantly smaller than either the NRPS or the ribosome complex. While both types of enzymes catalyze condensation reactions between a carboxylate (electrophile) and an amine (nucleophile), the mechanisms are distinct. ATP-grasp enzymes first activate the carboxylate through ATP hydrolysis to generate an acyl-phosphate, which can be attacked by the amine to form the amide bond; while ABS enzymes catalyze adenylation of carboxylate, followed by direct nucleophilic attack by the amine.¹⁰ ATP-grasp enzymes are more frequently found in primary metabolism, exemplified by glutathione synthetase¹² and d-Ala-d-Ala ligase.¹³ ABS enzymes belong to the ANL-superfamily enzymes that adenylate substrates,¹⁴ but are not typically found in primary metabolism. Most ANL-superfamily enzymes involved in amide bond formation are CoA-ligases which generate an acyl-CoA intermediate, followed by amidation catalyzed by a second enzyme which can be an N-acyltransferase. In recent years, ATP-grasp and ABS enzymes have been discovered in both bacterial and fungal biosynthetic pathways.^15–20 The versatility and substrate promiscuity of these standalone enzymes, together with the high importance of green synthesis of amide bonds, have resulted in significant interests in ATP-grasp enzymes and ABSs as biocatalysts, highlighted by bacterial enzymes TabS, McbA and CfaL.^17,20–27

With respect to the roles of ATP-grasp enzymes and ABSs in biosynthesis of PNPs, Walsh and coworkers were the first to demonstrate that a pair of ATP-grasp enzyme and ABS catalyzes formation of the two amide bonds in dapdiamides.¹⁵ Our lab discovered the putative fungal ATP-grasp enzyme, AnkG, that ligates l-aspartate to a pseudopeptide in the last step of the NK13650 biosynthetic pathway.¹⁸ Micklefield and coworkers identified a pair of ATP-grasp enzyme and ABS are involved in biosynthesis of cystargolide and belactosin in bacteria.¹⁷ The two enzymes ligate two amino acids in tandem to a β-lactone warhead. We recently reported a parallel biosynthetic logic in the biosynthesis of E-64 in fungi,²⁰ during which an ATP-grasp enzyme and an ABS work in tandem to condense three different building blocks (diacid, amino acid and amine) into the pseudotripeptide product. In both the cystargolide and E-64 pathways, the robust activities and substrate promiscuities of these amide bond forming enzymes led to the biocatalytic preparation of numerous PNP analogs. In the E-64 example, a high throughput colorimetric assay was coupled to enzymatic synthesis to identify new analogs that are potent cathepsin inhibitors.²⁰

These examples led us to hypothesize that both ATP-grasp enzymes and ABSs are widely spread in fungal PNP biosynthetic pathways, and therefore represent excellent leads in the search for new PNP scaffolds. Scanning fungal genomes using both sequence- and structure-based prediction tools, using these recently identified examples as queries, indeed unveiled many examples of these enzymes in putative biosynthetic pathways. In this work, we performed genome mining for new PNPs using fungal ATP-grasp enzymes and discovered a set of potential PNP biosynthetic gene clusters (BGCs). Reconstitution of three representative BGCs afforded three different isoindolinone-containing PNPs 1-3 (Figure 1C). Biosynthetic characterization of 1-3 not only showed these pathways generate a common biosynthetic fragment followed by structural diversification, but also revealed new enzymatic routes to nonproteinogenic amino acids with isoindolinone and oxindole side chains.

RESULTS AND DISCUSSION

Discovery of potential PNP BGCs using ATP-grasp enzymes.

