Unlocking the unexplored AMPSphere in marine rare species

Abstract

Background

Antimicrobial peptides (AMPs) have advantages over traditional antibiotics in fighting against drug-resistant bacterial infections. Natural microbial communities are considered as the priority targets for next-generation AMP bioprospecting initiatives. While progress has been made in characterizing AMPs from the dominant microbial taxa in natural ecosystems, current research largely overlooks the biosynthetic potential of rare species. Given their distinct evolutionary pressures, rare species likely produce AMPs with novel structures and unconventional mechanisms of action.

Results

In this study, enrichment cultivation of a marine biofilm was conducted in 138 carbon source- and oxygen level-based conditions, followed by metagenomic sequencing using both Illumina and Nanopore platforms. Analysis of 435 high-quality genomes derived from the metagenomes suggests that these bacterial strains are significantly underrepresented (< 0.01%) in global marine biofilm communities. Through multi-model prediction, we identified 3,054,472 candidate AMPs from the genomes, including 1048 high-confidence ones, thereby significantly expanding the previously known AMPSphere. Furthermore, AMPs derived from the rare bacterial species exhibit unique sequence characteristics, structural diversity, remarkable stability under diverse pH conditions and pepsin exposure, and strong therapeutic potential in animal models, reflecting their specialized adaptive and defensive strategies developed within ecological systems.

Conclusions

The features of the underexplored AMPs from low-abundance bacteria in marine biofilms provide valuable resources and theoretical foundations for the development of highly effective antimicrobial agents.

Download video file ^{(54.5MB, mp4)}

Video Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s40168-025-02326-0.

Introduction

Antimicrobial peptides (AMPs) have advantages over traditional antibiotics in fighting against drug-resistant bacterial infections [1–3]. AMPs can be synthesized through ribosomal routes, without complex post-synthesis modifications [4]. In natural environments, bacteria produce AMPs to combat competitors and to maintain their ecological niche, and thus environmental microbiomes are important AMP reservoirs. A study performed systematic mining of 63,410 metagenomic datasets and 87,920 microbial genome sequences spanning globally distributed microbial communities and established a unique AMPSphere database [5]. A recent survey further revealed the bioprospecting potential of microorganisms, especially focusing on those living in marine environments [6]. In addition, deep learning and Ribo-seq were applied to discover 341 AMPs from marine biofilms, which were previously undescribed and had unique molecular features [7]. These findings highlighted the natural microbial communities as the priority targets for next-generation AMP bioprospecting initiatives.

However, capturing bacterial information from complex microbiome samples from natural environments is challenging, primarily due to the presence of bacterial species with highly skewed distribution patterns [8]. In metagenomic sequencing, rare species are often completely overshadowed by the highly abundant bacteria or suffer from the low-quality genome assemblies [9, 10]. Similarly, bacterial isolation methods also face this issue, as bacterial cultivation is heavily inclined toward a few phylogenetic groups, and only abundant bacteria are usually isolated [11, 12]. Therefore, despite the previous global-scale bioprospecting efforts [5–7], AMPs from a vast number of rare species are still waiting to be discovered.

Due to their distinct functional traits, rare species in a natural ecosystem can perform functions complementary to those performed by other relatively abundant species [13]. These taxa act as ecological stabilizers under environmental stress and may encode unique metabolic pathways, including those for secondary metabolites, resistance genes, and adaptive traits [14, 15]. Given their distinct evolutionary pressures, rare microbes likely produce AMPs with novel structures and unconventional mechanisms of action. It is expected that focusing on this underexplored microbial fraction not only mitigates redundancy in current research pipelines, but also provides innovative strategies for addressing multidrug-resistant infections.

This study employed enrichment cultivation to access AMPs encoded by low-abundance bacteria within a marine biofilm community. Enrichment culture facilitates the proliferation of microorganisms rare in their original niche, thereby providing a much more comprehensive profile of the microbial community [16, 17]. A marine biofilm sample was collected from the surface of a common coastal stone and subjected to carbon-source-based enrichment cultivation under 138 distinct conditions. Metagenomic sequencing and bioinformatic approaches were then used to discover novel AMPs (3,054,472 candidate AMPs, including 1048 high-confidence ones). These AMPs significantly expanded the known repertoire of AMPSphere and demonstrated promising anti-infection activity in preclinical animal models.

Results

Taxonomic composition and diversity in enrichment cultures of a marine biofilm

The entire workflow of the present study is shown in Fig. 1. The seed biofilm sample used for enrichment cultivation was collected from the subtidal rock surfaces in the coastal area of Qingdao, China. Parallel enrichment culture experiments were conducted using 69 distinct carbon substrates as the sole carbon source (Table S1). The enrichment was conducted under two oxygen conditions: low-oxygen generated by N₂-purged headspace and controlled atmospheric conditions with 20% O₂ headspace. The selection of these conditions was based on the correlations between carbon-utilization patterns and redox environments [18], as well as the prevalence of anaerobic micro-niches in natural biofilms [19]. Using this experimental design, 138 distinct microbial communities (69 carbon sources × 2 oxygen conditions) were obtained for comprehensive omics analysis. The 138 samples were then subjected to metagenomic sequencing to result in the generation of 3.2 TB of Illumina NovaSeq data (average 22.8 ± 0.8 GB per metagenome, Table S2) and 37.4 GB of Nanopore long-read data (average 12.5 ± 0.6 GB per metagenome, Table S3).

Fig. 1.

Schematic workflow of cAMP discovery and validation. 1 Enrichment cultivation: rock-associated biofilms were collected from Qingdao subtidal zone and used for enrichment cultivation under aerobic and microaerobic conditions. 2 Metagenomics: Illumina–Nanopore hybrid sequencing to generate 435 MAGs assembly with completeness > 90% and contamination < 5%. 3 AMP prediction: cAMP prediction based on dual-model screening and prioritization based on score curve inflection analysis. 4 Experimentals: cAMPs were synthesized through solid-phase peptide synthesis and subjected to antimicrobial testing against bacterial pathogens. Promising candidates were further tested for cytotoxicity, hemocompatibility, pH/protease stability, and efficacy in murine infection models (skin abrasion and peritonitis)

Furthermore, 16S rRNA gene tags (miTags) [20] were extracted from the metagenomes, with 183,028 ± 46,301 per metagenome, and classified into operational taxonomic units (OTUs) at a similarity index of 97%. A total of 9,497,699 OTUs were obtained and further categorized into 62 phyla (with Pseudomonadota classified at the class level), 434 families, and 2027 genera. With a maximum abundance of 99.62%, Gammaproteobacteria emerged as the most dominant phylum across all the metagenomes, followed by Alphaproteobacteria (86.31%) and Bacteroidia (29.05%) (Fig. S1). At the genus level, the prevalent taxa included Vibrio, Marinomonas, Pseudolateronomas, Marinobacterium, Halomonas, and Ruegeria (Fig. S2). These results indicate that enrichment cultivation results in a substantial recovery of microbial diversity from the marine biofilm community. To further support this notion, we compared the taxonomic composition of the enrichment culture-derived biofilms with the global marine biofilm dataset. The dataset includes 101 biofilms collected from eight locations of three oceans (Pacific, Indian, and Atlantic oceans) [21], 6 subtidal rock samples from Qingdao [22], and 24 biofilms on the microplastic surfaces collected in Huangdao [23], in total 131 metagenomes. The result revealed 7 families and 117 genera that were specific to the enrichment-cultured biofilms (Fig. S3).

