Shiga toxin glycosphingolipid receptors in microvascular and macrovascular endothelial cells: differential association with membrane lipid raft microdomains.

Vascular damage caused by Shiga toxin (Stx)-producing Escherichia coli is largely mediated by Stxs, which in particular, injure microvascular endothelial cells in the kidneys and brain. The majority of Stxs preferentially bind to the glycosphingolipid (GSL) globotriaosylceramide (Gb3Cer) and, to a lesser extent, to globotetraosylceramide (Gb4Cer). As clustering of receptor GSLs in lipid rafts is a functional requirement for Stxs, we analyzed the distribution of Gb3Cer and Gb4Cer to membrane microdomains of human brain microvascular endothelial cells (HBMECs) and macrovascular EA.hy 926 endothelial cells by means of anti-Gb3Cer and anti-Gb4Cer antibodies. TLC immunostaining coupled with infrared matrix-assisted laser desorption/ionization (IR-MALDI) mass spectrometry revealed structural details of various lipoforms of Stx receptors and demonstrated their major distribution in detergent-resistant membranes (DRMs) compared with nonDRM fractions of HBMECs and EA.hy 926 cells. A significant preferential partition of different receptor lipoforms carrying C24:0/C24:1 or C16:0 fatty acid and sphingosine to DRMs was not detected in either cell type. Methyl-β-cyclodextrin (MβCD)-mediated cholesterol depletion resulted in only partial destruction of lipid rafts, accompanied by minor loss of GSLs in HBMECs. In contrast, almost entire disintegration of lipid rafts accompanied by roughly complete loss of GSLs was detected in EA.hy 926 cells after removal of cholesterol, indicating more stable microdomains in HBMECs. Our findings provide first evidence for differently stable microdomains in human endothelial cells from different vascular beds and should serve as the basis for further exploring the functional role of lipid raft-associated Stx receptors in different cell types.

In this study, we performed for the fi rst time a compositional analysis of microdomains obtained as detergentresistant membranes (DRMs) with special reference to the distribution of Stx receptors in detergent-resistant and detergent-soluble membrane fractions of HBMECs and EA.hy 926 endothelial cells. We report here on remarkable differences in microdomain composition with respect to the occurrence of Stx receptors and, moreover, on differential lipid raft stability toward cholesterol-depletion of the two endothelial cell types, where raft disintegration was found to be accompanied by loss of Stx receptors in DRM fractions. The newly developed technique combining thin-layer chromatography (TLC) separation and immunodetection of Stx receptors on the TLC plate with infrared matrix-assisted laser desorption/ionization orthogonal time-of-fl ight mass spectrometry (IR-MALDI-o-TOF-MS) served as an indispensable tool for structural analysis of small GSL amounts directly on the TLC plate ( 15 ).

Methyl-␤ -cyclodextrin treatment of endothelial cells
The infl uence on the cellular integrity upon methyl-␤cyclodextrin (M ␤ CD)-mediated cholesterol depletion was controlled microscopically. For this purpose, endothelial cells were grown in 24-well tissue culture plates (Greiner Bio-One, Frickenhausen, Germany) until confl uence in media as described above and treated for 1 h with cell-culture-tested M ␤ CD (Sigma-Aldrich Chemie GmbH, Steinheim, Germany), with increasing concentrations from 1 mM to 50 mM M ␤ CD in serum-free medium. Additional control cell culture experiments without M ␤ CD were performed for 1 h with serum-supplemented medium (see previous section), under serum-free conditions and with phosphate-buffered saline (PBS). Cells were evaluated at 20× and 100× magnifi cation using an Axiovert 40 inverse microscope (Zeiss, Göttingen, Germany) fi tted with an AxioVision 3.1 digital camera (Zeiss, 1,360 × 1,040 pixel). Data were processed with Adobe Photoshop software (Adobe Systems Inc., San Jose, CA).
For lipid and protein analysis, endothelial cells were grown each in two 175 cm 2 tissue culture fl asks until 80% confl uence in serum-supplemented medium (see above), washed twice with prewarmed serum-free medium, and incubated for 1 h in serumfree medium with 10 mM M ␤ CD. The ensuing lipid and protein extraction was performed as described below.

Lipid extraction and isolation of neutral GSLs from endothelial cells
Confl uent grown cells from four 175 cm 2 tissue culture fl asks (Greiner) were trypsinized and washed with prewarmed PBS. ( 13 ). Thus, the specifi c GSL composition of lipid rafts and/ or caveolae of many cell types remains to be elucidated.
GSLs are built up from a hydrophobic ceramide moiety and a hydrophilic oligosaccharide residue ( 14,15 ). They are embedded with the ceramide moiety in the outer leaflet of the plasma membrane of animal cells, where they are clustered in a more ordered state within the generally disordered milieu of the membrane. The localization of GSLs in the exofacial leafl et of the plasma membrane makes them excellent candidates as recognition structures in cell-cell interactions ( 16 ) and as receptors for numerous pathogens, such as viruses ( 17,18 ) and bacteria ( 19,20 ), as well as for bacterial toxins, including Shiga toxins (Stxs) ( 21,22 ). Much effort is currently spent employing (glyco)sphingolipidomics (23)(24)(25)(26) by use of various mass spectrometry technologies ( 15,(27)(28)(29)(30)(31) to better understanding their nature and functional role ( 32 ).
Human endothelial cells from various vascular beds compose globotriaosylceramide (Gb3Cer/CD77) and globotetraosylceramide (Gb4Cer) as major neutral GSLs, known as high-and low-affi nity receptors, respectively, for Stx1 and Stx2. These GSLs have been detected in various types of endothelial cells ( 22 ), and their structures have been characterized in detail; e.g., human umbilical vein endothelial cells (HUVECs) (33)(34)(35). Vascular damage is largely mediated by Stx, which particularly injure microvascular endothelial cells in the kidney and brain (36)(37)(38). Stx are released by enterohemorrhagic Escherichia coli (EHEC) in the gut, translocated across the intestinal epithelium into circulation ( 39 ), and then transported to endothelial cells (39)(40)(41). The pentameric B-subunit of Stx binds to the cell surface, followed by internalization and retrograde transport via the Golgi apparatus to the endoplasmic reticulum ( 42 ). After translocation into the cytosol, the enzymatically active A-subunit exerts its toxic function through inhibition of protein biosynthesis ( 43,44 ).
Several studies demonstrated the clustering of Gb3Cer in lipid rafts ( 45 ), the density-dependent binding of Stx with raft-localized receptors ( 46 ), and, moreover, raftassociation of Stx receptors as a requirement for the retrograde transport ( 47,48 ) and retro-translocation across the endoplasmic reticulum ( 49 ). Thus, according to present knowledge, only GSLs that associate strongly with lipid rafts can sort AB 5 toxins (including Stx) backward from the plasma membrane to the endoplasmic reticulum ( 50,51 ). Lipid raft association of GSLs has been described so far in different cell types, such as intestinal ( 45,50 ), HeLa, and Vero cells (47)(48)(49)46 ), whereas the membrane localization of GSLs of human endothelial cells has so far not been analyzed in detail and remains largely unknown. To this day, the structural characterization of the different lipoforms of Stx GSL receptors of the macrovascular HUVEC-derived EA.hy 926 cell line and human brain microvascular endothelial cells (HB-MECs) has been reported by us ( 52 ), indicating that HB-MECs express both Gb3Cer and Gb4Cer, whereas EA.hy 926 cells were found to synthesize Gb3Cer but not elongated Gb4Cer. luted 1:500 in PBS with 1% BSA and 0.3% Triton X-100. The slides were then washed three times with PBS and incubated for 1 h under light protection with fl uorochrome-labeled secondary antibodies diluted 1:1000 in PBS with 1% BSA and 0.3% Triton X-100. After threefold washing in PBS, nuclear DNA was stained for 10 min with 4´,6-diamidino-2-phenylindol (DAPI, Sigma-Aldrich) in PBS (20 ng/ml), and the slides were embedded in antifading mounting medium (Dianova, Hamburg, Germany). Controls without primary antibodies were processed in parallel. Bound fl uorochrome-labeled antibodies and stained nuclei were visualized under a fl uorescence microscope (Axio Imager.A1, Zeiss), original magnifi cation × 100, with fi lter sets adequate to the maxima of absorption/emission of Cy5 ® (649 nm/670 nm), Alexa Fluor ® 488 (495 nm/519 nm), Alexa Fluor ® 546 (556 nm/573 nm), and DAPI (368 nm/488 nm). The fl uorescence was recorded with an AxioCam CCD camera (Zeiss), documented with AxioVision 4.8, and processed with Adobe Photoshop software (Adobe Systems Inc.).

