A total of three types of samples were collected from a poultry slaughterhouse in a district of Guangzhou: animals, occupational personnel, and environmental samples. The environmental samples included air, soil, sludge, swabs from transportation vehicles, and swabs from the slaughterhouse workshop. The collection protocols were as follows: (1) Animal Samples: Sterile cotton swabs were inserted into the oral cavity and cloaca of chickens or ducks, rotated three times, and then removed. The swab’s tail was discarded, and the swab was immersed in sterile 0.5% BSA-PBS buffer for preservation. Three chickens or ducks from each of the three spaces (caged area, pre-slaughter area, slaughter area) had their oral and cloacal swabs mixed to form one sample. (2) Occupational Personnel Nasal Swab Samples: To collect nasal swab samples from occupational personnel, a sterile cotton swab was gently inserted into the nasal pharynx of the participating volunteer. After a few seconds, the swab was gently rotated and removed. The swab’s tail was discarded, and the swab was immersed in sterile 0.5% BSA-PBS buffer for preservation. Nasal swab samples from a single person with both nostrils were placed in an individual sample collection tube. Written informed consent was obtained from all participants. (3) Air Samples: BioSamplers KIT (225-9595, SKC, Eighty Four, PA) were installed at approximately 1.5 meters above floor at the ventilation points in the slaughter area, pre-slaughter area, and caged area (3 sampling points in total). Using 0.5% BSA-PBS buffer at a flow rate of 8 mL/h, sampling was conducted for 12 h per day at 110 V. Each day’s collection was considered one air sample, and this process was repeated continuously for 3 days. The collected samples were stored in PBS buffer. (4) Soil Samples: Soil samples were collected using the quincunx sampling method at various spaces (Xiao et al., 2023), including the entrance of the poultry slaughterhouse, the slaughter workshop, and the pre-slaughter caged area. Each sample weighed 5–10 g. (5) Sludge Samples: Sludge samples were collected at the four corners of the sewage discharge pool, with approximately 10 mL of sewage collected per sample. (6) Environmental Swab Samples: Sterile cotton swabs were used to collect environmental samples from the slaughter workshop, pre-slaughter area, caged area, and poultry transportation vehicles. Five swab samples were collected from each space or vehicle, discarding the swab tails and placing them in sterile 0.5% BSA-PBS buffer for preservation (Table 1). After collection, all samples were stored at 4°C, transported to the laboratory in a cooler, and then stored long-term at −80°C in an ultra-low-temperature freezer. This study was approved by the Medical Ethics Committee of the School of Public Health, Sun Yat-sen University (Permit No. [2018] No. 001).

In order to analyze the virus content and types in samples from different spaces (i.e., slaughter area, pre-slaughter area, caged area) and different types (i.e., air, animals, sludge.) within the slaughterhouse, we combined samples of the same type collected from the same space to prepare sample pools: (1) Combined oral and cloacal swabs from 20 ducks in each of the three spaces (caged area, pre-slaughter area, slaughter area) to create a pool (total of 60 ducks). Combined oral and cloacal swabs from 30 chickens in each of the three spaces to create a pool (total of 90 chickens). (2) Combined nasal swab samples from 20 frontline slaughterhouse workers into one pool. (3) Combined air samples collected continuously for 3 days from each sampling point, creating one pool per sampling point. (4) Combined soil samples collected from each space (mixed samples with four or more points) into one pool. (5) Combined sludge samples collected from each sewage discharge pool into one pool. (6) Combined swab samples from seven slaughterhouse process points in the workshop into one pool. (7) Combined swab samples collected from three poultry transportation vehicles into one pool. Sample pool information is provided in Supplementary Table S2.