Rare species recovered from the enrichment cultures

Given that similar carbon sources can select for microbial communities with overlapping structures [24, 25], samples within the same carbon source category were pooled prior to Nanopore sequencing to maintain representativeness while improving efficiency and reducing costs. This strategy enhanced sequencing data utilization and minimized redundancy. Through hybrid assembly of Illumina short-read and Nanopore long-read sequencing data, a total of 435 dereplicated, high-quality metagenome-assembled genomes (MAGs) were reconstructed from the 138 enrichment cultures. After stringent quality filtering (average nucleotide identity [ANI] ≤ 99%, completeness > 90%, and contamination < 5%), these MAGs demonstrated minimal genomic overlap (Fig. S4), and more than half were nearly complete (Fig. S5). Assembly metrics and genomic characteristics of MAGs have been comprehensively cataloged in Table S4.

Using the GTDB-Tk database, 435 MAGs were classified into 12 phyla, 62 families, and 144 genera. At the phylum level, Pseudomonadota dominated the community composition, followed by Bacteroidota, Planctomycetota, Bdellovibrionota, Actinomycetota, and Campylobacterota (Fig. 2A). At the genus level, Aequorivita (Bacteroidota), Thalassospira (Pseudomonadota), and Celeribacter (Pseudomonadota) were the most abundant taxa (Fig. S6). Notably, at the species level, around 35.9% (n = 156) of the MAGs remained unclassified (Fig. 2A). Thus, there is a substantial increase in the microbial diversity after enrichment cultivation.

Fig. 2.

Taxonomic affiliation and in situ abundance of the enrichment culture-derived bacteria. A Sankey diagram of taxonomic resolution of the enrichment culture-derived MAGs, with 35.9% unclassified to species level. B, C Phylogenetic overlap analysis: Venn diagrams comparing phylum-level (B) and genus-level (C) composition of the enrichment culture-derived MAGs and genomes of the isolated bacteria. D Metagenomic read recruitment patterns of the 435 enrichment culture-derived MAGs and the 713 genomes of the isolated bacteria. This figure shows the relative abundance of the MAGs and the genomes across 131 biofilm metagenomes collected from three ocean provinces, quantified via standardized read mapping

Moreover, in comparison to the 713 isolate genomes from coastal biofilms in our prior work [7], MAGs from enrichment culture-derived biofilms demonstrated a threefold increase in phylum-level diversity in this study (12 vs. 4 phyla; Fig. 2B), with 115 unique genera and 29 shared taxa (Fig. 2C). To validate the rarity of the MAGs, read mapping analysis was performed against the normalized metagenomes (1 × 10⁶ sequences/sample) of the global marine biofilm dataset, which includes 131 metagenomes. Abundance-based coverage estimates indicated that the MAGs recovered through enrichment culture exhibited significantly lower abundance than the genomes of the isolates (Fig. 2D). Specifically, over 95% of these MAGs showed relative abundances below 0.01% in more than half of the environments (Fig. 2D) and were thus defined as rare species according to earlier criterion [26]. In contrast, only around 50% of the 713 genomes of bacterial isolates met this rarity threshold (Fig. 2D).

The 435 MAGs derived from enrichment culture exhibited greater inter-strain variability and significantly higher numbers of ORFs (3,594 ORFs/genome), along with higher dispersion compared to the genomes obtained from isolation (Fig. S7A). In the KEGG annotation results, although the MAGs recovered from enrichment cultivation contained slightly higher absolute numbers of annotated genes, they showed an overall lower annotation rate (52.4% and 54.5% for enrichment and isolation cultures, respectively) (Fig. S7B), suggesting an abundance of diverse genes with unknown functions. Searching against the broader and more inclusive eggNOG database revealed comparable median annotation rates between the two groups (86.1% and 86.9%), yet a markedly higher dispersion among the enrichment-derived genomes than the isolated genomes (Fig. S7B). Notably, genomes from Bdellovibrionota have a minimum annotation rate of only 68.2%.

To further characterize the functional potential of the genomes obtained from enrichment cultures, we analyzed a genome from the phylum Desulfobacterota. Notably, Desulfobacterota was not detected in a previous global survey of marine biofilm microbiomes based on metagenomic approaches [21]. Members of this phylum are generally rare across diverse environments and have been implicated in mercury methylation [27, 28]. In this study, we acquired the strain Desulfopila ganghwensis ET04-09, which has the potential to encode thousands of cAMPs. Compared to the available reference strain D. ganghwensis IMCC35004 [29], ET04-09 possesses 120 unique genes (Fig. S8). These genes exhibit diverse functional roles, including motility-related components such as the flagellar proteins FlgC and MotB, as well as key enzymes involved in carbon metabolism, amino acid metabolism, and microbial secondary metabolism—examples include transketolase (tktA/B), glutamate decarboxylase (gadA/B), aspartate-ammonia ligase (asnA), and acetylornithine deacetylase (argE) (Table S5). The presence of these genes implies an expanded metabolic versatility in this strain, likely enabling the utilization of a broader substrate range and enhancing its adaptability to biofilm environments, thereby contributing to the metabolic repertoire of Desulfobacterota. These findings indicate that the genomes obtained from enrichment cultivation possess greater functional diversity and potential novelty, further supporting their distinctiveness in expanding the known functional repertoire of marine microorganisms.

Unique AMP sequences encoded in MAGs derived from the enrichment culture

To investigate the AMP-coding potential of MAGs obtained through enrichment cultures, we comprehensively analyzed the 78,562,450 sORFs (encoding 5–40 amino acid residues) extracted from 435 MAGs. AMP identification at this scale requires substantial computational resources and time, with a high risk of resource allocation to non-relevant sequences. Since most known AMPs are cationic peptides that interact with microbial membranes through electrostatic interactions [30], an initial filtering criterion of net positive charge > 2 was implemented. Through sequential filtering steps, the dataset of 78,562,450 sORFs was refined to 23,718,912 non-redundant sORFs. For subsequent AMP prediction, two complementary approaches were employed: the deep learning hybrid model [7] and the random forest-based Macrel model [31]. Both models were configured to prioritize the prediction precision over recall, ensuring high-confidence identification at the potential cost of comprehensive sequence detection. After intersection analysis of predictions from both models using default parameters, 3,054,472 candidate AMPs (cAMPs) were obtained (Fig. 3A).

Fig. 3.

