The GFP-GPI plasmid (a kind gift from Jennifer Lippincott-Schwartz, National Institutes of Health, Bethesda, MD) was constructed from the GPI signal sequence of the human folate receptor in the eukaryotic expression vector pJB20. Details on the construct and the cell cultures used are provided under supplemental “Materials and Methods.”

Detergent-resistant microdomain separation was performed for CHO cells via sucrose gradient centrifugation (24) and for Jurkat T cells using the flow cytometric assay of detergent resistance (25). Detailed descriptions are provided under supplemental “Materials and Methods.”

CHO cells were washed two times with a 37 °C Hanks' buffered salt solution (HBSS; PAA Laboratories) and mounted in a POCmini chamber system (LaCon, Staig, Germany) filled with HBSS. BODIPY FL C₅-ganglioside G_M1 (B-13950, Invitrogen) was diluted in HBSS and used to label CHO cells at various concentrations ranging from 50 to 500 nm at 4 °C for 20 min on ice. Additional washing steps were applied to remove unbound BODIPY-G_M1.

Labeling of Jurkat cells was performed as described previously (26) and under supplemental “Materials and Methods.” For cholesterol depletion experiments, cells were washed twice with HBSS and incubated with 10 mm methyl-β-cyclodextrin (MβCD; C4555, Sigma) at 37 °C for 20 min. Cholesterol oxidase (C8649, Sigma) was applied at 37 °C for 30 min at 2 units/ml (CHO cells) or at 10 units/ml (Jurkat T cells). For BODIPY-G_M1 experiments, cholesterol depletion was performed before staining. The extent of cholesterol depletion was determined using the Amplex Red cholesterol assay kit (A12216, Invitrogen).

For cholesterol replenishment experiments, cells were first depleted using MβCD via the above protocol. Cholesterol (700000P, Avanti Polar Lipids, Alabaster, AL) was replenished by incubation with 10 mm cholesterol-loaded MβCD for 30 min at 37 °C in HBSS following the protocol of Yancey et al. (27). For experiments with BODIPY-G_M1, cells were first replenished with cholesterol and subsequently labeled with the lipid analog.

For cleavage of GPI-anchored proteins, we incubated mGFP-GPI-expressing CHO cells for 3 h with 0.5 units/ml phosphatidylinositol-specific phospholipase C (P5542, Sigma) in HBSS at 37 °C. Trypan blue (33595, Sigma) was added at 0.4% in HBSS at 37 °C and measured immediately.

Purified His-mGFP-GPI(CD55) was a kind gift from Christoph Metzner (Institute for Virology, University of Veterinary Medicine Vienna, Wien, Austria). The protein was purified as described (28), except for using a HEK293 cell line as the expression system. Supported lipid bilayers were formed from vesicles as described under supplemental “Materials and Methods” (29).

mGFP-GPI-expressing CHO cells at 60–70% confluency were subjected to heat at the indicated temperatures (±0.1 °C) in a water bath for 1 h. After a 20-h incubation at 37 °C in a 5% CO₂ atmosphere, cells were washed with PBS and lysed with Laemmli sample buffer. Western blotting was performed as described under supplemental “Materials and Methods.”

Single molecule experiments were performed on a setup described in detail under supplemental “Materials and Methods” (21). Temperature was controlled using a home-built incubator box equipped with a heating unit, a temperable stage insert, and an objective heater (PeCon, Erbach, Germany). For heat shock experiments, cells were grown at 37 °C and adjusted to the indicated temperature immediately before the experiments.

After recording a pre-bleach image with an illumination time of 1 ms, samples were bleached with a laser pulse applied for 200–450 ms. After a variable recovery time of 600–2400 ms, a sequence of up to 10 images was recorded at an illumination time of 1 ms with a typical delay of 20 ms between subsequent images. The first image after bleaching was used for brightness analysis and the consecutive images for determination of mobilities. Because of the small bleaching area, multiple runs could be performed on a single cell without significant reduction of the total amount of active fluorophores on this cell.

The single molecule brightness was obtained after multiple runs of thinning out clusters while conserving the stoichiometry of labeling (TOCCSL) by extensively photobleaching the same cell for seconds. In this case, the probability for observing more than one active mGFP-GPI molecule per spot became negligible. Alternatively, we used the brightness values obtained in the last image of a sequence, which, due to photobleaching, contained a reduced number of peaks; also in this case, the likelihood that two molecules of dimer remained active was negligible (supplemental Fig. 3A). Both methods yielded the same results.