To search for candidate fungal BGCs that could produce PNPs, we performed a sequence-based search for new ATP-grasp enzymes that are conserved among multiple biosynthetic pathways. Recently studied fungal ATP-grasp enzymes, including FsqD,¹⁶ AnkG¹⁸ and Cp1B²⁰ from the fumisoquin, NK13650 and E-64 pathways, respectively, were used as queries. Because of the low sequence similarity between the three enzymes, each search yielded a different set of hits (Tables S6–S8). From the FsqD search, 500 homologues retrieved from UniProtKB database with lengths between 400 and 800 residues were analyzed by the Enzyme Function Initiative-Enzyme Similarity Tool (EFI-EST) to generate a sequence similarity network (SSN, Figure 2).^28,29 Members from each major cluster were analyzed for colocalization with conserved biosynthetic enzymes, which could indicate similar biosynthetic functions. One large cluster (Cluster II in Figure 2) contains 14 members that are present in BGCs containing multiple homologous genes (Figure S1). Six additional homologous BGCs, of which the corresponding ATP-grasp enzymes are not represented in the SSN, were found by further genome search using tblastn (Figure S1). Three representative BGCs are shown in Figure 3, and include the ave BGC from Aspergillus avenaceus NRRL 4517, the unc BGC from Uncinocarpus reesii UAMH 1704, and the coc BGC from the human pathogen Coccidioides immitis RS that is the causative agent for valley fever³⁰ (Table S1).

Figure 2.

Sequence similarity network (SSN) analysis of FsqD homologues from the UniProt database. The retrieved proteins with 400 to 800 residues were analyzed by the EFI-EST with an alignment score threshold of 180.

Figure 3.

The three BGCs studied in this work. The six conserved genes are shown in shaded arrows. Abbreviation: PLP-Dep, pyridoxal-5′ phosphate-dependent enzyme; P450, cytochrome P450; FMO, flavin-dependent monooxygenase; FMN, flavin mononucleotide; ABS, amide bond synthetase; MFS, major facilitator superfamily; HP, hypothetical protein.

Each of the BGCs contains six genes that are conserved as shown in Figure 3. The putative functions of protein products from the conserved genes include a nonreducing polyketide synthase (NRPKS, AveA, UncA and CocB), an ATP-grasp enzyme that was identified by SSN (AveB, UncK and CocI), a single-module NRPS (AveC, UncD, CocH) with the domain arrangement of inactive condensation (C⁰)-adenylation (A)-thiolation (T), a major facilitator superfamily transporter (AveD, UncC and CocD), an alanine racemase (AveF, UncE and CocA), and a pyridoxal-5′-phosphate (PLP)-dependent enzyme (AveG, UncF and CocG). The PLP-dependent enzymes have sequence homologies to a set of recently discovered, cystathionine γ-synthase like enzymes that are involved in nonproteinogenic amino acid biosynthesis.^18,31,32 These PLP-dependent enzymes, including AnkD¹⁸ and CndF,³² catalyze the γ-substitution of O-acetyl-l-homoserine (OAH) via an electrophilic vinylglycine-PLP ketimine intermediate. We hypothesized that these five conserved enzymes (excluding the MFS transporter) biosynthesize a common biosynthetic fragment, that is elaborated and diversified by other enzymes associated with each BGC (Figure 3). For example, the ave BGC encodes a flavin-dependent halogenase (AveE), while the coc BGC encodes redox enzymes such as P450 monooxygenase (CocC), flavin-dependent monooxygenase (CocE) and FMN oxidoreductase (CocF). Notably, unc and coc BGCs both encode an enzyme that is annotated as a putative CoA-ligase, with UncG and CocJ sharing 52% sequence identity. CoA-ligases and ABSs both belong to the ANL-superfamily that performs adenylation reactions, and are generally difficult to distinguish based on sequence analysis alone.¹⁴ In addition, the unc BGC encodes an additional ATP-grasp enzyme (UncJ), a peptidase (UncH) and an epimerase (UncI), which further indicate the likelihood that the biosynthetic product could be a modified oligopeptide. No related natural products have been reported from any of the three organisms.

Reconstitution of ave cluster generated an isoindolinone dipeptide.