Landscape of cAMPs identified from the enrichment culture-derived MAGs. A Prediction consistency between two computational models for cAMP identification. B Phylum-level taxonomic classification of 3,054,472 cAMPs co-identified by both models. C Phylogenetic mapping of cAMPs across MAGs (Circular dendrogram layers (inner to outer) indicate phylum classification, total ORFs per genome, sORFs count, and cAMP abundance per genome, respectively; the red dashed line indicates the median value). D Prediction scores of both models, with dashed lines marking the threshold cutoffs (Hybrid model: 0.9; Macrel model: 0.8). E Phylum-level distribution of 1,048 high-confidence cAMPs.

Phylum-level analysis revealed substantial variations in cAMP abundance (Fig. 3B). Among the dominant phyla, Bacteroidota demonstrated a significantly higher median cAMP abundance compared to Pseudomonadota (Fig. 3B). Both phyla exhibited marked intra-phylum heterogeneity, suggesting strain-specific AMP production (Fig. 3B). Actinomycetota, a phylum renowned for the biosynthesis of secondary metabolites [32], however, showed comparatively low abundance of cAMP (Fig. 3B). Phylogenetic reconstruction revealed the broad taxonomic distribution and remarkable sequence diversity of cAMP-producing bacteria (Fig. 3C). These findings position enrichment-cultured marine bacteria as underexplored reservoirs of novel AMPs.

To enhance the reliability of AMP identification, sequence clustering was performed using a family-based approach. AMP families were defined through pairwise sequence identity analysis, which required a minimum of three homologous sequences per cluster. This hierarchical clustering strategy effectively prioritized functional AMP groups and mitigated the false-positive predictions inherent to large-scale screening. From the resulting clusters, 24,956 representative sequences corresponding to distinct AMP families were identified. The subsequent quality-control process employed model-specific prediction score thresholds determined by inflection point analysis of score distribution curves (Fig. 3D). Based on the thresholds (hybrid model 0.9; Macrel model 0.8), 1,048 high-confidence cAMPs were obtained, which met the dual-model criteria. Taxonomic analysis revealed Pseudomonadota and Bacteroidota as the primary contributors to this refined dataset (Fig. 3E), with Epibacterium and Aequorivita emerging as particularly abundant genera (Fig. S9A). Notably, most of the source strains of high-confidence cAMPs were unclassified at the species level (Fig. S9B), representing 17% of the total, which suggests a considerable unexplored biosynthetic potential.

To investigate the properties of high-confidence cAMPs, their sequences were compared with the sequences of previously reported AMPs (Fig. 4A). The reference dataset includes 14,803 AMPs collected from five publicly accessible AMP databases, 341 AMPs identified from culturable marine biofilm-derived bacteria in our earlier study [7], and 592,082 sequences from the global AMPSphere generated in a previous study [5]. The maximum sequence identity of only 62% was observed between enrichment-derived cAMPs and reference AMPs, while > 95% of the sequences exhibited < 50% identity (Fig. 4B), indicating their substantial divergence from established AMP sequences. The low conservation extended to the composition of amino acids, where marine biofilm-derived cAMPs (both enrichment-cultured and isolate-originated) showed significantly higher proportions of serine, arginine, and cysteine residues, compared to other datasets (Fig. 4C). Moreover, the cAMPs derived from marine biofilms exhibited specific structural fingerprints, including reduced aliphatic index and global hydrophobicity (Fig. 4D). Interestingly, except for molecular weight and sequence length, the compositional profiles of enrichment-cultured cAMPs were consistent with the profiles of their isolate-derived counterparts. This finding suggests that the niche-driven evolutionary constraints predominantly shape the molecular features of enrichment-cultured cAMPs. Despite the functional convergence suggested by physicochemical profiles, the remarkably low sequence conservation highlighted the enrichment-derived cAMPs as a distinct evolutionary cohort. This phylogenetic divergence indicates potential novel mechanisms that were unexplored in the AMPs discovered previously.

Fig. 4.

Sequence identity and predicted physicochemical properties of high-confidence cAMPs. A Reference datasets used for sequence comparison: AMPSphere [5], marine biofilm-derived AMPs from our previous study [7], and a non-redundant AMP collection integrating DRAMP, CAMP, dbAMP, DBAASP, and APD3. B Maximum sequence similarity distribution between 1,048 novel cAMPs and existing AMP databases, with most cAMPs showing < 50% similarity to known AMPs. C Comparative analysis of amino acid profiles across the AMPs from distinct sources. D Comparison of the physicochemical properties (e.g., hydrophobicity, charge, isoelectric point) of cAMP sequences.

Identification of potent cAMPs against drug-resistant pathogens

To bridge computational predictions with biological functionality, 55 peptides were randomly selected from the high-confidence cAMPs for chemical synthesis. Among these, one peptide could not be synthesized. The remaining 54 peptides (amino acid sequences are given in Table S6) were tested at a concentration of 64 μg/mL against seven clinically relevant pathogens, including three drug-resistant strains of Staphylococcus aureus, one Pseudomonas aeruginosa strain, one Klebsiella quasipneumoniae strain, and two drug-resistant strains of Acinetobacter baumannii. A preliminary screening revealed universal bioactivity, where 53 tested cAMPs inhibited at least one pathogen, while 49 cAMPs inhibited more than three pathogens (Fig. 5A). Subsequently, minimum inhibitory concentration (MIC) assays were conducted on nine leading cAMPs, which revealed exceptional potency (Fig. 5B). In several cases, the cAMPs (e.g., cAMP-18-nc against the three S. aureus strains) showed MIC values as low as 2 μg/mL (Fig. 5B). Three cAMPs (919-nc, 1111-nc, 1229-nc) demonstrated conserved inhibition efficiency against both P. aeruginosa (MIC range 16–32 μg/mL) and A. baumannii (2–16 μg/mL) (Fig. 5B), suggesting evolutionary pressure for conserved molecular targets in intrinsically resistant pathogens.

Fig. 5.

