Environmental Viral Genomes Shed New Light on Virus-Host Interactions in the Ocean

1Viruses outnumber microbes such as bacteria in the oceans, and the destructive lytic infections caused by viruses are thought to have crucial effects on energy and nutrient cycles driven by marine microorganisms. Genomics-based research has been a powerful approach used to clarify the biology of viruses, including their infection strategies as well as their ecological significance. However, the diversity of viral genomes is still underrepresented in publically available genome databases. For example, SAR11 (Pelagibacterales) and SAR116 are major marine prokaryotic components, but only four and one phage genomes have been sequenced for these bacteria, respectively. Cyanophages, for which about 100 genomes have already been characterized, are the sole exception.

2To address the issue of the paucity of viral genomic data, Roux et al. analyzed publicly available prokaryotic genome sequence data to mine marine and nonmarine viral genomes that have been sequenced along with the genomes of their hosts. They identified 12,498 viral DNA sequences (either long fragments or whole circular genomes) representing 264 predicted new genera.

3Culture-independent viral metagenomics is also an effective research option for analyzing viral genomes in complex marine microbial communities. A decisive advantage of viral metagenomics stems from the small genomes of viruses. Viral genomes have so far been assembled from the metagenomes of the following viral types: RNA viruses, single-stranded DNA (ssDNA) viruses, and double-stranded DNA (dsDNA) viruses. Among these viruses, the genomes of dsDNA viruses have been the most difficult to assemble from metagenomes because of their relatively large genomes. However, recent advances in the construction of libraries, sequencing technologies, and bioinformatics software have resulted in the generation of larger assemblies. For example, 7 complete dsDNA viral genomes have been reported for a hypersaline lake, 18 for the deep-sea hydrothermal vent plumes, and 54 for glacial cryoconite holes. An interesting approach involved the construction of metagenomic fosmid libraries from virus-infected prokaryotes, which revealed 1 and 42 complete viral DNA genomes for solar salterns and 208 marine tailed-phage genomes. These studies indicated that marine viral metagenomics investigations have advanced from focusing on environmental genetics (i.e., collections of genes) to analyzing environmental genomics (i.e., collections of complete genomes), helping to unveil the evolutionary histories, life cycles, and metabolic strategies of individual viruses. In this study, we analyzed nine novel marine viral metagenomes (i.e., viromes) generated using a benchtop Illumina/MiSeq sequencer as well as previously published large-scale viromes. We identified 1,352 nonredundant complete viral genomes, the vast majority of which corresponded to previously unidentified viral lineages.

4We generated nine viromes (Osaka Bay viromes [OBVs]; 8.5 M read pairs; 2.4 Gbp) from water samples collected over a 24-h period in Osaka Bay, Japan (see Materials and Methods). We first compared four assemblers (SPAdes, metaSPAdes, IDBA-UD, and Ray Meta) regarding their ability to assemble viromes. SPAdes, metaSPAdes, and IDBA-UD clearly outperformed Ray Meta in terms of the total size of >10-kb contigs (Table 1). Of the first three assemblers, SPAdes (11.9 Mb) produced the largest assemblies (i.e., metaSPAdes, 6.8 Mb; IDBA-UD, 5.3 Mb). Regarding assembly error rates assessed by REAPR, SPAdes (8.48 regions/kb), metaSPAdes (8.73), and IDBA-UD (8.80) had similar error rates, which were slightly higher than that of Ray Meta (6.42). Most (99.97%) of these errors were short insertion/deletions (REAPR type 1 and type 3 errors), while there were very few (0 to 0.00662 regions/kb) scaffolding errors (type 2 and type 4 errors) (Table 1). On the basis of these results, we chose SPAdes as the best assembler for the following analyses.

5Given that the nine samples were collected at the same location over a short period and that the reads were relatively long (i.e., 2 × 150 or 2 × 300 bp), a coassembly consisting of the pooled nine samples was also prepared. The coassembly resulted in 879 contigs (>10 kb) that likely originated from dsDNA viruses (see Materials and Methods). Of these, 46 (28.5 to 192 kb; average, 54.2 kb) were assembled in a circular form (see Fig. S1 in the supplemental material). Thus, we refer to these 46 contigs as environmental viral genomes (EVGs).

