Proteomic and Biochemical Analyses of Human B Cell-derived Exosomes

Exosomes are 60–100-nm membrane vesicles that are secreted into the extracellular milieu as a consequence of multivesicular body fusion with the plasma membrane. Here we determined the protein and lipid compositions of highly purified human B cell-derived exosomes. Mass spectrometric analysis indicated the abundant presence of major histocompatibility complex (MHC) class I and class II, heat shock cognate 70, heat shock protein 90, integrin α4, CD45, moesin, tubulin (α and β), actin, Giα2, and a multitude of other proteins. An α4-integrin may direct B cell-derived exosomes to follicular dendritic cells, which were described previously as potential target cells. Clathrin, heat shock cognate 70, and heat shock protein 90 may be involved in protein sorting at multivesicular bodies. Exosomes were also enriched in cholesterol, sphingomyelin, and ganglioside GM3, lipids that are typically enriched in detergent-resistant membranes. Most exosome-associated proteins, including MHC class II and tetraspanins, were insoluble in 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonic acid (CHAPS)-containing buffers. Multivesicular body-linked MHC class II was also resistant to CHAPS whereas plasma membrane-associated MHC class II was solubilized readily. Together, these data suggest that recruitment of membrane proteins from the limiting membranes into the internal vesicles of multivesicular bodies may involve their incorporation into tetraspanin-containing detergent-resistant membrane domains.

Maturing endosomes accumulate vesicles in their lumen, resulting in their transformation into multivesicular bodies (MVB) 1 (1). These vesicles are formed by inward budding of the endosomal limiting membrane and contain a selected cargo. Proteins that are sorted to the internal vesicles of MVB potentially may have three distinct fates. The first possibility is exemplified by ligand-activated epidermal growth factor receptor, which is ultimately transferred to lysosomes for degradation (2). A second possibility is that proteins may be stored temporarily in MVB, as observed for MHC class II in immature dendritic cells (3). MHC class II-carrying MVB in dendritic cells have also been termed MHC class II compartments (MIIC), in accordance with similar structures in B cells (4). MIIC play a crucial role in peptide loading of MHC class II. In pathogen-stimulated dendritic cells, the internal vesicles of MVB fuse back with their limiting membrane, thereby allowing subsequent transfer of peptide-loaded MHC class II to the plasma membrane (3). The third potential fate of vesicles within MVB is their release into the extracellular environment as a consequence of fusion of the MVB-limiting membrane with the plasma membrane. These secreted MVB-derived vesicles have been called exosomes, which, depending on their source, may serve a multitude of functions (5)(6)(7).
Exosomes are released by a great number of cell types, including reticulocytes (5), cytotoxic T cells (8), B lymphocytes (9,10), dendritic cells (11)(12)(13), mast cells (14), platelets (15), and intestinal epithelial cells (16). The biological functions of exosomes are generally unclear. Increasing evidence, however, suggests that exosomes from hematopoietic cells may serve as intercellular communication vehicles that assist immune responses (6,7). For example, B cell-derived exosomes that carry peptide-loaded MHC class II were demonstrated to stimulate CD4 ϩ T cells (17) and to specifically bind follicular dendritic cells (18) in vitro. Furthermore, exosomes derived from cultured dendritic cells that were loaded in vitro with tumorderived peptides on MHC class I stimulated cytotoxic T lymphocytes both in vitro and in vivo (11).
Functions of exosomes should be reflected by their protein composition. Given that exosomes are formed as the internal vesicles of MVB, exosomes can be expected to also contain factors required for MVB formation and protein sorting therein. Immunoelectron microscopic studies, Western blot analyses, and peptide mass mapping of exosomes derived from dendritic cells (12,13), B lymphocytes (9,10), intestinal epithelial cells (16), and other cell types revealed the presence of common, as well as cell type-specific, proteins. For example, MHC class II is especially enriched in exosomes derived from B lymphocytes, dendritic cells, mast cells, and intestinal epithelial cells. Ubiquitous proteins in exosomes include cytoplasmic proteins, such as tubulin, actin, and actin-binding proteins, the heat shock proteins hsc70 (also named hsc71 or hsp73) and hsp90, and trimeric G proteins, as well as membrane proteins, such as members of the tetraspanin family (CD9, CD63, CD81, CD82). Sorting of a number of membrane proteins into the MVB pathway involves ubiquitination of their cytoplasmic domain (19) and binding of these acquired ubiquitin moieties to Tsg101 (20). Indeed, Tsg101 (12), as well as c-Cbl (21), a ubiquitin ligase required for ubiquitination of activated epidermal growth factor receptor (2), have been detected in isolated exosomes. Alternatively, membrane proteins may rely on ubiquitinated adaptors for their sorting in MVB (20,22). Importantly, sorting of at least some proteins into the MVB pathway occurs independently of the ubiquitin system (23). The molecular factors and mechanism(s) behind such alternative sorting processes in MVB are unknown, and the analysis of exosomes may help their discovery.
In an approach to understand more about the formation and function of exosomes we developed a protocol that yielded highly purified exosomes from human B cells and studied their molecular content and biochemical properties. Based on exosome characteristics, we propose a model in which the incorporation of proteins into tetraspanin networks and detergentresistant membranes (DRM) at the limiting membrane of MVB may be conditional for their sorting into the internal vesicles of MVB.
