Introduction

The extensive use of conventional antibiotics has led to the emergence of antimicrobial resistance that is spreading faster than the development of new antibiotics, posing a great medical challenge in the 21st century (Ling et al, 2015; Sierra et al, 2017). To tackle this challenge, one needs to explore various new sources of antibiotics with diverse chemical structures and modes of action to delay the occurrence of resistance. As the effectors of the innate immunity of multicellular organisms, antimicrobial peptides (AMPs), also known as peptide antibiotics (Hancock, 1997), provide the first line of defense against microbial infection. Because of their evolutionary success as a defensive weapon of multicellular organisms across 1.8-billion-year history (Futuyma & Kirkpatrick, 2017), these molecules have been considered as a class of promising alternatives of conventional anti-infective agents (Zasloff, 2002, 2019; Fox, 2013; Mylonakis et al, 2016; Magana et al, 2020). AMPs are a class of cationic molecules of normally < 100 amino acids and broadly present in plants, fungi, and animals (Mygind et al, 2005; Fjell et al, 2012; Mylonakis et al, 2016; Sierra et al, 2017). Their antimicrobial activity is mostly associated with their ability to electrostatically attach to an accessible bacterial membrane with a strong negatively charged surface and some are found to bind to other bacterial components (e.g., Lipid II and intracellular targets) (Brogden, 2005; Oppedijk et al, 2016). Besides merely serving as endogenous peptide antibiotics (Hancock, 1997), AMPs in vertebrates (also known as host defense peptides, HDPs) also function as multifaceted modulating molecules involved in both the innate and adaptive immune responses (Mansour et al, 2014).

Despite intense studies over several decades, very few AMPs have reached to the stage of clinical applications to treat systemic infection (Fox, 2013; Zasloff, 2019; Magana et al, 2020). This frustration is mainly because of the intrinsic non-specific cytotoxicity stemmed from their fundamental detergent-like properties causing membrane (particularly erythrocytes) disruption, in a dose required for treatment (Gao et al, 2009, 2018; Vaara, 2009; Mansour et al, 2014; Zasloff, 2019; Mahlapuu et al, 2020). In addition, human and other vertebrate-derived AMPs often exhibit multifaceted modulating properties that can interfere with the human immune system function when used as systemically administered drugs (Vaara, 2009; Mansour et al, 2014; Zasloff, 2019). Alternatively, AMPs from unicellular, prokaryotic organisms that have the most remote evolutionary distance from humans may exhibit more advantages in this aspect (Vaara, 2009). Firstly, most naturally occurring antibiotics clinically available are produced by actinomycetes and myxobacteria, the two most important natural sources of active metabolites, and only very few by fungi (Geddes, 2008; Waksman et al, 2010; Wright, 2012). Secondly, no multicellular organisms-sourced AMPs have been clinically approved for systemic therapy and several animal-sourced AMPs show a promise only for topical therapy (Zasloff, 2002, 2019). This may be explained by the fact that most of these AMPs are evolved to defend a microenvironment (e.g., skin, oral cavity, and eyes, etc.) and not to be delivered into the systemic circulation (Zasloff, 2019). Indeed, it has been reported that human serum and blood proteins as well as fetal bovine serum (FBS) are able to inactivate the antimicrobial activity of human HDPs, for example, LL-37 and defensins (Panyutich et al, 1995; Johansson et al, 1998; Kudryashova et al, 2021). This property might become an intrinsic cause bedeviling them as systemically administered drugs. Thirdly, like bacteria-produced chemical antibiotics (Hibbing et al, 2010), AMPs in bacteria should be evolved to fight their producers’ competitors (i.e., antagonism) as “pure” peptide antibiotics other than to optimize an inherent property to interfere with the human immune system. We, thus, inferred that exploring bacteria-derived AMPs might be a more promising strategy than using currently available AMPs from multicellular organisms.

