Efficient nonenzymatic cyclization and domain shuffling drive pyrrolopyrazine diversity from truncated variants of a fungal NRPS

Significance

Nonribosomal peptide synthetases (NRPSs) synthesize the core peptide scaffold of many natural products. These include small cyclic dipeptides such as peramine, which is a potent insect feeding deterrent synthesized by the 2-module NRPS PpzA-1 from grass endophytic fungi. Here we identify several new PpzA variants lacking the C-terminal product release domain of PpzA-1 that instead utilize efficient nonenzymatic cyclization of the NRPS-tethered dipeptidyl-thioester intermediate to release a range of diketopiperazine-containing products. The metabolic diversity generated from these truncated variants is the result of altered biosynthetic activities combined with recombination-mediated domain shuffling. This work highlights that allelic variants of a single NRPS can result in a surprising level of secondary metabolite diversity comparable to that observed for some gene clusters.

Abstract

Nonribosomal peptide synthetases (NRPSs) generate the core peptide scaffolds of many natural products. These include small cyclic dipeptides such as the insect feeding deterrent peramine, which is a pyrrolopyrazine (PPZ) produced by grass-endophytic Epichloë fungi. Biosynthesis of peramine is catalyzed by the 2-module NRPS, PpzA-1, which has a C-terminal reductase (R) domain that is required for reductive release and cyclization of the NRPS-tethered dipeptidyl-thioester intermediate. However, some PpzA variants lack this R domain due to insertion of a transposable element into the 3′ end of ppzA. We demonstrate here that these truncated PpzA variants utilize nonenzymatic cyclization of the dipeptidyl thioester to a 2,5-diketopiperazine (DKP) to synthesize a range of novel PPZ products. Truncation of the R domain is sufficient to subfunctionalize PpzA-1 into a dedicated DKP synthetase, exemplified by the truncated variant, PpzA-2, which has also evolved altered substrate specificity and reduced N-methyltransferase activity relative to PpzA-1. Further allelic diversity has been generated by recombination-mediated domain shuffling between ppzA-1 and ppzA-2, resulting in the ppzA-3 and ppzA-4 alleles, each of which encodes synthesis of a unique PPZ metabolite. This research establishes that efficient NRPS-catalyzed DKP biosynthesis can occur in vivo through nonenzymatic dipeptidyl cyclization and presents a remarkably clean example of NRPS evolution through recombinant exchange of functionally divergent domains. This work highlights that allelic variants of a single NRPS can result in a surprising level of secondary metabolite diversity comparable to that observed for some gene clusters.

Nonribosomal peptide synthetases (NRPSs) are large multimodular proteins that produce a huge variety of bioactive nonribosomal peptide (NRP) natural products, including small cyclic dipeptides such as the 2,5-diketopiperazines (1) and the pyrazinones (2–4). NRP biosynthesis occurs via a thiotemplate mechanism, with each module responsible for the colinear incorporation of a single amino acid substrate into the growing peptide chain (5–7). Each module contains an adenylylation (A) domain, which binds and activates a specific amino acid substrate using adenosine 5′-triphosphate, and a thiolation (T) domain with a 4′-phosphopantetheinyl cofactor to which the activated aminoacyl substrate is tethered via a thioester bond. Modules downstream of the N-terminal “initiation” module also begin with a condensation (C) domain, which catalyzes peptide bond formation between adjacent thiotethered aminoacyl or peptidyl intermediates, and all modules may optionally contain one or more accessorizing domains that modify the aminoacyl intermediate, such as the N-methyltransferase (M) domain. Most NRPSs end with a termination domain, such as a thioesterase or reductase (R) domain, which catalyzes release of the mature peptidyl chain (8).

Peramine (1a; Fig. 1) is a pyrrolopyrazine-1-one metabolite (PPZ-1-one) synthesized by fungi of genus Epichloë (9), which are symbiotic endophytes of cool-season grasses (10). Epichloë spp. produce a range of bioactive secondary metabolites that protect their grass hosts against herbivory (10), and production of 1a is agriculturally desirable due to the protection it provides against the pastoral pest insect Listronotus bonariensis (Argentine stem weevil) (9, 11, 12). Biosynthesis of 1a is catalyzed by the 2-module NRPS “PpzA-1,” which has the domain structure A₁-T₁-C₂-A₂-M₂-T₂-R₂, where numbers indicate the module to which each domain belongs (2). PpzA-1 also has a partial C domain at its N terminus that is thought to be an evolutionary relic required to maintain A-domain stability (13). The first module of PpzA-1 is thought to incorporate a pyrrolidine-containing amino acid substrate, while the second module incorporates and N-methylates arginine (Arg). A peptide bond is then formed between these aminoacyl substrates, and the resulting dipeptidyl thioester is reductively released by the R₂ domain as a dipeptide aldehyde intermediate that is then thought to undergo spontaneous cyclization, rearrangement, and oxidation reactions to form 1a.

Fig. 1.

Functionally divergent regions of PpzA-1 and PpzA-2 proteins. (A) Sliding-window analysis (width = 100, step = 50) of protein sequence conservation across an alignment of Efe_PpzA-1 and Efe_PpzA-2 protein sequences. The Efe_PpzA-1 protein map is annotated to show approximate domain boundaries and the A₁-domain substrate binding region (A₁-SBR). The positions of substitutions that are both conserved among PpzA-2 proteins and unique to PpzA-2 proteins are annotated on the Efe_PpzA-2 protein map with black pins. Gray pins indicate residues which are always substituted in PpzA-2 proteins relative to PpzA-1 proteins but whose identities are not fully conserved among PpzA-2 proteins. (B) A₁- and A₂-domain NRPS code residues for representative PpzA-1 and PpzA-2 protein sequences. Residues that differ from Epichloë spp. PpzA-1 codes are highlighted in red. (C) Structure of peramine (1a).

