Rule-based omics mining reveals antimicrobial macrocyclic peptides against drug-resistant clinical isolates

SPECO-based genome mining expands Avi(Me)Cys-containing RiPP families

Ribosomally synthesized post-translationally modified peptides are biosynthesized using a conserved mechanism in which post-translational modification (PTM) enzymes modify a ribosomal precursor peptide, followed by protease-mediated trimming to release the mature product¹⁸. Similarly, ACyP biosynthesis relies on a flavoprotein (Pfam number: PF02441) that catalyzes the oxidative decarboxylation of the C-terminal cysteine residue of the precursor. This process leads to the formation of an enethiol group, which can react with unsaturated amino acids, facilitated by oxidative thioaldehyde formation of the cysteine residue and parallel decarboxylation through tautomerization^{30,41,42,43,44}. The requirement of the flavoprotein for thioenol formation and the indispensability of the C-terminal cysteine residue in precursor peptides provide a unified rule for genome mining of ACyP-encoding BGCs (Fig. 1b). We noticed that ACyPs were mainly found from firmicutes and actinobacteria, the genomes from these two phyla were then selected and analyzed.

Following this rule, we analyzed 21,911 genomes of actinobacteria and 30,666 genomes of firmicutes for the ACyP BGC identification using the SPECO pipeline²⁴ (Fig. 1b). This analysis uncovered 1172 unique putative precursor-flavoprotein pairs (Supplementary data 1). Our findings indicate that most putative precursors were not clustered with known precursors (Fig. 2a) as shown in the precursor sequence similarity network (SSN), highlighting the unexplored chemical space of ACyPs. As expected, known precursors such as epidermin, thioviridamide, lexapeptide, and cypemycin precursors were included in the top rank (Fig. 2a and S2). We further analyzed the putative BGCs through sequence logos (Supplementary data 2) and phylogenetic analysis (Supplementary data 3) to prioritize candidates with unknown precursors or tailoring enzymes. Among the identified BGC families, one family caught our attention due to the presence of two oxidoreductases, which is uncommon in ACyP BGCs (Fig. 2b).  Further analysis of the genomic context revealed that multiple modifying enzymes were well conserved in this family. We named this family the mat family and hypothesized that the additional oxidoreductase might introduce alternative modifications to the precursor peptide. We were also drawn to the largest cluster in the SSN, which contains pristinin A3⁴⁵, and have designated this cluster as sis family. Similar to the mat family, the sequence logos of the sis family exhibited a characteristic C-terminal Avi(Me)Cys ring formation motif (T-X1-X2-X3-X4-C, T is threonine, C is cysteine, X represents a variable residue), and an additional conserved cysteine at the N-terminus (Fig. 2c). However, analysis of the precursor sequence logo for the sis family revealed an approximately 50% occurrence of Cys residues in the middle of the core region, which could be further categorized into two subfamilies with either two or three Cys residues (Fig. 2c). We propose that a single maturation system within the sis family could generate ACyPs with diverse thioether rings, owing to the propensity of Cys residues to react with Dha or Dhb. Overall, our analysis prioritized potential BGC candidates for exploring the chemical diversity of Avi(Me)Cys-containing RiPPs.

Linking Avi(Me)Cys biosynthetic gene clusters with peptide antibiotics

To discover new ACyPs from the BGCs of interest, we first focused on the mat cluster from Streptomyces leeuwenhoekii DSM42122 (Fig. 3a). This mat cluster contains several enzymes responsible for the formation of the (methyl)lanthionine ((Me)Lan) ring, including MatKYX enzymes, which exhibit similarities to the class V lanthionine (Lan) synthetase LxmKYX found in the lexapeptide BGC^33,46. Other enzymes in the cluster include MatM, which functions as a methyltransferase, and MatT1-T3 and MatP, which act as transporters and peptidase, respectively. Additionally, MatR1 and MatR2 are putative oxidoreductases involved in the hydrogenation of Dha or Dhb^33,36. These intricate modifications, encompassing decarboxylation, dehydration, thioether cyclization, hydrogenation, and methylation, give rise to a diverse range of possible monoisotopic masses for both intact and fragmented precursors (Fig. 3b, Supplementary data 4). Consequently, the identification of specific metabolites in the metabolic analysis of crude extracts can be challenging, as numerous mass-to-charge ratio (m/z) signals are observed, making it difficult to pinpoint the metabolites of interest.

