Structural Basis for a Dual Function ATP Grasp Ligase That Installs Single and Bicyclic ω-Ester Macrocycles in a New Multicore RiPP Natural Product

Abstract

Among the ribosomally synthesized and post-translationally modified peptide (RiPP) natural products, “graspetides” (formerly known as microviridins) contain macrocyclic esters and amides that are formed by ATP-grasp ligase tailoring enzymes using the side chains of Asp/Glu as acceptors and Thr/Ser/Lys as donors. Graspetides exhibit diverse patterns of macrocylization and connectivities exemplified by microviridins, that have a caged tricyclic core, and thuringin and plesiocin that feature a “hairpin topology” with cross-strand ω-ester bonds. Here, we characterize chryseoviridin, a new type of multicore RiPP encoded by Chryseobacterium gregarium DS19109 (Phylum Bacteroidetes) and solve a 2.44 Å resolution crystal structure of a quaternary complex consisting of the ATP-grasp ligase CdnC bound to ADP, a conserved leader peptide and a peptide substrate. HRMS/MS analyses show that chryseoviridin contains four consecutive five- or six-residue macrocycles ending with a microviridin-like core. The crystal structure captures respective subunits of the CdnC homodimer in the apo or substrate-bound state revealing a large conformational change in the B-domain upon substrate binding. A docked model of ATP places the γ-phosphate group within 2.8 Å of the Asp acceptor residue. The orientation of the bound substrate is consistent with a model in which macrocyclization occurs in the N- to C-terminal direction for core peptides containing multiple Thr/Ser-to-Asp macrocycles. Using systematically varied sequences, we validate this model and identify two- or three-amino acid templating elements that flank the macrolactone and are required for enzyme activity in vitro. This work reveals the structural basis for ω-ester bond formation in RiPP biosynthesis.

Graphical Abstract

INTRODUCTION

Natural products exhibit enormous chemical and functional diversity and have been an invaluable resource in the discovery and development of therapeutics.^1–3 The genes encoding natural product biosynthetic enzymes are colocalized within dedicated biosynthetic gene clusters (BGCs), allowing for detection of BGCs and prediction of natural product structures through genome sequencing and cluster analysis.^4–6 This genes-to-molecules paradigm fueled the prospect of genome mining as an approach to natural products discovery.^7–11 Among the major classes of natural products, the ribosomally synthesized and post-translationally modified peptides (RiPPs) have received considerable attention in genome mining efforts. RiPPs are a ubiquitous group of structurally diverse natural products, examples of which include cyanobactins, lantibiotics, lasso peptides, and microviridins. Their respective BGCs are relatively simple and encode a precursor peptide that consists of a conserved leader and more variable core region, tailoring enzymes that post-translationally modify the core often converting linear peptides into complex structures, and proteases and/or transporters responsible for the cleavage and transport of the final product.¹² The compact architecture of RiPP BGCs, the presence of highly conserved leader motifs for each class of RiPP, and the recent findings that core regions can be modified by diverse enzyme classes not previously considered in RiPP biosynthesis have facilitated innovative genome mining efforts that are leading to the discovery of new natural products.^13–22

Among this class, microviridins are tricyclic peptides with the caged structure shown in Figure 1a.^23,24 Originally discovered in cyanobacteria using traditional isolation methods, microviridins are known for being potent inhibitors of serine proteases and playing a major role in the ecology of aquatic systems.^25–27 The microviridin BGCs contain genes for a ribosomally encoded precursor peptide mdnA, tailoring enzymes mdnB and mdnC, and the acetylase mdnD (Figure 1b). The MdnA precursor peptide contains a conserved C-terminal core with the sequence TxKYPSDx(E/D)(D/E). MdnC and MdnB (in Microcystis sp.) are ATP-grasp proteins that insert two ω-ester bonds between the side chains of Thr/Ser and Asp/Glu residues and one ω-lactam bond between the Lys and C-terminal Glu or Asp residues, respectively.^23,24 Acetylation of the N-terminus is catalyzed by MdnD to produce the mature microviridin although the mechanism of cleavage of the core from the full length MdnA has not been determined.

Figure 1.

Overview of microviridins. (A) Chemical structure of tricyclic RiPP microviridin-B, with ω-esters shown in red, the ω-lactam in blue, and the N-terminal acetyl unit in brown. (B) Biosynthetic gene clusters of microviridin-containing RiPPs in two cyanobacteria and a Chryseobacterium species. The five conserved genes (A-E) are colored as in the legend, and unidentified or uncharacterized genes are shown in white. (C) Amino acid sequence alignments of the precursor peptides MdnA in Microcystis aeruginosa NIES298 (BAG02233.1) and Anabaena sp. PCC7120 (WP_010999573.1) and CdnA3 in C. gregarium DSM19109 (WP_027385893.1). The conserved leader motif is shown in purple and microviridin or microviridin-like sequences are underlined. In C. gregarium CdnA3, the Ser/Thr and Asp residues are shown in blue and red, respectively. Potential Gly-Gly cleavage sites in the cores are italicized. NCBI accession numbers and Pfam identiers are listed in Table S2 and a sequence alignment of CdnA1–4 appears in Figure S1.

