Association of Major Histocompatibility Complex II with Cholesterol- and Sphingolipid-rich Membranes Precedes Peptide Loading*

Major histocompatibility complex class II protein (MHC II) molecules present antigenic peptides to CD4-positive T-cells. Efficient T cell stimulation requires association of MHC II with membrane microdomains or-ganized by cholesterol and glycosphingolipids or by tetraspanins. Using detergent extraction at 37 °C combined with a modified flotation assay, we investigated the sequence of events leading to the association of MHC II with cholesterol- and glycosphingolipid-rich membranes (DRMs) that are distinct from tetraspanins. We find two stages of association of MHC II with DRMs. In stage one, complexes of MHC II and invariant chain, a chaperone involved in MHC II transport, enter DRMs in the Golgi stack. In early endosomes, these complexes are almost quantitatively associated with DRMs. Upon transport to late endocytic compartments, MHC II-bound invariant chain is stepwise proteolyzed to the MHC class II-associated invariant chain peptide (CLIP) that remains MHC II-bound and retains a preference for DRMs. At the transition between the two stages, CLIP is exchanged against processed antigens, and the result-ing MHC II-peptide complexes are transported to the cell surface. In the second stage, MHC II shows a lower overall association with DRMs. However, surface MHC II molecules occupied with peptides that induce resistance to denaturation by SDS are enriched in DRMs relative to SDS-sensitive MHC II-peptide complexes. Like-wise,

equated with lipid rafts; however, they may serve as a tool to estimate the affinity of a protein for a cholesterol-and sphingolipid-rich, liquid-ordered membrane environment (26). In addition to Triton X-100, detergents such as CHAPS (27), Lubrol WX (28), and Brij 98 (29) have been used for DRM isolation. Among these, Brij 98 is highly selective for non-raft lipids even at 37°C. Moreover, it produces relatively small DRMs, which may therefore resemble lipid rafts more closely.
There is evidence for a second type of membrane microdomain characterized by the presence of tetraspanins (30). In contrast to lipid rafts, which are critically dependent on lipidlipid interactions, tetraspanin microdomains are thought to form networks governed by protein-protein interactions. These involve not only members of the tetraspanin family but many other proteins. It is interesting that a subpopulation of MHC II has been shown to associate with tetraspanin microdomains (31). The interaction with tetraspanins seems to be significant because the MHC II molecules involved contain only intermediate-affinity peptides of limited diversity. Tetraspanindependent clustering of MHC II is thought to pre-concentrate them for efficient antigen presentation (31), similar to what was suggested for MHC II in lipid rafts (15). Interactions of tetraspanins with MHC II, H2-M/HLA-DM, and HLA-DO (a regulator of H2-M/HLA-DM) are only preserved in mild detergents such as CHAPS or Brij 98 (30,32). After extraction with these detergents, tetraspanins are not separated from cholesterol-and sphingolipid-rich membranes on conventional flotation gradients (27,33). Herein, we describe a novel Brij 98based procedure that results in the separation of MHC IIcontaining DRMs from tetraspanins. We demonstrate that MHC II passes through two consecutive stages of DRM association on its way to the cell surface defined by its interaction with invariant chain or processed antigens. We also show that the association of MHC II with DRMs in the second stage is modulated by peptides bound to MHC II.
Cholesterol Extraction of Post-nuclear Membranes-A post-nuclear membrane pellet was prepared from 3 ϫ 10 7 M12.C3F6 cells as described above and resuspended in H 2 O dd containing 10 mM methyl-␤cyclodextrin. To extract cholesterol from membranes, the sample was incubated for 30 min at 37°C followed by centrifugation at 40,000 rpm in a TLA45 rotor (Beckman Coulter) for 45 min at 4°C. After aspiration of the supernatant, the extracted post-nuclear membrane pellet was then processed for raft isolation as described.
Brefeldin A Treatment-BFA was added at 5 g/ml to a culture of M12.C3F6 cells 4 h before harvest. Subsequent steps were as described above.