Cell Culture and Exosome Isolation-RN cells (HLA-DR15 ϩ ) were cultured as described (9). We observed that fetal calf serum contains exosomes (data not shown). To exclude bovine exosomes, cells were cultured in medium supplemented with fetal calf serum that had been ultracentrifuged for 60 min at 141,000 ϫ g max . RN-derived exosomes were isolated routinely from 800 ml of culture medium containing ϳ10 8 RN cells. As a first isolation step, exosomes were collected from the medium by differential centrifugation, as described (9). In short, cells were removed by centrifugation for 10 min at 200 ϫ g. Supernatants were collected and centrifuged sequentially twice for 10 min at 500 ϫ g max , once for 15 min at 2,000 ϫ g max , once for 30 min at 10,000 ϫ g max , and once for 60 min at 70,000 ϫ g max using a SW27 rotor (Beckman Instruments, Inc., Fullerton, CA). Exosomes were pelleted at the final centrifugation step and were resuspended in PBS and re-pelleted at 70,000 ϫ g max . The final pellet routinely contained ϳ100 g of protein.
Re-pelleted exosomes were resuspended in 5 ml of 2.6 M sucrose, 20 mM Tris-HCl, pH 7.2, and floated into an overlaid linear sucrose gradient (2.0 -0.25 M sucrose, 20 mM Tris-HCl, pH 7.2) in a SW41 tube for 16 h at 270,000 ϫ g max . Gradient fractions of 1 ml were collected from the bottom of the tube and analyzed for the presence of MHC class II and tetraspanins by Western blotting. When indicated, gradient fractions were diluted with 3 ml of PBS each and centrifuged for 60 min at 350,000 ϫ g max , and the pellets were analyzed by SDS-PAGE and Coomassie Blue staining. As a final purification step, 750-l samples of pooled exosome-containing gradient fractions were added to 200 l of Dynabeads M-450 (ϳ8 ϫ 10 7 beads) coated with monoclonal mouse anti-human MHC class II (Dynal Biotech, Oslo, Norway). As a negative control, Dynabeads M-450 coated with goat anti-mouse IgG (Dynal) were used. The Dynabeads that were added to the exosomes suspensions were first extensively washed with and resuspended in PBS supplemented with 3 mg/ml bovine serum albumin. For adsorption, samples were rotated end-over-end for 16 h at 4°C. The beads were collected and washed once with PBS with the aid of a magnet (Dynal). Non-adsorbed membranes were diluted with PBS and collected by centrifugation for 30 min at 200,000 ϫ g max in a SW50 tube.
Mass Spectrometric Protein Analyses-Proteins from Dynabead-associated exosomes were segregated by SDS-PAGE, stained with Coomassie Blue, excised from the gels, and analyzed by mass spectrometry by Protana (Denmark). In-gel tryptic digestion of proteins was performed as described (25). Approximately 2% of the tryptic digest was analyzed on a Bruker Reflex MALDI-TOF mass spectrometer (Bruker, Bremen, Germany), and the obtained peptide maps were queried against a non-redundant sequence data base. Search criteria was as follows: mass accuracy, 50 ppm; tryptic peptides, allowed missed cleavage sites, 1. Samples not unambiguously identified by peptide mass fingerprints were purified and concentrated using home built Poros R2 (Applied Biosystems) microcolumns before sequence analysis. The tryptic peptides were sequenced on a QSTAR quadrupole-TOF mass spectrometer (Sciex) equipped with a nanoelectrospray source (Protana-Engineering). Prior to analysis, the mass spectrometer was calibrated to a mass accuracy of 20 ppm and a resolution of 9500. The data were processed with PPSS2 (Protana's Proteomics Software Suite), and the peptide sequence tags obtained were queried against a non-redundant sequence data base (26). Search criteria were as follows: MS mass accuracy, 1.1 Da; MS/MS accuracy, 0.1 Da; tryptic peptides, allowed missed cleavage sites, 1. For verification of a retrieved peptide sequences theoretical patterns were compared with the obtained collisioninduced dissociation mass spectra.
