A comprehensive RNA virome from molluscan transcriptomes reveals extensive diversity and modular genome evolution Authors: Jiaxiang Li, Kewei Chen, Xin Chen, Kaiyang Zheng, Shujuan Sun, Hao Yu, Yue Dong, Ziyue Wang, Yixuan Xu, Andrew McMinn, Yeong Yik Sung, Wen Jye Mok, Li Lian Wong, Yantao Liang, Min Wang Published: 2026 ABSTRACT Mollusca, the second-largest animal phylum, includes many aquaculture species used as important food resources for humans. While DNA viruses that threaten molluscan aquaculture have received much attention, molluscan RNA viromes are still understudied. Here, using a multi-stage RdRP discovery pipeline combining six-frame translation, profile-based homology search, and phylogenetic validation, 203 RNA viruses spanning five viral phyla and fifteen viral orders were identified, based on 223 molluscan metatranscriptomes covering eight classes. Phylogenetic analysis, combined with structural modeling, revealed a significant increase in the number of Pisuviricota-related lineages. Extensive modular evolution in viral genomes was observed, including gene rearrangements and co-evolution of capsid and RdRP genes. Host prediction linked 76% of the RNA viruses to a range of eukaryotes. These findings expand the diversity of RNA viruses associated with molluscs and shed light on their phylogenetic relationships, highlighting that molluscs might serve as an important reservoir of diverse RNA viruses infecting a range of eukaryotic hosts. Introduction Mollusca inhabits a diverse range of environments, ranging from terrestrial to coastal and deep sea. It is the second most diverse animal phylum, with over 100,000 described species (El-Bawab, 2020). Molluscs exhibit high levels of morphological, behavioral, and ecological diversity, which underpin key roles in ecosystems and comprise central positions in food webs. Economically, molluscs account for approximately 27% of global aquaculture production (Kocot et al., 2011). They also serve as important research models for studies of environmental adaptation (Sun et al., 2017), nervous system evolution (Albertin et al., 2015), genomic dynamics (Li et al., 2022), and disease transmission (Young et al., 2022). Viruses are obligate intracellular parasites and represent the most abundant biological entities on Earth (Mushegian, 2020). Advances in genome sequencing and metagenomics have revealed vast DNA-virus diversity (Call et al., 2021). High-throughput RNA sequencing of global metatranscriptomes has expanded known RNA virus diversity five-fold, identifying new phyla and challenging host associations by identifying putative prokaryotic infections by viruses previously thought to only infect eukaryotes (Neri et al., 2022). Metagenomic surveys of invertebrates have found far greater RNA virus diversity than in vertebrates (Shi et al., 2016). Starting from 2017, the analysis of transcriptomic datasets produced from molluscan species has been used to retrieve an increasing abundance of nearly complete genomes of RNA viruses, both focusing on economically relevant marine species (Rosani and Gerdol, 2017; Rosani et al., 2019) or on less studied taxonomic groups like nudibranchs (Rosani, 2022). Likewise, Shi et al. discovered 1445 RNA viral genomes by mining transcriptomes from 220 invertebrate species across nine animal phyla (Shi et al., 2016). However, the research on molluscan RNA viruses has lagged behind research on DNA viruses that threaten aquaculture, such as oyster herpesvirus OsHV-1 (Oyanedel et al., 2023), Papovaviridae-like agents (Renault and Novoa, 2004), and Iridoviridae (Elston and Wilkinson, 1985), which cause mass mortalities. A small number of RNA virus families (Togaviridae, Reoviridae, and Picornaviridae) have been characterized in bivalves (Renault and Novoa, 2004) through histopathology and electron-microscopy analyses. Recently, 18 and 154 RNA viruses were identified from oysters based on metatranscriptomics, and have been classified into five families or orders (Sobelivirales, Picornavirales, Leviviridae, Durnavirales, and Yanvirus) and three phyla Lenarviricota, Pisuviricota, and Kitrinoviricota, respectively (Zhu et al., 2022; Wu et al., 2024). However, the RNA viromes associated with molluscs remain underexplored. To address this shortcoming, this study undertook a large-scale meta-transcriptomic survey across eight molluscan classes (Bivalvia, Solenogastres, Gastropoda, Polyplacophora, Cephalopoda, Caudofoveata, Scaphopoda, and Monoplacophora) to identify RNA viruses. This study specifically focuses on RdRP-encoding RNA viruses (Orthornavirae), which constitute the vast majority of known RNA viral diversity and are defined by the presence of RNA-dependent RNA polymerase as their hallmark replication enzyme. We note that this definition excludes retroviruses (Retroviridae) and other RT-encoding viruses (Pararnavirae), which utilize reverse transcriptase (RT) and replicate via DNA intermediates, as these require distinct detection strategies not captured by RdRP-based pipelines. By mining molluscan viromes and comparing