To investigate metabolites produced by these BGCs, a combination of heterologous expression and direct enzymatic assays were used. First, the entire ave BGC (aveABCEFG) was expressed in A. nidulans A1145 ΔSTΔEM followed by metabolite analysis.³³ Two new metabolites 4 (80 mg/L) and 5 (3 mg/L) were detected (Figure 4C, i) and isolated. 4 was characterized as R-dihydroxy-methylisoindolinone by NMR (Table S12, Figures S43 and S44). The 8R stereochemistry in 4 was established by comparing the specific optical rotation (OR) value to that from a previous isolation and characterization.³⁴ 5 was characterized to be the C-6 chlorinated product of 4 (Figure 4A) (Table S13, Figures S45–S48). Expression of different combinations of ave genes in A. nidulans (Figure S4) indicated the combined activities of NRPKS (AveA), NRPS (AveC) and racemase (AveF) are responsible for the formation of 4 (Figure 4C, ii), while the predicted flavin-dependent halogenase (AveE) catalyzes C-6 chlorination to give 5 (Figure 4C, iii). The source of the nitrogen atom in 4 was probed with feeding of 2,3-¹³C₂-l-alanine to the producing host, which led to +2 mu in the molecular weight (MW) of 4 (Figure S5). This indicated l-alanine is a building block and epimerization of C_α to d-alanine is required to result in the 8R stereochemistry in 4. AveF is predicted to be a PLP-dependent alanine racemase and shows 70% sequence identity to alanine racemase TOXG.³⁵ In vitro assay using recombinant AveF followed by derivatization with Marfey’s reagent³⁶ confirmed AveF catalyzes the interconversion between l-alanine and d-alanine (Figures S6 and S7).³⁷

Figure 4.

Biosynthesis of isoindolamide A (1). (A) Proposed biosynthetic pathway of 5; (B) Proposed biosynthetic pathway to 1 and 8 by the ave pathway. l-Isd: O-isoindolinone-l-homoserine; OAH: O-acetyl-l-homoserine. Dash arrow indicates the exact timing of AveE is not resolved and may take place at different stages of the pathway; (C) Heterologous expression of ave BGC in A. nidulans; (D) Intact protein mass spectrometry analysis of UncD in the presence of d-alanine. Assay conditions for i) sodium phosphate buffer (NaPB, pH 7.6), UncD (20 μM), ATP (2.5 mM) and MgCl₂ (2.5 mM); ii) NaPB (pH 7.6), UncD (20 μM), d-Ala (1 mM), ATP (2.5 mM) and MgCl₂ (2.5 mM). Reactions were analyzed after 5 hours; (E) LC-MS analysis of products from enzymes assays performed with purified AveB or AveG. Assay conditions for i) NaPB (pH 7.6), AveG (7 μM), 5 (250 μM), OAH (500 μM) and PLP (100 μM). Reaction was quenched after 20 minutes; ii) NaPB (pH 7.6), AveB (20 μM), 6 (250 μM), l-Leu (500 μM), ATP (2.5 mM) and MgCl₂ (5 mM). Reaction was quenched after 4 hours.

Attempts to express the single module NRPS AveC from Escherichia coli BAP1³⁸ led to poor soluble protein yield, while the homolog UncD from the unc BGC expressed well (Figure S6). Intact protein mass spectrometry analysis of UncD following incubation with d-alanine, ATP and MgCl₂ showed a mass increase of 71 mu, which is consistent with the loading of d-alanine to the phosphopantetheinyl thiol of the T domain in UncD (Figures 4A and 4D). Under the same condition, l-alanine was activated and loaded by UncD at a much lower level (Figure S8). Based on these results, we propose the next step in formation of 4 is transfer of the d-alanyl unit from the NRPS to the starter-unit:ACP transacylase (SAT) domain of AveA to initiate polyketide biosynthesis (Figure 4A). Three chain extension cycles with malonyl-CoA, followed by aldol cyclization catalyzed by the product template (PT) domain, and hydrolytic release catalyzed by the thioesterase (TE) domain can give 4 as the product. Most isoindolinone natural products are formed as a racemic mixture via nonenzymatic condensation between aldehyde and amine precursors.^39,40 In contrast, both NRPKS and NRPS machineries are recruited to construct chiral isoindolinone 4 in this pathway. The NRPKS AveA is proposed to be primed by an amino acid via an in trans NRPS and PKS interaction.