Antimicrobial activity and mechanisms of synthesized cAMPs. A Antibacterial activity of cAMPs (64 μg/mL) against seven pathogens: S. aureus ATCC 12600 (dark blue), S. aureus ATCC 33591 (light blue), S. aureus ATCC 33592 (dark red), P. aeruginosa PAO1 (light red), A. baumannii ATCC 19606 (yellow), A. baumannii Bio-53272 (sky blue), and K. quasipneumoniae (green). Bacterial growth rates (bubble size) were normalized to the control group (set as 1). Bubble color indicates the significance levels. B MICs of nine broad-spectrum cAMPs against seven bacterial strains across a 2–64 μg/mL concentration gradient. C CD assay for analysis of secondary structure of cAMPs in 30% trifluoroethanol: CD spectra (190–260 nm) were recorded at 25 °C. Peptides (20 μM) were analyzed with three technical replicates. D, E PI and DISC3(5) assays demonstrating permeabilization and transmembrane potential disruption of S. aureus ATCC 12600 treated by cAMPs. F SEM images of S. aureus 12,600 treated with cAMPs (5 × MIC), with treated cells showing membrane disruption, cytoplasmic leakage, and morphological deformation compared to controls. All experiments included three biological replicates. Statistical significance was determined using Student’s t-test. *P-value < 0.05; **P-value < 0.01; ***P-value < 0.001

Antimicrobial mechanism of the cAMPs

To elucidate the antimicrobial mechanism of cAMPs, computational structural prediction was integrated with experimental conformation validation. Based on the initial secondary structure prediction of the nine unmodified cAMP sequences, seven candidates with characteristic α-helical domains were identified. The presence of α-helical domains is a structural hallmark of membrane-disrupting AMPs that degrade the cell membrane of pathogens through porin formation [33]. On the other hand, cAMP-1116-nc and cAMP-77-nc displayed divergent β-antiparallel or coil configurations (Fig. S10). The predictions were validated via circular dichroism (CD) spectroscopy in membrane-mimetic conditions (30% trifluoroethanol). Except for cAMP-1116-nc, all cAMPs had varying degrees of helical structures, as indicated by a positive peak around 192 nm and two negative peaks near 208 nm and 222 nm (Fig. 5C). In particular, the ideal helical structures of cAMP-18-nc, cAMP-919-nc, cAMP-1111-nc, cAMP-1229-nc, cAMP-502-nc, and cAMP-77-nc aligned with their exceptional antimicrobial activities (Fig. 5C).

To explore the antimicrobial action mode of cAMPs, a tiered experimental approach was employed using four representative cAMPs (1111-nc, 1229-nc, 502-nc, 77-nc) against S. aureus (ATCC 12600) at different concentrations (1 ×, 2.5 ×, 5 ×, and 10 × MIC). For the assessment of membrane integrity in S. aureus, propidium iodide (PI) influx assays were performed. The results revealed concentration-dependent membrane permeabilization of cAMP-1229-nc (Fig. 5D). For the assessment of transmembrane potential dissipation, DiSC3(5) depolarization assays were performed, which demonstrated the concentration-dependent effect of all four cAMPs (Fig. 5E). In particular, the dose–response profiles of DiSC3(5) fluorescence exhibited nonlinear amplification between 5 × and 10 × MIC for cAMP-1111-nc, cAMP-1229-nc, and cAMP-77-nc (Fig. 5E), likely due to pore formation beyond critical peptide density. Scanning electron microscopy (SEM) was used to directly visualize the damage to the pathogen's membrane. After treatment with 5 × MIC cAMPs for 1 h, S. aureus cells exhibited significant morphological changes, including membrane shrinkage and perforation (Fig. 5F).

High biocompatibility and stability of the cAMPs

The biocompatibility of AMPs was examined by systematically assessing their cytotoxicity and hemolytic activities. Given their potent antimicrobial efficacy, the potential effects of cAMPs on human keratinocytes (HaCaT), normal liver cells (LO2), kidney epithelial cells (HK-2), and fresh sheep red blood cells were investigated. At a concentration of 64 μg/mL (a dose within the effective antimicrobial range for all tested peptides), most cAMPs demonstrated negligible cytotoxicity against the eukaryotic cells (Fig. 6A) and minimal hemolytic activity with less than 10% erythrocyte lysis (Fig. 6B). The biocompatibility profile of cAMPs closely matched the profile of the clinical reference agent polymyxin B, underscoring their potential for therapeutic applications.

Fig. 6.

Biocompatibility and stability of biosynthesized cAMPs. A Eukaryotic cell toxicity screening using human HaCaT, LO2, and HK-2 cells. B Hemolytic activity assay using sheep red blood cells. C Changes in MICs of cAMPs against S. aureus 12,600 at different pH (2, 7.4, and 8) reflecting their pH stability. D Changes in MICs of peptides against S. aureus 12,600 after pepsin treatment, reflecting their resistance to enzymatic hydrolysis. All experiments were performed on three biological replicates

The previously reported AMPs often exhibited instability under extreme pH conditions or protease exposure [34]. In contrast, the cAMPs derived from enrichment-cultured marine biofilms in this study demonstrated remarkable resilience to such conditions. To systematically evaluate the stability of these cAMPs, we analyzed their antimicrobial activities against S. aureus 12,600 under three pH conditions (2.0, 7.4, and 8.0 in PBS). Across the tested pH spectrum, all cAMPs except cAMP-18-nc displayed minor fluctuations in antimicrobial activities (Fig. 6C). To analyze the protease resistance of cAMPs, their antimicrobial activities against S. aureus 12,600 were analyzed after 1-h incubation with pepsin in acidic PBS (pH = 2). Vancomycin and melittin were included as stable and unstable controls, respectively. All cAMPs and vancomycin retained full antimicrobial efficacy after pepsin digestion, showing only minor changes in MIC values compared to untreated controls, whereas melittin completely lost its antibacterial activity following treatment (Fig. 6D). This robust resistance of cAMPs to enzymatic hydrolysis surpassed the stability profiles of typical linear peptides.

Anti-infective efficacy of the cAMPs in mouse models

The clinical translation of AMPs is often hindered by the discordance between in-vitro potency and in-vivo performance. To address this challenge, we evaluated the anti-infective efficacy of four cAMPs (1111-nc, 1229-nc, 502-nc, and 77-nc, with MICs of 2–8 μg/mL) against S. aureus in a murine skin abrasion model infected with a polymicrobial inoculum of three antibiotic-resistant S. aureus strains (ATCC 12600, ATCC 33591, ATCC 33592). The cAMPs (32 μg/mL in PBS) were topically administered 2 h after infection (Fig. 7A). At 24 h, AMP-treated mice exhibited a slower rate of body weight loss compared with the PBS control (Fig. 7B). All cAMPs achieved a significant 0.5–1 log reduction in the bacterial colony-forming unit (CFU) counts (Fig. 7C), demonstrating direct in-vivo antimicrobial activity.

Fig. 7.

Anti-infective activity of biosynthesized AMPs in two mouse models. A Schematic diagram of the mouse skin abrasion infection model for assessing the anti-infective efficacy of AMPs against S. aureus infection. B Changes in the body weight of mice during the experiment. C Bacterial load at the infection site around 24 h after infection. D Schematic diagram of the mouse peritonitis infection model for assessing the AMP efficacy against S. aureus infection. E–G Bacterial load in peritoneal lavage fluid, liver, and kidneys 24 h after infection. H, I Representative H&E-stained histological images of the liver and kidney from mice treated with saline, AMPs, or vancomycin (scale bars 200 μm). All experiments were conducted with three biological replicates (three mice). Statistical significance was determined using Student’s t-test. *P-value < 0.05; **P-value < 0.01; ***P-value < 0.001

To further validate the therapeutic potential of cAMPs beyond localized infections, we assessed their systemic efficacy in a polymicrobial peritonitis model induced by intraperitoneal inoculation of three S. aureus strains mentioned above. All mice received a single intraperitoneal dose of cAMPs (4.5 mg/kg, equivalent to 4–16 × in-vitro MIC) 1 h post-infection (Fig. 7D). All four cAMPs significantly reduced the bacterial loads in the peritoneal cavity, liver, and kidneys of mice (Fig. 7E–G), achieving 0.5–1.5 log reductions in CFU compared to the untreated controls. Hematoxylin and eosin (H&E) staining revealed infection-driven pathology in control mice, indicated by inflammatory cell infiltration and vacuolization in the liver and kidneys (Fig. 7H, I). In contrast, tissue architecture was preserved in certain cAMP- (e.g., cAMP-77-nc) or vancomycin-treated groups with minimal inflammatory damage (Fig. 7H, I).