ATP-grasp proteins are generally characterized as ligating a variety of nucleophilic and phosphoryl-activated carboxylic acid substrates.²⁸ In RiPP biosynthesis, ATP-grasp ligases are used in N-terminal to C-terminal cyclization in head-to-tail bacterial peptides,¹² for ligation of an α-guanidino amino acid to a ribosomally synthesized peptide,²⁹ and to introduce intrachain macrocycles in the form of lactones and lactams in the graspetides, which include the microviridins.³⁰ ATP binds between the C-terminal and central domains, as exemplified by RimK, a ligase involved in poly alpha-glutamic acid synthesis,³¹ while the binding site of the amino acid or peptide substrates is thought to be catalyzed at the interface of the N-terminal and central domains.³² Bruner and co-workers previously determined the structures of the microviridin ligases MdnC and MdnB bound to their leader peptides; however, a cocrystal structure of a RiPP-associated ATP-grasp protein in complex with ATP and a core peptide susbtrate has yet to be determined.

Recent genome mining studies greatly expanded the diversity of the graspetide family and the types of macrocycles and side-chain connectivities seen in the mature peptides.²¹ Furthermore, bioinformatics analysis revealed numerous microviridin core sequences encoded in the genomes of phylogenetically diverse bacteria from the phyla Cyanobacteria, Proteobacteria, and Bacteroidetes, a much broader distribution than previously realized.³³ In particular, genes that encode unusually long precursor peptides (ca. 100 amino acids in length) distinguished by the presence of nonconserved and repetitive sequences, enriched in Thr, Ser, and Asp residues, followed by a conserved microviridin core sequence have been identified (Figure 1c). Despite the presence of extended core regions, additional ATP-grasp ligases were not detected in the corresponding BGCs suggesting plasticity in substrate recognition by MdnC homologues.

Given the importance of ATP-grasp ligases in natural product biosynthesis and the value that structural information on enzyme–substrate interactions would provide, we investigated a microviridin-containing RiPP from Chryseobacterium gregarium DSM 19109. Here, we characterize chryseoviridin, a new multicore graspetide, and solve the crystal structure (2.44 Å resolution) of a quaternary complex comprising the ω-ester-forming ATP-grasp ligase CdnC bound to ADP, a conserved leader peptide and a core peptide substrate. The X -ray structure and orientation of the peptide-bound complex is consistent with a model in which ω-ester bond formation occurs in an N-terminal to C-terminal direction for core peptides containing multiple Thr/Ser-to-Asp macrocycles. We provide experimental evidence supporting this path through enzymatic transformations of select peptides with systematically varied sequences. Furthermore, we identify templating elements comprising two or three amino acids N-terminal and C-terminal to the lactone ring-forming residues that are required for the enzyme recognition or activity in vitro. We were unable to produce the putative ω-lactam-forming ligase CdnB; nevertheless, this work reveals the structural basis for ω-ester bond formation in RiPP biosynthesis.

C. GREGARIUM CdnC CATALYZES CdnA CYCLIZATION AND GENERATES MULTICYCLIC RIPPS

We were interested in the RiPP BGCs detected in several strains of Chryseobacterium spp. because they possessed extended precursor peptides (ca. 90–100 amino acids) containing Thr-, Ser-, and Asp-rich sequences terminating with a microviridin-homologue core region (Figure 1). The proteins encoded by the B and C genes from these BGCs have 35% and 38% amino acid sequence identities to the prototype microviridin tailoring enzymes MdnB and MdnC. The number and arrangement of Ser, Thr, and Asp residues in the core region precluded prediction of the locations and connectivities of the macrolactones. ω-Ester bond formation could yield individual macrocycles, folded macrocycles having a “hairpin topology”,²¹ or those with the microviridin structure. Because a product of a multicore graspetide has not been identified, characterization of the mature peptide or peptides could lead to the identification of a new natural product family and provide insights into the specificity and activities of the putative ATP-grasp proteins. We first sought to detect by LCMS a predicted core region from aqueous and organic extracts of two Bacteroides and two Proteobacteria containing multicore A genes grown in a variety of media (Table S1). The extract from C. gregarium DSM 19109 displayed a m/z of 6323.7 (Figure S2). This corresponds to the predicted core encoded by cdnA3 with loss of four water molecules and proteolysis occurring after the glycine doublet (residues 27–28) as observed for other precursor peptides.^34,35 After scaling up fermentations, we were unable to detect this or similarly predicted core peptide ions from C. gregarium. We thus opted to produce and characterize CdnA3 and the proposed ATP-grasp tailoring enzyme CdnC using recombinant proteins and in vitro enzyme assays.

