Precursor-centric genome-mining approach for lasso peptide discovery

Abstract

Lasso peptides are a class of ribosomally synthesized posttranslationally modified natural products found in bacteria. Currently known lasso peptides have a diverse set of pharmacologically relevant activities, including inhibition of bacterial growth, receptor antagonism, and enzyme inhibition. The biosynthesis of lasso peptides is specified by a cluster of three genes encoding a precursor protein and two enzymes. Here we develop a unique genome-mining algorithm to identify lasso peptide gene clusters in prokaryotes. Our approach involves pattern matching to a small number of conserved amino acids in precursor proteins, and thus allows for a more global survey of lasso peptide gene clusters than does homology-based genome mining. Of more than 3,000 currently sequenced prokaryotic genomes, we found 76 organisms that are putative lasso peptide producers. These organisms span nine bacterial phyla and an archaeal phylum. To provide validation of the genome-mining method, we focused on a single lasso peptide predicted to be produced by the freshwater bacterium Asticcacaulis excentricus. Heterologous expression of an engineered, minimal gene cluster in Escherichia coli led to the production of a unique lasso peptide, astexin-1. At 23 aa, astexin-1 is the largest lasso peptide isolated to date. It is also highly polar, in contrast to many lasso peptides that are primarily hydrophobic. Astexin-1 has modest antimicrobial activity against its phylogenetic relative Caulobacter crescentus. The solution structure of astexin-1 was determined revealing a unique topology that is stabilized by hydrogen bonding between segments of the peptide.

With the increase in published genome sequences, many searches for natural products are rooted in bioinformatic approaches. Such genome-mining approaches have been particularly fruitful in the discovery of natural products that are derived from gene-encoded, ribosomally synthesized precursors (1). One such class of ribosomal natural products, the lasso peptides (2), is characterized by a unique “threaded lasso” fold (Fig. 1). In lasso peptides, a single posttranslationally installed isopeptide bond between the N-terminus of the peptide and an aspartate or glutamate side chain form an 8- or 9-amino acid macrocycle or ring through which the C-terminal portion is threaded forming a loop and a tail (Fig. 1A). This topologically constrained structure endows lasso peptides with protease resistance and tremendous stability toward chemical and thermal denaturation (3, 4). Lasso peptides have been divided into classes based on the presence or absence of disulfide bonds. Class I lasso peptides have a pair of disulfide bonds and class II lasso peptides are devoid of disulfide bonds. The founding members of class I are three lasso peptides, siamycin I, RP71955, and siamycin II, with highly similar sequences and structures (5–7). In contrast, the class II lasso peptides, of which there are four structurally confirmed examples (RES-701-1, microcin J25, lariatin, and capistruin) (8–13) and two probable lasso peptides (anantin and propeptin) without published structures (14, 15), have widely varying sequences and topologies. A lasso peptide with only a single disulfide bond, BI-32169, has been described and its structure has been determined (16, 17).

Fig. 1.

Lasso peptide topology and gene clusters. (A) A lasso peptide is composed of an N-terminal isopeptide-bonded macrocycle or ring (red) followed by a linear segment that threads the ring forming a loop (green) and a tail (orange). The end of the tail is the C terminus of the peptide. (B) Four-gene architecture of the microcin J25 gene cluster.

Of the lasso peptides described above, the biosynthetic gene clusters of only microcin J25 (MccJ25), capistruin, and lariatin are known (13, 18, 19). A recent peptidogenomics study has uncovered two new lasso peptide gene clusters, one each from class I and class II (20). These clusters all share four genes, and the functions of these genes have been elucidated by studies on the antimicrobial lasso peptide MccJ25. The mcjA gene in the MccJ25 cluster encodes a 58-aa precursor that is posttranslationally modified into the lasso structure by the action of the enzymes encoded for by the mcjB and mcjC genes (Fig. 1B) (21). The precursor protein McjA contains a 37-aa N-terminal leader sequence that is removed during the processing of the lasso peptide. The requisite peptidase activity for this transformation has been mapped onto the mcjB gene product (22, 23). The McjC enzyme has homology to ATP-dependent amide bond-forming enzymes (10, 22, 23), and is proposed to catalyze the formation of the isopeptide bond in MccJ25. In vitro studies on these enzymes revealed that the McjB and McjC enzymes must function in a concerted fashion to convert McjA into mature MccJ25 (21, 23). The fourth gene in the cluster, mcjD, encodes an ABC transporter that pumps the mature MccJ25 out of cells, and thus also serves as an immunity factor for the producing cells (24).

