Partitioning of Lipid-Modified Monomeric GFPs into Membrane Microdomains of Live Cells

See allHide authors and affiliations

Science  03 May 2002:
Vol. 296, Issue 5569, pp. 913-916
DOI: 10.1126/science.1068539


Many proteins associated with the plasma membrane are known to partition into submicroscopic sphingolipid- and cholesterol-rich domains called lipid rafts, but the determinants dictating this segregation of proteins in the membrane are poorly understood. We suppressed the tendency of Aequoreafluorescent proteins to dimerize and targeted these variants to the plasma membrane using several different types of lipid anchors. Fluorescence resonance energy transfer measurements in living cells revealed that acyl but not prenyl modifications promote clustering in lipid rafts. Thus the nature of the lipid anchor on a protein is sufficient to determine submicroscopic localization within the plasma membrane.

Subcellular compartmentalization of signaling increases the specificity and efficiency of signal transduction. Caveolae and lipid rafts are related microdomains of the plasma membrane that are enriched in cholesterol, sphingolipids, and many signaling proteins (1, 2). Whereas protein-protein interactions maintain many signaling complexes (3, 4), specific lipid modifications are believed sufficient to sequester proteins in lipid rafts and caveolae. In particular, acylated proteins may preferentially partition into these compartments (5, 6). Unambiguous observation of these small (<100-nm) microdomains in living cells is beyond the resolution of visible light microscopy; thus destructive assays such as cellular fractionation (7) or immunolocalization by electron microscopy (8) have been relied upon to study lipid rafts and caveolae. We now use fluorescence resonance energy transfer (FRET) between nondimerizing cyan (CFP) and yellow (YFP) variants of Aequorea green fluorescent protein (9, 10) to show which lipid modifications are sufficient to cause such test proteins to aggregate within lipid rafts inside living cells. FRET from CFP to YFP is advantageous, because it nondestructively detects proximities at nanometer scales and because CFP and YFP have no targeting signals aside from the lipid anchors we added.

We fused CFP and YFP onto short peptides containing consensus sequences for acylation (11) or prenylation (12) (Fig. 1A). These constructs (13) were transfected with Lipofectin (GibcoBRL) into Madin-Darby canine kidney (MDCK) cells, and the cells were imaged at 80 to 90% confluency. Regions of interest were chosen from plasmalemmal sites where there was obvious cell-cell contact and little or no interference from fluorescence from intracellular membranes (Fig. 1B). The associative properties of coexpressed lipid-modified CFP and YFP pairs were determined by measuring the FRET efficiency (FRETE%) and fitting those values to the simplest possible saturable binding model, in which E% is a hyperbolic function of the surface density or local fluorescence intensity Fof the acceptor, YFP.Embedded Image(1)We preferred such a saturable isotherm over a linear fit with adjustable slope and intercept because the hyperbola also has two free parameters yet intrinsically satisfies the physical constraints that E% must approach zero as F approaches zero and that E% must level off no matter how much acceptor is expressed. Also, the parameter K is analogous to a dissociation constant and provides a natural criterion for the degree of clustering at any given concentration of acceptor (illustrated inFig. 2). When F ≪ K, FRET efficiencies are approximately proportional to acceptor surface densities, because the donors and acceptors are mutually randomly distributed. When F ≫ K, FRET efficiencies are nearly saturated and independent of further increases in acceptor density (14), because each donor is already clustered with at least one acceptor. FRET E% was determined by selectively photobleaching the acceptor YFP and measuring the resulting increase in brightness of CFP emission as a percentage of the final CFP intensity (15).

Figure 1

Lipid-modified fluorescent proteins used in this work and representative images of their expression in MDCK cells. (A) MyrPalm (myristoylated and palmitoylated), GerGer (geranylgeranylated), PalmPalm (tandemly palmitoylated), and caveolin (full-length bovine caveolin-1, triply palmitoylated with a putative membrane-embedded hairpin loop) fusion constructs were generated by polymerase chain reaction primer extension. Lipid attachment sites are shown in bold. (B) Typical fluorescence images of MDCK cells expressing each construct. Arrows denote representative sites for defining regions for data collection, on membrane sites of cell-cell contact remote from intracellular membranes. mCFP-GerGer and MyrPalm-mYFP images were taken from the same cells.

Figure 2

Theoretical expectations for unclustered (A and C) versus clustered (B andD) donors and acceptors (cyan and yellow circles representing mCFP and mYFP) in a region of interest of a cell membrane (black boxes).

