1 Introduction

Mammalian viruses form a large class of nanometer-scale, small proteome organisms that use the target host cells to replicate and propagate. In spite of progress made with antivirals and vaccines, viral pathogens still represent a large public health burden in many developing countries with continued emergence of new strains (Woolhouse and Gowtage-Sequeria 2005; Jones et al. 2008; Howard and Fletcher 2012). Drug and vaccine design against viruses are driven by conventional ‘screening’ as well as rational design-based approaches (Quan et al. 1998; Finco and Rappuoli 2014). In both the cases, mechanistic understanding of molecular function, dynamics and interactions can provide profound and useful insight to inhibit key viral processes. In this review, we make a case for application of single-molecule fluorescence as a complementary tool to structural biology, functional activity assays and ensemble spectroscopic techniques. We review the current progress in the field of single-molecule virology, highlighting the studies that generated mechanistic understanding of viral protein, genome activity and function, viral lifecycle processes and their interactions with the host components.

Virus lifecycle involves a complex interplay of virus and host cell interactions that determines the infection outcome (figure 1a). The virus infection cycle starts by binding to receptors and/or membrane on the host cell surface. Most viruses enter the cell through endocytosis and, despite variations in structure and size, display commonalities in trafficking (Pelkmans and Helenius 2003; Marsh and Helenius 2006). After internalization, the viral genome release from endosomes into the cytoplasm occurs either by viral membrane fusion to the endosomal membrane (most enveloped viruses) or by disruption (or fusion) of the cellular membrane (most non-enveloped viruses). This is followed by key viral lifecycle processes, that is, replication of the viral genome to produce its progeny and translation of the additional viral proteins that are required for new rounds of replication, packaging of the new virus and other essential processes. The order and sites for these processes varies depending on the virus with viral proteins continuously interacting with cellular organelles and host proteins during their lifecycle. Finally, assembly and packaging of the new viral genomes into virus particles is followed by budding, exocytosis or cell lysis. The new virions thus released go on to infect new cells.

Figure 1
figure 1

Single-particle fluorescence-based tracking can monitor major steps in virus trafficking to and inside a host cell. (a) Schematic overview of the events of viral entry into a mammalian cell that can be explored with single virus or protein imaging experiments (not to scale) (i) Diffusion of the virus on a mammalian cell surface can be monitored in real time by fluorescently labelling the viral outer coat (e.g. with Quantum dots, QD). (ii) Attachment of the virus to the mammalian cell surface mediated by host–receptor interactions can be studied by high-resolution fluorescence co-localization. (iii) Endocytosis of the virus in clathrin-coated pits can be similarly tracked by co-localization. (iv) Endosomal maturation–associated virus fates can be measured using co-localization with endosomal markers. (v) Membrane fusion tracked by dequenching of virus membrane labels followed by release of the viral genome can be monitored with markers that bind the genome specifically. (vi) Budding and egress can be followed with labelled coat proteins along with host factors. (b) Representative particle tracks where the virus centroid (interferometric signal, black) and the peripherally attached tag (quantum dot fluorescence, green) signals are observed as it diffuses along the cell surface. The scattering from the virus could be used along with QD to monitor the orientation in which the virus binds to the cell surface. Simultaneous host protein tracking (red) and co-localization of all tracks can be used to infer specific interactions and kinetics of the processes.

One might ask, what is the need to study viruses at the single-molecule level (i.e. to characterize each molecule or virus individually)? Technically, single-molecule measurements represent the highest level of sensitivity and hence do not require perturbations like over-expression or excessive labelling of molecules to probe systems. More importantly, they do not require synchronization of biological reactions and processes, and can detect transient, yet significant, intermediates. In fact, the intrinsic stochastic nature of reactions is best evident at the single-molecule level. This also allows one to dissect reaction pathways and kinetics and identify on- and off-pathway mechanisms (Weiss 1999; Walter et al. 2008). Additionally, kinetics and thermodynamics can be measured simultaneously since the molecules continue to undergo forward and backward reactions at the rates defined by the thermodynamic barriers even under equilibrium conditions.

Single-molecule methods become more relevant when the biological process is inherently initiated and dependent on the activity and function of a small number of molecules as in the case of virus infections. Viruses comprise small number of structural proteins that form their coat and a few copies of enzymes and usually consist of a single copy of nucleic acid as its genome. Viral infections also start from small number of infectious units, and hence stochastic effects likely control the infection kinetics and outcome. Additionally, virus infections are intrinsically heterogeneous processes (i.e. variable over several scales both spatially and in composition) with their properties and distributions changing with time. Combined with similarly heterogeneous host cell interactions and changes induced by the viral infection, inferences from ensemble averaged data is extremely challenging (Snijder et al. 2009; Heldt et al. 2015; Ramanan et al. 2016). Several prior reviews have illustrated the differences between single-molecule and ensemble experiments elegantly and highlighted various single-molecule-based case studies in biology and we refer the reader to those for brevity (Weiss 1999; Moerner and Fromm 2003; Joo et al. 2008; Tinoco and Gonzalez 2011). In line with the focus of the special issue, we have reviewed the single-molecule- and single-virion-based fluorescence methods commonly employed in virology and underscored the new insights generated that were otherwise difficult to establish with conventional methods. We focus on viral entry, virus membrane fusion, viral protein interactions with nucleic acids relevant to replication and translation and finally virus assembly and packaging highlighting the common single-molecule fluorescence strategies used to study such processes and the understanding generated in some select cases.

2 Single-virus particle dynamics

Visualization of single viruses as they undertake infection of the cell has been a long-coveted goal that has become possible with the advent of multiple ways to label viruses without loss of infectivity and advent of highly sensitive and fast detectors (Seisenberger et al. 2001; Brandenburg and Zhuang 2007; Rust et al. 2011; Sivaraman et al. 2011). Usually virus imaging employs a fluorescently labelled virus particle that is monitored over different phases of its lifecycle in the live cell. To ensure efficient labelling of the virus without hampering its infectivity, several approaches have been demonstrated successfully (Lakadamyali et al. 2003; Finke et al. 2004; van der Schaar et al. 2008; Rust et al. 2011; Lelek et al. 2012; Sun et al. 2013; Wang et al. 2013). When isolation of the viruses is possible in high concentrations and purity, one can resort to organic fluorescent dyes to label the viral membrane (lipophilic dyes) or viral coat proteins (covalent conjugation chemistries that target lysine or cysteine residues). Quantum dots (Q-dot) or gold (Au) nanoparticles are also used occasionally to label the outer coat or envelope protein if the size of the probe does not hamper virus infection. Additionally, genomic (nucleic acid) probes such as those used in fluorescence in situ hybridization (FISH) are employed to directly follow the sites of genome processing (Chou et al. 2013). In some cases, infectious viruses with fluorescent protein fusions to viral proteins have been possible, enabling generation of genetically tagged viral particles (Finke et al. 2004; Kobiler et al. 2011; Avilov et al. 2012; Schoggins et al. 2012; Granstedt et al. 2013; Hoornweg et al. 2016; Maier et al. 2016; Sood et al. 2017; Vanover et al. 2017). The host cell is further tagged either on the membrane or on the corresponding virus receptors to enable simultaneous imaging of the interacting partners.

Epi- (including highly inclined thin illumination (HILO) mode), confocal- and total internal reflection-based microscopy are the most popular methods for single-particle detection and tracking (Axelrod 1981; Sako et al. 2000; Stephens and Allan 2003; Tokunaga et al. 2008; Rust et al. 2011). Rapid movement of the virus particle in the vicinity of the cell is captured in movies using fluorescence microscopy. Diffraction-limited images of the virus particles are identified by intensity thresholding, followed by fitting the image to a mathematical function (usually a 2D Gaussian) to estimate the centroid of the particle. Centroids of the virus particle from subsequent frames of the movie are then connected to generate a particle trajectory (figure 1b). Next, the particle trajectories are analysed for the type of motion (random diffusion, restricted diffusion, or directed motion), speed of movement and time-dependent intensity changes revealing the underlying transport processes and interactions. Co-localization of the trajectory with cellular components like receptors can further reveal the time scale and nature of contacts (figure 1b). Furthermore, similar fluorescence trajectories of the virus can be co-localized with other trajectories including those from other imaging modalities (figure 1b). Using such an approach with scattering interferometry that reports on virus particle centroids and comparing it to centroid of Q-dots located on virus periphery, the position and orientation of the virus could be measured with sub-nanometer spatial and less than 10 ms temporal resolution (Kukura et al. 2009).