For the precise control of all laser pulse trains, an acousto-optical modulator (1205C, Isomet, Springfield, VA) was used. Timing protocols were generated and controlled by an in-house program package implemented in LABVIEW (National Instruments, Austin, TX).

For single molecule analysis, images were analyzed using in-house algorithms implemented in MATLAB (MathWorks, Natick, MA) (30). Individual diffraction-limited signals were selected and fitted with a Gaussian profile, yielding the single molecule position, the brightness (B), and the full width at half-maximum of the Gaussian function. Single molecule mobility and brightness were analyzed as described under supplemental “Materials and Methods” and in Refs. 21, 30, and 31.

The surface density of BODIPY-G_M1 was determined by dividing the pre-bleach signal by the average single molecule brightness (ρ₁(B)). α-Distributions were fitted with a generalized Poisson model that accounts also for observation of non-raft resident free probes (Equation 1). In this model, λ specifies the expectation value of the Poisson distribution, P the probe partitioning into the non-raft phase, and H the unit step function. Mobile fractions were measured via standard fluorescence recovery after photobleaching experiments: the equilibrium ensemble signal obtained after photobleaching was normalized by the pre-bleach signal.

We generated a CHO cell line stably expressing the consensus raft marker mGFP-GPI (8, 14, 32). Using sucrose density gradient centrifugation or a flow cytometric assay of detergent resistance (25), we found similar mGFP-GPI partitioning into the raft phase as reported previously (compare supplemental Fig. 1 and Ref. 32). The probe was specifically targeted to the plasma membrane, yielding the expected homogenous staining without enrichment in resolvable microdomains (Fig. 1, A and B). We were interested in whether the apparent homogenous signal was caused by hundreds of mobile nanoscopic platforms per diffraction-limited area. In our approach, we aimed at detecting such platforms via their property to assemble specific markers over time scales of seconds. It was therefore necessary to populate each platform with multiple markers, so we selected for cell lines with high mGFP-GPI surface staining. To virtually dilute stable mobile fluorescent assemblies without altering their properties, we made use of a recently introduced photobleaching protocol (TOCCSL) (21).

Fig. 1 (B–D) illustrates the method: the upper panels show the sample during the TOCCSL sequence, and the lower panels sketch the effect on putative membrane platforms. After recording an image of the undisturbed plasma membrane (Fig. 1B), a narrow stripe is irreversibly photobleached (Fig. 1C). High contrast in photobleaching is achieved by imaging a field stop onto the sample, which ensures minimal photobleaching of structures located outside the illuminated region; we used here a slit aperture with a width of 7 μm in the specimen plane. Diffusional broadening is minimized by applying short photobleaching times and high laser power. Brownian motion of the platforms leads to the recovery of the fluorescence signal (33). At the onset of this process, the first platforms diffusing into the field of view can be imaged directly as individual isolated fluorescent spots (Fig. 1D). To minimize contributions from the cytosol or the top cell membrane, we performed all experiments in total internal reflection configuration at the glass slide/cell interface. Fig. 1D further shows that the recovery proceeds from the masked region, with the central region of the field of view being devoid of signals. This observation confirms that the detected signals correspond to plasma membrane-associated mGFP-GPI; a potential vesicular pool would have traversed the evanescent field directly from the top, thereby repopulating the photobleached area uniformly over the field of view (see also supplemental “Discussion,” Subsection 1, for further controls).

We studied the mGFP-GPI association state by analyzing the brightness of individual fluorescent spots observed in the TOCCSL images (Fig. 1D). A broad distribution of brightness values (B) was obtained as expected from the stochastic nature of the photon emission process (Fig. 1E; see supplemental Table 2 for information on sample sizes). For comparison, we also plotted the brightness distribution of single mGFP-GPI molecules, which was determined from the same experiments upon additional photobleaching (see Microscopy under “Materials and Methods” and supplemental Fig. 3A). Single molecule signals showed a substantially reduced brightness compared with spots obtained in the TOCCSL images, indicating that the latter represent a mixed population of mGFP-GPI monomers and homo-associates.

For quantification of the association state of mGFP-GPI, we fitted ρ(B), the brightness distribution obtained in the TOCCSL images, by a linear combination of the n-mer brightness distributions (ρ_n(B)); the weights (α_n) specify the fraction of n-mers (see Data Analysis under “Materials and Methods”). The majority of spots were found to correspond to mGFP-GPI monomers (α₁ = 67.8 ± 0.8%) (Fig. 1E). Importantly, α₂ = 32.3 ± 0.8% of the spots contained two mGFP-GPI molecules, indicating the presence of stable homo-association of GPI-anchored proteins in this system. Coefficients (α_n>2) representing three or more molecules per spot did not yield detectable contributions here.