All known lasso peptides are produced by bacteria, and several of them have narrow-spectrum antimicrobial activity (12, 13, 25). Thus, the likely role of these molecules in nature is as defense molecules against other organisms in their environment. However, lasso peptides also have a host of pharmacologically relevant functions, including receptor antagonism (8, 26), enzyme inhibition (15), and inhibition of viral fusion (27). This rich spectrum of activities motivates searches for new lasso peptides in the ever-expanding set of bacterial and archaeal genome sequences. Several groups have used homology searches to the mcjB and mcjC gene products to identify putative lasso peptide gene clusters (21, 28). This approach was successfully used in the discovery of the lasso peptide capistruin (13). As mentioned above, a combined genome-mining/MS approach led to the discovery of two new lasso peptides and their clusters (20). Here we describe a unique genome-mining approach, to broadly search for novel class II lasso peptides across all bacterial and archaeal genomes, that leverages information about sequence requirements in the precursors to lasso peptides. This approach allows us to gain a global, quantitative view of the frequency of lasso peptide gene clusters in bacterial and archaeal genomes. We validate this approach by heterologously expressing a unique lasso peptide, astexin-1, from the aquatic bacterium Asticcacaulis excentricus and determining its structure.

Results

Precursor-Centric Approach to Lasso Peptide Gene Cluster Discovery.

All known class II lasso peptides contain an N-terminal glycine residue that is ultimately covalently linked to an aspartate or glutamate side chain. Besides these conserved features of the lasso peptide, the sequence and length of the matured peptide vary significantly (Fig. S1A). The N-terminal leader peptides of lasso peptide precursors are similarly varied in length and sequence composition (Fig. S1B), making conventional homology searches for precursors useless. One element of leader peptides that is conserved in each of the five verified gene clusters is a threonine residue in the penultimate position of the leader sequence (Fig. S1B). Our group has recently investigated the role of threonine in the leader peptides of MccJ25 and capistruin, and found that it is the optimal residue for the production of these peptides in vivo (29, 30). Other amino acids similar in size and shape to threonine can be substituted and still lead to efficient production of MccJ25 and capistruin in cells. Combining the sequence requirements for the mature lasso peptide with the requirement for threonine in the penultimate position of the leader peptide results in a rudimentary pattern for lasso peptide precursors (Fig. 2). We developed a pattern-matching algorithm to search all bacterial and archaeal genomes, both annotated and unannotated, for lasso peptide precursors. Once a potential precursor was identified, a 10,000 base region centered on the precursor was searched for ORFs that contain conserved motifs from the lasso peptide maturation enzymes using the MAST algorithm (31). We chose the MEME (32) algorithm to search for such motifs in McjB and McjC and their homologs in capistruin and other putative lasso peptides identified previously (13, 28). The full MEME training set contained 11 known or putative lasso peptide clusters (Fig. S1C). Four conserved motifs were identified in the mcjB/capB homologs (Fig. 2), including the cysteine-histidine-aspartate catalytic triad of cysteine proteases, which is required for the function of McjB (22, 23). Three motifs were found in mcjC/capC homologs (Fig. 2), including a serine- and aspartate-rich ATP-binding pocket reminiscent of those found in asparagine synthetases (33). An overview of the genome mining workflow is presented in Fig. S2.

Fig. 2.