Biotinylation and GSH-stripping Procedures-Surface and endosome-specific biotinylation were done with membrane-impermeant, cleavable sulfosuccinimidyl 2-(biotinamido)ethyl-1,3-dithiopropionate. Detailed protocols were described previously (5). In brief, for surface biotinylation, 3 ϫ 10 7 cells were washed three times with ice-cold HBSS, resuspended in 300 l of HBSS on ice, and biotinylated by adding sulfosuccinimidyl 2-(biotinamido)ethyl-1,3-dithiopropionate in three 10-min intervals to a final concentration of 0.5 mg/ml. The reaction was terminated by washing the cells with stop medium (ice-cold CM containing 50 mM glycine but no ␤-mercaptoethanol) followed by incubation in the same medium for 30 min on ice. After two washes with ice-cold HBSS, surface-biotinylated cells were ready for further processing. The endosome-specific biotinylation protocol consists of two parts. First, endosomes and cell surface were biotinylated at 37°C; second, surface-attached biotin was removed by the membrane-impermeant reducing agent GSH on ice. For the first part, 3 ϫ 10 7 cells were washed three times with ice-cold HBSS and resuspended in 300 l of pre-warmed (37°C) HBSS containing 0.5 mg/ml biotinylation reagent. The suspension was incubated for 10 min at 37°C. Endocytosis was terminated by addition of 20 ml of ice-cold stop medium, a brief spin, followed by a 30-min incubation with 20 ml of stop medium on ice and HBSS washes. In the second part, all steps were strictly done on ice. Cells were incubated twice for 30 min in 10 ml of ice-cold GSH buffer (10% fetal calf serum, 50 mM GSH, and 100 mM NaCl, pH 8.4) to remove surface biotin. The GSH buffer was prepared as described previously (5). After washes with chilled 10% fetal calf serum, 1 mM CaCl 2 , and 0.5 mM MgCl 2 in phosphate-buffered saline, the cells were incubated for 10 min with 27 mM iodoacetamide in HBSS on ice, followed by one or two final washes with HBSS. The cells were then directly lysed for immunoprecipitation or processed for DRM isolation. The efficiency of the stripping procedure was checked by applying the stripping protocol to half of 6 ϫ 10 7 surface-biotinylated cells. The other half remained untreated. The cells were then lysed with TBST (50 mM Tris, 100 mM NaCl, 1% Triton X-100, and 0.02% NaN 3 , pH 7.2) and protease inhibitors for 1 h on ice. From the cleared lysates, first invariant chain and then MHC II were immunoprecipitated as detailed below (sequential IP). Biotinylated polypeptides were detected by Western blotting against neutravidin-horseradish peroxidase and quantified by densitometry.
Immunoprecipitation-Before IP, gradient fractions were diluted 1:2 in MB ϩ and lysed by addition of 60 mM ␤-octylglucoside overnight on ice. After a pre-clearing step with 15 l of protein A-Sepharose, the samples were incubated with 5 g/ml anti-A k (40F) or with anti-invariant chain (In-1 and MAR 18.5, each 5 g/ml) for 1 or 2 h on ice. The antibodies were adsorbed to 15 l of protein A-Sepharose by slow rotation for 1 h at 4°C. Washed beads were eluted with reducing SDS-sample buffer at 37°C for 1 h to preserve SDS-stable MHC II (metabolic labeling experiments). For biotinylation and re-precipitation experiments, samples were eluted with non-reducing SDS sample buffer for 5 min at 95°C. In re-precipitation experiments, the eluate was diluted 1:20 in TBST buffer plus protease inhibitors, followed by a second IP with mAbs In-1 and MAR 18.5 (first IP was with mAb 40F). For sequential IPs of biotinylated samples, the first IP was done with mAbs In-1 and MAR as described above. The supernatants of this first IP were pre-cleared and subjected to a second IP with mAb 40F. MHC II retrieved in this way was considered to represent peptide-loaded MHC II (peptide-free MHC II aggregates and is removed in the preclear steps). In some experiments, the supernatant of the second IP was again pre-cleared with protein A-Sepharose and then slowly rotated with 30 l of streptavidin-agarose beads for 1 h at 4°C to retrieve all the remaining biotinylated proteins. After washes, the samples from the first two IPs were eluted with non-reducing SDS-sample buffer. The biotin retrieval samples were eluted with reducing SDS-sample buffer (5 min at 95°C).
Gel Electrophoresis, Blotting Procedures, and Quantification-Standard SDS gel electrophoresis was performed in mini gels containing 11 or 15% polyacrylamide. Samples from metabolic labeling experiments were electrophoresed in 9 -11% polyacrylamide midi gels, incubated with En 3 Hance, and dried for autoradiography. For Western blotting, primary antibodies were used as recommended by the manufacturer or at 1 g/ml. CTB was also used at 1 g/ml and the anti-CTB antiserum at 1:1000 dilution. Secondary reagents were diluted according to the recommendations of the manufacturers. For dot-blotting, 3 l of gradient fractions were spotted on nitrocellulose, air-dried, and subjected to the standard immunoblotting procedure. All blots were developed with ECL reagent and exposed to film. Suitable exposures were scanned and quantified using NIH Image.
Iodination of HEL and Pulse-chase Uptake into M12.C3F6 -Detailed protocols have been published previously (45). In brief, 125 I-labeled HEL was prepared by denaturing 18 nmol of HEL in 6 M guanidinium hydrochloride without reduction and iodinating it with 3 mCi of 125 I according to the chloramine-T procedure. The reaction was stopped by p-hydroxyphenylacetic acid (final concentration, 10 mg/ml) and NaI (final concentration, 10 mM). Refolding was induced by rapid 10-fold dilution followed by desalting on a PD-10 column equilibrated with phosphate-buffered saline. For loading A k , 5 ϫ 10 7 M12.C3F6 cells were pulsed with 6 nmol of 125 I-labeled HEL in 2 ml of CM for 10 min at 37°C, washed three times with ice-cold HBSS and then re-cultured in CM at 37°C for 2h. After several washes with HBSS, the cells were subjected to standard DRM isolation. From individual fractions, IPs were performed using mAb 40F. Electrophoretic separation was on 15% polyacrylamide gels that were subsequently dried and autoradiographed using BioMax MS film.
