HIV-1 Assembly Differentially Alters Dynamics and Partitioning of Tetraspanins and Raft Components

Colocalization of CD9 and Gag in live cells

Multiple reports have shown that HIV-1 Gag and Env colocalize with tetraspanins at the PM. However, because all of these studies utilized bivalent antibodies to detect tetraspanins in fixed cells, we set out to determine whether colocalization of Gag and tetraspanins could still be detected in the absence of potential cross-linking or fixation artifacts. If no Gag was expressed, live cells labeled with monovalent Fab fragments exhibited a more diffuse distribution of CD9 compared to cells labeled with bivalent antibodies (Figure 1A,B), even if the bivalent labeling was carried out after fixation (data not shown). A distribution similar to monovalent labeling was obtained with CD9 fused to green fluorescent protein (GFP) (data not shown). To visualize Gag, HeLa cells were transfected with HIV-1 provirus (NL4-3) harboring matrix protein (MA) fused to GFP or monomeric red fluorescent protein (mRFP) (39), and live cells were labeled with anti-CD9 Fab and secondary Fabs. While much of the CD9 signal remained relatively diffuse in Gag-expressing cells, enrichment of CD9 at Gag assembly sites was readily observed (Figure 1C). Similar results were obtained in infected primary CD4+ T cells (Figure 1D). Overall, these results show that endogenous CD9 clusters at Gag assembly sites. They also document that bivalent labeling still represents a valid approach to study Gag–tetraspanin association, and this method, because of its convenience, was thus still used in some of the colocalization analyses in this study.

Because Gag assembly appeared to be the driving factor behind CD9 redistribution and clustering, we tested whether multimerization of Gag was required for these phenomena. Cells expressing a multimerization-deficient Gag that can still be targeted to the PM (40) were labeled for CD9. In the absence of Gag multimerization, no clustering of CD9 by Gag was observed and the overall distribution of CD9 resembled that in untransfected cells (Figure 1E).

Kinetics of Gag colocalization with tetraspanins

To investigate the kinetics of Gag–tetraspanin association, HeLa cells expressing HIV-1 GagGFP [which colocalizes with tetraspanins in the absence of other viral components (14)] were fixed at various time-points post-transfection, stained with antibodies against CD9 and assessed for colocalization between these two antigens at the PM. Only cells displaying detectable PM Gag puncta [which are known to represent assembling and budding structures (41-47)] were included in the analysis. Colocalization of Gag puncta with CD9 increased over time (Figure 2A). However, because the level of GagGFP expression varied greatly from cell to cell, even at the same time-points, we refined our analysis by pooling data from different time-points and sorting cells by their total GagGFP intensity (Figure 2B), which, on a single cell level, is proportional to expression time (Figure 2C). We found that cells displaying more intense GagGFP signal (and thus those cells that have been expressing Gag for a longer time and/or at a faster rate) exhibited higher colocalization with CD9 (Figure 2B). Similar results were obtained for surface CD63 (data not shown). Representative images of cells displaying low and high colocalization of Gag with CD9 are shown (Figure 2D,E). Because colocalization was expressed as percentage of Gag+ pixels colocalizing with CD9 and not vice versa, it was not biased by the overall amount of Gag signal. Overall, these data strongly suggested that tetraspanin–Gag colocalization increases over time, i.e. as virus assembly proceeds, and that Gag assembly is not targeted to preexisting TEMs.

HIV-1 Gag influences mobility of CD9 but not of the raft marker GM1

The observation that HIV-1 Gag clusters CD9 over time suggested that Gag induces CD9 trapping at assembly/budding sites. To test this hypothesis, we measured the mobility of CD9 at the population level by fluorescence recovery after photobleaching (FRAP) analysis. FRAP is commonly utilized to study the mobility of membrane proteins and lipids, and can yield important information about membrane compartmentalization (reviewed in 48). Here, FRAP measurements were used to calculate two parameters, the diffusion coefficient, D, and the mobile fraction, Mf, indicative of the fraction of molecules available for diffusion, for both CD9 and GM1. Vero cells were used in the FRAP experiments, because they express relatively high levels of CD9 (and GM1, see below) allowing for robust fluorescence bleaching and recovery. Further, colocalization between Gag and the former antigens was more apparent in these cells because of the low numbers of microvilli on the ventral side. Live cells were labeled with anti-CD9 Fab fragments on ice to visualize surface CD9, then FRAP experiments were carried out at 37°C. Representative FRAP analyses of single cells and average recovery curves (from multiple cells) for CD9 are shown in Figure 3. The mean Mf and D values from all FRAP experiments are summarized in Table S1.