To pinpoint the BGC-encoded ACyP from metabolites, the wild-type strain Streptomyces leeuwenhoekii DSM42122 was cultivated for 7 days and the high-resolution mass spectrometry (HRMS) data of the crude extract was collected. We then matched calculated mass data with ly detected metabolomic data to target natural products (Fig. 3b). This process involved enumerating the m/z values of theoretical peptide fragments with various modifications, which were predicted based on post-translational modification rules (Fig. 3b). We then compared these theoretical values with the HRMS data to find matches. For instance, UPLC-HRMS analysis of the culture extract showed typical peptidic signals and a mass matching pipeline revealed a hit of [M + 2H]²⁺ = 817.8949 (Fig. 3c, Supplementary note, Supplementary data 4). Further mass calculation suggested that the core region of the precursor peptide “TTVPCTVASFATGYFSC” (calculated [M + 2H]²⁺ = 817.8965, Δppm = 2.0) underwent six dehydrations (−6H₂O), one decarboxylation (−1CO₂H₂), three hydrogenations (+6H), and two methylations (+2CH₂). To solidify the association of the metabolite of [M + 2H]²⁺ = 817.8949 (1) with the mat BGC, we cloned and heterologously expressed the entire mat BGC in Streptomyces albus J1074 (Table S3). UPLC-HRMS analysis revealed a new peak with [M + 2H]²⁺ = 817.8963 in the S. albus/pCZMAT by comparison with the host strain harboring the empty pJTU2554 vector (Fig. 3c and S3). This result provided definitive evidence that the mat BGC encodes compound 1, with MatA as the crucial precursor for structural elucidation. To delve deeper into the bioactivity of the mat-encoded product, we conducted an antibacterial assay utilizing a fermentation culture of the heterologous expression strain. The findings demonstrated that S. albus/pCZMAT exhibited inhibitory effects on the growth of Micrococcus luteus (Fig. 3e), establishing a direct correlation between the biosynthetic capacity and the antibiotic potential of the ACyPs.

With the confirmed core peptide sequence and calculated modifications in place, we aligned tandem mass data with calculated monoisotopic mass data of fragment ions (Figure S4, Supplementary data 5). The results showed that b5 to b11, y6, and y8 to y11 fragments with modifications matched well with calculated ones. The matched modifications at ions b11 (−4H₂O_ + 2CH₂_ + 4H) and y6 (−2H₂O_−1CO₂H₂_ + 2H) are consistent with the total of six dehydrations (−6H₂O) and C-terminal decarboxylation (Fig. 3b). Upon comparison of b5 to b6 and b8 to b9 ions, it was observed that two hydrogenated residues were localized at Thr6 and Ser9, respectively. The absence of fragment ions in the Thr1-Cys5 and Thr12-Cys17 motifs strongly suggests the formation of Thr1-to-Cys5 and Thr12-to-Cys17 linkages, with Ser16 being the only remaining residue that can be hydrogenated. It is noteworthy that hydrogenation during RiPP biosynthesis can potentially generate D-amino acids, as previously reported^33,47,48. Based on the involvement of biosynthetic enzymes like methyltransferase and decarboxylase, we proposed a tentative structure for 1, which we named massatide A (Fig. 3b).