This study focuses on allelic variants of ppzA that contain a transposable element insertion into the 3′ end of the gene, resulting in deletion of all sequence encoding the C-terminal R₂ domain (14–16). While initially assumed to be pseudogenes, these truncated “ppzA-2” alleles were subsequently found to be widely distributed across multiple Epichloë spp. (14) and are still expressed at levels comparable to ppzA-1 (17, 18). The ppzA-1 and ppzA-2 alleles also exhibit transspecies polymorphisms (TSP), suggesting that balancing selection has maintained both alleles since the emergence of ppzA-2 in a common ancestor of most Epichloë spp. (14). We therefore hypothesized that ppzA-2 alleles may encode functional NRPSs with novel biosynthetic activities. As bioprotective Epichloë secondary metabolism genes such as ppzA are typically only expressed by Epichloë spp. in planta (2, 19), we use a heterologous expression system to identify the PPZ metabolites produced by different PpzA variants. We also investigate the biosynthetic mechanisms underpinning PPZ biosynthesis and demonstrate how recombination between different ppzA alleles has driven PPZ biosynthetic diversity. Note that while the peramine synthetase-encoding gene was originally abbreviated as “perA” (2), we suggest a modification to “ppzA” to better reflect the class of metabolites produced, with allele ppzA-1 being synonymous with perA.

Results

Identifying Divergent Regions in PpzA-2 Proteins.

Although PpzA-2 proteins lack the C-terminal R₂ domain found in PpzA-1, the domain structures of PpzA-1 and PpzA-2 proteins are otherwise identical. Sliding-window analysis showed that conservation between Epichloë festucae PpzA-1 (Efe_PpzA-1) and PpzA-2 (Efe_PpzA-2) proteins approaches 100% sequence identity in many regions (Fig. 1A). However, discrete regions within the A₁, C₂, M₂, and T₂ domains exhibit increased divergence, and substitutions that are both conserved among and unique to PpzA-2 proteins (PpzA-2–conserved substitutions) are exclusively located within these divergent regions (Fig. 1A and SI Appendix, Figs. S1–S4). Efe_PpzA-2 sequence divergence within the A₁ domain primarily affected the substrate-binding region (SBR), which is located between conserved A-domain motifs A4 and A5 (SI Appendix, Fig. S1) (20). This SBR contains 9 of the 10 “NRPS code” residues, which are the primary determinants of A-domain substrate specificity (21). The NRPS code residues of PpzA-1 A₁ domains are absolutely conserved across all Epichloë spp.; however, 2 of these residues are always substituted in PpzA-2 proteins, and additional substitutions are observed in PpzA-2 proteins from Epichloë bromicola and Epichloë typhina strains (Fig. 1B and SI Appendix, Fig. S1). This lack of NRPS code conservation suggests that the A₁ domains from PpzA-1 and PpzA-2 proteins may recognize different amino acid substrates.

Recombinational Shuffling Has Generated ppzA Allelic Diversity.

Sliding-window phylogenetic comparison between different ppzA alleles indicated that the Efe_ppzA-2 is actually a chimeric allele resulting from several recombinational cross-over events between ancestral ppzA-1 and ppzA-2 sequences (Fig. 2A). Approximately 25% of Efe_ppzA-2, including the A₁ SBR, the 5′ end of the M₂ domain, the T₂ domain, and a small part of the A₂-encoding region, appears to be inherited from an ancestral ppzA-2 allele. The remaining 75% of Efe_ppzA-2, including the T₁ and C₂ domains, the majority of the A₁ and A₂ domains, and the 3′ end of the M₂ domain, appears to descend from an ancestral ppzA-1 allele (Fig. 2A). Very similar patterns of recombination were observed in ppzA-2 alleles from E. bromicola (Ebr_ppzA-2); however, the ppzA-1–derived regions of Ebr_ppzA-2 and Efe_ppzA-2 alleles still grouped with extant Ebr_ppzA-1 and Efe_ppzA-1 sequences, respectively (SI Appendix, Fig. S5). This phylogeny therefore supports a convergent evolution model where Ebr_ppzA-2 and Efe_ppzA-2 were generated by independent yet equivalent recombination events that occurred after divergence of the E. bromicola and E. festucae lineages. No evidence of widespread recombination between ppzA alleles was observed within the E. typhina clade (SI Appendix, Fig. S5).

Fig. 2.

Recombination between ppzA alleles generated additional diversity. (A) Sliding-window comparison of listed alleles to consensus sequences for Ety_ppzA-1, Ety_ppzA-2, Efe_ppzA-1 (for Efe_ppzA_2, Eba_ppzA-3, and Esi_ppzA-4), or Ebr_ppzA-1 (for Ebr_ppzA-2 and Ebr_ppzA-3) using DualBrothers 1.1.5 (52) to identify spatial variations in phylogenetic topology (200-bp window, 50-bp step). Graph lines are slightly offset on the vertical axis to aid visual clarity. (B) Proposed order of ancestral meiotic recombination events between ppzA alleles in the E. festucae clade lineage, where E. baconii is defined as an E. festucae clade member. Ancestral and recombination-derived ppzA-2 alleles are differentiated as ppzA-2^A and ppzA-2^R, respectively.

Some ppzA alleles from Epichloë baconii, E. bromicola, and E. festucae isolates contained the same 3′ deletion found in ppzA-2 alleles yet exhibited recombination patterns that were different from their ppzA-2 counterparts and were therefore defined as “ppzA-3” and “ppzA-4” alleles (Fig. 2). The ppzA-3 allele was identified from some E. bromicola and E. baconii strains and differs from its ppzA-2 counterparts in that it retains a ppzA-1–type A₁ SBR-encoding region (Fig. 2A). Phylogenetic sequence analysis suggests that these Ebr_ppzA-3 and Eba_ppzA-3 alleles represent intermediates during the evolution of Ebr_ppzA-2 and Efe_ppzA-2, respectively (Fig. 2B and SI Appendix, Fig. S5); however, both alleles have also been retained within their respective host populations. The ppzA-4 allele was only identified in Epichloë siegelii, which is an asexual allopolyploid hybrid species of E. festucae and E. bromicola (22). This Esi_ppzA-4 allele is from the E. festucae-derived portion of the E. siegelii genome and appears to be an Efe_ppzA-2 sequence that underwent an additional recombination event that replaced most of the M₂-encoding region with sequence from an Efe_ppzA-1 donor (Fig. 2 A and B and SI Appendix, Fig. S5). Because E. siegelii is asexual and lacks a suitable Efe_ppzA-1 donor sequence, this additional recombination event likely predates the emergence of E. siegelii, although an equivalent ppzA-4 allele has not been observed in any E. festucae isolates to date.