Lipid Analysis-PBS-washed RN cells, Dynabead-associated exosomes, and ultracentrifuged non-adsorbed membranes were suspended in a total volume of 3 ml of chloroform/methanol (1:1) (v/v), and lipids were extracted overnight at 40°C. The suspensions were then centrifuged for 10 min at 2000 ϫ g, and the clear supernatants were dried in a stream of nitrogen. The residue was dissolved by first adding 60 l of chloroform followed by 0.96 ml of methanol and 0.94 ml of water. Each resulting solution was freed of salts and sucrose by reverse phase chromatography. For this procedure small pieces of silanized glass fiber wadding were introduced into glass Pasteur pipettes, and a suspension (1.5 ml; 1:3 (v/v) in methanol) of LiChroprep RP18 (40 -63 m; Merck) was added. These columns were subsequently washed three times with 1 ml of chloroform/methanol (1:1) (v/v), 3 ϫ 1 ml of methanol, and 3 ϫ 1 ml of water. Then the sample solution was applied, and the column was washed with 3 ϫ 1 ml of water. Bound lipids were eluted with 3 ϫ 1 ml of methanol and 3 ϫ 1 ml of chloroform/methanol (1:1) (v/v). The eluate was dried under a stream of nitrogen, dissolved in 0.10 ml of chloroform/methanol (1:1) (v/v), and analyzed by thin-layer chromatography (TLC). TLC was performed in a horizontal development chamber (Camag, Muttenz, Switzerland) using 10 ϫ 10-cm silica gel-coated thin-layer plates (Merck) and a mixture of chloroform/methanol/water (65:25:4, by volume) as the mobile phase. Lipids were visualized by fine spraying the developed plates with copper sulfate (10%) (w/v) in phosphoric acid (8%) (w/v) followed by charring for 8 min at 180°C. For quantitation the charred plates were scanned, and the data were analyzed by phosphorimaging using the TINA software, version 2.0 (Raytest) and compared with known amounts of reference lipids (cholesterol, dioleoyl phosphatidyl ethanolamine, bovine heart cardiolipin, dioleoyl phosphatidyl choline, and sphingomyelin containing stearic or nervonic acid). For the distinction between cholesterol and its esters thin-layer plates were developed in n-hexane/diethyl ether/acetic acid (60:40:1, by volume). For an unambiguous identification extracted lipids from B lymphocytes, which were grown in quantity in the presence of [1-14 C]sodium acetate (Amersham Biosciences), were separated into uncharged and acidic lipids prior to preparative TLC. Lipids were traced by their radioactivity, scraped, and extracted with chloroform/methanol (1:1) (v/v). The lipid extracts were then subjected to mass spectrometry. The separation according to charge was by DEAE-Sephadex A 25 chromatography in chloroform/methanol/water (3:7:1, by volume). Acidic lipids were eluted with chloroform/methanol/1 M ammonium acetate (3:7:1, by volume) and freed of salts, short chain fatty acids, and other less hydrophobic material by reverse phase chromatography on LiChroprep RP18 as described above.
For electrospray-MS, mass spectra were recorded in the negative or positive ion mode for acidic or uncharged lipids, respectively, on a Q-TOF 2 mass spectrometer (Micromass, Manchester, United Kingdom) equipped with a nanospray source. Lipids were dissolved in chloroform/methanol (1:1) (v/v). Solutions were injected into the mass spectrometer by glass capillaries (long type; Protana, Odense, Denmark) using a capillary voltage of 1000 V and a cone voltage of 50 V at 70°C. Instrument calibration was done with a mixture of sodium iodide and cesium iodide in 50% aqueous acetonitrile with 0.1% formic acid. For MS/MS experiments argon was used as collision gas, and fragmentation was observed at energy values from 20 -50 eV.
For MALDI-TOF-MS, measurements were done on a TOFSpec E (Micromass, Manchester, United Kingdom) in positive or negative ion mode with an accelerating voltage of 20 kV. For lipid samples matrix solutions of 2,5-dihydroxybenzoic acid were used in a concentration of 10 mg/ml in methanol. Spectra were calibrated externally by using suitable reference substances.
Density Gradient Electrophoresis and Cell Surface Biotinylation-RN cells were collected and washed with PBS by centrifugation at 4°C for 10 min at 200 ϫ g. The cell surface was biotinylated for 15 min at 0°C using 5 mg/ml sulfo-NHS-biotin (Pierce). Excess sulfo-NHSbiotin was quenched with 10 mM NH 4 Cl in PBS for 20 min at 0°C after which the cells were collected and washed twice with homogenization buffer (10 mM triethanolamine, 10 mM acetic acid, 1 mM EDTA, 0.25 M sucrose, pH 7.4) by centrifugation. Cells were homogenized using an EMBL-cell cracker (ball 8.011 applying 10 strokes), and nuclei were removed by centrifugation at 900 ϫ g max for 3 min. The post-nuclear supernatant was treated with trypsin (25 g/mg protein) for 15 min at 37°C after which trypsin inhibitor (100 g/mg protein; Sigma) and protease inhibitor mix (Roche Molecular Biochemicals) were added. Density gradient electrophoresis was performed as described (27) for 30 min at a constant current of 10 mA. Fractions of 500 l were collected from the anodic side and analyzed by Western blotting for MHC class II ␤-chain. Plasma membrane proteins were detected by probing Western blots with streptavidin-peroxidase (Sigma) and enhanced chemiluminescence. ␤-Hexosaminidase was determined as described (28), and total protein was assayed using the Bio-Rad protein assay (Bio-Rad).
To compare plasma membrane-associated MHC class II with exosomes-associated MHC class II, intact cells and isolated floated exosomes were biotinylated at 0°C as described above. After detergent extraction and ultracentrifugation, pellets were suspended in 0.5 ml of PBS containing 1% Triton X-100, 10 mg/ml bovine serum albumin, 2 mM EDTA, 0.2 mM phenylmethylsulfonyl fluoride, and 30 l of Neutravidin-Sepharose beads (Pierce) for 16 h. The beads were washed with PBS and analyzed for bound MHC class II by Western blotting.
Solubility Assay and Floatation-Gradient fractions (3 ml in total) containing floated exosomes were diluted with 26 ml of PBS. Aliquots of 3.6 ml were mixed with 0.4 ml of PBS containing 10% CHAPS, 10% Triton X-100, 10 mM EDTA, 10 mM MgCl 2 , and/or 100 mM methyl-␤cyclodextrin (Sigma) as indicated. The samples were then centrifuged in SW60 tubes for 1 h at 350,000 ϫ g max . Pellets were analyzed by SDS-PAGE followed by Coomassie Blue staining or Western blotting.