To further validate the structure of compound 1, we conducted a large-scale fermentation of the S. albus/pCZMAT strain, and 5 mg of 1 was purified for 1D and 2D nuclear magnetic resonance (NMR) analysis (Figures S5–S12, Table S4). In the ¹H and ¹³C spectra, characteristic signals of the Phe and Tyr side chains and the modified residues aminobutyric acid (Abu), Dhb, and AviMeCys were identified. The C-S formation in the AviMeCys motif was supported by key HMBC correlations from Cys17*-Hβ (5.29 ppm, d, * denotes modification) to Thr12*-Cβ (45.3 ppm) and from Thr12*-Hβ (3.23 ppm) to Cys17*-Cβ (105.2 ppm) (Table S4 and Figure S12). A Z-geometry of the double bond in the AviMeCys residue was determined based on the corresponding ³J_H,H value of 8.5 Hz (Table S4). Additionally, an N,N-dimethyl MeLan was observed based on HSQC and HMBC spectra, which is rarely reported except for a lanthidin RiPP³⁶. To confirm absolute configurations for all the amino acid residues, we conducted advanced Marfey’s analysis⁴⁹ for 1 using L/D-FDLA (5-fluoro-2,4-dinitrophenyl)-L/D-(leucinamide). The results showed the presence of both L– and D-Ala, with D-Abu also observed (Table S8). Further investigation showed that Ala8 and Ala11 were genetically encoded and in the L-configurations. Based on these results, we proposed that Thr6, Ser9, and Ser16 were converted to D-Abu and D-Ala in 1, while the remaining unmodified residues all existed as L-configuration. To the best of our knowledge, massatide A (1) represents the first Avi(Me)Cys-containing peptide with D-amino acids in the C-terminal AviMeCys ring.

Most previously reported lanthipeptide BGCs have only one reductase for Dha/Dhb hydrogenation^33,47,48. However, in the mat BGC, both matR1 and matR2 encode LLM class oxidoreductases containing the conserved F420 binding domain, which belong to the LanJ_C family of lanthipeptide reductases³³. To determine the essentiality of both MatR1 and MatR2, we constructed a single gene knockout strain S. albus/pCZMAT_ΔR1. UPLC-HRMS data showed that compound 1a ([M + 2H]²⁺ = 816.8880) from the knockout strain, named massatide B, was 2 Da lighter than 1 (Fig. 3c). Tandem mass analysis suggested that Thr6 was converted to Dhb in massatide B (1a) instead of D-Abu in 1 (Fig. 3d and S13). Thus, we confirmed that in the mat BGC, MatR1 catalyzes D-Abu6 formation, while MatR2 may be responsible for both D-Ala9 and D-Ala16 formation.

Two crosslinking patterns catalyzed by one Avi(Me)Cys biosynthetic machinery

We next investigated the sis family, which is anticipated to produce ACyPs containing 2 and 3 thioether rings. This is due to the existence of precursor peptides containing 2 and 3 cysteine residues. We examined a sis BGC from Streptomyces kasugaensis DSM40819, which belongs to class V lanthipeptides with methylation and AviMeCys motif (Fig. 4a, Table S3). Two putative precursors, namely SisA1 and SisA2, contain three and two cysteines, respectively. This composition of cysteines suggests that SisA1 can potentially form a three-ring peptide, while SisA2 may give rise to a two-ring peptide. Utilizing the same PTM rule-guided metabolomic analysis workflow mentioned above, we pinpointed two putative peptide ion peaks corresponding to the precursors SisA1 (compound 2, observed [M + 3H]³⁺ = 1085.2083, calculated [M + 3H]³⁺ = 1085.2083, Δppm = 0.1) and SisA2 (compound 3, observed [M + 3H]³⁺ = 816.7481, calculated [M + 3H]³⁺ = 816.7487, Δppm = 0.7), respectively (Figure S14). The modifications of SisA1 occurred in the C-terminal 34-mer core region “TTYICASVAISRTSSVKCSAAASAISGATYEWTC”, including eleven dehydrations (−11H₂O), one decarboxylation (−1CO₂H₂), five hydrogenations (+10H), and two methylations (+2CH₂), resulting in the production of 2 (Supplementary data 6). Similarly, modifications of the SisA2 core peptide “STPACGAATVSWIVSQFSAKTVKDGC” involved seven dehydrations (−7H₂O), one decarboxylation (−1CO₂H₂), and three hydrogenations (+6H), leading to the formation of 3 (Supplementary data 6). To validate the ribosomal origins of compounds 2 and 3, we cloned and heterologously expressed the entire sis BGC in S. albus (Table S3). Comparative UPLC-HRMS analysis revealed two new peaks in S. albus/pCZSIS, i.e., [M + 3H]³⁺ = 1085.2076 and [M + 3H]³⁺ = 816.7487, which are respectively identical to 2 and 3 from the wild-type strain (Figure S14). The results confirmed that 2 and 3, named sistertide A1 and A2, were indeed encoded by the sis BGC. In the antibacterial assay using a fermentation culture of the heterologous expression strain, S. albus/pCZSIS exhibited growth inhibition against M. luteus, confirming the antibiotic potential of these sistertides (Fig. 4a).

