Remarkable virus diversity and organized community in lter-feeding oysters varied from those of the ocean virome


 Background:Viruses play critical roles in the marine environment because of their interactions with an extremely broad range of potential hosts. Many studies of viruses in seawater have been published, but viruses that inhabit marine animals have been largely neglected. Oysters are keystone species in coastal ecosystems, but as filter-feeding bivalves with very large roosting numbers and species co-habitation, it is not clear what role they play in marine virus transmission and coastal microbiome regulation.Results:Here we report a Dataset of Oyster Virome (DOV) that contains 728,784 nonredundant viral operational taxonomic unit (vOTU) (≥800 bp) contigs and 3,473 high-quality viral genomes, enabling the first comprehensive overview of both DNA and RNA viral communities in oysters (Crassostrea hongkongensis). We discovered tremendous diversity among novel viruses that inhabit oysters using multiple approaches, including reads recruitment, vOTUs, and high-quality virus genomes. Our results show that these viruses are very different from viruses in the oceans or other habitats. In particular, the high diversity of novel circoviruses that we found in oysters indicates that oysters may be potential hotspots for circoviruses. Notably, the viruses that were enriched in oysters are not random but are well organized communities that can respond to changes in the health state of the host and the external environment at both compositional and functional levels.Conclusions:In this study, we generated a first knowledge landscape of the oyster virome, which has increased the number of known oyster-related viruses by tens of thousands of times. Our results suggest that oysters provide a unique habitat that is different from that of seawater, and highlight the importance of filter-feeding bivalves for marine virus exploration as well as their essential but still invisible roles in regulating marine ecosystems.

Many studies have focused on free viruses in seawater, whereas viruses in marine animals have been largely neglected. Marine animals are teeming with viruses that inhabit the hosts' surfaces, body spaces, and blood (Scanes et al., 2021). Virome of marine animals form connections with their host, which is vital to the interaction of the microbe community both in and outside the host's body. Among these, bivalves, such as oysters and mussels, may play a particularly important role to the coastal and estuarine environment. That is because bivalves are sedentary and thus imposing a stable and lasting ecological effect to a given area. The population characteristics of bivalves, including very large roosting numbers and species co-habitation, provide the perfect setting for viral transmission with the ow of ocean currents. Most importantly, as lter-feeding animals, they can draw up to 5 L of water per hour through their gills and concentrate suspended microbes and particles by factors of a thousand to a hundred thousand times their seawater concentrations (Bedford, 1978;Olalemi et al., 2016). Indeed, the enrichment of human enteric viruses (Newell et al., 2010) and mimiviruses (Andrade et al., 2015) in oyster gill or gut tissues is clearly a lter-feeding effect. However, the circulatory system of bivalves is open because, unlike vertebrates, they have no vessels to hold blood. In bivalves, the blood (hemolymph) ows freely throughout the body cavities, and therefore it is interesting to speculate what the virus communities in bivalves look like. For example, whether bivalves provide a similar environment or a unique habitat for marine viruses and whether bivalves have the potential to spread viruses and regulate coastal microbial communities are important questions that are yet to be answered.
Oysters, in family Ostreidae, which widely distributed in the intertidal zone all over the world, are the most highly produced seafood in the world. China is the largest oyster producer, accounting for 85.3% of the world's total production (FAO, 2019). As keystone species, oysters provide essential bene ts to coastal and estuarine ecosystems, improving water quality and providing a critical habitat for various organisms (Zhang et al., 2012;Powell et al., 2018). Here we report an extensive Dataset of Oyster Virome (DOV) that consists of 54 sequencing libraries from different tissues, sampling sites, and sampling times of Crassostrea hongkongensis, the most farmed species of oyster along the South China coast. We used viral-like particle (VLP) enrichment and targeted ampli cation strategies, and built a knowledge landscape of the oyster virome community, its function, and in uencing factors of both RNA and DNA viruses, which provides a good foundation to address questions about the connections between bivalves and marine viruses.