The Truncated Allele ppzA-2 Encodes a Pyrrolopyrazine-1,4-Dione Synthetase.

The fungus Penicillium paxilli strain PN2013 was utilized as a naive heterologous expression host to identify the metabolite products of the enzymes encoded by different ppzA alleles, with 1a-producing PN2013/Efe_ppzA-1 control strains obtained from a previous study (23). As expected, PN2013 transformants expressing Efe_ppzA-2 from E. festucae strain E189 or Ety_ppzA-2 from E. typhina subsp. poae strain E1022 did not produce 1a (SI Appendix, Table S1); however, metabolomic analysis identified a 254.16 m/z metabolite exclusively produced by these ppzA-2–expressing strains (SI Appendix, Supplementary Note). Liquid chromatography-coupled tandem mass spectrometry comparison to a synthetic standard and NMR analysis of the purified natural product determined that this metabolite was the PPZ-1,4-dione molecule cyclo(l-Pro, l-Arg) (2a) (Fig. 3 A and B and SI Appendix, Supplementary Note). Analysis of extracts from grass material confirmed that 2a was present in all samples infected with ppzA-2-genotype Epichloë isolates, whereas 2a was not observed in any samples infected with ppzA-1-genotype Epichloë isolates or in uninfected plants (Table 1). The Efe_ppzA-2 gene was deleted in E. festucae strain E189 (SI Appendix, Fig. S6), and loss of 2a was confirmed in grass samples infected with these ∆Efe_ppzA-2 mutants (Fig. 3C and SI Appendix, Table S2).

Fig. 3.

Identification and characterization of PPZ-1,4-diones synthesized by PpzA-2 proteins. (A) Polymorphic structures showing all PPZ metabolites described in this study. (B) Comparison of MS² spectra generated at 35% normalized collision energy for a synthetic 2a standard vs. a 254.16 m/z metabolite extracted from E. typhina-infected plant material and PN2013/Ety_ppzA-2 mycelia. (C) Concentration of 2a in blade (BLD) or pseudostem (PST) tissue from 2 different grass hosts infected with wild-type E. festucae E189 or 1 of 2 ∆ppzA-2 mutants. Error bars show the SEM; “nd” indicates where 2a was not detected in a sample (limit of detection 0.05 µg/g). (D) Concentration of metabolites 2a and 2b in extracts from cultures of P. paxilli strains that express the PpzA proteins shown to the left of the image. Protein maps are annotated with domain boundaries, and synthetic hybrid proteins are colored to indicate the natural PpzA protein from which each region is derived. Concentrations are averaged across at least 3 independent transformants for each ppzA expression construct, with the SEM shown in parentheses. Metabolites that were not detected are annotated “nd” (limit of detection 0.05 µg/g). Colored bars illustrate the relative between-sample concentration for each metabolite. Concentrations for 2b are estimates based on the response factor of synthetic 2a. **P < 0.01 and ***P < 0.001.

Table 1.

PPZ profile of grasses infected with representative Epichloë strains^*

				PPZ concentration†, µg/g
Allele‡	Endophyte	Strain	Host	[1a]	[1b]§	[1c]§	[2a]	[2b]¶	[2c]¶	[2d]¶
ppzA-1	E. bromicola	NFe1	Hordeum bogdanii	26.4	0.2	nd	nd	nd	1.0	0.4
ppzA-1	E. festucae var. lolii	AR5	Lolium perenne	19.1	nd	nd	nd	nd	6.0	nd
ppzA-1	E. typhina	E8	Lolium perenne	347	0.3	LOQ	4.3	4.0	58.3	nd
ppzA-1	E. typhina ssp. poae	NFe76	Bromus laevipes	101	LOQ	nd	nd	nd	18.5	nd
ppzA-2	E. festucae	E189	Festuca rubra ssp. rubra	nd	nd	nd	19.3	1.2	nd	nd
ppzA-2	E. typhina	AL1218	Dactylis glomerata	nd	nd	nd	699	nd	nd	0.6
ppzA-2	E. typhina ssp. poae	AL9921/1	Poa nemoralis	nd	nd	nd	32.9	nd	nd	nd
ppzA-3	E. baconii	E424	Agrostis tenuis	nd	nd	nd	nd	nd	nd	0.4
ppzA-3	E. bromicola	NFe7	Hordeum brevisbulatum	nd	nd	nd	nd	nd	nd	9.6
ppzA-4	E. siegelii	e915	Festuca arundinacea	nd	nd	nd	nd	515	0.7	nd
ppzA-5	E. uncinata	e167	Festuca pratensis	nd	nd	nd	nd	nd	2.5	nd
N/A	uninfected	N/A	Bromus laevipes	nd	nd	nd	nd	nd	nd	nd

^*Results from additional associations and some replicates are in SI Appendix, Table S2.

^†Metabolites that were not detected are annotated “nd” (limit of detection 0.05 µg/g). LOQ indicates detection of a metabolite at a concentration below the limit of quantitation (0.2 µg/g).

^‡As determined by sequence analysis.

^§Approximate concentration based on response factor of synthetic 1a.

^¶Approximate concentration based on response factor of synthetic 2a.

Mean production of 2a was 10-fold higher in PN2013/Ety_ppzA-2 cultures compared to PN2013/Efe_ppzA-2 cultures (P = 0.038; Fig. 3D), and plant material infected with Ety_ppzA-2-genotype endophytes also exhibited the highest 2a concentrations (Table 1 and SI Appendix, Table S3), suggesting that Ety_PpzA-2 may be a more efficient 2a synthetase. The absence of an Arg N-methyl group in the structure of 2a (Fig. 3A) suggested that the PpzA-2–specific M₂-domain substitutions may have eliminated N-methyltransferase activity (Fig. 1A and SI Appendix, Fig. S3). Targeted analysis revealed that the N-methylated PPZ-1,4-dione cyclo(Pro, meArg) (2b; Fig. 3A) was also produced by Efe_ppzA-2 strains, but not by Ety_ppzA-2 strains (Fig. 3D and SI Appendix, Supplementary Note). Synthetic allele ppzA-S1 was generated by replacing the M₂-encoding sequence of Efe_ppzA-2 with its Efe_ppzA-1 equivalent to investigate whether the production of 2a by Efe_PpzA-2 was due to reduced N-methyltransferase activity. This abolished production of 2a and massively increased production of 2b (Fig. 3D), demonstrating that N-methyltransferase activity has been reduced in Efe_PpzA-2 by substitutions within the M₂ domain. Additional lineage-specific substitutions may explain the complete loss of M₂-domain function in Ety_PpzA-2 (SI Appendix, Fig. S3).

