Pyridoxal-5′-phosphate–dependent bifunctional enzyme catalyzed biosynthesis of indolizidine alkaloids in fungi

Identification of the Curvulamine Biosynthetic Gene Cluster and Its On-Pathway Intermediates.

In an attempt to get an insight into the curvulamine (1) biosynthesis, we performed the whole genome sequencing of Curvularia sp. IFB-Z10, which produces 1 as a major polyketide alkaloid. We analyzed the sequenced genome by the antibiotics and secondary metabolite analysis shell (antiSMASH) (fungal version) to reveal that this strain contains at least 27 biosynthetic gene clusters (BGCs) (SI Appendix, Fig. S2) (19). In reminiscence of the fact that PKSs are essential for the skeletal construction of polyketide alkaloids (13⇓⇓–16), our attention was first focused on the 10 pks-containing BGCs (designated as pks1-10) (SI Appendix, Fig. S2). Thus, the RT-qPCR experimentation was conducted to analyze the mRNA expression levels of the pks genes in the Czapek-Dox liquid medium that allows the wild-type (WT) strain of Curvularia sp. IFB-Z10 to produce 1 as an abundant secondary metabolite (SI Appendix, Fig. S3A). As a result, the mRNA expression levels of pks4, pks5, and pks10 (designated as cuaA) were obviously higher than those of others (SI Appendix, Fig. S3B). Adjacent to cuaA is a gene (designated as cuaB) that encodes a PLP-dependent aminotransferase (Fig. 2A and SI Appendix, Table S1). Deletion of cuaA and cuaB completely abolished the production of 1 (SI Appendix, Fig. S3C). But the ∆pks4 and ∆pks5 mutants formed 1 as did the WT strain (SI Appendix, Fig. S3C). Thus, the biosynthesis of 1 requires the cuaA and cuaB genes, but not pks4 or pks5 genes.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Identification of the curvulamine (1) biosynthetic gene cluster. (A) The putative cua biosynthetic gene cluster in Curvularia sp. IFB-Z10. CuaA: polyketide synthase; CuaB: PLP-dependent bifunctional aminotransferase; CuaC: short-chain alcohol dehydrogenase; CuaD: FAD-dependent monooxygenase; CuaE: major facilitator superfamily permease; CuaF: Zn(II)Cys₆ transcription factor. (B) HPLC profiling for alkaloids produced by the 6 mutants (∆cuaA−∆cuaF) and the WT strain. *, unidentified minor metabolites.

In view of the disappearance of 1 in ∆cuaA and ∆cuaB mutants, our attention was drawn to the cuaA- and cuaB-containing BGC (hereafter referred to as the cua gene cluster) which was found in scaffold 8. The cua gene cluster includes 4 additional genes that encode respectively a short-chain alcohol dehydrogenase (CuaC), a FAD-dependent monooxygenase (CuaD), a major facilitator superfamily permease (CuaE), and a Zn(II)Cys₆ transcription factor (CuaF) (Fig. 2A and SI Appendix, Table S1). To ascertain gene cluster boundary, we constructed ∆ORF1-3, ∆cuaC-F, and ∆ORF10-12 mutants (SI Appendix, Fig. S4A). The high-performance liquid chromatography (HPLC) analysis of the mutant cultures showed that the ∆ORF1-3 and ∆ORF10-12 strains produced 1 as did the WT strain (SI Appendix, Fig. S4B), underscoring that ORF1 and ORF12 were the left and right boundaries of the cua gene cluster, respectively. Moreover, 1 failed to be detected in the ∆cuaE and ∆cuaF mutant cultures (Fig. 2B, traces iv and v), reinforcing their functional roles in the efficient transportation and up-regulated expression of 1 as annotated by the bioinformatics analysis.