Discussion

The escalating crisis of antibiotic resistance in microbes has intensified the global demand for novel therapeutic agents capable of countering multidrug-resistant pathogens. To address this challenge, AMP discovery has expanded to ecologically diverse environments for combating the evolving microbial threats. Although recent large-scale AMP explorations have encompassed marine habitats [5–7], these efforts predominantly relied on the metagenomic or genomic data from isolated strains and were biased towards the dominant microbial taxa. This led us to employ an enrichment cultivation strategy for the recovery of AMPs from the low-abundance species, which constitutes less than 0.01% of the original population in a marine biofilm community. From an ecological perspective, rare bacterial species in complex microbial communities are typically at a competitive disadvantage, yet they are seldom eliminated entirely, maintaining persistence through niche-specific adaptive mechanisms [13]. To maintain their ecological niches, they may secrete AMPs with selective killing abilities, enabling them to survive under resource-limited or high-competition conditions. These AMPs often possess unique physicochemical properties—such as high stability and broad-spectrum inhibitory activity—reflecting their long-term adaptation to specific ecological pressures. Consistent with this, the cAMPs identified in this rare microbial biosphere were structurally and functionally distinct from the conventional AMPs. Furthermore, these cAMPs showed antimicrobial activity against antibiotic-resistant pathogens, as well as good biocompatibility and in-vivo efficacy in murine infection models.

The results of this study demonstrated the unique capacity of enrichment cultivation to resuscitate rare taxa that evade conventional metagenomic sequencing and isolation-based approaches. By leveraging multi-substrate cultivation and genome assembly, the genomes from 12 distinct phyla were successfully reconstructed, with a threefold increase in taxonomic diversity compared to conventional single-substrate isolation [7]. Notably, this dataset revealed underexplored microbial lineages with potential biotechnological or ecological significance. For example, Desulfobacterota was not discovered during the global survey of marine biofilm microbiomes based on metagenomics [21]. The species belonging to this phylum are often rare in various environments and are suggested to have mercury methylation function [27, 28]. In this study, a strain (Desulfopila ganghwensis ET04-09) from this phylum showed the potential to encode thousands of cAMPs and a number of enzymes for secondary metabolism, thereby expanding the metabolic capacity of Desulfobacterota. Moreover, four genomes belonging to the group Casp-alpha2 were recovered through enrichment cultivation, and members of this group were suggested to live exclusively in deep-sea environments, participating in sulfur oxidation [35]. The discovery of Casp-alpha2 in our data indicates that its ecological distribution is much wider than previously thought. In addition, DEV007, detected in this study, was identified as an uncultured bacterial family with facultative anaerobic characteristics and sensitivity to multiple metals [36, 37]. Its successful recovery through enrichment culture demonstrates that this bacterial group can be cultivated in the laboratory.

Integrated metagenomic-genomic analysis revealed that the majority of enriched strains exhibited very low abundance in global biofilm metagenomes, further confirming their ecological rarity. This finding aligns with the ecological principle that certain low-abundance species may thrive through metabolic cross-feeding or specialized substrate utilization [13, 14]. Consistently, our recent study suggested that cross-feeding is a prevalent strategy for marine biofilm-associated strains of the Rhodobacteraceae family to expand their metabolic niche [38]. Moreover, strategic cultivation under high- and low-oxygen conditions generated physiologically distinct subcommunities, which aligns with the prevalence of anaerobic micro-niches and subpopulations in natural biofilms [22]. This oxygen-driven functional divergence also provided ideal metadata for differential coverage binning, reducing the complexity of the community and improving the completeness of MAG. To this end, MAGs of bacterial strains with phylogenetic novelty were obtained, indicating that the enrichment culture strategy effectively taps into the underexplored microbial biosynthetic reservoirs.

Through integrated application of two machine learning algorithms coupled with stringent intersection criteria, 1,048 high-confidence cAMP sequences were successfully curated. Comparative sequence alignment demonstrated striking divergence of these cAMPs from the global AMPSphere and the public databases. In particular, cAMPs reported here exhibited conserved enrichment of cysteine, serine, and arginine content. This signature likely confers enhanced biostability to AMPs through cysteine-mediated disulfide bridges and serine-rich domains capable of forming extensive hydrogen-bonding networks [39, 40]. In support of this notion, the cAMPs derived from enrichment culture displayed exceptional pH tolerance and pepsin resistance, offering a pharmacological advantage over the pH-sensitive and protease-vulnerable conventional AMPs. The multimodal validations established membrane disruption as the dominant bactericidal mechanism of cAMPs, which aligns with the previous understanding regarding the structure of potent AMPs [33]. Furthermore, in murine infection models, all four cAMPs demonstrated significant bactericidal efficacy, highlighting their potential as promising candidates for combating drug-resistant infections. In certain aspects, cAMP-77-nc matched the efficacy of vancomycin in combating S. aureus infection. It also displayed lower cytotoxicity than polymyxin B, suggesting significant potential for drug development application.

From a clinical perspective, AMPs with broad-spectrum antibacterial activity have long been considered ideal therapeutic candidates. However, issues related to their stability and biocompatibility remain major obstacles to further application [41, 42]. AMPs derived from rare bacteria exhibit unique sequence characteristics and structural diversity, reflecting their specialized adaptive and defensive strategies developed within ecological systems. These features provide valuable resources and theoretical foundations for the development of highly effective and low-toxicity antimicrobial agents. Owing to their small molecular size and ease of modification and synthesis, such peptides can be optimized to enhance target specificity, improve stability, and modulate immune responses, offering new avenues for the development of next-generation anti-infective drugs, antibiofilm agents, and combination therapies.

Altogether, by bridging the gap between environmental microbial ecology and targeted drug discovery, this study has established a blueprint for the discovery of evolutionarily refined AMPs from the rare biosphere in nature. The carbon-source-based enrichment cultivation and bioinformatic approaches effectively increased the abundance of rare taxa, thereby enabling the detection of numerous novel AMPs that had not been previously described. This ecological intelligence-driven paradigm not only redefines antimicrobial discovery roadmaps but also opens new exploration avenues for the discovery of other bioactive metabolites (e.g., anticancer agents) from microbial dark matter.