Myristoyl and palmitoyl chains were attached to CFP and YFP by genetically fusing an acylation substrate sequence from the 13 NH2-terminal residues of the kinase Lyn to the NH2-termini of the fluorescent proteins. FRET efficiencies in regions of interest (ROIs) of cells expressing these constructs (MyrPalm-CFP and -YFP) showed that E% was saturated, indicating strong clustering (Fig. 3A). However, this clustered distribution survived extraction with 10 mM 5-methyl-β-cyclodextrin (MβCD) (Fig. 3B), a treatment known to disrupt lipid rafts and caveolae by depletion of cholesterol (16).

Figure 3

Determination by FRET of clustering between lipid-anchored fluorescent proteins. The least squares fits of the experimental data (each black square representing one region of interest) to Eq. 1 are shown as red curves, with 95% confidence intervals for the fit shown as flanking blue curves. The fitted value for K in Eq. 1 is given for each graph. The Kvalues should be compared to the acceptor intensities within the same experiment rather than to K's from other experiments, whose absolute intensity scales are not directly comparable. (A) FRET E% between MyrPalm-wild-type-CFP and MyrPalm-wild-type-YFP (K = 0.2 ± 1.2). (B) MyrPalm-wild-type-CFP and MyrPalm-wild-type-YFP after treatment with 10 mM MβCD (K = 10 ± 10). (C) MyrPalm-mCFP and MyrPalm-mYFP (L221K) in untreated cells (K = 6 ± 8). (D) MyrPalm-mCFP and MyrPalm-mYFP after MβCD (K = 1300 ± 1100). (E) Full-length caveolin-1–mCFP (L221K) and MyrPalm-mYFP L221K (K = 19 ± 9). (F) Caveolin-1–mCFP and PalmPalm-mYFP F223R (K = 3 ± 2). (G and H) mYFP-GerGer and mCFP-GerGer (F223R), (G) before (K = 0.3 ± 0.3) and (H) after 10 mM MβCD (K = 0.1 ± 4). (I) mYFP-GerGer and caveolin-1–mCFP F223R (K = 1100 ± 300). (J) mCFP-GerGer F223R and MyrPalm-mYFP F223R (K = 123 ± 90).

We suspected the residual clustering to be an artifact of the known tendency of all forms of GFP to dimerize at high concentrations, which would be particularly likely for proteins confined to two dimensions. To determine accurately the degree to which GFPs interact, we measured the homoaffinity of purified recombinant YFP (not membrane-anchored) by sedimentation equilibrium analytical ultracentrifugation (15, 17) and found aK d of 0.11 mM (Fig. 4, A and C) (18). We replaced hydrophobic residues at the crystallographic interface of the dimer (19) with positively charged residues (A206K, L221K, or F223R) (15). These mutants were likewise expressed and purified, and their self-association was measured by analytical ultracentrifugation. YFP homoaffinity was reduced to the point that dimerization was essentially eliminated (Fig. 4, B and D, and Table 1). We call these novel monomeric GFP variants mCFP, mGFP, and mYFP. These mutations caused no significant alterations in the spectral properties. When the NH2-terminus of Lyn was fused to mCFP and mYFP to cause myristoylation and palmitoylation, the proteins were clustered (Fig. 3C), but this clustering was destroyed by MβCD (Fig. 3D, showingE% nearly linearly proportional to F, i.e.,FK in Eq. 1). The latter result showed that the antidimerization mutations and MβCD extraction procedure were effective and that the clustering seen in unperturbed cells could now be attributed to lipid partitioning into microdomains rather than CFP-YFP affinity. mCFP and mYFP were used for all subsequent experiments. It is noteworthy that the visible distribution of MyrPalm-GFPs to plasma membranes was similar with or without MβCD, indicating the utility of FRET for discrimination and analysis of submicron-size domains of plasma membrane. These data also suggest that the domains must be fairly small, as larger domains would cause intradomain, density-dependent FRET, and no clustering would be seen.

Figure 4

Sedimentation equilibrium analytical ultracentrifugation of wild-type and monomeric YFP. Sedimentation equilibrium experiments were performed by standard protocol (15, 17). (A and B) wtYFP (A) and monomeric YFP L221K (B) data sets (gray) were fit globally at multiple speeds and concentrations to the monomeric (green) and dimeric (red) molecular weights. (C and D) The residual differences between the predicted distribution and the data for wtYFP (C) and mYFP L221K (D). Data shown are for 20,000 rpm at 230 μM.

Table 1

: Fluorescence and dissociation of wild-type versus mYFPs without lipid modifications. Dissociation constantK d (mM) derived from the association constant (K a) was measured by sedimentation equilibrium analytical ultracentrifugation. Variance (goodness of fit) was determined by global analysis with a nonlinear least-squares algorithm in the software provided by Beckman. Each value is statistically significant.

