The Origins and Evolution of the RNA Virome in Land Flora

Abstract

The colonization of land by plants was a major transition in the history of life, fundamentally reshaping the evolution of terrestrial organisms and Earth’s ecosystems. Plants are hosts to an enormous diversity of viruses, but the origins and evolution of these viruses are not thoroughly understood. We mined the transcriptomes and genomes of 2,402 species sampled across the plant kingdom and discovered about 35,000 plant-associated RNA viruses, including several potential families not described previously, resulting in a 2.2-fold increase in the known plant-associated RNA virus diversity, greatly expanding the virus host ranges, and revealing many groups of atypical, apparently persistent viruses. Phylogenetic analysis shows that plant viruses emerged on numerous independent occasions, mostly, via horizontal virus transfer facilitated by close ecological associations of plants with other terrestrial organisms, such as invertebrates and fungi.

SHORT SUMMARY

By mining transcriptomes and genomes across 2,402 species, this study discovered ~35,000 plant-associated RNA viruses, substantially expanding the known plant virosphere. Phylogenomic analyses reveal that plant viruses emerged on numerous independent occasions, primarily driven by horizontal virus transfer from invertebrates and fungi within the holobiont.

INTRODUCTION

The origin of land plants (Embryophyta) represents one of the major evolutionary transitions in the history of life (Delwiche and Cooper, 2015; Kenrick and Crane, 1997; Morris et al., 2018). This Paleozoic “greening” of terrestrial environments fundamentally changed the Earth’s atmosphere, climate, landscapes, and global ecology (Delwiche and Cooper, 2015; Gibling and Davies, 2012; Kenrick and Crane, 1997; Morris et al., 2018). Land plants originated from Charophyta, a group of freshwater algae, with Zygnematophyceae as the sister lineage of extant land plants (Clark et al., 2023; Delaux and Schornack, 2021; Han, 2019; Morris et al., 2018; Tihelka et al., 2022). Over 450 million years elapsed since the start of the transition to terrestrial environments, plants have evolved enormous morphological diversity (Clark et al., 2023) and ubiquitous symbiotic associations with microbes (primarily, bacteria and fungi) and invertebrates (in particular, nematodes and various arthropods), ranging from parasitism to mutualism (Delaux and Schornack, 2021; Tihelka et al., 2022). These interactions with microbes and invertebrates had far-reaching effects on the evolution of land plants (Delaux and Schornack, 2021).

Since the discovery of tobacco mosaic virus (TMV), the first virus ever described, a wide variety of viruses have been isolated and characterized in plants, especially angiosperms of agricultural importance. These viruses span multiple phyla, including Kitrinoviricota (including, among others, well-studied orders Martellivirales, Tymovirales, and Hepelivirales, and the large family Tombusviridae), Pisuviricota (including the largest family of plant viruses, Potyviridae, Amalgaviridae, and Partitiviridae), and Negarnaviricota (including Rhabdoviridae, Fimoviridae, Phenuiviridae, Tospoviridae) (Dolja et al., 2020; Lefeuvre et al., 2019). It is generally accepted that land plants and algae host distinct virus repertoires: land plants are predominantly infected by RNA viruses and single-stranded DNA viruses, whereas algae primarily host large double-stranded DNA viruses (Dolja and Koonin, 2011; Dolja and Koonin, 2018a; Dolja et al., 2020), with only a limited diversity of RNA viruses (Coy et al., 2018; Mifsud et al., 2022; Mushegian et al., 2016; Van Etten et al., 2019). Crucially, in nature, plants do not exist in isolation but function as complex holobionts, intimately associated with endophytic fungi and diverse invertebrates (Bordenstein and Theis, 2015; Poupin and González, 2024). This close ecological proximity blurs strict host boundaries and provides a continuous arena for cross-kingdom viral exchanges. Indeed, host switching between different kingdoms of eukaryotes, particularly from arthropods to plants and from fungi to plants, known as horizontal virus transfer (HVT), has been hypothesized to have been a major driver of the plant RNA virome evolution (Dolja and Koonin, 2018b; Dolja et al., 2020; Roossinck, 2019). For instance, virus families such as Rhabdoviridae, Fimoviridae, and Tospoviridae show clear phylogenetic links between plant- and arthropod-infecting members (Kormelink et al., 2021). By contrast, other families, such as Partitiviridae and Endornaviridae, include cryptic, persistent viruses that lack movement proteins (MPs) and are apparently transmitted vertically, showing strong evolutionary connections between plant and fungal viromes (Roossinck, 2012; Roossinck et al., 2015). Systematic study of the phylogenetic distribution and genomic features of these diverse groups of viruses within a comprehensive transcriptome dataset is crucial for testing hypotheses on their origins and adaptation to plant hosts. However, in spite of the long history of plant virus research, the current knowledge of RNA virus diversity in plants and algae remains limited, fragmentary, and biased towards agriculturally important species (Charon et al., 2021; Dolja and Koonin, 2018a; Lefeuvre et al., 2019; Roossinck, 2019), lagging behind the overall explosion of global RNA virome studies (Edgar et al., 2022; Hou et al., 2024; Neri et al., 2022; Shi et al., 2018; Shi et al., 2016a).

To address the fundamental gap in the characterization of the diversity of plant RNA viruses, we mined the transcriptomes of 1,857 species sampled across land plants and algae for RNA viruses and report here the discovery of 22,750 plant-associated RNA viruses. We also identified 12,248 endogenous viral elements (EVEs) that document past viral infections in the genomes of 813 land plants. Phylogenomic analyses of the plant RNA virome within the context of the eukaryotic RNA virosphere as a whole provide insights into the origins and evolution of diverse RNA virus lineages in land plants. In particular, these analyses provide strong support for numerous instances of HVT, primarily, from invertebrates and fungi to plants, and reveal diverse atypical, probably, latent plant-associated viruses. Thus, cross-kingdom host switching and subsequent change in genome architectures seem to have shaped the diversity and evolution of the plant RNA virome. This work substantially expands the known diversity of plant RNA viruses, provides a major resource for studying the plant RNA virome, and contributes to understanding the origins and natural history of plant RNA viruses.

RESULTS

Expanded diversity of RNA viruses associated with land plants

The initial screening of 2,977 transcriptomes from 1,857 plant species yielded 22,750 contigs harboring RNA-directed RNA polymerase (RdRP), the hallmark protein of RNA viruses (kingdom Orthornavirae) used for their phylogeny-based classification (Wolf Yuri et al., 2018) (Figure 1A–B). Based on the initial screening, a set of 3,404 representative contigs was selected for further analysis by requiring >70% coverage of the RdRP domain and excluding phylogenetic singletons. These viruses are operationally defined as “plant-associated” because they were identified in plant transcriptomes. These contigs were combined with a known plant virus genome sequence dataset (comprising plant viruses retrieved from NCBI, recent studies (Mifsud et al., 2022), and the Serratus palmprint database (Edgar et al., 2022)) and clustered based on amino acid identity (AAI) thresholds of 0.9, 0.7, 0.5, and 0.3. The numbers of these clusters (hereafter, total data) were 5629, 2957, 1135 and 312, compared to 2005, 1330, 620 and 186, respectively, obtained for the known plant virus dataset (Figure 1C). Notably, the magnitude of discovery increased substantially at deeper taxonomic levels: the increase in the diversity of the identified viruses changed from 1.55-fold at the 0.9 AAI to 2.48-fold at the 0.3 AAI threshold. This trend shows that our study did not merely add new variants to known viral groups but, rather, primarily uncovered deep, previously uncharacterized lineages that fill major gaps in the plant virome. Based on the adjusted Rand index (ARI), we found that the AAI thresholds of 0.5 and 0.3 aligned well with the genus (ARI = 0.84) and order (ARI = 0.93) ranks classified by the International Committee on Taxonomy of Viruses (ICTV), respectively. For the five large Orthornavirae phyla (Kitrinoviricota, Pisuviricota, Duplornaviricota, Negarnaviricota, and Lenarviricota) (Koonin et al., 2024), the number of 0.5 AAI virus clusters (roughly at the genus level) in the total data was 1.8 to 4.0 times those in the known data (Figure 1D). Altogether, our study resulted in a 2.2-fold increase in the known diversity of the RNA viruses associated with land plants, at the genus level.

Figure 1. Discovery of RNA viruses in land plants.

A. Pipeline for mining RNA viruses in land plants. B. Number of transcriptomes from different land plant groups analyzed in this study. C. Number of viral clusters identified at RNA-dependent RNA polymerase (RdRP) clustering thresholds of 0.9, 0.7, 0.5, and 0.3, based on the known virus dataset and the total dataset (this study combined with the known virus dataset). D. Number of viral clusters across viral phyla and land plant groups at RdRP AAI threshold of 0.5 (K: known dataset, T: total dataset). E-G. Rarefaction curves for viral cluster discovery at different RdRP AAI thresholds (E), across land plant groups at 0.5 AAI (F), and across viral phyla at 0.5 AAI (G). Solid lines indicate rarefaction, dashed lines indicate extrapolation, and shading represents 95% confidence intervals.

The rarefaction curves showed that the number of virus clusters based on the RdRP AAI thresholds of 0.9, 0.7, and 0.5 increased monotonically with the number of samples (transcriptomes), whereas the number of virus clusters based on the 0.3 AAI threshold (roughly, at the order level) appeared to nearly saturate (Figure 1E). For each of the five major land plant groups (angiosperms, gymnosperms, ferns, lycophytes, and bryophytes), the number of virus clusters increased steadily with the number of transcriptomes analyzed (Figure 1F). Notably, bryophytes exhibited a higher new cluster discovery rate than the other four plant groups, in accord with limited previous sampling of this plant lineage (Figure 1F). For the virus phyla Kitrinoviricota, Pisuviricota, Lenarviricota, and Negarnaviricota, the number of 0.5 AAI virus clusters increased gradually with sampling (Figure 1G). Kitrinoviricota included the largest number of 0.5 AAI virus clusters (Figure 1G). In contrast, for Duplornaviricota, the number of 0.5 AAI clusters was much lower, nearing saturation (Figure 1G) and implying relatively low diversity of duplornaviruses in land plants. These results reveal uneven representation of RNA virus phyla in the global plant virome.