PpzA-1 and PpzA-2 Proteins Bind Different A₁-Domain Substrates.

Structural conservation between 1a and 2a-b (Fig. 3A) suggests that the A₂ domains of PpzA-1 and PpzA-2 proteins share l-Arg as substrate, and this is supported by the absolute conservation of the substrate-defining NRPS code between A₂ domains from all PpzA proteins (Fig. 1B). The PpzA-2 A₁-domain substrate can similarly be inferred as l-Pro from the structures of 2a-b; however, the spontaneous pyrrolidine oxidation step proposed during 1a biosynthesis (2) and different NRPS codes exhibited by A₁ domains from PpzA-1 vs. PpzA-2 proteins (Fig. 1B) mean that the PpzA-1 A₁-domain substrate cannot be determined with certainty. Identification of the cognate PpzA-1 A₁-domain substrate was therefore attempted by feeding several candidate amino acids to PN2013/Efe_ppzA-1 cultures to assess their effects on 1a production. Of these, only trans-4-hydroxy-l-proline (T4HP) and cis-4-hydroxy-d-proline (C4HP) had a substantial impact, with each of these 4-hydroxy-proline (4HP) stereoisomers increasing 1a production 50-fold relative to water-fed controls (Table 2). Modeling the PpzA-1 A₁-domain structure places the NRPS code residues Thr500 and Glu521 in a position that would enable hydrogen bonding interactions with the pyrrolidine-hydroxy group of T4HP and/or C4HP (SI Appendix, Fig. S7). In contrast, Thr500 is substituted for Ser and Glu521 is substituted for Gly/Ala in PpzA-2 proteins, presumably altering the binding pocket shape and charged microenvironment to favor Pro as substrate. Additional supplementation of PN2013/Efe_ppzA-1 cultures with l-Arg did not further increase 1a production over 4HP feeding alone (Table 2), indicating that l-Arg availability was not limiting 1a production. T4HP feeding was therefore used in subsequent experiments to improve production of 4HP-derived metabolites.

Table 2.

Substrate feeding effects on PPZ production

		PPZ concentration*, µg/g
Allele	Medium†	[1a]	[1b]‡	[1c]‡	[2a]	[2b]§	[2c]§	[2d]§
Efe_ppzA-1	CD + H2O	0.3	nd	nd	nd	nd	nd	nd
Efe_ppzA-1	CD + l-Glu	LOQ	nd	nd	nd	nd	nd	nd
Efe_ppzA-1	CD + l-Pro	1.0	nd	nd	nd	2.4	nd	nd
Efe_ppzA-1	CD + P2C	0.5	nd	nd	nd	nd	nd	nd
Efe_ppzA-1	CD + C4HP	21.1	LOQ	nd	nd	nd	3.6	nd
Efe_ppzA-1	CD + T4HP	20.8	LOQ	nd	nd	nd	4.7	nd
Efe_ppzA-1	CD + T4HP, l-Arg	15.3	LOQ	nd	nd	nd	3.4	nd
Efe_ppzA-2	CD + H2O	NT	NT	NT	28.6	8.0	nd	nd
Efe_ppzA-2	CD + l-Pro	NT	NT	NT	23.0	6.9	nd	nd
Efe_ppzA-2	CD + P2C	NT	NT	NT	26.5	8.3	nd	nd
Efe_ppzA-2	CD + T4HP	NT	NT	NT	18.3	3.4	1.2	3.4
neg. ctrl	CD + l-Pro	nd	nd	nd	nd	nd	nd	nd
neg. ctrl	CD + T4HP	nd	nd	nd	nd	nd	nd	nd

^*PPZ concentrations are averaged across cultures of 3 independent P. paxilli strains transformed with Efe_ppzA-1, Efe_ppzA-2, or empty pRS426 vector (negative control). Metabolites that were not detected are annotated “nd” (limit of detection 0.05 µg/g). LOQ indicates detection of a metabolite at a concentration below the limit of quantitation (0.2 µg/g). NT, not tested.

^†Cultures were grown in 50 mL Czapek Dox liquid medium under standard conditions with feeding of 2.4 × 10⁻⁴ mol each substrate after 4 and 5 d growth.

^‡Approximate concentration based on response factor of synthetic 1a.

^§Approximate concentration based on response factor of synthetic 2a.

PN2013/Efe_ppzA-1 cultures fed with 4HP also produced the PPZ-1,4-dione cyclo(4HP, meArg) (2c; Fig. 3A and SI Appendix, Supplementary Note), which contains a hydroxy group attached to the pyrrolidine ring. Although PpzA-1 had previously only been reported to synthesize 1a (2), 2c was subsequently identified in all ppzA-1-genotype samples with sufficiently high 1a concentrations. The hydrogenated PPZ-1-one 1b (Fig. 3A and SI Appendix, Supplementary Note), which is likely an intermediate that exists in weak equilibrium with 1a, was also detected in 4HP-fed cultures (Table 2). Additionally, feeding PN2013/Efe_ppzA-1 cultures with l-Pro induced production of the PPZ-1,4-dione 2b (Table 2), indicating that the Efe_PpzA-1 A₁ domain exhibits some weak specificity toward this substrate. Feeding PN2013/Efe_ppzA-2 cultures with T4HP similarly induced production of the hydroxylated PPZ-1,4-diones 2c and cyclo(4HP, Pro) (2d; Fig. 3A and SI Appendix, Supplementary Note), indicating the Efe_PpzA-2 A₁ domain retains residual specificity toward 4HP (Table 2). Unlike production of 1a by PN2013/Efe_ppzA-1 cultures, PN2013/Efe_ppzA-2 cultures did not appear to be substrate-limited for production of 2a and 2b, as feeding l-Pro and l-Arg did not dramatically increase 2a concentrations (41.3 µg/g with l-Pro/l-Arg feeding vs. 26.5 µg/g water-fed control). Collectively, these results show that 4HP is likely the substrate of the PpzA-1 A₁ domain, and that PpzA-1 can synthesize both PPZ-1-one and PPZ-1,4-dione products.

