Role of Lipid Modifications in Targeting Proteins to Detergent-resistant Membrane Rafts

Sphingolipid and cholesterol-rich Triton X-100-insoluble membrane fragments (detergent-resistant membranes, DRMs) containing lipids in a state similar to the liquid-ordered phase can be isolated from mammalian cells, and probably exist as discrete domains or rafts in intact membranes. We postulated that proteins with a high affinity for such an ordered lipid environment might be targeted to rafts. Saturated acyl chains should prefer an extended conformation that would fit well in rafts. In contrast, prenyl groups, which are as hydrophobic as acyl chains but have a branched and bulky structure, should be excluded from rafts. Here, we showed that at least half of the proteins in Madin-Darby canine kidney cell DRMs (other than cytoskeletal contaminants) could be labeled with [3H]palmitate. Association of influenza hemagglutinin with DRMs required all three of its palmitoylated Cys residues. Prenylated proteins, detected by [3H]mevalonate labeling or by blotting for Rap1, Rab5, Gβ, or Ras, were excluded from DRMs. Rab5 and H-Ras each contain more than one lipid group, showing that hydrophobicity alone does not target multiply lipid-modified proteins to DRMs. Partitioning of covalently linked saturated acyl chains into liquid-ordered phase domains is likely to be an important mechanism for targeting proteins to DRMs.

Cholesterol and sphingolipid-rich detergent-resistant mem-branes (DRMs) 1 can be isolated from mammalian cells (23). DRM lipids are in a state similar to the liquid-ordered (l o ) phase (3, 24 -26). The l o phase, which requires cholesterol to form, is favored by lipids like sphingolipids, whose long saturated acyl chains give them a high degree of order and a high acyl-chain melting temperature (3). Acyl chain order explains the detergent-insolubility of DRMs (3). We hypothesize that DRMs are an in vitro correlate of rafts in intact membranes. It is important to note that detergent insolubility can underestimate the association of proteins and lipids with the l o phase; some proteins and lipids that are in rafts can be solubilized (25). Nevertheless, DRM association provides a powerful tool for identifying molecules that are likely to have a high affinity for rafts. DRMs isolated from cells contain a number of proteins (27)(28)(29) which are undoubtedly crucial for the function of the domains in vivo. For this reason, it is important to determine how proteins associate with DRMs. Three DRM targeting signals have been defined. First, glycosylphosphatidylinositol (GPI)anchored proteins are targeted to DRMs through acyl chain interactions (23)(24)(25)30). An N-terminal Met-Gly-Cys motif that is present in some Src family kinases and heterotrimeric G protein ␣ subunits, in which Gly is myristoylated and Cys is palmitoylated, can also serve as a DRM targeting signal (31,32). Third, dual palmitoylated Cys residues are required for raft association of the T cell adaptor protein LAT (15) and the neuronal protein GAP-43 (33).
The finding that DRM lipids are in an l o -like phase suggests a unifying mechanism for targeting of proteins to DRMs. Proteins with a high affinity for the ordered lipid environment of the l o phase might spontaneously partition into the domains. In agreement with this model, all three of the DRM targeting signals listed above contain two closely spaced acyl chains. Myristate and palmitate, as well as most of the acyl chains on GPI-anchored proteins (34) are saturated, and thus should fit well into ordered lipid domains. This suggests that acylation, especially multiple acylation, may be a general DRM targeting signal.
Alternatively, however, it might be imagined that lipid modifications could target proteins to DRMs simply through hydrophobic interactions. Both models predict that many DRM proteins would be linked to lipids. The behavior of prenylated proteins should distinguish between the models, because pre-nyl groups are as hydrophobic as acyl chains, but have a bulky branched structure that should not fit well into the l o phase. Because of the possibility that multiple hydrophobic modifications are required for DRM targeting, the behavior of dually or multiply lipid-modified proteins that are prenylated is especially informative in this regard. To test these models, we examined the lipid modifications of DRM proteins and the DRM association of prenylated proteins.

