Single Virus Genomics: A New Tool for Virus Discovery

Whole genome amplification and sequencing of single microbial cells has significantly influenced genomics and microbial ecology by facilitating direct recovery of reference genome data. However, viral genomics continues to suffer due to difficulties related to the isolation and characterization of uncultivated viruses. We report here on a new approach called ‘Single Virus Genomics’, which enabled the isolation and complete genome sequencing of the first single virus particle. A mixed assemblage comprised of two known viruses; E. coli bacteriophages lambda and T4, were sorted using flow cytometric methods and subsequently immobilized in an agarose matrix. Genome amplification was then achieved in situ via multiple displacement amplification (MDA). The complete lambda phage genome was recovered with an average depth of coverage of approximately 437X. The isolation and genome sequencing of uncultivated viruses using Single Virus Genomics approaches will enable researchers to address questions about viral diversity, evolution, adaptation and ecology that were previously unattainable.


Introduction
Whole genome amplification and sequencing of single microbial cells is a powerful new tool in the field of microbial genomics, enabling direct examination of the genomic contents of individual cells without the need of cultivation [1][2][3].Microbes are found in nearly all environments (e.g., human microbiome, rhizosphere, aquatic ecosystems, soils, air) performing essential roles in processes such as biogeochemical cycling [4], degradation [5], metabolism [6], and forming the foundation of environmental food webs [7].Sequencing of single cells permits the study of previously uncharacterized microbes from virtually any environment, thus enabling the direct and comprehensive analysis of a microbe's genetic and metabolic repertoire.Flow cytometry [1,8] and micromanipulation [9,10] have aided in the advent of single cell isolation and sequencing by providing access to individuals from naturally occurring consortia or pure cultures.A reaction called multiple displacement amplification (MDA) [11][12][13], which uses the high-fidelity processive capabilities of the phi29 DNA polymerase, can amplify the genome of a bacterial cell more than a billion-fold generating the micrograms of genomic DNA typically required for DNA sequencing either via Sanger sequencing [14], 454 pyrosequencing [15], and/or Illumina platforms [16].While some sequence is lost due to non-specific amplification or damage to the single genome copy, as much as .90% of the genome has been recovered from single cell sequencing [16,17].Small MDA reaction volumes were shown to improve amplification from single viral DNA molecules [18] and single cells [15].Recently, Rodrigue et al., showed a consistent increase in the total genome coverage of Prochlorococcus single-celled amplified genomes by using a duplex-specific nuclease to degrade highly abundant sequences apparent after amplification; thereby increasing the coverage of low abundant sequences.
While most single cell studies have focused on bacteria and cyanobacteria, single virions have yet to be isolated and genomically described using similar mechanisms.Viruses are ubiquitous and the most numerous and diverse biological entities on our planet [19].Nearly all aspects of our lives are influenced by viruses through shaping the environments that surround us [20], our immune responses [21] and even our genomes [22].The field of environmental viral metagenomics has gained momentum over the past several years [23][24][25][26][27][28]; however, sequencing of individual environmental viral genomes is currently dependent on the establishment of cultivable virus-host systems.With this in mind, if less than one percent of microbial populations can be cultured using standard microbiological techniques due to incongruencies in direct counts versus cultivatable microbes [29][30][31][32], then only a very small number of viruses have the likelihood of being genomically described.Currently, viral genomic sequences are lacking in public databases, with the exception of human viruses and those of agricultural and industrial significance (e.g.Lactococcal phages).Clearly, a better understanding of virus diversity and evolution will not be achieved until the genomes of a broad range of viruses are available.Here we introduce an approach for isolating and characterizing the genomes of viruses called ''Single Virus Genomics'' (SVG) (Figure 1).The benefits of SVG will be far-reaching, enabling novel virus discovery in a variety of clinical and environmental settings, altering our understanding of virus evolution, adaptation and ecology and facilitating the interpretation of viral genomic and metagenomic data by providing suitable reference genomes.