To demonstrate association of low density lipid material with MHC class II after detergent solubilization we performed floatation assays. Floated exosomes from sucrose density gradients were pelleted as above and suspended in 500 l of 2.5 M sucrose, 0.5 mM CaCl 2 , 0.5 mM MgCl 2 , 1 mM phenylmethylsulfonyl fluoride, 20 mM Hepes/NaOH, pH 7.2, in the presence or absence of 1% CHAPS by 10 passages through a 23gauge needle mounted on a syringe. The suspension was overlaid with a 2-0.4 M sucrose gradient in 20 mM Hepes/NaOH, pH 7.2, and centrifuged for 4 h at 200,000 ϫ g max . Fractions were collected from the bottom of the tube and analyzed by SDS-PAGE followed by Coomassie Blue staining or Western blotting.
Immunoelectron Microscopy-Perfringolysin O modified with subtilisin Carlsberg and subsequent biotinylation was kindly obtained from Dr. Ohno-Iwashita (29). Cells were prepared for cholesterol labeling as detailed elsewhere (30). Perfringolysin O-labeled sections were fixed in 1% glutaraldehyde and labeled with rabbit polyclonal anti-DR (24) or anti CD63 followed by protein A coupled to 15 nm of gold. After another fixation with 1% glutaraldehyde, sections were incubated with antibiotin antibodies (4 g/ml) followed by incubation with protein A coupled to 10 nm of gold. Sections were then examined and photographed at 80 kV with a JEOL 1200 EX electron microscope (Tokyo, Japan).

RESULTS
Purification of Secreted Human Exosomes-Exosomes secreted by the human B cell line RN were purified in three sequential steps. In the first step we performed differential centrifugation to collect membranes from the culture medium that sedimented between 300,000 and 4,200,000 g max ϫ min (10,12). In the second purification step pelleted exosomes were washed and floated into sucrose density gradients to remove non-membranous serum protein (complexes). Exosomes, as identified by the presence of MHC class II, floated up to a density of ϳ1.15 g/ml (Fig. 1B), consistent with previous observations (9,10). The gradient fractions were diluted with PBS and ultracentrifuged, and the pellets were analyzed by SDS-PAGE and Coomassie Blue staining (Fig. 1A). The sample buffer used for SDS-PAGE was supplemented with urea (see "Materials and Methods") as we found that this greatly enhanced the separation of exosome-associated proteins by SDS-PAGE. A number of proteins, possibly originating from fetal calf serum in the culture medium, remained in the bottom fractions (indicated by the arrows) whereas the majority of proteins co-distributed with MHC class II in the gradient. In addition to MHC class II, these proteins include CD86 and the tetraspanins CD37, CD53, CD63, CD81, and CD82 (data not shown) as identified previously (10) by Western blotting. As a third purification step, gradient fractions containing MHC class II (fraction 6 -8) were pooled, and exosomes were immunoadsorbed onto anti-MHC class II-coupled magnetic beads. Membranes that did not associate with the beads were collected from the bead supernatant by ultracentrifugation. Beadassociated and non-associated proteins were analyzed by SDS-PAGE and Coomassie Blue staining (Fig. 2). The anti-MHC class II antibody-conjugated beads recovered nearly all proteins whereas beads coated with negative control antibodies did not collect any of these proteins. This indicates that all detected proteins were linked physically to MHC class II-carrying exosomes.
Protein Composition of Exosomes-To analyze the identity of FIG. 1. Isolation of exosomes on sucrose density gradients. Exosomes were collected from RN culture medium by differential centrifugation. Membranes that pelleted at 420,000 g max ϫ min were floated up into a sucrose density gradient. Gradient fractions (from bottom to top, indicated 1-11) were diluted with PBS, and membranes were collected by ultracentrifugation and analyzed by 12.5% SDS-PAGE and Coomassie Blue staining (A) or Western blotting for MHC class II-␤ (B). Exosomes peaked in fractions 6 -8. Arrows in A at the left indicate example proteins that did not co-migrate with exosomes; numbers at the right indicate molecular weight markers. exosome-associated proteins, discernable Coomassie Bluestained bands in Fig. 2, lane 1 were excised from the gel and analyzed by mass spectrometry. Identified proteins are indicated in Fig. 2 and listed in Table I. Only proteins with a minimum of two matching peptides are shown. HLA-encoded proteins, including MHC class I heavy chain and several MHC class II subtypes, are dominantly present in exosomes. Other membrane proteins that were identified include Na ϩ /K ϩ -ATPase, the receptor tyrosine phosphatase CD45, integrin ␣4, and the receptor-associated inhibitory signaling molecule G i ␣ 2 , which is linked to the cytoplasmic face of membranes by a palmitoyl anchor. Other identified proteins can be grouped in heat shock proteins (hsp90␣ and hsc70), cytoskeletal proteins (␣ and ␤ tubulin and actin), a member of the ERM (ezrinradixin-moesin) family of cytoskeleton-associated proteins (moesin) and a set of enzymes involved in glycolysis (glyceraldehyde-3-phosphate dehydrogenase, pyruvate kinase, ␣-enolase, and fructose-bisphosphate aldolase A). Clathrin heavy chain-1 and elongation factor 1A were detected, as well. Murine immunoglobulin heavy chain originated from the anti-MHC class antibody that was conjugated to the Dynal beads. Tetraspanins were not detected by mass spectrometry, possibly because of low abundance and/or their poor resolution characteristics by SDS-PAGE (10).