For the non-enveloped viruses, labelling of the capsid proteins has allowed tracking of the viruses, and genome release could be monitored post-infection. One of the first reports of single-virus tracking of Adeno-associated virus (AAV) in HeLa cells displayed heterogeneous diffusive behaviour on cell membrane and post-cell entry (D ~ 0.4-7 µm2/s) (Seisenberger et al. 2001). After rapid endocytosis, AAV motion was characterized by anomalous diffusion in the endosomes or directed motion possibly under the influence of motor proteins. Such heterogeneous movement, a common feature observed in virus tracking experiments, is masked in other ‘coarser’ approaches and highlights the underlying need of probing particle movements individually. More recent work has described how viruses use cortical actin cytoskeleton and cell surface attachment factors to ‘surf’, search and activate binding to specific receptors and cell entry sites (Lehmann et al. 2005; Coyne and Bergelson 2006). Combining multi-colour imaging with virus tracking has helped resolve the underlying principles for cellular entry and transport. When the simian virus 40 (SV40) was co-visualized with the caveolae, a two-step pathway for delivery of the virus to the smooth endoplasmic reticulum (ER) was discovered (Pelkmans et al. 2001). After entrapment of the virus in the small caveolin vesicles, these vesicles would merge with larger caveolin-rich organelles (caveolosome). Viruses would then be sorted in caveolin-free vesicles and undertake microtubule-guided movement to the ER for genome delivery. On the other hand, Polio virus (PV), tracked in a similar fashion, entered the cell via the endocytic pathway that was independent of clathrin or caveolin and did not involve the microtubules (Brandenburg et al. 2007). On the other hand, depletion of ATP, inhibition of tyrosine kinases and disruption of actin microfilaments led to a complete loss of infection events. By simultaneously labelling the viral RNA, they further showed that the PV genome was released from the encapsulating vesicles close to the cell membrane surface after inducing conformational changes in the viral capsid and formation of the viral pore complex. Therefore, non-enveloped viruses employ a multitude of clathrin-independent endocytic pathways instead of disrupting the plasma membrane directly to achieve productive entry into the host cell.

Enveloped RNA viruses that require an obligatory membrane fusion step have been studied using viral membrane labelling with lipophilic dyes and co-localizing them with endocytic pathway markers. The influenza virus displayed distinct and highly heterogeneous entry and transport pathways in the cell (Lakadamyali et al. 2003; Rust et al. 2004; Lakadamyali et al. 2006). The motion of the influenza particles on the cell surface was found to be actin dependent in the initial stages that switched to dynein-dependent movement on microtubules towards the perinuclear region post endocytic entry (Lakadamyali et al. 2003). About two-thirds of the virus particles underwent Clathrin-dependent endocytosis by de novo formation of clathrin-coated pits and the rest entered the cell using clathrin- and caveolin-independent pathways (Rust et al. 2004). The virus was also selectively sorted into a highly mobile pool of early endosomes (Lakadamyali et al. 2006). Eventual fusion of the viruses measured by dequenching of the viral membrane dye occurred after this microtubule-dependent transport and was dependent on this transport. Co-localization with Rab5 and Rab7 endosome markers showed how the fusion typically happened during the endosomal maturation process. In case of the dengue (DENV) virus, another enveloped virus, binding to cells was extremely poor, which could explain its poor infectious unit-to-viral particle ratio (van der Schaar et al. 2007). Dengue virus gains cell entry exclusively via clathrin-mediated endocytosis (van der Schaar et al. 2008). However, unlike influenza, DENV particles diffusively searched the cell membrane till they were arrested on clathrin-coated pits. Once endocytosed, only one-sixth of the DENV viruses displayed complete fusion that occurred in Rab5/Rab7-rich late endosomes. Therefore, such single-virus tracking studies in combination with inhibitors and antibodies can reveal cell receptor-mediated virus interactions and viral cell entry pathways which are potent targets for intervention.

3 Protein–membrane interactions

Another crucial interaction that modulates viral infection is the virus interaction with cellular membranes. After the initial search for receptor and binding (as discussed earlier), enveloped viruses must undergo membrane fusion either with the cytoplasmic membrane or with the endosome membrane (after endocytosis) to deliver their genome into the cytoplasm for viral translation and replication. The viral membrane fusion process is a heterogeneous multi-step process with large kinetic barriers (Chernomordik and Kozlov 2003). Despite differences in the structure and size, all enveloped viruses employ dedicated envelope proteins that reside on the viral membrane in a broadly similar mechanism to achieve membrane fusion (Harrison 2008, 2015; White and Whittaker 2016). Envelope proteins, often triggered by a ligand binding (such as protons in the endosome or receptors/co-receptors on the cell surface), undergo large conformational changes and expose lipophilic ‘fusion’ regions. Insertion of these hydrophobic segments into the cellular membrane links the two membranes. Further structural changes in the envelope protein bring the two membranes closer together and thereby catalyse the fusion process. In vitro fluorescence assays that probe membrane fusion have been used with fluorescently labelled virus particles that bind and fuse to a cell membrane and/or artificial bilayer when triggered by envelope protein activation (Otterstrom and van Oijen 2013). Using TIRF microscopy to limit background, various stages of the virus membrane interaction can be studied where the viral membrane as well as the content (for example, the genome) is labelled with two distinct fluorophores. Approach and binding of the virus to the host cell (or membrane-bound receptors on supported bilayer) is observed as an abrupt appearance of a fluorescent signal (in a diffraction limited spot) on the membrane (figure 2a). This abrupt change in fluorescence intensity is largely due to the limited lateral mobility of the virus on the lipid membrane compared with 3D diffusion in the solution as well as selective illumination of the membrane using evanescent field from total internal reflection. Similarly, unbinding events as (abrupt) disappearance of the spots can be used to study the binding affinity of the membranes/receptors to the virus. Next, fusion of the viral membrane to the target membrane can be followed with a change in the intensity of the lipophilic dye. Several lipid dyes can be incorporated on the virus membrane at high self-quenching concentrations (~ 2–5 mole percent of total viral lipids) without compromising virus infectivity. Mixing and subsequent diffusion of these quenched lipid dyes upon viral membrane fusion with the target membrane causes a transient enhancement of the fluorescence signal, followed by a gradual drop at the site of fusion (figure 2b). If the acquisition is fast compared to the fusion kinetics, fusion intermediates like the hemi-fusion state (fusion of outer leaflets of the bilayer membrane) can also be captured as intermediate steps in the fluorescence changes observed. Finally, the release of the viral content can be measured by drop in content (water-soluble) dye fluorescence (‘turn-off’ assay) due to diffusion after release or fluorescence enhancement due to interaction of the contents with the external component (figure 2c). The water-soluble dyes are usually incorporated by long incubation of the dye with the virus, followed by gel filtration or dialysis. Several recent advances like the ability to generate supported bilayer platforms from cellular membranes (containing all relevant receptors and local photo-uncaging based changes in the pH and other molecules) are enabling increasing levels of sophistication in such measurements (Costello et al. 2012, 2013).

Figure 2
figure 2

Single-viral membrane fusion and genome release assays (Floyd et al. 2008). (a) Schematic of a fluorescent virus particle binding and fusing onto an artificial lipid membrane. The viral membrane is labelled with a quenched dye (red) and fusion to the artificial lipid membrane will re-distribute the dye, resulting in dequenching of the fluorescence signal. The nucleic acid (genome, blue) is labelled with spectrally distinct fluorophores and its release and diffusion can be monitored separately. (b) Time trajectory of a single virus fusion event is shown schematically. (i) When the virus binds to the cell surface receptors, appearance of a fluorescence spot (red) could be obtained due to restricted diffusion on the surface. This signal could be from a fluorescent tag on the Env or a lipophilic dye in the membrane of the virus or signal from a fluorescently tagged genome. If the quenched lipophilic dye is present on the viral membrane, fusion of the membranes can be tracked in further detail. (ii) As the two membranes fuse together, rapid dequenching due to mixing of the membranes will increase fluorescence (yellow) with time signifying the hemi-fusion step. Full fusion will lead to redistribution of the dye within the target bilayer, resulting in decline in this intensity level to values prior to fusion. (iii) Dye on the nucleic acid is released in bulk along with the capsid–nucleic acid complex and a drop in this intensity at the fusion site reports on genome release (blue). tLigand represents time interval when the virus diffuses near the membrane till it is captured by the host receptor. tHF and tF represent the time intervals for hemi-fusion post trigger and full-fusion from hemi-fusion state. (c) Fluorescence images of virus fusion assay. In a three-dye experiment, a pH sensor, the labelled viral genomic content and labelled membrane are simultaneously observed with time. (d) Representative time traces from the single viral fusion assay, viz. the pH drop (blue), hemi-fusion (green) and fusion (red) states can be acquired and estimated. (e) Decoupling the fusion kinetic pathway is possible by fitting lag times between the pH trigger and hemi-fusion (green, tHF) and pore formation (red, tHF+tF) to gamma functions. (f) The distribution of time delays obtained between hemi-fusion and pore formation (tF) are best fit to an exponential decay, indicating a single rate-limiting step in achieving full fusion post hemi-fusion. Figure 2c–f is adapted with permission from Floyd et al. by The National Academy of Sciences of the USA.

