GPI-anchored proteins are confined in subdiffraction clusters at the apical surface of polarized epithelial cells

Spatio-temporal compartmentalization of membrane proteins is critical for the regulation of diverse vital functions in eukaryotic cells. It was previously shown that, at the apical surface of polarized MDCK cells, glycosylphosphatidylinositol (GPI)-anchored proteins (GPI-APs) are organized in small cholesterol-independent clusters of single GPI-AP species (homoclusters), which are required for the formation of larger cholesterol-dependent clusters formed by multiple GPI-AP species (heteroclusters). This clustered organization is crucial for the biological activities of GPI-APs; hence, understanding the spatio-temporal properties of their membrane organization is of fundamental importance. Here, by using direct stochastic optical reconstruction microscopy coupled to pair correlation analysis (pc-STORM), we were able to visualize and measure the size of these clusters. Specifically, we show that they are non-randomly distributed and have an average size of 67 nm. We also demonstrated that polarized MDCK and non-polarized CHO cells have similar cluster distribution and size, but different sensitivity to cholesterol depletion. Finally, we derived a model that allowed a quantitative characterization of the cluster organization of GPI-APs at the apical surface of polarized MDCK cells for the first time. Experimental FRET (fluorescence resonance energy transfer)/FLIM (fluorescence-lifetime imaging microscopy) data were correlated to the theoretical predictions of the model.

(B) Histograms show the distribution of mean squared errors between the data and the fitted models for n distinct regions of 4x4 μ obtained from m independent experiments and an average of k cells per experiment. The blue histogram corresponds to the random model, the red histogram corresponds to the clustered model. Overlaps between the two histograms appear in grey. A Kolmogorov-Smirnov test is used to compare the two error distributions and assess if the clustered model (equation [2] in Experimental procedures) provides a significantly better fit to the data than the random model. The resulting p-value is reported in each case and does not reveal significantly different distributions (p>0.1). This indicates that the data are equally well fit by a random model than a more complex clustered model.  Pointillist images (upper panels) and corresponding pc-PALM curves (lower panels) of simulated random (left panels) and clustered (right panels) molecular distribution.
While pointillist images are very similar for random and clustered distribution, pc-PALM analysis shows that the two molecular distributions are different. Indeed, in the case of simulated random distribution (left panel) random model (blue curve) fits better than the clustered model (red curve) the pc-PALM curve (circles), the contrary in the case of simulated clustered distribution (right panel).

Figure S3: Evaluation of protein diffusion in STORM experiments
Boxplots show evaluation of the clusters size in MDCK GFP-FR experiments for the full set of frames (1-30,000) or by selecting either the first 10,000 frames or the last 10,000 frames. The cluster sizes were measured from 42 pc-PALM curves (corresponding to 42 regions of size 4 μ μ ). Note that there is no difference in the cluster size of GFP-FR when we considered either the whole set of images or the first or last 10,000 frames. This indicates that potential diffusion of molecules during acquisition time has a negligible influence on cluster size.

Figure S4: Evaluation of temporal appearance of localizations for random and clustered distributed proteins in STORM experiments
Colour coded localization images reflecting the time frames of detected localizations for p75-NTR and PLAP (in MDCK and CHO cells, respectively). To generate this coloured image, we first merged localizations occurring in consecutive frames within a radius of 100 nm into a single localization, in order to reduce the apparent clustering due to repeated detection of the same molecule. However, our molecules typically blink until the fluorophore bleaches permanently and this can happen multiple times. Therefore, each of these ON times could results as distinct pixel because of localization errors and consequently multiple ON/OFF switching events can lead to apparent cluster of pixels. To avoid this, we implemented 2D colour coding in which pixels are characterized by both hue (from the average of the frame number of localizations within that pixel) and saturation (from the standard deviation of these frame numbers). Early time frames correspond to blue/violet hue, and later time frames to red hue. Saturated colours (left of the colour bar) indicate that localizations occurring within the pixel are temporally close, whereas unsaturated (whitish) colours (right of colour bar) indicate that localizations occurred at different times during STORM imaging. Because multiple fluorophores within a protein cluster are expected to be stochastically activated at independent time points, they should appear as mixtures of pixels with different colours, or white (unsaturated) pixels, whereas isolated molecules should appear as pixels of homogeneous saturated colour. Moreover, because we observed that the time between the first fluorescent activation of the molecule and its permanent photobleaching is apparently much smaller than the duration of the image sequence acquisition multiple ON/OFF switching events (due to the blinking) will result as pixels of homogeneous saturated colour.
For p75-NTR, the image shows isolated clusters with predominantly homogeneous colours, consistent with pc-PALM analysis, which identified these molecules as having a random distribution (see Figure S1). For PLAP, the image shows predominantly clusters with a mixture of colours and/or less saturated colours, consistent with pc-PALM analysis, which identified this protein as organized in clusters. The pixel size is 36 nm and the scale bar is 2 μm.