Advancing RNA phage biology through meta-omics

Abstract

Bacteriophages with RNA genomes are among the simplest biological entities on Earth. Since their discovery in the 1960s, they have been used as important models to understand the principal processes of life, including translation and the genetic code. While RNA phages were generally thought of as rare oddities in nature, meta-omics methods are rapidly changing this simplistic view by studying diverse biomes with unprecedented resolution. Metatranscriptomics dramatically expanded the number of known RNA phages from tens to tens of thousands, revealed their widespread abundance, and discovered several new families of potential RNA phages with largely unknown hosts, biology, and environmental impact. At the same time, (meta)genomic analyses of bacterial hosts are discovering an arsenal of defense systems bacteria employ to protect themselves from predation, whose functions in immunity against RNA phages we are only beginning to understand. Here, I review how meta-omics approaches are advancing the field of RNA phage biology with a focus on the discovery of new RNA phages and how bacteria might fight them.

Graphical Abstract

Graphical Abstract.

Introduction

In all ecosystems, bacteria are constantly facing the deleterious actions of viruses preying on them. These bacterial viruses, called bacteriophages, represent an enormous genetic and functional diversity. With this diversity, phages drive the evolution and composition of their ecosystems, broadly influencing, e.g. nutrient cycles [1, 2] and human health [3–8]. More than 100 years of phage research revealed that phages encompass a diverse range of morphologies, life cycles, genome sizes, and genome compositions [9]. Yet, most studies on phage biology, their application, and their impact on their environments focus on tailed double-stranded DNA (dsDNA) phages belonging to the class Caudoviricetes. This strongly limits our understanding of the biology and importance of nontailed phages that come in a range of flavors, including such with RNA genomes.

A major bottleneck limiting our understanding of RNA phage biology is the comparably small number of RNA phages that have been isolated and can be experimentally studied in concert with their hosts. Only ∼200 RNA phages belonging to the class Leviviricetes [10] and ∼14 RNA phages belonging to the class Vidaverviricetes [11] have been isolated to date. Still, an even smaller number of RNA phages were studied in detail: most research has been carried out using a handful of model phages such as MS2, Qβ, and φ6. Moreover, and perhaps even more frustratingly, the overwhelming majority of these RNA phages were isolated in the mid-to-late 20th century [10], their genomes were never sequenced, and they seem to have been lost to science [12]. These issues notwithstanding, RNA phages historically have been important models for the development of both the field of molecular biology and biotechnological tools. For instance, the gene encoding the coat protein of phage MS2 was the first gene to be sequenced [13], and its genome was the first to be fully sequenced [14]. An equally remarkable success was the discovery of an RNA aptamer structure in the MS2 genome, which is bound by a dimer of the phage coat protein with high specificity and affinity to inhibit translation of the downstream replicase gene [15]. This enabled the development of molecular tools for, e.g. the imaging of RNA in live cells [16] and the purification of RNA–protein complexes from cell lysates [17]. The coat proteins of RNA phages are also used to produce virus-like particles, which are nanotools with tremendous potential for application in vaccines, drug delivery, and more [18].

With the recent advent of powerful metagenomics methods, phage research has started to undergo a renaissance, and it is now possible to broadly investigate the presence of phages in virtually all biomes, ranging from soil to the ocean to the human gut. Metagenomics led to the identification of a wealth of as-of-yet uncultured phage species, including dsDNA phages with enormous genomes reaching up to 735 kb [19]—more than six times the size of the smallest known bacterial genome [20]. Analysis of these genomes further annotated >100 million phage proteins deposited in the Integrated Microbial Genomes and microbiomes/Viral Resources (IMG/VR) v4 database [21], most of which are of unknown function. Additionally, functional genomics of bacterial hosts has become a powerful tool to understand bacterial-phage warfare and uncovered >250 defense systems bacteria use to protect themselves against phage infection [22].

The shotgun sequencing of DNA employed by metagenomics, however, is blind to RNA phages, which has not only hindered the discovery of new species and families, but also limited our appreciation of their ecological importance. Only recently, metatranscriptomics studies have started to overcome these limitations to gain insights into the distribution, diversity, and potential functions of RNA phages. In this review, I briefly discuss the biological features of RNA phages and their traditional characterization. I then focus on how metatranscriptome mining is transforming our understanding of RNA phage biology and how bacterial genomics can be exploited to study RNA phage–host interactions.

General biology of RNA phages

Leviviruses

Virulent RNA phages belonging to the class of Leviviricetes are among the simplest known viruses. They only encode four proteins in their highly compact ∼4000-nt positive-sense single-stranded RNA (+ssRNA) genomes [23, 24]. The genome is packaged into a small (∼27 nm) nontailed, nonenveloped icosahedral capsid, which is made up from the coat protein and a single copy of the maturation protein, the latter of which is involved in adsorption and host entry (Fig. 1A) [23]. The other two encoded proteins are the replicase, which is an RNA-dependent RNA polymerase (RdRp) that is involved in transcription and replication of the genome, and the lysis protein, which is involved in host lysis (Fig. 1B and C) [23]. All characterized leviviruses adhere to bacterial pili, such as the type IV secretion system pilus of the Escherichia coli F plasmid or the type IV pilus of Pseudomonas aeruginosa, via the maturation protein to start the infection process (Fig. 1D) [10]. After adsorption to the pilus, the virion is brought into proximity of the bacterial outer membrane by pilus retraction. Then, the maturation protein is ejected together with the genome and both enter the host via a poorly understood mechanism [23]. Ejection might be driven by tension in the virion caused by the incorporation of the maturation protein during assembly [25]. The +ssRNA genome is then translated by host ribosomes in a highly regulated fashion [23]. Following virion assembly, the progeny particles are released into the environment by lysis of the host. To date, of the six families of leviviruses recognized by the International Committee on Taxonomy of Viruses (ICTV) [26], only members of the Fiersviridae have been isolated. Evidence for the existence of the other five families exclusively comes from metatranscriptome data (Table 1). Furthermore, most isolated leviviruses infect enterobacteria [10], with no known example infecting a Gram-positive bacterium.