A major challenge of transcriptome mining is to confidently assign hosts for the newly identified viruses. Even though transcriptomes analyzed here came from plants, newly identified viruses potentially could originate from contamination (e.g., soil or herbivore material) or could be derived from organisms that constitute plant holobionts (e.g., ubiquitous symbionts such as fungi). To assess the confidence of plant-virus associations, we developed a quality control (QC) framework based on two bioinformatic metrics: viral abundance (Log10 RPM) and plant contig ratio (Figure S1). We first established a benchmark dataset of 1,450 “trusted” plant viruses, that is, viruses with strong prior evidence of plant infection, such as those belonging to well-characterized plant virus families and/or those encoding MPs (Data S1 and see Methods). We then defined stringent QC criteria based on the distribution of these trusted viruses. When applied to our data, 911 (62.8%) of the benchmark viruses and 680 (40.3%) of the other putative plant-associated viruses passed these criteria. The high stringency of this filter, indicated by the exclusion of more than one-third of the known plant viruses in the benchmark set, ensures that sequences that pass it are highly likely to represent reliable plant-virus associations. In the following sections, we combine this QC with additional evidence from taxonomic library profiling (SRA STAT) to assess the host association for each major viral clade we analyzed.

We combined newly described and known RNA viruses associated with plants in accord with the current taxonomy and mapped these taxa (mostly, at the family rank) onto 91 orders of land plants (Figure 2). Overall, our analysis expanded the plant host ranges for all the studied virus taxa by 1 to 28 plant orders, except for Potyviridae. In particular, virus families previously known to infect land plants, such as Benyviridae and Botourmiaviridae, as well as viruses identified in plant transcriptomes, such as Deltaflexiviridae (Bartholomäus et al., 2017; Bejerman and Debat, 2022; Chen et al., 2016) and Narnaviridae (Yang et al., 2022), were found to be associated with many more land plant orders than previously recognized (Figure 2). Therefore, our study greatly expands both the diversity and the host ranges of RNA viruses associated with land plants.

Figure 2. Distribution of RNA viruses across plant clades.

The matrix illustrates the presence of different viral taxa (columns) across 91 orders of land plants (rows). The phylogenetic tree on the left displays the evolutionary relationships among the plant orders. A circle at an intersection indicates that at least one virus belonging to that viral taxon was identified in at least one transcriptome from that plant order. Filled circles represent novel viral groups identified in this study, while empty circles denote previously known groups from the reference dataset. The number of transcriptomes sampled for each plant taxon is shown in parentheses. Colors correspond to different viral phyla. Shading highlights plant orders in which over 80% of the associated viral groups were newly discovered in this study. Viral groups are abbreviated as follows: K., Kitrinoviricota; P., Pisuviricota; D., Duplornaviricota; L., Lenarviricota; N., Negarnaviricota; U., unclassified taxa.

Our analysis included transcriptomes from plants of agricultural importance, in which we identified 125 previously undescribed RNA viruses in 25 species (Figure S2). Among these, at least eight novel viruses (viruses in new lineages or new hosts) were identified in rice (Oryza spp.), wheat (Triticum spp.), maize (Zea mays), sunflower (Helianthus annuus), banana (Musa acuminata), and tea plant (Camellia sinensis) each. Furthermore, the host ranges of the plant virus lineages known to infect crop plants, such as Tymoviridae, Kitaviridae, Endornaviridae, Partitiviridae, Botourmiaviridae, and recently characterized jiviviruses (Candresse et al., 2022; Kinsella et al., 2022), were substantially expanded (Figure S2). The potential pathogenicity and prevalence of these crop-associated viruses merit further study.

The diversity of plant-associated RNA viruses showed a notably uneven distribution among major evolutionary lineages of the plant kingdom (Archaeaplastida) (Figure 2). Besides land plant transcriptomes, we also mined for RNA viruses 454 transcriptomes of deep-branching lineages within Archaeplastida, including glaucophytes (Glaucophyta), red algae (Rhodophyta), and green algae (Chlorophyta and Charophyta). Among these, analysis of four Glaucophyta transcriptomes did not yield any RNA virus. We further found that Rhodophyta, Chlorophyta, and Charophyta host only a limited diversity of orthornaviruses (Figure 2), in agreement with previous observations (Dolja et al., 2020; Mifsud et al., 2022; Mushegian et al., 2016). A high proportion of alga-associated viruses, many of which were assigned with high confidence (see Methods), belong to the orders Durnavirales and Ghabrivirales that include viruses with double-stranded RNA genomes and are highly enriched for fungal viruses (Kondo et al., 2022). Thus, a major fraction of algal viruses might have been acquired from fungi via recurrent inter-kingdom HVT.

The deepest branching order of non-vascular land plants that exhibited substantial orthornavirus diversity is Jungermanniales (liverworts; Figure 2). However, only the flowering plants (Angiosperms) host virtually the entire diversity of orthornaviruses identified in this and previous studies, with several plant orders (e.g., Poales, Asterales, Asparagales, Lamiales, Malpighiales) being particularly “welcoming” virus hosts (Figure 2). Although this pattern of virus distribution could be in part due to more thorough sampling of angiosperms, the much greater diversity of the flowering plants themselves compared to other plant lineages likely plays a major role. Furthermore, because the diversity of RNA viruses in land plants is but a fraction of that of invertebrate (and in particular, arthropod) viromes, it has been proposed that the plant RNA virome originated from the invertebrate viromes during the phase of explosive evolution of plants and their tight association with arthropods at the terrestrialization stage (Dolja et al., 2020). The rest of this work not only provides extensive phylogenomic evidence in support of this concept but also expands and clarifies its evolutionary context.

The origins of RNA viruses in land plants

Expanded diversity of RNA viruses associated with eukaryotes outside Embryophyta.

To investigate the provenance of the plant RNA viruses in a broader evolutionary context, we mined orthornaviruses in the transcriptomes of diverse eukaryotes outside Embryophyta (hereafter non-Embryophyta eukaryotes or nEEs). In addition, we screened the nuclear genomes of both land plants and nEEs for EVEs derived from RNA viruses (Figure 3A–C). A total of 152,826 RNA virus contigs from 5,228 nEE transcriptomes were identified, among which 43,910 contigs encoded RdRPs. Additionally, 89,878 EVEs were identified in 813 plant genomes and 10,010 nEE genomes, with 9,417 EVEs encoding RdRPs, after assembly and de-duplication (Figure 3A–C). All these RdRPs were clustered into established viral taxa, resulting in a 1.8 to 2.4-fold increase in the known diversity of RNA viruses of eukaryotes compared to the NCBI Virus database (as of November 19, 2022) at the AAI level of 0.5 (Figure 3D).

Figure 3. Relationship of plant RNA viruses with viruses of other eukaryotes.

A. Pipeline for mining RNA viruses in non-Embryophyta eukaryotes (nEEs) and endogenous viral elements (EVEs) in eukaryotes. B-C. Number of transcriptomes (B) and genomes (C) mined. D. Number of viral clusters across viral phyla and eukaryote groups based on RdRP sequences at a 0.5 clustering threshold (K: known dataset from the NCBI Virus database; T: total dataset combining this study and the NCBI Virus database). E. NMDS ordination of RNA viruses across eukaryote groups, based on the composition of viral communities within different eukaryotes. F. Similarity network of RNA viruses constructed from BLASTp analysis of RdRP sequences. Circles represent centroid sequences (0.5 AAI threshold), with size reflecting cluster size (0.9 AAI threshold). Lines indicate shared similarity (e-value threshold of 10⁻¹⁰, identity threshold of 30%, and length coverage threshold of 100 amino acids). Clusters within the same phylum are circled in the same color. Representative plant viral groups are indicated, where viral phyla are abbreviated as follows: K., Kitrinoviricota, P., Pisuviricota, D., Duplornaviricota, L., Lenarviricota, and N., Negarnaviricota. G. Robustness of network modularity. Comparison of the modularity score of the original network with the distribution of scores from 100 balanced sub-networks, generated by stratified subsampling of viral clusters across all major phyla. H. Number and host distribution of RNA viruses in viral communities (CM) identified in the network, with phylum-specific communities circled in matching colors.

Virus diversity analysis.

We further used non-metric multidimensional scaling (NMDS) to investigate sharing of virus communities (identified within the RdRP similarity network graph as densely connected subgraphs) among different eukaryote hosts. Permutational multivariate analysis of variance (PERMANOVA) revealed a significant overall difference in the RNA virome composition among the major groups of eukaryotes (P-value = 0.001), indicating substantial dissimilarity among their viromes (Figure 3E). Subsequent pairwise PERMANOVA showed that the composition of RNA virome associated with land plants was significantly different from those of the major nEE taxa, including red and green algae (Table S1). By contrast, no significant differences in RNA viromes were observed among algae, fungi, and protists, suggesting that these deep-branching lineages of eukaryotes might have shared their ancestral RNA viromes (but also possibly reflecting insufficient sampling, particularly, of algal and protist diversity) (Table S1). These findings imply that the RNA virome of land plants is not a simple derivative of the algal virome, and was not shaped primarily by virus-host co-divergence during plant terrestrialization.

To explore the evolutionary relationships between well-characterized plant and nEE RNA viruses, we analyzed RdRP similarity networks (Figure 3F). All RNA viruses clustered into 302 communities with modularity of 0.826, far above the threshold of 0.3 which is considered an indication of a pronounced community structure (Newman, 2006), and the number of members between 1 and 1,296. To test whether the observed network modularity was affected by sampling bias, we performed stratified subsampling analysis. We computationally generated 100 sub-networks, each with an equal number of viral clusters (nodes) from every major phylum. The mean modularity of these balanced networks was 0.830 ± 0.005, which is statistically indistinguishable from the 0.826 calculated for the original, complete network (Figure 3G). Thus, the highly modular structure of the RNA virome is likely to be a genuine biological feature reflecting deep evolutionary divergence, rather than an artifact of uneven sampling.

