Analysis of Different Size Fractions Provides a More Complete Perspective of Viral Diversity in a Freshwater Embayment

Most studies of aquatic virus communities analyze DNA sequences derived from the smaller-size “free-virus” fraction. Our study demonstrates that analysis of virus communities using only the smaller-size fraction can lead to erroneously low diversity estimates for many of the larger viruses such as Mimiviridae, Phycodnaviridae, Iridoviridae, and Poxviridae, whereas analyzing only the larger->0.45-μm-size fraction can lead to underestimates of Caudovirales diversity and relative abundance.

ABSTRACT

Inspired by recent discoveries of the prevalence of large viruses in the environment, we reassessed the longstanding approach of filtering water through small-pore-size filters to separate viruses from cells before metagenomic analysis. We collected samples from three sites in Hamilton Harbour, an embayment of Lake Ontario, and studied 6 data sets derived from <0.45-μm- and >0.45-μm-size fractions to compare the diversity of viruses in these fractions. At the level of virus order/family, we observed highly diverse and distinct virus communities in the >0.45-μm-size fractions, whereas the <0.45-μm-size fractions were composed primarily of Caudovirales. The relative abundances of Caudovirales for which hosts could be inferred varied widely between size fractions, with higher relative abundances of cyanophages in the >0.45-μm-size fractions, potentially indicating replication within cells during ongoing infections. Many viruses of eukaryotes, such as Mimiviridae, Phycodnaviridae, Iridoviridae, and Poxviridae, were detected exclusively in the often-disregarded >0.45-μm-size fractions. In addition to observing unique virus communities associated with each size fraction from every site we examined, we detected viruses common to both fractions, suggesting that these are candidates for further exploration because they could be the product of ongoing or recent lytic events. Most importantly, our observations indicate that analysis of either fraction alone provides only a partial perspective of double-stranded DNA (dsDNA) viruses in the environment, highlighting the need for more comprehensive approaches for analyzing virus communities inferred from metagenomic sequencing.

IMPORTANCE Most studies of aquatic virus communities analyze DNA sequences derived from the smaller-size “free-virus” fraction. Our study demonstrates that analysis of virus communities using only the smaller-size fraction can lead to erroneously low diversity estimates for many of the larger viruses such as Mimiviridae, Phycodnaviridae, Iridoviridae, and Poxviridae, whereas analyzing only the larger->0.45-μm-size fraction can lead to underestimates of Caudovirales diversity and relative abundance. Similarly, our data show that examining only the smaller-size fraction can lead to underestimations of virophage and cyanophage relative abundances that could, in turn, cause researchers to assume their limited ecological importance. Given the considerable differences we observed in this study, we recommend cautious interpretations of environmental virus community assemblages and dynamics when based on metagenomic data derived from different size fractions.

INTRODUCTION

Environmental viruses play key roles in shaping microbial communities and structuring ecosystems vital to the health of the planet and our species. They act as major drivers of microbial community succession (1, 2), influence global biogeochemical cycles (3), and may play important roles in the evolution of eukaryotes (4, 5). Metagenomics has helped uncover a remarkable and diverse array of previously undiscovered viruses, leading to the rapid expansion of virus sequence databases (6–14). As new analytical tools and pipelines are developed, constant exploration and scrutiny of new and existing bioinformatic and methodological approaches are necessary when addressing primary questions in environmental virology.

Early metagenomic studies of aquatic virus communities were conducted using DNA extracted from water samples passed through filters with 0.2-μm or smaller pore sizes (15, 16). More recently, the discoveries of viruses with increasingly larger capsid sizes (as reviewed in reference 17) necessitated reevaluation of common filtration methods when attempting to capture entire aquatic virus communities. It is now clear that the <0.2-μm-size fraction excludes many diverse larger viruses, prompting some researchers to use 0.45-μm-pore-size filtration in attempt to sample these larger viruses and obtain a more complete picture of the total virus community (18–21). For example, in one study of soils, use of 0.45-μm-pore-size filters approximately doubled the number of observable virus-like particles compared to that with the 0.2-μm-pore-size filters (22). Thus, the choice of filtration schemes to enrich aquatic samples for viral community analysis is nontrivial and is exacerbated by the presence of larger viruses, particularly those in the Mimiviridae family, that can be captured on 0.45-μm-pore-size filters. Many of these larger Mimiviridae can be excluded from analyses of total virus communities when examining filtrates from 0.22-μm or 0.45-μm filtration, with the potential for slightly higher mimivirus recovery with 0.45-μm-pore-size filters (22, 23). Additionally, viruses within or attached to larger bacterial cells, eukaryotic cells, or particulate material are likely to be captured on the filters and excluded from the analysis of filtrates used to obtain a picture of the viral community. Some cell-associated viruses may pass through 0.22-μm or 0.45-μm filters and could be lost during the viral DNA extraction process if they are within ultrasmall bacteria that are common in aquatic environments (24).