With the gene function ascertained, we were curious to identify the intermediates that may be suggestive of biosynthetic steps toward 1. Our attention was first directed to the products of the ∆cuaC and ∆cuaD mutants. The ∆cuaC mutant was unable to produce 1 but capable of forming 3 and 4 (Fig. 2B, trace vii). Due to its instability, 3 could only be isolated under the protection of dry ice and without any light exposure. Compound 3 was evidenced to have a molecular formula of C₁₀H₁₃NO₂ from its Na⁺-liganded molecular ion at m/z 202.0843 (C₁₀H₁₃NO₂Na requires 202.0839) in its high-resolution electrospray ionization mass spectrometry (HR-ESI-MS) spectrum. Subsequent interpretation of its 1D (¹H and ¹³C) and 2-dimensional (2D) NMR spectra (¹H-¹H chemical shift correlation spectroscopy [COSY], heteronuclear singular quantum correlation [HSQC], and heteronuclear multiple-bond coherence [HMBC]) indicated that 3 was 2-hydroxy-2-methyl-5-((1E,3E)-penta-1,3-dien-1-yl)-1,2-dihydro-3H-pyrrol-3-one (Fig. 3 and SI Appendix, Table S5 and Figs. S16–S20). Compound 4 was found to have a molecular formula of C₁₀H₁₃NO₃ according to its Na⁺-liganded molecular ion at m/z 218.0784 (C₁₀H₁₃NO₃Na requires 218.0788) in its HR-ESI-MS spectrum, underscoring that 4 was a monooxygenated derivative of 3. The same set of 1-dimensional (1D) and 2D NMR spectra of 4 confirmed that it was 2,3-epoxide of 3 (Fig. 3 and SI Appendix, Table S6 and Figs. S21–S25). Complementation of 3 and 4 to ∆cuaA and ∆cuaB mutants restored the production of 1, indicating that both are the on-pathway intermediates (SI Appendix, Fig. S5). The ∆cuaD mutant generated 3 as well, but along with a reduced abundance of 1 (Fig. 2B, trace vi). The residual amount of 1 produced by the ∆cuaD strain (Fig. 2B, trace vi) highlighted the presence of presumable CuaD homolog(s) outside the cua gene cluster as encountered in the l-allo-isoleucine biosynthesis (20).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Proposed curvulamine (1) biosynthetic pathway based on gene mutation, heterologous expression, and enzymatic assay. CuaB, a bifunctional aminotransferase, catalyzes the formation of both C−C (via Claisen condensation) and C−O bonds in the presence of O₂.

Heterologous Expression of CuaA, CuaB, and CuaD in Aspergillus oryzae and Saccharomyces cerevisiae.

The production of 3 and 4 by the ∆cuaC mutant prompted us to hypothesize that CuaA and CuaB might be involved in the nascent stage of the 1 biosynthesis. To confirm the assumption, we conducted the heterologous coexpression of cuaA and cuaB in Aspergillus oryzae NSAR1 and Saccharomyces cerevisiae BJ5454-NpgA (cuaA and cuaB without introns) (21⇓–23). As expected, 3 formed in the 2 hosts after a 3-d fermentation (Fig. 4A, traces ii and iv). The structural feature of 3 agrees with the bioinformatic analysis suggesting that methyltransferase and enoyl reductase domains of CuaA are inactive (SI Appendix, Fig. S6) (24). These results confirmed that the cuaA and cuaB genes are responsible for the early-stage biosynthesis of 1 with the on-pathway intermediate 3 produced via a consortium of CuaA and CuaB (Fig. 3). This is another example of PLP-dependent polyketide chain release involved in the natural product biosynthesis (25).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Heterologous expression of cuaA, cuaB, and cuaD. (A) HPLC profilings (recorded at 310 nm) for 3 and 4 produced in A. oryzae (AO) and S. cerevisiae (S-BJ5464). (B) CuaD (100 μM) catalyzes the conversion of 3 into 4 in the 50-mM PBS buffer (pH 7.2).

