Metagenomic investigation reveals bacteriophage-mediated horizontal transfer of antibiotic resistance genes in microbial communities of an organic agricultural ecosystem

ABSTRACT Agricultural microbiomes are major reservoirs of antibiotic resistance genes (ARGs), posing continuous risks to human health. To understand the role of bacteriophages as vehicles for the horizontal transfer of ARGs in the agricultural microbiome, we investigated the diversity of bacterial and viral microbiota from fecal and environmental samples on an organic farm. The profiles of the microbiome indicated the highest abundance of Bacteroidetes, Firmicutes, and Proteobacteria phyla in animal feces, with varying Actinobacteria and Spirochaetes abundance across farm animals. The most predominant composition in environmental samples was the phylum Proteobacteria. Compared to the microbiome profiles, the trends in virome indicated much broader diversity with more specific signatures between the fecal and environmental samples. Overall, viruses belonging to the order Caudovirales were the most prevalent across the agricultural samples. Additionally, the similarities within and between fecal and environmental components of the agricultural environment based on ARG-associated bacteria alone were much lower than those of total microbiome composition. However, there were significant similarities in the profiles of ARG-associated viruses across the fecal and environmental components. Moreover, the predictive models of phage-bacterial interactions on bipartite ARG transfer networks indicated that phages belonging to the order Caudovirales, particularly in the Siphoviridae family, contained diverse ARG types in different samples. Their interaction with various bacterial hosts further implied the important role of bacteriophages in ARG transmission across bacterial populations. Our findings provided a novel insight into the potential mechanisms of phage-mediated ARG transmission and their correlation with resistome evolution in natural agricultural environments. IMPORTANCE Antibiotic resistance has become a serious health concern worldwide. The potential impact of viruses, bacteriophages in particular, on spreading antibiotic resistance genes is still controversial due to the complexity of bacteriophage-bacterial interactions within diverse environments. In this study, we determined the microbiome profiles and the potential antibiotic resistance gene (ARG) transfer between bacterial and viral populations in different agricultural samples using a high-resolution analysis of the metagenomes. The results of this study provide compelling genetic evidence for ARG transfer through bacteriophage-bacteria interactions, revealing the inherent risks associated with bacteriophage-mediated ARG transfer across the agricultural microbiome.

F oodborne outbreaks are ongoing public health issues around the world.In 2015- 2018, a total of 4,237 foodborne pathogen-related outbreaks occurred in the United States, causing 76,120 illnesses (1).Most fresh produce-associated foodborne outbreaks derive from cross-contamination through direct or indirect contact with contaminants, such as animal feces or contaminated water, in pre-harvest and post-harvest environ ments.Pathogens from animal wastes could be disseminated into the agricultural environment through run-off from animal farms, which often end up in the soil and water reservoirs used for irrigation and initial post-harvest washing (2).A recent survey showed that animal feces was the primary source of contamination that spread pathogenic Escherichia coli in soil, water samples from dairy farms, and even unprocessed milk (3).
Antibiotic resistance has become a serious food safety issue, primarily resulting from the overuse and misuse of antibiotics in the meat and dairy industries (4).The continu ously occurring and wide-spreading bacterial "superbugs, " resistant to multiple types of antibiotics in the agricultural environment, have jeopardized the effectiveness of antimicrobial intervention strategies (5)(6)(7).Antibiotic resistance genes (ARGs) are found to be prevalent in various agricultural environments, and their ecological and epidemio logical consequences pose legitimate risks to human health.For instance, Salmonella enterica, Escherichia coli, and Shigella were the most resilient foodborne pathogens to intervention through antibiosis due to the rapid evolution of resistance mechanisms (i.e., antimicrobial resistance or AMR).It was reported that AMR has been most prevalent in chicken meat compared to other meat products, resulting from the heavy use of antibiotics in the poultry industry (8).Similarly, a study on 33 strains of pathogenic E. coli isolated from milk, cheese, and dairy farms showed that more than 42% of the strains had resistance to at least one type of antibiotic (9).Another study revealed the prevalence of multiple antibiotic resistance mechanisms conferred by different ARGs among the Salmonella enterica populations isolated from farms (10).Moreover, these ARGs are ubiquitous in soil amended with livestock manure rather than in soil without manure amendment (10).Indeed, these studies provide strong evidence that livestock as well as animal-associated environments are the major reservoirs of microbial inocula for the continuous dissemination of bacterial strains with ARG-mediated persistence mechanisms (11).
Genetic mutation and horizontal gene transfer (HGT) mediated by different mobile genetic elements (MGEs) in prokaryotes are the major drivers of ARG evolution and dissemination for the continuous emergence of antibiotic-resistant strains (12,13).HGT in bacteria occurs through conjunction, transformation, and transduction that are facilitated by MGE, including retrotransposons, plasmids, and bacteriophages.Plasmidmediated ARG transfer is common in nature (14,15).For example, IncP-1ε plasmids, which have a broad host range, mostly harbor class-1 integrons, often carrying the ARG sul1 and can transfer among bacterial communities in manure and soil at high frequencies (16).Another study examined 193 ARGs of plasmid origin in agricultural reclaimed water using the metagenomic approach, providing evidence regarding the role of plasmid-mediated ARG transfer among bacterial communities in the agricultural water reservoir (17).
The critical roles of plasmids, integrons, and transposons for HGT-mediated dissem ination of ARGs have been well established.The important contributions of bacter iophage-mediated ARG transfer among foodborne pathogens have recently gained more attention.The increased resolution of natural microbial ecological dynamics has revealed bacteriophages as a major component of different environmental microbiomes through metagenomic approaches (i.e., metagenomic next-generation sequencing or mNGS).Regarding bacteriophages, ARG transfer often occurs via transduction, either a generalized (i.e., virulent and temperate phages) or specialized (i.e., temperate phages) mechanism (18).Several metagenomic studies have revealed the role of bacteriophages as major vehicles for the spread of ARG in different environments.A previous study on hospital wastewater samples indicated a significant abundance of ARGs in the phage DNA fraction (0.26%) compared to the bacterial fraction (0.18%) (19).Moreover, several phage-related MGEs have been found to facilitate horizontal transfer of ARGs, including plasmid-phage, transposon-phage, genomic island (GI)-phage, and gene transfer agents (GTAs) (20).
The interaction between viruses (mostly bacteriophages) and their bacterial hosts is crucial to the dynamic evolutionary processes in environmental microbiomes, which are important aspects of the success of pathogens in nature.The cyclical spikes of pathogen outbreaks in fresh produce as well as in the production and processing chain are believed to be associated with the rapid evolution of AMR due to various selection pressures.Therefore, an important question that needs to be addressed is how bacteriophage-mediated mechanisms contribute to the dynamic process of ARG dissemination.Here, we performed a metagenomic analysis of the most likely sources and vehicles (i.e., animal feces and surrounding environments) of pathogen dissemina tion from a natural agricultural field, highly focusing on the dynamics between the viral communities and their bacterial hosts.The results of this study provide insights into the potential impacts of natural ARG evolution and contribute to the growing body of evidence supporting the hypothesis that bacteriophages are significant contributors to the evolution of ARGs in bacterial communities.These findings will serve as a back bone for understanding the origins of future foodborne pathogen outbreaks across the agricultural pre-harvest and post-harvest production and processing pipelines.