We previously studied virus communities in Hamilton Harbour (14), an urban eutrophic freshwater embayment of Lake Ontario, Canada, that is well studied due to its economic importance and long-term remediation efforts. We mined viral sequences from metagenome libraries derived from >0.22-μm-size fractions originally collected to study cellular communities. Typically, to avoid confounding viral and cellular sequences, viral sequences within this size fraction are filtered out and, therefore, are not included in viral community analyses (25–28). Our study revealed diverse, seasonally dynamic viral communities that were distinct at different sites within the harbor. In most samples, virophages and Mimiviridae were highly abundant, contrasting other studies of similar environments that focused on viral communities in the <0.22-μm-size fraction. This prompted us to examine both <0.45-μm- and >0.45-μm-size fractions collected from the same water samples in attempt to capture the entire virus community and to evaluate the diversity and relative abundances of different groups of viruses partitioning into each fraction. To achieve this objective, we used two common methods for targeting aquatic virus communities: (i) water filtration through 0.45-μm-pore-size filters and concentration of viruses in the filtrate using iron chloride flocculation (29), followed by DNA extraction and sequencing of liberated viruses, and (ii) mining viral DNA from cellular metagenomes, i.e., DNA extracted and sequenced from the material captured on 0.45-μm-pore-size filters. The resulting size-fractionated double-stranded DNA (dsDNA) viral communities were then analyzed using established metagenomics techniques to address the hypothesis that the virus communities observed in each fraction would differ, as each fraction would include unique viral taxa.

(Data presented here were included in a manuscript submitted to a preprint archive [30].)

RESULTS

Virus community composition in <0.45-μm- and >0.45-μm-size fractions.

Contigs annotated as viruses comprised 15% to 20% of all contig annotations in the <0.45-μm-size fractions, while fewer than 1% of annotated contigs were of viral origin in the >0.45-μm-size fractions (Fig. 1). Diverse virus contigs were observed in both size fractions belonging to the virus groups Caudovirales, virophages (Lavidaviridae), Mimiviridae, Phycodnaviridae, other bacterial viruses, other dsDNA viruses, single-stranded DNA (ssDNA) viruses, and unclassified viruses. The viral communities in <0.45-μm-size fractions were similar at the 3 sites and clustered closely together in the Bray-Curtis dissimilarity dendrogram (Fig. 1). While many virus groups were found in both size fractions, most were present at only low relative abundances in the <0.45-μm-size fractions; Caudovirales, other bacterial viruses, and unclassified viruses comprised >99% of all virus contigs in the smaller-size fractions. Caudovirales relative abundances were remarkably consistent in the smaller-size fractions at 82% but were more variable in the larger-size fractions ranging from 2% to 42%. Additionally, there was more variability in the relative abundances of viral groups between sites in the >0.45-μm-size fractions than the <0.45-μm-size fractions (Fig. 1).

FIG 1.

Relative abundances of all virus taxa from the <0.45-μm-size fractions and >0.45-μm-size fractions at each site. The table above the bars shows the number of virus contigs that comprise each bar and the percentage of all contigs that were annotated as viral for each sample. Sample similarity is represented by the Bray-Curtis dissimilarity dendrogram, where blue branches represent <0.45-μm-size fraction samples and red branches represent >0.45-μm-size fraction samples.

Relative abundances of virophages in >0.45-μm-size-fraction samples ranged from 27% to 79%, whereas the relative abundance of Mimiviridae ranged from 10% to 14% (Fig. 1). In contrast, virophages and Mimiviridae made up <0.1% at all 3 sites in the <0.45-μm-size fraction (Fig. 1). Algal viruses in the family Phycodnaviridae were present in all samples but represented a small fraction of the viral community. In <0.45-μm-size fractions, Phycodnaviridae constituted 0.1% of the virus community at all 3 sites and were slightly more abundant in the >0.45-μm samples ranging from 1% to 2% relative abundance. Relative abundances of unclassified viruses ranged from 15% to 16% in the <0.45-μm-size fractions and from 6% to 16% in the >0.45-μm-size fractions (Fig. 1).

Caudovirales community composition and host inference.

Viruses belonging to the order Caudovirales were predominated by members of the Siphoviridae family in all samples (Fig. 2). In <0.45-μm samples, the Siphoviridae comprised >88% of all Caudovirales at the 3 sites, while the relative abundance of Siphoviridae ranged from 54% to 76% in >0.45-μm-size-fraction samples (Fig. 2). Myoviridae was the second most abundant family within Caudovirales in the >0.45-μm-size fractions, making up 14% to 36% of their respective viral communities. In the <0.45-μm-size fractions, Podoviridae was the second most abundant Caudovirales family making up 4% to 5% of Caudovirales in their respective communities (Fig. 2). Overall, Podoviridae were present at low relative abundances in every sample, but with slightly higher values in the <0.45-μm-size fractions. As shown in the Bray-Curtis dissimilarity dendrogram, community composition was similar among sites in the <0.45-μm-size fraction, whereas in the larger-size fraction, only sites 1 and 2 clustered together; site 3 was more similar to the smaller-size fractions than to the other larger-size fractions (Fig. 2).

FIG 2.