EXPERIMENTAL PROCEDURES
Materials-Mouse monoclonal anti-Rab5 antibodies were the gift of A. Wandinger-Ness or were from Transduction Laboratories (Lexington, KY). Mouse monoclonal anti-Rap1, anti-Ras, and anti-caveolin antibodies (Ig fraction) were from Transduction Laboratories. Rabbit polyclonal anti-influenza hemagglutinin (HA) antibodies (35) were used. Rabbit anti-G ␤ antibody B600, against a C-terminal G ␤ peptide, (36) was the gift of S. Mumby. Mouse monoclonal anti-transferrin receptor antibodies were a gift of I. Trowbridge. Rabbit polyclonal antibodies to p62 yes (Yes) were generated by immunization of rabbits with a TrpE-Yes fusion protein as described (37). A purified Ig fraction was obtained using an immobilized Protein A column according to instructions from the supplier (Pierce, Rockville, IL). Rabbit anti-placental alkaline phosphatase (PLAP; Ig fraction) was from Dako (Carpinteria, CA). Horseradish peroxidase-conjugated goat anti-mouse Ig(G ϩ M) was from Jackson Labs (West Grove, PA), and horseradish peroxidase goat anti-rabbit IgG was from Sigma. The enhanced chemiluminescence (ECL) reagent and Amplify (fluorography enhancement reagent) were from Amersham. Ampholytes were from Bio-Rad and prestained protein molecular weight standards were from Bio-Rad or Life Technologies (Gaithersberg, MD). EXPRE 35 Koke et al. (38) with modifications as described (39). All other materials were purchased from Sigma or Fisher Scientific (Pittsburgh, PA).
Cells and Plasmids-MDCK strain II (40), COS-1 (33), and CV1 cells (35) were maintained as described previously. MDCK cells stably expressing PLAP (detected without butyrate induction) have been described (41). Met-18b-2 cells (Ref. 42; the gift of J. Faust) were grown in Ham's F-12 medium containing 5% iron-supplemented calf serum. Mutant HA cDNAs have been described (35). Briefly, oligonucleotide-directed mutagenesis was used to mutate Cys residues at positions 536, 543, and/or 546 near the C terminus of HA. Mutants were named by three letters referring to the amino acids at these positions, respectively; e.g. SCC contains Ser at 536 and Cys at 543 and 546. All mutants incorporate less [ 3 H]palmitate than wild type, and SSS incorporates none (35). Wild-type and mutant HA proteins were expressed in CV1 cells as described (35). A cDNA encoding H-Ras in the pEXV-3 expression vector (43) was the gift of K. Cadwallader (Addenbrooke's Hospital, Cambridge, United Kinddom). H-Ras was expressed transiently in COS-1 cells as described (44).
Protein Radiolabeling-Metabolic labeling of MDCK cell proteins with [ 35 S]methionine (50 Ci/ml) to steady-state was as described (29). CV-1 cells expressing HA mutants were starved in media lacking Met and Cys for 30 min, pulse labeled for 20 min in media containing 0.6 mCi/ml Tran 35 S-label, and chased for 40 min in complete medium. Lysis in 1% Triton X-100, immunoprecipitation, analysis, and quantitation were as described (45 Preparation of Total Cell Membranes and DRMs-For total cell membranes, cells were scraped from dishes, washed in phosphate-buffered saline (150 mM NaCl, 20 mM sodium phosphate, pH 7.4) and then in hypotonic buffer (10 mM Hepes, pH 7.4, 0.5 mM EDTA), resuspended in 1 ml of hypotonic buffer, and broken by passage through a 25-gauge needle 40 times. Nuclei and debris were removed by centrifugation at 300 ϫ g for 10 min at 4°C, and light membranes were collected from the supernatant by centrifugation at 120,000 ϫ g for 1 h. Membranes were solubilized directly in gel loading buffer for analysis by SDS-PAGE. For DRMs, cells in 1 confluent 10-cm dish were lysed in 1 ml of TNE buffer (25 mM Tris-Cl, pH 7.5, 150 mM NaCl, 5 mM EDTA) containing 1% Triton X-100 (TNE/Triton X-100), and DRMs were isolated by flotation on sucrose gradients as described (40). Unless otherwise indicated, the lysis buffer and all sucrose gradient solutions were adjusted to pH 11 with 0.1 M sodium carbonate. In most cases, the floating membrane band was harvested and diluted to about 12 ml with TNE. Membranes were harvested by centrifugation for 1 h at 120,000 ϫ g. Alternatively, where indicated 1-ml fractions were collected from the bottom of the sucrose gradient with an ISCO (Lincoln, NB) Model 185 density gradient fractionator. All buffers were ice-cold and contained the following protease inhibitors: 0.5 g/ml leupeptin, 0.7 g/ml pepstatin, and 0.2 mM phenylmethylsulfonyl fluoride.