Conserved elements of lasso peptide biosynthesis machinery used for genome mining. (A) Lasso peptide precursor pattern. A conserved threonine residue (red) is found in the second-to-last position of the leader peptide. The conserved glycine at the N terminus of the core peptide is covalently linked to an Asp (D) or Glu (E) side chain. (B) Sequence logo representation of conserved motifs in lasso peptide maturation enzymes as predicted by MEME software. Asterisks in the B motifs refer to catalytic triad residues. Scale representation of the B and C genes show the location of the motifs.

This genome-mining approach has the advantage of providing a global perspective on the prevalence of lasso peptides in bacterial and archaeal genomes, because all possible ORFs are considered as potential precursors, not just annotated sequences. In 1,490 fully sequenced and 1,678 partially sequenced genomes, we found 86,986,699 ORFs that are within the correct length range of a lasso peptide precursor. Of these ORFs, 336,425 fit the pattern in Fig. 2, and 98 precursors are associated with 79 different lasso peptide gene clusters that contain five or more of the seven conserved motifs described above (Table S1). Twenty-one of the clusters have matches to all seven of the conserved motifs, including the cluster encoding capistruin. In fact, the capistruin gene cluster is found in four different species of Burkholderia. The recently identified class II lasso peptide in Streptomyces roseosporus (20) also appears in our genome mining and has matches to six of the seven motifs. Only 2 of the 79 clusters come from archaeal genomes. Of the 79 clusters identified, 15 have multiple precursor genes associated with the cluster. All together, 76 different organisms were found to contain one or more lasso peptide gene clusters. These organisms include both Gram-positive and Gram-negative bacteria and are broadly spread across nine different bacterial phyla and a single archaeal phylum. Of the 76 putative lasso peptide-producing organisms, 8 contain two gene clusters, and no organisms are observed with more than two clusters. The freshwater bacterium A. excentricus contains two clusters, each containing two precursors (Fig. 3). As validation of our genome-mining approach, we focused on the isolation of a single lasso peptide from A. excentricus.

Fig. 3.

Lasso peptide gene clusters in A. excentricus. (A) A. excentricus contains two different lasso peptide clusters, each with two precursors. (B) Minimal engineered gene cluster for heterologous production of astexin-1 in E. coli. The stem-loop region found in the native cluster is replaced with an ideal E. coli ribosome binding site.

Heterologous Expression of A. excentricus Lasso Peptide in Escherichia coli.

A. excentricus is an α-proteobacterium that is in the Caulobacteraceae family and the strain that is sequenced, A. excentricus CB48, contains two circular chromosomes, each encoding a lasso peptide gene cluster with two distinct precursors (Fig. 3A). The cluster on chromosome 1 has one precursor gene [atxA1, National Center for Biotechnology Information (NCBI) Gene ID 10053804] within approximately 50 bases of the mcjB/capB homolog. The second precursor gene is more than 900 bases upstream from the mcjB/capB homolog (atxB, NCBI gene ID 10053805) and is not annotated in the NCBI (Fig. 3A). Of particular note is that this cluster lacks an McjD homolog, and therefore is devoid of a dedicated transporter. We cultured A. excentricus in its recommended medium and examined both the culture supernatant and boiled cell lysates for the presence of lasso peptides by MS, but none of the predicted masses were observed. We turned our focus instead to the heterologous production of a single putative lasso peptide in E. coli, a strategy that was successful in the production of capistruin (13, 30). The lasso peptide encoded by the atxA1 gene is longer (23 aa) than any known lasso peptide. This peptide also contains highly polar segments, a feature not observed in most lasso peptides, so we focused our initial efforts on this peptide. We constructed a minimal gene cluster (Fig. 3B) consisting of the atxA1, atxB, and atxC genes (NCBI gene IDs 10053804–10053806). There is a 54 base region between the atxA1 and atxB genes that includes an inverted repeat sequence, a feature that has been observed previously in lasso peptide gene clusters (28), including the cluster that codes for capistruin (30). This sequence was removed from the cluster and replaced with a short sequence that includes an optimized E. coli ribosome binding site (Fig. 3B). Upstream of the atxA1 gene, a 23-base region from the A. excentricus genome including a putative ribosome binding site was maintained. This minimal cluster was harbored on a tetracycline-inducible plasmid, allowing for controlled expression of the three-gene cluster. This plasmid was named pMM32.