Materials and methods

Biofilm sampling and enrichment cultivation

The seed community used for the enrichment culture was sampled from the subtidal zone around Maidao Island (36.04°N, 120.26°E) in Qingdao, China. The subtidal zone is connected to the open ocean. Biofilms were collected from the surface of immersed rocks using sterile cotton swabs. The collected biofilm samples were placed in sterile seawater and immediately transported to the laboratory. The bacterial cells were resuspended using vortex and washed with sterile seawater to remove the excess carbon sources. Fungi, algae, and other impurities were excluded by filtration through a cellulose ester membrane, with a pore size of 3 μm (MilliporeSigma). The bacterial cells were then incubated at 4 °C for 16 h for the consumption of any remaining carbon sources.

The media used for enrichment cultivation were prepared by adding 69 different sole carbon sources (0.1%, w/v), which could be classified into four categories (alcohols, amino acids, organic acids, and saccharides; Table S1). The carbon sources were individually added to solutions containing macro- and microelements. The media were filtered through 0.22 μm polycarbonate membrane filters (MilliporeSigma) to prevent contamination. The pH of the media was adjusted to 6.5–7.5 and stabilized using HEPES buffer (1 mol/L). Starved bacteria were inoculated into these media and then incubated in the dark at 25 °C. For aerobic cultivation, the space of each container was filled with 80% N₂ and 20% O₂. For anaerobic culture, container space was filled with high-purity nitrogen (99.999%). The bacterial consortia were harvested when the cell density reached an OD₆₀₀ of approximately 1.0, after 20–30 days of incubation. For DNA extraction, bacterial culture samples were snap-frozen in liquid nitrogen and stored at − 80 °C.

Metagenomic sequencing

DNA was extracted from the bacterial cells using the QIAamp PowerFecal DNA Kit (QIAGEN). The quality of extracted DNA was assessed via gel electrophoresis (YEASEN) and NanoDrop Lite spectrophotometer (Thermo Fisher Scientific). For Illumina sequencing, 138 individual DNA libraries with 350 bp insert lengths were prepared using the NEBNext® Ultra™ DNA Library Prep Kit (New England Biolabs), according to the manufacturer’s protocol. Sequencing was conducted on the DNASEQ-T7 platform at Novogene, which generated over 20 Gb of clean data per sample through paired-end sequencing.

For Nanopore sequencing, DNA samples were classified into the four carbon categories, and samples from the same category were pooled to create four metagenomes. Libraries were prepared by fragmenting DNA with G-tubes and then repairing the ends using the NEBNext FFPE DNA Repair Mix and NEBNext Ultra II End Repair/dA-Tailing Module (New England Biolabs). Barcoding was performed using the Native Barcode Expansion Kit (Oxford Nanopore Technologies). Sequencing was carried out on the PromethION48 platform at Novogene, resulting in the generation of over 10 Gb of clean data per metagenome.

Taxonomic profiling

Illumina reads were subjected to quality control using NGS QC Toolkit (version 2.0) [43]. In this process, adaptor-containing reads, low-quality (score < 20) reads, or unpaired reads were removed. Taxonomic classification was conducted based on miTags extracted using Parallel-META3 (version 3.5.3) [44]. OTUs were clustered at 97% similarity. Bowtie2 (version 2.4.4) [45] was employed for annotation of OTUs using GreenGenes [46], RDP [47], and SILVA databases [48]. Only the OTUs detected in both paired reads were retained for taxonomic profiling at the community level.

Genome binning and annotation

Prior to genome binning, Illumina reads, obtained from aerobic and anaerobic culture with the same carbon source, were assembled into contigs using MEGAHIT (version 1.2.9) [49]. Binning was performed on the contigs longer than 5000 bp using MaxBin (version 2.2.6) [50]. The completeness and contamination levels of the resulting MAGs were assessed using CheckM2 (version 1.0.1) [51] and the MAGs exhibiting > 5% contamination or < 80% completeness were excluded. MAGs that met these criteria underwent a second round of binning using MetaBAT (version 0.32.5) [52], followed by further refinement using Concoct (version 1.1.0) [53]. All MAGs were then reassembled with Nanopore long reads. The raw MAGs were indexed using Bowtie2. The Illumina reads were subsequently mapped to this index, and matched reads were extracted by Samtools (version 1.9) [54]. Nanopore reads were aligned to the raw MAGs with BLASTn, retaining those with > 99% identity. Illumina and Nanopore reads were then combined and reassembled using SPAdes (version 3.12.0) in single genome mode [55]. The resulting MAGs were evaluated with CheckM2, and only those with < 5% contamination and > 90% completeness were retained. Subsequently, the pairwise ANI between genomes was calculated using fastANI (version 1.33) [56], and the genomes with ANI ≤ 99% were retained.

Taxonomic classification of MAGs was conducted using GTDB-Tk (version 2.4.1) [57]. A phylogenetic tree generated based on 120 bacterial universal marker gene sets revealed by GTDB‐Tk and visualized in iTOL [58]. For functional annotation, ORFs were predicted using Prodigal (version 2.5) [59] and searched against the KEGG database (version 2022) [60] via BLASTp (1e− 7), while simultaneously performing a more comprehensive gene annotation using eggNOG (version 2.1.13) [61]. To determine the genome abundance in global marine biofilm metagenomes (n = 131), Bowtie2 was used to align the metagenomic sequencing data normalized to 1 million reads. The resulting alignment files were then processed using Samtools, and the reads mapped to each genome were counted. The read count was used as an indicator of the relative abundance of a genome in the metagenomes.

AMP prediction

The sORFs were predicted using the “getorf” function of EMBOSS (version 6.6.0.0) [62], with the length of amino acid sequence set to a range of 5 to 40. The “‐find” argument was set to 2, and the universal bacterial codon table was applied. Duplicate sequences were removed using the Seqkit (version 2.1.1) [63]. The resulting sORFs were further filtered based on their net charge. Since most active AMPs are cationic peptides, a cutoff threshold of net charge was applied, and the sORFs with a net charge greater than 2 were retained. For AMP prediction, both hybrid model [7] and Macrel model [31] were applied to sORFs with default parameters to identify the cAMPs. The intersection of predictions from the two models was then analyzed to identify the final set of cAMPs.

AMP refinement and sequence analysis

Peptide sequences, with a similarity threshold of 0.9 and a minimum sequence length of 5, were clustered using CD-HIT (version 4.8.1) [64]. Clusters containing at least three sequences were considered valid, and the representative sequence of each cluster was selected. The sequence similarity of the selected cAMPs was calculated using the ‘needle’ function in EMBOSS with default parameters. This function is based on the Needleman-Wunsch algorithm. The physicochemical properties of cAMPs were analyzed using the “GlobalDescriptor” and “PeptideDescriptor” module of modlamp (version 4.3.0) [65]. The analyzed properties included molecular weight, length, charge, charge density, isoelectric point, instability index, aromaticity, aliphatic index, global hydrophobicity, hydrophobic moment, hydrophobic ratio, and Boman index.