Relative abundances of Caudovirales families from the <0.45-μm-size fractions and >0.45-μm-size fractions at each site. Numerical values above the bars are the percentage of Caudovirales in the entire virus community for each sample. Sample similarity is represented by the Bray-Curtis dissimilarity dendrogram, where blue branches represent <0.45-μm-size-fraction samples and red branches represent >0.45-μm-size-fraction samples.

Phages of Methylophilaceae, Methylophilales, and Pontimonas were most abundant in the <0.45-μm-size fractions, together comprising 53% to 60% of all Caudovirales with inferred hosts (Fig. 3). L5 viruses that infect Mycobacterium and Rhodococcus species were the most abundant phages of heterotrophic bacteria in the >0.45-μm-size fractions at sites 1 and 2. Phages inferred to infect Lactococcus were the most abundant phages in the >0.45-μm-size fraction at site 3, comprising 16% of all Caudovirales annotations in that sample. Of the inferred cyanophages, phages of Synechococcus were the most abundant in the <0.45-μm-size fractions and in the >0.45-μm-size fraction at site 3, whereas the >0.45-μm-size fractions at sites 1 and 2 were dominated by phages of Microcystis. Phages of Aphanizomenon were detected in every sample, with higher relative abundances in the larger-size fractions than in the smaller-size fractions. Across all samples, the “other bacteriophages” category contained 85 unique phages, many of which were viruses of heterotrophic bacteria. In the Bray-Curtis dissimilarity analysis, distinct clusters formed for the <0.45-μm-size fractions and the >0.45-μm-size fractions. For both size fractions, Caudovirales communities from sites 1 and 2 clustered together, and the communities at site 3 were less similar to those at sites 1 and 2.

FIG 3.

Bubble plot showing relative abundances of Caudovirales for which hosts could be inferred with abundances denoted by the size of the bubble. Shading represents the percentage of all Caudovirales that each bubble represents. Sample similarity is represented by the Bray-Curtis dissimilarity dendrogram, where blue branches represent <0.45-μm-size-fraction samples and red branches represent >0.45-μm-size-fraction samples.

Virus diversity in the <0.45-μm- and >0.45-μm-size fractions.

Of 694 unique contig assignments for sequences generated from all 6 samples, 93% were detected in at least one sample from the <0.45-μm-size fractions, whereas only 24% were detected in at least one of the >0.45-μm-size fractions (Fig. 4A). Seventy-six percent of unique contig assignments were found only in the <0.45-μm-size fractions, while 7% of unique contig assignments were found only in the >0.45-μm-size fractions. The two size fractions shared 17% of unique contig assignments, most of which were Caudovirales. Ninety-one percent of the unique contig assignments that were identified solely in the <0.45-μm-size fractions were Caudovirales compared to 35% in the >0.45-μm-size fractions. The majority of unique Caudovirales and other bacterial viruses were detected only in the <0.45-μm-size fractions, whereas Iridoviridae and Poxviridae were only detected in the >0.45-μm-size fractions. In addition to the 11 unique Mimiviridae contigs detected in both size fractions, 7 unique Mimiviridae contigs were detected in the <0.45-μm-size fractions, and 11 unique Mimiviridae contigs were detected in the >0.45-μm-size fractions. There were 18 unique Phycodnaviridae contigs detected only in the <0.45-μm-size fraction compared to 6 unique Phycodnaviridae contigs detected only in the >0.45-μm-size fraction. Seven unique virophage contigs were detected in both size fractions, and an additional 3 were detected in the <0.45-μm-size fractions.

FIG 4.

Unique and shared elements in the samples pooled into <0.45-μm- and >0.45-μm-size fractions. (A) Venn diagram of the number of unique contig annotations in each size fraction. Bar charts show the relative abundance of virus groups in each category of the Venn diagram. (B) Venn diagram of the number of unique inferred Caudovirales hosts in each size fraction.

There were 86, 87, and 89 unique Caudovirales contigs with inferred hosts in the smaller-size fractions at sites 1, 2, and 3, respectively. We observed lower richness in the larger-size fractions, which contained 15, 31, and 12 unique Caudovirales with inferred hosts at sites 1, 2, and 3, respectively. All inferred Caudovirales hosts except Salinibacter were captured in the <0.45-μm-size fractions (Fig. 4B). All Caudovirales with inferred hosts with >1% relative abundance (i.e., those in Fig. 3) were captured by both size fractions, except Pontimonas and Pseudoalteromonas phages, which were not detected in the larger-size fractions. Many of the 85 viruses with inferred hosts within the “other bacteriophages” category were only detected in the smaller-size fraction and at low relative abundances.

DISCUSSION

Data considerations and assumptions.

Beyond factors related to data processing programs and pipelines, there are considerations and assumptions related to our methods and results that merit some discussion. The sequencing methods used in this study only capture DNA viruses; thus, the dsRNA and ssRNA virus communities were not considered. Furthermore, sequencing would only capture ssDNA viruses in dsDNA form, for instance, during replication inside their hosts. This is reflected in our results, which showed that most ssDNA viruses were detected in the larger->0.45-μm-size fractions that would include viral DNA within cells. Therefore, the data and conclusions presented in this study apply almost exclusively to the dsDNA virus communities.

