Goat gut microbiome as a reservoir for microorganism-encoded short peptides: regulation by host development age and nematode challenge

Abstract

Short peptides are a promising class of antimicrobial agents with efficacy against multidrug-resistant pathogens. While host-derived defense peptides from animals and plants have been widely characterized, the landscape of microbial-encoded short peptides, particularly those with antimicrobial activity that can mitigate antibiotic resistance, remains underexplored. In this study, we identified 1147 novel microorganism-encoded short peptides (MSPs) from the goat gut microbiome, most of which lack homology to existing database entries. The microbial communities producing these MSPs were altered by the developmental stage of the goat, whereas the overall composition and diversity of these populations remained unchanged during nematode infection, despite significant correlations between certain MSPs and type 2 immunity. Notably, two MSPs demonstrated potent antimicrobial activity with lower cytotoxicity, exhibiting remarkable efficacy against methicillin-resistant Staphylococcus aureus and Streptococcus agalactiae. Collectively, our findings establish the goat gastrointestinal tract microbiome as a distinctive and promising reservoir for the discovery of antimicrobial and immunoregulatory peptides.

Introduction

Short peptides, defined as those with less than 50 amino acid residues, represent a promising class of antimicrobial agents effective against multidrug-resistant pathogens^1,2. These peptides can be encoded by either the host or microorganisms. Host-derived peptides, often referred to as host-defense peptides, are widely distributed across the animal and plant kingdoms. Their antimicrobial activities, immunoregulatory characteristics, and therapeutic potential in the context of drug-resistant infections and inflammatory diseases have been well documented^3–7. More recently, peptides encoded by microorganisms, along with their antimicrobial activities and clinical applicability, have also been reported^8–11.

Although host-defense peptides and microorganism-derived short peptides hold promise as clinical therapeutics, few have reached commercial availability due to challenges in their identification and the absence of standardized discovery procedures. Various approaches, such as screening natural sources, introducing amino acid residue substitutions^12–16, applying machine learning^10,17–20, genetic algorithms^21,22, and pattern recognition algorithms^23,24, have been successively employed to identify candidate antimicrobial peptide (AMP). Limited efforts, however, have been directed toward mining metagenomes^8,9, which indicate that gut symbiotic microbes in humans and honeybees represent a significant reservoir of such peptides. Notably, AMPLiT, a lightweight AI-driven computational pipeline, demonstrates high efficiency in screening functional AMPs from metagenomic datasets of ancient gut microbiomes²⁵. Similarly, EBAMP, a generative-discriminative framework for de novo design of broad-spectrum AMPs, shows a high positive rate in screening functional AMPs²⁶.

The ruminant gastrointestinal tract (GIT), consisting of ten compartments, including four foregut chambers with distinct environmental conditions, hosts a highly complex and diverse microbiome^27,28. It is reported that 39 out of 27,192 peptide candidates identified from seven varieties of ruminants, including goat, demonstrate antimicrobial activity against Staphylococcus aureus (S. aureus), with one peptide showing effectiveness against methicillin-resistant S. aureus (MRSA)⁸. Distribution of biosynthetic gene clusters potentially associated with the production of lasso peptides with antimicrobial activity in the rumen bacterial genomes has also been characterized²⁹. Additionally, 195 rumen lanthipeptide biosynthetic gene clusters have been identified by metagenomic binning approaches³⁰. These researchers infer that the ruminant gut microbiome presents considerable untapped potential for the discovery of novel antimicrobial peptides⁸.

Production of antimicrobial peptides by bacteria is stimulated by nutrient depletion or by specific signals linked to reduced growth rates³¹. It remains unclear whether signals originating from nematode infection or changes associated with developmental age influence antimicrobial peptide production. In the present study, we therefore first sought to identify microorganism-encoded short peptides (MSPs) from 3975 metagenome-assembled genomes (MAGs) obtained across the gastrointestinal tract (GIT) of goats. For this purpose, computational algorithms were employed to predict the antimicrobial potential of putative MSP sequences, yielding a final set of 1147 candidates. Analysis further revealed that the MSPs-producing microbial community and MSPs abundance in the goat GIT were altered by developmental stage, but not by nematode infection. From these 1147 candidates, along with differentially expressed MSPs identified in the developmental and nematode infection trials, seventeen candidate MSPs were chemically synthesized and subsequently subjected to in vitro testing against clinically prevalent pathogens. Two peptides demonstrated antimicrobial and bacteriocidic activities and lower cytotoxicity and hemolysis, among which MSP_121684_1 showed high potency against Streptococcus agalactiae (S. agalactiae), and MSP_375782_4 was effective against MRSA. Taken together, our study implemented a progressive, hypothesis-driven workflow to systematically identify, validate, and contextualize MSPs with dual in vivo and in vitro relevance. By integrating metagenomic prediction, host-physiology-informed in vivo screening, and mechanistic in vitro functional assays, this approach establishes a robust link between MSPs functionality and host-relevant outcomes, thereby ensuring the efficacy and safety of identified active MSPs for translational applications.Table 1

Table 1.

The minimum inhibitory concentrations of two candidate MSPs

Peptide	Minimum inhibitory concentration (μmol/L)
	S. agalactiae	MRSA	E. coli	S. enterica	S. aureus
MSP_121684_1	128	>128	>128	>128	>128
MSP_375782_4	64	128	>128	>128	>128

S. agalactiae: Streptococcus agalactiae.

E. coli: Escherichia coli.

S. enterica: Salmonella enterica.

MRSA: Methicillin-resistant Staphylococcus aureus.

S. aureus: Staphylococcus aureus.

Results

Constructing a microbial metagenome-assembled genomes catalog in the gastrointestinal tract of goats

We collected 99 content samples covering four gastrointestinal compartments, including rumen (n = 28), abomasum (n = 29), caecum (n = 30), and colon (n = 12) of goats. DNA of these samples was extracted, quality-evaluated, and sequenced on the DNBSEQ-T7 platform. Finally, over 1.96 terabytes (Tb) of metagenomic data were obtained with an average of 20.3 gigabytes (Gb) per sample, totaling 14.06 billion sequencing reads. Following the removal of host (Capra hircus) DNA contamination, a total of 10.80 billion sequencing reads of the gastrointestinal tract (GIT) microbial data remained for the subsequent analyses (Additional file 1: Table S1).

We generated a total of 38.54 million contigs longer than 500 bp and 87.73 million open reading frames (ORF) with an average length of 707 bp (Additional file 1: Table S1). 14,899 medium- and high-quality (more than 50% completeness and less than 10% contamination) metagenome-assembled genomes (MAGs) were obtained. After clustering at 95% nucleotide sequence identity³², 3975 medium and high quality non-redundant MAGs were retained as the goat GIT microbial genome catalog (GMGC) and used for the downstream analysis (Additional file 2: Table S2). Therein, 81% of these non-redundant MAGs were assessed as high-quality (more than 90% completeness and less than 5% contamination).

