|Home | About | Journals | Submit | Contact Us | Français|
Over the past decade researchers have begun to characterize viral diversity using metagenomic methods. These studies have shown that viruses, the majority of which infect bacteria (bacteriophages), are likely the most genetically diverse components of the biosphere. Here we briefly review the incipient rise of a phage biology renaissance catalyzed by recent advances in next generation sequencing. We explore how work characterizing phage diversity and their lifestyles in the gut is changing our view of ourselves as supra-organisms. Finally, we discuss how a new appreciation of phage dynamics may yield new applications for phage therapies designed to manipulate the structure and functions of our gut microbiomes.
From Alfred Hershey and Martha Chase’s studies indicating that DNA was the genetic material 1, to Francis Crick and Sydney Brenner’s experiment establishing the triplet nature of the genetic code 2, bacteriophages (“phages”) have helped define fundamental components of modern biology. Most of the tools for early molecular biology arose from the work of phage biologists 3. The first genomes sequenced were from phages and other viruses. The first comparisons of multiple genomes were carried out on Lactobacillus and Mycobacteria phages. These early studies showed that there was extensive diversity in essentially every phage community. Now it is clear that viruses are the most diverse and uncharacterized components of the major ecosystems on Earth 4, and that viruses have intricate roles in ecosystem function, far beyond simple predator-prey dynamics 5 (Box 1).
Most of life on Earth exists as Bacteria and Archaea in the ocean, sediments, land, and potentially the deep biosphere 117. In the early 1980s, people became aware that there are literally millions of actively growing microbes per milliliter of seawater. Marine phages were subsequently rediscovered in 1989 and efforts to characterize the impact of phage lifecycles on planetary-scale biogeochemistry were initiated 118–121. As an example of how dramatic phage effects can be, consider the widespread cyanobacteria clades Prochlorococcus and Synechococcus. These two unicellular algae carry out about half of the world oceans’ primary production. They are infected by cyanophages: variations in cyanophage and bacterial concentrations are tied to daily and seasonal cycles 120. In the 1990s, isolation and characterization of phages infecting Synechococcus and later Prochlorococus not only revealed that cyanophages are widespread, often infecting 40–50% of cyanobacteria, but that they kill 10–50% of their hosts daily, 119, 120 rapidly driving the diversification of their hosts as they co-evolve resistance 122 and simultaneously driving carbon into a dissolved form when bacterial cells lyse. Under lower resource conditions, cyanophages may enter a lysogenic state. A growing list of genes that are important for bacterial host metabolism and function have been found in marine phages, including photosystem genes that can increase photosynthetic output and maintain energy production during infection so that phages can lyse their host cells 51, 123–125. Lysogeny may be an important lifestyle under a number of suboptimal conditions, including when host abundance or nutrient abundance is low 126.
Bacteria can use temperate phage to enable invasion of new habitats by sacrificing part of the population through phage lysis. The released phage will target competitors but allow bacterial kin harboring the prophage to survive since they are resistant to attack by a process called superinfection exclusion127–130.
The concept that a virus can have a beneficial effect beyond that experienced by its host cells is not conceptually novel. Three-way symbioses have been well described in macro-and micro-ecosystems. For example, a symbiotic bacterium that inhabits pea aphids protects them from a wasp that can otherwise lay eggs in the haemocoel of the aphids; a phage-encoded toxin expressed by the bacteria confers this protection 131, 132. Drought, heat and cold tolerance are conferred to plants through viruses 127, 133. In Yellowstone National Park, a fungal endophyte infecting panic grass confers thermal tolerance, allowing the grass to grow in hot geothermal soils. The fungus alone is not heat tolerant without the virus that infects it 134. In a more phage-related example, there is a reported case of phage-associated corynetoxin synthesis in the bacterium Rathayibacter toxicus (formerly Clavibacter toxicus) that colonizes ryegrass plants; the toxin makes the grass toxic to grazing animals such as sheep 135.
At the same time, the clinical world has become increasingly interested in phage-based therapeutics because of the increased prevalence of antibiotic resistant bacteria 6. The idea of using phages as therapeutic tools is not new. Félix d’Herelle, co-discoverer of phage, recognized the potential medical applications nearly a century ago 7 and his first phage therapies were tested as early as 1919 8. However, a rudimentary understanding of the composition and dynamic operations of the human microbiome, a lack of knowledge of phage biology, and poor quality control during production of phages made this therapeutic approach unreliable 9.
