Assembly of Complete Genomes

Even though sequences from Microviridae are found in a large number of viromes, assembly of complete Microviridae genomes was described in only one dataset, the Sargasso Sea virome [26]. Indeed, a consensus sequence of a Chp1-like Microviridae was first created by Angly et al., [26] and two complete genomes affiliated to Microviridae were assembled in a recent analysis of the same dataset using up-to-date assembly software [9]. To further decipher the evolution and distribution of this family, we assembled all available public viromes (with a threshold of 98% identity on 35 bp) and screened them for the presence of complete Microviridae-like circular genomes. As a result, 81 contigs representing putative complete Microviridae genomes were obtained from 25 out of the 95 viromes tested (Table S1). Out of these 81 new genomes, 15 new Microviridae were assembled from freshwater virome reads, 2 from a marine sample, 2 from microbialites, 1 from coral, 2 from human lung and 59 from human gut samples. Obviously, the number of Microviridae genomes assembled from a virome depends on the length of the sequences (Table S1) but the assembly also depends on the relative abundance of a given virotype in the sampled ecosystem. The sizes of the different genomes generated were quite homogeneous (Table S2), with the smallest genome being 3,989 bp-long and the longest 6,723 bp. This size range is consistent with the genome sizes of known Microviridae (between 4.4 and 6.1 Kb).

In order to detect potential new prophages (viral genomes integrated into bacterial chromosomes), the newly assembled Microviridae sequences were used as queries in searches against the complete bacterial genomes from the NCBI and the Human Microbiome Project [27] databases. The five previously described complete prophages [17] were fully retrieved, alongside a new one, detected in the recently sequenced Prevotella multiformis strain, highly similar to other Microviridae prophages detected in Prevotella.

Phylogeny and Genome Organization of the Microviridae Family

In order to gain insight into the diversity of the Microviridae family and its genome evolution, a phylogenetic tree was computed from the major capsid protein (VP1) sequences and correlated with the corresponding genome maps (Fig. 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Phylogenetic tree drawn from the major capsid protein multiple alignment.

Linearized genomes are represented for each virus. The open reading frames in each genome are color-coded following the nomenclature used for Chlamydia phage genomes (i.e.VP1 : major capsid protein, VP2 : DNA pilot protein, VP3 : internal scaffolding protein, VP4 : genome replication initiation protein, and VP5 : DNA binding protein). Striped-colored genes encode proteins possessing features characteristic of VP2 proteins, but displaying no significant sequence similarity, as assessed by BLAST. The four Microviridae subgroups are highlighted on the tree. Bootstrap scores greater than 80 are marked with gray dots.

https://doi.org/10.1371/journal.pone.0040418.g001

Detailed Analysis of the Conserved Microviridae Proteins

Major capsid protein (VP1).

Virions of microviruses and gokushoviruses display distinct structural features and molecular composition; although both possess icosahedral capsids composed of 60 copies of the major capsid protein, MCP (F and VP1, respectively), only those of microviruses are decorated with pentameric major spike protein complexes positioned at each of the 12 five-fold vertices [11]. Electron cryo-microscopy (cryo-EM) study of the SpV4 virions revealed that gokushoviruses instead possess 55 Å-long ‘mushroom-like’ protrusions located at the 3-fold symmetry axes of their capsids [21]. These protrusions are formed by insertion loops coming from three subunits of the VP1 protein (Fig. 2A), and were suggested to participate in receptor recognition and binding on the host cell surface. The protrusion-forming insertion is not present in the MCPs of ΦX174-like microviruses and is largely accountable for the size differences between the MCPs of microviruses and gokushoviruses (Fig. 2B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Major capsid protein (MCP) variation within the Microviridae.