Oyster sampling
The oyster samples collected in this study span ve years, from June 2014 to July 2019. We divided the samples into nine time batches according to the chronological order. In addition, all the samples were divided into four other groups: four ampli cation groups based on the ampli cation methods (whole genome ampli cation (WGA), whole transcriptome ampli cation (WTA), reverse transcription and WGA (RT-WGA), and double-stranded DNA (dsDNA)); two tissue groups based on tissue origin (mixed tissues and hemolymph of adults); two status groups based on health status (diseased and moribund), and seven Site groups based on sampling sites (BH, HD, LJ, SZ, TS, YJ, and ZH) (Fig. 1E). In total, we constructed 54 sequencing libraries with 35 samples. For more information, see Table S1.
For eight of the nine time batches, the tissues without the gonad from three oysters were mixed into one sample; the seventh batch was the exception. The rst batch, dCh, contained samples of dying adult C. hongkongensis collected from an oyster farming area in Beihai (BH) of Guangxi Province in June 2014. The second batch had two groups, YJd and YJr, and contained samples of healthy adult C. hongkongensis collected in September 2015 from an oyster farming area in Yangjiang (YJ), Guangdong. The downstream ampli cation method for YJd was WGA (to detect mainly DNA virus genomes) and for YJr it was WTA (to detect mainly RNA virus genomes). The third batch had eight groups, LJd, LJr, QZd, QZr, TWd, TWr, ZHd, and ZHr, and contained healthy adult C. hongkongensis collected from oyster farming areas in the Qinzhou area (QZ) of Beihai (BH), Tanwei area (TW) of Huidong (HD), Zhuhai (ZH), and Lianjiang (LJ) of Guangdong Province in November 2015. The fourth batch had two groups, SZd and SZr, and contained healthy adult C. hongkongensis collected from the Shenzhen (SZ) oyster farming area in Guangdong in April 2016. (The letters "d" and "r" indicate WGA and WTA, respectively, in the third and fourth batches.) The fth batch, ML, contained healthy adult C. hongkongensis collected from SZ in May 2016. The downstream ampli cation method for ML was RT-WGA (to detect both DNA and RNA virus genomes). The sixth batch, BH, contained moribund adult C. hongkongensis collected from BH in July 2016 . The seventh batch had nine groups. GX, K1ZY, K2ZY, T2S, T4S, T5S, T6S, T8S, and ZH, and contained adult C. hongkongensis that were separately collected from BH in Guangxi Province, Kaozhouyang (K#ZY) of Huidong (HD), Taishan (T#S), and Zhuhai (ZH) in Guangdong Province in May 2017. K1ZY, K2ZY, and T8S contained healthy adult C. hongkongensis, and the other groups contained moribund adult C. hongkongensis. The method of sampling in this batch was different from the method used for all the other batches. A 1-mL syringe was used to draw hemolymph from the pericardial cavity of oysters, and samples from 5-8 oysters were mixed into one sample. The eighth batch, os, contained adult C. gigas collected in July 2018. The samples in these eight batches were collected and preserved by the South China Sea Fisheries Research Institute (Guangdong, China). The ninth batch had two groups, HSd and HSr, and contained healthy adult C. hongkongensis purchased from the Huangsha Aquatic Product Market in Guangzhou (GZ), Guangdong, in July 2019; their original farming location was ZH. The samples in this batch were collected and preserved by Guangdong Magigene Technology Co., Ltd (Guangzhou, China). All the samples were quickly frozen in liquid nitrogen, temporarily stored during transportation, and placed in an ultra-low temperature freezer at −80°C for long-term storage.