Analysis of DRM Proteins, SDS-PAGE, and Blotting-For identification of GPI-anchored proteins, DRMs recovered from sucrose gradients were solubilized in TNE containing 1% Triton X-114. GPI-anchored proteins were detected as described (46). Briefly, phase separation was induced by warming lysates to 37°C (47). The aqueous phase was removed, and the detergent phase containing hydrophobic proteins was mixed with fresh TNE and incubated with 10 units/ml phosphatidylinositol-specific phospholipase for 1 h at 37°C. The phase separation was repeated, and proteins in the aqueous phase were incubated overnight with phenyl-Sepharose and then precipitated with 15% trichloroacetic acid. DRMs harvested from sucrose gradients and collected by centrifugation as described above were incubated with 1 ml of TNE/ Triton X-100 at 37°C for 5 min. Lysates were subjected to centrifugation at top speed in a microcentrifuge for 5 min. Proteins in the supernatant were precipitated with 15% trichloroacetic acid. Precipitates were washed with acetone and diethyl ether to remove all lipids, solubilized in gel loading buffer, and analyzed by sodium dodecyl sulfatepolyacrylamide gel electrophoresis (SDS-PAGE; Ref. 48) on 11% acrylamide gels. Where indicated, replicate lanes from the same gel were incubated with either buffer containing 1 M hydroxylamine and 1 M Tris-Cl (pH 7.5), or 1 M Tris-Cl (pH 7.5) alone for 4 h with one change of reagent before processing for fluorography. When gradients were fractionated, 75 l of each 1-ml fraction was mixed with 25 l of 4 ϫ gel loading buffer and subjected to SDS-PAGE and Western blotting as described (25). Quantitation of signals in the linear range of the film was performed using a Bio-Rad GS-670 imaging densitometer as follows: % of protein in DRMs ϭ cpm in DRMs (gradient fractions 9 -12))/ cpm in lysate (fractions 1-5) ϩ cpm in DRMs. Although only the region of the blot showing the protein of interest is shown, no other bands were present. Two-dimensional gel electrophoresis, using non-equilibrium pH gradient electrophoresis in the first dimension and SDS-PAGE in the second dimension, was performed according to Jones (49,50) after precipitation of proteins with acetone. In some cases, radiolabeled proteins on two-dimensional gels were transferred to nitrocellulose. After detection of Yes or caveolin by Western blotting, the ECL signal was quenched with 0.05% sodium azide. Blots were then air dried, sprayed with Amplify, and exposed to film to detect the [ 35 S]methionine signal.

RESULTS
Acylated Proteins in DRMs-To determine how many DRM proteins from MDCK cells were palmitoylated, cells were incubated with [ 3 H]palmitic acid before preparation of DRMs. For comparison, DRMs were also prepared from cells incubated with [ 35 S]methionine to metabolically label all proteins. DRM proteins were separated by SDS-PAGE and visualized by fluorography (Fig. 1). As previously reported (29), about 20 -25 major proteins were observed (lane 1). Many of the proteins were labeled with palmitate (lane 2). The same pattern of [ 3 H]palmitate-labeled proteins was also observed in DRMs prepared without sodium carbonate at pH 7.5 (data not shown).
