Phage-mediated expansion of the virulence gene types and enhanced ecological integration of pathogens in wild mice from human-impacted environments

Abstract

Wild mice are crucial in the transmission of infectious diseases; however, quantitative indicators for evaluating risk of virulence factors transmission are still lacking. We combined metagenomics and network analysis to evaluate ecological connectivity and functional gene profiles of microbial communities in wild mice from human-impacted environments (HE) and woodland environments (WE). We found that the pathogen centrality was significantly higher in HE than in WE (p < 0.001). Random Forest Model suggested habitats, phage abundances, and antibiotic resistance genes (ARGs) counts were crucial factors influencing virulence factor genes (VFGs) counts (p < 0.05). Structural Equation Model revealed that habitats affected VFGs (p < 0.01) via phages mediation (p < 0.05), while ARGs directly affected VFGs (p < 0.001). Although VFG counts were significantly higher in HE (p < 0.001), their expression levels did not differ between two habitats (p = 0.2952), indicating that VFG diversity was not necessarily accompanied by higher virulence expression. This study highlights the mediating role of phages and the direct contribution of ARGs in shaping the virulence-associated genetic repertoire, underscoring the importance of a One Health perspective that considers human impacts on microbial communities in infectious disease surveillance.

Introduction

Wild mice are significant in the transmission of infectious diseases, such as plague, hemorrhagic fever with renal syndrome¹, and leptospirosis², posing critical threats to public health. Their habitation in diverse environments and close contact with human settlements shape their body’s microbiomes differently^3–5, causing them to harbor various pathogens⁶. Metagenome sequencing and microbiome analysis have emerged as powerful tools for predicting the functional potential and ecological traits of microbial communities in zoonotic hosts^7,8. Wild mice thus represent a critical node in the One Health continuum, where the health of humans, animals, and ecosystems converge. Human activities that alter natural habitats can reshape the microbial communities carried by these animals, potentially affecting the reservoir and transmission dynamics of virulence and resistance genes. Understanding how environmental disturbance translates into shifts in the ecological integration of pathogens and the mobilization of virulence traits is therefore essential for predicting disease risks at the human-animal-environment interface.

Examining the complex community structures of microorganisms could reveal patterns indicative of interaction complexity and ecological adaptation. However, conventional metagenomic methods offer limited insight into higher-order microbial interactions. Network analysis is a critical tool for deciphering the structure and function of microbial communities⁹. In a microbial network, microorganisms are presented as nodes, and potential pairwise associations or interactions between microbes are presented as edges¹⁰. Thus, microbial co-occurrence networks could be established to reveal potential microbial interactions and unique co-occurrence patterns. The complexity of a microbial community is affected by interactions among microbes¹¹. Microbial communities from different environments exhibit environment-specific co-occurrence patterns¹². Microbial network analysis also enables the identification of keystone taxa, including non-pathogens and pathogens, based on centrality metrics such as degree centrality (DC), which quantifies the connectedness of each node within the network.

Virulence factors are substances secreted by pathogens that are related to pathogenicity and include two categories, namely invasiveness and toxins¹³. These factors enable pathogenic microorganisms to cause disease in the host¹⁴. The conditions of microbial communities promote specific pathogen colonization and minimize virulence loss, favoring disease development¹⁵. The co-occurrence patterns of virulence factor genes (VFGs) and antibiotic resistance genes (ARGs) have been extensively documented in recent metagenomic studies, for example, in aquatic sediments¹⁶, agricultural soils¹⁷, and coastal sediments¹⁸. Understanding the relationship between VFGs and ARGs is crucial for predicting and controlling the spread of infectious diseases. To our knowledge, in microbial communities, the environment or habitat is crucial in shaping VFG and ARG classification and expression¹⁷. In human-impacted environments (HE), phages act as vectors for ARG transfer¹⁹, complicating efforts to control their spread. In addition, phages drive the expansion of ARG-related bacteria in wild rodent populations²⁰. Phages influence VFG’s transmission, affecting the pathogenic potential of the bacterial hosts²¹. However, the complex relationships among VFGs, ARGs, habitats, and phages of microbial communities in the wild mice microbiome remain rarely reported.

In this study, we collected wild mice from HE and woodland environments (WE). Metagenomic approaches were applied to characterize the compositions of microbes and genes in communities. By integrating metagenomic data and causal inference models, we investigated the ecological integration of pathogens and the risk of horizontal gene transfer of virulence and resistance across habitats. We further examined the complex relationships among VFGs, ARGs, phages, and habitats. Grounded in the One Health framework, this study aims to develop a quantifiable network metric for pathogen ecological integration and disentangle the direct and indirect relationships among habitats, phages, ARGs, and VFGs in wild mouse microbiomes.