Truncated Alleles ppzA-3 and ppzA-4 also Encode Functionally Distinct PPZ-1,4-Dione Synthetases.

As the A₁ domains of PpzA-3 proteins retain PpzA-1–like NRPS codes (SI Appendix, Fig. S1), these were predicted to bind 4HP substrate. Expression of Eba_ppzA-3 from E. baconii strain As6 in P. paxilli induced production of the hydroxylated PPZ-1,4-diones 2d and 2c (Fig. 4). Replacement of the A₁ SBR-encoding region of Efe_ppzA-2 with the equivalent sequence from Efe_ppzA-1 was used to generate ppzA-S2, which is a synthetic analog of ppzA-3. Expression of ppzA-S2 also induced production of 2c and 2d in P. paxilli (Fig. 4). Analysis of Epichloë-infected plant material also showed that 2d was exclusive to samples infected with ppzA-3-genotype E. baconii or E. bromicola strains (Table 1), although concentrations were much higher in the E. bromicola-infected material.

Fig. 4.

PPZ synthesis profiles of different PpzA proteins. Each row shows the PPZ profile of T4HP-fed P. paxilli strains expressing each of the proteins shown to the left of the image. Protein maps are annotated with domain boundaries, and synthetic hybrid proteins are colored to indicate the natural PpzA protein from which each region is derived. Concentrations are averaged across at least 3 independent transformants for each ppzA expression construct, with the SEM shown in parentheses. Where PPZ production was not observed for all strains, the number (n) of strains used to generate the listed concentration is shown instead of the SE. Metabolites that were not detected are annotated “nd” (limit of detection 0.05 µg/g). LOQ indicates detection of a metabolite at a concentration below the limit of quantitation (0.2 µg/g). NT = not tested. The 1a concentrations for a small subset of these analyses, which are indicated by an asterisk (*), were determined using slightly less sensitive method (limit of detection 0.1 µg/g). Colored bars illustrate the relative between-sample concentration for each metabolite. The concentrations listed for 1b–c are estimates based on the response factor of synthetic 1a, while the concentrations listed for 2b–d are estimates based on the response factor of synthetic 2a.

Unlike PpzA-3, the A₁-domain NRPS code of Esi_PpzA-4 is identical to that of Efe_PpzA-2 (Fig. 2 and SI Appendix, Fig. S1) and was therefore predicted to bind l-Pro substrate. However, while Esi_ppzA-4 is derived from Efe_ppzA-2, much of the M₂-domain-encoding region of Esi_ppzA-4 has been replaced with sequence from an Efe_ppzA-1 donor (Fig. 2 and SI Appendix, Fig. S3). This suggested that Esi_PpzA-4 may be a dedicated 2b synthetase, as much of the weakly functional Efe_PpzA-2 M₂-domain was replaced. This functionality was confirmed by analysis of Epichloë-infected plant material, which showed that high concentrations of 2b were exclusive to E. siegelii-infected samples (Table 1 and SI Appendix, Table S3). Furthermore, the synthetic allele ppzA-S1 can be considered an analog of Esi_ppzA-4, and PpzA-S1 was also shown to be a dedicated 2b synthetase (Fig. 3D). However, unlike PpzA-S1, Esi_PpzA-4 retains all of the conserved PpzA-2–conserved substitutions located at the N-terminal end of the M₂ domain (Fig. 2 and SI Appendix, Fig. S3). Given that the M₂ domain of Esi_PpzA-4 appears to be fully functional, this implies a causative role for one or both of the PpzA-2–conserved M₂-domain substitutions that are replaced in Esi_PpzA-4 (D1935A and P2003A; SI Appendix, Fig. S3).

PPZ-1,4-Dione Synthesis Is Proposed to Occur through Nonenzymatic Dipeptidyl-Thioester Cyclization.

In the absence of the R₂ domain, it is conceivable that release of PPZ-1,4-diones from truncated PpzA variants could be catalyzed by separate protein, such as a type II thioesterase. However, candidate genes for this function are not found coclustered with any ppzA variant (2, 16, 23). Furthermore, expression of these truncated ppzA variants alone was sufficient to achieve PPZ-1,4-dione production in the naive host P. paxilli at levels comparable to 1a production by PN2013/Efe_ppzA-1 strains (Figs. 3D and 4 and SI Appendix, Table S1). Given that transcription of all ppzA variants in P. paxilli was controlled by identical regulatory sequences, this suggests that all PpzA variants have similar biosynthetic efficiencies. This would therefore require a hypothetical endogenous P. paxilli protein that can catalyze dipeptidyl release at a rate similar to the integrated R₂ domain of PpzA-1, which seems unlikely. We therefore propose that PPZ-1,4-dione production by PpzA proteins occurs through a spontaneous nucleophilic substitution reaction where the dipeptidyl-thioester sulfur atom is replaced by the pyrrolidine nitrogen atom, releasing a product with a 2,5-diketopiperazine (DKP) core (Fig. 5). This mechanism is equivalent to the pathway proposed by Stachelhaus et al. (24) for biosynthesis of the DKP cyclo(Phe, Pro) by an artificially truncated 2-module variant of the gramicidin synthetase NRPS complex and would not require a termination domain.

Fig. 5.