Figure 1.

Characteristics of the Fiersviridae. (A) Virion structure of the Fiersviridae. Left: Cross-section of the virion. Right: Icosahedral capsid structure. (B) Genome organization of phage MS2 (3569 nt, NCBI: NC_001417.2). mat: maturation protein; cp: coat protein; lys: lysis protein; rep: replicase (RdRp). (C) Genome organization of phage Qβ (4215 nt, NCBI: NC_001890.1). mat/lys: maturation protein with additional lysis protein function; cp: coat protein; A1: minor capsid protein that is a translational read-through product of cp; rep: replicase (RdRp). (D) Life cycle of the Fiersviridae. Following adsorption to the pilus via the maturation protein, the pilus retracts. Via an unknown mechanism, the ssRNA genome enters the cell and replication initiates. The new virions are assembled and released via cell lysis. Virion images modified from ViralZone [181] under a CC BY 4.0 license.

Table 1.

Overview of recognized and putative RNA phage families

Phylum	Class	Order	Family	Isolated?	Reference
Lenarviricota	Leviviricetes	Norzivirales	Atkinsviridaea	No	[26]
Lenarviricota	Leviviricetes	Norzivirales	Duinviridaea	No	[26]
Lenarviricota	Leviviricetes	Norzivirales	Fiersviridaea	Yes	[30]
Lenarviricota	Leviviricetes	Norzivirales	Solspiviridaea	No	[26]
Lenarviricota	Leviviricetes	Timlovirales	Blumeviridaea	No	[26]
Lenarviricota	Leviviricetes	Timlovirales	Steitzviridaea	No	[26]
Pisuviricota	Duploviricetes	Durnavirales	Cystoviridaea,b	Yes	[34]
Pisuviricota	Duploviricetes	Durnavirales	“f.0112” (Cysto-like)	No	[55]
Pisuviricota	Duploviricetes	Durnavirales	“f.0114” (Cysto-like)	No	[55]
Pisuviricota	Duploviricetes	Durnavirales	“f.0115” (Cysto-like)	No	[55]
Pisuviricota	Duploviricetes	Durnavirales	? (Cysto-like)	No	[56]
Pisuviricota	Duploviricetes	Durnavirales	Partitiviridaec	No	[55, 56, 77]
Pisuviricota	Duploviricetes	Durnavirales	Picobirnaviridaec	No	[55, 56, 69]
Pisuviricota	Duploviricetes	Durnavirales	“Paraxenoviridae”	No	[80]
“Artimaviricota”	?	?	?	No	[77]
Lenarviricota	Howeltoviricetes	?	?	No	[56]
“p.0002”	?	?	“f.0278”	No	[55]
“RvANI90_0011770”	?	?	?	No	[55]

^aRecognized by the ICTV [179, 180].

^b Cystoviridae were recently proposed to be re-classified [55, 56]. The ICTV assigns Cystoviridae to the phylum Duplornaviricota, class Vidaverviricetes, order Mindivirales [179, 180].

^cRecognized by the ICTV as eukaryote-infecting viruses [179, 180].

Cystoviruses

In addition to RNA phages belonging to the Fiersviridae family, a handful of virulent RNA phages belonging to a second family, Cystoviridae (class Vidaverviricetes), have been isolated and characterized [11]. The virions of cystoviruses are nontailed and showcase a unique structure involving two icosahedral capsid layers and a lipid envelope (Fig. 2A) [27]. In contrast to leviviruses, cystoviruses have a ∼13-kb double-stranded RNA (dsRNA) genome that is segmented into three molecules of different sizes and which encode for ∼13 proteins (Fig. 2B). The three dsRNA molecules are packaged together with the RdRp into an inner capsid that is surrounded by an enveloped outer capsid, resulting in a ∼85-nm virion [24]. Spike proteins that facilitate the recognition of the host receptor are embedded into the lipid envelope. Following adsorption to host pili or lipopolysaccharides [28], the envelope of the cystoviral virion fuses with the host outer membrane to enter the periplasm (Fig. 2C). The virion then enters the cytoplasm, where the outer capsid is shed and transcription of the genome engages from within the inner capsid, which stays intact and thereby protects the dsRNA genome from host access [27]. The phage messenger RNAs are translocated into the cytosol, where the phage proteins are translated. During assembly of new virions, so-called pac sites located at the 5′ ends of the phage transcripts are used as packaging signals to load them into the inner capsid [27]. Second strand synthesis then occurs within the inner capsid to generate the mature dsRNA genome and is followed by the assembly of the outer capsid. In the final maturation step, the new virions acquire host membrane via a poorly understood mechanism and leave the host via lysis [27]. The majority of isolated cystoviruses infect pseudomonads, especially the important plant pathogen Pseudomonas syringae [11], which is why cystoviruses have been investigated as potential crop protectants [29]. As for the leviviruses, no cystovirus infecting a Gram-positive bacterium has been isolated to date.

Figure 2.

Characteristics of the Cystoviridae. (A) Virion structure of the Cystoviridae. Left: Cross-section of the virion. Right: Icosahedral structures of both capsid layers. (B) Genome organization of φ6. The S segment (2948 bp, NCBI: NC_003714.1) encodes for outer capsid protein P8, morphogenic protein P12, major membrane protein P9, and muralytic enzyme P5. The M segment (4063 bp, NCBI: NC_003716.1) encodes for membrane protein P10, spike proteins P6/P3, and membrane protein P13. The L segment (6374 bp, NCBI: NC_003715.1) encodes for nonstructural protein P14, assembly cofactor protein P7, RdRp P2, packaging NTPase P4, and inner capsid protein P1. Annotation of P14 according to [182]. (C) Life cycle of the Cystoviridae. Following adsorption to the pilus via the spike protein, the pilus retracts. This brings the virion in proximity to the cell and its membrane fuses with the host outer membrane. The virion then enters the cytosol and transcribes its genome from within the inner capsid, thereby initiating replication. Following assembly, the ssRNA is packaged and second strand synthesis engages inside the virion. Finally, the assembled virion acquires its lipid envelope from the host outer membrane via a poorly understood mechanism and leaves the cell via lysis. Virion images modified from ViralZone [181] under a CC BY 4.0 license.

