Glycosphingolipids Are Not Essential for Formation of Detergent-resistant Membrane Rafts in Melanoma Cells

Recent data suggest that membrane microdomains or rafts that are rich in sphingolipids and cholesterol are important in signal transduction and membrane trafficking. Two models of raft structure have been proposed. One proposes a unique role for glycosphingolipids (GSL), suggesting that GSL-head-group interactions are essential in raft formation. The other model suggests that close packing of the long saturated acyl chains found on both GSL and sphingomyelin plays a key role and helps these lipids form liquid-ordered phase domains in the presence of cholesterol. To distinguish between these models, we compared rafts in the MEB-4 melanoma cell line and its GSL-deficient derivative, GM-95. Rafts were isolated from cell lysates as detergent-resistant membranes (DRMs). The two cell lines had very similar DRM protein profiles. The yield of DRM protein was 2-fold higher in the parental than the mutant line, possibly reflecting cytoskeletal differences. The same amount of DRM lipid was isolated from both lines, and the lipid composition was similar except for up-regulation of sphingomyelin in the mutant that compensated for the lack of GSL. DRMs from the two lines had similar fluidity as measured by fluorescence polarization of diphenylhexatriene. Methyl-β-cyclodextrin removed cholesterol from both cell lines with the same kinetics and to the same extent, and both a raft-associated glycosyl phosphatidylinositol-anchored protein and residual cholesterol showed the same distribution between DRMs and the detergent-soluble fraction after cholesterol removal in both cell lines. Finally, a glycosyl phosphatidylinositol-anchored protein was delivered to the cell surface at similar rates in the two lines, even after cholesterol depletion with methyl-β-cyclodextrin. We conclude that GSL are not essential for the formation of rafts and do not play a major role in determining their properties.

Recent studies suggest that plasma membrane lipids do not always mix homogeneously but that membranes may contain microdomains or rafts that are rich in sphingolipids and cholesterol (1)(2)(3). Rafts may be concentrated or stabilized in caveolae but also exist in cells that lack caveolae. Rafts have been proposed play important roles in signal transduction; for instance, recruitment of signaling proteins to rafts in T cells (4 -8) and basophils (9 -11) appears to be required for signaling. Rafts may also play a role in intracellular sorting. For instance, depletion of cholesterol, sphingolipids, or certain raftassociated proteins affects sorting of apical proteins in epithelial cells (12)(13)(14)(15)(16)(17) and axonal proteins in neurons (18,19). Recent studies also suggest that rafts play a role in sorting in the endocytic pathway (20,21).
Two models for the organization of lipids in rafts have been proposed. The first was developed by Simons and colleagues (1,22,23) as part of a model for sorting of apical and basolateral proteins in the trans-Golgi network of polarized epithelial cells. In this model, clusters of GSL 1 form spontaneously in the trans-Golgi network. Apical proteins partition into these clusters, and the resulting rafts are packaged into apical transport vesicles. In early versions of the model, it was suggested that GSL might self-associate through hydrogen bonds between GSL headgroups and between the hydroxyl groups of the sphingosine base and the hydroxy fatty acid present on many sphingolipids (22). More recently, it was suggested that sphingolipids might interact through weak interactions between the GSL carbohydrate headgroups (1). Because cholesterol is now known to be an important component of rafts, it was proposed that cholesterol, with its small headgroup, is recruited to rafts because of its ability to fill gaps in the bilayer created by the discrepancy in size between the large GSL headgroups and their acyl chains (1).
The second model for the structure of rafts was developed from studies of rafts isolated by their insolubility in nonionic detergents such as Triton X-100 (3,24,25). This model postulates that interactions between lipid acyl chains play a key role in raft formation and that rafts exist in membranes as domains in the liquid-ordered (l o ) phase (3). In particular, the high acyl chain melting temperature (T m ) of sphingolipids was proposed to promote phase separation and formation of l o phase domains in the presence of high amounts of cholesterol.