To understand the phylogenetic position of all MAGs, genome information, and biosynthetic gene clusters (BGCs) classification in the GMGC, we then built a maximum-likelihood phylogenetic tree. It showed that the GMGC was dominated by two large clusters representing Bacillota_A and Bacteriodota (Fig. 1a), which aligns with previous findings in the rumen and colon of goats (e.g., Jiao et al.³³). Using GTDB-Tk, 98.39% and 1.58% of the MAGs were assigned to bacteria and archaea, and the rest (0.03%) MAGs were unclassified (Fig. 1b and Additional file 3: Table S3). These MAGs were most abundant in the caecum (1333), followed by the rumen (1194), abomasum (901), and least abundant in the colon (547, Fig. 1c). Based on the number of GIT samples used in this study, this means that colon content produced the highest number of MAGs (45.6 MAGs per sample), while the rumen content generated the lowest number of MAGs (only 31.1 MAGs per sample). Diversity results showed that Shannon indices of the rumen, abomasum, and colon were greater than that of the caecum (P < 0.05, Fig. 1d), indicating that the species richness and evenness of the microbial community in these three segments were higher than that in the caecum. Meanwhile, principal coordinate analysis (PCoA) based on the Bray–Curtis distance algorithm also showed that the microbial community in the caecum is significantly different from that of other intestinal segments from the first dimension (P < 0.05), this dissimilarity was observed between the large intestine and the foregut from the second dimension (P < 0.05, Fig. 1e).

Fig. 1. 3975 metagenome-assembled genomes (MAG) catalog was obtained from the gastrointestinal tract (GIT) of goat (GMGC).

a The maximum-likelihood tree of the MAGs. The inner layer is the taxonomy at the domain level, second circle layer is the MAGs completeness, third layer is the MAGs contamination, fourth layer is the GC content of the MAGs, fifth layer is biosynthetic gene clusters-nonribosomal peptides (BGC-NRP), sixth layer is BGC-NRP-Polyketide, seventh layer is BGC-Polyketide, eighth layer is BGC-ribosomally synthesized and post-translationally modified peptide, ninth layer is BGC-Terpene, tenth layer is BGC-Saccharide and the eleventh layer (outermost) is the genome size. b Taxonomic annotations for the GMGC. c Distribution of MAGs in each GIT compartment; d Shannon indexes of each GIT compartment. Statistical analysis: Kruskal–Wallis test. e Beta-diversity of each GIT compartment. Statistical analysis: permutational multivariate analysis of variance. Different lowercase letters within the same gastrointestinal tissue indicate statistically significant differences.

Identifying more than one thousand MSPs in the GIT of goats

Using three prediction methods with AMPir³⁴, AmPEP³⁵, and APIN³⁶, 1147 MSPs were identified (Additional file 4: Table S4). The length of these cMSP starts from 12 amino acids, and ends with 49 amino acids, therein the percentage of MSPs with length between 41–49 amino acids accounts for 73.41% (Fig. 2a). The distribution of MSPs -based bins in the rumen was the greatest (34.2%), followed by caecum (31.0%) and abomasum (24.6%), while this distribution in the colon was the lowest (10.3%, Fig. 2b). The distribution pattern of MSPs-based bins is not entirely consistent with that of MAGs. This result infers that the rumen and caecum are the important segments of the genome that encode MSPs. If the total amount of MAGs in each segment is taken into account, the efficiency of encoding MSPs in the foregut is greater than that in the hindgut. To figure out the microbial origin of MSPs, we quantified the microbial annotations of the MAGs that produce more than one MSP, and found that the proportion of Bacillota_A- and Bacteriodota-encoded MSPs exceeded 50% (Fig. 2c), contributing the main source of MSP-encoding genes. Additionally, Pseudomonadota, Bicillota_C, and Verrucomicrobiota were also the microbial sources of MSPs.

Fig. 2. 1147 microorganism-encoded short peptides (MSPs) were identified from the GMGC.

a Length distribution of all the identified MSPs. b Proportion of MSPs-based bin in each GIT compartment. c Classification and distribution of MSPs-encoding bacteria at the phylum level. d Network of the negative correlation between MSPs and bacterial species in the rumen, abomasum, and caecum community.

The relationship between the microbiome and MSPs was further investigated. In the rumen, there are 17 MSPs negatively correlated with 677 MAGs (Additional file 5: Table S5). In the abomasum, the abundance of seven MSPs was negatively correlated with 152 MAGs (Additional file 6: Table S6). In the cecum, 34 MSPs abundance was negatively correlated with 61 MAGs (Additional file 7: Table S7). Unfortunately, the negative correlation between MSPs' abundance and MAG’s abundance in the colon was not observed. Specifically, the interaction network in the rumen was more complex and extensive than that in the abomasum and caecum, wherein five types of MSPs occupy a central position and have a close interaction with other MAGs. Among these MAGs, Bacteroidota and Bacillota_A and Bacillota_C had a tight interaction with MSPs in each community. Altogether, a total of 57 MSPs abundance was correlated with the relative abundance of bacterial species (Fig. 2d). Among these MSPs, only 13 most potent MSPs were selected for further activity testing.

MSPs-producing microbial community and MSPs abundance in the GIT of the goat are altered by the developmental stage

To gain a deeper understanding of developmental changes of MSPs and related microorganisms, we collected 50 gastrointestinal digesta samples from goats with different developmental periods, including 14 days (D14, breast milk suckling period), 42 days (D42, milk and starter mix feeding period) and 105 days (D105, solid diet feeding period) and corresponding DNA samples were subjected to metagenomic sequencing. The body weight of the experimental goats increased gradually with the increase of age, and there was a significant difference between each two groups (P < 0.05, Fig. 3a). The PCA analysis showed that MSPs-encoding MAGs in the D105 group were obviously different from those of D14 and D42 groups, while there was no difference between D14 and D42 group (Fig. 3b). For the alpha diversity parameters, both microbial richness and Shannon indexes in the rumen were higher in the D42 and D105 groups compared to the D14 group (P < 0.05). In the abomasum, the D105 group exhibited greater microbial richness and Shannon indexes than the D42 group (P < 0.05). Notably, microbial richness and Shannon indexes in the caecum displayed dynamic changes across developmental stages (P < 0.05, Fig. 3c). At the phylum level, except Bacillota_C, Bacillota_I, and Elusimicrobiota, relative abundance of other major microorganisms, including Bacillota_A, Bacteroidata, Pseudomonadota, Verruconmicrobiota, Synergistota, Bacillota, and Spirochaetota, in the rumen, abomasum and caecum was altered with the developmental period (Fig. 3d). At the species level, relative abundance of UBA3857 sp900313585, Lawsonibacter sp000177015, JAFUYV01 sp016280565, Pelethomonas sp944387775 in the rumen, abomasum and caecum was changed with the age of goats (Fig. 3e). The heatmap of MSPs in each segment showed that the abundance of two MSPs (MSP_69132_1 and MSP_1844_1) in the rumen, abomasum and caecum was uniformly declined at D105. It seemed that most MSPs' abundance tendency in the rumen and abomasum was quite similar. Eleven MSPs abundances in the cecum were increased at D105 compared to D42 and D14 (Fig. 3f). The function of these differential MSPs s affected by the ages of goats needs to be further clarified.