Among the 302 communities, 50 included RNA viruses of land plants (Figure 3H). The proportion of RNA viruses associated with land plants varied among these communities. The well-recognized plant viruses are significantly distinct from nEE viruses, with 19 communities including more than 80% plant-associated viruses. Notably, plant-associated viruses frequently formed communities with arthropod viruses, including 11 in which the arthropod viruses accounted for more than 50% (Figure 3F–G). These results imply close evolutionary relationships between plant and arthropod RNA viruses and indicate that the well-separated clusters of plant-associated viruses have likely originated from arthropod viruses on many independent occasions. Because explosive diversification of arthropods and their RNA virome occurred in marine environments during the Cambrian period prior to the diversification of land plants, it has been argued that the emerging terrestrial plant virome was “inseminated” by that of arthropods and other invertebrates (Dolja et al., 2020). The observed strong, network-level association between plant and arthropod viruses is in agreement with this concept. Furthermore, at least 10 communities encompassed substantial fractions of viruses associated with plants and fungi, including three where arthropod and fungal viruses were both well represented (Figure 3F–G). This latter combination seems to be biologically relevant given that fungi are common symbionts of both plants and arthropods (Raman et al., 2012), therefore forming the closely associated, ménage à trois communities well poised for HVT via multiple routes.

Phylogenetic analysis of orthornaviral RdRPs.

To further explore the implications of the similarity network analysis, we generated RdRP-based phylogenetic trees for several major lineages within each of the five orthornavirus phyla (Figure 4, Figures S3–S7). These analyses revealed distinct phylogenetic patterns for different groups of plant RNA viruses. Some plant virus lineages are highly divergent, with few nEE viruses clustering with them, exemplified by the order Tymovirales within phylum Kitrinoviricota (Figure S3). In contrast, many plant-associated virus lineages exhibited complex phylogenetic patterns, being lodged within the diversity or even intermixed with viruses of arthropods and other invertebrates at different taxonomic levels including order, family, or even genus. The most relevant examples of such mixing of plant and animal viruses include order Picornavirales (Figures 4A and S4), order Martellivirales and the family Kitaviridae and Clade M5 within it (Figures 4B and S5), order Bunyavirales and families Phenuiviridae and Fimoviridae within it (Figures 4C and S6), family Rhabdoviridae (Figures 4D and S7), and genera Oryzavirus and Fijivirus within family Spinareoviridae (Figure 4E). Analogously, some plant-associated virus lineages were lodged within deeper rooted diversity or were intermixed with fungal (and in some cases also oomycete) viruses, as within family Gammaflexiviridae (Figure 4F) and order Ghabrivirales (Figure 4G–H). In agreement with the similarity network analysis described above, the phylogenetic patterns for some lineages were more complex, possibly due to multiple, multidirectional HVT events among fungi, arthropods and plants, including potential transfers from plants back to other eukaryotes, as illustrated by family Partitiviridae (Figure 4I). Notably, fungal, arthropod and plant viruses were intermixed in several genera, including Gammaflexivirus (Figure 4F), Alphapartitivirus and Betapartitivirus (Figure 4I), and Alphachrysovirus (Figure 4G). In only a few instances exemplified by deep-branching clades within the order Durnavirales, namely, Clades D2 and D1, algal viruses appeared to be basal to land plant viruses (Figure 4J).

Figure 4. Phylogenetic analysis of RNA viruses from land plants and other eukaryotes.

Phylogenetic trees were constructed using RdRP core region sequences. Collapsed tip counts are indicated to the right of each triangle. The proportional host composition for each clade is shown as a pie chart. Colored triangles highlight clades in which a single host type accounts for >60% of the sequences, and white triangles label clades with no dominant host type. Yellow stars label high-confidence alga viruses. SH-aLRT support values of >0.80 are displayed at major nodes. Scale bars represent the number of amino acid substitutions per site. Key viral clades and hosts are annotated. The viral groups include Picornavirales (A), Virgaviridae, Kitaviridae, negevirus, and Clade M5 in Martellivirales (B), Bunyavirales (C), Rhabdoviridae (D), Spinareoviridae (E), Gammaflexiviridae (F), Ghabrivirales (G-H), Partitiviridae (I), Durnavirales (J), and Yueviridae (K), and Flaviviridae (L). Viral phyla are abbreviated as follows: K., Kitrinoviricota, P., Pisuviricota, D., Duplornaviricota, L., Lenarviricota, and N., Negarnaviricota. EVE, endogenous viral element.

To quantitatively assess the inter-kingdom HVT, we estimated host transition rates using a Bayesian approach (Pagel et al., 2004) and identified 16 significant pathways of likely virus transfer to land plants (Data S2). The primary sources for HVT to land plants were invertebrates (eight pathways) and fungi (five pathways). The host transition rate estimates should be interpreted with caution because it is technically challenging to distinguish between host switching and other, more complex biological scenarios, such as vector-mediated virus transmission. Nevertheless, these results provide further support for the importance of invertebrates and fungi in the origin and evolution of the plant virome.

Collectively, our findings are in agreement with the concept according to which algal progenitors of land plants carried a limited diversity of RNA viruses during the conquest of terrestrial environments, whereas emergent close ecological associations with other terrestrial organisms, particularly arthropods and fungi, shaped the RNA viromes of land plants via continuous and frequent inter-kingdom HVT (Dolja et al., 2020). The phylogenomic analysis presented in the following sections further solidifies this concept.

Previously unknown clades of plant-associated RNA viruses

Novel clades within Martellivirales.

The order Martellivirales within phylum Kitrinoviricota includes eight recognized families, six of which are large families of plant viruses. We identified five previously uncharacterized or emerging virus clades roughly at the family level within this order (Figure S5). The largest of these, Clade M1, which corresponds to the proposed family Ambsetviridae initially characterized in fungi (Deakin et al., 2017), forms a sister group to Endornaviridae, a family of capsid-less viruses reproducing mostly in plants, fungi, and oomycetes (Valverde et al., 2019) but also found in invertebrates (Shi et al., 2016a). Despite sharing closely related RdRPs, endornaviruses and Clade M1 viruses have strikingly different genome organizations. The large, monopartite endornavirus RNAs encode up to six protein domains involved in RNA replication and carbohydrate metabolism (Valverde et al., 2019), with genome sizes ranging from 9,639 to 23,635 nucleotides (accessions NC_030938 and NC_031463, respectively). In contrast, the genomes of diverse Clade M1 viruses consist of at least two segments, one of which encodes RdRP (the largest genome segment comprising 2,060 nucleotides) whereas the other one encodes a capping methyltransferase (Mtr, the largest genome segment comprising 1,997 nucleotides) shared with some endornaviruses (Figure S5). Although it cannot be ruled out that additional components of Clade M1 genomes were missed, this distinct group of plant-associated viruses seems to include minimal endorna-like replicons. Notably, these replicons were found in 45 orders across Archaeplastida, including green algae as well as diverse non-vascular and vascular land plants (Figure 2). Of the 233 Clade M1 viruses, 43 passed the QC, including 35 from libraries with negligible fungal reads (<1%) and 8 from libraries with negligible nEE reads (⩽1%). The phylogenetic affinity with endornaviruses, combined with the ubiquitous presence of Clade M1 viruses in plant transcriptomes, supports their ability to reproduce in plant hosts. Both plant endornaviruses and Clade M1 viruses lack MPs and are likely transmitted exclusively vertically, through gametes. Given their documented fungal hosts and pervasive association with plants, this clade likely represents a multi-host lineage or a transitional state spanning the plant-fungal ecological interface.

The unclassified Clade M5 of Martellivirales is a sister group of Kitaviridae, a family of plant viruses closely related to insect negeviruses (Lu et al., 2024; Ramos-González et al., 2023) (Figure S5). Both negeviruses and kitaviruses appear to be in evolutionary flux, showing substantial variation in genome and virion architectures (Ramos-González et al., 2023). In contrast to the monopartite negevirus genomes, the genomes of kitaviruses consist of two to four genome segments and encode either a 30K MP broadly distributed among diverse plant viruses (Butkovic et al., 2023) or an unrelated MP tandem found in Higrevirus genus (Morozov and Solovyev, 2022) (Figure S5). Kitaviruses are not typical plant viruses: despite being able to move from cell to cell and to be transmitted by arachnid mites, they lack the ability to establish systemic infections (Ramos-González et al., 2023). Therefore, kitaviruses likely evolved from the nege-like arthropod viruses via cross-kingdom HVT and are gradually transitioning to plant host-based lifestyle. The related Clade M5 viruses appear to be undergoing a similar evolutionary transition. Although the majority of these viruses were identified in arthropods, Clade M5 viruses were also observed in transcriptomes of twelve plant orders (Figure 2). Of the 19 plant-associated Clade M5 viruses, 9 passed the QC, including 8 from libraries with negligible metazoan reads (⩽1%). In addition to Mtr, superfamily 1 helicase (S1H), and RdRP, monopartite genomes of the plant-associated Clade M5 viruses also encode duplicated or triplicated TMV coat protein (CP)-like proteins that might occupy different positions and perform distinct functions in the virion, similarly to the case of closteroviruses (Dolja et al., 2006). Notably, viruses in a distinct lineage within Clade M5, including viruses experimentally demonstrated to infect rubber trees (Hevea brasiliensis) (Zhao et al., 2023), also encode a canonical 30K MP (Figure S5). In addition, these viruses encode the AlkB demethylase previously found in diverse plant viruses and implicated in counter-acting host defense that involves viral RNA methylation (van den Born et al., 2008; Yue et al., 2022) (Figure S5). Thus, Clade M5 viruses further enrich the dynamic picture of arthropod virus adaptation to plant hosts among Martellivirales.

The relatively small Clade M3, which is represented in five genera of gymnosperms and angiosperms, is a sister group to the vertebrate-infecting Togaviridae. Two of the seven plant-associated Clade M3 viruses passed the QC, both originating from libraries with negligible fungal and metazoan reads (≤1%). The genomes in this clade encode the replication polyprotein similar to those in the majority of the Martellivirales and a Superfamily 2 helicase (S2H) of unknown function, similar to the recently described plant-associated jiviviruses with multipartite genomes (Candresse et al., 2022; Kinsella et al., 2022). The viruses in the Clade M4 that were identified in 23 plant orders encode a similar replication polyprotein and a TMV-like CP. Although related viruses within this broad assemblage have been characterized in fungi (Yoshioka et al., 2025), 19 of the 93 viruses of Clade M4 passed the QC, including 13 from libraries with negligible fungal and metazoan reads (≤1%).