Pilot experiments on BL21 E. coli cells transformed with pMM32 were carried out to determine whether heterologous production of the lasso peptide was possible. Cells were grown in M9 minimal media supplemented with amino acids and supernatants, and boiled cell lysates from these cultures were subjected to solid-phase extraction (SPE). The eluents from the SPE columns were examined by MALDI-TOF MS. Although only faint signals corresponding to lasso peptides were detected in the cell lysates, strong signals were observed in the supernatant preparations. In addition to a mass corresponding to the full-length lasso peptide, two truncated products were observed. These products correspond to the removal of 3 or 4 amino acids from the C terminus of the peptide (Fig. S3A). Similar C-terminally truncated products have been previously observed for the lasso peptides propeptin and lariatin (12, 34). All three of these products contained a single dehydration relative to their corresponding linear peptides, indicative of cyclization. To determine conditions for the maximal production of the full-length lasso peptide, cells were induced at a range of different OD₆₀₀ values. The optimal OD₆₀₀ for full-length lasso peptide production was found to be 0.2–0.4 (Fig. S4). The cultures were scaled up to 1-L shake flasks, and a crude preparation containing the lasso peptides was obtained by SPE. Subsequent purification by semipreparative HPLC yielded a pure sample of the full-length lasso peptide, which we have named astexin-1 (Fig. S3B).

MS Analysis of Astexin-1.

We carried out an extensive MS analysis of astexin-1 to confirm the identity and sequence of this peptide and to get a preliminary view of its topology. Using a MALDI-TOF/TOF instrument, we acquired full MS spectra of the peptide mass range and performed MS² fragmentation on purified astexin-1 to confirm its sequence: G¹LSQGVEPD⁹IGQTYFEESRINQD²³ (Fig. 4). The singly charged parent ion was detected at 2563.96 Da, which is in agreement with the theoretical mass of 2564.19 Da, taking into account a loss of 18 Da after a single dehydration. Previous MS studies show that the fragmentation pattern in MSⁿ analyses can be used to partially elucidate the structure of lasso peptides (9–11, 13, 35). Because of the cyclic component of the structure, MS² experiments typically do not fragment the macrolactam ring. Because the amide bonds in the tail are readily broken in MS², the point of cyclization can be determined by looking for the last b-series ion. The sequence of astexin-1 contains a glutamate at position 7 and an aspartate at position 9, which makes it chemically possible for astexin-1 to have either a 7- or a 9-membered ring, although a lasso peptide with a 7-membered ring has never been observed. The b-series product ions that we observed were limited to b₉–b₁₇ and b₁₉–b₂₂ (Fig. 4). This absence of b₇ and b₈ ions provided strong evidence that cyclization is indeed between Gly1 and Asp9. We also observed signals from y-type product ions consisting of y₁₄ through y₂, consistent with cyclization at Asp-9.

Fig. 4.

Tandem MS analysis of astexin-1. The fragmentation pattern is consistent with cyclization between Gly-1 and Asp-9.

The threading of the tail also contributes characteristic features to the MS² spectrum (35, 36). Sterically linked b- and y-series fragments have previously been reported in MS² spectra of MccJ25 (9–11). Although these fragments correspond to a loss of several residues from loop of MccJ25, its steric lock prohibits the tail from dissociating from the peptide after the loop has been broken. In the MS² spectrum of astexin-1, we matched fragment ions, not belonging to the a/b/c/x/y/z-series, to species that have lost portions of the loop. Specifically, we observed 2278.06 g/mol (astexin-1 missing G11-T13) and 2172.02 g/mol (missing Q12-Y14) fragments (Fig. S5). This finding suggested that astexin-1 is threaded between bulky residues F15 and R19, which can act as a steric lock.