Chemical synthesis of selected peptides

The peptides used in this study were synthesized by Sangon Biotech using the solid-phase peptide synthesis method. During the synthesis process, chemical modifications, including N-terminal acetylation and C-terminal amidation of peptide sequences, were performed. The precise molecular weight and purity of the peptides were determined by mass spectrometry and high-performance liquid chromatography. All peptides showed a purity of greater than 95%. The representative quality-inspection reports based on chromatographic and mass spectroscopic analyses are given in Fig. S11 and Fig. S12, respectively.

Antimicrobial assay

The antibacterial experiment involved seven bacterial strains, including Gram-positive strains S. aureus ATCC 12600, S. aureus ATCC 33592, and S. aureus ATCC 33591, as well as Gram-negative strains P. aeruginosa PAO1, K. quasipneumoniae, A. baumannii ATCC 19606, and A. baumannii bio-53272. All strains were cultured overnight in cation-adjusted Mueller–Hinton broth (CAMHB) at 37 °C with constant shaking at 180 rpm. The bacterial suspensions were serially diluted in fresh medium to obtain a cell density of 5 × 10⁵ CFU/mL for subsequent experiments. The antibacterial assays were conducted in 96-well plates. The final concentration of cAMPs in a well was set to 64 μg/mL. Four experimental groups were established: (1) blank control group, with 200 μL CAMHB; 2) Positive control group, with 180 μL bacterial suspension and 20 μL ultrapure water; (3) negative control group, with 180 μL CAMHB and 20 μL cAMPs solution; and (4) experimental group, with 180 μL bacterial suspension and 20 μL cAMPs solution. The OD₆₀₀ values of suspensions were measured after incubation at 37 °C for 12 h and normalized using the following formula:

$$\text{Normalized OD}=\left({\text{OD}}_{\text{experimental group}},-,{\text{OD}}_{\text{negative control}}\right)/\left({\text{OD}}_{\text{positive control}},-,{\text{OD}}_{\text{blank control}}\right)\times 100\%$$

To measure the MICs, all strains were cultured overnight at 37 °C in CAMHB. After incubation, the bacterial suspensions were diluted to 5 × 10⁵ CFU/mL. The group setup was the same as in the antibacterial assay, with cAMPs’ concentration adjusted to a gradient of 2 to 64 μg/mL. OD₆₀₀ values were measured after incubation of suspensions at 37 °C for 24 h. The MIC was defined as the lowest concentration of cAMPs at which no visible bacterial growth was observed, after accounting for the negative control (i.e., OD₆₀₀ of experimental group minus OD₆₀₀ of negative control approached zero). All experiments were performed in triplicate.

AMP structures prediction

The structures of the AMPs were predicted using ColabFold [66]. The prediction was performed with the following parameters: num_relax = 1, template_mode = pdb100, and msa_mode = mmseqs_uniref_env.

Circular dichroism (CD) assay

CD assay was performed at room temperature using a CD spectrophotometer J‐1500 (Jasco, Japan). A micro-quartz cuvette with a path length of 1 mm was used, and the wavelength range was set to 190–260 nm. The scanning speed was 200 nm/min, and the bandwidth was fixed at 1 nm. The spectra obtained from the CD spectrophotometer were smoothed using a Means-Movement smoothing filter with a value of 25. For all cAMP solutions (final concentration of 20 μM), 30% trifluoroethanol was used as the solvent.

Membrane integrity assay

Cells of S. aureus ATCC 12600 in the logarithmic growth phase were washed three times with HEPES buffer (5 mM, pH 7.4) and diluted to adjust the OD₆₀₀ to 0.1. PI (5 μM) was then added to the bacterial suspension and incubated in the dark at 37 °C for 1 h. Subsequently, cAMPs were added to the suspensions at concentrations of 0 ×, 1 ×, 2.5 ×, 5 ×, and 10 × MIC, and then the suspensions were incubated again at 37 °C for 1 h. After completion of the incubation period, fluorescence was measured at an excitation wavelength of 535 nm and an emission wavelength of 615 nm. Detected fluorescence signals were corrected by subtracting the values of PI-containing untreated cells. The assay was performed in triplicate.

Transmembrane potential dissipation assay

Cells of S. aureus ATCC 12600 in the logarithmic phase were washed three times with HEPES buffer (5 mM, pH 7.4) and diluted to adjust the OD₆₀₀ to 0.1. Thereafter, DiSC₃(5) (5 μM) was added to the cell suspension, and the mixture was incubated in the dark at 37 °C for 1 h. The cAMPs were then added to the mixture at different concentrations (0 ×, 1 ×, 2.5 ×, 5 ×, and 10 × MIC), followed by incubation at 37 °C for an additional 1 h. After incubation, the fluorescence of the mixture was recorded at an excitation wavelength of 622 nm and an emission wavelength of 670 nm. The fluorescence values were corrected by subtracting the fluorescence of untreated controls. The assay was performed in triplicate.

Scanning electron microscope (SEM) observation

S. aureus ATCC 12600 was cultured overnight at 37 °C in LB medium and then diluted to a cell density of 10⁸ CFU/mL. The cells were treated with cAMPs (at concentrations corresponding to 4 × MIC) and vancomycin at 37 °C for 1 h. Untreated S. aureus was used as a negative control. After treatment, the bacterial suspensions were centrifuged at 5000 rpm for 3 min, and the supernatant was discarded. The bacterial pellet was washed twice with sterile PBS at pH 7.0, resuspended in 200 μL of 2.5% glutaraldehyde solution, and fixed overnight at 4 °C. After fixation, the bacterial cells were centrifuged again at 4000 rpm for 10 min, and the supernatant was discarded. The bacterial cells were then uniformly spread on glass slides and washed three times with 0.1 M PBS buffer. The slides with bacterial cells were dehydrated using a series of ethanol solutions (30%, 50%, 70%, 80%, and 100%), followed by a 10-min treatment in 100% ethanol. The slides were then sequentially treated with 50% isopentyl acetate (isopentyl acetate: ethanol, 1:1) and 100% isopentyl acetate. After the dehydration and treatment steps, the slides were subjected to critical point drying using CO₂. Finally, the cells on the slide were gold-coated and observed using SEM Tescan Vega3 (TESCAN, Czech Republic).