Virus-like particles (VLPs) were enriched separately based on the different properties of the samples. Approximately 0.4 g of sludge and soil samples were added to about 2–5 volumes of sterile SB buffer (0.2 M NaCl, 50 mM Tris–HCl, 5 mM CaCl2, 5 mM MgCl2, pH 7.5). For air samples and swab samples, they were directly added to 2–5 volumes of sterile SB buffer to dissolve the virus particles. After three cycles of freeze-thawing, the particles were completely resuspended in 10 times the volume of pre-chilled SB buffer. All samples were centrifuged at 1,000, 3,000, 5,000, 8,000, 10,000, and 12,000 × g for 5 min at 4°C using a Sigma 3 K30 centrifuge (Sigma Laborzentrifugen GmbH, Germany), and the supernatant was collected. Subsequently, the supernatant was filtered through 0.22 μm Millipore filters (Burlington, MA) to further remove any cell debris and organelles. The filtrate was transferred to 28% sucrose solution and ultra centrifuged at 300,000 × g for 2 h in a Himac CP 100WX ultracentrifuge (Hitachi, Tokyo, Japan). The supernatant was discarded, and the pellet was re-suspended in 720 μL of water, 90 μL of 10 × DNase I Buffer, and 90 μL of DNase I (1 U/μl) (TAKARA, Japan). The suspension was thoroughly re-suspended, incubated at 37°C with shaking for 60 min, stored overnight at 4°C, and then transferred to a 2 mL centrifuge tube.

Total nucleic acids were extracted using the HP Virus DNA/RNA Kit (R6873; Omega Bio-Tek, Norcross, USA), and carrier RNA was not used during the process to avoid potential interference with sequencing results. The concentration of RNA was quantified using the Qubit^™ dsDNA HS Assay Kit (Q32851) and Qubit™ RNA HS Assay Kit (Q32855) (Thermo Fisher Scientific, Waltham, USA).

Virome research heavily relies on amplification, as the viral biomass in natural samples is often very low. Due to variations in most amplification methods, quantitative studies of viral data present challenges at present (Hosono et al., 2003; Pan et al., 2013). In the current study, uniform genome amplification (WGA) and transcriptome amplification (WTA) were performed using the repi-g Cell WGA and WTA Kit (150,052, Qiagen, Hilden, Germany), based on the multiple displacement amplification (MDA) method (Lasken, 2009; Picher et al., 2016; Stepanauskas et al., 2017; Tithi et al., 2018). Compared to other amplification methods, MDA has several significant advantages. It can replicate fragments up to 70 kb, provides more uniform genome coverage, and has a fidelity 1,000 times higher than Taq polymerase amplification. Most importantly, MDA has the capability to preferentially amplify circular ssDNA genomes, which is crucial for the study of viruses in the Microviridae (Paez et al., 2004; Roux et al., 2016; Wei et al., 2018).

The amplified DNA was quantified using gel electrophoresis and Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA). Ultrasonic random shearing (Covaris M220) was performed to generate fragments with lengths ≤800 bp. Fragment ends were repaired using T4 DNA Polymerase (M4211, Promega, Madison, Wisconsin), Klenow DNA Polymerase (KP810250, Epicentre, Madison, Wisconsin), and T4 Polynucleotide Kinase (EK0031, Thermo Fisher Scientific, Waltham, MA). Fragments in the range of 300–800 bp were collected after electrophoresis. After amplification, the libraries were pooled, and paired-end sequencing of 150 bp, 250 bp, or 300 bp was performed on the Novaseq 6000, HiSeq X ten, and Miseq platforms (Illumina, San Diego, California) (Patch et al., 2018; Ravi et al., 2018; Modi et al., 2021).

All samples underwent virome sequencing, resulting in approximately 700 million raw sequencing reads. High-quality clean reads were generated using Fastp v0.20.0 (Chen et al., 2018) (options: --correction, --trim_poly_g, --trim_poly_x, --overrepresentation_analysis, --trim_front1 = 16, --trim_tail1 = 2, and --length_required = 50). Reads matching the Illumina sequencing adapters were removed (option: –detect_adapter_for_pe). The clean reads in libraries that were in the same assembly group were pooled and assembled using MEGAHIT v1.2.9 (Li et al., 2015) with the default settings. Contigs shorter than 500 bp were discarded. To detect low-abundance contigs, clean reads not mapped to contigs from the initial assembly were reassembled for two additional rounds, and all remaining reads were merged and assembled together. Contigs from all four assembly rounds were pooled and clustered at 97% global average nucleotide identity with at least 90% overlap of the shorter contig using Cd-hit-est v4.8.1 (Li and Godzik, 2006) (options: -aS 0.9 -c 0.97 -G 1 -M 0 -T 0 -g 1), resulting in 169,216 nonredundant contigs.

