Genome Mining and Discovery of Imiditides, a Family of RiPPs with a Class-defining Aspartimide Modification

Abstract

Ribosomally synthesized and post-translationally modified peptides (RiPPs) are a large and diverse class of natural products of ribosomal origin. In the past decade, various sophisticated machine learning-based software packages have been established to discover novel RiPPs that do not resemble the known families. Here we show that tailoring enzymes that cluster with various RiPP families can serve as effective bioinformatic seeds, providing a complementary approach for novel RiPP discovery. Leveraging the fact that O-methyltransferases homologous to protein isoaspartyl methyltransferases (PIMTs) are associated with lasso peptide, graspetide, and lanthipeptide biosynthetic gene clusters (BGCs), we utilized a C-terminal motif unique to RiPP-associated O-methyltransferases as the search query to discover a novel family of RiPPs, the imiditides. Our genome-mining algorithm reveals a total of 670 imiditide BGCs, distributed across Gram-positive bacterial genomes. In addition, we demonstrate the heterologous production of the founding member of the imiditide family, mNmaA^M, encoded in the genome of Nonomuraea maritima. In contrast to other RiPP-associated PIMTs that recognize constrained peptides as substrates, the PIMT homolog in the mNmaA^M BGC, NmaM, methylates a specific Asp residue on the linear precursor peptide, NmaA. The methyl ester is then turned into an aspartimide spontaneously. The substrate specificity is achieved by extensive charge-charge interactions between the precursor NmaA and the modifying enzyme NmaM suggested by both experiments and an AlphaFold model prediction. Our study shows that PIMT-mediated aspartimide formation is an emerging backbone modification strategy in the biosynthesis of multiple RiPP families.

Graphical Abstract

Introduction

Ribosomally synthesized and post-translationally modified peptides (RiPPs) are a class of natural products of ribosomal origin.^1–3 RiPP biosynthesis begins with a ribosomally synthesized precursor peptide, the N-terminal leader peptide of which recruits the tailoring enzymes to post-translationally modify the C-terminal core sequences.^1,2 As the precursor sequences of RiPPs are gene-encoded, genome-mining approaches have proven to be an exceptionally well-suited strategy for RiPP discoveries.^2,4–9 In fact, genome mining for RiPPs has advanced to the point where well-known RiPPs such as lanthipeptides, lasso peptides, and thiopeptides can be readily identified from genome sequences using tools like antiSMASH.^4,10 Improvements in genome annotation have also led to the assignment of many genes involved in RiPP biosynthesis in the genome sequences deposited in GenBank.

One frontier in genome mining for RiPPs has focused on the discovery of entirely new classes of RiPPs that do not resemble the known families. Sophisticated machine learning-based software packages have been developed for this task.^11,12 Previous work by our group and the van der Donk group reported homologs of protein isoaspartyl methyltransferases (PIMTs) are associated with biosynthetic gene clusters (BGCs) of several RiPP families, including lanthipeptides,^13,14 lasso peptides,^15,16 and graspetides.^17,18 We reasoned that these shared auxiliary enzymes associated with known RiPPs could also be used as a bioinformatic handle to identify novel RiPP BGCs. In particular, the PIMT homologs previously reported strictly recognize constrained macrocyclic RiPPs as substrates to install the aspartimide post-translational modification. Therefore, we investigated whether these PIMT homologs can directly modify linear precursors and become a class-defining enzyme for an uncharacterized family of RiPPs.

Here, we report the genome mining and experimental validation of a novel family of RiPPs in which aspartimide formation is the class-defining modification, for which we propose the name imiditide. The results of our genome mining efforts reveal 670 putative imiditide BGCs, all of which are in Gram-positive bacterial genomes. Representative genera that encode imiditides include Streptomyces, Actinomadura, and Nonomuraea. In addition, we report the heterologous production of an imiditide from Nonomuraea maritima in E. coli. The PIMT homolog in the BGC, NmaM, directly methylates a specific aspartate residue on the linear precursor, NmaA. The aspartimide is then spontaneously formed. The aspartimide in the resulting imiditide exhibits some stability, but can be hydrolyzed regioselectively to incorporate the isoaspartate residue into the polypeptide. Our experimental evidence along with an AlphaFold model of the NmaA-NmaM complex suggest that the specificity of the modification is achieved through charge-charge interactions. Our findings demonstrate the feasibility of discovering novel classes of RiPPs by using shared tailoring enzymes across different RiPP families as bioinformatic seeds. Furthermore, we illustrate that PIMT-mediated aspartimide formation is an underappreciated backbone modification strategy in RiPP biosynthesis, compared to the well-studied backbone rigidification chemistries, such as thiazol(in)e and oxazol(in)e formation.^19,20