Results

Microbial compositions and diversity in different habitats

Shotgun metagenomic sequencing was performed to determine the microbial compositions of the eukaryal-bacterial and viral communities. Eukaryotic-bacterial communities were analyzed based on DNA-derived taxonomic profiles, while viral communities were characterized using RNA-derived taxonomic profiles.In the eukaryal-bacterial communities, Mannheimia, Colletotrichum, Fusarium, Pyricularia, Botrytis, Kluyveromyces, Candida, Saccharomyces, Thermothelomyces, and Pasteurella were the top 10 abundant genera (Fig. 1A). Pasteurella was the predominant genus in the WE, with an average relative abundance of 22.50% (95% confidence interval (CI): 21.58–23.42%); however, in the HE, its relative abundance was 0.85% (95% CI: 0.74–0.96%) (p < 0.0001) (Fig. 1C). Colletotrichum was abundant in the HE, with a relative abundance of 13.69% (95% CI: 13.29–14.09%); however, it had a relative abundance of 9.13% (95% CI: 8.91–9.36%) in the WE, indicating significantly lower abundance than that in the HE (p < 0.0001) (Fig. 1C). The Venn plot showed that 884 genera (80.00%) existed in both environments in the eukaryal-bacterial communities (Fig. 1B). There were 135 genera (12.22%) only in the HE and 86 unique genera (7.78%) in the WE. In the viral communities, Cyprinivirus, Gammaretrovirus, Rtpvirus, Tobamovirus, Macavirus, Rhadinovirus, Pestivirus, Retroviridae, Phietavirus, and Siphoviridae were the top 10 genera with the highest relative abundances (Fig. 1D). Cyprinivirus was predominant in the HE (64.80%, 95% CI: 57.43–72.17%) and WE (55.76%, 95% CI: 51.78–59.75%), and its relative abundance in the HE was significantly higher than it in the WE (p < 0.01) (Fig. 1F). Gammaretrovirus was the second most abundant genus in the WE, with a relative abundance of 43.18% (95% CI: 39.27–47.09%); however, it exhibited a lower relative abundance of 28.41% (95% CI: 21.83–34.99%) in the HE (p < 0.0001) (Fig. 1F). A Venn plot showed that 23 genera (53.49%) existed in both environments (Fig. 1E). There were 10 genera each in the HE and WE.

Fig. 1. Taxa composition of different communities in different environments.

The top 10 taxa of eukaryal-bacterial (A) and viral (D) communities in different habitats at the genus level. The Venn plot shows the number of genera unique or shared by different habitats in the eukaryal-bacterial (B) and viral (E) communities. C The relative abundances of Pasteurella and Colletotrichum in different habitats. F The relative abundances of Cyprinivirus and Gammaretrovirus in different habitats.

The α diversity parameters, including the Shannon, Simpson, and evenness indices, were used to measure the microbial diversity and evenness in the communities. In eukaryal-bacterial communities, the Shannon, Simpson, and evenness indices were higher in the HE than in the WE (all p < 0.01) (Fig. 2A). The HE exhibited higher values for the Shannon index (3.37, 95% CI: 3.29–3.44) (p < 0.01), indicating a more even distribution of genera than those in the WE (3.22, 95% CI: 3.18–3.27). The Simpson index was also higher in the HE (0.93, 95% CI: 0.92–0.93) than in the WE (0.91, 95% CI: 0.90–0.92) (p < 0.0001), suggesting a more equitable genus distribution, with no single genus dominating the community in the HE. Furthermore, the evenness index was significantly higher in the HE (0.56, 95% CI: 0.54–0.57) than in the WE (0.49, 95% CI: 0.48–0.50) (p < 0.0001), underscoring the uniformity in genus abundance within the HE. These results suggest that the HE harbored more diverse and evenly distributed communities of eukaryotic microorganisms and bacteria than did the WE. Pairwise Wilcoxon rank-sum tests indicated that habitat type was significantly associated with differences in α diversities (all p < 0.01) (Fig. 2A). In the virial communities, the Shannon and evenness indices showed no significant differences between the HE and WE (all p > 0.05) (Fig. 2B). However, the Simpson index showed lower values in the HE (0.35, 95% CI: 0.31–0.40) than in the WE (0.45, 94% CI: 0.43–0.47) (p < 0.05). This result suggests that Cyprinivirus was the dominant genus in the HE (Fig. 1D, F). The pairwise Wilcoxon rank-sum tests results showed that different habitats were related with different Simpson indices (p < 0.01) (Fig. 2B). Non-metric multidimensional scaling (NMDS) based on Bray–Curtis distance was performed to measure β-diversity differences in all samples. NMDS based on Bray-Curtis distance was performed to visualize β diversity among samples. NMDS ordination revealed a clear separation between HE and WE eukaryal-bacterial communities (Fig. 2C). PERMANOVA confirmed that the composition differed significantly between the two habitats (R² = 0.70, p = 0.001). A beta-dispersion test indicated heterogeneous within-group variances (p = 0.001), suggesting that community dispersion also differed between habitats. Nevertheless, the large effect size (R² = 0.70) and the clear visual separation in the NMDS ordination together support that the compositional difference between HE and WE is substantial and not solely attributable to dispersion effects. For viral communities, the compositional difference, while statistically significant (PERMANOVA R² = 0.06, p = 0.002), was not visually distinctive (Fig. 2D); beta-dispersion was also heterogeneous (p = 0.001).

Fig. 2. α and β diversities of microbial communities in different habitats.

The box plots of α diversity indices in eukaryal-bacterial (A) and viral (B) communities. The results of non-metric multidimensional scaling in eukaryal-bacterial (C) and viral (D) communities.

