RNA-Mediated Virus Assembly: Mechanisms and Consequences for Viral Evolution and Therapy

Abstract

Viruses, entities composed of nucleic acids, proteins, and in some cases lipids lack the ability to replicate outside their target cells. Their components self-assemble at the nanoscale with exquisite precision—a key to their biological success in infection. Recent advances in structure determination and the development of biophysical tools such as single-molecule spectroscopy and noncovalent mass spectrometry allow unprecedented access to the detailed assembly mechanisms of simple virions. Coupling these techniques with mathematical modeling and bioinformatics has uncovered a previously unsuspected role for genomic RNA in regulating formation of viral capsids, revealing multiple, dispersed RNA sequence/structure motifs [packaging signals (PSs)] that bind cognate coat proteins cooperatively. The PS ensemble controls assembly efficiency and accounts for the packaging specificity seen in vivo. The precise modes of action of the PSs vary between viral families, but this common principle applies across many viral families, including major human pathogens. These insights open up the opportunity to block or repurpose PS function in assembly for both novel antiviral therapy and gene/drug/vaccine applications.

Introduction

Viral diseases are major threats to human health and agricultural productivity. Yet, viruses also play positive roles in health and ecology—for example, within our intestinal microbiome (78) and in regulating biomass turnover in the oceans (94). The presence of virally derived sequences within the human genome attests to our coevolution (31,79). Regular viral epidemics (e.g., by the Zika and Ebola viruses) and the emergence of novel threats due to zoonotic transfer (59, 124) [e.g., of a simian retrovirus adapting to human hosts as a basis of the HIV epidemic (112, 115)] make the development of novel antiviral solutions paramount. The shear diversity of the virosphere, however, presents problems for clinical intervention that are compounded by the large number of strain variants for each virus due to error-prone replication. This is particularly the case for the RNA viruses discussed here, as RNA genomes exhibit larger mutation rates than DNA viruses (33, 41) and typically occur as families of genetically related variants called quasispecies (32, 40, 74, 134) (i.e., the population of genetically related viral phenotypes) during a viral infection. As a result, alternatives for treatment and prevention are still rather limited, with prophylactic vaccination being the most successful option (16).

Viral resilience to intervention is somewhat surprising, because the vast majority of viral genomes contain very few genes in comparison to their cellular hosts. One viral family produces virions as large, if not larger, than the smallest bacterial species and also carries significant genetic information (23, 137). However, in their simplest forms, virions consist of a relatively short genetic sequence encoding only a replicase and the coat protein (CP) subunit. Multiple copies of the latter self-assemble to form a protective protein layer, the capsid, around the genome. Some satellite viruses, such as Satellite tobacco necrosis virus (STNV), even “borrow” their replicase from a helper virus, allowing them to encode only their CP subunits (87). Many human pathogens, such as the picornaviruses (e.g., polio), are not much more complicated than this, encoding only additional protein machinery to facilitate replication, translation, and assembly. More sophisticated viral families encapsulate similar nucleocapsids within membrane-bounded compartments, the latter being decorated with additional virally encoded glycoproteins involved in cell entry. Despite these variations in form and complexity, all viruses face the same challenge of assembling their (nucleo)capsids efficiently and faithfully against a backdrop of cellular competitor RNAs. All viral infections have the ultimate goal of generating increased numbers of infectious virions, and a loss of the ability to assemble severely impacts their infectivity. This critical dependence on the precise assembly of their transport vehicles highlights the enormous therapeutic potential of reagents directed at assembly.

Our lack of understanding of the detailed mechanisms leading to virus assembly has hampered the harnessing of this essential step in antiviral strategies. Recent novel insights into the assembly mechanisms of viral pathogens packaging their genetic material in the form of single-stranded (ss), positive-sense RNA genomes is in the process of changing this situation for one class of viruses (116, 118). Simple viruses have highly symmetrical virion structures based on icosahedral or helical geometry, as predicted from considerations of the coding capacity of nucleic acids (26). This idea was later refined to encompass quasiequivalent structures (20). We have recently passed the 40th anniversary of the determination by X-ray crystallography of the first atomic-resolution spherical virus structure (56)—namely, that of Tomato bushy stunt virus (TBSV) by Harrison and coworkers. This was a landmark structure, the TBSV virion being far larger than any other molecular structure determined up to that point. Recording and scanning diffraction data from small viral crystals, combining the data from many crystals, as well as the enormous—for that time—computational power required to calculate the electron density map represent both an intellectual and technical tour de force. The structure revealed many features and structural principles of biological assemblies for the first time. The globular fold of the CP subunit has a jelly-roll topology subsequently seen multiple times in other viruses, even where there is very limited sequence identity. Slight conformational differences between TBSV CPs in differing quaternary structure environments within the icosahedral asymmetric unit in the shell, in part, account for its quasiequivalent T = 3 geometry. Viral geometries were defined by Caspar & Klug (20) in terms of triangulation numbers, referring to the subdivision of the basic triangular face of an icosahedron into triangular facets. They are determined by the equation T = h² + hk + k² , where h and k are zero or positive integers. However, a totally unexpected and dominant feature regulating the assembly is the presence of extended arms of polypeptide that are located toward the N-terminus of the subunits and that take up radically different conformations dependent on their symmetry positions. Such features are seen in many virus structures and other macromolecular assemblies. Caspar (19) and Harrison (55) have coined the term tentacles for these polypeptide regions, and the intermolecular contacts they can form tentacular interactions. In isolated CP structures, these can be disordered, but it is argued that these regions are “inherently ordered,” with the ordering occurring with the correct molecular partners (55).