To determine whether CdnC functions as an ATP-grasp ligase, we first expressed and purified full-length His₆-tagged proteins CdnA3 and CdnC in Escherichia coli (Tables S2 and S3). Following a two-step chromatographic separation, we analyzed CdnC in the apo, ADP-bound, and ADP/CdnA3 precursor-bound states by size exclusion chromatography and found that CdnC is a homodimer (Figure 2a and Figure S3). Incubation of precursor CdnA3 with CdnC and ATP yielded a peptide with the mass of 11611.297 Da, indicating loss of six water molecules (Δ6) compared to untreated CdnA with an observed mass of 11719.619 Da (Figure 2b,c, Table S4). This suggested that, in addition to the proposed cyclization of the microviridin core, four additional macrocycles were formed, yielding six macrocycles in total. To locate these sites, we attempted to proteolyze the multicyclic product but never observed smaller fragments with a strong MS/MS signal. Thus, we instead analyzed the core sequence of CdnA3 and identified three regions that could represent independent subdomains based on the following reasoning. First, as observed in a number of RiPP BGCs, the CdnA3 precursor peptide can be expected to undergo proteolysis by an as yet unidentified protease that cleaves C-terminal of the sequence Gly-Gly which appears twice in CdnA3 (Figure 3a, GG sequences appear in italics). Second, the macrocyclic region of microviridins begins with the sequence Thr-Leu-Lys, which is present in two locations in CdnA3 (Figure 3a).^23,24,36 This led us to construct three truncated core peptide variants with TEV protease cleavage sites inserted at the beginning of each predicted core region. We incubated each of the three truncated constructs CdnA3 (1), (1–2) and (1–3) with CdnC followed by cleavage with TEV protease to release peptides 1–3 (P1, P2, and P3), and analyzed each mixture by UPLC-HRMS/MS. Molecular ions corresponding to loss of one and two water molecules were observed for each construct, and MS/MS sequencing located the positions of two individual Thr-to-Asp macrocycles in P1 and P2, and a Thr-to-Asp and Ser-to-Asp pair in P3 (Figure 3b–g and Tables S5–S10). Treatment of each variant for longer times or with higher amounts of CdnC yielded the same molecular ions and fragmentation suggesting that additional macrocycles are not formed. MS/MS sequencing of the Δ1 and Δ2 ions of peptide 3 showed that it has the expected bicyclic microviridin core structure containing the overlapping 7- and 4-membered lactone rings between Thr2 and Asp8 and Ser7 and Asp10. Thus, each of the core subregions in our constructs contains two ω-esters. Notably, each of the first four macrocycles (Figure 3h) is separated by two amino acids, and despite the presence of 12 Thr and Ser residues in this part of the core, only four are used to form the ω-esters. This finding implies that CdnC encompasses recognition elements to install multiple macrocycles in a sequence-specific or topologically distinct manner.³⁷ These structural features present a new arrangement in multicyclic and multicore graspetides illustrated in Figure 3h. We refer to the fully cyclized and full length core region of CdnA3 as chryseoviridin, and note that this product lacks tailoring by CdnB.

Figure 2.

CdnC binds and post-translationally modifies precursor peptide CdnA3. (A) Size exclusion chromatography shows that CdnC free (red) and in the presence of ADP (cyan) elutes as a homodimer and forms a stable complex when bound to precursor peptide CdnA3 (blue). Calibration curve and SDS PAGE gels are shown in Figure S2. (B, C) CdnC post-translationally modifies native CdnA3 resulting in a loss of six water molecules detected by LC-HRMS. TIC in part B shows coinjection of CdnA3 and the Δ6 product, and the deconvoluted mass of each is shown in part C.

Figure 3.

Structure determination and ω-ester bond locations in chryseoviridin A3 by LC-HRMS/MS. (A) CdnA3 constructs produced to probe Thr/Ser to Asp ω-ester bond formation. Constructs contain 1, 2, or 3 core regions with the homologous microviridin core colored blue. TEV protease cleavage sites and introduced stop codons are shown with black arrows and asterisks, respectively. Core region peptides released after TEV cleavage are labeled as P1, P2, and P3. An additional Gly residue remains at the N-terminus of each peptide post cleavage. The hydrophobic C-terminus does not contain nucleophilic amino acids. (B–G) Each of the three constructs in part A were treated with CdnC (37 °C, 10 min) followed by cleavage with TEV protease (30 °C, 1 h) to release each of the three core region peptides 1–3 for HRMS/MS analysis. The molecular ions with loss of one or two water molecules are labeled P1Δ1, P2Δ1, and P3Δ1 (left panels) and P1Δ2, P2Δ2, and P3Δ2 (right panels), respectively. The TEV site introduces an N-terminal Gly in peptides 2–3. Fragment ions were identified by automated sequencing using BioConfirm software (Agilent Technologies) and manual assignments. Locations of ω-esters are indicated with horizontal brackets; fragment ions within each macrocyclic lactone are generally not observed. Ions labeled with Cyc refer to the macrocyle fragment. (H) Schematic showing chryseoviridin macrocycles with ester bonds in red and nucleophilic amino acids in green. The acyclic C-terminal region is shown as a black tail.

OVERALL STRUCTURE OF CDNC

Having characterized a multicyclic chryseoviridin product, we next sought to determine the structural basis for substrate recognition and ω-ester bond formation; in particular, we were interested in solving a cocrystal structure of CdnC bound to the precursor peptide or a core peptide substrate. We carried out crystallization trials on numerous discrete samples including apo CdnC, CdnC in the presence of ADP, and CdnC:ADP in the presence of either full length precursor peptide CdnA3 or synthetic peptides corresponding to leader and various core regions. We eventually obtained small, plate-shaped crystals (0.4 × 0.3 × 0.05 mm) of a complex comprising CdnC, ADP, a conserved leader peptide KEPFFAAFLEKQ, and core peptide 3, TLKYPSDSDEG. The crystals diffracted to 2.44 Å at the synchrotron, and the structure was solved and refined at this resolution (Table S11).

Consistent with our biochemical studies, CdnC is a homodimer in the crystal (Figure 4a). Each monomer displays a conserved ATP-grasp fold comprising domains A–C, also referred to as the N-terminal domain, the central domain, and the C-terminal domain, respectively.²⁸ In CdnC, the three structural domains consist of residues 1–114, 141–211 and 212–333. A span of two helices, α4 and α5, formed by residues 115–140, forms an additional C-2 domain that packs against the surface of the C-domain and occurs only in a handful of other ATP-grasp enzymes (examples include PurD, PurK, PurT, and PurP). Although ATP-grasp proteins share similar overall topologies, they show limited sequence conservation and accommodate a variety of secondary structural elements. A structural similarity search using the DALI server³⁸ identified MdnC from Microcystis aeruginosa NIES298 as the most similar to CdnC with 315 of the 333 observed residues structurally aligned (rmsd = 2.6, PDB ID: 5IG9, Z = 32.9, 39% sequence identity).³⁹ The structure of RimK from E. coli is the second best match (PDB: 4IWX, Z = 20.3, RMSD = 3.7) with more than 265 aligned residues and 15% sequence identity,³¹ followed by Thermus thermophiles LysX (PDB: 3VPD, Z = 19.7, RMSD = 3.1) with ca. 250 aligned residues and 19% sequence identity.⁴⁰

Figure 4.

