Redistribution of caveolae during mitosis

Membrane trafficking from and to the plasma membrane is responsible for the transport of many macromolecules and for plasma membrane homeostasis, which is essential during interphase, cellular migration and division (reviewed by Doherty and McMahon, 2009; Scita and Di Fiore, 2010). Caveolae represent an important pathway that links the plasma membrane with intracellular organelles (Echarri et al., 2007; Hansen and Nichols, 2009; Parton and Simons, 2007). Caveolae cycle between the plasma membrane and early endosomes without disassembly of their main protein caveolin-1 (Cav1) (Nichols, 2002; Pelkmans et al., 2004; Pelkmans and Zerial, 2005). Recent work has uncovered a role for membrane trafficking during mitosis (reviewed by Montagnac et al., 2008). For example, clathrin-mediated endocytosis remains constant during all stages of mitosis, but recycling back to the cell surface halts during the rounding-up phase (prophase to metaphase), leading to effective membrane accumulation within the cell and a decrease of plasma membrane surface area (Boucrot and Kirchhausen, 2007). Recycling resumes once the cell enters anaphase and the stored membrane traffics back to the cell surface, where rapid fusion mediates the recovery of plasma membrane area and the subsequent formation of envelopes of the two daughter cells (Boucrot and Kirchhausen, 2007). Caveolae are another important endocytic route, but their dynamics has not been investigated during cell division. In this report, we have used advanced live-cell imaging techniques and quantitative electron microscopy to study caveolae during mitosis. We show that the vast majority of caveolae are dynamic during interphase, with lifetimes ranging from seconds to several minutes. We also reveal an extensive redistribution of caveolae from the plasma membrane to endosomal structures during mitosis. This has profound implications for understanding caveolae dynamics during the important cellular event of division.

Caveolae were observed by TIRF microscopy and are reported to exist in two pools at the plasma membrane: a majority that is immobile, and a minority displaying a fast ‘kiss-and-run’ behavior (transient fusion without full collapse of the vesicle) with lifetimes of 2–5 seconds (Pelkmans and Zerial, 2005). By substantial extension of the acquisition period (an image collected every 2 to 10 seconds, for up to 30 minutes) we now report that the majority of caveolae in HeLa or BSC1 cells are not immobile, but instead are endocytosed when observed over periods of several minutes (Fig. 1A and supplementary material Fig. S1A). We noted that the temperature at which the acquisitions were performed had a strong influence on caveola budding. Cooling the cells from 37°C to 30°C was enough to totally block the long-lived (>2 seconds) caveolae in HeLa cells but had little impact on short-lived (≤2 seconds) caveolae (supplementary material Fig. S1B). We note that previous TIRF live-cell imaging studies of caveolae were performed using high imaging frequency (5 Hz) for just a few minutes and not always at 37°C, possibly explaining why the slow, long-lived caveolae were reported as being immobile and non-endocytic (Pelkmans and Zerial, 2005).