Enzymatic Reactions-For detection of alkaline phosphatase, 50 l of substrate containing 20 mM p-nitrophenylphosphate, 2 M diethanolamine, and 2.5 mM MgCl 2 , pH 9.8, were added to a 50-l sample in a microtiter plate and incubated for 2-4 h at 37°C. Color development was read at 404 nm in an enzyme-linked immunosorbent assay reader.

MHC II and Invariant Chain Partition to DRMs
Independently of Tetraspanins-Current DRM isolation protocols based on mild detergents do not differentiate between DRMs and tetraspanin microdomains (27,33). To solve this problem, we established a procedure based on Brij 98 extraction at 37°C (29) and linear 0 -40% sucrose flotation gradients containing the detergent throughout. A typical profile for the mouse B-cell line M12.C3F6 is depicted in Fig. 1. Profiles for human B-cell lines (Raji, T1.A k ) were very similar (data not shown). DRMs floated to fractions 8 -14 as indicated by the distribution of the ganglioside G M 1 and alkaline phosphatase, a glycosylphosphatidylinositol-anchored protein enriched in lipid rafts (22,46,47). It is interesting that hardly any DRM marker was detected in the pellet, demonstrating that post-nuclear membranes were quantitatively detergent-extracted under our conditions. Membrane proteins with no preference for microdomains such as B7-2 (16), CD 45 (48), and lamp-1 (32) peaked in fractions 20 -23 with only little tendency of spreading into the gradient. To investigate how tetraspanins fractionate under these conditions, the steady state distribution of the mouse tetraspanin CD81 was analyzed. This molecule behaved similarly to soluble proteins by remaining in the bottom fractions of the gradient. We also assayed the human tetraspanin CD63 in the B-cell line T1.A k . It displayed a distribution similar to that of mouse CD81, suggesting that tetraspanins do not significantly float under our conditions and thus were readily separated from DRMs ( Fig. 1). By contrast, MHC II and its two chaperones, invariant chain and HLA-DM/H2-M, showed different patterns: SDS-stable, peptide-loaded MHC II partially localized to DRM fractions in the mouse B-cell line M12.C3F6 (63 Ϯ 6% in seven experiments) (Fig. 1). We found a considerable variation in the degree of steady state MHC II-DRM association between different cell lines and primary cells expressing the same MHC II molecule (A k ), ranging from a few percent to about 70% (data not shown). This suggests that cell-specific factors influence the association of MHC II with DRMs. For invariant chain, two glycosylation variants were detected in M12.C3F6 cells at steady state, the slower migrating, mature invariant chain, and the faster migrating but more

FIG. 1. Separation of DRMs and tetraspanins by flotation on
Brij 98-containing sucrose gradients. Brij 98-resistant membranes (DRMs) prepared from post-nuclear membranes of 3 ϫ 10 7 human T1.A k cells (CD63 only) or 6 ϫ 10 7 mouse M12.C3F6 cells (other markers) were floated on linear 0 -40% sucrose gradients. The DRM markers G M 1 and alkaline phosphatase were quantified by dot blot and enzyme assay, respectively. The distribution of the non-DRM markers B7-2 and B220 and of the non-tetraspanin microdomain marker lamp-1 was quantified by dot blot (mAb B7-2) or visualized by Western blot (lamp-1, B220). The distribution of peptide-loaded MHC II (A k ) resistant to denaturation by SDS (SDS-stable), invariant chain (Ii), and H2-M/ HLA-DM was determined by Western blotting. The tetraspanin CD81 was also quantified by Western blot. In addition to mouse markers, the distribution of human CD63 in the cell line T1.A k was assessed by dot blot. Note that a fraction of SDS-stable A k and most mature Ii, but not immature Ii (Ii imm. ) co-floated with DRMs. p, pellet; a.u., arbitrary units.
abundant, ER-localized, immature invariant chain. The latter mostly distributed near the bottom of the gradient with peak levels in fractions 19 -20. By contrast, the slower migrating post-Golgi form of invariant chain was strongly enriched in DRM fractions (Fig. 1). A similar enrichment was found in other B-cell lines (data not shown). Our results suggest that invariant chain obtains the potential to associate with DRMs during or after passage through the Golgi. Unexpectedly, under our conditions, almost all H2-M/HLA-DM and a fraction of immature invariant chain were found in the pellet. The same result was obtained for bone marrow-derived immature dendritic cells arguing that this is not a cell type-specific phenomenon. 2 When a similar fractionation was performed with postnuclear membranes after extraction with 1% Triton X-100 at 4°C, some H2-M/HLA-DM remained soluble, consistent with work published by Cheng et al. (48). However, even under this condition, the large majority of H2-M/HLA-DM pelleted (data not shown), suggesting that H2-M/HLA-DM displays a high tendency to aggregate once its lipid environment is disturbed. Because by definition DRM proteins are insensitive to such perturbations and because no markers for cholesterol-and sphingolipid-rich membranes were detected in the pellet, our data suggest that H2-M/HLA-DM does not associate with cholesterol-and sphingolipid-rich membranes.