In cells not expressing Gag, bleached regions of interest (ROIs) in planar membranes recovered relatively rapidly, though not completely, with an apparent D for CD9 of 0.213 μm/s², and Mf of 76%. FRAP of microvilli-like structures enriched in CD9 yielded a lower (though not significantly, p > 0.05) apparent D (0.144 μm/s²) and a similar Mf (77%). However, in Gag-expressing cells, at planar membrane areas containing Gag puncta, CD9 exhibited a considerably lower D (0.137 μm/s²) and a drastically decreased Mf (29%), indicating that CD9 is trapped at Gag assembly sites. Consistent with this idea, CD9-enriched areas containing Gag puncta within the bleach ROI failed to recover to their initial intensity, whereas areas adjacent to Gag puncta within the same ROIs recovered most of their (diffuse) CD9 signal (see insets in Figure 3B,C). The amount of recovery in Gag-containing ROIs decreased with increasing amounts of Gag expression (cf. Figure 3, panels B and C). Typically, the ventral side of the cell was used for FRAP analysis, because it provided a large, flat and uniform membrane area. However, because previous studies have shown that HIV particles released from ventral side of cells can get trapped between the cell and the coverslip (41,43,47), we also performed FRAP experiments on the dorsal side of cells to rule out the possibility that the Gag puncta analyzed in our experiment represent detached, budded particles that cannot exchange proteins with the rest of the membrane. As on the ventral side, recovery of CD9 at Gag-containing structures on the dorsal side was also strongly reduced (data not shown).

Membrane rafts have been proposed to serve as platforms for many membrane-based processes, including budding of HIV-1 and other viruses (recently reviewed in 49). The glycosphingolipid GM1 was used as a marker of raft-like domains, because it is a prototypical component of membrane rafts (50), and has been reported to colocalize with HIV-1 Gag (37,38,51,52). Because Gag appeared to cluster and colocalize with GM1 (as previously reported) comparably to CD9 (also see Figure 4B), we also measured the mobility of this raft marker [labeled with fluorescently labeled cholera toxin subunit B (CTxB)] in the presence or absence of Gag. Overall, GM1 exhibited relatively slow diffusion and low recovery, likely because of the clustering and/or internalization induced by pentavalent CTxB, as previously reported (53,54). Interestingly, neither diffusion nor recovery of GM1 was significantly altered by the presence of Gag assembly sites (Figure 4 and Table S1). Further, GM1, like CD9, showed apparent accumulation at Gag-rich areas, but unlike CD9, this enrichment could be recovered after photobleaching (see insets in Figure 4B). Together, these results imply that, while both GM1 and CD9 are recruited to Gag assembly sites, CD9 becomes trapped but GM1 remains free to diffuse. Importantly, the fact that GM1 is still in exchange with the rest of the PM also shows that most of the Gag puncta analyzed in our experiments represent viral assembly sites that are still connected to the PM, as opposed to being released particles.

HIV-1 Gag differentially influences CD9 and CD55 dynamics and membrane partitioning at the single-molecule level

To corroborate and expand upon the population-level mobility measurements obtained by FRAP, single-molecule fluorescence experiments were performed to analyze the behavior of single CD9 molecules. Single molecules were tracked using Atto647-labeled Fab fragments with a high temporal (100 ms) and spatial (~50 nm) resolution, as in our previous work (18). As observed in PC3 cells, CD9 molecules exhibited three main types of diffusion in control GFP-expressing HeLa cells, namely Brownian (48%), confined (14%) and mixed (38%), a combination of Brownian and confined behavior (Figure 5; Table S2 for the distribution of diffusion coefficient and detailed values). In the absence of Gag expression, the mean value of CD9 apparent diffusion coefficient (ADC) was 0.24 ± 0.03 μm/s², which is comparable to that previously obtained in PC3 cells and in concurrence with the D obtained in the FRAP measurements in this study. Upon expression of Gag, the percentage of CD9 molecules exhibiting confined motion increased significantly (from 14 to 34% of the total number of trajectories) at the expense of Brownian trajectories (from 48 to 27%) (Figure 5; Table S2). The amplitude of this increase was comparable to that of the mobile fraction measured in our FRAP experiments. Reflecting the increase in CD9 confinement, the mean ADC decreased to 0.16 ± 0.03 μm/s². Comparable results were obtained with single molecule tracking (SMT) analysis of CD81, another tetraspanin recruited to Gag assembly sites, where expression of Gag induced trapping of CD81 (data not shown). Note that, while the confined fraction observed by single-molecule analysis was not quite as large as that observed by FRAP, the SMT analysis included the entire cell, which contained large areas of membrane not containing Gag, whereas the FRAP analyses focused on an ROI containing large amounts of Gag.