Cytotoxicity assay

HaCaT, LO2, and HK-2 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum, 100 U/mL penicillin, and 100 μg/mL streptomycin at 37 °C in a humidified incubator with 5% CO₂. When the cells reached 80–90% of confluence, the medium was removed and the cells were washed 2 times with sterile PBS. After removing PBS, the cells were treated with 1 mL of trypsin for 1–3 min. After stopping the digestion with 3 mL of complete medium, the cell suspension was transferred to a 15 mL centrifuge tube. The cells were centrifuged at 1000 rpm for 5 min, and the supernatant was discarded. The cell pellet was resuspended in 3 mL of complete medium and subjected to subculture at a 1:3 ratio in new culture dishes. Cells at 80–90% confluence were digested with trypsin, centrifuged at 1000 rpm, and resuspended in complete medium. The cells were then seeded into 96-well plates at a density of 5000 cells per well. AMPs were added to the wells at a concentration of 64 μg/mL to set up the experimental groups. DMEM medium served as the blank group, while untreated cell suspensions were used as the control group. Each group had three replicates. The plates were gently shaken and incubated for 24 h. Subsequently, 10 μL of Methylthiazolyldiphenyl-tetrazolium bromide solution (5 mg/mL) was added to each well, and the plates were further incubated for 3 h in the incubator. After carefully removing the medium, 150 μL of dimethyl sulfoxide was added to each well. After mixing the solutions thoroughly, OD₅₇₀ was measured using a microplate reader. Cell viability was calculated using the following formula:

$\text{Cell viability}=\left({\text{absorbance}}_{\text{experimental}},-,{\text{absorbance}}_{\text{blank}}\right)/\left({\text{absorbance}}_{\text{control}},-,{\text{absorbance}}_{\text{blank}}\right)\times 100\%$

Hemolysis assay

A 50 mL aliquot of 10% sheep red blood cell suspension was centrifuged, and the supernatant was discarded. The obtained red blood cells were washed twice with physiological saline until the supernatant became clear. The red blood cells were then diluted with 250 mL of physiological saline to obtain a 2% red blood cell suspension. In separate centrifuge tubes, 400 μL of AMP solutions were taken, and then 100 μL of 2% red blood cell suspension was added to obtain a final AMP concentration of 64 μg/mL as the experimental group. A 2% Triton X-100 solution in physiological saline was used as the positive control, while physiological saline was used as the negative control. The samples were mixed thoroughly and incubated at 37 °C in a water bath for 1 h. After incubation, the samples were centrifuged at 3000 rpm for 10 min at room temperature. Subsequently, 200 μL of supernatants was transferred to a 96-well plate, and their OD values were measured at 595 nm using a microplate reader. The hemolysis rate for each sample was then calculated as follows:

$$\text{Hemolysis}=\left({\text{absorbance}}_{\text{experimental}},-,{\text{absorbance}}_{\text{negative}}\right)/\left({\text{absorbance}}_{\text{positive}},-,{\text{absorbance}}_{\text{negative}}\right)\times 100\%$$

AMP stability assay

The AMPs with a concentration of 6.4 mg/mL were added to PBS buffer solutions with different pH (pH 2, pH 7.4, and pH 8) respectively, resulting in a final peptide concentration of 640 μg/mL. The mixtures were then incubated at 37 °C for 1 h. After incubation, the MIC of AMP against S. aureus 12,600 was determined as previously described. To assess the proteolytic resistance, AMPs with a concentration of 6.4 mg/mL were mixed with pepsin in PBS buffer (100 μg/mL, pH 2), resulting in a final AMP concentration of 640 μg/mL. Vancomycin (Letco Medicine) and melittin (Sigma-Aldrich) were included as stable and unstable controls, respectively. The mixtures were then incubated at 37 °C for 1 h to allow enzymatic digestion. The stability of AMP after pepsin digestion was evaluated by measuring the change in its MIC against S. aureus ATCC 12600.

Murine infection experiments

Three S. aureus strains (ATCC 12600, ATCC 33591, and ATCC 33592) were inoculated into LB medium and cultured. Harvested bacterial cells were washed twice with sterile PBS (pH 7.4) and then resuspended in PBS to obtain an OD₆₀₀ of 0.6. The three strains were then mixed (1:1:1), and the concentration of bacterial cells in the mixed suspension was adjusted to the desired level using a McFarland nephelometer. Six-week-old female Kunming mice were used for this experiment. Prior to experimentation, all mice underwent a 5-day acclimatization period upon arrival from a commercial supplier. The mice were anesthetized using isoflurane. Each experimental group consisted of three mice (n = 3), and all mice were housed individually to prevent cross-contamination.

To construct the skin abrasion infection model, fur was shaved from the dorsal area of the mouse, specifically the region that could not be reached by mouth or limbs. A 12 × 12 mm superficial abrasion wound was created. After cleaning the wounds with sterile saline, 20 μL of mixed bacterial suspension (1 × 10⁸ CFU/mL) was applied to the wounded area. Two hours post-infection, 20 μL of a peptide solution (32 μg/mL of cAMP-1111-nc, cAMP-1229-nc, cAMP-502-nc, or cAMP-77-nc) was administered to the wound. In the control group, sterile PBS was administered instead of the peptide solution. After 24 h post-infection, the mice were euthanized. A 5 × 5 mm section of infected tissue from the wounded area was excised, homogenized at 25 Hz, and serially diluted 10 times. Each dilution was spread on three replicate agar plates. After 24 h of incubation, CFU counting was performed. Additionally, the body weight of the mice was monitored at each stage of the experiment.

To construct the peritonitis infection model, mice were intraperitoneally injected with 200 μL of mixed bacterial suspension (1 × 10⁹ CFU/mL). One hour after infection, the control group received sterile PBS via intraperitoneal injection, while the treatment groups received intraperitoneal injections of a peptide solution (4.5 mg/kg cAMP-1111-nc, cAMP-1229-nc, cAMP-502-nc, cAMP-77-nc, or vancomycin). Around 24 h post-infection, mice were injected intraperitoneally with 5 mL of sterile PBS for peritoneal lavage. The lavage fluid was collected along with liver and kidney tissues. The tissue samples were serially diluted 10 times, and the diluted solutions were plated on agar to analyze the bacterial load. In addition, the harvested organs were subjected to histological analysis using H&E staining, and representative images were captured.

Authors’ contributions

W. Z. designed the project. S. F. and J. L. were involved in experimental design and performed the majority part of experiments and computational analyses. H. C. was involved in the experiments. Y. L., S. L., W. D., and J. S. provided experimental guidance or constructive comments. W. Z. and S. F. wrote the manuscript.

Funding

This work was supported by the Original Exploration Project from Qingdao (25–1-1–254-zyyd-jch), the Blue Medicine Bank Project from the Marine Biomedical Research Institute of Qingdao (YZ1-25–05), the Research Fund from Shandong Province (862205033005), and the Fundamental Research Fund for Central Universities (842441005).

Data availability

The 138 enrichment culture-derived metagenomes have been deposited in the China National Center for Bioinformation (CNCB) under accession number CRA006934. Please see Table S1 for the accession number for each metagenome. The 435 MAGs can be obtained from NCBI under the accession numbers PRJNA1211307. The 131 global marine biofilm metagenomes are available from the NCBI database under the BioProject accession numbers PRJNA438384, PRJNA753157, and PRJNA902427.

Declarations

Ethics approval and consent to participate

All experiments were conducted with the approval of the Institutional Animal Care and Use Committee of Ocean University of China, in accordance with the governmental regulations of China.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.