Observation of HIV-1 particle assembly using live-cell microscopy requires the attachment of a suitable label to a component of the virus particle. The fact that Gag is not only the main orchestrator of the HIV-1 assembly process, but also the most abundant component of the virion, renders it an attractive target for labeling approaches. The most widely used labeling strategy for live-cell applications, which has also been used to investigate HIV-1 assembly, is fusion to autofluorescent proteins (FPs). FPs comprise a color palette ranging from blue to the far red as well as photoactivatable and photoswitchable derivatives suitable for pulse-chase analyses or sub-diffraction fluorescence microscopy [3,4,5,6]. In contrast to chemical labeling with synthetic dyes, employment of genetically encoded FPs results in attachment of the label at a well-defined position and avoids staining steps that may be difficult to control. Hence, the effect of the label on the functionality of the virus is reproducible and can be characterized. However, one should be aware that, depending on the fusion context, a significant fraction of FPs may be non-fluorescent due to misfolding of the protein, immature chromophores, or chromophores in a dark state. Since approximately 2400 Gag molecules are incorporated into a single HIV-1 particle [7,8], labeling with FPs still permits sensitive detection of individual assembly sites even when spiking the virions with non-fluorescently labeled Gag molecules. Codon-optimized versions of HIV-1 Gag, which can be overexpressed in the absence of other HIV-1 proteins and are tagged at the C-terminus by fusion to FPs, have been described [9]. FPs have also been inserted between the N-terminal matrix (MA) domain and capsid (CA) domain of the Gag polyprotein, allowing labeling of Gag in the context of the complete viral genome [10,11]. While FP labeled virus derivatives have been instrumental in studying kinetics of Gag assembly at the plasma membrane (see e.g., Section 3.2), one of the disadvantages of FPs is their size of ~27 kDa, resulting in reduced replication efficiency of the modified virus under most conditions. The requirement for chemical maturation of the fluorophore also limits detection of the earliest steps in trafficking of newly synthesized proteins. An alternative labeling approach is the small tetracysteine-tag (TC-tag), which comprises only 6–12 amino acids and can be stained using biarsenical dyes [12]. A drawback of this method is the limited selection of cognate dyes currently available and the requirement for an additional chemical labeling step, which can result in background staining. More specific labeling is usually achieved with larger protein based tags that can react intracellularly with an added fluorescently labeled ligand. An example of this approach is the SNAP-Tag, a 20 kDa variant of the human O6-alkylguanosyltransferase (hAGT), which undergoes covalent self-labeling with various benzylguanine (BG) derivatives [13]. Both TC-tag [12] and SNAP-tag [13,14] have been employed to generate labeled Gag derivatives, expressed either alone or in the viral context, but use of these derivatives in live-cell imaging analyses has so far been limited. The large variety of FPs with different spectral properties makes it possible to perform multi-color experiments, which allow a combined detection of labeled Gag with other fluorescently-labeled virion components or cellular factors involved in the assembly process. Surrogates of the HIV-1 genomic RNA can be labeled for live-cell imaging by including sequence motifs specifically recognized by RNA binding proteins such as the bacteriophage MS2 coat protein. FP-tagged versions of the RNA binding protein attach to the cognate RNA motif, thereby labeling the RNA in live cells [15,16]. This approach opened the possibility for quantitative analyses of Gag-RNA interactions by fluorescence imaging [17].

Once viral and/or cellular components have been fluorescently labeled, a number of fluorescence microscopy approaches can be used to investigate the dynamics of protein complexes or single viruses during assembly. Image based techniques like Total-internal Reflection Fluorescence (TIRF) microscopy, Spinning-disc Confocal Microscopy (SDCM) or super-resolution fluorescence microscopy exploit the spatial distribution of the fluorescence intensity to extract the position and/or brightness of a fluorescence object. In contrast, Förster Resonance Energy Transfer (FRET) uses the spectral or fluorescence lifetime information at a single position or per pixel to gain information on the conformation or interaction of biomolecules. Fluorescence fluctuation spectroscopy (FFS) methods analyze the temporal changes of the fluorescence intensity to gather information over the number of particles in the observation volume, their brightness and mobility. Interactions between proteins can be investigated using cross-correlation methods. Within this classification, Raster Image Correlation Spectroscopy (RICS) plays a special role as it analyzes both the spatial and temporal signal fluctuations within a raster scanned image, combining the strengths of FFS methods with imaging information. An overview of the various methods discussed in this review is given in (Figure 1) and (Table 1).