Traditional isolation of RNA phages

RNA phages are generally thought of as being rare in the environment. This is likely because only a small number of RNA phages has been cultured in the laboratory, especially when compared with the number of isolated dsDNA phages. Historically, leviviruses and cystoviruses were discovered serendipitously. The first leviviruses, called f2 through f7, and with it the first RNA phages, were discovered in a screen intended to identify phages able to specifically infect E. coli strains harboring the F plasmid [30, 31]. f2 and MS2 (which was isolated around the same time as f2 [32]), quickly became important models for the study of levivirus biology and molecular biology in general. While ∼200 leviviruses were isolated in the decades following the discovery of f2 [10], the true diversity of these isolates is unknown since only a few of them were sequenced. For instance, a single study isolated ∼100 (i.e. half of all) leviviruses infecting E. coli from sewage samples in Korea and classified them into three groups based on serological analysis [33]. It remains unclear, however, if the same phages were isolated over and over again or whether these indeed represent a larger diversity. Similar to f2, the first isolated cystovirus, φ6, was discovered in a screen for phages infecting P. syringae [34]. Other cystoviruses were discovered later on [11], but φ6 remains the most important model for cystoviral research [35].

A major reason for the limited diversity of isolated RNA phages is the lack of isolation protocols able to specifically enrich for RNA phages, contrasting with the situation for dsDNA phages, which can readily be isolated from virtually all environments [36]. In addition, several biases in the common phage isolation protocols prevent the discovery of new RNA phages [37, 38]. To provide a few examples, most protocols add chloroform during the isolation of phages in order to sterilize the sample [36]. While leviviruses resist chloroform [31], the enveloped cystoviruses rapidly become inactivated in its presence [34] and thus cannot be isolated using standard protocols. Similarly, the choice of the indicator bacterial strain in plaque assays used for phage isolation often excludes RNA phages: since all known leviviruses and the majority of cystoviruses require pili for adsorption, standard models such as the pilus-devoid E. coli MG1655 are poor choices for the isolation of RNA phages. Thus, new isolation methods must be developed to sample the full scope of RNA phages.

One option to enrich for RNA phages is to take advantage of their pilus-specificity, similar to the study that discovered the first RNA phages [31]. This procedure can be further refined by addition of RNase, which inhibits the propagation of leviviruses [39], to the plate of pilus-expressing bacteria in order to only select pilus-dependent, RNase-sensitive phages [33, 40]. A recently developed method, Phage Discovery by Co-culture (Phage DisCo), employs mixed indicator lawns of fluorescently labeled bacteria to rapidly identify phages from environmental samples that are dependent on a pilus of interest (Fig. 3A) [12]. To that end, a GFP-labeled Pseudomonas putida strain is co-cultured with an RFP-labeled Salmonella enterica strain, both of which harbor the same plasmid-encoded pilus. The mixed lawn is then infected with environmental samples and screened for plaques that do not show a fluorescent signal, indicating that both bacterial species have been lysed by the same phage. This approach was very successful and discovered numerous new pilus-dependent phages, including eight new leviviruses [12]. Going one step further, the same group used Phage DisCo to screen for phages that use specific receptors by employing up to three fluorescently tagged knockout strains of known phage receptors [41]. This modified protocol should also be able to isolate RNA phages specific for genome-encoded pili such as the type IV pilus of P. syringae. Yet, Phage DisCo is limited to bacteria with genetic tractability and necessitates downstream analyses of the isolated phages to determine their genomic material. Moreover, only known receptors can be included or knocked out in the assay, which excludes RNA phages with unknown receptors from being isolated. A possible way to overcome this limitation is to enrich RNA phages without the use of genetic tools—e.g. by inhibiting DNA synthesis using small molecules, which would specifically inhibit the replication of DNA phages and thereby favor RNA phage propagation [38].

Figure 3.

RNA phage isolation methods. (A) Phage DisCo co-cultures fluorescently labels bacterial species expressing the same plasmid-encoded pilus. Phages dependent on the pilus will infect and lyse both indicator strains, resulting in a clear plaque. To identify RNA phages, pilus-dependent phages are isolated, and their genomic material is analyzed. (B) Chronic cystovirus infection can be detected by isolating dsRNA from single bacterial colonies. When dsRNA is present, the infected colony can be used to isolate and characterize the phage. Cystovirus image modified from ViralZone [181] under a CC BY 4.0 license.

Plaque assay-based isolation protocols are robust and practical and remain the gold standard for phage isolation. Yet, they are unable to detect phages that do not form clear lysis zones in a bacterial lawn. Therefore, other methods should be considered to isolate RNA phages from the environment. Since the presence of dsRNA in an infected cell is a hallmark of RNA virus infection—either as a replication intermediate of ssRNA viruses or as the genomic material of dsRNA viruses [42]—dsRNA from single bacterial colonies can be isolated using CF11 cellulose columns (Fig. 3B) [43]. Using this approach, phage RNA can be detected in infected bacteria even if no plaque formation can be observed. This method led to the discovery of cystovirus φNY, which causes chronic infection of its host Microvirgula aerodenitrificans and does not form plaques on agar plates [43]. Following its discovery, the structure and genome of φNY could be characterized, showcasing how alternative isolation methods can aid the discovery of new RNA phages [43].