Results

Identification of Biosynthetic Gene Clusters for a Putative Novel RiPP Family

Our laboratory has previously characterized several RiPPs with aspartimide post-translational modifications, including lasso peptides cellulonodin-2 and lihuanodin,¹⁵ as well as graspetides fuscimiditide and amycolimiditide.^17,21 In their biosynthesis, a dedicated O-methyltransferase related to protein isoaspartyl methyltransferase (PIMT) installs the aspartimide on the matured lasso peptides or graspetides as the last step of the reaction series. Since PIMT homologs cluster with various families of RiPPs (Figure 1a), we reasoned that these PIMT homologs can serve as a bioinformatic handle to unveil uncharacterized families of RiPPs. We first constructed a sequence similarity network (SSN) of ThfM (Figure S1), the O-methyltransferase responsible for installing the aspartimide in the graspetide fuscimiditide.¹⁷ A total of 2000 sequences were queried to construct the SSN. ThfM surprisingly does not cluster with any other sequence. On the other hand, AmdM, the methyltransferase that installs an aspartimide in the tetracyclic graspetide amycolimiditide, clusters with methyltransferases from strains that are also near other graspetide BGCs. Similarly, OlvS_A, the PIMT homolog responsible for isoaspartate incorporation in OlvA(BCS_A) through the aspartimide formation, clusters with other methyltransferases near lanthipeptide BGCs. TceM and LihM, the lasso peptide methyltransferases, are not a part of this sequence similarity network, showing they are more distant from ThfM. Additionally, we noticed a cluster that contains a large number of PIMT homologs from Nonomuraea and Actinomadura strains. These PIMT homologs appear to be isolated in the genome context at first glance. However, by manual inspection of the genome sequence, we observed unannotated open reading frames (ORFs) of 30–75 amino acids that neighbor the PIMT homolog. To investigate their potential as class-defining enzymes for a novel family of RiPPs, we collected a set of 50 putative BGCs by performing a BLASTP search on an O-methyltransferase from Nonomuraea jiangxiensis found in the ThfM SSN (Table S1). The putative precursor sequences in these BGCs all contain at least one Asp residue required to be recognized by the PIMT homolog. The precursors are also rich in Gly, Lys and Pro residues (Figure 1b). Apart from genes encoding the putative precursor and the PIMT homolog, there are no other conserved genes in these BGCs. Therefore, we hypothesized that the PIMT homolog in these BGCs may be the class-defining enzyme for a novel family of RiPPs, where the O-methyltransferase directly acts on the linear precursor to install a post-translational modification.

Figure 1: Identification of a Novel RiPP Family.

a) Biosynthetic gene clusters containing protein isoaspartyl methyltransferase (PIMT) homologs. Red, precursor peptides; brown, PIMT homologs. In the cellulonodin-2 BGC, tceB1 and tceB2 code for a bipartite cysteine peptidase, while tceC encodes the lasso cyclase (orange). In the amycolimiditide BGC, amdB encodes an ester-installing ATP-grasp enzyme. In the OlvA(BCS_A) BGC, olvB encodes a dehydratase, and olvC codes a cyclase. Identified here first, BGCs corresponding to the novel RiPP family only harbor genes encoding a precursor and an O-methyltransferase. b) Sequence alignments of 5 putative precursor sequences in the novel RiPP family. Nonomuraea strains are colored blue and Actinomadura strains are colored red. The putative site of modification (Asp) is colored red. Conserved residues include Gly, His, Pro, and Lys, which are marked with asterisks.

Genome Mining of the BGCs for a Putative Novel RiPP Family

O-methyltransferases near RiPP BGCs are often annotated as “methyltransferase domain-containing protein” in NCBI and consist of an N-terminal region that is homologous to canonical PIMT (~220 aa) and a C-terminal domain of 110–180 aa (Figure S2). In our previous study on cellulonodin-2 and lihuanodin, a conserved WXXXGXP motif was identified in the C-terminal domain of PIMT homologs near lasso peptide BGCs. Amino acid substitutions in this motif in the TceM and LihM methyltransferases have a deleterious effect on aspartimidylation of pre-cellulonodin-2 and pre-lihuanodin, showing that the C-terminal domain plays an important role in substrate recognition.¹⁵ Additionally, the length of the C-terminal domain of these methyltransferases varies based on the type of RiPP families with which they cluster, and the sequences of methyltransferases associated with the same family show much more resemblance to each other (Figure S2).

To probe if the 50 PIMT homologs discussed above (Table S1) have a similarly conserved sequence motif, we first subjected these sequences to Multiple EM for Motif Elicitation (MEME).²² MEME analysis revealed a conserved 41 aa sequence motif within the C-terminal domain of the PIMT homologs (Figure 2a). A limited BLASTP search with this motif then revealed that it is a unique signature of methyltransferases encoded within minimal BGCs for the novel RiPP family. As such, we reasoned that we could use this conserved motif as a bioinformatic handle for genome mining of this new class of PIMT-associated RiPPs.

Figure 2: Genome Mining of Imiditide Biosynthetic Gene Clusters.

a) Sequence logo of a conserved 41 aa C-terminal motif of 50 methyltransferases associated with putative imiditide BGCs. A complete list of 50 methyltransferases is shown in Table S1. b) Overall strategy for the genome mining of imiditide BGCs. Blasting the 41 aa conserved sequence against the NCBI protein database returns a total of 5839 hits. Then, we search short open reading frames that contain putative precursor sequences near these PIMT homologs. c) An example workflow to identify the putative precursor sequence near a PIMT homolog. All 6 frames are translated, and putative precursor sequences are searched within 500 bp upstream or 1000 bp downstream of the PIMT homologs. Only the ORFs that contain at least one Asp residue and have a size between 30 to 75 aas will suffice as putative precursors. d) Applying additional filters based on other PIMT-associated RiPP clusters further eliminates potential false positive hits, resulting in a total of 670 identified imiditide BGCs.