(A) Three-dimensional model of the SpV4 virion (PDB ID:1KVP). Three capsomers donating long insertion loops to form the ‘mushroom-like’ protrusions at the three-fold axes of symmetry of the icosahedral capsid are highlighted in blue, green, and red. (B) A boxplot illustrating the variation of MCP sizes between the four subgroups of the Microviridae. (C) Three-dimensional models of the MCPs from viruses representing the four subgroups of the Microviridae: Microvirus (ΦX174 protein F; PDB ID:1CD3), Gokushovirinae (SpV4 VP1, PDB ID:1KVP), Alpavirinae (Prevotella bucalis prophage BMV5 protein VP1; GI:282877220), Pichovirinae (Pavin_279 protein VP1). The insertions within the VP1 proteins of gokusho-, alpa- and pichoviruses relative to the F protein of ΦX174 are highlighted in green.

https://doi.org/10.1371/journal.pone.0040418.g002

Comparison of the MCP sequences from the four subgroups of the Microviridae revealed that the average identity at the protein level varies from 40 to 80% within a group, and from 20 to 40% between the groups, with an exception of the Enterobacteria-infecting phages, which generally present no significant sequence identity with MCPs from other clades of the Microviridae (Fig. S2). Analysis of the VP1 size variation including the newly discovered members of the Microviridae revealed that the average MCP size in the four different sub-families is significantly different (one-way ANOVA, p-value 1e-43). ΦX174-like microviruses possess smaller MCPs (average length 427 aa), while those of gokusho-, alpa- and pichoviruses are significantly larger (Fig. 2B). Notably, among the latter three groups, pichoviruses possess the smallest MCPs (average length 512 aa), while the MCPs of alpaviruses are the largest (average length 630 aa) and also the most variable in terms of size (ranging from 541 to 780 aa). Multiple sequence alignment of VP1 homologues revealed that the MCP size difference is a result of variation in the number and size of insertions in VP1-like proteins (Fig. S3).

To gain insights into possible architecture of viruses from the newly identified groups Alpavirinae and Pichovirinae and to understand the effect that the insertions within their MCPs might have on virion architecture, we constructed three-dimensional VP1 models for representative viruses. These were compared to the available X-ray structure of ΦX174 protein F (PDB ID:2BPA; [11]) and the cryo-EM-based model of SpV4 protein VP1 (PDB ID:1KVP; [21]). Structural modeling and model quality assessment are described in Materials and Methods. Comparison of the structural models (Fig. 2C) revealed that VP1 homologues from viruses belonging to all four groups of the Microviridae possess a conserved eight-stranded β-barrel core (also known as viral jelly-roll; [29]) and all, but ΦX174-like microviruses, possess an extended loop that forms a mushroom-like protrusion in SpV4. Consequently, it is likely that virions of gokusho-, alpa- and pichoviruses, unlike those of microviruses, possess characteristic receptor-binding spikes at the three-fold axes of the icosahedral capsids (Fig. 2A).

Further analysis has revealed that the size of the putative receptor-binding spike-forming insertions differs between different subgroups of Microviridae (Fig. S4A): the shortest are found in pichoviruses (average length 60 aa), while those of alpaviruses are the longest (average length 110 aa). The insertion length also varies considerably within Gokushovirinae (from 53 to 114 aa) and Alpavirinae (from 45 to 209 aa). Interestingly, this variation appears to be ecosystem- rather than virus subgroup-dependent. The insertion length variation was much less pronounced for VP1 proteins of viruses residing in aquatic environments (including both gokushoviruses and pichoviruses) than it was for viruses from human samples (gokushoviruses and alpaviruses) (Fig. S4B). The same tendency was also true for the full length MCPs, with VP1s from human samples being larger (average length 604 aa) than those of viruses thriving in aquatic environments (average length 533 aa). It therefore appears that evolution towards acquisition of insertions within the MCPs of viruses isolated from human samples might be driven by the need to cope with additional factors (e.g., immune system, low pH of the human gastrointestinal tract, etc.) that are not present in aquatic environments.