Field for metagenomic survey of agricultural microbiomes
Our study location serving as an agricultural environment field was an organic farm in California, USA, with a more than 20-year history of vegetable and fruit production.Dairy and meat animals (cattle, goats, and pigs) were also seasonally reared at various locations across the farm that are also alternately used to produce leafy greens.Alternating the use of land on this farm has been a strategy to increase soil fertility organically.
To investigate the ARG profiles of an agricultural microbiome, this organic farm was used as a field for an ecosystem with minimal influence from the otherwise common farm strategy of antibiotic application.Since the common route of pathogen dissemina tion was from animal feces to surrounding environments, a total of 15 closely located sampling sites, including the animal feces from cattle, goats, and pigs, surrounding soil, and surface water, were investigated in this study (Fig. 1).To further understand the inherent interactions between bacteria and bacteriophages in ARG transmission, both bacterial and viral communities were recovered from each sample and subjected to metagenomic next-generation sequencing.

Diversity of bacterial microbiota in the organic agricultural environment
To have an overview of the microbiome profile, the metagenomic data sets of various environmental samples (Table S1) were transformed into taxonomic classification.Bacteria (71.97%-99.94%)and Archaea (0.09%-28.01%) are the most frequently occur ring taxa in the microbiome profile among our agricultural-related samples.With a specific focus on bacterial microbiota, samples were analyzed to obtain bacterial distribution at the phylum and genus levels (Fig. 2a).The top three most abundant bacterial phyla across the animal fecal samples were Bacteroidetes, Firmicutes, and Proteobacteria, based on their total occurrences ranging from 70.96% to 89.93% relative abundance in the total microbiota.Our results were consistent with the major phylum profile within the gut microbiomes of these representative animals reported in previous studies (21)(22)(23)(24).In addition, bacteria belonging to Actinobacteria (0.01%-10.35%),Candidatus Saccharibacteria (1.52%-3.89%),and Spirochaetes (0.43%-2.34%) were also represented in the bacterial population of fecal samples with varying abundance across farm animals.The bacterial populations in soil samples comprised primarily Proteobacte ria (44.08%),Verrucomicrobia (21.11%), and Chloroflexi (10.17%).On the other hand, water samples predominantly contained Proteobacteria (69.64%) and Bacteroidetes (28.20%) (Fig. 2a).Similar results were reported that Proteobacteria was the most common dominant phylum in the environmental microbiome, with varying relative abundance depending on the isolated sources (25)(26)(27)(28).
At the genus level, the most predominant groups of bacteria in the cattle fecal microbiota were composed of unclassified genomic bins under Firmicutes (56.42%),Oscillospiraceae (7.68%), Bacteroidetes (3.89%), and Proteobacteria (1.91%).Goat fecal microbiota included unclassified Bacteroidetes with a relative abundance of 46.56%, complemented by Campylobacter (15.91%) and Escherichia (14.84%).Additionally, unclassified Firmicutes (31.94%),Lactobacillus (29.14%), unclassified Bacteroidetes (7.75%), Prevotella (5.29%), and Bifidobacterium (4.49%) were detected in the pig fecal samples (Fig. 2a).Our current metagenomic data sets indicated that Pseudomonas (42.48%),Flavobacterium (23.22%),Comamonas (6.14%), and Janthinobacterium (5.58%) represent the most predominant genera in the microbiota of agricultural water samples.Unclassified genera under four phyla Proteobacteria (39.15%),Verrucomicrobia (21.11%),Chloroflexi (10.17%), and Acidobacteria (8.85%), as well as genus Escherichia (3.46%), were dominant in the microbiota of agricultural soil samples (Fig. 2a).The alpha diversity was conducted at the genus level and indicated that pig fecal microbiota had the highest alpha indices, followed by water and soil samples (Fig. S1a).The principal coordinates analysis (PCoA) further suggested that the compositional dissimilarity of bacterial communities between animal fecal and environmental samples was greater than the distance among the three animal fecal types (Fig. S2a).It was also shown that the distance between cattle and pigs was less than their distance from the goat group.Most importantly, a large proportion of unclassified taxa at the genus level emphasized the diversity of agricultural-related microbiomes and required more effort to comprehen sively characterize and expand the coverage of those microbial genomes in recent databases.
Overall, the consistency of the microbiome profile between our study and previous studies indicated that the current metagenomic data sets represent sufficient bacterial richness and coverage of data sampling that can be integrated and interpreted in the context of our central hypotheses on bacteriophage and host bacterial dynamics.

Diversity of viral microbiota in an organic agricultural environment
The same field, used to profile the diversity of bacterial microbiota in the agricul tural ecosystem, was also examined in parallel to establish the viral composition and enrichment signatures (Fig. 2b).Overall, metagenomic sequence tags corresponding to known taxonomies were assigned to estimate the viral composition across the agricul tural samples.
While the order Caudovirales represents a common signature across all metage nomic samples, the trends in individual viral representation signatures are quite diverse and more specific across the fecal samples relative to the profiles observed for fecal bacterial microbiota (Fig. 2b).Among known viruses, Geplafuvirales (1.91%) and Petitvirales (1.18%) were the most predominant signatures in the cattle fecal sample, whereas Tubulavirales (60.25%) and Cirlivirales (24.29%) represent the most predomi nant signatures in goat and pig fecal samples, respectively.Other viral orders, such as Petitvirales (0.01%-0.02%) and Chitovirales (around 0.01%), were also detected in the majority of animal fecal samples, albeit in relatively low abundance.The results were also aligned with earlier observations by other researchers that phages belonging to Tubulavirales, Petivirales, and Caudovirales were detected in pig feces (29).
The order Caudovirales represents the most predominant group of viruses in the agricultural soil and water samples, accounting for 82.76% and 34.08% of the total viral taxonomic groups detected in soil and water samples, respectively.Additionally, Petitvirales, with a relative abundance ranging from 11.76% to 19.00%, was detected in both the agricultural soil and water samples, while Geplafuvirales, with a 17.51% relative abundance, was present in the water sample.
In this study, we detected metagenomic signatures corresponding to 16 viral families as well (Fig. 2b).Of these, three phage families, Myoviridae, Podoviridae, and Sipho viridae, were commonly detected in all animal fecal samples, with a relative abun dance ranging from 0.1% to 3.4%.These trends are consistent with the results of previous studies showing that bacteriophages under families Myoviridae, Podoviridae, and Siphoviridae are known to occur ubiquitously in the fecal microbiomes of ruminant and non-ruminant farm animals, including poultry, cattle, pigs, goats, and sheep (30,31).Additionally, the families Inoviridae (60.26%) and Circoviridae (24.30%) were the most predominant in the goat and pig microbiomes, respectively.In contrast, a distinct composition of the cattle microbiome revealed that the families Genomoviridae (1.91%) and Microviridae (1.81%) represented its most predominant components.Moreover, we found that the phages belonging to the families Myoviridae (16.87%-48.16%)and Podoviridae (16.94%-18.42%)were the primary group of viruses in the agricultural soil and water samples.In addition, Siphoviridae (16.21%) and Microviridae (11.76%) were relatively predominant in the soil virome, while the families Microviridae (19.00%) and Genomoviridae (17.51%) were quite endemic in the virome of water samples.
Overall, the water virome had the highest alpha diversity, followed by cattle fecal samples (Fig. S1b).Based on the results of this survey, it appears that different types of bacteriophages are endemic in both the animal fecal and environmental samples collected in this study (Fig. S2b).Moreover, the abundance and diversity of viruses varied widely between animal species, even though the animals were grown under the same conditions (31,32).Compared to the "core" bacterial population, the higher variability of virome observed among farm animal samples serves as a further basis to investigate the underlying bacteriophage-bacterial interactions and the potential influence of bacteriophages in shaping bacterial communities.