The functions of ATP-grasp enzyme (AveB) and PLP-dependent enzyme (AveG) were not discernable from the A. nidulans heterologous expression experiments, as coexpression of these two enzymes did not lead to new products (Figure 4C and Figure S4). No related products could be detected from the native strain A. avenaceus under laboratory culture conditions (Figure S9). As a result, we switched to a yeast-based biotransformation to assay the function of AveB and AveG. Feeding of 5 to Saccharomyces cerevisiae RC01⁴¹ expressing aveB and aveG led to formation of two new metabolites 1 (m/z 428 [M+H]⁺) and 6 (m/z 315 [M+H]⁺) at low titers (Figure S10). Based on the putative function of AveG, and the increase in MW of 6 compared to 5, we hypothesized that AveG could catalyze γ-replacement of acetate in OAH with 5 to generate the amino acid chloro-O-isoindolinone-l-homoserine (l-Cl-Isd) 6 (Figure 4B and Figure S11). To test this, AveG was expressed from E. coli (Figure S6) and assayed in the presence of 5, OAH and PLP. Complete conversion of 5 to 6 was observed within 20 minutes (Figure 4E, i and Figure S12). The identity of 6 was confirmed through a scaled-up reaction and NMR analysis (Table S14, Figures S49–S53). The mechanism of AveG is proposed to follow that of other PLP-dependent, γ-substitution enzymes (Figure 4B), in which OAH serves as the latent electrophile and forms the vinylglycine ketimine intermediate in the enzyme active site. Upon deprotonation, the 3-phenolate of 5 serves as the nucleophile to attack the vinylglycine, leading to formation of a new C-O bond in the product that is released as 6.

To identify the structure of product 1, AveB was expressed and purified from E. coli and was incubated with 6 in the presence of ATP, MgCl₂, and either l-Leu or l-Ile, based on the MW difference between 6 and 1. In the presence of l-Leu, 1 was formed (Figure 4E, ii, and Figure S13). Scaled-up reaction followed by NMR analysis revealed 1 is the dipeptide l-Leu-l-Cl-Isd (Table S9, Figures S27–S31), and is named isoindolamide A. The structure of 1 shows AveB is indeed an amide bond-forming enzyme that activates the carboxylate of l-Leu for condensation with the amino group in l-Cl-Isd to give the new isoindolinone-containing PNP 1. All the biosynthetic enzymes in the ave BGC are accounted for in the biosynthesis of 1, making it the likely end-product of the pathway.

Unc BGC uses ATP-grasp enzymes and an ABS to produce a tripeptide product.

Reconstitution of ave BGC showed the product of the five conserved enzymes and the halogenase is the dipeptide 1. The unc and coc pathways do not encode the halogenase homolog, suggesting that the unchlorinated dipeptide l-Leu-l-Isd 8 could be the product of the five conserved enzyme in these pathways (Figure 5A). We tested this by directly assaying the function of the PLP-dependent enzyme UncF, which readily converted 4 to 7 in the presence of OAH and PLP (Figure 5D, i, and Figure S12). The structure of 7 was confirmed to be l-Isd by NMR (Table S15, Figures S54–S58) and X-ray crystallography (CCDC 2416685). Next, the ATP-grasp enzyme UncK was confirmed to catalyze the ligation of l-Leu to 7 and yielded dipeptide 8, which was structurally confirmed through scaled-up reaction and NMR analysis (Figure 5D, ii) (Table S16, Figures S59–S63). Based on this result, we revisited the ave pathway and assayed the activity of AveB with 7 (Figure 4B and Figure S13), which led to the formation of 8 with the same efficiency as UncK. Although the timing of the chlorinase AveE was not studied due to protein expression difficulties, it is possible that chlorination enzyme is promiscuous and can catalyze chlorination of different intermediates, including the conversion of 8 to 1 (Figure 4B). This would make 8 the common biosynthetic intermediate between ave and unc pathways.

Figure 5.

Discovery and biosynthesis of isoindolamide B (2) from the unc BGC. (A) Proposed biosynthetic pathway of 2; (B) Discovery of unc metabolites from the native strain U. reesii UAMH 1704; (C) Reconstitution of the biosynthetic steps from 4 to 2 using S. cerevisiae RC01-based biotransformation; (D) LC-MS analysis of enzymatic transformation steps from 4 to 2. Assay conditions for i) NaPB (pH 7.6), UncF (9 μM), 4 (250 μM), OAH (500 μM) and PLP (100 μM). Reaction was quenched after 10 minutes; ii) NaPB (pH 7.6), UncK (11 μM), 7 (250 μM), l-Leu (500 μM), ATP (2.5 mM) and MgCl₂ (5 mM). Reaction was quenched after 4.5 hours; iii) NaPB (pH 7.6), UncI (30 μM) and 8 (250 μM). Reaction was quenched after 4 hours; iv) NaPB (pH 7.6), UncG (16 μM), 10 (250 μM), ATP (2.5 mM) and l-Ala-l-Gln (500 μM). Reaction was quenched after 4 hours; v) NaPB (pH 7.6), UncH (14 μM) and 11 (250 μM). Reaction was quenched after 2 hours.

