Multi-protein assemblies underlie the mesoscale organization of the plasma membrane

Abstract

Most proteins have uneven distributions in the plasma membrane. Broadly speaking, this may be caused by mechanisms specific to each protein, or may be a consequence of a general pattern that affects the distribution of all membrane proteins. The latter hypothesis has been difficult to test in the past. Here, we introduce several approaches based on click chemistry, through which we study the distribution of membrane proteins in living cells, as well as in membrane sheets. We found that the plasma membrane proteins form multi-protein assemblies that are long lived (minutes), and in which protein diffusion is restricted. The formation of the assemblies is dependent on cholesterol. They are separated and anchored by the actin cytoskeleton. Specific proteins are preferentially located in different regions of the assemblies, from their cores to their edges. We conclude that the assemblies constitute a basic mesoscale feature of the membrane, which affects the patterning of most membrane proteins, and possibly also their activity.

Figure 1. Proteins are organized into assemblies in the plasma membrane of living cells.

(a) The general principle of the protein labelling. The amino acid L-AHA is metabolically incorporated into cells for 72 h. The fluorophore DIBAC-KK114 is then covalently coupled to the proteins on the outer surface of COS-7 cells through a copper-free click reaction, allowing live investigation by various imaging approaches at 37 °C. (b) Confocal and STED micrographs of the protein organization in the membrane of living cells. Note the areas with high abundance of proteins (protein assemblies). Scale bar, 500 nm. (c) STED micrographs of plasma membrane proteins (DIBAC-KK114) and a lipid probe, cholesterol-polyethyleneglycol(PEG)-Atto490LS. The lipid probe is homogeneously distributed, few exceptions likely representing organelles or membrane infoldings. The correlation of the protein and lipid labelling is low; the Pearson correlation coefficient (PCC) is 0.040±0.009 (mean±s.e.m., n=9 cells). Scale bar, 2 μm. (d) Analysis of protein mobility by FRAP. A confocal image sequence displays the protein signal before (‘pre-bleach’) and immediately after (‘post-bleach’) photobleaching, and after 90 s of recovery. Scale bar, 1 μm. The change of intensity over time is plotted after background subtraction and correction for imaging-induced bleaching (mean±s.e.m., n=87 recovery traces from 30 cells in three independent experiments). (e) Analysis of protein mobility by scanning STED-FCS measurements. The confocal micrograph shows the overview of the bottom membrane of a living COS-7 cell. Scale bar, 2 μm. The detection spot (90 nm in diameter) was continuously scanned (4 kHz) on a circular trajectory (red circle) over the membrane. (f) The scanning trajectory was subdivided into 50 pixels, and a temporal autocorrelation analysis was performed on each pixel. The correlation curves were fitted (red) according to the standard two-dimensional FCS diffusion model. From the fit for each position along the scanning trajectory, the amplitude and the half-time of decay were obtained, which were used to calculate the local protein density and diffusion coefficient, respectively. (g) Scatter plot of the protein density against the protein mobility (diffusion coefficient D) shows a significant negative correlation. The PCC is −0.29±0.02 (mean±s.d., n=500 data points from 10 cells in three independent preparations).

Figure 2. Cell–cell fusion experiments show that protein assemblies are long-lived.

(a) COS-7 cells that incorporated methionine analogues, AHA (red) or HPG (green) for 60 h were plated together and cell–cell fusion was induced by HVJ-E. Membrane sheets were generated from the fused cells, fixed, and labelled by click reaction. (b) Two possible scenarios are anticipated. Top, in absence of stable assemblies, the fused membrane would contain both red and green proteins with similar distributions. A line scan through the membrane would show that the green and red signals correlate. Bottom, if long-lived assemblies exist, then the fused membrane would initially contain red assemblies that exclude green proteins, and vice versa. The areas between the assemblies should contain more mobile proteins that intermix faster. (c) Two-colour STED image of a typical membrane sheet obtained after fusion. Following fusion, cells were incubated at 37 °C for 5 min before generation and fixation of membrane sheets. The three-dimensional views (prepared using a metallic, non-transparent colourmap in Matlab) show the signal distributions for an HPG-dominated area (top) and AHA-dominated area (bottom). (d) 4.65 μm long (three-pixels wide) line scans were drawn through green (top) or red (bottom) protein assemblies, at the positions indicated by the dashed lines in c. Note the low correlation of red proteins with green assemblies, and similarly of green proteins with red assemblies. (e) Alternatively, following fusion, cells were further incubated for 60 min. Line scans were drawn as in d, at the position marked by the dashed line. Note the increased correlation. Scale bars, 2 μm. (f) Using line scans, we analysed the enrichment of red- and green-labelled proteins in the assemblies of the opposite colour (normalized to their own baseline). At 5 min, the dominant protein is substantially more enriched in its own assemblies than the subordinate colour (for the green side of the fused membrane, green is dominant and red is subordinate). The differences were statistically significant (*P<0.05, t-tests). At 60 min, there is no clear dominance, so the same regions of interest were analysed either with normalization to green or red signal. The enrichment was much more even (n=15–19 fused cells from three independent experiments).