In initial experiments, we also labeled cells with  3 and 4).
Not All Acylated Proteins Are in DRMs-We next determined how many of the palmitoylated proteins in the cell were present in DRMs. Cells were incubated with [ 3 H]palmitate and lysed. DRMs were then prepared after saving a small fraction of the whole cell lysate. Both fractions were analyzed by SDS-PAGE and fluorography (Fig. 3). Very different patterns of labeled proteins were seen in whole cell lysate and DRMs. Thus, only a subset of palmitoylated proteins associated efficiently with DRMs.
Two-dimensional Gel Analysis of DRM Proteins-We extended our analysis of DRM proteins using two-dimensional gels, separating [ 35 S]methionine-labeled proteins in the first dimension by non-equilibrium pH gradient electrophoresis, and in the second by size by conventional SDS-PAGE. We first compared total cell proteins ( Fig. 4A) with DRM proteins prepared from cells with or without carbonate during detergent extraction (Fig. 4, B and C). For orientation, MDCK cells stably expressing the GPI-anchored protein PLAP were used in this experiment. PLAP was prominent among the DRM proteins (Fig. 4, B and C), and could be detected in the whole cell lysate after long exposure (not shown).
Ten proteins that were especially abundant in the whole cell lysate (Fig. 4A, arrows) were detected in DRMs prepared without carbonate (Fig. 4B, asterisks). These are likely to be cytoskeletal or other structural proteins. Consistent with this possibility, fibrous material could sometimes be observed in DRM preparations by electron microscopy (not shown). Most of these proteins were present in lower abundance in DRMs prepared with carbonate (Fig. 4C, asterisks). Actin, the most abun-dant cellular protein, was among these proteins. Actin could also be detected in DRMs on one-dimensional gels by Western blotting (not shown). Although these abundant proteins may associate specifically with DRMs, it is likely instead that they adhere nonspecifically during DRM preparation.
Finding these 10 abundant proteins in DRMs raised the possibility that most of the spots on our two-dimensional gels were contaminants. Proteins that associated specifically with DRMs, and were enriched there, might be present at such low levels that they could not be seen. To test this idea, as described next, we identified several proteins that are known to associate specifically with DRMs. Our ability to detect these proteins suggested that many of the other spots also corresponded to bona fide DRM proteins.
We first examined [ 35 S]methionine-labeled GPI-anchored proteins released from DRMs by phosphatidylinositol-specific phospholipase treatment (Fig. 5B). (No proteins were observed when phosphatidylinositol-specific phospholipase was omitted from the reaction (not shown).) These were analyzed in parallel with total [ 35 S]methionine-labeled DRM proteins (Fig. 5A). (For orientation, major cell proteins identified in Fig. 4 are labeled with asterisks.) As found by others (53), GPI-anchored proteins of 50 and 80 kDa could be detected in the whole DRM pattern by alignment of the films. These are labeled GPI in Fig.  5, A and C.
We next identified two known DRM proteins on two-dimensional gels by Western blotting. DRM proteins from [ 35 S]methionine-labeled MDCK cells were separated on two-dimensional gels and transferred to nitrocellulose. The positions of the Src family non-receptor tyrosine kinase Yes and of caveolin, a marker for caveolae (4,54), were determined by Western blotting (not shown). Blots were then exposed to film for detection of all [ 35 S]methionine-labeled DRM proteins. Alignment of the films allowed identification of Yes and caveolin in the [ 35 S]methionine-labeled pattern. Spots corresponding to Yes and caveolin (labeled Cav) are labeled in Fig. 5, A and C. (The caveolin spot is faint because much of the caveolin was in a high molecular weight oligomeric form (55). Oligomeric caveolin was detected by Western blotting, but could not be aligned unambiguously with an [ 35 S]methionine-labeled spot.) Thus, although these proteins are not the most abundant DRM pro-teins, they are easily detectable, increasing our confidence that many of the unidentified proteins are also specific.