The TBSV structure was calculated using noncrystallographic symmetry averaging (27, 101), which allowed significant phase improvement and mimics an effect that occurs naturally in the process of crystallization, where the symmetric surfaces determine interactions with the neighbors in the crystal. This has the consequence of emphasizing the molecular components (i.e., the CPs) that follow icosahedral symmetry, while paying the price of removing density for those without it. In particular, it does not contain any information on the viral genome and the termini of the TBSV CP tentacles that are rich in basic amino acids and interact with the genomic RNA. Today, we are all familiar with images of viral structures determined following the TBSV precedent (100). However, recent advances in cryo-electron microscopy and tomography (cryo-EM/T) have opened up a new window on viral structures, allowing imaging of particles with either natural (62, 72), or artificially imposed—owing to complexation (30, 53)— asymmetry, to be determined without symmetry averaging (28, 29, 52, 69, 141). These structures reveal radically new insights that are allowing the functional roles of genomes over and above simple coding capacity to be appreciated. Here, we review the roles genomic ssRNA play in mediating efficient formation of viral particles and the consequences for viral evolution and control.

Discovery of RNA-Mediated Virus Assembly: The Model System Bacteriophage MS2

Since the male-specific, fraction 2 (MS2) bacteriophage of Escherichia coli was first isolated from sewage in New York City (76), this relatively simple virus has played multiple pioneering roles in the development of molecular biology. It is the first biological entity to have its complete genome sequenced (45) and was the first example of translational repression and overlapping genes. Its symmetry-averaged X-ray structure at 2.8-Å resolution, completed in 1990, was modeled as a T = 3 capsid in the Caspar–Klug nomenclature (129). It encompasses 90 noncovalent CP dimers that exceptionally lack extended tentacles. These adopt the two quasiequivalent conformations A/B (60 copies) and C/C (30 copies) required to form the capsid. They differ principally in the conformations of their FG-loops within their dominant β-sheet fold (Figure 1a ). As expected from previous structures, the encapsidated RNA is not visible, nor is the single copy of an additional structural protein, the maturation protein (MP). The latter is an essential part of the infection cycle, binding the phage to the bacterial F pilus (male-specific) and being the only protein from the virion to enter the infected bacterium (88).

Figure 1. Asymmetry and packaging signal (PS)-mediated assembly of the male-specific, fraction 2 (MS2) bacteriophage.

(a) RNA stem-loops (SLs) sharing their coat protein (CP) recognition motif with a translational repression operator (TR) (shown) act as allosteric effectors, triggering formation of the asymmetric A/B quasiconformer (blue/green) of the CP dimer from an unliganded, symmetric C/C dimer (magenta) (36, 99, 117). The X-ray structure showing the CP surface lattice (left), and a cross-section of the icosahedrally averaged cryo–electron microscopy (cryo-EM) structure revealing the multiple contacts to genomic RNA (right) (125). (b) Cross-section of the high-resolution asymmetric cryo-EM reconstruction of MS2 at 3.6 Å (29). The CP layer is shown in yellow, RNA in blue, and single-copy maturation protein (MP) in orange, and in a colour gradient in the inset. (c) The RNA backbone model built into the density in panel b with the MP also shown, together with SLs in contact with CPs (i.e., PSs) built as complete models, adapted from the data file associated with Reference 29. Density is color coded from TR toward the 5′ and 3′ ends (purple and green, respectively). (d) The data in panel c confirm the predicted division of genome organization in MS2 into the distinct-hemispheres-model prediction (35) illustrated, color coded as in panel c.

Olke Uhlenbeck and colleagues (17, 18) showed that the translational repression complex is composed of a CP dimer bound to a 19-nucleotide (nt) stem-loop (SL). This translational repression operator (TR) encompasses the replicase AUG start codon in the 3′ leg of its stem. CP binding can be studied with short RNA oligonucleotides in vitro, allowing this complex to become a paradigm for sequence-specific RNA-protein recognition. Hundreds of sequence variants (17, 18), and RNAs encompassing chemically variant groupings incorporated via solid-phase synthesis, were tested for their effects on CP binding (119, 120). In collaboration with Lars Liljas in Uppsala, we showed that it was possible to soak RNA oligonucleotides into MS2 virus-like particles (VLPs) that form in E. coli when the CP is expressed recombinantly. Such VLPs crystallize easily, leading to a plethora of high-resolution X-ray structures for RNA SLs bound to CP (25, 58, 104, 130, 131), including the first structures for RNA aptamer-protein complexes (25, 104).