VLP enrichment
All 35 samples were processed to enrich for VLPs as described by Wei et al. (2017) and using the online protocols (dx.doi.org/10.17504/protocols.io.m4yc8xw). First, 500 mg of mixed tissue, or 14-34 mg spat mixture, was dissected and ground to powder in liquid nitrogen. The powder was further homogenized in approximately 2-5 volumes of sterile SB buffer (0.2 M NaCl, 50 mM Tris-HCl, 5 mM CaCl2, 5 mM MgCl2, pH 7.5). After three rounds of freezing and thawing, the pellets were resuspended entirely in 10 volumes of pre-cooled SB buffer. For the hemolymph sample, 10 mL hemolymph was mixed with an equal volume of 2×SB buffer, then directly subjected to three rounds of freezing and thawing. The following steps were the same for the tissue, spat, and hemolymph samples. All the samples were centrifuged at 1,000, 3,000, 5,000, 8,000, 10,000, and 12,000 × g for 5 min each at 4°C using a 3K30 centrifuge (Sigma, Osterode am Harz, Germany), and the supernatants were retained. Cell debris, organelles, and bacterial cells were further removed using a Millex-HV 0.22 μm lter. The ltrates were transferred to ultracentrifuge tubes containing 28% (w/w) sucrose using a syringe. The tubes were transferred to an ice bath for 10 min before centrifugation in a Himac CP 100WX ultracentrifuge (Hitachi, Tokyo, Japan) at 300,000 × g for 2 hr. Supernatants were discarded and the precipitates were fully resuspended in 720 μl of water, 90 μl 10 × DNase I Buffer, 90 μl DNase I (1 U/μl), and incubated at 37°C with shaking for 60 min, followed by storage overnight at 4°C, and transfer to 2-ml centrifuge tubes. To better compare the RNA and DNA virus communities, we used WGA and WTA methods to construct libraries in four batches of mixed tissues, which accounted for 70% (38/54) of all libraries (Table S1). RT-WGA is a modi ed protocol that simultaneously ampli es DNA and RNA (Wei et al., 2018b;Li et al., 2019). In this study, 14 libraries were constructed based on RT-WGA, including hemolymph and mixed tissue samples (Table S1). The steps for the WGA, WTA, and RT-WGA were according to the online protocols (dx.doi.org/10.17504/protocols.io.m5vc866). For WTA, there is a "DNA wipeout" step before reverse transcription that aims to remove DNA altogether, but this step is not part of the WGA and RT-WGA protocols. Compared with WTA and RT-WGA, the WGA protocol skips the reverse transcription reaction to avoid amplifying RNA in the downstream reaction. In addition, two other samples were directly subjected to random shotgun library preparation using a Nextera XT DNA Library Preparation Kit (Illumina) following the standard manufacturer's protocol. Because of the limited data quality and sample number, these two libraries were not included in the following diversity analysis.

Library construction and sequencing
Ampli ed DNA was quanti ed by gel electrophoresis and Nanodrop 2000 spectrophotometer (Thermo Fisher Scienti c) and randomly sheared by ultrasound sonication (Covaris M220) to produce fragments ≤800-bp long. The sticky ends were repaired, and adapters were added using T4 DNA polymerase (M4211, Promega, USA), Klenow DNA Polymerase (KP810250, Epicentre), and T4 polynucleotide kinase (EK0031, Thermo Fisher Scienti c, USA). Fragments of 300-800 bp were collected after electrophoresis. After ampli cation, libraries were pooled and subjected to 150 bp, 250 bp, or 300 bp paired-end sequencing on Novaseq 6000, HiSeq X ten, and Miseq platforms (Illumina, USA). Considering the RT-WGA libraries were likely to have higher virus diversity than the WGA and WTA libraries (Wei et al., 2018a), they were sequenced with higher depth and also produced better assembly results (Table S1).

Virus detection and quanti cation based on de novo assembly (vOTU annotation)
High-quality clean reads were generated using Fastp (version 0.20.0) (Chen et al., 2018), (options: -correction, --trim_poly_g, --trim_poly_x, --overrepresentation_analysis, --trim_front1=16, --trim_tail1=2, and --length_required=50) and reads that matched 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 (version 1.2.9) (Li et al., 2015) with the default settings. Only contigs longer than 800 bp were kept. To detect low abundant contigs, clean reads that did not map back to the rst round of assembled contigs were reassembled for two additional rounds, then all remaining reads were pooled 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 ( Among them, 728,784 (21.77%) of the total contigs were annotated as viral origin (i.e., vOTUs). Among them, 7.68% were Eukaryota, 0.34% were Archeae, 21.59% were bacteria, and 0.82% were unclassi ed cellular organisms, and 47.89% unknown origin (Fig. 1A). FastViromeExplorer was used with default settings to map the clean reads against the vOTU contigs to obtain the vOTUs abundance table.

Viral genome integrity, taxonomy, and auxiliary metabolic genes analysis
The viral genome completeness of assigned contigs was tested using CheckV (version 0.7.0) and its associated database . After removing false positive contigs that matched more host genes than viral genes, 3,473 nearly complete viral genomes were obtained.
Three methods (Diamond, vContact2, and PhaGCN) were used to determine the taxonomy of the viral contigs at the family level. Diamond annotations were further processed using two scripts (daa2rma and rma2info) in MEGAN6 (Huson et al., 2016) with default parameters, and parsed to taxonomy annotations. The advantage of Diamond is that there is no minimum length requirement for query sequences; however, it has three drawbacks: low accuracy, low annotation rates, and inaccurate taxonomy of NCBI. PhaGCN is a novel semi-supervised learning model that combines the strengths of a BLAST-based model and learning-based model using a knowledge graph (Shang et al., 2017). For comparison purposes, only vOTUs that were longer than 10 kb were analyzed using PhaGCN and vContact2 with default parameters.