6The EVGs did not contain any scaffolding errors (REAPR type 2 and type 4 errors), indicating high structural integrity for the contigs. To further assess the integrity of these EVGs, we mapped the contigs assembled from individual viromes on the EVGs. Of the 46 EVGs, 16 were totally covered by the contigs from individual assemblies, thus decreasing the possibility of artefactual chimeras due to coassembly for these 16 EVGs. The remaining 30 EVGs contained 1 to 24 regions (229 in total) that were supported only by coassembly and were not observed in the individually assembled contigs. We randomly selected 21 such weakly supported regions and tested the coassemblies by PCR assays (using the environmental DNA samples as a template) and sequencing. The results verified all of the tested regions of the coassembled contigs (Fig. S2A). Furthermore, 18 of the 46 EVGs exhibited complete or nearly complete genomic colinearity with closely related reference genomes (Fig. S2B; see Materials and Methods for the definition of genomic colinearity) or with the other independently determined EVGs described below (Fig. S2C). These results further corroborated the accuracy of the overall structure of the EVG assemblies.

7Each of the individual EVGs likely corresponds to genomes of closely related viruses because the sequence assemblies were obtained from environmental viral populations. To assess the genetic diversity of each EVG, we analyzed single nucleotide polymorphisms (SNPs) and calculated the nucleotide diversity of each EVG. Nucleotide sites containing SNPs that were supported by at least one read were present in genomes at a rate of 0.558 to 7.897% (median, 2.473%) (see Table S1A in the supplemental material). The nucleotide diversity of EVGs was 0.073 to 1.734% (median, 0.423%). These results are within the ranges for genomes from the same viral species. We conclude that each of the EVGs represents a consensus genome of a viral species.

8Prompted by the detection of 46 OBV-EVGs in a modest sequencing effort, we applied our genome assembly and complete genome identification protocol to the Tara Oceans viromes (TOV), which consist of 43 viromes representing 26 oceanic locations. Given the wide geographic areas and seasons covered by these samples and the large volume of sequence data for individual TOV samples (i.e., average, 50 M reads; 2 × 100 bp), we assembled these 43 viromes individually. We obtained 1,554 TOV-EVGs (i.e., circular complete contigs, 10 to 211 kb) with a predicted viral origin. Only 64 were detected as complete in the previously reported original TOV assemblies, and 85.6% of the remaining EVGs (i.e., 1,275 EVGs) were detected in the original assemblies as smaller contigs with less than half the size of the contigs in these new assemblies. Clustering on the basis of the nucleotide sequence identity among the OBV-/TOV-EVGs resulted in 1,352 nonredundant complete genomes (i.e., 46 OBV-EVGs and 1,306 TOV-EVGs).

9After discarding possible eukaryotic virus genomes, we obtained 1,567 complete genomes that were likely of prokaryotic dsDNA viral origin (45 OBV-EVGs and 1,522 TOV-EVGs; see Materials and Methods). Of these genomes, 1,404 (89.6%) were predicted to encode homologs of tailed-virus hallmark proteins (i.e., terminase large subunits [89.5%], major capsid proteins [34.4%], or portal proteins [60.2%]), suggesting that the genomes were derived from tailed viruses. Of the remaining 163 EVGs, 72 were predicted to encode integrase homologs.

10To investigate the global novelty offered by culture-independent viral genome sequencing efforts, we compiled a set of 1,811 EVGs (>10 kb) composed of the 45 OBV-EVGs, the 1,522 TOV-EVGs, and 244 EVGs from other studies. We also compiled a set of 2,429 prokaryotic dsDNA viral genomes (>10 kb) from cultured viruses, which are referred to here as reference viral genomes (RVGs) (Fig. S3; Table S1B).

11We first generated a viral proteomic tree on the basis of genomic similarity scores (denoted by SG) derived from tBLASTx scores. The SG value is 1 when two genomes in a comparison are identical and decreases to 0 when a tBLASTx search fails to detect any sequence similarities. The viral proteomic tree revealed a clear separation between EVG and RVG clades, with most of the EVGs grouped with other EVGs and not with the RVGs (Fig. 1). We also used average linkage clustering of the EVGs/RVGs to delineate operational taxonomic units (i.e., genomic OTUs [gOTUs]) on the basis of the SG value, with six different clustering cutoff values (Fig. 2 for cutoff SG = 0.15 and Fig. S4 for all cutoff values from 0.1 to 0.9). The EVG-containing gOTUs outnumbered the RVG-containing gOTUs at five of six tested SG cutoff values. For example, we observed a 1.6-fold EVG-to-RVG gOTU overrepresentation ratio at SG = 0.3 (Fig. S4A). The proteomic tree and comparative genome maps are available at http://www.genome.jp/viptree/EVG2017.