Distinct members of the Alpavirinae (a group exclusively associated with human samples; Fig. 1) possess insertions at different locations within their VP1 proteins, suggesting that VP1 proteins within this Microviridae subgroup are indeed evolving rapidly. Such species-specific insertions were found to be up to 231 aa-long (Human_feces_B_007). In order to verify whether such extensive insertions would not interfere with normal virion formation, we fitted our three-dimensional model of the alpaviral VP1 into the pseudoatomic model of SpV4 (PDB ID:1KVP) and mapped the location of all the insertions exceeding 15 aa. We identified 6 MCP hot-spots where large insertions were tolerated in alpaviruses (Fig. S5). Notably, all of these insertions occurred in the loop regions of the MCP facing outwards from the virion surface and are therefore expected not to affect virion assembly. Interestingly, the 231 aa-long insertion in the MCP of Human_feces_B_007 is predicted to be rich in β-strands and is likely to fold into an independent domain. Peculiarly, the major spike protein G of ΦX174-like microviruses, which forms protrusions at the virion five-folds of these viruses is also rich in β-strands. Unfortunately, the sequence of this insertion in the MCP of Human_feces_B_007 does not share significant similarity with proteins in extant databases and its provenance therefore remains obscure.

Horizontal Acquisition of New Genes

It has been previously suggested that genes encoding novel functions in microviral genomes emerge from pre-existing genomic regions through accumulation of point mutations [19], [33]. This conclusion has been supported by the lack of identifiable cases of horizontal acquisition of new genes by the Microviridae. Analysis of the complete microviral genomes assembled in this study has unexpectedly revealed 11 genes from human-associated Microviridae (5 Gokushovirinae and 6 Alpavirinae - 10 from human gut and 1 from human lung; Fig. 1) encoding a putative peptidase of the M15_3 family (Pfam Id : PF08291). M15 family peptidases are widespread in bacteria and are involved in cell wall biosynthesis and metabolism; they catalyze hydrolytic cleavage of the amide bond within peptide bridges that cross-link glycan strands of the bacterial cell wall [34].

The closest homologues detected by BLAST for these 11 genes are from bacterial genomes, except for Human_feces_A_016, for which the closest homologue is found in a tailed dsDNA phage genome (Table S3). Half of the Alpavirinae peptidase genes are affiliated to Bacteroidetes and a Bacteroidetes phage, the three others are associated with Burkhorderiales (Leptothrix and Collimonas). Gokushovirinae peptidases are affiliated to Firmicutes : Faecalibacterium prausnitzii (4 of 5), and Gamma-proteobacteria (Providencia alcalifaciens). These closest homologues of the microviral proteins are found next to phage-like genes in several bacterial genomes (Fig. 3A). For example, phage-like integrase genes are proximal to the M15 peptidase genes in Bacteroides vulgatus ATCC 8482 and Bacteroides vulgatus PC510 genomes, indicating a likely phage origin for these genomic regions. Consistently, the peptidase gene from Providencia alcalifaciens DSM30123 genome is retrieved within a complete prophage region, and next to a putative holin gene. The peptidase gene from Faecalibacterium prausnitzii M21/2 is present within a three-gene cassette, with all three genes having homologues in bacteriophages (Fig. 3A). Notably, besides the M15 peptidase, the cassette includes a putative holin gene (hit to Lactococcus phage ul36.t1, ABD63797; 47% identity, E = 1e-20) and a gene of unknown function, with a homologue present as part of the lysis gene cluster in Streptococcus phage 858 (Fig. 3A). All this indicates that peptidase genes are likely to be frequently exchanged between viruses and their hosts, probably through prophage integration. Interestingly, 5 of the 11 peptidase genes detected in Microviridae genomes are adjacent to an “unknown” predicted ORF, which displays no sequence similarity to proteins in the extant databases. Peculiarly, this unknown gene (153 codons) in the Human_gut_33_003 genome is encoded on the complementary strand. Such orientation is highly unusual; to our knowledge, no other cases of complementary strand genes have been reported in Microviridae.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Genomic context and phylogenetic analysis of the Peptidase M15 Microviridae sequences.