As the first step in our genome mining pipeline, we comprehensively identified all PIMT homologs harboring the unique 41 aa motif by BLASTP search of the motif against the entire non-redundant NCBI protein database. Using an e-value cutoff of 1000, 5839 hits were obtained. For each of the PIMT homologs identified, we searched for an associated RiPP coding sequence in the vicinity by scanning for short open reading frames using a custom Python script. We initially filtered for ORFs encoding a 30–75 aa product within 500 bp upstream or 1000 bp downstream of the PIMT homolog and having at least one Asp residue in the protein sequence (Figure 2b).⁶ Given the operon nature of many RiPP BGCs, the short ORF and PIMT homolog coding sequence were also required to be transcribed in the same direction. When multiple short ORFs were identified within the defined windows of a PIMT homolog, only the ORF that has the highest alignment score with the putative precursor sequence from Nonomuraea maritima was considered, as the cluster was experimentally verified (see below). This analysis generated 1002 potential PIMT-associated RiPPs, hereafter referred to as imiditides because of the class-defining cyclic imide or aspartimide modification (hit rate of 17.2%). To further reduce potential false positive hits, we applied two additional filters to eliminate predicted imiditide BGCs that lack a close PIMT homolog. Based on observation of experimentally characterized RiPP-associated methyltransferases, we limited the length of PIMT homologs to 300–450 aa (Figure S2). Additionally, we required that the PIMT homologs harbor a GXGXG motif upstream of the 41 aa motif used in our initial protein BLAST search; the GXGXG motif has been shown to bind SAM in human PIMT (Figure S3). After applying these additional filters, we identified 670 putative imiditide BGCs containing a PIMT homolog, all in Gram-positive bacteria. The most represented genus from which these BGCs are found include Streptomyces, Actinomadura, Nonomuraea, and Nocardiopsis (Figure S4).

Additionally, we analyzed the amino acid compositions of the putative imiditide precursors. Consistent with what we observed in the original set of 50 precursors, these sequences are particularly rich in Gly and Lys residues, with Lys residues accounting for more than 5% of all residues in 294 precursors and Gly residues representing more than 10% of all residues in 444 precursors (Figure S5). A subset of the putative imiditide precursors also contain a tetracysteine motif that surrounds the potential modification site. A further BLASTP search on these cysteine-containing precursors reveals a total of 37 putative cysteine-containing imiditide clusters (Table S2). A structure prediction of the cysteine-containing imiditide precursors using HHpred suggested they are homologous to a part of the chaperone protein DnaJ.²³

Heterologous Expression of an Imiditide from Nonomuraea maritima

To validate that these putative BGCs indeed encode RiPPs, we attempted to heterologously express a putative imiditide BGC from Nonomuraea maritima in E. coli. In the native BGC, the ORF that encodes the precursor, nmaA, is immediately upstream of the gene encoding the methyltransferase, nmaM (Figure 3a, c). The precursor sequence NmaA is 46 aa long, and the methyltransferase NmaM has a size of 373 aa. To enhance the expression level of NmaA, we refactored the cluster by adapting a co-expression system. His₆-SUMO-NmaA was placed under an IPTG-inducible T5 promoter, while untagged NmaM was put under an IPTG-inducible T7 promoter on a separate plasmid (Figure 3b). Then, we coexpressed these two proteins using an E. coli BL21(DE3) ΔslyD strain at room temperature for 20 h (Figure 3b). Products were purified using nickel affinity chromatography under both the native and the urea denaturing conditions. Chromatography of the purified fraction on LC-MS revealed the formation of two additional species, with one carrying an addition of 14 Da and the other one bearing a loss of 18 Da (Figure 3d). These species were absent when we expressed His₆-SUMO-NmaA alone, indicating that the formation of these species was due to the action of NmaM (Figure 3d). Additionally, the SDS-PAGE gel of the purified protein fractions revealed that modified His₆-SUMO-NmaA pulled down untagged NmaM in the native condition as well as in the presence of 8M urea (Figure 3e, Figure S6), suggesting either a strong binding interaction between the two proteins or non-specific binding of untagged NmaM to the nickel resin. To distinguish between these two possibilities, we expressed untagged NmaM alone and subjected it to nickel affinity chromatography under native conditions. Untagged NmaM was retained on the column until the elution step with 250 mM imidazole (Figure S7), showing that NmaM has an inherent affinity for the nickel resin.