After performing time resolved imaging in two or three spatial dimensions, a wealth of information can be extracted using different analysis methods. Given that individual virions or subviral particles can be resolved, single virus tracing (SVT) is an invaluable tool for following the dynamics of the observed processes (Figure 1F). There are two basic approaches for tracking single particles: one is to track a particle in real-time using a feedback loop [69], while the second, currently preferred method involves tracking ex post facto [20] from image sequences, either manually or using automated tracking software. Tracking involves first locating the individual particles in each frame and channel collected during a movie. If the particles are smaller than the point-spread-function of the microscope, this is often done with the aid of a spot enhancing filter and the use of a Kalman filter (e.g., [45,70,71,72,73]). The main difficulty in SVT is the correspondence finding, which maps the movement of each particle between consecutive frames. This is challenging since new particles can assemble or appear, signals from two particles can merge or split, or particles can disappear from the observation plane. Hence, the verification of trajectories by visual inspection is imperative. Based on the coordinates and intensities determined for each particle for all consecutive time points, its intracellular localization, co‑localization with cellular organelles or structures as well as changes in fluorescence intensities indicative of dissociation or association processes can be analyzed. In addition, the 2D or 3D instantaneous velocity of the particle and the type of motional behavior (diffusion, corralled diffusion or transport) can be determined by a mean-squared-displacement analysis [21,74,75,76] providing detailed insight into the mechanisms and dynamics of cell-virus interactions.

The HIV-1 Gag polyprotein, consisting of four functional domains, is the only viral protein required for assembly and budding of VLPs from the plasma membrane of virus producing cells. The N‑terminal MA domain is responsible for membrane targeting and membrane binding, while the CA domain provides protein-protein interactions essential for formation of the immature Gag shell and the mature conical capsid. The nucleocapsid (NC) domain of the HIV-1 Gag polyprotein is essential for viral RNA binding [77], which in turn plays an important role in the assembly process by concentrating Gag molecules and promoting conformational changes relevant for membrane association. The C-terminal p6 domain recruits the cellular ESCRT machinery required for virus release. During HIV-1 assembly, Gag polyproteins are transported to the plasma membrane and assemble into an outward curved structure, which grows into a spherical membrane enveloped bud that finally pinches off from the cell membrane (Figure 2). HIV-1 particles do not contain a precisely defined number of Gag molecules and HIV-1 preparations show a rather broad distribution of virion diameters. Electron tomography analyses have revealed that the protein shell of immature HI virions is an incomplete sphere. Based on these results, it was estimated that a virion comprises an average number of ~2400 Gag molecules [7,8]. Quantitative measurements of VLPs using fluorescence brightness analysis showed that the number of Gag molecules within one VLP varied with the expression level of Gag in the host cells from 750 to 2500, while the diameter of VLPs did not change, also indicating varying degrees of shell completeness [27].

The retroviral assembly process starts with the oligomerization of Gag. In the HIV-1 replication cycle, this step is coupled to the recognition of dimeric viral RNA genome, which is recruited by Gag to the viral assembly site [78,79]. With the aid of two-photon FRET, Larson et al. [19] demonstrated the oligomerization of Rous Sarcoma Virus (RSV) Gag not only at the plasma membrane but also within the cytosol. Initiation of HIV-1 assembly at the plasma membrane is apparently governed by an intricate relationship between Gag-Gag, Gag-RNA and Gag-lipid interactions. Analyzing cytosolic and membrane diffusion of Gag.GFP by FCS confirmed the formation of cytosolic Gag-Gag complexes and allowed analysis of the role of the NC domain and the membrane-binding domain of MA within this process [19]. The interaction of Gag with RNA appears to be an important step in the initiation of the assembly process with RNA acting as a scaffold for the binding of Gag molecules [77]. Gag-RNA interactions also appear to convert Gag from a compact conformation into an extended structure required for virus assembly [80]. However, Gag-RNA complexes formed in the cytosol contain only a low number of Gag molecules. Co-immunoprecipitation assays verified that Gag predominantly forms low order oligomers in complex with viral RNA in the cytosol whereas higher order Gag complexes are only detected at the host cell membrane and are dependent on the presence of a membrane interaction domain. In a recent study, fluorescence brightness analysis based on two-photon FFS [34] revealed that the oligomerization of cytosolic Gag is concentration dependent and leads to large cytosolic complexes when myristoylation of Gag is prevented by mutagenesis of the accepting glycine residue.