Lipid Composition of B Lymphocytes and Exosomes-Lipids extracted from [1-14 C]sodium acetate-labeled lymphocytes were separated according to charge and relative mobility by TLC, stained, and quantified (see Fig. 3, lane 2 and Table II). The nature of these lipids was identified in parallel experiments in which the position of 14 C-labeled lipids was determined on film rather than by staining. These were then extracted from the TLC plates and analyzed by electrospraytime-of-flight-mass spectrometry and MALDI-TOF-MS (Tables   III and IV). Among the acidic lipids were ganglioside GM3 with palmitoyl or nervonoyl residues. The majority of acidic lipids were, however, composed of phosphatidylinositol (PI), phosphatidylserine (PS), phosphatidic acid (PA), bismonoacyl glycerophosphate (BMP), and cardiolipin (CL) with varying fatty acyl moieties. Their fatty acid composition was deduced from tandem mass spectra collected in the negative ion mode. The amount of BMP exceeded that of CL (see Fig. 3A). The uncharged lipids were comprised of sphingomyelin (SM), cholesterol (Chol), phosphatidyl choline, phosphatidylethanolamine (PE), and ether lipids with an ethanolamine phosphoryl head group. Like GM3, sphingomyelin contained either palmitoyl or nervonoyl residues, resulting in two distinct bands by TLC. SM containing C16:0 was not separated from ganglioside GM3 by TLC (lower band of the double band in Fig. 3A). Dynabeadbound exosomes and membranes from exosome-containing sucrose gradient fractions that did not bind to anti-MHC class II-coated Dynabeads were analyzed for their lipid content and compared with total cell membranes (see Fig. 3 and Table II). Exosomes were enriched in cholesterol (42 versus 20 mol % in total cell membranes) and in sphingomyelin and ganglioside GM3 (23 versus 13 mol % in total cell membranes) on the expense of the presence of phosphatidyl ethanolamine and its respective ether lipids, as well as phosphatidyl choline, phosphatidyl inositol, phosphatidyl serine, and phosphatidic acid. As expected, cardiolipin, a lipid that is predominantly found in mitochondria, was absent from exosomes. BMP, also referred to as lysobisphosphatidic acid (LBPA), could also not be detected in exosomes.
Electron Microscopic Detection of Cholesterol on Exosomes-To determine the morphological distribution of cholesterol, cryosections of RN cells were labeled for cholesterol with 10 nm of colloidal gold using Perfringolysin O and examined by electron microscopy (30). The sections were double-labeled for either MHC class II (15 nm of gold) or CD63 (15 nm of gold). MHC class II was detected on the plasma membrane, in MIIC, and on exosomes within MIIC-plasma membrane fusion profiles. Consistent with previous observations (10), CD63 was found predominantly on the internal membranes of MIIC and exosomes rather than on the plasma membrane. Fig. 4 shows fusion profiles of MIIC with the plasma membrane. Consistent with the lipid analyses, cholesterol was predominantly present on secreted exosomes and much less abundant on the MIIClimiting membrane and plasma membrane. We conclude, based on the morphological and biochemical characterization of exosomes, that exosomes are relatively enriched in cholesterol.
Exosomes Are Detergent-resistant-The enrichment in sphingomyelin, GM3, and cholesterol is a characteristic of so-called DRM or raft domains (31). Such domains are unusually resistant to solubilization by non-ionic detergents (32). To investigate whether exosomes display DRM-like properties we determined their solubility in the presence of 1% Triton X-100 or 1% CHAPS (Fig. 5). Sucrose gradient fractions containing exosomes (as in Fig. 1) were pooled and washed with PBS. Aliquots were incubated as indicated for 30 min at 0 or 37°C in the presence or absence of 1% CHAPS, 1% Triton X-100, 1 mM EDTA, 1 mM MgCl 2 , and/or 10 mM methyl-␤-cyclodextrin and then centrifuged. Pellets were analyzed for total protein content by SDS-PAGE and Coomassie Blue staining (Fig. 5A) and for the presence of MHC class II by Western blotting (Fig. 5B). Many exosomal proteins, but not all, were resistant to solubilization by CHAPS, independently of divalent cations or temperature. The solubility of some proteins slightly increased when cholesterol was chelated with methyl-␤-cyclodextrin at 37°C, suggesting that cholesterol is important for exosomal DRM. The solubility of exosomes was higher in Triton X-100  Table I. mIgHC is immunoglobulin-derived from the mouse anti-MHC class II antibody from the Dynal beads. compared with CHAPS but remained incomplete. The same observations were made for exosome-associated MHC class II. MHC class II was solubilized to a significant degree only in the concomitant presence of CHAPS and methyl-␤-cyclodextrin at 37°C or in the presence of Triton X-100. DRM are characterized by the stable association of both proteins and lipids and, consequently, have a relative low buoyant density in sucrose density gradients. To test whether DRM are associated with exosomes, we performed floatation experiments for CHAPStreated exosomes (Fig. 6). Pelleted exosomes were resuspended in 2.5 M sucrose containing 1% CHAPS and overlaid with a sucrose density gradient. After ultracentrifugation, gradient fractions were analyzed for the presence of MHC class II and the tetraspanins CD81 and CD63. About one-third of each of these markers floated up into the gradient, indicating their association with DRM.