To uncover how the unc pathway can further modify 8, extracts from the native producer U. reesii UAMH 1704 were examined for possible downstream metabolites using the characteristic UV absorbances of isoindolinone (Figure 5B and Figure S14). Two compounds, 2 (0.2 mg/L, m/z 480 [M+H]⁺) and 9 (0.2 mg/L, m/z 281 [M+H]⁺), were identified and isolated (Figure 5B). The planar structure of 2 was determined to be the tripeptide Isd-Ala-Gln (Table S10, Figures S32–S37). This compound is named isoindolamide B. Stereochemistry of the side chains of alanine and glutamine residues in 2 were verified to be in the l configurations by Marfey’s method (Figure S15).³⁶ However, the configuration of the Isd residue could not be determined by the Marfey’s method, as the adduct was not observed. Unexpectedly, the absolute structure of 9 was solved to be d-Isd by NMR analysis (Table S17, Figures S64–S68) and comparison of the specific OR value {[α]^25.2_D +155 (c 0.04, CH₃OH)} with that of l-Isd 7 {[α]^25.6_D −32 (c 0.25, CH₃OH)}.

The isolation of tripeptide 2 from the native host suggests this is likely the bona fide PNP of the unc BGC. To test if 4 is an intermediate of the pathway, a yeast host (RC01) expressing the dipeptide 8 forming genes uncFK, as well as the remaining genes in the pathway, uncGHIJ, was constructed. Feeding of 4 to this host indeed led to the formation of 2 and 9 after 24 hours (Figure 5C, iv). The transformation of dipeptide 8 to tripeptide 2 requires one proteolytic step to remove the N-terminal l-Leu, and successive ligation of l-Ala and l-Gln to the carboxylate end. To examine the role of the remaining enzyme in the pathway in biosynthesis of 2, we used a combination of direct enzyme assays (Figure 5D) and combinatorial unc gene expression in yeast with feeding of 4 (Figure 5C). The next enzyme in the pathway was determined to be the epimerase UncI, as both yeast biotransformation (Figure 5C, i) and enzyme assay (Figure 5D, iii) showed UncI converts 8 to a new product 10 with the same MW (Figure S16). UncI shows sequence homology to bacterial l-Ala-d/l-Glu epimerases that belong to the enolase superfamily. Members of this family use a two-base mechanism to perform epimerization via C_α proton abstraction and reprotonation.⁴² Multiple sequence alignment showed UncI contains the two conserved catalytic lysine residues (Figure S17). Scaled up reaction led to NMR characterization of 10 (Table S18, Figures S69–S73), which showed it to have the same planar structure as 8. Therefore, one of the two C_α atoms is epimerized by UncI.

UncH is annotated as a peptidase, and displays structural homology to members of the prolyl oligopeptidase (POP) family⁴³ based on AlphaFold 3 prediction (Figure S2).⁴⁴ Adding recombinantly expressed UncH to the dipeptides 8 or 10 led to hydrolysis to l-Isd 7 or d-Isd 9, respectively, along with l-Leu (Figure S18). This led us to conclude that 10 is the dipeptide l-Leu-d-Isd, in which the C_α of the Isd residue is epimerized from l in 8 to d by UncI (Figure 5A). While the activity of UncH explains the accumulation of 9 as a byproduct of the pathway, direct assays with 9 and remaining enzymes did not lead to any further modifications, suggesting this compound is a dead-end shunt product. This is consistent with the accumulation of this compound in both the natural and the biotransformation hosts (Figures 5B and 5C). Hence, the timing of UncH to hydrolyze the l-Leu residue to give 2 is likely after ligation of the dipeptide 10 with l-Ala and l-Gln.