Diamond (Buchfink et al., 2015), an advanced sequence annotation tool known for its high accuracy and speed, was employed. BLAST searches of the NCBI NR database against non-redundant protein sequences significantly reduce the number of false positive results compared to BLAST searches against virus databases only (Nouri et al., 2018; Yao et al., 2020). However, BLAST accuracy decreases for short fragments and cannot be used for dissimilar sequences (Jiang et al., 2011). Nevertheless, combining multiple viral mining tools such as CheckV (Nayfach et al., 2021) and VirSorter2 (Guo et al., 2021) improves the success rate of predicting unknown viral sequences. Identifying and classifying putative viral sequences is challenging due to a lack of reliable annotations (Handley and Virgin, 2019). Given these challenges, we used Diamond v0.9.24.125 (options: -e 1e-10, --max-target-seqs 50), for annotation against the NCBI NR database.

CheckV (version 0.7.0) and its associated databases (Nayfach et al., 2021) were used to assess the completeness of assigned viral genomes. After removing false-positive contigs that matched more host genes than viral genes, 1,224 nearly complete viral genomes were obtained. Diamond were used to determine the taxonomy of the viral contigs at the family level. Diamond annotations were further processed in MEGAN6 (Huson et al., 2016) using default parameters with two scripts (daa2rma and rma2info) and parsed into taxonomic annotations. A total of 98 viral sequences (Contig IDs are shown in Supplementary Table S1) were identified as complete genomes and annotated as belonging to the Microviridae for further in-depth analysis.

Cenote-Taker 2 (Tisza et al., 2020) was used in the current study to predict open reading frames (ORFs) in 98 virus genomes (use default parameters, add ‘-am True’ at the end of the command line to output genome map files). Cenote-Taker 2 is a virus discovery and annotation tool that can be used on the command line. It utilizes a highly sensitive model of signature viral genes to identify familiar or novel virus sequences from the user’s input of contigs. Cenote-Taker 2 performs better in discovering virus sequences in complex datasets, with lower false positive and false negative rates compared to similar tools (Tisza et al., 2020). The major capsid protein or capsid protein sequence (major capsid protein is preferred if available, otherwise capsid protein is chosen. These proteins are collectively referred to as “Cap”) was selected from the predicted results of each viral sequence. NCBI BLASTP (Altschul et al., 1990, 1997) was used to compare ORF sequences with the NR database, with an Expect threshold (e-value) set to 10⁻⁵. For each ORF in the alignment results, the top ten protein sequences with their complete genomic sequences were downloaded based on identity. Duplicate sequences were removed from all downloaded sequences. SnapGene¹ was utilized to open the Cenote-Taker2 output file for visualizing the genomic structure.

Cap sequences predicted for DSV microvirus were collected, along with top 10 ranked Cap sequences from the aforementioned BLASTP results, and introduced 20 Cap sequences from microviruses that have been definitively classified by the ICTV. A total of 577 sequences were aligned with each other using DIAMOND (Buchfink et al., 2015) (version 0.9.14.115, options: --evalue = 0.00001) to build a matrix of sequence similarities. Based on the Score value output by DIAMOND alignment, Gephi (Bastian et al., 2009) (version 0.9.7, Layout selected as Fruchterman Reingold, Rescale weight = True, with other parameters set to default) was used to construct a clustering network graph [appropriate score values were selected based on clustering effect to form clear clusters as demonstrated in previous studies (Jiang et al., 2023; Zhu et al., 2024)]. The nodes were colored on different sequence sources, hosts, or virus classification results. Furthermore, the Cap sequences used by Kirchberger et al. (2022) were integrated with the above data. The same method was employed to construct a similarity clustering network graph and color it, aiming to compare the network clustering resolution of our research method with that of Kirchberger et al.