Several pieces of evidence support the importance of acyl chain interactions in raft formation. First, sphingomyelin (which has the same phosphocholine headgroup as phosphatidylcholine) and GSL are equally enriched in DRMs isolated from cells (26). Further support came from studies of sphingolipid-and cholesterol-rich model membranes that contain l o phase domains and can form DRMs upon detergent extraction. When two phosphatidylcholine species, one with saturated acyl chains and a high T m and the other with unsaturated acyl chains and a low T m , were incorporated into these membranes and subjected to detergent extraction, the high T m lipid but not the low T m lipid associated with DRMs (25). Finally, a GPIanchored protein associated with detergent-insoluble gel phase dipalmitoyl phosphatidylcholine (DPPC) domains in model membranes (27), showing that sphingolipid-specific hydrogen bonding is not required for organization of these domains.
The use of detergent insolubility in studying rafts has raised concern about potential detergent-induced artifacts. Indeed, we found that detergent insolubility can underestimate raft association of proteins and lipids (27,28) and is therefore not a quantitative measure of raft association. However, several potentially serious detergent effects (including artifactual creation of domains from previously homogeneous bilayers and recruiting of nonraft associated proteins and lipids into rafts during lysis) do not occur (3,27,29). Thus, enrichment of proteins or lipids in DRMs appears to be a good indication that they associate with rafts in vivo, and DRMs have proven to be very useful tools for studying rafts.
Glycosphingolipids play important roles in cell recognition, signaling, and other processes (30,31) and are essential for mammalian development (32). However, some mammalian cells that lack glycosphingolipids can be grown in culture (33).
In this study, we tested the role of GSL in raft formation in vivo using the GM-95 melanoma cell line, which is deficient in GSL because it lacks ceramide glucosyltransferase, the first enzyme in GSL synthesis (34). Sphingomyelin is up-regulated in GM-95, and the total amount of sphingolipid is similar in GM-95 and the parental MEB-4 line (35). Up-regulation of sphingomyelin in GM-95 is a direct effect of the lack of glycosphingolipids, as shown by the fact that expression of recombinant ceramide glucosyltransferase restores both GSL and sphingomyelin to wild-type levels (35). We compared DRMs and several other parameters that might reflect differences in raft structure between MEB-4 and GM-95 cells.
We also examined the transport kinetics of placental alkaline phosphatase (PLAP), a GPI-anchored protein, through the secretory pathway to the cell surface. GPI-anchored proteins have a high affinity for rafts and are apically targeted in epithelial cells (36). Although melanoma cells are not polarized, two separate transport pathways, cognates of apical and basolateral pathways, have been shown to exist in nonpolarized cells (37)(38)(39). Thus, if the presence of GSL in rafts were crucial for the incorporation of GPI-anchored proteins into transport vesicles, transport would be slowed in GSL-negative cells. However, PLAP was delivered to the cell surface at the same rate in the two cell lines. Depletion of cholesterol with MBCD did not affect PLAP transport in either line.
Other Materials-MBCD was from Sigma. The Renaissance chemiluminescence reagent, [1,2,6,  DRM Preparation-DRMs were prepared by ultracentrifugation of cold Triton X-100 cell lysates in sucrose step gradients, either harvesting the DRM band or fractionating the gradient, as described (43) with the following minor changes. The lysis buffer but not gradient solutions contained 0.1 M sodium carbonate, pH 11. Lysates were passed through a 22 gauge needle to shear DNA before mixing with Tris-NaCl-EDTA (TNE) buffer (43) containing 80% sucrose and placing at the bottom of the centrifuge tube. These conditions allowed more reproducible recovery of DRMs, which otherwise stuck variably to cytoskeleton and pelleted in the sucrose gradients. Under these conditions, sucrose gradient pellets were small and contained very little protein. Except where noted, cells in each 10-cm dish were lysed in 1 ml of TNE/carbonate (43) containing enough Triton X-100 to give the indicated w/w ratio of detergent to total cell protein, determined by first measuring cell protein in a parallel dish. In a pilot experiment, equal amounts of DRM protein were recovered if 10-cm dishes of MEB-4 cells were extracted with 1 ml of TNE/carbonate containing 2% Triton X-100/dish or with 2 ml of TNE containing 1% Triton X-100/carbonate/dish, showing that the detergent:protein ratio and not the concentration of Triton X-100 determined DRM yield. For lipid analysis, ultracentrifugation was overnight at 24K RPM in an SW28 rotor (Beckman, Palo Alto, CA); for other experiments, it was for at least 3 h, and sometimes overnight at 28K RPM in an SW41 rotor. To measure [ 3 H]cholesterol, 0.1-ml aliquots of each 1-ml fraction were analyzed by scintillation counting. The fraction of [ 3 H]cholesterol in DRMs was determined by dividing the cpm in gradient fractions 9 -12 by the total cpm in the gradient.