The Solubility of MHC Class II Depends on Its Location-Exosomes are formed as the internal vesicles of MVB/MIIC. Because the majority of MHC class II in MIIC of RN cells localizes to these internal vesicles (9), we hypothesized that MHC class II in MIIC, like exosomes, should be resistant to solubilization in CHAPS. To test this idea, we isolated MIIC from surface-biotinylated cells by density gradient electrophoresis (Fig. 7), a technique that segregates MIIC from other cellular membranes (33). Each fraction was tested for the presence of total protein, ␤-hexosaminidase (a marker for MVB and  Fig. 2 as identified by mass spectrometry Coomassie Blue-stained bands in Fig. 2; lane 1 were excised from the gel, trypsinized, and analysed by mass spectrometry for peptide mass fingerprints and, when indicated, peptide sequences. a When mass fingerprints did not unambiguously match a specific protein, the peptides were sequenced by mass spectrometry.

FIG. 3. Analysis of lipids from RN cells and exosomes.
Lipids from total cellular membranes and Dynal bead-associated exosomes were extracted, subjected to thin-layer chromatography, and stained. Lane 1, reference lipids, their nature is indicated on the left side of the figure. Chol, cholesterol; DOPE, dioleoyl phosphatidyl ethanolamine; BHCL, cardiolipin from bovine heart; DOPC, dioleoyl phosphatidyl choline; SM, sphingomyelin from bovine brain containing stearic or nervonic acid, resulting in a double band by TLC. Lane 2, lipids from total RN membranes. Their nature is indicated at the right side of the figure and was determined in a parallel experiment in which metabolically 14 C-labeled lipids were extracted from cells, separated by TLC, and analyzed by mass spectrometry. Spot X is a non-identified compound, spot Y is not identified but has the mobility of monohexosylceramides, and spot Z at the front could not be associated with any known lipid and is most likely related to compound(s) from plastics or other paraphernalia that were extracted by the solvents. Lane 3, lipids from Dynal bead-associated exosomes (isolated as in Fig. 2, lane 1). Lane 4, lipids from non-bound material (as in Fig. 2, lane 3). The quantified relative amounts of lipids are depicted in Table II.   TABLE II Relative lipid compositions as determined from Fig. 3 Lipids from RN cells (Fig. 3, lane 2), immunoadsorbed exosomes (Fig.  3, lane 3), and membranes that did not bind to the immunobeads (Fig.  3, lane 4) were stained and quantified by comparison with known amounts of reference lipids. The data are representative of two independent experiments. Values shown are Ϯ5%. The numbers represent relative amounts of lipid with respect to the total of identified lipids in the same membrane preparation. Because of the low amount of material, some lipids, especially those of the non-bound material derived from unlabeled cells, could not be determined (ND). Cholesterol, SM, and ganglioside GM3 are relatively enriched in adsorbed exosomes as compared with total cell membranes on the expense of the relative amount of PE and its respective ether lipids, as well as PC, PI, PS, and PA. Note that BMP (LBPA) and the mitochondrial lipid CL comigrate and were not detected in exosomes. lysosomes), and biotinylated proteins (positioning the plasma membranes) and Western blotted for MHC class II. As expected, most MHC class II was found in plasma membranecontaining fractions (fractions 16 -21), but a significant amount was associated with MIIC, as indicated by its co-migration with ␤-hexosaminidase (fractions 5-10).
To compare exosomes with MIIC for the solubility of MHC class II, samples of pooled MIIC-containing fractions from density electrophoresis gradients and pooled exosome-containing sucrose gradient fractions were diluted with excess PBS, containing either CHAPS or Triton X-100 or lacking detergent. After 30 min at 0 or 37°C, DRM were pelleted by ultracentrifugation and analyzed for the presence of MHC class II by  6 and 7). The samples were incubated for 30 min at either 0°C (lanes 1-4, 6, and 8) or 37°C (lanes 5 and  7) and ultracentrifuged at 4°C to sediment DRM. Pellets were analyzed either by SDS-PAGE and stained with Coomassie Blue (A) or by Western blotting for MHC class II ␤ (B). Closed arrows indicate example proteins that were entirely solubilized by detergents, open arrows indicate partially solubilized proteins, and arrow heads indicate nonsolubilized proteins. Molecular weight markers are on the right. The data are representative of three independent experiments.  Western blotting (Fig. 8). Exosomes and MIIC were similar with respect to the solubility of MHC class II (Fig. 8A), consistent with the notion that the majority of MHC class II in MIIC localizes to internal vesicles and that these vesicles are released as exosomes upon exocytic fusion of MIIC with the plasma membrane.