Over a period of 10 minutes, the vast majority (85±12%) of Cav1-positive structures were dynamic (Fig. 1B and supplementary material Movie 1). Among the dynamic pool, short-lived caveolae (≤2 seconds) corresponded to 30.8%, whereas the remainder had lifetimes ranging from >2 seconds to over 7 minutes (Fig. 1C). There was no correlation between lifetime and maximum fluorescence intensity reached (supplementary material Fig. S1C,D), suggesting that the global population of caveolae is more heterogeneous than previously appreciated (Pelkmans and Zerial, 2005). Short-lived caveolae have been shown to be pre-assembled structures displaying a ‘kiss-and-run’ behavior with the plasma membrane (Pelkmans and Zerial, 2005). To demonstrate that the fast caveolae were fusing with the plasma membrane, an elegant assay taking advantage of pH-sensitive quenching of loaded fluorescently tagged Cholera Toxin B (CTB) was used in a previous study (Pelkmans and Zerial, 2005). We tried to use a similar assay to assess whether the long-lived caveolae were fusing with the plasma membrane. Unfortunately, maintaining the cells at pH 5.5 or 6.2 for periods longer than 1–2 minutes, which is required to follow the dynamics of the long-lived caveolae, induced a significant enlargement of Cav1–EGFP-positive endosomes (results not shown), suggesting a perturbation of Cav1 trafficking and rendering the assay unusable to study the behavior of the long-lived caveolae. Analysis of 500 caveolae imaged fast (5 Hz) showed similar appearance and disappearance kinetics (Fig. 1D and supplementary material Fig. S1E). The vast majority (97±3%) of caveolae that appeared during image acquisition reached their full intensities within 1 second, irrespective of their subsequent lifetime. Similarly, most caveolae (89±7%) disappeared with the same kinetics (Fig. 1D and supplementary material Fig. S1E). Considering that the short-lived caveolae represent less than a third of the dynamic caveolae, we concluded that short-lived and long-lived caveolae arrive at, and depart from, the plasma membrane with similar dynamics. In addition, treatment for 30 minutes with 80 μM dynasore, a cell-permeable small inhibitor of dynamin (Macia et al., 2006), totally blocked caveolae dynamics at the plasma membrane (Fig. 1E), consistent with the reported dependency of Cav1 endocytosis on dynamin activity (Henley et al., 1998; Oh et al., 1998). Finally, both short- and long-lived caveolae were virtually all positive for polymerase I and transcript release factor (PTRF, also known as and hereafter referred to as cavin-1) (Fig. 1F), a cytoplasmic protein that is associated with mature plasma membrane caveolae (Hill et al., 2008). Altogether these results suggest that ‘short-lived’ and ‘long-lived’ caveolae have a similar composition and mechanism of formation.

To investigate whether the repartition between short- and long-lived caveolae varies upon different cellular environments, we investigated the dynamics of these distinct pools of caveolae in confluent versus migrating cells, a state reported to result in the accumulation of caveolae within a specific cellular region (Parat et al., 2003). We first verified that in BSC1 cells, as reported in other cell types (Howes et al., 2010; Parat et al., 2003), caveolae were enriched at the rear of migrating but not confluent cells (Fig. 1G,H, gray bars). We then measured the repartition of short-lived (<2 seconds, blue bars) and long-lived (>2 seconds, yellow bars) caveolae at both the front and rear halves of confluent and migrating cells (Fig. 1G,H). The repartition between short- and long-lived caveolae changed significantly upon cell migration, when both at the front and rear of the cells, the proportion of short-lived caveolae increased (Fig. 1G,H), even though the long-lived caveolae were always the most abundant population. Altogether, our data suggest that the caveolae turnover time, that is the time spent by a caveola on the plasma membrane, varies from a few seconds to several minutes, depending on the particular cellular settings.

We next wondered whether caveolae have the same dynamics during mitosis as in interphase, because recent work highlighted changes in membrane trafficking during cell division (Boucrot and Kirchhausen, 2007). Compared with interphase, the density of caveolae at the plasma membrane increased by a factor of about two when the cell was rounding up during prophase to metaphase (called hereafter pro-metaphase), decreased by a factor of two during metaphase, and increased again by a factor of four as the daughter cells form during anaphase to cytokinesis (called hereafter cytokinesis) (Fig. 2A,B). Such changes in caveolae density suggested variations in turnover time. During interphase, the pool of caveolae at the surface is maintained by a balance between caveolae that arrive and those that depart from the plasma membrane (Fig. 2C). During pro-metaphase, 17±6% more structures were internalized than appeared (Fig. 2C). During cytokinesis, the reverse occurred: 23±8% more caveolae arrived than departed (Fig. 2C and supplementary material Movie 2). During pro-metaphase and cytokinesis, caveolae were more dynamic (i.e. stayed for less time at the plasma membrane and turned over more often) than during interphase (Fig. 2D, gray and white bars). The average lifetime of 146 caveolae from three cells in pro-metaphase and 210 caveolae from four cells in cytokinesis cells were 17.8 and 15.1 seconds, respectively, around twofold faster than during interphase (30.9 seconds). The short-lived caveolae (≤2 seconds) were more abundant during both phases (43.8% and 49.0%, respectively, compared with 30.8% during interphase), although the majority of structures remained long-lived (Fig. 2D). Altogether, these studies reveal profound changes in the dynamics of caveolae during mitosis.