Preparation of DRMs
The preparation of DRMs was done according to Brown and Rose ( 65 ) with minor modifi cations. Detergent extractions were done on ice with prechilled solutions. Membranes from confl uent grown cells were disintegrated by cell lysis in 10 mM HEPES pH 7.4, 42 mM KCl, 5 mM MgCl 2 , and protease-/phosphataseinhibitors (Complete, Mini cocktail tablets, Roche Diagnostics) under sonication (Sonifi er Bandelin Sonopuls HD 2070, 20 counts, each 2 s, 80% power). Cell debris was removed by centrifugation at 400 g for 10 min at 4°C, followed by ultracentrifugation of the supernatant at 150,000 g for 30 min at 4°C to separate cytosol from remaining membranes. The membrane sediment was resuspended with a 20G × 11/2 drain tube 0.9 mm × 40 mm (Microlance TM 3, BD, Heidelberg, Germany) in 1 ml of 1% Triton X-100 (v/v) in TNE-buffer (10 mM Tris pH 7.5, 150 mM NaCl, 5 mM EDTA) and kept on ice for 30 min. The resuspension procedure was repeated twice with 30 min breaks between GSLs were sequentially extracted from the sediments with 15 ml each of methanol, chloroform/methanol (1/2, v/v), chloroform/methanol (1/1, v/v), and chloroform/methanol (2/1, v/v). The combined supernatants were dried by rotary evaporation, and phospholipids were saponifi ed by incubation in 20 ml of aqueous 1N NaOH for 1 h at 37°C. After neutralization with 2 ml of 10 N HCl, the samples were dialyzed against deionized water and dried by rotary evaporation. Neutral GSLs were separated from gangliosides by anion-exchange column chromatography employing DEAE-Sepharose CL-6B (GE Healthcare, Munich, Germany) in the acetate form and purifi ed by silica gel 60 (particle size 0.040-0.063 mm, Merck, Darmstadt, Germany) chromatography according to standard procedures ( 55,56 ). The neutral GSL fractions were adjusted to defi ned volumes of chloroform/ methanol (2/1, v/v) corresponding to 1 × 10 5 cells per µl.
Human heart whole-cell lysate containing caveolin proteins (Catalog No. ab29431, Abcam, Cambridge, MA) and HeLa wholecell lysate containing fl otillin proteins (Catalog No. sc-2200, Santa Cruz Biotechnology) were used as positive controls.

Immunofl uorescence microscopy
HBMECs and EA.hy 926 cells were seeded in 8-well polystyrene chamber slides (tissue culture chambers, Permanox TM , Nunc GmbH) in amounts of 1.5 × 10 4 and 3 × 10 4 cells/chamber, respectively, and grown for 24 h until ‫ف‬ 80% confl uence. Medium was aspirated, and cells were washed with prewarmed PBS (Lonza) and fi xed for 30 min at room temperature with 3.7% paraformaldehyde (Merck). If not otherwise stated, the following steps were performed at ambient temperature. The fi xed cells were washed with PBS and quenched for 15 min with 0.2 M glycine (Roth, Karlsruhe, Germany; pH 7.2). After twofold washing with PBS, cells were blocked and permeabilized with 5% (w/v) BSA (Serva Electrophoresis, Heidelberg, Germany), 0.2% (v/v) fi sh skin gelatin (Sigma-Aldrich) and 0.3% Triton (v/v) X-100 (Serva Electrophoresis) in PBS for 2 h. After washing with PBS, the slides were incubated overnight at 4°C with anti-Gb3Cer antibody 38.13, anti-caveolin-1, or anti-fl otillin-2 antibody, each di-µm, ‫ف‬ 100 ns; Spektrum, Berlin). The focal laser spot size was about 200 µm × 300 µm. After soaking of the BCIP-stained TLC stripes for 2 h in 10 mM ammonium acetate buffer (pH 3.6) and drying at room temperature, Plexigum P28 was removed by threefold consecutive dipping in distilled chloroform. For MS analysis, the TLC stripes were cut to pieces of ‫ف‬ 15 mm × 40 mm to fi t into the custom-made MALDI sample plate. A matrix of 0.3 µl drops of glycerol was applied with a pipet across the positive GSL bands. Spectra were acquired in the positive ion mode. About 300 laser shots at a fi xed laser fl uence were applied to the GSL bands to obtain a measure for relative variation of GSL abundances. Absolute quantities cannot be derived from the IR-MALDI-MS spectra.

