Survey of Early-Diverging Lineages of Fungi Reveals Abundant and Diverse Mycoviruses

Viruses are key drivers of evolution and ecosystem function and are increasingly recognized as symbionts of fungi. Fungi in early-diverging lineages are widespread, ecologically important, and comprise the majority of the phylogenetic diversity of the kingdom. Viruses infecting early-diverging lineages of fungi have been almost entirely unstudied. In this study, we screened fungi for viruses by two alternative approaches: a classic culture-based method and by transcriptome-mining. The results of our large-scale survey demonstrate that early-diverging lineages have higher infection rates than have been previously reported in other fungal taxa and that laboratory strains worldwide are host to infections, the implications of which are unknown. The function and diversity of mycoviruses found in these basal fungal lineages will help guide future studies into mycovirus origins and their evolutionary ramifications and ecological impacts.

are unknown. The function and diversity of mycoviruses found in these basal fungal lineages will help guide future studies into mycovirus origins and their evolutionary ramifications and ecological impacts. KEYWORDS mycovirus, dsRNA virus, Chytridiomycota, Blastocladiomycota, Neocallimastigomycota, Zoopagomycota, Mucoromycota, double-stranded RNA virus, mycoviruses F ungal viruses (mycoviruses) have been reported from all major fungal taxonomic groups (1,2). However, the overwhelming majority of mycoviruses have been identified in hosts belonging to just two phyla-Ascomycota and Basidiomycota (known together as "Dikarya")-though the kingdom Fungi is comprised of at least eight phyla (3)(4)(5) (Fig. 1). Known mycovirus infections in non-Dikarya are limited to an early report of virus-like particles from ultrastructure studies of Allomyces arbusculus (6) and sequence-based identification in Rhizopus oryzae (7), Rhizopus microsporus (8), six isolates of arbuscular mycorrhizal fungi (9,10), Umbelopsis ramanniana (11), and Entomophthora muscae (12,13). While it has become clear that mycoviruses are widespread, disproportionate sampling across the fungal kingdom has resulted in an incomplete understanding of mycovirus diversity and prevalence. By sampling unexplored and phylogenetically diverse fungal lineages, we predicted to find equally diverse viruses that could empower meaningful inquiries into the origins and ecological functions of mycoviruses.
The conventional approach to mycovirus discovery exploits the observation that double-stranded RNA (dsRNA) genomes predominate among known mycoviruses (1). The most common method uses cellulose chromatography to purify dsRNA from total RNA extracts and effectively isolate mycoviruses with dsRNA genomes and the dsRNA replicative intermediates of single-stranded RNA (ssRNA) genomes (14). This approach is quick and inexpensive; however, its exclusive use ignores DNA viruses and reinforces the bias that mycoviral genomes are predominantly composed of RNA. Furthermore, viral RNA is detected visually by gel electrophoresis (see Fig. 2), which could result in false negatives in instances of low-titer infections. To our knowledge, estimates of the proportion of false negatives based on chromatography have not been reported. We tested the relative accuracy of cellulose chromatography for identifying fungal isolates with viral infection and compared the results to more indirect sequence-based alternative approaches.
In silico approaches to mycovirus discovery have become more common (15)(16)(17). These approaches search transcriptomic data for sequences similar to RNA-dependent RNA polymerase (RdRp), which is the only conserved gene among RNA viruses and is diagnostic of infection. Conceptually, these approaches should be more sensitive to low titer viral infections because they do not rely on gel visualization and high transcriptome sequencing depth is readily achieved. Indeed, RNA sequencing has revealed cryptic coinfection in isolates previously thought to be singularly infected (18) (J. Myers, unpublished data). As sequencing costs have dropped, the availability of transcriptome (RNA) sequencing data has grown, which presents a novel opportunity to probe existing data sets to characterize viruses for the first time. Gilbert et al. (17) developed a data-mining pipeline for mycovirus identification, applied it to public transcriptomes of fungi in the subphylum Pezizomycotina, and discovered 52 novel mycoviruses, demonstrating the utility of this approach. However, such approaches have yet to be applied widely to non-Dikarya lineages.