(A) Organization of the Peptidase M15_3 region in the 11 newly assembled Microviridae genomes. The regions encompassing homologous peptidase genes in three bacterial and two phages (noted with a black circle) genomes are also shown. P2 GpR stands for the P2 phage tail completion protein. (B) Maximum-likelihood tree computed from the multiple alignment of peptidase M15 sequences of the Microvidae and their closest homologues in viral and bacterial genomes. Bootstrap support values are indicated on each node. A fully expanded view of this tree is available as Fig. S9.

https://doi.org/10.1371/journal.pone.0040418.g003

In order to shed light on the evolutionary event(s) leading to the acquisition of M15 peptidase genes by Microviridae phages, a phylogenetic tree was computed from a multiple alignment including the Microviridae peptidases and their closest homologues in both bacterial and viral genomes (Fig. 3B, Fig. S9). The topology of the peptidase tree is consistent with the VP1-derived tree of Microviridae, with a clear separation between the peptidase genes from the Alpavirinae (highlighted in pink) and the five peptidase genes from Gokushovirinae (in blue). Within the Alpavirinae, the peptidase tree topology can be associated with the location of the peptidase gene integration within these viral genomes. Peptidases genes are inserted in two different positions in Alpavirinae genomes : between VP2 and VP4 (CF7ML001, Human_feces_B_039, Human_gut_22_017 and Human_gut_33_017), and between VP4 and VP1 (Human_feces_A_016 and Human_gut_33_005) (Fig. 1). Consistently, these two groups are retrieved on the peptidase tree: Human_feces_A_016 and Human_gut_33_005 are found near the Fusobacterium gene within the Bacteroidetes group, whereas the other Alpavirinae peptidases are retrieved at the base of the Bacteroidetes group. Within the Gokushovirinae, the Human_feces_C_014 peptidase is separated from the rest of gokushoviral peptidases, similarly to the topology of the VP1 tree (Fig. 3B, Fig. 1). In the peptidase tree, the closest neighbors of the Gokushovirinae peptidases are from Firmicutes (Faecalibacterium) and Proteobacteria (Ahrensia, Providencia). The most likely explanation for such clustering is that peptidase genes were horizontally acquired by several members of the Microviridae on multiple occasions. The presence of dsDNA phage peptidases near the Microviridae sequences on the tree suggests that ds and ssDNA phages might be engaged in gene exchange, either directly during a co-infection or via infection of a prophage-bearing host cell.

Finally, the possibility of horizontal gene transfer between Microviridae genomes was investigated in light of this acquisition of a peptidase gene by several human gut Microviridae. For that, we performed a phylogenetic analysis of the two most conserved Microviridae proteins, VP1 and VP4, for each of the four subgroups (Microvirus, Alpavirinae, Gokushovirinae and Pichovirinae). No signs of recent gene transfer event could be detected, confirming the hypothesis that gene transfer between Microviridae are rare, even within the temperate members of the group [14], [17], [19].

Microphages Diversity in Environment

Viral metagenome sequences similar to VP1.

A total of 498 sequences were found to be significantly similar (BLASTx bit score greater than 50) to the VP1 protein of at least one of the Microviridae complete genomes. These 498 metagenomic sequences span 36 of the 95 viromes, from 5 different types of ecosystem (human microbiome, other eukaryote, seawater, freshwater, microbialites). Microviridae remains undetected in hypersaline and hyperthermophilic environments (Fig. 4). In order to analyze the dispersal of each Microviridae subgroup, the presence of each subgroup was checked in the 36 viromes containing Microviridae sequences. The Gokushovirinae subgroup is the most widespread among Microviridae (28 viromes out of 36), and is found in all Microviridae-containing biomes (Fig. 4). Pichovirinae are less frequently detected (12 viromes), but are also retrieved from different types of biomes. On the contrary, Alpavirinae are exclusively detected in human sample viromes (16 samples). Finally, only one sequence affiliated to the genus Microvirus was detected in a seawater virome. Yet, the low BLAST bit score (50.1) and the fact that this is the only microvirus-like sequence retrieved indicates that microviruses are likely to be extremely rare in such environments.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Abundance and distribution of VP1-like sequences in the environment.