Detection of Stx receptors in neutral GSL fractions of total cells using TLC overlay assay
GSLs were extracted from both HBMECs and EA.hy 926 cells. Neutral GSLs were isolated by anion-exchange chromatography. Aliquots of neutral GSLs equivalent to identical cell quantities were separated by TLC followed by orcinol staining and immunodetection ( Fig. 1 ). Stx receptors of the globo-series were immunostained with anti-Gb3Cer antibody ( Fig. 1A ) and anti-Gb4Cer antibody ( Fig. 1B ) in addition to orcinol staining ( Fig. 1C ). The anti-Gb3Cer antibody bound to Gb3Cer of HBMECs and EA.hy 926 cells ( Fig. 1A , lane b and c, respectively) where upper bands harbor Gb3Cer structures with mostly C24 and lower bands Gb3Cer with C16 fatty acid, which were verifi ed by IR-MALDI-o-TOF-MS (see next section). The anti-Gb4Cer assay demonstrated that Gb4Cer is present in HBMECs ( Fig. 1B , lane b), but undetectable in EA.hy 926 cells ( Fig. 1B , lane c), which is in agreement with the results of the orcinol stain ( Fig. 1C ) and previous investigations ( 52 ). the steps. The homogenous solution was then overlayed with a discontinuous sucrose bottom-to-top gradient of 42.5%, 30%, and 5% sucrose (each w/v) in TNE-buffer. DRMs were separated from the bulk of detergent-soluble membrane lipids by ultracentrifugation at 200,000 g for 19 h at 4°C owing to their lower buoyant density whereby the light DRMs fl oat to the interface between 5% and 30% sucrose in the density gradient. Eight fractions of 1.5 ml volume each were collected from top to bottom and the sediment was dissolved in 1.5 ml of PBS (= fraction 9). A scheme of the procedure is provided as supplementary Fig. I.

Isolation of GSLs, cholesterol, and proteins from density gradient fractions
Lipids, including GSLs and cholesterol, were extracted from 1.2 ml of each fraction. Phospholipids were saponifi ed by adding 120 µl 10 M NaOH under gentle agitation for 1 h at 37°C. After neutralization with 120 µl of 10 M HCl, the samples were dialyzed at 4°C for two days against deionized water and dried by rotary evaporation. The extracts were dissolved under short sonication in chloroform/methanol (2/1, v/v) and adjusted to defi ned volumes of chloroform/methanol (2/1, v/v) corresponding to 1 × 10 4 cells per µl. GSLs and cholesterol were analyzed by TLC as described above.
Volumes of 0.3 ml of each density gradient fraction (see above) were dialyzed for two days at 4°C against deionized water to remove sucrose. Protein concentrations were determined in triplicate in maxi-sorp 96-well microtiter plates (Nunc) according to Lowry et al. ( 66 ) at an absorbance wavelength of = 630 nm using an EL800 microwell plate reader (BioTek Instruments, Bad Friedrichshall, Germany; software Gen5 TM ) and BSA as reference protein. Proteins were precipitated with 5% (w/v) trichloroacetic acid, washed with absolute ethanol, and applied to SDS-PAGE. Three independent experiments were performed, and representative measurements are shown in the results section.

SDS-PAGE and Western blotting
The precipitated proteins of the density gradient fractions were dissolved in 1 × sample buffer, heated for 5 min at 95°C, and loaded onto the gels. SDS-PAGE separation was done under reducing conditions using 4% stacking and 12% separating gels of 0.75 mm thickness ( 67 ). Standard proteins in a molecular range from 14 kDa to 66 kDa were used as references (Dalton Mark VII-L; Sigma-Aldrich). Proteins were fi xed with 20% methanol in 25 mM Tris buffer (pH 8.25) supplemented with 192 mM glycine. SDS-PAGE-separated proteins were transferred by means of a semi-dry blotting system onto a nitrocellulose membrane (Amersham Hybond TM -C Extra; GE Healthcare, Freiburg, Germany). Nonspecifi c binding was blocked by incubating the membrane for 1 h in Tris-buffered saline with 0.1% (v/v) Tween-20 (TBS/T) supplemented with 3% (w/v) BSA (= blocking solution, pH 7.6). After washing with TBS/T, the membrane was overlayed with anti-caveolin-1 or anti-fl otillin-2 antibodies (diluted 1:4000 in blocking solution) with gentle agitation overnight at 4°C. The membrane was then washed three times for 5 min each with TBS/T, followed by incubation with alkaline phosphataselabeled anti-rabbit IgG secondary antibody (diluted 1:2000 in blocking solution) for 1 h. After three washings with TBS/T, bound antibodies were visualized with 0.005% (w/v) BCIP and 0.01% (w/v) nitroblue tetrazolium (NBT/BCIP kit; Invitrogen).

IR-MALDI-o-TOF-MS
The specifi cations of the orthogonal time-of-fl ight mass spectrometer (MDS Sciex, Concord, ON, Canada) have been outlined in detail in previous studies ( 64,68 ). Direct TLC-IR-MALDI-o-TOF-MS analysis was performed in situ from immunopositive GSL bands utilizing a Q-switched Er:YAG laser (Speser, = 2.94 In addition, traces of ions in the spectrum at m/z 1186.80 gave evidence for the presence of Gb3Cer (d18:1, C26:0) in EA.hy 926 cells not found in the upper band spectrum of HBMECs. A synopsis of the relative m/z intensities attributable to the different Gb3Cer lipoforms is presented in Table 1 .
Direct ionization of immunodetected GSLs on chromatograms allows in situ structural characterization by IR-MALDI-o-TOF-MS, omitting scratching of the silica gel from the plate and ensuing extraction of GSLs. Moreover, GSLs can be comparatively analyzed without any loss of analyte of particular interest in case of small GSL amounts, e.g., isolated from DRM fractions (see below). These are the major advantages of this approach compared with previous investigations, where the Stx receptors of HBMECs and EA.hy 926 cells were analyzed from pooled silica gel extracts of several TLC runs ( 52 ).

IR-MALDI-o-TOF-MS of Stx receptors in the neutral GSL fractions of total cells
The immunodetected Gb3Cer bands of total HBMECs and EA.hy 926 cells were analyzed by in situ IR-MALDI-o-TOF-MS directly on the TLC plate as shown in   Table 1 . EA.hy 926 cells, HUVEC descendant cell line; Gb3Cer, globotriaosylceramide; Gb4Cer, globotetraosylceramide; GSL, glycosphingolipid; HBMEC, human brain microvascular endothelial cell; IR-MALDI, infrared matrix-assisted laser desorption/ionization; o-TOF, orthogonal time-of-fl ight. thelial cell lines. Interestingly, the cholesterol content was found in average 1.4-fold higher in EA.hy 926 cells (calculated for DRM fractions 1 to 3).