The TR site was suspected to be the phage assembly initiation site on the genome, coupling assembly to the cessation of replicase expression (5). Fluorescence resonance energy transfer studies of the folding behavior of oligonucleotides encompassing the TR sequence are consistent with it acting as a molecular switch (49). In addition, an icosahedrally averaged cryo-EM reconstruction at ~8-Å resolution shows RNA density underneath many CP dimers (125) (Figure 1a ). We therefore proposed that other SLs from within the genomic RNA may act in a similar way to TR (117), consistent with expectations based on in vitro affinity assays with oligonucleotides encompassing two copies of TR (136). Noncovalent mass spectrometry then revealed that the CP dimer by itself and the saturated CP2:TR complex were kinetically trapped for many hours, while mixtures of these two conformers rapidly lead to VLP formation. This suggested that TR must act as an allosteric switch, converting the symmetric C/C dimer into the asymmetric A/B conformer. Nuclear magnetic resonance spectroscopy provided vital clues as to the direction of this effect (68, 117), and in silico modeling of assembly intermediate formation again suggested that multiple TR-like effectors would be advantageous (80).

The nature of the mechanism underlying this effect was still obscure. We therefore used normal-mode analysis to dissect the mechanism of the allosteric conformer switching of the CP dimer. Upon TR binding, the motions of the two loops connecting the F and G β-strands in each monomer of the dimer, the sites of largest conformational change between quasiconformers, become distinct. This facilitates one of them folding to form the B conformer (36, 130). We concluded that any SL binding to CP in the same binding site as TR would have this effect, suggesting that many more SLs than previously anticipated could act in this way. Indeed, SLs derived from the genome other than TR prove equally effective at triggering assembly. Their CP affinities are lower, making it difficult to saturate the complex with TR, thus leading to assembly across a wide range of concentrations (4, 80).

We created a model to analyze the consequences of such binding for genome organization. We term the CP-binding, multiple, dispersed, and sequence-degenerate RNA SLs within the viral genomes packaging signals (PSs) by analogy to the high-affinity PS TR, and the consequences of these binding events PS-mediated assembly. We predicted that the multiple dimer switching events triggered by PSs during MS2 assembly would result in an assembly pathway in which dimers binding to PSs 5′ or 3′ of TR are located in different halves of the virion shell (35) (Figure 1d ). This has recently been confirmed by asymmetric cryo-EM reconstruction (29) (Figure 1b,c ). We later also combined bioinformatics with geometric information on the positions of the PS-binding sites at the inner capsid surface (the Hamiltonian paths approach, in which genome organization in proximity to capsid locally resembles a segment of a path connecting proximal binding sites) to identify all TR-like PSs in the MS2 genome (37, 127). The predicted ensemble contains all 15 SLs that remain so tightly bound in the virion that the asymmetric reconstruction allows atomic models to be built directly into the electron density. These results have also been independently validated by CLIP-Seq (cross-linking immunoprecipitation sequencing) and lead ion footprinting within the virion (99).

Packaging-Signal-Mediated Assembly Overcomes Mechanistic Challenges

Every virus faces three key challenges during assembly. It must confine its genome into the capsid, ensure efficient capsid formation in the arms race against the immune system, and guarantee specific packaging against a backdrop of cellular competitor RNAs. We, and others, have addressed these issues with a combination of experimental and theoretical techniques. The secondary structures of nucleic acids depend on their ability to form base pairs. This impacts the conformations of genomic RNAs, with viral ssRNAs being typically more compact than other RNA molecules (51, 138) or synonymously mutated RNAs (126), implying that viral genomes have evolved the ability to be confined into their capsids. Compactness of a viral genome is typically described in terms of its hydrodynamic radius (R _h).