We further performed statistical analysis on the genome length and GC content of each cluster. Statistical analysis was performed by one-way ANOVA and Turkey test at a significant level of p < 0.05.

hostG and cherry are software tools specifically designed for predicting bacteriophage hosts, and they are considered superior to existing virus host prediction software (Shang and Sun, 2021; Shang and Sun, 2022; Howell et al., 2024). All complete genome sequences included in the analysis in section 2.7 were subjected to host prediction using hostG (Shang and Sun, 2021) (output results taking genus, genus_score > 0.7) and cherry (Shang and Sun, 2022) (output results taking Top_1_label, Score_1 > 0.7), analyzing the relationship between these viruses and their hosts, as well as the proportion of these hosts.

Cap is a conserved gene of microviruses (Quaiser et al., 2015) with approximately 500 amino acids in length, and is commonly used as a phylogenetic marker for the classification of evolutionary branches or subfamilies within the Microviridae. Multiple sequence alignment was performed using MAFFT (Katoh and Standley, 2013) ambiguous regions were removed using TrimAl (Capella-Gutiérrez et al., 2009), and a maximum likelihood phylogenetic tree based on Cap sequences was constructed using IQtree (Minh et al., 2020) (version 2.1.4). ModelFinder (Kalyaanamoorthy et al., 2017) was set to MFP (for ModelFinder Plus), and 1,000 ultrafast bootstrap replicates were used. The tree was visualized using iTOL (Letunic and Bork, 2016) (version 6.5.2).²

According to the clustering in Figure 1 and cherry host prediction results, DSV-related viral sequences are named, respectively. Taking cluster_1 as an example, if a sequence has host prediction results, it is named based on the host, such as the contig sequences CD-SXS-WGA-1-k141_397009 and CD-SXD-WGA-1-k141_230904 are named Bdellovibrio microvirus C1_1 and Bdellovibrio microvirus C1_2, respectively. Similarly, CD-SXG-WGA-1-k141_33139 and CD-SXG-WGA-1-k141_32996 are named Escherichia microvirus C1_1 and Escherichia microvirus C1_2. If the sequence has no host prediction results, contig sequences like CD-SXD-SXG-WGA-all--k141_113185 and CD-SXD-SXG-WGA-all--k141_328845 are named DSV microvirus C1_1 and DSV microvirus C1_2, and so forth. The original sequence ID and their corresponding names are listed in Supplementary Table S1.

According to the ICTV standards, Microviridae includes two subfamilies (Bullavirinae and Gokushovirinae) and seven described genera (Walker et al., 2021). Among them, the subfamily Bullavirinae has three genera, comprising 14 species. The subfamily Gokushovirinae has four genera, consisting of 8 species. We selected 98 complete DNA viral genomes from the DSV dataset, annotated as Microviridae, with a genome integrity exceeding 90%, for in-depth analysis. All DSV genomes have lengths ranging from 4 to 6 kb, consistent with the genome size of Microviridae (2012). As predicted, these viruses all have Cap with lengths of 450–600 amino acids (AA). According to the Cap similarity clustering network graph (Figure 1), the Microviridae sequences from DSV, along with the related sequences aligned in NR and the Microviridae sequences from ICTV (totaling 577 sequences), roughly cluster into 9 clusters (cluster_1 to 8 and Bullavirinae). Among them, 14 Bullavirinae sequences recognized by ICTV formed a separate cluster (light green). The remaining 6 ICTV- recognized Gokushovirinae sequences are distributed in cluster_1 (C1) and cluster_2 (C2). According to this classification criterion, C1 and C2 should belong to the Gokushovirinae. While the other 6 clusters (cluster_3 to 8) do not contain ICTV sequences, indicating they may represent new subfamilies within the Microviridae.

Based on the similarity of these 9 clusters, we can divide them into 5 major clusters, tentatively referred to as Family_Red, Family_Blue, Family_Green, Family_Yellow, and the independent Bullavirinae cluster. Family_Red contains 6 sequences recognized by the ICTV as Gokushovirinae; therefore, we tentatively equate Family_Red with the Gokushovirinae. If using Bullavirinae as the standard, then the separate cluster represents at least one subfamily level, while a major cluster formed by multiple clusters may represent a higher level of classification. However, using Gokushovirinae as the standard, one subfamily can be distributed in different clusters (Family_Red). As for whether the major clusters of Family_Blue, Family_Green, and Family_Yellow should be regarded as new subfamilies or elevated to new families requires further discussion.