In an early influenza virus membrane fusion study that employed several of the strategies outlined earlier, a pH-dependent membrane fusion assay was used to monitor the hemi-fusion and fusion kinetics by van Oijen and Harrison groups (Floyd et al. 2008). In addition to the virus membrane and content labelling, the lipid bilayer was labelled with a pH sensor dye and supported on dextran polymer cushion to monitor pH change and promote efficient virus fusion, respectively. Gamma distribution function fitting of the hemi-fusion time distributions (time taken to observe dequenching of lipophilic dye from the point of pH change) revealed at least three intermediate steps before the formation of hemi-fusion stalk, suggesting the requirement for three hemagglutinin (HA) trimers for successful hemi-fusion (figure 2d–2f). Extending the same approach to measure the fusion kinetics of mutant HA proteins, Ivanovic et al. determined that the release of the fusion peptide from its pocket near the threefold axis regulated the formation rate of a long-lived extended intermediate (Ivanovic et al. 2013). They also compared dwell-time distributions of various fusion intermediates observed in the assay to simulations of molecular events underlying the fusion and described how the hemi-fusion proceeded rapidly upon the availability of three or four ‘adjacent’ HA trimers in the membrane-inserted state. Single-virus particle studies with West Nile virus (WNV, a representative of the flavivirus family) have also shown that the envelope protein, which is present as a dimer in the mature virus membrane, undergoes sequential steps of dimer dissociation, conformational change to form the extended state that exposes the fusion loop, followed by a much slower trimer formation that regulated the fusion process (Chao et al. 2014). The last step was controlled by the availability of the adjacent extended monomers akin to the case of influenza, but the critical number required for hemi-fusion was determined to be two molecules for flaviviruses, such as WNV and Kunjin virus. In a slightly different configuration, Wessels et al. first reported the pH-induced change in the Sindbis virus fusion protein, and observed rapid binding and fusion kinetics to target lipid membranes (Wessels et al. 2007). Sindbis virus fusion was not only dependent on the membrane composition (cholesterol levels) but also the fusion kinetics slowed down at low pH. On the other hand, Influenza virus fusion was unaffected under the same range of conditions. This suggests that the role of membrane lipid groups, solution conditions and the fusion protein play a role in regulating the fusion kinetics even at the refolding stage of the fusion protein activity. Floyd et al. also demonstrated that bilayer mixing preceded full pore formation by measuring the elapsed time between hemi-fusion and content release (figure 2e) (Floyd et al. 2008). The single exponential dwell-time distribution of the hemi-fusion state prior to fusion was consistent with a single rate-limiting step controlling the transition from membrane fusion to content release.

Figure 3
figure 3

Nucleic acid interactions with viral proteins. (a) Quenching assay in combination with smPIFE to monitor nucleic acid remodelling by a protein. Increase in inter-dye distance leads to dequenching of dye (placed originally in the vicinity of a quencher molecule) as the polymerase disrupts base-pairing locally (i → ii) and further enhancement of intensity in signal is observed due to fluorescent enhancement induced by the protein (ii → iii). (b) Real-time FRET-based monitoring of a helicase unwinding of a nucleic acid stem-loop structure. A reversible FRET signal decrease is observed when the inter-dye distance increases due to rapid disruption of the stem-loop structure. (c) Protease activity can be similarly monitored on a peptide substrate labelled with a fluorescent protein (FP) FRET pair. The abrupt decrease in FRET efficiency is used to determine the kinetics of protease activity. (d) Labelled protein spontaneously oligomerizing upon binding to nucleic acid can be estimated with tracking the step-wise changes in the intensity of the diffraction-limited spots that is indicative of the number and binding/dissociation of protomers. (e) Scheme for the real-time measurement of polymerization by an RNA polymerase. Formation of the dsRNA by an RNA polymerase can be measured in real time via nucleic acid binding–induced emission (enhancement of a dsRNA selective dye). The three model trajectories (blue traces) represent the activity of individual polymerases on three RNA molecules with slopes corresponding to rate of polymerization. An intermediate (arrow) no-activity regime in a trace might represent a polymerase fall-off or a stalled complex.

Unlike simple fusion triggers, like acidic pH discussed earlier, the HIV-1 envelope (Env) a protein that is organized as a mushroom-shaped trimer of heterodimers of gp120/gp41, requires receptor (CD4) binding that induces an initial conformational change and exposes a co-receptor–binding surface (Blumenthal et al. 2012). Upon further co-receptor (CXCR4 or CCR5) binding and activation of Env, full fusion is achieved. Single-molecule FRET studies demonstrated that native HIV-1 Env trimers can exist in a dynamic equilibrium of three distinct pre-fusion structures representing the ground-state conformation, one that was stabilized by CD4 with a co-receptor mimic and an uncharacterized structural state that was an obligatory intermediate during the activation of Env by CD4 (Munro et al. 2014). This enabled not only the direct validation of the two-step activation by CD4 and co-receptors of gp120 but also the characterization of modulation of these states by neutralizing antibodies and inhibitors. Several questions still remain regarding the catalytic mechanism of HIV-1 Env mediated fusion like the conformational dynamics and stoichiometry of extended gp41 trimers required for fusion that shall be addressed in the future with the development of new labelling strategies using non-canonical amino acids (Sakin et al. 2017).

Membrane composition and heterogeneity also can determine the protein-induced fusion kinetics. For example, cholesterol-rich ‘raft’ domains are known to be vital for insertion of fusion peptide of HIV gp41 and HIV cell entry (Yang et al. 2015, 2016). Using the HIV fusion peptide and pseudotyped HIV binding to artificial membranes, Yang et al. showed that phase-segregated lipid bilayers that displayed co-existing ordered (Lo) and disordered (Ld) phases were more amenable to fusion and the HIV particles interacted and fused preferentially at boundaries between co-existing Lo and Ld lipid phases (Yang et al. 2015). Single-particle tracking of the HIV fusion peptide also displayed interconversion between different diffusive states that further confirm association of different membrane phases and possible peptide structures that might regulate the fusion dynamics (Ott et al. 2013). The exact molecular role of cholesterol or the phase-segregated bilayers in HIV and other envelope virus cell entry still needs to be evaluated.

4 Protein–nucleic acid interactions

Viral nucleic acid-binding proteins like polymerases and helicases form an important class of druggable targets because of their enzymatic nature. There have been continued efforts to develop functional assays to measure their binding, activity and structure–function relationship in the context of their interaction with nucleic acids. A key feature of any successful biological life process is an ability to exclusively and rapidly recognize and bind to its cognate partner to achieve functional outcome. This especially holds true for nucleic acid proteins from the virus, which despite favourable thermodynamics, must outcompete the high copy number host components for successful infection, a recurrent theme discussed here.