To determine the generality of membrane protein trapping by Gag assembly, we also investigated the membrane behavior of CD46, a non-raft transmembrane protein, which is not confined by TEMs (18). The mean ADC of this protein (0.16 ± 0.02 μm/s²) as well as the percentage of the different diffusion modes were in the same range than those previously observed in PC3 cells (Figure 5; Table S1). However, compared to CD9, CD46 membrane behavior was mostly unaffected by Gag expression, with only a slight but non-significant increase in the number of confined trajectories (from 11 to 13% of the total number of trajectories) as well as slight decrease in the mean ADC value (from 0.16 to 0.15 μm²/s). These results show that trapping of CD9 at Gag assembly sites is a specific process.

Because Gag assembly also appeared to recruit raft-like domains and in order to parallel the FRAP experiments with GM1, we investigated the behavior of CD55, a putative raft-associated protein (18). CD55 ADC was slightly but significantly reduced in the presence of Gag (p < 0.01 using the Mann–Whitney test, Figure 5), as was the proportion of Brownian trajectories (from 43 to 36%, p < 0.01), while the percentage of confined and mixed trajectories was increased (respectively from 16 to 19% and 41 to 45%) but not significantly (Figure 5, upper panel). Taken together, these results show that CD9 and CD55 distribution of ADC were both modified upon Gag assembling but with a large difference in magnitude.

To confirm these findings, we analyzed the subpopulations of CD9, CD46 and CD55 trajectories that coincided with Gag-enriched areas (Figure 5, lower panel). Sixteen percent of the total CD9 trajectories colocalized with Gag (see Materials and Methods for details). As expected from the increase in confinement described above, most of these trajectories displayed a confined behavior (68%) (note that confinement areas not colocalized with Gag were often observed, presumably the same areas that were seen in the cells not expressing Gag). Interestingly, CD55 trajectories were less often associated with Gag (11%) as compared to CD9. More importantly, membrane behavior of CD55 in Gag areas was also different, with most of the trajectories (71%) exhibiting mixed diffusion, which corresponds to a transient confinement. As expected, CD46 trajectories that colocalized with Gag-enriched regions exhibited similar membrane behavior to that obtained with GFP-expressing control cells (upper panel of Figure 5). Altogether, these data show that, while CD9 and CD55 are specifically recruited into assembly sites, tetraspanins interact with Gag in a very different manner compared to raft components.

Differential dependence on cholesterol for enrichment and confinement of CD55 and CD9 at HIV-1 assembly sites

Because raft components and tetraspanins were clustered to HIV-1 Gag assembly sites, we tested whether cholesterol, a key structural component of membrane rafts and TEMs, was required for this clustering. Cells expressing HIV-1 GagGFP were briefly treated with cholesterol oxidase (COase) to decrease membrane cholesterol content, labeled with Fabs to visualize CD9, CD46 or CD55 (ensemble labeling conditions) and processed for colocalization analysis (Figure 6). In agreement with our results on trajectories that overlapped with Gag-enriched areas (Figure 5, lower panel), this colocalization analysis indicated that CD9 is associated with these areas (57%) as was CD55, although to a lesser degree (36%). Interestingly, when cells were treated with COase, the fraction of Gag that colocalized with CD9 partially decreased (from 57 to 38%). However, the overlap of Gag and CD55 was almost completely abolished (from 36 to 9%). In contrast, CD46 showed little colocalization with Gag-enriched areas (6%) and this association was not significantly affected by COase treatment (from 6 to 5%). These results suggest that the recruitment of CD9 to assembly sites only partially depends on cholesterol, while this lipid is fully required for CD55 association with Gag-enriched areas.

The effect of COase treatment was then investigated at the single-molecule level (Figure 7). When membrane cholesterol content was decreased, the mobility of CD9, CD55 and CD46, as measured by their Brownian diffusion coefficient in the absence of Gag was slightly decreased, as previously observed with methyl-β-cyclodextrin treatment (18). This decrease was attributed to a general effect on membrane dynamics upon cholesterol depletion. While the effect of Gag expression, namely the increase in confined CD9 trajectories at the expense of Brownian trajectories, was reduced by cholesterol depletion, the expression of Gag still clearly increased the percentage of confined trajectories under these conditions (upper right panel of Figure 7 and Table S2). Meanwhile, the increase in CD55 confinement by Gag was almost completely abolished by cholesterol depletion (lower panel of Figure 7). The behavior of CD46 was not affected by cholesterol depletion because no effect of Gag expression was observed in the first place (middle panel of Figure 7). Taken together, these data suggest that CD9 confinement at Gag assembly sites is partially dependent on cholesterol, whereas the transient recruitment of CD55 in these areas is completely dependent on cholesterol.