Proposed PPZ biosynthetic pathways. Predicted reactions are shown after amino acid substrates have been selected, activated, and thio-tethered. Hatched shading is used to indicate the noncatalytic N-terminal partial C domain (13) and the weakly or nonfunctional PpzA-2 M₂-domain. Residual M₂-domain activity means that some PpzA-2 proteins produce both 2a and 2b. T4HP is shown as the A₁-domain substrate for PpzA-1 and PpzA-3, but feeding experiments suggest this may be replaceable with C4HP. Biosynthesis of 2d by PpzA-3 is equivalent to biosynthesis of 2a by PpzA-2, except the l-Pro substrate is replaced with 4HP. Biosynthesis of 2b by PpzA-4 is equivalent to PpzA-2-catalyzed biosynthesis of 2a, except the M₂ domain of PpzA-4 is functional. In the absence of the R₂ domain, PpzA-5 biosynthesis of 2c occurs via the competing autocatalytic dipeptidyl release pathway shown in red for PpzA-1.

Interestingly, biosynthesis of the PPZ-1,4-diones 2b and 2c was also observed for PpzA-1 proteins (Fig. 4, Table 2, and SI Appendix, Table S1), including the PpzA-1 ortholog from the insect pathogen Metarhizium rileyi, which belongs to the same family (Clavicipitaceae) as the Epichloë genus. This suggests that nonenzymatic cyclization of the dipeptidyl-thioester intermediate is not a novel innovation of the truncated PpzA proteins. Rather, nonenzymatic cyclization appears to compete with R₂-catalyzed reduction for dipeptidyl thioester release from PpzA-1, and both activities were likely present in PpzA-1 from the last common ancestor of the Epichloë and Metarhizium genera. The transposable element insertion observed in all ppzA-2, ppzA-3, and ppzA-4 alleles could therefore have eliminated the competing reductive release pathway through deletion of the R₂-encoding sequence, resulting in immediate subfunctionalization of the encoded PpzA protein into a dedicated PPZ-1,4-dione synthetase.

To investigate this hypothesis, synthetic allele ppzA-S3 was generated by truncating Efe_ppzA-1 to the same length as Efe_ppzA-2. This R₂-domain deletion abolished production of 1a but did not affect production of 2c (Fig. 4). Exchanging the T₂-encoding sequence of ppzA-S3 for its Efe_ppzA-2 equivalent did not further improve 2c production (PpzA-S4; Fig. 4), nor did exchanging the Efe_ppzA-2 T₂-encoding sequence for its Efe_ppzA-1 equivalent reduce production of 2a (PpzA-S5; Fig. 3D). These results demonstrate that 2c release from PpzA proteins occurs through an R₂-independent mechanism, with R₂-domain truncation resulting in immediate subfunctionalization of PpzA-1 into a dedicated 2c synthetase. Furthermore, although PpzA-2–conserved substitutions were identified within the T₂ domain (Fig. 1 and SI Appendix, Fig. S4), these impart no specific contribution to the efficiency of DKP formation. It was also observed that the artificially truncated ppzA-S3 allele is similar to 1 of the 2 ppzA alleles present in Epichloë uncinata strain e167, which is an asexual allopolyploid hybrid species of E. bromicola and E. typhina subsp. poae (22, 25). Both E. uncinata alleles were previously thought to be pseudogenes (14); however, the premature stop codon in the E. typhina-derived allele only affects translation of the R₂ domain. We therefore predicted this “Eun_ppzA-5” allele may encode a 2c synthetase analogous to PpzA-S3. Analysis of Epichloë-infected plant material confirmed that E. uncinata-infected material contains 2c, and this was the only plant material tested that contained 2c in the absence of 1a (Table 1). These results support the hypothesis that Eun_PpzA-5 is a dedicated 2c synthetase, meaning that subfunctionalization of PpzA-1 into a dedicated PPZ-1,4-dione synthetase appears to have occurred at least twice in the evolutionary history of Epichloë.

The rate of nonenzymatic dipeptidyl-thioester cyclization would depend on the strength of the attacking nitrogen nucleophile, and the pyrrolidine hydroxy group could reduce this nucleophilicity in 4HP relative to Pro. To investigate if nonenzymatic cyclization of Pro-meArg thioesters is more efficient than for 4HP-meArg thioesters, synthetic allele ppzA-S6 was generated by PCR-induced mutagenesis of Efe_ppzA-1 to replace 3 A₁ SBR NRPS code residues of PpzA-1 with their Pro-binding PpzA-2 equivalents (Fig. 1B and SI Appendix, Fig. S1). This significantly reduced production of 1a by PpzA-S6 relative to Efe_PpzA-1, and PN2013/ppzA-S6 cultures instead produced substantial quantities of 2b (Fig. 4). Levels of 1a in PN2013/ppzA-S6 cultures were also still strongly inducible through T4HP feeding (SI Appendix, Table S1), indicating that incomplete A₁-domain binding specificity conversion rather than oxidation of a Pro-incorporating intermediate was the source of this residual 1a biosynthesis. Furthermore, no metabolite corresponding to a hypothetical Pro-derived reduced analog of 1a (C₁₂H₂₁N₅O, 252.18 m/z) was detected in any PN2013/ppzA-S6 samples. These results suggest that nonenzymatic cyclization of Pro-incorporating dipeptidyl thioesters occurs much more rapidly than R₂-catalyzed reductive release from PpzA proteins. In contrast, nonenzymatic release of 4HP-incorporating dipeptidyl thioesters appears to be slower, allowing R₂-catalyzed reductive release to dominate. However, an alternative hypothesis where Pro-incorporating dipeptidyl intermediates are incompetent as R₂-domain substrates cannot be excluded.

Attempts to further reduce 4HP binding specificity were made by exchanging the A₁ SBR-encoding region of Efe_ppzA-1 for that of Efe_ppzA-2 or Ety_ppzA-2 to generate synthetic alleles ppzA-S7 and ppzA-S8, respectively. However, concentrations of both 1a and 2b were significantly reduced in PN2013/ppzA-S7 cultures, while PN2013/ppzA-S8 cultures did not produce detectable levels of any product (Fig. 4), indicating these subdomain exchanges were detrimental to protein function. This suggests that targeted mutagenesis of NRPS code residues may be a superior strategy to subdomain swaps when attempting to modify binding specificity toward an amino acid that is closely related to the cognate substrate for that A domain.

Restoring 1a Biosynthesis to Efe_PpzA-2 Defines All Functionally Divergent Regions.