Host prediction was performed on the 98 newly discovered microvirus sequences from this study, along with their associated 459 NR sequences and 20 ICTV sequences, using hostG (Shang and Sun, 2021) and cherry (Shang and Sun, 2022). In the hostG results, only NC_002643.1 from ICTV was accurately predicted to have a host (Bdellovibrio). However, in the cherry results, the majority of hosts were consistent with the ICTV results, indicating that the success rate and accuracy of cherry predictions were higher than hostG. According to the hostG results, the main hosts for DSV were Bdellovibrio and Chlamydia (Figures 2A,B). For NR sequences, the hosts were mainly distributed in the Bdellovibrio, Chlamydia, and Parabacteroides. Although these results align well with the current understanding of microvirus hosts, results from cherry suggest (Figures 2C,D) that the hosts of microviruses may be far more diverse than these three genera.

Although Bdellovibrio bacteriovorus and Escherichia coli are the predominant hosts in the NR-derived virus hosts, respectively. Cherry also predicted hosts such as Caulobacter vibrioides and Bacillus halmapalus, indicating numerous microvirus hosts that have not been previously reported. Among the DSV-derived virus hosts, Bdellovibrio bacteriovorus still dominated, followed by Caulobacter vibrioides and Candidatus Pelagibacter ubique, representing novel hosts. In addition, NR data revealed the presence of human and animal pathogens such as Helicobacter pylori and Enterobacter cancerogenus. The DSV data also identified potential hosts including Vibrio cholerae and Pseudomonas aeruginosa. This suggests that the host range of microviruses within the Microviridae may be extensive, and that there are likely more potential hosts yet to be discovered.

From the perspective of sample types, the highest abundance of microviruses was observed in soil and sludge samples, corresponding to a higher diversity and quantity of their respective hosts (Figure 2E). Bdellovibrio bacteriovorus, as a typical host for microviruses, showed a higher proportion across various samples. Caulobacter vibrioides also exhibited high abundance in sludge, soil, and the slaughterhouse workshop (Figure 2E). While Escherichia coli, Gordonia terrae, and Pseudomonas aeruginosa were predicted only in soil samples, Vibrio cholerae was exclusively found in sludge samples. Other host bacteria were detected across different sample types. This indicates a close relationship between the detection of microviruses and the distribution of their host bacteria, displaying certain characteristics in various sample types.

The genome sizes and GC content of viruses within the same family or genus are usually relatively consistent (Roux et al., 2019; Yuan et al., 2022). Based on the identification of 9 clusters in the previous sections, we further created boxplots illustrating their genome size and GC content (Figure 3). Both genome size and GC% exhibited high consistency within each of the 9 clusters, while significant differences were observed among different clusters. For instance, Bullavirinae showed distinct genome sizes and GC content compared to other groups. Individual scattered black dots outside the boxes in the figure represent sequences from the NR data. It has been reported that there is an association between the GC content of viral genomes and that of the hosts (Ha and Denver, 2018; Yuan et al., 2022). The differences in GC content between clusters may imply host-specific characteristics of each group. This phenomenon may be due to the close evolutionary history of members within each group, which has not led to significant host crossings or changes in genomic features. This similarity will also be observed in the subsequent host analysis of each cluster. These results indicate that the genome characteristics of microviruses from different taxonomic groups exhibit good consistency and indirectly validate the reliability of our classification method based on the similarity clustering network graph.

To better illustrate the diversity of DSV-related microviruses and their evolutionary origins, phylogenetic trees were constructed for each cluster based on the results in Figure 1. The sequences of DSV-related microviruses were classified and named according to the sample source and host type of the viruses (see Materials for reference).