Cells, plasmids and transfections

HeLa and Vero cells were maintained in DMEM with 10% FBS. Primary human CD4+ T cells were obtained from Ficoll-separated whole blood (from healthy donors) by negative selection (Miltenyi Biotec, according to manufacturer’s instructions). Purity was typically 90–95%. CD4+ T cells were activated with 5 μg/mL phytohemagglutinin (PHA) for 24 h, then infected with NL4-3(MA-GFP) virus (see below). CD4+ T cells were maintained in RPMI with 20% FBS and 20 units/mL of interleukin-2 (IL-2).

The CD9-enhanced green fluorescent protein (EGFP) (in pEGFP-N1 vector, containing a mutation preventing the dimerization of EGFP) plasmid kindly provided by M. Hemler (DFCI) (68). The NL4-3(MA-EGFP) and NL4-3(MA-mRFP) proviruses were kindly provided by B. Müller (University of Heidelberg) (39). The codon-optimized untagged Gag and GagGFP-fusion constructs has been previously described (69). Venus-fused monomeric Gag construct was kindly provided by P. Spearman. This construct is myristoylated and can bind cellular membranes, albeit weakly, and contains mutations in the D (C-Terminal Domain) and N-terminal domain that completely abolish multimerization (40).

Transfections were performed using Lipofectamine 2000 (Invitrogen), according to manufacturer’s instructions.

Epifluorescence microscopy

HeLa cells were grown on glass-bottom dishes (Mattek). Cells were either labeled on ice or after fixation, as indicated in figure legends. Fixation was done in 4% paraformaldehyde (PFA) in PBS for 10 min, unless otherwise indicated. The following antibodies were used: mouse monoclonal anti-CD9 K41 (BACHEM), mouse monoclonal anti-CD63, H5C6 (Developmental Studies Hybridoma Bank); anti-Gag p6 rabbit serum, kind gift of D. Ott (NCI). Secondary antibodies were AlexaFluor 488, 594, or 647-conjugated and directed against the appropriate species (Invitrogen). Note that, while full-length anti-CD9 antibody K41 induces clustering of CD9 at physiological temperature (70,71), its Fab fragment completely lacks this activity (71), and was therefore suitable for live imaging, and the anti-CD9 SYB-1 antibody used in SMT experiments also lacks this activity.

Anti-CD9 Fab fragments were prepared using the ImmunoPure Fab Preparation Kit (Pierce), and their purity was tested by SDS–PAGE. When monovalent labeling was performed, fluorescein isothiocyanate (FITC)- or DyLight (488, 594 or 649)-conjugated anti-mouse Fab secondary antibodies (Jackson Immunoresearch) were used.

For GM1 labeling, cells were incubated on ice using AlexaFluor 488-conjugated CTxB (Invitrogen) for 30–60 min, fixed and incubated with anti-CTxB rabbit antibody (Invitrogen) to enhance clustering. For TfnR labeling, cells were incubated with transferrin conjugated to AlexaFluor 594 (Invitrogen) for 30 min on ice, then fixed.

Image acquisition and deconvolution were performed using either a DeltaVision deconvolution microscopy station (Applied Precision, Inc), as previously described (14), or a Nikon Eclipse Ti-E equipped with a Nikon 60x PlanApo 1.4 numerical aperture (NA) objective, Qimaging EXi Blue camera, image software NIS Elements 3.10 and AUTOQUANT 2.1.0 (deconvolution). For live imaging, cells were labeled on ice, then imaged at 37°C and 5% CO₂ in a custom-built WeatherStation climate chamber (Applied Precision, Inc).

Colocalization analysis and image processing

Colocalization analysis was carried out using Volocity software essentially as described (14) or CoLocalizer Pro software. Briefly, deconvolved images representing a single bottom optical Z-section of a single cell were converted to tiffs and imported into Volocity software v3.7 (Improvision). The classifier module set to 2 or 3 SD above the mean pixel intensity was used to detect and define the bright puncta for each channel, while excluding the dimmer fluorescent signals. The total number of pixels overlapping between two different channels was divided by the total number of pixels for a chosen channel (as indicated in the figures), yielding percent colocalization. Colocalization values from at least 10 cells were averaged to give mean percent colocalization. Average pixel intensity measurements were obtained from corresponding images, also using the classifier module.

Final image assembly, cropping and resolution adjustments for figures, including magnification of boxed regions, was done using Adobe Photoshop.