There have been several single-molecule fluorescence approaches that have been successfully employed in a variety of molecular systems to probe protein interaction with nucleic acids (Joo et al. 2008; Koh et al. 2016). One simple implementation of single-molecule fluorescence that utilizes the change in photophysical properties of the dye is protein-induced fluorescence enhancement, PIFE (Luo et al. 2007; Hwang and Myong 2014). By measuring the increase in the fluorescence emission upon protein binding in the vicinity of the dye, one can report on binding kinetics, equilibrium or non-equilibrium interactions between protein and the corresponding nucleic acid–binding site (figure 3a). Importantly, in spite of employing a single fluorophore, the 1 to 4 nm distance sensitivity and linearity of the effect on dye–protein separation ensures that local protein dynamics can be probed with high precision (Hwang and Myong 2014). Similarly, dye quenching has been employed to report on nucleic acid structure where proximity of dyes can report on the status of the conformation in real time (figure 3a). Förster resonance energy transfer (FRET) remains a popular single-molecule fluorescence method because of its ratiometric nature and nanometer sensitivity (Roy et al. 2008). In FRET, a donor fluorophore, when in its excited electronic state, can transfer its excitation energy to a nearby acceptor chromophore in a non-radiative fashion through long-range dipole–dipole interactions and efficiency of this energy transfer is mostly linear over 2–8 nm for most popular dye pairs. Placement of the dye pair (or any combination of donor and acceptor) on different parts of the potential nucleoprotein complex, and dynamic changes in the conformation of the molecular system can be measured (figure 3b). Another potential application of smFRET relevant to viral protease activity is monitoring the real-time cleavage activity of target proteins or peptides carrying a dye pair at their two ends (figure 3c). Because of the high signal-to-noise ratio achievable using current microscopy techniques like TIRF-microscopy (TIRFM) and use of photo-stabilizing agents, one can quantify binding of proteins to nucleic acid scaffolds directly (if the binding affinity is high and photo-physics of the dye is unaltered upon binding) as step-wise changes in intensity at the binding site (figure 3d). Finally, more functional assays like polymerization of the nucleic acids by the viral proteins can be captured at the single-molecule level by using dyes specific to a double-stranded form of the product (figure 3e). These single-molecule fluorescence methods still primarily remain in vitro in nature, but several of them (e.g. smFRET) can be adapted for measurements inside live cells.

4.1 Binding and oligomerization of viral proteins on nucleic acid

Specific binding and oligomerization of the HIV-1 Rev protein on the highly conserved rev response element (RRE) of viral mRNA is critical to the activation of nuclear export of unspliced and partially spliced viral RNA (Daly et al. 1993; Mann et al. 1994). To probe the kinetic pathway and assembly intermediates of Rev oligomerization on RRE RNA, single-molecule fluorescence intensity measurements of labelled Rev proteins as they bind to the immobilized (truncated) RRE RNA was measured (Pond et al. 2009). Intensity jumps in the fluorescence time trajectories for single RNA corresponded to step sizes of a single Rev monomer, hence supporting the sequential binding model. This indicated that binding of a single Rev unit allows further assembly of additional monomers in contrast to a pre-formed assembly model in which pre-assembled Rev oligomers bind to the RRE. The high-affinity Rev-binding site in stem-loop IIB of the RRE displayed a maximum of four binding states and its deletion abolished the nucleation of Rev completely on the RRE. In a subsequent study, dwell time distributions of Rev oligomerization on the full-length RRE revealed two kinetic phases for the initial binding step, while dissociation had a single rate-limiting step (Robertson-Anderson et al. 2011). In the presence of human DEAD-box protein 1 (DDX1) helicase, a cellular host cofactor of the HIV-1 Rev, the second kinetic phase attributed to non-productive Rev nucleation events, was removed and it resulted in generation of higher-order Rev–RRE complexes (up to 8 Rev proteins per RRE). In a recent study, they further demonstrated that the DDX1 stimulation of Rev oligomerization was achieved by specific enhancement of the Rev nucleation step on the RRE (Lamichhane et al. 2017). In the light of smFRET experiments that demonstrated that DDX1 and Rev co-occupy the same RRE RNA with low probability, the proposed role of DDX1 is to remodel the RRE RNA conformation to prime it for Rev monomer binding.

The human immunodeficiency virus type 1 (HIV-1) reverse transcriptase (RT) plays an indispensable role in converting the ss-RNA viral genome to dsDNA required for insertion into the host genome. The physiologically characterized HIV-1 RT consists of two polypeptide chains, the 66 kDa (p66) which contains a polymerase and RNaseH domain and 51 kDa (p51) formed from the proteolytic cleavage of the p66 subunit during viral maturation, lacking the RNaseH domain (Starnes et al. 1988; Kohlstaedt et al. 1992). Additionally, the homodimers, p66-66 and p55-55 have been purified and studied, however little was understood about nucleic acid recognition and binding of the homodimers versus the functional heterodimer (Le Grice and Grüninger-Leitch 1990; Maier et al. 1999). Marko et al. used smPIFE to characterize binding of the heterodimeric and homodimeric HIV-1 reverse transcriptase (RT) with its Cy3-labelled primer-template (Cy3-P-T) RNA substrate (Marko et al. 2013). The RTp66-p51 and RTp66-p66 showed 50–100 times higher affinities for the P-T than the RTp51-p51. Moreover, the RTp51–p51complex with Cy3-P-T was found to be less stable showing faster dissociation kinetics than the other complexes. The RTp51 has a weak concentration dependence to dimerize, however dimerizes rapidly upon the addition of a non-nucleoside RT inhibitor (NNRTI), Efavirenz (Venezia et al. 2006). Another significant experimental observation made was that the unbound/bound ratio of the RTp51 at high concentrations reduced drastically once the NNRTI was added to induce homo-dimerization, indicating dimers have a stronger tendency to bind the Cy3-P-T. Hence, the smPIFE assay indicated a new mechanistic possibility that the dimerization of the RT dictates affinities for substrate interaction.

4.2 Viral helicase activity measurements

Single-molecule FRET experiments with HCV NS3 helicase have revealed several interesting features of the viral NS3 helicases (Myong et al. 2007). Upon binding of the HCV NS3 helicase in the presence of ATP to initiate dsDNA unwinding, the FRET changes were monitored for dsDNA modified with a donor–acceptor dye pair at the duplex junction. Earlier, force-assisted optical tweezers-based single-molecule detection of NS3 helicase unwinding yielded a characteristic repetitive pattern, in which the helicase would unwind 3–4 bp rapidly followed by long pauses (Dumont et al. 2006). The smFRET results indicated six distinct plateaus of FRET decrease for unwinding an 18 bp dsDNA. An automatic step-finding algorithm indicated discrete 3 bp unwinding events, which upon dwell-time analysis revealed hidden 1-bp steps coupled to the hydrolysis of a single ATP molecule per step. At the molecular level, this spring-loaded behaviour was explained by modelling an unwinding process in which domain 3 of the HCV NS3 helicase remains fixed, while domains 1 and 2 translocate and tension is built up in the nucleo-protein complex, which is eventually released in a burst of unwinding activity of three nucleotides (Lin and Kim 1999).

A series of smFRET experiments with NPHII, an SF2 family helicase of the Vaccinia virus, provided interesting insights to the mechanism of ATPase-coupled RNA unwinding (Fairman-Williams and Jankowsky 2012). Previous studies on NPHII-mediated unwinding indicated that the helicase hydrolyses multiple ATP molecules before initiating strand separation. The inability to decouple distinct conformations upon unwinding was attributed to high ATP turnover numbers (Jankowsky et al. 2001). Hence, smFRET experiments were performed using a duplex RNA labelled with a proximal donor–acceptor pair in which the unwinding complex was stalled by various ATP analogs to represent intermediate stages of ATP hydrolysis. Distinct FRET populations corresponding to the unbound RNA and two conformations of bound NPHII to RNA were detected when the ADP-BeFx and ADP-AlFx analogs were titrated; however, the relative ratios of the conformations (FRET states) varied across the two stages of the hydrolysis cycle. In contrast, the ADP-bound NPHII seemed to exclusively prefer an intact duplex to a single-stranded RNA, unlike the ADP-bound state of the HCV NS3. Dwell-time analysis from time trajectories of binding to the non-hydrolyzable ATP analogs revealed multi-phase kinetics of interconversion between the two ssRNA-bound conformational states, indicating a dynamic equilibrium among these NPHII conformations.