Lipid Analysis-For extraction of whole cell lipids, cells were lysed with methanol after rinsing with phosphate-buffered saline (PBS) (43). Lysates were transferred to glass tubes with Teflon-lined caps with several methanol rinses, brought to 1:1 chloroform:methanol, sonicated, and incubated overnight at room temperature with rocking. In harvesting DRM bands from the interface between 5 and 38% sucrose gradient steps for lipid analysis, care was taken to avoid a triglyceride-rich fat cake that floated on the 5% layer. DRMs were extracted with 1:1 chloroform:methanol as for whole cell lipids. Neutral and acidic lipids were separated (45,46) for quantitation of DRM lipids but not for quantitation of cholesterol from whole cell extracts. In all cases, lipids were desalted and (except for GM3) analyzed by quantitative TLC by charring in the presence of cupric acetate and densitometry in comparison with standards on the same plate (45, 46) with minor modifications as described (24). Quantitation was performed using a Bio-Rad GS-670 imaging densitometer.
Because GM3 is the only major ganglioside in MEB-4 cells, it was quantitated (assuming a molecular weight of 1182) by measuring organic sialic acid by a standard resorcinol assay as follows. Lipids were dried under N 2 and dissolved in 500 l of H 2 O. 500 l of resorcinol reagent (made as follows: 50 mg of resorcinol were dissolved in 2.5 ml of H 2 O; 20 ml of concentrated HCl containing 0.0625 ml of 0.1 M CuSO 4 was added; volume was brought to 25 ml with H 2 O) was added, and samples were boiled for 15 min, chilled on ice 2 min, and extracted with 0.8 ml of butyl acetate/butanol 85:15. The absorbance of the upper phase was measured at 580 nm and compared with a standard curve (2-30 g/ml sialic acid).
Fluidity Measurements-Fluorescence polarization of DPH in large unilamellar liposomes prepared by suspension of dried lipids in PBS, sonication, and three freeze-thaw cycles was measured as described (25). DRMs isolated from four confluent 10-cm dishes of each cell type were suspended in 2 ml of PBS with sonication and divided in half. DPH was added to one tube to a final concentration of 100 nM from a 100 M stock in tetrahydrofuran; the other was used as a background control. Fluorescence polarization was measured as described (25). Control experiments showed that DPH fluorescence polarization was not affected by the ratio of DPH to DRM. Labeling with [ 3 H]Cholesterol-Cells in 10-cm dishes were incubated with 3 Ci of [ 3 H]cholesterol/10 ml normal growth media for 2 days, trypsinized, and transferred to a fresh dish to avoid contamination of lysates with [ 3 H]cholesterol that adhered to the dish (47). After growth for about 6 h to allow adhesion to the dish, medium was replaced with serum-free DMEM plus 1% bovine serum albumin, to minimize esterification of [ 3 H]cholesterol (47), for growth overnight before the experiment.
Cholesterol Depletion-Except for surface delivery experiments, (described separately below), cells were grown to confluency in 10-cm dishes. Medium was removed, and monolayers were rinsed once with PBS. 2.5 ml of DMEM without serum was added to each dish. At various times, this medium was replaced with 2.5 ml of the same medium containing 10 mM MBCD. Times of addition of MBCD were staggered so that all cells were exposed to serum-free medium for the same time. Cells remained adherent to the dish for up to 2 h of exposure to MBCD, although they often could be detached easily after 60 min. (Most cells detached after 150 min.) After MBCD treatment, cells were processed as described above (for cholesterol determination) or placed on ice and extracted with TNE/Triton X-100 for DRM preparation.