Compared to prokaryote and eukaryote sequences, virus sequences are underrepresented in reference databases, influencing the reported diversity of viruses in metagenomes (discussed in detail in reference 14). This inevitably leads to a large portion of sequences in environmental metagenomes remaining unclassified and being discarded at the annotation step in the data processing pipeline. To increase the likelihood of annotating virus sequences, we used the NCBI-nonredundant (nr) database which encompasses reference sequences from both curated and noncurated data sources. Of the virus sequences available in reference databases, some groups are more comprehensively represented than others. For example, Caudovirales genomes are 2 orders of magnitude more numerous than virophage genomes in NCBI RefSeq (data from https://www.ncbi.nlm.nih.gov/genomes/GenomesGroup.cgi). This leads to a higher likelihood of annotating Caudovirales sequences as originating from discrete taxa than virophage sequences. Our observations of disproportionally high virophage abundances compared to those in reference databases aligns with previous observations of their high relative abundances in larger-size fractions (14, 31) and reinforces the notion that sequence annotations are only as good as the databases from which they are derived.

In this study, we inferred the presence and abundance of viruses using the normalized relative abundances of assembled contigs. There are many viral genes present in prokaryotic and eukaryotic genomes due to the intimate evolutionary relationship between viruses and their hosts. For example, it is well known that many bacterial genomes are littered with cryptic prophage sequences (32), and endogenized retrovirus sequences are a common and surprisingly abundant feature of mammalian genomes (33). Unless viral homologues within host genomes are specifically annotated as such, they could be incorrectly annotated as viruses (34). Furthermore, metagenomic data cannot distinguish between viruses that are involved in active infections and inactive viruses that are adsorbed to cells or particles, which could also lead to an overestimation of viral influence. Therefore, a caveat of this study is the fact that we could not distinguish viral DNA integrated into host genomes versus viral DNA packaged within virions.

Comparisons of the <0.45-μm- and >0.45-μm-size fractions.

We observed major differences between the <0.45-μm-size fractions and >0.45-μm-size fractions at almost all levels of taxonomic analysis. DNA extractions from the smaller-size fractions yielded more DNA than the larger-size fractions. DNA was extracted from only one-quarter of the filters for the larger-size fractions, and DNA yield may have been impacted by incomplete lysis, adsorption to particles, or incomplete recovery of supernatant after bead beating. Sequencing library concentrations were higher in the smaller-size fractions, but despite this, there were no apparent trends in the number of reads generated before or after quality control or in the total number of contigs assigned. The number of virus contigs assigned was much higher in the <0.45-μm-size fractions than in the >0.45-μm-size fractions, as expected due to the enrichment of virus particles in the <0.45-μm-size fractions.

For the sake of brevity, we will use only the virus name or taxonomic group when referring to contigs annotated as those viruses. Across all levels of taxonomic analysis, the smaller-size fractions clustered closely together in the Bray-Curtis dissimilarity dendrogram, indicating few differences in the relative abundances of virus groups between sites. The larger-size fractions had higher Bray-Curtis dissimilarity values at all levels of analysis. At the order and family levels, the smaller-size fractions consisted almost entirely of Caudovirales, other bacterial viruses, and unclassified viruses, whereas the larger-size fractions were more diverse, with lower relative abundances of Caudovirales but higher relative abundances of virophages, Mimiviridae, Phycodnaviridae, Iridoviridae, Poxviridae, other dsDNA viruses, and ssDNA viruses. Virus community composition in the smaller-size fractions is typical of freshwater viromes, which are predominated by Caudovirales and other phages. Viruses associated with eukaryotes such as Mimiviridae and virophages are often undetected or reported at very low abundances in these viromes (35–41). However, Roux et al. (31) demonstrated that diverse virophage and nucleocytoplasmic large DNA virus (NCLDV) communities can be identified by mining sequences derived from the larger-size fraction that is typically used to study cellular organisms.

At the level of Caudovirales families, the larger-size fraction from site 3 clustered more closely with the smaller-size fractions than with the other larger-size fractions. While Siphoviridae were the most abundant Caudovirales family in every sample, relative abundances of Myoviridae in the larger-size fractions were higher and more variable between the sites. Previous metagenomic studies of Lake Ontario and nearby Lake Erie found that Myoviridae was the dominant Caudovirales family (39). Likewise, metagenomic data from Lake Baikal showed a predominance of Myoviridae (35, 42), whereas Siphoviridae and Myoviridae relative abundances were similar in Lakes Michigan, Pavin, and Bourget (38, 43). Therefore, our findings of highly abundant Siphoviridae relative to Myoviridae in Hamilton Harbour contrasts with previous studies of Caudovirales communities in other freshwater lakes.