X-ray crystallography reveals distinct CdnC conformations in the free and core peptide-bound subunits. (A) Overall structure of the CdnC homodimer in the asymmetric unit, with molecule A colored blue and molecule B gold. The leader peptide is shown as a red ribbon and the bound core peptide and ADP as gray and orange sticks with N, O, and P atoms colored blue, red, and orange, respectively. (B) Molecule A bound to ADP and a leader peptide. The A-, B-, and C-domains of the ATP-grasp fold are labeled. (C) View of molecule B bound to ADP, leader peptide, and core peptide 3 (P3) that is homologous to microviridin cores.

In the CdnC homodimer, two different monomeric conformations are observed, giving rise to an asymmetric homodimer. The two conformations result from capturing the substrate bound in one monomer of the homodimer; in molecule A, CdnC is bound to ADP and the leader peptide (Figure 4b), whereas in molecule B, CdnC is bound to ADP, the leader peptide, and core peptide 3 (Figure 4c). In each protomer, we observed electron density for 333 of the 340 residues; ADP, the conserved leader peptide, and peptide 3 were all successfully built into the electron density map (Figure S4). The CdnC homodimer harbors an extensive interface (3177 Ų buried surface area) created by interactions between β3 of the N-domain and β8’ of the central domain which forms an extended β-sheet. These two strands are structurally conserved in all ATP-grasp enzyme family members.²⁸ In addition, the twenty-one residue α3 helix extends along the neighboring central domain interacting with the C2-subdomain on one end and β10 at the other, contributing to the stabilization of the dimer. Compared with other reported ATP-grasp ligases (e.g., ecRimk), the long α3 helix in CdnC is atypical, as other members contain a shorter helix at this position. This extended helix is also present in the ATP-grasp ligase MdnC from M. aeruginosa.³⁹

ADP AND LEADER PEPTIDE BINDING

Among ATP-grasp proteins, the B-domain is connected to the stable core through flexible loops of varying length and is positioned distal from the protein core. Upon ATP binding, the B-domain folds over the central domain to form the ATP binding site. Due to the lability of ATP, CdnC was crystallized with ADP, a leader peptide, and a core peptide. ADP and the leader peptide are bound in both protomers, with ADP positioned in the cleft between the B- and C-domains. Key, conserved interactions between residues in the B-domain and ADP are observed in both monomers and include 12 hydrogen bonds between the side chains of Lys 167, Thr 185, Gln 207, Lys 208, Lys 212, and Glu 215 and ADP (Figure 5a). In molecule B, which contains the bound substrate, the α8 helix and loop region fold inward toward the nucleotide, the C-domain, and the substrate (see below). This conformational change extends the ADP binding site and leads to additional interactions dominated by hydrogen bonds between the side chains of Glu 294 and Arg 243 with O1, O2, and O3 of the β phosphate group of ADP (Figure 5a,b). Seven of these eight residues are conserved between CdnC and MdnC; only Lys 208 is replaced by an acidic Glu or uncharged Ala residue in the cyanobacterial proteins (CdnC:ADP interactions are summarized in Table S12).

Figure 5.

Comparison of CdnC:nucleotide and CdnC:leader peptide interactions in free (left) versus substrate-bound (right) conformations. Molecules A and B are colored light blue and gold, respectively, and peptide substrate is shown in white sticks. In the core peptide-bound conformation of molecule B, the ADP binding site is extended as the B-domain moves closer to the C-domain. Additional hydrogen-bonds to ADP are formed with Glu294 and Arg243. (C, D) Similarly, changes in surrounding amino acids create an extensive hydrogen bonding network to main chain and side chain N and O atoms of the leader peptide, especially near the C-terminus. ADP and the leader peptide are rendered as sticks, and hydrogen bonds are shown with black dashed line.

LEADER PEPTIDE INTERACTIONS

Crystallization screening with full length CdnA3 or full length leader peptide (residues 1–28) and CdnC did not yield crystals. Given that Bruner and co-workers showed that a conserved region of the leader peptide in MdnA could act in trans to activate MdnC-catalyzed macrocyclization,³⁹ we used a synthetic peptide having the partial CdnA3 sequence KEPFFAAFLEKQ, which includes the DSM19109 leader motif PFFAAFL, for crystallization. In the crystal, the leader peptide is bound in each monomer and is lodged deep within the groove between α7 and the β9/β10 hairpin of the B-domain (Figure 5c,d). The conserved motif FFAAF (residues 4–8) forms a short 3₁₀-helix with all intrachain i to i+3 backbone hydrogen bonds observed in this region (CO to NH hydrogen bonds between Pro 3/Ala 6, Phe 4/Ala 7, Phe 5/Phe 8 and Ala 6/Leu9; leader residues rendered in italics). The helix immediately meets a short β-strand formed by residues 9–11 in both complexes. Hydrophobic residues line the binding pocket (Table S13). At its N-terminus, the ε amino group and main chain NH and O atoms of Lys 1 are hydrogen bonded to the carboxylate and O atom of Glu 87 and Asn 89. At the C-terminus, the short β-strand is stabilized by numerous intermolecular hydrogen bonds including those between the main chain atoms of Glu 10 and Gln 12 to β-strand 11, and the side chains of Lys 11 and Asn 181. As observed for ADP, the presence of bound substrate and the inward movement of the α8 helix and loop extends the leader peptide binding site and creates many additional hydrogen bonds (Table S14). For example, the side chains of Asn 186/B and Lys 244/B make the full complement of hydrogen bonds to Glu 10; and the side chains of Asn 181/B and Glu 180/B interact with Lys 11 and Gln 12. This hydrogen bonding pattern creates a small β sheet surface formed by the leader peptide and CdnC. When compared to the structure of MdnC bound to a synthetic leader peptide, the conserved motifs form the same helical structure and bind to the ligases with surface areas ca. 715–750 Ų. In MdnA the conserved Arg 17 makes key electrostatic contacts with Glu 191 and Asp 192 of MdnC. In CdnA3, this Arg residue is substituted by Ala, and residue 192 is a Lys whose side chain does not interact with the CdnA3 leader peptide.