To confirm this, we used defensins as an example to exploit the structural diversity and therapeutic potential of bacterial AMPs. Defensins are a class of small cationic AMPs stabilized by multiple intramolecular disulfide bridges. Based on the connectivity and orientation of their disulfide bridges, defensins are categorized into two distinct structural classes: cis- and trans-defensins (Shafee et al, 2016). The former includes peptides with the cysteine-stabilized α-helix and β-sheet (CSαβ) motif and is also the only AMPs shared by plants, fungi, and invertebrates (Dimarcq et al, 1998; Mygind et al, 2005; Zhu et al, 2005), which can be further classified into three major subgroups: ancient invertebrate-type defensins (AITDs); classical insect-type defensins (CITDs); and plant/insect-type defensins (PITDs) (Zhu, 2008). The latter includes vertebrate-derived α-, β-, θ-defensins and the big defensins from invertebrates (mollusks, arthropods, and chordates) (Shafee et al, 2016). Evolutionarily, they both could originate from a common ancestor (Zhou et al, 2019).

By surveying the microbial genomes, we found that AITD-like peptides are restrictedly distributed in two microbial subgroups (actinobacteria and myxobacteria). On this basis, we measured the selective pressure driving the evolution of bacterial defensins by codon-substitution models and conducted a series of experiments to determine the structure of actinomycesin (abbreviated as AMSIN), one representative of the actinomyces-sourced defensins, and to evaluate its antibacterial activity, action mode, and interacting target as well as therapeutic potential. Our work reveals that: (i) Different from the defensins from fungi and animals, defensins from the actinomyces colonized in human oral cavity as well as ruminant rumen and dental plaque are evolving under positive Darwinian selection, in favor of their role in ecological adaptation to the complex oral and rumen multispecies communities. (ii) AMSIN is an inhibitor of cell-wall synthesis with a significant therapeutic efficacy on the systemic infections caused by Streptococcus pneumoniae (abbreviated as SP) and methicillin-resistant Staphylococcus aureus (abbreviated as MRSA) in two experimental mouse models. Our studies indicate that AMSIN and its bacterial homologs from actinobacteria and myxobacteria represent a new source of peptide antibiotics for systemic therapy.

Results Discovery and evolution of bacterial defensins

To explore potential defensin-like peptides in bacteria, we conducted a systematic mining of the microbial genome database using eukaryotic defensins as queries (Fig EV1). The database contained 39 microbial subgroups (Dataset EV1) with genome assembly at four different levels, from which we identified 48 defensin-like peptides. These peptides were named according to their origins at the genus level with an ending as “sin” considering their potential antibiotic feature (Dataset EV2). They are restrictedly distributed in actinobacteria and myxobacteria (Fig 1A and B; Appendix Figs S1 and S2; Datasets EV2 and EV3). The former includes Actinomyces, Corynebacterium, and Micromonospora and the latter includes Melittangium and Cystobacter. In each Actinomyces species, multiple defensin genes are clustered together in chromosomes and located between two ABC transporter genes (Fig 1A; Appendix Fig S1). Copy number variation, a mechanism broadly contributing to both virus evolution (Bayer et al, 2018) and pesticide resistance in mosquitos (Weetman et al, 2018), occurs in the genome of Actinomyces oris among different strains (MMRCO6-1, S24V, and CCUG 34286) (Appendix Fig S1). These bacterial peptides exist as a precursor form and their organization varies depending on their origins (Fig 1C). For the actinobacteria-derived peptides, their maturation most requires excision at an acidic residue (Glu or Asp) and few at the polar Asn (Appendix Fig S2), suggesting that they belong to a class of non-classically secreted proteins, as evidenced by their SecP score exceeding ≥ 0.5 (Appendix Fig S3A) (Bendtsen et al, 2004). For the myxobacteria-derived peptides, their precursors share the same organization as those from fungi and ticks but different from those from scorpions, spiders, and mussels (Fig 1C). In comparison with the eukaryotic defensins, these mature bacterial peptides share detectable to high sequence similarity to AITDs from fungi and invertebrates, both having a relatively short N-loop between Cys1 and Cys2 (Fig 1D; Appendix Fig S2). At the physiological pH, 80% mature bacterial defensins carry net positive charges (Appendix Fig S3B; Dataset EV3), a typical feature required for the antibacterial function in eukaryotic defensins. Several defensin pseudogenes have also been identified in Corynebacterium and Actinomyces, in which mutations lead to a frame shift or a premature stop codon formation (Dataset EV4).

Click here to expand this figure.