FRAP analysis

Vero cells were plated on Mattek dishes coated with fibronectin (to minimize detachment of cells during photobleaching) and transfected with the GagGFP and untagged Gag expression vectors (1:1 ratio); 16–24 h posttransfection cells were labeled on ice with anti-CD9 Fab and DyLight 649-labeled secondary Fab (Jackson Immunoresearch), or with CTxB conjugated to Alexa 647 (Invitrogen). All FRAP experiments were performed on a Zeiss LSM 510 Meta confocal microscope at the Microscopy Imaging Facility at the University of Vermont. Cells were maintained at 37°C with a stage heater. Each dish was imaged for 20–40 min. A 100× PlanApochromat 1.4 NA oil immersion objective was used. GFP was excited with the Argon 488 laser line (3–15% transmission), while Alexa 647 and Dylight 649 were excited with the 633-nm He–Ne line (5–20% transmission), and emission was split and collected simultaneously for both channels. Single color experiments were performed to ensure no cross-talk between the two channels. Photobleaching was performed using the 633-nm laser line at full laser power for 200 iterations on a circular ROI approximately 3.5 μm in radius. Subsequent recovery images were collected using low laser power at 5-second intervals for approximately 2–3 min, until ROI recovery had reached a plateau. Only a single Z-section was bleached and imaged, typically the ventral side of the cell contacting the cover slip.

For FRAP calculations, the mean intensity over time of the bleach ROI, as well as a background ROI (containing no cells), and an unbleached ROI in the cell of interest were determined using the Zeiss LSM analysis software and exported into Microsoft Excel. The bleach ROI data were converted to fractional recovery (72), normalized by the overall fluorescence loss in the cell, as measured by the unbleached ROI. This data was imported into GraphPad Prism and fitted using non-linear regression (an exponential decay algorithm, Y = span × exp(−K × X) + plateau, with the half-time to recovery, t_1/2 = 0.69/K). The span and plateau were used to calculate Mf. D was calculated using a simple two-dimensional diffusion model for a circular bleach ROI: D = 0.224 × (r²/t_1/2) (73,74). This analysis was performed for each individual cell and the D and Mf values were averaged to get mean D and Mf. To yield average recovery curves, mean fractional recovery over time was averaged between multiple cells.

SMT experiments and analysis

SMT experiments were carried out as previously described (18). Briefly, cells expressing GFP or GagGFP [cotransfected with untagged Gag (1:1 ratio)] were incubated in DMEM at 37°C for 10 min with Atto647N-labeled Fab fragments of mAbs raised against CD9 (SYB-1), CD81 (TS81), CD46 (11C5) and CD55 (12A12) (18,75). For COase treatment, cells were incubated for 30 min with 1 UI/mL of the enzyme before labeling. A home-made objective-type TIRF setup allowing multicolor single-molecule imaging and equipped with a Plan Fluor 100x/1.45 NA objective (Zeiss, Le Peck, France Brattleboro, VT) was used. All the experiments were performed with a 100 ms integration time. For some experiments, to achieve a better specificity in the detection of the two fluorescent signals, alternating-laser excitation (ALEX) was performed using an acousto-optical tunable filter and controller (AOTF; AA Optoelectronics) (76).

All the movies were analyzed using a home-made software (named ‘PaTrack’) implemented in visual C++. Trajectories were constructed using the individual diffraction limited signal of each molecule. The center of each fluorescence peak was determined with subpixel resolution by fitting a two-dimensional elliptical Gaussian function. The two-dimensional trajectories of single molecules were constructed frame per frame. Only trajectories containing at least 40 points were retained. Diffusion coefficient values were determined from a linear fit to the MSD (mean square displacement)-τ plots between the first and the fourth points (D1–4) according to the equation MSD(t) = 4Dt.

The determination of the motional modes (Brownian, confined or directed) and parameters was performed as described by Kusumi et al. (77). For each trajectory, we first linearly fitted the MSD on the 10% first points in order to use sufficiently populated curves. If the MSD-τ plot shows positive or negative deviation from a straight line with a slope of 4D (Brownian diffusion), the MSD is respectively adjusted with a quadratic curve (4Dt + v²t²) (directed diffusion) or with an exponential curve $\frac{L^2}{3}\left[1-\exp \left(\frac{-12\mathit{Dt}}{L^2}\right)\right]$ (confined diffusion where L is the side of a square domain, the confinement diameter being related to L by dconf = (2/√π)L). For the mixed trajectory exhibiting a combination of Brownian and apparent confined motion mode, the trajectory was split and the MSD of each segment was adjusted with a linear or an exponential curve as described above.