Surface Delivery of PLAP-Determination of surface delivery rate was as essentially as described (48,49). Briefly, MEB-4 or GM-95 cells stably expressing PLAP in 35-mm dishes were incubated 45 min with or without 10 mM MBCD in 1 ml of DMEM lacking methionine and serum; pulse labeled 20 min with 100 l of DMEM containing 2 mCi/ml [ 35 S]methionine without methionine or serum; chased for 0, 15, 30, 45, 60, 120, or 180 min in DMEM without serum containing 5 mM methionine and 0.1% bovine serum albumin; subjected to four rounds of cell surface biotinylation with 0.5 mg/ml sulfo-NHS-biotin or sulfo-NHS-LC-biotin (with similar results) in PBS; quenched; and subjected to sequential immunoprecipitation and collection of biotinylated proteins on streptavidin-coated agarose beads and fluorography. 2 l (10% of the total) of the eluted immune complexes were removed before addition to immobilized streptavidin and analyzed on a separate gel to determine the relative amounts of total labeled PLAP at each time point, to control for differences in labeling efficiency, cell loss, and degradation. Quantitation of radioactivity in both total immunoprecipitated and streptavidin-bound PLAP was performed with a Storm model 840 Phospho-rImager (Molecular Dynamics, Sunnyvale, CA). The values of streptavidin-bound PLAP were normalized for the total amount of labeled PLAP at each time point.
Cholesterol Synthesis-GM-95 cells in 35-mm dishes were incubated with DMEM without methionine or serum containing 10 mM MBCD for 45 min, and then for 20 min with methionine-free DMEM, whereas control dishes were maintained in growth medium. One MBCD-treated and one control dish were used for protein determination; protein varied by less than 10% between samples. Other dishes were then rinsed with PBS and incubated with 100 l of DMEM containing 0.1% bovine serum albumin, 5 mM methionine, and 2 mCi/ml [ 3 H]acetate (dried under nitrogen before dissolving in medium) for 20, 40, or 60 min. Cells were then lysed with methanol and transferred to glass tubes, and dishes were rinsed twice with methanol. An equal volume of chloroform was added, and tubes were incubated overnight at room temperature. Cell debris was pelleted by centrifugation at 2,000 RPM for 10 min. Supernatants were transferred to new tubes, dried under nitrogen gas, and desalted (50). Desalted lipids were analyzed by thin layer chromatography (hexane/isopropyl ether/acetic acid 65:35:2). The cholesterol band was visualized with iodine vapor and scraped from the plate, and radioactivity was quantitated in a scintillation counter.

Detergent:Protein Ratio Affects Recovery of DRM Protein-To
compare recovery of DRMs from MEB-4 and GM-95 cell lines, we initially extracted confluent monolayers of each cell line in 10-cm dishes in 1 ml of TNE containing either 1, 2, or 3% Triton X-100, prepared DRMs from the lysates, and measured the protein concentration. Although DRMs were recovered efficiently from both cell types, recoveries were quite variable. We speculated that the ratio of detergent to cellular protein and lipid might affect DRM recovery, because MEB-4 cells generally grew to higher density than GM-95 cells (not shown). To test the role of detergent in raft protein yield, we plotted the fraction of total cell protein recovered in DRMs in each of our initial experiments versus the w/w ratio of detergent to whole cell protein (Fig. 1, circles). As expected, more DRM protein was recovered at lower detergent:protein ratios. Recovery appeared to plateau at detergent:protein ratios greater than 5 or 6. Slightly less DRM protein was recovered from GM-95 than MEB-4 cells at each detergent:protein ratio. We repeated the experiment, lysing MEB-4 and GM-95 cells at a detergent: protein ratio of 10, after determining protein concentration from a dish grown in parallel. As shown in Fig. 1 (squares), we recovered 0.18 Ϯ 0.04% of total cellular protein in DRMs from MEB-4 cells and 0.09 Ϯ 0.02% from GM-95 (n ϭ 4, Ϯ SD). In further experiments, we prepared DRMs from cells lysed at known detergent:protein ratios.