This Review details recent advances on the rising field of viral metagenomics with an emphasis on our current views of phage communities associated with the human gut. It does not discuss many of the resistance mechanisms such as Clustered Regularly Interspaced Short Palindromic Repeats (CRISPRs), or the extremely large field of eukaryotic viral discovery that is being propelled by metagenomics, since these topics have been reviewed elsewhere 10–14. Instead, we focus on viruses that infect bacteria. Box 2 provides examples of some challenges facing phage metagenomics at the present time, some of which are discussed below.
The introduction of small subunit (SSU) ribosomal RNA (e.g. 16S rRNA) as a reliable prokaryotic phylogenetic marker 15–17 opened the door to remarkable insights about microbial community diversity and dynamics. Phylogenetic marker ‘envy’ rapidly ‘infected’ the psyche of phage biologists: they did not have an SSU rRNA equivalent and there was, and is, no conserved protein or gene enabling a similar characterization of all or the majority of phages present in a sample. Efforts to characterize phage diversity focused on characterizing partially conserved fragments of phage genes such as polymerases. However, this method was useful only within certain viral families 18–20. Horizontal gene transfer further complicates the use of marker genes in phages. For example, most Caudovirales have identifiable conserved genes including terminases, portal proteins, capsid proteins, but horizontal transfer and recombination events generate extensive genome mosaicism that challenges phylogenic characterization 21. The arrival of next generation sequencing (NGS) together with methods for purifying virus-like particles (VLPs) set the stage for defining viral diversity based on shotgun sequencing.
Although viral particles outnumber microbial cells 10:1 in most environments, viral DNA represents 2–5% of total community DNA 22–24. For this reason, it is often desirable to separate viruses from microbial cells. If sample volume is large and viral density low (such as in ocean environments), tangential flow filtration can be used to remove large particles and concentrate VLPs. For solid samples with high viral density e.g., feces, a common approach is to resuspend the material in an osmotically neutral buffer followed by one or more steps designed to remove large particles (e.g., cesium chloride density gradient ultracentrifugation 25 and subsequent filtration) (Fig. 1). This procedure has been successfully applied to fecal material that had been stored at −80°C for several years, without any pre-processing of the sample, indicating that VLP structures are quite stable to freezing and thawing 22.
After VLPs are purified, non-encapsulated free nucleic acids are removed by treatment with DNase and RNase, and VLP nucleic acids are then isolated. The methods chosen will determine purity, influence DNA and RNA yields, and represent a selection step that can bias interpretation of virotype abundance and viral community diversity 26. Unfortunately, the yield of DNA following extraction of nucleic acids from purified VLPs is often below the required minimum for sequencing. Therefore, a variety of amplification methods have been developed such as random amplified shotgun library RASL 27, linker-amplified shotgun library LASL 28 and others 29–32. Caveats concerning random PCR amplification of viral DNA include inherent bias due to exponential amplification of mixed templates, uneven coverage of viral genomes, and its limitation to dsDNA templates. Another common method uses the phage-derived phi29 polymerase for multiple displacement amplification (MDA) 33. MDA takes advantage of the high processivity of this DNA polymerase (>70,000 nucleotides) and its strong strand-displacement capability, which permits amplification of complete viral genomes. The result is a very fast method that can efficiently amplify minute amounts of both ssDNA and dsDNA. While fast, the method is not without flaws, including over-amplification of small circular ssDNA viruses 34 and potential chimera formation 35, 36. Procedures for avoiding some of these limitations continue to be developed 37, 38, including a novel transposon-based method for rapidly generating DNA libraries from small quantities of dsDNA 39. RNA viruses can be sequenced by reverse transcription followed by application of the protocols described above. Alternatively, Whole Transcriptome Amplification (WTA) approaches can also be used 40.
Although sequencing costs are falling at an astonishingly rapid rate as newer technologies offer higher degrees of parallelism (greater numbers of reads per run; multiplex sequencing using sample-specific DNA barcodes), read length matters 41. When characterizing a viral community where most of the sequences obtained are novel and enriched in regions of low complexity repeats, accurate assembly and taxonomic assignment benefit from the longest reads possible 41, 42. The earliest NGS analyses were powered by the 454 GS20 instrument with ~100 nt reads. Advances in pyrosequencing technology, including today’s FLX+ platform, have produced average read lengths that exceed 800 nt 22, 23, 43–46. Most recently, analysis of metagenomic datasets generated by deep sequencing of total microbial community DNA with highly parallel Illumina instruments showed that the percentage of reads with similarity to known viral sequences was generally less than 0.01% 30, 47, 48. This low value is in part due to the short read length (≤100 nt). However, the percentage increases when VLPs are purified 32. Other studies have obtained better assignments by pre-assembling the short reads 49, 50.