Diversity of ARGs in the bacterial microbiomes of agricultural environments
To illustrate the diversity and abundance of ARGs in the bacterial populations across the fecal and environmental components of the agricultural environment, we performed a more in-depth analysis of ARGs in the metagenomic data sets using both the trimmed reads and assembled metagenome contigs.The results of this analysis revealed a total of 24 types of ARGs in bacterial metagenomes from diverse samples with different abundances (Fig. 3).ARGs that were predominant in the bacteriome of environmental samples included genes conferring resistance to aminoglycosides, macrolide-lincosa mide-streptogramin (MLS), and tetracyclines.In detail, the predominant types of ARGs in the bacterial microbiomes of the fecal samples appeared to encode aminoglycosides (0.54%, 0.46%, and 0.60% in cattle, goat, and pig feces, respectively), MLS (0.28%, 0.24%, and 0.18% in cattle, goat, and pig feces, respectively), and oxazolidinone (0.07%, 0.07%, and 0.05% in cattle, goat, and pig feces, respectively).Fewer ARGs were detected in the soil samples, which included the genes encoded for resistance to aminoglycosides (0.17%) and MLS (0.07%) among the predominant types.In contrast, the bacterial microbiome of the water samples had a relatively higher abundance and diversity of ARGs that include genes encoding for aminoglycosides (0.35%), MLS (0.318%), cationic antimicrobial peptides (0.11%), and drug and biocide resistance compounds (0.10%).These results are consistent with earlier findings based on the analysis of animal gastrointestinal microbiomes and also indicate that water is a potential route for the dissemination of ARG-associated bacteria in agricultural pipelines (33)(34)(35).Moreover, these trends indicate that ARGs that are predominant in the bacterial microbiomes across the agricultural environment confer resistance to most of the major classes of antibiotics used in human and animal medicine (36)(37)(38).

Diversity of ARGs across the viral microbiomes of the agricultural environ ment
A total of 19 different ARG types were identified across the viral communities of the fecal and environmental samples from this agricultural field (Fig. 3).The patterns of ARG abundance in the virome were quite distinct from the abundance in the bacteriome.ARGs detected in the fecal viral microbiome of cattle encoded resistance to MLS (0.01%) and elfamycins (0.01%) as the most predominant.ARGs for resistance to beta-lactams (8.80% in cattle and 1.2% in goats), aminoglycosides (0.93% in cattle and 0.16% in pigs), and MLS (0.27% in cattle and 0.14% in goats) were the three most predominant.These trends indicate a peculiarly high abundance of ARGs in the viral microbiomes of goats and pigs.The most predominant types of ARGs in the viral microbiome of soil include those encoding for drug and biocide resistance (0.09%), cationic antimicrobial peptides (0.06%), and resistance to MLS (0.06%).In addition, resistance to MLS (0.01%) represents the most predominant ARG in the viral microbiome of water samples.

Bacterial and viral microbiome profiles and ARG profiles in the agricultural environmental components
Based on taxonomic enrichment, we compared the compositional similarities and differences between each animal's fecal (cattle, goat, and pig) and environmental (soil, water) components of the agricultural environment field as a first step in understanding the spatial dynamics across the total microbiome.The bacterial microbiome profiles had 70%-77% overlap across three animal fecal sample (cattle, goats, and pigs) metage nomes and 68%-82% overlap between the soil and water sample metagenomes (Fig. 4a).The bacterial metagenomic profile of all three fecal samples combined was 60% and 72%, similar to the bacterial metagenomic profiles of water and soil samples, respec tively.
In contrast to the high levels of similarity observed for the bacterial metagenomic profiles, distinct profiles across the fecal and environmental components are much more evident based on the composition of viral metagenomes (Fig. 4a).The viral profiles across the animal fecal components (i.e., animal fecal samples combined) were only 17%-32% similar to each other, while the profiles of environmental components (soil, water) had 14%-68% similarities.The qualitative evaluation indicated that the viral metagenomic profiles of animal feces shared 48% and 10% similarity with the profiles of water and soil samples, respectively.Overall, these results indicate that viral communities are much more diverse between the fecal and environmental components of an agricultural environment, with strong indications of unique composition in the biological and physical aspects of the environment.
We further examined the specific components of the total microbiome (mostly the bacteriome) that are known to be associated with ARG to assess similarities and differences across the animal fecal and environmental components of the agricultural environment field (Fig. 4b).In the animal feces, the composition of the ARG-associated bacteria in the microbiomes had an average of 43% similarity across the farm animals of cattle, goats, and pigs, while the environmental components (water, soil) had 35% similarity.Overall, the similarities within and between animal fecal and environmental components based on ARG-associated bacteria alone were much lower than those in the total bacterial composition of the microbiomes.This suggests that ARGs are acquired not only from lateral transfers but also from other mechanisms and sources, potentially including plasmids and bacteriophages.
We further classified the viruses and bacteriophages in the total metagenomic data sets in terms of their known association with ARGs based on literature and database information (Fig. 4b).Enrichment analysis indicated that the composition of ARGassociated viruses in the animal fecal samples had 28%-44% similarities to the total viral composition of the fecal metagenomes.Similarly, the composition of ARG-associated viruses was found to be 42% and 71% similar in water and soil samples, respectively.The composition of ARG-associated viruses in the animal fecal samples was 57% and 50%, similar to that of water and soil samples, respectively.The significant similarities in the profiles of ARG-associated viruses across the animal fecal and environmental compo nents of the agricultural environment field suggest that certain types of bacteriophages in the agricultural environment are likely to play critical roles in disseminating ARGs in nature.