We introduced single-molecule fluorescence correlation spectroscopy (smFCS) to follow the real-time R_h of dye-labeled genomes during titration with CPs (9) (Figure 2a ). Working at the single-molecule level allows low reagent concentrations to be used (~1–100 nM), significantly lower than in previous in vitro assembly assays (3, 47). Under such conditions, addition of sufficient cognate CP to encapsidate all copies of the genomes of phage MS2 or STNV results in a sudden collapse in the R _h of protein-free RNA (~30% in each case). Over time, these values increase, plateauing at the size of capsids, consistent with the VLPs seen in electron micrographs. This is in stark contrast to the results with noncognate interactions, where the collapse does not occur, and R _h is either unaffected or increases due to lack of CP binding or binding and aggregation, respectively. Similar results occur when labeled CP is titrated into unlabeled RNAs. Compaction of the viral genome and virion assembly in vitro, at least, must therefore be a consequence of sequence-specific interactions. Such specificity mirrors the outcomes of natural viral infections that incorporate specific RNAs at close to 100% efficiency (102). This was the first in vitro reassembly experiment to demonstrate such specificity. We concluded from the smFCS data that the MS2 CP dimer switching events triggered by CP binding the PSs occur in two stages. A subset of the CP-free genomic PSs are competent (i.e., folded into SLs) to bind protein rapidly and cooperatively, promoting formation of the CP–CP contacts in the virion shell and hence in the collapse (Figure 2b ). A second group of PSs are induced to fold into SLs during this process, creating a continuous RNA-mediated assembly pathway, consistent with RNA probing (99).

Figure 2. Packaging-signal (PS)-mediated assembly of Satellite tobacco necrosis virus (STNV) and mathematical modeling.

(a) Single-molecule fluorescence correlation spectroscopy (smFCS) traces of STNV coat protein (CP) interacting with dye-labeled STNV (orange) or MS2 (green) genomes. A drop in hydrodynamic radius occurs only with the cognate genome, demonstrating sequence specificity of this interaction (9). (b) Molecular basis of PS action in STNV. The N-terminal tail of the CP subunit shown in green-magenta (left) becomes more ordered as a PS binds and triggers assembly, as can be seen in the corresponding purple section of a tail in the virus-like particle (right). (c) Cooperativity between PSs is revealed by smFCS with oligonucleotides encompassing one (navy) or five (black) PSs from the 127-mer 5′ genomic fragment. Note the larger fragment undergoes a collapse in hydrodynamic radius (R _h) similar to that seen in the full-length RNA (90). (d) Mathematical modeling of PS-mediated assembly for a dodecahedral model system formed from 12 pentagonal units that can associate with, and disassociate from, any one of 12 binding sites (PSs) on model RNAs (top). These are represented here and in (e) as strings of colored beads indicating positions of PSs with high (green; ≥−12 kcal/M and <−8 kcal/M), intermediate (blue; ≥−8 kcal/M and < −4 kcal/M), and weak (red; ≥−4 kcal/M) affinities for CP (39); here all beads are shown in green as an example for illustration purposes. The assembly process is modeled via a Gillespie algorithm based on a set of reactions (bottom) describing the association and disassociation to and from RNA and between bound CPs on neighboring PSs. (e) If a complete aliquot of CP, able to assemble around each RNA, is added at the start of the simulation, capsid yield (green) is less than two-thirds of the expected yield and is characterized by significant amounts of malformed species, as well as variable nucleation at most PS pairs (bottom; percentages indicating propensity to act as a nucleation site). If CP is added according to a protein ramp (right), capsid yield is almost 100% and nucleation is highly localized.

To understand the consequences for assembly of cooperative CP binding, we developed a mathematical model of capsid assembly for a dodecahedral model system (Figure 2d ). We tracked assembly of pentagonal building blocks (CP) into dodecahedral capsid shells around genomic RNAs that were modeled as sequences of PSs with varying affinities for CP (38, 39). Assembly efficiency varied strongly with the PS–CP affinity distributions across the genome, and for most sequences improved when the CP concentration in the model was allowed to rise slowly, in contrast to having a complete stoichiometric aliquot available from the start. Such a protein ramp mimics measurements of phage CP expression in vivo (40), and the model suggests that the true impacts of PSs on assembly efficiency can only be seen under such conditions (Figure 1e ). Mapping the order in which PSs are bound by CP dimers as they enter the growing RNA-CP complex suggests that there are a limited number of ways to assemble efficiently via this mechanism. These solve the viral equivalent of Levinthal’s paradox in protein folding. The protein ramp, in combination with variation in PS affinities for CP, favors specific assembly pathways, preventing kinetic traps created by multiple assembly initiation events on the same RNA. These results in turn suggest that the path of RNA genomes adjacent to the protein shell in virions should be highly conserved, which is consistent with our previous predictions (35, 37, 50) and the fact that genomic RNA density is visible in nonaveraged structures (29, 52). This model also demonstrates that with a protein ramp multiple PSs result in selective assembly around model cognate genomes, even in the presence of a massive excess of competing nonspecific RNAs [in a factor 1:300, typical of in vivo scenarios (39), with their prospective PS affinities all set to the lowest values of test RNAs]. It also implies that under these conditions, both PS affinities and PS positions are important for efficient virus assembly.