In summary, viral metagenomics has opted for technologies that prioritize long read length over platforms with short read length and higher throughput. However, as the latter platforms approach 250nt read lengths and 250 million reads per lane (Illumina HiSeq 2500), and the cost per read falls, we will undoubtedly see a very rapid transition to these types of sequencers — as long as improvements in assembly algorithms keep pace (Fig. 1).
To address the question of viral community composition, shotgun metagenomic sequences are typically compared to individual viral genomes. Although public sequence databases have expanded considerably — from 500 viral genomes as of 2007 to over 3000 full viral genomes to date — the number of deposited genomes is far less than the expected number of virotypes present in 100 liters of seawater 51. Compounding the problem, existing databases include very few viral proteins in their training sets, meaning many taxonomic assignments are based on proteins transferred between a virus and a microbial host or that are present in prophages and described as part of a microbial genome. Databases with a particular focus on viruses are under development and include a CLAssification of Mobile genetic Elements (ACLAME) 52 and the Phage SEED 53. Novel data analysis pipelines are also being constructed to improve the accuracy and efficiency of homology searches 54–61.
Once taxonomic and functional assignments have been made for a given sample, a viral community profile can be created that characterizes the diversity present in that sample. Multi-dimensional reduction methods such as Principal Coordinate Analysis (PCA) and hierarchical clustering have been used to visualize similarities among viral communities, and methods such as supervised learning can help to identify discriminatory features.
Given that most of the viral metagenomic data lack similarity to entries in databases, similarity-independent methods have been developed to better understand viral community structure. One example, PHACCS (Phage Communities from Contig Spectrum), was designed to quantify virotypes 62, 63 based on the assumption that if a virotype is present in high abundance in a VLP sample, it is more likely to be assembled into a large contig. Moreover, we can posit that if assembly of a single sample dataset allows prediction of community structure and diversity, pooling two samples together and performing a cross-assembly analysis can determine inter-sample diversity (e.g., using MaxiPhi 64). Another alternative for identifying shared viruses among different samples comes from crAss 65, an algorithm that allows for the simultaneous cross-assembly of all the samples in a dataset as opposed to the pairwise assemblies used in MaxiPhi. As more tools are developed, special attention should be given to the assembly parameters to prevent mixed assemblies and chimeras between viral genomes (Fig. 1).
The gut provides an enticing place to characterize the role of phages in community assembly and dynamics. Assembly of the human gut microbiota begins at birth with evolution toward an adult-like configuration during the first three years of life 66. The importance of early environmental exposures is emphasized by the fact that the overall phylogenetic composition of the gut microbiota of adult monozygotic twins is not significantly greater than that of dizygotic twins, and family members share a higher degree of similarity than unrelated individuals living in different households. These patterns are robust to different cultural traditions, and the observations about mono- versus dizygotic twins apply to infants and children as well as teenagers and adults 66, 67. Microbes in this densely populated ecosystem are engaged in a constant fight for nutrients and survival. Peristalsis moves an ephemeral menu of dietary components along the cephalocaudal axis of the intestine and microbial members face the omnipresent threat of washout from the gut ‘bioreactor’. Maintaining a foothold in this ecosystem depends not only on physical interactions with the perpetually renewing mucus layer and partially digested food particles, but also on functional interactions with other community members. Preserving functional redundancy contributes to community resilience, with horizontal gene transfer providing an opportunity to constantly refashion the genomes (and by extension the pan-genomes) of species-level phylotypes. Each adult appears to harbor a persistent collection of one or at most a few hundred species in their intestines, although strain level diversity is great 24, 68,69. While the proportional representation of taxa changes as the community responds to various environmental perturbations, intrapersonal variation in species content is considerably less than interpersonal differences 66, 67, 70, 71.