Bacteriophage interaction with the bacterial population as a potential mechanism for ARG transmission
The process of transduction is an important mechanism by which bacteriophages can facilitate the horizontal transfer of ARGs across bacterial populations (39,40).How much this process contributes to the dissemination of ARGs in the agricultural environment is a controversial topic without definitive evidence, if not contrasting results from independ ent investigations (41)(42)(43).To address this important question, we used our spatial metagenomic data sets across the animal fecal and environmental components of the agricultural environment field to develop predictive models of the potential interactions between bacteriophage and bacterial populations and to make inferences regarding bipartite ARG transfer networks.
For the microbiomes of the environmental components, a total of 91 putative bacteriophage-bacterial interactions were detected in the soil microbiome (Fig. 6a).Viruses belonging to the Siphoviridae family contained the greatest number of ARGs that are commonly found in bacterial populations (57.43%), followed by Myoviridae (12.16%) and Podoviridae (6.75%).Similar results were observed in the viral microbiome of soil, where both Myoviridae and Podoviridae shared similar ARGs associated with the MLS mechanism with their host bacteria.However, the ARG types shared by Siphoviridae and their hosts were quite diverse.Specifically, a total of 8 putative ARG types were identified in the network of Siphoviridae with different bacterial phyla, in which resistance to aminoglycosides (45.27%) and tetracyclines (7.43%) accounted for the majority of ARGs in the network.Additionally, 17 bacterial phyla showed potential ARG overlaps within the viral microbiome, including Actinobacteria-Siphoviridae, Firmicutes-Siphoviridae, and Proteobacteria-Siphoviridae, with an average of 6.75%.On the other hand, a total of 108 bacteriophage-bacterial interactions were found in the water samples (Fig. 6b).The three most common families were Myoviridae (13.82%),Podoviridae (5.85%), and Siphoviridae (50.53%).The other two viral families-Ackermannviridae (7.97%) and Herelleviridae (1.59%)-were also in the network of putative phage-bacterial interaction with low frequency and statistical power.Similarly, the interactions associated with Siphoviridae showed diverse ARGs in the network, including resistance to aminogly cosides (28.19%), elfamycins (11.17%), and tetracyclines (6.91%).The most common bacterial interactions with Siphoviridae in the water microbiome included Proteobac teria-Siphoviridae (9.57%), Firmicutes-Siphoviridae (9.04%), Actinobacteria-Siphoviridae (8.51%), and Bacteroidetes-Siphoviridae (7.97%).In contrast, Herelleviridae within the viral microbiome of water samples were all related to resistance to rifampin, while Ackermannviridae were associated with resistance to either aminoglycosides or MLS.Similar results were obtained in the Podoviridae and Myoviridae families, but most Podoviridae viruses were associated with resistance to MLS.Myoviridae shared the most ARGs associated with resistance to aminoglycosides and MLS.
Noticeably, among the putative bacteriophage-bacteria associations, the tailed dsDNA phages under the order Caudovirales, Siphoviridae, and Myoviridae in particular appeared to be the most involved in the putative interactions predicted in the network models by virtue of their interactions with diverse bacterial phyla and their correlations with corresponding types of ARG across the samples.Although information regarding the association between phage types and ARG types is lacking, our results suggest a positive correlation between ARG types and certain phage-bacterial interactions.Moreover, these tailed phages and their bacterial hosts are widely present in various environments, such as animal feces, sewage water, and plants, likely suggesting underes timated ARG transfer events in natural environments (44)(45)(46)(47).