them to public seawater RNA-virus datasets and other known molluscan RNA viruses (Shi et al., 2016; Wolf et al., 2020; Zhu et al., 2022; Wu et al., 2024), this study identified 203 novel RNA virus genomes and generated a comprehensive portrait of the molluscan RNA virosphere. A systematic phylogeny and host-lineage predictions were also applied to these RNA viruses. Together, the results redefine the molluscan RNA virosphere and shed new light on key patterns and processes driving the evolution of molluscan RNA viruses. Results RNA viruses identified from molluscan metatranscriptomes A total of 223 metatranscriptomes, covering a wide range of molluscs (Fig. 1A) and spanning eight classes (Bivalvia, Solenogastres, Gastropoda, Polyplacophora, Cephalopoda, Caudofoveata, Scaphopoda, and Monoplacophora), were collected and used to identify RNA viruses (Fig. 1B). These metatranscriptomes yielded 9,876,760 contigs (≥1 kb) and 12,086,615 predicted proteins. A total of 203 RNA viruses were identified based on our custom pipeline for the identification of RNA viruses (Supplementary Table S1, Fig. S1). Taxonomic annotation via the combination of VITAP, geNomad, and BLASTP against the NCBI nr database assigned these RNA viruses to five viral phyla, i.e. Pisuviricota (n = 122, 60.1%), Kitrinoviricota (n = 33, 16.3%), Lenarviricota (n = 18, 8.9%), Negarnaviricota (n = 15, 7.4%), and Duplornaviricota (n = 15, 7.4%). Positive-sense single-stranded RNA (+ssRNA, n = 173) viruses were the dominant RNA viruses (Supplementary Table S2). In addition, ten RNA viruses could not be classified into known viral families (Fig. 1C). A comparison with the invertebrate molluscan RNA viruses dataset (Shi et al., 2016) showed that both studies recovered members of Pisuviricota and Kitrinoviricota (Supplementary Fig. S2). Compared with earlier studies (Rosani and Gerdol, 2017; Rosani et al., 2019; Zhu et al., 2022; Wu et al., 2024), viruses not only belonging to Negarnaviricota and Lenarviricota but also to identified eight viruses within the order Ghabrivirales of Duplornaviricota from Neomeniomorpha, representing the first report of this viral lineage in molluscan viromes were recovered.Fig. 1The distribution of molluscan metatranscriptomic data is analyzed in this study. A Overview of the samples analyzed in this study. Pie charts show the composition of mollusc species collected at each location. The total area of each pie chart is proportional to the number of individuals collected. Colors represent different mollusc species; only samples with clear geographic information are shown. B phylogenetic trees of some mollusc samples were constructed using maximum likelihood (ML) based on nucleotide sequences of the COI gene. The color scheme matches Fig. 1A. C Distribution of 203 viral contigs by length, taxonomic family. Segments represent the spread between the minimal and maximal lengths of contigs assigned to the taxonomic family rank. Dot size represents the number of contigs within the taxonomic family rank. The distribution of molluscan metatranscriptomic data is analyzed in this study. A Overview of the samples analyzed in this study. Pie charts show the composition of mollusc species collected at each location. The total area of each pie chart is proportional to the number of individuals collected. Colors represent different mollusc species; only samples with clear geographic information are shown. B phylogenetic trees of some mollusc samples were constructed using maximum likelihood (ML) based on nucleotide sequences of the COI gene. The color scheme matches Fig. 1A. C Distribution of 203 viral contigs by length, taxonomic family. Segments represent the spread between the minimal and maximal lengths of contigs assigned to the taxonomic family rank. Dot size represents the number of contigs within the taxonomic family rank. Phylogeny and three-dimensional structural clustering network of RdRP expanded the diversity of molluscan RNA viruses Molluscan RNA viruses displayed broad phylogenetic diversity when classified using the RdRP marker gene. The RdRP tree splits into five major clades—Lenarviricota, Negarnaviricota, Kitrinoviricota, Pisuviricota, and Duplornaviricota (Supplementary Fig. S3). Further subtree analysis revealed that most mollusc-derived viruses identified in this study formed distinct, well supported clades (bootstrap ≥80%) divergent from known viral supergroups (Zhu et al., 2022). Notably, previously undersampled lineages including Picorna, Toli, Martelli, Hepeli, Amarillo, and Ghabri exhibited greatly expanded clusters, especially Picorna, strongly indicating that the phylogenetic breadth of molluscan RNA viruses has been severely underestimated (Rosani and Gerdol, 2017; Rosani et al., 2019; Zhu et al., 2022; Wu et al., 2024). Three previously unreported candidate order-level viral lineages (SG01, SG02, SG03) were identified in the phylogenetic tree. which further enrich the phylogenetic diversity of molluscan RNA viruses and provide new evolutionary units for RNA virus taxonomy. In total, 19 viral