The two remaining enzymes in the BGC are the second ATP-grasp enzyme UncJ and the annotated CoA-ligase UncG. We hypothesized that UncG may be an ABS because of common misannotation of ANL-superfamily members. Both enzymes were expressed and individually assayed in the presence of 10, ATP and l-Ala. No product can be observed under these conditions. However, using the dipeptide l-Ala-l-Gln as the amine nucleophile, UncG converted 10 into a tetrapeptide 11 (Figure 5D, iv) in an ATP-dependent, but CoA-independent manner. This compound was also detected at low level from the yeast strain expressing uncFK-uncGI upon feeding of 4 (Figure 5C, ii). Isolation of 11 followed by NMR characterization showed it is indeed the expected tetrapeptide l-Leu-d-Isd-l-Ala-l-Gln (Table S19, Figures S74–S79). The function of UncG as an ABS was suggested by detection of pyrophosphate (PPi) from the enzyme assay (Figure S19). AlphaFold 3 predicted structure of UncG shows a typical tertiary structure of ANL superfamily enzymes, with a large N-terminal domain and a small C-terminal domain (Figure S3).¹¹ However, UncG shows no sequence similarity to the recently discovered fungal ABS Cp1D from the E-64 biosynthetic pathway,²⁰ indicating these two enzymes belong to separate groups of fungal ABSs.

Interestingly, repeating the same assay with 8, l-Ala-l-Gln, ATP and UncG led to the formation of a shunt product 12 while no tetrapeptide can be detected (Figure S20). 12 was structurally characterized to be the diketopiperazine (DKP) cyclo-l-Leu-l-Isd (Table S20, Figures S80–S84). 12 can be formed from the intramolecular cyclization following adenylation of the terminal carboxylate in 8 (i.e. 8-AMP), which outcompetes intermolecular tetrapeptide formation. Formation of the corresponding heterochiral DKP cyclo-l-Leu-d-Isd is greatly diminished in the reaction of UncG with 10 and ATP (Figure S20). While the substrate binding mode of UncG will be revealed by structural investigation, we proposed that this could be due to greater steric constraint following epimerization of the Isd residue in the dipeptide, or different binding modes in UncG that disfavor cyclization of 10-AMP. These observations suggest UncG can adenylate the carboxylate in both 8 and 10 despite the opposite stereochemical configurations of the Isd residue. It is possible that the epimerization step by UncI is strategically incorporated to suppress DKP cyclization and to favor tetrapeptide formation. UncG therefore has potential as a biocatalyst in the enzymatic synthesis of tetrapeptides, given that the stereochemistry of the electrophile is taken into consideration.

When the ATP-grasp enzyme UncJ was coexpressed in the yeast biotransformation host that afforded 11, we detected significant increase in the titer of 11 (Figure 5C, iii). This suggested a role of this enzyme in the pathway, which can be logically assigned as the dipeptide synthetase that forms l-Ala-l-Gln from l-Ala and l-Gln. BLASTP search showed UncJ has low homology to murine carnosine synthase 1 and bacterial d-Ala-d-Ala ligase (Figure S21), even though it was initially annotated as a FsqD homolog (Table S1). To confirm the function of UncJ, in vitro assays of UncJ in different conditions were performed, and the products were derivatized with dansyl chloride for detection. The dansyl adduct of l-Ala-l-Gln (dansyl-l-Ala-l-Gln) could be observed in the reaction of UncJ with l-Ala, l-Gln, MgCl₂ and ATP in potassium phosphate buffer (KPB), while omittance of MgCl₂ abolished function (Figure S21). When this reaction was performed in NaPB to exclude K⁺, dansyl-l-Ala-l-Gln could not be detected (Figure S21). These results confirmed UncJ is an l-Ala-l-Gln ligase that requires both Mg²⁺ and K⁺. The requirement of potassium cation in the reaction is consistent with those reported for bacterial ATP-grasp enzymes that phosphorylate alanine prior to amide bond formation, including d-Ala-d-Ala ligase¹³ and BelU from belactosin pathway.¹⁷

Finally, hydrolysis of 11 by UncH led to a product that is identical to 2 (Figure 5D, v and Figure S18). Based on the intermediates and shunt products characterized for this pathway, the stereochemistry of the tripeptide 2 is assigned as d-Isd-l-Ala-l-Gln (Figure 5A). Compared to ave pathway, dipeptide 8 undergoes significant modification via a sequence of epimerization-dipeptide ligation and N-terminus hydrolysis to arrive at the final product 2 (Figure 5A).