Cholesterol has been ascribed an essential role for the targeting of GPI-anchored proteins to membrane rafts (1, 14, 17). We therefore tested whether cholesterol depletion affects the observed mGFP-GPI homo-association. Indeed, a reduction of the cellular cholesterol content by 30% using either MβCD or cholesterol oxidase fully abolished mGFP-GPI homo-association (Fig. 1, F and H), consistent with the interpretation that the platforms were disrupted or the partitioning to the platforms was abolished. The effect was reversible, as cholesterol replenishment recovered the dimer fraction essentially to the level of the untreated control samples (Fig. 1G). In further controls, we ruled out that laser excitation (supplemental Fig. 3B) or GFP self-association could account for the observed results (see supplemental “Discussion,” Subsection 2).

We next addressed the dependence of α₂ on the surface density of mGFP-GPI (σ). It was reported previously using standard wide-field microscopy that the degree of protein nanoclustering did not depend on the average cell brightness (14, 15). We attempted to obtain a more accurate measure for σ by using the total internal reflection fluorescence pre-bleach image from each cell. A high cell-to-cell variability of both α₂ and σ was found (Fig. 1H), and interestingly, we observed a small but clear increase in α₂ with σ. Apparently, cells expressing higher amounts of mGFP-GPI in their plasma membrane also show a higher cargo load per platform.

In recent years, there was an opinion shift under raft proponents from long- toward short-lived rafts, with lifetimes speculated in the submillisecond range (4, 5). We therefore used our method to address the cargo exchange dynamics between different platforms. All fluorescent structures observed in the TOCCSL image had multiple encounters with photobleached structures during the recovery time, with multiple chances for cargo exchange. In particular, subsecond disintegration and reformation of structures hosting non-bleached mGFP-GPI would lead to cargo equilibration with photobleached platforms. The result would be a decrease in the single spot brightness (B) and concomitantly the dimer weight (α₂) with increasing recovery times. However, we found no dependence of α₂ on the recovery time (correlation coefficient = 0.037) (supplemental Fig. 5A). We conclude that the lifetime of the observed platforms substantially exceeds the maximum recovery time used in our experiments of 2.2 s.

We next questioned whether the loading of the observed structures correlates with their size. In particular, we were interested in whether the observed mGFP-GPI monomers resided in similar platforms as clusters of higher order or represented a population leaking into the non-raft bulk phase. The mobility of a diffusing tracer in a lipid membrane decreases with increasing size (34, 35), rendering the diffusion constant (D) a valid measure for the size of the moving structure. On average, the motion of mGFP-GPI was characterized by D = 1.30 ± 0.08 μm²/s (supplemental Fig. 6A), in agreement with the recent literature (17). We detected no difference in the mobility of the mGFP-GPI species observed in the TOCCSL images versus results obtained on single molecule data recorded after extensive photobleaching of the cells (D = 1.20 ± 0.17 μm²/s) (supplemental Fig. 6B), indicating that the molecules observed in the TOCCSL images are representative for the mobile fraction of plasma membrane-associated mGFP-GPI (76 ± 5%) (supplemental Table 1). Consistent with recent literature (36, 37), the mobility of mGFP-GPI decreased substantially upon cholesterol depletion (supplemental Fig. 6C). To check for different populations of mGFP-GPI, we compared the brightness with the mobility of each diffusing spot (supplemental Fig. 5B; see supplemental Fig. 7A for the correlation plot on cholesterol-depleted cells). We found hardly any correlation between the two parameters (correlation coefficients of −0.085 and −0.089 for untreated and cholesterol-depleted cells, respectively), indicating that the observed spots are similar in mobility and therefore in size, irrespective of whether they are loaded with one or two mGFP-GPI molecules. Our finding is thus consistent with the interpretation that all observed signals correspond to similar platforms.

Having shown the homo-association of an ectopically expressed protein, we next attempted to confirm the results using an externally applied probe. We used BODIPY-G_M1, a fluorescent lipid analog that was reported recently to be enriched in the liquid-ordered phase of plasma membrane vesicles (23). When applied to CHO cells, we found that BODIPY-G_M1 homo-associated in a cholesterol-dependent manner (Fig. 2A). We measured the association at different probe densities in the plasma membrane: at a surface density (σ) of 2600 μm⁻², we determined α₂ = 0.26 ± 0.02, which decreased linearly with decreasing probe densities. The linear dependence of homo-association on surface density is plausible for the externally applied probe: BODIPY-G_M1 appears to be incorporated stochastically into pre-existing platforms.