Biomarker taxa and pathogenic genera in different habitats

The LEfSe showed that in the eukaryal-bacterial communities, Colletotrichum, Mannheimia, Pyricularia, Botrytis, and Fusatium were significantly enriched at the genus level in the HE (Fig. 3A), whereas Pasteurella was primarily enriched in the WE. Ascomycota was the primary phylum in the HE, whereas Pseudomonadota was the enriched phylum in the WE. In the viral communities, Rtpvirus and Pestivirus were significantly enriched in the HE (Fig. 3B), and Gammaretrovirus was the dominant genus in the WE. At the phylum level, Uroviricota and Kitrinoviricota were enriched in the HE, and Artverviricota was significantly enriched in the WE. Uroviricota is a phylum of bacteriophages that infects bacteria.

Fig. 3. Biomarker taxa and pathogen compositions in different habitats.

The linear discriminant analysis effect size showing the biomarker taxa of eukaryal-bacterial (A) and viral (B) communities in different habitats. In eukaryal-bacterial communities, taxa with the top 30 linear discriminant analysis scores are shown. The pathogen compositions at the kingdom level (C). The UpSet plot of the common and unique pathogens between the human-impacted environment (HE) and woodland environment (WE) (D). The top 30 pathogens with high relative abundance between the HE and WE (E).

To determine the composition and abundance of human-transmitted pathogenic microorganisms in the microbiomes of mice from different habitats, we used a list of human-transmitted pathogenic microorganisms published by the National Health and Wellness Commission of China. We identified 78 pathogen genera in all samples (Fig. 3D), including bacteria (82.05%), viruses (10.26%), and eukaryotes (7.69%) (Fig. 3C). There were 75 pathogen genera in both habitats. There were three genera distributed only in the WE (Fig. 3D), namely Dermatophilus, Mastadenovirus, and Orthoflavivirus. Among these pathogens, Gammaretrovirus, Fusarium, and Candida were genera with high relative abundances in both habitats, and Pasteurella was the pathogen with a high relative abundance in the WE (Fig. 3E). These four genera were the primary taxa of eukaryal-bacterial and viral communities in the different habitats (Fig. 1A, D).

Co-occurrence network analysis of eukaryal-bacterial and viral communities

To better understand the potential interactions and co-occurrence patterns of eukaryal-bacterial and viral communities, we performed a co-occurrence network analysis for all samples (Fig. 4). For eukaryal-bacterial communities, we selected genera with top 100 relative abundances to construct co-occurrence networks; other genera were classified into “Others” (Fig. 4A, B). There were 101 nodes in each network (Table 1). The numbers of edges in the HE and WE eukaryal-bacterial networks was 941 and 526, respectively. The number of edges in the HE eukaryal-bacterial network was higher than that in the WE eukaryal-bacterial network (1.79-fold). The average path lengths of these networks were 1.90 (HE) and 2.23 (WE) (Table 1). The shorter average path length of the HE eukaryal-bacterial network indicates that the relationships in this network were tightly correlated, implying more frequent interactions and efficient communication (Fig. 4A). In contrast, the lower values of edges and average path length in the WE eukaryal-bacterial network indicate a sparse but complex network structure (Fig. 4B). The degree centralities of nodes in the HE eukaryal-bacterial network were significantly higher than those in the WE eukaryal-bacterial network (p < 0.001) (Fig. 4C). The pathogen centrality (PAC) of the nodes were higher in the HE eukaryal-bacterial network than in the WE eukaryal-bacterial network (p < 0.001) (Fig. 4D), indicating that pathogens were more deeply ecologically integrated and formed more numerous connections within the HE microbial community. Nodes with a high DC may be crucial in communication within a network. Nodes with the highest degree centrality in the eukaryal-bacterial networks were Ustilago (DC = 0.31 in the HE eukaryal-bacterial network) and Mannheimia (DC = 0.20 in the WE eukaryal-bacterial network). Nodes with the highest pathogen centrality were Ustilago, Riemerella, and Bartonella (pathogen centrality = 0.12 for each in the HE eukaryal-bacterial network), Micromonospora and Sugiyamaella (pathogen centrality = 0.08, for each in the WE eukaryal-bacterial network). The clustering coefficient measures the tightness of interactions within microbial communities. The values of the average clustering coefficient in HE and WE eukaryal-bacterial networks were 0.23 and 0.14, respectively (Table 1); this index was higher in the HE eukaryal-bacterial network than in the WE eukaryal-bacterial network (p < 0.0001). The closeness centrality index was used to measure the average distance from one microbe to the other microbes in the network. The closeness centrality of the HE eukaryal-bacterial network differed from that of the WE eukaryal-bacterial network (p < 0.0001). The betweenness centrality measures the frequency with which a microorganism acts as an intermediary node in a network. Microorganisms with a high betweenness centrality may serve as bridges between various microbial interactions. The average betweenness centrality of the WE eukaryal-bacterial network was higher than that of the HE eukaryal-bacterial network (p < 0.01).

Fig. 4. Co-occurrence networks and main indices of networks in different habitats.

The co-occurrence networks of the eukaryal-bacterial communities (A, B) and bacterial-viral community (E, F). The node colors indicate different microbe types, and the node sizes represent the degree of microbes. Edges between nodes represent the relationships between them. The Violin plots showing the degree centrality of eukaryal-bacterial (C) and viral (G) networks, the pathogen centrality of eukaryal-bacterial networks (D), and the phage centrality of viral networks (H).

