Ras microlocalization was examined on intact 2-D sheets of apical plasma membrane, ripped off from adherent cells directly onto EM grids (Sanan and Anderson, 1991
; Parton and Hancock, 2001
). First, we investigated the distributions of GFP fused to the minimal plasma membrane targeting motifs of H-ras, GFP-tH (). GFP-tH, targeted by a combination of palmitoylation and a farnesylated CAAX motif, is an excellent lipid raft marker because even on high expression, it is localized exclusively to low density fractions on sucrose gradients (Prior et al., 2001
). We visualized GFP-tH in plasma membrane sheets with anti-GFP–5 nm gold ( a). GFP-tH is extensively distributed in small, morphologically featureless patches. To analyze the complete gold patterns as rigorously and objectively as possible, we used statistical methods for point pattern analysis that are well established, but rarely used in cell biology. Ripley's K-function (Ripley, 1977
; Philimonenko et al., 2000
) evaluates an exhaustive map of all interparticle distances over the study area and compares the observed distribution with that expected from complete spatial randomness (CSR).*
The analysis, which focuses on the second-order properties of the gold pattern, has the advantage that it analyzes spatial structure at multiple ranges simultaneously (Ripley, 1977
; see Materials and methods and supplementary data).
Figure 1. Visualizing lipid rafts using electron microscopy and spatial point pattern analysis. (a) Anti-GFP labeling is specific, ending at the edge of a typical GFP-tH sheet. (b) 60 min of 1% cyclodextrin treatment depletes cell surface cholesterol, detected (more ...)
K-function analysis of GFP-tH sheets shows that the gold pattern is clustered ( c, red line), i.e., the curve shows significant positive deviation from the L
) − r
= 0 value expected for a random point pattern. The maximum deviation of the GFP-tH L
) − r
curve from CSR occurs at a radius of 22 nm. From these data, we can model the size and distribution of GFP-tH microdomains (Figs. S1 b and S2, available at http://www.jcb.org/cgi/content/full/jcb.200209091/DC1
); we estimate that they are domains with mean radius of 22 ± 4 nm that occupy 35% of the plasma membrane. The mean radius of 22 nm is within the range of lipid raft size derived by other methods, which estimated diameters of <70 nm (Friedrichson and Kurzchalia, 1998
; Varma and Mayor, 1998
; Pralle et al., 2000
). Detection of clustering is extremely sensitive to fixation and labeling procedures. Glutaraldehyde fixation must be used to eliminate short-range antibody-induced aggregation into larger clusters (Fig. S3 a), and clustering is only evident when gold particles ≤6 nm in diameter are directly conjugated to primary antibody (Fig. S3 b).
We tested if GFP-tH microdomains are cholesterol-dependent by treating cells with methyl-β-cyclodextrin. Depletion of cell surface cholesterol, visualized by filipin staining ( b), did not cause any loss of GFP-tH from the plasma membrane, assessed qualitatively by fluorescence or quantitatively by immunogold labeling (not depicted). However, K-function analysis of the gold patterns reveals a time-dependent loss of GFP-tH clustering in cyclodextrin-treated cells ( c). After 60 min of cyclodextrin treatment, L
) − r
tracks at zero over most of the r
range analyzed, indicating a random distribution. These results confirm that GFP-tH is localized to cholesterol-rich lipid rafts and reveals that their disruption disperses GFP-tH over the plasma membrane, rather than driving association with other microdomains. The presence of rafts in the extracellular leaflet of the plasma membrane was supported by studies showing cholesterol-dependent clustering of glycophosphatidylinositol (GPI)-anchored proteins (Friedrichson and Kurzchalia, 1998
; Harder et al., 1998
; Varma and Mayor, 1998
), but similar data for rafts in the intracellular leaflet have been lacking until now.
Next, we examined the relationship between inner- and outer-leaflet lipid rafts using a variation of the K-function analysis. When plasma membrane sheets are labeled for two different antigens with 2 nm and 4–5 nm gold, colocalization can be assessed using bivariate K-functions that determine whether one gold population is clustered with respect to the other (Diggle, 1986
; see Materials and methods and supplementary data). We compared the distribution of GFP-tH with the outer-leaflet raft marker GFP-GPI; ). Both proteins are GFP-tagged, but because only one membrane surface is exposed at any point in the labeling and rip-off procedure, no leakage of gold probes occurs (unpublished data). We used two protocols to induce different degrees of GFP-GPI aggregation, as revealed by univariate K-function analysis of the 2-nm gold patterns ( c). The semi-patched technique induces relatively little GFP-GPI aggregation (univariate K-function shows a mean cluster radius of 50 nm), whereas the patched protocol, routinely used to visualize lipid rafts by immunofluorescence, induces very large GFP-GPI aggregates (univariate K-function shows a mean radius of 180 nm). It is not possible to completely evaluate unpatched GFP-GPI because this necessitates ripping off apical membranes from prefixed cells, a technique that has to date proven unsuccessful. The bivariate K-function shows that there is significant colocalization of GFP-tH with GFP-GPI only when GFP-GPI is aggregated into very large patches ( d); this is indicated by significant positive deflections of the Lbiv
) − r
curve from zero. Interestingly, despite the major reorganization of GFP-GPI, GFP-tH remains in small clusters ( b; univariate K-function for GFP-tH). This result indicates that inner leaflet rafts, when aggregated by cross-linking GPI-anchored proteins, still retain their modular structure. Incomplete colocalization of inner- and outer-leaflet raft markers has been observed previously (Harder et al., 1998
; Prior et al., 2001
), but our new data show that inner and outer leaflet rafts are only loosely associated at steady state. This offers a potential new mechanism for regulating signaling by modulating the extent of coupling between inner and outer leaflet rafts.