To compare the detergent solubility of MHC class II at the plasma membrane with that of exosomes, intact cells and isolated exosomes were first biotinylated. This procedure allowed selective labeling of the exoplasmic domain of plasma membrane or exosome-associated MHC class II. After the addition of CHAPS, insoluble MHC class II was pelleted by centrifugation and solubilized in Triton X-100-containing PBS. Biotinylated MHC class II was collected using Neutravidin-conjugated Sepharose beads and analyzed by Western blotting. The solubilization of biotinylated MHC class II from exosomes in CHAPS was inefficient (Fig. 8B), possibly even more so than total MHC class II from non-biotinylated exosomes (Fig. 8A). In contrast to exosome-related and MIIC-derived MHC class II, however, plasma membrane-derived MHC class II was entirely solubilized by CHAPS (Fig. 8B). These data indicate that MHC class II is associated with DRM in exosomes and MIIC but not at the plasma membrane. Possibly, the incorporation of MHC class II in DRM at the MIIC-limiting membrane plays a role in its sorting into the MVB internal vesicles. DISCUSSION We developed a method to purify B cell derived exosomes to homogeneity, thus allowing determination of their protein and lipid composition. In a previous study we already demonstrated the presence of MHC class I and class II on B cell-derived exosomes (9, 10); here we show that they are among the most prominent proteins. MHC class I and class II have also been detected on exosomes from dendritic cells (11), intestinal epithelial cells (16), and T cells (21). MHC class I was also found to associate with tumor-derived exosomes (34), and MHC class II was found to associate with mast cell-derived exosomes (14). Previously, we also demonstrated by immunoelectron microscopy and Western blotting that the tetraspanins CD63, CD37, CD53, CD81, and CD82 are heavily enriched on B cell-derived exosomes and on the internal vesicles of MVB (10). One of these tetraspanins, CD63, is in fact also known as lysosome-associated membrane protein 3 (Lamp 3). Tetraspanins have also been demonstrated on exosomes derived from dendritic cells (11,12), intestinal epithelial cells (16), T cells (21), and platelets (15). In the current study we failed to detect tetraspanins by the mass spectrometric analyses. Despite their relative enrichment, the only tetraspanin detected previously (12) in exosomes by mass spectrometric analyses is CD9, most likely because tetraspanins cannot be recovered as discrete bands from acrylamide gels (10,13). Tetraspanins comprise a large group of ubiquitously expressed 25-50-kDa proteins that contain a number of conserved residues (35). Tetraspanins associate with each other, as well as with many Ig superfamily proteins, proteoglycans, integrins, growth factor receptors, and signaling enzymes, to form large transmembrane protein networks. Such networks are involved in a variety of processes at In A, isolated exosomes and MIIC were incubated in the absence or presence of CHAPS or Triton X-100 (TX100) at 0 or 37°C as indicated. The samples were then ultracentrifuged, and the pellets were analyzed by Western blotting for MHC class II ␤. MHC class II from MIIC and exosomes has similar solubilization characteristics. In B, biotinylated intact cells and biotinylated exosomes were incubated in the presence or absence of detergents as above. Biotinylated proteins were extracted with Neutravidin beads from pelleted detergent-resistant DRM that were resuspended in Triton X-100 and analyzed for the presence of MHC class II ␤-chain by Western blotting. Biotinylated MHC class II was almost entirely extracted by detergents from the plasma membrane but not from exosomes or MIIC, indicating that the incorporation of MHC class II in DRM depends on its subcellular location. The data are representative of three independent experiments. the plasma membrane such as cell adhesion, cell motility, and signaling (36). Many of the interactions within these networks are relatively stable in the presence of detergents. Moreover, detergent solubility assays demonstrated the association of detergent-resistant lipids with tetraspanin networks providing them with raft/DRM-like properties (37).
It has been well established that at least fractions of MHC class II and MHC class I localize to membrane microdomains, together with tetraspanins and integrins, as measured by coimmunoprecipitation from detergent lysates (37)(38)(39)(40), flow cytometric energy transfer methods (41), and competition assays (42). In mild detergents tetraspanins remain associated with microdomains resembling lipid rafts/DRM whereas at relatively harsh conditions they can be solubilized as protein webs that remain stable independently of lipid microdomains (37). The association of MHC class II with lipid rafts (43), as well as with tetraspan microdomains distinct from lipid rafts (44), has been proposed to facilitate antigen presentation. Although several of the above mentioned studies indicated the presence of such microdomains at the plasma membrane, their subcellular distribution has not been investigated systematically. MHC class II may associate with distinct sets of tertraspanins depending on its subcellular location (45). For example, CD82 associates with MHC class II in MIIC (46). Here we demonstrate that MHC class II is in detergent-resistant DRM/protein webs at MVB rather than at the plasma membrane.
We found that exosomes are enriched in cholesterol, sphingomyelin, and GM3. These lipids are characteristically enriched in rafts/DRM. These features, together with the presence of tetraspanins and the stable association of lipids with CHAPS-solubilized exosomal protein webs, indicate that exosomes contain protein/lipid complexes that can be described as webs or DRM/rafts. These webs or DRM/rafts may contribute to protein sorting in MVB but may also play a role in the generation of membrane buds and even in membrane fission (31,47). We did not observe detectable amounts of BMP by TLC on exosomes. This is seemingly inconsistent with other studies in which BMP was detected either immunocytochemically on exosomes (18) or biochemically in a subcellular fraction containing the internal membranes of MVB (48). Possibly, the amount and distribution of BMP is cell type-dependent. Furthermore, we observed by immunoelectron microscopy that in RN cells BMP is enriched on multilaminar lysosomes rather than on MVB. 2 Certain viruses, such as human cytomegalovirus (49) and HIV in macrophages (50), assemble at MVB in a process resembling the formation of MVB internal vesicles. T cells, in contrast to macrophages, assemble HIV predominantly at the plasma membrane in a process that requires elements of the same molecular machinery involved in MVB biogenesis (51). Interestingly, the membrane of HIV, type 1 is, like exosomes, enriched in cholesterol and sphingomyelin (52), and lipid rafts have been implicated in HIV, type 1 assembly and release (53).