12We analyzed the viral taxonomic classification of the RVGs and evaluated the correspondence between viral genera and gOTUs using different SG cutoff values. The SG values between 0.07 and 0.2 were associated with relatively high adjusted Rand index values (i.e., > 0.79), and SG = 0.15 (adjusted Rand index = 0.847) was determined to be the most accurate cutoff value for a genus-level classification (Fig. S5). With this cutoff value, we obtained 1,087 gOTUs for the EVGs/RVGs. The 2,429 RVGs were distributed across 487 gOTUs, whereas the 1,811 EVGs were distributed across 617 gOTUs (i.e., 1.27-fold-higher richness), with only 1.4% of the total gOTUs containing both EVGs and RVGs (Fig. 2B). Therefore, the EVGs potentially represent 600 new viral genera. Of the 600 gOTUs, 497 were composed exclusively of OBV-/TOV-EVGs. To complement this analysis, we added 11,779 mined viral genomes (MVGs; genome sizes, >10 kb). We observed only a limited overlap of gOTUs among the EVGs, RVGs, and MVGs (i.e., only two gOTUs with sequences from all three sets), and 590 genus-level gOTUs remained specific to the EVGs.

13Because of the dissimilarity between EVGs and RVGs, host predictions on the basis of similarities to known viral genomes (i.e., RVGs) were difficult to make. Using information regarding RVG hosts, we calculated an optimal SG threshold that separated viruses into those that infect similar hosts and those that do not. The threshold was a SG value of >0.2937 (>90% precision) for the prediction of pairs of viruses infecting host organisms that are evolutionarily related at the genus level (Fig. S6). With this cutoff, we predicted host groups for only 29 of 1,811 EVGs (2 OBV-EVGs, 13 TOV-EVGs, and 14 other EVGs; Table S1C). Of the 29 EVGs, 18, 10, and 1 were predicted to be cyanophages, Pelagibacter phages, and Pseudoalteromonas phages, respectively. Two additional host prediction methods based on tRNA genes and clustered regularly interspaced short palindromic repeat (CRISPR) spacer sequences failed to predict possible hosts for the EVGs. However, the physical linkage of genes on the EVGs provided additional clues about their hosts and biology. In the following sections, we describe virus-host interactions inferred from the genomic contexts of EVGs.

14Four previously undescribed lineages that likely infect unculturable marine group II (MGII) Euryarchaeota species were revealed in the proteomic tree. These four clades were exclusively composed of OBV/TOV-EVGs, with 18, 13, 23, and 4 EVGs in clades 1, 2, 3, and 4, respectively (Fig. 3). Phylogenetic analyses of the DNA polymerases encoded in those EVGs strongly support the existence of the four clades identified in the proteomic tree (Fig. 4A). These clades were grouped with homologs from haloviruses and euryarchaea. Identifications of archaeal hosts for the 58 EVGs were also supported by their gene content. Of the genes in the EVGs with homologs in cellular organisms, an average of 36.1% (14.3 to 60.0%) were most closely matched to archaeal proteins. Additionally, one to five tailed-virus structural protein homologs were detected in each of the EVGs (Table S1D). Archaeal tailed viruses have been detected only in Euryarchaeota species, with the exception of a provirus of Nitrososphaera viennensis (Thaumarchaeota) isolated from soil.

15We observed that the EVGs contained chaperonin genes (Fig. 3A). Thirty-eight of the 58 EVGs encode chaperonin homologs, even though chaperonin genes have rarely been identified in sequenced viral genomes (i.e., only 7 of the 2,429 RVGs encode chaperonins). In some viruses, chaperonins, which are usually provided by the hosts, are responsible for the correct assembly of viral particles. All 18 EVGs in clade 1 encode archaeon-type chaperonin homologs (i.e., thermosome; group II chaperonin), while 20 EVGs in clades 2 to 4 encode bacterium-type chaperonin homologs (i.e., GroEL; group I chaperonin). We detected both groups of chaperonin genes in the MGII genomes. The group I and group II chaperonin sequences from the EVGs were grouped with these MGII chaperonins (Fig. 4B), suggesting that MGII species serve as hosts for these environmental viruses.

16The following three archaeal taxa are abundant in the marine water column: marine group I Thaumarchaeota (MGI), MGII, and marine group III Euryarchaeota (MGIII). Of these, currently cultivated representatives exist only in MGI. The members of MGII are abundant in particle-rich surface waters, while those of MGIII have been observed almost exclusively in deep seas. A recent study revealed that MGII members can temporarily become the most abundant (up to 40%) prokaryotic components in the days following a spring bloom. The 58 EVGs were derived from surface or deep chlorophyll maximum viromes, suggesting their photic-zone habitat. These observations and the genomic context described above suggest that the 58 EVGs represent genomes of tailed viruses infecting MGII Euryarchaeota species.