With the finding, we speculated the FAD-dependent monooxygenase CuaD might be responsible for the conversion of 3 into 4. The recombinant CuaD protein was overexpressed in Escherichia coli transetta (DE3) and purified as a yellow protein with FAD as the cofactor (SI Appendix, Fig. S7). Exposure of CuaD to 3, FAD, and NADH led to the complete transformation of 3 into 4 over a prolonged reaction time (Fig. 4B, traces i–iii). We also coexpressed cuaA, cuaB, and cuaD in A. oryzae NSAR1 to reinforce the in vivo production of 4 (Fig. 4A, trace v). The above experiments established CuaD as a monooxygenase that catalyzes the 2,3-epoxidation of 3 to form 4 (Fig. 3). Next, we tried to characterize the function of CuaC, annotated as short-chain alcohol dehydrogenase. Unfortunately, the CuaC protein could not be obtained in a soluble form after many attempts. However, the accumulation of 4 by the ∆cuaC mutant (Fig. 2B, trace vii) could be rationalized only by assuming that CuaC is likely involved in the further modification of 4 (Fig. 3). Presumably, a proton-promoted dehydration of 4 gives 5, which is transformable into 6 after ketoreduction (26). The keto–enol tautomerization of 6 allows a second-round ketoreduction to form 7, whose nitrogen atom attacks the epoxide carbon to yield 8 after dehydration. Further tailoring of 8 affords 1 (Fig. 3). The proposed function of CuaC gained a reinforcement from its homolog RhlG (31% identity to CuaC), a NADPH-dependent β-ketoacyl reductase of short-chain alcohol dehydrogenase/reductase (SDR) superfamily catalyzing the ketoreduction steps in the biosynthesis of rhamnolipid (27).

Biochemical Characterization of CuaB.

As confirmed by the heterologous expression in A. oryzae and S. cerevisiae BJ5454-NpgA, the coinvolvement of CuaA and CuaB in the production of 3 motivated us to understand the biochemical feature of CuaB. Therefore, we expressed CuaA (276 kDa) in S. cerevisiae BJ5454-NpgA and the recombinant CuaB in E. coli transetta (DE3) (SI Appendix, Fig. S8A). The purified CuaB is a yellow holoenzyme with a characteristic absorbance maximum at 425 nm, indicating the formation of internal aldimine by combining with the PLP cofactor (SI Appendix, Fig. S8B) (28). Then equimolar amounts (100 μM) of CuaA and CuaB were incubated at 30 °C for 10 h with the substrates (l-alanine [1 mM], acetyl-CoA [2 mM], and malonyl-CoA [2 mM]) in the presence of NADPH (4 mM) and PLP (1 mM). After being quenched by adding acetonitrile, the liquid chromatography–mass spectrometry (LC-MS) analysis of the reaction mixture indicated the production of 3 (Fig. 5A, trace vi). At this stage, our attention was drawn to the unusual hydroxylation of C9 in 3 (Fig. 3). The reported PLP-dependent Claisen condensation between polyketide chain or acyl-CoA and amino acids results in the formation of a new C–C bond as discerned in the biosynthesis of undecylprodigiosin, fumonisin B1, and saxitoxin (29⇓–31). Nevertheless, the α-hydroxylation of amino acid residue was not observed. The α-hydroxylation of the on-pathway intermediate 3 suggests an unusual catalytic mechanism of CuaB, which might be fundamental to the construction pattern for the indolizidine skeleton. In order to verify whether CuaB is responsible for the α-hydroxylation of alanine residue of 3, we synthesized the substrates, (4E,6E)-3-oxoocta-4,6-dienoyl-N-acetylcysteamine (numbered S5) (SI Appendix, Figs. S26–S28) and (4E,6E)-3-oxoocta-4,6-dienoyl-SCoA (numbered S13) (SI Appendix, Figs. S29–S33), to mimic the acyl carrier protein (ACP)-bound tetraketide (3′, Fig. 3). The reaction of S5 and l-alanine was performed with exposure to CuaB and PLP, but no production of 3 was revealed by the LC-MS analysis (Fig. 5A, trace i), suggesting that S5 could not efficiently mimic the biosynthetic step. Therefore, we expressed the ACP domain of CuaA and phosphopantetheinyl transferase (PPTase) NpgA in E. coli transetta (DE3) (SI Appendix, Fig. S9A). A 340-Da shift was observed by the matrix-assisted laser desorption/ionization time-of-flight mass spectrometer (MALDI-TOF-MS) analysis after the incubation of the apo CuaA-ACP and CoA using NpgA, indicating that apo CuaA-ACP was converted into its holo form (SI Appendix, Fig. S9B). In addition, S13 was also confirmed to be accommodated by holo CuaA-ACP by the observed 136-Da increment in the molecular weight (SI Appendix, Fig. S9B). The substrate S13 and l-alanine were incubated with CuaB in the presence of apo CuaA-ACP, NpgA, and PLP. Satisfactorily, 3 was produced as detected by the LC-MS analysis (Fig. 5A, trace iv). These results indicated that the coparticipation of CuaA and CuaB was important for the recognition of substrate and subsequent catalysis.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Biochemical characterization of CuaB with PLP as a cofactor. (A) LC-MS profilings for the CuaB catalyzed transformation of S13 or S5. (B) LC-MS analysis of the CuaB-transformed products (3 with and without ¹⁸O atom) in exposure to ¹⁸O₂ (i) and H₂¹⁶O (ii), and to ¹⁶O₂ (iii) and H₂¹⁸O (iv).