Distribution of Gb3Cer and Gb4Cer to sucrose gradient fractions of HBMECs
The lipid extracts and protein preparations of the sucrose density gradient fractions of HBMECs were subjected to TLC immunodetection of Gb3Cer and Gb4Cer and Western blot analysis of caveolin-1 and fl otillin-2 ( Fig. 4 ). DRM fraction 2 exhibited the highest relative quantity of Gb3Cer (54%), followed by fractions 1 and 3 with 21% and 14%, respectively ( Fig. 4A , upper panel). Thus, 89% of Gb3Cer fl oated in the DRM fractions; the residual 7% were made up of soluble Gb3Cer, which distributed more or less equally to the nonDRM fractions 4 to 8, and 4% were found in the sediment fraction. The determination of the relative quantities of Gb3Cer with long-(upper band) to short-chain (lower band) fatty acids [i.e., of

Cholesterol and protein content of sucrose density gradient fractions
Lipids were extracted from 1.2 ml and proteins were prepared from 0.3 ml aliquots of the nine fractions (1.5 ml each), which were obtained from HBMEC and EA.hy 926 cell homogenates by sucrose density gradient centrifugation using 1% Triton X-100 (supplementary Fig. I).
The cholesterol and protein contents were calculated as micrograms per 1 × 10 6 cells. The distribution of cholesterol and protein to the gradient fractions of both cell lines is shown in Fig. 3 . The highest quantities of cholesterol were determined in the DRM fraction 2 and adjacent fractions 1 and 3 as expected ( Fig. 3A ). The intermediate fractions 4 to 7 and the bottom fraction 8 are characterized by very low cholesterol concentrations, whereas the sediment fraction 9 exhibited a slight enrichment compared with the nonDRM fractions. Notably, cell cholesterol contents of the DRM fraction 2 and contiguous fraction 3 from EA.hy 926 cells were 1.8-fold and 1.4-fold higher, respectively, and equal in fraction 1 compared with HBMECs ( Fig. 3A ). Compared with cholesterol, the protein distribution was inversed in the gradient fractions. Proteins increased in nonDRM fraction 7, and the bulk of proteins accumulated in the bottom fraction 8 and the sediment fraction 9 ( Fig. 3B ), whereas DRM fraction 2 and fl anking fraction 3 harbored only minor quantities of proteins and were almost indistinguishable from the intermediate fractions 4 to 6 with regard to the protein content. These gradient profi les demonstrate predominant distribution of microdomain-associated cholesterol to DRM fractions and preferential occurrence of soluble membrane proteins in nonDRM fractions in both endo-   compared with lower band Gb4Cer (d18:1, C16:0) was virtually identical in DRM fraction 2 and total cells, where Gb4Cer (d18:1, C22-C24) amounted to 57% and 59%, respectively ( Table 2 ). In conclusion, these data suggest that both Gb3Cer and Gb4Cer reside preferentially, but not exclusively, in the DRM fractions. The calculated mean values of Gb3Cer and Gb4Cer residing in DRM fractions from three individual experiments gave 83% (SD ± 5.0) and 54% (SD ± 15.1), respectively, indicating that DRM association was more pronounced for Gb3Cer, corresponding to a remarkably higher ratio of soluble Gb4Cer in the nonDRM fractions. Moreover, a subtle enrichment of Gb3Cer with long-chain fatty acids was detectable in the major DRM fraction 2 of HBMECs.
Western blot analysis of the dissemination of the caveolae marker protein caveolin-1 revealed low but exclusive occurrence of caveolin-1 in DRM fraction 2 ( Fig. 4B , upper panel). Interestingly, the higher level of caveolin-1 in the protein preparation of total cells indicated that only a minor quantity of caveolin-1 resided in membrane microdomains of DRM fraction 2. The lipid raft marker protein fl otillin-2 was found to preferentially partitioned, like caveolin-1, to DRM fraction 2, but on an apparently higher level than caveolin-1 compared tp fl otillin-2 in total cells ( Fig. 4B , lower panel). In summary, the distribution pattern of Gb3Cer and Gb4Cer TLC overlay assays and the Western blots of caveolin-1 and fl otillin-2 clearly evidenced colocalization of both GSLs with lipid raft marker proteins in DRM fractions. Furthermore, the data suggested a preferential occurrence of lipid rafts harboring fl otillin-2 and only minor existence of caveolin-1-containing caveolae in DRM fractions of HBMECs.

Distribution of Gb3Cer to sucrose gradient fractions of EA.hy 926 cells
TLC immunodetection of Gb3Cer in lipid extracts and Western blots of caveolin-1 and fl otillin-2 in protein preparations of the gradient fractions obtained from EA.hy 926 cells were performed as with HBMECs ( Fig. 5 ). Gb3Cer was found preferentially in DRM fraction 2 and adjacent fraction 3 amounting to 28% and 15%, respectively, accompanied by only 3% in fraction 1 ( Fig. 5A ). Thus, 46% of Gb3Cer fl oated in the DRM fractions. The remaining 32% were made up of soluble Gb3Cer, which distributed almost equally to the nonDRM fractions 4 to 8, and 22% were detected in the sediment fraction. The relative quantities of Gb3Cer (d18:1, C22-C26) to Gb3Cer (d18:1, C16) were almost identical in DRM fraction 2 and total cells where Gb3Cer (d18:1, C22-C26) constituted 70% and 67%, respectively, indicating a minor relative increase of Gb3Cer with long-chain fatty acids in DRM fraction 2 ( Table 2 ). In conclusion, Gb3Cer was found to reside preferentially, but not exclusively, in the DRM fractions. Importantly, EA.hy 926 cells exhibited only half of the fl otation rate of Gb3Cer in DRM fractions and 3-fold higher amounts of soluble Gb3Cer in the nonDRM fractions compared to HBMECs.
A similar gradient distribution pattern was observed for Gb4Cer with predominant occurrence in DRM fraction 2 (50%) followed by fractions 3 and 1 with 25% and 3%, respectively ( Fig. 4A , lower panel). Thus, 78% of Gb4Cer fl oated in the DRM fractions. The residual 21% constituted soluble Gb4Cer in the nonDRM fractions 4 to 8, and only 1% was detectable in the sediment fraction. The relative distribution of upper band Gb4Cer (d18:1, C22-C24) upper Gb3Cer positive band of total cells ( Fig. 2B , left  panel). The m/z values of 1146.67 acquired from the lower immunopositive Gb3Cer band gave evidence for Gb3Cer (d18:1, C16:0) as the only Gb3Cer variant with short-chain fatty acid in DRM fraction 2 ( Fig. 6B , right panel) as observed for the neutral GSL fraction of total EA.hy 926 cells ( Fig. 2B , right panel). A synopsis of the relative m/z intensities attributable to the different Gb3Cer lipoforms is presented in Table 1 .