Characterization of RNA phages in the metagenomic era

Identification of new leviviruses and cystoviruses using metatranscriptomics

With sequencing becoming ever cheaper and more powerful it has become possible to study the genomic content of virtually any environment in great detail. In contrast to isolating single organisms and studying them in culture, metagenomics is able to accurately chart the diversity of a biome of interest [44]. For example, we now know that the majority of microbial species in the environment have never been cultured [45]. Functional analysis of metagenomic data then allows the prediction of the roles that uncultured species might play in their environment [46].

The untargeted shotgun sequencing applied in metagenomics studies not only provides information about the microbial communities of the studied samples. It also contains sequences from viruses, including phages. Only through these datasets we have begun to understand the true diversity and abundance of phages in the biosphere, making metagenomics the main source for the discovery of new phage genomes [21, 47]. However, DNA sequencing naturally cannot capture information about the presence of phages with RNA genomes. The solution to this is metatranscriptomics, which performs shotgun sequencing of the RNA content of a sample. To identify genomic sequences of RNA phages in metatranscriptomic data, the data are typically searched for signatures of RNA phages, most importantly the RdRp gene, which is found in all RNA viruses that lack a DNA stage, including all known RNA phages (Fig. 4A) [48].

Figure 4.

Identification of RNA phages using metatranscriptomics. (A) Simplified workflow of RNA phage detection using metatranscriptomics. (B) Left: Of the 530 603 RNA viruses (realm Riboviria) in the IMG/VR v4 database [21], 95 241 (∼18%) are RNA phages of the classes Leviviricetes and Vidaverviricetes. Right: Distribution among the seven named RNA phage families as well as the RNA phages without assigned family: Atkinsviridae (n = 4911), Duinviridae (n = 327), Fiersviridae (n = 18 558), Solspiviridae (n = 1929), Blumeviridae (n = 5388), Steitzviridae (n = 30 952), Cystoviridae (n = 833), and RNA phages with unknown family (n = 32 343).

Around 10 years ago, the first metatranscriptome studies achieved a major leap forward for our understanding of RNA phage diversity by studying transcriptome datasets originating from a diverse set of environments including oceans, sewage, soil, and animal-associated sources [48–51]. More than 150 new leviviral genomes were identified in these studies, whereas only a few new cystoviral genomes were discovered. Excitingly, however, the full genome of a cystovirus was discovered in a dataset of a pure culture of Streptomyces avermitilis [50]. While experimental validation of this cystovirus is still pending, this is the first indication that Gram-positive bacteria can also be infected by RNA phages.

Analysis of soil metatranscriptomes further expanded the diversity of RNA phages with thousands of leviviral genomes and a handful of cystoviruses, suggesting a potential important role of RNA phages in this environment [52]. Similarly, aquatic and activated sludge metatranscriptomes revealed several thousand additional leviviruses [53, 54], further verifying the wide spread of this family of phages. This massive expansion of known levivirus diversity led to the taxonomic restructuring of these phages, increasing the number of recognized families from one to six (Fig. 4B) [26]. These studies were followed by the analysis of thousands of highly diverse metatranscriptomes to further expand the RNA virome [21, 55–63]. With it, levivirus diversity approaches an astonishing number of 40 000 genomes listed as high quality in the IMG/VR v4 database [21]. Similarly, the diversity of the Cystoviridae was also dramatically expanded to >200 high-quality, albeit incomplete, genomes, revealing that dsRNA phages are present in a broad range of environments, such as soil, freshwater, and animal feces [55, 56, 60]. Based on RdRp phylogeny, the now available sequence information within the Cystoviridae further suggests that they belong to the phylum Pisuviricota instead of Duplornaviricota, thus placing them together with eukaryote-infecting RNA viruses in the order Durnavirales [55, 56]. Finally, four new cysto-like families were discovered, which could represent new RNA phage families that have not been isolated thus far (Table 1) [55, 56].

Based on metatranscriptomic analyses, the number of known leviviruses and cystoviruses has increased by several orders of magnitude within only a few years. While this number is still dwarfed by the number of known tailed dsDNA phages (13 499 462 Caudoviricetes genomes in the IMG/VR v4 database [21]), we now understand that RNA phages are more abundant and widespread than previously anticipated, which also raises new questions: What are their hosts? How do they impact their natural habitats? Are they involved in nutrient cycling? Are they involved in human health? Longitudinal studies investigating the microbiota in concert with the corresponding virome might be able to provide first answers to these questions by correlating how the abundances of RNA phages and bacteria change over time [64, 65].

Identification of putative new families of RNA phages

Based on genetic searches for the hallmark protein of RNA viruses, the RdRp, metatranscriptomics is not only able to expand the diversity of known RNA phage families, but also bears the potential to discover entirely new families of RNA phages. A particularly interesting candidate for a new family of RNA phages is the family of Picobirnaviridae. This group is represented by small, nonenveloped viruses with ∼4.3-kb bisegmented dsRNA genomes that are thought to primarily infect mammals [66]. Yet, their true host remains elusive, and no culture model exists to study these viruses. Identification and analysis of new picobirnaviruses using metatranscriptomics, however, has challenged the assumption that they primarily infect mammals, as there are strong indicators suggesting that they might actually infect bacteria [67, 68]. First, the majority of genes assigned to picobirnaviruses have bacterial Shine–Dalgarno (SD) motifs upstream of their coding sequences [55, 56, 69–73], which are strong indicators for bacterial translation [74]. Secondly, genes encoding for lysins, i.e. phage proteins involved in host lysis [75], were identified in the genomes of picobirnaviruses [55] and later experimentally validated to induce lysis in E. coli [76]. These findings indicate that at least a subset of picobirnaviruses infects bacteria and that they could therefore represent a new family of RNA phages, although experimental validation is currently lacking.

