A molecular dissection of caveolin-1 membrane attachment and oligomerization. Two separate regions of the caveolin-1 C-terminal domain mediate membrane binding and oligomer/oligomer interactions in vivo.

Caveolins form interlocking networks on the cytoplasmic face of caveolae. The cytoplasmically directed N and C termini of caveolins are separated by a central hydrophobic segment, which is believed to form a hairpin within the membrane. Here, we report that the caveolin scaffolding domain (CSD, residues 82-101), and the C terminus (residues 135-178) of caveolin-1 are each sufficient to anchor green fluorescent protein (GFP) to membranes in vivo. We also show that the first 16 residues of the C terminus (i.e. residues 135-150) are necessary and sufficient to attach GFP to membranes. When fused to the caveolin-1 C terminus, GFP co-localizes with two trans-Golgi markers and is excluded from caveolae. In contrast, the CSD targets GFP to caveolae, albeit less efficiently than full-length caveolin-1. Thus, caveolin-1 contains at least two membrane attachment signals: the CSD, dictating caveolar localization, and the C terminus, driving trans-Golgi localization. Additionally, we find that caveolin-1 oligomer/oligomer interactions require the distal third of the caveolin-1 C terminus. Thus, the caveolin-1 C-terminal domain has two separate functions: (i) membrane attachment (proximal third) and (ii) protein/protein interactions (distal third).

more than structural proteins. They interact with cholesterol and membrane lipids (15)(16)(17)(18), and an array of lipid modified and integral membrane proteins (4). The bulk of caveolininteracting proteins are signaling molecules, and many of these caveolin-interacting proteins bear a common caveolin binding motif that is recognized by a 20-aminoacyl residue domain of the caveolin molecule (residues 82-101 in caveolin-1). We have termed this portion of the caveolin molecule the caveolin scaffolding domain (CSD), 1 and identified its consensus binding motif through phage display (19 -21).
Caveolins have an unusual structure. Both N and C termini of caveolin-1 are directed toward the cytoplasm, and it is believed that a central hydrophobic segment inserts into the membrane (residues 102-134 in caveolin-1), effectively splitting the molecule into two cytoplasmic domains (22). Several lines of evidence support this model. First, antibodies directed against the extreme N and C termini fail to stain caveolin-1 in unpermeabilized cells (23,24). Likewise, artificial glycosylation sites introduced at the extreme N and C termini are not glycosylated (23). Third, caveolin-1 is not labeled when cells are subjected to surface biotinylation, suggesting that there is no extracellular domain (25,29). Finally, the N terminus of caveolin-1 undergoes phosphorylation (26 -28), and the C terminus of caveolin-1 undergoes palmitoylation (24), both cytoplasmic modifications.
Caveolin-1 and -3 can each form homotypic high molecular mass oligomers containing ϳ14 -16 individual molecules (8,23,29). In the endoplasmic reticulum, homo-oligomerization is mediated by a 40-aminoacyl residue domain (residues 61-101 in caveolin-1), which contains the CSD (29). In the Golgi, adjacent homo-oligomers undergo a second stage of oligomerization; they interact with one another through contacts between the C-terminal domains of caveolin, in a side-by-side packing scheme (30). These oligomer/oligomer interactions then produce an interlocking network of caveolin molecules that gives rise to the striated caveolar coat seen by scanning electron microscopy (30,31).
Recently, we characterized a panel of caveolin-1 deletion mutants to assess whether the putative transmembrane (TM) domain is required for membrane attachment (32). Interestingly, we found that Cav-1-(1-101), a truncation mutant that retains the complete N terminus (but lacks the TM domain and the palmitoylated C terminus) binds tightly to membranes, targets to caveolae membranes with high efficiency, and interacts with signaling molecules. In contrast, Cav-1-(1-81) (a truncation that is missing the TM domain, the C terminus, and the CSD) was completely soluble and failed to regulate signaling. These results suggested a direct role for the CSD (residues 82-101) in membrane binding.
Using reconstituted lipid vesicles (17), we found that a glutathione S-transferase (GST) fusion protein of Cav-1-(1-101) bound membranes directly. Interestingly, a GST fusion protein to the complete C terminus of Cav-1 (residues 135-178) also bound lipid vesicles. Since these GST fusion proteins are expressed in bacteria, our results also suggest that palmitoylation is not required for the C-terminal domain of caveolin-1 to associate with membranes. Surprisingly, a GST fusion protein to the TM domain had no affinity for lipid vesicles. Thus, caveolin-1 may be anchored to membranes by two unconventional membrane-binding domains (32).
Here, we identify and characterize the membrane attachment domains of caveolin-1 in vivo using a panel of caveolin-1 domains fused to a soluble reporter protein, GFP. We show that the first 16 residues of the C-terminal domain are necessary and sufficient to bind GFP to trans-Golgi network (TGN) membranes. We also use these fusion proteins to map the domains required for oligomer-oligomer interaction in co-immunoprecipitation studies with full-length caveolin-1. Since the terminal ϳ10 residues are required for oligomer/oligomer interaction, we conclude the C terminus of caveolin-1 has two separate protein/protein and protein/lipid interaction domains at opposite ends.
Construction of GFP Fusion Proteins-Standard polymerase chain reaction strategies were used in constructing the various fusion proteins shown in Figs. 1A and 3. Canine caveolin-1 (25), mouse caveolin-2 (36), and rat caveolin-3 (8) cDNAs were used as the templates for these reactions. Fusions were constructed by subcloning the polymerase chain reaction products in-frame into the pEGFP-C1 vector (CLON-TECH) as HindIII/BamHI fragments within the polylinker. Intended mutations were confirmed by DNA sequencing.