Figure 2. Analysis of inner- and outer-leaflet lipid raft markers. (a) Inner-leaflet GFP-tH (5 nm gold) and outer-leaflet GFP-GPI (2 nm gold) were specifically labeled to visualize individual microdomains. Univariate K-function analysis of GFP-tH (b) and GFP-GPI (more ...)
Full-length GFP–H-ras and GFP–H-rasG12V also have clustered distributions when expressed in BHK cells (). However, in contrast to GFP-tH, cyclodextrin treatment has relatively small effects on the clustering of GDP-bound H-ras ( a), GTP-bound H-rasG12V ( b) or GFP-targeted by the COOH-terminal 25 amino acids of H-ras (GFP-CTH; unpublished data). These data are consistent with a model where full-length H-ras has affinity for, and is resident in at least two plasma membrane microdomains; lipid rafts, and a noncholesterol-dependent nonraft microdomain (Jaumot et al., 2001
; Prior et al., 2001
). To formally test this hypothesis, we used bivariate K-function analysis to study the interaction of H-ras with lipid rafts. GFP-tH was coexpressed with untagged H-ras and plasma membrane sheets labeled with anti-GFP–2 nm and anti-Ras–4 nm gold. Bivariate analysis of the gold patterns ( d) reveals extensive colocalization of GDP-bound H-ras with GFP-tH in serum-starved cells, and shows that this decreases with serum stimulation. Constitutively activated H-rasG12V gives an even clearer picture of this phenomenon in that Lbiv
) − r
trends around zero, indicating no detectable colocalization of GFP-tH with H-rasG12V ( d). Together with the recent FRAP analysis of GFP-H-Ras in living cells, our results strongly suggest that wild-type H-ras is normally in a dynamic equilibrium between lipid rafts and other noncholesterol-dependent microdomains. Interaction with the nonraft site is favored experimentally by disrupting lipid rafts, i.e., removing one of the H-ras interaction sites, or physiologically by GTP loading. Thus, constitutively active H-ras is predominantly resident in the nonraft microdomain. The interaction of H-ras with the nonraft domain requires protein sequences in the hypervariable region because GFP-tH clustering is abolished by cyclodextrin treatment, whereas GFP-CTH is not.
Figure 3. H-ras also occupies nonraft microdomains. Clustering of GDP-bound H-ras (a; GFP-HG12) and activated H-ras (b; GFP-HG12V) changes little with cyclodextrin treatment. (c) Plasma membrane sheets expressing GFP-H-ras were labeled with anti-GFP–2 (more ...)
Recent studies have shown that galectin-1 interacts with the processed COOH terminus of activated H-ras and may stabilize plasma membrane binding (Paz et al., 2001
). Therefore, we examined whether galectin-1 might also be involved in regulating the microlocalization of activated H-ras. Transfection with antisense galectin-1 reduces endogenous galectin-1 expression ( e; Paz et al., 2001
) and significantly reduces the clustering of H-rasG12V ( f). These data suggest that galectin-1 may have a critical role in stabilizing H-ras interactions with nonraft microdomains that in turn may also be required for stable plasma membrane association of activated H-ras.