MHC II and Invariant Chain Localize to Cholesterol-and Sphingolipid-rich Membranes-If flotation of mature invariant chain and MHC II results from their association with cholesterol-and sphingolipid-rich membrane domains, then the removal of cholesterol should interfere with their flotation in our gradients. To test this, cholesterol was extracted with methyl-␤-cyclodextrin from post-nuclear membranes before detergent extraction. Under this condition, flotation of mature invariant chain was completely prevented, and MHC II in DRM fractions was markedly diminished (Fig. 2).
To obtain evidence for the association of MHC II with cholesterol-and sphingolipid-rich membranes independent of detergent-based approaches, we analyzed the co-patching of MHC II or highly expressed membrane proteins such as MHC I with the ganglioside G M 1 at the surface of M12.C3F6 cells. As shown in Figure 3, all clusters of MHC II (labeled in red) co-localized with patched G M 1 (labeled in green). In addition, some patches of G M 1 were observed that did not contain MHC II. By contrast, only very few clusters of MHC I (labeled in red) showed colocalization with patched G M 1, whereas most remained separate. Thus, by this qualitative, detergent-independent method, MHC II co-localized with the lipid raft marker G M 1, whereas MHC I largely remained separate. So far, we have not detected patching of invariant chain at the surface of M12.C3F6 cells.
However, this is most probably a result of the insensitivity of the technique because invariant chain is expressed at low levels on the surface of M12.C3F6 cells (5).
Invariant Chain Enters DRMs in the Golgi-Because invariant chain binds to MHC II, DRM association of mature invariant chain in M12.C3F6 cells suggests that MHC II may already enter DRMs long before the deposition of MHC II peptide complexes at the cell surface. Therefore, we set out to delineate a sequence of events, first by investigating DRM association of invariant chain and MHC II in organelles along the pathway to peptide loading compartments (5). To probe for DRM association of invariant chain in the Golgi, we blocked transport beyond the trans-Golgi cisternae with Brefeldin A, which also induces a Golgi-ER hybrid organelle (49). After 4 h of treatment, the level of invariant chain was strongly increased relative to the control, most likely because it could no longer reach MVBs, where it is quickly degraded (Fig. 4). In the BFA-treated sample, about 50% of invariant chain was found in DRM fractions, indicating that the BFA-induced ER-Golgi hybrid organelle permitted at least partial DRM association of invariant chain. It is noteworthy that proteins beyond the early stages of biosynthesis were not affected by BFA, neither peptide-loaded MHC II nor mature lamp-1 showed any change in their distribution, confirming the specificity of the BFA effect. We also detected a faster migrating form of lamp-1 in the soluble fractions of the BFA-treated sample, most probably representing immature lamp-1 accumulated in the ER-Golgi hybrid organelle. Its distribution to soluble fractions indicates that in a BFA-induced ER-Golgi hybrid organelle, non-DRM proteins do not associate nonspecifically with cholesterol-and sphingolipid-rich membranes. Because no evidence was found for DRM association of invariant chain in the ER (see Figs. 1 and 7), the ability to enter DRMs in a BFA-induced hybrid organelle must have been contributed by Golgi-derived material. This suggests that invariant chain enters DRMs somewhere along the Golgi stack.
Partitioning to DRMs in the Endosomes and at the Cell Surface-Post-Golgi trafficking of MHC II-invariant chain complexes in B-cells involves endosomes and the plasma membrane; they are targeted from the trans-Golgi network to early endosomes, from where the majority of the complexes travel down the endocytic pathway. A fraction of them escapes from early endosomes to the plasma membrane and is quickly re-2 S. Reese and R. Lindner, unpublished data. trieved back to early endosomes (5). We investigated DRM association of MHC II-invariant chain complexes, peptideloaded MHC II, and transferrin receptor at these subcellular locations in M12.C3F6 cells. A GSH-cleavable, membrane-impermeant biotin derivative was used for both surface labeling (on ice) or endosome-specific labeling; the latter was done by first biotinylating cell surface and endosomes at 37°C for 10 min. In the course of this incubation, biotinylated surface proteins as well as the biotinylation reagent were endocytosed and transported to early endosomes. In a second step, the surface label was selectively removed by treatment with GSH on ice. To demonstrate that GSH-stripping was quantitative and thus that endosomal proteins were specifically labeled, we used M12.C3F6 cells and biotinylated either the cell surface or both cell surface and endosomes. Thereafter, half the cells of each group were stripped with GSH on ice, and the other half remained untreated. We confirmed that biotin attached to MHC II-invariant chain complexes and to peptide-loaded MHC II at the cell surface was quantitatively removed by GSH (Fig. 5A,  lane 2). In cells labeled for 10 min at 37°C, a large fraction of MHC II-invariant chain complexes and a very small fraction of peptide-loaded MHC II resisted GSH-stripping and were therefore considered genuinely endosomal (Fig. 5A, lane 4). In a next step, biotinylation was combined with DRM isolation and sequential immunoprecipitation (Fig. 5, B-D). We found very high DRM association of MHC II-invariant chain complexes in endosomes and at the plasma membrane (Fig. 5B). In three different experiments, DRM association of endosomal MHC II-invariant chain complexes varied from 85 to 98%. By contrast, peptide-loaded MHC II was DRM-associated to a lower extent in endosomes and at the cell surface (Fig. 5C). Cell surface MHC II-invariant chain complexes and peptide-loaded MHC II rapidly exchange with their respective endosomal pools (5). We did not find a large difference in DRM association of MHC II-invariant chain complexes and peptide-loaded MHC II between the two locations, although DRM association always seemed to be slightly lower at the plasma membrane. A different result was obtained with transferrin receptor in endosomes and at the cell surface (Fig. 5D); about 97% of transferrin receptor at the plasma membrane distributed to the bottom fractions of the gradient, consistent with the idea that this receptor is not associated with cholesterol-and sphingolipidrich membranes at this location (46). However, in endosomes, about 41% of transferrin receptor was present in DRMs, in line with its recent co-localization with lactosylceramide in recycling endosomes by Sharma et al. (50).