Similar MEB-4 and GM-95 DRM Protein Profiles-Although similar amounts of DRM protein were isolated from the two cell types, it was possible that some proteins might require GSL for DRM association. To test this idea, we examined the protein profile of DRMs from the two cell types. Cells were labeled to steady state with [ 35 S]methionine and extracted at detergent: protein ratios of 2.5:1, 5:1, and 10:1. DRMs prepared from the lysates were subjected to SDS-PAGE and proteins detected by fluorography (Fig. 2). DRMs from the two cell types had very similar protein profiles, showing that most proteins do not require glycosphingolipids to associate with DRMs. Consistent with differences in overall DRM protein recovery (Fig. 1), more protein was recovered from each cell type at a detergent:protein ratio of 2.5:1, whereas recovery was similar at ratios of 5:1 and 10:1. However, the differences were more quantitative than qualitative; similar protein profiles were seen under all conditions, but more of each protein was recovered when less Triton X-100 was used.
DRM Marker Proteins Are Equally Abundant in MEB-4 and GM-95 DRMs-Because cytoskeletal contaminants can be prominent DRM proteins (43), it was important to examine individual proteins known to associate specifically with DRMs (51,52). We first chose three endogenous proteins, the Src family kinase Yes and the heterotrimeric G proteins G␣ s and G␣ i . 5 g of DRM protein from cells extracted at detergent: protein ratios of 3, 5, or 9 were subjected to SDS-PAGE and transferred to nitrocellulose, and Yes, G␣ s , and G␣ i were detected by Western blotting (Fig. 3A). All three proteins were present at similar levels in DRMs from the two cell types. We next examined DRM association of a GPI-anchored protein in the two cell lines, using cells stably expressing PLAP. PLAP was detected in sucrose gradients by Western blotting. The protein was highly concentrated in DRM fractions in both cell lines (Fig. 3B).
We could not unambiguously identify caveolin, another DRM marker protein, in MEB-4 or GM-95 cells. Caveolin was not visible on Western blots under conditions in which it was easily detected in another mouse cell line (Fig. 3C). When cell lysate containing 80 g of protein was loaded on the gel, and the blot was exposed for a longer time, a band comigrating with caveolin was detected (not shown). However, a large number of background bands were also visible on the blot under these conditions. Because another group (53) readily detected caveolin in the F10 subline of the B16 cell line from which MEB-4 was derived, we were concerned that MEB4 cells might have changed in culture and might contain substantially less caveolin than the parental B16 line. For this reason, we examined both the B16-F1 and F10 sublines (of lower and higher metastatic potential respectively (54)) for caveolin. The protein was equally difficult to detect in all these lines (Fig. 3C). The fact that Iwabuchi et al. (53) immuno-isolated caveolin-containing membranes may have facilitated detection of the protein in their study by concentrating the protein. DRMs isolated by our method should include both the GM 3 -rich "glycosphingolipid signaling domain" and the caveolin-containing membrane fraction identified by those workers.

MEB-4 and GM-95 DRM Lipid
Compositions-We next measured MEB-4 and GM-95 DRM lipids. The total yield from the two cells types was very similar (Table I). Sphingomyelin was more abundant in DRMs from GM-95 than from MEB-4, and the total DRM sphingolipid was very similar in the two cell types. The lipid composition of DRMs from the two cell types was otherwise similar.
Similar "Fluidity" of MEB-4 and GM-95 DRM Lipids-Because GSL can have unusually high acyl chain melting temperatures (25), they might be imagined to alter the physical properties of DRMs. A rough measure of membrane fluidity or acyl chain mobility can be obtained from the fluorescence polarization of DPH. We have used this method to show that the fluidity of lipids in detergent-insoluble sphingolipid and cholesterol-rich model membranes (25) and in DRMs isolated from MDCK cells 2 is similar to that of lipids in the l o phase.