Our primary goal was to improve understanding of prevalence and sequence diversity of RNA mycoviruses in the early-diverging lineages of Fungi, specifically Mucoromycota, Zoopagomycota, Chytridiomycota, Blastocladiomycota, Neocallimastigomycota, and Cryptomycota/Rozellomycota. We screened early-diverging lineages with both a culture-based chromatography approach ("in vitro") and a transcriptome data-mining approach ("in silico"). When possible, we compared methods by screening the same isolates used for transcriptome-generation with the in vitro method. In total, we screened 333 hosts, of which 72 (21.6%) harbored viruses. These results demonstrate that mycoviruses are abundant in the early-diverging lineages of the fungal kingdom, including in research laboratory cultures around the world.

RESULTS AND DISCUSSION
Viral prevalence. Through deep-sequencing approaches, recent studies have revealed the abundance of viruses in pathogenic fungal species, marine environments,  and soils (15,19,20). We build on this momentum by exploring viruses hosted in species from the deep branches of kingdom Fungi. In doing so, we more completely characterized the abundance of viruses across the kingdom and uncovered novel viral sequence diversity linked to specific fungal hosts. We determined the first sequences of mycoviruses in the particularly under-explored phyla Blastocladiomycota, Neocallimastigomycota, and Chytridiomycota ( Fig. 1). In total, we screened 333 fungi spanning six phyla by either or both in vitro (cellulose chromatography) and in silico (transcriptomemining) methods (see Table S1 in the supplemental material) and found one or more viruses in 72 isolates (21.6%), 65 of them not previously known as mycoviral hosts (Fig. 1). Of the 36 hosts for which we obtained viral sequence data-either by assembling from host transcriptomes or direct sequencing-we generated 154 complete or partial viral sequences which, using the criteria described above, we conservatively reduced to 85 unique viruses.
All sampled phyla were hosts to viruses except the poorly sampled Cryptomycota. Infection prevalence at the phylum level ranged from 15.9 to 40.0% and 0 to 89% at the subphylum and order levels but was highest in Glomeromycotina, Entomophthoromycotina, and Cladochytriales (Table 1 and Fig. 3). Glomeromycotina are also notorious hosts of endosymbiotic bacteria which may have originated by the invasion of freeliving bacteria following hyphal damage by herbivores or other fungi (21); it is possible that mycoviruses similarly originated in these fungi by invasion through damaged cell walls. In the zoosporic fungi, we predicted that polycentric organisms would be disproportionately virally infected since this mode should support viral transmission more frequently than monocentric growth. In monocentric development a zoospore encysts on a substrate and develops into a single zoosporangium-the structure which will ultimately cleave into many zoospores. Polycentric development, on the other hand, involves a single zoospore producing multiple zoosporangia with cytoplasmic continuity via aseptate rhizomycelium. Thus, a single viral infection in a polycentric organism could lead to many spores being infected, while monocentric development may allow for clearing of viral infection through selection between zoospores. Within the Chytridiomycota, the percentage of polycentric organisms that were viral positive (46.2%, n ϭ 26) was significantly greater (P Ͻ 0.00004, Fisher exact test) than monocentric organisms (9.2%, n ϭ 120). Ten of the twelve positive polycentric isolates were from the order Cladochytriales, however, and consequently we were unable to determine whether this trend is a result of the morphological trait or susceptibility traits that track the host phylogeny without additional data from non-Cladochytrialean polycentric organisms. Although the sample size is inadequate for definitive results regarding Neocallimastigomycota, viral infection did not favor polycentric over monocentric organisms in this phylum (polycentric n ϭ 4 and monocentric n ϭ 4, 25% of each was viral positive). Early in the study of mycoviruses, Bozarth (22) estimated 10 to 15% infection prevalence for fungal cultures predominantly made up of dikaryotic lineages but including some organisms in the Mucoromycota and Blastocladiomycota. A recent in silico survey of mycoviruses in Pezizomycotina (Ascomycota) revealed infection prevalence ranging up to 50% in some classes but an overall rate of about 8% (47/569) for the subphylum (17). Our estimate is approximately 22% (Table 1) across the earlydiverging phyla. Interestingly, we also found substantial variation across taxa (Table 1 and Fig. 3). Among the phyla with the highest