Figure 3: Heterologous Expression of mNmaA^M.

a) Native gene cluster of a putative imiditide from Nonomuraea maritima. nmaA encodes the putative precursor sequence (sequence shown in part c), and nmaM encodes an O-methyltransferase that is a PIMT homolog. b) Refactoring strategy for mNmaA^M expression. An IPTG-inducible T5 promoter was introduced upstream of a N-terminally His₆-SUMO tagged precursor peptide gene. An IPTG-inducible T7 promoter was used to drive untagged methyltransferase (NmaM) expression. c) Amino acid sequence of the putative precursor NmaA. d) NmaM was responsible for the formation of the putative methylated and aspartimidylated species. The heterologous expression of His₆-SUMO-NmaA alone gave a species with a mass of 17277.1 Da (expected mass: 17277.1 Da). Coexpression of His₆-SUMO-NmaA and NmaM resulted in additional putative methylated (+14 Da) and aspartimidylated species (−18 Da). The putative aspartimidylated species accumulated over time, showing it was likely the intended product in vivo (also shown in Figure S9). e) SDS-PAGE gel of mNmaA^M native purification with pull-down of untagged NmaM. FT: flowthrough, W: washes, E: elutions. Figure S7 shows that untagged NmaM binds efficiently to Ni resin despite lacking a His-tag. f) Proposed reaction pathway for mNmaA^M. NmaM methylates a specific aspartate residue using SAM. The intramolecular aspartimide is then spontaneously formed.

Since NmaM is predicted to be a PIMT homolog, we hypothesized that NmaM methylates a specific Asp side chain, followed by nucleophilic attack of the adjacent amide to form an aspartimide (Figure 3f). The masses of these two new species agreed with this hypothesis, as one was putatively methylated (+14 Da) and the other one was putatively aspartimidylated (−18 Da) (Figure 3d). Additionally, to confirm that the SUMO tag did not interfere with the post-translational modification, we coexpressed His₆-NmaA with NmaM, and also observed the emergence of the putative methylated and aspartimidylated species (Figure S8). Since the SUMO tag greatly improved the expression level of NmaA, we used His₆-SUMO-NmaA for all subsequent experiments.

To determine the kinetics of the reaction, we conducted co-expressions of His₆-SUMO-NmaA and NmaM at additional various time points (1 h, 2 h, 3 h, 4 h and 40 h). We observed a faster accumulation of the putative methylated species than the putative aspartimidylated species at shorter time points (Figure S9). As mentioned earlier, after 20 h of expression, we detected a ratio of approximately 0.4:0.2:0.4 (the starting material is normalized to 1) among the putatively aspartimidylated, unmodified and methylated NmaA (Figure 3c). With a longer expression time (40 h), the ratio shifted to approximately 0.5:0.3:0.2 among putatively aspartimidylated, unmodified and methylated species (Figure 3d). Since the putatively aspartimidylated species accumulated when longer expression time was allowed, we considered this species to be the intended product in vivo, similar to what we observed in graspetides and lasso peptides.^15,17,21 We named this imiditide mNmaA^M following the nomenclature used in our study on amycolimiditide, where m stands for modified, and superscripted M stands for the methyltransferase.²¹ Our data indicated that the rate of methylation was faster than the rate of aspartimidylation for the biosynthesis of mNmaA^M (Figure S9), also observed in the graspetide and lasso peptide aspartimidylation reactions.^15,17,21

Since NmaM is homologous to PIMT, we wondered whether Pcm, the housekeeping PIMT in E. coli, could alter the intended product in vivo. To investigate this, we constructed E. coli BL21(DE3) ΔpcmΔslyD strain, and expressed mNmaA^M using this strain. The product distribution remained the same using either E. coli BL21(DE3) ΔpcmΔslyD or E. coli BL21(DE3) ΔslyD, suggesting that Pcm was not involved in the process (Figure S10). We used E. coli BL21(DE3) ΔslyD for all subsequent experiments.

To determine the effect of the stoichiometry between NmaA and NmaM on the biosynthesis of mNmaA^M, we constructed a plasmid where the precursor protein NmaA was placed under an IPTG-inducible T5 promoter and NmaM was placed under the constitutive promoter native to the biosynthetic enzymes needed for lasso peptide microcin J25 biosynthesis.^15,24,25 Expressing NmaM constitutively resulted in a significant decrease in the protein expression level, which led to only a small fraction of NmaA being modified (Figure S11). It also suggests that NmaM may catalyze only one or a few turnovers of NmaA.

mNmaA^M Contains an Aspartimide Moiety

Based on our previous studies on aspartimide-containing lasso peptides and graspetides, we hypothesized that mNmaA^M also contains an aspartimide moiety (Figure 1a, 3d). Aspartimides are often considered as a reaction intermediate catalyzed by canonical PIMTs.²⁶ However, heterologous expression of mNmaA^M shows a slow accumulation of the putative aspartimidylated product in vivo, illustrating the relative stability of the moiety in this linear peptide when produced in bacteria. To validate that mNmaA^M indeed contains an aspartimide moiety, we reacted 0.1 mM mNmaA^M species (a mixture of unmodified peptide, methylated peptide, and aspartimidylated peptide, Figure 3d) with 2M hydrazine in 50 mM Tris-HCl buffer at pH 8. The reaction mixture was incubated at room temperature for 30 mins. If mNmaA^M contains an aspartimide, a mass shift of 32 Da was expected (Figure 4a,b). The incubation resulted in species with a +32 Da mass shift, confirming the presence of an aspartimide in the mNmaA^M mixture.