We have shown that the Efe_PpzA-2 A₁ SBR and M₂ domains are functionally divergent compared to PpzA-1 proteins, and that the R₂ domain is required for 1a biosynthesis. However, it was not clear if all functionally divergent regions within Efe_PpzA-2 had been identified. To investigate this, synthetic hybrid alleles ppzA-S9 through ppzA-S12 were generated by iterative replacement of the divergent Efe_ppzA-2 A₁ SBR, C₂-, M₂-, and R₂-encoding regions with their Efe_ppzA-1 equivalents. These alleles were expressed in P. paxilli PN2013 and assayed for 1a production. Replacement of the A₁ SBR and R₂-encoding sequences alone in PpzA-S9 did not result in synthesis of 1a or 2c, although small amounts of the nonmethylated analogs 1c and 2d were detected in some strains (Fig. 4 and SI Appendix, Supplementary Note). Equivalent results were observed when the divergent C₂-encoding region was additionally replaced in PpzA-S10 (Fig. 4). Replacement of the A₁ SBR, M₂-, and R₂-encoding sequences in PpzA-S11 resulted in low but consistent production of 1a (Fig. 4), and the additional replacement of the divergent portion of the C₂-encoding region in PpzA-S12 fully restored 1a production to the same level as wild-type Efe_PpzA-1 (Fig. 4). This shows that most Efe_PpzA-2–derived regions in PpzA-S12 (55%) are functionally equivalent to their Efe_PpzA-1 counterparts, although the C₂ domain may have functionally diverged. However, no PpzA-2–conserved substitutions are present within the C₂ domain (Fig. 1A), and modeling the Efe_PpzA-1 and Efe_PpzA-2 C₂-domain structures did not reveal any clustering of differential residues (SI Appendix, Fig. S8). The cause of this apparent differentiation between Efe_PpzA-1 and Efe_PpzA-2 C₂-domains therefore remains unclear.

Discussion

Fungi typically contain a large number of secondary metabolism gene clusters, each of which encodes proteins involved in biosynthesis of a specific class of natural products (26). Within-class natural product diversity is usually a function of gene gain or loss polymorphisms within these clusters (27), as has been well documented for Epichloë species (16, 28). It is understandable that the PPZ product diversity encoded by the orphaned ppzA gene in Epichloë spp. was thought to be restricted to 1a, especially given that PPZ clusters containing ppzA-1 orthologs alongside at least 6 accessory genes were recently identified in Metarhizium and Cladonia spp. (23). However, the results presented here show that allelic variants of a single NRPS can generate metabolite diversity comparable to many gene clusters, with 4 functionally distinct ppzA alleles described here in addition to the previously characterized ppzA-1 (2, 23). Of these, the ppzA-2, ppzA-3, and ppzA-4 alleles all derive from the same 3′ sequence deletion caused by a transposable element insertion (14, 15, 18), whereas ppzA-5 contains a premature stop codon near the 3′ end of the gene (14). In the absence of the C-terminal R₂ domain, these truncated PpzA proteins are proposed to utilize nonenzymatic dipeptidyl-thioester cyclization to release a range of DKP-containing PPZ-1,4-dione products. Functional divergence of some PpzA-2 domains relative to their PpzA-1 progenitors is also demonstrated; PpzA-2 A₁ domains have evolved to bind l-Pro as substrate instead of 4HP, and PpzA-2 proteins produce nonmethylated products due to their weakly active or inactive M₂ domains. Recombination between ppzA alleles is also shown to have further enhanced the PPZ diversity encoded by this locus.

Fungi are known to produce a huge variety of DKP metabolites (29), and previous studies have described a number of fungal NRPS pathways dedicated to DKP biosynthesis (1, 30–35). While the mechanism of DKP formation in many of these NRPS pathways has not yet been resolved, this could conceivably occur through nonenzymatic cyclization of the dipeptidyl thioester. However, spontaneous trans → cis isomerization of the dipeptidyl peptide bond would be required before cyclization could occur, which is usually suppressed by steric repulsions between consecutive C^α-linked side chains (36). Gao et al. (37) demonstrated that fungal NRPSs can utilize a C-terminal condensation-like “C_T” domain to catalyze peptidyl cyclization, and all of the dedicated DKP-synthesizing fungal NRPSs described to date terminate with a C-like domain that could conceivably catalyze DKP release (1, 30–35). Indeed, Baccile et al. (38) recently demonstrated that the 2-module NRPS GliP requires a C-terminal C_T–T₃ didomain to catalyze synthesis of the DKP cyclo(Phe, Ser), which is the first step in gliotoxin biosynthesis, and proposed that this is a general mechanism for DKP biosynthesis by fungal NRPSs. Nonenzymatic dipeptidyl cyclization has been observed for some larger NRPS systems when peptidyl chain elongation is stalled by downstream substrate starvation, incorrect substrate loading, or artificial protein truncation (24, 39, 40). Formation of these DKP side-products typically involves Xaa-Pro (Xaa = any amino acid) or N-methylated “tertiary” dipeptidyl intermediates, as the influence of a third N-linked carbon atom in these tertiary peptide bonds significantly reduces the energy differential between the cis and trans isomers. However, spontaneous trans → cis isomerization of tertiary dipeptidyl peptide bonds still appears to be considerably slower than other NRPS-catalyzed reactions (41). Biosynthesis of the PPZ-1,4-diones by PpzA proteins therefore presents an enigma, as 2a and 2d biosynthesis does not occur via a tertiary dipeptidyl intermediate. These biosynthetic pathways should therefore be severely bottlenecked by trans → cis isomerization, yet our results suggest that the efficiency of PpzA-2-catalyzed 2a biosynthesis is comparable to that of PpzA-1–catalyzed 1a biosynthesis. This suggests that PpzA proteins are able to predispose the dipeptidyl thioester toward the cis conformation, for example by utilizing a C₂ domain that specifically catalyzes cis peptide bond formation. While such a mechanism would obviously benefit PPZ-1,4-dione biosynthesis by removing the peptide isomerization bottleneck, it could also ensure that reductively released dipeptide-aldehyde intermediates are rapidly cyclized during 1a biosynthesis, inhibiting potentially harmful interactions with external nucleophiles. Nonenzymatic dipeptidyl cyclization is likely also facilitated by the extraordinarily high nucleophilicity of the Pro-derived nitrogen atom in Pro–Arg dipeptidyls (42), which may be reduced by the 4-hydroxy group of 4HP. This could explain why 4HP–meArg dipeptidyl intermediates are primarily—but not exclusively—released via R₂-catalyzed reduction during 1a biosynthesis, whereas nonenzymatic cyclization completely dominated release of Pro-containing dipeptidyl intermediates even when a functional R₂ domain was present.