Fig. 3. Developmental changes of MSP-encoding bacteria and MSP abundance.

a Average body weight of the goats with different ages. Data are shown as mean ± SD (n = 12 for each group). Statistical significance was assessed using one-way ANOVA with the least significant difference test. b PCA analysis of the GIT microbiome at the MAG level based on Bray–Curtis distance. c Alpha-diversity of the GIT microbiome for each developmental age (n = 6). Different lowercase letters in each segment indicate significant differences (P < 0.05). d Relative abundance of MSP-encoding microbiome at the phylum level. The symbol (*) in the parentheses of the legend means P < 0.05, ** indicates P < 0.01, and *** means P < 0.001. e Relative abundance of MSP-encoded microbiome at the species level. f Top50 MSP abundance among different developmental ages in the rumen, abomasum, and caecum, respectively. The top 50 MSPs with the highest accumulated RPKM (Reads Per Kilobase per Million mapped reads‌) abundance were listed. Statistical analysis: Kruskal-Wallis test. Different lowercase letters within the same gastrointestinal segment indicate statistically significant differences (P < 0.05).

MSPs-producing microbial community and MSPs abundance in the GIT of goats were slightly affected by the disturbance from the nematode infection

To figure out whether MSPs-encoding microorganism and MSPs abundance in the GIT of goats were altered by the nematode infection, 49 gastrointestinal digesta samples from uninfected goats (CK group) and goats infected with Haemonchus contortus (Infected group) were collected for analysis. The PCA analysis showed that MSPs-encoding MAGs in the CK group did not differ from that of the infected group (Fig. 4a). The microbial indexes of Richness, Shannon and Simpson in both groups were not changed in each compartment (P > 0.05, Fig. 4b). In the rumen, Bacteroidota and Bacillota_C abundance in the infected group was greater than that of CK group (P < 0.05). Meanwhile, Bacillota_I abundance in the cecum of the infected group was also greater than that of the CK group (P < 0.05, Fig. 4c). At the species level, abundance of UBA1189 sp017622495, Prevotella sp900319305, and sp902776195 in the rumen of the infected group was greater than that of the CK group (P < 0.05). Abundance of Prevotella sp902776195 and Sodaliphilus sp902762385 in the abomasum of the infected group was also increased compared to that of the CK group (P < 0.05). Abundance of UBA3766 sp902765365 in the colon of the infected group was increased compared to the CK group (P < 0.05, Fig. 4d).

Fig. 4. Effects of nematode infection on the abundance of MSP-encoding bacteria and MSP abundance.

a PCA of the GIT microbiome at the MAG level based on Bray-Curtis dissimilarity. b Alpha-diversity of GIT microbiome in CK (n = 6) and infected group (n = 6). Data are shown as mean ± SD. c Relative abundance of MSP-encoding microbiome at the phylum level. The symbol (*) in the parentheses of the legend means P < 0.05. d Relative abundance of MSP-encoding microbiome at the species level. * means P < 0.05 and **P < 0.01. e The Top50 MSP abundance in the rumen, abomasum, and caecum, respectively. Different lowercase letters within the same gastrointestinal segment indicate statistically significant differences (P < 0.05). f Correlation between MSP abundance and phenotypic parameters. Spearman’s rank correlation analysis was employed. The symbol (+) in the dot means P < 0.05, while ++ indicates P < 0.01. Statistical analysis: Mann–Whitney U test.

Heatmap illustrated that MSP_121684_1 abundance in the rumen and MSP_175466_13 abundance in the colon were decreased in the infected group (P < 0.05). The abundance of MSP_375782_4 and MSP_265523_11 in the abomasum and MSP_485477_52 in the colon was stimulated in the infected group (P < 0.05). No significant changes in the caecum were observed between the two groups (Fig. 4e). Due to the important role of type 2 cytokines and immunoglobulins in stimulating type 2 immunity in the GIT mucosa, we performed the correlation analysis of MSPs abundance with phenotypic indicators of goats. It turned out that MSP_435445_18 and MSP_378620_1 abundance in the rumen was positively correlated with the dry matter intake of goats. MSP_87582_1 abundance was negatively correlated with type II immune cytokines (IL-4/5/10) and IgA concentration in the blood. MSP_268_4 abundance in the rumen was positively correlated with IgG concentration in the blood. Abundance of MSP_308587 _1 and MSP_471768_3 in the rumen was negatively correlated with IgE concentration in the blood. In the abomasum, MSP_451638_1 and MSP_237663_1 abundance were negatively correlated with dry matter intake and IL-10 in the blood, respectively. MSP_53734_1 and MSP_731_1 abundance were negatively correlated with IgE concentration in the blood. In the caecum, MSP_626962_1, MSP_164690_1, and MSP_348284_7 abundance were negatively correlated with IL-5 concentration in the blood. In the colon, MSP_471768_3 abundance was negatively correlated with IgG concentration in the blood (Fig. 4f). It seems that the number of MSPs associated with phenotypic indicators was the greatest in the rumen, followed by the colon and the abomasum, and the least in the caecum.

Two candidate MSPs showed high efficacy against Streptococcus agalactiae and methicillin-resistant Staphylococcus aureus

To assess the antimicrobial potential of the candidate MSPs, we chemically synthesized 17 sequences (Supplementary Table S8) and tested their activity against five clinically relevant pathogenic strains: two Gram-positive strains (S. agalactiae ATCC 12386 and S. aureus ATCC 29213), two Gram-negative strains (Escherichia coli ATCC 25922 and Salmonella enterica ATCC 13076), and one antibiotic-resistant strain (methicillin-resistant S. aureus ATCC 43300). We observed that 8–128 μmol/L MSP_121684_1 inhibited the growth of methicillin-resistant S. aureus by 43.3–62.1% (Fig. 5a). The same peptide at 32–128 μmol/L also suppressed the growth of S. agalactiae by 74.4–86.3% (Fig. 5b). Notably, 8–128 μmol/L MSP_375782_4 exhibited a stronger inhibitory effect against S. agalactiae, with an inhibition rate of 59.7–78.6% (Fig. 5c). Three-dimensional imaging and secondary structure analysis of these two MSPs revealed distinct structural features. Unlike MSP_121684_1, MSP_375782_4 predominantly adopted extended strand and random coil conformations, with no evidence of α-helix or β-sheet (Fig. 5d, g).