Of the Caudovirales for which hosts could be inferred, the larger-size fractions from sites 1 and 2 had higher relative abundances of cyanophages than the smaller-size fractions and the site 3 larger-size fraction. These higher relative abundances in the larger-size fractions may indicate active cyanophage infections, as Hamilton Harbour is known to experience algal blooms during the summer and late autumn (44). Microcystis phage sequences were particularly abundant in the >0.45-μm-size fractions from sites 1 and 2, suggesting ongoing active infections of Microcystis spp. Microcystis aeruginosa is a well-known bloom-forming cyanobacterium in Hamilton Harbour (45), Lake Ontario (46), and diverse freshwater lakes around the globe (47, 48); therefore, high relative abundances of Microcystis phages were unsurprising. However, in the <0.45-μm-size fractions, contribution of Microcystis phages to the total cyanophage community was low, demonstrating the different insights into the virus communities that the two size fractions provide. Moreover, Microcystis phages were not detected in the >0.45-μm-size fraction from site 3, reinforcing previous findings of distinct virus communities over relatively small spatial scales in Hamilton Harbour (14). We observed more pronounced differences between the size fractions at site 3, where Lactococcus phage sequences were highly abundant in the >0.45-μm-size fraction but undetectable in the <0.45-μm-size fraction. Across all samples and Caudovirales with known hosts, the Lactococcus phages in the larger-size fraction from site 3 were the most abundant of all Caudovirales annotated in that sample. The most abundant Lactococcus phage was Lactococcus phage 16802, a phage that infects Lactococcus lactis, the most abundant lactic acid bacterium in 7 Japanese lakes (49). The discovery of highly abundant Lactococcus phages in a eutrophic embayment of the Great Lakes is intriguing given previous reports that Lactococcus species may mitigate harmful algal blooms by acting as cyanobacteriacidal bioagents (50). Regardless of their potential ecological interactions with bloom-forming species, the high relative abundance of Lactococcus phages is indicative of their ecological importance in this system.

In all <0.45-μm-size-fraction samples, phages infecting Methylophilales, Methylophilaceae, and Pontimonas were the most abundant Caudovirales with inferred hosts. Methylophilaceae is a ubiquitous and metabolically diverse bacterial family (51), with observed enrichments in the Great Lakes surface waters (52). Members of this bacterial family play important roles in global carbon (53) and nitrogen cycling via methanol oxidation linked to denitrification (54) and seem to be ecologically important in Hamilton Harbour, a nutrient-rich environment that supports high microbial biomass. Therefore, the observation of these phages is interesting as they may indirectly influence nitrogen cycling in Hamilton Harbour by acting as a major source of Methylophilaceae mortality. Furthermore, some phages of Methylophilaceae carry auxiliary metabolic genes (AMGs) that enhance methanol oxidation by their hosts (55), which may in turn act as a viral control mechanism of denitrification; however, the most abundant Methylophilaceae phage in our samples (P19250A) does not appear to contain these AMGs (56). Methylophilaceae phage P19250A was originally isolated from the <0.2-μm-size fraction from an oligotrophic freshwater lake but has been detected in the viromes of 8 other diverse freshwater systems, including eutrophic lakes and even Lake Ontario (56). Methylophilaceae grow rapidly and have high biomass yields (53), potentially contributing to our observations of high relative abundances of their phages during a major lytic event. Because of their potential relevance to nitrogen cycling in Hamilton Harbour, further study of these apparently abundant phages may be warranted.

The Pontimonas genus contains only one described strain, Pontimonas salivibrio strain CL-TW6^T, which was recently isolated from a coastal marine environment along with its virus and was hypothesized to occupy environments that experience significant stresses (57), such as Hamilton Harbour. The researchers concluded that its lytic virus was poorly adapted to replicate in this strain, since repeated attempts to sequence its genome were unsuccessful. Given the high abundances of this Pontimonas phage in all <0.45-μm-size fractions from Hamilton Harbour, we presume that it has a permissive host supporting its replication, yet its absence in all the >0.45-μm-size fractions warrants further investigation of this virus-host system that may be prevalent in Hamilton Harbour.

The results of this study demonstrated that Caudovirales were more effectively captured in the smaller-size-fraction samples, whereas the larger, eukaryote-infecting viruses were predominantly observed in the larger-size fractions. Interestingly, a subset of viruses was detected in both size fractions, suggesting that they are highly abundant and likely ecologically important in this environment. These viruses were sufficiently abundant to be detected as free viruses in the water column as well as in the larger-size fraction, potentially indicating ongoing or recent lytic events. Of particular importance to Hamilton Harbour, we detected a variety of cyanophages in both size fractions, including phages of Microcystis (Ma-LMM01 and MaMV-DC), Prochlorococcus (P-HM2), Aphanizomenon (vB_AphaS-CL131), and Synechococcus (ACG-2014, S-PM2, S-CAM7, S-T4, S-CBS2, S-SKS1, S-PRM1, S-CAM22, S-CBS4, S-WAM2, S-CAM8, S-TIM5, and KBS-S-2A). We also detected viruses of eukaryotic primary producers in both size fractions, including Chrysochromulina ericina virus, unclassified chloroviruses, Yellowstone Lake phycodnavirus 2, and Micromonas sp. RCC1109 virus MpV1. Identifying overlap between the virus communities in different size fractions provides the impetus for further exploration, because the presence of these viruses in both fractions allows us to hypothesize that they may be the product of recent or ongoing active viral infections in aquatic environments.

