Separation and Characterization of Late Endosomal Membrane Domains*

Very little is known about the biophysical properties and the lipid or protein composition of membrane domains presumably present in endocytic and biosynthetic organelles. Here we analyzed the membrane composition of late endosomes by suborganellar fractionation in the absence of detergent. We found that the internal membranes of this multivesicular organelle can be separated from the limiting membrane and that each membrane population exhibited a defined composition. Our data also indicated that internal membranes may consist of at least two populations, containing primarily phosphatidylcholine or lysobisphosphatidic acid as major phospholipid, arguing for the existence of significant microheterogeneity within late endosomal membranes. We also found that lysobisphosphatidic acid exhibited unique pH-dependent fusogenic properties, and we speculated that this lipid is an ideal candidate to regulate the dynamic properties of this internal membrane mosaic.

Evidence is accumulating that some lipids contribute to the organization and functions of the vacuolar apparatus. Microheterogeneity of the plasma membrane could be studied using morphological and biophysical approaches, because the cell surface is readily accessible to probes and agents added from the medium. These studies revealed that, in addition to the fluid glycerophospholipid-rich regions of the bilayer, the plasma membrane also contains liquid-ordered regions rich in cholesterol and glycosphingolipids (rafts), in which glycosylphosphatidylinositol-anchored proteins and double-acylated proteins seem to preferentially partition. Lipid rafts are believed to act as cell surface platforms involved in internaliza-tion, signaling, and infection by pathogens and also to facilitate protein sorting in the biosynthetic pathway (1)(2)(3). However, it has been extremely difficult to study putative intracellular lipid domains without using perturbing agents, e.g. methyl-␤cyclodextrin to deplete cholesterol, or detergents, including CHAPS 1 (4,5), or Triton X-100 (6). Hence, very little is known about the biophysical properties and lipid or protein composition of membrane domains presumably present in endocytic and biosynthetic organelles.
In the endocytic pathway, molecules internalized into early endosomes are either recycled back to the plasma membranes or transported to late endosomes and lysosomes for degradation. Evidence is accumulating that some lipids are not randomly distributed in endosomal membranes along these recycling and degradation routes, contributing to the notion that endosomes contain a mosaic of structural and functional membrane domains (7). In particular, raft lipids seem to be abundant in recycling endosomes, at least in some cell types (8). However, lipid analogs with a preference for ordered membrane domains were reported to be transported toward late endosomes, in contrast to those with a preference for more fluid domains (9). Phosphatidylinositol 3-phosphate, which interacts with proteins containing a FYVE or a PX domain (10,11), appears to be more abundant at early stages of the endocytic pathway (12), and lysobisphosphatidic acid (LBPA) is restricted to late endosomes (13). Immunogold labeling of cryosections shows that LBPA is abundant within the internal membranes of this multivesicular or multilamellar organelle (13) and cannot be detected on the cytoplasmic surface of the limiting membrane (14). In this paper, we report that late endosomal membranes can be separated without detergents into distinct populations, each with a unique composition, corresponding to limiting and internal membranes. Our data also suggest that internal membranes can be subfractionated into more than one population, including LBPA-rich membranes, and that LBPA itself exhibits unique pH-dependent fusogenic properties. We conclude that late endosomal internal membranes contain different mem-brane domains and that their dynamic interplay is regulated by LBPA. ,4adiaza-s-indacene-3-dodecanoyl)-1-hexadecanoyl-sn-glycero-3-phosphocholine (␤-BODIPY FLC12-HPC) was from Molecular Probes (Eugene, OR). Dioleoylphosphatidylcholine (DOPC) was purchased from Avanti Polar Lipids, Inc. (Alabaster, AL). DEAE-Sephadex A-25 was from Amersham Biosciences AB. Pre-swollen carboxymethylcellulose CM52 was from Whatman. Monolayers of baby hamster kidney (BHK) cells were grown and maintained as described. When necessary to determine the protein composition of late endosome subfractions, cells were metabolically labeled by incubation for 16 h before the experiment with 0.5 mCi/ml [ 35  Purification of 3,3Ј-and 2,2Ј-LBPA-3,3Ј-LBPA was purified from BHK lipid extracts by preparative TLC after silica gel (Iatrobeads, Sigma, and DEAE-Sephadex A-25 column chromatography); the purified lipid migrated as a single spot on high performance TLC plates and was resistant to phospholipase A 2 digestion. 2,2Ј-LBPA was purified using DEAE Sephadex A-25 and CM52 column chromatography.