Fig. 5. Evaluation of the in vitro antimicrobial efficacy of the candidate MSPs.

a Inhibitory rate curve of MSP_121684_1 addition on the growth of methicillin-resistant Staphylococcus aureus. b Inhibitory rate curve of MSP_121684_1 addition on the growth of Streptococcus agalactiae. c Inhibitory rate curve of MSP_375782_4 addition on the growth of Streptococcus agalactiae. d Three-dimensional ribbon structure of MSP_121684_1. e Three-dimensional ribbon structure of MSP_375782_4. f The proportion of secondary structure of MSP_121684_1. g The proportion of secondary structure of MSP_375782_4. Values are expressed as means ± SD of replicates (n = 3). Statistical significance was assessed using one-way ANOVA with Least Significant Difference test. Comparison of different concentrations for the same MSP treatment: Different letters indicate a statistically significant different (P < 0.05).

Two candidate MSPs demonstrated potent bactericidal activity while maintaining low cytotoxicity and hemolytic activity

To further evaluate the biosafety of the two candidate MSPs (MSP_121684_1 and MSP_375782_4), we performed a comprehensive series of in vitro experiments, including cytotoxicity assays, hemolysis tests, time-dependent killing kinetics, and membrane permeability analyses. Both MSPs exhibited no significant cytotoxicity at concentrations ranging from 32 to 256 μmol/L (P > 0.05, Fig. 6a, b). However, cell viability sharply declined at 512 μmol/L, indicating a concentration-dependent toxicity threshold. Treatment with 32–512 μmol/L of either MSP resulted in hemolysis rates below 20%, suggesting minimal red blood cell damage. Notably, MSP_375782_4 exhibited lower hemolysis than MSP_121684_1 at equivalent concentrations (Fig. 6c, d). The bactericidal efficiency of both MSPs was evaluated against S. agalactiae and MRSA at 1×, 2×, and 4× MIC. Both MSPs rapidly reduced bacterial viability within 0.5 h (Fig. 7a, b). MSP_375782_4 exhibited detectable killing activity only after 3 h of exposure (Fig. 7c). Both MSPs induced a concentration-dependent increase in SYTOX Green fluorescence (P < 0.05, Fig. 7d, e), confirming membrane disruption. All tested concentrations of MSP_121684_1 and MSP_375782_4 generated significantly stronger fluorescence signals than the negative control (P < 0.001), indicating effective bacterial membrane permeabilization.

Fig. 6. Assessment of the in vitro cytotoxicity and hemolysis activity of candidate MSPs.

a The cytotoxicity of MSP_121684_1 against HEK293 cells. b The cytotoxicity of MSP_375782_4 against HEK293 cells. c Hemolysis rates of MSP_121684_1 on mouse red blood cells, with the 2% Triton X-100 group serving as the positive control (PC) and the PBS group serving as the negative control (NC). d Hemolysis rates of MSP_375782_4 on mouse red blood cells. Values are expressed as means ± SD of replicates (n = 3). Statistical significance was assessed using one-way ANOVA with least significant different test. Comparison of different concentrations for the same MSP treatment: Different letters indicate a statistically significant different (P < 0.05).

Fig. 7. Antibacterial curve and cell membrane permeabilization of candidate MSPs.

a Time kill kinetics of MSP_121684_1 against S. agalactiae. b Time kill kinetics of MSP_375782_4 against S. agalactiae. c Time kill kinetics of MSP_375782_4 against Methicillin-resistant S. aureus. d Cell membrane permeability of MSP_121684_1 effect on S. agalactiae. e Cell membrane permeability of MSP_375782_4 effect on S. agalactiae. f Cell membrane permeability of MSP_375782_4 effect on methicillin-resistant S. aureus. Triton-X 100 (%) was used as a positive control and PBS buffer as a negative control. Data are expressed as means (±SD) of replicates (n = 3).

Discussion

Microbiome-derived antimicrobial peptides (AMPs), with their unique structural features and pathogen-specific targets, hold therapeutic promise, driving high-throughput mining efforts from diverse sources (e.g., human/cockroach gut microbiomes^9,11,17,25 and rumen bacteria^29,37), including recent deep-learning approaches⁸. Our study advances AMP discovery via an integrated computational-experimental platform combining metagenomic analysis, in silico antimicrobial prediction, and targeted in vitro validation. Leveraging 3975 goat gut MAGs, we identified 1147 candidate MSPs, of which 17 (selected based on metagenomic and animal trial data) were synthesized and tested. This yielded potent peptides (e.g., MSP_121684_1 against S. agalactiae and MSP_375782_4 against MRSA) with low cytotoxicity. Notably, our findings suggest MSPs may have dual antimicrobial-immunomodulatory roles, implicating their role in host-microbe interactions. This ecologically informed, mechanism-driven strategy enables scalable, translationally relevant AMP discovery.

We analyzed the 410 entries of bacterial antimicrobial peptides with known activity from Antimicrobial Peptide Database³⁸ and found that approximately 53% of these peptides have a length of 20–50 amino acids. About 15% of these peptides have a length of more than 50 amino acids, and among them, a very small proportion (about 1.5%) have a length of more than 100 amino acids. The cMSPs with lengths of from 12 to 49 amino acids in this study are in line with the bacterial antimicrobial peptide clusters in terms of amino acid length. Our cMSPs shared no common peptide with the antimicrobial peptides in the global microbiome (Supplementary Fig. 1). Most of our cMSPs (43.9%) were previously unknown and lacked detectable homologs in other databases. The absence of detectable homologs may stem from limitations in our capacity to reliably detect such homologous relationships in short sequences. However, it is also plausible that short proteins are more prone to be generated de novo than longer proteins and may have undergone convergent or independent evolution across diverse taxa³⁹.

It is reported that Firmicutes and Bacteroidota were the most phyla to produce secretory antimicrobial peptides, and the distribution of antimicrobial peptides-encoding MAGs was region-specific along the GIT of ruminants⁸. We also observed that the distribution of the MSP-encoding genomes in different GIT segments shows regional specificity, but the main phyla encoded are a little different. Primarily, it is Bacillota_A and Bacteroidota, which encode nearly half of the total MSPs. It is certain that Bacteroidota, which encode MSPs, are an important type of bacteria in the GIT of ruminants. Perhaps the difference in bioinformatics analysis methods (deep learning-driven approach vs. computational algorithms) leads to the inconsistency in the MSP-encoding microorganisms.

The GIT microbiota of goats undergoes rapid assembly after birth, with its abundance and functional potential changing substantially throughout development. Early-stage microbial communities (before day 10) form distinct clusters, while later stages (days 25, 30, and 90) exhibit greater similarity, regardless of the GIT region (rumen, ileum, or colon)³³. The microbial composition and metabolism of the GIT are strongly influenced by diet. In dairy cattle, a forage-based diet increases fiber-degrading taxa and decreases hydrogen sinks in propionate in the four-chambered stomach⁴⁰. Similarly, pelleted-hay alfalfa boosts rumen bacterial richness, particularly among fibrolytic bacteria, compared to loose-hay alfalfa⁴¹. A fiber-rich diet promoted the growth of fibrolytic bacteria and methanogens, leading to improved fiber utilization but also elevated methane emissions. In contrast, a starch-rich diet favored amylolytic bacteria and lactate utilizers, helping to maintain rumen homeostasis and reduce methane production in beef cattle⁴². These dietary effects on stomach microbiota highlight the role of substrate availability in shaping microbial community structure and function, a principle that also applies to non-ruminant species during dietary transitions. The gut microbiota becomes more complex during the transition from infancy to childhood, especially with the introduction of solid foods⁴³. Compared to milk-fed infants, solid-food-fed children show greater microbial diversity and the emergence of new carbohydrate-digesting species, reflecting a significant ecological shift in the gut⁴⁴. Consistent with these findings, we also observed an increased bacterial diversity and altered abundance of phyla and species after the introduction of the pelleted starter feed (after day 30), further supporting the critical role of diet type in shaping the composition of the GIT microbiota.