The mobility of BODIPY-G_M1 was similar to that of mGFP-GPI (D = 1.37 ± 0.08 μm²/s) (supplemental Fig. 6D). In particular, the same diffusion constant was obtained for BODIPY-G_M1 irrespective of whether the experiment was performed in TOCCSL mode or at low surface density using conventional single molecule tracking (supplemental Fig. 6, compare D and E). Considering the size dependence of diffusion (34, 35), we conclude that both experiments addressed similar diffusing objects. This finding further confirms that pre-existing platforms were labeled by the lipid analog.

Because rafts were particularly discussed as important structures in T cell signaling (38), we repeated the experiments on Jurkat T cells. For both mGFP-GPI and BODIPY-G_M1, we found clear cholesterol-dependent homo-association, similar to the results obtained with CHO cells (Fig. 2B and supplemental Fig. 4, C and D; data on diffusion constants can be found in supplemental Fig. 6, F–I).

When reducing the temperature to 25 °C, we observed a shift in the population toward larger structures, including up to 15% trimers and even tetramers, using again BODIPY-G_M1 as a probe on Jurkat T cells (Fig. 3A). The relative fractions of the observed n-mers closely followed a Poisson distribution (Fig. 3B), as expected when the probe was randomly distributed over pre-existing platforms. A fit yielded the average number of fluorescent cargo molecules per platform (λ); note that this parameter also accounts for structures that by chance remained empty. A priori, however, it remained to be shown whether a subfraction of BODIPY-G_M1 might also have partitioned into the bulk phase, where molecules would have moved as free monomers. We therefore tested also a generalized model that takes probe partitioning into the bulk phase into account. We reasoned that a potential non-raft fraction of BODIPY-G_M1 would appear as additional monomers and that the resulting n-mer distribution would deviate from the Poisson model. However, fitting the distributions shown in Fig. 3B by Equation 1 yielded partition coefficients close to zero (p = 0.1 ± 0.1), indicating that there is hardly any free BODIPY-G_M1 in the non-raft fraction of the Jurkat T cell plasma membrane at 25 °C.

The platform loading (λ) was found to increase linearly with the probe density (σ) (Fig. 3C, triangles; see also supplemental Fig. 7B). From the reciprocal slope of the curve, we were able to estimate the surface density of platforms, ζ = σ/λ = 1300 μm⁻². For comparison, also data obtained at 37 °C with Jurkat T cells and CHO cells were included, assuming the Poisson model for platform loading. Also, linear increases were observed yet with reduced slope. Note that in the latter cases, BODIPY-G_M1 partitioning into the bulk phase could not be excluded because analysis of the stoichiometric composition did not reveal cluster sizes (n) >2, rendering the calculation of a partition coefficient unfeasible. The reduction at 37 °C thus may either indicate a higher surface density of platforms (ζ) or may also be the consequence of BODIPY-G_M1 leakage into the bulk phase.

There is a long history of speculations for the involvement of lipid rafts in cell signaling (1). However, up to now, hardly any report has been available providing a direct link between raft properties and signal initiation. As a first field of application of our new technology, we attempted to test whether a rather general signaling mechanism, the expression of heat shock proteins upon mild heat stress, is correlated with a structural change in the plasma membrane. The rich phase behavior observed in binary or ternary model membranes (39) rendered the assumption of a temperature-induced switch in the formation of plasma membrane domains attractive. Indeed, mild fever-type heat shock response was speculated to be initiated by membrane raft reorganization (40, 41); a recent report showed that mild heating is sensed in the plasma membrane of plants cells (42). Up to now, however, there has been no direct experimental evidence for the influence of temperature on raft properties. For the following experiment, we reverted to our CHO cell line overexpressing mGFP-GPI. Consistent with previous reports (40, 43), we found differential heat shock response of small and large heat shock proteins: upon exposing the cells to mild heat shock of 39–42 °C, we observed elevated levels of the small heat shock protein Hsp27, whereas the expression of the large heat shock protein Hsp70 remained at basal levels (Fig. 4A). In the TOCCSL experiment, already a mild heat shock (39 °C) led to the almost total disappearance of mGFP-GPI homo-association, which became even more pronounced at 41 °C (Fig. 4B). This is a switch of remarkable steepness, centered exactly at the transition from physiological to fever-type temperatures. It is thus attractive to speculate that the plasma membrane may be involved in sensing small temperature elevations. Interestingly, in artificial bilayer systems, atomic force microscopy studies have shown that GPI-anchored proteins can be released from the liquid-ordered phase by an increase in temperature (44); a similar mechanism may account for the observed dissociation of mGFP-GPI homo-associates in the CHO cell membrane.