Cells and Reagents
Gas-Liquid Chromatography and Mass Spectrometry-Fatty acids in LBPA purified from BHK cells were converted to their methyl esters by treatment with 5% HCl in methanol for 1 h in boiling water and were analyzed by gas-liquid chromatography using a column of 10% EGSS-X on a Gas Chroma Z (100 -120 mesh). Electrospray ionization time of flight/mass spectrometric analysis was performed with Mariner (PerkinElmer Life Sciences). The samples were dissolved in acetonitrile/methanol (1:1) containing 1% ammonium formate and injected to the instrument. Mass spectrometric analysis was performed in a negative mode with a nozzle potential of Ϫ180 eV at a nozzle temperature of 180°C.
Liposome Fusion-Liposomes were prepared in 20 mM Hepes-Na, pH 7.3, by sonication; donor liposomes (19 M) contained 74% DOPC, 21% 2,2Ј-LBPA, and 5% ␤-BODIPY FLC12-HPC, and acceptor liposomes (120 M) contained 67% DOPC and 33% 2,2Ј-LBPA. In the assay, 50 l of acceptor liposomes was added to 2.5 ml of Hepes-Na, pH 7.3, and the pH of the solution was adjusted by adding 1 N HCl. Fusion was started by adding 50 l of donor liposomes and analyzed by fluorometry after excitation at 500 nm. The time course of increased fluorescence emission was measured at 515 nm and recorded.
Subcellular and Suborganellar Fractionation-Late endosomal fractions were prepared as described (13,15,16). Briefly, cells were homogenized, and then a post-nuclear supernatant was prepared. The postnuclear supernatant was adjusted to 40.6% sucrose, 3 mM imidazole, pH 7.4, loaded at the bottom of an SW60 tube, and overlaid sequentially with 35 and 25% sucrose solutions in 3 mM imidazole, pH 7.4, and then homogenization buffer (HB; 250 mM sucrose, 3 mM imidazole, pH 7.4). The gradient was centrifuged for 60 min at 35,000 rpm using an SW60 rotor. Early and late endosomal fractions were collected at the 35/25% and 25%/HB interfaces, respectively (13,15,16). We used two protocols to disrupt late endosomes, osmotic shock, or sequential freezing and thawing cycles. When the first protocol was used, fractions were first centrifuged for 30 min at 55,000 rpm in a TL55 rotor, and the pellet containing endosomes was resuspended under hypotonic conditions in 3 mM imidazole, pH 7.4 (osmotic shock), by 3 passages through a 22-gauge needle fitted onto a 1-ml tuberculin syringe. When the second protocol was used, fractions recovered from the gradient were frozen in liquid nitrogen and then thawed at 37°C, and this freezing and thawing cycle was repeated 3 times. In both cases, the suspension containing broken endosomes was then centrifuged for 30 min at 55,000 rpm in a TL55 rotor. The pellet was resuspended in 500 l of 40% sucrose in 3 mM imidazole, pH 7.4, loaded into an SW40 tube at the bottom of a linear 8 -40% sucrose gradient in the same buffer, and centrifuged at 4°C in the SW40 rotor for 16 h at 35,000 rpm. Fractions (1 ml each) were collected from the top of the tube.