Next, we were curious about the origination of the hydroxyl group in 3. Thus, S13 was subjected to an anaerobic conversion test, but it failed to form 3 (Fig. 5A, trace ii). This suggests that dioxygen was likely indispensable for the catalytic ability of CuaB. To reinforce the assumption, we conducted separately the isotope-labeling experiments using ¹⁸O₂ and H₂¹⁸O. As indicated by the LC-MS analysis, ¹⁸O-3* (3 labeled with an ¹⁸O atom) generated in the presence of ¹⁸O₂ and displayed its [M+Na]⁺ ion at m/z 204, which has a 2-Da increment from that of 3 formed in exposure to ¹⁶O₂ (Fig. 5 B, i and iii). Therefore, the hydroxyl oxygen of 3 is derived from dioxygen, not from water. PLP is a versatile cofactor which is involved in a wide range of reactions as reviewed recently (32). To our knowledge, a total of 4 PLP-dependent enzymes are noticed to activate O₂ to complete challenging reactions including the oxidation of inert C–C bonds (SI Appendix, Fig. S10) (33⇓⇓–36). To validate the possible catalytic residues of CuaB, the site-directed mutagenesis experiments were conducted. For the PLP-dependent AONSs, the active site residues including His, Ser, Aps, Thr, and Lys are important for the AONSs to catalyze the C–C bond formation through the decarboxylative condensation between an amino acid and an acyl-CoA (28, 37). Similarly, the multiple sequence alignment indicated H158, S214, D243, H246, T274, and K277 in CuaB are also conserved (SI Appendix, Fig. S11). The purified CuaB-K277A mutant is colorless and inactive (Fig. 5A, trace iii and SI Appendix, Fig. S8B), indicating K277 covalently binds the PLP cofactor to generate internal aldimine (Fig. 3). The H158A, S214A, and T274A mutations extremely decreased the activity of CuaB (SI Appendix, Fig. S13). However, H246 seems inessential for the enzymatic function since the CuaB-H246A mutant was shown to be as active as CuaB (SI Appendix, Fig. S13). Interestingly, a total of 12 polar amino acids (K199, K202, R203, D266, H337, H367, E368, D369, H373, E374, T378, and H379) were distinctly conserved in CuaB and its homologous proteins, but not in PLP-dependent AONSs (SI Appendix, Fig. S11). Furthermore, E368 and H373 are particularly essential for the discerned CuaB function because CuaB-E368A and CuaB-H373A mutants were demonstrated to have sharply reduced catalytic activity (SI Appendix, Fig. S13). Collectively, the amino acid residues H158, S214, T274, K277, E368, and H373 are important for the CuaB function. In addition, our phylogenic analysis indicated that CuaB and its homologs are located in the same clade (Fig. 6) but separated from AONSs and other PLP-dependent oxidases (33⇓⇓–36, 38, 39). Accordingly, CuaB showcases an undescribed family of PLP-dependent bifunctional enzymes that catalyzes both the C–C bond formation and activates O₂ via a hydroperoxide-PLP adduct to complete the α-hydroxylation of amino acid residue. With the finding, we were curious about whether other enzymes in this clade can perform the same/similar catalytic function as does CuaB. Therefore, we overexpressed and purified the XP_007700279.1 protein from Bipolaris sorokiniana ND90Pr (designated as CuaB-H1 with 90% similarity and 83% identity to CuaB) and the XP_007711337.1 protein from Bipolaris zeicola 26-R-13 (designated as CuaB-H2 with 90% similarity and 84% identity to CuaB). As expected, the 2 enzymes did catalyze the conversion of S13 into 3 as indicated by LC-MS analysis (SI Appendix, Fig. S8C).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