Our results show that very few mutations would have been required during the transition of PpzA-1 to PpzA-2. Interestingly, the A₁-domain Pro binding, nonmethylated product synthesis, and nonenzymatic dipeptidyl cyclization activities that characterize PpzA-2 proteins were also observed to be reactions that are weakly catalyzed by PpzA-1 proteins. PpzA-2 proteins therefore appear to have evolved through a series of domain subfunctionalizations that cumulatively resulted in an NRPS that efficiently synthesizes 2a, with our results suggesting this evolutionary cascade could have been triggered by the transposon-mediated R₂-encoding sequence deletion observed in all ppzA-2 alleles (14, 15, 18). We also showed that further PPZ diversity was generated in the E. festucae and E. bromicola lineages by recombination-mediated shuffling of functionally divergent sequences between ppzA-1 and ppzA-2 alleles. This generated the ppzA-3 and ppzA-4 alleles, each of which encodes a protein with novel biosynthetic activity. Due to their modular nature, recombination-vectored domain exchange has long been proposed as a dominant mechanism driving NRPS evolution and diversity, although relatively few examples have been described (43–45). The ppzA loci of Epichloë spp. represent a remarkably clean example of this process, as even the most divergent alleles share ≥95% DNA sequence identity. Interestingly, Ebr_ppzA-2 and Efe_ppzA-2 were also shown to have undergone extensive recombination, yet these alleles remain functionally equivalent to their nonrecombinant Ety_ppzA-2 counterparts. This suggests that the E. bromicola and E. festucae lineages may have inherited a nonfunctional ppzA-2 pseudogene that was repaired through recombination with a functional ppzA-1 donor. Surprisingly, the equivalent recombination patterns exhibited by Ebr_ppzA-2/Efe_ppzA-2 and Ebr_ppzA-3/Efe_ppzA-3 alleles appear to have arisen after divergence of the E. bromicola and E. festucae lineages in an example of convergent evolution.

Interestingly, the ppzA-2–derived regions of the recombinant ppzA-2, ppzA-3, and ppzA-4 alleles exhibit TSP, meaning that the donor ppzA-2 alleles diverged from ppzA-1 before emergence of the E. bromicola, E. festucae, and E. typhina lineages. Each of these species also has conspecific strains with functionally distinct ppzA alleles, suggesting that the ppzA locus is subject to negative frequency-dependent or “balancing” selective pressures that reduce the fitness of the most common allele. The major histocompatibility complex (MHC) in mammals is a well-characterized example of TSP, with balancing selection driven by pathogenic coevolution thought to have prevented fixation of any one MHC allele (46, 47). Analogous balancing selective pressures on the ppzA locus could be driven by coevolution of resistance against the dominant PPZ chemotype in target insects, or by displacement of susceptible insect species by resistant competitors. For example, previous experiments have demonstrated that although 1a exhibits potent feeding deterrent activity against L. bonariensis it is not active against several other species (12, 48, 49), which could potentially be susceptible to other PPZs. The N-methyl group is known to be essential for the bioactivity of 1a (12), suggesting that while N-methylated PPZ-1,4-diones 2b and 2c may exhibit activities similar to 1a, the nonmethylated PPZ-1,4-diones 2a and 2d likely do not. This is supported by the results of Rowan et al. (12), who coincidentally tested 2a due to its structural analogy with 1a and demonstrated that 2a does not deter feeding by L. bonariensis. However, 2a has previously been shown to be a potent inhibitor of chitinase activity (50, 51) and might thus interfere with insect ecdysis or suppress fungal competitors instead. Future studies into 2a–d would be useful to determine if these PPZ-1,4-diones exhibit agriculturally relevant bioactivities.

The results presented here demonstrate that nonenzymatic dipeptidyl thioester cyclization can be utilized by NRPSs in vivo for the efficient synthesis of DKP-containing metabolites, meaning that this mechanism should be considered when encountering NRPSs without an obvious termination domain. Further studies of PpzA proteins could reveal if these enzymes exhibit features that facilitate this process, such as the proposed cis-peptide-forming C domain. Although reports of allelic neofunctionalization and TSP in fungi are rare, we demonstrate that these processes can generate considerable diversity and suggest that both processes may be more prevalent than currently realized. These processes may be particularly relevant at allelic loci where gene conversion is inhibited by the insertion of large homology-breaking DNA elements such as transposons. TSP and lateral gene transfer can also produce similar phylogenetic patterns, particularly between closely related species, meaning that both hypotheses need to be considered carefully.

Materials and Methods

A comprehensive description of the materials and methods used in this study is available in SI Appendix. Homologous recombination was used to generate E. festucae ∆ppzA-2 strains. Constructs for ppzA expression in P. paxilli were assembled with all ppzA variants placed under transcriptional control of the same regulatory sequences from the native P. paxilli secondary metabolism gene paxM. These constructs were introduced into the P. paxilli genome via nontargeted integration, with RT-PCR used to select at least 3 independent transformants expressing each ppzA variant. Cultures for metabolite analysis were grown for 6 d under standardized conditions, with substrate feeding performed at 4 and 5 d postinoculation. PPZ metabolites were extracted from lyophilized P. paxilli mycelia or Epichloë-infected plant material and were analyzed using hydrophilic-interaction chromatography-coupled positive electrospray ionization mass spectroscopy (LCMS).

Data Availability.

DNA sequence data generated during this study have been deposited in the GenBank database under accession numbers MN605951 to MN605962. Raw PPZ concentrations from all LCMS analyses, MSⁿ spectra for all PPZ metabolites, and NMR spectra for 2a are available in SI Appendix.