Figure 4: mNmaA^M Contains an Aspartimide Moiety.

a) Scheme of hydrazinolysis and hydrolysis of aspartimide. Aspartimides can react with hydrazine at room temperature for 30 mins to form hydrazides, resulting in an increase of 32 Da. In addition, aspartimides can also react with water in neutral or weak alkaline conditions (pH 7–8), forming a mixture of aspartate and isoaspartate. b) Mass spectrum of mNmaA^M reacted with hydrazine. A +32 Da mass shift was observed upon reactions with hydrazine, demonstrating mNmaA^M contains aspartimide. c) Hydrolyzed mNmaA^M contains isoaspartate. EIC spectra of hydrolyzed mNmaA^M (10–46) and unmodified NmaA (10–46) are shown in black and blue, respectively. mNmaA^M (10–46) had a retention time of 3.8 min. After treating this sample with human PIMT and SAM, the isoaspartate residue was reverted back to aspartate, as suggested by the same retention time as NmaA (10–46).

Aspartimides are known to be labile and can be hydrolyzed at neutral pH in buffer solutions (Figure 4a). Therefore, to investigate the hydrolyzed products of mNmaA^M, we incubated it in 50 mM Tris-HCl, pH 8 at room temperature for 16 h, and then digested the protein with endopeptidase GluC. This digestion generated a C-terminal 37-aa fragment with a missed GluC digest site at E31 in the NmaA sequence (Figure 3c), which we refer to as hydrolyzed mNmaA^M (10–46). LC-MS analysis of this species revealed that the aspartimide was indeed hydrolyzed under this reaction condition. Moreover, the hydrolyzed mNmaA^M(10–46) had a retention time of 3.8 min, while the GluC-digested unmodified His₆-SUMO-NmaA control NmaA(10–46) had a retention time of 4 min (Figure 4c). This observation suggested that the aspartimide in mNmaA^M is converted regioselectively to form a β-amino acid, isoaspartate (Figure 4a). To confirm this hypothesis, we reacted this species with human PIMT in vitro, as human PIMT can revert isoaspartate residues back to aspartate via aspartimide. Indeed, the human PIMT recognized this species, fully converting the species to a retention time that matched with the NmaA(10–46) control, which only contained the proteinogenic α-amino acids (Figure 4c).

Next, we sought to determine the location of the aspartimide in mNmaA^M. The NmaA sequence contains six Asp residues (Figure 3c). However, only 4 Asp residues can be aspartimidylated, as P8 and P18 lack the amide proton to form aspartimides with the D7 and D17 sidechains, respectively. We first digested mNmaA^M with endoproteinase GluC, which generated a C-terminal 37-aa peptide bearing the modification, which we refer to as mNmaA^M(10–46). Subjecting this peptide to MS/MS fragmentation (Figure S12) revealed that the modification was located between Gln22 and His29. However, because this region was located right in the middle of the fragment, the b/y-ion coverage was not sufficient for suggesting the exact location of the aspartimide. Therefore, to determine the exact location, we generated D23N, D25N, and D28N variants. Coexpressing these variants with NmaM suggested that neither D23N nor D25N variant affect the aspartimide formation (Figure S13). In comparison, substituting D28 to N resulted in no modification on the precursor (Figure S13), indicating that aspartimide formed between the sidechain of D28 and the backbone amide of H29.

To provide spectroscopic evidence of aspartimide formation, we acquired the FTIR spectrum of the GluC digested mNmaA^M D23E variant. The mNmaA^M D23E variant can be modified equally as well as the wild-type substrate (Figure S14). The GluC digestion of the mNmaA^M D23E variant generates a 23 aa C-terminal fragment that can be purified to homogeneity using HPLC. We denote this fragment as mNmaA^M (24–46). The FTIR spectrum of mNmaA^M (24–46) has an additional shoulder peak at 1708 cm⁻¹ compared to the unmodified NmaA (24–46) control (Figure S15).^27,28 To validate that this peak corresponds to an aspartimide formation, we also acquired FTIR spectra of graspetides amycolimiditide and pre-amycolimiditide reported by our previous studies.²¹ Both amycolimiditide and pre-amycolimiditide have 4 ester linkages, but amycolimiditide has an additional aspartimide moiety as shown by its solution NMR structure (PDB: 8DYM).²¹ Indeed, both spectra showed a peak at 1742 cm⁻¹, indicating the presence of esters. Additionally, the FTIR spectrum for amycolimiditide also contains an additional peak at 1708 cm⁻¹, indicating this peak corresponds to aspartimide formation (Figure S15).

His is the Preferred Residue C-terminal to the Aspartimide in mNmaA^M

Clarke and coworkers have reported several experimental studies on aspartimide formation from aspartyl peptides, suggesting that the sidechain of the residue that contributes its backbone to form the aspartimide (i.e. the n + 1 residue) greatly affects the rate of aspartimide formation.^26,29 In lasso peptides with an aspartimide modification, the n + 1 residue is highly conserved to be Thr, and our studies have shown that the n + 1 residue in these peptides plays a major role in determining the rate of methylation and the subsequent aspartimide formation.¹⁵ Moreover, we have recently reported that the n + 1 residue, Thr7, in lasso peptide lihuanodin also controls the regioselectivity of the aspartimide hydrolysis and subsequently the threadedness of the lasso peptide.¹⁶ In comparison, the n + 1 residue for aspartimides in lanthipeptides and graspetides are predominantly Gly,^13,17,18 which has been shown to be the preferred residue for spontaneous aspartimide formation in both computational and experimental studies. ^21,30