clades were defined, encompassing 5 phyla, 15 orders, and 3 candidate order-level lineages (Fig. 2); these results collectively illustrate the taxonomic composition and evolutionary potential of RNA viruses in molluscs.Fig. 2Phylogenetic diversity of RNA viral supergroups. Each phylogenetic tree was estimated using a maximum likelihood method based on an amino acid alignment of the RdRP domain. Newly identified viruses are marked in yellow, and viral RdRP sequences from published metagenomic studies are marked in blue. The name of each supergroup is shown at the top of each phylogeny. All trees are midpoint-rooted for clarity only, and the scale bar indicates 1 aa substitutions per site. Phylogenetic diversity of RNA viral supergroups. Each phylogenetic tree was estimated using a maximum likelihood method based on an amino acid alignment of the RdRP domain. Newly identified viruses are marked in yellow, and viral RdRP sequences from published metagenomic studies are marked in blue. The name of each supergroup is shown at the top of each phylogeny. All trees are midpoint-rooted for clarity only, and the scale bar indicates 1 aa substitutions per site. The RdRP gene plays a crucial role in the replication and transcription processes of RNA viruses (Wu et al., 2023; Du, 2024). Amino acid residues involved in key processes, such as catalysis and nucleotide selection, are often highly conserved (Neri et al., 2022). To verify the reliability of the phylogenetic analyses at the molecular structural level, three-dimensional structure prediction and structure-based similarity clustering were performed on the sample RdRp sequences, reference RdRp sequences, and RT sequences. After modeling with ESMFold, pairwise structural comparisons were conducted using FoldSeek (e-value ≤ 1e-5), and the top 100 sequences ranked by bitscore from the structural alignments were selected for visualization. Results showed that RdRp sequences from each viral order generally formed distinct structural clusters (Fig. 3), reflecting the structural conservation of RdRp proteins at the order level. However, a small number of viral orders could not be fully separated in three-dimensional space, which may be attributed to convergent evolution or sequence similarity in structural features between some members of these orders and those of other orders. These structural clustering results were highly consistent with our phylogenetic tree (Supplementary Fig. S3), confirming the scientific validity and reliability of the phylogenetic relationships of molluscan RNA viruses at the molecular structural level, and providing dual lines of evidence to support viral taxonomic annotation.Fig. 3RdRP-based structure network analyses of molluscan RNA viral contigs. Three-dimensional structure similarity network of RdRP and Reverse Transcriptase protein domain structures. Each node represents a different structure. For clearer visualization, the top 100 best hits associated with each node are selected as significant nodes, with the structure bit as the weight. Different colored nodes represent the commented virus order. RdRP-based structure network analyses of molluscan RNA viral contigs. Three-dimensional structure similarity network of RdRP and Reverse Transcriptase protein domain structures. Each node represents a different structure. For clearer visualization, the top 100 best hits associated with each node are selected as significant nodes, with the structure bit as the weight. Different colored nodes represent the commented virus order. Exploration of potential hosts of RNA viral contigs Viruses significantly influence a wide range of microbial-mediated processes through their interactions with host organisms. To identify the hosts of RNA viruses in the mollusc samples, three approaches were employed (Chen et al., 2024): (i) CRISPR spacer matching, which identifies prokaryotic hosts by detecting matches between viral contigs and CRISPR arrays in bacterial and archaeal genomes; (ii) taxonomic inference from known virus-host pairs, which assigns putative hosts by cross-referencing viral taxonomic assignments against established International Committee on Taxonomy of Viruses (ICTV) virus-host records; and (iii) machine-learning prediction using RNAVirHost, which integrates sequence homology and genome features to classify viral host taxa. The combination of these methods allowed for comprehensive host prediction across diverse taxonomic groups: CRISPR-based analysis primarily identified potential bacterial hosts, while taxonomic inference and RNAVirHost predominantly linked viruses to eukaryotic lineages. Overall, 76% of the 203 identified RNA viruses were associated with potential hosts (Supplementary Table S3). Host classification results showed that the potential hosts of mollusc-associated RNA viruses were predominantly eukaryotes, including invertebrates, vertebrates, and protists (Fig. 4). Combined with host association data for specific viral families, their distribution patterns are as follows: Invertebrates represented the dominant host