Immunofl uorescence microscopic detection of Gb3Cer, caveolin-1, and fl otillin-2
The subcellular localization of Gb3Cer and DRMassociated marker proteins caveolin-1 and fl otillin-2 in HBMECs and EA.hy 926 cells was evaluated after Triton X-100 permeabilization ( Fig. 7 ). Antibody detection of Gb3Cer revealed a broad distribution in both endothelial cell lines, whereby HBMECs exhibited a more regular and homogenous distribution of Gb3Cer clusters compared to EA.hy 926 cells. The caveolae and lipid raft marker proteins, caveolin-1 and fl otillin-2, respectively, appeared distinguishable from each other in differently clustered structures and obviously did not localize in identical fl uorescence spots in HBMECs ( Fig. 7A ) or EA.hy 926 cells ( Fig. 7B ), indicating their subdistribution to different types of microdomains. Furthermore, flotillin-2 and Gb3Cer seemed to reside in more punctate structures than caveolin-1 in both types of endothelial cells. Higher-resolution imaging using confocal microscopy is required to address the question of colocalization of Gb3Cer with the two lipid raft marker proteins. However, it is noteworthy that fl otillin-2 exhibits a subcellular distribution different from that previously observed in kidney epithelium ( 69 ) and oligodendroglial cells ( 70 ).
To address the issue of surface localization of Gb3Cer, we produced immunofl uorescence micrographs of nonpermeabilized HBMECs and EA.hy 926 cells (supplementary Fig. II). HBMECs seem to harbor considerable amounts of Gb3Cer on the cell surface, whereas EA.hy 926 cells obviously express less abundant Gb3Cer molecules in and adjacent fraction 3, respectively ( Fig. 5B , upper  panel). Compared with the protein preparation of total cells, it was obvious that high amounts of caveolin-1 localized to membrane microdomains. Notably, fl otillin-2 was undetectable in DRM fraction 2 and in accompanying gradient fractions and appeared only in traces in the bottom fraction 8 ( Fig. 5B , lower panel). In summary, our results unequivocally evidenced colocalization of Gb3Cer with the caveolae marker protein caveolin-1 in DRM fractions and suggested the absence of fl otillin-2-harboring lipid rafts in EA.hy 926 cells.
Interestingly, in the DRM fraction 2 of EA.hy 926 cells, the dominant monosodiated ions at m/z 1158.80 related to Gb3Cer (d18:1, C24:0) were not accompanied by ions indicative of Gb3Cer (d18:1, C24:1) and Gb3Cer (d18:1, C22:0) ( Fig. 6B , left panel), which were clearly detectable as minor ions in the mass spectrum of total cell neutral GSLs ( Fig. 2B , left panel). Furthermore, ions at m/z 1186.83, which correspond to Gb3Cer (d18:1, C26:0), were found considerably enriched in the DRM fraction 2 ( Fig. 6B , left panel) but were almost undetectable in the Relative distribution of Gb3Cer and Gb4Cer lipoforms with long-chain (C22-C26, upper bands) and shortchain (C16, lower bands) fatty acids in double bands was determined by TLC scanning densitometry using GSL extracts of total endothelial cells and DRM F2 (see supplementary obvious effect observed was the considerably diminished cholesterol content in DRM fraction 2 and adjacent fractions 1 and 3 of HBMECs and EA.hy 296 cells, respectively ( Fig. 8A ) compared with untreated cells (see Fig. 3A ). The average reduction of cholesterol in DRM fraction 2 and adjacent fractions 1 and 3 of HBMECs after M ␤ CD-treatment was 55%, leaving 45% in intact DRMs. The simultaneous increase in the other gradient fractions, particularly in fractions 7 to 9, suggested a low release of approximately 25% of cholesterol, most likely into the environmental culture medium. In contrast, the average cholesterol content of DRM fraction 2 and fractions 1 and 3 of EA.hy 926 cells was reduced to 96%, indicating an almost complete M ␤ CDmediated destruction of cholesterol-containing microdomains, which harbored only 4% of the original cholesterol before M ␤ CD incubation ( Fig. 8A ). The slight concomitant increase in fraction 7 and the almost entire loss of cholesterol in the sediment fraction 9 pointed to a high release of approximately 87% of cholesterol, most probably into the cell culture medium as calculated from the total remaining cholesterol content of the gradient fractions. Importantly, the primary cholesterol content was 1.8-and 1.4-times higher in DRM fraction 2 and adjacent fraction 3 of EA.hy 926 cells, respectively, compared with HBMECs (see Fig.  3A ), whereas after M ␤ CD exposure, the remaining cholesterol in the corresponding fractions of HBMECs was found to be 6 and 7.5 times higher, respectively ( Fig. 8A ), indicating a greater loss of cholesterol in EA.hy 296 cells. These fi ndings clearly evidenced a pronounced refractiveness and thus a more stable molecular assembly of microdomains in HBMECs toward M ␤ CD-mediated cholesterol depletion compared to EA.hy 926 cells.
Determinations of the protein concentrations ( Fig. 8B ) revealed for both endothelial cell lines an overall diminishment in the DRM fraction 2 and contiguous fraction 3 and the intermediate gradient fractions 4 to 6, similar amounts in fraction 7, and reduction of the protein content in the bottom fraction 8 and the sediment fraction 9 after M ␤ CD-exposure (see Fig. 3B ). The lower protein quantity of the latter two fractions was more pronounced for EA.hy 926 cells. The total recovery rate of proteins in the gradient fractions after M ␤ CD-treatment was higher for HBMECs, and the calculated environmental release of membrane proteins of DRM fraction 2 and fl anking fraction 3 amounted on average of both fractions to 34% for HBMECs and 52% for EA.hy 926 cells. Thus, the higher segregation rate of cholesterol that was calculated for EA.hy 926 cells coincided with a concomitant enhanced loss of cholesterol-associated membrane proteins, giving further evidence for the existence of microdomains with higher resistance to M ␤ CD treatment in HBMECs.