DRM Association of MHC II-bound Degradation Intermediates of Invariant Chain-Transport of MHC II-invariant chain complexes down the endocytic pathway to MVBs results in the proteolytic removal of invariant chain from MHC II. Degradation of invariant chain proceeds in several steps, first to p22 and then to p12, both of which remain bound to MHC II and still retain their N-terminal membrane anchor (51,52). Degradation of this N-terminal segment generates the CLIP peptide, which binds to the peptide-binding groove of MHC II (8) and which is the direct substrate for the MHC II-peptide loading reaction (53). To test whether MHC II-p22 and MHC II-p12 partition to DRMs, MHC II-associated invariant chain degradation products from flotation gradient fractions were Western blotted with an antibody recognizing the cytosolic tail of invariant chain (mAb In-1). Only MHC II-bound p12 was detected, and it distributed completely to DRMs (Fig. 6A). In an alternative approach, M12.C3F6 cells were labeled by endosomespecific biotinylation and subjected to DRM flotation. MHC II and associated proteins were retrieved in a first immunoprecipitation. After elution from the immune matrix, invariant chain and its degradation products were re-immunoprecipitated with mAb In-1. Biotinylated invariant chain-derived polypeptides were then detected by Western blotting with neutravidin. Fig. 6B shows that endosomal, MHC II-associated p22 and p12 quantitatively partition to DRMs. This suggests that C-terminal truncation of invariant chain up to the position of the CLIP peptide does not affect DRM association of the respective MHC II complex.
In a next step, we determined the distribution of MHC II-CLIP complexes on flotation gradients. For this purpose, we used an antibody that recognizes human CLIP bound to MHC II (HLA-DR) in the B-cell line Raji. This cell line shows a gradient distribution of MHC II very similar to that of the mouse B-cell line M12.C3F6 as analyzed by dot-blotting with two different antibodies (data not shown). MHC II-CLIP complexes or total MHC II were immunoprecipitated from lysed pools of DRM and soluble gradient fractions prepared from Raji cells. MHC II retrieved in this way was then visualized by Western blotting with an antibody against the ␣-chain of human MHC II (mAb 1B5). Both total MHC II and MHC II-CLIP complexes showed a similar, high degree of DRM association, although this seems to be somewhat lower compared with the still membrane-anchored invariant chain degradation products analyzed in Fig. 6, A and B.
Dynamics of MHC II Association with DRMs-Peptide loading of newly synthesized MHC II involves the exchange of CLIP peptide against processed antigens (53). To investigate how MHC II distributes in the flotation gradients during peptide loading and to obtain a more complete picture of the time course of the association of MHC II with DRMs, we performed metabolic pulse-chase labeling experiments. For this purpose, M12.C3F6 cells were labeled for 10 min with [ 35 S]methionine and chased for up to 4 h as described previously (5). After DRM flotation, MHC II and associated material was immunoprecipitated from lysed fractions and analyzed (Fig. 7). In the pulse, most MHC II-invariant chain complexes were found in the pellet, some were found in bottom fractions, but none were found in DRMs. This implies that ER-located, immature MHC II-invariant chain complexes partly aggregate in Brij 98, similar to observations in Fig. 1. At 20 -60 min of chase, the now-prevalent mature MHC II-invariant chain complexes largely co-floated with DRMs. Mature MHC II sensitive to dissociation by SDS was readily detected by its ␤-chain. It moved through a stage of up to 80% DRM association (Fig. 7G). Later, DRM association of mature, SDS-unstable MHC II decreased to ϳ60%. This drop was paralleled by the loss of invariant chain from MHC II ( Fig. 7; 60 -120 min). This confirms that high DRM association of MHC II is maintained as long as mature invariant chain is bound to MHC II. At 60 -120 min, the first MHC II-peptide complexes that resisted dissociation by SDS were detected. These complexes are indicative of loading of MHC II with processed antigens. At first, a higher fraction of SDS-stable MHC II partitioned to DRMs compared with SDSunstable MHC II, but this difference seemed to decrease slightly over time. Both populations are known to have different sets of peptides bound, one of which resists dissociation by SDS at room temperature and thus confers SDS-stability to MHC II whereas the other does not and therefore allows the subunits of MHC II to dissociate (54). It was surprising that a second SDS-stable MHC II complex of ϳ100 kDa was detected that remained highly DRM-associated from ϳ20 to 240 min. This complex dissociated into MHC II ␣and ␤-chains upon boiling (Fig. 7F) and did not seem to be prevalent at the cell surface (Fig. 8A). Similar complexes have been described by other groups (55) and may represent MHC II molecules tethered via longer peptides containing at least two SDS stabilityinducing epitopes (56). Regardless of this issue, SDS-stable MHC II-peptide complexes displayed high DRM association from early on, suggesting that the formation of these complexes occurs in DRMs or only very shortly before DRM association.