To test that idea experimentally, we worked with STNV, whose CP subunits have basic N-terminal tentacles and are therefore more representative of wider families of viruses. STNV, a T = 1 capsid, shows two-phase behavior in the smFCS assay, even though its biology is different from that of MS2. To identify its PS sites, we used an approach that has subsequently proven successful for other viruses. The CP subunits were used as a selection target for RNA systematic evolution of ligands by exponential enrichment (SELEX) (13). The resulting aptamer sequences were screened for matches against the STNV genome, identifying up to 30 SLs across each of the genomes of all three known strain variants with a common loop motif of AXXA (X = any nucleotide) (12). We then used smFCS to confirm that the best match to an aptamer, located in the 5′ 127 nts of the genome, triggered sequence-specific assembly of STNV capsids by promoting formation of a CP trimer (90). The crystal structure of the resulting VLP shows that the N-terminal tails, which form a short helix, become more ordered in the presence of cognate RNA—an example of Caspar’s specific tentacular interactions but with an RNA, not a protein partner (46). For STNV binding, PS RNA neutralizes the positive charges in the N-terminal tail of its CP, allowing a trimeric capsomer to form.

Five PSs are predicted to occur within the first 127 nts of the STNV genome, with the highest-affinity site being central. We used this fragment to confirm the minimal CP recognition motif within the loop of each PS; showed that multiple PSs significantly enhanced in vitro assembly (i.e., showed the cooperativity inherent in PS-mediated assembly) (Figure 1c ); confirmed that the relative spacing of PSs was critical for this effect; and showed that folding propensity of each PS SL contributed to their functions in assembly (90, 92). These experiments allowed us to extract the assembly instructions in the form of SLs, recognition motifs, and relative spacing and introduce them to a nonviral RNA sequence. The latter can be tuned to be a better or worse assembly substrate than the wild-type viral RNA. It was also possible to construct genome-length variant RNAs lacking all PSs, or do so with only PSs in the 127-mer 5′ cassette, and show that complete assembly requires PSs throughout the RNA. It is important to recognize that these in vitro experiments may differ from the mechanism of assembly in vivo. For other ssRNA plant viruses, such as Brome mosaic virus, assembly may be restricted to nascent genomes, consistent with the detection of CP-binding sites on viral polymerases/replication complexes (109). A similar mechanism is thought to occur in picornaviruses (see the section titled Picornaviruses below and Reference 110). While these details may alter the mechanism by which SLs fold to reveal CP-binding motifs, the basic CP–RNA interactions are probably the same as those identified in vitro.

A Paradigm Shift in Our Understanding of Virus Assembly

The results of in vitro reassembly assays with MS2 and STNV do not fit the previous paradigms of virus assembly. These were based on experiments at concentrations that were in retrospect too high and failed to show any genome packaging specificity [i.e., the opposite of most biological outcomes (102), as well as the assays at single-molecule concentration]. CPs alone can assemble VLPs or even package host RNAs (102, 103), anionic polymers (3, 48, 60), or a functionalized gold core (22). The first models of virus assembly were protein-centric and focused on the local rules (7, 106), kinetics (42, 81, 133, 143, 144), thermodynamics (11), and electrostatic contributions (132) that regulate protein container formation. The energetic contributions accounting for the formation of different quasiequivalent conformations (97) or T-numbers (139) have been analyzed and solid printed models of capsid assembly created to illustrate the roles of intercapsomer interactions (85). Protein-centric models are appropriate for double-stranded DNA viruses that typically package their genomes into preformed containers via an ATP-driven packaging motor (21). Assembly of those containers often relies on additional components, such as scaffold proteins (122), rather than interactions between genomic material and capsid protein. By contrast, ssRNA viruses exploit their genomic RNA to promote concomitant encapsidation with capsid assembly (i.e., their formation relies on a coassembly process).

The impact of CP–RNA interactions on assembly has been studied by weakening CP–CP interactions (47) or by coarse-grained modeling (54, 67, 96), and different contributions from hydrophobic and electrostatic interactions and so-called Caspar carboxylate pairs in the formation of a helical virus (65) have been studied. Several models have been proposed for the roles of the genomic RNA (e.g., via charge neutralization of negative charges on the genomic RNA via interactions with multiple copies of CP) (6, 15, 43, 123), and dependence of viral RNA encapsidation on its sequence (66) and biophysical properties [e.g., stiffness (75)] has been modeled. Electrostatic contributions do play roles in assembly, as seen in the formation of the STNV capsomer, but these are not the dominant driving force. The affinities of the highest-affinity PSs in the MS2 and STNV genomes for their respective CPs is in the low nanomolar range (18, 67, 71, 90) when measured in isolation. Modeling the concentration of a single PS oligonucleotide within the volume of a capsid (e.g., MS2) suggests that the local concentration is in the millimolar range. It is not surprising, therefore, to find that PSs can encompass considerable sequence variation (29), especially as they all benefit in affinity terms by being part of the same molecule. There may well be additional features within viral RNAs that contribute to their ability to be compacted into the small volumes of their capsids [e.g., the MP in the case of MS2 (see above)].