Site comparisons and similarities with a previous study of Hamilton Harbour viruses.

As noted above, across all levels of analysis and both size fractions, sites 1 and 2 were generally more alike than site 3, and the most noticeable differences between sites were in sequences from the >0.45-μm-size fractions. These observations reinforce previous observations (14), allowing us to conclude that distinct virus communities can be observed over small spatial scales in Hamilton Harbour. Furthermore, this pattern of community structure appears to be a stable ecological phenomenon rather than a chance singular observation from a snap-shot study of the environment. In 2015, >0.22-μm-size fractions from Hamilton Harbour were analyzed from samples collected biweekly from July to September from sites 1 and 3. Different methodologies were used to study viruses from larger-size fractions in Hamilton Harbour in 2015 and 2019, including different approaches for water sampling, filtration, and DNA extraction. Furthermore, in these separate studies, samples were collected and processed by different researchers several years apart, and different sequencing machines, raw read lengths, and data analysis parameters were used. Despite these large differences in methodology, in both studies, we observed similar patterns of high virophage abundances in the larger-size fractions, which were particularly pronounced at site 3. As in the 2015 samples, Mimiviridae relative abundances in the >0.45-μm-size fractions did not fluctuate to the same extent as the virophages, despite their presumably intimate association. Like the study of samples collected in 2015, in the present study of 2019 samples, we observed low relative abundances of Phycodnaviridae even though Hamilton Harbour is known to support high algal biomass during the sampling period. Again, this indicates that Mimiviridae may be the dominant alga-infecting viruses in this system.

Experimental and ecological relevance.

In this study, we opted to use 0.45-μm-pore-size filters rather than 0.2-μm-pore-size filters to maximize the likelihood of capturing large, free virus particles in the filtrate in attempt to distinguish between free viruses in the <0.45-μm-size fraction and viruses associated with larger particulate material in the >0.45-μm-size fraction. Mimiviridae and other large DNA viruses are an exception to this generalization, as they could be captured on the >0.45-μm-pore-size filters even if present as free viruses in the water column. Traditional viromics studies that look only at the <0.2-μm- or <0.45-μm-size fractions may exclude large portions of the Mimiviridae, Phycodnaviridae, Iridoviridae, and Poxviridae communities, potentially leading to an underestimation of their influence on the microbial communities. Additionally, members of the Caudovirales community involved in ongoing infections of larger bacterial cells may be overlooked. Conversely, mining viral sequences in the larger-size fractions may not sufficiently capture the diversity of Caudovirales, many of which may be present at low abundances but acting as a seed bank (58) for future infection. The >0.45-μm-size fractions captured the diversity of larger viruses more effectively than the <0.45-μm-size fractions; however, many viruses in the >0.45-μm-size fractions may also be missed by sequencing efforts due to the low representation of viruses in these data sets. Larger cellular genomes comprise most of the DNA in the >0.45-μm-size fractions, leading to low sequencing coverage of viruses and, therefore, likely reducing diversity estimates by only permitting sequencing of the most abundant viruses. While all discrete virophage taxa were observed in the <0.45-μm-size fractions, virophage contigs observed in the >0.45-μm-size fractions were often longer and more numerous. Therefore, high virophage relative abundance estimates from the larger-size fractions were not simply a result of low detection of other groups of viruses. The high relative abundance of virophages in the larger-size-fraction samples could be the result of their intimate associations with cellular or giant virus hosts, as demonstrated in Mavirus and Sputnik virophages (59, 60). Whether these examples are widely applicable to virophage-virus-host systems remains to be assessed, yet the high relative abundance of virophages in all >0.45-μm-size fractions is indicative of their potential influence on the Mimiviridae and their eukaryotic host communities.

These unequivocal results from all samples highlight the potential biases associated with prefiltering water samples to concentrate viruses prior to metagenomic sequencing, demonstrating that the entire virus community cannot be captured in either size fraction alone. While we observed highly diverse and different virus communities in the >0.45-μm-size fractions, the <0.45-μm-size fractions were similar and composed primarily of bacteriophages belonging to the Caudovirales family. However, higher relative abundances of cyanophages were observed in the larger-size fractions, potentially indicating replication within cells during ongoing infections. As noted previously, larger viruses, including Mimiviridae, Phycodnaviridae, Iridoviridae, and Poxviridae, were also captured in the >0.45-μm-size fractions. While some of these larger viruses were small enough to pass through the 0.45-μm-pore-size filters, they may have been attached to cells, including nonhost cells, or particles.

We expect our observations of clear taxonomic differences in virus relative abundances in different size fractions to apply broadly to other environments. Our data show that examining only the smaller-size fraction can lead to underestimations of the abundance and diversity of certain virus groups such as virophages and cyanophages that could, in turn, lead to incorrect assumptions about their limited ecological importance in some environments. Furthermore, the fact that virus community profiles were vastly different when comparing viruses in the <0.45-μm- versus >0.45-μm-size fractions indicates that analysis of either size fraction alone provides only a partial perspective of environmental viruses. Conversely, viruses detected in both size fractions may be candidates for further exploration to discern active viral infections in the environment. Given our observation of unique ecological patterns associated with distinct size fractions, we recommend caution when interpreting environmental virus community assemblages and dynamics based on metagenomic data derived from different size fractions.