As illustrated above, in mNmaA^M, the n + 1 residue is His29, which is uncommon in predicted imiditide precursors (Figure 1b) as well as other PIMT-associated RiPPs. To probe whether histidine can serve as an effective n + 1 residue for methylation and aspartimide formation in mNmaA^M biosynthesis compared to the well-characterized Thr and Gly, we carried out site-directed mutagenesis at this position. We first constructed the H29T variant to mimic the aspartimidylation site seen in lasso peptides cellulonodin-2 and lihuanodin (Figure 5). Coexpressing this variant with NmaM at room temperature for 20 h showed an increase in the accumulated methylated species compared to the wild-type peptide. This observation suggests that having Thr as the n + 1 residue in mNmaA^M leads to a fast methylation step, followed by a relatively slower aspartimide formation step. This observation is also supported by the rate measurements of aspartimide formation in lihuanodin.¹⁶ Nevertheless, the percentage of aspartimidylated product formed in H29T variant was lower than the wild-type under the same heterologous expression condition. Additionally, we constructed the H29G variant to resemble the aspartimides in lanthipeptides and graspetides. The coexpression of this variant with NmaM showed nearly no accumulation of the methylated species (Figure 5), suggesting that aspartimide formation happened readily as soon as the methylated species was formed. This result also agreed with the heterologous expressions of aspartimide-containing graspetides, where having Gly as the n + 1 residue led to no accumulations of methylated species.^17,21 Similarly, the amount of aspartimidylated H29G variant was also lower compared to the wild-type peptide, suggesting that the H29G variant was not as good of a substrate for NmaM. These mutagenesis measurements collectively illustrated that H29 in wild-type mNmaA^M is the preferred n + 1 residue for for maximizing aspartimide formation compared to the commonly adopted Thr or Gly in other RiPPs. As the n + 1 residue for aspartimidylation has been constrained to mainly Gly and Thr in previous studies, this finding helps to expand the list of potential n + 1 residues for predicting the site of aspartimidylation bioinformatically.

Figure 5: Effects of Single Amino Acid Substitutions of n + 1 Residue on Methylation and Aspartimidylation.

Both mNmaA^M H29T and H29G variants were heterologously expressed under the same conditions as the wild-type substrate (20 h). The H29T substitution caused a larger buildup of the methylated species compared to the wild-type, showing that the rate of aspartimidylation was slower than methylation. In addition, less aspartimidylated products were produced for the H29T variant compared to the wild-type. On the other hand, the H29G substitution led to no buildup in the methylated species, showing that rate of aspartimidylation was much faster than methylation when Gly was the n + 1 residue. Similarly, less aspartimidylated product was observed, showing the H29G variant was not as good of a substrate compared to the wildtype peptide. Collectively, His29 is the preferred n + 1 residue for aspartimidylation for NmaA.

Portions of the Leader and Follower Peptides of mNmaA^M are Dispensable

It is fascinating how NmaM achieves specificity for aspartimidylating Asp28 on NmaA out of an abundance of aspartate residues present not just in the precursor peptide but in the rest of the bacterial proteome. To probe the substrate specificity of NmaM, we first attempted to decipher the elements on the precursor peptide that are important for substrate recognition. While the core peptide of mNmaA^M is not well-defined, there are 27 aa N-terminal to the aspartimidylation site and 18 aa C-terminal to it. These portions of NmaA comprise the leader and follower peptides respectively with the caveat that any subsequent proteolytic processing of mNma^M is unknown. We first constructed variants with truncations of the NmaA leader peptide, removing 5 aa at a time, up to 20 aa in total (Figure 6a). The Met residue coded by the start codon was not included as the precursor was fused to a His₆-SUMO tag. Coexpressing these variants with NmaM showed that removing the first 5 and 10 amino acid of the leader had no effect on aspartimidylation of NmaA (Figure 6b). In contrast, a decrease in aspartimidylation of NmaA was noted when the first 15 aa of the leader peptide was removed. Removing a total of 20 aa of the leader peptide resulted in a more severe drop in aspartimidylated product formation (Figure 6b). It is worth noting that there is a high content of basic amino acids in the region of NmaA(12–21), which can play an important role in substrate recognition through electrostatic interactions. Nevertheless, the aspartimidylation could still occur even when the first 20 aa of leader peptide was removed, showing that much of the leader peptide of mNmaA^M is dispensable.

Figure 6: Effects of Leader and Follower Truncations on mNmaA^M Biosynthesis.

a) Sequence of His₆-SUMO-NmaA and schematic of the leader and follower truncation variants. The site of modification, Asp28, is colored blue. b) Much of the leader is dispensable for mNmaA^M biosynthesis. 5 aas were removed at a time. Removing the first 10 aa in the leader had minimal effect on aspartimidylation. Removing the next 10 aa was detrimental to the extent of aspartimidylation, likely due to the removal of basic amino acids in this region. c) Some of the follower peptide is also dispensable for mNmaA^M biosynthesis. Removing the 5 C-terminal aa of the follower sequence had minimal effect on aspartimidylation. Removing 10 C-terminal aa significantly reduced the extent of the reaction, showing this region was important for substrate recognition.