Similar to picobirnaviruses, bacterial SD motifs were also detected in partitivirus genomes, suggesting they might infect bacteria [55, 56, 77]. Partitiviruses are small, nonenveloped bisegmented dsRNA viruses with ∼3–4.8 kb genomes known to infect plants, fungi, and protozoa [78]. In addition to the detected SD motifs, CRISPR spacers specific for a newly discovered genus within the Partitiviridae were found in the genome of Roseiflexus sp. RS-1, which is a dominant member of microbial mats in a hot spring in Yellowstone National Park [79]. These CRISPR spacers were exclusively present the array of a type III CRISPR–Cas system with an encoded reverse transcriptase, suggesting spacer acquisition directly from RNA (discussed below). Dynamic loss and acquisition of spacers targeting the newly discovered partitivirus genus could be tracked in metagenomic data of Roseiflexus sp. RS-1 over a period of 9 years, suggesting extensive phage–host interaction [55]. Pending experimental verification, partitivirus members might thus represent a new family of RNA phages.

Interestingly, both Picobirnaviridae and Partitiviridae belong to the order Durnavirales, in which, as mentioned above, recent phylogenetic analyses also place Cystoviridae as well as new cysto-like families [55, 56]. Additionally, the full genomes of a new family of putative phages with bisegmented dsRNA genomes, provisionally termed “Paraxenoviridae”, was identified in metatranscriptomic ocean samples and classified within the order Durnavirales [80]. These putative phages are characterized by SD motifs upstream of their genes as well as unique capsid protein structures. Overall, this suggests that Durnavirales comprise a growing number of segmented dsRNA virus families that infect bacteria.

Metatranscriptomic analysis of dsRNA isolated from Japanese hot springs identified a putative new group of bisegmented dsRNA phages that feature a unique type of RdRp related to reverse transcriptases [77]. These hot spring riboviruses (HsRV) form a distinct phylum, which was provisionally named “Artimaviricota”. Barely any eukaryotic signatures were detected in the HsRV-containing samples, which, together with the presence of conserved SD motifs in HsRV-associated genes, strongly suggests that HsRV infect bacteria [77]. Another putative new family of RNA phages that was identified based on SD motifs is a clade basal to the class Howeltoviricetes [56], which currently includes the capsidless mitochondria-infecting Mitoviridae as the only family. Matching spacers from reverse transcriptase-containing type III CRISPR–Cas systems further link “base-Howeltoviricetes” to Aliarcobacter cryaerophilus as a potential host [56]. Finally, two putative new phyla of RNA phages (“p.0002” and “RvANI90_0 011 770”) were identified with genomes harboring SD motifs and/or lysis proteins [55].

Overall, these studies highlight the power of metatranscriptome mining for the expansion of RNA phage diversity by adding new members to existing families as well as detecting putative new families (Table 1). Importantly, current analyses of RNA virus diversity do not seem to have reached saturation [55], suggesting that we are still missing much of the RNA phage diversity.

“Unencapsidated RNA phages”: viroid-like replicators infecting bacteria

Viroids are unencapsidated covalently closed circular ssRNAs (cccRNAs) known to infect plants [81–84]. With only a few hundred nucleotides in length, they are the smallest known infectious agents but still able to cause disease. While viroids are not considered viruses, they are conceptually similar in that they are obligate intracellular parasites with no own metabolism. Viroids do not encode proteins and rely on sophisticated RNA structural elements to interact with host factors, recruit host RNases or employ ribozymes to process their genomes, and co-opt host RNA polymerases to replicate [85]. Additionally, several classes of viroid-like cccRNAs exist, such as satellite RNAs, retrozymes or ribozyviruses [81, 83]. Similar to viroids, these viroid-like cccRNAs feature extensive RNA structures and typically encompass ribozyme activities necessary for replication.

In recent years, several metatranscriptomic studies searched for new viroid-like cccRNAs [57, 86–89], increasing their total number to >11 000 [87]. Importantly, and contrary to prior belief, these studies suggested that viroid-like cccRNAs might infect a diverse range of hosts [86, 87]. This leads to a simple question: Are bacteria also being parasitized by unencapsidated infectious RNAs? Similar to partitiviruses discussed above, type III CRISPR spacers targeting viroid-like cccRNAs were detected in the genome of Roseiflexus sp. RS-1 and other hot spring-resident bacteria as well as in metagenomes from other sources [87].

The presence of such CRISPR spacers suggests that bacteria are hosts of viroid-like cccRNAs. Indeed, a member of a newly discovered class of viroid-like cccRNAs called obelisks was shown to reside in the oral commensal Streptococcus sanguinis SK36 [86]. This obelisk is a highly structured, ∼1100-nt cccRNA encoding for one “oblin” protein of unknown function. Spontaneous loss of this obelisk in a culture of S. sanguinis SK36 did not lead to any observable phenotypes under the tested conditions, thus leaving its functions elusive [86]. While obelisks and other viroid-like cccRNAs are not RNA phages per se, their ability to infect and replicate in bacteria suggests they could have a broad impact on microbial communities such as in the human gut. To understand these potential biological functions and to understand how viroid-like cccRNAs spread and replicate in bacteria, detailed mechanistic studies in defined model systems will be required.

Functional genomics of RNA phages

The massive expansion of available RNA phage genome data not only enables the analysis of RNA phage distribution and diversity, but also the functional characterization of their genes. For example, leviviruses encode single-gene lysis proteins (Sgls) to lyse the infected host and release the progeny virions [90]. In contrast to the enzymatic lysins of tailed dsDNA phages that actively lyse the infected bacterium [75], Sgls of leviviruses are small proteins that inhibit peptidoglycan synthesis to passively lyse the cell [90]. For this reason, Sgls can be considered as “protein antibiotics” and understanding their mechanism of action could lead to the identification of targets for the development of novel antibiotics [91]. To identify new Sgls, metatranscriptomic levivirus genomes were systematically screened for Sgls, predicting >150 new ones, 33 of which could be experimentally validated in E. coli [92]. It is important to note that the hosts of the phages from which the tested Sgls were derived are unknown, possibly explaining why most of the Sgls were not active in E. coli. Simple protein BLAST analysis of the 33 verified Sgls in leviviral genomes revealed only few hits, indicating that Sgls are highly diverse at the sequence level [92]. A more detailed analysis by, e.g. hidden Markov model profiling or structural predictions should therefore be considered to further expand the number of known Sgls.