MATERIALS AND METHODS

Sampling sites and collection.

Water samples were collected from three sites in Hamilton Harbour on 20 September 2019. Site 1 (43°17.253′′N, 79°50.583′′W) is located in the middle of the harbor at its deepest point, approximately at the location of the mid-harbor site from our 2015 sampling (14). Site 3 (43°16.904′′N, 79°52.527′′W) is closer to the shoreline and corresponds to the nearshore site from our sampling in 2015, and site 2 (43°17.135′′N, 79°51.419′′W) is located at the approximate midpoint between sites 1 and 3. Surface water samples were collected in 20-liter carboys using a bucket and were prefiltered through 100-μm-pore-size Nitex mesh to remove terrestrial plant debris and metazoan tissues, since this biomass could overwhelm the sequencing library. An RBRmaestro³ multichannel logger (RBR Ltd., Canada) was used to measure environmental physiochemical parameters, including temperature, conductivity, dissolved O₂, chlorophyll a, photosynthetically active radiation (PAR), salinity, and pH, and the Secchi depth at each site was also recorded.

Upon collection, water samples were covered and kept in the dark for immediate transportation and processing. After transportation to the lab, approximately 11 liters of each sample was pressure filtered (100 mm Hg maximum pressure) through 0.5-μm-nominal-pore-size GC50 glass fiber filters placed atop 0.45-μm-pore-size polyvinylidene fluoride (PVDF) filters (see Fig. S1 in the supplemental material). Following filtration, filters were stored at −20°C until DNA extraction. Filtrates, or water passing through the GC50 and PVDF filters, were treated with FeCl₃ to flocculate and precipitate viruses according to the protocol of John et al. (29), except 0.45-μm-pore-size PVDF filters were used to capture iron flocs after incubation. Filters with iron flocs were stored in the dark at 4°C until resuspension of viruses was performed according to the standard protocol (29). Resuspended viruses were transferred to polypropylene centrifuge tubes on top of a 1-ml sucrose cushion and pelleted via centrifugation at 182,000 × g for 3 h at 4°C. Supernatants were decanted, and 100 μl of Tris-EDTA (TE) buffer was added to each pellet, which was covered with Parafilm to soften at 4°C before resuspension and nucleic acid extraction.

To fulfill our goal of comparing virus communities in the <0.45-μm- and >0.45-μm-size fractions, we utilized two different capture methods to maximize nucleic acid recovery from these size fractions. For the >0.45-μm-size fractions, there was no additional concentration step beyond the filtration of water through the GC50 and PVDF filters. However, for the <0.45-μm-size fractions, comparable filter extraction methods are not possible. Filters with pore sizes small enough to capture viruses have very low porosity, allowing filtration of only small volumes; therefore, only minute amounts of DNA can be recovered from these filters, prohibiting effective library preparation for metagenomic sequencing. Therefore, for the <0.45-μm-size fractions, viruses were concentrated using FeCl₃ flocculation, a commonly used approach for concentrating free viruses that uses redox chemistry to flocculate and subsequently liberate viruses in water samples (29). FeCl₃ flocculation was selected for its benefits over other widely used methods for concentrating aquatic viruses such as tangential flow filtration (TFF), which is expensive, time consuming, and produces highly variable results (29). Although using a flocculation approach was an essential aspect of this study that allowed us to compare virus communities in the <0.45-μm- versus >0.45-μm-size fractions, it is possible that this method concentrated certain types of viruses more effectively than others, thereby influencing estimates of virus diversity and relative abundance. However, such taxon-specific biases have not previously been identified with this viral recovery method.

DNA extraction and sequencing.

For the >0.45-μm-size fractions, one-quarter of each PVDF and GC50 filter was cut into pieces that were transferred to separate screw-cap centrifuge tubes. Community DNA was extracted separately from each filter using the DNeasy plant minikit (Qiagen, Germany) according to the manufacturer’s protocol, except that bead beating instead of the TissueRuptor was used to disrupt samples. Approximately 0.25 g each of small (212 to 300 μm) and large (425 to 600 μm) glass beads was added to the centrifuge tubes with AP1 buffer and RNase A. The tubes were shaken in a FastPrep 96 bead beater (MP Biomedicals, USA) for 3 cycles of 90 s followed by 1 min on ice between cycles. Final DNA concentrations were measured using a Qubit dsDNA HS assay kit (Life Technologies, USA). Rather than combining equal volumes or amounts of DNA from each filter, we chose to pool the DNA proportionally to avoid biasing the “total” DNA sample with DNA extracted from only one filter. Ultimately, the different amounts of DNA recovered from each filter type were proportionally represented in the DNA sample that was sent for sequencing. For the <0.45-μm-size fractions, virus pellets were resuspended by vortexing, and community DNA was extracted using the QIAamp viral RNA minikit (Qiagen, Germany) according to the manufacturer’s protocol.