In this study, we observed a higher proportion of MSPs in the caecum at D105, accompanied by increased bacterial richness and elevated abundance of CAG-791, UBA1067, JAFUYV01, and Pelethomonas, suggesting a potential functional linkage. CAG-791 has been identified as a dominant genus in the rumen of breastfed buffalo, where its abundance correlates positively with carbohydrate-active enzymes involved in lactose degradation, amino acid synthesis pathways, and antibiotic-producing pathways⁴⁵. In the meat sheep subjected to rumen microbiota regulation, CAG-791 showed a negative correlation with TNF-alpha and a positive correlation with total antioxidant capacity⁴⁶. Collectively, these observations indicate that the abundance of MSPs across the rumen, abomasum, and caecum is developmentally regulated and dynamically responsive to dietary transitions. This finding aligns with broader evidence of functional heterogeneity among microbiome constituents. For example, single-cell transcriptomic profiling across 2534 microbial species in the rumen has demonstrated that even closely related taxa can exhibit distinct transcriptional and functional profiles⁴⁷, highlighting the potential for age-dependent expression of microbial-encoded molecules such as short peptides. Moreover, our findings imply that the transition to a fully solid diet after weaning likely fosters microbial community maturation (e.g., CAG-791 expansion) and cellular proliferation, thereby creating a favorable environment for the synthesis of MSPs associated with these functionally relevant taxa. The causal relationship and underlying mechanisms between the introduction of a fully solid diet post-weaning and the synthesis of CAG-791 and associated MSPs should be investigated in the near future.

Our previous research^48,49 and Haemonchus contortus infection in sheep⁵⁰ revealed a significant reduction in alpha diversity in the rumen and abomasum, as assessed by 16S rRNA sequencing. In contrast, the current study did not detect statistically significant changes in alpha diversity in these foregut compartments, even in the caecum and colon. These inconsistencies suggest that metagenomic approaches may provide a more comprehensive assessment of microbial community diversity, as they encompass a broader taxonomic range, including archaea and a greater number of operational taxonomic units (OTUs). Furthermore, our findings indicate that foregut microbiota may exhibit a more rapid adaptive response to nematode infection compared to hindgut communities, a pattern that aligns with observations from sheep infection experiments^50,51.

The lack of significant change in MSP abundance during nematode infection (Fig. 4) suggests functional resilience in the goat gut microbiome. Two hypotheses may explain this: (1) Functional redundancy among MSP-producing microbes could maintain collective antimicrobial capacity despite community shifts, as different taxa may produce similar peptides. (2) Compensatory mechanisms might dynamically adjust MSP expression or activity under infection stress, preserving function even if taxonomic diversity remains stable. These hypotheses align with microbial ecology concepts where functional redundancy (e.g., feed degradation in the rumen⁵² or honeybee-Varroa systems⁵³) ensures robustness despite taxonomic or niche variations. Applying relative entropy⁵⁴ or ultra-deep meta-proteomics⁵⁵ to analyze functional redundancy at the single-trait or proteome level would provide a powerful approach to experimentally validate these proposed mechanisms.

Nutrient depletion or specific signals associated with reduced growth rates are known to trigger bacterial antimicrobial peptides production³¹. Furthermore, a nematode-derived antimicrobial peptide, specifically a metridin-like ShK toxin from Haemonchus contortus, has been shown to lyse cells and reduce the growth rate of Bacillus subtilis in vitro⁵⁶. This suggests that antimicrobial peptides may act as key mediators in the direct interactions between nematodes and gut microbiota. Our previous work demonstrated that nematode infection increases free amino acid levels in the foregut digesta, likely because the parasites require more host amino acids to maintain infectivity⁴⁹. While these elevated free amino acids also support microbial growth, they may explain the lack of bacterial MSP stimulation in the present study.

It is the first report describing a correlation between MSPs and type 2 immunity. Under conditions of nematode infection, MSPs appear to shift their roles from shaping microbial communities to modulating immune responses. Previous studies have shown that host-derived antimicrobial peptides, such as resistin-like molecule beta (RELMβ)⁵⁷ secreted by activated Paneth cells, contribute to the expulsion of intestinal nematodes (e.g., Nippostrongylus brasiliensis and Heligmosomoides polygyrus). The expression of RELMβ in the epithelium is synergistically induced by IL-4/IL-13 and parasitic pathogens⁵⁸. This peptide may enhance type 2 immune response by prolonging IL-4/IL-13 bioactivity through stabilization of IL-4Rα surface expression and promotion of STAT6 phosphorylation, facilitating T cell polarization⁵⁹, as well as by directly activating eosinophils and promoting their recruitment to infection sites for parasite clearance. In humans, prior research has demonstrated that the increases in salivary antimicrobial peptides (such as Lac and HBD-2) following exercise are associated with changes in salivary sIgA⁶⁰, possibly compensating for a transient decline in sIgA secretion⁶¹. In the present study, we observed a significant positive correlation between type 2 cytokines (IL-4 and IL-5) and MSPs abundance, along with a marked negative correlation between IgE and MSPs levels across all gastrointestinal segments. These findings suggest that bacterially derived MSPs may participate in the precise regulation of type 2 cytokine secretion from Th2 cells, eosinophils, and mast cells, as well as in immunoglobulin production by plasma cells, either through direct interactions or indirect mechanisms. However, it is critical to acknowledge that these conclusions are based on correlative analyses, which inherently preclude definitive causal inferences. While our data implicate MSPs as potential regulators of type 2 immune responses, an alternative mechanistic scenario-where type 2 cytokines (e.g., IL-4 and IL-13) modulate the epithelial expression or release of antimicrobial peptides^62,63, could conversely foster a microenvironment that selectively promotes the growth of MSP-producing bacteria. This reverse causality may underlie the observed co-occurrence of MSP-abundant taxa with elevated type 2 cytokine levels in our dataset. Future investigations are warranted to disentangle these bidirectional relationships.