Cells, Media, and Transfection Methods-Human embryonic kidney 293T cells and Cos-7 cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 2 mM glutamine, 100 units/ml penicillin, 100 g/ml streptomycin. 293T cells were transfected using the calcium phosphate precipitation method. Cos-7 cells were transfected using Fugene 6, as described by the manufacturer (Roche Molecular Biochemicals). Optimal transfection of Cos-7 cells was achieved by mixing 2 g of DNA with 3 l of Fugene 6 reagent diluted in 97 l of medium.
Protein Assay-Protein concentrations were determined using a commercially available bicinchoninic acid protein assay kit, exactly as described by the manufacturer (Pierce).
Immunoblotting-Samples were subjected to SDS-PAGE under reducing conditions and transferred to nitrocellulose membranes. The protein bands were stained with Ponceau S (Sigma). Membranes were then washed, blocked, incubated with primary antibody, washed again, and incubated with a secondary antibody conjugated with horseradish peroxidase (Transduction Laboratories). Bound IgG were detected using a chemiluminescent substrate (Pierce).
FIG. 1. Construction and characterization of GFP/caveolin-1 fusion proteins. A, caveolin-1 may be divided into three domains. The N terminus (residues 1-101) and the C terminus (residues 135-178) are separated by a hydrophobic transmembrane domain (TM, residues 102-134)). GFP-FL-Cav-1-(1-178) is a fusion of the entire caveolin-1 protein to GFP; GFP-N-Cav-1-(1-81) is a fusion of the N-terminal domain of caveolin-1, up to but excluding the CSD. GFP-N-Cav-1-(82-101) is fusion of the caveolin-1 CSD to GFP; GFP-TM-Cav-1-(102-134) is a fusion of the complete transmembrane domain of caveolin-1 to GFP; and GFP-C-Cav-1-(135-178) is a fusion of the complete C-terminal domain of caveolin-1 to GFP. 293T cells were transiently transfected with the indicated cDNAs. Thirtysix hours post-transfection, cells were subjected to hypotonic lysis. Proteins were fractionated into soluble (S) and particulate (P) components. In each experiment, the particulate material was resuspended in the same volume as the soluble fraction, and equal volume aliquots of each were separated by SDS-PAGE. Proteins were transferred to nitrocellulose, and all blots were probed with anti-GFP IgGs. Note that the first 81 residues of caveolin-1, and residues 102-134 do not alter GFP's solubility, whereas the CSD (residues 82-101), and the C-terminal domain (residues 135-178) conferred affinity for membranes. B, like Cav-1(FL), the CSD, and the C-terminal domain of caveolin-1 targeted GFP to the particulate fraction-in an alkaline sodium carbonate and Triton X-100 inextractable manner (I).
Hypotonic Lysis-Post-nuclear lysates were separated into soluble and insoluble components, as described previously (32,37). Cells grown to confluence in a 60-mm diameter dish were scraped into cold PBS, pelleted by centrifugation, and resuspended in 750 l of hypotonic buffer (5 mM Tris, pH 7.5, 1 mM MgCl 2 , 1 mM EGTA, 0.1 mM EDTA) containing protease inhibitors. Following a 30-min incubation on ice, the lysate was passed through a 26-gauge needle 10 times and centrifuged at 1000 ϫ g to remove nuclear debris. The post-nuclear lysate was transferred to an 11 ϫ 34-mm polycarbonate tube and centrifuged at 56,000 rpm for 30 min in a TLA-100.2 rotor (Beckman). The supernatant was collected, and the pellet was resuspended in an equal volume of 1% SDS. An aliquot of 25 l from each fraction was subjected to SDS-PAGE and immunoblot analysis, as described above.
Alkaline Carbonate Extraction-Carbonate extraction was performed, as we described previously (32,34). Briefly, cells were grown to confluence in a 60-mm diameter dish and washed twice in ice-cold PBS and once in 150 mM NaCl. After aspiration of the NaCl solution, 1 ml of 200 mM NaCO 3 , pH 11.3, containing protease inhibitors was used to scrape the cells off the dish. The sample was transferred to a tightly fitting 1-ml Dounce homogenizer and homogenized with 5 strokes. Following a 30-min incubation on ice, the sample was transferred to an 11 ϫ 34-mm polycarbonate tube and centrifuged at 100,000 rpm in a TLA-100.2 rotor. The pellet was resuspended in an equal volume of 1% SDS. Both fractions were sonicated on ice, and 50 l of each were subjected to SDS-PAGE and immunoblot analysis, as described above.
Triton Solubility-Extraction of Triton-soluble proteins was performed, as we described previously (25,32). Briefly, cells were grown to confluence in 35-mm diameter dishes and washed twice with ice-cold PBS. Three hundred microliters of cold MBS (25 mM Mes, pH 6.5, 150 mM NaCl), containing 1% Triton X-100 plus protease inhibitors was then added. Following a 30-min incubation on ice without agitation, the soluble fraction was collected. An equal volume of 1% SDS was added to the plate to dissolve the remaining Triton X-100-insoluble material. The latter fraction was passed through a 26-gauge needle 10 times in order to lower its viscosity. Twenty microliters of the Triton X-100soluble and -insoluble fractions were each separated by SDS-PAGE and subjected to immunoblot analysis, as described above.