As shown in Table II temperature was characteristically low in control fluid phase dioleoyl phosphatidylcholine liposomes and high in gel phase DPPC liposomes. As found previously (25), fluorescence polarization was slightly lower in the l o phase (DPPC:cholesterol 2:1) (55) than in the gel phase (Table II). Fluorescence polarization of DPH in both MEB-4 and GM-95 DRMs was similar to that in control l o phase liposomes at room temperature. Probably because of the effect of cholesterol on promoting phase separation (3,29,56), fluorescence polarization remained high in both cell lines even at 50°C, above the T m of sphingomyelin (Table II). By contrast, fluorescence polarization in DPPC above its T m of 41°C declined to a value similar to that of dioleoyl phosphatidylcholine, as expected. Isolating DRMs at a Triton X-100: protein ratio of 2.5:1 instead of 10:1 had little effect on DPH fluorescence polarization (not shown). Thus, DRMs from the two cell lines have the same fluidity.
Cholesterol Removal with MBCD-Cyclodextrins efficiently remove cholesterol from cell membranes and can perturb raft structure and function (11,13,(57)(58)(59)(60)(61)(62)(63). Sphingomyelinase treatment of fibroblasts profoundly increased the rate of cholesterol removal by a cyclodextrin (64), suggesting that interaction of cholesterol with rafts might slow removal. We reasoned that if GSL have a specific effect on the organization of raft lipids, they might affect the efficiency or kinetics of cholesterol removal. Thus, we compared cholesterol removal by MBCD from MEB-4 and GM-95 cells (Fig. 4). Cholesterol removal from the two cell lines occurred with very similar kinetics and to the same extent.
To further probe potential differences in raft structure with MBCD, we examined the effect of the drug on DRM formation in the two cell types. We focused first on the DRM association of PLAP. MEB-4 or GM-95 cells expressing PLAP were treated with MBCD for varying times, and then extracted at a detergent:protein ratio of 10:1. (In control experiments, no significant loss of PLAP from whole cell lysates was detected after 60 min MBCD treatment (not shown)). Lysates were subjected to sucrose gradient ultracentrifugation, fractionation, and detec-tion of PLAP by Western blotting (Fig. 5). Progressive redistribution of PLAP from DRMs into the Triton-soluble fractions was observed. After 10 or 30 min of MBCD treatment, however, most PLAP was still present in the DRM-containing fractions in both cell types. Results were somewhat more variable after 60 min of treatment, but the variation did not correlate with cell type. We conclude that the requirement for DRM association of PLAP in MEB-4 and GM-95 cells is similar.
We next used MBCD to study the effect of cholesterol depletion on the organization of raft lipids. To do this, we examined the DRM association of the cholesterol that remained in cells after MBCD treatment. Monolayers were prelabeled with [ 3 H]cholesterol and then treated with MBCD for varying times before extraction at a Triton X-100:protein ratio of 10:1. Lysates were subjected to sucrose gradient centrifugation and fractionation, and radioactivity in the fractions was measured. The kinetics and extent of [ 3 H]cholesterol removal by MBCD (Table III, [ 3 H]cholesterol remaining) agreed well with removal of bulk cholesterol as determined by TLC (Fig. 4) Biosynthetic Transport of PLAP Is Similar in MEB-4 and GM-95 Cells-We next examined biosynthetic transport of PLAP from the endoplasmic reticulum to the plasma membrane in stably transfected MEB-4 and GM-95 cells using a pulse-chase surface biotinylation procedure. An autoradiogram is shown in Fig. 6A (MEB-4, panel 1; GM-95, panel 3). The results were quantitated using a PhosphorImager, normalizing for the total amount of labeled PLAP immunoprecipitated at each time point. The value for the highest amount of cell surface PLAP (obtained at 120 min in both cell types) was set

and GM-95 DRM lipid composition
Recovery of DRM lipids (nmol/mg whole cell protein, Ϯ S.D.; n ϭ 3 for MEB-4, n ϭ 4 for GM-95) from MEB-4 or GM-95 cells after Triton X-100 extraction at a detergent:protein ratio of 10:1. Phosphatidylinositol and phosphatidylserine were present but made up less than 1% of the total DRM lipid in each cell type. PC, phosphatidylcholine; PE, phosphatidylethanolamine; SPL, sphingolipids; ND, not detected.   Cholesterol depletion was reported to slow transport of influenza hemagglutinin (which associates with rafts and is apically targeted in epithelial cells) from the trans-Golgi network to the cell surface in a fibroblast and an epithelial cell line (13). By contrast, transport of vesicular stomatitis virus glycoprotein (which does not associate with rafts and is basolaterally targeted in epithelial cells) was not affected. Because raftassociated GPI-anchored proteins are also targeted apically, we examined the effect of cholesterol depletion with MBCD on transport of PLAP in MEB-4 and GM-95 cells. Treatment with 10 mM MBCD for 45 min, to remove about 60% of the total cholesterol (Fig. 4), did not affect transport in either cell line (Fig. 6, A, panels 2 and 4, and B, open circles and squares).