As expected, except for faint labeling of a protein of about 30 kDa (Fig. 5C, marked 10,*), the major cell proteins in DRMs were not labeled with [ 3 H]palmitate. Neither could we detect [ 3 H]palmitate labeling of the two GPI-anchored proteins. Although palmitate is present in GPI anchors, the anchors are preassembled and added to proteins en bloc (56). Thus, a 2-h labeling period was probably not sufficient to allow detectable labeling of these proteins. However, in agreement with data from the one-dimensional gels, a large fraction of the other proteins were labeled with [ 3 H]palmitate. Twenty-three palmitoylated proteins (including caveolin) and only 12 non-palmitoylated proteins (other than cytoskeletal or GPI-anchored proteins) are labeled in Fig. 5C, the former with numbers corresponding to [ 3 H]palmitate-labeled proteins in Fig. 5D, and the latter with capital letters. The fact that we did not detect [ 3 H]palmitate labeling of caveolin, although it is triply palmitoylated (57), suggests that other less-abundant DRM proteins might also be palmitoylated, but not detectable in Fig. 5D.
Finding that so many DRM proteins were palmitoylated was consistent with the idea that palmitoylation is a DRM targeting signal. Thus, we next examined several known palmitoylated proteins for DRM association, in order to test the role of palmitoylation in DRM targeting directly. Vesicular stomatitis virus glycoprotein does not associate with DRMs (23,33). Similarly, we found that endogenous MDCK cell transferrin receptor (another palmitoylated transmembrane protein (58)) was not in DRMs (data not shown). In contrast, influenza HA, which is triply palmitoylated (35), associates with DRMs in several cell types (45,59,60). In agreement with this result, we found that 50% of HA expressed in MDCK cells associated with DRMs (not shown).
DRM Association of Influenza Hemagglutinin Requires Palmitoylated Cys Residues-To test the role of the palmitoylated Cys residues in DRM targeting, HA proteins mutated in one, two, or all three Cys residues were expressed in CV1 cells. After detergent extraction and separation of soluble and insoluble fractions, HA was recovered from both fractions by immunoprecipitation as described (60), and the percent insoluble was determined. Using this procedure, 29% of wild-type HA was insoluble. (The difference between this and the 50% of HA found in DRMs in transiently transfected MDCK cells may reflect cell type or other procedural differences. It is also possible that the exogenously expressed HA was incompletely palmitoylated.) Mutation of any Cys, or any combination of Cys, essentially abolished detergent insolubility (Fig. 6). Thus, all three palmitoylated Cys residues are essential for targeting HA to DRMs.
Prenylated Proteins in DRMs-We examined DRM proteins for possible prenylation by incubating MDCK cells with [ 3 H]mevalonate, a precursor of prenyl groups. Cells were then lysed as usual, except that sodium carbonate was omitted and lysis was performed at pH 7.5.
To detect all cellular prenylated proteins, proteins in 10% of the lysate were precipitated with trichloroacetic acid and analyzed by SDS-PAGE (Fig. 7A, WCL). DRMs were isolated from the remaining 90% of the lysate (Fig. 7A, DRM). Although [ 3 H]mevalonate-labeled proteins were easily seen in whole cell lysates, they were virtually undetectable in DRMs.
Most mammalian cells take up mevalonate poorly, making it difficult to detect prenylated proteins by [ 3 H]mevalonate labeling. For this reason, we repeated the experiment shown in Fig.  7A using met-18b-2 cells (42), which express a mutant mevalonate transporter that allows faster uptake of mevalonate (61). Results are shown in Fig. 7B. As expected, [ 3 H]mevalonate labeling was more efficient than in MDCK cells. As was seen for MDCK cell DRMs, very few met-18b-2 cell DRM proteins were labeled.