CDNC:SUBSTRATE P3 INTERACTIONS

We were unable to crystallize CdnC in complex with full length CdnA3 or truncated CdnA3 variants in spite of trying numerous combinations and conditions. However, we were able to solve the structure of a quaternary complex with a linear peptide TLKYPSDSDEG corresponding to P3, a microviridin homologue. In the CdnC crystal structure, P3 is bound only on one side of the asymmetric homodimer. Substrate binding leads to conformational changes that accommodate the peptide and also affect interactions with ADP and the leader peptide (Figure 5b). When bound, P3 sits in the groove between the A-domain and central C-domain, and its N-terminus extends toward the B-domain and leader peptide (Figure 6). Binding is dominated by polar contacts with at least 20 hydrogen bonds observed (Table S15). The interactions fall into three groups including stabilizing interactions, hydrophobic interactions and those that appear to be key recognition elements. In the first group, backbone and hydroxyl atoms of Thr 1 are hydrogen bonded to the O atom of Val 183 and NH1 and NH2 of Arg 243 and the backbone O atoms of both Ser 6 and Ser 8 are hydrogen bonded to the side chain heteroatoms of Arg 76, Glu 300, and Tyr 78. Interactions that appear to be key to specificity include those from Lys 3 to Glu180; Asp 7 OD1/OD2 to Arg 217 NH1/NH2 and Glu 300; Asp 9 OD1/OD2 to Arg 123 NH1/NH2 and Tyr 74’s OH group; and Glu 10 to Arg 123. The conserved Pro 4 binds in a hydrophobic pocket created by Trp 242, Ile 250, Phe 302, and Trp 303. In addition to the large movement of helix α8, binding of substrate P3 leads to other smaller changes in the position of the Arg 75/Arg 76 loop (connecting β6 and α2) and residues in the first helix of the C-2 domain. In particular, in molecule A, Arg 76 is hydrogen bonded to Asp 126 located in the first helix of the C-2 domain; and Arg 75 is hydrogen bonded to Glu 300 and Ser 15, located in the C- and A-domains, respectively. Arg 123, positioned close to Asp 126 in the C-2 domain helix, is directed toward solvent. These tight interactions effectively cap the groove between the B- and C-domains. Upon substrate binding and as shown in molecule B, the χ₁ angle of Arg 123 changes from 178° to −54° creating hydrogen bonds with Asp 9 and Glu 10 of substrate. The substrate pushes the Arg75/Arg76 loop away from the central domain breaking hydrogen bonds described above, making way for CdnC:P3 hydrogen bonds mediated by Arg 76 and Tyr 78 to conserved substrate residues. Interestingly, subtle and concomitant changes preserve the hydrogen bonds between Arg 75 and Ser 15 and Glu300, and an additional hydrogen bond is formed with OD1/OD2 of Asp12, rigidifying the substrate binding groove across the C-domain.

Figure 6.

CdnC interactions with core peptide 3 and substrate specificity. (A) Detailed interactions of peptide 3 (gray sticks) with molecule B (gold) of CdnC homodimer. The hydrogen bonds between Arg123 of CdnC and Asp9 of peptide 3 in chryseoviridins diverge from corresponding residues in MdnC and microviridin core sequences. (B) Sequence alignment for ten structurally characterized microviridins and chryseoviridin; * and : symbols denote identical and conserved residues. The Asp residue unique to C. gregarium is shown with a red arrow. (C) Sequence alignment of MdnC from three microviridin-producing cyanobacteria to CdnC; Arg 123 (red arrow) is unique to C. gregarium.

PROBING THE DETERMINANTS OF CDNC MACROCYCLIZATION

Synthetic version of peptides 1–3 (Table 1) were used for crystallization. When incubating each of the synthetic peptides with CdnC, we found that only peptide 1 served as a substrate for the ligase to produce the bicylic structure with loss of two water molecules (Figure 3c, Table 1). Neither peptide 2 or 3 was cyclized even with longer incubation times or increased enzyme concentrations. Besides the core sequences, the major difference between peptide 1 versus peptides 2 and 3 is the presence of three amino acids N-terminal to the threonine nucleophile that begins each macrocycle. We hypothesized that an N-terminal handle is required for ligase activity. More specifically, GVS could function as a recognition or initiation sequence.

Table 1.