C1 is the cluster with the highest number of members among the eight clusters and exhibits the most diverse range of host sources (Figure 4). Displayed are partial positions of the phylogenetic tree. Some phylogenetic branches have been collapsed, the complete phylogenetic tree is detailed in Supplementary Figure S1. Notably, in addition to typical hosts such as Escherichia coli and Bdellovibrio bacteriovorus, this cluster has hosts that were previously unreported, such as Caulobacter vibrioides, Pseudomonas aeruginosa, and Helicobacter pylori. Caulobacter vibrioides is a Gram-negative oligotrophic bacterium widely distributed in freshwater lakes and streams, serving as an important model organism for studying cell cycle regulation, asymmetric cell division, and cell differentiation (Abraham et al., 1999). Pseudomonas aeruginosa is a common multidrug-resistant pathogen, characterized by its capsule, Gram-negative nature, and aerobic or facultatively anaerobic growth, causing diseases in plants and animals, including humans (Diggle and Whiteley, 2020). Helicobacter pylori is a Gram-negative, flagellated spiral bacterium, classified as a class I carcinogen, responsible for approximately 89% of gastric cancer cases and associated with 5.5% of cancer cases worldwide (Violeta Filip et al., 2018; de Brito et al., 2019; Marghalani et al., 2020). In general, hosts within the same clustering branch are relatively homogeneous. For example, in Figure 4, the hosts in the purple-colored block branch are primarily Bdellovibrio bacteriovorus, while the hosts in the deep blue-colored block branch are mainly Caulobacter vibrioides.

From the sample sources perspective, DSV viruses in this cluster mainly originate from soil (CD-SXG) and sludge (CD-SXD), with a few from swab samples taken in the slaughterhouse workshop environment (CD-SXS) (Supplementary Table S1). In comparison, the sources of NR viruses are more diverse, including animal metagenomes, wastewater metagenomes, human metagenomes, and blackflies. As seen in the genomic structure diagram in Figure 4, members of C1 typically possess signature genes such as Major capsid protein or capsid protein (Cap), and Replication associated protein (Rep). Moreover, the genomes in this cluster often exhibit a sequential arrangement of Cap, Minor capsid protein (MinCP), Scaffold protein (SP), Rep, and DNA binding protein (DBP). However, a few viruses in this cluster have genome organization sequences that deviate from this pattern, such as DSV microvirus C1_4. Furthermore, Replication-associated protein was not predicted in DSV microvirus C1_1 and DSV microvirus C1_3. Coincidentally, these two sequences also lack host prediction results, likely suggesting the novelty of these viral genomes. Overall, sequences with closer phylogenetic relationships tend to exhibit more apparent consistency in host specificity and genomic structure.

C2 is a small viral cluster with a consistent host source, all being Bdellovibrio bacteriovorus, and a highly consistent genomic structure (Figure 5). The viruses in this cluster exhibit a sequential arrangement of Cap, MinCP, SP, DBP, and Rep, with Hypothetical protein (HYP) inserted on either side of DBP. Bdellovibrio microvirus C2_2 and Bdellovibrio microvirus C2_1 are closely related to AZL82867.1 and QJB19506.1, respectively. The genome of AZL82867.1 is derived from Honey bees, while QJB19506.1 originates from wastewater metagenome. This observation suggests the widespread presence of microviruses and their Bdellovibrio bacteriovorus hosts in various environmental settings. Only three sequences in cluster_3 (C3) were predicted to have hosts (Supplementary Figure S2), indicating that this group lacks sufficient host information and is a relatively novel group compared to C1 and C2. The predicted hosts for these three sequences are Azospirillum brasilense (A. brasilense) and Enterobacter cancerogenus (E. cancerogenus). A. brasilense is a microaerophilic nitrogen-fixing bacterium widely present in the rhizosphere worldwide, promoting plant growth (Tarrand et al., 1978; Tien et al., 1979). E. cancerogenus is a significant pathogen commonly found in human clinical specimens such as blood and cerebrospinal fluid. It is not sensitive to penicillin and cephalosporin (Farmer et al., 1985), and exploring bacteriophage targeting such multidrug-resistant pathogens is meaningful for developing phage therapy methods.