Despite the array of studies focusing on the kinetics and mechanism of unwinding by viral helicases, and the emerging role of the NS3 in viral packaging and its translocation, behaviour of helicase on long nucleic acid substrates has not been examined (Dumont et al. 2006; Myong et al. 2007; Ma et al. 2008; Gu and Rice 2010). In a recent study, a hybrid scheme coupling single-molecule fluorescence (TIRFM) with optical tweezers for sub-pixel localization of the helicase in motion on nucleic acids revealed a hitherto unknown interaction phenomenon (Lin et al. 2017). The long nucleic acids, maintained as stretched tracks on which fluorescent NS3 helicase molecules can translocate, undergo gradual shortening followed by a force jump and relaxation. This property termed as repetitive translocation was observed exclusively in NS3h–ssDNA interactions, and the shortening rate decreased monotonically with an increase in ssDNA extension force. Furthermore, in an smFRET assay using a partial duplex DNA with a Cy3 donor at the 5’ terminus of the longer strand and Cy5 acceptor at the 3’ terminus of the shorter strand, the addition of NS3h led to gradual increase in FRET efficiency followed by an abrupt drop. This phenomenon indicated ‘repetitive looping’ in the DNA, wherein the donor and acceptor strands were periodically brought into proximity over cycles of translocation. A comparison between the translocation rates of NS3 helicase on an ssRNA substrate versus a dsRNA substrate revealed a threefold faster translocation in the former case. This estimate corroborated previous ensemble fluorescence stopped-flow experiments tracking translocation rates of NS3h on polyU and polydU substrates (Khaki et al. 2010). This difference can be partially explained by the interaction between the amino acids and the sugar moieties of the nucleic acids in differing conformations, and by the compact structure of ssRNA – a property which decreases the affinity of NS3h.

4.3 Dynamics of HIV reverse transcriptase on nucleic acid substrates

Binding to a specific site on the nucleic acid site involves weak binding followed by local search along the length of the polymer, and final reorganization with exploration of multiple alternative configurations of the protein contacts to achieve binding in the ‘correct orientation’ (Halford and Marko 2004; Blainey et al. 2009). In a study conducted by Ganji et al. the mechanism of this final reorganization (protein flipping) of the HIV-1 RT on a 19-bp double-primed dsDNA (dpdsDNA) was explored by smFRET. Such protein flipping is a fast event in which the protein moves along a vector perpendicular to the axis of the duplex and specific contacts between the polymerase and nucleic acid duplex are temporarily disrupted and reformed. The RT was observed to bind the symmetric dp duplex DNA in two distinct conformations (exhibiting high and low FRET values of 1 and 0.3 respectively), spending roughly equal times in both conformations but flipping between the two binding states. The dwell time as well as number of binding events of the RT to dpdsDNA was higher at low salt concentrations of 50 nM; at physiologically relevant salt concentrations (150–200 nM) the koff increased and kon decreased. Crowding molecules like PEG 8000, which compact the RT–DNA complex, were found to entropically stabilize the binding. By modelling the effect of macromolecule crowders using the scaled particle theory (Zhou et al. 2008), it was shown that at physiological levels of crowding, sub-nanomolar affinities can be achieved for the RT–DNA interactions. Short-range interactions between incoming cognate nucleotides to the RT–DNA complex were found to decrease koff without affecting the Pflip, that is, the likelihood of a flipping event versus dissociation from nucleic acid. Using simulations, they demonstrated that a hopping model could describe the long-range interactions under physiological salt and crowding effects to facilitate the RT from rebinding its DNA template. The exploration of multiple configurations between two macromolecules is posited to kinetically aid assembly and function.

A similarly detailed understanding of nucleic acid substrate recognition and polymerization activities of HIV-1 RT was obtained through smFRET-based alternating laser excitation (ALEX) experiments (Liu et al. 2008). The HIV-1 RT was labelled with the Cy3 FRET donor on either the RNaseH or the finger domain in the p66 subunit and RNA-DNA hybrids of different lengths were labelled with the acceptor Cy5 dye. FRET histograms with longer hybrid (38 bp) substrates revealed two distinct binding modes for the interaction: a ‘polymerization-competent’ mode with the polymerase active site located at the front-end of the hybrid terminus and a ‘polymerization-incompetent’ mode in which the RNaseH domain is positioned at the back-end of the terminus, based on crystal structures (Jacobo-Molina et al. 1993; Sarafianos et al. 2001). The distinct FRET values were indicative of the sliding of the HIV-1 RT on the substrate, as the separation of the FRET peaks increases with the length of the RNA–DNA hybrid. Analysis of the FRET signal from a single binding event suggested that the HIV-1 RT shuttles between two ends of the hybrid, a thermally driven diffusion phenomenon. Sliding observed on DNA/DNA duplexes showed higher rate constants for escaping the back-end of the hybrid and lower FRET values accounting for the larger inter-base distance of DNA. The establishment of a kinetic model for sliding of the HIV-1 RT provided significant insights. For example, addition of the initiating nucleotide, dGTP, was found to stabilize the complex of the front-end bound hybrid duplex with HIV-1 RT. In contrast, the NNRTI nevirapine was found to kinetically increase kfront→back by loosening the ‘clamp’ of the fingers and thumb domains of the RT at the front-end of the hybrid duplex (Huang et al. 1998; Quan et al. 1998). On long DNA–RNA hybrid tracts, the HIV-RT was found to bind to DNA adjacent to the polymerization site and slide towards the primer terminus, often flipping at the terminus, to achieve the polymerization-competent orientation. These experiments demonstrated that the HIV-1 RT does not adopt a purely one-dimensional search for the polymerization start site, and instead uses sliding and flipping to improve its efficiency.

4.4 Understanding structure–function relationship in influenza replication

Influenza A virus has an eight-segment negative-sense single-stranded RNA genome (denoted as vRNA) with complementary regions at the two ends of the genome. The base-pairing between two ends of the RNA genome generates a promoter with a unique structure, a phenomenon similarly proposed for flaviviruses, that enables its RNA-dependent RNA polymerase (RdRP) to recognize the promoter (Robertson 1979; Desselberger et al. 1980). In the absence of structural details of the RdRP-bound dsRNA promoter of the influenza A virus, Tomescu et al. resorted to single-molecule FRET assay using ALEX to map the structural changes induced in the promoter and dynamics of the interaction with high sensitivity (Kapanidis et al. 2005; Tomescu et al. 2014). In the first experiment, a synthetic dsRNA promoter was labelled with donor Cy3 at position U18 on the 5′ strand with the acceptor ATTO647N dye at position U4 on the 3′ strand of dsRNA (figure 4a and 4b). The uncorrected mean FRET efficiency (E*) of the wild-type dsRNA promoter in solution of 0.57 changed on the titration of RdRP with the promoter. The distribution of the RdRP-bound dsRNA promoter complex became bimodal, where the polymerase-bound promoter population had an E* value of 0.79 (figure 4a). A second smFRET assay was designed to decouple changes between the proximal (residues 1–9) and distal (residues 11–18) promoter changes induced by polymerase binding. In this quenchable smFRET (quFRET) assay (Cordes et al. 2010), the proximity of the Cy3 donor dye on residue 3 at the 5′ end and the ATTO647N acceptor dye on U4 at the 3′ end resulted in a quenched low FRET state of the wild-type promoter. The quenching was significantly reversed to a high value of observed FRET (E* ~0.84) on titration with RdRP, indicative of opening and rearrangement of the double-helical dsRNA promoter (figure 4b and 4d). To study the structural changes at the distal end, a dsRNA promoter with Cy3 at position 18 on the 5′ end and the ATTO647N dye at position 13 on the 3′ end gave a mean FRET value of E* ~0.82, irrespective of RdRP binding, indicating no major structural alteration upon polymerase binding. Structural details were further refined by correcting the E* values for factors such as background, cross talk and γ-factor effects (to compensate differences in quantum yield and detection efficiency of the fluorophores), to estimate the ‘true’ donor–acceptor distances. The FRET-restrained positioning and screening, FPS (Kalinin et al. 2012) algorithm helped in modelling dyes on the ab initio 3D models of the dsRNA promoter in its free and RdRP complexed forms, and was found to be consistent with nuclear magnetic resonance (NMR) structures (figure 4c and 4e). The 3D model generated using corrected FRET efficiencies for the polymerase-bound promoter generated a corkscrew topology as proposed previously (Flick et al. 1996). Another key insight from the solution-based assays suggested a dynamic equilibrium between the double-helical and corkscrew conformations. The presence of the corkscrew conformation has been considered significant based on the fact that the hairpin loops are essential for the endonuclease activity of RNAP (Leahy et al. 2001a; b) and the polymerase–corkscrew promoter complex is conformationally stable (Brownlee and Sharps 2002). In a similar follow-up study, the promoter was processively unwound during de novo replication rather than displaying melting in a single step (Robb et al. 2016). The outcome has been a new model for the influenza A replication in which the switch to initiation from its pre-initiation stage is achieved by translocation of the 3’ vRNA through the active site, hence destabilizing the processive unwinding the dsRNA promoter.