In some cases of RiPP biosynthesis, such as bottromycin^31,32 and cyanobactin,^33,34 a C-terminal follower sequence to the core peptide has been shown to be important for RiPP maturation. Since the modification on mNmaA^M occurs in the middle of the precursor sequence, we asked whether a C-terminal recognition sequence exists for imiditide biosynthesis. We constructed variants with 5 aa and 10 aa truncations of the precursor from the C-terminus (Figure 6a). Removing the last 5 aa from the C-terminal domain had minimal effect on aspartimidylation of NmaA (Figure 6c). In comparison, removing 10 aa from the C-terminus was detrimental to the extent of aspartimidylation on NmaA (Figure 6c). Therefore, our data suggest that at least 5 aa of the follower sequence is also dispensable for aspartimidylation to occur in the mNmaA^M biosynthesis.

Importance of the C-terminal Domain of NmaM in Substrate Recognition

Our methodology of genome mining suggests that the O-methyltransferases that carry out class-defining aspartimidylation in imiditide BGCs share conserved C-terminal domains that are absent in canonical PIMTs (Figure 2a, S2). In our previous work on the PIMT homologues that aspartimidylate lasso peptides, we showed that this C-terminal extension to the canonical PIMT is important for the substrate recognition.¹⁵ Therefore, to examine the importance of the identified C-terminal motif (Figure 2a) in substrate recognition, we first constructed the 50 aa C-terminal truncated variant of NmaM (NmaMΔC50) that lacked this motif. We coexpressed NmaMΔC50 with His₆-SUMO-NmaA, and observed no modification on the precursor (Figure S16). We also observed no non-specific pulldown of NmaMΔC50 (Figure S16), suggesting either that the protein was not well-expressed or that the C-terminal 50 aa of NmaM contained the motif that non-specifically bound to the nickel resin. We confirmed that NmaMΔC50 expressed well but no longer bound non-specifically to nickel resin (Figure S16). The C-terminal 50 aa of NmaM is rich in His residues, including an HHQH sequence (Figure S16) that may result in its non-specific binding to nickel resin.

In addition to the conserved motif, we noticed that the ~150 aa C-terminal domain of NmaM is rich in negatively charged residues and has an overall pI value of 5.4 (Figure S17). As demonstrated above, the leader and follower peptides in the precursor peptide are rich in positively charged residues, and removing these residues is detrimental to substrate recognition (Figure 6b). Therefore, we hypothesized that the substrate recognition is likely achieved through charge-charge interactions. To test this hypothesis, we first identified a region in C-terminal domain of NmaM that has a very concentrated presence of negatively charged residues, namely DEDGD from position 289–293 in NmaM. Then, we constructed a NmaM variant where this region was substituted with SGSGS to remove 4 negative charges from the C-terminal domain, which we named NmaM(SGSGS). Coexpressing the NmaM(SGSGS) variant with His₆-SUMO-NmaA resulted in predominately unmodified His₆-SUMO-NmaA, indicating that these negative charged residues in NmaM are important for mNmaA^M maturation (Figure S18). Untagged NmaM(SGSGS) was still pulled down by the nickel resin in this coexpression (Figure S18), and we confirmed that when expressed on its own NmaM(SGSGS) also binds non-specifically to nickel resin (Figure S18).

AlphaFold Model Predicts Mode of Substrate Recognition by NmaM

To further understand the nature of the substrate specificity of NmaM on NmaA, we generated an AlphaFold model of the NmaA-NmaM complex using ColabFold,^35,36 assuming a 1:1 stoichiometry between the two proteins (Figure 7a, b). To our surprise, the model nearly perfectly supports our experimental results. Even though the precursor NmaA is predicted to be a random coil, various hydrogen bonding as well as charge-charge interactions ensure the correct positioning and substrate specificity between NmaA and NmaM (Figure S19). The side chain of Asp28 sits right outside of the predicted SAM binding pocket in NmaM, aligning with our observation that Asp28 is the site of modification (Figure 7c). The C-terminal domain of NmaM binds to both the leader and core portion of the leader peptide, supporting this additional C-terminal domain compared to canonical PIMTs in NmaM is essential for substrate recognition (Figure 7b). Additionally, the N-terminal 50 aa domain (Figure S2) that is absent in canonical PIMTs seems to be involved in binding to the core peptide (Figure 7b). The side chain of the n + 1 residue His29 appears to interact with F284 in NmaM through a CH−π interaction (Figure S20),³⁷ further promoting substrate specificity. For the NmaA leader peptide, the model suggests that NmaA(1–10) does not interact with NmaM, agreeing with our experimental result that removing NmaA (1–10) had minimal effect on mNmaA^M maturation. On the other hand, NmaA(11–20) binds to the C-terminal domain of NmaM, and experimental results showed that removing NmaA (11–20) had deleterious effect on the extent of modification on NmaA (Figure 6). In contrast, the model indicates the follower sequence doesn’t interact with NmaM. However, based on experimental results above, we hypothesized that the follower sequence that is rich in positively charged residues should be engaged with the acidic loop DEDGD on NmaM (289–293), as they are spatially adjacent based on the model (Figure 7a, Figure S21).