The results described above with the MS2 and STNV examples clearly demonstrate the importance of sequence-specific RNA–CP interactions for assembly, establishing the new paradigm of PS-mediated assembly. These results beg the following question: Are these phenomena confined to the bacterial and plant viruses, or do the principles of multiple RNA packaging signal-mediated assembly apply much more widely in nature?

Hallmarks of Packaging-Signal-Mediated Assembly in Human Viral Pathogens

Animal and human viruses are more challenging to work with, because they are difficult to prepare on the large scale required for biophysical assays. Their biology is also often more complex, and they may also require biocontainment. However, in the last few years, we have been able to adapt the successful approaches used with model viruses to probe whether PS-mediated assembly is a widely occurring paradigm.

Picornaviruses

The picornaviruses are a very large and important group of ssRNA viruses that infect many different organisms. The best-known exemplars are poliovirus and human rhinovirus (HRV), the causes, respectively, of poliomyelitis, which results in paralysis and sometimes death in infants, and the common cold. The symptoms of HRV are short lived and relatively mild, except for patient cohorts with preexisting respiratory conditions, but it remains the most frequent viral infection in people, with estimates of more than two billion cases per year (140).

Parechoviruses are a subgroup of picornaviruses that are the leading cause of infant death by sepsis (84). Unlike for polio, there are no vaccines for their prevention, and up to nine strain variants are known. Most picornavirus structures, characterized either by X-ray or cryo-EM, are symmetrically averaged and as expected show little or no density for their RNA genomes [~7,500 nts long (57)]. Our attention was therefore drawn to an ~8.5-Å EM structure of Human parechovirus 1 (HPeV1; Figure 3a ) from the Butcher group in Helsinki (107). It reveals significant electron density around all the fivefold vertices that can belong only to ordered segments of the genome, an arrangement suggestive of PS-mediated assembly. Jointly with the Butcher group, we carried out SELEX against the pentameric capsomer of HPeV1 and developed a novel bioinformatics analysis of the data, identifying multiple PS-like sequences distributed across the genome (110). Each of these sites can fold into an SL with a similar consensus loop motif: GXUXXX, suggesting that they are PSs. A later X-ray structure showed that the ordering of the RNA density encompassing a hexanucleotide segment extends to at least 3.1Å (64). Building the SELEX consensus motif into this density revealed an excellent fit (Figure 3b ) and permits multiple hydrogen bonds to form between amino acid side-chains and the bases at positions G1 and U3, explaining their specificity. The rest of the oligonucleotide is contacted along the phosphodiester backbone. Mutation individually to alanine of each amino acid side-chain contacting the hexanucleotide RNA motif results in million-fold drops in viral titer. Individual PSs are mostly within coding regions and are therefore more difficult to ablate synonymously. However, single PS substitutions result in drops in titer from zero to a million-fold. In each of these experiments, the amounts of viral RNA and CPs produced by the mutant viruses were roughly similar. These results are consistent with the primary target of the mutations being virion assembly and are good evidence that 60 RNA-CP contacts occurring at PSs within the genome are critical for assembly. This is the first identification of RNA assembly determinants for a picornavirus, which have been thought to assemble purely via a CP–replication complex interaction (61). Picornavirus RNA PSs are controversial, but there would be no need for sequence-specific interaction in these contacts were they merely secondary RNA sites assisting genome compaction, as has been proposed in Reference 114.

Figure 3. Packaging-signal (PS)-mediated assembly in human pathogens.

(a, left) The Human parechovirus-1 (HPeV1) capsid seen along a fivefold axis from the outside and (right) a pentamer viewed from the inside. The latter shows density for ordered hexanucleotides (gray) corresponding to its PSs (VP0, VP1 & VP3 correspond to the three structural proteins). (b) Electron density map showing the specific recognition of G1 of the GXUXXX recognition motif (top left) and a view of one hexanucleotide and its interactions with residues in the oligo-binding pocket (top right). The generic PS motif is shown below (left), with the specific PS (lower right) that encodes part of the PS-binding site (110). This coupling of genetic code and assembly instructions perhaps explains how the mechanism persists in error-prone replicators. (c) Single-molecule fluorescence correlation spectroscopy trace of PS-mediated assembly of HBV nucleocapsids in the presence of its most conserved PSs and its immediate flanking 3′ sequence that can fold into distinct secondary structures (black, red, and navy). (d, top) The proposed sequence-specific charge neutralization of the arginine-rich domains by RNA, permitting interaction between CP dimers as the first step of assembly. (bottom) A cryo-EM reconstruction of the principal product of the assembly reaction shown in panel c, a T=4 virus-like particle (VLP); and (lower right) a cross-section of an asymmetric cryo–EM reconstruction of this VLP (91).