NMR Analysis of Astexin-1 Structure.

To confirm our hypothesized topology for astexin-1, we acquired COSY, TOCSY, and NOESY spectra on a sample of astexin-1 in DMSO-d₆ at a concentration of 4.4 mg/mL. All residues in the peptide were fully assigned (Table S2). Amide to α-proton correlations were observed for all residues except Pro-8 in the fingerprint region. The Hβ protons of Asp-9 had a strong cross-peak with the amide proton of Gly-1 in the NOESY spectum (Fig. 5), strongly suggesting cyclization between these residues (36). Strong sequential NOE cross-peaks were observed for all residues in the peptide. In the case of Pro-8, a strong cross-peak in the NOESY spectrum between its Hδ protons and the Hα proton of Glu-7 indicated that Pro-8 is in the trans conformation. Several long-range cross-peaks were observed in the NOESY spectrum as well. Specifically, through space magnetization transfer between protons of Tyr-14, Glu-16, Glu-17, Ser-18, and protons of residues in the ring provided evidence for threading of the C-terminal tail of the peptide threading through the N-terminal ring. Furthermore, NOE cross-peaks between Asn-21 and Ile-10 and between Gln-22 and Thr-13 suggest that the C-terminal tail may have a specific interaction with residues in the loop portion of the lasso peptide. The volumes of all assigned NOESY spectrum cross-peaks were measured by integration and calibrated to the Tyr Hδ-Hε cross-peak, yielding a set of upper distance restraints (Table S3). Eight additional constraints were introduced to properly define the geometry of the Gly-1_NH–Asp-9_Cγ bond (13).

Fig. 5.

Fingerprint region of astexin-1 NOESY spectrum. A strong cross-peak between the amide proton of Gly-1 and the β-protons of Asp-9 (boxed) demonstrate the connectivity of the ring. Cross-peaks between the amide proton and α-protons for each residue are labeled.

Simulated annealing was performed using CYANA 2.1 (37). All structures had good covalent geometry after energy minimization (Fig. S6), with the average structure confirming the lassoed topology of astexin-1. Consistent with the six through-space NOE cross-peaks observed for Ser-18 (Table S3), the final energy-minimized structures all have Ser-18 at the position that crosses through the ring (Fig. 6A). The Arg-19 residue serves as the steric lock residue that prevents the tail from slipping through the ring. The association of the tail portion of the peptide with the loop formed by residues 10–17 is retained in most of the structures after energy minimization (Fig. 6B), and there are hydrogen bonds between the Ile10-Thr13 segment of the loop and the Asn21-Asp23 segment of the tail. Although the presence of the Ser-Arg dyad as the residues that pierce the ring and serve as a steric lock, respectively, is reminiscent of the structure of capistruin (13), the overall structure of astexin-1 differs from previously observed lasso peptides. Because of its relatively large size (23 aa), astexin-1 has both a long loop and a long tail. The structural data presented here indicate that these portions of the peptide interact to form a structure resembling a three-pronged claw in which the three prongs are the tail and the two sides of the loop (Fig. 6). The structural models of astexin-1 have been deposited in the BioMagResBank (accession no. 18481) and the Protein Data Bank (PDB ID code 2LTI).

Fig. 6.

Solution structure of astexin-1. In all figures the ring portion of the peptide is presented in gray as a reference. (A) Lowest energy conformer of astexin-1 showing the Arg-19 steric lock (red). (B) Alignment of the 20 lowest energy structures. (C) Hydrogen bonding (dashed green lines) between the loop and the tail of astexin-1 is contrasted to the disulfide bond (solid green line) formed between the ring and tail of BI-32169. (D) Electrostatic surface potential of astexin-1 reveals an overall polar structure with a small hydrophobic patch delineated by the Val-6 and Ile-11 side chains.