Figure EV1. Database search for discovery of bacterial homologs of eukaryotic defensins

Queries extracted from the APD3 database included cis- and trans-defensins (Shafee et al, 2016; Zhou et al, 2019) and their redundancy was removed by CD-HIT to 50% identity. The sign “★” denotes the microbial subgroups identified as the sources of bacterial defensin-like peptides. “Actin.” and “DES” are the abbreviations of actinobacteria and delta/epsilon subdivisions, respectively. For other abbreviations, see Dataset EV1.

Figure 1. Comparisons of AITDs from prokaryotic and eukaryotic organisms

Circular genome plot for Actinomyces sp. oral taxon 171 str. F0337 with genes encoding AITD-like peptides. In this GView (Petkau et al, 2010) image, red and blue tracks show genes on forward and reverse strands, respectively. AMSIN genes and their immediate upstream and downstream genes are shown. ABC, ATP-binding cassette. HP, hypothetical protein. The BioProject database accession number for the genomes is PRJNA43131. The AMSIN gene cluster was verified by PCR amplification (inset) and DNA sequencing. Arrows in green mark binding positions of the two specific primers. A neighbor-joining tree of bacterial defensins with the fungal defensin micasin as outgroup (Zhu et al, 2012). The evolutionary distances were computed by the p-distance method. Numbers on the nodes represent bootstrap values (from 500 replicates) and only > 50% are shown. The scale bar represents the number of amino acid differences per site. Branches in green and red indicate peptides from actinobacteria and myxobacteria, respectively. Two clusters comprising members with structural variations are boxed in different colors (orange, dimer; cyan, vicinal disulfide). Comparison of precursor organization of AITDs with different origins. NCSS, non-classical signal sequence; SP, signal peptide; PP, propeptide; MP, mature peptide. Sequence comparison between unicellular and multicellular AITDs. For the latter, only the top BLASTP hits obtained by AMSIN as query searching against the GenBank database were chosen for alignment. The completely conserved cysteines are shaded in yellow and the non-conserved underlined once. Identical non-cysteine residues are shaded in different colors based on their side-chain nature (red, acidic; purple, aromatic; blue, basic; cyan, polar and uncharged; green, hydrophobic). Residues in the N-terminus and the C-loop of AMSIN-1, which form a continuous hydrophobic surface in its dimer, are underlined twice in red. LRT statistics for M1/M2 (df = 2) and M7/M8 (df = 2) and the corresponding tail probability P. Weblogo of bacterial defensins showing the evolutionary conservation (the sign ♦ in yellow and green, respectively, denoting completely and highly conserved positions). Positions under positive selection are denoted by red arrows and their structural location is indicated.

As shown in Dataset EV2, most bacterial defensins described here are derived from Actinomyces colonized in human oral cavity (A. johnsonii, A. naeslundii, A. oris, and Actinomyces sp. oral taxon 171 str. F0337) and a few in ruminant rumen (A. succiniciruminis) and dental plaque (A. denticolens). Therefore, it is logical to assume that these complex multispecies bacterial communities (Kolenbrander & Periasamy, 2011; McCann et al, 2017; Verma et al, 2018; Chevrette & Currie, 2019) have driven the adaptive evolution of these defensins, which may be different from the evolution of AITDs in fungi and invertebrates. To test this assumption, we compared selective pressure for driving their evolution using the maximum likelihood-based models of codon substitutions (Yang, 2006). For the actinomycetes-derived defensins, the likelihood ratio test (LRT) statistic between models M1 and M2 is 2Δl = 19.68915, much greater than χ2 distribution critical values (df = 2, P = 0.00005). The maximum likelihood estimate of parameters (MLEs) under M2 suggests that about 20% of sites are under positive selection with ω = 3.17. The test with models M7 and M8 yielded 2Δl = 15.587266 (df = 2, P = 0.00041) (Fig 1E; Table 1). MLEs under M8 suggests that about 18% of sites are under positive selection with ω = 2.49 (Table 1). Both tests obtained very similar conclusions. Using the empirical Bayes method computing the posterior probabilities (P), six common positively selected sites (PSSs) with P > 95% under M2 and M8 were identified (Table 1). These sites are distributed in two evolutionarily variable regions (the α-helix and the C-terminus) (Fig 1F) and they all are located on the molecular surface (Fig EV2), suggesting that natural selection has intensified the antibiotic activity of these bacterial peptides through adjusting and optimizing their bacterium-contacting surface, and thus highlighting their potential value as peptide antibiotics. On the contrary, no signatures of natural selection were detected in the fungal and animal defensins (both P = 1) (Fig 1E; Tables EV1 and EV2). This differential might reflect a difference in their physiological roles to the hosts. Our observations also hint at the possible involvement of these bacterial peptides in species competition as the consequence of the actinomycetes adapting to the complex oral and rumen microflora, a phenomenon analogous to the soil bacteria-produced antibiotics (Hibbing et al, 2010).