Single virus isolation
Flow cytometric methods have been optimized for [33] and used on natural viral populations for enumeration purposes [34][35][36].Therefore, for this study, flow cytometry was used to sort a mixed viral assemblage consisting of two known viruses; E. coli bacteriophages lambda and T4.To increase the accuracy of detecting a single viral particle, a fluorescence-activated cell sorter (FACS) AriaII with a forward scatter photo multiplier tube (PMT) was used, which enabled more sensitive detection and lower size thresholds (Figure 2).While 96-and 384-well microtiter plates would have been optimal for high-throughput processing of viral assemblages, we were unable to reliably recover single virions from plate wells.The majority of wells (98%) contained no viruses evidenced via polymerase chain reaction (PCR) amplification of specific loci for each bacteriophage.Therefore, as an alternative strategy, viruses stained with SYBR Green (Invitrogen) were sorted directly onto cooled agarose beads applied to 'multi-well' polytetrafluoroethylene (PTFE) microscope slides to increase virus capture efficiency, as well as, to maximize the recovery of highquality template DNA required for the MDA reaction.PTFE slides were chosen due to the ability of defining each sorting event since they possess distinct regions (wells) where agarose beads were positioned.Overlaying the nanoliter droplet containing the virions with additional agarose simultaneously embedded and stabilized the sorted viral particles in preparation of downstream processing.

Single virus validation: Confocal microscopy and locispecific PCR
To confirm isolation of single viruses, Confocal Laser Scanning Microscopy (CLSM) was performed to detect the fluorescently stained virions embedded in agarose [37,38].CLSM was chosen to obtain greater confidence that only a single viral particle was contained within an agarose bead through 3D rendering of stacked images surrounding the viral particle (Figure 3A).Additionally, Figure 3B demonstrates the utility of CLSM to detect the relative fluorescence of a single stained virus particle above background.Once successful candidates were identified, whole genome amplification via MDA was performed in situ in order to minimize the potential for virus loss, reduce genomic shearing, and contamination.Multiplex PCR using T4 (gp23, major capsid gene) and lambda (integrase gene) specific primers was performed on the amplified genomic material to confirm the genotype of the virus and indicated that the isolated viral particle used in this study was phage lambda (Figure 4A).To further confirm this result, two additional loci specific to lambda were targeted including the superinfection exclusion protein B and repressor genes (Figure 4B).The results confirmed our initial finding that we had isolated and amplified the genome of phage lambda.A subsequent experiment to quantify virus particles within agarose droplets using CLSM indicated that 75% contained 1 or .1 (1-5), viruses (Table S2); and amplification of genomic material via MDA was successful in 92% of virus-containing droplets.Multiplex PCR using T4 and lambda-specific primers on amplified genomic material was successful for 25% of the droplets and positive results were only found for those droplets containing one or more virus.

Sequencing, reference mapping and De novo assembly
The 48,502 bp double stranded DNA phage lambda was sequenced using 454 GS FLX titanium pyrosequencing to an  average coverage of 437X across the genome (Figure 5C), (ranging from 0-2000X).With the exception of the first 5 bp, the complete genome of lambda was recovered (Table S3).Lacking the first 5 bp is likely due to a reported artifact of MDA reactions where the ends of linear DNA segments are underrepresented [12,39].It has been reported that MDA is biased against genomic areas of high GC content [40], however, our data suggests otherwise as there was higher coverage in the regions of greater %GC, shown in Figure 5A where the bars indicate the GC above or below the average (average GC of phage lambda is 49%).We expected to achieve .600Xcoverage from the 99,911 sequencing reads (mean read length of 361.6 bp) that were generated if all sequences produced were from the lambda template DNA.Reference mapping to the lambda genome (NCBI Accession J02459) indicated that 38,505 (38.5%) reads did not recruit to the genome and were further termed 'unmapped'.BLASTX analysis was performed on these sequences, which resulted in 22,411 reads (58.2% of the unmapped set) annotated based on homology to public sequences (Table S4).The stringent settings used during reference mapping prevented the recruitment of 116 sequences classified as E. coli lambda phage through BLAST.The majority of the annotated unmapped sequences were classified within the Pseudomonas genera (12.9% of total reads).To assess errors within the amplified lambda genome SNP analysis was completed, which resulted in the detection of 17 SNPs across the genome (Table S6), however, it is difficult to determine if these errors arose during amplification, 454 pyrosequencing, or maintenance of ATCC cultures.Deletion-Insertion Polymorphisms (DIPs) are also given in Table S5; Two DIPs corresponding to reference positions 31619 and 39143 are deletions causing a frameshift within essential phage proteins.The deletions (-) are found in 37.5 and 38.6 percent of the reads covering the corresponding positions in the lambda repressor (cl) and DNA replication proteins respectively.It is likely, therefore, that the polymorphism would not be present in the phage population but perhaps are a result of MDA and/or 454 pyrosequencing artifacts.
De novo assembly was performed with the GS De Novo Assembler Software (i.e., Newbler, 454 Life Sciences) to assess the utility of these methods for use on unknown SVGs (Table S6).Optimal coverage depth for assembly is between 15-25X (personal communication 454 Life Sciences; Newbler manual); therefore the number of reads was randomly reduced to yield 30X (4,700 reads) and 22X (3,400 reads) coverage, with the latter generating the highest quality assembly (Figure 6A).Although reducing coverage resulted in the highest quality assembly based on assembly metrics (i.e., fewer contigs, longer length, greater N50 [41]-see methods for details), using all reads (Figure 6B) resulted in near complete coverage of the genome, however with shorter contigs.In an effort to increase contig size while retaining genome coverage, the redundancy of reads among overrepresented contigs from this assembly was reduced and the data re-assembled, as seen in Figure 6C [16].This method of bioinformatic normalization of the data resulted in larger contigs coupled with almost complete coverage of the genome (.99%).The utility of SVG approaches for the study of uncultivated viruses will ultimately depend on the success of de novo assembly due to the lack of suitable reference genomes.Recently, SVG approaches were applied to virioplankton samples collected from the Southern California Current (Zeigler Allen, et al., in prep) followed by bioinformatic normalization during de novo assembly procedures, which similarly improved assembly statistics.