The number of viromes and the origin of the samples used for virome preparation are indicated. VP1 sequences were affiliated by best BLAST hit against a database including VP1 sequences from both the previously published Microviridae genomes and the complete genomes assembled in this study.

https://doi.org/10.1371/journal.pone.0040418.g004

Technical Issues and Potential Bias of the de novo Metagenomic Assembly of Viral Genomes

Several points have to be discussed regarding both bioinformatical and biological issues in order to better understand the results obtained in this study. First, sequences reconstructed here from metagenomic data represent consensus sequences of individual microviral populations and thus, DNA sequence variations within each population are masked. Nevertheless, the assembly criteria used here (98% similarity on 35 bp) are considered stringent enough to gather only sequences from the same viral species [26]. Such criteria both limit sequence variations within each assembled genome and mask pyrosequencing errors. Second, even if Microviridae seems more abundant and frequently retrieved in particular ecosystem types (as in the human gut), all biomes were not evenly sampled. Indeed, the published viromes used in this study were generated independently and with different sequencing technologies. For instance, viromes from the human gut contain approximately 2 times more base pairs than viromes from sea water, and 6 times more than those from hypersaline ponds. Third, the methodology used to generate a virome greatly influences the type of viruses which will be retrieved. Viromes prepared through LASL (Linker Amplified Shotgun Library) are not supposed to contain any ssDNA genomes as this technique only recover dsDNA fragments. Not surprisingly, no Microviridae were detected in the two viromes prepared through LASL in this study, both sampled from hyperthermophilic environments. Last, the quantitative importance of ssDNA viruses in general, and of Microviridae more specifically, remains an open question. The abundance of Microviridae was first considered to be higher than predicted when the affiliation of virome reads was normalized by the mean genome length, i.e. when viruses were compared in terms of “number of viral particles in the original samples” [5]. Nevertheless, this quantitative importance was balanced by a potential bias of the whole-genome amplification methodology, which would preferentially amplify small circular DNA templates [35]. Although the real relative abundance of the Microviridae is still unknown, the fact remains that a considerable part of reads in numerous viromes could be affiliated to Microviridae, leading to the assembly of 81 complete Microviridae genomes.

Microviridae is a Coherent Family of ssDNA Phages

The new genomes assembled in this study confirmed that Microviridae is a consistent and homogeneous viral family. All but two genomes contained ORFs significantly similar to the three core genes of Microviridae (the major capsid protein VP1, the replication protein VP4, and the DNA pilot protein VP2). Moreover, the division of the Microviridae family into sub-families, which has been previously presented [20], was both confirmed and complemented by this study. The microphage diversity deduced from the analysis of the environmental samples confirmed the distinction between Enterobacteria-infecting phages (genus Microvirus), and the other known Microviridae. Members of the genus Microvirus are considered as Microviridae archetypes, and are the most studied Microviridae so far, yet none of the genomes assembled in this study is associated with this genus. This apparent paradox might be due to culture bias, as Microviruses are cultivated on E. coli strains, the most widely used and studied prokaryotic model organism. The present study indicates that this genus is rare among the biosphere, and likely constitute a very specific type of Microviridae especially in terms of gene content and capsid structure. The other Microviridae appear to be internally divided into three main subgroups, namely Gokushovirinae, Alpavirinae, and a new subgroup, which we propose to name Pichovirinae. This new clade appear to be more related to Gokushovirinae than to Enterobacteria phages, and could represent an intermediate group, like Alpavirinae, which could fill the gap between the two currently recognized sub-families. Moreover, this new group has a unique genome organization, as VP2 is located between VP1 and VP4 for all the genomes in this group. Based on this synteny break, and the fact that Pichovirinae-like genomes have been assembled from very different samples (namely seawater, freshwater, microbialites, and coral samples), this clade has probably diverged from the common Microviridae ancestor a long time ago. Consistently, significant sequence similarity between pichoviruses and the other Microviridae is confined to the major capsid protein VP1 and the replication protein VP4, while the pilot proteins (VP2) displays only limited sequence similarity to corresponding proteins from the Alpavirinae prophages.