Viral contamination assessment
The experimental preparation for viromic sequencing involves the use of various reagents, many of which have been proved to carry contaminated viral sequences of unknown origin (Holmes, 2019 genomes were used as queries in the same BLASTN search, but no matches were found. We also used Salmon (v1.5.2) to map all the clean reads in the DOV libraries to the contaminant viral sequences. The mapping rates for most of these libraries were <0.01% (Additional le 2), which is consistent with the BLASTN results.

Viral community and statistical analysis
In this study, the FPKM (Fragments Per Kilobase per Million) value was used to represent the relative abundance of the reference viral genomes, vOTUs, and AMGs. On the basis of the FPKM-transformed abundance table, R and Excel were used to analyze the corresponding viral diversity and community structures. The vegan and ggplots R packages were used to calculate α-diversity indexes and plot the nonmetric multidimensional scaling (NMDS). Analysis of variance (ANOVA) and TukeyHSD were used to test the differences between groups with the signi cance level set at 0.05. For the procrustes analysis, the characteristic axis coordinates of NMDS were extracted as the input of the procrustes function, and the protest function was used to perform the substitution test to evaluate the signi cance of the results. For this study, we collected 35 samples of mixed tissue or hemolymph from Crassostrea hongkongensis at nine time points from seven major oyster farming areas along the coast of South China (Fig. 1, Table  S1). Fifty-four oyster virome libraries were constructed using three primary ampli cation methods (WTA, WGA, and RT-WGA) and sequenced (Table S1). A total of 3,347,421 nonredundant contigs (≥800 bp) were obtained after assembly. Among them, 728,784 (21.77%) were annotated as viral origin by comprehensive blast (Fig. 1A), which we called the DOV. The viral contigs were assembled mainly from the RT-WGA libraries of hemolymph samples with higher sequencing coverages (Fig. 1B). Rarefaction curves (Fig. 1C) show that the sequencing depths were su cient, and the vOTU numbers in the WTA libraries were the lowest among all the libraries.
Notably, the ratio of viral reads (mapping rate) varied a lot depending on the reference databases that were searched (Fig. 1E). The mapping rate of de novo assembled vOTUs (29.81%) was much higher than the mapping rates of the RefSeq (NCBI viral reference genomes) (3.50%) and the RefSeq plus two other public virus datasets (GOV and IMG/VR) (12.06%) (Fig. 1E, Table S1). The higher mapping rates of vOTUs con rmed that the VLPs enrichment protocol was effective (Wei et al., 2018b; Liu et al., 2019), and lterfeeding oysters can e ciently accumulate environmental viruses. The low mapping rates of the reference genomes (3.50% and 12.06%) imply that the viruses found in oysters are largely unknown. To our knowledge, this is the biggest viral metagenomic dataset for any marine animal currently available.  Fig. 2A-C). Considering the primary bias of MDA, circular ssDNA viruses (including Microviridea and Circoviridea) accounted for only 2.23% of all the viruses (Fig. S1), which means their diversity is lower than the diversity of the dsDNA viruses in DOV.

Viruses in Oysters
BLAST-based taxonomy of short contigs has limited accuracy (Jiang et al., 2011), and a large proportion of them (34.88%) could not be assigned at the family level (Fig. S1). In view of this, PhaGCN was used and successfully classi ed 6,362 out of 8,760 large vOTUs (≥10 kb) (Fig. 2B), which exceeded the number classi ed by vContact2 (214/8,760) (Fig. 2C), and the unassigned vOTUs decreased to 11.46% (Fig. 2D). Impressively, the DOV nodes (vOTUs) accounted for 74.58% of the total nodes, whereas the RefSeq nodes account for only 25.42% in the vContact2 network (Fig. 2E), indicating that current knowledge about the ocean virosphere is far from su cient.