View this table:

We next measured the associations of various lipid modifications with caveolin, a marker for caveolae, a type of lipid raft. MyrPalm-mYFP clustered with full-length caveolin-1 fused to mCFP (Fig. 3E), and MyrPalm-mCFP likewise clustered with caveolin-1–mYFP. mYFP fused to the NH2-terminal 20 residues of GAP-43 to cause dual palmitoylation (PalmPalm-mYFP) and was also clustered with caveolin-1–mCFP (Fig. 3F). Thus, acylation by fatty acids suffices to cause clustering of test fluorescent proteins in lipid rafts and caveolae.

To examine the effectiveness of prenyl adducts in causing clustering, we fused the CAAX box (12) from the guanosine triphosphatase Rho to the COOH-terminus of mCFP and mYFP to cause the addition of a single geranylgeranyl group (mCFP-GerGer and mYFP-GerGer). These constructs clustered with each other (Fig. 3G), even after extraction of cholesterol with 10 mM MβCD (Fig. 3H). However, mYFP-GerGer did not cluster with caveolin-1–mCFP (Fig. 3I), nor did mCFP-GerGer congregate with MyrPalm-mYFP (Fig. 3J), suggesting that geranylgeranylation promotes clustering but not in cholesterol- and sphingolipid-rich lipid rafts.

We also used traditional biochemical methods to examine partitioning of lipid-modified YFPs into detergent-soluble or detergent-insoluble membranes, or into low- or high-density membranes after carbonate extraction (15). These techniques confirmed the results of the FRET experiments on live cells. MyrPalm-mYFP (Fig. 5A) partitioned primarily into a detergent-resistant membrane fraction (DRM) and almost exclusively with low-density caveolae-rich membranes (CRM), consistent with the partitioning of endogenous caveolin (Fig. 5C). Conversely, mYFP-GerGer was relatively excluded from DRM and CRM (Fig. 5B) (7). These experiments further confirm that lipid modifications alone are sufficient to confer specific sublocalization into or outside of lipid rafts within the plasma membrane.

Figure 5

Fractionation of MDCK plasma membrane reveals differential partitioning of acyl- and prenyl-modified YFP. MyrPalm-mYFP L221K and mYFP-GerGer F223R stably transfected MDCK cells were selected with 200 ng/ml G418 (Gibco). Fractionations were performed as described (5, 7) with the addition of a Percoll gradient to separate plasma membrane (PM) from intracellular membranes in a postnuclear supernatant (PNS) (22, 23) before carbonate or detergent fractionation. One-half of the PM was fractionated with detergent (5) and one-half with carbonate (7,24). Equal volume fractions were collected of detergent-soluble membrane (DSM), detergent-resistant membrane (DRM), noncaveolar membrane (NCM), and caveolae-rich membrane (CRM). (A) Western blot with antibody against GFP (Covance) of MyrPalm-mYFP fractions shows enrichment in detergent-insoluble and caveolae-rich fractions. (B) Blot by means of antibody against GFP of mYFP-GerGer fractions shows enrichment in detergent-soluble and noncaveolar membranes. (C) Immunoblot of endogenous caveolin by antibody against caveolin (Transduction Labs), with partitioning identical to that of the MyrPalm-mYFP.

Our data examining fluorescent proteins attached to the cytosolic side of the plasma membrane complement previous studies using FRET between dye-labeled antibodies, toxins, or ligands to look for clustering of proteins anchored by glycosylphosphatidylinositol (GPI) linkages to the extracellular leaflet of the plasma membrane (14,15, 20). Until now, allAequorea-derived GFPs and mutants of any color have contained the hydrophobic patch of Ala206, Leu221, and Phe223 responsible for dimerization of the beta barrels. Our new mutations to positively charged residues should prevent dimerization of all colors and are advisable whenever assessing intermolecular interactions of pairs of GFP fusion proteins.

These mutants allowed direct determination in living cells that acylated proteins associate in a manner predicted by models for clustering of proteins into the liquid-ordered phase (5). They cluster with each other in lipid rafts and with full-length caveolin-1, a marker for caveolae, suggesting similar lipid compositions and environments in the two structures. Raft disruption with MβCD not only disaggregates acylated fluorescent proteins but also alters numerous signaling events (15, 16,21), demonstrating the importance of these domains for cellular function.

  • * These authors contributed equally to this work.

  • Present address: Merck Research Laboratories, 3535 General Atomics Court, MRLSDB1, San Diego, CA 92121, USA.

  • To whom correspondence should be addressed. E-mail: rtsien{at}


Stay Connected to Science

Navigate This Article