The changes in caveola dynamics and variations in caveolae densities during mitosis measured above by TIRF suggested profound modifications in caveolae localization. Global cellular repartition of Cav1 was measured by fast three-dimensional (3D) spinning-disk confocal imaging on live cells naturally undergoing interphase, metaphase or cytokinesis. During interphase, Cav1 is present at the plasma membrane, cytoplasmic vesicles and Golgi complex (Parton and Simons, 2007) (Fig. 3A). Cav1 at the plasma membrane displays a punctate or continuous signal at the sides of adherent cells (Fig. 3A), which correspond to a high density of individual caveola structures (as seen by electron microscopy) (Parton and Simons, 2007). During metaphase, plasma membrane Cav1 levels showed a striking decrease and several large intracellular structures containing Cav1 were observed (Fig. 3A). We verified that these structures were not connected with the plasma membrane, because they failed to stain with the red membrane-impermeant dye FM 4-64 (supplementary material Fig. S2A). Internal Cav1 significantly colocalized with the early endosomal marker EEA1 (supplementary material Fig. S2B), suggesting that Cav1 internalizes during mitosis. Cav1 reappeared at the cell surface during cytokinesis (Fig. 3A). Whole caveolae are probably internalized during mitosis because not only Cav1, but also cavin-1, a recently identified constituent of caveolae (Hill et al., 2008), displayed similar relocalization (Fig. 3C). Two lines of evidence suggested that caveolae redistribute by membrane trafficking during mitosis. First, treatment for 30 minutes with 80 μM dynasore, which blocked Cav1 dynamics during interphase (Fig. 1E), also inhibited the decrease of Cav1 levels at the plasma membrane of pro-metaphase cells (supplementary material Fig. S2C). Second, treatment with the Golgi-dispersing drug Brefeldin A (BFA) did not perturb the reappearance of Cav1 during cytokinesis (supplementary material Fig. S2D), showing that reappearance of Cav1 at the cell surface results from the fusion of the previously stored structures with the plasma membrane, rather than from a newly synthesized pool. Time-lapse spinning-disk confocal microscopy performed on a middle section plane of dividing cells illustrated the disappearance of the plasma-membrane-localized Cav1 during the rounding-up phase and its reappearance as the two daughter cells form (supplementary material Fig. S2E,F and Movie 3). The specific redistribution of Cav1 during metaphase was also observed for endogenous Cav1 in BSC1, A431 and HeLa cells. Cav1 subsequently returned to the plasma membrane during cytokinesis (Fig. 3B and supplementary material Fig. S2H).