The combination of fluorescently labeled virus derivatives with sensitive TIRF microscopy allowed the quantitative analysis of HIV-1 budding site formation at the plasma membrane of HeLa cells. Gag expressing cells display a diffuse cytoplasmatic staining followed by the appearance of fluorescent punctae at the plasma membrane that accumulate over a period of ~2 h (Figure 3A). The maximum intensity of the fluorescent punctae is comparable to that of individual fluorescent virions indicating that punctae represent individual assembly sites. By following the fluorescence intensity over time for many individual assembly sites, distinct phases during the assembly process could be distinguished (Figure 3B,C). At the majority of budding sites, the fluorescence intensity initially follows a saturating exponential, indicating gradual recruitment of Gag molecules. Delivery of larger pre-assembled structures to the membrane was not observed while the existence of smaller pre-assembled structures (consisting of 5% of the total labeled Gag signal or less) cannot be ruled out. Quantitative evaluation of 309 traces revealed an average rate constant of 0.0043 s⁻¹ for this initial phase when Gag.GFP is expressed in the viral context resulting in 90 % completion within ~9 min. An average time of 7 min was determined for the assembly of Gag.GFP expressed alone [37]. Whether this small difference is due to a difference in expression context or other experimental conditions remains to be determined. Assembly rates remained unaltered when virion release was inhibited by mutation of the late domain motif or upon blocking virus maturation by introduction of a mutation in the PR active site [38]. Live‑cell TIRF imaging of Gag together with labeled viral RNA revealed that the nucleation event of assembly site formation involves a Gag dependent immobilization of viral genomes at the plasma membrane. Initial RNA recruitment apparently occurs through a small number of Gag molecules (i.e., below the detection limit of the live-cell analyses performed) and the immobilized complex then serves as a nucleation site for the recruitment of further Gag molecules [17]. To investigate whether Gag arriving at the nascent budding sites was derived from the cytosolic pool or from plasma membrane associated Gag molecules, experiments were performed employing the photoconvertible FP mEOS. Using a wavelength of 405 nm for TIRF excitation, plasma membrane associated Gag.mEOS was photoconverted from green to red fluorescence. Analysis of budding sites formed shortly after photoconversion showed that Gag recruitment to the assembly site occurred mainly from the cytosolic pool rather than by lateral membrane-diffusion of Gag molecules binding to the surrounding membrane [38]. At the end of the exponential assembly phase, fluorescence intensity reaches a plateau. Photobleaching experiments revealed that Gag molecules are no longer recruited to the assembly site during this phase [37]. Characterization of viral release is hampered by a limitation inherent to TIRF microscopy, namely, that only the plasma membrane facing the coverslip can be visualized. Thus, many newly formed virions in the observation plane become trapped between membrane and glass surface and the budding process is not easily detectable.

Jouvenet et al. induced acidification of the cytosol by raising the external CO₂ pressure and used the pH sensitivity of pHluorin-tagged Gag to show that the majority of virions were disconnected from the cytosol within approximately 30 min [37]. Unfortunately, this approach does not allow visualization of actual budding events in real time. A different approach was taken by Ivanchenko et al. [38], who inferred HIV-1 release through a change in particle motility or by the sudden disappearance of the particle. This has the advantage that it can be performed with high temporal resolution; however, fission events are not directly visualized and care needs to be taken to distinguish release from endocytotic uptake. Release of individual particles can be verified by a mean-squared displacement analysis, provided the virus could be tracked for several frames after release (Figure 3D). For the particle analyzed in Figure 3D, random Brownian motion (without active transport) was observed with a diffusion coefficient of 1.6 × 10⁻³ μm²/s, a motion that is much too fast to describe free diffusion of a particle the size of HIV-1 in the cytoplasm. Based on these experiments, release was inferred to occur with an average delay of ~15 min after the completion of Gag assembly, suggesting that additional processes may take place during the plateau phase [38].