Antimicrobial Activity of Astexin-1.

We carried out a series of spot-on-lawn assays against a panel of Gram-negative bacteria to determine whether astexin-1 has narrow-spectrum antimicrobial activity found in other lasso peptides such as MccJ25, lariatin, and capistruin. A 2.2-mM solution of astexin-1 in water was spotted on E. coli BL21, Vibrio harveyi, Vibrio fischeri, Burkholderia thailandensis (producer of capistruin), Salmonella newport (hypersensitive to MccJ25), and Caulobacter crescentus CB15. Astexin-1 only had observable antimicrobial activity against C. crescentus (Fig. S7), a closely related bacterium in the same taxonomic family as A. excentricus.

Discussion

Here we have described a unique genome-mining approach for the discovery of lasso peptides that leverages sequence pattern information in lasso peptide precursors. This approach has several advantages over conventional homology-based genome-mining approaches, including the ability to be completely automated and to provide a more global view of the prevalence of lasso peptide gene clusters in bacterial and archaeal genomes. Our approach identifies several lasso peptide clusters (and putative clusters) that have been previously described (13, 20, 21, 28), but also uncovers dozens of unique clusters that are broadly distributed across bacteria. In addition, we find two archaeal lasso peptide clusters.

To validate the genome-mining approach, we searched for lasso peptides in cultures of the α-proteobacterium A. excentricus, a bacterium with no precedence as a natural product producer, but that contains two lasso peptide clusters. No lasso peptides were detected when A. excentricus was cultured under laboratory conditions, indicating that the lasso peptide clusters may be silent (or at least strongly repressed) under these conditions. Fortunately, heterologous expression of one of the predicted lasso peptides, astexin-1, was successful in E. coli. The approach used to heterologously produce astexin-1 involves construction of a minimal gene cluster for lasso peptide production in which any potential regulatory sequences are stripped out and replaced with optimized elements for production in E. coli. Because lasso peptides require only two posttranslational processing steps (cleavage of the leader peptide followed by formation of the isopeptide bond), we believe that similar heterologous expression approaches will be useful for the production of other poorly expressed or silent lasso peptides.

Many of the lasso peptide clusters identified in our genome mining study, including the astexin-1 cluster, lack a dedicated ABC transporter that is found in the MccJ25 and capistruin clusters. Despite this, heterologously produced astexin-1 was found in the culture supernatant. E. coli producing astexin-1 continue to grow, and are alive after 24 h of induction, suggesting that astexin-1 does not cause lysis of the producing cells. This finding implies that astexin-1 may be using an existing ABC transporter in E. coli. There is precedence for this notion, because MccJ25 can use the transporter YojI as an alternative to McjD (38). An alternative explanation is that astexin-1 is not actively transported out of E. coli, and is simply able to diffuse through the bacterial membrane.

The solution structure of astexin-1 was determined and found to have a nine-membered macrolactam ring followed by an 8-aa loop and a 6-aa tail, forming the largest lasso peptide described to date. The extended length of the loop and tail regions allow for hydrogen-bonding interactions between the backbone and polar side chains of these regions of the peptide. Interactions between different segments of the lasso peptide are found in class I lasso peptides and BI-32169, where a disulfide bond covalently links the tail to the ring. In contrast, in astexin-1, the loop and tail interact via noncovalent hydrogen bonds (Fig. 6C). The electrostatic surface potential of astexin-1 (Fig. 6D) reveals a structure that is primarily polar with a shallow hydrophobic cleft bracketed by Val-6 and Ile-10. There is increasing evidence that lasso peptides are highly redesignable (39). Multiple amino acid substitutions can be tolerated in MccJ25, particularly in its loop (40, 41), and single and some double substitutions are tolerated within the ring and tail portion of capistruin (42). The large size and high polarity of astexin-1 make it attractive as a scaffold in which both the loop and tail regions can be modified to generate lasso peptides with new functions.

Methods

Bioinformatics.