Phospholipid Analysis-Cells were labeled to equilibrium for 16 h with 32 P i or [ 14 C]acetate and then fractions were prepared as above. Lipids of subcellular or suborganellar fractions were extracted in CHCl 3 /MeOH and then separated by two-dimensional chromatography (13). The first direction was run with chloroform, methanol, 32% ammonia (65:35:5, v/v) and the second direction with chloroform/acetone/ methanol/acetic acid/water (50:20:10:12.5:5, v/v). Radioactive lipids were detected by autoradiography and then quantified using the Molecular Imager System (Bio-Rad GS-363).
G-protein of Vesicular Stomatitis Virus-Incorporation of the G-protein of vesicular stomatitis virus (VSV) into the plasma membranes and subsequent internalization were as described (17)(18)(19). Briefly, virions were bound to the cell surface at 4°C, and then the viral envelope was fused with the plasma membrane at low pH. Then G molecules were cross-linked on ice with a polyclonal antibody, and cells were incubated at 37°C in the presence of 0.5 mg/ml leupeptin, to limit G-protein degradation (19). Cross-linked G molecules are endocytosed and transported to late endosomes and lysosomes for degradation (18). At the desired time, cells were processed for immunofluorescence or subcellular fractionation.
Electron Microscopy-To analyze the morphology and composition of endosomes, HeLa cells were fixed for 1 h at room temperature in a solution of 2% paraformaldehyde and 0.2% glutaraldehyde. The fixative was rinsed three times with PBS, and the cells were processed for cryosectioning essentially as described (20). Briefly, the cell pellet was infiltrated with sucrose and frozen in liquid nitrogen. Frozen sections were cut with a Leica FCS cryotome, transferred to grids, and incubated with an antibody to GFP (A-6455, Molecular Probes) and then with gold-coupled protein A. Alternatively, cells were fixed with 4% paraformaldehyde, and the grids were incubated with an antibody to CD63 (M1544 CLB) and then with a gold-coupled goat antibody to mouse immunoglobulin. Grids were examined in a Philips CM10 transmission electron microscope.

RESULTS
2,2Ј-Dioleoyl-lysobisphosphatidic Acid-Biochemical analysis of LBPA purified from BHK cells revealed that oleic acid accounted for Ͼ90% of LBPA fatty acyl chains (Table I). This was confirmed by mass spectrometry, because the most abundant ion peak was at m/z 773.54 (theoretical value, 773.53), corresponding to [M Ϫ H] Ϫ of LBPA containing two oleic acids (Fig. 1). These data show that Ͼ90% of the lipid is present as dioleoyl-LBPA in BHK cells. However, the position of the fatty acyl chains is unclear, either the ␤-(24) or the ␣-position (25,26) of the glycerol backbone. Because the ␤-position is thermodynamically unstable, fatty acids may migrate to the ␣-position during purification by silica column chromatography (27). Indeed, LBPA purified by silica chromatography (13) was completely degraded by R. arrhizus lipase ( Fig. 2A), which is well known to cut preferentially at the ␣-position. In contrast, an analysis of total lipids extracted from late endosomal fractions, hence without silica column chromatography, showed that LBPA was fully resistant to lipase, whereas PC and PE were almost completely degraded (Fig. 2, B and C). These data strongly suggest that acyl chains are predominantly esterified to the ␤-position in vivo and that LBPA is thus mostly present in living cells as the 2,2Ј-dioleoyl isoform (herein referred to as 2,2Ј-LBPA, Fig. 2D). LBPA is a structural isomer of phosphatidylglycerol with an uncommon stereoconfiguration, which presumably accounts for its resistance to phospholipase and degradation (28). The lipid is related to cardiolipin and is predicted to be cone-shaped. In addition, the presence of free hydroxyl groups at the 3,3Јposition ( Fig. 2D) may influence its biophysical properties. Indeed, these groups are presumably not buried in the hydrophobic regions of the bilayer but exposed on its outer surface, and thus likely to face the acidic endosomal milieu. We therefore measured the pH-dependent fusion properties of donor liposomes containing 74 mol % dioleoylphosphatidylcholine (DOPC), 21 mol % 2,2Ј-LBPA, and 5 mol % ␤-BODIPY FLC12-HPC, as a fusion tracer, with acceptor liposomes containing 67 mol % DOPC and 33 mol % 2,2Ј-LBPA. As shown in Fig. 3, these liposomes were highly fusogenic at pH 5.6, a value in the range of late endosomal and lysosomal pH (29). Fusogenicity decreased at pH 6.0 and was essentially abolished at pH 6.5 or 7.3. In contrast, liposomes lacking LBPA did not show pH-dependent fusion activity (not shown). These observations show that 2,2Ј-LBPA is endowed with unique properties and suggest that the lipid might contribute to the dynamic properties of late endosomal internal membranes.