Core Peptide Sequences Probed for ω-Ester Bond Formation by CdnC Detected by HRMS

Entrya	Peptide sequences probed	Peptide m/z (ppm)	CdnC
	Sequence and core region		ΔH2O (−18 Da) m/z (ppm)	Δ2 H2O (−36 Da) m/z (ppm)
	Core region 1 and variantsb
1	GVSTSLKDVVTSPLGD	1574.8273 (0)	1556.8176 (−0.6)	NDc
2	GVSTSLKDVVTSPLGDTL	1788.7879 (95.0)	ND	1752.7780 (91.0)
3	GVSTSLKDVV	1004.5639 (−1.6)	986.5517 (0)	n/a
4	VSTSLKDVVTSPLGDTL	1732.5995 (220.0)	1713.9259 (0.6)	1695.9180 (0.8)
5	STSLKDVVTSPLGDTL	1632.8701 (−0.6)	1614.8594 (−0.6)	d1596.8471 (-0.1)
6	TSLKDVVTSPLGDTL	1545.8382 (−0.8)	1527.8289 (−1.6)	ND
7	SLKDVVTSPLGDTL	1444.7895 (−0.01)	1426.7806 (−1.2)	n/ae
8	LKDVVTSPLGDTL	1357.7579 (−0.4)	1339.7485 (−1.3)	n/a
9	KDVVTSPLGDTL	1244.6734 (−0.1)	1226.6633 (−0.5)	n/a
10	DVVTSPLGDTL	1116.5784 (0)	1098.5731 (−4.8)	n/a
11	VVTSPLGDTL	1001.5518 (−0.4)	983.5433 (−2.5)	n/a
12	VTSPLGDTL	903.0261 (0.1)	885.0141 (0.5)	n/a
13	TSPLGD	589.2876 (−8.1)	ND	n/a
	Core region 2
14	TLTLKTLDNGTTPAADTPV	1930.0284 (−5.6)	1912.0081 (0.6)	1894.0084 (−6.4)
	Core region 3: Native or hybrid microviridin coresa,f
15 g	TPVTLKYPSDSDEG	1509.6195 (60)	d1491.7009 (0)	1473.6174 (50)
16	GVSTLKYPSDSDEG	1454.6649 (−0.3)	d1436.6524 (1.1)	1418.6416 (1.3)
17	GVSTLKYPSDSDEGTL	1668.7973 (−0.6)	d1650.7954 (−5.8)	1632.7760 (−0.5)
18	TPVTLKYPSDSEEGG	1579.6968 (32.0)	1561.7375 (0.3)	ND

^aBold entries 2, 14, and 15 correspond to the synthetic versions of peptides 1–3 from the CdnA3 recombinant peptides.

^bUnderlined letters correspond to amino acids flanking the macrocycles.

^cND, not detected.

^dMinor product; ca. 1–5% abundance compared to the major cyclic or bicyclic product.

^en/a, not applicable; peptide sequence allows for a single macrocycle to be formed.

^fThe three-amino acid leaders GVS and TPV correspond to the N-termini of core domain peptides 1 and 3, respectively. The bold Glu residue in peptide 18 occurs in all cyanobacterial microviridin cores.

^gNative C. gregarium microviridin sequence.

To address this possibility, we designed and tested a panel of synthetic peptides with systematic additions or deletions to the N- and C-terminal residues of peptide 1 (Table 1, Figure S5) by incubating them with CdnC followed by LC-MS/MS analysis. We found that core 1 region peptides must contain two or three amino acids (VS or GVS) N-terminal to the nucleophilic Thr residue, as well as two amino acids C-terminal to the last Asp residue. To further explore these requirements, we synthesized the peptide TPVTLKYPSDSDEG that contains three amino acids (TPV) N-terminal to the initiating Thr residue and two C-terminal amino acids (underlined), as well as a hybrid peptide that begins and ends with the amino acids GVS and TL from core 1 flanking the bicyclic chryseoviridin sequence TLKYPSDSD (entries 15–17, Table 1). Incubation with CdnC followed by HRMS showed the loss of two water molecules from both peptides and thus validated this hypothesis.

CHRYSEOVIRIDINS AND PROTEASE INHIBITION

Microviridins were first isolated and characterized as protease inhibitors.^25,36 The C. gregarium precursor peptide differs from those yielding microviridins due to the presence of the additional macrocycles in the core. We tested each bicyclic core region (P1 to P3) and the full length chryseoviridin core containing four macrocycles and the C-terminal microviridin against four serine proteases including chymotrypsin, elastase, thrombin, and trypsin, and found that chryseoviridin (CdnA3) and P3Δ2 potently and specifically inhibited chymotrypsin with respective IC₅₀ values of 70 nM and 100 nM (Figure S6 and S7). Inhibition constants for 16 characterized microviridins range from 0.1 to 40 μM (using an arbitrary cutoff of 50 μM as inactive), making chryseoviridin one of the most potent RiPP chymotrypsin inhibitors reported to date.⁴¹

MECHANISTIC MODEL FOR ω-ESTER BOND FORMATION: ROLE OF ATP, MULTICORE MACROCYCLIZATION, AND SUBSTRATE SPECIFICITY OF CHRYSEOVIRIDINS

ATP Participation.

The microviridin ATP-grasp ligases are ATP-dependent, ω-ester bond-forming enzymes. To gain further insight into the mechanism of ω-ester bond formation, we generated a model of ATP bound in the nucleotide binding site of molecule B. The model places the γ-phosphate oxygen atoms O2G and O1G, in close proximity (ca. 2.8 Å) to the carboxyl side chain of Asp 7 consistent with proposals that phosphoryl transfer from ATP to Asp (or Glu) carboxylate groups activates the γ-carbon of Asp (or the δ-carbon of Glu) for nucleophilic attack by the hydroxyl group of Thr or Ser (Figure 7). In our current structure, Thr 1 is the first residue of peptide 3 and its hydroxyl group is oriented toward the N-terminus of the peptide, taking the place of the backbone O atom of what would be an upstream residue. This binding mode places the nucleophile ca. 9 Å away from Asp 7. This distance has to be reduced for nucleophilic displacement and lactone formation by the Thr hydroxyl and likely co-occurs with several concomitant changes; the ATP-Asp 7 intermediate might pull this residue closer to the nucleotide binding site and the B-domain, and the β9/β10 hairpin and/or the Arg75/Arg76 loop might move further toward the core. It is noteworthy that this site and the respective motions by the surrounding domains must have some plasticity because CdnA3 is able to generate macrocyles with different ring sizes and amino acid compositions.