Clade M2, which formed a deep branch in the RdRP tree of Martellivirales (Figure S5), was found in association with 23 plant orders (Figure 2). Viruses in this clade encode a replication polyprotein with Mtr, S1H, and RdRP domains typical of Martellivirales and a single jelly-roll (SJR)-CP. Because Clade M2 viruses do not encode any of the signature plant virus proteins such as MPs, they could be persistent, vertically transmitted plant viruses although their origin from plant endophytic fungi could not be excluded. In contrast to Clade M4, only three of the 48 Clade M2 viruses and four of the 34 jiviviruses passed the QC although all the passing viruses originated from libraries with negligible fungal and metazoan reads. We also detected association of jiviviruses with 15 additional plant orders (Figure 2). Although recent studies have identified jiviviruses in axenic fungal cultures (Kinsella et al., 2022) and noted uncertainties regarding their definitive plant hosts (Candresse et al., 2022), our analysis reveals a subset of plant-associated jiviviruses encoding previously undescribed CPs distantly related to those forming flexuous filamentous particles (CPf) of closteroviruses (Figure S8B). Nonetheless, the association of viruses in these groups with plants requires further evidence. Together, these findings expand the diversity of Martellivirales and reveal substantial variability in the complexity of their genomes (Figure S5).

Expanded diversity of Picornavirales.

Within the order Picornavirales of phylum Pisuviricota, the plant-infecting Secoviridae are closely related to Iflaviridae and to the unclassified Clade P3. Because Clade P3 is predominantly invertebrate-associated, its sparse detection in our plant transcriptomes likely reflects micro-arthropods within the plant holobiont (Shi et al., 2016a). Nonetheless, this phylogenetic proximity implies the possibility of the origin of secoviruses via HVT from invertebrates to plants accompanied by genome segmentation and acquisition of the 30K MP (Figure S4). Furthermore, several plant-associated viruses within predominantly non-plant families passed the QC. These included members of the Dicistroviridae (20 of the 48 passed QC), Iflaviridae (14 of the 26), and Polycipiviridae (10 of the 19). Taxonomic profiling confirmed that many of these viral genomes originated from libraries with negligible estimated-host contamination. Specifically, libraries in which 15 dicistroviruses, eight iflaviruses, and eight polycipiviruses were identified lacked metazoan reads (⩽1%), further highlighting the intimate associations between viruses infecting different hosts within the plant-arthropod holobiont, likely, facilitating cross-kingdom HVT.

New plant-associated lineages of Negarnaviricota.

Order Mononegavirales (phylum Negarnaviricota) includes a subfamily of plant viruses Betarhabdovirinae that encode a typical 30K MP. Phylogenetic analysis of the RdRP domains (embedded in L, the large replication protein of most negarnaviruses) clearly suggests the origin of this subfamily from arthropod-hosted Deltarhabdovirinae or a group of related nematode viruses (identified here as EVEs) (Figures 4D and S7). Furthermore, we identified a small Clade R2 of plant-associated viruses encoding 30K MP and closely related to Gammarhabdoviridnae, indicating a relatively recent origin from invertebrate viruses dominating the entire Rhabdoviridae family (Figure S7). Finally, although some members of the unclassified Clade R1 have been recently shown to propagate in plants (e.g., Maize suscal virus and a Disporopsis-associated virus (Li et al., 2024)), our analysis systematically expanded this lineage across 10 orders of flowering plants (Figures S1 and S7). Of the 20 plant-associated Clade R1 viruses, 11 passed the QC, including seven from libraries with negligible metazoan reads (≤1%). This broad ecological distribution confirms Clade R1 as a pervasive component of the plant holobiont.

The order Bunyavirales (phylum Negarnaviricota) with segmented genomes is dominated by arthropod viruses, with a healthy proportion of vertebrate and plant viruses (Figure S6). This order appears to be particularly prone to spawning clades of plant viruses on multiple, independent occasions (Kormelink et al., 2021). Family Phenuiviridae includes arthropod-transmitted vertebrate and plant virus genera nested within a greater diversity of arthropod- and other invertebrate-specific viruses (Sasaya et al., 2023). The major difference in the genome structures of the arthropod and plant phenuiviruses is the loss of the envelope glycoproteins and acquisition of the additional genome segments encoding 30K MP and plant-specific RNAi suppressors by the latter. Basal to Phenuiviridae, we identified a clade (initially designated Clade B4) that corresponds to the recently established family Konkoviridae (Kuhn et al., 2024) (Figure S6). This family includes the known plant-infecting Tulip streak virus. Our analysis expands the known diversity of this family to include viruses associated with diverse additional plant orders (e.g., Poales, Zingiberales). Among the nine analyzed viruses of this clade, four passed the QC, all originating from libraries with negligible metazoan reads (≤1%). The four-component genomes of plant-associated viruses in this clade also apparently have lost glycoproteins, while capturing the 30K MP-encoding genome segment. Similar evolutionary processes apparently were at work in the evolution of the other bunyavirus families of plant viruses, Fimoviridae (Digiaro et al., 2024) and Tospoviridae (Kormelink et al., 2021). All well-characterized plant members of Bunyavirales are transmitted by arthropod vectors and are capable of replication in these vectors (Kormelink et al., 2021), again pointing to their origin from arthropods.

The L-protein of the deep-branching, recently proposed family Viridisbunyaviridae of moss-associated viruses (Mifsud et al., 2022) is closely related to those of the putative family Deltamycobunyaviridae associated with plant-pathogenic oomycetes (Nerva et al., 2019). Thus, viridisbunyaviruses could have originated via HVT from oomycetes analogously to a recently discovered rice virus (Wang et al., 2022) related to fungal and oomycete Discoviridae (Kuhn et al., 2023). Both the rice virus and discoviruses lack membrane glycoproteins whereas only the former encodes a divergent MP homolog (Figure S6). Thus, a common evolutionary scenario for the plant viruses within the order Bunyavirales is their origin via HVT, primarily, from arthropods but also from fungi or oomycetes, accompanied by acquisition of MPs and often by the loss of the envelope glycoproteins.

Atypical plant-associated viruses

We also identified a variety of plant-associated viruses in lineages not previously demonstrated to infect plants. These viruses do not encode proteins involved in plant virus-specific functions such as MPs. Many of these viruses passed our stringent QC criteria and could represent bona fide ‘cryptic’ plant viruses that establish persistent infections supported by vertical transmission through gametes, similar to Endornaviridae, Partitiviridae or Mitoviridae previously demonstrated to propagate in plants without causing detectable disease symptoms (Roossinck, 2012). However, their definitive host assignment within the holobiont require experimental validation.

Narnaviridae.

The family Narnaviridae (phylum Lenarviricota) that consists of capsid-less RNA replicons encoding only RdRP is broadly represented in phytopathogenic and other fungi, protists (Chiba et al., 2023; Kondo et al., 2022), and invertebrates (Zhou et al., 2025), with some reports of association with plants (Mifsud et al., 2022; Yang et al., 2022). Here, we identified such plant-associated narnaviruses in as many as 32 Archaeplastida orders including green algae as well as vascular and non-vascular plants (Figure 2). Of the 156 plant-associated narnaviruses, 46 passed the QC, and 10 of these were identified in libraries where all nEE reads were negligible (≤1%). Notably, several cases exhibited exceptionally high viral abundance (RPM > 100) along with an extremely low fungal content (<0.01%). Although these data demonstrate a robust association of narnaviruses with highly purified plant transcriptomes, definitive proof of plant infection requires future experimental validation. Taken together with the recent demonstration of a narnavirus replication in a plant host (Gao et al., 2024), our findings suggest that these narnaviruses might replicate in multiple hosts or represent transitional states within the plant holobiont.

Gammaflexiviridae and Ormycoviruses.

Gammaflexiviridae (phylum Kitrinoviricota) is a small family of viruses so far limited to fungal hosts (Roossinck, 2019). Here, we identified multiple gammaflexiviruses associated with 11 orders of land plants (Figure 2). Two of the 14 plant-associated gammaflexiviruses passed the QC, including one from a library where combined metazoan and fungal reads were negligible (<1%), although reads assigned to other unicellular eukaryotes were present.

We also detected viruses related to the recently discovered ormycoviruses (Forgia et al., 2022) in association with 23 plant orders (Figure 2). The ormycoviruses encode extremely divergent RdRPs that could not be assigned to any of the established six phyla of orthornaviruses (Forgia et al., 2022). These viruses have been reported to replicate in fungi, oomycetes, and recently, apicomplexans (Petrone et al., 2024). Sixteen of the 59 plant-associated ormycoviruses passed the QC, including one highly abundant ormycovirus (Log10 RPM = 3.89) originating from a library with negligible nEE reads (⩽1%). Given the known host ranges of these viral families and the complex composition of plant transcriptomes, we conservatively classify these lineages as plant-associated, perhaps, representing transitional components of the plant holobiont.

Yueviridae.

Since their discovery in crustaceans (Shi et al., 2016a), Yueviridae (order Goujianvirales in class Yungchangviricetes, phylum Negarnaviricota) have been identified in protists (specifically, stramenopiles), insects (Krupovic et al., 2023) and a lycophyte, one of the most ancient vascular land plants (Mifsud et al., 2022). Our study greatly expanded the diversity of plant-associated yue-like viruses that were identified in mosses (two orders), lycophytes (three orders), ferns (two orders), gymnosperms (four orders), and angiosperms (19 orders) (Figure 2). In total, 57 of the 108 plant-associated yueviruses passed the QC, including 36 from libraries with negligible metazoan reads (⩽1%) and two libraries where all nEE reads combined were negligible (⩽1%). The bi-segmented RNA genomes of yueviruses encode only L-protein and nucleocapsid protein, lacking not only the lipoprotein envelope typical of the majority of negarnaviruses but also MP, a hallmark of ‘genuine’ plant viruses. Furthermore, phylogenetic analysis reveals an interesting evolutionary pattern: viruses associated with deep-branching plant lineages (e.g., Zygnemophyceae, bryophytes, and gymnosperms) generally occupy basal positions relative to the large radiation of angiosperm-associated viruses (Figure 3K). Additionally, we identified two Yueviridae-derived endogenous viral elements (EVEs) integrated into the nuclear genomes of an angiosperm (Arctostaphylos glauca, JAHSPW010000002.1: 26957925–26959439) and a charophyte green alga (Chara braunii, BFEA01000106.1: 1705403–1706581), providing historical evidence of their association with plants. Therefore, we conservatively classify these viruses as plant-associated, representing persistent components of the plant holobiont.

Stelpaviricetes.