Subfractionation of Late Endosomal Lipids-Next, we investigated whether late endosomal lipids could be fractionated into different membrane populations, using a "suborganellar fractionation" protocol. We have shown previously (13,15,16) that late endosomes are easily separated from other endosomal or biosynthetic membranes by floatation in a step sucrose gradient. These purified late endosomal fractions do not contain the mitochondrial lipid cardiolipin and are depleted in phosphatidylserine (PS), cholesterol, and sphingomyelin, which are enriched in recycling endosomes (8,13). Purified late endosomes contain a defined subset of phospholipids (PLs) and are highly enriched in LBPA. We showed previously (13,30) that LBPA accounts for Ϸ0.5% of cellular 32 P-labeled PLs after metabolic labeling with 32 P for 20 h but Ϸ14% of late endosomal 32 P-PLs (see Table II). In order to ensure that these values were not biased by the lipid phosphate group turnover, experiments were repeated after metabolic labeling with [ 14 C]acetate. As shown in Table II, very similar values were obtained when comparing 32 P-and 14 C-labeled lipids.
Purified late endosomes were then ruptured under hypotonic conditions by 3 freeze/thaw cycles. Endosome rupture was assessed by measuring the amounts of endocytosed horseradish peroxidase, a content marker, that remained endosome-associated after sedimentation at high speed (latency) (31). After the treatment, latency was reduced to 10% of the untreated control (not shown). This treatment was gentle, because 3 freeze/thaw cycles had hardly any effect on the latency of an early endosomal content marker (not shown). Ruptured late endosomes were then loaded at the bottom of a continuous sucrose gradient (8 -40%) and centrifuged to equilibrium for 16 h at 4°C. After metabolic labeling with 32 P, 32 P-PLs distributed across the entire gradient ( Fig. 4)  late endosome (not shown). As a control, early endosome 32 Plabeled PLs after the same freeze/thaw cycles distributed as a single peak in fraction 7 (d ϭ 1.1031 g/cm 3 ), at the expected density of early endosomes (see Fig. 6). These experiments show that late endosomal lipids can be subfractionated into different membrane populations.
Lipid Composition of Suborganellar Fractions-We then analyzed the PL composition of each fraction, and we found that the distribution of each PL varied significantly across the gradient (Fig. 5A). Phosphatidylcholine (PC) was mostly (Ϸ55%) found in fractions 3-5 at d ϭ 1.0571-1.0772 g/cm 3 . In these fractions, PC accounted for Ͼ70% of the total PLs, while accounting for Ϸ48% of total 32 P-PLs in late endosomes (Table II) and other membranes (13). Phosphatidylethanolamine (PE, 30%) and phosphatidylinositol (PI, 40%) were recovered predominantly in fraction 7 at d ϭ 1.1031 g/cm 3 . In this fraction, PE and PI accounted for 42 and 16% of total PLs, while representing roughly 19 and 4% of total late endosomal 32 P-PLs, respectively (Table II), again much like in other membranes (13). Finally, the bulk of LBPA (Ϸ40%) was found in fractions 11 and 12 (d ϭ 1.1545-1.1654 g/cm 3 ), where it was highly enriched, corresponding to Ϸ60% of total PLs in these fractions, and representing Ϸ15% of total late endosomal PLs (Table II). Some LBPA was also found in fraction 1, perhaps reflecting LBPA partial association with neutral lipids, in particular cholesterol esters (30). This PL distribution did not result from the fragmentation protocol, because it was essentially identical after osmotic rupture of late endosomes (not shown). When using early endosomal fractions as starting materials, no difference in the PL composition of fractions across the peak (fraction 7, d ϭ 1.1031 g/cm 3 ) was detected (Fig. 6).