Discussion
The Single Virus Genomics approach described here enabled, for the first time, isolation and whole genome sequencing of an individual virus; a significant technical achievement that has the potential to alter the course of virological research.Further optimization of SVG will pave the road to high throughput processing of uncultivated viral assemblages, advancing studies of viral diversity, evolution, adaptation and ecology.These include efforts to improve the occurrence of single virus particles in agarose droplets.Although genotyping of the single lambda phage particle that yielded the sequence data for this study was successful (Figure 2) and can clearly be accomplished, the overall low success rate (25%) of specific PCR post-MDA evidenced was possibly due to a lack of purification of the MDA products prior to genotyping with T4 and lambda-specific primers, which is recommended (see materials and methods).In addition, the high success rate of MDA (92%), as evidenced by gel electrophoresis, could represent nonspecific amplification in addition to amplified viral DNA.Optimization of flow-sorting parameters should increase the likelihood of capturing individual viruses.
While complete coverage of the lambda phage genome was our goal, the first 5 bp of the genome were missing, perhaps due to DNA breakage or the linear nature of the molecule.To date, complete coverage of bacterial genomes has not been reported, suggesting that single cell genomics projects suffer from similar obstacles.We make an effort to reduce DNA breakage through the immobilization of viral particles in an agarose matrix which minimizes DNA damage during viral particle isolation and genome amplification.When applying SVG techniques to unknown viruses, it may be difficult to determine if the ends of linear genomes have been captured.However, approaches for genome closure such as primer walking could be attempted on the amplified viral genomic material if complete coverage is critical.A recent study of MDA on phage lambda genomic DNA also showed underrepresentation of DNA termini and reported using a ligation reaction prior to MDA to generate circular molecules, thereby overcoming this bias [42].A similar approach can be adopted if future data suggests it is necessary.
Background DNA synthesis or nonspecific amplification is commonly reported during amplification using the MDA reaction [18].Nonspecific amplification has been attributed to contaminating DNA emerging from reaction reagents and/or through a mechanism that enables amplification from the random hexamers within the reaction mixture.The average coverage retrieved here was lower than expected, most likely due to non-specific amplification.As mentioned previously, steps were taken to reduce the likelihood of contaminating DNAs being introduced into our sample following flow cytometric sorting.However, during the sorting process we acknowledge that free DNA as well as multiple viral particles may be co-transported.Treatment of viral assemblages with DNase I and examination of virus containing agarose beads using confocal microscopy was used to address these issues.Additionally, there is a higher likelihood of nonspecific DNAs preferentially amplified due to the lower quantity of template viral DNA as opposed to single bacterial cells as a result of the significant difference in particle (cell) size and genomic DNA content (25-100 nm; ,1.5femtograms for viruses, as opposed to 0.2-1.5 um; ,14femtograms for bacteria).To address this potential shortcoming the incubation time of genome amplification was reduced and we took advantage of the massively parallel, high-throughput capabilities of pyrosequencing to ensure both adequate coverage of the lambda genome and to examine the nature of any nonspecific amplification.A potential source of contaminants is high molecular weight DNA fragments present in commercially available phi29 polymerases.A recent report found no manufactured enzyme to be contaminant-free and that levels of contamination varied among enzyme and buffer reaction lots [43].Specific 16S bacterial DNA sequences were detected as contaminants in our process.The identity of the microbial contaminants present in no template control (NTC) MDA reactions using 16s rDNA PCR and sequencing were determined and the most abundant taxa are similar to those found in our taxonomic classification of unmapped reads, in particular Burkholdaria and Pseudomonas (Table S7).We have not attempted to distinguish between the particle handling, MDA, 16S PCR and PCR product sequencing steps as the potential source of the contaminants.While bioinformatic curation of data can be performed to identify potential contaminants that are not related to the target viral molecule, conservative approaches are necessary in their removal so that pertinent data is not lost.Therefore, steps to reduce the amount of exogenous DNA-based contamination prior to sequencing are imperative and are especially relevant when working with unknown viral isolates.For example, testing new enzyme and reagent lots prior to use and the reduction of free DNA through nuclease treatment should help to reduce nonspecific amplification.
A number of important factors must be taken into consideration when applying SVG approaches to natural, unknown assemblages of viruses.Although it is possible to capture and immobilize RNAcontaining viruses using flow cytometry, the MDA reaction will not work on RNA templates.However, a reverse-transcription step prior to amplification would address this issue and we are currently evaluating the utility of whole transcriptome amplification (WTA) to amplify individual RNA viral genomes.Genotyping of previously unknown viruses is another topic that deserves careful consideration since PCR using virus-specific primers conserved across all viral groups is not option.While certain techniques such as randomly amplified polymorphic (RAPD) PCR may successfully produce unique viral genomic fingerprints [44], we are also evaluating alternative strategies such as optical mapping [45] and automated artificial neural networks using known morphological characteristics and fluorescence data gathered from reference phage for training [46].
Single virus genomics has the potential to dramatically influence a wide variety of fields that will benefit from whole genome sequence data produced from previously uncultivated viruses; including (but not limited to) viral and microbial ecology, evolutionary biology, epidemiology, immunology and other clinical and agricultural-based sciences.In addition to enabling novel virus discovery and facilitating comparative genomic analyses, SVG will also provide an 'anchor' for metagenomic studies by supplying relevant reference genomes.Reference viral genomes will not only assist assembly of metagenomic data, but will help to address questions surrounding genetic and functional biodiversity, as well as the representation of individual viruses within a community.Lastly we anticipate that the production of new reference viral genomes will improve our ability to classify previously unidentified sequences within viral metagenomes, effectively bridging the gap between genomic and metagenomic studies.