infection rates, Mucoromycota and Zoopagomycota, viral prevalence at the subphylum level ranged from 19.7 to 88.9% (Table 1), a surprising difference from the 8% prevalence found for the same taxonomic level, Pezizomycotina, in Ascomycota. These data sets were not explicitly controlled for geographic origin of isolates, which could influence prevalence if, for example, some geographic regions had a higher proportion of viruses relative to others. However, the isolates in our study predominantly originate from distinct populations across the globe (see Table S1 in the supplemental material). Our results indicate higher prevalence in some subphyla of basal fungal lineages compared to Pezizomycotina. From our comparison of viral prevalence at the subphylum level between Pezizomycotina (Ascomycota) versus the early-diverging lineages, it is tempting to speculate on differing viral prevalence at the phylum level. However, a taxonomically thorough and geographically controlled study of viral prevalence in the other subphyla of Ascomycota and in Basidiomycota is needed in order to make a direct comparison.
In addition, anecdotal evidence suggests mycoviruses are commonly lost through culturing. Our estimates from the in vitro assays are thus perhaps underestimates of the true infection load in nature. Nonetheless, it is compelling to consider the implications of years or decades of stable maintenance of mycoviruses in culture on how the viruses affect the fungi. Future studies are needed to better characterize the effects of these mycoviruses on these phylogenetically diverse hosts.
Taken together, these findings indicate a strong phylogenetic component to viral prevalence. The earlier-diverging fungal lineages surveyed in this study typically share the trait of coenocytic mycelia, which may benefit the transmission of obligate symbionts and thus facilitate mycoviral infection; if true, it is possible that septa evolved, at least in part, as a viral defense mechanism to limit the spread of viruses throughout a mycelium.
Comparison of screening approaches. The results of screening the same isolate by chromatography and transcriptome-mining were mostly in agreement (i.e., at least one band of dsRNA was present by chromatography and RdRp sequence(s) were identified in silico or no dsRNA was present by chromatography, and no RdRp sequences were identified) (85.7%, n ϭ 21). Of the three isolates where the two methods varied in detection, two were found positive in vitro but negative in silico, and one with the inverse result. For Mortierella humilis (PMI 1414), where the in silico method revealed viral sequences but the in vitro method did not, RT-PCR confirmed viral presence in our culture. From the transcriptome data, we determined the virally derived sequences made up 2.58 transcripts per million. If we consider this a proxy of viral titer, the low viral abundance in the host likely accounts for the initial negative result obtained with the in vitro method. From Umbelopsis nana TLT204, which was positive by in-house chromatography but negative in silico, we sequenced purified dsRNA and assembled a toti-like viral contig containing a complete RdRp domain and an L-A virus major coat-protein domain, confirming a viral presence in our culture. Most likely, the virus was lost through subculture of this isolate in the laboratory in which transcriptome analysis was performed.
Paired comparison of screening approaches suggests minimal discrepancies, but more viruses overall were detected by the in silico method (33.3% positive) than in vitro (17% positive). The in silico method is likely to be more sensitive than the in vitro method, but it is possible that the higher virus detection rate in silico was a result of phylogenetic skews of the data sets rather than methodology. Approximately 73% of the cultures screened in vitro were in the Chytridiomycota or Blastocladiomycota, the lesser-infected phyla. Accordingly, 64% of the isolates screened in silico were in the Mucoromycota or Zoopagomycota, the greater-infected phyla. Thus, we recommend screening by transcriptome-mining when the data are available, but the chromatography approach is a good alternative for very cost-effective initial screening. Mycoviral diversity. We found mycoviruses in all five "branches" of the Riboviria (RNA virus) tree published by Wolf et al. (23) (Fig. 4 to 8). Many of these newly found viruses will need to be described as new taxa at the levels of genus, family, and order. In addition, some viruses, including a group of related, endogenous virus-like sequences, represent novel diversity such that they were unable to be assigned to a branch, though they show sequence similarity to other unplaced mycoviruses (Fig. 9).