Figure 7.

Model of ATP-dependent cyclization in peptide-bound CdnC. (A) Superposition of the two CdnC protomers shows differing orientations of the β-phosphate group in the absence (blue) or presence (gold) of peptide substrate. (B, C) Model and close up view of ATP superimposed on ADP of protomer B places the γ-phosphate within 2.8 Å of the carboxylate side chain OD1 of Asp 7. Substrate peptide 3 is colored green. The model was generated by superimposing the coordinates of ATP from PDB ID 5ZUA onto ADP of CdnC subunit B and refined to fit the ADP density map using the program COOT.⁴²

Multicore Macrocyclization.

Chryseoviridin is unusual among the graspetide RiPPs, in that it contains four individual macrolactones followed by a C-terminal microviridin core. Given that the timing and control of macrocyclizations and post-translational modifications can vary among different classes of RiPPs, we sought to investigate the timing of ω-ester bond formation in the chryseoviridins. This information would be of particular interest in light of the structure and orientation of a bound substrate. Previously, Ding and co-workers characterized macrocycle formation in vitro of the MdnA gene of Anabaena sp. PCC7120 using low concentrations of MdnC.³² We used a similar strategy and performed CdnC-mediated cyclizations in a matrix format on solutions (1 μM) of the three truncated CdnA3 constructs shown in Figure 3a. Starting with extremely low concentrations of CdnC (0.001 nM) and short incubation periods (5 min), we systematically increased the enzyme concentration to 10 nM. The reaction products were immediately cleaved by addition of TEV protease (10 U, 100-fold excess over substrate) for 1 min, quenched with 50% aqueous EtOH and analyzed by HRMS for detection of the peptide substrate (starting material) and the Δ1 and Δ2 products. The results suggested that macrocycle formation occurs starting with the N-terminal-most macrocycle and proceeds in the C-terminal direction ending with the microviridin bicyclic core (Table 2). This result is consistent with our initial sequencing studies (Figure 3) wherein formation of the C-terminal macrocycle was never observed in the absence of the N-terminal ring. Although Ding et al. observed distributive processing of a multimicroviridin core, they too only observed the second lactone after formation of the upstream ring. We performed a parallel run with native CdnA3; although we did not sequence or identify the macrocycle locations, we observed the same pattern with an increase in the number of water molecules lost with increasing CdnC concentrations, suggesting that the cyclizations occur in the same order with all constructs. Noting that additional mutagenesis studies are required to unambiguously prove the order of macrolactone formation, these results are consistent with the N-to-C orientation of the bound peptide substrate observed in the crystal structure: with each cyclization, the cyclic product has to move out of the binding site to allow occupation by the next participating Thr-to-Asp sequence for the formation of the next macrocyle (see below).

Table 2.

CdnC-Catalyzed ω-Ester Bond Formation^a,b,c

^aSequence detected after TEV proteolysis; proteolysis is quantitative under experimental conditions used; Δ1 corresponds to the first upstream macrolactone; Δ2 corresponds to both macrolactones. The C-terminal macrolactone is never detected in the absence of the N-terminal one.

^b‘+’, 10–50% abundance compared to Δ2;

^c‘++’, 50–100% abundance compared to starting substrate.

^d“nd”, not detected.

^e“sm”, starting material only, no cyclic product is detected.

Substrate Specificity in Chryseoviridins.

Finally, our structure provides information on elements of specificity for the C. gregarium ligase and microviridin core. C. gregarium contains a variant of the known microviridin concensus sequence TxKYPSDx(D/E)(D/E)xx. Sequence alignments of ten cyanobacterial microviridins (Figure 6b) show divergence at residues 8 and 9, where peptide 3 contains Ser and Asp, whereas microviridins contain a Trp at position 8 and a Glu at position 9, the location of the second lactone formed with Ser 6. In the substrate-bound structure, Arg 123 located on the α4 helix makes two hydrogen bonds (3 and 3.14 Å) to Asp 9 of peptide 3. The sequence alignment of CdnC to MdnC of M. aeruginosa and O. agardhii shows that Arg 123 is not conserved among these ligases and is replaced by Asp or Glu in the cyanobacterial enzymes (Figure 6c and Figure S8). To experimentally determine the role of Arg 123 in chryseoviridin specificity, we generated CdnC mutants R123D and R123G and carried out cyclization reactions with the peptide substrates shown in Table S16 corresponding to the second and third core regions. Both amino acid changes abolished the CdnC activity. Furthermore, CdnC treatment of a microviridin core containing Glu in place of Asp 9 resulted in the formation of the first ω-ester bond but not the second, further supporting the specificity for the Arg 123/Asp 9 pairing. These results demonstrate a critical, defining role for Arg 123 in C. gregarium compared to the cyanobacterial ligases.