As our knowledge of inter- and intrapersonal variations in the microbiota expands, a lagging question has been the role of phages in shaping community properties. Although a number of individual phages have been extensively characterized (providing an important genomic context against which metagenomic data can be interpreted), recently much more attention has been given to phage dynamics at a microbial community level. In 2003, the first report of a human-associated gut virome was published; it described the results of shotgun (Sanger) sequencing of VLPs isolated from a fecal sample obtained from a single healthy adult. The identifiable fraction of the virome was dominated by phages, including temperate phages (‘prophages’ are defined here as temperate phages in their host-incorporated state). This report estimated that there were 1200 different virotypes in the single sample analyzed, with the majority assigned to the Siphoviridae family 28. Siphoviridae and temperate phages have subsequently been reported to be the most abundant identifiable viruses in other sampled fecal viromes, followed by members of the Podoviridae 22, 23, 72.
The prominence of Microviridae in adult human gut microbiota was initially dismissed as an artifact of the MDA method, which has a preference for ssDNA. However, a novel branch of the Microviridae has been identified recently from prophages present in the genomes from members of two genera in the Bacteroidetes: Bacteroides and Prevotella 73. These two genera are prominently represented in the microbiota of adult human populations living in a number of diverse geographic areas 66, 74. Another study characterizing Microviridae from healthy human donors also clustered these novel viruses with Bacteroides and Prevotella prophages 72, suggesting that Microviridae could be an important viral family in the human gut and that what was previously considered to be an exclusively lytic phage can integrate into bacterial hosts in an environment that encourages a temperate (lysogenic) viral-host lifestyle (see Fig. 2 and below).
Marine environments can contain millions of different virotypes in a single sample 51. None of the human fecal samples characterized thus far has had greater than 1500 virotypes. Moreover, the ratio of virotypes to species-level bacterial phylotypes in the ocean is 10:1 but closer to 1:1 in the gut 22. Microscopy counts have further validated these estimated ratios, demonstrating an average of 108–109 VLPs per gram of feces compared to ~109 bacterial cells per gram of feces72. These findings also support the notion that phage exhibit a more temperate lifestyle in the gut, in contrast to the active kill-the-winner viral-bacterial dynamic manifest in marine environments.
The temperate lifestyle of phages observed in the gut, along with bacterial CRISPR elements involved in conferring immunity to infection with foreign DNA (including phage), have facilitated bioinformatics efforts to discover new phages in this body habitat. Recently, Stern et al, 75 used datasets obtained from deep shotgun sequencing of human fecal community DNA24 to extract CRISPR spacers present in bacterial genomes. These spacers were then used to query datasets of pyrosequencer reads produced by shotgun sequencing of fecal VLP-derived DNA22, 23 as well as contigs assembled from VLP and total fecal community DNAs. The results led to identification of contigs in shotgun sequencing datasets of fecal microbiomes that were not previously recognized as viral, assignment of these sequences to bacterial hosts, and an appreciation of the wide distribution of some of these novels phages across human gut communities.
To date, only three reported metagenomic studies of the gut DNA virome have characterized temporal variation 22, 23, 76. One of these studies used VLPs isolated from frozen fecal samples collected from four adult female monozygotic twin pairs and their mothers at three time points over a 12-month period 22. VLP-derived virome datasets were compared to datasets of sequenced bacterial 16S rRNA genes and shotgun reads from total fecal community DNA generated from the same fecal samples used to prepare VLPs. The results disclosed that the viromes of co-twins and their mothers exhibited a significantly greater degree of interpersonal variation than did the corresponding bacterial communities. Despite the marked interpersonal variation in gut viromes and their encoded gene functions, intrapersonal diversity was extremely low, with >95% of virotypes retained over the period surveyed, and with DNA viromes dominated by a few temperate phages that exhibited remarkable genetic stability (>99% sequence conservation). These observations suggested that a temperate viral lifestyle is more prevalent in the distal intestine than a kill-the-winner dynamic (see Fig. 2).
Another study of temporal variation involved adults subjected to a defined diet for a period of eight days 23. During this time, both fecal bacterial and viral communities changed in a comparable manner. Importantly, interpersonal variation at the late time points was reduced among individuals consuming the same diet, suggesting that diet has an important effect in shaping both bacterial and viral communities.
A third study examining temporal variation characterized the DNA virome of a one week old healthy infant and used DNA microarrays to compare relative viral abundances in the fecal microbiota between postnatal weeks one and two76. The results showed that at early stages of life, the viral population changes drastically: over half of the virotypes present at one week were undetectable at two weeks. While contrasting with the stability documented in the fecal DNA viromes of healthy adults, these results are consistent with the dynamic and rapid nature of assembly of the infant bacterial microbiota 66, 77.