MHC II-bound Peptides Modulate Association with DRMs-The differences in DRM association between SDS-stable and -unstable complexes of MHC II provided a first hint that MHC II-bound peptides might modulate the association of MHC II with DRMs. Because it was difficult to quantify the low amount of SDS-stable MHC II-peptide complexes in the pulse-chase experiments (Fig. 7), we decided to analyze the distribution of the two forms with another method. We restricted our analysis to peptide-loaded MHC II molecules on the surface of M12.C3F6 cells that were labeled by biotin, floated into gradients, and retrieved from gradient fractions by immunoprecipitation (Fig. 8A). Again, we observed a small but reproducible difference between the distribution of SDS-stable and -unstable MHC II molecules: always a higher fraction of SDS-stable MHC II partitioned to DRM fractions (64 Ϯ 5% versus 51 Ϯ 7%). This suggests that peptides that induce SDS-stability indeed slightly increase the likelihood that MHC II will associate with DRMs. However, peptides bound to either SDSstable or -unstable MHC II are very heterogeneous, and this heterogeneity may obscure the detection of a larger difference in DRM association between both subpopulations. To resolve this issue, we investigated the distribution of a series of 125 Ilabeled HEL processing products (both peptides and full-length HEL) that were bound to the mouse MHC II molecule A k via the immunodominant 48 -61 epitope (45). For this purpose, 125 I-labeled HEL was fed to M12.C3F6 cells for 10 min followed by a chase of 2 h. After DRM flotation and lysis, A k and associated HEL processing products were immunoprecipitated FIG. 6. High preference of MHC II-bound invariant chain degradation products for DRMs. A, steady state DRM association of MHC II-bound p12 in M12.C3F6 cells. P12 still contains the cytosolic In-1 epitope and was visualized in immunoprecipitates of MHC II from gradient fractions by Western blot. B, DRM association of endosomal MHC II-bound invariant chain degradation products p22 and p12. p22 and p12 labeled by endosome-specific biotinylation and isolated by immunoprecipitation of MHC II and re-immunoprecipitation with an antibody against the cytosolic tail of invariant chain were visualized by Western blot with neutravidin. C, steady state DRM association of MHC II-CLIP. DRMs extracted from Raji cells were floated on sucrose gradients. MHC II (left) or CLIP (right) was immunoprecipitated from pooled gradient fractions with the mAbs L243 or CerCLIP.1, respectively. Elution was at 60°C under non-reducing conditions (left). Aliquots of the eluates were also reduced with ␤-mercaptoethanol at 100°C (right). Samples were probed for MHC II by Western blotting with an antibody against MHC II ␣-chain (mAb 1B5). Con, fractions 2 and 3 not containing MHC II; DRM, fractions 5 and 6; sol, fractions 10 and 11 containing Brij 98-soluble proteins. Note that in addition to the MHC II ␣-chain, the immunoprecipitating antibody is detected by Western blotting as revealed by comparison with lanes denoted "con." Also note that a disulfide-linked form of ␣ at an apparent MW of ϳ70 kDa was precipitated (denoted ␣*) by both antibodies (compare non-reducing with reducing elution conditions). Ig, immunoprecipitating antibody; HC, heavy chain of Ig; LC, light chain of Ig; Ii, invariant chain. and analyzed for SDS-stable complexes. As shown in Fig. 8B, DRM association of these complexes was 82 Ϯ 4%, which is significantly above the mean value for DRM association of the total population of SDS-stable A k (64 Ϯ 5%).