High resolution microbiome profiles in an organic agricultural field
Agricultural environmental microbiota, including bacteria, archaea, fungi, and viruses, represent a complex and dynamic environment of great significance to food safety and human health.To minimize the occurrence of foodborne outbreaks, studies routinely explore the microbiome profiles of diverse agricultural samples using different analytical approaches (48,49).Metagenomic next-generation sequencing has been established as a powerful approach for understanding population dynamics in agricultural micro biomes (49)(50)(51)(52)(53)(54).As a major source of foodborne pathogens, the gut microbiota of farm animals, including bovines, swine, and poultry, have become a hotspot for metagenomic studies, with primary emphasis on eukaryotic pathogens causing infectious diseases in humans and animals (55)(56)(57).Here, we established a field for an agricultural environ ment that adequately represents the microbial population dynamics and the possible mode of ARG transmission across the agricultural processing pipeline.Using this field, we examined the metagenomes of both the animal fecal (cattle, goats, and pigs) and environmental (water, soil) components of the ecosystem to illuminate the bacterio phage-bacterial interactions and the potential effects on the ARG transfer across the agricultural ecosystem.The metagenomic sequencing of this study generated more than 62 and 65 Gb of raw reads for microbial and viral populations, respectively.The trends and signatures revealed from the analysis of these metagenomic data sets echoed many of the results previously reported for similar agricultural samples, indicating the high degree of interpretability of the biological implications of the data sets we generated from such fields.The current metagenomic data sets contribute to the enrichment of publicly available environmental microbial metagenomic profiles for further use in the hypothesis-driven comparative investigation of microbial population dynamics across different types of agricultural ecosystems.

Bacteriophage-bacterial interactions in an agricultural environment
The nature of bacteriophage-bacterial interactions has been explored in different environmental samples.General trends indicate that bacteriophages drive the diversity and evolution of the microbial population (58).Bacteriophage-bacterial infection occurs mainly in a random, one-to-one, nested, and modular network, which illuminates the major influence of bacteriophages on complex microbial communities (59).It has been shown that different bacteriophage families are positively correlated with the infection network with bacteria due to differences in morphology.Bacteriophages with a long tail possessed a broader host range compared to the short-tailed types as a result of the interaction of the specific tail structure with bacterial host receptors (60).Current virus-host databases include 2,307 Siphoviridae-bacteria pairs and 862 Myoviridae-bac teria pairs, whereas only 369 Podoviridae-bacteria pairs have been reported so far (61).Similarly, results showed that bacteriophages belonging to the family Podoviridae have a narrow host range, while most siphophages and myophages show a relatively broad host range (62,63).These findings provide additional support for the biological significance of Siphoviridae-bacteria pairs and Myoviridae-bacteria pairs as the most predominant trends in the bacteriophage-bacterial host interaction networks that we have uncovered in this study.

The crucial role of bacteriophage in ARG dissemination
The critical roles of bacteriophages as ARG reservoirs for dissemination in the environ ment have long been appreciated.Metagenomic studies previously showed a direct correlation between ARG distribution and diversity across bacteriophage populations from animal wastewater (swine).The study revealed that tetracycline resistance genes had the highest abundance, while aminoglycosides had the highest diversity among the ARGs (42).In addition, studies on ARGs in bacteriophages from hospital wastewater showed highly detectable levels of β-lactamases in bacteriophage DNA fractions (19).Among the samples we examined in this study, the viral population had a relatively high abundance of ARGs encoding for elfamycin, aminoglycosides, and MLS resistance proteins.Elfamycins have a broad antimicrobial spectrum against pathogens such as Yersinia, Haemophilus, and Streptococcus.They have been commonly used to enhance the growth of farm animals at different growth points (64,65).Aminoglycoside-type antibiotics are a critically important antimicrobial compound to treat human infections, as the World Health Organization reported.With their high efficacy and broad-spec trum antibacterial effects, aminoglycoside-type antibiotics are potent and powerful treatments for severe illnesses caused by Gram-negative bacterial pathogens (36).Due to the frequent use of MLS for treating staphylococcal infections, the MLS B resistance phenotype has been widely reported in clinical isolates of Staphylococcus aureus (37,38).However, several studies regarding the investigations of antibiotic resistome in human-impacted areas present a need for critical considerations on the continuous use of aminoglycoside-and MLS-type antibiotics (66,67).Notably, one study reported a total of 22 ARG types as significant components of the bacterial microbiome of urban river sediments; the abundance of these ARGs was positively correlated with the selective pressure of antibiotics extensively used in the nearby areas (33).Another recent study identified 71 ARG subtypes that were detectable in different sources of potable water in the Huaihe River Basin (China).It was also found that ARG profiles were determined by the number of livestock and health facilities around that area (68).Coincidentally, these findings further confirm that agricultural operations and anthropo genic activities highly affect antibiotic resistome in agricultural environments.Antibiotics used within a certain vicinity have major impacts on the microbial communities around the proximal areas.These correlations have apparent significance for the restructuring of the antibiotic resistome in environmental bacterial microbiomes.Nevertheless, the underlying mechanisms of ARG transmission within the microbiome are still not clear.
In this study, we examined the taxonomic similarity of the microbiome components carrying ARGs between animal fecal and environmental samples.Viruses carrying ARGs were found to share higher similarity among all the sample types than bacteria carrying ARGs only, suggesting the potential roles of bacteriophages in the mechanisms for ARG transfer across environments and ecosystems.Bacteriophage-bacterial interactions facilitate horizontal gene transfer both within bacteriophage communities and into bacterial host communities.This process contributes to the rapid evolution of microbial populations across microorganisms in the agricultural ecosystem (69).However, the precise mechanisms that explain the importance of bacteriophages in MGE horizontal gene transfer of ARG are still unclear from the available data (19,41,70).Therefore, the role of bacteriophage-mediated ARG transfer was estimated in this study by screen ing MGE genetic markers within the assembled contigs carrying ARGs.Notably, the results showed that an average of 60% of the total assembled contigs with ARGs in the viral library were identified as plasmid-related markers (Table S3), indicating the potential gene exchange between plasmids and bacteriophages as well.Our findings also correlate with previously reported data that showed that the armA gene, which is usually only harbored by multicopy plasmids (MCPs), as well as other fragments from plasmid DNA, could use the bacteriophage capsids to disseminate, especially as most MCPs do not possess mobility genes (71).In addition, it was suggested that MCP-borne ARGs are encapsidated up to 10,000 times more efficiently than the ARGs carried by low-copy plasmids that are significantly larger in size.Together, these findings support that a distinct form of bacteriophage phage-mediated ARG transfer by encapsidation of ARGs from plasmid DNA and the transduction of the bacteriophage genome are likely to occur in nature.It is important to highlight that, based on the results of this study on the bacteriophage-bacterial interaction on the ARG transfer, Siphoviridae carrying various ARGs have a diverse host range, whereas Podoviridae carrying MLS only have a relatively narrow host range.Although no information is available regarding the association between bacteriophage types and ARG types, our findings provided a novel insight into the correlation between bacteriophage-bacterial interactions and the dissemination of certain ARG types.Thus, further investigations are necessary to explore the underlying mechanism of bacteriophage-mediated transduction for ARG mobilization and its relationship with the emergence of certain AMR pathogens.
Through high-resolution metagenomic sequencing, our study provided valuable genetic information regarding the microbiome and virome from various agricultural-rela ted samples.However, some limitations need to be further improved.Specifically, in this study, water and soil viral metagenomic sequences were less than three replications due to insufficient quantities of viral DNA.Therefore, the rarefaction of viral metagenomic data from each sample type was conducted and suggested that our sequences met adequate depth in the viral richness at the family level (Fig. S3).The lack of replications should be considered in future studies through protocol optimization, such as increasing sample size and improving methods for viral concentration, to yield enough DNA for metagenomic sequencing.In addition, our findings indicated the important role of bacteriophage in the horizontal transfer of antibiotic resistance genes.The findings bring attention to a deep investigation regarding the underlying mechanism of phage-bacte rial interactions on ARG transfer and their association with agricultural-related environ mental factors for future direction.