An obvious puzzle is how such RNA–CP contacts promote assembly. Empty picornaviral capsids assemble in vivo routinely. It is thought that they act as a reservoir for subsequent virion assembly (83). The empty VLPs are unstable, so RNA encapsidation is required to stabilize the particle. This is consistent with the very extensive buried surface area involved in the PS–CP contacts in HPeV1. Additionally, RNA may stabilize the conformation of the CP pentamer. These are composed of five copies of viral protein 1 (VP1), VP3 and VP0, each of which has an N-terminal tail (tentacle). The RNA-binding sites and specific recognition motifs are formed from tentacles from the nonadjacent CP subunits, which are therefore sequestered into a particular conformation (110).

Similar RNA–CP contacts have been seen in the structures of two other Parechoviruses, HPeV3 (111) and Ljungan virus (142), suggesting a common phenomenon across the genus. The SELEX aptamers raised against HPeV1 CPs show extensive matches to the genomes of these viruses, suggesting that PS-mediated assembly is a conserved feature in this group of pathogens. In most picornaviruses, the VP0 subunit is cleaved to form VP2 and VP4, leading to large-scale rearrangements of CP tails and RNA in the capsid. This may mask RNA–CP contacts involved in initial assembly. Assembly of picornaviruses occurs in replication factories producing nascent RNA chains. It is thought that protein–protein-specific contacts between the virally encoded 2C protein and VP1 allow this RNA to interact with the CP (61). We have postulated that both the protein- and the RNA-mediated aspects of assembly could coexist in a viral machine we named an assemblysome (110). This would be consistent with the vital roles in assembly played by RNA–CP contacts in Foot-and-mouth disease virus, which is one of the picornaviruses that cleave their VP0 subunits (77). Indeed, such cleavages have been postulated to occur in a genomic RNA-dependent manner. This implies that multiple RNA–CP contacts must be formed, i.e. the RNA components act as PSs (1). Indeed, similar virion assembly on a virally encoded complex may be a common feature when nascent transcripts are being packaged (i.e., when transcription and assembly occur concurrently). Brome mosaic virus, a ssRNA plant virus, has a CP-binding site on one of its two replicase subunits, which could act as an assemblysome in this way (109). Structure probing of STNV RNA conformation in virio, or following deproteinization, is also consistent with its assembly initiating on incomplete RNAs (73).

Hepatitis B virus—a DNA virus that uses RNA to guide assembly

Hepatitis B virus (HBV) is an enveloped pararetrovirus that assembles an initially ssRNA version of its genome, the pregenomic RNA, inside a T = 3 or T = 4 nucleocapsid, along with a copy of its polymerase (108). Since the formation of the nucleocapsid occurs around ssRNA, we used our SELEX protocol against the HBV CP and our bespoke bioinformatics approach to explore the possibility that this process might make use of RNA PSs (91). This identified evolutionarily conserved genomic fragments that appeared like the PSs in ssRNA viruses (i.e., they can fold into SLs displaying a conserved sequence motif in the loops). Single-molecule FCS of dye-labeled oligonucleotides encompassing individual PS-like sites revealed efficient sequence-specific triggering of nucleocapsid VLPs under conditions where CP alone self-assembles very poorly (Figure 3c ). RNA sequence variation experiments confirmed the importance of the consensus motif and the secondary structure for CP recognition and assembly. The HBV CP includes a C-terminal tentacle rich in arginine residues known as the arginine-rich domain (ARD). Removal or phosphorylation of the ARD results in protein-alone assembly of VLPs and abolishes RNA binding. This suggests a role for PSs in HBV akin to those in STNV— namely, that sequence-specific RNA binding overcomes repulsive charge interactions between CPs (Figure 3d ). There are many fewer PS-like sites across the HBV pregenome compared to ssRNA viruses. This may hint at the very different biology of this virus. HBV virions contain a double-stranded DNA genome created by the action of its polymerase on the pregenomic RNA inside the nucleocapsid, which is permeable to nucleotides. Polymerase must therefore move within the nucleocapsid, traveling along the RNA like a track (135). PS-like sequences, in addition to their role in promoting assembly, may therefore also be useful in ensuring that the pregenome is not entangled and is thus a good encapsidated replication substrate.

Therapeutic Prospects for Rna-Mediated Virus Assembly

Assembly Inhibitors

Zlotnick and colleagues have developed small molecular weight drugs disrupting nucleocapsid formation in HBV (105). They showed that HAP (heteroaryldihydropyrimidine) compounds act as allosteric effectors promoting the over-rapid assembly of the nucleocapsid. The presence of RNA PSs that play critical roles in HBV nucleocapsid and picornavirus virion assembly opens up a novel set of targets for the identification of antivirals. Preliminary experiments with a library of RNA-binding ligands (see (24)), has identified those specific for viral PSs. Some of these inhibit HBV in culture, offering a novel therapeutic target for a viral infection for which there is a significant clinical need. These results show that PS-mediated assembly is a druggable mechanism.