Tandem mass analysis of sistertide A1 (2) fragments revealed N-terminal methylation and C-terminal decarboxylation (Fig. 4a and S15). Furthermore, the analysis indicated that Ser7, Ser11, Ser19, Ser23, and Ser26 respectively underwent a 16 Da mass loss, suggesting that they were first dehydrated (−H₂O) and then reduced (+2H). Notably, no fragment ions were observed around three cysteines, which implied that Cys5, Cys18, and Cys34 might cyclize with Thr1, Ser14, and Thr29 (Fig. 4a and S15). We next performed a large-scale fermentation of S. albus/pCZSIS to determine the crosslinking patterns and obtained 4 mg pure sistertide A1 (2) for NMR analysis (Figures S16–S23, Table S5). Characteristic signals of the side chains of Tyr, Trp, Thr, and modified residues Dha and Dhb were observed in 2. NMR analysis also showed that 2 contains an AviMeCys scaffold with a Z-geometry (³J_H,H 8.4 Hz) at the double bond, which was derived from the Thr29/Cys34 motif of the core peptide (Table S5). The presence of N,N-dimethyl MeLan was confirmed by key HMBC correlations (Figures S20 and S23). However, due to its low yield, we could not collect enough sistertide A2 (3) for NMR analysis. Mass calculation and tandem mass data of 3 suggested that the core peptide of SisA2 underwent dehydration, decarboxylation, hydrogenation, and ring formations, which led to the formation of one lanthionine ring and one AviMeCys ring (Fig. 4a and S24).

Sistertide A1 is methylated, while A2 lacks this modification despite being modified by the same enzymes. This suggests that the methyltransferase in the sis BGC may exhibit selectivity in substrate recognition. The presence of a putative oxidoreductase SisR in this BGC suggests possible D-amino acid formation in sistertides. Advanced Marfey’s analysis⁴⁹ for 2 revealed the presence of both L- and D-Ala in the structure (Table S9), with the D-Ala residues likely converted from Ser7, 11, 19, 23, and 26 in the precursor. The remaining unmodified residues were L-configurated. Similarly, for sistertide A2 (3), Ala11, Ala15, and Ala18 likely resulted from dehydration and reduction of Ser residues in the precursor, suggesting that they may also be D-Ala in the mature compound based on biosynthetic logic.