Distribution of Gb3Cer and Gb4Cer to sucrose gradient fractions of HBMECs after M ␤ CD-mediated cholesterol depletion
To elucidate the infl uence of cholesterol depletion on the distribution of Gb3Cer and Gb4Cer as well as caveolin-1 and fl otillin-2 to sucrose gradient fractions of HBMECs, TLC overlay assays and Western blots were performed the plasma membrane. Owing to their localization to the inner leafl et of the plasma membrane bilayer, caveolin-1 and fl otillin-2 were undetectable in nonpermeabilized cells (data not shown). The lipid extracts and protein preparations of sucrose density gradient fractions from HBMECs and EA.hy 926 cells were examined as described above after cholesterol depletion of the cell cultures with 10 mM M ␤ CD. The most Gb4Cer distributed with 51%, 20%, and 1% to DRM fraction 2 and contiguous fractions 3 and 1, respectively ( Fig. 9A , lower panel). Thus, 72% of Gb4Cer fl oated in the DRM fractions. The residual 16% constituted soluble Gb4Cer in the nonDRM fractions 4 to 8, and 12% were present in the sediment fraction, indicating a slight relative increase of Gb4Cer in the sediment fraction but, in general, a similar pattern compared with untreated HBMECs (see Fig. 4A , lower panel). A maintenance level of 60% after cholesterol depletion was calculated for summed DRM fractions 1 to 3 in relation to the control immunostains on the same TLC plates. Interestingly, whereas the ratios of long-(upper band) and short-chain (lower band) fatty acid harboring Gb4Cer variants remained largely stable in total cells, a considerable rise of Gb4Cer (d18:1, C22-C24) from 57% to 73% could be determined in DRM fraction 2 after M ␤ CD-mediated cholesterol depletion ( Table 2 ). As a conclusion, treatment of HBMECs with after exposure to 10 mM M ␤ CD and gradient fractionation ( Fig. 9 ). Cholesterol depletion resulted in a pattern of Gb3Cer bands residing with 53%, 22%, and 3% in DRM fraction 2 and adjacent gradient fractions 3 and 1, respectively ( Fig. 9A , upper panel). Thus, 78% of Gb3Cer fl oated in the DRM fractions. The remaining 12% accounted for soluble Gb3Cer in the nonDRM fractions 4 to 8, and 10% were detected in the sediment fraction, indicating slight relative reduction of Gb3Cer in the DRM fractions compared with untreated HBMECs (see Fig. 4A , upper panel). Quantitation of summed fractions 1 to 3 in relation to the control immunostains on the same TLC plates revealed 94% maintenance of DRM-associated Gb3Cer after cholesterol depletion. The ratios of Gb3Cer species with long-(upper band) to short-chain (lower band) fatty acids upon M ␤ CD exposure were almost identical to those determined in DRM fraction 2 and total cells of untreated HBMECs ( Table 2 ). fractions. The remaining 30% accounted for soluble Gb3Cer in the nonDRM fractions 4 to 8, and 17% were attributable to the sediment fraction after cholesterol removal, evidencing similar relative distribution of Gb3Cer in the DRM fractions compared with EA.hy 926 cells without cholesterol depletion (see Fig. 5A ). Importantly, quantitation of summed fractions 1 to 3 in relation to the control immunostains on the same TLC plates revealed an extremely low maintenance of DRM associated Gb3Cer of only 9% after cholesterol depletion. The ratios of Gb3Cer species with long-(upper band) to short-chain (lower band) fatty acids upon M ␤ CD exposure were similar to those obtained from DRM fraction 2 and total cells of untreated EA.hy 926 cells ( Table 2 ). In conclusion, incubation with M ␤ CD resulted in a tremendous reduction of DRMs accompanied by an almost entire loss of Gb3Cer in the corresponding gradient fractions, suggesting an extremely labile GSL-cholesterol assembly in microdomains of EA.hy 926 cells.

Cholesterol and protein content of sucrose density gradient fractions after M ␤ CD-treatment
Despite the approximately 90% reduction in DRMs, the caveolae marker caveolin-1 was still clearly detectable in DRM fraction 2 along with a considerable increase of soluble caveolin-1 in the bottom fraction 8 after M ␤ CD treatment ( Fig. 10B , upper panel). Flotillin-2 was again undetectable in DRMs ( Fig. 10B , lower panels) as observed in gradient fractions obtained from cells without M ␤ CD exposure ( Fig. 5B , lower panel). Together the data evidenced an almost complete destruction of lipid rafts caused by cholesterol depletion, along with a 90% decline of Gb3Cer in the DRM fraction with rising soluble caveolin-1, which suggested a rather labile lipid assembly in microdomains of EA.hy 926 cells.

GSL expression of EA.hy 926 cells after medium change
Because culturing cells in different types of media (e.g., DMEM:F12 or RPMI 1640) may alter their biochemical M ␤ CD demonstrated high maintenance levels for both Gb3Cer and Gb4Cer, distinguishable by signifi cantly higher content of Gb3Cer in the DRM fractions pointing to a particularly stable association of Gb3Cer with remaining cholesterol in the DRM fractions.
The Western blot analysis of caveolae and lipid raft marker proteins revealed moderate appearance of caveolin-1 in DRM fraction 2, accompanied by enhanced content in the nonDRM fractions 7 and 8 as a consequence of M ␤ CD treatment ( Fig. 9B , upper panel) compared with untreated cells (see Fig. 4B , upper panel). In addition, a strong rise of fl otillin-2 was observed in DRM fraction 2, and mildly increased levels were found in gradient fraction 3 and sediment fraction 9 after cholesterol elimination ( Fig. 9B , lower panel) as the most impressive change compared with HBMECs, which were not exposed to M ␤ CD (see Fig. 4B , lower panel). Together the data suggested a partial destruction of lipid rafts caused by cholesterol depletion, indicated by decline of GSLs in the DRM fraction along with rising soluble caveolin-1 in nonDRM fractions, as well as an obvious dynamic replenishment of DRM fraction 2 by fl otillin-2 and its exceptionally fi rm association with the remaining cholesterol and GSLs.