In contrast to our findings, three prenylated proteins, Rab5 (62), Rap1 (27,62), and the ␤␥ component of heterotrimeric G proteins (63), have been detected by others in DRMs. For this reason, we next examined MDCK cell DRMs for the presence of these proteins. Proteins in total cell membranes or in DRMs prepared from at least 10 times as many cells were separated by SDS-PAGE and transferred to nitrocellulose. Although both Rab5 and Rap1 were easily detected in total cell membranes (Fig. 8, WM), they were barely detected in DRMs (Fig. 8, DRM). Approximately 1% of the total cellular Rap1 was present in DRMs.
The prenylated G ␥ subunit is responsible for membrane targeting of the G ␤␥ complex. 30% of this complex was found in DRMs isolated from a neuroblastoma cell line grown with serum (63). In contrast, very little G ␤␥ was found in chicken gizzard DRMs (27). We subjected lysates of MDCK cells stably expressing PLAP to sucrose gradient ultracentrifugation, fractionated the gradients, and examined the distribution of G ␤␥ and PLAP between the Triton-soluble lysate fractions and floating DRMs (Fig. 9A, panels 1 and 2). Although 85% of PLAP was recovered in the DRM fractions, G ␤␥ was barely detectable. Similar results were obtained whether or not the Triton X-100 lysis buffer contained sodium carbonate.
Most members of the Ras family are both farnesylated and palmitoylated (43). Thus, if all lipid modifications enhance DRM targeting, then Ras should be highly enriched there. In contrast, if packing into l o phase rafts is important, then the prenyl group might inhibit association of Ras with DRMs. In agreement with the latter model, Ras is excluded from mouse lung DRMs (62). We repeated the gradient fractionation procedure used for G ␤␥ to show that Ras is also excluded from MDCK cell DRMs (Fig. 9A, panels 3 and 4).
Although H-ras, N-ras, and K-rasA are both palmitoylated and prenylated, K-RasB is not palmitoylated, but is targeted to membranes by prenylation in combination with a polybasic domain (43). Because we do not know how much MDCK cell Ras is of this type, we confirmed that H-Ras is excluded from DRMs by expressing the protein exogenously in COS-1 cells. Western blotting of 1% of a whole cell lysate or DRMs prepared from the remaining 99% of the lysate showed that less than 1% of the total was in DRMs (Fig. 9B). DISCUSSION Several findings suggest that cholesterol and sphingolipidrich l o phase microdomains or rafts can exist in cell membranes and can be isolated as DRMs (1,3). First, GPI-anchored proteins and gangliosides, which are enriched in DRMs, can exhibit a clustered distribution in the plasma membrane (64 -69). Furthermore, the physical properties of DRMs isolated from cells are very similar to those of the l o phase (24). In a complementary approach, we demonstrated the plausibility of phase separation in biological membranes by showing that l o phase microdomains form spontaneously at 37°C in liposomes containing physiologically reasonable levels of sphingolipids and cholesterol (26).
Emerging functions for rafts in the structure and function of caveolae, in signal transduction, and in sorting in the secretory and endocytic pathways (1,2,5) highlight the importance of determining how lipids and proteins associate with them. If rafts are l o phase microdomains, as we propose, then acyl chain order should be a key determinant of their formation. As expected, lipids with saturated acyl chains, whose extended structure fits well into an ordered environment, are enriched in DRMs isolated from cells (23) and model membranes (24,25). Proteins, too, might be targeted to DRMs by modification with saturated acyl chains. In agreement with his idea, the best defined DRM-targeting signals (GPI anchorage (30,39), tandem myristoylation and palmitoylation (31,32), and dual palmitoylation (15,33) all consist of saturated acyl chains. The role of palmitoylation in targeting proteins to DRMs was tested further in this paper. An important finding of this work is that a high fraction of the proteins in DRMs is acylated. This suggests that acylation is a commonly used signal for DRM targeting of proteins. Other proteins might be targeted to DRMs indirectly, by binding to more tightly associated proteins or lipids.