Final DNA concentrations were measured using a Qubit dsDNA HS assay kit (Life Technologies, USA), and DNA samples were sent for sequencing at the Microbial Genome Sequencing Centre (MiGS, Pittsburgh, PA). Library preparation was performed using a Nextera Flex DNA sample preparation kit (Illumina, USA) and sequenced from paired ends on the NextSeq 550 (2 by 150 bp) system (Illumina, USA).

Metagenome data processing.

Demultiplexing and adapter removal were performed by MiGS using bcl2fastq. Raw reads were processed according to Palermo et al. (14) with minor alterations to account for the different read lengths. Briefly, quality control was performed using Sickle version 1.33 (61) using a quality cutoff value of 30 and read length cutoff value of 50. Reads were assembled using IDBA-UD version 1.1.3 (62) with k values from 20 to 100 in increments of 20 and a minimum k-mer count of 2. BLASTx searches against the December 2019 NCBI-nr database (downloaded from https://ftp.ncbi.nlm.nih.gov/blast/db/FASTA/nr.gz) were performed using DIAMOND version 0.9.29 (63) with frameshift alignment and very sensitive modes activated. MEGAN6-LR version 6.14.2 (64) was used to annotate contigs using the November 2018 protein accession mapping file (downloaded from https://software-ab.informatik.uni-tuebingen.de/download/megan6/welcome.html) with long-read mode activated and bit score and E value cutoffs of 100 and 10⁻⁶, respectively. Bowtie 2 version 2.3.5.1 (65) was used to map quality-controlled contigs back to the assembled contigs in very sensitive and end-to-end modes. SAMtools version 1.9 (66) was used to extract the mapping information for each contig. Table 1 summarizes the number of reads and contigs at each step in the processing pipeline for each sample, while Table S1 provides statistical information for the subset of contigs annotated as viral in each sample. Figure S2 shows the length distribution of contigs annotated as viral in the samples from each size fraction.

TABLE 1.

Summary of sample parameters throughout the processing pipeline

Sample	Concn (ng/μl)		No. of reads		No. of contigs
	Initial DNA	Library	Pre-QC	Post-QC	Post-assembly (mean length [bp])	Annotated
<0.45 μm
Site 1	61.2	7.22	34,682,792	26,909,942	350,572 (699)	102,574
Site 2	82.7	6.53	33,438,900	25,831,282	329,713 (714)	97,968
Site 3	99.6	6.07	36,189,024	28,542,680	377,348 (745)	134,843
>0.45 μm
Site 1	43.8	4.24	35,222,034	25,937,694	290,644 (714)	164,583
Site 2	34.0	5.82	37,718,578	28,401,490	356,847 (756)	217,580
Site 3	45.8	5.27	40,863,560	28,521,306	192,849 (537)	45,468

Metagenome data analysis.

Data analysis methods were according to those previously described in detail by Palermo et al. (14). Briefly, contig relative abundances were calculated by mapping quality-controlled reads back to assembled contigs and dividing the number of mapped reads by the length of the contig to normalize by contig length. These values were summed for all contigs in a sample and divided by 1,000 to generate a normalized sum that was used to account for differences in sequencing depth between samples. All virus contig assignments were categorized as one of the following based on NCBI taxonomic classifications: Caudovirales, virophages (Lavidaviridae), Mimiviridae, Phycodnaviridae, Iridoviridae, Poxviridae, other bacterial viruses, other dsDNA viruses, ssDNA viruses, and unclassified viruses. Taxonomic assignments for several contigs were less specific than information provided in the literature or NCBI’s taxonomy browser; therefore, these contigs were manually reassigned to more-specific groups (Table S2). Additionally, there were some contigs originally annotated as Phycodnaviridae that are now considered to be more closely related to Mimiviridae and are placed within the proposed subfamily “Mesomimivirinae” (17, 67). These contigs were manually reassigned to the Mimiviridae category. Marseilleviridae and Tectiviridae were present at low relative abundances (<0.01% and < 0.02%, respectively) and were therefore grouped together with the “other dsDNA viruses” (Table S2).

To assess the potential impact of Caudovirales on the prokaryotic community, Caudovirales contigs with inferred hosts that could be inferred from database information were extracted and analyzed separately. Host inferences were determined based on the name of the annotated virus. For example, a virus named “Synechococcus phage” was presumed to infect Synechococcus, whereas a virus named “Mediterranean phage” was presumed to infect an unknown host and was not included in the analysis of Caudovirales with inferred hosts.

Various packages in RStudio were used for data analysis and visualization as follows: virus relative abundances were presented as bar charts and bubble plots using the ggplot2 package (68), unique contig annotations were compared between the size fractions using Venn diagrams generated using the VennDiagram package (69), and Bray-Curtis dissimilarity analyses with the unweighted pair group with arithmetic mean (UPGMA) clustering were generated using the vegan and dendextend packages (70, 71).

Data availability.

DNA sequences generated in this study were deposited in the NCBI Sequence Read Archive under the BioProject accession number PRJNA670357.