We did not incubate cells with an inhibitor of cholesterol synthesis during the assay, and we cannot exclude the possibility that newly synthesized cholesterol rescued a transport defect in GM-95. However, cholesterol synthesis is almost completely repressed in cells grown in serum-containing medium (65). We found that cholesterol synthesis was induced less than 2-fold from this low repressed level following MBCD removal under the conditions of our delivery assay (Fig. 6C), suggesting that newly synthesized cholesterol did not contribute significantly to cellular cholesterol pools. Similarly, others have shown that cellular cholesterol levels do not increase appreciably in a 2-h incubation following MBCD removal (11). DISCUSSION DRMs from MEB-4 and GM-95 cells were very similar in protein and lipid yield, protein composition, membrane fluidity, and response to cholesterol removal with MBCD. DRM lipid composition was also very similar in the two cell lines, except for the loss of GSL and a corresponding increase in sphingomyelin in GM-95 DRMs. We conclude that GSL headgroup interactions are not required for raft formation.
Cholesterol and Membrane Phase Behavior-Before starting this study, one reason to suspect different behavior of MEB-4 and GM-95 DRMs was the very high T m of many GSL. Our data have shown that T m is an important determinant of whether lipids will associate with rafts (25,27,29), so the high T m GSL might be expected to affect the properties of rafts and DRMs. If so, this should have been especially evident in the DPH fluorescence polarization at 50°C, above the T m of sphingomyelin. However, this parameter was very similar in the two cell types (Table II). This similarity can be explained by the profound effect of cholesterol on membrane phase behavior. Cholesterol can markedly enhance both detergent insolubility (27) and phase separation (29,56) of order-preferring lipids, in a crude sense effectively raising their T m . Thus, the effect of cholesterol on the temperature dependence of DRM acyl chain order was much greater than that of any difference between the T m of pure GSL and sphingomyelin. This phenomenon was previously observed in phospholipid-sphingolipidcholesterol model membranes (29). The phase behavior of these membranes was the same whether the sphingolipid consisted of a mixture of sphingomyelin and cerebrosides or was entirely sphingomyelin.
Sphingolipid Levels and Membrane Phase Behavior-Sphingomyelin expression is up-regulated in GM-95 cells, so that the total sphingolipid levels in GM-95 and MEB-4 cells (35) and DRMs prepared from them (our data) are very similar. This similarity supports the idea that maintaining the correct plasma membrane phase behavior is important to cells and is an important function of sphingolipids and sterols. Studies in yeast also support this idea; the lethality of sphingolipid deficiency can be suppressed by novel phospholipids that contain unusually long, saturated acyl chains (66).