An equally exciting opportunity for functional genomics of RNA phages is to mine leviviral genomes for RNA regulators. In MS2 and other leviviruses, the RdRp gene is preceded by an RNA aptamer that tightly binds to a dimer of the coat protein, leading to shutoff of RdRp translation late during infection when the RdRp is no longer needed [15, 23]. This minimal aptamer–coat protein system was repurposed as a powerful molecular tool in all domains of life to, e.g. localize RNA molecules in live yeast cells [16] or purify RNA–protein complexes from bacteria [17]. Importantly, these systems are also highly specific, thus enabling the combined application of, e.g. the MS2 and PP7 systems in the same experimental setup [93]. To further expand this molecular toolbox, new aptamer–coat protein pairs could be searched for in the now available leviviral genomes. Tools such as AlphaFold 3 [94], which can predict the interactions between protein and RNA, could be employed for this purpose. Indeed, for MS2, AlphaFold 3 accurately predicts the interaction between the aptamer and the coat protein (Fig. 5A and B and Supplementary Fig. S1A), while it performs worse for the PP7 system (Supplementary Fig. S1B–D). This indicates that structure prediction tools might be helpful to discover new aptamer–coat protein pairs. As more genome sequences become available, the functional analysis of RNA phage genes, including those of putative new RNA phages, is expected to strongly contribute to the understanding of their biology.

Figure 5.

Structural prediction of the MS2 coat protein–aptamer complex. (A) Crystal structure of the MS2 coat protein–aptamer complex. Only a single coat protein dimer and one aptamer RNA are shown. PDB: 1ZDH [183]. (B) AlphaFold 3 [94] prediction of the MS2 coat protein–aptamer complex. A hexamer of the coat protein (UniProt ID: J9QBW2) was folded with three MS2 aptamer RNAs (sequence: ACAUGAGGAUCACCCAUGU). Confidence values for the AlphaFold 3 prediction: ipTM = 0.83 and pTM = 0.85. Calculated RMSD between the crystal structure and the predicted structure = 0.538 Å using the cealign method from PyMOL version 3.0.0. Only a single coat protein dimer and one aptamer RNA are shown. Visualization was performed using UCSF ChimeraX version 1.7.1 [184].

Limitations of metatranscriptomics in the characterization of RNA phages

Metatranscriptomics is a powerful way to identify new species and potentially even new families of RNA phages and helps to characterize their diversity. Yet, metatranscriptomics has several limitations that restrict its translation into biological understanding. Most importantly, the hosts of newly discovered RNA phages cannot readily be identified from the underlying data. This is because most datasets are generated from mixed samples from diverse biomes such as soil or wastewater, where many bacterial species coexist with their phages. There are some solutions to this limitation. For example, some hosts of new RNA phages can be predicted based on the presence of specific CRISPR spacers in their genomes, although this appears to be rare in nature [54–56, 63]. Alternatively, hosts of new RNA phages can be predicted by searching for phage genomes in data of bacterial monocultures as was shown for a potential cystovirus of S. avermitilis [50]. It is also possible to “reboot” RNA phages by generating virions using a synthetic genome in an in vitro transcription–translation system or by expression in a heterologous host [38, 95, 96]. The unavailability of the natural hosts of rebooted RNA phages complicates the experimental study of their biology, however, as they obviously are unable to replicate on their own. Overall, to my knowledge, no RNA phage originating from metatranscriptomic data has been successfully rebooted and cultured in the lab to date.

Analysis of RNA phages through metatranscriptomics further is limited by several technical biases. For instance, sample preparation, ribosomal RNA depletion, and library preparation all can bias the recovery of certain sequences, leading to overrepresentation or depletion of certain phage genomes [47, 97–100]. Similarly, the computational analysis of metatranscriptomics data introduces several biases. Most studies search for new RNA phages based on the presence of an RdRp gene, the hallmark gene of RNA viruses [48]. This immediately prohibits the discovery of virus-like sequences such as viroid-like cccRNAs that do not encode for an RdRp. The same is true for hypothetical “retrophages” that, similar to eukaryotic retroviruses, would employ a reverse transcriptase to copy their RNA genome into a DNA molecule before integrating it into the bacterial host genome. Further, unless dedicated sample preparation methods, such as fragmented and primer-ligated dsRNA sequencing, are used [77, 80, 101], searches for the RdRp gene will only retrieve the segment of the phage genome on which it is located [102]. This means that in segmented RNA phages like cystoviruses, the other segments are missed by the analysis and their full genomes cannot be recovered. At the same time, this also limits the potential for rebooting of segmented RNA phages from metatranscriptome data. Segmented genomes also facilitate splitting of the RdRp gene between two different segments, as was shown for some fungal dsRNA viruses [103, 104]. It cannot be excluded that certain RNA phages also have split RdRp genes, which would not be detectable by regular searches for the RdRp gene. Likewise, metatranscriptomics is only able to detect the RdRp-encoding segments of multipartite viruses, which package each of their segments into distinct virions that need to consequently coinfect a host cell to allow replication [105]. While no multipartite phages have been described to date, their existence cannot be ruled out. It is therefore important to develop new strategies to detect all segments of RNA phages and obtain their complete genomes.