This taxonomic structure of the Microviridae family confirms that the microphage diversity was under-estimated. The detection of sequences homologous to Pichovirinae and Gokushovirinae in very different environments (Fig. 4) suggests that the viruses giving rise to these clades have probably diverged from their common ancestor in a distant past. The latter possibility is supported by the differential gene order conservation in the Pichovirinae on one side (VP1-VP4-VP2) and the remaining Microviridae subgroups on the other (VP1-VP2-VP4). The finding that microphages belonging to different subgroups occupy a variety of different ecological niches suggests that the association of microphages with bacteria is ancient, possibly predating the divergence of this cellular domain into the contemporary lineages.

Major Capsid Protein Structure and Evolution

Based on similarity in virion architecture, it has been previously suggested that members of the Microviridae might share a common origin with eukaryotic viruses from the families Circoviridae, Geminiviridae and Parvoviridae [10]. Indeed, all these ssDNA viruses utilize eight-stranded β-barrel capsid proteins to build their icosahedral (T = 1) virions [36], [37]. However, while the capsid proteins of Microviridae and parvoviruses possess long insertion loops connecting the β-strands (although at different locations; [36]), those of geminiviruses and circoviruses are much more compact [37], [38]. Consequently, if the structural relationship indeed testifies for the common origin of these viruses, the evolution of Microviridae virion structure most likely proceeded through acquisition of insertion loops within the eight-stranded β-barrel core. As revealed through comparative analysis and structural modeling of the MCPs presented in this study, such dynamics within the loop regions of microviral MCPs appears to be an ongoing process, possibly assisting host-range expansion and adaptation to new environments in this viral family. This is especially obvious for microphages associated with human microbiota (all alpaviruses and certain gokushoviruses) that on average possess larger and more numerous insertions within their MCPs (Fig. 1, Fig. S4B). Paradoxically, although ΦX174-like phages are also known to infect hosts isolated from human samples [39], their MCPs are the most compact among the Microviridae. Interestingly, the putative receptor-binding spikes present at the three-fold symmetry axes of gokushovirus capsids ([21]; Fig. 2A) are also likely to decorate the virions of alpaviruses [17] and pichoviruses (Fig. 2C). The presence of this protrusion in all gokusho-, alpa- and pichoviruses, suggests that this feature is ancestral to the spikes present at the five-fold vertices of ΦX174-like microvirus capsids.

The number and the size of insertions within the microviral MCPs were similar in both prophages and free-living viruses, suggesting that these sequence modifications do not preclude the formation of viable virions. This specific evolutionary pattern of human microbiome Microviridae MCPs is reminiscent to co-evolution consequences described for cultivated phages. As described in the experiment of Paterson et al. [40], the basis of co-evolution is the absence of “non-adapted” host for the phage. This is consistent with the restriction of these viruses to human gut flora, where the highest bacterial densities for a microbial habitat were found [41]. Thus, human gut Microviridae are likely to be exposed to a higher host-phage encounter frequency compared to other Microviridae, thereby increasing the evolution rate of their MCP.

Horizontal Gene Acquisition of a Possible Endolysin Gene

Until now, genes encoding novel functions in microviral genomes were thought to emerge from pre-existing genomic regions through accumulation of point mutations [19], [33]. However, the discovery in the Microviridae genomes of peptidase coding genes that were clearly acquired by horizontal gene transfer (HGT), and more likely through at least two independent transfer events, shows that Microviridae are able to integrate genes of interest from external sources into their genomes, even if such transfers are rare. The different uncharacterized ORFs detected in the new Microviridae genomes are then of great interest, since they could represent other horizontally acquired genes. On the contrary, no direct gene transfers between two Microviridae genomes could be detected in our dataset. Notably, phylogenetic analysis of the 47 closely-related Escherichia coli-infecting microviruses illuminated a few cases of HGT between these viruses that probably occurred by homologous recombination [14]. It is possible that HGT in Microviridae is limited by the genetic distance between the donor and the recipient virus species. Consequently, larger datasets of closely related virus genomes might be needed to better understand the prevalence of homologous recombination-driven HGT events in Microviridae.