Purification of Caveolae-enriched Membrane Fractions-Caveolaeenriched membrane fractions were purified, essentially as we described previously (25,32). Cells grown to confluence in two 100-mm diameter plates were washed twice in ice-cold PBS, scraped into 750 l of MBS containing 1% Triton X-100, passed 5 times through a tightly fitting Dounce homogenizer, and mixed with an equal volume of 80% sucrose (prepared in MBS lacking Triton X-100). The sample was then transferred to a 4.5-ml ultracentrifuge tube and overlaid with a discontinuous sucrose gradient (1.5 ml of 30% sucrose, 1.5 ml of 5% sucrose, both prepared in MBS, lacking detergent). The samples were then subjected to centrifugation at 200,000 ϫ g (44,000 rpm in a Sorval rotor TH-660) for 18 h. A light scattering band was observed at the 5/30% sucrose interface. Twelve 375-l fractions were collected, and 50-l aliquots of each fraction were subjected to SDS-PAGE and immunoblot analysis.
Immunofluorescent Staining-Thirty-six hours after transfection, cells grown on coverslips were fixed in an ice-cold mixture of methanol and acetone (1:1, v/v) for 10 min at Ϫ20°C. Cells were air-dried, rehydrated in phosphate-buffered saline supplemented with 0.1 mM CaCl 2 and 1 mM MgCl 2 (PBS-CM), washed twice in PBS-CM supplemented with 1% (w/v) BSA (PBS-CM-BSA), and incubated with primary antibody. After washing, cells were incubated with fluorophore-conjugated secondary antibodies and washed again. Primary and fluorophore-conjugated secondary antibodies were diluted in PBS-CM-BSA. After immunostaining, coverslips were mounted on glass slides, and cells were viewed with an Olympus IX70 inverted microscope using a 60ϫ objective. Images were collected with a Photonics cooled CCD camera. All microscopy was performed at the Analytical Imaging Facility of the Albert Einstein College of Medicine.
Co-immunoprecipitation Studies-Cells were subjected to lysis in immuno-precipitation buffer (10 mM Tris, pH 8.0, 150 mM NaCl, 1% Triton X-100, 60 mM n-octylglucoside, supplemented with protease inhibitors). DNA was sheared by brief sonication on ice, and cellular debris was removed by centrifugation at 20,000 ϫ g for 10 min. Lysates were pre-cleared by incubation with Protein A-Sepharose for 1 h at 4°C and then transferred to fresh tubes containing 30 l of a 1:1 slurry of Protein A-Sepharose and immunoprecipitation buffer. Two micrograms of anti-caveolin-1 IgG (monoclonal antibody 2234) were added to the mixture. Following a 3-h incubation at 4°C, immune complexes were collected by centrifugation, washed six times with 1 ml of immunoprecipitation buffer, and disrupted by boiling in 1% SDS.

Caveolin-1 Has Two High Affinity
Membrane-binding Domains-We previously determined that both the complete N terminus (i.e. residues 1-101) and the complete C terminus (i.e. residues 135-178) of caveolin-1 can anchor a heterologous protein (GST) to lipid vesicles in vitro, whereas the TM domain of caveolin-1 (i.e. residues 102-134) does not. Furthermore, removal of the final 20 aminoacyl residues of the caveolin-1 N terminus (i.e. residues 82-101) restores GST's solubility (32). To assess whether these caveolin-1 domains have similar properties in vivo, a panel of GFP fusion proteins was constructed (Fig. 1A).
When fused to caveolin-1, GFP behaves as a silent reporter of localization, and does not misdirect this protein away from caveolae. More specifically, we have demonstrated that fusion of GFP to either the N or the C terminus of FL caveolin-1 does not alter caveolin-1's ability to form oligomers, to resist nonionic detergents, to target to caveolae membranes, or to interact with signaling molecules (38,39). Additionally, recent reports suggest that GFP is well suited as a reporter for the subcellular localization of membrane attachment sequences from diverse proteins; fusion of the N-terminal myristoylated, palmitoylated, or polybasic sequences from a number of Src family tyrosine kinases and G␣ subunits (as little as 10 residues in some cases) results in efficient anchoring of GFP to membranes (39,40). Similar results are obtained when the polybasic domains and C-terminal isoprenylation sequences of two Ras isoforms are coupled to GFP (41).
To eliminate the possible interaction of these fusion proteins with endogenous caveolins, we chose to express their cDNAs in human embryonic kidney 293T cells. Like many transformed cell lines, 293T cells do not express any known caveolins (42).
FIG. 4. GFP is rendered membranebound when fused to the C terminus of caveolin-1. 293T cells transiently expressing the indicated fusion proteins were subjected to hypotonic lysis and fractionation as in Fig. 1. Immunoblot analysis revealed that all GFP fusion proteins shown were in the particulate fraction (P), suggesting membrane association (Upper panel). Blots were reprobed with antibodies against endogenous cytosolic (GDI) and integral membrane (LAMP-1) proteins to validate the fractionation method. Both GDI and LAMP-1 partitioned as expected (Lower panels).