There are a number of examples of known phage-encoded host fitness factors in gut bacteria (e.g., lambda bor and lom, 933W Stx2, 78, 79), but most of these appear to be virulence determinants of one kind or another. When comparing purified VLP viromes and fecal microbiomes in the study of monozygotic twins, phage exhibit enrichment for genes involved in anaerobic nucleotide synthesis, as well as cell wall biosynthesis and degradation 22. Other distinctive features of phage genomes include genes that can alter bacterial receptors and prevent superinfection 80. Interestingly, many phage receptors may be involved in carbohydrate transport and utilization. In an environment such as the gut, where carbohydrate utilization is an important fitness factor, mobilization of these genes by phages could endow their bacterial host with benefits (Fig. 2). There are likely a great number of bona-fide (metabolic) fitness factors encoded in phages that have yet to be characterized.
Intriguingly, new evidence suggests that carbohydrate-binding components of the human gut virome may change at an extremely high rate. A recent metagenomic study examining VLPs purified from fecal samples collected from 12 humans identified 51 hypervariable loci, areas with mutation rates that are much higher than the rest of their viral genomes 50. Protein structural predictions revealed that some do not have homology to known folds, some have similarity to Ig-superfamily proteins, and others have similarity to C-type lectin folds, which play a key role in carbohydrate binding. Moreover, these loci appear to be specifically targeted for mutation by a reverse transcriptase-based mechanism, perhaps suggesting a critical functional advantage provided by these hypervariable loci. It is tempting to speculate that these loci confer a selective advantage to phages, enabling immune evasion through IgA binding, or improving the chances of infecting a host cell in the rapidly changing conditions of the gut through adaptable binding to relevant environmental materials or bacterial surface receptors. There is a well-documented precedent for hypervariable loci conferring a fitness advantage in Bordetella phage by allowing tropism switching in the phage receptor-binding protein 81. Is important to emphasize that the hypothesis that these loci may allow a phage to bind IgA or environmental ligands is speculative and needs experimental validation; non-receptor phage structural proteins have been found to contain Ig-like domains, which may aid in host binding by weakly interacting with the cell surface 82.
The RNA virome of two healthy adults has been characterized by purifying VLPs 83. This study showed that most RNA viruses appeared to be consumed together with food. A pepper-associated virus (PMMV), comprised over 80% of the identifiable gut viruses. The only animal RNA virus observed was a picobirnavirus that had previously been found in the feces of healthy individuals as well as in patients with diarrhea: it has not been associated with any particular disease.
Comparative studies of different mammalian species represent a source of information about the effects of environmental factors, including diet 84, 85, and various host factors on phage diversity in the gastrointestinal tract. Extensive surveys have been conducted of viruses associated with different mammalian species in search of potential sources of zoonosis, the etiology of animal diseases, and to identify common mammalian viruses 13. These studies, which includes mammals occupying very distinct habitats 29, 31, 32, 86–91, identified viruses from the same families identified in the human gut virome, further underscoring the prevalence and long-standing nature of the evolved viral-mammalian host relationship (Box 3). In all these studies, ssDNA viruses were ubiquitous, accompanied in some cases by positive-sense ssRNA enteric viruses 43, 90, 91.
The eukaryotic virome can be considered from at least three different perspectives: viruses associated with the eukaryotic component of a gut microbiota, viruses associated with various human cell populations exposed to this microbiota, and viruses associated with ingested food. Metagenomic studies of healthy individuals are dominated by bacterial viruses with eukaryotic viruses either absent 23 or present at very low abundance 22, 28, 50, 72, 76. RNA viruses are an apparent exception: while abundant, they appear to be largely derived from ingested food 83. Our limited view of the eukaryotic gut virome is largely derived from studies that apply viral metagenomics to identify agents associated with gastrointestinal diseases 136–142. These studies identified known enteric viruses (Adenovirus, Rotavirus, Enterovirus and Norovirus), novel members of Bocavirus, Picobirnavirus, Cosavirus, and Anellovirus that are potential human pathogens, as well as novel viruses that may be related to diet (Gyrovirus, Nodavirus and members of the Dicistroviridae, Vigaviridae and Partitiviridae families). The high prevalence of eukaryotic ssDNA viruses in metagenomic datasets has led to new perspectives about the potential importance and the diversity of these small viruses 143. Although identification of these viruses has come from symptomatic individuals, they have also been identified, at similar prevalence, in asymptomatic contacts of those with disease. The broad representation of these eukaryotic viruses in the human gut as well as other body habitats has prompted a call to consider the functional significance of the eukaryotic human virome 12.The almost ubiquitous presence of human viruses that are not phages and that have not been associated with any disease suggests that viruses, especially those acquired in early childhood, may be essential for proper immune system development, and that particular host genotypes or immunologic constraints may cause normally benign viruses to induce disease states.