group, accounting for 72% of all associated hosts, encompassing arthropods, molluscs themselves, and other invertebrates. Vertebrates comprised 21%, including teleost fish, mammals, and birds, suggesting the potential for cross-taxon transmission among some of these viruses.Fig. 4The connectivity between molluscan RNA viral contigs and potential hosts. The Sankey plot illustrates the data sources of RNA viruses and shows host predictions based on three host prediction methods. The connectivity between molluscan RNA viral contigs and potential hosts. The Sankey plot illustrates the data sources of RNA viruses and shows host predictions based on three host prediction methods. Patterns of RNA virus genome evolution and protein features Based on the analysis of viral genomes from key clades, extensive modularity, including segment fusions and protein rearrangements, was observed (Fig. 5A). RdRP-encoding segments ranged broadly in length but clustered around 1200 nt. Whereas Tombusviridae, Virgaviridae, Picornaviridae, and Marnaviridae typically separate the capsid protein (CP) and RdRP on different segments, instances where both genes co-occur on the same segment were identified here. Several examples of non-homologous replacements of structural modules were also detected: one unknown-family RNA virus contig carried a CP similar to Marnaviridae and Dicistro_VP4, while another swap replaced a CP with an IgA-protease-like domain, eliminating the CP entirely. In Picornaviridae RNA viruses, distinct CP variants nonetheless shared high 3D-structural similarity (Supplementary Fig. S4A) and conserved β-sheet and loop motifs (Supplementary Fig. S4B), supporting a common fold despite sequence divergence. Single-jelly-roll capsid proteins (SJR CPs) are often found in basal lineages, so these early viruses likely encoded only an RdRP and an SJR CP. Most eukaryotic RNA viruses probably evolved from these simple ancestors (Koonin et al., 2020).Fig. 5Genomic features and phylogenetic comparison of molluscan RNA viral contigs. A Homologous domains are shown as boxes of the same color (see key on the right). Domains not commonly found in RNA viral contigs are shown in red and are labeled above the corresponding boxes. Virus taxa and contig identifiers are noted to the left of each virus genome. At the bottom, a scale indicating the length in nucleotides. B Comparison of the topologies of RdRp- (left) and capsid protein-based (right) phylogenetic trees of picornaviruses identified in this study. Trees were inferred with IQ-TREE v2.4.0 using ModelFinder (MF) and 3000 ultrafast bootstrap replicates; bootstrap values > 70 are shown. Tip labels are color-coded by data source. Genomic features and phylogenetic comparison of molluscan RNA viral contigs. A Homologous domains are shown as boxes of the same color (see key on the right). Domains not commonly found in RNA viral contigs are shown in red and are labeled above the corresponding boxes. Virus taxa and contig identifiers are noted to the left of each virus genome. At the bottom, a scale indicating the length in nucleotides. B Comparison of the topologies of RdRp- (left) and capsid protein-based (right) phylogenetic trees of picornaviruses identified in this study. Trees were inferred with IQ-TREE v2.4.0 using ModelFinder (MF) and 3000 ultrafast bootstrap replicates; bootstrap values > 70 are shown. Tip labels are color-coded by data source. Picornavirales are the most abundant RNA viruses in coastal waters (Culley et al., 2006), and several mollusc-associated members were identified in this study (Fig. 5B). Most were closely clustered with Beihai picorna-like virus, Wenzhou picorna-like virus, and Bivalve RNA virus G5, whereas one picorna-like virus (TRINITY_DN175790_c0_g1_i1_len_3060) was markedly divergent. Comparison of RdRP and capsid-based phylogenies of picornaviruses revealed no evidence of recombination among major clades, yet significant topological incongruence between the RdRP and capsid trees was observed in several branches, suggesting that their capsids may have been lost and subsequently reacquired from other sources. To characterize RNA virus protein features in molluscs, RdRP sequences from the Pisoviricota evolutionary branch (Supplementary Fig. S5) were analyzed. Comparative analysis of physicochemical properties, including hydrophobicity, polarity, mutability, transmembrane propensity, and refractivity, revealed distinct patterns between mollusc derived RNA viruses, LucaProt, and RVMT (Supplementary Table S4, Fig. 6). Notably, molluscan RdRPs exhibited statistically significant differences in refractivity and mutability compared to both LucaProt and RVMT (Student’s t-test, P < 0.05). In contrast, no significant divergence was observed in polarity or transmembrane propensity across these groups.Fig. 6Comparison of physicochemical property indices of proteins from molluscan RNA viral contigs with those from LucaProt and RVMT. A The comparisons of refractivity of RdRPs proteins among different habitats. B The comparisons of the polarity of RdRPs