Figure 3. Protein assemblies in plasma membrane sheets.

(a) Left, STED micrograph of the protein organization in PC12 membrane sheets. AHA was metabolically incorporated into PC12 cells for 72 h before generation of membrane sheets. The sheets were then fixed and subject to copper-catalysed click reaction with alkyne-functionalized Atto647N. Right, a high-zoom view of the area marked with the white square on the left. Scale bars, 2 μm and 500 nm. See Supplementary Fig. 1a–c for controls showing the protein labelling specificity, and the reproducibility of the protein pattern under different fixation schemes. (b) The graph shows the assembly size (apparent diameter) for protein assemblies from live and fixed cells, and from membrane sheets (NS, non-significant; ****P<0.0001, 1-way analysis of variance, Bonferroni’s multiple comparison test, n=573–6,812 protein assemblies. See Supplementary Methods for details of the analysis.

Figure 4. Protein assemblies in membrane sheets contain proteins of different ages and classes.

(a) Two-colour STED imaging of proteins of different ages. PC12 cells were incubated first with AHA for 72 h (old proteins, green), then starved of methionine for 1 h, and incubated with HPG for 3 h (new proteins, red). Two click reactions were applied sequentially coupling first HPG to Atto647N-azide, then AHA to Chromeo494-alkyne. To analyse the relative distributions of the two labels line scans centred on the green AHA spots were performed. After normalization to the baseline, the scans were aligned using their maxima in the green channel as reference. The graph on the left shows averaged line scans from 856 protein assemblies (mean±s.e.m.). Note that the newer proteins are less strongly enriched in the centres of assemblies. The same observation was made when the two fluorophores are swapped (graph on the right), coupling AHA to Atto647N-alkyne (old, red) and HPG to Chromeo494-azide (new, green), n=768 assemblies. (b,c) Distributions of different classes of post-translationally modified proteins (red) carrying various membrane anchors (b) or glycosylations (c), with respect to general protein assemblies (green). In addition to AHA or HPG, PC12 cells were treated with clickable versions of the indicated post-translational modifiers, and were sequentially labelled by click reactions. A typical two-colour STED image for palmitoyl labelling is shown. For each condition, a plot was generated as described above. n=3–4 independent experiments, with a total of 629–1,373 analysed protein assemblies. A reference trace for the protein assemblies (black hollow circles) was obtained by averaging the protein profiles from all the conditions represented in each graph. See Supplementary Fig. 2 for the analysis of the GPI-anchor. (d) Distribution of lipid probes in relation to protein assemblies. After click labelling of proteins with Chromeo494, membrane sheets were incubated with cholesterol-PEG-KK114 or Atto647N-PE at RT for 30 min (see Supplementary Methods for more information about the lipid probes). A typical image for cholesterol analogue is shown. Alternatively, phophatidyl-inositol 4,5-bisphosphate (PIP2) was probed through immunostaining (see Supplementary Methods). For the two fluorescent lipid analogues, n=3–4 independent experiments, with a total of 1,373–1,675 protein assemblies. For PIP2 immunostaining, n=285 assemblies. The error bars are occasionally smaller than the data point symbols. Scale bar, 750 nm.

Figure 5. Protein assembly organization persists after many harsh treatments.