In this study, we conducted further evaluation of the antimicrobial potential of 17 candidate MSPs by testing their efficacy against pathogenic bacterial strains. Our results revealed that only two MSPs demonstrated growth-inhibitory effects against S. agalactiae and methicillin-resistant S. aureus. S. agalactiae, a beta-hemolytic Gram-positive streptococcus, is a significant pathogen in the Western world, capable of causing neonatal septicemia that may result in mortality or long-term sequelae^64–66. Methicillin-resistant S. aureus, as the most prevalent Gram-positive bacterium, is responsible for various infections ranging from superficial skin lesions to life-threatening bloodstream infections^67,68. Notably, MSP_121684_1 exhibited inhibitory activity against both strains, while MSP_375782_4 specifically suppressed S. agalactiae growth. The encoding organisms of these MSPs were identified as Streptococcus penaeicida (located in the cecum) for MSP_121684_1 and Selenomonadaceae (residing in the rumen) for MSP_375782_4. These findings suggest that the Bacillota in the cecum and Bacillota_C in the rumen represent important reservoirs of bioactive antimicrobial peptides. To validate these observations, in-situ isolation of these bacterial species should be performed to confirm their antibacterial efficacy both in vitro and in vivo.

The observed antimicrobial activities of candidate MSPs likely primarily reflect their ecological role in interbacterial competition or microbiome modulation rather than direct anti-parasite action. Nevertheless, accumulating evidence demonstrates that AMPs can contribute to parasite control through both direct and indirect mechanisms. Direct effects include membrane disruption (as seen with OH_KR34 inducing cytoskeletal damage, apoptosis-like processes, mitochondrial dysfunction, and DNA fragmentation in Acanthamoeba⁶⁹), while other AMPs like tyrocidine A affect multiple parasite survival pathways⁷⁰, cysteine-rich mucosa-derived AMPs alter chitin exoskeleton development in copepod ectoparasites⁷¹, and RCB-4 impairs Toxoplasma cruzi proliferation with ultrastructural alterations including Golgi disorganization and mitochondrial swelling⁹³. Indirect mechanisms involve immune modulation, such as cathelicidin’s TLR2/VDR-dependent enhancement of anti-leishmanial macrophage activity⁷². Future studies should explore whether MSPs exhibit activity against parasite-associated microbes.

Beyond antimicrobial activity, MSPs, functioning as AMPs, play critical ecological roles in the gut microbiome. They mediate microbial competition by displacing rival strains to secure niches⁷³, as demonstrated by pathogens like Shigella spp.⁷⁴, Staphylococcus spp.⁷⁵, Vibrio cholerae⁷³, and Listeria spp.^76,77, which produce AMPs to eliminate competitors (including conspecifics) and dominate ecological space. Concurrently, MSPs shape microbial communities by influencing bacterial composition and spatial organization through host-derived AMPs that selectively target taxa⁷⁸. In symbiotic systems, such as insect-protist associations with vertically transmitted endosymbionts, AMPs enhance metabolic integration by facilitating metabolite exchange across symbiont membranes⁷⁹. These dual roles in competition and symbiosis highlight MSPs’ ecological significance, though their impact on microbial resilience under stress (e.g., infection, dietary shifts) requires further study. Notably, such microbial regulatory functions may extend to host metabolism, as demonstrated by gut microbiota-derived metabolites (e.g., L-theanine⁸⁰) that modulate branched-chain amino acid catabolism, implying a potential interplay between MSPs and host metabolic pathways.

To conclude, more than one thousand MSPs were successfully identified through analysis of the goat GIT microbial genome catalog. The microbial communities responsible for producing these peptides in the rumen, abomasum, and caecum were influenced by the developmental age of the goats. Several MSPs showed correlation with type 2 immunity, even though their host bacterial populations remained unchanged following nematode challenge. Two MSPs derived from the rumen and caecum exhibited efficacy against methicillin-resistant S. aureus and S. agalactiae. These findings highlight that the goat GIT microbiome is a distinctive and promising source for discovering antimicrobial and immunoregulatory peptides. Testing these MSPs against a broader range of gut pathogens or clinically relevant drug-resistant strains would strengthen the translational potential of our findings. Since in vitro assays cannot fully replicate the complex gastrointestinal environment, future studies employing more physiologically relevant models, such as animal infective experiments or ex vivo intestinal segment systems, will be important to further validate and extend the clinical relevance of these results.

Methods

All animal care and procedures were approved by the Animal Care and Use Committee of the Institute of Subtropical Agriculture in compliance with the Regulations for the Administration of Affairs Concerning Experimental Animals (No.ISA-2019-0116). The experiment was conducted at the Institute of Subtropical Agriculture, Chinese Academy of Sciences, Changsha, China.

Experimental design and sample collection

For the de novo generation of metagenomic sequencing data, 99 luminal digesta samples were collected from four gastrointestinal (GI) regions (rumen, abomasum, colon, and caecum) of Xiangdong black goats. Animals were slaughtered 3 h later, following the morning feeding. GI regions were tied off separately by using atraumatic hemostatic forceps. Then luminal contents from each region were transferred to a sterile container for homogenization and collected via DNase- and RNase-free tubes. All collected fresh samples were frozen in liquid nitrogen and transported to the laboratory in a dry-ice pack, where they were immediately stored at −80 °C before total DNA was extracted.

DNA extraction and metagenomic sequencing

DNA was extracted from each sample (100–200 mg) using MagPure Stool DNA KF Kit B (MAGEN, Guangzhou, China) according to the manual instructions. The DNA integrity was controlled by electrophoresis on 1.0% agarose gels, and the concentration and quality were determined by Gemini XPS microplate reader (Molecular Devices, San Francisco, USA). Samples with high-quality DNA of concentration more than 12.5 ng/μL and total DNA amount more than 1 μg were used for further library construction. A metagenomic library with an average insert size of 300 bp was constructed using the MGIEasy Universal DNA Library Prep Set (MGI, Shenzhen, China) following the manufacturer’s instructions and then sequenced on a DNBSEQ-T7 platform (BGI, Shenzhen, China) with a read length of 150 bp.

Metagenomic assembly, binning, and taxonomy

Adapters and low-quality reads from the raw data were first trimmed using Trimmomatic⁸¹ (v.0.33). To decrease potential host DNA contamination, we mapped the sequence data to the goat (Capra hircus, GCF_001704415.2_ARS1.2) genome using BWA-MEM (v.0.7.17). Then, the clean reads after host genome removal were used for constructing the goat GI microbial gene catalog. We assembled the high-quality reads from each sample using MEGAHIT⁸² (v.1.1.1) to generate contigs. Prodigal⁸³ (v.2.6.3) was used to predict open reading frames (ORFs). All the predicted gene sequences of each sample were clustered using the CD-HIT⁸⁴ software with parameters of 95% identity and 90% coverage. The longest gene in each cluster was taken as the representative sequence to construct a non-redundant gene set. Contigs longer than 2.5 kb were selected for single-sample binning into metagenome-assembled genomes (MAGs) using metaBAT2⁸⁵ with default parameters. The completeness and contamination of each MAG were then estimated with CheckM2⁸⁶. All genomes were annotated using GTDB-Tk⁸⁷ (v.2.4.1) based on the Genome Taxonomy Database (http://gtdb.ecogenomic.org/, version r226), and the proportion of taxonomic annotations at the domain level was created with the plotting functions in the R-ggplot2 package. CoverM (v.0.6.1; https://github.com/wwood/CoverM) was used to calculate the MAG abundance⁸⁸. Biosynthetic gene clusters (BGC) within MAGs were identified using antiSMASH (v.5). A maximum-likelihood phylogenetic tree was constructed using PhylophlAn3 with 3975 MAGs and visualized and annotated using the interactive Tree of Life⁸⁹ (iTOL, v.7.2.1, https://itol.embl.de/) online platform. Reads were taxonomically annotated using Kraken2 with the core NCBI nucleotide database to generate sample-level species abundance tables. Based on the resulting species abundance matrix, alpha diversity indices were calculated to assess within-sample microbial diversity. All alpha diversity indices were computed by the Kruskal–Wallis test using the R-vegan package. Boxplots were subsequently generated to visualize the distribution and differences of alpha diversity indices across groups. The final scatter plot with boxplot overlays of the Shannon index was visualized by integrating the tidyverse and boxplot packages. Distribution of MAGs in each GIT compartment and beta-diversity (Bray-Curtis distance) were illustrated by the R-ggplot2 package.