ΦX174 is the only microvirus for which the mechanism of host cell lysis has been elucidated. Unlike dsDNA phages that typically encode a holin-endolysin system, where holin perforates the cytoplasmic membrane and endolysin digests the peptidoglycan, microviruses depend on a single-gene lysis system [42]. It has been shown that protein E of ΦX174 induces lysis by inhibiting cell wall biosynthesis [43]. It was therefore surprising to discover that gene for M15_3 peptidase identified in several gokushoviral and alpaviral genomes is associated with canonical lysis genes (for holin and endolysin) in dsDNA (pro)phages (Fig. 3A), suggesting that phage-encoded M15 peptidases might play a role in cell lysis during virus progeny release. Indeed, endopeptidase PLY500 (family VanY; PF02557), which is structurally related to M15 family proteases (families M15_3 and VanY belong to the same clan – Peptidase_MD; CL0170), acts as an endolysin at the end of the infection cycle of Listeria phage A500 [44]. We therefore suggest that Microviridae M15 peptidase might also be involved in dissolution of the host cell wall at the end of the phage life cycle.

Microviridae Life Cycle and Putative Bacterial Hosts

From the current knowledge on Microviridae, only one subgroup (Alpavirinae) was found to contain temperate members (i.e. detected as prophages). This could be linked to a relatively low number of complete bacterial genomes from aquatic environments. However, the absence of Microviridae prophages in Enterobacteria, which have been far more thoroughly studied, as well as the absence of detection of any new prophage even with the new Microviridae genomes described here, suggests that the use of the lysogenic cycle is likely to be rare among Microviridae.

The lysogenic cycle of some of the Alpavirinae and the presence of a horizontally transferred peptidase in several of their genomes made it possible to deduce potential host organisms for these phages. As Alpavirinae prophages have been found only within Bacteroidetes genomes, and the Alpavirinae peptidase genes are most similar to genes from Bacteroidetes genomes as well, members of the Alpavirinae group are likely to infect members of this bacterial phylum. Interestingly, free-living Alpavirinae closely-related to Prevotella prophages were only found in a lung sample, whereas Alpavirinae related to Bacteroides prophages were found in different human stool samples. This finding is consistent with the fact that most of the sequenced Prevotella strains have been isolated from oral samples, while Bacteroides are thought to be primarily associated with gut flora.

The absence of described prophage for Gokushovirinae makes it impossible to be conclusive regarding the potential host(s) of these viruses. Yet, 4 peptidases from human gut gokushoviruses form a common clade with genes from a marine bacterium (Ahrensia sp. R2A130) and 7 human gut bacteria belonging to genera Faecalibacterium and Anaerofustis (Fig. 3B, Fig. S9). Both of these genera are members of the order Clostridiales (phylum Firmicutes), thus it is tempting to speculate that at least some members of the Gokushovirinae might infect Gram-positive bacteria.

Viromes Data Set

A set of 95 viromes available in public databases were downloaded and used in this study (Table S1). Lake Pavin and Lake Bourget viromes were previously described in [45], viromes identified with a number from 12 to 87 in [46], Lake Limnopolar in [6], human lung viromes in [47], human gut and human faeces viromes in [25] and [18], hot springs viromes in [48] and virome JCVI_mv858 is part of the GOS dataset [49]. The 95 viromes span viral communities from the 3 main aquatic ecosystems studied so far (i.e. seawater, freshwater and hypersaline) as well as communities associated with different eukaryotes (fish, coral and mosquito), human lungs and human gut.