Lamp1 and CD63-We then analyzed the distribution of two late endosomal proteins, Lamp1 and CD63, which colocalize with LBPA in late endosomes (13,30). In contrast to LBPA, the bulk of Lamp1 is present on the organelle-limiting membrane, even though the protein can occasionally be detected within internal membranes (13,16,32). After suborganellar fractionation, small amounts of Lamp1 (Ϸ15%) were found in fraction 12 containing LBPA, perhaps corresponding to some Lamp1 molecules present within LBPA-rich membranes. The bulk of Lamp1, however, peaked in fraction 7 with a complex PL composition (Fig. 7A) and did not co-fractionate with LBPA, indicating that Lamp1-and LBPA-containing membranes could be separated after late endosome subfractionation.   (13,15,16), and lipids were analyzed by two-dimensional chromatography and autoradiography. Phospholipids were quantified using the Molecular Imager System, and amounts are expressed as a percentage of the total late endosomal content. 32 4. Total late endosomal phospholipid distribution after subfractionation. After metabolic labeling with 32 P i , late endosomes were prepared, ruptured by 3 freeze/thaw cycles, and then subfractionated by floatation in a sucrose gradient (suborganellar fractionation). After collecting fractions (1, top; 12, bottom), lipids were extracted in CHCl 3 /MeOH, and 32 P-lipids were counted. The same distribution was obtained after osmotic rupture of late endosomes, indicating that the PL distribution did not depend on the fragmentation protocol. Intact late endosomes were recovered at their expected density as a single peak (not shown).
The tetraspanin CD63 is abundant within internal membranes, in marked contrast to Lamp1, but is also present on the organelle-limiting membrane (33,34) (Fig. 8B). Consistently, the protein distributes not only to vesicles but also to highly dynamic late endosomal tubules visualized by time-lapse video microscopy, like Lamp1 (35). We have shown recently (35) that both endogenous CD63 and green fluorescent protein (GFP)tagged CD63 colocalize with both Lamp1 and LBPA in several cell types, including HeLa and BHK cells. We then analyzed the distribution of transiently expressed CD63-GFP by electron microscopy in HeLa cells because of their very high transfection rate. CD63-GFP distributed to perinuclear late endosomes (35) (Fig. 8A) with a characteristic multivesicular ultrastructure (Fig. 8, C and D). Immunogold labeling of cryosections showed similar labeling patterns using anti-GFP antibodies (Fig. 8C) or anti-CD63 antibodies (Fig. 8D), which reveal both endogenous CD63 and CD63-GFP in these cells. CD63-GFP was abundant within internal membranes and also present on the limiting membrane, much like endogenous CD63 (33, 34) (Fig. 8B).
We therefore used CD63-GFP in BHK cells in our experiments, because existing anti-CD63 antibodies do not crossreact with BHK cells used in these fractionation studies. The overall distribution of CD63 was different from that of Lamp1 after subfractionation (Fig. 7B), because the bulk (Ϸ65%) of CD63-GFP was found in fractions 3-5 (containing mostly PC) and 12 (containing mostly LBPA) (Fig. 7B). CD63 (Ϸ30%) was also found in fractions 7 and 8 that contained Lamp1, presumably reflecting the presence of CD63 within both the limiting membrane and late endosomal tubules visualized by time-lapse FIG. 5. Phospholipid composition of late endosome subfractions. 32 P-Phospholipids from late endosomal subfractions were prepared as in Fig. 4 (density of each fraction is indicated in g/cm 3 ) and analyzed by two-dimensional TLC and autoradiography. Phospholipids were then quantified using the Molecular Imager System. A, the values for each lipid in a representative experiment are expressed as a percentage of the total 32 P-PL content of each fraction. B, pie-like representation of the peak fractions. PS, phosphatidylserine; SM, sphingomyelin; UNK, unknown lipid. video microscopy (35). Because the bulk of Lamp1 is present in the late endosome limiting membrane in vivo, and in fractions 6 -8 after separation in our gradients, we conclude that the two membrane populations containing CD63 and either PC or LBPA, but not Lamp1, are likely to be derived from internal membranes.