Bioinformatic identification of candidate MSPs and MSP-excreting microorganisms

We used metaprodigal⁹⁰ to predict short peptides from goat microbial genomes, and only genes that encoded peptides between 5 and 50 amino acids were selected. Then AMPir³⁴, AmPEP³⁵ and APIN³⁶ were combined to predict MSPs. Amino acid sequences that meet at least two criteria, including being successfully identified by more than two methods, and each MSP should pass the lowest score (AMPir score greater than 0.8, AmPEP score ≥ 0.5, and APIN = 1), were referred to as MSP candidates^9,10. The abundance of candidate MSPs across the GIT samples was calculated by the ARGs-OAP pipeline (v3.2.4). Considering the number of amino acids for each MSP, the hit cutoffs for step two within the ARGs-OAP pipeline were ‘--id 60 --qcov 50 --length 10’. RPKM-based abundance (Reads Per Kilobase per Million mapped reads) was used as the normalized copies for MSPs in the further analysis. MSP-secreting MAGs at the phylum level in different GIT regions were constructed using the R-ggalluvial package. Spearman’s rank correlation analysis was used to test the correlation between MSP abundance and MAG abundance. The p-value was calculated using the cor. test function in the R software. The network of bacterial species and MSP was illustrated by Cytoscape⁹¹ (https://cytoscape.org/).

Variation of candidate MSPs and MSP-encoding bacteria in the GIT of goat kids of different ages

Twenty-eight neonatal goats were divided into three developmental groups: 14 days (D14, breast milk suckling period, n = 10), 42 days (D42, milk and starter mix feeding period, n = 10), and 105 days (D105, solid diet feeding period, n = 8). Fifty DNA samples from rumen, abomasum, and caecum digesta were extracted and subjected to metagenomic sequencing. The PCA analysis of cMSP-excreting bacteria was plotted directly using ggplot2. For alpha-diversity, data transformation was first performed with the reshape2 package, followed by Kruskal-Wallis testing using the R vegan package, and the scatter plot with overlayed boxplot was then visualized using functions from tidyverse and boxplot. Composition of cMSP-excreting bacteria at the phylum and species level was created using the ggalluvial package to draw Sankey diagrams, and statistical significance was assessed with the Kruskal-Wallis test in the vegan package. cMSP abundance in different GIT and ages involved multiple mean comparisons among groups, which were performed with the multcomp package, and then the significance results were visualized using multcompView. Finally, the corresponding heatmap was generated with the tidyverse package.

Changes of candidate MSPs and MSP-encoding microorganism in the GIT of growing goats infected with nematode

Seventeen growing and female goats with similar body weight were randomly divided into a control group (CK, n = 8) and an infected group (Infected group, n = 9). Goats in the Infected group were infected with 10,000 third-stage Haemonchus contortus larvae. The administration of nematodes and management of these experimental goats were the same as in our previous study⁹². Phenotypic indicators, including dry matter intake, IL-4, IL-5, IL-10, IgA, IgG, and IgE in the blood, were cited from our previous research^48,92. Forty-nine DNA samples from ruminal, abomasal, caecal, and colonic digesta were extracted and then subjected to metagenomic sequencing. The analyses of PCA and alpha-diversity were the same as those of the developmental trial. The Mann–Whitney U test was used to compare the composition of cMSP-encoding bacteria at the phylum and species levels. Finally, correlation data between phenotypic indicators and cMSP were restructured using reshape2, correlation analysis was performed using vegan, and visualized with ggcor, while the color scheme was designed with R-RColorBrewer, and the correlation picture in different GIT were combined into one image using the R-patchwork package.

Laboratory evaluation of antimicrobial activity of candidate MSPs

Among the 57 MSPs correlated with bacterial abundance, only 13 initially met the following three selection criteria: (1) a significant negative correlation (R < −0.3, P < 0.05) with a specific bacterial community; (2) being encoded by a high-quality metagenome-assembled genome (MAG); and (3) a grand average of hydropathicity (GRAVY) below zero. These 13 candidate MSPs were subsequently synthesized and tested for antimicrobial activity. However, no significant antimicrobial activity was detected in minimum inhibitory concentration (MIC) assays. Subsequently, we screened differentially expressed MSPs among the Top50 MSPs from the developmental and nematode infection trials, again requiring a GRAVY score less than zero, and selected four additional candidate MSPs. In total, 17 peptides were ultimately chosen for further functional testing. Each cMSP was synthesized through solid-phase peptide synthesis at Dangang Biotech, Wuhan, China. The purity of the synthesized peptides was assessed through high-performance liquid chromatography (Waters 2695, USA), and their molecular weights were determined by mass spectrometry (Agilent-6125 B, China). The purity of all cMSPs was more than 90%. The mass spectrometry report of each MSP is listed in the Supplementary Figs 2–18. The antimicrobial activities of seventeen synthesized peptides were initially evaluated using five strains of bacteria, including S. agalactiae ATCC 12386, S. aureus ATCC 29213, and methicillin-resistant S. aureus ATCC 43300 as Gram-positive bacteria, and E. coli ATCC 25922 and S. enterica ATCC 13076 as Gram-negative bacteria. Individual colonies of the bacterial strains were inoculated into cation-adjusted Mueller-Hinton broth (CAMHB) medium after being transferred from Luria–Bertani (LB) medium and incubated overnight at 37 °C. Fetal calf serum (FCS) was added to the CAMHB medium for S. agalactiae. After incubation overnight in a 37 °C shaker, the bacterial suspensions in CAMHB were diluted to achieve an optical density (OD) of 0.6 at 620 nm. The stock solution of each peptide was prepared at a concentration of 256 μmol/L by dissolving in double-distilled water. Working solutions with concentrations ranging from 128 to 8 μmol/L were obtained through serial dilution. For each peptide, the testing experiment comprised four groups: (1) a blank control group containing 200 μL of CAMHB solution; (2) a negative control group with 200 μL of bacterial suspension; (3) a test group with 200 μL of peptide working solution; and (4) a positive control group with 200 μL of antibiotic solution. Vancomycin served as the positive control antibiotic for Gram-positive bacterial strains, while chloramphenicol was used as the positive control antibiotic for Gram-negative bacteria. The experiment was conducted in a 96-well plate and incubated overnight at 37 °C in a temperature-controlled incubator. Each experiment was repeated four times independently. The OD620 value was measured using a microplate reader (MULTISKAN FC, Thermoscientific, USA), and cellular sedimentation was manually broken up by mild pipetting. The growth percentage of the cells was calculated by (OD₆₂₀ of peptide − OD₆₂₀ of blank)/(OD₆₂₀ of control − OD₆₂₀ of blank). The minimum inhibitory concentration of each candidate MSP against five tested bacterial strains was determined as the lowest concentration at which no visible bacterial growth was observed. Swiss-model (https://swissmodel.expasy.org/) was used to generate the PDB files of candidate MSPs and PyMOL (v.3.1.6.1) was used to illustrate the 3D pictures. SOPMA was used to predict the protein secondary structures.