DISCUSSION AND CONCLUSIONS

The biosynthesis and enzymatic logic involved in the formation of RiPP natural products continue to attract the attention of researchers as genome mining approaches reveal an abundance of possible post-translational modifications and new natural product families. In comparison, structural studies on RiPP modifying enzymes have received less attention. This could be due to the potentially transient and conformationally distinct nature of a free versus activated complex between the leader peptide, cofactors and substrate, which hampers crystallographic analysis of these complexes. This notion is supported by the greater than 90° arc that the B-domain moves upon substrate binding (Figure 8a). In addition to this change in orientation, residues 230–252 undergo a reversal in their secondary structure, with the locations of helical and loop regions changing positions (Figure 8b).

Figure 8.

Conformational changes at the α8 helix and loop occur with core peptide binding. (A) The superposition of protomers A (blue) and B (gold) in the asymmetric unit highlights the ca. 90° rotation of the α8 helix/loop in CdnC with peptide 3/substrate binding. Symmetry mates for CdnC appear in Figure S9. (B) Changes in secondary structure for residues 230–252.

The CdnC ligase studied here catalyzes the formation of successive ω-ester-mediated macrocycles in an unusually long Thr/Ser and Asp-rich core peptide. Together with the crystal structure of the complex, extensive HRMS/MS studies revealed that macrocycle formation occurs in the N-to-C direction and that a two or three amino acid sequence must be present at both the N- and C-termini of any given core peptide substrate for the cyclization to occur. The crystal structure and model of successive ω-ester ester bond formation support our experimental findings, in that, the substrate binds in an orientation that would allow the C-terminus of the bound leader peptide and the N-terminus of the bound core peptide to be connected. Recognition of the participating Asp residue (Asp 7 in the bound substrate used) by the strictly conserved Arg 217 appears to direct the substrate into the active site positioned for phosphorylation by ATP. We propose that, once ω-ester bond formation of the first macrocyle is complete, the B-domain swings away from the central domain, releasing ADP and the cyclic product. The open binding site is then ready to accept the next proximal substrate sequence and a new ATP equivalent to form the second macrocycle. HRMS/MS studies on a cyanobacterial microviridin BGC whose MdnA precursor contains three copies of a canonical microviridin core led to the conclusion that the macrocyclizations took place in a distributive manner because cyclic intermediates were detected at multiple locations in the core.³² The microviridin core regions in that precursor are separated by considerably longer sequences of 7 and 15 residues compared to the multicore of C. gregarium. The chryseoviridin ligase appears to act differently on the CdnA3 core because we have never detected ω-ester bond formation of C-terminal macrocycles in the absence of upstream cyclic products, and all six lactones are formed rapidly. Indeed, to observe formation of successive macrocycles (first through third), it was necessary to reduce the amount of CdnC to pM concentrations (Table 2). The “threading” of the tandem core regions through the substrate binding site and the very short linkers separating each macrocycle suggest a nondistributive mechanism in chryseoviridin biosynthesis although experiments beyond the scope of this work will be required to establish this.

Genome mining of RiPP BGCs is revealing the presence of precursor peptides that contain multiple core regions in otherwise unrelated RiPP families as observed previously in the cyanobactins⁴³ and fungal and plant peptides.⁴⁴ Recently, Kim and co-workers have greatly expanded the number of examples of multicore graspetides by directing their searches to BGCs containing ATP-grasp ligases and core regions with Ser/Thr nucleophiles and the receiving Asp/Glu amino acids.²¹ Biochemical and mass spectrometric characterization of those BGCs revealed new multicore ω-ester-containing peptides with diverse connectivities or topologies. For example, plesiocin contains multiple TTxxxxEE motifs, each forming a hairpin structure with adjacent cross-strand ω-esters,⁴⁵ and the thuringins display a similar cross-linked hairpin topology for the consensus core sequence TxxTxxxExxD.⁴⁶ Chryseoviridin is a novel hybrid RiPP whose structure consists of four macrocycles that appear as “beads on a string”, followed by a bicyclic core with the microviridin connectivity. Full length chryseoviridin and the microviridin-like product inhibited chymotrypsin with respective IC₅₀ values of 70 nM and 110 nM. This suggests that the fourth macrocycle located two amino acids upstream from the microviridin core region augments chryseoviridin binding to the protease. The upstream macrocycles on their own were inactive as protease inhibitors. This is in contrast to the multicore RiPP plesiocin that has a hairpin toplogy; in that study, residues at the crown of the cross-linked hairpin seem to control protease binding as amino acid changes could switch an inactive peptide to a protease inhibitor.⁴⁷ In the case of chryseoviridin, we did not determine the role of the multicore regions. The in situ functions of these multicore ω-ester-containing RiPPs remain to be determined, leaving open the possibility that their repeating macrocycles possess activities beyond the inhibition of proteases.

The structural and chemical studies described here expand our view of how ATP-grasp ligases function in the process of RiPP tailoring. Typically classified as enzymes that ligate discrete nucleophilic and carboxylate substrates, graspetide BGCs appear to have co-opted this activity to introduce intrachain macrocycles, and although there is plasticity among individual multicore precursor peptides, the results presented here and by Lee et al. suggest that macrocyclization could be highly species-specific. As with most RiPP families, many interesting questions remain. For instance, how is the mature RiPP cleaved and transported? Chryseoviridin contains two double glycine motifs posing the question whether a single or multiple mature peptides are produced in the source organism. The multicore domains in the cyanobactins are surrounded by conserved recognition sequences that are required for cleavage and further processing. In contrast, the C. gregarium multicore graspetide lacks extensive motifs between individual macrocycles. Last but not least, the functions of these multicore, multicyclic products in nature remain unknown. The chryseoviridin family of RiPPs can serve as a valuable model system for answering some of these questions.