17Iron-sulfur (Fe-S) cluster proteins are involved in a variety of biological processes, including gene regulation, electron transfer, catalytic reactions, and oxygen-iron sensing. In a previous study, Fe-S cluster assembly protein genes (e.g., sufA and iscU) were identified as auxiliary metabolic genes (AMGs) of photic-zone viromes. However, the lack of complete genome data hindered further characterizations of the viruses carrying these genes. We identified 16 OBV/TOV-EVGs with Fe-S cluster assembly protein genes, including 14 EVGs containing an Fe-S cluster A-type carrier (ATC) gene and 6 EVGs carrying the IscU gene (Fig. 5A). These genomes are scattered across four groups of viruses in the proteomic tree, and many of their close relatives (i.e., other EVGs and Pelagibacter phage HTVC008M in Fig. 5A) do not contain these genes. The ATC and IscU proteins function as scaffolds in which Fe and S atoms are assembled into Fe-S clusters. Phylogenetic trees of IscU (Fig. 5B) and ATC (Fig. S7) revealed that all six EVG-encoded IscU genes form a clade with gammaproteobacterial homologs. Of these, an IscU gene from OBV_N00005 was phylogenetically closely related to homologs from SAR86, suggesting that SAR86 members represent potential hosts for OBV_N00005. The prevalence of these viral genes in photic-zone viromes appears to be linked to the wide distribution of these bacteria.

18In addition to the Fe-S scaffolding proteins, some of the EVGs encode several Fe-S cluster proteins that use Fe-S clusters as prosthetic groups, such as radical S-adenosylmethionine (SAM) superfamily enzymes and CRISPR-associated Cas4 exonucleases. The EVGs also encode proteins involved in the metabolism of Fe-S cluster proteins, such as glutaredoxins (Grx), the phenylacetyl-coenzyme A oxygenase component PaaD, and ClpP, which is a serine protease targeting Fe-S cluster proteins. A notable example is the T4-like TARA_ERS488813_N000010 (183 kb; group iv in Fig. 5A), which includes an ATC gene, 12 genes for radical SAM superfamily enzymes, and cas4, grx, and paaD (16 genes in total; Table S1E). Other T4-like EVGs encoding ATC and/or IscU proteins contain two to seven additional Fe-S-related genes. Of these genes, paaD has not been previously associated with a virally encoded protein and thus represents a novel AMG. These observations suggest that Fe-S cluster assembly proteins encoded in these viral genomes function as a part of Fe-S cluster-related metabolic processes involving not only host proteins but also many virally encoded proteins.

19The RVG set included 114 cyanophage genomes, which were grouped into 17 viral-genus-level gOTUs. There were no other RVGs classified into these gOTUs. Of these 17 gOTUs, 5 included 34 EVGs (i.e., 3 OBV-EVGs, 16 TOV-EVGs, and 15 previously described EVGs), which are likely to have been derived from cyanophages or their relatives. Screening all EVGs with 11 photosynthesis-related AMGs (see Materials and Methods) led to the identification of 11 predicted cyanophage EVGs, of which 10 were included in the gOTUs mentioned above (Table S1F). The remaining EVG (i.e., TARA_ERS489084_N000023; gOTU G241), which carries psbA and hli, formed a singleton gOTU and represents a new cyanophage group. To characterize the approximate abundances of these 18 cyanophage gOTUs (149 genomes; Table 2), we mapped the TOV and OBV reads on these putative cyanophage genomes. The following five most abundant gOTUs represented >98% of the total cyanophage content: (i) G386, including T4-like myoviruses (35.1%); (ii) G14, including podoviruses (33.7%); (iii) G234, including a siphovirus and dwarf myoviruses (23.4%); (iv) G238, including Synechococcus phage S-EIV1 (3.3%); and (v) G15, including Prochlorococcus phage P-RSP2 (3.2%) (Tables 2 and S1B for the list of genomes). Thus, marine cyanophage genomes are well represented in the current databases.