Coinfection of a single host by multiple viruses was very common (mean and median number of viruses per host ϭ 2.4 and 1.5, respectively). Even after filtering by our conservative criteria, a single host, Kickxella alabastrina (Kickxellomycotina), was found to harbor at least 11 unique viruses. Such a result would be unsurprising in macrobes, such as humans but is staggering for a microfungus.
(i) Riboviria: "branch 1." Branch 1 (Fig. 4) is composed of positive-sense ssRNA viruses, including the bacterium-infecting Leviviridae, the mitochondrion-infecting mitoviruses (Narnaviridae), and the cytoplasmic narnaviruses (Narnaviridae) and ourmiaviruses (Botourmiaviridae). In total, 22 viruses were assigned to branch 1 based on blast similarity. A complete RdRp gene from the transcriptome of Chaetocladium brefeldii was most similar to viruses in family Leviviridae. These viruses are currently only known from bacteria, so it is possible, and perhaps likely, that this contig derives from the viruses of a bacterial endosymbiont of C. brefeldii.
Interestingly, the new viruses in this group are mostly hosted by fungi in the Mucoromycota or Zoopagomycota. Previous evidence suggests the origin of mitoviruses was a common ancestor of Mucoromyota and Zoopagomycota, which was followed by cospeciation of host and mitoviruses and horizontal transmission to plants (13,24). Our findings support this hypothesis, but sequence data are still limited for viruses in Chytridiomycota, Blastocladiomycota, and Neocallimastigomycota. Additional sequencing of the viral positives in those groups found by chromatography in this study will more strongly support or refute this hypothesis.
(ii) Riboviria: "branch 2." The picornavirus supergroup makes up branch 2 (Fig. 5), which also includes Partitiviridae, Amalgaviridae, Barnaviridae, and Potyviridae. In all, nine viruses were assigned to branch 2 based on BLAST similarity. The hosts of these contigs represent the Mucoromycota, Zoopagomycota, Neocallimastigomycota, and Chytridiomycota. A virus of Entomophthora muscae nested within the picorna-like viruses, specifically in the genus Iflavirus, which is only known to infect insects. Thus, it may have originated by cross-kingdom horizontal transfer. A virus of the arbuscular mycorrhizal fungus Rhizophagus irregularis (Glomeromycotina) is nested within the plant-infecting Potyviruses and likely also arose by horizontal transmission given the tight mutualism and nutrient exchange between R. irregularis and its plant hosts.
(iii) Riboviria: "branch 3." Branch 3 (Fig. 6) includes the alphavirus and flavivirus supergroups. In total, six contigs were assigned to this branch based on BLAST similarity. We found three new tombusvirus-like viruses, hosted by Mortierella selenospora (Mucoromycota), Syncephalis fuscata (Zoopagomycota), and Anaeromyces sp. (Neocallimastigomycota). Two arbuscular mycorrhizal fungi, R. irregularis and Paraglomus brasilianum, contained viruses related to the predominantly plant-hosted viruses in Virgaviridae. Another Glomeromycotina fungus, Geosiphon pyriformis, was host to a tymo-like virus positioned basally in the order Tymovirales and may represent a new family in this order.
(iv) Riboviria: "branch 4." The largest number of new contigs were assigned to branch 4 (Fig. 7), which includes Totiviridae, Chrysoviridae, Quadriviridae, and Reoviridae. In total, our analysis assigned 41 contigs to this group. Hosts are largely represented by Chytridiomycota and Blastocladiomycota, including substantial coinfection within the same individual, but also by Mucoromycota and Zoopagomycota.