The phylogenetic tree of CuaB and its homologs. CuaB, XP_007711337.1, and XP_007700279.1 (solid circle in red) located in a distinct clade can catalyze both the C−C bond formation (via Claisen condensation) and α-hydroxylation of alanine residue employing O₂. This is functionally different from other PLP-dependent enzymes such as PLP-dependent oxidases CcbF, MppP, RohP, and Ind4 (solid circle in blue) that catalyze oxidation (33⇓⇓–36), and l-serine palmitoyltransferase (SPT) and AONS (solid circle in yellow) catalyze the C−C bond formation via the Claisen condensation of acyl-CoA with α-amino acids (38, 39). The sequence alignment was analyzed by Clustalw and the phylogenetic tree was constructed by the neighbor-joining method using MEGA 7.0 software.

In view of biotin biosynthesis (28), the K277 residue in CuaB could be proposed to bind to the PLP cofactor to form the internal aldimine (Fig. 3), which is subsequently exchanged with l-alanine to generate external aldimine. Specifically, K277 despoils the C_α-proton of the external aldimine to yield the quinonoid intermediate, which primes the tetraketide chain-tethering ACP to form intermediate I via Claisen condensation (Fig. 3). Decarboxylation of I gave the quinonoid carbanion complex II. An electron derived from II is transferred to O₂ to produce radical aldimine and superoxide anion complex III, which tends to form a hydroperoxide-PLP adduct IV. These processes were adopted for reasoning the biosynthetic pathways of celesticetin and capuramycin (33, 34). Most likely, the water attacks the peroxy bond of IV to complete the hydroxylation of V with the elimination of H₂O₂ in analogy with the proposals for the PLP- and O₂-codependent oxidations (32). The amino nitrogen atom in V was apt to nucleophilically attack the conjugated carbonyl group to give 3 after the release of PLP and water (Fig. 3). This rationalization was reinforced by the reaction assay based on the principle of the H₂O₂ oxidation-triggered fluorescent liberation within a few seconds and the production of H₂O₂ in stoichiometric amounts (SI Appendix, Fig. S12) (40).

Genome Mining for New Antibacterial Indolizidine Alkaloids.

Many new bioactive polyketide alkaloids are thought to exist in nature as “hidden” natural products, since their gene clusters keep silent in laboratory conditions (41). Inspired and guided by the distinctness of CuaB involved in the biosynthesis of curvulamine (1), an antibacterial polyketide alkaloid (17) (Fig. 6), a genome mining effort was performed to search for the homologous BGCs in fungal genomes available in the National Center for Biotechnology Information (NCBI) and Joint Genome Institute (JGI) databases. After AntiSMASH analysis, 5 BGCs were shown to be similar to the cua gene cluster that encodes the typical enzymes (SI Appendix, Fig. S14A) required for the biosynthesis of curvulamine-like indolizidine alkaloids. Furthermore, such genome mining allowed us to recognize 5 more distinct BGCs that encode, in addition to the CuaB-like enzyme, a set of different tailoring enzymes including cytochrome P450 monooxygenase, cofactor F₄₂₀-dependent oxidoreductase, and α-ketoglutarate–dependent oxygenase (SI Appendix, Fig. S14B). These BGCs were therefore anticipated to synthesize novel polyketide alkaloids. Among them, the homologous bip gene cluster in Bipolaris maydis ATCC48331 (Fig. 7A and SI Appendix, Table S2) was expected to be prominent in constructing polyketide alkaloids. In contrast to our expectation, no polyketide alkaloid was detectable in the B. maydis culture in different media (Fig. 7B, traces i–viii and SI Appendix, Table S16), suggesting the assumed bip gene cluster was likely silent in the laboratory condition (Fig. 7B, traces i–viii). After scrutinizing the cua and bip gene clusters (SI Appendix, Table S2), we noticed and overexpressed the putative transcription factor bipF (52% identity to cuaF) using a strong α-amylase (amyB) promoter to activate the bip gene cluster (Fig. 7B, trace ix). The cultivation of the B. maydis (OE::bipF) strain was scaled up, and the ethyl acetate extract was fractionated to afford 9 new polyketide alkaloids we have named bipolamines A−I (14−22) (Fig. 7B). The structure and absolute stereochemistry of 14 to 22 were unequivocally assigned by a combination of their 1D (¹H and ¹³C) and 2D NMR spectra (¹H-¹H COSY, HSQC, HMBC, and NOESY) and single crystal X-ray diffraction (SI Appendix, Tables S7–S15 and Figs. S35–S92). In agreement with the gene organization, 14 to 22 are all polyketide alkaloids (Fig. 7B).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 7.