20Bacteroidetes is one of the most abundant bacterial phyla in the oceans (e.g., 30% of the bacterioplankton during phytoplankton blooms). Members of this phylum are involved in the decomposition and remineralization of phytoplankton biomass. A recent study revealed that an algal bloom is followed by the presence of a rapid succession of diverse Flavobacteriaceae bacteria. To the best of our knowledge, the genomes of the following 38 phages infecting marine Bacteroidetes (Flavobacteriaceae) have been described: psychrophilic Flavobacterium phage 11b, Croceibacter phage P2559S, 2 Persicivirga phages, 31 Cellulophaga phages, Flavobacterium phage 1/32, and 2 Polaribacter phages. Polaribacter was reported to be abundant following a spring phytoplankton bloom, while Cellulophaga phages (31 of 38) likely represent a “rare biosphere” rather than abundant marine phages. We detected two groups (i.e., groups 1 and 2) of putative Flavobacteriaceae phage genomes (i.e., 5 RVGs, 8 OBV-EVGs, 222 TOV-EVGs, and 9 EVGs from another study; Fig. 6). Group 1 and group 2 consisted of 29 and 25 gOTUs, respectively. Of these, 23 and 21 gOTUs were exclusively composed of OBV/TOV-EVGs. Of the genes in the OBV/TOV-EVGs having homologs in cellular organisms, 64.4% (15.8% to 92.3%) on average for the members of group 1 and 32.4% (10.5% to 59.1%) on average for the members of group 2 were most similar to Bacteroidetes genes. For example, the gene20 sequence of OBV_N00025 (group 2, G506; Fig. 6B) was most similar to the RNA polymerase sigma-70 factor sequence of a Flavobacteria strain from marine surface water (WP_009781949; Leeuwenhoekiella blandensis; E value = 1e-30). Genomes of these groups also encode conserved virion structural or morphogenetic proteins. For the members of group 1, we detected putative portal gene homologs in 148 EVGs (93.7%) and prohead protease homologs in 145 EVGs (91.8%). For the members of group 2, we detected homologs of two to six structural proteins of Cellulophaga phage phi38:1 (i.e., a member of group 2) in 78 EVGs (100%). Additionally, we detected GroEL homologs in 36 EVGs of the members of group 2 (Fig. 6B) which were phylogenetically related to the homologs in Cellulophaga phages (Fig. 4B). Therefore, these EVGs probably correspond to viruses of Flavobacteriaceae species and may prove to be useful genetic markers for studying viruses affecting bacterial decomposer communities.

21Isocitrate lyase (AceA) and malate synthase (AceB) catalyze two reactions in the glyoxylate shunt, which bypasses the CO2-generating steps of the tricarboxylic acid cycle and enables the net assimilation of carbon from acetyl-coenzyme A (acetyl-CoA), leading to gluconeogenesis (i.e., generation of glucose) and cell growth. We identified an aceBA operon in a TOV-EVG (TARA_ERS478052_N000008; 179 kb; see Table S1G for gene description) that included homologs of three structural genes from T4-like phages. Our genomic similarity and gene composition analysis did not provide any clue about the host of this virus. A previous study detected aceA and aceB in ocean viromes, but this is the first time, to our knowledge, that an aceBA operon has been observed in a complete viral genome. The genome also encoded six enzymes (i.e., Gmd, WcaG, ManC, NeuA, KdsA, and WaaG) for the biosynthesis of lipopolysaccharides (LPS) and capsular polysaccharides, important components of bacterial cell wall and capsule. Previous studies identified LPS synthesis genes in temperate and lytic phages and proposed that these genes function to modify cell surface compositions to prevent other viruses from attaching to the cell during the lysogenic or pseudolysogenic phase, in the latter of which a lytic process is halted due to suboptimal host cell growth. Following this “lock out” hypothesis, the aceBA-carrying virus (i.e., TARA_ERS478052_N000008) should have a provirus phase, and AceA and AceB may function to promote the growth of host cells. gene40 of the TOV-EVG was predicted to encode a homolog of zeta toxin proteins (Table S1G) thought to be involved in a toxin-antitoxin system. Toxin-antitoxin systems enhance the stability of plasmids and prophages by postsegregational killing. This corroborates the existence of a lysogenic phase of this virus, though there was no other evidence for lysogeny in the viral genome. It should be further noted that the function of LPS is not limited to protection of the cell from viral infection but that LPS on bacterial outer membrane confers tolerance of temperature and oxidative stresses as well as resistance to antibiotics. Therefore, aceBA and the cell wall biogenesis genes in the TOV-EVG may contribute to a host’s survival and environmental adaptation by altering carbon metabolism and cell surface compositions during the lysogenic phase.

22Our analysis also unveiled phage genomes likely infecting members of the SAR116 clade, which is one of the most abundant marine bacterial lineages. OBV_N00085 (40 kb) and three closely related TOV-EVGs (40 to 41 kb; SG for OBV_N00085 = 0.25 to 0.26) exhibited clear collinearity with an approximately 40-kb genomic segment from “Candidatus Puniceispirillum marinum” IMCC1322 of the SAR116 clade (class: Alphaproteobacteria) (Fig. 7 for OBV_N00085). This suggests that these EVGs are derived from temperate phages infecting SAR116 or related bacteria. These genomes consistently encode integrases.