Hemolytic effect of the candidate MSPs

Red blood cells of the mouse, at a percentage of two, were dispensed into 96-well plates. The candidate peptide was properly dissolved in PBS and added to the 96-well plates to achieve final concentrations of 32,64,128, 256, and 512 μmol/L. PBS and a 2% TritonX100 solution were also introduced to the cells, serving as the negative and positive controls, respectively. After a 1-h incubation at 37 °C, the plates underwent centrifugation at 1500 rpm for 10 min, and the supernatant was retrieved for OD₄₅₀ determination. Subsequently, the hemolysis percentage was calculated using the formula (OD₄₅₀ of Peptide − OD₄₅₀ of PBS)/(OD₄₅₀ of TritonX100 − OD₄₅₀ of PBS). This experimental procedure was independently repeated three times.

Cytotoxicity of the candidate MSPs

The cytotoxicity of the candidate MSPs against HEK293 cells (CL-0001, Procell Life Science & Technology Co., Ltd, Wuhan, China) was assessed using the Cell Counting Kit-8 (HY-K0301, MedChemExpress, New Jersey, USA). In brief, HEK293 cells were cultured in 96-well plates in DMEM medium. After a 24-h incubation at 37 °C with 5% CO₂, the candidate MSPs solution was added to achieve final concentrations of 32, 64, 128, 256, and 512 μmol/L, followed by a further 24-h incubation at 37 °C. Blank control was set without cell addition. Normal control was incubated in DMEM medium with PBS. Cell viability was assessed by the addition of CCK8 solutions according to the manufacturer’s instructions, and the OD₄₅₀ was measured. Lastly, cytotoxicity was calculated as (OD₄₅₀ of MSP − OD₄₅₀ of blank)/(OD₄₅₀ of normal − OD₄₅₀ of blank). Each experiment was carried out with three replicates.

Membrane permeabilization assay

The membrane permeabilization was assessed by the Sytox green assay. The strains of S. agalactiae and Methicillin-resistant S. aureus were streaked on a Brain Heart Infusion and Tryptic Soy Agar plates, respectively, and then incubated under aerobic conditions at 37 °C overnight until obvious colonies could be observed. The fresh plate cultures were washed with sterile PBS, and the OD₆₀₀ value was adjusted to 0.6 for use as the stock solution. Bacterial cells in PBS (100 μL) were dispensed into wells of 96-well plates along with 5 μM Sytox green (Maokang Biotechnology Co., Ltd., Shanghai, China). Plates were incubated for 15 min in the dark at room temperature, and then 100 μL of candidate MSP was added to achieve the concentrations of 128, 256, and 512 μmol/L. Fluorescence was measured spectrophotometrically (at an excitation wavelength of 480 nm and an emission wavelength of 522 nm) every 5 min up to 30 min, and then after 150 min. Triton X-100 1% (v/v) in PBS was used as a positive control to disrupt the cytoplasmic membrane of bacteria. Each experiment was carried out with three replicates.

Time-dependent killing assay

Methicillin-resistant S. aureus and S. agalactiae were streaked on a Tryptic Soy Agar and Brain Heart Infusion agar plates, respectively, and then incubated under aerobic conditions at 37 °C overnight until obvious colonies could be observed. The bacterial suspension was then collected and adjusted to an OD₆₀₀ value of 0.6 and a concentration of 10⁶ CFU/mL using PBS. Peptide powders were dissolved in double-distilled water before being added to the bacterial suspension to achieve final peptide concentrations of 1× MIC, 2× MIC, and 4× MIC. Subsequent incubation took place in a shaker at 37 °C. Samples were collected at 0, 0.5-, 1-, 2-, 3-, and 5-h time points, where 100 μL aliquots were taken, diluted tenfold, and plated onto Mueller-Hinton agar plates for overnight incubation at 37 °C. Finally, the CFU per mL was calculated. Each experiment was carried out with three replicates.

Statistical analysis

For comparative analysis of microbial α-diversity (Shannon index) across different GIT compartments, the non-parametric Kruskal–Wallis test was employed to account for potential non-normal distribution of diversity metrics. To assess differences in β-diversity (community composition), Permutational Multivariate Analysis of Variance (PERMANOVA) was performed.

In the developmental trial, one-way ANOVA followed by the least significant difference (LSD) test was used to compare average body weights across goat age groups. For microbial data within the same GIT segment (diversity indices, phylum- and species-level composition, and Top50 MSP abundance), the non-parametric Kruskal-Wallis test was applied.

In the nematode infection experiment, Mann–Whitney U tests were used to compare microbial α-diversity, phylum/species-level composition, and Top50 MSP abundance between control and infected groups. To examine relationships between MSP abundance and phenotypic parameters, Spearman’s rank correlation analysis was employed, followed by calculation of correlation coefficients (r) and P-values using R’s cor. test function.

For in vitro experiments evaluating MSP effects (growth inhibition, survival rates, and hemolytic activity across concentrations), one-way ANOVA with LSD tests was conducted to compare means among treatment groups. All statistical analyses declared significance at P < 0.05.

Author contributions

Qiongxian Yan: Writing-original draft, Writing-review & editing, Methodology, Validation. Zhixiong He: Writing-review & editing, Investigation, Project administration. Wenxun Chen: Sample collection, Investigation, Visualization. Xiaoling Zhou: Software, Data curation. Jinhe Kang: Resources, Investigation. Zhiliang Tan: Conceptualization.

Data availability

Metagenomic sequencing data have been deposited in the China National Gene Bank Database with accession number CNP0007333. A total of 3975 MAGs and 1147 MSPs sequences have been deposited in the figshare website with the link (https://doi.org/10.6084/m9.figshare.31701583).

Code availability

Parts of the code are available at https://github.com/zd200572/GGM. Other unpublished codes may be made available to qualified researchers at the corresponding author’s reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41522-026-01051-2.