Table 1.

Properties of microbial networks for eukaryal-bacterial and viral communities in the human-impacted environment (HE) and woodland environment (WE)

Properties	Eukaryal-bacterial community		Viral community
	HE	WE	HE	WE
Nodes	101	101	33	33
Edges	941	526	44	49
Degree Centrality	0.19 ± 0.06	0.1 ± 0.03	0.12 ± 0.06	0.11 ± 0.07
Pathogen/Phage Centrality	0.06 ± 0.03	0.03 ± 0.02	0.04 ± 0.03	0.02 ± 0.02
Clustering Coefficient	0.22 ± 0.05	0.14 ± 0.07	0.14 ± 0.3	0.09 ± 0.14
Closeness Centrality	0.53 ± 0.04	0.45 ± 0.03	0.34 ± 0.06	0.36 ± 0.05
Betweenness Centrality	44.79 ± 26.76	61.67 ± 38.53	23.42 ± 29.28	25.97 ± 34.04

For viral communities, we used all identified genera to construct co-occurrence networks (Fig. 4E and F). Each network contained 33 nodes (Table 1). The numbers of edges in the HE and WE viral networks was 44 and 49, respectively. The average path lengths of these networks were 3.04 (HE) and 2.84 (WE) (Table 1). The longer average path length of the HE viral network indicates that the relationships in this network were sparsely correlated, implying fewer interactions and inefficient communication. There were no significant differences in the DC, clustering coefficient, closeness centrality, or betweenness centrality indices between the HE and WE viral networks (all p > 0.05) (Fig. 4G and Table 1). However, the phage centralities of the HE viral network were significantly higher than those of the WE viral network (p < 0.05) (Fig. 4H). This finding implies that phages may be crucial in the microbial communities in HE habitats. The nodes with the highest phage centrality were Rtpvirus (phage centrality = 0.11 in the HE viral network), Rtpvirus, Rhadinovirus, and Pestivirus (pathogen centrality = 0.07 for each in the WE viral network).

Phages’ mediation effect expanded types of the virulence factors in the HE

To reveal the relationships among habitats, phages, ARGs, and VFGs of microbial communities, we annotated VFGs and ARGs using Abricate (v1.0.1). The VFG and ARG counts in the HE were higher than those in the WE (p₁ < 0.001, p₂ < 0.01, respectively) (Fig. 6A, C). Linear regression analysis revealed a strong positive relationship between ARG counts and VFG counts (R² = 0.80, p < 0.001) (Fig. 5A). Subsequently, we investigated the crucial factors influencing the VFG counts using an RFM. The RFM suggested that the ARG counts, different habitats, and the abundance of phages were significant factors contributing to VFG counts (R² = 0.56, p < 0.01), while the abundance of pathogens and non-pathogens were not identified as significant factors in the model (Fig. 5B).

Fig. 6. Comparison of ARGs and VFGs between HE and WE.

Comparison of ARGs counts (A) and abundance (B) between HE and WE. Comparison of VFGs counts (D) and abundance (E) between HE and WE. The heatmaps of ARGs (C) and VFGs (F) show the expression levels and differences between different habitats.

Fig. 5. Relationships among VFGs, ARGs, habitats, and phages.

The correlation between VFGs and ARG counts (A). The Random Forest Model showed the key factors that increased VFGs counts (B). The structural equation model showed relationships among VFGs, ARGs, habitat, and phage abundance (C). The Mantel test revealed the associations between VFGs and phages (D) and ARGs and phages (E) in different habitats.

We implemented an SEM to determine the relationships among VFGs counts, habitats, phage abundance, and ARGs counts (Fig. 5C and Table 2). The category variables of HE and WE were labeled as “1” and “2,” respectively. Habitat negatively affected phage abundance from the HE to the WE (standardized coefficient = −0.27, p < 0.05); however, there was no significant correlation between habitat and ARGs (standardized coefficient = −0.22, p > 0.05). Habitat was negatively correlated with VFGs (standardized coefficient = −0.37, p < 0.01). ARGs directly affected VFGs through a positive effect (standardized coefficient = 0.85, p < 0.001); however, ARGs did not play a mediator in the “Habitat → ARGs → VFGs” path (standardized coefficient = −0.18, p > 0.05). An increased abundance of phages increased the VFG counts (standardized coefficient = 0.18, p < 0.001). From the HE to the WE, phages played a mediating role through a negative effect (standardized coefficient = −0.05, p < 0.05) (Table 2). These results suggest that ARGs counts and different habitats directly influence VFGs counts, and phages play a mediating role in the “Habitat → Phage → VFGs” path. The Mantel test was implemented to examine the associations between phage abundance and VFGs or ARGs. Twenty highly expressed VFGs and 20 highly expressed ARGs were selected as the gene expression matrix. In the taxon abundance matrix, all taxa were classified into three categories: non-pathogens, pathogens, and phages (Supplementary Fig. 1). The VFGs and ARGs associated with the phages were identified (Fig. 5D, E). For VFGs, gspF, gspJ, gspK, gspM, and entS showed strong correlations with phage abundance in the HE, whereas these genes, except gspF, were not correlated with phage abundance in the WE (Fig. 5D). For ARGs, emrA, emrB, mdtP, pmrF, ampH, and APH(3’)-Ia showed correlation with phage abundance in the HE and WE (Fig. 5E).