Metagenomic sampling across the agricultural environment
A total of 15 samples for metagenomic studies were collected from animal fecal and environmental components of the agricultural environment field selected for this study.This field is an organic farm in California, United States.Sample collection was conducted in September 2018.Fecal samples included the feces from three types of animals (cattle, goats, pigs) that are reared on the farm.Environmental samples included the water and soil from different locations across the farm, as illustrated in the location map in Fig. 1.Sampling was done according to our previous study (47).In brief, each type of sample was collected into three different sterile Whirl-pak bags (Nasco, WI, USA) or sterile jars in three replications.A total of 400 g of soil samples collected 5 cm below the soil surface were collected using a sterile sampling spatula.A total of 300 g of animal feces were hand-collected by wearing sterile gloves and placed into sterile Whirl-pak bags.Due to the low moisture content of goat feces, multiple goats' feces were combined to obtain the 300 g of the required sample size.A total of 1 L of surface water from a river were sampled using a sampling pole with a wide-mouth bottle and transferred into sterile jars.All samples were stored on carbon dioxide ice and transported to the laboratory within 3 h of collection for further processing.

Microbial and viral DNA extraction
For soil and animal feces, a total of 50 g of each sample were weighed, transferred to filtered bags, and further mixed with 100 mL of PBS buffer for 2 min.Fifty milliliters of the sample-PBS mixture or 1 L of water samples was aliquoted and centrifuged for 10 min at 5,000 × g.The supernatant was filtered through a 0.45-µm pore size filter membrane and again by a 0.22-µm pore size filter membrane, which allowed the passage of virus particles in the filtrate, while microbiota were retained on a 0.22-µm membrane surface.The microbiota was collected from a 0.22-µm membrane and further subjected to washing with PBS prior to the concentration centrifuge at 10,000 × g for 2 min to avoid virus contamination.All viral particles from a 0.22-μm-filtered solution were concentra ted to 500 µL using 50 kDa Amicon ultra centrifugal filter units at 4°C.Concentrated microbial and viral samples were treated with DNase (100 U/mL) to eliminate free DNA contamination.The microbial and viral DNA was extracted using the QIAamp DNeasy PowerSoil Kit and Power Virus DNA&RNA extraction Kits, respectively.Three replicates of microbial DNA were obtained; however, for viral libraries, only two replicates and one replicate were obtained for water and soil viral metagenomics, respectively, due to the low concentration of virome.