Figure 4
figure 4

Adapted with permission from The National Academy of Sciences of the USA.

The influenza vRNA promoter adopts a corkscrew structure upon binding by the Influenza A polymerase (Tomescu et al. 2014). (a) Bimodality in the smFRET distribution upon titrating the vRNA promoter (top panel) with influenza RdRP indicates structural rearrangements in the promoter. (b) Representation of the duplex panhandle conformation of the vRNA and alternative labelling sites of the donor (green) and acceptor (red) dyes used to infer distances based on smFRET. (c) Three-dimensional (3D) model of the native duplex panhandle vRNA promoter with inter-dye distances indicated. (d) Two-dimensional schematic of the expected corkscrew vRNA structure adopted when RdRP opens and restructures the promoter. (e) 3D model of the RdRP-bound influenza promoter in the corkscrew configuration.

4.5 Kinetics of viral IRES-mediated translation

Translation is a multi-step stochastic process and has been difficult to kinetically understand due to the lack of control over synchronization of ribosomes. Eukaryotic translation has been even more evasive to study by fluorescence due to the complex interactions between the eukaryotic 80S ribosome and its various associated translation factors, the need to incorporate fluorescent labels while minimally perturbing the natural translation system. Sensitive assays for quantifying translation would be informative especially in virology since viruses have evolved multiple strategies for hijacking the host translation machinery. An example of such a strategy is the internal initiation using conserved Internal Ribosome Entry Site (IRES) on viral RNA competent for translation in the host (Firth and Brierley 2012). Bugaud et al. tracked how a single mammalian ribosome elongates using the viral IRES, a complex translation initiation sequence (Bugaud et al. 2017). Surface-immobilized mRNA (imaged by TIRF microscopy) with either the cricket paralysis virus (CrPV) or the HCV IRES were annealed with two sets of spectrally distinct fluorescent RNA probes at fixed locations from the IRES to report on the translation status (figure 5a). Since the ribosomal entry channel is unable to accommodate double-stranded RNA, the helicase action of the 80S ribosome ensures that the annealed fluorescent probes are unwound as the ribosome translates across the RNA. The placement of the two primers such that one reports on five elongation cycles (+5) and other on nine elongation cycles (+9) post initiation enabled a distinction between initial translation kinetics and elongation kinetics (figure 5b and 5c). A single elongation cycle for CrPV IRES with purified pre-incubated ribosomes was determined to be ~ 1.4 ± 0.2 s, which was slower in comparison with the cycle time of 0.2 s per codon determined by ribosomal profiling (Ingolia et al. 2012). An interesting contrast was observed for HCV IRES-mediated translation, which showed a relatively higher efficiency with free ribosomes iterating its dependence on eukaryotic factors like IF2 and IF3 (Borman et al. 1995). Furthermore, the distribution of departure times for the (+5) primer could be only fitted to a gamma distribution function with two time parameters indicating different kinetics for intermediate steps. In contrast to studies performed with heterogeneous cell-free translation systems (Zhang et al. 2016) where the first four elongation steps were found to be slow (~80–200 s), CrPV IRES-mediated initiation kinetics was found to be slow in the first two steps (~40 s) but proceeded more rapidly (~1.4 s) in subsequent translocation steps. The molecular basis for the appearance of the slower steps may be rationalized by the complex interaction between the ribosome, factor eEF2 and the PKI domain of the CrPV IRES, as studied by cryo-EM (Fernández et al. 2014; Murray et al. 2016). The ribosome assembles upstream the IRES with an offset of 1 codon, with PKI blocking its A-site. The factor eEF2 interacts with PKI and causes a translocation of the ribosome to correct the offset, the rate-limiting step of translation initiation. Though the A-site is free to accept incoming tRNA, the IRES is shifted to occupy the P- and E-sites, possibly resulting in a second kinetically slow translocation. Once the ribosome translocates beyond the complex IRES, the rest of the elongation steps proceed at a rapid rate. The ribosome assembled on an HCV IRES does not bear an offset (Filbin et al. 2012); however, the interactions between the ribosome and the IRES remain like that of CrPV IRES. In addition, the HCV IRES is dependent on multiple initiation factors. A hallmark feature of the IRES-dependent translation as established by these studies is that the first few elongation cycles are rate limiting. This establishes a useful tool to understand translational kinetics and its implications in defining viral fitness in the host.

Figure 5
figure 5

Adapted with the permission of the RNA Society.

Single ribosome viral IRES translation assay (Bugaud et al. 2017). (a) A ribosome-bound mRNA is immobilized onto a PEG-neutravidin coverslip through a biotinylated probe complementary to the 5′ end of the mRNA. Each immobilized mRNA is visualized by two hybridized probes, the ATTO647 N UP and ATTO565 DOWN primers. (b) Design of the mRNA and fluorescent probes: A-site with the IRES initiator codon is indicated in red; fluorescent probes have three non-complementary nucleotides to act as a spacer between the dye and the mRNA. (c) TIRF microscopy images to visualize the co-localization (yellow) of the UP(+5) and DOWN(+14) probes hybridized to each mRNA at the start of the experiment. As translation proceeds, the time difference in probe detachment due to helicase activity of the ribosome can be monitored to estimate translation kinetics.

4.6 Host–protein interactions with viral RNA in immune response

Single-molecule fluorescence methods have also found use in probing host factor–viral RNA interactions, a fundamental theme in host immunity to viruses. Viral RNA is initially recognized by pattern recognition receptors (PRRs), and PRRs then induce type I interferons (IFNs) and other pro-inflammatory cytokines, but mechanistic understanding of how viral RNA are discriminated is poor. The Retinoic acid-inducible gene I (RIG-I) is a cytosolic host PRR that provides immunity against many negative-strand RNA viruses by sensing viral RNA (Gack 2014). RIG-I activation occurs only in the presence of pathogenic RNA, despite the ability of RIG-I to bind endogenous RNA in the cytoplasm meriting the question how RIG-I discriminates self- and non-self-RNA. Translocation of the central RNA helicase domain of the RIG-I protein was observed on a dsRNA of a non-viral origin using smPIFE (Myong et al. 2009). This truncated RIG-I showed a steady binding to the dsRNA, which changed to periodic fluctuations in fluorescence once ATP was introduced to the protein–RNA complex indicative of repetitive translocation along the length of the dsRNA without unwinding it. A marked dsRNA length and ATP concentration dependence was observed for the translocation of truncated RIG-I, whereas wild-type RIG-I showed low-frequency translocations only. In a splice variant of RIG-I that is incapable of generating antiviral response in the host due to a deficient CARD domain (Gack et al. 2008), ATP-dependent translocation on dsRNA resembled that of the truncated RIG-I variant, which was deficient of CARDS. RIG-I displayed larger dwell times on double-stranded RNA/DNA than on ssRNA substrates with 5’triphosphate, which is responsible for activating the ATPase domain of RIG-1 and could be a verification signal for dsRNA to suggest its viral nature. Such a mode of translocation might also assist in removal of RIG-I from low-affinity RNA ligand sites.

5 Viral assembly, packaging and architecture

Viruses must assemble its proteinaceous coat from viral structural proteins, package its genome in it and acquire a lipid membrane encompassing it (if an enveloped virus) in the cytoplasm of the infected cell to generate the virion particles for subsequent infection rounds (Sun et al. 2010; Stockley et al. 2013; Perlmutter and Hagan 2014; Lakdawala et al.2016). This elaborate and extensive process is orchestrated at different locations, time points and through numerous heterogeneous intermediates. Broadly, the virus assembly and packaging progresses through two major pathways. Many single-stranded viruses employ electrostatics and ‘packaging’ sequences present in their genome to bind capsid (coat) protein and use this nucleation to drive the assembly process while simultaneously packing the genome. On the other hand, when the charge densities and bending flexibility of the nucleic acid set limits on spontaneous assembly, an empty coat shell is pre-assembled and the genome is packaged with the help of devoted nucleic acid motor proteins.