We next used quantitative electron microscopy to examine the distribution of Cav1 and caveolae in mitotic cells. Immunogold labeling for Cav1 using an N-terminal antibody, which exclusively labels the cell surface caveolae of interphase cells (Dupree et al., 1993), showed strong labeling on cytoplasmic vesicles (including structures with caveola morphology) and tubules of pro-metaphase cells (Fig. 3D,E and supplementary material Fig. S3), confirming the internalization of Cav1 during the mitotic round up. Internalized Cholera Toxin B coupled to horseradish peroxidase (CTBHRP), a non-specific surface marker, labeled tubulovesicular endocytic structures, but strikingly also labeled numerous structures with the morphology of budded caveolae (Fig. 3D and supplementary material Fig. S3A–D). Gold-labeled CTB (CTBgold), which in contrast to CTBHRP is highly concentrated in caveolae (Parton et al., 1994) (see also supplementary material Fig. S3E), was used in a dual-labeling strategy (Parton et al., 1994) with CTBgold as internalized caveolar marker, and subsequently CTBHRP at 4°C as surface marker (Fig. 3D). This confirmed the intracellular redistribution of Cav1 during metaphase. Compared with interphase cells, pro-metaphase cells showed a decrease in morphologically distinguishable caveolae at the cell surface and an increase in internalized CTBgold-positive, CTBHRP-negative structures (Fig. 3D,F). Quantification of surface caveolae in metaphase showed a decrease of 28±11% (P<0.001) compared with interphase cells (Fig. 3F). Conversely, the percentage of internalized caveolae increased by 49±10% (P<0.0001) in metaphase compared with interphase cells (Fig. 3F). To confirm the change in caveola dynamics during mitosis reported in Fig. 2, the number of caveolae that budded in 1 minute at 37°C as a percentage of total plasma membrane caveolae was quantified. We used a recently developed method that relies on the quenching of the activity of surface CTBHRP by ascorbic acid and recognition of budded caveolae by their characteristic size and morphology (Kirkham et al., 2005; Le Lay et al., 2006). In interphase cells, an average of 3±2% of the total caveolae budded in 1 minute (Fig. 3G). At this rate, the whole plasma membrane pool of Cav1 will be turned over within ~20–30 minutes, consistent with the kinetics reported above. In pro-metaphase cells, 19±7% of the total caveolae budded in a minute (Fig. 3G), which corresponded to a ~sixfold increase of caveolae dynamics. Collectively, our results show that caveolae redistribute intracellularly during metaphase to return at the surface during cytokinesis. These fluxes are mediated by an activation of caveola dynamics, resulting in increased trafficking of these organelles during cell division.

To identify possible mechanisms involved in the transient redistribution of caveolae during metaphase, we next investigated the involvement of the microtubule cytoskeleton. Mitosis is associated with profound cytoskeletal rearrangements mediating the proper separation of the chromosomes pairs. In particular, microtubules depolymerize to reassemble the mitotic spindle in an organized array (Goshima and Scholey, 2010). Because recent work linked microtubule dynamics to plasma membrane targeting of caveolae during interphase (Wickstrom et al., 2010), we wondered whether microtubule depolymerization during cell division could mediate caveolae redistribution. Microtubule perturbation could not be investigated during mitosis because it blocks proper cell division (Zieve, 1984). However, upon treatment of interphase BSC1 cells with the microtubule-depolymerizing drug nocodazole, levels of Cav1 decreased significantly at the plasma membrane and redistributed in part to early endosomes labeled with EEA1 (Fig. 4A–C), which occurs naturally during prophase to metaphase (Fig. 3). Washout of the drug (and subsequent repolymerization of microtubules) caused a significant shift of Cav1 back to the cell surface (Fig. 4A–C). As observed during the prophase to metaphase transition, the progressive departure of Cav1 from the plasma membrane in nocodazole-treated cells was due to a reduction of the number of caveolae that arrive at the plasma membrane (Fig. 4D). During washout of the drug, the number of caveolae that arrived at the plasma membrane increased beyond the number that departed, mediating the progressive restoration of the caveolae pool (Fig. 4D). Because microtubule depolymerization during interphase was enough to induce a similar redistribution to that observed during mitosis, we conclude that natural microtubule rearrangements during cell division can mediate changes in caveola localization, perhaps by trapping caveolae intracellularly and thus decreasing the number of caveolae that can recycle from internal stores to the plasma membrane.

This study revealed that the vast majority of caveolae are dynamic. In addition to the short-lived (~2–5 seconds) caveolae previously described (Pelkmans and Zerial, 2005), we observed that a large proportion of the caveolae previously described as immobile are in fact long-lived (10 seconds to over 7 minutes). Endocytosis of the long-lived structures was only observed in cells strictly kept at 37°C. At 30°C, only the short-live caveolae were observable, the rest being seemingly static. Long-lived caveolae all appeared and disappeared with similar rates as the equivalent steps of the short-lived ones, but because many individual fluorescence traces of long-lived caveolae showed significant intensity variations (not shown), it suggests that some might be multi-caveolae assemblies, potentially formed by several individual caveolar vesicles docking on top of each other, as observed previously by electron microscopy (Parton et al., 1994).