An early survey of coliphages in cows, pigs, and humans 92 showed that they were present in titers of up to 107 VLPs per gram of feces and that temperate phages were the most common. Interestingly, humans and pigs (omnivores with simple guts) had higher counts of temperate coliphages than cows (herbivores with foregut fermentation chambers). In an independent study, estimates of phage diversity from bovine rumen fluid 93 suggested that up to 28,000 different virotypes could be present in titers as high as 109 VLPs per ml of sample, hinting at strikingly higher diversity and abundance compared to humans. In contrast to the large interpersonal variation observed in human gut viromes 22, 23, this latter study showed a high degree of similarity between the phage communities of cohabitating animals on a similar diet. Metagenomic studies in horses (herbivores containing a hindgut fermentative chamber) revealed an intermediate level of phage diversity between that documented in herbivorous foregut-fermenting ruminants and omnivorous humans with simple guts.
Together, these studies suggest that diet, gut physiology, and potentially the transit time of food, play important roles in determining the lifecycle (and diversity) of phages in the mammalian gut. Further dissection of these relationships requires manipulable, representative, and defined in vivo models. Moving in this direction, Maura and colleagues used mice to study the effects of a lytic enteric phage, observing stable long-term replication over three weeks 94. As noted below, gnotobiotic mouse models may also be very informative.
Much has already been written about the history, successes, and failings of phage therapy. Most of the studies have focused on the use of lytic phages to destroy pathogenic bacteria 95,96 (Fig. 3). Clinically oriented phage research began very soon after the discovery of phages, with Felix d’Herelle using phages to treat bacillary dysentery in a number of human patients 8. This optimistic start, however, led to a number of misconceptions and missteps, both scientific and political, regarding the use of phages. d’Herelle incorrectly assumed that there was only one universally efficacious strain of lytic phage 97, though we now know phages exhibit exquisite host cell specificity. In the 1930s, pharmaceutical companies began distributing enormous amounts of lytic phages as generic antibacterial therapies, but in part because of the perceived universality of phages, they had very little knowledge of their product’s components.
In retrospect, many of the commonly used phage preparations were destroyed by the organomercury preservatives added to the vials that contained them, or were contaminated with bacterial exotoxins secreted by the cultures used to generate them 98. Inevitably these problems, along with manufacturing inconsistencies (the supposedly standardized strains of phages would change from batch-to-batch) led to distrust among the medical and scientific community.
The recent resurgence of phages as possible therapeutics has been driven by a number of factors. The alarming prevalence of antibiotic-resistant strains of pathogenic bacteria, combined with the inexorable spread of antibiotic-degrading enzymes, such as the New Delhi metallo-beta-lactamase (NDM-1), have led to calls for new therapeutic strategies 99. From a practical standpoint, antibiotic discovery efforts have produced few novel compounds over the past decade 100. Phages are a promising tool as they are easy to manufacture, have good host specificity, and can be readily genetically manipulated. Moreover, resistance to phages may develop more slowly than to antibiotics, though the reasons for this are multifaceted 101. Phage resistance can occur spontaneously in cultures (as frequently as 1 in 105 cells), but there can be fitness costs associated with resistance. In contrast, many forms of antibiotic resistance cannot occur spontaneously, but instead require introduction of a foreign DNA element. In many ways, addressing bacterial resistance is much easier with phages than with antibiotics because one can isolate different phages, or phages may spontaneously mutate to overcome host resistance.
Perhaps a more interesting question, in the context of community dynamics and our growing understanding of the virome and microbiome, is whether we can produce more subtle phenotypic shifts in an ecological niche. Rather than destroy a single pathogenic member of a community, lysogenic phages could be introduced to promote a community structure that is beneficial to both the human host and microbial community members (Fig. 3). For example, one could expand the capacity of the gut microbiome to degrade dietary components102. Similarly, phage could be used to introduce novel, beneficial traits to community members, such as those involving nutrient biosynthesis. In the latter circumstance, it may be difficult to introduce traits that are not purely beneficial to a lysogenic phage’s bacterial host, as the energetic effects of synthesizing an unnecessary protein impose a selection pressure.