proteins among different habitats. C The comparisons of the hydrophobicity of RdRPs proteins among different habitats. D The comparisons of the relative mutability of RdRPs proteins among different habitats. E The comparisons of the transmembrane tendency of RdRPs proteins among different habitats. Comparison of physicochemical property indices of proteins from molluscan RNA viral contigs with those from LucaProt and RVMT. A The comparisons of refractivity of RdRPs proteins among different habitats. B The comparisons of the polarity of RdRPs proteins among different habitats. C The comparisons of the hydrophobicity of RdRPs proteins among different habitats. D The comparisons of the relative mutability of RdRPs proteins among different habitats. E The comparisons of the transmembrane tendency of RdRPs proteins among different habitats. Relative abundance and activity of RNA viruses The relative abundance of RNA viruses exhibits considerable variability (Supplementary Fig. S6). The log-transformed TPM values indicated that 61 viruses exceeded the threshold of five. This indicates that the reads corresponding to these viruses comprise at least 10% of the total aligned reads. The 61 viral contigs were assigned to 15 families and were dominated by Picornaviridae, Steitzviridae, Orthomyxoviridae, Dicistroviridae, Marnaviridae, and Phenuiviridae (Supplementary Table S5). Picorna-like RNA viruses have previously been reported to infect molluscs (Carella et al., 2023), and a candidate calicivirus designated Congeria kusceri calicivirus 1 (CkCV1; contig TRINITY_DN175790_c0_g1_i1) was identified in the freshwater bivalve Congeria kusceri, a lineage that also includes major human gastrointestinal pathogens (Glass et al., 2009; Zhuo et al., 2021), providing additional support that C. kusceri may serve as a potential calicivirus host. However, this contig accounted for less than 0.6% of total reads, indicating limited environmental distribution or low persistence in C. kusceri relative to the highest-abundance viruses. Metatranscriptomes offer a unique opportunity to further understand which viruses are active. Strand-specific mapping information was leveraged to identify actively replicating viruses on the basis of detecting both coding and non-coding genome strands. This enabled a comparative analysis of activity levels for RNA viruses in molluscs. RNA-virus activity was detected across the surveyed molluscs (Supplementary Fig. S7). Picornaviridae were active in Congeria kusceri, Tonicella lineata, Pholidoskepia, Proneomenia sp., Yoldia eightsii, and Chrysomallon squamiferum. Marnaviridae were active in Congeria kusceri, Tonicella lineata, and Pholidoskepia (Supplementary Table S6). In strand-specific datasets, concurrent detection of coding and non-coding strands further supported active replication. Collectively, picornaviruses appear widespread and frequently active across molluscs, while marnaviruses detected in benthic grazers most likely reflect algal infections present in the diet or biofilms. Discussion This study examines the diversity and evolutionary features of RNA viruses through an in-depth analysis of 223 molluscan metatranscriptomes, spanning eight classes, significantly expanding the known RNA viral diversity of molluscs. Phylogenetic reconstruction showed that most new viruses formed distinct clades, divergent from established taxa (Fig. 2) (Shi et al., 2016), suggesting that molluscs may serve as a reservoir for RNA viruses. Three-dimensional structural clustering of RdRPs corroborated phylogenetic relationships, demonstrating conservation at the viral order level (Fig. 3), though structural overlaps in minor clades may reflect evolutionary convergence or horizontal gene transfer. Comparison with earlier studies underscores the novelty of our dataset (Supplementary Fig. S2). While both our work and previous invertebrate surveys recovered Pisuviricota and Kitrinoviricota, our larger-scale analysis across 223 metatranscriptomes identified 203 high-confidence RNA viruses, providing a standardized baseline for cross-library comparisons. Notably, eight Duplornaviricota (Ghabrivirales) lineages from Proneomenia, that has not been documented in molluscs, were detected and a strongly skewed composition dominated by Pisuviricota was observed. These additions extend both the breadth and depth of molluscan RNA viromes and provide a new phylum-level dimension that was missing from previous records. Modular evolution of RNA viral genomes was prominent, particularly in the organization of capsid protein and RdRP coding regions (Neri et al., 2022). Instances of CP and RdRP co-encoded on a single genomic segment were observed in families such as Tombusviridae (Fig. 5A), alongside non-homologous structural gene displacement, supporting the hypothesis of modular recombination as a driver of host adaptation (Neri et al., 2022). For example, while CP sequences in Picornaviridae exhibited variability, their three-dimensional structures remained highly conserved (Supplementary Fig. S4A), indicating functional