Furthermore, 293T cells do not express members of another class of integral caveolae membrane structural proteins, cavatellin-1 or -2 (34), which we demonstrated can hetero-oligomerize with caveolin-1 (43). Importantly, we found previously that expression of caveolin-1 in these cells results in proper oligomerization, palmitoylation, and localization (32). In summary, 293T cells are suited for an unbiased analysis of the membrane affinity of GFP when fused to distinct caveolin-1 domains because these cells contain the necessary sorting machinery for assembling and directing caveolin-1 oligomers to caveolae, yet they do not express caveolae structural proteins endogenously.
Using an established fractionation scheme, we separated soluble and membrane-bound proteins from 293T cells expressing our panel of GFP fusion proteins using hypotonic lysis. When fused to FL caveolin-1, GFP was found predominantly in the pellet (Fig. 1A). Similarly, fusion of GFP to the C terminus of caveolin-1 rendered GFP membrane-bound. Likewise, the majority of the CSD-fused GFP was directed to the pellet; but the majority of GFP-TM-Cav-1-(102-134) was soluble. As a control, we verified that fusion of GFP to the N-terminal domain up to but excluding the CSD (i.e. residues 1-81) did not alter the solubility of GFP. In summary, the CSD of caveolin-1 (residues 82-101) and the C-terminal domain of caveolin-1 (residues 135-178) were sufficient to anchor GFP to membranes in vivo, with efficiencies comparable to that of the full-length caveolin-1 molecule.
We also assessed whether the membrane-bound GFP fusion protein were resistant to alkaline sodium carbonate extraction.
Caveolae and related lipid domains are resistant to solubilization in cold Triton X-100. This physical property reflects the concentration of saturated lipid chains, sphingolipids, cholesterol, and lipid-modified proteins in these microdomains (45). Furthermore, these detergent-resistant membranes are buoyant in sucrose density gradients. We have exploited this physical property to purify membranes enriched in caveolins from the bulk of cellular proteins (25,46,47). Interestingly, the CSD and the C-terminal domain of caveolin-1 each rendered GFP insoluble in Triton X-100, as well (Fig. 1B).
To examine the subcellular localization of the various GFP fusion proteins, we performed fluorescence microscopy in transfected Cos-7 cells (Fig. 2). This cell line, like 293T, shows little or no endogenous caveolin-1, and was selected for these studies because it adheres tightly to glass coverslips, whereas 293T cells are less adherent. Like untagged GFP, GFP-N-Cav-1-(1-81) was diffusely distributed throughout the cell, confirming our fractionation results that the first 81 residues of caveolin-1 do not anchor GFP to membranes. As expected, GFP-Cav-1(FL) yielded a punctate distribution pattern at the cell periphery. GFP-N-Cav-1-(82-101) also had a punctate distri- FIG. 5. GFP resists extraction by alkaline sodium carbonate when fused to the C terminus of caveolin-1. 293T cells transiently expressing the indicated GFP-C-Cav-1 cDNAs were extracted with 200 mM NaCO 3 , pH 11.3, homogenized, and fractionated as in Fig. 1. Immunoblot analysis demonstrates that fusion of GFP to all fragments of the caveolin-1 C terminus that include residues 135-150 rendered the protein resistant (P) to alkaline sodium carbonate (Upper panel). As a control, we verified that the peripherally associated membrane protein EEA1 is completely solubilized (S) by this treatment (Lower panel). bution pattern. GFP-TM-Cav-1-(102-134) was found throughout the cell in a diffuse pattern, but we also detected some small stipples at the cell surface, again confirming the fractionation results that the majority of this protein is soluble. Surprisingly, GFP-C-Cav-1-(135-178) was found in a perinuclear ring, as well as in a punctate pattern at the cell periphery.
Detailed Mutational Analysis of the Caveolin-1 C Terminus in the Context of a GFP Reporter-Since GFP-C-Cav-1-(135-178) was avidly bound to membranes, and was localized in a pattern that did not overlap with full-length caveolin-1, we examined this fusion protein in greater detail. For this purpose, a panel of 11 GFP-C-Cav-1 mutants were constructed (Fig. 3), including 6 truncation and 5 point mutants. The residues selected for point mutation are 3 of 12 residues conserved across all caveolin isotypes from all species characterized to date. Since two familial forms of autosomal dominant limbgirdle muscular dystrophy, type 1C (LGMD-1C) are due to mutations in other conserved caveolin residues of human caveolin-3 (48), we suspected that mutation of the three conserved C-terminal residues to alanine might unmask important functions. Ser 168 was also mutated to Glu. For a number of proteins, introduction of a negatively charged residue effectively mimics chronic phosphorylation. Thus, this mutant was made to examine the possible effect of phosphorylation of this residue. Finally, two cysteine residues that undergo palmitoylation were mutated to serine (C143/156S) to prevent thioacylation in the "pal Ϫ " mutants.
To determine if our fusion proteins were membrane-bound, we performed hypotonic lysis and fractionation. Fig. 4 shows that all fusion proteins examined were found predominantly in the pellet, suggesting that they are bound to membranes in vivo. This binding is independent of palmitoylation and is insensitive to mutation of the three conserved residues. These results also show that a 16-aminoacyl residue segment of the caveolin-1 C terminus (i.e. residues 135-150) is sufficient to bind a heterologous protein to membranes. As internal controls to validate our fractionation methods, blots were re-probed with antibodies directed against endogenous cytosolic (GDI) and glycosylated integral membrane (LAMP-1) proteins. Fig. 4 shows that GDI was found in the soluble fraction (34), and LAMP-1 was in the pellet (49), confirming that our fractionation method effectively separates soluble proteins from membrane-bound proteins.