Table 2.

Results of the structural equation model for VFGs

Latent relationships	Total effect	Direct effect		Indirect effect
		Standardized coefficient	P value		Standardized coefficient	P value
Habitat → ARGs	−0.22	−0.22	0.051		-	-
Habitat → Phage	−0.27	−0.27	0.015		-	-
Habitat → VFGs	−0.37	−0.14	0.002		−0.23	<0.001
Habitat → ARGs → VFGs	0.66	0.85	<0.001		−0.18	0.053
Habitat → Phage → VFGs	0.13	0.18	<0.001		−0.05	0.040

Phage-mediated enrichment of virulence factor diversity in the HE

The ARG count reflects the richness (the number of different types) of ARGs within the microbial community, whereas the abundance (expression level) of ARGs indicates the community’s actual resistance capacity. Our analysis revealed that both the count and abundance of ARGs were significantly higher in the HE compared to the WE (p₁ < 0.01, p₂ < 0.001) (Fig. 6A, B). Similarly, the VFG count serves as an indicator of the diversity of VFGs, while the expression level of VFGs is a more direct proxy for the community’s pathogenic potential. Intriguingly, we observed a dissociation between these two metrics for virulence: although the VFG count was significantly elevated in the HE (p < 0.001) (Fig. 6D), the overall expression abundance of VFGs showed no statistically significant difference between HE and WE habitats (p = 0.2952) (Fig. 6E). This finding is consistent with the phage-mediated expansion of the VFG types in the HE environment (Fig. 5B, C). The observed discrepancy suggests that while this process has equipped the HE microbial community with a broader spectrum of virulence genes, it nonetheless maintains these genes at a baseline level of expression, thereby not immediately elevating the pathogenic threat.

Notably, when we specifically analyzed the expression of phage-associated VFGs and ARGs, we found that these genes were expressed at significantly higher levels in the HE habitat compared to the WE habitat (all p < 0.001) (Fig. 6C, F). For VFGs, gspF, gspJ, gspK, gspM, and entS were positively correlated in both habitats (all p < 0.001) (Fig. 6C). For ARGs, emrA, emrB, mdtP, pmrF, and ampH were positively correlated in both habitats (all p < 0.001) (Fig. 6F). According to the RFM and SEM (Fig. 5 and Table 2), ARGs were the direct and primary contributors to the increase in the VFGs of the microbial communities. We further applied a Linear Regression Model (LM) to assess the relationship between phage-related ARGs and VFGs (Supplementary Fig. 2). The model revealed that in the HE habitat, the coefficient of determination (R²) between ARGs (emrA, emrB, mdtP, pmrF, and ampH) and VFG (gspM) was over 0.9; however, these R² values were below 0.8 in the WE (Supplementary Fig. 2A–E(3)). In the HE, the R² exceeded 0.9, suggesting a strong relationship where the ARGs explained a substantial proportion of the variance in gspM. Conversely, in the WE, the R² dropped below 0.8, indicating a weaker relationship between ARGs and gspM than that in the HE. To explore the potential gene arrangement patterns of phage-related VFGs and ARGs, we recovered high-quality metagenome-assembled genomes (MAGs) and determined the phage-related VFGs and ARGs. In total, 48 high-quality MAGs were recovered, all of which were identified as Escherichia coli. Gene positional analysis indicated that phage-related VFGs and ARGs were distributed more diffusely in the HE habitat, whereas in the WE habitat they appeared in tighter clusters (Supplementary Fig. 3). This number of MAGs is consistent with the sequencing depth achieved (averaging approximately 208 thousand assigned reads per DNA sample) and is sufficient for the genomic comparison of phage-associated VFG and ARG distributions. Notably, VFGs and ARGs analyses were conducted on the full non-redundant gene catalog, ensuring comprehensive detection independent of MAG completeness.

Discussion

There were two genera with high abundances in the HE, Colletotrichum (Fig. 1C) and Cyprinivirus (Fig. 1F). Colletotrichum is a phytopathogenic fungus with diverse species and broad host range^22,23. Fungi belonging to the Colletotrichum genus may also affect mouse health. A study indicated that Colletotrichum gloeosporioides promotes the accumulation of phthalic acid esters in mice, compromising their kidney function²⁴. Impaired kidney function and other adverse health statuses affect the composition of microbiota in wild mice, influencing their disease transmission capabilities^25,26. Several species of the genus Cyprinivirus are associated with infections in the common carp, goldfish, and freshwater eel²⁷. However, the relationship between Cyprinivirus and wild mice remains unknown. Differences in the abundances of Colletotrichum and Cyprinivirus in wild mice from different habitats may be owing to variations in their food sources and the impact of human activities.

The α diversity results indicated that the genus richness of eukaryal-bacterial communities in the HE was higher than that in the WE (Fig. 2A). However, viral community results showed no significant differences in the viral αdiversities between the two habitats (Fig. 2B). These findings suggest complex interactions within microbial communities in the HE. Moreover, the results of β diversity indicate that the formation of these complex interactions among microbes is related to varying habitats (Fig. 2C, D). For bacterial communities, higher α diversity may indicate more complex and frequent interactions between microbial communities, such as horizontal gene transfer and quorum sensing, increasing virulence^28,29.