Class Stelpaviricetes of phylum Pisuviricota is currently split into two orders, with a single family each, Stellavirales (Astroviridae) that infect vertebrates and plant-infecting Patatavirales (Potyviridae). Although the affinity between astroviruses and potyviruses supported by comparisons of the RdRPs and trypsin-like proteases has long been appreciated (Jiang et al., 1993), the exact evolutionary relationships between these families with contrasting host ranges remained unclear. The phylogenetic tree for the RdRPs of expanded Stelpaviricetes (Figure S9A) implies that both astroviruses and potyviruses are embedded within a far greater diversity of related viruses from various environments (French et al., 2022; Sadiq et al., 2024; Zell et al., 2023), including ‘protopotyviruses’ from aquatic metaviromes (Wolf et al., 2020), plant-associated ‘plastroviruses’ (Lauber et al., 2019), and a few arthropod-associated viruses (Shi et al., 2016a) (Figure S9A). These viruses share similar domains with astroviruses, including RdRP, trypsin-like proteases, and highly divergent SJR-CPs (Figure S9; Figures S8B and S10A, Astrovirus-like clades). Nineteen of the 46 plant-associated astro-like viruses passed the QC, including 14 from libraries with negligible metazoan reads (≤1%). Given that diverse plant-associated, astro-like viruses were found in 26 plant orders (Figure 2), they represent a widespread component of the plant holobiont, although definitive evidence of their reproduction in plants remains to be obtained.

Both potyviruses and a well-supported sister group that consists of a few viruses and EVEs from fungi dubbed “mycopotyviruses” encode a distinct CPf related to those of several Alsuviricetes families, particularly Alpha- and Betaflexiviridae (Figures 5A, S8 and S10B). Notably, potyviruses and their fungal ‘siblings’ also share S2H that is lacking from all other Stelpaviricetes (Figures 5A and S9A). Although the results of this analysis are preliminary due to the limited number of representative genomes in the mycopotyvirus group, they suggest a tentative evolutionary scenario according to which potyviruses evolved via HVT from plant-associated fungi. Under this scenario, mycopotyviruses already possessed S2H and CPf, thus enabling further complexification and plant adaptation of the ancestral potyviruses. This process involved acquisition of the potyvirus-specific genes encoding helper component-protease, P3, genome-linked virus protein (VPg) and other proteins and domains that function in virus arthropod transmission, cell-to-cell movement, RNAi suppression and other aspects of virus-plant and virus-vector interactions (Pollari et al., 2024).

Figure 5. Evolution of genome complexity of plant RNA viruses.

Phylogenomic analysis includes Stelpaviricetes (A), Kitaviridae, negevirus, and Clade M5 in Martellivirales (B), Benyviridae in Hepelivirales (C), Endornaviridae (D), and Rhabdoviridae (E). Phylogenetic trees are displayed on the left, with representative genome organizations on the right. Segmented or multipartite genomes are marked by dashed blocks, and clade-specific domains are annotated near each clade. Potyvirus-specific domains (e.g., HC-Pro, NIa-VPg, P3) are shaded in gray. Asterisks (*) and double vertical lines (||) denote stop codons and frameshifts, respectively. S1H, superfamily 1 helicase; Mtr, viral methyltransferases; SJR-CP, single jelly-roll CP; CPf, CP that forms filamentous virions; TMV-CP, tobacco mosaic virus-like CP; 30K MP, 30K superfamily MP; TGB, triple gene block; BMB, binary movement block; AlkB, Alkylated DNA repair protein; OTU, ovarian tumor-like cysteine protease; S2H, superfamily 2 helicase; SP24, putative virion membrane protein; FtsJ-Mtr, FtsJ-like methyltransferase; Macro, macrodomain; TM, transmembrane domain; N, nucleoprotein; P, phosphoprotein; M, matrix protein; G, glycoprotein; L, the large polymerase. Viral groups are labeled, with viral phyla abbreviated as follows: K., Kitrinoviricota; P., Pisuviricota; N., Negarnaviricota.

Flaviviridae.

Within the family Flaviviridae, we identified a distinct clade of plant-associated viruses that appears to be phylogenetically nested within the Jingmenvirus group (Shi et al., 2016b) (Figure 4L). Three of six plant-associated flaviviruses passed the QC, including one from a library with negligible metazoan reads (⩽1%). The phylogenetic placement of these plant-associated viruses suggests independent origins from arthropod-associated ancestors, expanding the known host range of flaviviruses to land plants.

Ghabrivirales.

The majority of dsRNA viruses in Chrysoviridae family (Duplornaviricota, Chrymotiviricetes, Ghabrivirales) infect fungi whereas some have been reported in plant (Kotta-Loizou et al., 2020; Li et al., 2013; Zhang et al., 2017) and arthropod (Shi et al., 2016a) hosts. We identified chrysoviruses in association with 9 land plant orders (Figure 2), including two bryophyte orders. The viruses in the recently reorganized Orthototiviridae family from the same order Ghabrivirales have been isolated from a variety of fungi (Sato Y, 2024). Here, we expanded their distribution to 15 previously unreported orders of green algae and land plants (Figure 2). Twenty of the 37 orthototiviruses and 15 of the 26 chrysoviruses passed the QC. Of these, 17 orthototiviruses and 10 chrysoviruses originated from libraries with negligible fungal reads (≤1%). In contrast, none of the 13 identified viruses of the strictly fungal/invertebrate family Pseudototiviridae passed the QC. Thus, the newly identified members of Orthototiviridae and Chrysoviridae are unlikely to represent uniform fungal contamination, supporting their association with the plant holobiont. Together with the growing evidence of plant-associated, disease-causing orthototiviruses, which form a clade within a greater diversity of fungal viruses (Alvarez-Quinto et al., 2024; Reyes-Proaño et al., 2024), these observations strongly suggest that orthototiviruses have adapted to replication in plants following HVT from fungi. Similar to the other atypical plant viruses considered above, neither chrysoviruses nor orthototiviruses encode MPs, suggesting that these are persistent, vertically-transmitted viruses. Alternatively, some of these viruses could hijack host mechanisms or use other virus proteins such as CPs to enable intercellular transport (Simon et al., 2024; Ying et al., 2024).

Evolution of genome architectures of plant RNA viruses

From the dawn of RNA virus phylogenomics (Ahlquist et al., 1985; Morozov et al., 1989; Shi et al., 2018; Shi et al., 2016a) to the development of the comprehensive framework of virus macroevolution (Harvey and Holmes, 2022; Koonin et al., 2022), it transpired that virus genome evolution is modular, occurring primarily via gain and loss of individual genes and gene blocks (Koonin et al., 2015; Krupovic and Koonin, 2026). To assess this concept in the face of the current major expansion of plant virus diversity, we compared genome architectures of known plant viruses and those discovered here.

First, we examined the evolutionary and taxonomic provenance of the main signature plant virus proteins, the MPs, using similarity network analysis (Figure S8A). The most broadly represented 30K MP superfamily occurred in orthornaviruses from Kitrinoviricota, Pisuviricota, Negarnaviricota, and Lenarviricota phyla, in accord with previous observations (Butkovic et al., 2023; Koonin et al., 1991). Notably, two large, tightly knit 30K MP communities (middle-top and right-top of Figure S8A) combined families from three phyla indicating pervasive horizontal transfer of MP genes, likely from the most broadly represented members of Kitrinoviricota to Pisuviricota and Negarnaviricota. Such transfer apparently continues to this day as illustrated by 30K MP acquisition in emerging clades of plant viruses where 30K MP gene is present in only a fraction of viruses, including Kitaviridae and the related Clade M5 (Figure S5) or Rhabdoviridae (Figure S7). The triple and binary gene block MPs (Morozov and Solovyev, 2022; Verchot-Lubicz et al., 2010) exhibited more limited distribution among viruses of multiple families within Kitrinoviricota (Figure S8A). However, it should be emphasized that the unrelated 30K and triple gene block MPs are functionally equivalent as illustrated by multiple replacements between these types of MPs in Virgaviridae, Alpha- and Betaflexiviridae (Figures S3, S5, and S8A).

Second, we explored, again, by similarity network analysis the phylogenomic distribution of several unrelated folds of another ubiquitous type of plant virus proteins, the CPs, using known and predicted structures. The SJR-CP that forms icosahedral capsids is the most common CP in plant viruses (and orthornaviruses in general) (Krupovic and Koonin, 2017), found in multiple families of Pisuviricota, Kitrinoviricota, including Clade M2 described here, and several unclassified families (Figure S8B). The SJR CPs of the diverse plant-associated astro-like viruses were distributed among four distinct communities of which three consisted of proteins that are only distantly related to the CPs of bona fide astroviruses, suggestive of multiple, independent acquisitions of the SJR CP genes within this virus lineage (Figures S8B, S9A, and S10A). The CPf is another capsid protein found in diverse plant viruses (Dolja et al., 1991) and particularly common in Kitrinoviricota (Figure S8B). Because Potyviridae and mycopotyviruses constitute a single clade of Pisuviricota encoding CPf, it seems likely that this gene was acquired horizontally from a kitrinovirus by a potyvirus ancestor (see above). Structurally, CPf is related to nucleocapsid proteins of some animal viruses of Negarnaviricota phylum such as phleboviruses (Agirrezabala et al., 2015; Zamora et al., 2017). Thus, CPf adoption by plant viruses likely involved at least two inter-phylum horizontal transfers of the CPf gene. The Kitrinoviricota order Tymovirales provides a particularly striking case of the CP diversity within the same lineage, where alpha-, beta-, and gammaflexiviruses encode CPf, whereas several plant-associated gammaflexiviruses encode unrelated TMV-like CP forming rod-shaped virions. Additionally, deltaflexiviruses, previously thought to be capsidless (Bartholomäus et al., 2017; Bejerman and Debat, 2022; Chen et al., 2016), and Tymoviridae encode SJR-CPs (Figures S3 and S8B). Thus, consistent with the modular nature of the orthornavirus genomes, both MPs and CPs exhibit extreme evolutionary mobility within the plant RNA virome and beyond.