Endocytosis of the VSV-G Glycoprotein-Internal membranes of multivesicular endosomes presumably contain proteins destined for degradation in lysosomes (36). To characterize further the membrane populations obtained after subfractionation, we used the trans-membrane glycoprotein G of VSV as marker of the degradation pathway. VSV-G can be incorporated into the plasma membrane by low pH-mediated fusion of the viral envelope with the plasma membrane (17)(18)(19). After cross-linking with antibodies, VSV-G is efficiently endocytosed and then degraded in lysosomes (18,19). Fig. 9 shows that endocytosed VSV-G was transported to LBPA-positive late endosomes and that transport was inhibited by microtubule depolymerization (Fig. 9), as expected (19).
Late endosomes were then analyzed as above, after internalization of cross-linked VSV-G in the presence of leupeptin in the medium to limit degradation (Fig. 10B). VSV-G distributed across the entire gradient, with the bulk (Ϸ70%) in fractions 2-4 (d ϭ 1.0388 to 1.0655 g/cm 3 ) and 10 -12 (d ϭ 1.1423 to 1.1654 g/cm 3 ). VSV-G was also present in fractions 7 and 8, containing Lamp1, consistently with its presence at the limiting membrane (18,19). The broader VSV-G distribution, when compared with that of Lamp1, cannot be accounted for by differences in the amounts of the two proteins; very low amounts of VSV-G were incorporated into the plasma membrane (Ͻ1% of total plasma membrane protein) (18), whereas lysosomal glycoproteins are very abundant (Ϸ50% of total late endosomal/lysosomal proteins) (32). Interestingly, despite the presence of leupeptin, top and bottom fractions contained processed, trans-membrane forms of VSV-G, revealed using the P5D4 antibody against G-protein cytoplasmic domain (Fig.  10B). When using intact late endosomes as a control, VSV-G was recovered as a single peak (d ϭ 1.0772 g/cm 3 ) at the late endosome density (not shown), as expected (Fig. 2C). Similarly, when early endosomal fractions were prepared 5 min after VSV-G internalization (18, 19), VSV-G was also recovered as a single peak around fraction 7 (d ϭ 1.1031 g/cm 3 ), whether early endosomes had been subjected to the same freeze/thaw cycles FIG. 10. Subendosomal fractionation of the G-protein. A, VSV-G was endocytosed for 5 min so that the G-protein reached early endosomes (19). Early endosomes were subjected to the same protocol as in Fig. 6, and G-protein distribution was analyzed by gel electrophoresis followed by Western blotting with the P5D4 monoclonal antibody against the G-protein cytoplasmic domain. B, as in Fig. 8, VSV-G was endocytosed for 45 min but in the presence of leupeptin in the medium to limit degradation. VSV-G distribution was then analyzed after subfractionation, as above. (In the absence of leupeptin, G molecules that had not been degraded exhibited a similar distribution.) Molecular mass markers are indicated.  9. Distribution of endocytosed G-protein. VSV-G was incorporated into the plasma membrane by low pH-mediated fusion of the viral envelope with the plasma membrane and cross-linked with a rabbit polyclonal antibody against the G-protein. Cells were incubated for 45 min at 37°C with or without 10 M nocodazole to depolymerize microtubules, as indicated, and then processed for double immunofluorescence using antibodies against LBPA (6C4) and against rabbit IgG (to detect endocytosed G-protein).