23Seven EVGs (OBV_N00073, three TOV-EVGs, and three other EVGs; 39 to 42 kb) exhibited high genome-wide sequence similarities to Pelagibacter podovirus HTVC019P (SG = 0.34 to 0.44; 42 kb; a dot plot comparing OBV_N00073 and HTVC019P is presented in Fig. S2B). On the basis of the SG values (i.e., >0.2937; estimated precision, >90%), we predict that these EVGs infect host species in the genus Pelagibacter (Table S1C). Another Pelagibacter podovirus (i.e., HTVC010P), which is believed to be a member of the most abundant virus subfamily in the biosphere, was classified in a different group of the proteomic tree together with 102 EVGs (OBV_N00107, 77 TOV-EVGs, and 24 other EVGs; 31 to 73 kb; Fig. S8). These 102 genomes carry homologs of HTVC010P structural protein genes. The G+C content of the HTVC010P genome is 32%, while the EVGs of this group contain higher levels of G+C content (i.e., 31 to 57%). Low levels of G+C content (i.e., 28.6 to 32.3%) are a common genomic feature of the SAR11 clade members. Since high levels of correlation between the G+C content of prokaryotic viruses and that of their hosts were previously reported, the variation in the levels of their G+C content suggests that the viruses in this group infect a wide range of host species.

24We mapped protein sequences and raw reads from independently generated photic virome data (i.e., the Pacific Ocean viromes [POV]) on the RVGs and EVGs. The RVG set recruited 4.70% of the POV proteins, while the EVG/RVG union set recruited 22.6% of the proteins (i.e., a 4.8-fold increase; Fig. 8A). At the nucleotide sequence level, the RVG set recruited 1.02% of the POV reads, while the EVG/RVG union set recruited 4.20% of the reads (i.e., a 4.1-fold increase; Fig. 8B). Thus, the EVGs serve as an effective additional reference viral genome data set for exploring viromes from photic oceans.

25From the assemblies of 52 marine viromes, we obtained 1,567 circular complete genomes that are most likely of prokaryotic dsDNA viral origin. The acquisition of the complete genome sequences helped classify the viral lineages and provided important clues about their hosts and metabolisms. The genome-based clustering of the metagenome-derived viral genomes together with previously reported ones suggests that 600 of the 617 gOTUs represent new genera of prokaryotic viruses. Additionally, they contain greater genome richness than the reference genomes of cultured prokaryotic viruses that have so far been sequenced. Our analyses also predicted the relationships among the EVGs and the major groups of marine prokaryotes, for which no viruses have been isolated (i.e., MGII and SAR86). Given the lack of isolation of viruses, the physiological features of the sequenced EVGs are unclear. However, some of the newly identified EVGs carried functionally related AMGs, such as those encoding proteins related to Fe-S clusters (16 genes) and to carbon assimilation/cell wall biogenesis enzymes (8 genes). These AMGs may function to coordinate the supply/recycling of Fe-S clusters and to enhance host adaptation during the lysogenic cycle. Previous studies also revealed that cyanophages carry multiple functionally linked photosynthesis and lipopolysaccharide synthesis genes for their efficient replication. Therefore, viral survival strategies in marine viruses involving many functionally related AMGs appear to target not only the biosynthesis of molecular building blocks (e.g., nucleotides) but also diverse metabolic and cellular processes.

26Seawater samples (9 × 4 liters) were collected at a 5-m depth at the entrance of Osaka Bay (34°19′28″N, 135°7′15″E), Japan, every 3 h for 24 h on 25 and 26 August 2014. Seawater was filtered through a 142-mm-diameter (3.0-μm-pore-size) polycarbonate membrane (Millipore, Billerica, MA) and then through a 142-mm-diameter (0.22-μm-pore-size) Durapore polyvinylidene fluoride membrane (Millipore). The filtrates were stored at 4°C prior to treatments. The viruses in the filtrate were concentrated by FeCl3 precipitation and purified using DNase and a CsCl density centrifugation step. The DNA was then extracted as previously described. Libraries were prepared using a Nextera XT DNA sample preparation kit (Illumina, San Diego, CA) according to the manufacturer’s protocol, except that we used 0.25 ng viral DNA. Samples were sequenced with a MiSeq sequencing system and MiSeq V2 (2 × 150 bp; five samples) or V3 (2 × 300 bp; four samples) reagent kits (Illumina, San Diego, CA).