Distribution of Gb3Cer to sucrose gradient fractions of EA.hy 926 cells after medium change
TLC immunodetection of Gb3Cer in lipid extracts and Western blots of caveolin-1 and fl otillin-2 in protein preparations of the gradient fractions after medium shift of EA. hy 926 cells are shown in Fig. 11 . Gb3Cer was prevalent in DRM fraction 2 and also present in accompanying fractions 1 and 3, composing 52% compared with the remaining gradient fractions with higher sucrose density ( Fig.  11A ). This result was comparable with data from EA.hy 926 cells grown in original DMEM:F12, where 46% of Gb3Cer distributed to DRM fraction 2 and the two adjacent fractions (see Fig. 5A ). nature ( 71 ), we cultivated EA.hy 926 cells in the same medium as HBMECs (RPMI 1640 medium) and investigated the possible infl uence of the cell culture medium on GSL expression, particularly on the inability of Gb4Cer biosynthesis. The medium shift from DMEM:F12 to RPMI 1640 did not cause any changes related to EA.hy 926 cell morphology, growth behavior, adhesiveness, or confl uence degree as demonstrated for cells grown in original DMEM:F12 and after 10 passages in RPMI 1640 medium (supplementary Fig. IV). Furthermore, triple experiments of EA.hy 926 cells propagated in the two different media revealed the same GSL profi le concerning Gb3Cer abundancy and lack of Gb4Cer (see supplementary Fig. V). The Gb3Cer profi les of EA.hy 926 cells propagated in original DMEM:F12 and after switching to RPMI 1640 were comparable (supplementary Fig. V-A, lanes c and d). Moreover, the lack of Gb4Cer expression was observed in both media (supplementary Fig.  V-B, lanes c and d), indicating this failure as an intrinsic feature of EA.hy 926 cells as previously reported ( 52 ).   Table 2 . B: Western blots. Protein preparations were separated by SDS-PAGE and transferred onto nitrocellulose membranes. Caveolin-1 and fl otillin-2 were detected with anti-caveolin-1 antibody (upper panel) and anti-fl otillin-2 antibody (lower panel). Control (C): 10 µg of human heart lysate and 20 µg of HeLa cell lysate for caveolin-1 and fl otillin-2 immunodetection, respectively; protein amounts of the sucrose gradient fractions 1-9 and total cells (T) correspond to 1 × 10 6 cells, respectively. Vertical white lines indicate areas of noncontiguous lanes assembled. DRM, detergent-resistant membrane; Gb3Cer, globotriaosylceramide; Gb4Cer, globotetraosylceramide; GSL, glycosphingolipid; HBMEC, human brain microvascular endothelial cell; M ␤ CD, methyl-␤ -cyclodextrin; S, sediment fraction.
Gb3Cer were detectable in all of the gradient fractions as deduced from the TLC overlay assay ( Fig. 12A ). The overall reduction in distribution of Gb3Cer to DRM fractions was the same as observed for EA.hy 926 cells cultivated in original DMEM:F12 followed by cholesterol depletion (see Fig. 10A ), which indicated same lability toward M ␤ CD treatment after the medium switch.
Caveolin-1 dominated in DRM fraction 2 and the gradient fractions with high sucrose density ( Fig. 12B , upper panel), while fl otillin-2 remained largely undetectable with the exception of the sediment fraction 9 ( Fig. 12B , lower panel), which was quite similar to the fi ndings in the original DMEM:F12 after cholesterol depletion (see Fig.  10 ). Together the data confi rmed a less stable lipid assembly in microdomains of EA.hy 926 cells, which was only marginally infl uenced by the cell culture medium.