Near-complete viral genomes
A total of 3,473 viral contigs with >90% genomic completeness (including 27 RNA viral genomes) were identi ed (Figs. 3, S2, Table S2). The genomes were 1,206-60,277 bp long, and the GC content was 24.74%-65.70% (Fig. S2). The encoded proteins shared only 0%-40% identity with known viral proteins (Fig. 3), which again indicated that most of the genomes represented new viral categories. Only 16 of them clustered with nonredundant reference genomes of CheckV with 95% average nucleotide identity and 70% alignment fraction of contigs. We considered both unknown and unclassi ed sequences (67.1% (2,330) of the total) as representing novel viruses at the family level (Table S2).

Oyster-related circoviruses
Circoviridae have been commonly found in some virome studies, especially those based on the MDA method; however, most of the samples analyzed in these studies were environmental samples. Finding so many circovirus-like genomes in oysters was quite unexpected because all currently known hosts of circoviruses are in clade Bilateria in kingdom Animalia (Virus-Host Database, May 2021). Although this nding suggests that the circoviruses in oysters were mostly hosted by oysters, it is unlikely that other marine microorganisms commonly host circoviruses. Furthermore, we used the circovirus replicase protein sequences recorded by the International Committee on Taxonomy of Viruses (ICTV) as queries, and mined out 1,390 and 8,763 nearly complete circovirus-related replicase protein sequences from NCBI nr and DOV respectively by iterative BlastP searches. Similarity clustering of the identi ed replicase protein sequences (Fig. S3) shows that the circovirus-related sequences are very diverse. With the exception of two standard Circoviridae genera, Circovirus and Cyclovirus, which have been recorded by the ICTV, most of the other clusters contain sequences that have not been clearly classi ed (Fig. S3).
Among them, the sequences from DOV accounted for 86.3% (6.3 times the percentage from NCBI nr) and were widely distributed and present in all the clusters. Some clusters even contained only DOV sequences, which indicates that these sequences may be unique to oysters (Fig. S3).
We also constructed a phylogeny tree (Fig. 4) using the replicase sequences that clustered with the circoviruses and cycloviruses (Fig. S3). The results showed that most of the replicase sequences from DOV were on an independent branch that was separate from the Circovirus and Cyclovirus branches, and distant from the branches of contaminant sequences (excluding the possibility of reagent contamination). We considered that these replicase sequences from DOV represented a new oyster-or bivalve-speci c genus under Circoviridae, and temporarily named it Crasscircovirus (Fig. 4). Five of the DOV sequences were scattered in different Circovirus and Cyclovirus branches (Fig. 4). These ndings suggest that oysters (and possibly bivalves) may be hotspots of circoviruses. Whether these circoviruses are pathogens or live as symbionts in oyster hosts, and whether they will spill-over to other marine animals like coronavirus in bats need further study (Van Brussel and Holmes, 2021).

RNA viruses versus DNA viruses
Most previous virome studies focused only on DNA or RNA viruses. Quantitatively comparing the diversity and abundance among RNA and DNA viruses in real environments will likely be very interesting (Holmes, Firstly, our study shows different ampli cation strategies can e ciently target different genomes, because the vOTUs of RNA viruses in the WTA libraries signi cantly outnumber those in the WGA libraries, and vice versa for the DNA viruses (Fig. S4A, B). Secondly, it seems to be common that the diversity of DNA viruses in nature and public databases is higher than the diversity of RNA viruses Rosario et al., 2018;Levin et al., 2017). Although the differences in α-diversity indexes of WGA and WTA libraries were not signi cant, similar rules were still observed (Fig. S4D-F), which is consistent with previous obversions (Figs. 1C, S1). However, further studies are needed to con rm the conclusion that the diversity of DNA viruses is higher than that of RNA viruses. Furthermore, the extremely high mutation rates of RNA genomes challenged their detection recall of alignment-based annotations Notably, although the diversity of RNA viruses seemed low, their abundance in the WTA libraries was not low, reaching nearly 100% in some libraries (Fig. S5) where no active transcription products of DNA virus were detected. Similar results were not observed in the RT-WGA libraries (Fig. S5). The samples that contained a high proportion of RNA viruses need further investigation to determine which kinds of RNA viruses are dominant in these samples, and to understand why RNA and DNA viruses seem to occupy different niches and have different lifestyles. To compare the activity of RNA and DNA viruses in the same sample more objectively and effectively avoid the in uence from host cells, metatranscriptome analysis may be an appropriate method.