Ten of these viruses group within the genus Totivirus, and three new viruses grouped within the genus Victorivirus, both genera belonging to the family Totiviridae. By our analysis, these two genera are composed of fungal viruses truly spanning the Kingdom, including hosts in Chytridiomycota, Blastocladiomycota, Zoopagomycota, Mucoromycota, Ascomycota, and Basidiomycota. This finding may align with the cospeciation hypothesis for this viral family: Göker et al. presented evidence that Totiviridae speciate through codivergence with their hosts (25). The predominance of totiviridae-like sequences in early-diverging lineages supports this hypothesis and suggests that an early fungal ancestor harbored ancestors of Totiviridae. However, this origin hypothesis does not preclude occasional horizontal transfer, many instances of which are suggested by our phylogenetic tree. Further, this viral family also include two genera, Leishmaniavirus and Giardiavirus, that infect protozoa; since protozoans and Fungi are polyphyletic, ancient horizontal transfer between host groups is likely.
Basal to Victorivirus, a clade formed including viruses from K. alabastrina, Operculomyces laminatus, the ampicomplexan Eimeria tenella, and multiple-diatom colonies. This finding aligned with our prediction of close-relatedness between formerly reported diatom-associated viruses and the newly reported sequences from Chytridiomycota, a group that contains diatom parasites, but lacked bootstrap support. Viruses from multiple isolates of Allomyces sp. form a clade sister to Botybirnavirus, likely a new Mycoviruses in Early-Diverging Fungi ® genus. Generally, many of the new sequences in this branch represent novel diversity that will require deeper viral characterization to phylogenetically classify with confidence.
(v) Riboviria: "branch 5." We identified just one viral contig in branch 5, which includes the negative-sense ssRNA viruses Mononegavirales and Bunyavirales (Fig. 8). The virus found in Mortierella minutissima is most closely related to the undersampled yueviruses known from insect hosts. Viruses with fungal hosts have not previously been (vi) Riboviria: unassigned. Some viral contigs either did not have significant BLAST hits or hit to known viruses currently unassigned to a taxonomic grouping (n ϭ 12) (Fig. 9). These new viral contigs appear to be related to unplaced "bipartite mycoviruses" and Curvularia thermotolance virus, known to form a tripartite mutualism with its fungal host and a plant (26). Among these viruses we uncovered is a novel lineage found endogenous in hosts' genomes. Intriguingly, we identified a highly supported clade of virus-like sequences only in the genomes of fungi in the Mucoromycota, including two species of Phycomyces, Dissophora ornata, Lobosporangium transversale, and multiple Mortierella spp. (Fig. 9). Conserved RdRp domains were identifiable in all nine of these sequences; all lacked specific hits to reverse transcriptome domains and thus are not believed to be retroviruses. Rather, the phylogenetic conservation of these endogenous virus-like sequences suggests that, before the divergence of Mucorales and Mortierellales, an endogenization event, in which a viral gene was reverse transcribed and integrated into the host genome, occurred that was conserved in members of Phycomycetaceae and Mortierellales. Integration of viral genes into host genomes is a well-known phenomenon generally, and the transfer of genes from dsRNA viruses to Fungi has been reported by others (27)(28)(29). We identified the new sequences via their mRNA transcripts with our in silico approach, which confirms their expression. Thus, this previously unreported instance of viral endogenization seemingly resulted in a novel fungal gene, the function of which remains to be tested.
Conclusions. The results of this study provide a new perspective on mycoviral prevalence in Fungi and the phylogenetic diversity of viruses. The results also aid in the identification of hosts of environmentally derived virus samples. For example, many of our mitovirus-like viral contigs appear to be related to Mitovirus species from soils; we suggest that these soil-derived mitoviruses are likely hosted by fungi in Mucoromycota.