DISCUSSION
To estimate the preference with which MHC II and its cofactors invariant chain and HLA-DM/H2-M associate with liquidordered, densely packed membranes, we developed a modified, detergent-based isolation procedure. Complete extraction with the highly selective detergent Brij 98 (29) was combined with fractionation on linear sucrose gradients containing the detergent throughout. This method reliably separated DRM from non-DRM markers (Fig. 1). Unlike the widely used Triton X-100 method (47,57), it also revealed the transient DRM association of the transferrin receptor (Fig. 4D), a non-raft marker at the cell surface (46) that clusters in sphingolipid-rich membranes in early and in recycling endosomes (50). In addition, the glycosylphosphatidylinositol-anchored protein alkaline phosphatase, which was found to be quantitatively associated with Triton X-100 DRMs (47,57), was only partially DRM-associated in our Brij 98-based procedure, reminiscent of what was found for glycosylphosphatidylinositol-linked fluorescent proteins (22). Thus, the characteristics of DRMs isolated by our method are more consistent with the results obtained by biophysical measurements. This is also reflected by the size and shape of the DRMs, which in the case of the Triton X-100 DRMs are sheets and vesicles with diameters in the 100 -1000 nm range (47), whereas our protocol yields exclusively flat membrane disks in the 30 -300 nm range. 3 It is possible that our DRM preparation consists of several membrane subpopulations, as evidenced by the slight flotation difference of peptide-loaded MHC II and mature invariant chain (Fig. 1). Recent evidence for such subpopulations was obtained by electron microscopy on native plasma membrane sheets and by immunoisolation procedures (58,59) Our work also demonstrates that DRMs can be separated from tetraspanins with a Brij 98-based isolation procedure. With previously described fractionation protocols (27,33), it has been not possible to decide whether a given protein floats because of its affinity to cholesterol-and sphingolipid-rich membranes or to tetraspanins. With our procedure, we could demonstrate that a part of peptide-loaded MHC II and a large fraction of mature invariant chain float because of their intrinsic affinity to DRM-forming lipids. This was confirmed by the cholesterol-dependence of flotation (Fig. 2). We also used a detergent-independent microscopical method to demonstrate that surface MHC II specifically co-distributed with G M 1 after patching of individual components (Fig. 3). Our results confirm data by others on localization of surface MHC II to DRMs/ patched G M 1-containing membranes (12,(15)(16)(17). Other groups have found similar properties for surface MHC II after antibody-mediated cross-linking of this molecule (11,13,14). In an attempt to detect any large-scale confinement in the plasma membrane, translational diffusion coefficients for MHC II loaded with fluorescent peptide have been determined by single molecule fluorescence microscopy (60). No deviation from twodimensional Brownian motion was detected for native E k molecules within a resolution limit of 60 nm and a timescale of seconds. Given the current estimates for the size and lifetime of non-cross-linked lipid rafts (5 nm; submillisecond) (21,22), their results do not argue against an association of MHC II with lipid rafts.
Dolan et al. (61) recently reported that surface A k in M12.C3F6 cells is not associated with DRMs based on a 0.05% Triton X-100 extraction procedure. Their finding contradicts our data shown in Figs. 5, 7, and 8. However, we find that with 0.05% Triton X-100, only ϳ50% of total A k are extracted from M12.C3F6 cells (data not shown). Thus, it is conceivable that surface MHC II outside of cholesterol-and sphingolipid-rich membranes is preferentially extracted by 0.05% Triton X-100, possibly explaining why they could not detect MHC II in DRMs after surface labeling. The same group also reported that invariant chain does not partition to DRMs. Probably again because of inefficient extraction, they did not detect mature invariant chain. Instead, they based their conclusions the highly abundant immature ER form, which we also found not to be associated with DRMs ( Figs. 1 and 7). Our data show that a complex of mature invariant chain and MHC II associates with cholesterol-and sphingolipid-rich membranes (Figs. 5A and 7). To our knowledge, this is the first report on the DRM association of this complex. Thus, not only do peptide-loaded MHC II at the cell surface show a preference for DRMs but also MHC II-invariant chain complexes on the transit through intracellular compartments before peptide loading.
By following the pathway of newly synthesized MHC II from the ER to the cell surface, we provided evidence for two consecutive stages of its association with DRMs. In the first stage, MHC II associates with DRMs in the Golgi, distributes quantitatively to DRMs in endosomes, and is retained in DRMs in late endocytic compartments. No DRM association was detected in the ER, in agreement with observations that most proteins that partition to cholesterol-and sphingolipid-rich membranes first attain detergent resistance in the Golgi (19). The first stage of DRM association of MHC II correlated with its binding to mature invariant chain and its degradation products (Figs. 5-7). We do not know yet what causes MHC II to associate with DRMs at this stage. Our preliminary data on the DRM association of invariant chain in MHC II-negative B-cell lines suggest that mature invariant chain itself shows only a low preference for DRMs (data not shown). Another clue comes from the observation that the MHC II-CLIP complex displays a DRM association similar to total immunoprecipitated MHC II (ϳ60%; Fig. 6C). This suggests that the association of MHC II-invariant chain complexes with DRMs is largely caused by MHC II-CLIP. The increase to almost quantitative DRM association observed for MHC II complexed to invariant chain or its degradation products p22 and p12 (Figs. 5 and 6, A and B) may be caused by the higher number of membrane anchors present in their nonameric structures (1,51,52).