constraints despite sequence plasticity. These observations align with the SJR CP origin hypothesis (Koonin et al., 2020), which proposes that eukaryotic RNA viruses likely arose from ancestors encoding RdRP and simplified CPs. Viruses significantly influence a wide range of microbial-mediated processes through their interactions with host organisms. From an ecological perspective, host prediction linked 76% of RNA viruses to eukaryotic organisms (e.g., Vertebrata, and Invertebrata, Fig. 4), potentially due to RNA viruses primarily targeting eukaryotic hosts (Wu et al., 2022). RNA viruses that infect protists have been found in Congeria kusceri, Meretrix petechialis, Panopea generosa, Perna viridis, Pinctada maxima, Reishia clavigera, and Tonicella lineata. Plant-infecting RNA viruses have been detected in Mimachlamys nobilis, Tridacna squamosa, and Euprymna berryi. This suggests that molluscs may act as ecological bridges for viral transmission by filtering and accumulating viruses from the environment. In the future, integrating cross-host metagenomics spanning diverse molluscs, in vitro assays, and single-cell technologies will verify mollusc-virus interactions and their ecological and evolutionary implications. Notably, the identification of these diverse RNA viruses, including picorna-like and calicivirus-related lineages, suggests that they could serve as potential indicators of mollusc health or environmental stressors. Given that some detected viruses share phylogenetic proximity with known pathogens, their presence may pose uncharacterized risks to aquaculture biosecurity, necessitating future monitoring of viral loads during mollusc mass mortality events. Conclusions This study reports 203 RNA viruses from 223 molluscan metatranscriptomes spanning eight classes, expanding the known RNA virosphere of Mollusca across five phyla and fifteen orders. Phylogenetic and structural analyses of RdRPs confirm that molluscs harbor extensive, previously underestimated viral diversity, particularly within Pisuviricota. Extensive modular genome evolution, including gene rearrangements and capsid-RdRP co-evolution, highlights the evolutionary plasticity of these viruses. Host prediction links 76% of identified viruses to diverse eukaryotes, indicating that molluscs serve as important ecological reservoirs and potential vectors for cross-taxon transmission. These findings establish a genomic baseline for molluscan RNA virology and underscore the need for continued surveillance to inform both aquaculture biosecurity and broader understanding of RNA virus ecology and evolution. Materials and methods Metatranscriptome data acquisition The pipeline for the identification of RNA viruses leveraged 223 publicly available, de novo assembled metatranscriptomes deposited in MolluscDB (Liu et al., 2025), spanning eight molluscan classes and diverse geographic locations (Fig. 1A). The complete workflow for RNA virus identification is illustrated in Supplementary Fig. S1. For detailed information, refer to the Supplemental tables. (If available, refer to the study where the sample was originally published.) Preliminary filtering process To identify RNA virus–derived sequences from molluscan transcriptomes, a multi-stage RdRP discovery pipeline was implemented following the framework of Wolf et al. (2020), beginning with a six-frame end-to-end translation of all contigs. Each nucleotide contig was translated in all six frames without internal stop filtering and the most coding-compatible translation was retained as the representative amino acid sequence. Previously published multiple sequence alignments of viral RdRPs and reverse transcriptases were formatted as HMMER v3.4 and sequence-based subject databases [the Uri Neri et al. Dataset (Neri et al., 2022), global marine RNA data from Zayed et al. (2022), and aquatic virion profiles from Wolf et al. (2020)], and used as queries against the six-frame translation database using hmmsearch, PSI-BLAST (v2.16.0+), DIAMOND (v2.1.10), and MMseqs2 (v16.747c6). High-stringency criteria (profile coverage ≥ 50%, e-value ≤ 1e−10, score ≥ 70) were applied, calibrated using a true-negative set constructed from non-viral palm-like proteins retrieved via HHsearch against PDB70 and collapsed using MMseqs2 (v16.747c6). Candidate hits that passed these thresholds were subjected to Palmscan (v2.0) to validate the presence, order, and coordinates of conserved polymerase motifs A–C–D, yielding high-confidence palmprints. For each validated hit, the complete RdRP open reading frame was reconstructed by locating the nearest upstream and downstream in-frame stop codons within its six-frame translation, producing accurate RdRP ORFs rather than full-frame translations. Corresponding nucleotide contigs were retrieved from the transcriptome assembly. To obtain a non-redundant RdRP set, palmprint regions were clustered using CD-HIT (v4.8.1) (90% identity, ≥ 75% coverage) (Fu et al., 2012), generating core representative sequences. These high-confidence RdRP ORFs and their nucleotide contigs