DRM Protein Yield-The major difference between MEB-4 and GM-95 DRMs (aside from the sphingolipid composition) was the 2-fold higher yield of protein in MEB-4 DRMs. We do not know the basis of this difference. Some proteins may have a slightly higher affinity for DRMs that contain GSL, although we did not see significant differences in overall DRM protein composition or enrichment of specific DRM marker proteins in MEB-4 DRMs. Alternatively, it is known that cytoskeletal proteins can associate with DRMs, possibly as contaminants (43).  H]cholesterol were treated with MBCD for the indicated times and then extracted at a Triton X-100:protein ratio of 10:1. Lysates were subjected to sucrose gradient ultracentrifugation. The protein concentration of each lysate was determined, and results were normalized for protein to correct for cell loss during MBCD treatment. Twelve 1-ml fractions were collected, and the radioactivity in 0.1 ml of each fraction was measured in a scintillation counter. The percentage of initial [ 3 H]cholesterol remaining ϭ (total cpm in gradient, normalized for protein)/(total cpm in the no-MBCD gradient) ϫ 100. The percentage of [ 3 H]cholesterol in DRMs ϭ (cpm in fractions 9 -12)/(total cpm in gradient) ϫ 100. Averages and average deviation of the mean are shown (n ϭ 3 for MEB-4; n ϭ 2 for GM-95). MEB-4 and GM-95 cells have markedly different morphologies (33), presumably reflecting differences in cytoskeletal organization. At least part of the difference in DRM protein recovery might simply reflect variability in nonspecific adherence of cytoskeleton to DRMs. Effect of Detergent:Protein Ratio on DRM Protein Yield-We recovered more protein in DRMs as the detergent:protein ratio was lowered (Fig. 1). This may simply mean the low detergent concentrations were insufficient to solubilize bulk nonraft-associated membrane. However, the selectivity of the increase in DRM protein at low detergent concentrations argues against this interpretation. Instead of seeing a more complex DRM protein profile at low detergent concentrations (reflecting increasing amounts of contaminants), we saw better recovery of proteins that were also prominent in high detergent DRMs (Fig. 2). We observed similar behavior of the DRM-associated neuronal protein GAP-43 (28). These results suggest that detergent may partially solubilize rafts, and this effect may be more pronounced with more detergent. In agreement with this idea, detergent can sometimes partially solubilize l o phase lipids in two phase model membranes containing l o and liquiddisordered domains (27). We conclude that detergent insolubility can demonstrate qualitatively that a protein or lipid has a high affinity for rafts. However, detergent insolubility may not be a reliable quantitative measure of the in vivo raft association of proteins or lipids.
Rafts and Intracellular Sorting-The raft model for sorting postulates that organization of proteins and lipids into rafts in the trans-Golgi network facilitates their sorting into a special class of transport vesicles in polarized and nonpolarized cells (1,22). Because PLAP was transported to the cell surface as rapidly in GM-95 as in MEB-4 cells, efficient packaging of this protein into the correct transport vesicles does not seem to require GSL.
In contrast with results of Keller and Simons on influenza hemagglutinin transport (13), we found that cholesterol depletion with MBCD did not slow cell surface transport of PLAP. This discrepancy could be explained by a differential effect of MBCD on raft association of GPI-anchored and transmembrane proteins. Sheets et al. (11) showed that two molecules that are anchored in the outer leaflet of the bilayer (Thy1, a GPI-anchored protein, and the ganglioside GD1b) associated with DRMs even after MBCD treatment. (Similarly, PLAP still associated with DRMs after depletion of cholesterol by inhibition of synthesis (67) and by MBCD treatment as shown in this study.) In contrast, the Src family kinase Lyn (which associates with the inner leaflet) and the high affinity immunoglobulin E receptor, a transmembrane protein that is normally recruited to rafts when clustered, no longer associated with DRMs after MBCD treatment (11). Influenza hemagglutinin, a transmembrane protein that associates with rafts via its transmembrane domain (59) and its three palmitate chains (43), may behave in a similar manner.
The fraction of total labeled PLAP on the cell surface after 2 and 3 h of chase was about the same in GM-95 cells, whereas in MEB-4 cells this value declined by about 35% in the same time (Fig. 6). This suggests that rates of internalization and/or recycling of the protein to the surface may differ in the two cell types, because GPI-anchored proteins can recycle more slowly than other membrane markers (20). Further work will be required to determine the basis of this effect.
Acknowledgments-We thank E. London for help measuring DPH fluorescence polarization, Y. Hirabayashi for useful discussions, and E. London and B. Haltiwanger for critical reading of the manuscript.