Viral communities
Viral community studies can help determine whether the viruses enriched in oysters can be regarded as an organic whole, similar to viruses in the marine environment, or are simply a random and accidental assembly, as well as whether the community can respond to external in uences. Although MDA introduces bias by prioritizing circular ssDNA genome (Binga et al., 2008), and this may have led to the >80% abundance of circular ssDNA virus in several libraries in this study (Fig. S4). But Parras-Moltó et al. (2018) found that ordination plots based on dissimilarities among vOTU pro les showed perfect overlapping of related ampli ed and unampli ed viromes and strong separation from unrelated viromes, which showed that MDA can be used for community studies.
We rstly evaluated the correlation among various community parameters, including the vOTU counts, the ratio of viral reads, varies diversity indexes, and the quantity and quality of sequencing reads (Fig.   S6). The α-diversities correlated well among three community deciphering approaches (based on the OTU and AMG reference datasets) (Fig. S6), which indicates that the methodologies we used for community analysis veri ed each other. Secondly, as we expected, targeted ampli cation plays a decisive role in the virus community (Fig. 5A), and this was further veri ed by the community results based on reference genomes (Fig. 5B). Besides the ampli cation, the obviously different virus abundance patterns, as shown by the heatmap (Fig. 5C) and the F-value ranks (Fig. 5A), showed prominent differences between tissue groups. Even in an open circulatory system, the virus community in the tissue submerged by hemolymph was quite different from that in the hemolymph itself, which shows that different host tissues had a selective effect on the viruses. Importantly, although the in uence of health status, sampling sites, and sampling times on the whole community did not seem to be signi cant (Fig. 5A, low F-valve), we still found there were signi cant differences in both the α-and β-diversity (NMDS) between all healthy and diseased samples (Fig. S7A, C).
The α-diversity of moribund groups was relatively high, perhaps indicating that the decrease of immunity caused by diseases leads to an increase of opportunistic pathogens and their bacteriophages in the host.
However, the expected differences between moribund and healthy groups were not detected in the parallel cohorts (Fig. S7B, C).
Geographical origin (sampling site) also substantially in uenced the community. Samples from the same location tended to aggregate, and signi cant differences in α-diversity were observed from the WGA and WTA groups separately ( Figure S8). The in uence of the habitat on the microbiome of the host has been . The in uence of site on the viromic community was weaker than that of the time point (Fig. 5A, lower F-value), and this was also re ected in the proportion of unique vOTUs (i.e., those that were detected only in one group) (Fig. S9). The relatively high proportion of unique vOTUs in the time period groups implies that viral communities are dynamic with time, and the low proportion of unique vOTUs indicates that viruses actively communicate among locations. However, because of the limited sample number and the diversity of host species, these results need further veri cation.

Auxiliary metabolic genes (AMGs)
Viruses play essential roles in metabolic regulation in the marine ecosystem (Suttle, 2005;Breitbart et al., 2012;Breitbart et al., 2018). Like marine viruses, a large number (9,091) of AMGs were identi ed from DOV. They were assigned to 12 KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways, and 98 pathways (Table S3) Fig. S10B. Importantly, the AMG community ( Fig.   S10C) showed consistency with the vOTU community (Fig. S10D), and the richness and Shannon index showed positive correlations between the two communities (Figs. S6, S10E, S10F). These ndings indicate that the oyster viromic function was closely related to that of the species community. Although it is di cult to know which of them is the cause and which is the result, this discovery provides clues that can help further understand the ecological function of the virome in oysters.

Conclusions
Here we report a comprehensive viromic dataset (DOV) with high resolution that provides a new resource for studying and understanding the marine virome. We have shown from multiple aspects, including reads recruitment, vOTUs, high-quality virus genomes, and circovirus-related replicases, that oysters undoubtedly harbour a large, diverse, and unique array of viruses, which are very different from viruses in the ocean or any other habitat. Oysters can be considered as repositories and transmission hotspots of marine viruses, which may be caused by its lter-feeding lifestyle and high density of natural population.
Notably, the viral communities in oysters are not random but well organized, and can respond to changes in host tissues and health state, and in the external environment at both compositional and functional levels. Further studies on the viral community structure and function of bivalves will greatly contribute to an understanding of their role in coastal microbiome regulation, in disease transmission, and in protecting and restoring coastal ecosystems.