Genome mining for new polyketide alkaloids from the silent bip cluster in B. maydis ATCC48331. (A) Comparison between the cua and bip clusters. (B) HPLC profilings of the EtOAc extracts derived from the B. maydis cultures in different media. The silent bip cluster was activated by overexpressing transcription factor BiF, leading to characterization of new alkaloid bipolamines A−I (14−22). *, Nonalkaloidal secondary metabolites (a−f) ascertained by the LC-HR-ESI-MS analysis (SI Appendix, Table S18).

Inspired by the bacterial growth inhibition of curvulamine (1) (17), 14 and 17 to 22 were evaluated for the antibacterial action against pathogens Veillonella parvula, Actinomyces israelii, Streptococcus sp., Peptostreptococcus sp., and Bacteroides vulgates. Gratifyingly, bipolamine G (20) was shown to be more potent than 1 (SI Appendix, Table S17). Due to the instability, 15 and 16 decomposed prior to the antibacterial assay. This result is supportive of the conserved mechanism for the polyketide alkaloid construction in fungi and demonstrates that the genome mining strategy is effective for the discovery of new bioactive natural products biosynthesized by silent gene clusters.

Indolizidine alkaloids are widespread in nature since they can be biosynthesized by many plant species and microbial strains (3, 4). Some animal species such as toads also “produce” toxic indolizidine alkaloids by utilization of alkaloid-containing prey for the sake of maintaining the chemical defense (6). Thus, there might be an unforeseeable versatility of the indolizidine alkaloid biosynthesis in different organisms. This is substantiated by the present study that reveals an undescribed construction of the indolizidine nucleus of curvulamine (1) in Curvularia sp. IFB-Z10. Our attention to the biosynthetic gene cluster of 1 led to the identification of CuaB as a bifunctional PLP-dependent aminotransferase capable of catalyzing both the Claisen condensation to form new C–C bonds and the α-hydroxylation of the amino acid residue in the resonance-stabilized quininoids II and III (Fig. 3). As the active form of vitamin B6, pyridoxal 5-phosphate is a cofactor for a great variety of enzymes catalyzing a long list of reaction processes, including primarily the transamination or desamination, decarboxylation, and racemization of α-amino acids (42). However, CuaB is functionally distinct from the elsewhere-sourced homologs such as the plant prokaryotic-type enzyme with aspartate and prephenate aminotransferase activities (43) and l-aspartate 4-decarboxylase catalyzing the aspartate decarboxylation and transamination reactions (44). This agrees with the phylogenetic analysis of CuaB and its homologs, establishing its distinctness as a bifunctional aminotransferase (Fig. 6).

As an efficient way of obtaining new bioactive natural products, the genome mining effort can be facilitated by the BGC organization of almost all types of compounds such as meroterpenoids (45). This strategy enables as well the characterization of structurally diverse secondary metabolites encoded by orphan and neglected biosynthetic gene clusters (46). Herein, we used the genome mining approach to identify a cryptic bip gene cluster in B. maydis ATCC48331 under the inspiration of the particular function of CuaB as a PLP-dependent aminotransferase. Activation of the silent bip cluster resulted in a family of previously uncharacterized polyketide alkaloids, including bipolamine G (20) which is more antibacterial than curvulamine (1).