At the plasma membrane, glycosylphosphatidylinositol-anchored proteins have the tendency to be incorporated into DRM/rafts. The fact that glycosylphosphatidylinositol-anchored proteins are also enriched in reticulocytes exosomes (54,55) is consistent with the idea that exosomes derive from DRM at the MVB-limiting membrane. G␣ subunits of heterotrimeric G proteins undergo palmitoylation and/or myristoylation on their amino-terminal ends and as a consequence are also targeted to lipid rafts (56). Given our current finding that exosomes contain G i ␣ 2 , it is possible that G i ␣ 2 is targeted to the internal vesicles of MVB because of its incorporation into lipid rafts/DRM at the MVB-limiting membrane. The presence of G i ␣ 2 in exosomes has also been reported for dendritic cellderived exosomes (12). The association of G i ␣ 2 with exosomes may be related either to general MVB functions or specifically to sorting of G protein-coupled receptors at MVB (57,58).
The presence of hsc70 in B cell-derived exosomes is consistent with its detection in exosomes from dendritic cells (12), tumors (34), and maturing reticulocytes (59,60). Maturing reticulocytes dispose of their transferrin receptors by incorporating them into exosomes, and it has been proposed that a direct interaction between hsc70 and the transferrin receptor cytoplasmic domain is associated with its targeting to exosomes (59,60). In addition to such chaperone functions, hsc70 is also acting in other processes involving protein folding and unfolding. For example, hsc70 has been demonstrated to regulate the disassembly of clathrin coats (61). Clathrin is most often associated with outward budding of membranes into the cytoplasm. However, clathrin has also been detected in non-curved lattices on MVB (62), (22) including MIIC (63). Clathrin in these coats has been demonstrated to associate with Hrs, an adaptor protein that binds directly to clathrin and ubiquitinated membrane proteins and is involved in the sorting of such proteins into the internal vesicles of MVB (22) (see also below). This process may result in the incorporation of clathrin, which we also detected in exosomes, and hsc70 into MVB internal vesicles. hsc70 forms complexes with other chaperones, including hsp90 (64), another heat shock protein that we detected in exosomes. Consistent with their presence in exosomes, these molecular chaperone complexes have been demonstrated to function in the translocation of cytoplasmic protein substrates into a subset of lysosome-like organelles (64). Although this process has been interpreted to reflect protein translocation across the lysosomal outer membrane, targeting to the internal vesicles of MVB cannot be excluded.
In addition to actin we detected moesin, an actin-binding protein of the ERM family in exosomes. Moesin has been demonstrated to play a role in de novo actin assembly on phagosomal membranes (65). ERM proteins may play a role in budding processes as they are incorporated into rhabdoviruses (66) and HIV (67, 68).
As indicated above, ubiquitination of membrane proteins serves as a signal for their sorting in MVB. Sequential association of sorting complexes, termed ESCRT 1-3, to their ubiquitin moiety is thought to select membrane proteins for sorting into the MVB pathway (19,69,70). All components that are required to recruit proteins into the MVB pathway, including ubiquitin, ESCRT complexes, and the clathrin coat, are released from assembled cargo prior to the actual packaging into inwardly budding vesicles at the MVB-limiting membrane. Dissociation of the ESCRT complexes seems to be regulated by the AAA-ATPase SKD1/Vps4, and interference with this process results in aberrant sorting in MVB (20,71). It should be noted, however, that not all proteins require ubiquitination for their sorting into the MVB pathway (23). Such proteins may partition into the inwardly budding vesicles because of intrinsic properties and preference to partition into raft-like microdomains or associate with tetraspanins. Endocytosed proteins that normally recycle, like the transferrin receptor and acetylcholinesterase, are, when aggregated by antibodies, mistargeted into the MVB pathway and subsequently secreted in association with exosomes (72). The hypothesis that protein clustering is an important determinant for entry into the MVB pathway is also supported by the observation that interference with transferrin receptor binding to hsc70 increases its aggregation and association with exosomes (60). Similarly, MHC class II (73) and associated invariant chain (74) have been shown to bind directly to hsc70, and this interaction may be important for the trafficking of MHC class II in MVB.
The transmembrane protein tyrosine phosphatase CD45 modulates the signal that is transduced via the B cell antigen receptor by regulating the phosphorylation state of Src family kinases and is required for normal B cell development, tolerance induction, and responsiveness to antigen (75). Interestingly, CD45 was found to be absent on T cell-derived exosomes (21). The relevance of the specific association of CD45 with B cell-derived exosomes is unclear. Na ϩ /K ϩ -ATPase is generally present at the plasma membrane. However, its cell surface expression can be regulated by endocytosis, and it plays a regulatory role in the acidification of endosomes and lysosomes (76).
The presence of a ␣4-integrin on B cell-derived exosomes is intriguing. Exosomes released by maturing reticulocyte also contain ␣4 ␤1-integrin (77). B cell selection involves their homing to follicular dendritic cells in the germinal center in a process that is dependent on the interaction of ␣4 ␤1 with VCAM-1 (78). B cell (RN)-derived exosomes bind in vitro to follicular dendritic cells and not to other cell types, suggesting that follicular dendritic cells are physiological targets for B cell-derived exosomes (18). As for B cells, binding of exosomes to follicular dendritic cells may require integrin ␣4. In the germinal center B cells recognize native antigens that are held in immune complexes at the surface of follicular dendritic cells by a set of different complement receptors. Binding of B cells to these antigens is an essential selection step for their differentiation into memory B cells. A prerequisite for the stimulation of T helper cells by follicular dendritic cells is their interaction with MHC class II. However, follicular dendritic cells do not synthesize MHC class II molecules themselves but rather passively acquire peptide-loaded MHC class II (79), possibly by binding B cell-derived exosomes through the interaction of ␣4 ␤1 with VCAM-1.