Next, we determined if the GFP-C-Cav-1 fusion proteins could be solubilized by extraction with alkaline sodium carbonate. Fig. 5 shows that all of the fusion proteins were resistant to extraction. As internal controls for this experiment, we verified that the peripherally associated membrane protein EEA1 is rendered completely soluble by alkaline sodium carbonate (50). Taken together, these results again suggest that a 16aminoacyl residue segment (i.e. residues 135-150) is sufficient for tight association of the C terminus of caveolin-1 to membranes, and that this binding is independent of thio-acylation.
Next, we assessed the Triton solubility of various GFP fusion FIG. 7. Defining a minimal membrane attachment domain within the C terminus of caveolin-1: residues 135-150 are sufficient, whereas residues 151-178 are not required. A and B, to demonstrate that only residues 135-150 are necessary for membrane attachment, cells were transfected with the indicated constructs and subjected to hypotonic lysis and fractionation as in Fig. 1. Immunoblot analysis with anti-GFP antibody revealed that GFP-C-Cav-1-(151-178) was present in both fractions in nearly equal abundance, whereas GFP-C-Cav-1-(151-178pal Ϫ ) was soluble (left). C, cells were transfected as in Fig. 2, fixed, and viewed by fluorescence microscopy. Note that GFP-C-Cav-1-(151-178pal Ϫ ) was distributed throughout the cell. GFP-C-Cav-1-(151-178) manifested both diffuse and punctate distribution patterns, confirming the fractionation result that this protein is present in both membrane and soluble fractions. D and E, to assess the contribution of palmitoylation to the membrane avidity of residues 135-150, we constructed GFP-C-Cav-1-(135-150pal Ϫ ), and found that this protein is still associated with membranes upon hypotonic lysis, and resists extraction with alkaline sodium carbonate and Triton X-100.

Caveolin-1 Membrane Attachment and Oligomerization
proteins to the C-terminal domain of caveolin-1. Fig. 6 shows that GFP was rendered insoluble in Triton X-100 by fusion to as little as the first 16 residues of the C-terminal domain. These findings corroborate the results of the hypotonic lysis and alkaline sodium carbonate extraction experiments. As a control, we verified that GFP was soluble in Triton X-100 (38).
Subcellular Localization of GFP Fused to the C Terminus of Caveolin-1-Although GFP-C-Cav-1-(135-178) resisted solubilization in Triton X-100, its pattern of subcellular localization, as assessed by microscopy, did not overlap with GFP-Cav-1(FL) (Fig. 2). Using an established sucrose density gradient centrifugation method to purify caveolae membranes from the bulk of cellular proteins, we determined whether the C-terminal fusion proteins were targeted to caveolae. Fig. 8 (A and B) demonstrates that most of the cellular proteins remain at the bottom of the gradient during caveolae purification. As a control, we verified that GFP targets to low density membranes when fused to FL caveolin-1 (38) (Fig. 8C). However, GFP-C-Cav-1-(135-178) did not target to these low density domains, suggesting that the Triton X-100 domain occupied by these proteins is not caveolae. Additionally, GFP-TM-Cav-1-(102-134) was excluded from these low density Triton-resistant domains.
Interestingly, only the CSD was able to target GFP to low density Triton-insoluble domains; however, this targeting was 293T cells transiently expressing the indicated cDNAs were subjected to lysis in 1% Triton X-100 as described under "Experimental Procedures." The lysate was mixed with an equal volume of 80% (w/v) sucrose (1.5 ml final volume). The mixture was placed at the bottom of an ultracentrifuge tube and then overlaid with a discontinuous sucrose gradient. Following ultracentrifugation, twelve fractions were collected and equal volume aliquots of each were subjected to SDS-PAGE. Following separation, proteins were transferred to nitrocellulose. A, the membrane was stained with Ponceau S to confirm that the bulk of cellular proteins remain at the bottom of the gradient (fractions 9 -12). B, protein assays of each fraction revealed that the concentration of protein in the high density fractions of the gradient was over 100-fold higher than in the lower density fractions. C, immunoblot analysis demonstrates that, unlike wild-type caveolin-1, GFP-C-Cav-1-(135-178) did not target to low density fractions (4 and 5), suggesting exclusion from caveolae membranes. GFP-TM- Cav-1-(102-134) was also excluded from low density fractions, whereas GFP-N-Cav-1-(82-101) was detected in these domains. significantly less efficient than that of full-length caveolin-1. These results support our previous report that the N terminus of caveolin-1 targets to caveolae membranes provided that the CSD is intact (32).
Since GFP-C-Cav-1-(135-178) was concentrated in a perinuclear location (a ring in some cases) and also had a punctate distribution pattern in the cell periphery (Fig. 2), we suspected that this protein targets to the TGN. To test this hypothesis directly, we doubly immunostained cells with antibodies directed against the marker protein TGN38. Fig. 9A shows dramatic colocalization of TGN38 and GFP-C-Cav-1-(135-178), suggesting that the C terminus of caveolin-1 bears a trans-Golgi membrane attachment signal. In contrast, GFP-Cav-1(FL) did not co-localize with this protein (not shown). Additionally, GFP-C-Cav-1-(135-178) showed significant colocalization with another trans-Golgi marker protein, ␥-adaptin (Fig. 9B).