Additionally, within the sis family, we identified two other ACyPs encoded by a sis BGC homolog, the keb cluster from Streptomyces kebangsaanensis DSM42048 (Fig. 4b). Our initial analysis of the wild-type DSM42048 using the workflow revealed no discernible target peaks from the fermentation. Therefore, we hypothesized that the BGC might be either inactive or expressed at a very low level in the native host. Subsequently, we tried to heterologously express the entire keb BGC in Streptomyces albus, which resulted in the production of two new compounds, 4 (named kebanetide A1, [M + 3H]³⁺ = m/z 921.4331) and 5 (named kebanetide A2, [M + 3H]³⁺ = m/z 793.7395) (Fig. 4c and S25). Using mass calculation, tandem mass spectrum analysis (Supplementary data 7), ¹H NMR spectroscopy, and advanced Marfey’s analysis, we determined that kebanetide A1 (4) features one C-terminal AviMeCys ring and two thioether crosslinks, while kebanetide A2 (5) resembles sistertide A2 (3), containing a C-terminal AviMeCys ring and an N-terminal thioether ring (Fig. 4d and S26–S33, Table S6–S7 and S10–S11).

Although the leader peptide remains conserved across multiple precursors within the sis family, significant variations are observed in the core regions of precursors A1 and A2, particularly in terms of their length and the number of essential cysteine residues involved in ring formation. Notably, the AviMeCys ring of sistertide A1 is characterized by the presence of two bulky hydrophobic aromatic side chains (Tyr30 and Trp32) and one negatively charged residue (Glu31), while sistertide A2 features a positively charged lysine residue (Lys23). These substantial disparities in the core peptide imply that the tailoring enzymes within the sis BGC family display broad substrate selectivity. It is worth noting that the discovery of 42 BGCs in this family indicates that the “one tailoring enzyme system constructs multiple crosslinking patterns” are widespread among actinobacteria. RiPP biosynthetic cassettes with multiple precursors were found in two-component class II lantibiotics, such as lichenicidin⁵⁰ and haloduracin⁵¹. However, in contrast to the sis family, multiple LanM enzymes are encoded in these two-component lantibiotics BGCs to catalyze precursors in a one-LanM-one-precursor manner. The ribosomal peptide maturases in the sis family resemble promiscuous lanthipeptide synthetase ProcM, which has been widely used to generate cyclic peptide libraries⁵².

To explore the catalytic potential of the sis family maturase, we conducted a leader-core mixing and matching assay. We designed two chimeric precursor peptides (Syn1 and Syn2), each combining the KebA leader with the SisA core region, to investigate the substrate tolerance in ACyP biosynthesis (Fig. 4f and S34). Co-expression of these two chimeric precursors with the maturases from the keb BGC (constructs keb-syn1 and keb-syn2) led to the production of compound 2a, sistertide A2 (3) and 3a (Fig. 4g and S35–36). Tandem mass analysis supported that 3a is monomethylated 3 at the N-terminus (Figure S36), indicating the keb maturase can not only modify the core peptide of SisA2, but also generate new-to-nature analogs via methylation. HRMS data revealed that 2a is a new derivative compared to sistertide A1 (2). Further tandem mass spectrometry analysis indicated that the new modification was localized at the N-terminus (Figure S35). However, no new peaks were observed when chimeric precursors Syn1 and Syn2 were co-expressed with the sis maturases (constructs sis-syn1 and sis-syn2, Figure S34), implying that leader-maturase mismatching hinders core peptide modification. Taken together, these findings suggest that the tailoring enzymes from the keb BGC exhibited selectivity for leader peptide but tolerance for core peptide.