Bacterial immunity against RNA phages

Expansion of the bacterial immune system by (meta)genomics

Bacteria are not simply at the mercy of the phages infecting them. For instance, phage adsorption can be inhibited by adaptation of the host receptor [106]. Additionally, bacteria have evolved a vast arsenal of defense systems to actively protect themselves against their predators. At the same time, phages have evolved anti-defense mechanisms to overcome these defenses and allow productive replication. Yet, until a few years ago, our knowledge of this evolutionary arms race was limited to only a few examples of bacterial defense and phage anti-defense, most notably restriction-modification [107], abortive infection [108], and CRISPR–Cas systems [109]. Powerful genomics and metagenomics methods have changed this, dramatically increasing the number of both CRISPR–Cas systems [110] and non-CRISPR defense systems [22, 111, 112]. However, these newly discovered defense systems almost exclusively defend against DNA phages, with only a handful of examples that target RNA phages, which I will highlight in the following sections.

RNA-targeting CRISPR–Cas systems

CRISPR–Cas systems represent one of the most common classes of defense systems [111, 113]. As the only known adaptive bacterial defense systems, CRISPR–Cas systems “memorize” pieces of invading foreign nucleic acids, such as phages or plasmids, as spacers in CRISPR repeats [109]. Upon second encounter with the memorized sequence, an immune response is elicited that either leads to cleavage of the foreign nucleic acid or abortive infection/growth arrest of the infected cell. Metagenomic analyses have uncovered an ever-growing diversity of CRISPR–Cas systems that are currently categorized into seven types within two major classes based on the organization of their Cas effector proteins [110, 114]. Several types of CRISPR–Cas systems are RNA-targeting, suggesting they could defend their host bacteria against RNA phage infection [115]. The most prevalent RNA-targeting CRISPR–Cas systems are type III systems that, upon RNA target recognition, produce cyclic oligoadenylate second messengers to activate indiscriminate accessory nucleases to induce abortive infection or growth arrest (Fig. 6) [116–118]. Under laboratory conditions, type III systems can be programmed to defend against the model RNA phage MS2 [119, 120], and, as mentioned above, spacers against putative RNA phages have been identified in metagenomic data [54–56]. Yet, no natural example of type III-mediated defense against RNA phages has been reported.

Figure 6.

Putative defense against RNA phage infection by type III CRISPR–Cas systems. Following RNA phage infection, the RT-Cas1 (reverse transcriptase-Cas1)/Cas2 acquisition machinery memorizes the invading phage RNA in the CRISPR array. The memorized sequence is expressed as crRNA (CRISPR RNA) and loaded into the type III effector complex. Upon secondary infection, the phage RNA gets cleaved by Csm3 (type III-A) or Cmr4 (type III-B) in a crRNA-dependent manner. This also activates Cas10, which converts ATP to coA (cyclic oligoadenylates) that activate Csm6 (type III-A) or Csx1 (type III-B) to indiscriminately cleave cytosolic RNA, leading to abortive infection or growth arrest and inhibition of phage replication.

Similar to type III systems, type VII [114, 121], type VI [122–124], as well as subtypes of type II [125] and type V systems [126, 127] target foreign RNA. All of these RNA-targeting CRISPR–Cas systems were shown to protect against MS2 infection when programmed to target the MS2 genome [121, 122, 125, 128–130], with the exception of type V systems, which have not been tested yet. In contrast to type III systems, no naturally occurring spacers targeting RNA phages have been identified for type VII, II, V, and VI systems, however, suggesting that they mainly target transcripts of DNA invaders. A possible explanation for this could be that the spacer acquisition machinery (Cas1 and Cas2 proteins) of RNA-targeting CRISPR–Cas systems primarily acquires spacers from DNA, although some systems encode Cas1 proteins with fused reverse transcriptase domains that are able to acquire spacers directly from RNA [55, 128, 131–133]. Overall, while CRISPR–Cas systems have the potential to defend against infecting RNA phages, it remains elusive whether this occurs outside the laboratory.

Innate defense systems targeting RNA phages

In addition to the adaptive CRISPR–Cas systems, genomic searches revealed an astonishing number of new innate defense systems present in bacterial genomes. These searches relied on the observation that defense systems tend to cluster together in the genomes of bacteria in so-called defense islands, which can facilitate the prediction of new defense systems (Fig. 7A) [134]. With more and more sequencing data becoming available, this approach proved very successful and has led to the discovery of numerous diverse defense systems [113, 135, 136]. Additional strategies further expanded this diversity by identifying defense systems in, e.g. genomic [137] or metagenomic DNA libraries [138], integrons [139–141], and phages [142]. Bioinformatics tools, such as PADLOC [143], DefenseFinder [144], and Deepdefense [145] can search for known defense systems in any genome of interest, including metagenomic data [146]. Most recently, protein large language models are now being employed to search for new defense systems outside of defense islands [147, 148]. This expanding knowledge of the bacterial immune system further established evolutionary connections between bacterial and eukaryotic immunity, revealing that many innate immune systems of humans originate in bacteria [149–152]. Surprisingly, however, barely any of these innate bacterial defense systems were shown to protect against infection by RNA phages. The reasons for this could be manifold, but perhaps the primary reason is that most studies that predicted new defense systems simply did not investigate the activity of these systems against RNA phages.

Figure 7.

Innate immunity against RNA phages. (A) Simplified workflow of defense system identification using (meta)genomics. (B) Defense against RNA phages by bNACHT25 [154]. Following infection, the coat and lysis proteins of the phage are translated and activate bNACHT25 through a DnaJ-dependent mechanism. Activated bNACHT25 cleaves the host genome to induce abortive infection and stop phage replication.