The extensive trafficking (internalization, followed by reappearance at the surface) of most caveolae during cell division reported here uncovers a mitosis-specific redistribution of Cav1. A decrease in caveolae arrivals coupled to consistent budding results in an imbalance between arrivals versus departures (Fig. 2), and a progressive net decrease of surface caveolae (Fig. 3). What could be the function of such trafficking? One hypothesis is that it could contribute to the variations of the plasma membrane area during cell division that we have previously reported (Boucrot and Kirchhausen, 2007). However, considering the predicted low capacity of membrane uptake by caveolae endocytosis (Parton et al., 1994; Parton and Simons, 2007), it is unlikely to be a major membrane uptake pathway, consistent with the viability of Cav1^−/− mice and the ability of Cav1^−/− cells to divide. Nevertheless, the importance of Cav1 in the fine regulation of the cell cycle has been revealed by detailed studies of Cav1-null cells or cells overexpressing Cav1–GFP (Cerezo et al., 2009; Fang et al., 2007; Galbiati et al., 2001). These studies link Cav1 to Rac1- and p53/p21-dependent pathways, which regulate of the progression through the G1 phase of the cell cycle. In addition, Del Pozo and colleagues have implicated caveolin internalization in this process (Cerezo et al., 2009). Thus, the precise and regulated redistribution of Cav1 during mitosis reported here might be required for proper progression through the subsequent G1 phase.

Internalization of caveolae during mitosis could be required for the correct distribution of Cav1 between the two daughter cells. The localization of Cav1 at the plasma membrane of many interphase cells is highly asymmetrical and polarized, especially during cell migration (Lentini et al., 2008; Parat et al., 2003). If the distribution of Cav1 between the two daughter cells is not equal, one cell would inherit an excess of Cav1. This would result in the same consequences as Cav1 overexpression, which is known to inhibit cell proliferation by blocking the progression through G1 (Fang et al., 2007; Galbiati et al., 2001). Therefore, redistribution of Cav1 to intracellular endosomal structures, which partition stochastically during mitosis (Bergeland et al., 2001), might ensure the equal distribution of Cav1 between the two daughter cells. Use of specific inhibitors of caveola endocytosis will be required to pinpoint the precise role of the changes in caveola dynamics revealed here in Cav1-mediated regulation of the cell cycle.

The precise mechanisms inducing Cav1 redistribution during mitosis are yet to be elucidated, but this study provides some potential new clues. Our results suggest that microtubule depolymerization and reorganization, which occurs during prophase and metaphase, could explain, at least in part, caveolae redistribution because microtubule depolymerization in interphase cells was enough to induce a similar change (Fig. 4). In both cases, a decrease in the rate of arrivals of caveolae to the surface coupled to a sustained budding induced the net decrease in the surface pool. A microtubule-based transport has been recently shown to mediate traffic of caveolae to the cell surface (Wickstrom et al., 2010). Mitosis provides a natural state of activated and directed trafficking of caveolae and thus future studies during this stage of the cell cycle might also be useful to elucidate molecular mechanisms of caveolae trafficking.

Monkey kidney epithelial cells BSC-1 (ATCC CCL-26, referred to here as ‘BSC1’) and human HeLa (ATCC CCL-2) and A-431 (ATCC CRL 1555, referred to here as ‘A431’) cells were grown as adherent cells in DMEM supplemented with 10% fetal bovine serum (FBS). We have generated a HeLa and a BSC1 cell line stably expressing an EGFP-fusion chimeras of Cav1 (Cav1–EGFP) by transfecting the plasmid using FuGene 6 (Roche Diagnotics) or Lipofectamine 2000 (Invitrogen), respectively, followed by selection and maintenance with complete medium supplemented with geneticin (G418, 0.5–0.7 mg/ml). Transient transfections of cavin-1–mRFP and EEA1–mRFP were performed on the Cav1–EGFP stable BSC1 cell line using Lipofectamine 2000. Cells were imaged 24–48 hours later.