The ecological implications of viruses in these deep branches of the fungal tree are currently only a matter of speculation, but their role in natural ecosystems may be of great importance. The host organisms studied embody a broad diversity of ecological niches including saprotrophs, plant mutualists, obligate and opportunistic pathogens,  By searching unexplored and underexplored fungal lineages, we uncovered novel mycoviral diversity and discovered that fungal viruses are indeed ubiquitous throughout the fungal kingdom, detected now in nearly every phylum. Our data suggest that early-diverging lineages may harbor greater viral prevalence than the Dikarya, but there is wide variation across lineages. A caveat of comparisons across major taxonomic groups is that without broad and deliberate taxonomic and geographic sampling, comparison of rates of infection are subject to high error in estimation. Further, by searching publicly available transcriptomes, as well as cultures from collections that are distributed to researchers globally, we learned that mycoviruses are abundant in research organisms used in laboratories worldwide. Mycoviruses are known not only to be persistent and often asymptomatic but also to cause variable phenotypic alterations such as in pigmentation, growth rate, and virulence. The implications for cryptic mycoviral infection in laboratory cultures are currently unknown but provide a guide for future studies into mycovirus origins and ecological functions.  Table S1 in the supplemental material). We grew isolates in media appropriate for their nutritional needs until sufficient biomass accumulated (3 days to 2 weeks, depending on the species), harvested tissue, and ground it by sterile mortar and pestle in liquid nitrogen. With every batch of fungi screened, we harvested and processed a mycovirus-infected strain of Ustilago maydis as a positive control for degradation of mycoviruses by RNases.

FIG 9
Maximum-likelihood tree of new mycovirus RdRps with top blast hits (included in tree) unassigned by current viral taxonomy. The best model of amino acid substitution for this model was determined to be LG ϩG according to Prottest v3.4. All viruses have fungal hosts. New sequences are indicated by purple tip labels. Blue circles indicate nodes with Ն70% bootstrap support. Triangles indicate novel virus-like sequences determined to be endogenous in the host genome. A DNA-based genome for Mortierella elongata has yet to be sequenced, and so we cannot conclude that Mortierella elongata UA virus is endogenous, although it appears likely.
Mycoviruses in Early-Diverging Fungi ® (ii) Column preparation and dsRNA extraction. We screened cultures for RNA mycoviruses by dsRNA extraction and purification by cellulose chromatography as described by Okada et al. (30) with slight modifications. Before RNA extraction, we prepared columns by piercing the bottom of a 0.5-ml tube with an 18-gauge needle, packing it with ϳ90 mg of cellulose D powder (Advantec), placing it in a 2-ml microcentrifuge tube, and adding 400 l of freshly prepared 1ϫ sodium chloride-Tris-EDTA (STE) with 16% ethanol. Immediately before use, we centrifuged the columns briefly and discarded the flowthrough. We extracted RNA by adding one ml of TRIzol reagent (Invitrogen) to ϳ0.5 mg finely ground frozen tissue and either freezing at -20°C for later processing or incubating for 10 min at room temperature, adding 200 l of chloroform, followed by mixing by inversion, incubation for 3 min at room temperature, and centrifugation at 12,000 ϫ g at 4°C for 15 min. We made a 16% ethanol solution with the supernatant, added it to the cellulose column, collected and discarded the flowthrough, washed the column three times with 400 l of 1ϫ STE with 16% ethanol, thoroughly dried it by centrifugation, and eluted the columns with 400 l of 1ϫ STE. We precipitated dsRNA by adding 40 l of 3 M sodium acetate and 1 ml of ethanol, centrifugation at 15,000 ϫ g for 5 min, pipetting off the supernatant, and allowing the tubes to dry before reconstitution with water. We treated samples with S1 nuclease and DNase 1 according to the manufacturer's instruction before visualization using agarose gel electrophoresis. We considered samples positive if a band was visible (Fig. 2).
In silico screening. We obtained unassembled RNA-Seq data from the SRA database (see Table S2 in the supplemental material), stringently filtered raw reads for quality (minimum [min] phred score 20) using the fastxtoolkit (31), and assembled contigs de novo with Trinity assembler (32). We predicted open reading frames (ORFs) with Transdecoder (33) using default parameters and queried the protein translations with hmmscan (34) against a custom RdRp HMM database. We constructed the RdRp database as in Gilbert 2019: we downloaded entire alignments of Pfam families RdRp_1, RdRp_2, RdRp_3, RdRp_4, and Mito_RdRp and generated HMM profiles from each using hmmbuild (HMMER2; hmmer.org). We further queried each RdRp profile hit found in the transcriptome ORFs with TBLASTN and BLASTP to the NCBI nt and nr databases (downloaded 18 September 2019), respectively, and considered isolates positive for viral infection if the resultant hits were viral sequences with an E value Ͻ e Ϫ10 . To ensure the virus was exogenous to the host genome, we blasted viral contigs against the hosts' genome when available or, if this was unavailable, the genome of the closest relative.