In the phylogenetic tree of C1 (Figure 4), the two green boxes belong to the same level of evolutionary branch. The ICTV sequence NC_003438.1 belongs to Spiromicrovirus, located in the branch of the first green box. The four ICTV sequences NC_002194.1, NC_001998.1, NC_002180.1, NC_001741.1 belong to Chlamydiamicrovirus, located in the branch of the second green box. It can be inferred that the branches at this level belong to members of the same genus. The C2 (Figure 5) group follows the same logic, NP_073538.1 and the ICTV sequence NC_002643.1 are in the same level evolutionary branch, likely also belonging to Bdellomicrovirus. According to this criterion, we can further classify a large number of microviruses at the genus level and discover their evolutionary patterns. However, to ensure its accuracy, further investigations may be needed based on their genome structure and host. In conclusion, the DSV-related microviruses dataset and its phylogenetic analysis can further help explore new evolutionary pathways and evolutionary significance.

Cluster_4 (C4) generally exhibits a relatively tidy genome structure (Supplementary Figure S3). It is noteworthy that, despite being in the same cluster, there are significant differences in the host sources between DSV and NR viruses in C4. The majority of NR viruses are derived from wastewater metagenome samples, and their hosts are predominantly Bacillus halmapalus. Bacillus halmapalus, a halophilic bacterium, is a Gram-positive, alkaliphilic, alkalitolerant, facultative anaerobe. It is commonly isolated from soil, and its pathogenicity is not well understood (Nielsen et al., 1995). In DSV, only two viruses have Bacillus halmapalus as their host, and both are derived from sludge samples, aligning with the source of this bacterial species. Unlike NR, the primary hosts for DSV viruses are Candidatus Pelagibacter ubique. Except for Candidatus Pelagibacter microvirus C4_1, which originates from a swab of the transportation vehicle (CD-SXt), the rest are all from sludge samples (CD-SXD) (Supplementary Table S1). Studies suggest that Candidatus Pelagibacter genus may be among the most abundant bacteria globally and play a crucial role in the carbon cycle on Earth. Croceibacter atlanticus belongs to the phylum Bacteroidetes and is a species isolated from the Atlantic Ocean (Cho and Giovannoni, 2003). Croceibacter microvirus C4_1 is the only virus in this cluster derived from swab of duck oral and cloaca (D-SXAO) (Supplementary Table S1), and it is specifically associated with the host Croceibacter atlanticus. This observation once again confirms the conclusion from Figure 2E that the detection of microviruses is closely related to the distribution of their host bacteria and the source of the samples.

Cluster_5 (C5), as shown in Supplementary Figure S4. Although most NR sequences include Cap and Rep, DSV sequences, such as Bdellovibrio microvirus C5_2/4/7, Escherichia microvirus C5_1/2/13, only predict 2 ORFs: capsid protein and MinCP. We have not observed a correlation between this situation and sample sources, indicating a potentially higher novelty and lower conservation of genes in DSV sequences. Cluster_6 (C6) (Supplementary Figure S5), similar to C3 (Supplementary Figure S2), is a smaller cluster without predicted hosts, indicating a need for further research on this cluster. The genomic structure of the cluster_7 (C7) sequences is primarily arranged in the order of Cap, MinCP, SP, Rep, and DBP (Supplementary Figure S6). Croceibacter microvirus C7_1 and Croceibacter microvirus C7_2, two viruses within the same major branch, share Croceibacter atlanticus as their host (previously introduced in C4). This branch is the only one with predicted host results, while other branches lack host predictions. Therefore, C7 is also a potentially interesting virus cluster worthy of in-depth research. Cluster_8 (C8) (Supplementary Figure S7) has two distinct hosts, with Ralstonia solanacearum being the dominant host and Achromobacter xylosoxidans as the second host. Ralstonia solanacearum is considered one of the most important plant pathogens due to its lethal nature, persistence, wide host range, and extensive geographical distribution (Denny, 2006). Achromobacter xylosoxidans belongs to the genus Achromobacter and is commonly found in moist environments, causing diseases such as bacteremia, pneumonia, pharyngitis, and urinary tract infections (Igra-Siegman et al., 1980; Duggan et al., 1996).