The cells were not synchronized by any chemical means. The stage along the cell cycle was determined in the population of asynchronous cells using phase-contrast bright-field illumination according to the following criteria: cells in prophase contain condensed chromosomes surrounded by the nuclear envelope; cells in metaphase appear round, lack their nuclear envelope and display condensed chromosomes aligned at the metaphase plate; cells undergoing cytokinesis (but before abscission) display a deep furrow, still have condensed chromosomes and start to show their nuclear envelope.

The following antibodies were used: rabbit anti-Cav1, anti-VIPN (Dupree et al., 1993) and RbTL C13630 (BD Transduction Laboratory). The following reagents were used: human transferrin–HRP (Rockland Inc), CTBHRP (Sigma), nocodazole (Sigma), Hoechst 33342 (Sigma) and dynasore (Macia et al., 2006). CTBgold was prepared as described previously (Parton, 1994) and each batch of CTBgold was tested for specificity for labeling of surface caveolae at 4°C as described previously (Parton, 1994).

Cav1 immunolabeling was performed with the anti-VIPN, anti-Cav1 antibody raised against the N-terminus of canine Cav1 that does not recognize the Golgi pool of caveolin (Dupree et al., 1993). For internalization experiments two different schemes were used based on previously described methods (Parton et al., 1994). CTBgold (14 nm) was bound to the cell surface at 4°C with or without 10 μg/ml transferrin–HRP and internalized for 20 minutes at 37°C before surface labeling with CTBHRP at 4°C (Parton, 1994). In other experiments, CTBHRP was bound to the cell surface at 4°C for 30 minutes and the cells were warmed at 37°C for 20 minutes before fixation. For quantification of caveola dynamics in A431 cells using the DAB and ascorbic acid method, experiments were performed and quantified exactly as described previously (Kirkham et al., 2005). Budded caveolae were defined by their content of DAB reaction product and morphology, as in previous studies (Le Lay et al., 2006). Metaphase cells were identified by their characteristic morphological appearance under the light microscope, marked, and then ultrathin sections of the same areas, cut parallel to the culture substratum, were prepared. No synchronization protocol was used and only cells that could be unequivocally identified as being in metaphase were further analyzed. Interphase cells in the same dishes were analyzed in the same manner.

Cells grown on No. 1.5 glass coverslips (25 μm in diameter) were washed with DMEM supplemented with 10% FBS and put in imaging buffer (α-MEM without Phenol Red supplemented with 20 μm HEPES, pH 7.4 and 5 % FBS). The coverslips were transferred to a sample holder (20/20 Technology, Inc., Wilmington, NC) and kept on the microscope at 37°C with 5% CO₂ and 100% humidity located inside an environmental chamber set at 37°C also containing the objective lenses. In some experiments, dynasore (80 μM in 0.4 % DMSO final) was directly dissolved in PBS supplemented with 0.1 μm CaCl₂, 1 μm MgCl₂, glucose (4.5 g/l) and 1% Nu-Serum (BD Biosciences). The cells were washed three times with pre-warmed PBS-supplemented solution (without dynasore) before addition of dynasore-containing medium to the cell for 30 minutes at 37°C before imaging in the same medium (with dynasore).

Live-cell time series or 3D imaging were done using a spinning disk confocal head (Perkin Elmer, Boston, MA) coupled to a fully motorized inverted microscope (Axiovert 200M, Carl Zeiss) equipped with a 63× oil-immersion lens (Pan Apochromat, 1.4 NA, Carl Zeiss). A 50 mW solid-state laser (473 nm; Crystal Laser) coupled to the spinning head through an acoustic-optical tuneable filter (AOTF) was used as light source. The imaging system operates under control of SlideBook 4.2 (Intelligent Imaging Innovations) and includes a computer-controlled spherical aberration correction device (SAC, Intelligent Imaging Innovations) installed between the objective lens and the back illuminated 16-bit CCD camera (Cascade 512B; Roper Scientific, Photometrics) used with a binning of 1×1. Rapid acquisition of sequential optical sections spaced 0.25 μm apart was achieved with the aid of a piezo-driven stage (Applied Scientific Instrumentation). To reduce photo-bleaching, the illumination was turned off during the readout period from the CCD to the computer.