Yet, there are a few studies that investigated the activity of defense systems against RNA phages. Bacterial NACHT module-containing (bNACHT) proteins are related to human nucleotide-binding domain and leucine-rich repeat containing gene family (NLR) proteins, which are crucial components of the inflammasome [153]. bNACHT proteins are single-protein defense systems whose activation by phage infection leads to abortive infection [153–155]. Excitingly, of the 27 bNACHT homologs tested for antiphage defense, 4 protected against infection by the model RNA phage MS2 as well as 3 related phages of the Emesvirus genus [153, 154], thus establishing bNACHT as the first innate defense system active against RNA phages (Fig. 7B). The defensive properties of bNACHT homologs were also tested against the closely related model RNA phage Qβ as well as four related phages of the Qubevirus genus. In contrast to the MS2-like phages, no defense against Qβ and related phages could be observed, suggesting the bNACHT proteins to be highly specific [153, 154]. While bNACHT proteins sense DNA phage infection by direct binding of phage-encoded proteins [155], the mechanism of RNA phage defense seems to be indirect: activation of bNACHT proteins by the coat and lysis proteins of MS2 does not rely on direct interaction and instead depends on the host chaperone DnaJ. Following activation of bNACHT, DNA degradation of the host is triggered, thus causing abortive infection and stopping the infection [154].

A second defense system active against RNA phages is type I Zorya [135], which was shown to protect P. aeruginosa against infection by the model RNA phage PP7 [156, 157]. During infection with DNA phages, the ZorAB proteins of the type I Zorya system sense membrane perturbations, which leads to recruitment of the ZorCD effector proteins that cleave the invading phage genome [158]. Since PP7 is known to remove the membrane-anchored pilus of P. aeruginosa following adsorption [159], the type I Zorya system possibly senses this perturbation and then restricts the invading RNA, potentially by means of ZorCD activity. Still, the mechanism of PP7 defense by the type I Zorya system remains to be studied. Additionally, several new systems with the potential to protect against MS2 infection have been identified in mobile integrons [141]. Finally, protection by a selection of additional defense systems against MS2 [160], Qβ [161], or both [136, 162] was tested but no defense was observed.

In addition, several systems have been identified or predicted that could facilitate defense against RNA phages based on their domain architectures. To provide a few examples, coiled-coil nuclease tandems are a branch of type IV restriction systems that are predicted to target RNA [163], suggesting they could target and cleave the genomes of infecting RNA phages. Similarly, the Mokosh system contains a predicted RNA helicase domain and was suggested to recognize phage RNA [113], which indicates Mokosh could defend against RNA phages as well. Moreover, a type I CBASS (cyclic oligonucleotide-based antiphage signaling system) from Staphylococcus schleiferi was shown to specifically sense phage RNAs in order to elicit an immune response [164]. These CBASS-activating bacteriophage RNAs are highly structured and recognition by CBASS relies on double-stranded regions within them. This suggests that CBASS could potentially also sense RNA phage infection, either by structured regions within phage transcripts or dsRNA replication intermediates. Finally, several defense systems were shown to be activated by direct interaction with components of the infecting phage, such as capsid or portal proteins [155, 165–170]—a mechanism that defense systems could also exploit to target invading RNA phages.

As mentioned above, bacteria apparently use type III CRISPR systems to target infecting viroid-like elements [87]. No innate defense systems targeting viroid-like cccRNAs have been described so far. Yet, given their unique circular structures, these “RNA plasmids” might be recognized by specific defense systems, similar to DNA-targeting systems that restrict invading DNA plasmids [107, 135].

Going forward, to better understand bacterial defense mechanisms against RNA phage infection, it will be crucial to include RNA phages in the routine panel of phages in studies screening for new candidate systems. Given that bNACHT proteins can discriminate between two closely related genera of RNA phages [153, 154], the RNA phages employed in such screens should be as diverse as possible.

Anti-defense mechanisms of RNA phages

Although bacteria appear to be armed to the teeth, phages can overcome these defenses by employing diverse anti-defense mechanisms [107, 171–173]. This raises the question how RNA phages could counteract potential defense systems that would otherwise inhibit their replication. The compact genomes of RNA phages, especially those of leviviruses, leave no room for dedicated anti-defense proteins, indicating that they would likely need to rely on other mechanisms. Perhaps the most probable such mechanism are the high mutation rates of ∼10⁻⁴ and ∼10⁻⁶ mutations per nucleotide per cell infection for leviviruses [174, 175] and cystoviruses [176], respectively, which are considerably higher than those of dsDNA phages (∼10⁻⁸–10⁻⁷ mutations per nucleotide per cell infection) [177]. Indeed, in case of bNACHT proteins, a coat protein mutant of MS2 was able to escape detection and hence overcome defense [154]. Alternatively, it cannot be excluded that RNA phage proteins moonlight as defense system inhibitors, similar to eukaryotic RNA viruses like influenza A, which expresses several proteins with such a dual function [178]. The compartmentalization of cystoviruses during infection represents another possible anti-defense mechanism that hides the dsRNA genome in the inner capsid after host entry and thereby protects it from host access (Fig. 2C) [27]. Yet, to study potential anti-defense mechanisms of RNA phages in detail, the obvious first step will need to be the identification of additional defense systems able to prevent RNA phage propagation.

Conclusion

RNA phages have been important model systems since their discovery more than six decades ago. Yet, only through recent studies leveraging the power of meta-omics we have begun to appreciate their diversity, their abundance, and how their hosts counteract the infection. We now know that the RNA viromes of certain biomes such as soil or wastewater are dominated by RNA phages, indicating their environmental importance [55, 56]. Within only a few years, we went from two RNA phage families recognized by the ICTV to seven [26, 179, 180], and multiple new families of RNA phages with unknown hosts and biology have been predicted (Table 1). On top of these insights obtained through metatranscriptomics, (meta)genomic analyses of bacterial genomes have identified a remarkable diversity of antiphage defense systems, whose roles in RNA phage defense we are just beginning to understand. This progress notwithstanding, only a small number of RNA phages has been isolated and studied to date. The major challenge for future RNA phage research will therefore be to develop methodologies to identify their natural hosts, as well as to culture and experimentally characterize them.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

No external funding.

Data availability

No new data were generated or analyzed in support of this research.