Complete Genome Identification

All viromes were assembled (Table S1), and screened for circular contigs with significant sequence similarity with Microviridae genes (tBLASTx, threshold of 50 on bit score). Viromes were assembled using Newbler 2.6 (454 Life Sciences), using the stringent threshold of 98% identity on 35 bp. In addition to the contigs assembly, Newbler software detect putative links between different contigs, usually used to create scaffolds. The contigs linked to themselves (i.e. the end of the contig is similar to the start of it) were thus considered as circular DNA sequences, and searched for Microviridae-like genes via tBLASTx (threshold of 50 on bit score). After a first iteration of this search step, all Microviridae genomes retrieved were used as query in a second iteration, to detect more distant homologies (i.e. contigs with genes not significantly similar to known Microviridae, but significantly similar to contigs retrieved in the first iteration). A fasta file containing the raw sequences of the 81 Microviridae genomes assembled in this study, alongside the annotation of each genome in separated genbank-formatted files, in a zip archive (available through Dryad Digital Repository, doi:10.5061/dryad.8ht80; http://dx.doi.org/10.5061/dryad.8ht80).

In addition, complete bacterial genomes from Refseq database and genomes currently assembled from the NCBI were looked for Microviridae-like genes via tBLASTx (threshold of 50 on bit score) in order to identify new prophages related to Microviridae.

Annotation of Complete Genomes

An ORF prediction was processed using Glimmer 3.02 [50] for each circular contig identified as a Microviridae genome. The predicted ORFs were compared to the sequence database NR using BLASTp and best BLAST hit were conserved. In order to identify and annotate genes not predicted by Glimmer, intergenic regions of the genomes were also compared to NR using BLASTx.

Analysis of Proteins from the New Circular Genomes

Sequences similar to the Major Capsid Protein (VP1 in Gokushovirinae, Protein F in Microvirus) were retrieved from known sequenced genomes and complete genomes generated from viromes. The VP1 sequence retrieved as prophage in Prevotella bergensis genome was not included in the analysis, since the prophage is split among two scaffolds [17], and thus a genome map is difficult to draw from it. Still, its presence did not modify the tree topology.

A multiple alignment of these VP1 protein sequences was done using Muscle [51]. Mega 5 [52] was used to generate a Neighbor-joining phylogenetic tree from this alignment. A custom-designed Perl script was used to calculate the percentage of identity between each pair of protein sequences, based on the multiple alignments computed with Muscle [51]. Jalview [53] was used to visualize RCR I motif manually on the multiple alignment.

Structural Modeling and Model Quality Assessment

VP1 homologues from each of the analyzed Microviridae subgroup were aligned using PROMALS3D [54] and analyzed for the presence and location of insertions with respect to the sequence of ΦX174 protein F [11]. VP1 sequences of Pavin_279 and BMV5 prophage from Prevotella bucalis (GI:282877220; [17]) were chosen as representatives of subgroups Pichovirinae and Alpavirinae, respectively. Three-dimensional model of the Pavin_279 VP1 was generated using a multi-template (ΦX174 F, PDB ID:1CD3 and SpV4 VP1, PDB ID:1KVP) modeling with MODELLER v9.10 [55] The BMV5 VP1 model was obtained using I-TASSER, which uses a combination of ab initio and homology-based approaches for structural modelling [56]. The initial Pavin_279 and BMV5 VP1 models were optimized via multiple rounds of loop refinement with MODELLER v9.10. The stereochemical quality of the models was then assessed with ProSA-web [57]. The final Pavin_279 and BMV5 VP1 models had the quality Z-scores of −6.57 and −6.73, respectively, which were comparable to those of the template structures (-6.4 for ΦX174 F and −6.14 for SpV4 VP1). Comparison and visualization of the structural models was performed with VMD [58] and UCSF Chimera [59].

Peptidase Phylogenetic Tree

Reference peptidase sequences were taken from the NR database, based on a BLASTp of the Microviridae peptidases (threshold of 90 on bit score). Peptidases from Bacteroides and Prevotella genomes were added to the dataset, as Microviridae prophages had been detected in each of these genera. The multiple alignment was computed using Muscle [51], and the maximum-likelihood phylogenetic tree was computed with FastTree [60].

Major Capsid Protein Detection and Affiliation from Linear Sequences

The unassembled reads from the set of viromes used in this study were screened for sequences homologous to major capsid protein, and these sequences were affiliated via a best BLASTx hit against a database formed of all VP1 from the complete Microviridae genomes both published and assembled in this study (threshold of 50 on BLAST bit score).