Table 1. Maximum likelihood estimates of parameters and sites inferred to be under positive selection in the actinomyces-sourced defensins. Model S p l Estimates of parameters PSSs M0 (one-ratio) 15.03 1 −1,545.53 ω = 0.64 None M1 (Nearly Neutral) 16.65 2 −1,473.64 p0 = 0.37 (p1 = 0.63) Not allowed ω0 = 0.07 (ω1 = 1.00) M2 (Positive Selection) 17.40 4 −1,463.80 p0 = 0.22 10E**, 13R**, 16V* p1 = 0.58 (p2 = 0.20) 30R*, 32T*, 34H** ω0 = 0.00 ω1 = 1.00 (ω2 = 3.17) M7 (beta) 17.18 2 −1,465.13 P = 0.24, q = 0.21 Not allowed M8 (beta&ω > 1) 17.63 4 −1,457.34 P = 0.18, q = 0.14 10E**, 13R**, 16V** p0 = 0.82 (p1 = 0.18) 30R**, 32T*, 34H** ωs = 2.49 S represents the tree length; p is the number of parameters in the ω distribution; l is the log likelihood. PSSs under M2 and M8 are identified by the Bayes Empirical Bayes (BEB) method (*P > 95%; **P > 99%). The Naive Empirical Bayes (NEB) method produced similar results. The ω values as indicators of positive selection are boldfaced. Residues are numbered according to AMSIN. Click here to expand this figure.

Figure EV2. Mapping of PSSs of the actinomyces-sourced defensins on the structure of AMSIN

The ribbon drawing of AMSIN with PSSs shown as ball-and-stick model. The molecular surface of AMSIN. PSSs are marked in red. Structural diversity of bacterial defensins

All the bacterial defensins contain six cysteines identical to those of eukaryotic AITDs, which can form three disulfide bridges. In some members, single or a pair of additional cysteines were evolved in their N-termini (Fig 1; Appendix Fig S2). We explored these cysteine-based structural variations through nuclear magnetic resonance (NMR), homology modeling, and molecular dynamics (MD) simulations. AMSIN is a defensin encoded by the genome of the oral-associated species Actinomyces sp. oral taxon 171 str. F0337 and shares about 50–65% sequence identity to several eukaryotic defensins (Fig 1D). For the determination of its solution structure, we carried out the chemical synthesis and oxidative refolding (Fig EV3A and B). The structural statistics of AMSIN from the nuclear magnetic resonance (NMR) analysis are summarized in Table EV3 and the summary of its nuclear Overhauser effect (NOE) data in Fig EV3C. The NMR signals of all residues except Ile29, Arg30, and Gln31 were assigned. More than 11.7 constraints per residue were employed for the structure calculation. For the final 20 structures, no violations were found in distance (> 0.3 Å) and angle (> 5 degrees) restraints. The unassigned region (Ile29 to Gln31) is located in the C-loop region connecting the two β-strands. This is very similar to the results for micasin (Zhu et al, 2012). The NMR ensemble, ribbon model, and the molecular surface colored with charge distribution are shown in Fig 2A–C. AMSIN adopts a typical structure of the cysteine-stabilized α-helix and β-sheet (CSαβ) superfamily (Zhu et al, 2005), in which the helical region spans residues 8–16 and the antiparallel β-sheet is formed by residues 23–26 and 32–35 (Fig 2B). This structure is highly similar to those found in fungi and animals, particularly in their secondary structure elements (Appendix Fig S4).

Click here to expand this figure.