To better understand the relationship between the identified microviruses in the poultry slaughterhouse and other reported microviruses in the Microviridae family, we expanded our focus beyond the 577 viral genomes highlighted in this study (Figure 1). To this end, an additional set of 4,077 microvirus Cap sequences (utilizing 4,007 sequences for this study) studied by Kirchberger et al. (2022) were incorporated into our analysis for a more comprehensive clustering analysis. In the study by Kirchberger et al., microviruses were broadly classified into 19 families, corresponding to the 19 color-coded clusters in Figure 6. Upon comparing the clustering results between Figures 1, 6, there is a good overall agreement between the two figures. Specifically, the four families identified in Figure 1 are concentrated within the purple cluster in Figure 6, representing the largest cluster in Family 3, as defined by Kirchberger et al. DSV viral sequences are predominantly distributed within Gokushovirinae A (Kirchberger et al., 2022), Shukshmavirinae (Tikhe and Husseneder, 2017) and Group D (Rosario et al., 2012) of Family 3. Additionally, three scattered sequences are found in Pichovirinae (Roux et al., 2012), Gokushovirinae B and Gokushovirinae C (Kirchberger et al., 2022), indicating that microviruses in the poultry slaughterhouse environment primarily belong to these groups. This result suggests that, although microviruses in the poultry slaughterhouse environment exhibit high diversity and novelty, they may still be relatively underrepresented in the entire microvirus virosphere. Family 3, possibly due to its close association with the human environment, is the largest group within the Microviridae. The expansion of other groups awaits further supplementation with samples from different sources and microbial hosts.

On the other hand, according to the clustering results in Figure 6, the 19 major families delineated by Kirchberger et al. can be further subdivided into approximately 45 smaller clusters. Particularly, within family 3 (the purple clusters), our clustering method can split it into as many as 20 smaller clusters. Specifically, Group D, Shukshmavirinae, Pichovirinae, and Gokushovirinae C each form independent cluster, while Gokushovirinae A and Gokushovirinae B can be further divided into multiple clusters at the same hierarchical level. These clusters are at least equivalent in status to Group D, Shukshmavirinae, and Pichovirinae. Additionally, families of other colors can also be further subdivided into smaller clusters. For example, family 5 identified as Alpavirinae (Krupovic and Forterre, 2011) can be distinctly clustered into 4 different clusters in this study. This suggests that these smaller clusters may represent novel subfamilies or families that require further identification.

According to the cherry (Shang and Sun, 2022) results, the points in Figures 1, 6 are colored coded on host types in Figure 7. As shown in Figure 7A, clusters C2, C4, C5, a portion of C7, C8, and the Bullavirinae cluster exhibit clear host specificity, while the host colors in cluster C1 are highly mixed. Specifically, Bdellovibrio bacteriovorus is mainly the host for C2 and C5, the hosts for C4 are primarily Bacillus halmapalus and Candidatus Pelagibacter ubique, the main host for C8 is Ralstonia solanacearum, and only some sequences in C7 have host results, all of which are associated with Croceibacter atlanticus. C1 includes a significant number of Bdellovibrio bacteriovorus viruses, as well as viruses infecting Escherichia coli, Caulobacter vibrioides, Croceibacter atlanticus, and other bacteria. This may suggest that this group of viruses is more prone to host jumping compared to other viruses group.

Similar to the results in Figure 7A, the host sources of Family_Red members in Figure 7B remain diverse, primarily involving Bdellovibrio bacteriovorus, Caulobacter vibrioides, Croceibacter atlanticus, Escherichia coli. As the number of members increases, Family_Green shows an expanded range of host types, mainly associated with Caulobacter vibrioides, Bdellovibrio bacteriovorus, and Candidatus Pelagibacter ubique. Family_Blue continues to be dominated by Bacillus halmapalus and Candidatus Hamiltonella defensa. Although the number of Family_Yellow members has increased significantly, a majority still lacks predicted hosts. Apart from the four main groups focused on in this study, the host types of most smaller groups are relatively singular, such as Bacillus halmapalus (family4), Escherichia coli (family5), Rhodobacter capsulatus (family7) (Figure 7B). These results suggest variations in host specificity among different viral groups. Moreover, the good correspondence between similarity clustering networks and host prediction results is evident.