To improve assemblies, we used the nucleotide sequence of the Trinity-assembled viral contig as the seed for contig extension with PRICE (35) with the parameters of a minimum 30-nucleotide overlap for mini-assembly, a minimum 80% identity for contig-edge assembly, a 90% identity to starting contigs, and for 10 cycles, using loosely filtered raw reads (min phred score 5) (36). This most often resulted in contigs that were unchanged after 10 cycles, which we considered complete. If contigs were updated in the procedure, we ran 10 additional cycles with the updated contig as the starting seed.
As a final quality control check of our virus genomes, we reassembled the viromes of a subset of isolates by first aligning loosely filtered raw reads (min phred score 5) to a reference genome using STAR (37), assembled unmapped reads de novo using Trinity and then continued the pipeline exactly as described above. In all cases, at least one viral contig was extended by this method, but the RdRp region was most often unchanged. For additional informatics details and code, see https://github.com/jimyers/ Mycoviruses-in-early-diverging-fungal-lineages.
Mycovirus sequencing. (i) Pacific Biosciences sequencing. We prepared purified dsRNAs obtained from in vitro screens of Allomyces sp. strain JMM01, Allomyces sp. strain DJ-02, Allomyces sp. strain DJ-07, Zopfochytrium polystomum WB228, and Ustilago maydis (as a control) as described by M. J. Roossinck in 2010 (43) with slight modifications. Briefly, we purified dsRNAs by cellulose chromatography as described above, except the final elution was performed with 20 l of 2ϫ STE (pH 8.0). We mixed 1 l of dsRNA with 7 l of H 2 O and 2 l of tagged random 12-mer at 20 M (5=-ACCTTCGGATCCTCC-N12-3=), placed the tube in boiling water for 2 min to melt strands, immediately quenched the sample on ice, used the Omniscript reverse transcription kit according to the manufacturer's protocol, incubated the tubes on ice for 10 min followed by 60 min at 50°C, removed the unreacted template with 2 l of RNase A (5 mg/ml), and then incubated the sample at room temperature for 15 min, followed 3 min at 80°C to denature the enzymes. We used a Qiagen PCR purification kit according to the manufacturer's protocols to clean up PCR products and amplified them by PCR with individually barcoded primers using GoTaq polymerase, followed by gel extraction using a Qiagen gel extraction kit and the manufacturer's protocol. For each isolate, we sequenced 120 ng of product on a PacBio RSII at the University of Michigan Advanced Genomics Core (UM-AGC). We analyzed reads with SMRT Portal (parameters: minimum barcode score 22, minimum of 5 passes, minimum 90% accuracy) and removed adapters with cutadapt, followed by read correction, trimming, and assembly with Canu (v1.3) (parameters: minOverlapLength ϭ 100, minRead-Length ϭ 150, errorRate ϭ 0.035, estimated genome size of 10,000 bp). For quality control, we compared a U. maydis contig to the GenBank U. maydis H1 cap-pol fusion nucleotide sequence (NC_003823; fungal isolate unknown) which matched with a 99% query cover and a 93% identity.