Given the potential power and replicating nature of phages, a number of questions must be addressed before they can be more widely adopted including issues related to bio-containment 101. Although phages are frequently sold as viruses that ‘can only infect bacteria’, their safety has yet to be completely defined. The intravenous administration of phage (e.g., in the case of bacterial sepsis) is particularly complex given the immunogenicity of some preparations and rapid clearance of phage particles by the reticuloendothelial system of the spleen 103. It is tempting to assume that other routes of administration, such as oral cocktails of phage to target the human gut microbiome, would not have such effects. However, phage DNA is detectable by PCR and FISH in serum shortly after oral consumption 104. Other studies have provided evidence of trans-placental passage of phage 105. There is data suggesting that enzymes transcribed from phage DNA can be expressed in mammalian cells 106, this finding has even led to attempts to use phage as gene therapy vectors 107, 108.
Despite these concerns, we are exposed to millions of phages every day, including the ones from our own microbiota, without significant observable harm. In this spirit, it is interesting to consider the potential ‘therapeutic’ use of phages in the context of current efforts to apply microbiome-directed therapies 109. Questions that can be asked include whether it is beneficial or detrimental for bacterial taxa being considered as candidate probiotics to possess or lack prophages or whether phages should be deliberately administered coincident with or preceding introduction of a probiotic consortium to help create niches that promote successful invasion and engraftment of the consortium.
There has been little experimental work done on the ecology of phages in vivo. Germ-free mice and mice mono-colonized with different strains of E. coli, including strains isolated from children with diarrhea, have been used to examine replication of T4 and T7 phages 110, 111. Gnotobiotic mouse models of the human gut microbiota may not only provide better understanding of phage-host dynamics, but may also represent a potentially valuable tool for establishing a preclinical pipeline designed to evaluate the feasibility of phage therapy. Recent work has shown that transplanting intact uncultured human gut (fecal) microbial communities into gnotobiotic mice is efficient, capturing the majority of microbial diversity and microbiome-encoded functions present in the human donor’s community in the recipient animals 112, 113. Mice with replicated human gut microbiomes can be fed diets resembling those of the human donor to explore diet-microbiome-phage interactions. An additional refinement to this approach is to transplant sequenced collections of bacteria (some containing prophage) cultured from a given donor’s fecal sample into recipient mice 113. The effects of various perturbations of gnotobiotic mice harboring these microbiota on phage-bacterial dynamics can be studied over time under highly controlled conditions.
Yet another envisioned approach is to assemble defined communities of sequenced members of the human gut microbiota in formerly germ-free mice and then to deliberately stage phage attacks. VLPs prepared from human fecal samples and introduced into mice containing these sequenced collections of cultured microbes from human donors would allow investigators to directly determine the bacterial host specificity of the VLP-associated phage, as well as the effects of (i) the presence or absence of prophage in community members, (ii) the diet administered to the animals, and (iii) the phage’s contribution to mammalian host physiology, including immune function. The impact of a staged phage attack on community structure and functions can also be defined in gnotobiotic animal models over time and as a function of location along the length of the gut, where available nutrient and energy resources vary considerably.
These model systems, coupled with observational data in human fecal samples, should help us understand how phages influence metabolism in the gut and possibly how to manipulate it. Obtaining answers will almost certainly demand new and more efficient methods for deliberately curing bacterial hosts of their prophage. It may also require the application of whole genome transposon mutagenesis methods that are married to next generation sequencing platforms 114 to identify the functional contributions of genes in a given prophage. While the journey ahead will certainly be demanding, approaches are in hand or can be envisioned that will help propel the ‘new age of phage’ forward so that long standing questions can be addressed and new insights can be obtained.
Work from the authors’ labs described in this report was supported by grants from the NIH (DK78669, DK30292, DK70977 to JIG and GM095384 to FLR), and the Crohn’s and Colitis Foundation of America. A.R. is the recipient of an International Fulbright Science and Technology Award. N.P.S is a member of Washington University’s Medical Scientist Training Program (MSTP) funded by NIH grant GM007200. Due to space limits we were not able to cite many wonderful studies relevant to the topics covered.