Microbial network analysis revealed a higher pathogenic centrality in the HE microbial community than in the WE microbial community (Fig. 4A, B, D). At the gene level, the counts of VFGs and ARGs were higher in the HE than in the WE (Fig. 6A, D). The abundance of ARGs was also significantly elevated in the HE habitat (Fig. 6B), whereas no significant difference was observed in VFGs abundance between the two habitats (Fig. 6E). These results indicate that microbial communities in the HE habitat possess a heightened potential for antibiotic resistance. Moreover, although the HE communities harbor a broader diversity of virulence factors, the expression level of these genes did not significantly differ from those in the WE habitat. The LEfSe suggested that Uroviricota (a phylum of bacteriophages) is a biomarker taxon in the HE (Fig. 3B). Similarly, in the viral community, the phage centrality in the HE was higher than that in the WE (Fig. 4E, F, H). Based on these findings, we propose that phages play a crucial role in the dissemination of virulence factors and the expansion of virulence factor type within the microbial community. Several studies have reported that phage-mediated horizontal gene transfer promotes the spread of virulence genes and antibiotic resistance within microbial communities^30,31, particularly in HE¹⁹. This study revealed that ARGs directly affected VFGs (Fig. 5A–C and Table 2), whereas phages played a mediating role in increasing the number of VFGs in the HE (Fig. 5B, C and Table 2). Human activities that promote phage abundance further alter the diversity of VFGs in microbial communities of wild mice. These results suggest that the management of infectious diseases transmitted by mice should consider multiple aspects, particularly antibiotic pollution, which may indirectly facilitate the dissemination of virulence genes via phage-mediated gene transfer. In addition, an improved understanding of the mechanisms by which phages modulate VFGs and ARGs across various environmental conditions may facilitate the development of more effective phage-based therapeutic strategies.

In this study, we selected five phage-related VFGs (gspF, gspJ, gspK, gspM, and entS) and five phage-related ARGs (emrA, emrB, mdtP, pmrF, and ampH) (Fig. 6 and Supplementary Fig. 3). gspF encodes the inner membrane protein GspF, gspJ encodes minor pseudopilin GspJ, gspK encodes GspK, which participates in the signal recognition particle pathway, and gspM encodes the inner membrane platform protein GspM³². These “Gsp” proteins belong to the Type II secretion system (T2SS), which is significant to bacterial pathogenicity^32,33. The filamentous phage encodes the cholera toxin, which is secreted through the mediation of T2SS, playing a significant role in the enhancement of Vibrio cholerae virulence^34,35. In contrast, the phage-encoded protein PIT2, which regulates bacterial quorum sensing, inhibits T2SS and decreases the virulence of Pseudomonas aeruginosa³⁶. However, only a few studies have reported the relationship between phages and T2SS and virulence genes in Escherichia coli. Among the ARGs, emrA and emrB encode components of a multidrug efflux pump in Escherichia coli, which contribute to resistance against fluoroquinolones³⁷. The mechanisms underlying the trade-off and co-evolution between phage-mediated ARGs and VFGs are partially understood. Our study provides potential clues and research directions for investigating these mechanisms.

The pathogen centrality index proposed in this study provides a valuable metric for assessing the ecological integration of pathogens within microbial communities, reflecting their potential influence through connectivity rather than direct pathogenic risk. Our analysis demonstrated a significantly higher pathogen centrality in HE compared to WE environments. indicating a more embedded and interactive ecological role of pathogens in HE. This heightened ecological integration of pathogens in human-impacted habitats is a hallmark of the One Health paradigm: anthropogenic environmental change alters the structure of wildlife-associated microbial communities, potentially creating conditions that favor pathogen persistence and interaction. At the genetic level, both the richness (count) and expression (abundance) of ARGs were markedly elevated in HE (p_₁ < 0.01, p_₂ < 0.001; Fig. 6A, B), suggesting enhanced antibiotic resistance capacity. Similarly, the richness of VFGs was significantly greater in HE (p < 0.001; Fig. 6D). However, no significant difference was observed in VFG expression between HE and WE (p = 0.2952; Fig. 6E), indicating a dissociation between genetic capacity and actual expression. This discrepancy suggests that while phage-mediated horizontal transfer significantly enriches the VFG types in HE (Fig. 5B, C), these acquired genes are maintained at baseline expression levels, thereby not immediately elevating pathogenic threat. Phages, facilitated by human activities, appear to serve as key agents in disseminating both VFGs and ARGs, enhancing the genetic repertoire without consistently elevating virulence expression. The significant co-occurrence of phage-associated VFGs and ARGs in HE habitat points to a coordinated mechanism, potentially involving systems such as the T2SS, which may mediate the trade-off between virulence and resistance traits. These mechanistic links close a critical knowledge gap in the One Health framework by demonstrating how antibiotic pollution, a quintessential human environmental pressure, can indirectly amplify the virulence gene repertoire of wildlife microbiomes through phage-mediated horizontal gene transfer.