Little is known about the microlocalization of K-ras targeted by a polylysine domain and a farnesylated CAAX motif. Therefore, we investigated the distribution of GFP fused to the minimal plasma membrane targeting motifs of K-ras, GFP-tK. Analysis of the GFP-tK–labeled gold patterns reveals that they are clustered, but with different characteristics from GFP-tH ( a and Fig. S1 a). Modeling establishes that the GFP-tK domains have a mean radius of 16 ± 3 nm and occupy 20% of the plasma membrane (Fig. S1 c). In contrast to GFP-tH, cholesterol depletion causes a small rise in GFP-tK clustering after 60 min of cyclodextrin treatment. The subtle effect of cyclodextrin on GFP-tK microdomains may reflect a general role of cholesterol in maintaining overall plasma membrane integrity. Wild-type and constitutively active K-ras show identical clustering to GFP-tK, both in the presence or absence of cyclodextrin (unpublished data), and bivariate analysis of plasma membranes coexpressing GFP-tH and activated K-rasG12V showed no significant colocalization of the lipid raft marker with K-ras ( b). Clustering of GFP-tK and K-ras was unexpected, although biophysical studies have shown that myristoylated polybasic peptides can sequester negatively charged lipids to generate novel membrane domains (Murray et al., 1999
). Intriguingly, clustering was strikingly reduced when the wild-type K-ras CVIM motif was replaced with CCIL, to direct geranylgeranylation rather than farnesylation ( c, GFP-tKCCIL). Therefore, our data show that nature of the prenoid group profoundly affects the ability of polybasic K-ras to organize into specific microdomains, an observation that may have functional implications. Overall, our results clearly demonstrate for the first time the existence of K-ras microdomains within disordered plasma membranes that are distinct from classical lipid rafts.
Figure 4. K-ras clusters in nonraft microdomains distinct from H-ras microdomains. Peak clustering of GFP-tK occurs at 16 nm; clustering is increased slightly with cyclodextrin treatment (a; t = 0 min, open squares; t = 15 min, open triangles; (more ...)
Our analysis shows that neither activated H-ras nor K-ras localize to lipid rafts ( d and b). A direct test of colocalization in cells coexpressing activated H- and K-ras is not possible at present because of a lack of suitable antibodies, and the fact that short NH2
-terminal epitopes on Ras proteins attached to plasma membrane sheets are not visible to gold-conjugated antibodies (unpublished data). To indirectly address whether activated H-ras and K-ras occupy the same nonraft signaling domain, we used bivariate K-functions to examine colocalization to microdomains defined by the presence of GFP-tK. In this analysis, the Lbiv
) − r
curve for K-rasG12V coexpressed with GFP-tK shows significant deviation from zero, whereas the curve for H-rasG12V coexpressed with GFP-tK does not leave the 95% confidence interval ( d). This result indicates that a significant fraction of activated K-ras, but not H-ras, resides in microdomains labeled by GFP-tK. We can conclude from these data that the majority of activated H-ras does not colocalize with activated K-ras, indicating distinct (if not exclusive) locations for Ras microlocalization. The localization of activated H- and K-ras to different microdomains is important because if the distribution of effectors and/or concentration of cofactors for effector activation also varies across these distinct microenvironments, then differential signal output from H- and K-ras can be readily accounted for. For example, K-ras is a more potent activator of Raf than H-ras (Yan et al., 1998
). Therefore, we might expect cofactors in Raf activation (such as phosphatidyl serine) to be more enriched in K-ras compared with H-ras microdomains. The challenge now is to better define the protein and lipid content of these microdomains.
A recent study has shown GFP dimerization compromises FRET analysis of protein–protein interactions in intact cells and results in lipid raft–targeted CFP and YFP continuing to display FRET when cells are treated with cyclodextrin (Zacharias et al., 2002
). In contrast, we show here that GFP-tH clustering detected by immunogold labeling is completely abolished with cyclodextrin treatment; thus, if GFP is dimerized, then the gold-labeled antibody must detect only one GFP epitope in the dimer. Therefore, this loss of clustering is an important validation of the methodology, and clearly demonstrates that potential antibody or GFP dimerization–induced artifacts are not being observed. A similar argument applies to the loss of GFP-tK and GFP–H-rasG12V clustering induced by changing the prenoid group or reducing galectin expression, respectively, and an analysis of non-GFP tagged Ras clustering with anti-Ras–5 nm gold that yields very similar results to those observed with GFP-tagged Ras (unpublished data). Finally, we can also detect marked differences in the clustering of different transmembrane-anchored plasma membrane proteins, e.g., GFP-tagged EGF receptor is not clustered when expressed at relatively low levels (Fig. S3 c), whereas GFP-tagged Angiotensin II, Type 1 receptor, a G protein–coupled receptor, is highly clustered when expressed at comparable levels (unpublished data).
In summary, we have integrated powerful statistical analysis tools with high resolution imaging of morphologically featureless plasma membrane microdomains. This has allowed us to characterize both the size and distribution of lipid rafts and Ras signaling domains at steady state. Ras proteins are clustered irrespective of activation state, and occupy multiple raft and nonraft microdomains that are extensively distributed over the cell surface. K-ras requires the farnesyl moiety for correct microlocalization, whereas galectin-1 stabilizes H-ras interactions with nonraft microdomains. Our analysis shows that activated H-ras and activated K-ras largely occupy spatially distinct plasma membrane signaling domains; an observation that could account for differential signal outputs from these highly homologous proteins. Finally, the technique we have described has the potential to spatially map any epitope-tagged or endogenous signaling protein with respect to validated microdomain signposts. This will greatly facilitate studies dissecting the spatial organization and regulation of plasma membrane–based signaling pathways.