Several single-molecule approaches exist to ‘visualize’ assembly and packaging of viruses which are reviewed here. While ATP-driven, motor-based packaging of double-stranded nucleic acids into viral procapsid has been studied in detail using single-molecule optical tweezer experiments (Smith 2011), we have limited our discussion to only fluorescence methods for brevity. One fluorescence technique that is sensitive to growth of capsid and that can report on genome packaging without getting overwhelmed by the diversity of intermediates is fluorescence correlation spectroscopy (FCS). FCS relies on detecting mathematical ‘likeness’ (correlation) in the time scales of fluorescence fluctuations of freely diffusing molecules across a illumination volume (Maiti et al. 1997; Ries and Schwille 2012). Since, the time scale of decay for the correlation function is dependent on the diffusion coefficient (which scales with the approximate size) of the particle, virus assembly growth kinetics can be measured in solution (figure 6a). Fluorescently tagged protein, when allowed to assemble in vitro or in cells, display increased intensities of particles over time that can be used to track assembly sites and kinetics in real time (figure 6b), though separating single-particle intensities unambiguously becomes challenging in most cases when number of particles per diffraction limited spot goes beyond one. If a singly labelled protein component assembles and can be spatially separated, one can allow irreversible photo-bleaching of the dyes to estimate the stoichiometry of the assembled complex by counting the number of observed steps in the intensity trace (figure 6c). To determine the identity or number of packaged nucleic acid in the virus, one can use fluorescently labelled complementary oligos (smFISH) that can be hybridized to the nucleic acid in the virion to ‘image’ the genome (figure 6d).

Figure 6
figure 6

Understanding virus assembly and packaging with fluorescence correlation spectroscopy (FCS), single-molecule fluorescence in situ hybridization (smFISH), single-particle photo-bleaching analysis of capsid–RNA complexes. (a) Representative auto-correlation functions from fluorescence fluctuations captured in a confocal volume of diffusing molecules can monitor virion assembly. Changes in the hydrodynamic radius of the capsid or the RNA molecule(s) binding is estimated from changes in the diffusion coefficient of the single particles manifesting in decreasing τ½ values. (b) Photo-bleaching analysis (in a diffraction-limited spot) of the capsid–nucleic acid complex quantifies the number of labelled capsid protein units enabling examination of the assembly intermediates. The step-wise changes in intensity levels can be directly correlated to the number of labels. (c) Fluorescence intensity histograms from single spots depicting time-dependent assembly of capsid protein can report on viral assembly kinetics. As more labelled capsid proteins come together, the intensity of each diffraction-limited spot increases with time. (d) Strategy to infer genomic packaging of viruses by smFISH. Fluorophore-labelled FISH probes are designed against different segments of the viral genome and co-localization of probes with virus coat proteins and/or other genome segments allows examination of genome packaging efficiency.

Finally, the ‘super-resolution’ avatars of single-molecule imaging like stochastic optical reconstruction microscopy (STORM), photoactivated localization microscopy (PALM) and fluorescence photoactivation localization microscopy (FPALM) are providing new insights regarding virus architecture and its interaction with host cell components. These methods take advantage of image reconstruction from the particle location estimates recovered from mathematical fitting of the point-spread-function of single molecules when induced to stochastically turn-on or undergo photo-blinking (Betzig et al. 2006; Hess et al. 2006, ‘Method of the Year 2008’ Nature Methods). We recommend the readers to comprehensive reviews that already cover the developments in this field (Gould and Hess 2008; Ji et al. 2008; Hell et al. 2009; Huang et al. 2009; Heilemann 2010; Patterson et al. 2010). Compared to electron microscopy, which can reveal exquisite structural detail (at <10 nm), fluorescence super-resolution methods (with lateral resolution of ~ 10–40 nm) allow access to imaging of multiple species (colour) and dynamical information in physiologically relevant solution conditions. Hence, these methods are already finding applications in understanding of virus architecture, its spatial distribution and organization and its interaction with cellular components that we expect will grow with development of new tools (Müller and Heilemann 2013; Roy 2013; Gray et al. 2016) and the information is expected to be significantly enriched in the coming years.

5.1 Super-resolution microscopy of viruses

Using photoswitchable fluorescent proteins or photo-blinking dyes tagged to genetic protein fusion tags, various aspects of HIV architecture, assembly and membrane interactions have been extensively studied with super-resolution microscopy (Müller and Heilemann 2013). For example, HIV virion assembly that is coordinated by the virus structural polyprotein Gag on the plasma membrane of the host cell has been studied using Gag’s ability to self-assemble into virus-like particles (Betzig et al. 2006; Manley et al. 2008). Initial stages of Gag assembly was demonstrated to be fast (within 10 min) but proceeded in two phases as a significant (~ 40%) fraction of the spontaneously forming Gag clusters would be limited to low numbers of Gag proteins (<5% compared to numbers in immature virions) suggesting kinetic barriers to formation of HIV-Gag virions (Ivanchenko et al. 2009; Gunzenhäuser et al. 2012). Recruitment of HIV Env protein to the Gag clusters could be visualized by co-localization in two-colour super-resolution imaging (Muranyi et al. 2013; Roy 2013). Env accumulation was shown to be dependent on direct interaction with the Gag protein, yet it localized to the periphery of the Gag clusters and only low levels of Env were directly associated with the nascent Gag clusters, indicating viral protein-induced restructuring of the membrane composition to assist in virus assembly. Similarly, assembling virus interaction with host cellular proteins can be measured using multi-colour super-resolution imaging. HIV-1 relies on the cellular endosomal sorting complex required for transport (ESCRT) for its final budding step during cell egress. It was previously suggested that ESCRT proteins induced membrane fission from around or below the nascent budding virus. Using super-resolution microscopy, two groups independently co-localized the ESCRT proteins with Gag protein clusters on the membrane (Van Engelenburg et al. 2014; Prescher et al. 2015). Both studies demonstrated that when ESCRT proteins did co-localize with the Gag, ESCRT protein clusters were significantly narrower than the HIV-1 bud, suggesting that ESCRT proteins promoted membrane scission inside a narrow structure within the developing bud and did not act from outside. Tetherin, a GPI-anchor membrane cellular restriction factor that is known to inhibit the release of enveloped viruses, was shown to co-localize to HIV assembly sites (Lehmann et al. 2011). Tetherin clustering was mediated by the transmembrane domain interactions but was not associated with lipid raft domains unlike popular belief, and it likely tethered the developing HIV-1 virions directly to the plasma membrane, suggesting the mechanistic basis for tetherin inhibition of viral action. Stimulated emission depletion (STED)-based super-resolution microscopy has also aided in understanding of the HIV maturation process. Using dual-colour STED, Stefan Hell’s group demonstrated maturation of the HIV-1 virion was associated with rearrangement and clustering of Env proteins close to the cluster of receptor molecules (Chojnacki et al. 2012). This has led to the emergence of the idea that Gag maturation signalling primes the onset of events necessary for cell entry and post-entry events. Using photo-cleavage of an HIV maturation inhibitor, time-resolved induction of Gag polyprotein cleavage and virion protein relocalization was studied in a synchronized fashion allowing direct visualization of the virus maturation at the sub-viral level over the course of tens of minutes (Hanne et al. 2016). We predict that such sub-viral resolution afforded by fluorescence super-resolution techniques will be employed increasingly to understand virus processes in the future.

5.2 Single-molecule FISH to study viral RNA lifecycle and genome packaging

Using single-molecule RNA (sm)FISH against Influenza genome, it was shown that virus packages its segmented genome at one copy per RNA segment per viral particle (Chou et al. 2012). The mechanism for such precise single-copy packaging of each segment was further revealed to be orchestrated in the cytoplasm itself. When the viral genome segments were observed with two-colour single-molecule FISH, co-localized signals for genome segments revealed that the viral RNA was transported collectively to the nucleus post infection (Chou et al. 2013). The newly replicated viral RNA would then be distributed spatially in the cell cytoplasm but would start to reassemble with its counterparts in Rab11-enriched recycling endosomes. On the other hand, smFISH with the Rift valley Fever virus, a bunya virus that carries a tripartite RNA genome showed the absence of one or more genomic segments in the virus particles, indicating that RVFV packaging was a non-selective process (Wichgers Schreur and Kortekaas 2016). Interestingly, RVFV RNA imaged in the cells post infection showed how the virus would start replicating locally at the site of infection, then spread to cytoplasm, before being localized at the Golgi with the help of Gn glycoprotein.