Metagenomic sequencing and analysis
DNA samples were sequenced on the NextSeq at 150 bp paired-end reads.An average of 25 and 40 million paired reads were generated for the microbial and viral metagenom ics, respectively.Raw reads were preprocessed using sickle v1.33 with a quality cutoff of 30 and a minimum length of 100 bp prior to further metagenomic analyses (72).Specifically, trimmed reads of each sample type were combined with the minimum length of a contig of 300 bp and further de novo assembled by megahit v1.2.9 (73).The taxonomic classification of the microbiome was performed using Metaphlan 4 (version 4.0.6) with the default setting (74).Viral taxonomic classification was conducted by mapping against an annotated DNA viral database from MiCop via bwa-mem v0.7.17-r1188 (75,76), followed by reassignment of multi-mapped reads and normalization of taxonomic abundance levels using MiCop (75).The alpha and beta diversity was analyzed using the R package vegan (version 2.6-4) and agricolae (version 1.3-5) based on the Shannon index and Bray-Curtis matrix, respectively, prior to the visualization via ggplot2 (version 3.4.1).To verify the richness and coverage of viral family level, virus reads were rarefied with 1 million increments using SeqKit v2.4.0 with seed value 99 and further processed for abundance estimation via MiCop as described above (75,77).Detection and quantification of ARG abundance were performed using blastn against the MEGARes v2 database (accessed on 28 September 2020) (78).The presence of the common ARG between bacterial and viral contigs with an alignment blast E-value <0.05 was used as a potential association and transfer between them.The cumulative coverage of ARG by contigs was calculated with bedtools v2.29.2 (79).All tools were performed with default settings unless otherwise specified.

FIG 1
FIG 1 Spatial map of environmental and animal fecal sampling for metagenomic analysis across the organic farm.A total of five sample types in close proximity to each other, including animal feces from cattle, goats, and pigs, soil, and surface water around the animal activity, were used for the analysis of metagenomes.Each sample type was collected in triplicate.P1 = cattle feces; P2 = goat feces; P3 = pig feces; P4 = soil; P5 = surface water (stream).

FIG 2
FIG 2 Heat maps showing the composition of the total bacterial (a) and viral (b) microbiomes across the agricultural environment field as revealed by metagenomic analysis of animal feces (cattle, goats, and pigs) and environmental samples (soil and water).Taxonomic interpretation of metagenomic sequence tags was accomplished by mapping bacterial sequence reads using Metaphlan4 (version 4.0.6) and viral sequence reads against the DNA viral database from Micop via bwa-mem (v0.7.17-r1188).In the bacterial profile, only the top 30 most abundant genera, which represent the most significant and quantifiable enrichments of bacterial sequence tags, are included in the heat map (a).The viral profile includes all known families of DNA viruses that were detected with significant statistical confidence across the libraries of the fecal and environmental samples (b).

FIG 3
FIG 3 The ARG profile within total bacterial and viral communities across the agricultural environment field was established by metagenomic analysis of animal feces (cattle, goats, and pigs) and environmental samples (soil and water).The abundance of ARGs was estimated by blastn against the database from MEGARes v2.Different ARGs were detected and present in the bacterial and viral populations among the animal fecal and environmental samples by blastn with an E-value <0.05.

FIG 4
FIG 4 Qualitative evaluation of taxonomic diversity and similarities between the animal feces (cattle, goats, and pigs) and environmental samples (water, soil) of the agricultural environment field as measured by the number of unique accessions from bacterial and viral contigs (a) and those with known associations to ARGs (b), respectively.Venn diagrams compare the similarities between the animal fecal and environmental samples.Contigs were used to detect bacterial or viral species (blastn with the setting of E-value <0.05) based on bacterial RefSeq and non-redundant viral database.

FIG 5
FIG 5 Hypothetical model of potential ARG transfer across bacterial and bacteriophage populations based on the analysis of occurrences in the metagenomic profiles of (a) cattle, (b) goat, and (c) pig fecal microbiomes.Putative interactions were established by the presence of common ARGs between bacterial and bacteriophage metagenomic assemblies from the same sample.The corresponding bacterial phyla and bacteriophage families were shown on the left and right tracks.The width of the box corresponds to the number of predicted or putative ARG transfers for a given bacterial and viral species.The color of the tracks shows the general mechanisms of the given ARG.Each color represents an ARG type, as shown in the figure legend.

FIG 6
FIG 6 Hypothetical model of the potential ARG transfer across bacterial and bacteriophage populations based on the analysis of occurrences in the meta genomic profiles of (a) soil and (b) water microbiomes.Putative interactions were established by the presence of common ARGs between bacterial and bacteriophage metagenomic assemblies from the same sample.The corresponding bacterial phyla and bacteriophage families were shown on the left and right tracks.The width of the box corresponds to the number of predicted or putative ARG transfers for a given bacterial and viral species.The color of the tracks shows the general mechanisms of the given ARG.Each color represents an ARG type, as shown in the figure legend.