Additional illuminating examples of gene capture by plant orthornaviruses via horizontal transfer include the RNA-demethylase AlkB domain that is encoded by Alphaflexiviridae, Betaflexiviridae, Closteroviridae, and Potyviridae (van den Born et al., 2008). Here, we identified AlkB domains in some plant viruses from Clade M5, Benyviridae (Figure 5B and C) and, unexpectedly, in arthropod-associated viruses of the Picornavirales Clade P1 (Figure S4). In the Endornaviridae, several plant and arthropod viruses acquired glycosyltransferase domains, whereas a lineage found primarily in plants also acquired a capsular polysaccharide synthesis domain (Figure 5D). Both of these domains appear to be most closely related to bacterial homologs (Song et al., 2013). Cysteine proteases of the ovarian tumour-like (OTU) superfamily were identified in Betaflexiviridae, Kitaviridae, and Clade M5 (Figures S3 and S5). In most of these cases, horizontal gene transfer occurred on multiple, independent occasions, resulting in only a fraction of genomes within each lineage possessing a new domain, in accord with a recently described common pattern in the kingdom Orthornavirae (Mutz et al., 2024).

Neofunctionalization following gene duplication is a special case of pervasive gene gain processes involved in RNA virome evolution. Within Clade M5 that is closely related to Kitaviridae, several plant-associated viruses encode three diverged copies of the TMV-like CPs (Figures 5B and S5). The triplication and neofunctionalization of CPf was previously observed in Closteroviridae where each of the divergent CPf copies is required for the virion morphogenesis and virus cell-to-cell movement (Figure S5) (Dolja et al., 2006; Liu et al., 2021), suggestive of analogous neofunctionalization of the TMV-like CPs in Clade M5. By contrast, arthropod-associated Clade M5 viruses encode either envelope glycoprotein or one or two copies of the TMV-like CP (Figures 5B and S11A). Along the same lines, both binary and triple gene block MPs encompass an S1H gene that apparently independently originated from duplication of the replication-associated S1H genes (Figure S11B). Furthermore, the 30K MP originally evolved via duplication of SJR-CP followed by neofunctionalization (Butkovic et al., 2023). Thus, gene duplication appears not to be as rare in orthornavirus evolution as previously thought (Simon-Loriere and Holmes, 2013).

Adaptation to plant hosts upon HVT can also result in gene loss. Such recurrent loss of genes encoding proteins with apparent specific roles in animals occurred in enveloped viruses that shed the glycoprotein (G) and/or matrix (M) protein genes. In particular, plant kitaviruses lost the G protein, whereas viruses in the related Clade M5 lost both G and SP24, another membrane protein (Figure 5B). Among the Rhabdoviridae members, some of the cytorhabdoviruses lost either G or both G and M (Figures 5E and S8) (Bejerman et al., 2023; Zhang et al., 2021).

Genome segmentation is a widespread feature of the plant orthornavirus genome architecture and an evolutionary process that can occur in both directions, fusion and fission. Typically, the genomes of viruses in Benyviridae contain two to five genome components encoding RdRP and TMV-like CP in different RNA segments. However, in this work, we identified many viruses closely related to benyviruses that encode RdRP and CP in the same RNA segment, suggestive of genome segment fusion (Figure 5C) (Solovyev and Morozov, 2022). The inverse transition from monopartite to multipartite genomes was observed among Kitaviridae, Secoviridae, and at least twice in Rhabdoviridae (Figures 5, S4, S5, and S7).

Taken together, our comparative analyses illuminate a range of evolutionary processes that drive the diversification of genome architecture in plant orthornaviruses, including gene gain via horizontal gene transfer or gene duplication, gene loss, and genome segmentation.

DISCUSSION

To investigate the composition of the global RNA virome of plants, we performed large-scale mining of transcriptomes and genomes covering nearly the entire diversity of the plant kingdom (97 orders of Archaeaplastida) including red and green algae, bryophytes, lycophytes, ferns, gymnosperms and angiosperms (Initiative, 2019). Our analysis resulted in a 2.2-fold increase in the diversity of known plant-associated orthornaviruses roughly at the genus level and revealed several previously uncharacterized virus clades at the family level, providing a rich resource for further inquiry into the biology and evolutionary history of plant orthornaviruses. Among the 22,750 discovered viruses, 125 were identified in crop plants although their prevalence and potential pathogenicity remain to be investigated.

This study validates, substantially expands and enriches the general concept of the origins and evolution of the land plant RNA virome (Dolja and Koonin, 2018b; Dolja et al., 2020). The central tenet of this concept is the origin of the majority of extant plant orthornaviruses via HVT from plant-associated invertebrates and fungi, with a limited contribution from the comparatively short-scale co-divergence of viruses with their plant hosts. Indeed, our similarity network analysis revealed distinct RNA virome compositions between land plants and their green algal ancestors. We identified only a few algal RNA viruses phylogenetically related to land plant viruses that likely coevolved with plants during the conquest of terrestrial environments. However, this observation must be interpreted with caution. Given the vast undersampling of the algal virosphere compared to land plants, the possibility remains that more extensive sampling of algae could reveal basal viral lineages that would refine our understanding of plant virome evolution. In contrast, a high degree of community sharing was observed between plant and arthropod viromes and to a lesser extent with the fungal virome. Phylogenetic analyses supported the major role of HVT in shaping the plant virome. The topology of the RdRP phylogenetic trees implies that the majority of plant RNA virus lineages emerged from within a greater diversity of arthropod or fungal viruses on numerous independent occasions within each of the five main phyla of the orthornaviruses. Although the exact number of the HVT events in the evolution of the plant virome is difficult to estimate due to the uncertaintyof some nodes in phylogenetic trees, tens of such events are traceable within the currently known viromes, and many more remain to be discovered. Thus, the barriers for interkingdom HVT appear to be much lower than previously anticipated.

The evolutionary transition of invertebrate viruses to replication in plants typically involved acquisition of the MPs and other functionalities involved in virus-plant interactions as well as frequent replacement of the capsid proteins with unrelated ones. As a result, many plant virus lineages possess elongated, filamentous or rod-shaped virions that are not known to exist or are rare among invertebrate viruses.

Arguably, the most unexpected outcome of this study that potentially changes the overall picture of the plant virome is the discovery of the previously neglected, substantial part of this virome that consists of atypical viruses lacking MPs and other known adaptations to reproduction in plant hosts. Similar to previously identified partitiviruses and mitoviruses, these are likely persistent cryptic viruses that provide a large pool for emergence of full-fledged, infectious viruses via acquisition of MP and additional plant-specific functions. Most of the cryptic virus clades, including narnaviruses (Lenarviricota), astrovirus-like (Pisuviricota), Clade M1 endornavirus-like (Kitrinoviricota), yuevirus-like (Negarnaviricota), and orthototivirus-like (Duplornaviricota) possess small genomes with minimal replication requirements that are amenable to gene gain and complexification. The phylogenetic placement of some of these highly divergent lineages carries a higher degree of uncertainty due to the inherent limits of deep sequence conservation. Nevertheless, their consistent association with plant transcriptomes, supported by independent lines of evidence, such as taxonomic profiling, points to a substantial ‘dark matter’ component of the plant virome. It seems most likely that this cryptic plant virome still remains severely undersampled, and its true dimensions remain to be elucidated through comprehensive transcriptome mining.

Furthermore, the rigid categorization of these cryptic or atypical clades as exclusively ‘fungal’ or ‘plant’ viruses likely represents a false dichotomy within the context of the plant holobiont. Recent experimental studies highlight the permeability of inter-kingdom barriers and the mechanisms driving HVT in nature. For example, it has been demonstrated that Cryphonectria hypovirus 1 (CHV1), a fungal virus, can systemically infect plants when complemented by the MP of a co-infecting plant virus, and that physical plant-fungus interactions mediate the horizontal transfer of this virus into plant tissues (Bian et al., 2020). Similar trans-kingdom virome sharing and trans-complementation have been documented for typical plant viruses, such as cucumber mosaic virus, that has been shown to replicate in phytopathogenic fungi, both in nature and in the laboratory (Andika et al., 2017). Therefore, although clades such as Ambsetviridae (Clade M1) or Jiviviruses have been documented in axenic fungal cultures (Deakin et al., 2017; Kinsella et al., 2022), their pervasive, high-confidence association with diverse plant transcriptomes in our study suggests they are not mere contaminants. Instead, they might be transitional or multi-host viruses that actively navigate the plant-fungal ecological interface. They likely constitute the dynamic evolutionary reservoir from which classical systemic plant viruses ultimately emerge via HVT, following the acquisition of autonomous movement capabilities.

During terrestrialization, plants were continually exposed to viruses of invertebrates and fungi, two types of organisms that developed close ecological associations with plants (Figure 6). Many of these viruses crossed interkingdom barriers through the evolutionary process that likely involved several overlapping steps: i) ability to reproduce in plant cells including gametes and thus to establish sustained infections via vertical plant-to-plant transmission; ii) ability to move from cell-to-cell by hijacking intrinsic mechanisms of plant intercellular communication (e.g., via host phloem proteins (Ying et al., 2024)); iii) capability of efficient systemic transport within plants via the emergence of dedicated MPs followed by pervasive horizontal MP gene transfer during virus coinfections; iv) capability of efficient plant-to-plant transmission either by retaining infectivity in their ancestral hosts (plant-associated invertebrates or fungi that become transmission vectors) or by evolving non-propagative transmission mechanisms mediated by virion-associated receptors and helper factors. This process of ‘becoming’ a plant virus seems to be proceeding uninterrupted to this day via evolutionary adaptation to plant hosts of the atypical, persistent viruses that constitute the ‘dark matter’ of the extant plant virome. Rather than a single linear trajectory, this adaptation likely encompasses divergent evolutionary strategies. Following initial cross-kingdom transfer, some lineages establish persistent, vertically transmitted associations, constituting the substantial ‘dark matter’ of the extant cryptic plant virome, which likely often represent evolutionary dead-ends. In contrast, other transitional lineages acquire MPs and vector-transmission mechanisms to become classical infectious agents.

Figure 6. Evolutionary scenario for the origins of the RNA virome of land plants.