constitute the curated set of RNA virus–derived sequences and were used for downstream phylogenetic and evolutionary analyses (Supplementary Table S1). Contig set augmentation with published genomes To benchmark the newly predicted viral genomes against the known diversity of RNA viruses, a “reference set” of published RdRP-encoding sequences was used. This collection includes all RdRP hits in NCBI’s NT database (NT; accessed on February 28, 2025) to ensure comprehensive coverage of viral RdRP diversity at the nucleotide level, plus non-redundant genomes from major RNA virus surveys: the Leviviricetes catalog of Callanan et al. (2020), the Yangshan assemblage (marine RNA viruses from the Yangshan aquaculture area, China; Wolf et al., 2020), Plastroviruses described by Lauber et al. (2019), the RNA Viruses in Metatranscriptomes (RVMT) database from Neri et al. (2022), LucaProt from Hou et al. (2024), Oysters RNA data from Wu et al. and Jiang et al. (Zhu et al., 2022; Wu et al., 2024) and ocean-derived RdRPs from the Tara Oceans gene atlases (Salazar et al., 2019). These sequences, together with the novel contigs for downstream phylogenetic and domain analyses, were merged and coherently labeled. All reference and newly generated data are publicly available without restriction. The newly generated dataset is named the Mollusca RNA Viruses (MRV). Taxonomic annotation of RNA virus Virus taxonomic assignment used three complementary approaches: (1) the Viral Taxonomic Assignment Pipeline (VITAP v1.7) with default settings (Zheng et al., 2025), (2) geNomad (v1.9) with default settings (Camargo et al., 2024), and (3) BLASTp of our RdRP sequences against NCBI’s non-redundant protein database (NR; accessed on March 15, 2025) for protein-level homology-based taxonomic inference. The results from all three methods were then reconciled to derive consensus taxonomic annotations for the mollusc-associated viruses. Phylogenetic reconstruction The phylogenetic placement of mollusc-associated RNA viruses against the “reference set” was inferred as follows. First, an environmental RNA virus database under the RCR90 (Representative Cluster at 90% amino acid identity) standard (≥ 90% pairwise identity, ≥ 30% coverage) was screened by BLASTx (e-value 1e–5) to recruit additional RdRP homologs (Buchfink et al., 2015). Mollusc RdRPs and their reference matches were clustered at 30% identity via MMseqs2 (v16.747c6), then subjected to all-vs-all BLASTp (e-value 1e–5) to generate similarity scores. MCL clustering (v.22.282, with an inflation value of 3) (Enright et al., 2002) was applied to these scores, yielding 60 protein clusters; after removing clusters lacking mollusc RdRPs or containing reverse transcriptases, each cluster was aligned with MUSCLE (v5.3). For each alignment, HMM profiles were constructed using HH-suite (v3.0.0. beta3)’s hhmake (Steinegger et al., 2019), performed HMM–HMM comparisons with hhalign, and converted scores to distances (dAB = –ln[SAB/min(SAA, SBB)]) to produce a 29 × 29 distance matrix. A maximum-linkage dendrogram (R hclust) was cut at height 1.5 to define 15 subtrees. Each subtree was realigned with MUSCLE5’s “-super5” option (four guide-tree permutations, three perturbation seeds), then trimmed for ≤ 20% gaps with trimAl (v1.5) (Capella-Gutiérrez et al., 2009). Finally, all processed RdRP alignments were concatenated to infer an approximate maximum-likelihood tree in IQ-TREE (v3.0.1) (Nguyen et al., 2014) (1000 bootstraps), with ModelFinder Plus selecting the best-fit substitution model (PFAM + F + R9). Three-dimensional structure network analysis Representative RdRP structures from each phylogenetic clade were modeled using ESMFold (v1.0) (Lin et al., 2023). Pairwise structural alignments were performed with FoldSeek (v10.941cd33) (e-value ≤ 1e–5) (van Kempen et al., 2024), including reverse-transcriptase models as outgroups. Alignment bitscores were used to build a 3D-similarity network in Gephi (v0.1.0) (Bastian et al., 2009) with the ForceAtlas layout. To simplify visualization, the top 100 virus-associated RdRPs by bitscore plus the outgroups were retained. Inference of virus-host associations RNA virus hosts were predicted by applying three complementary methods: (i) CRISPR Spacer Matching. Viral genomes were queried against CRISPR spacers predicted from bacterial and archaeal genomes using BLASTn v2.11.0 (-dust no -word_size 7) (Neri et al., 2022). Arrays with ≥ 3 spacers, each matching a viral sequence with ≤ 1 mismatch and spaced by ≥ 20 bp, were retained. Matched repeat sequences were then blasted against IMG/M bacterial genomes (perc_identity ≥ 90, -dust no, -word_size 7) to confirm host candidates and verify adjacent Cas genes indicative of functional CRISPR-Cas systems. (ii) Taxonomic Inference from Known Virus-Host Pairs. Hosts for annotated RNA viruses were inferred from cross-referencing their VITAP, geNomad, and BLASTn taxonomic assignments against established RNA virus taxa (Zayed et al., 2022). Putative host lineages were assigned via a Last Common Ancestor approach and