In conclusion, this study successfully established pathogen centrality as a quantitative indicator of pathogen ecological integration, which was significantly elevated in human-impacted environments. We further disentangled the complex relationships among habitats, phages, ARGs, and VFGs, revealing that ARGs directly and strongly drive the accumulation of VFG types, while phages mediate the habitat effect by expanding the virulence gene repertoire. Notably, the dissociation between enriched VFG counts and their unchanged expression levels suggests that phage-mediated horizontal transfer broadens the genetic capacity for virulence without immediately elevating pathogenic threat. These results collectively achieve our goal of providing both a novel quantifiable metric and a mechanistic understanding linking human activities to the shifting pathogenic potential of wildlife microbiomes. Grounded in the One Health paradigm, our findings underscore that surveillance and management of rodent-borne infectious diseases must move beyond pathogen detection to include environmental monitoring of antibiotic pollution and molecular insights into phage-bacteria interactions, thereby enabling more accurate prediction and mitigation of emerging health risks at the human-animal-environment interface.

Methods

Sampling and sequencing

Wild mice samples were collected at YunHe County, LiShui City, Zhejiang Province, China (119.20 E, 27.53 N). All animal procedures were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the Eastern Theater Center for Disease Control and Prevention of PLA. The study was conducted in accordance with the Chinese National Guidelines for the Ethical Review of Animal Welfare (GB/T 35892-2018) and the AVMA Guidelines for the Euthanasia of Animals. Following capture, mice were humanely euthanized by cervical dislocation, a method recommended by the AVMA Guidelines for the Euthanasia of Animals and approved by the institutional IACUC. This procedure was performed rapidly by skilled personnel to ensure immediate and irreversible loss of consciousness. Death was verified by the cessation of heartbeat and respiration, along with the absence of pedal and corneal reflexes. Spleen and liver tissues were then immediately collected to minimize post-mortem RNA degradation. All mice were divided into two groups: 56 samples were collected from HE, including human residential areas and agricultural land, and 70 samples were collected from WE, where mice were rarely affected by human activities. For each group, spleen and liver tissues were collected and stored at −80 °C until nucleic acid extraction. Total DNA and RNA were simultaneously extracted from the pooled spleen and liver tissues of each mouse using the AllPrep DNA/RNA Mini Kit (Qiagen, Germany) following the manufacturer’s instructions. The quality and concentration of the extracted DNA and RNA were assessed using a Qubit fluorometer (Thermo Fisher Scientific, USA) and an Agilent 2100 Bioanalyzer (Agilent Technologies, USA). For DNA metagenomic sequencing, sequencing libraries were prepared using the MGI Easy Universal DNA Library Prep Kit (MGI Tech, China). For RNA sequencing, ribosomal RNA was depleted, and strand-specific RNA sequencing libraries were constructed using the MGI Easy RNA Library Prep Kit (MGI Tech, China). Both DNA and RNA libraries were sequenced on the DNBSEQ-T7 platform (MGI Tech, China) using a paired-end 150 bp strategy. The raw sequence data reported in this article were deposited in the Genome Sequence Archive³⁸ of the National Genomics Data Center³⁹, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA021663), and are publicly accessible at https://ngdc.cncb.ac.cn/gsa.

Metagenomic assembly and gene quantitation

After sequencing, quality control was performed using FASTP⁴⁰ (v0.18.0) to obtain clean data. Low-quality reads and adapter sequences were excluded. For DNA samples, we removed the sequence of the host mouse genome to reduce interference from the host sequence in subsequent analyses. Based on the final annotated taxonomic profiles, DNA samples yielded an average of 208 thousand assigned reads per sample (range: 98–345 thousand). RNA samples yielded an average of 5054 assigned reads per sample (range:1214–8001). While these numbers represent annotated reads after quality filtering and host sequence removal rather than raw sequencing depth, they directly reflect the effective data depth available for taxonomic profiling, functional gene annotation, and MAG recovery in this study. All clean reads were assembled into contigs using MEGAHIT⁴¹ (v1.2.9). Open reading frames (ORFs) were predicted using Prodigal⁴² (v2.6.3) on the assembled contigs. The ORF sequences were clustered into non-redundant gene sets using SeqKit⁴³ (v2.5.1) and CD-HIT⁴⁴ (v4.8.1). To quantify the relative abundance of an ORF, we used a method similar to transcripts per million using Salmon⁴⁵ (v1.10.3):

$$X_{OR{F}_i}=\frac{r_i/l_i}{{\sum }_{i=1}^n{r}_i/l_i}\times {10}^6,$$ 1

where X_ORFi is the relative abundance of i-th ORF (ppm, parts per million), r_i is the number of clean reads mapped to i-th ORF, and l_i is the length of i-th ORF.