Vast RNA virome that diversified in marine invertebrates during the Cambrian period is transferred to land together with more limited virome of fungi and emerging land plants during terrestrialization. Development of close ecological associations among invertebrates, fungi and plants results in gradual, multidirectional, cross-kingdom horizontal virus transfer (HVT) among invertebrates, fungi and plants. Formation of the extant global plant RNA virome occurs in four main steps: (i) A fraction of invertebrate and fungal viruses that entered plant cells through invertebrate-plant and fungus-plant interactions adapt to reproduction in these cells followed by evolving the ability to enter gametes and be vertically transmitted. (ii) Some of the viruses evolve to hijack host transport pathways expanding the extent of plant colonization. (iii) Emergence of specialized movement protein (MP) genes via gene acquisition, and neofunctionalization, and horizontal gene transfer facilitate cell-to-cell and systemic transport within plants. (iv) In parallel, efficient plant-to-plant transmission evolves either through maintenance of propagative transmission by ancestral vectors or via development of non-propagative transmission routes. Dashed arrows denote hypothesized, divergent evolutionary transitions rather than a strict linear progression, reflecting that some persistent viruses may represent stable evolutionary endpoints. The extant plant RNA virome encompasses fully host-adapted viruses as well as viruses on the evolutionary path to this state and the dark matter of persistent plant viruses from step (i). RdRP, RNA-dependent RNA polymerase; CP, coat protein. Icons were acquired from Bioicons and Freepik.

The framework to assess the confidence of the association between plants and viruses was intentionally set conservative, to minimize misassignment of viruses to plant hosts. Thus, some plant viruses are inevitably missed in this analysis. The conservative threshold notwithstanding, transcriptome analysis involves inherent uncertainty, and in many cases, contamination cannot be rigorously ruled out. Definitive host range assignment will require experimental identification of the respective viruses which can be highly problematic, especially, for non-model plants. In contrast to land plants, algae have been only sparsely sampled by both virus research and transcriptome analysis, and much of the available data for algae comes from laboratory cultures, in which natural viral diversity is likely to be underrepresented. Larger-scale studies might increase the fraction of plant viruses that were inherited from algal ancestors. Inference of HVT is prone to uncertainty due to the lack of confidence in the topology of parts of phylogenetic trees, particularly, for highly divergent viruses where standard sequence-based alignments approach the ‘twilight zone’ of sequence similarity. The directionality of some of the evolutionary processes described in this work is not always discernible with the available data. This uncertainty is well illustrated by the large Clade M1 of plant-associated viruses (Figure S5) whose minimal genomes could have evolved from larger monopartite endornavirus genomes by gene loss and genome segmentation. Alternatively, however, the more complex, larger endornavirus genomes could have emerged via gene gain and genome segment fusion from the Clade M1 virus genomes. The future discovery of evolutionary intermediates should help resolve this uncertainty.

METHODS

Virus discovery in land plant transcriptomes

Our analysis included 2977 transcriptomes of 1857 land plant species from three sources: assembled land plant transcriptomes from 1KP (Initiative, 2019), Transcriptome Shotgun Assembly (TSA) database, and transcriptomes de novo assembled in this study using SRA data from the NCBI online data repository. The information on all analyzed transcriptomes is provided in the Figshare repository (see Data and Code Availability). Given that it is computationally infeasible to assemble all the SRA data, we leveraged the palmprint (a short, well-conserved amino acid sub-sequence of RdRP) data from Serratus (Edgar et al., 2022) to strategically guide our de novo SRA assembly. We performed phylogenetic analyses of the palmprint sequences from Serratus and the full-length RdRP sequences from the NCBI Virus database using the approximate maximum likelihood method implemented in FastTree (Price et al., 2009). Based on the phylogenetic analyses, we identified and retrieved the SRA datasets of plants and nEEs, in which palmprint sequences fell outside (for plants) or are closely related and basal to (for nEEs) the diversity of well-established land plant virus clades. These SRA datasets were then subject to quality control using Trimmomatic-0.39 (Bolger et al., 2014) and de novo assembly using Trinityrnaseq v2.14.0 (Grabherr et al., 2011). Host sequences were removed using the diamond BLASTx program against the non-redundant (nr) protein sequences database, and rRNA sequences were removed by matching the IMG metarRNA data. The remaining sequences were filtered to retain only those at least 1,000 nucleotides in length. This step minimized false positives and ensured sufficient information for robust viral operational taxonomic unit (vOTU) definition and phylogenetic inference (Chen et al., 2022; Zayed et al., 2022), although it may have come at the cost of reduced viral diversity. The sequences were then clustered at 99% nucleotide identity using CD-HIT (Fu et al., 2012). The filtered sequences were translated in six frames and searched for RNA viruses using BLASTp (Altschul et al., 1990) with five iterations, reference protein sequences from the NCBI Virus database as the queries, and an e-value threshold of 10⁻⁵ as well as using hmmsearch (Finn et al., 2011) with Pfam profiles related to RdRP (Mistry et al., 2021) as the queries and an e-value threshold of 10⁻⁵. The virus-associated sequences identified in each iteration were screened to remove contamination. Specifically, when query sequences matched a wide range of non-orthornaviral eukaryotic sequences (tested using Diamond BLASTp against the nr protein database and InterProScan analysis), those results were removed, and the query sequences were not used in the next iteration.

The identified viral contigs were collected from different datasets and individually assembled using CAP3 (Huang and Madan, 1999), a sequence assembly program based on sequence overlaps, with an overlap length cut-off of 100 nucleotides and an overlap identity cut-off of 0.98. For putative multipartite and segmented viruses, different sequence segments were considered to originate from the same virus if: (i) the segments matched the same virus with similar identity value in a Diamond BLASTp analysis against the nr protein database, or (ii) the segments were consistently present across the majority of a virus clade. Viral singletons in terms of host groups in phylogenetic trees were considered to be likely contaminants and thus were excluded from further analyses. RdRP protein sequences that covered >70% of the conserved region of RdRP were retained for subsequent analysis. Sequence processing was performed using SeqKit (Shen et al., 2016). Computational analyses were accelerated using Parallel (Tange, 2011).

Assessing the confidence of plant-virus associations

We developed a quality-control framework to systemically assess the confidence of plant-virus associations (Figure S12). This framework consists of the following steps:

(i). Benchmark dataset and QC metrics:

We first curated a benchmark dataset of 1,450 trusted plant viruses with strong prior evidence of plant infection (e.g., from known plant virus families or those encoding MPs, Data S1). We then calculated two metrics for all viruses: plant contig ratio and viral abundance. Both metrics were significantly higher in the benchmark dataset compared to other putative plant-associated viruses (P < 0.05, one-sided Mann-Whitney U-test; Figure S1A), confirming their utility.

The plant contig ratio was defined as the proportion of plant-derived contigs in all assigned eukaryotic contigs in a transcriptome assembly. The sources were inferred based on the host taxonomy of the best hits generated through Diamond BLASTx search against the UniRef90 database. To mitigate the impact of reference database biases and the technical variance introduced by unassigned contigs (which predominantly represent host non-coding RNAs or short fragments), the denominator was restricted to assigned eukaryotic contigs. This enables a consistent comparison among the most likely sources of biological contamination (plants, fungi, metazoa, and protists). Viral abundance was calculated based on the RdRp-encoding contigs as mapped reads per million reads (RPM), normalized against the total number of non-ribosomal RNA reads in each library.

(ii). Filtering criteria:

Based on the distributions in the benchmark dataset (Figure S1B), we established two non-exclusive criteria for a virus to pass our QC filter. Criterion 1 (High-quality library): plant contig ratio > 89.26% (33rd percentile of benchmark) and Log10 RPM > 0.38 (10th percentile). Criterion 2 (High viral titer): Log10 RPM > 2.56 (75th percentile) and plant contig ratio > 81.27% (25th percentile). A virus was considered to have passed QC if it met either Criterion 1 or Criterion 2.

(iii). Application of the QC framework:

We applied the two filtering criteria to our entire dataset. For the benchmark set of 1,450 trusted plant viruses, 565 viruses passed only Criterion 1, 22 viruses passed only Criterion 2, and 324 viruses passed both criteria. In total, 911 (62.8%) benchmark viruses passed at least one criterion. For the 1,686 other putative viruses, 597 passed only Criterion 1, 1 passed only Criterion 2, and 82 passed both, for a total of 680 (40.3%) passing viruses.

This outcome reflects a deliberate trade-off in the design of our framework. We acknowledge that some bona fide plant viruses may appear at low titers (e.g., phloem-limited viruses) or in libraries with low plant contig ratios, making them computationally indistinguishable from contaminants based on these metrics alone. Therefore, we intentionally set stringent thresholds that accept a lower sensitivity (passing only 62.8% of trusted viruses) to achieve a higher specificity. By prioritizing the exclusion of ambiguous cases—even at the cost of excluding some genuine plant viruses from the benchmark set—we maximize confidence that the viruses passing our filter are strongly supported and represent reliable plant-virus associations.

For downstream analyses, we defined a high-confidence dataset comprising all 1,450 trusted benchmark viruses (which possess strong prior evidence of plant infection regardless of their individual transcriptomic metrics) and the 680 putative plant-associated viruses that successfully passed our stringent QC filter. Combined, these accounted for 67.92% of the total data. The remaining viruses were designated “putative,” reflecting their uncertain status while acknowledging their potential plant association.

To further address the complexity of plant holobionts and complement the contig-based metrics, we utilized SRA STAT (read-based taxonomic profiling) for the targeted exclusion of alternative hosts. We tailored this screening to the most biologically plausible confounder for specific viral clades based on their phylogenetic affinities. For example, we quantified metazoan reads for clades nested within arthropod-associated viruses (e.g., Clade M5, Dicistroviridae) and fungal reads for clades with known fungal associations (e.g., Narnaviridae). A library was considered to have negligible contamination from the targeted confounder if the relevant read proportion was ⩽1% (Data S1).

To empirically evaluate the filtering behavior of our criteria against environmental and holobiont sequences, we assessed the pass rates of known non-plant viral families detected in the initial dataset. Viruses from strictly fungal or invertebrate taxa, such as Pseudototiviridae (n=13) and Solinviviridae (n=4), were completely excluded (0% pass rate), as their presence typically coincides with sufficient non-plant RNA to lower the plant contig ratio below our thresholds. For clades with known fungal associations (e.g., Narnaviridae, Clade M1), the pipeline rejected 70.5% and 81.5% of the identified viruses, respectively. This retrospective evaluation confirms the pipeline’s capacity to filter out general background sequences, isolating viruses in purified plant environments for downstream SRA STAT validation.