What is the function of the association of MHC II-invariant chain complexes with DRMs? These complexes pass several sorting steps: in the trans-Golgi network, where they are di-rected to early endosomes (3)(4)(5), at the plasma membrane, where they are internalized via clathrin-coated pits (62), and in MVBs, where they are sorted to the internal vesicles (63). Although sorting at the plasma membrane involves clathrin and specific adaptors, DRMs may fulfill a specific function in sorting MHC II-invariant chain complexes at the trans-Golgi network or in MVBs. Our preliminary results on formation of SDS-stable MHC II-peptide complexes after cholesterol depletion would be consistent with a sorting function showing a slight delay relative to controls (data not shown). However, cholesterol or sphingolipid depletion may affect many cellular functions; therefore, these experiments are difficult to interpret.
At the transition between the two stages of DRM association, MHC II-bound CLIP peptides are replaced by processed antigens with the help of HLA-DM/H2-M molecules (53). This requires complex formation between MHC II-CLIP and HLA-DM/H2-M (64), suggesting that they should be able to enter the same membrane microdomain. However, we found that H2-M did not associate with DRMs ( Fig. 1 and "Results"), consistent with results by Cheng et al. (48). Using CHAPS or ␤-octylglucoside for solubilization, other groups have found co-flotation of HLA-DM with DRMs and tetraspanins (27) or co-immunoprecipitation with tetraspanins and MHC II (31,32). This apparent contradiction may be resolved by considering that many interactions of tetraspanins with other proteins are sensitive to detergents such as TX-100 at 4°C (30) and possibly Brij 98 upon warming. This would leave HLA-DM/H2-M in tetraspanins microdomains and MHC II-CLIP in DRMs distinct from tetraspanins. How could they meet? Many tetraspanins are palmitoylated, and this modification might help to target them and their associated proteins to cholesterol-and sphingolipidrich membranes/lipid rafts. This has recently been shown for the tetraspanin CD81 that is required for stable raft association of the CD81-associated CD19-CD21 complex upon crosslinking to the B cell receptor (65). The small size found for nanoclusters/lipid rafts of glycosylphosphatidylinositol-linked proteins restricts the number of proteins per lipid raft (Ͻ4) (22), thus exposing a large surface fraction of every cluster protein to the exterior. For these reasons, nanoclusters containing MHC II-CLIP complexes might be imagined to transiently team up with HLA-DM/H2-M-containing tetraspanin microdomains for peptide loading. Such an event may take place in the course of the backward fusion of internal vesicles with the limiting membrane of MVBs, a process that has been suggested to induce HLA-DM/H2-M-catalyzed peptide loading of MHC II in maturing dendritic cells (6).
In the second stage, MHC II loaded with processed antigen is transported to the cell surface. Its overall DRM association is lower than at the first stage (except for MHC II-CLIP complexes). We found differences in DRM association of SDS-stable and -unstable MHC II-peptide complexes and for MHC IIpeptide complexes loaded with a particular immunodominant peptide (Fig. 8). These data together with data from the pulsechase experiment (Fig. 7) suggest that MHC II-bound peptides modulate the affinity of MHC II to detergent-resistant lipids. No evidence for a distinct peptide content of DRM-associated versus soluble MHC II was found by Kropshofer et al. using 1% Triton X-100 for extraction (31). However, only a very small fraction of MHC II partitioned to DRMs under their conditions (Ͻ5%). Given this low percentage, it is difficult to rule out that a fraction of MHC II with intrinsic affinity to DRMs might have been solubilized by the stronger, less selective detergent precluding the detection of differences in peptide content. How might an MHC II-bound peptide influence the lipid preference of an MHC II molecule? The structure of MHC II excludes any direct effect of the bound peptide onto the transmembrane domain that is most likely to determine DRM association. In addition, no stoichiometrically bound proteins are known that could effect DRM association of MHC II. Thus, the most likely explanation is a peptide-induced conformational change in MHC II causing an altered affinity for DRMs. Although x-ray crystallography did not reveal significant conformational differences between different MHC II-peptide complexes (8), evidence for two MHC II conformations was provided by studies on a MHC II epitope that strictly depended on H2-M/HLA-DMmediated peptide selection (66).
DRM association of MHC II loaded with processed antigens may fulfill two functions: at the plasma membrane, DRM association of peptide-loaded MHC II has been shown to be important for presentation at low concentrations of antigen (15). Our results may provide a link between this effect and the intracellular selection of optimal MHC II-peptide complexes. In line with this, the naturally processed, immunodominant HEL 48 -61 epitope induces a high degree of DRM association, significantly above the mean association of SDS-stable and -unstable MHC II (Fig. 8). Further studies with synthetic peptides should help to decide whether this effect is caused by the abundance of certain peptides provided by the cellular processing machinery or by an intrinsic property of peptides possibly related to their immunodominance. We also found a small but reproducible increase in DRM association of SDS-stable MHC II-peptide complexes over SDS-unstable ones. It is interesting that peptides that confer SDS-stability to MHC II have longer half-lives than the ones that are incapable of doing so (67). To our knowledge, no molecular mechanism is known to cause this difference in the half-lives. A high degree of association with cholesterol-and sphingolipid-rich membranes might affect the trafficking of MHC II-peptide complexes by retaining them in recycling endocytic membranes rather than permitting further transport to degradative late endocytic compartments.