We have assessed the merits of such novel therapies directed at multiple dispersed PSs in the context of viral infection models that describe the evolution of the viral quasispecies. Using the modeling framework in Figure 2d above, we introduced fitness functions that capture the contribution of PSs to viral fitness (8, 34), so that the model can assess the impact(s) of drugs targeting PSs. Computationally, therapy directed at multiple PSs, rather than replicase— currently, the predominant antiviral drug target—is less likely to elicit resistance mutations (8). This suggests that PS-targeting drugs overcome two key bottlenecks in existing forms of therapy: They minimize therapy resistance and cover a broader spectrum of viruses in a viral genus.

Repurposing of the Assembly Manual for Virus-Like-Particles Production and Vaccination

Our understanding of the PS-mediated assembly mechanism also opens up another opportunity for translation. Since the PS assembly code can be isolated and repurposed for the production of VLPs (92), these can be designed for drug/gene delivery and synthetic vaccine applications.

Discussion

This is an exciting period for studying viral assembly mechanisms because several technical advances in different fields (10, 68, 93, 113, 128) in combination with modeling allow interrogation of the process with ever greater precision, while structural studies have been invigorated by advances in electron microscopy (29). Modeling suggests that there are distinct evolutionary advantages for ssRNA viruses assembling via the PS-mediated mechanism, consistent with its discovery in viruses infecting different hosts and from distinct viral families. Even though the precise molecular manifestations of the mechanism differ from example to example, the general principle is shared. The concept of a single PS in a viral genome is not new (95), but the idea that there are multiple versions of such sites that act cooperatively is. There is, therefore, much work yet to do in identifying PSs in other virus families. For example, it will be fascinating to see whether PSs occur in nonspherical viruses, such as the fullerene core of HIV (86), where nucleocapsid protein–genomic RNA interactions have been identified by CLIP-Seq (70). Modeling has already addressed some of the roles of viral genomes in the formation of conical retroviral capsids (44) and maturation pathways (82), and it will be interesting to see how far sequence-specific contacts between genomic RNA and capsid protein contribute to these processes.

PS-mediated assembly also has profound implications for viral cell entry and evolution. Most ssRNA viruses introduce their genomes into target cells following interaction with a cellular receptor. This occurs by extrusion of the genome across membranes rather than sudden release into the cytoplasm, thus allowing viruses to avoid silencing. This mechanism in turn requires an asymmetric association with the receptor that is coupled to the location of a genome end within a capsid. As we now see very clearly with MS2, asymmetry is a critical aspect of assembly of an infectious particle. Such asymmetry can be structural, such as with the unique MP component of the MS2 shell, or it could be dynamic, because CP–RNA interactions will vary across an otherwise uniformly repeating capsid structure, leading to asymmetric local motion. Such motions can be dramatic even within such extended biological assemblies (98). Such dynamical changes can alter the local affinities for receptors and may be the origin of the limited interaction of parvovirus with a unique transferrin receptor (53). Clearly, overly strong RNA–CP interactions would not facilitate these infection mechanisms, so PS–CP contacts must be weakened or lost during this process. Indeed, since they are only absolutely required for initial assembly, further maturation events—such as VP0 cleavage in the picornaviruses—may mask their presence.

A major evolutionary aspect to PS-mediated assembly is how such defined interactions manifest themselves across different strain variants and within a quasispecies of genomes. As we have demonstrated in parechoviruses, the CP recognition motif is shared by different strain variants. One explanation for this, which seems to apply in the three examples for which we have structural details (90, 110), is that one of the PS sites encodes at least part of the PS-binding site of the CP (Figure 3b ). Mutation of that PS would therefore result in loss of the binding site, affecting all PS–CP contacts and deleterious effects on viral fitness. Such mutations are therefore unlikely to survive into future generations of infectious virions, as we also demonstrated via modeling (8). As mentioned earlier, this hints at a novel drug target that is less likely to trigger resistance, implying that inhibiting this mechanism could be the basis of robust, broadly acting antivirals. The conservation of PS sites across known strain variants of HPeV1 supports that idea.

It is also possible that PS-mediated assembly extends beyond virology. The evolution of protein-based containers to traffic messenger RNAs (mRNAs) between cells is a remarkable event. It is perhaps, therefore, not surprising that capsid-like structures have evolved for this purpose from proteins that are evolutionarily related to viruses, such as the retrotransposon Gag-like Arc proteins (2, 89). Our understanding of such systems is also being extended by the evolution of self-assembling proteins that encapsidate their own mRNAs in a laboratory environment from first principles (14, 121). It seems we are only just at the beginning of appreciating the many roles of RNA motifs in biological assemblies.