Similar studies in combination with immuno-fluorescence employed in case of LCMV has also revealed how genome replication and pre-assembly would take place in Rab5c early endosomes (King et al. 2017). In HCV cell infection model, smFISH against the positive and negative strands of the viral RNA and their simultaneous imaging with ribosomes and viral proteins highlighted the dynamics of HCV lifecycle (Shulla and Randall 2015). Positive strands of the viral RNA were initially associated with the ribosomes but over time co-localized with virus ‘replication factories’ and finally virus assembly highlighting the spatio-temporal regulation of the viral RNA. Extending this approach, Ramanan et al. examined the effect of various antivirals on HCV viral dynamics (Ramanan et al. 2016). HCV antivirals displayed strand-specific decay kinetics suggesting differences in underlying mechanism of these drugs. For example, daclatasvir (DCV) treatment led to early dramatic drops in negative strand of vRNA, while IFN treatment produced a much weaker and variable response per cell. Intriguingly, IFN induction of interferon-stimulated genes (ISGs) that are known to play a role in suppression of vRNA was not only highly variable among cells, but also their RNA levels were positively correlated with viral RNA levels. Such quantitative single genome measurements in single cells allow an unprecedented level of accuracy that combined with kinetic modelling of virus lifecycles can provide crucial insights into virus–host cell interactions.

5.3 Virus assembly in vitro

Virus particles can be assembled in vitro from the coat proteins with and without the genome in optimized solution salt conditions (Zlotnick and Mukhopadhyay 2011; Bush and Vogt 2014). This, combined with lack of specific sequences shown to be associated with virus assembly, has led to the widely held belief that viral coat assembly is largely driven by electrostatics. However, most of such in vitro assemblies must rely on high protein concentrations (~ 1–10 μM), require long incubations and sample processing and purification to observe the assembled structures. In the cells, viral genome is able to compete with high concentrations of cellular RNA and is present exclusively in the mature virions (Routh et al. 2012). This has led to the search for virus assembly packaging principles that can explain this crucial aspect of its lifecycle. Since the virus assembly in vivo and in the initial stages occurs with sub-micromolar concentrations of coat proteins, Peter Stockley’s group adapted FCS to monitor assembly of virus particles in real time (Borodavka et al. 2012). FCS reports on the diffusion coefficient of the tagged molecule (either the RNA or the coat protein) which can be employed to determine the hydrodynamic radii (Rh) of the growing virus particle with time. Using two distinct ssRNA viruses, they showed that addition of capsid proteins to the labelled RNA would result in a rapid and dramatic condensation of the RNA (drop of ~ 20–30% of Rh). The capsid–RNA complex would then grow to resemble the size of fully formed viruses. Non-viral RNA would not display any such changes and produced only aberrant structures, suggesting that specific interaction between viral RNA and capsid was important for virus assembly. Interestingly, the RNA collapse also required inter-protein interactions, indicating that capsid assembly was being nucleated. Based on this, they proposed a two-stage mechanism for virus assembly: (i) a rapid and cooperative capsid binding at multiple ‘high’ nanomolar affinity locations on the RNA and compaction of the RNA by capsid–capsid interactions and (ii) addition of capsid proteins to this nucleoprotein complex. In a more recent study, they attempted to ‘redesign’ the assembly process using mutations of the RNA sites involved in capsid binding, that is, ‘packaging signals (PS)’ on the viral genome (Patel et al. 2017). Virion formation was compromised when all such capsid recognition motifs were removed from the viral RNA. Working with a smaller RNA region and introducing synthetic motifs that bind tightly to the capsid, they could restore assembly of the PS-less RNA or generate more efficient assembling genomes that could outcompete the wild-type genome. Such increased understanding of virus assembly can help in design and development of better vaccines, gene therapy carriers and aid in search of antivirals that target the viral assembly process.

6 Challenges and outlook for the future

In spite of its potential for revealing fresh insights into biomolecular systems, single-molecule methods have found limited adoption in virology and even in the biology scientific community. Apart from costs of initial investment on setups, the complexities in developing single-molecule assays requires a broad understanding of issues with sample preparation, limitations of the methods and challenges in data interpretation. Availability of integrated optical microscope systems and possible service-based business models that have helped genomics and proteomics approaches could alleviate some of the issues. Other technical issues are addressable but working closely or visiting operating labs in the field can help alleviate initial hiccups and learning-curve bottlenecks.

We expect several other technical innovations will drive progress in application of single-molecule fluorescence in virology and other areas. One of the common challenges in conducting single-molecule fluorescence experiments still remains to be shortage of fluorescently tagged biological reagents. Due to the demand for high photon flux, high photostability, genetic and/or orthogonal bioconjugation, single-molecule fluorescence probes are far and few and further development in this area will fuel increase in single-molecule applications. Currently, single-protein labelling with Cyanine, ATTO and Alexa family of organic dyes remain popular. Genetic tagging with fluorescent proteins like EYFP and RFP is employed though they suffer from lower photon budget and controlling their expression levels to get optimal density of molecules for single-molecule imaging is challenging. Photoswitchable and photoactivable proteins like mEOS2 and PA-GFP provide a better handle on managing the number of observable molecules due to control over the activation process. Self-labelling protein tags like SNAP-tag and Halo-tag provide an opportunity to use the photon budget of dyes with genetic encoding (Keppler et al. 2003; Los et al. 2008). Similarly, improved methods that label the viral nucleic acid at high signal-to-noise ratio with help of nucleic acid tags such as nuclease-resistant molecular beacons, GFP fusion proteins that bind to RNA/DNA secondary structure and/or multiply labelled tetravalent RNA imaging probes (MTRIPs) will help enhance virus tracking and cell-entry (Sivaraman et al. 2011; Alonas et al. 2016). Apart from development of new fluorophores, ways to conjugate smaller variants of Q-dots and other promising candidates like nitrogen-vacancy centres in diamond can provide photostable and high-contrast alternatives to dyes, especially for rapid tracking and long-term imaging. Convenience of introducing these tags also require plasmid-based cDNA infectious clones to enable straightforward molecular biology and purification, high-throughput ways of measuring activity and viability after tagging, and convenient methods for delivery of tagged molecules to cells.

However, advances in illumination and imaging schemes that provide advantages of enhanced signal-to-noise ratio, higher imaging speeds with wide-field imaging and better optical-sectioning for 3D imaging are already demonstrating promise (Gebhardt et al. 2013; Chen et al. 2014; Greiss et al. 2016). Parallel development of analytical tools and algorithms, tracking of fluorescent particles, stoichiometry analysis of molecular complexes and recovery of physical properties like diffusional and structural dynamics are expected to percolate and become ‘standards’ for single-molecule data analysis pipelines (McKinney et al. 2006; Greenfeld et al. 2012; Persson et al. 2013; Coltharp et al. 2014; Gray et al. 2016).

Another front for further development will be a combination of single-molecule fluorescence imaging with other non-fluorescence-based imaging and spectroscopy platforms that can enhance the informational output of the single-molecule assays. In vitro mechanical manipulation methods like optical and magnetic tweezers, atomic force microscopy (Neuman and Nagy 2008) and acoustic force spectroscopy (Kamsma et al. 2016) that can apply pN to nN range of forces on single molecules have already been combined with fluorescence detection (Zhou et al. 2010; Kemmerich et al. 2016). Apart from aiding in understanding of virus–receptor interactions (Alsteens et al. 2016) and mechano-chemical coupling of viral enzymes (Dumont et al. 2006; Cheng et al. 2011; Lin and Ha 2017), these methods can allow mapping of the cellular mechanics upon infection (Grashoff et al. 2010). Other high-throughput global analysis approaches like genomics and proteomics, when coupled to single-molecule methods, are finding use in evaluating the cell-to-cell heterogeneity and can become powerful tools in addressing stochasticity in viral infection and host-response (Lubeck and Cai 2012; Lovatt et al. 2014; Lee et al. 2015; Moffitt et al. 2016).

Single-molecule experiments have also remained largely in vitro in nature with studies primarily employing recombinant and truncated versions of the viral components to enable easy and ‘correct’ interpretation of data. Generation of single-molecule experiment capable of constructs using genetic approaches (Nelles et al. 2016; Khan et al. 2017), ability to visualize single molecules deep in animal tissue (Chen et al. 2014; Greiss et al. 2016; Shah et al. 2016), automation and increased user-friendliness of single molecule experiments (necessary for adaptation to enhanced biosafety level facilities) can increase the impact of single-molecule-based methods in virology and other infectious diseases.