Next, we performed fluorescence microscopy to examine where the various GFP-C-Cav-1 mutants reside. Fig. 10 shows that all proteins were found concentrated in the perinuclear region, and were also detected in a punctate pattern at the cell periphery. Like the membrane association findings, these results suggest that localization to the trans-Golgi network requires merely the first 16 residues of the C-terminal domain, and that this localization is insensitive to mutation of the invariant caveolin residues and the sites of palmitoylation.
Mapping the Caveolin-1 Oligomer/Oligomer Interaction Domain-Caveolin-1 undergoes a two-step oligomerization process. In the endoplasmic reticulum, homotypic oligomers of ϳ14 -16 subunits are formed through the interaction of the 40-aminoacyl residue oligomerization domain (61-101). How-ever, in the Golgi, adjacent oligomers interact with one another to form higher order structures, and this extended contact is mediated by C-terminal/C-terminal interactions (30,51). Additionally, this oligomer/oligomer interaction is isoform-specific, with the C terminus of caveolin-1 interacting only with caveolin-1 oligomers, but not with caveolin-2 and caveolin-3 oligomers (30).
To define the oligomer/oligomer interaction domain with greater precision, we co-expressed wild-type caveolin-1 with several GFP-C-Cav-1 proteins. Fig. 12 shows that GFP-C-Cav-1-(135-178) co-immunoprecipiated with FL caveolin-1. GFP-C-Cav-1(pal Ϫ ) also co-immunoprecipiated with FL caveolin-1, albeit to a lesser extent. This finding corroborates our previous result in vitro that GST-C-Cav-1-(135-178) expressed in bacteria binds to caveolin-1 oligomers from a mammalian cell lysate (30). These results also agree with our recent report that G i ␣, which bears a caveolin-binding motif and interacts directly with the CSD, is bound to caveolin-1 with greatest efficiency when both it and caveolin-1 are lipid-modified (39). Additionally, oligomerization of caveolin-1 in vitro is stabilized by palmitoylation (52). In summary, lipid modification appears to strengthen the interaction between caveolin-1 and other lipidmodified proteins, although it is not required for these interactions.
The final residues of the caveolin-1 C terminus are the most divergent from caveolins 2 and 3, and this lack of homology may account for the isoform-specific nature of the oligomer/ oligomer interaction (Fig. 13A). Similarly, the first 16 residues of the C-terminal domain of caveolin-2 diverge significantly from the corresponding segments of caveolin-1 and -3. This divergence may account for the unique subcellular distribution of GFP-C-Cav-2-(120 -162).  Fig. 9. Like GFP-C-Cav-1-(135-178), all fusion proteins appear concentrated in the peri-nuclear region of the cell, and in a punctate pattern at the cell periphery.

DISCUSSION
Here, we have systematically identified and characterized the membrane attachment domains of caveolin-1 in vivo. For this purpose, we engineered a panel of caveolin-1 domains fused to a soluble reporter protein, GFP, and examined their capacity for membrane attachment after transient expression in 293T cells. We showed that the CSD, the putative transmembrane domain, and the C terminus of caveolin-1 are each independently able to anchor GFP to membranes. Unexpectedly, the C terminus of caveolin-1 was able to attach GFP to membranes to the same extend as full-length caveolin-1, whereas both the CSD and TM domain of caveolin-1 conferred less membrane affinity to GFP.
We next examined the properties of the C-terminal domain in great detail by creating and characterizing a panel of deletion and point mutants. The first third of the C-terminal domain (residues 135-150) was sufficient to target GFP to membranes and was independent of palmitoylation. In contrast, residues 151-178 could not bind GFP to membranes in the absence of palmitoylation. Furthermore, fusion proteins containing caveolin-1 residues 135-150 were resistant to extraction with alkaline sodium carbonate, were insoluble in Triton X-100, and co-localized with markers of the trans-Golgi. Thus, our results suggest that caveolin-1 contains at least two competing membrane attachment signals in its primary sequence: the CSD (for caveolar targeting) and the C terminus (for trans-Golgi targeting).
In conjunction with these studies, we mapped the minimal caveolin-1 oligomer/oligomer interaction domain to the final third of the C terminus. Co-immunoprecipitation of the various GFP fusion proteins with full-length Cav-1 was prevented by deletion of the final 8 residues of the C terminus, and by point mutation of a conserved serine that is 10 residues from the end of the molecule. Thus, the C terminus of caveolin-1 has two separate protein/protein and protein/lipid interaction domains at opposite ends. We propose the terms N-MAD (N-terminal membrane-attachment domain, residues 82-101), C-MAD (Cterminal membrane-attachment domain, residues 135-150), and terminal domain (residues 168 -178) to refer to these protein domains. These findings are summarized in Table I.
Mechanistic Insights into the Association of Caveolin-1 with Membranes-How do the membrane attachment domains of caveolin-1 bind to lipid bilayers? For the CSD/N-MAD, at least, the answer may lie in studies of the unconventional secretion of homeodomain transcription factors (53). Prochiantz and colleagues noted that the caveolin-1 CSD bears remarkable resemblance to a highly conserved portion of the homeodomain transcription factor engrailed (Fig. 13B). This conserved portion of engrailed is responsible for the non-vesicular exocytosis of this protein. Indeed, this conserved domain can carry covalently linked oligopeptide and oligonucleotide cargo into cells by "penetrating" lipid bilayers (54). Interestingly, engrailed's membrane translocation occurs in caveolae and caveolae-like domains (53). Critical differences between sequences in and around the caveolin-1 CSD/N-MAD and the conserved region of homeodomain proteins may be responsible for the high affinity of caveolin-1 for membranes.