Potent antibacterial activity of isolated ACyPs

We next employed the Kirby-Bauer assay to evaluate the antibacterial properties of isolated ACyPs from the mat and sis families. Massatides A (1) and B (1a) showed strong inhibitory activity against Staphylococcus aureus and Micrococcus luteus (Fig. 3e). Similarly, kebanetides A1 (4) and A2 (5) exhibited growth inhibition against M. luteus, as shown in Fig. 4e. Moreover, we determined the minimal inhibitory concentrations (MICs) of massatide A, massatide B, sistertide A1, kebanetide A1, and A2 against seven bacterial strains to assess their bioactivity, as detailed in Table 1. Massatide A demonstrated a broad-spectrum antibacterial activity against gram-positive bacteria, with lower MICs than nisin and comparable activity to vancomycin. Its antimicrobial activities against S. aureus and M. luteus even surpassed those of vancomycin, highlighting its potential as a candidate for gram-positive antibiotics. The activity of massatide B was similar to that of massatide A, with only a slightly larger MIC, indicating that the presence of D-Abu6 has a minimal contribution to the antibacterial activity. Kebanetide A1 and A2 demonstrated weaker bioactivity than massatides but performed better than nisin against E. faecalis and E. faecium. The crude extract of the sistertides producer strain exhibited a zone of inhibition (Fig. 4a), even though sistertide A1 alone did not display significant inhibitory activity. This suggests that the combined action of sistertide A1 and A2 is necessary for their inhibitory activity. No bioactivity was detected against gram-negative bacteria, indicating that these ACyPs specifically target gram-positive bacteria.

Efficacy of massatide A in vitro and in vivo

In light of the promising antibacterial potential of massatide A, we conducted extensive research to investigate its antibacterial efficacy both in vitro and in vivo. The growth kinetics and time-dependent killing assay revealed that massatide A was bactericidal against S. aureus and was superior to vancomycin, the first line of defense antibiotic, in killing early exponential phase populations (Fig. 5a, c). We noticed that massatide A did not result in lysis of the cell culture, which is the same as vancomycin (Fig. 5b). In vitro stability assay showed that massatide A is resistant to trypsin, chymotrypsin, and many other proteases (Figure S37) and highly stable under high temperatures (Figure S38). The subsequent antibacterial assays yielded remarkable results, demonstrating that massatide A exhibited excellent activity against a range of clinically isolated resistant pathogens, as outlined in Table 2. Its potency against tested gram-positive pathogens, including vancomycin intermediate-resistant strains, remained below 4 μg/mL. Particularly noteworthy was its exceptional performance against linezolid-resistant S. aureus 20-1 and methicillin-resistant S. aureus ATCC43300, with a MIC of only 0.25 μg/mL. Over a span of 25 days, serial passaging in subinhibitory concentrations of massatide A resulted in only a slight increase in the MIC to S. aureus, from 0.5 to 4 μg/mL. No significant change in MIC was observed during the subsequent 19 days after day 6, suggesting that S. aureus did not develop significant resistance to this compound. This finding further supports the potential of massatide A as an effective antibacterial agent.

Massatide A exhibits cytotoxicity against Hela cells, with an IC50 value of 136.3 μg/mL, while massatide B did not exhibit any significant cytotoxicity against Hela cells. However, both massatide A and B demonstrated moderate toxicity against the human cell line Hek293T (Figure S39). To further assess its safety profile, an acute toxicity study was conducted in mice, where massatide A was administered at doses ranging from 10-50 mg/kg. After 5 days of injection, all treated mice survived without any signs of acute toxicity (Fig. 5e). These findings suggest that massatide A has a favorable safety profile, highlighting its potential as a relatively safe and effective antibacterial agent.

Expanding upon its potent antibacterial activity, low in vivo toxicity, and relatively low risk of resistance, we conducted further evaluations to assess the efficacy of massatide A in a mouse septicemia model. Mice were intraperitoneally infected with 7.2 × 10⁹c.f.u. of MRSA ATCC43300, and lethality was observed after 7 days of infection (Fig. 5f). Remarkably, the survival rate of mice significantly improved to 80% at a single dose of 10 mg/kg of massatide A, and further increased to 100% at a higher dosage of 50 mg/kg, in contrast to the control group with no survivors. The findings strongly suggest the efficacy of massatide A in vivo, showcasing its potential as a therapeutic agent against gram-positive pathogens. With its remarkable efficacy and favorable safety profile, massatide A is a candidate for further research and development in antibacterial treatment. These results underscore the tremendous significance of ACyP macrocyclic peptides in the discovery of new antibiotics, particularly in addressing the critical need for effective antibacterial agents to combat drug-resistant bacteria.