Investigation of PNP products from coc BGC.

The coc BGC is found in the pathogenic C. immitis RS and is conserved in all sequenced C. immitis and C. posadasii strains that cause coccidioidomycosis (valley fever). Transcriptomics analysis showed the entire pathway is transcribed in the pathogenic spherule form,⁴⁵ suggesting the product of this BGC may be a virulence factor. No natural product has been reported from this strain. Since the strain was not accessible, potential products of the BGC were investigated using purified enzymes expressed from synthetic genes (Figure S6). First, to establish that l-Leu-l-Isd (8) is a common biosynthetic fragment as in the ave and unc pathways, we expressed the ATP-grasp enzyme CocI. As expected, CocI displayed the same function as UncK and AveB, and readily amidated l-Isd (7) to 8 (Figure 6B, i, and Figure S22).

Figure 6.

Identification of compound 3 as an advanced PNP product of the coc pathway. (A) Formation of 3 from amide-bond formation between 8 and 13 catalyzed by the ABS homolog CocJ; (B) LC-MS analysis of assays with CocI, CocE and CocJ. AAs: natural amino acids. Traces i – v are total ion counts (TIC) and traces vi and vii are selected ion counts at m/z [M+H]⁺ = 221. Assay conditions for i) NaPB (pH 7.6), CocI (19 μM), 7 (250 μM), l-Leu (500 μM), ATP (2.5 mM) and MgCl₂ (5 mM). Reaction was quenched after 4 hours; ii) NaPB (pH 7.6), CocJ (15 μM), 8 (250 μM), 20 AAs (each 1 mM) and ATP (2.5 mM); iii) NaPB (pH 7.6), CocJ (15 μM), 8 (250 μM), l-Trp (1 mM) and ATP (2.5 mM); iv) NaPB (pH 7.6), CocJ (15 μM), CocE (14 μM), 8 (250 μM), l-Trp (1 mM), FAD (100 μM), NADPH (1 mM) and ATP (2.5 mM); v) NaPB (pH 7.6), CocJ (15 μM), 8 (250 μM), 13 (1 mM) and ATP (2.5 mM); vi) NaPB (pH 7.6), CocE (14 μM), l-Trp (1 mM), FAD (100 μM) and NADPH (1 mM); vii) standard of 13. Reactions with CocJ or CocE were quenched after 3 hours; (C) Two possible mechanisms for the formation of l-Oid (13) from l-Trp.

The next enzyme investigated was the annotated CoA-ligase CocJ. CocJ shares 52% sequence identity to UncG, but is separated into a different clade upon phylogenetic tree analysis (Figure S23). This indicated the nucleophile substrate of CocJ is most likely different from the l-Ala-l-Gln recognized by UncG. When recombinant CocJ was assayed with 8, ATP and twenty natural amino acids, a new product with MW 579 emerged at low conversion (Figure 6B, ii, and Figure S24). This MW is consistent with the addition of l-Trp to 8 to give the tripeptide l-Leu-l-Isd-l-Trp. This result was confirmed by directly using l-Trp in the CocJ assay (Figure 6B, iii). The low conversion, however, suggested that the native substrate of CocJ may be a modified l-Trp derived from the function of another enzyme in the BGC. Among the remaining enzymes encoded in the BGC, the predicted flavin-dependent monooxygenase (FMO) CocE shows 27% sequence homology to dimethylaniline N-monooxygenase FMO5.⁴⁶ We hypothesized that CocE could be a l-Trp modifying enzyme that reacts with the electron-rich indole moiety. To test this, we added CocE, FAD and NADPH to the CocJ assay with l-Trp, which led to the complete conversion of 8 to a new product 3 with MW of 595 (Figure 6B, iv). 3 was isolated from a scaled-up enzymatic reaction with CocJ and CocE, and was characterized by NMR (Table S11, Figures S38–S42) to be the tripeptide l-Leu-l-Isd-l-oxindolylalanine (Figure 6A), in which the indole side chain of the third residue is oxidized into the 2-oxindole.