Figure EV3. Oxidative refolding and identification of AMSIN

Reversed-phase high-performance liquid chromatographic (RP-HPLC) profile of oxidized AMSIN (marked by an asterisk). MALDI-TOF MS. The two main peaks correspond to the singly and doubly protonated forms of AMSIN, respectively. Summary of nuclear Overhauser effect (NOE) data of AMSIN. The thickness of the bars indicates the intensity of the NOEs.

Figure 2. Representative three-dimensional (3D) structures of bacterial defensins

A–C. The NMR structure of AMSIN. A. Ensemble of the 20 lowest energy conformers superimposed over the backbone atoms of residues 3–36. B. A ribbon representation with disulfide connectivities shown in sticks. C. Electrostatic potential map. Red, blue, and gray represent negatively charged, positively charged, and electrostatically neutral zones, respectively. D. Backbone-RMSDs for AMSIN-1 and CBSIN-2, shown as a function of time (30-ns simulations). The box marks the time for extracting the snapshots from the trajectories (49.9–50 ns) in (E) and (H). A total of 11 snapshots were taken from an MD simulation. E–J. Simulated 3D structures of AMSIN-1 and CBSIN-2. Their presentations are in the same pattern with that of AMSIN (E-G for AMSIN-1 and H-J for CBSIN-2). The interchain disulfide bridge in AMSIN-1 and the Cys1-Cys2 vicinal disulfide ring (VDR) in CBSIN-2 (shown as sticks) are denoted by arrows. The N- and C-termini are labeled.

Data information: All structure images are displayed by MOLMOL.

AMSIN-1 is a paralog of AMSIN and has the seventh cysteine in its N-terminus (Fig 1D). Using homology modeling combined with MD simulations (Fig 2D), we were able to build a homodimer structure for AMSIN-1, where the seventh cysteine is involved in an intermolecular disulfide bridge (Fig 2E–G). In this dimer structure, there exists a continuous hydrophobic surface formed by residues in the N-terminus and the C-loop (Fig 1D; Appendix Fig S5). Since some eukaryotic defensins function by forming a dimer to interact with the microbial target (Zhang et al, 2010; Song et al, 2011), it is likely that such form in AMSIN-1 is of functional significance. At the same time, using this strategy, we built the structure of CBSIN-2, a defensin from Corynebacterium sp. KPL1995, which contains two adjacent cysteines in its N-terminus. These two cysteines form a rare structural element called vicinal disulfide bridge (Fig 2H–J), as found in some native proteins or oxidative folding intermediates (Carugo et al, 2003; Cemazar et al, 2003).

Antibacterial activity and synergy of AMSIN with oral AMPs

Next, we conducted a series of experiments to assess the antibacterial activity, the interacting target(s) and the therapeutic potential of AMSIN. It showed a lytic effect on bacteria (Fig 3A). Using the classical inhibition-zone assay (Hultmark, 1998), we quantitatively evaluated its antibacterial potency on various bacteria, primarily including some oral Streptococcus strains and clinically relevant strains in human and animal infections (Table 2; Table EV4). The results are summarized as follows: (i) AMSIN exhibited excellent activity against gram-positive bacteria with the strongest potency toward SP (strains D39, R6, ST556, and TIGR4), including a streptomycin-resistant strain. The lethal concentrations (CL) determined were about 0.10 μM (Table 2); (ii) It was also highly potent to three oral Streptococcus bacteria (S. mutans, S. salivarius, and S. sanguinis; abbreviated as SM, SSAL, and SSAN, respectively) with a CL of 0.5–1.4 μM; (iii) AMSIN was highly active to the Staphylococcus bacteria, including a series of clinical isolates. For example, for methicillin-resistant strains (i.e., MRCNS P1369, MRSA P1374, and MRSA P1386), the CL ranged from 0.6 to 1.0 μM and for penicillin-resistant strains (e.g., PRSA P1383 and PRSE P1389), the CL ranged from 1.0 to 5.0 μM; (iv) Its antibacterial spectrum also includes Enterococcus faecium (Table 2), a major pathogen of enterococcal sepsis in the neonate (Nizet & Klein, 2011). Since in some AMPs (e.g., protegrin-1 and HNP1), the disulfide bridges are not functionally essential (Varkey & Nagaraj, 2005; Dawson & Liu, 2010), we assessed this case in AMSIN. The result showed that its reduced peptide exhibited an antibacterial activity comparable to the cyclic version (Appendix Fig S6).