Taxonomic and pathogen annotation

For each group, we obtained two sample types, namely DNA and RNA sequencing samples. Taxonomic annotation of DNA sequencing samples was based on non-redundant gene sets and was conducted using Kraken⁴⁶ (v2.1.2) with the NCBI nucleotide sequence database. Taxonomic annotation of RNA sequencing samples was based on clean reads and was conducted using MetaPhlAn⁴⁷ (v4.0.6). Based on the differences between the taxonomic community compositions of the DNA and RNA samples for each group, we identified two communities, namely eukaryal-bacterial and viral communities. DNA metagenomic sequencing, which robustly captures genomes of bacteria and eukaryotes, was used to profile the eukaryotic-bacterial community. The viral community was characterized using the RNA sequencing data, because RNA sequencing is more sensitive for detecting both RNA viruses and actively transcribed DNA viruses, thereby providing a more comprehensive view of the viral fraction in mouse tissues. For each mouse, DNA and RNA were co-extracted from pooled spleen and liver tissues, and all subsequent community analyses were performed on these composite samples. The pathogens in the communities were identified according to a list of human-transmitted pathogenic microorganisms published by the National Health and Wellness Commission of China.

The α diversity parameters, including Shannon, Simpson, Chao 1, and evenness indices, were calculated using R package “vegan”⁴⁸ (v2.6.4) to estimate the richness and diversity of microbial communities. Linear discriminant analysis (LDA) effect size (LEfSe) was performed to measure taxa enrichments in these communities with LDA scores of >3. The β diversity analysis based on the Bray-Curtis distance was conducted to assess the dissimilarity of taxonomic components among different communities. Non-metric multidimensional scaling (NMDS) ordination based on Bray-Curtis distance was performed using the “metaMDS” function in the R package “vegan”⁴⁸ (v2.6.4) to show the dissimilarity of taxonomic components among different communities. Permutational Multivariate Analysis of Variance (PERMANOVA) was conducted using the “adonis” function with 999 permutations. Beta-dispersion was assessed using betadisper and tested with permutest. A significant beta-dispersion test (p < 0.05) indicates heterogeneous within-group variances, warranting cautious interpretation of PERMANOVA results alongside visual ordination and effect size. Microbial co-occurrence networks were inferred using MicroNet-MIMRF⁴⁹ (v1.02). Network analysis and topology calculations were performed using the “igraph”⁵⁰ package (v2.1.2), and the network visualization was conducted using Gephi⁵¹ (v0.10.1).

Gene annotation and metagenome-assembled genome (MAG) binning

ARG and VFG annotations were conducted based on non-redundant gene sets using the ABRicate program (v1.0.1) with the criteria of identity >80% and coverage >80%. The database of ARGs was the Comprehensive Antibiotic Resistance Database (CARD, https://card.mcmaster.ca/)³⁷. The VFG database was the Virulence Factor Database (VFDB, https://www.mgc.ac.cn/VFs/)¹³.

MAGs were assembled to study the co-occurrence patterns of VFGs and ARGs at the genomic level. We collected 2081 bins using MetaBAT2⁵² and we used CheckM⁵³ to assess the quality of each bin. Forty-eight MAGs were selected based on the criteria of completeness of >70% and contamination of <5%. The taxonomy of these MAGs was determined using GTDB-Tk⁵⁴ (v2.0).

Statistical analysis and visualization

For network analysis, several topology metrics were applied to assess the ecological integration and potential influence of pathogens within the microbial communities. The node DC was calculated as follows:

$$DC=\frac{N_{\mathit{degree}}}{n-1},$$ 2

where N_degree is the degree of a node, and n is the total number of nodes in the network.

The pathogen centrality (PAC) and phage centrality (PHC) were calculated similarly to the DC.

$$PAC=\frac{E_{pathogen}}{n-1},$$ 3

where E_pathogen is the number of edges between a node and pathogens.

$$PHC=\frac{E_{phage}}{n-1},$$ 4

where E_phage is the number of edges between a node and phages.

All statistical analyses were performed using R (v4.2.3). The differences in taxonomic composition, α diversity, gene abundance, and network indices were tested using the Wilcoxon rank-sum test. Mantel tests were implemented to assess the associations between highly expressed genes and the components of microbial communities using the R package “vegan”⁴⁸ (v2.6.4). The Random Forest Model (RFM) was used to identify key factors that were sensitive to VFG abundance using the R package “randomForest”⁵⁵ (v 4.7.12). The Structural Equation Model (SEM) was performed to confirm the relationships between crucial factors and VFG abundance using the R package “lavaan”⁵⁶ (v 0.6.19). The results with p < 0.05 were considered statistically significant. “****” represents p < 0.0001, “***” indicates p < 0.001, “**” denotes p < 0.01, and “*” implies p < 0.05; “ns” indicates p > 0.05 (there is no statistical significance).

Author contributions

Chenqionglu Feng, Huanhuan Lu, and Jiaxuan Bian contributed equally as co-first authors. Ligui Wang and Chenqionglu Feng conceptualized the study and developed the methodology; Ligui Wang and Weilong Tan conducted the investigation and performed the experiments; Chenqionglu Feng carried out the formal analysis and data curation. Huanhuan Lu, Jiaxuan Bian, and Hongbin Song contributed to validation, resources, and manuscript review. Hui Wang, Huiqun Jia, Xiaoying Li, and Mingjuan Yang assisted in the investigation and visualization. Hongbin Song, Weilong Tan, and Ligui Wang provided supervision, project administration, and funding acquisition, and serve as corresponding authors. All authors reviewed and approved the final manuscript.

Data availability

All sequencing data are available at the Genome Sequence Archive of the China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number CRA021663.

Code availability

No custom code or software was generated for this study. All bioinformatics tools used are publicly available and cited in the Methods section.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41522-026-01054-z.