27Nine OBVs were individually assembled using the following four assemblers: SPAdes, metaSPAdes (http://bioinf.spbau.ru/spades), IDBA-UD, and Ray Meta. SPAdes 3.1.1 was used with default k-mer lengths as well as the accompanying BayesHammer and MismatchCorrector. The metaSPAdes 3.7.0 program was used with default k-mer lengths and BayesHammer. The IDBA-UD 1.1.1 program was used with fixed multiple k-mer lengths (24 to 124, increased by 10 for 2 × 300 bp reads; 24 to 84, increased by 10 for 2 × 150 bp reads) and the option of a preread correction with a minimum contig length of 300 bp. Ray Meta 2.3.1 was used with a fixed k-mer length (k = 41). Additionally, we used scaffolds of these assemblies, which we called contigs for simplicity. The REAPR 1.0.18 program was used to assess the quality of the assemblies. This program reports four types of errors categorized as short insertion/deletion errors (i.e., types 1 and 3) or scaffolding errors (i.e., types 2 and 4).

28Nine sets of OBV reads were also coassembled by SPAdes with the same settings as described above. We determined that a contig was circular (i.e., complete) if its 5′ and 3′ terminal regions were nearly identical (i.e., >94% and ≥50 bp). We identified 40 circular contigs (>10 kb) satisfying this condition. A coassembly involving the merged paired-end reads generated by FLASH was also prepared. We included the merged and remaining unmerged reads for the assembly. With this second coassembly, we detected 34 circular contigs (>10 kb), of which 6 were not detected in the first coassembly. We incorporated these 6 contigs in our data set, and we ultimately obtained 934 OBV contigs (>10 kb), including 46 circular ones. Forty-three TOV samples were similarly analyzed, except that the sequence assemblies were prepared sample by sample and only with raw reads (i.e., not from merged paired-end reads). Code for circular contig detection is downloadable at ftp://ftp.genome.jp/pub/db/community/EVG2017.

29Gene predictions were completed using MetaGeneMark. Homology searches were conducted using BLASTp against the NCBI-nr database (E value, <1e−5), RPS-BLAST against the COG database (as of April 2015; E value, <1e−4), and HMMER against the Pfam (as of May 2015; E value, <1e−4) and TIGRFAMs (release 15; E value, <1e−4) databases. For predictions of tailed-virus hallmark genes and integrase genes, we used HHsearch (E value, <1e−9) against the Pfam database after constructing query hidden Markov models (HMMs) using jackhmmer (part of the HMMER package) with default settings. We also used PSI-BLAST to identify homologs of specific genes.

30We used a newly developed method (see Text S1 in the supplemental material) and VirSorter to distinguish between viral and prokaryotic contigs. We discarded all contigs predicted to be of prokaryotic origin by either or both methods. Finally, 879 of the 934 OBV contigs (including 46 circular ones) and 1,554 of the 1,618 TOV circular contigs were considered to originate from viruses.

31We conducted PCR assays for 21 weakly supported regions in four randomly selected OBV circular contigs (i.e., OBV_N00005, OBV_N00020, OBV_N00021, and OBV_N00023; see Fig. S2A in the supplemental material). Primer sequences are provided in Table S1H in the supplemental material.

32Colinearity was evaluated on the basis of the percentage of OBV-EVG genes that had orthologous relationships with the most closely related genome (i.e., Bg in Fig. S2B). If ≥60% of the OBV-EVG genes had orthologs in the closest relative, we considered the OBV-EVG to exhibit nearly complete genomic colinearity. Eighteen OBV-EVGs (39%) were observed to exhibit complete or nearly complete colinearity with other viral genomes. Additionally, we identified colinear genomic regions using MCScanX and calculated the percentage of OBV-EVG genes in these regions (i.e., Cg in Fig. S2B).

33We used raw reads for the above assemblies, but the reads underwent a quality-control screening before being back-mapped to contigs with the following procedure: (i) duplicated reads were removed using FastUniq; (ii) paired-end reads were merged with FLASH, and the merged and unmerged reads were kept; (iii) reads were removed if the percentage of high-quality nucleotide positions (i.e., quality score >30) was <80%; and (iv) reads were removed if the sum of the lengths of ambiguous nucleotide positions and low-complexity regions detected by DUST was >40% of the total length. If one of the paired-end reads was removed in step iii or step iv, the mate was retained as a single read.

34To detect SNPs and assess nucleotide diversity, we mapped quality-controlled reads on contigs using the Bowtie 2 program. To minimize the inclusion of sequencing errors among the mapped nucleotides, we considered only high-quality nucleotides (i.e., quality score, >30). Nucleotide diversity was defined as previously described and was calculated using equation 1 of a published method. The SNPs were detected for positions with ≥5× sequence coverage using the following six criteria: (i) at least one read, (ii) at least two reads, (iii) more than 10% coverage, (iv) more than 20% coverage, (v) more than 10% coverage or at least two reads, and (vi) more than 10% coverage and at least two reads. These criteria were applied to the second-most-frequent nucleotide at each position.