To pinpoint the timing of 2-oxindole formation, CocE was directly assayed with l-Trp, which led to the formation of the amino acid l-oxindolylalanine (l-Oid, 13) as compared to an authentic standard (Figure 6B, vi and vii). Directly adding 13 to the CocJ and 8 led to the complete conversion of 8 to 3 (Figure 6B, v), while adding CocE to l-Leu-l-Isd-l-Trp did not lead to any modifications (Figure S24). These results confirmed i) CocJ is an ABS enzyme involved in tripeptide 3 biosynthesis; and ii) CocE can catalyze direct oxindole formation with free l-Trp, which has not been reported by a FMO enzyme.^47,48 Two proposed mechanisms of CocE are shown in Figure 6C, differing in the initial site of hydroxylation at either C-3 (path I) or C-2 (path II) of the indole ring. Because the 2-oxindole fragment is an important pharmacophore,⁴⁹ CocE could be a potential biocatalyst for preparing diverse 2-oxindole building blocks. It is intriguing to note that the tripeptide 3 contains two nonproteinogenic amino acids with heterocyclic side chains, isoindolinone in l-Isd and 2-oxindole in l-Oid. Further assays (Figure S25) with the two remaining enzymes CocC (P450) and CocF (FMN oxidoreductase) did not yield any additional modifications to 3. Therefore, the final PNP of the coc BGC remains undetermined.

Summary of ATP-grasp and ABS enzymes characterized in this work.

Three different fungal BGCs were discovered using a conserved ATP-grasp enzyme as the lead. Each BGC encodes a conserved five-gene cassette that we showed to generate the dipeptide l-Leu-l-Isd (8). Additional ATP-grasp and ABS enzymes were encoded in two of the three BGCs and were characterized. Figure 7 shows a summary of the activities of these enzymes. Even in this small set of BGCs examined, diverse amide bond-forming activities were observed, including the synthesis of di-, tri- and tetrapeptides. All three conserved ATP-grasp enzymes, AveB, UncK and CocI, are confirmed to condense l-Leu with l-Isd to give l-Leu-l-Isd (8). An additional ATP-grasp enzyme from the unc pathway condenses l-Ala with l-Gln in a K⁺-dependent fashion to give l-Ala-l-Gln. Both ABS enzymes, UncG and CocJ, were initially annotated as CoA-ligases, which belong in the same ANL-superfamily as ABSs. CocJ is a tripeptide synthetase that condenses l-Leu-l-Isd (8) with the nonproteinogenic amino acid l-Oid (13) to give the tripeptide l-Leu-l-Isd-l-Oid (3). UncG is a tetrapeptide synthetase that condenses two dipeptides, l-Leu-d-Isd (10) and l-Ala-l-Gln, to afford the tetrapeptide l-Leu-d-Isd-l-Ala-l-Gln (11). The differences in both electrophiles and nucleophiles recognized by these amide bond-forming enzymes showcase the potential of these and related enzymes as biocatalytic tools for assembling oligopeptides.

Figure 7.

Summary of amide bond-forming enzymes discovered in this work. Peptides are shown from N-terminus to C-terminus.

CONCLUSIONS

A fragment-guided genome mining approach was used to identify the natural products of three uncharacterized BGCs from diverse fungi, including the causative agent of valley fever. The common fragment is the dipeptide l-Leu-l-Isd (8) that is derived from a five-enzyme pathway, which includes a dedicated ensemble of enzymes for biosynthesis of the nonproteinogenic amino acid O-isoindolinone-l-homoserine (7). A conserved ATP-grasp enzyme, which was the beacon that led to these BGCs, catalyzes dipeptide formation. The common fragment is then elaborated and diversified into three different PNPs by the three pathways, resulting in the identification of a halogenated dipeptide isoindolamide A 1, a d-l-l tripeptide isoindolamide B 2, and a tripeptide 3 with heterocycle-containing amino acids, l-Leu-l-Isd-l-Oid. Although no biological activities have been identified with these PNPs after initial antifungal and antibacterial assays, the structures of these compounds, as well as the nonproteinogenic amino acid building blocks, are new to the natural product chemical space. Given the large number of cryptic BGCs predicted to encode ATP-grasp enzyme and/or ABS as core-forming enzymes, it is anticipated that genome mining using these leads will lead to a substantial increase in the inventory of new PNPs.

Supplementary Material

Additional experimental details, materials, methods, 1D and 2D NMR and UV spectra for all compounds.