Figure 3. Antibacterial activity of AMSIN and its synergy with HNP1&2 or lysozyme C

Lysis of SSAN by AMSIN. Vancomycin (abbreviated as VAN) was used as a control. Inhibition-zone-based assay for synergy between AMSIN and HNP1&2 or lysozyme C (abbreviated as hLZc) to two oral Streptococcus (SSAN and SSAL). Inset (upper panel, boxed), a representative example of the inhibition-zone test showing synergy. Peptide doses used here were 0.05 nmol/well for HNP1&2 and hLZc at which no visible inhibition zones were observed for both of them. At the same dose, AMSIN produced a small inhibition zone. For synergetic observation, two peptides were applied to the same well according to the doses indicated. Mean ± SD of three biological replicates is displayed. P values were obtained by Student’s t-test (***P < 0.001; exact P values are listed in Table EV7). Note: the nanomolar concentration of HNP1&2 was calculated from the average molecular weight of HNP1 and HNP2. The FIC index assay for synergy between AMSIN and HNP1&2 or hLZc to SSAL and SSAN. The dashed line represents the cutoff of the FIC index (0.5) for synergy.

Source data are available online for this figure.

Table 2. Lethal concentrations (CL) of AMSIN against various bacteria. Bacterial strains CL (µM) Bacillus megaterium CGMCC 1.0459 0.11 Bacillus megaterium CGMCC 1.0459 0.038a Bacillus megaterium CGMCC 1.0459 0.045b Bacillus subtilis CGMCC 1.2428 1.00 Staphylococcus aureus CGMCC 1.89 5.20 penicillin-sensitive Staphylococcus epidermidis P1111 5.20 penicillin-resistant Staphylococcus epidermidis P1389 1.14 penicillin-resistant Staphylococcus aureus P1383 4.28 Methicillin-resistant Staphylococcus aureus P1374 0.62 Methicillin-resistant coagulase negative Staphylococci P1369 1.07 Methicillin-resistant Staphylococcus aureus P1386 0.74 Staphylococcus aureus J685 10.40 Staphylococcus aureus J698 5.20 Staphylococcus aureus J700 5.20 Staphylococcus aureus J706 5.20 Staphylococcus aureus J708 1.14 Staphylococcus aureus J710 2.60 Streptococcus pneumoniae D39 0.09 Streptococcus pneumoniae R6 0.14 Streptococcus pneumoniae ST556 0.14 Streptococcus pneumoniae ST556 (StrR) < 0.14 Streptococcus pneumoniae TIGR4 0.14 Streptococcus mutans CGMCC 1.2499 (ATCC 25175) 1.01 Streptococcus mutans CGMCC 1.2499 (ATCC 25175) 0.38a Streptococcus mutans CGMCC 1.2499 (ATCC 25175) 0.56b Streptococcus salivarius CGMCC 1.2498 (ATCC 7073) 0.49 Streptococcus sanguinis CGMCC 1.2497 (ATCC 49295) 1.41 Streptomyces griseus NBRC 13350 2.89 Streptomyces scabiei CGMCC 4.1765 (ATCC 49173) 2.01 Lysinibacillus fusiformis 0.27 Enterococcus faecalis V583 (ATCC 700802) 4.65 StrR, streptomycin-resistant.

Because of the origin of AMSIN from an oral bacterium, we assessed its potential synergy with human oral-derived antibacterial factors (the α-defensins HNP1&2 and lysozyme C) isolated from human saliva (Appendix Fig S7). HNP1&2 is a mixture of HNP1 and HNP2 co-eluted in our purification process (Appendix Fig S7) (Note: HNP1 has an extra N-terminal Ala relative to HNP2, but this does not create an effect on their antibacterial activity) (Ericksen et al, 2005). When AMSIN and a sub-inhibitory dose of HNP1&2 or lysozyme C were jointly applied to the same well, the inhibition zone was significantly increased, compared with AMSIN alone (Fig 3B; Appendix Fig S8), and accordingly its CL values for Bacillus megaterium (abbreviated as BM) and SM descended (Table 2). Using the liquid fractional inhibitory concentration (FIC) assay, we confirmed that the FIC indexes for SSAL and SSAN (two representatives of oral bacteria)