(ii) (ds)RNA-Seq and total RNA-Seq. We submitted purified dsRNAs from in vitro screens of Cladochytrium sp. JEL861, Allomyces arbusculus North Carolina 2, Umbelopsis nana TLT 204, and U. maydis (as a control) to the UM-AGC for 2 ϫ 150 sequencing on Illumina MiSeq with the according to modifications to the manufacturer's instructions for library preparation as described by Sasai et al. (38): fragmentation was conducted with 87.5 ng of purified dsRNA for 20 min at 94°C in first-strand synthesis buffer with random primers, and the first-strand synthesis reaction was conducted for 30 min at 42°C (38). We submitted total RNA extractions of Allomyces sp. JMM01 and Z. polystomum WB228 to the UM-AGC for library preparation using the Illumina TruSeq stranded mRNA protocol and 2 ϫ 50 sequencing using an Illumina HiSeq 4000. All RNA-Seq was processed to remove adapters and low-quality sequences from paired-end data using trimgalore (min quality threshold 5) (https://github.com/FelixKrueger/TrimGalore), followed by assembly de novo with Trinity. We performed ORF prediction, HMM queries for RdRp homologs, and PRICE extension as described above.
For transcriptomes sequenced by JGI and not published previously (see Table S2), stranded cDNA libraries were generated using the Illumina Truseq stranded RNA LT kit. mRNA was purified from 1 g of total RNA using magnetic beads containing poly(T) oligonucleotides, fragmented, and reverse transcribed using random hexamers and SSII (Invitrogen), followed by second-strand synthesis. The fragmented cDNA was treated with end-pair, A-tailing, adapter ligation, and 8 to 10 cycles of PCR. The prepared library was quantified using KAPA Biosystem's next-generation sequencing library qPCR kit and run on a Roche LightCycler 480 real-time PCR instrument, multiplexed with other libraries, and the pool of libraries was then sequenced on an Illumina platform (HiSeq 2000/2500 or NovaSeq) following a 2 ϫ 150 indexed run recipe.
Comparison of screening approaches. One isolate, Mortierella humilis PMI 1414, produced negative results by in vitro virus screen but positive results by the in silico method. We designed virus-specific primers based on the in silico results and conducted RT-PCR using SuperScript IV reverse transcriptase (Thermo Fisher) according to the manufacturer's protocol. For a proxy of viral titer, we calculated the abundance of viral transcripts generated in the transcriptome with the align_and_estimate_abundance.pl script of the Trinity package (https://github.com/trinityrnaseq/trinityrnaseq), employing bowtie for alignment and the RSEM estimation method.
Mycovirus sequence analysis and phylogenetics. We classified each new mycoviral sequence by top blast hit and grouped them into clusters corresponding to "branches" described in the most recent RNA virus phylogeny (23). ORFs, including each RdRp gene, were predicted and translated using the universal genetic code except for sequences with top blast hits to mitoviruses, for which we used the standard mitochondrial genetic code.
Despite our best attempts to resolve assemblies, some remained fragmented. To avoid overreporting, we applied conservative criteria to limit the number of viruses reported for each isolate in a biologically relevant manner. The criteria for inclusion in our analyses were as follows: if only one viral contig was identified per strain per branch, we included it. For branches with more than one viral contig per isolate, all contigs with Ͼ60% coverage of the RdRp conserved domain were included, or if all contigs for that branch contain Ͻ60% of the RdRp, then we kept the contig with highest coverage. Three isolates had only two viral contigs assembled, one of which contained the C terminus of the RdRp and the other contained the N terminus (B. trispora, M. verticillata, and R. intraradices). In these three instances, we concatenated the two contigs.
For each branch, we inferred RdRp gene trees with viral contigs that met our criteria, their top BLAST hits, and reference sequences (Table S3). A sixth tree (Fig. 9) includes sequences with BLAST similarities to viruses currently unassigned by viral taxonomy. We aligned sequences with MAFFT version 7 using the E-INS-i algorithm (39) and trimmed the resulting alignments using the -automated1 method in TrimAl (40). We determined the best-fit model of amino acid substitution for each alignment with ProtTest 3.4 (41) and reconstructed trees with the maximum-likelihood approach implemented in RaxML by the rapid bootstrap analysis (-f a) with 100 replicates (42).
Data availability. The sequences generated for this study can be found in GenBank BioProject PRJNA657856. Sequence alignments, HMMs, and code can be found on github at https://github.com/ jimyers/Mycoviruses-in-early-diverging-fungal-lineages.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.