STED images of protein assemblies in PC12 membrane sheets subjected to different treatments either applied on sheets before fixation (a,b) or onto live cells before generation of sheets (c). (a) Membrane sheets were either incubated in sonication buffer (which mimics the intracellular conditions, leftmost image) as control or treated with buffers of high-salt, low ionic strength or high-calcium, for 20 min at 37 °C (see Methods for buffer compositions). Scale bar, 500 nm. (b) Lipid composition of the membranes was manipulated by applying 1–2 units ml⁻¹ phospholipases (PL) A2, C, D and phosphatidylinositol (PI)-specific PLC (not shown) or 20 units ml⁻¹ SPMase diluted in sonication buffer, for 15 min. Compare with the first image in a (albeit incubated for 20 min, rather than 15 min; an image of a 15-min incubated membrane sheet is shown in Fig. 6a). (c) For treatments on full cells, controls were prepared by chemically fixing the membrane sheets directly after sonication. For ionic perturbation, cells were treated for 5 min with 1 μM of ionomycin in HEPES-buffered medium containing 5 mM Ca²⁺, with or without 10 mM EGTA. In independent experiments, the abundant membrane proteins syntaxin 1 or SNAP25 were knocked down by siRNAs. See also Supplementary Fig. 3a–c for the effect of overexpression of proteins. To manipulate the cytoskeletal elements, cells were treated for 1 h with 0.1% dimethyl sulfoxide (as solvent control, not shown), 10 μM nocodazole, colchicine or cytochalasin D or 1 μM latrunculin A. (d) For all the conditions, the area occupied by assemblies was quantified and expressed as % of the respective control for each condition. (e) The graph indicates the assembly sizes (average assembly area) for each condition as fold increase over their respective control conditions. (d,e) n=3–5 independent experiments, means±s.e.m. for all treatments, except for the knockdown (KD) conditions, where means±range of values (from two independent experiments) are shown. Data analyses are explained in the Supplementary Methods. The strongest increases in size are seen under treatments of low ionic strength, SPMAse, latrunculin A or cytochalasin D, all of which affect the membrane actin cytoskeleton.

Figure 6. Cholesterol controls the formation of protein assemblies.

(a) The panels show STED images of PC12 membrane sheets treated with sonication buffer (left; 15 min at 37 °C), or subjected to cholesterol depletion with MBCD (middle; 5 mM, 5 min, followed by incubation in sonication buffer for 10 min at 37 °C). To replete cholesterol, after treatment with MBCD (5 mM, 5 min), cholesterol was loaded by applying MBCD-cholesterol (right; 5 mM, 10 min). Scale bar, 500 nm. To quantify the effects, we calculated the area occupied by assemblies as in Fig. 5d but here refer to the area poor in protein, as a an indication for the loss of assemblies. This fraction is predicted to be 0 on the complete dispersion of proteins from the assemblies. All values were expressed as % of control (buffer incubation alone). The bars show the following treatments: MBCD treatment as above, 0.2 units ml⁻¹ cholesterol oxidase (COase) for 15 min, 25 mM alpha- or beta-hydroxypropyl-CD for 5 min, cholesterol replenishment as above. n=3–7 independent experiments (*P<0.05; **P<0.01, 1-way analysis of variance, Dunnett’s multiple comparison test). The variance of the data was similar (0.34, 0.24, 0.44, 0.33, 0.37 and 0.74, for the six conditions, respectively). Note the loss of protein-free areas on cholesterol depletion, and their reformation (along with protein assemblies) on cholesterol repletion. (b) STED images of a normal (control) early endosome and of an early endosome treated with MBCD (left panels). STED images of a control mitochondrion and of a mitochondrion after loading with cholesterol (right panels). The images were processed by deconvolution for display purposes. Scale bar, 250 nm. The graph shows the quantification of the clustering of membrane proteins in endosomes and mitochondria, based on the coefficient of variation in fluorescence intensity along the line scans over the organelle membranes. n>110 organelles per condition, from multiple independent experiments (***P<0.001, two-sample Kolmogorov–Smirnov tests; this test was chosen since not all data sets passed a Jarque–Bera normality test). The variance of the data was similar (0.33, 0.54, 0.27 and 0.24, for the four data sets, respectively). See Supplementary Methods for the preparation and treatment of organelles.

Figure 7. Distributions of several proteins within the protein assemblies.

(a) Selected proteins were fluorescently labelled through immunostaining (labelled red) on AHA-containing (labelled green) PC12 plasma membranes. The STED images indicate that all of these proteins co-localize with the assemblies to a good extent, but are enriched in different positions relative to assembly centres. Scale bar, 750 nm. (b) Analyses of the distributions of specifically labelled proteins, relative to the assembly centres. Protein distributions were categorized based on the widths of their profile: proteins having a tighter distribution (smaller full width at half maximum) than the general protein labelling were placed in the ‘Centre preference’ category, while those that had similar distributions to the protein assemblies were described as ‘No preference’ (no preferred distribution within the assemblies). Protein distributions whose peak was away from the centre of the protein assemblies were categorized according to the ratio between their intensity at the centre of the protein assembly and their peak intensity. Those where the central point represented more than 60% of the peak value were described as having a ‘Moderate edge preference’, while those where the central point was below this value were described as having an ‘Edge preference’. The graphs indicate averaged normalized line scans, from 300–900 protein assemblies (mean±s.e.m.). The line scan profile for the assemblies (average for all stainings in each category) is shown in black (hollow circles), as a reference. See Supplementary Fig. 3b for stainings on COS-7 membranes, and Supplementary Fig. 5 for control stainings with different types of probes, and under different sample preparation conditions.