validated using ICTV’s curated virus-host records. (iii) Machine-Learning Prediction. RNAVirHost v1.0.5 (Chen et al., 2024), which integrates sequence homology and genome features through a trained classifier to predict viral host taxa, was then applied. Comparison of physicochemical property indices of RdRP in RNA viral contigs from different environments RdRPs from LucaProt, RVMT, and the mollusc dataset were analyzed for five physicochemical properties: hydrophobicity, polarity, relative mutability, transmembrane tendency, and refractivity using the ProtScale server (https://web.expasy.org/protscale/). For each property, scores were computed along the primary sequence with a nine–residue sliding window. Positional fluctuations in each index are presented as line graphs, and comparisons of the full set of physicochemical profiles across the three RdRP groups are summarized in violin plots generated in GraphPad Prism 8. Estimation of the relative abundance of RNA viruses Validation of viral contigs. To evaluate the prevalence of RNA viruses identified in molluscs, quality-trimmed reads (processed using BBDuk v37.62 to remove adapter sequences, trim low-quality bases, and filter reads with average quality score < 3 or containing ≥ 4 ambiguous ‘N’ bases, following the DOE JGI Metagenome Workflow) (Clum et al., 2021) were mapped to RdRP-bearing contigs using Bowtie2 (v2.5.4) (Langmead and Salzberg, 2012) with the “very sensitive” option (–local -D 20 -R 3 -N 0 -L 16 -i S, 1,0.50 -I 0 -X 2000 –non-deterministic). To minimize potential bias in abundance estimation caused by overrepresented polyA stretches, both the reads and viral contigs were trimmed at polyA or polyT ends using BBDuk (v37.62) (https://jgi.doe.gov/data-and-tools/bbtools/), without iterative trimming, as the second round removed no additional polyA/polyT reads. Mapped reads were converted into sorted BAM files using SAMtools (v1.21) (Li et al., 2009), which were then used to calculate overall genomic coverage and abundance. Read mappings were subsequently filtered at thresholds of ≥ 90% nucleotide identity and ≥ 75% read length coverage using CoverM (v0.7.0) (https://github.com/wwood/CoverM). Transcripts per million (TPM) were calculated with the “-m tpm” option in CoverM to provide an estimate of relative abundance. While the conventional approach of identifying expressed genes by comparing metagenomic and metatranscriptomic signals does not apply to RNA viruses due to their RNA-based genomes, the activity of the RNA viruses is assessed solely based on metatranscriptomic coverage and depth thresholds. Specifically, RNA viruses with ≥ 50% of the genome covered at 1 × or more with a median coverage depth > 0 on the coding strand were classified as actively transcribing, as defined by the Coclet et al. (2023) transcriptional activity criteria. Statistical analysis All statistical analyses were carried out in R (v4.3.2). Alpha- and beta-diversity metrics, rarefaction curves, and scatter plots were generated with ggplot2 (v3.4.4) and ape (v5.7.1). Violin plots were drawn in Python (v3.9) using seaborn (v0.13.2) and matplotlib. pyplot (v3.10.8). Protein–structure bitscores were used to build a 3D-similarity network in Gephi via the ForceAtlas layout. Phylogenies were embellished in iTOL (v6) (Letunic and Bork, 2024), and virus–host Sankey diagrams were created with SankeyMATIC (https://sankeymatic.com). Data availability All the data, code, and results produced in the course of this project are available at https://github.com/lijiaxiang111/mollusc-rna-viruses/tree/main and Sciecnce Data Bank: https://doi.org/10.57760/sciencedb.v.00026. All data and metadata for these sequence records are publicly available and can be accessed via the NCBI (US National Center for Biotechnology Information) at www.ncbi.nlm.nih.gov. For specific sequence IDs and result files, please refer to the supplementary tables. Ethics statement This article does not involve any research that utilizes animal subjects or human participants. Author contributions Jiaxiang Li: Methodology, visualization, writing-original draft. Kewei Chen: Investigation, writing-review & editing. Xin Chen: Software, validation, visualization. Kaiyang Zheng: Software, validation, visualization. Shujuan Sun: Formal analysis. Hao Yu: Formal analysis. Yue Dong: Methodology, software. Yixuan Xu: Methodology, software. Ziyue Wang: Methodology, software. Andrew McMinn: Conceptualization, funding acquisition, project administration, supervision, writing-review & editing. Yeong Yik Sung: Writing-review & editing. Wen Jye Mok: Writing-review & editing. Li Lian Wong: Writing-review & editing. Yantao Liang: Conceptualization, funding acquisition, project administration, supervision, writing-review & editing. Min Wang: Conceptualization, funding acquisition, project administration, supervision, writing-review & editing. Conflict of interest The authors declare that they have no competing financial, personal, or professional interests that could have influenced the research described in this manuscript.