We previously reported that the caveolin-1 transmembrane domain has no affinity for lipid vesicles in vitro (32). Here, we found that fusion of this domain to GFP results in minimal anchoring of GFP to membranes in 293T cells (Fig. 1). We retain the name "transmembrane," however, because this domain mediates hetero-oligomerization with caveolin-2 (36), and we cannot exclude the possibility that this interaction occurs within the lipid membrane. Additionally, caveolin-1 coupling of integrins to Src-like kinases appears to involve the transmem-brane domain of caveolin-1; Src-like kinases bind to the CSD, and the single-pass transmembrane anchor of integrins may interact with the caveolin-1 transmembrane domain (reviewed in Ref. 55).
Residues 135-150 of caveolin-1 could anchor GFP to membranes even in the absence of palmitoylation (Figs. 3 and 7). However, this C-MAD bears no homology to any known membrane-binding domains of soluble proteins like the phosphatidyl serine-binding domain of protein kinase C (56). Furthermore, residues 135-150 of caveolin-1 do not contain a polybasic stretch like those employed by some members of the Ras superfamily of GTPases (57) and of the Src family of tyrosine kinases, which facilitates their anchorage to membranes (58). Thus, this portion of caveolin-1 represents a novel membrane attachment motif.
Finally, we note that the N-and C-MADs map to regions of the caveolin-1 protein that bind to proteins (Table I; Refs. 4,59,and 60). This dual function is reminiscent of the pleckstrin homology superfamily of membrane and protein-binding domains. Members of this superfamily share a common threedimensional structure, yet individual pleckstrin homology domain containing proteins can bind either lipids or proteins (61). Future studies to determine the three-dimensional structure of the caveolin-1 N-and C-MADs will provide the strongest mech- FIG. 11. Caveolin-2 and caveolin-3 also contain C-terminal membrane attachment domains. To assess whether other caveolin proteins contain C-terminal membrane attachment domains, the C termini of caveolin-2 and -3 were fused to GFP. A, these fusion proteins were expressed transiently to equivalent levels in 293T cells. B, cells were transfected with the indicated cDNAs and subjected to hypotonic lysis and fractionation. The C terminus of caveolin-2 and the C terminus of caveolin-3 were able to attach GFP to membranes. C, fluorescence microscopy demonstrated that GFP-C-Cav-3 has a peri-nuclear distribution pattern, whereas GFP-C-Cav-2 was distributed in a vesicular pattern throughout the cell. anistic insights into how caveolin-1 can bind both membranes and proteins, using the same stretches of amino acids.
Multiple Membrane-binding Domains Dictate Caveolin-1 Localization to Distinct Subcellular Locations-At steady state, over 90% of caveolin-1 is at the plasma membrane (62). Yet, caveolin-1 molecules move between the plasma membrane and internal membrane compartments in a cholesterol-regulated manner (63)(64)(65), and caveolin-1 is even detected in a cytosolic complex with chaperone proteins and cholesterol (66). In order to uncover the molecular determinants of caveolin-1 traffic through the secretory pathway, Anderson and colleagues (67) performed alanine scanning mutagenesis of the caveolin-1 oligomerization domain, with recombinant expression in Chinese hamster ovary cells that endogenously express caveolin-1. They introduced a series of consecutive penta-alanyl codon substitutions into the caveolin-1 gene from the start of the oligomerization domain (residue 60) to the end of the oligomerization domain (residue 100), and examined the localization of the resulting proteins in Chinese hamster ovary cells. They found that Cav 66A70 (i.e. residues 66 through 70 were mutated to alanine) was localized to the endoplasmic reticulum, whereas Cav 71A75 and all consecutive mutants up to and including Cav 96A100 were localized to the Golgi apparatus. Several internal deletion mutants confirm the alanine scan results, and they conclude from these findings that certain segments of the caveolin-1 molecule are required for exit from the endoplasmic reticulum (residues 66 -70) and the Golgi (residues 70 -100).
Notably, correct membrane topology was maintained in these mutants.
For caveolins and many other oligomeric membrane proteins to pass through the secretory pathway, correct membrane topology must be achieved, and appropriate oligomerization must occur. Failure to satisfy these requirements leads to arrest in transport and proteosomal degradation (68). Indeed, a number of human genetic diseases are characterized by retention in the endoplasmic reticulum and Golgi of improperly assembled oligomeric proteins, which are degraded rapidly (69). For example, we have isolated and characterized mutant caveolin-3 molecules from patients with LGMD-1C (48). One mutant bears an in-frame microdeletion of three residues in the scaffolding domain that includes one of the 12 invariant caveolin residues. Another LGMD-1C caveolin-3 has a point mutation in the transmembrane domain that results in substitution of yet another invariant residue. Both of these proteins reside in the Golgi apparatus as high molecular weight aggregates and have shortened half-lives (70).