35To detect redundancies among TOV-EVGs and OBV-EVGs, an all-against-all BLASTn search was conducted. We merged high-scoring segment pairs (HSPs) for each resulting pair, and if the merged HSPs covered ≥80% of the shorter EVG, with ≥95% average identity, the EVGs were considered redundant. Nonredundant EVGs were obtained by single-linkage clustering of these redundant pairs.

36We first compiled 46 OBV-EVGs, 1,554 TOV-EVGs, and 247 EVGs from three projects, including 192 complete contigs, 54 circular consensus genomes, and a complete viral genome obtained from samples from single amplified genomes (SAG). The RVGs were retrieved from RefSeq (release 75; March 2016), EBI Genomes Pages (May 2015), and CAMERA. We selected dsDNA viral genomes that were larger than 10 kb. We then removed the genomes of eukaryotic viruses identified using the GenomeNet Virus–Host Database. Thirty-six EVGs (i.e., 1 OBV, 32 TOVs, and 3 others) were most similar to eukaryotic viral genomes among RVGs and were removed from the proteomic tree and gOTU analyses, which were used to compare the levels of diversity of the RVGs and EVGs of prokaryotic viruses.

37We constructed a proteomic tree as previously described. Briefly, the all-against-all distance matrix of the EVG/RVG data set was calculated on the basis of the normalized bit score of tBLASTx (SG), and the proteomic tree was built with BIONJ using the distance matrix. The proteomic tree, gene annotations, and genome alignment views are accessible at http://www.genome.ad.jp/viptree/EVG2017.

38The genus-level threshold value for gOTU clustering was estimated from a subset of the RVGs used in this study (i.e., 345 prokaryotic dsDNA viruses), each of which was assigned to a viral genus (i.e., 82 genera in total). We constructed gOTUs with different SG cutoffs (intervals of 0.01) and evaluated how closely the resulting gOTUs corresponded to the genus-level viral classifications using the adjusted Rand index.

39We attempted to predict host taxonomic groups for EVGs on the basis of viral genomic similarities measured with SG. We estimated the precision of our prediction method on the basis of RVGs (i.e., 1,285 prokaryotic dsDNA viruses), each of which was linked to a uniquely assigned host taxonomic group according to the Virus-Host Database. Regarding host taxonomic groups, Cyanobacteria (phylum) and Enterobacteriaceae (family) were regarded as individual host taxonomic groups because closely related viruses are known to infect hosts of different genera belonging to these host groups. The remaining viral hosts were grouped at the genus level. For each RVG, the best SG values for the members of the same host group, and for the members outside the host group, were recorded (i.e., 2,570 SG scores in total). A precision curve was generated using sliding SG cutoff values (Fig. S6). When the SG cutoff value was >0.3889 or >0.2937, the viral pairs were predicted to infect hosts in the same group at >95% or >90% precision, respectively.

40To detect photosynthetic genes in the EVG/RVG data set, we used PSI-BLAST (E value, 1e−6; inclusion_ethresh, 1e−6; num_iterations, 3) and the query sequences listed in Table S1I.

41Multiple sequences were aligned using the MAFFT program (version 7.245), with the FFT-NS-2 mode and a maximum of 1,000 iterations (--retree 2, --maxiterate 1000). Conserved positions in the alignments were selected with the trimAl program (version 1.3). Maximum likelihood trees with 100 bootstrap replicates were calculated with RAxML (version 8.2.4) using the fast bootstrapping mode, and models were selected by the use of ProteinModelSelection.pl (i.e., LGF for DNA polymerase B and LG for chaperonins, IscU, and ATC).

42Reads (3.68 M sequences) and proteins (2.78 M sequences) of 16 photic POV samples were downloaded from iMicrobe (http://data.imicrobe.us). These sequences were mapped on EVGs and RVGs using BLASTn (for reads; E value, <1e−3) and tBLASTn (for proteins; E value, <1e−3) if the alignment revealed ≥60% identity and covered ≥80% of the query sequence.

43Read and assembled sequences obtained from OBV were deposited at DNA Data Bank of Japan (DDBJ) under accession numbers DRR053207 to DRR053215 and SAMD00045684 to SAMD00045692. The sequence data for the OBV project are accessible under DDBJ BioProject accession number PRJDB4437. Sequences and additional data are available at ftp://ftp.genome.jp/pub/db/community/EVG2017.