It is important to note, however, that not all palmitoylated proteins are targeted to DRMs. This suggests that although palmitoylation can increase the affinity of proteins for DRMs, this effect is not always strong enough to mediate stable association. Other structural features (for instance, prenylation or membrane spans) that prefer a disordered environment will also contribute to the overall affinity of the protein for DRMs. The degree of partitioning of a protein into DRMs will reflect all these interactions. Thus, modification with multiple acyl chains should enhance DRM association. By contrast, a membrane-spanning peptide might not pack easily into such an environment. In agreement with this idea, to our knowledge all proteins examined to date that lack membrane-spanning domains but are modified with dual saturated acyl chains are targeted to DRMs. (This includes the myristoylated and palmitoylated protein endothelial cell nitric oxide synthase, although it associates with DRMs less efficiently than other such proteins (70).) The "rules" for how palmitoylation can target transmembrane proteins to DRMs are less straightforward. We showed here that dual palmitoylation is not sufficient for targeting of vesicular stomatitis virus glycoprotein G or the transferrin receptor to DRMs, and that three palmitate groups are required to target HA to DRMs. As HA is trimeric, each molecule is modified with a total of nine palmitate chains. This high concentration of saturated acyl chains may be required to overcome packing difficulties and allow efficient targeting to DRMs. In addition, transmembrane domain sequences undoubtedly play an important role in determining the affinity for DRMs, as has been shown for HA (60) and the hyaluronan receptor CD44 (71,72). (CD44 is palmitoylated (73), and the role of this modification in its DRM association has not been explored.) However, Zhang et al. (74) have shown that dual acylation is sufficient for targeting of LAT, an membrane-spanning adaptor protein that plays an important role in T cell signaling, to DRMs (15).
An alternative model is that lipid modifications target proteins to DRMs through hydrophobic interactions that do not depend on the structure of the lipid. Prenyl groups are hydrophobic, but would not be expected to fit in an l o domain. Thus, the finding that prenylated proteins are excluded from DRMs provides important support for our model. Nevertheless, it might be imagined that prenyl groups could aid in targeting proteins to DRMs via hydrophobic interactions, but that DRM targeting requires two lipid modifications. This model would predict that most Rab proteins would be enriched in DRMs, as most of them (including Rab5) are dually geranylgeranylated (75). However, we found that Rab5 was excluded from DRMs ( Fig. 8), and we found no enrichment there of [ 3 H]mevalonatelabeled proteins in the 20 -25 kDa range that is characteristic of Rabs (Fig. 7). Thus, dual geranylgeranylation does not target Rab proteins efficiently to DRMs.
Such a model would also predict an enrichment of Ras in DRMs, as most forms of Ras are both prenylated and palmitoylated. Our finding (in agreement with others) that Ras is excluded from DRMs supports the model that lipid modifications target proteins to DRMs via packing order, not hydrophobicity. Exclusion of prenylated proteins from rafts may have important functional consequences.
The high frequency of palmitoylation among DRM proteins suggests that acylation is a common means of targeting proteins to rafts. However, there must be additional targeting signals, as at least one protein, caveolin, associates with DRMs even after removal of palmitoylation sites by mutagenesis (57). Caveolin, which binds cholesterol tightly (76) and can induce the formation of caveolae when expressed in caveolae-negative cells (77), appears to be an unusual protein that may interact with rafts in a unique manner.
In summary, GPI-anchored proteins, Src family kinases, GAP-43, HA, and LAT are now known to be targeted to DRMs via saturated acyl chains, and the list continues to grow. The affinity of these lipid groups for an ordered environment, and not simply their hydrophobicity, is required for targeting. A substantial fraction of DRM proteins is acylated. Thus, acylation appears to be a common mechanism of increasing the affinity of proteins for DRMs, and may be the primary targeting mechanism for proteins without membrane spans. As the role of rafts in cellular function is becoming increasingly clear (1,2,15,16), it is becoming increasingly important to understand in molecular detail how proteins and lipids are organized into these domains.