( Fig. 10A) or not (not shown). These experiments thus show that a cargo protein destined for degradation distributed within all membrane populations separated on our gradient, including those containing markers of internal membrane domains. DISCUSSION Our observations show that different membrane populations with a defined lipid composition can be physically separated after late endosome rupture under conditions sufficiently gentle to preserve early endosome integrity. It is unlikely that this heterogeneity simply reflects selective lipid partitioning during endosome rupture by freezing and thawing, because the same distribution was also observed after osmotic shock. In addition, this heterogeneity cannot be accounted for by the presence of early endosomes, recycling endosomes, or biosynthetic organelles in the initial fraction used as starting material. BHK late endosomes are recovered at a very low buoyant density in sucrose (d Ϸ1.0772 g/cm 3 ), well separated from these organelles (13,15,16). More importantly, LBPA itself provides a convenient late endosome lipid marker, and all proteins we studied colocalize with LBPA in late endosomes (13,30) (Fig. 7).
Lamp1, which is restricted to late endosome limiting membrane by immunogold labeling of cryosections, is found mostly in fractions with a complex lipid composition and yet distinct from that of early endosomes. This composition presumably reflects the fact that the limiting membrane is directly connected by membrane traffic pathways to other compartments. Late endosomes also form highly dynamic tubular regions that contain CD63 and Lamp1 (35), and these tubules may well account for a significant portion of the membrane population containing both proteins (fractions 7 and 8). In contrast, internal membranes form a highly specialized and privileged environment. As opposed to other vacuolar membranes, internal membranes are not directly integrated into the vacuolar membrane flow, a situation reminiscent of chloroplast and mitochondria inner membrane systems.
Suborganellar fractionation reveals the presence of two membrane populations, which are both well separated from Lamp1 membranes, suggesting that they originate from intralumenal membranes. Both populations contain CD63, which is predominantly present in internal membranes (33), and transmembrane processed forms of VSV-G, a marker of the degradation pathway (18). However, only one of these populations contains the bulk of LBPA, and LBPA is dramatically enriched in these membranes (Ϸ70% of total PLs), whereas the second population is enriched in PC. This may seem a priori surprising. However, internal membranes seem to be highly heterogeneous when visualized by electron microscopy, because late endosomes appear multivesicular or multilamellar, sometimes with highly ordered membrane arrays (37). Essentially nothing is known about the biochemical or biophysical nature of these differences. These membranes are also heterogeneous in composition, because LBPA is abundant in late endosomes but not in multivesicular intermediates at earlier stages of the pathway (13). In addition, LBPA, although present in significant amounts, represents only Ϸ15% of the total PLs in the late endosomal fraction. LBPA is thus unlikely to be the single major PL species of internal membranes, because these are very abundant in multivesicular regions (16,32). In contrast, PC is the most abundant lipid (45% of total lipids; 3:1 molar ratio PC:LBPA) and, as such, should represent a major constituent of internal membranes.
In addition to morphological and compositional differences, internal membranes also appear to fulfill different functions. They accumulate molecules destined to be degraded (19,36), consistently with the findings that LBPA and other negatively charged PLs facilitate glycolipid degradation (38). However, internal membranes also accumulate Man-6-P receptor in transit (32), tetraspanins (33), and major histocompatibility complex class II receptor (39). In fact, the cycle of Man-6-P receptor is inhibited when interfering with LBPA functions, as is cholesterol transport (13,40). Conversely, cholesterol accumulation inhibits Man-6-P receptor and CD63 transport (30,34). Finally, our observations show that LBPA membranes separated on the gradient contain CD63 but also processed forms of VSV-G. It thus appears that LBPA membranes have a dual role in degradation and transport, perhaps suggesting that these membranes have turnpike functions within the complex system of late endosomal internal membranes. It is attractive to believe that late endosome internal membranes may be composed primarily of PC, a lipid with a preference for fluid regions that can accommodate the high curvature of internal vesicles and tubules but that the organization and dynamic properties of these membranes depend on LBPA domains. Indeed, the cone-shaped structure and pH-dependent fusion activity of LBPA make it an ideal candidate to regulate internal membrane biogenesis and interactions. We conclude that late endosome inner membranes form a mosaic of lipid domains and that their dynamic properties depend on LBPA.