Figure 7: AlphaFold Model of NmaA-NmaM using ColabFold. ^35,36.

a) AlphaFold model of NmaA-NmaM complex. A stoichiometry of 1:1 between NmaA and NmaM is assumed. The leader, core, and follower peptide of NmaA are colored pink, purple and gold, respectively. The N-terminal domain and the C-terminal domain of NmaM are colored light blue and cyan, respectively. The rest of NmaM is colored green. The sidechain of Asp28 is shown in sticks. b) Space filling model of NmaA-NmaM with the charge profile indicated in red and blue for the interacting region. The leader peptide binds to the C-terminal domain of NmaM. The core sequence is locked in near the active site of NmaM. The follower sequence was not engaged with NmaM based on the model, but our experimental data suggest that the follower peptide also binds with the C-terminal domain of NmaM. c) Asp28 sits right outside of the SAM binding pocket of NmaM (red), consistent with it being the site of modification.

Considering how well the AlphaFold prediction model agrees with our experimental data acquired for the mNmaA^M biosynthesis, we advocate that AlphaFold can serve as a powerful tool guiding novel RiPPs discovery. The putative precursor peptide can be readily tested to see whether it can serve as a substrate for nearby encoded putative RiPP enzymes when compared to experimental results based on previous studies.³⁸ Additionally, AlphaFold can also help indicate which reaction happens first when several enzymes are involved in the biosynthetic gene cluster (Figure S22–24).^17,38 We show that AlphaFold can correctly determine the order of modification in both aspartimidylated graspetides (Figure S22) and lasso peptides (Figure S23). We further show that AlphaFold correctly docks the lasso peptide precursor McjA near the catalytic triad of cysteine protease McjB in a McjB/McjC complex (Figure S24). This cleavage step must occur before the lasso cyclase McjC transforms the linear core peptide into a lasso peptide.

Conclusion

In this work we have described the genome mining and discovery of a novel family of RiPPs, imiditides. Utilizing the fact that PIMT homologs install the aspartimide moiety on peptides from various RiPP families, we used the conserved motif in the extended C-terminal domain of these PIMT homologs as a bioinformatic seed to identify 670 imiditide BGCs bioinformatically. Our genome mining algorithm suggests that imiditides are widely distributed in Gram-positive bacterial genomes. In addition, we report the discovery of the founding member of the imiditide family, mNmaA^M, from Nonomuraea maritima. The aspartimide in mNmaA^M exhibits some stability; the aspartimidylated species accumulated and was the major product in the heterologous expression in E. coli. In contrast to other RiPP-associated PIMT homologs that recognize constrained peptides as substrates (i.e. lasso peptides, graspetides, lanthipeptides),^13,15,17,21 imiditide-associated PIMT homologs recognize specific Asp residues on the linear precursor rather than on a matured RiPP. In mNmaA^M biosynthesis, the modifying enzyme NmaM methylates a specific Asp residue in the precursor sequence NmaA through extensive charge-charge interactions and hydrogen bonding networks. This specificity is crucial for the fitness of the host as random methylation on Asp residues will affect functions of numerous proteins and constantly consume precious SAM molecules. We note that both methylation and aspartimidylation of NmaA appear to be much slower than what we observed for lasso peptides and graspetides. The stability of aspartimides in RiPPs is explained by a kinetic framework;¹⁶ if the rate of aspartimide hydrolysis is small relative to the rate of aspartimide formation, the aspartimide product will persist. Based on the data we have acquired here, the aspartimide in mNmaA^M does not appear as stable as those in the lasso peptides lihuanodin and cellulonodin-2 or the graspetides fuscimiditide and amycolimiditide.^15,17,21 This variation in aspartimide stability likely also contributes to the as yet unknown bioactivity of these molecules.

The discovery of imiditides provided more insights in various roles of the aspartimide formation in cells. Aspartimides have been long considered as the unstable intermediate to isoaspartate resulted from spontaneous protein aging.^39–41 Recently, Woo and coworkers reported that C-terminal cyclic imide stemming from intramolecular cyclization of glutamine or asparagine residues is a target for the ubiquitin E3 ligase adapter cereblon (CRBN) for protein degradation.⁴² Our findings illustrate another route of aspartimide formation in vivo, through direct enzymatic methylation of specific aspartate residues by PIMT homologs. While the role of aspartimide formation in imiditides remains unknown, future discovery of aspartimides in proteomes should also consider other functionalities rather than solely regarding them as protein aging and degradation signals.

Moreover, we were pleased by the accurate AlphaFold prediction of NmaA-NmaM complex, which predicts our experimental results. It not only predicts that NmaA will be a substrate for NmaM, but also correctly predicts the site of modification. Other than the fact that the follower sequence should interact with the acidic loop in the C-terminal domain of NmaM based on our experimental results (Figure S21), it predicts the binding interface between NmaA and NmaM extremely well. We used AlphaFold as a validation tool in this study since much of our experimental results were gathered before the release of ColabFold. However, our results suggest that AlphaFold can be a powerful tool in predicting whether a biosynthetic gene cluster can form a RiPP, and can help design experiments in validating the gene cluster. In summary, we have discovered a novel RiPP family in which aspartimide formation is the class-defining modification. Future studies will be directed towards the discovery of other novel RiPPs with the aspartimide modification, as well as understanding the role of aspartimides in the bioactivity of these RiPPs.