Like the LGMD-1C mutant caveolin-3 molecules, all of the alanine scan mutants are retained in intracellular membrane compartments; and many have shortened half-lives when compared with wild-type caveolin-1 (67). Additionally, some of these mutants are not oligomeric, indicating assembly into high molecular weight complexes has been compromised by mutation of residues within the oligomerization domain (67). These results indicate that alanine scanning mutagenesis of FIG. 12. Mapping of the caveolin-1 oligomer/oligomer interaction domain. Cells were co-transfected with untagged full-length caveolin-1 and the indicated GFP/caveolin-1 cDNAs. A, cell lysates were subjected to immunoprecipitation with monoclonal antibody 2234, which recognizes an epitope in the aminoterminal 20 residues of caveolin-1. Recovered proteins were subjected to SDS-PAGE and transferred to nitrocellulose. Full-length caveolin-1 was detected with rabbit polyclonal IgGs (N20) which was raised against a synthetic peptide corresponding to residues 2-21 of caveolin-1 (Upper panel). Blots were also probed with anti-GFP IgGs (Middle panels). Note GFP-C-Cav-1-(135-178) and GFP-C-Cav-1(pal Ϫ ) co-immunoprecipitate with fulllength caveolin-1, whereas all three deletion mutants co-immunoprecipitate with far less efficiency. Two exposures are shown to better illustrate this point. Importantly, all GFP fusion proteins were expressed to comparable levels (Lower panel). B, compared with GFP-C-Cav-1(P158A), and GFP-C-Cav-1(G164), GFP-C-Cav-1(S168A) and GFP-C-Cav-1(S168E) showed decreased binding to full-length caveolin-1. the oligomerization domain did not uncover any caveolin residues as sorting determinants for transit through the secretory pathway. Rather, this approach identified where the "quality control" machinery of the secretory pathway halts and eliminates caveolin-1 proteins that have failed to fold and oligomerize properly (71). Since the LGMD-1C mutant caveolin-3 proteins block the normal traffic of wild-type caveolin-3, it will be interesting to see if the alanine scan mutants exert a similar dominant negative effect on the endogenous wild-type caveolin-1 that is expressed in Chinese hamster ovary cells.
In our previous study of the membrane-binding domains of caveolin-1, we characterized a deletion mutant that is free of the confounding phenotypes sometimes associated with alanine scanning and deletion mutagenesis. More specifically, we examined the consequences of deleting the transmembrane domain (residues 102-134) and found that the resulting protein, Cav-1(⌬TM), is oligomeric, and palmitoylated. However, Cav-1(⌬TM) targets to the plasma membrane proper, as assessed by immunofluorescence microscopy and subcellular fractionation (32). In hindsight, is comes as no surprise that juxtaposition of two localization signals that target to different membrane compartments (i.e. residues 82-101 targeting to caveolae and residues 135-150 targeting to the trans-Golgi) results in spurious residence at non-caveolar regions of the plasma membrane proper. In the present study, by using GFP fused to select protein domains (72), we have circumvented such possible artifacts induced by alanine scanning and deletion mutagenesis in the context of the entire caveolin-1 FIG. 13. Diagram summarizing the important functional domains identified within the caveolin-1 molecule. A, protein sequence alignment of the C-terminal domains of caveolin-1, -2 and -3. Residues conserved between caveolin-1 and caveolins 2 and 3 are boxed. Three C-terminal residues that are invariant in all caveolin species sequenced to date (from humans to C. elegans) are indicated by an asterisk (*). We show here that membrane attachment is mediated by the first 16 residues of the caveolin-1 C-terminal domain, whereas oligomer/oligomer interaction requires the final 10 residues of this domain. The first 16 residues of the C-terminal domain of caveolin-2 diverge from the corresponding residues in caveolin-1 and -3, perhaps accounting for the distinct localization pattern observed with GFP-C-Cav-2-(120 -162). Note that the final 8 residues of all three human caveolins are highly divergent, perhaps accounting for the known homotypic isoform specificity of caveolin oligomer/oligomer interactions. Additionally, note that caveolin-2 lacks the cysteinyl residues (bold) that are palmitoylated in caveolin-1 and -3. B, note that the caveolin-1 CSD/N-MAD bears remarkable resemblance to a highly conserved portion of the homeodomain transcription factor engrailed, as described previously (53). C, two adjacent caveolin-1 homo-oligomers (shown as dimers for simplicity) are drawn bound to the membrane. Note that the amino (NH 2 ) and carboxyl (COOH) termini are both cytoplasmic, and the entire molecule is intracellular. Membrane attachment is mediated by the three domains marked with arrows: the N-MAD (residues 82-101), the TMD (residues 102-134), and the C-MAD (residues 135-150) can each anchor a heterologous protein (GFP) to membranes. Palmitate groups are found (hatched marks) in the C-terminal portion of the molecule. The oligomer/oligomer interaction requires the final 10 residues of this domain (TD; residues 168 -178). The position of the homo-oligomerization domain (OD; residues 61-101) is also shown.

molecule.
Taken together with our previous results, the present study suggests that caveolin-1 contains at least two localization signals in its primary sequence (Fig. 13C). The CSD/N-MAD targets the protein to caveolae membranes, and C-MAD targets the protein to the TGN. These results are entirely consistent with the original identification of caveolin-1 as a component of trans-Golgi-derived transport vesicles (44,73). They also suggest that caveolin-1 may move among various membranous compartments because it has affinities for several distinct lipid micro-environments. In light of these results, we propose that the CSD and the first third of the C-terminal domain are bound to the membrane (Fig. 13C). We posit that the steady-state localization of caveolin-1 reflects an integration of at least two localization signals. Other work clearly demonstrates that cellular cholesterol levels (63-66, 74, 75), and reversible posttranslational modifications like phosphorylation and palmitoylation can alter caveolin-1 localization and function dramatically (39,76). The steady-state localization of caveolin-1 represents a vector sum of these forces.