HIV-1 particles are released by abscission of the enveloped virus bud from the plasma membrane of the virus-producing cell. This process is mediated by proteins that are part of the cellular endosomal complex required for transport (ESCRT) machinery, which are recruited to the viral budding site through interactions with Gag. ESCRT is a system of interacting protein complexes involved in various membrane fission events, most notably in the abscission of intralumenal vesicles forming a multivesicular body (MVB) and in cytokinesis [2,81,82,83]. While, in multivesicular budding, ESCRT-0 is responsible for the recruitment of monoubiquitinylated cargo proteins, ESCRT-I and -II apparently cooperate in a further translocation of the cargo to the vesiculation site and in the initial membrane bending. ESCRT-III proteins assemble into membrane-associated multimeric structures in the area of the bud neck and mediate membrane deformation, which promotes the fission event [84,85]. The AAA-ATPase VPS4 is required in the final stages of the budding process where it catalyzes dissociation of membrane bound ESCRT-III assemblies, thereby recycling the individual subunits for further use [86,87]. It has been well established that HIV-1 Gag can interact with the ESCRT components TSG101 and ALIX through ‘late domain’ motifs within p6 and that the virus exploits ESCRT to mediate bud abscission. However, parts of the machinery, in particular ESCRT-0, -II and some components of the ESCRT-III complex, have been shown to be dispensable for HIV-1 release [88,89]. While the assembly of a spherical Gag shell and the abscission of the membrane neck between virus and cell were originally assumed to be consecutive processes, the observation that the Gag shells of HI virions only line approximately 2/3 of the lipid envelope whereas the Gag shell is more complete in late-domain deficient mutants [7] suggested that Gag assembly and function of the ESCRT factors involved may be cooperative forces that mediate the membrane fission process. Recent live-cell imaging analyses of the recruitment of ESCRT components and ESCRT associated factors to the budding site have yielded information on the order of events occurring at retroviral budding sites [20,39].

These experiments are challenging, since fusion to FPs renders most ESCRT proteins dysfunctional or dominant negative (dn). Thus, their expression levels and functionality have to be carefully controlled in each case. By employing stable cell lines expressing low levels of individual FP-tagged ESCRT proteins, Jouvenet et al. showed that the ESCRT associated protein ALIX accumulated simultaneously with Gag at retroviral assembly sites [39]. In contrast, ESCRT-III proteins as well as VPS4 were only transiently recruited to the budding site after the accumulation of Gag molecules had ceased (Figure 4).

Based on in vitro reconstitution experiments using purified yeast ESCRT proteins and model membranes, it had been proposed that membrane abscission is mediated by the concerted action of ESCRT-I, -II -III assembling in the neck of the nascent vesicle, while VPS4 is only required to catalyze recycling of ESCRT-III complexes [90,91]. In contrast, data from live-cell imaging studies suggest a more active role of VPS4, at least in the case of HIV-1 release. First, ESCRT-III is still recruited to the viral budding site in the presence of dnVPS4 while failing to mediate HIV-1 release under these conditions [39]. Secondly, a detailed analysis of the time points of VPS4 recruitment indicated transient recruitment of VPS4 after completion of Gag assembly but before virus release is detected [20]. Both findings are compatible with a model in which VPS4 contributes to the formation or the transient stability of an ESCRT-III assembly required for membrane abscission (Figure 4A). This is also in agreement with the observation of electron dense structures, possibly representing non‑functional ESCRT assemblies, in the neck region of late HIV-1 buds by electron microscopy upon knock-down of the ESCRT-III component Chmp2 required for VPS4 recruitment [89].

Viral bud neck constriction involves a large conformational reshaping of the plasma membrane, leading to an energetically unfavorable curvature prior to the final release of the virus. Thus, it can be considered as an energy consuming process and the energy involved was modeled in terms of neck diameter, radius of curvature and protein contributions by Fabrikant et al. [84]. While Gag in the absence of ESCRT is capable of assembling into spherical buds arrested before the membrane neck fission event, it is still unclear to what degree the growing Gag lattice, membrane bound assemblies of ESCRT proteins and the specific lipid composition of the membrane at retroviral budding sites cooperate to achieve membrane remodeling, curvature induction and fission during the regular HIV-1 budding event [23,93,94]. Further significant advancements can be expected from the development of improved labeling strategies allowing sensitive detection of fully functional proteins of interest and in particular from the development of methods yielding a real-time microscopic readout of membrane fission.