Viral suspensions
Bacteriophage standards for T4 and lambda were obtained from ATCC (ATCC 11303-B4 and 23724-B2, respectively).Stocks were diluted in 0.1 mm-filtered TE (Tris-EDTA, pH 7.2, Invitrogen) followed by filtration through a 0.22 mm Pall syringe filter.The viral particles were not fixed prior to flow cytometry, as is typical, due to insufficient evidence that the fixative would not inhibit downstream reactions.Table S3 Reference mapping statistics (all sequence lengths are given in bp).

(PDF)
Table S4 BLAST analysis of unmapped read sequences following reference mapping.(PDF)

Figure 1 .Figure 2 .
Figure 1.Flow diagram depicting SVG methodology.Viral suspensions are sorted via flow cytometry onto PTFE slides with 24 distinct wells containing agarose beads.Viral particles are then embedded within the agarose bead by overlaying with an additional layer of agarose.Lastly, MDA is performed in situ.doi:10.1371/journal.pone.0017722.g001

Figure 3 .
Figure 3. Confocal laser scanning micrograph of sorted viral particle embedded in agarose bead.A) Three dimensional reconstruction of syber green I stained viral particle within depth of agarose bead verifying a single sorted event.Inset: higher magnification of viral particle.B) Profile plot of relative fluorescence for a stained viral particle in an agarose bead.The blue line through the viral particle (green) is the reference for the inset graph showing the relative fluorescence.doi:10.1371/journal.pone.0017722.g003

Figure 5 .
Figure 5. Lambda genome attributes and coverage.A) GC plot with bars indicating %GC above or below the average of 49%, B) Genome map of lambda (adapted from http://img.jgi.doe.gov), and C) Reference mapping of SVG reads to phage lambda, x-axis is genome position, y-axis is %coverage.doi:10.1371/journal.pone.0017722.g005

Figure 6 .
Figure 6.De novo assembly of reads followed by reference mapping to evaluate assembly.A) Filtered sequences randomly to 3400 reads, approximately 22X coverage of the lambda genome, B) All reads (99,911), C) Normalization of assembly by reducing redundancy of overrepresented sequences from (B). doi:10.1371/journal.pone.0017722.g006 Table S6 De novo assembly statistics.(PDF) Table S7 Contaminants found in 16S PCR analysis of MDA reactions.(PDF)