Virus discovery in nEE transcriptomes and quality control

To elucidate the origins of land plant RNA viruses, we also mined RNA viruses in 5228 transcriptomes of nEEs from three sources: assembled transcriptomes from the 1KP project and the TSA database, and de novo assemblies of SRA data. As described above, we prioritized SRA libraries in which palmprints are closely related and basal to known or newly discovered land plant viruses. We included all available SRA datasets from algae. These nEE transcriptomes were subjected to downstream virus identification pipeline as used for the land plant data.

To ensure the reliability of host assignments for these nEE viruses, we also performed quality control based on the taxonomic purity of each source library. For major nEE groups that are phylogenetically distant from plants (fungi, arthropods, other invertebrates, and protists), a virus was classified as high confidence only if >80% of the eukaryotic contigs in its source library were annotated to the expected host lineage. As many algal transcripts share high sequence similarity with land plant genes, standard taxonomic annotation tools (like Diamond BLASTx against nr) often misclassify algal contigs as “land plant” simply because land plant databases are vastly larger and better annotated than algal ones. This creates a systematic artifact where a pure algal transcriptome might appear to have a high “land plant contig ratio”, potentially leading to false host assignment. Thus, a specialized criterion was applied to algal datasets. An algal virus was classified as high confidence if (i) the combined proportion of algal and land plant sequences in its source library exceeded 80%, and (ii) the proportion of algal sequences was greater than that of land plant sequences.

EVE identification

We mined 813 land plant genomes and 10,010 nEE genomes (see data and code availability) for the presence of EVEs. The eukaryotic genomes were retrieved from the NCBI Genomes database. These selected species cover the major diversity of eukaryotes. To ensure the genome quality, most of the genomes we used are reference genomes. These genomes were screened for EVEs using tBLASTn (Altschul et al., 1990) with reference protein sequences from the NCBI virus database as the queries and an e-value threshold of 10⁻⁵. Due to the fragmentary nature of EVEs, the identified EVEs were further assembled based on their physical positions.

Virus classification and genome annotation

To assign taxonomic affiliations, a clustering approach was used. First, our newly identified RdRP sequences were pooled with a reference dataset of eukaryotic viral RdRPs retrieved from the NCBI Virus database. An all-vs-all sequence similarity search was performed on this combined dataset using BLASTp (e-value cutoff ⩽ 1e-5). The BLASTp results were then used as input for clustering with the MCL (Van Dongen, 2008) algorithm (inflation factor = 2.0). Each resulting cluster was provisionally annotated based on the taxonomy of the reference viruses it contained. Finally, based on these annotations and their placement in the global phylogeny (Wolf et al., 2020; Wolf Yuri et al., 2018; Zayed et al., 2022), clusters were manually curated and merged into major viral groups (e.g., families or orders). For each viral group, RdRP sequences were aligned using MAFFT (Katoh et al., 2002). Viral genomes were annotated using InterProScan (Paysan-Lafosse et al., 2023). Protein structures were predicted using AlphaFold3 (Abramson et al., 2024) and Phyre2 (Kelley et al., 2015). Protein structures were aligned using DALI (Holm et al., 2023).

Virus diversity analysis

In this work, viruses are defined as novel when they do not fall within the diversity of existing, officially classified viral taxa (e.g., families or genera) or when they are identified in a new host species.

To assess the diversity of the viruses identified in this study, we assembled a dataset of known plant-associated RNA viruses (designated known dataset). This known dataset includes the Orthornavirae protein sequences whose hosts were curated to be land plants from the NCBI Virus portal (a curated interface to GenBank and RefSeq; May 2025 version) and the full viral dataset from Mifsud et al (Mifsud et al., 2022). For convenience, we used the total dataset to refer to the combination of the representative RdRP sequences identified in our study and the known dataset. To maintain methodological parity with the unfiltered reference databases, this total dataset includes all representative RdRP sequences identified in the plant transcriptomes, rather than only those passing the strict abundance and contig-ratio QC filters. To quantify diversity at different taxonomic levels, all RdRP sequences in both the “known” and “total” datasets were clustered using MMseqs2 (Steinegger and Söding, 2017) based on AAI thresholds of 0.3, 0.5, 0.7, and 0.9, with 80% coverage of short sequences (--cov-mode 5).

To validate the biological relevance of the operational AAI thresholds, their congruence with the ICTV taxonomy was quantified using the ARI on the curated plant virus dataset. For each AAI threshold (0.9, 0.7, 0.5, and 0.3), the resulting cluster assignments were compared against the ICTV taxonomic ranks (e.g., genus, family, order) using the adjustedRandIndex function in the R package mclust (version 6.1) (Scrucca et al., 2023). The adjusted Rand index (ARI) score measures the similarity between two clustering (Hubert and Arabie, 1985), with 1 indicating perfect agreement. This validation confirmed that the 0.5 AAI threshold corresponded well with the genus rank (ARI = 0.84) and the 0.3 AAI threshold with the order rank (ARI = 0.93), supporting their use as operational proxies for different taxonomic ranks.

Rarefaction and extrapolation curves were generated using the iNEXT package (Hsieh et al., 2016) to assess viral discovery relative to sampling effort. The iNEXT provides statistically unbiased estimators for interpolated diversity. For extrapolation beyond the reference sample size, the curve is statistically guided by an analytic estimator for the asymptotic richness of the assemblage (Chao et al., 2014; Hsieh et al., 2016).

The similarity network of RdRP was reconstructed based on the e-values estimated by the all-versus-all BLASTp using the igraph package (Csardi and Nepusz, 2006) with the e-value threshold of 10⁻¹⁰, identity threshold of 30%, and length coverage threshold of 100 amino acids. Sequences used for the construction of the similarity network of RdRP were initially clustered with 0.9 AAI based on different host groups, which were further clustered at 0.5 AAI. The number of 0.9 AAI sequences within each 0.5 AAI cluster was standardized and represented by the size of the nodes. Virus communities were identified in the RdRP similarity network using the random walks algorithm (Pons and Latapy, 2005). To test for sampling bias, we performed a stratified subsampling analysis by generating 100 replicate sub-networks, each balanced with an equal number of viral clusters from every major phylum. Modularity was recalculated for each replicate and compared to the original network to assess the stability of the community structure.

The similarity network of CPs was constructed using structures from the PDB and predicted models, based on the Z-scores from DALI all-against-all structure comparisons (Holm et al., 2023), with thresholds of 100 structurally equivalent C-alpha atoms (lali), Z-score threshold of 5, and initial sequence identity threshold of 15%. For singletons and isolated viral clades in the CP structure network, structural alignments with sequence identity below 15% were included. Similarity network of MPs was reconstructed based on the e-values estimated by five-iteration psiBLAST, with an e-value threshold of 10⁻⁵ and a length coverage threshold of 50 amino acids. The communities of CP and MP networks were detected using propagating labels algorithm (Raghavan et al., 2007) with modularity of 0.64 and random walks algorithm (Pons and Latapy, 2005) with modularity of 0.77, respectively.

The Bray-Curtis distance was calculated based on the presence of different communities in each eukaryotic group. NMDS analysis and PERMANOVA were performed based on the Bray-Curtis distance using the Vegan package (Oksanen, 2015). Data were visualized using the tidyverse and ggplot2 packages (Wickham, 2011) in R.

Phylogenetic analysis

Core sequences of RdRPs were aligned using MAFFT (Katoh et al., 2002) with the localpair algorithm. The edges of the alignments were manually trimmed to retain only the core regions of the RdRPs. Sequences with incomplete conserved RdRP motifs A-C were removed. The alignment was further refined using trimAl v1.2 (Capella-Gutiérrez et al., 2009). Phylogenetic analyses were performed using the maximum likelihood method implemented in IQ-TREE2 (Minh et al., 2020). The ModelFinder algorithm was used to select the best-fit substitution model (Minh et al., 2020). The phylogenetic analyses of Stelpaviricetes were inferred with site-heterogeneous CAT and LG+G model. Branch supports were estimated using SH-aLRT with 1,000 replicates. The phylogeny of RdRP from different clades was rooted using outgroup sequences chosen from closely related branches in the global RdRP phylogenetic trees (Wolf et al., 2020; Wolf Yuri et al., 2018). The phylogeny for TMV-like CP was rooted using the minimal deviation of ancestors (MAD) method (Tria et al., 2017). Phylogenetic tree and corresponding viral genomes were visualized using iTOL (Letunic and Bork, 2021).

Quantitative analysis of host-switching dynamics

We estimated host transition rates for each major viral lineage using a Bayesian approach implemented in BayesTraits v4.0 (Pagel et al., 2004). For the host state assignments, viruses that did not pass our quality control framework (from both land plants and nEEs) were treated as an uncertain host state (NA) and were pruned from the trees. To ensure comparability across lineages, all trees were internally scaled to a mean branch length of 0.1 using the ScaleTrees command. For each lineage, a reversible-jump Markov chain Monte Carlo (RJ-MCMC) analysis was run with an exponential hyperprior (RJHP exp 0 100). Each chain was run for a minimum of 200 million generations. A conservative burn-in of the initial 60 million of each chain was discarded. Convergence of each MCMC chain was confirmed by visual inspection of the trace plots for stationarity in Tracer v1.7 (Rambaut et al., 2018). A transition rate was considered statistically significant when its 95% highest posterior density (HPD) intervals did not overlap with zero. Detailed results for this analysis are provided in the online data repository (Data S2).

Statistical analysis

All specified thresholds for analyses involving sequence searches or alignment procedures are provided in the main text or in the Methods section.

DATA AVAILABILITY

All original data and results generated in this study are publicly accessible. Code: All custom analysis scripts used for virus discovery, QC of host-virus association, and evolutionary analyses have been deposited in a GitHub repository (https://github.com/liuqiyan67/Plant_virome/). Data: All viral genome sequences identified in this study have been deposited in the Third Party Annotation (TPA) Section of the DDBJ/ENA/GenBank databases under the accession numbers BK076013–BK082814 (BioProject PRJNA1420164), which will be released upon publication. For immediate access during the peer review process, the sequence datasets and intermediate files are also fully accessible on Figshare (https://doi.org/10.6084/m9.figshare.30688322). Extended results: Detailed results for the benchmark plant virus dataset (Data S1), transition rate analysis (Data S2), representative library taxonomic composition and validation metrics (Data S3), and a per-clade summary of host associations and quality control results (Data S4) are available in the 05_Extended_Results_and_Figures folder of the Figshare repository. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.