The fasta (.fna) format sequence files for each genome were downloaded from ftp.ncbi.nih.gov/genomes. Scripts implemented in the Perl programming language were used to: (i) organize and parse genome fasta files, (ii) identify putative lasso peptide precursors and associated maturation enzymes, (iii) detect motifs in maturation enzymes, and (iv) rank biosynthetic loci. ORFs (both precursors and maturation enzymes) bracketed by a start and a stop codon in all six frames were identified using getorf, a program from the EMBOSS bioinformatics suite (43). Associated maturation proteins were searched for within a 10,000-bp region centered on the precursor gene. Forward and reverse conserved domain motif matching was implemented using the MEME software Suite (31, 44). Putative lasso peptide clusters were scored from 0 to 7 based on how many of the maturation protein motifs were present.

Heterologous Expression and Purification of Astexin-1.

E. coli BL21 cells harboring the plasmid pMM32, which contains the minimal gene cluster encoding astexin-1 production, were cultured in M9 minimal medium supplemented with amino acids. The culture was induced with anhydrotetracycline at an OD of 0.21–0.26 and induction continued for 48 h. The culture supernatant was collected by centrifugation and passed over a C₈ solid-phase extraction column (Strata). The column was eluted step-wise with increasing concentrations of acetonitrile in water. The peptide was purified on the semipreparative scale on an Agilent 1200 HPLC system. Detailed descriptions of plasmid construction, culture conditions, and astexin-1 purification can be found in SI Methods.

Mass Spectrometry.

Molecular weight determination was performed in the m/z 800–4,000 range using a 4800 Plus ABSciex MALDI TOF/TOF Analyzer (ABSciex). All samples were dried and reconstituted in a mixture of 50% (vol/vol) acetonitrile/water to the indicated concentration. One microliter of this sample was spotted onto an Applied Biosystems 384 Opti-TOF 123 mm × 81 mm SS plate and allowed to dry. A 2.5-mg/mL solution of α-Cyano-4-hydroxycinnamic acid (Sigma), dissolved in 50% acetonitrile/water (1 μL), was subsequently spotted on top of the dry sample and allowed to dry. MS/MS product ion analysis was performed with a 1-kV collision energy. Both MS and MS/MS analyses were performed in positive-ion mode.

NMR.

The NMR sample was dissolved in DMSO-d₆ (Cambridge Isotope Laboratories). All NMR experiments were run on a 500 MHz Bruker Avance-III NMR spectrometer equipped with a TCI cryoprobe (Bruker-Biospin) at 295 K temperature using pulse sequences either from the Bruker library (TopSpin 2.1) or in-house modified versions. The samples were prepared in a 5/2.5 mm OD precision NMR tube (New Era Enterprises). The 1D experiments with WET (45) suppression of the water signal in DMSO-d6 were run using 8.88-ms selective pulses on the water resonance in combination with gradient pulses to provide a narrow suppression profile. The time domain size was 32,000 time domain points over a 15-ppm spectral window, with 64 scans averaged. The spectrum was centered on the water resonance at 3.376 ppm. Phase-sensitive, gradient-selected COSY experiments were also run with WET solvent suppression. The data matrix was 4 K × 512 data points collected over 15 ppm in States-TPPI (46) mode with acquisition times of 0.273 and 0.034 s in t2 and t1, respectively, with eight scans for each t1 increment. Gradient-assisted TOCSY experiments (47) were collected with the same acquisition parameters using DIPSI2 (48) spin-lock sequence of 60–120 ms SL time. No particular solvent suppression was applied. Phase-sensitive NOESY experiments were collected using the excitation sculpting method (49) for solvent suppression, using a 4 K × 1 K data matrix, otherwise identical parameters. The selective inversion pulse for the excitation sculpting sequence was 2.4 ms. Various mixing times were applied between 80 and 700 ms. All data processing was done using MestReNova (MestreLab Research). Detailed information about processing of the spectra can be found in SI Methods.