TIRF experiments were performed using an inverted microscope (Axiovert 200M; Carl Zeiss) equipped with a Zeiss TIRF module, a 100× 1.45 NA Zeiss TIRF oil-immersion lens and computer-controlled SAC. Cells were kept on the microscope at 37°C with 5% CO₂ and 100% humidity located inside an environmental chamber set at 37°C also containing the objective lens. A 470 nm solid-state laser (Crystal lasers, Reno, NV) was the source of illumination in TIRF. A back-illuminated Cascade 512B (Roper Scientific) was used for image acquisition under no-gain magnification mode. To reduce photo-bleaching, the illumination was turned off during the readout period from the CCD to the computer.

Fluorescence imaging of fixed cells was done with an upright microscope (Carl Zeiss) equipped with manually-driven SAC and a 40× or 63× 1.4 NA oil-immersion Zeiss objective lenses. Epifluorescence illumination was provided through Lambda DG-4 illumination unit (Sutter Instruments). 12-bit digital images were obtained with a 12-bit cool CCD camera (Cool Snap HQ, Photometrics) with 2×2 binning. Images were acquired with exposure times between 100 and 500 milliseconds. Three-dimensional stacks of sequential optical sections were acquired 0.3 μm apart. No deconvolution was applied on any images shown in the figures.

The following sequential steps were used to automatically identify and track Cav1–EGFP clusters in TIRF captures: (1) Creation of a mask for every cluster. A no-neighbor deconvolution was applied to each raw image (using Slidebook 4.2), to increase the signal-to-noise ratio and to even out the signal fluctuations in the background. The deconvolved images were then smoothed by applying a Gaussian spatial two-dimensional radial transformation (radius=1 pixel) and fast spatial changes in signal were eliminated by applying a Laplacian two-dimensional derivative. Finally, a mask corresponding to each spot in the images was obtained by ‘AND’ logical operation applied between the images resulting from the Laplacian transformation and the deconvolved images segmented by intensity. (2) Particle identification: determination of the spatial coordinates for all the centroids of every masked spot in any given image. (3) Particle tracking: the masked spots were connected in time, by allowing the software (Slidebook 4.2) to explore the position of every spot along the time series. All particles tracked were individually validated. Every dynamic spot was accepted, as far as they were individual objects (diffraction-limited) and all their ‘history’ can be known (from appearance to disappearance). The relative fluorescence intensity of each spot masked in the raw image corresponds to the difference in fluorescence signal intensity defined by the mask minus the average background signal of the cytosol. Calculation of several descriptors (lifetime, maximal intensity, density, etc.) was performed using Slidebook 4.2. Most caveolae appeared and disappeared without x-y movements. For the caveolae that were moving in x-y for few frames while they appeared or disappeared, only the stationary phase at the plasma membrane was used for calculation. To determine the ratio between arrivals versus departures of caveolae, all objects appearing or disappearing during the course of the time-series were separately counted. Statistical significances (P values) were calculated by unpaired two-tailed Student's t-test. P values less than 0.05 were considered significant.

We are grateful to Brigitte Joggerst, Nicole Schieber (R.G.P.) and Saveez Saffarian (E.B.) for technical assistance. R.G.P. made the initial observation and performed and analysed the electron microscopy experiments together with M.T.H. All the other experiments were performed and analyzed by E.B. The project was designed and supervised by R.G.P. and T.K. The paper was written by E.B. and R.G.P. with input from all the other authors. This work was supported by grants from the National Health and Medical Research Council of Australia (R.G.P.), the Human Frontier Science Program (R.G.P.) and the US National Institutes of Health grants GM 075252 and GM 62566 (T.K.). E.B. was a long-term fellow of the International Human Frontier Science Program Organization (HFSP). The authors acknowledge the use of facilities in the Australian Microscopy and Microanalysis Facility (AMMRF) at the Centre for Microscopy and Microanalysis at The University of Queensland. Deposited in PMC for release after 12 months.