DISCUSSION
Since it was fi rst described as a simple barrier between extra and intracellular compartments, different plasma membrane structural and composition models have been proposed ( 2 ). Over the last decade, several works have provided evidence that the plasma membrane is indeed more mosaic than fl uid ( 32 ). It is now widely accepted that the lateral organization of cellular membranes is formed by the clustering of specifi c lipids, such as cholesterol and GSLs, into highly condensed microdomains, and that the formation and disassembly of lipid raft domains is a dynamic process ( 1 ).
The strong appearance of caveolin-1 in the Western blots of DRM fraction 2 and adjacent fraction 3 ( Fig. 11B , upper panel) was consistent with former results of EA.hy 926 cells grown in DMEM:F12 (see Fig. 5B , upper panel). Flotillin-2 was detectable only in trace quantities in DRM fraction 2 and fraction 8 ( Fig. 11B , lower panel), comparable to the Western blot obtained from EA.hy 926 cells before the medium shift (see Fig. 5B , lower panel). In conclusion, the medium switch did not signifi cantly change the expression of either Gb3Cer or the two lipid raft marker proteins.   type-specifi c mechanisms of endothelial cell injury beyond inhibition of protein biosynthesis ( 73 ). Furthermore, one would expect a higher susceptibility of HBMECs toward Stx due to higher content of DRM-associated Stx receptors as shown in this study, but this is only the case for Stx2 ( 73 ). Thus, with the current knowledge regarding Stxreceptor content and plasma membrane localization as well as lipid raft association, we do not know the reasons for the observed differential cytotoxic action of Stx1 and Stx2. Further investigations are required using confocal laser microscopy, which has been employed successfully for subcellular routing of Stx and localization of fl otillin by Sandvig and coworkers ( 74,75 ).
However, there seems to be no question that Gb3Cer within DRMs, an index of GSL-cholesterol-enriched lipid rafts, is required for in vitro cytotoxicity ( 47,49,76,77 ) and may defi ne glomeruli-restricted pathology ( 78 ), whereas randomly distributed Gb3Cer in the nonordered liquid state may have a protective effect. The most plausible reason for this is that the internalized Stx B-subunit-Gb3Cer complex derived from soluble Gb3Cer is traffi cked to the lysosomes where the toxin is degraded, resulting in survival of the cells ( 47 ). It is well accepted that the presence of Gb3Cer within DRMs is central for the ensuing transport of the Gb3Cer/Stx agglomerate via endosomes and the trans-Golgi network to the endoplasmic reticulum. In addition, the translocation of the A-subunit from the endoplasmic reticulum into the cytosol depends largely on Gb3Cer assembly in DRMs ( 49 ). However, the molecular arrangement of soluble Gb3Cer and Gb4Cer in non-DRM fractions remains obscure and raises the question of whether the GSLs distribute as single molecules or whether a so-far unknown cholesterol-free association with soluble proteins might exist, especially in the liquid-disordered phase of the gradient bottom fraction containing the highest sucrose concentration.
A relevant specifi c enrichment of certain lipoforms (i.e., GSL variants with long-and short-chain fatty acids) in DRMs compared with total cell extracts could not be detected, although a subtle change in favor of Gb3Cer with C24 fatty acid was arguable in DRM fraction 2 of HBMECs as deduced from TLC overlay assays, accompanied by a small relative increase of monounsaturated C24:1 fatty detectable by in situ TLC-IR-MALDI-o-TOF-MS. Lingwood et al. proposed that the heterogeneity in fatty acid composition of GSLs may mediate aglycone (ceramide) regulation of GSL membrane receptor function by a differential interaction with cholesterol and other membrane components that may be differentially organized within plasma membrane lipid domains ( 79,80 ). These concepts have been illustrated in model membrane studies, particularly in relation to Gb3Cer-Stx-interaction. Importantly, mixtures containing Gb3Cer with C18 fatty acid were not recognized by Stx1, but the addition of the Gb3Cer variant with C24:1 fatty acid resulted in the generation of presentation and binding platforms for Stx1, whereas Stx2 bound Gb3Cer species largely irrespective of fatty acid chain length ( 81 ). Those studies indicated that fatty acid-mediated fl uidity within simple GSL and cholesterol-containing The tight packaging of cholesterol, GSLs, and phospholipids confers to lipid rafts their resistance to solubilization by nonionic detergents in the cold. Detergent resistance refl ects a stronger lateral interaction between membrane components, which allows their separation and isolation as DRMs from the rest of the plasma membrane using sucrose-density gradients ( 65,72 ). We found that Stx receptors preferentially localized to the liquid-ordered phases of DRM fractions of HBMECs and EA.hy 926 cells, where they colocalized with prevalent fl otillin-2 and caveolin-1, respectively. Note that the ratio of insoluble versus soluble Gb3Cer was higher in HBMECs, whereas there was a considerably lower Gb3Cer association with DRMs and a concomitant increase of soluble Gb3Cer in nonDRM fractions of EA.hy 926 cells. We previously described that EA.hy 926 cells exhibited a higher sensitivity to Stx1 and a higher mortality rate in vitro compared with HBMECs ( 52 ). A very recent study of our group on the direct comparison of both types of endothelial cells using quantitative apoptosis/necrosis assay ( 73 ) revealed different susceptibilities to Stx1 and Stx2: EA.hy 926 cells were slightly, but signifi cantly ( ‫ف‬ 10 times), more sensitive to Stx1, whereas HBMECs were strikingly ( у 1,000 times) more susceptible to Stx2. These fi ndings were quite surprising because Stx1 and Stx2 are known to exhibit identical receptor binding specifi city, and both EA.hy 926 cells and HBMECs express the high-affi nity receptor Gb3Cer in relevant quantities ( 52 ), suggesting the existence of yet-to-be delineated Stx A: TLC overlay assays of lipid extracts from gradient fractions 1-9 (F1-9) and of total cells (T) were performed as described in Fig. 10A . B: Western blots of SDS-PAGE separated protein preparations from gradient fractions 1 to 9 and total cells were performed as described in Fig. 10B . Vertical white lines indicate areas of noncontiguous lanes assembled. DRM, detergent-resistant membrane; EA.hy 926 cells, HUVEC descendant cell line; Gb3Cer, globotriaosylceramide; M ␤ CD, methyl-␤cyclodextrin; S, sediment fraction. tubular membrane invaginations which hint to a perplexing diversity of endocytic routes ( 84 ), have been performed with nonendothelial cells, and generalizations for endothelial cells are therefore limited.
Thus, despite the hypothetical functional role of GSL association with lipid rafts in endothelial cells, our study provides some useful information and the basis for future investigations to clarify the involvement of raft-associated GSLs in uptake and traffi cking of Stx to intracellular targets. Our study may contribute to a deeper understanding of Stx-mediated endothelial cell injury.
DRMs can selectively regulate GSL carbohydrate ligand binding. We found the classical TLC-separated Gb3Cer doublets in HBMECs and EA.hy 296 cells consisting of Gb3Cer (d18:1, C24:1/C24:0) in the faster and Gb3Cer (d18:1, C16:0) in the slower moving band in both cell lines but no Gb3Cer lipoforms with C18 and C20 fatty acids ( 52 ). Because the latter have been reported to fail in binding to Stx1 in Gb3Cer-cholesterol vesicles ( 81 ), this suggests involvement of at least Gb3Cer (d18:1, C24:1/C24:0), but the participation of Gb3Cer (d18:1, C16:0) in the recognition process of Stx1 and Stx2 in HBMECs and EA.hy 926 cells is unknown. We aim therefore to unravel in the near future the functional role of these DRM-associated Gb3Cer species to explain the differential cytotoxic actions of Stx1 and Stx2 toward the different types of endothelial cells ( 73 ).
As Stx cytotoxicity and retrograde traffi cking have been shown for HeLa and glomerular microvascular endothelial cell cultures to depend on Gb3Cer presence within DRMs ( 47,82 ), whereas randomly distributed Gb3Cer in the plasma membrane was found to mediate binding and subsequent traffi c to lysosomes for Stx degradation in HeLa and bovine epithelial cells ( 47,76 ), we investigated the destabilization of endothelial DRMs employing cholesterol-sequestering M ␤ CD. Cholesterol, as a dynamic "glue" holding lipid rafts together, was only incompletely removed from the plasma membrane of HBMECs by M ␤ CD treatment, giving evidence for highly refractive microdomains in the liquid-ordered membrane phase. The remaining and rather high cholesterol content of 45% in the DRMs correlated with only marginally diminished Gb3Cer retaining the same C24 to C16 fatty acid ratio compared with DRMs of untreated cells. In contrast to Gb3Cer, a disproportionately high loss of Gb4Cer accompanied by a simultaneous substantial loss of Gb4Cer with C16 fatty acid indicated less stable raft-insertion of Gb4Cer with the short-chain fatty acid as the most plausible explanation for high extraction effi cacy. In contrast to HBMECs, exposure to M ␤ CD resulted in almost completely abolished cholesterol content of 9% in macrovascular EA.hy 926 cells, in accordance with exceptionally low Gb3Cer content of 10%. We do not know the membrane constituents that might confer structural stability to lipid rafts of HBMECs or why microdomains of HBMECs and EA.hy 926 cells are distinct in terms of sensitivity toward cholesterol depletion. However, it is tempting to speculate that, for example, the observed replenishment of DRM fractions by fl otillin-2 may support the stability of lipid rafts through cholesterol removal, pointing to a dynamic process of membrane reconstitution by auxiliary membrane proteins. Moreover, besides unique GSL requirements for functionally intact lipid rafts ( 26,83 ), it is unclear whether GSL-GSL interaction may contribute to lipid raft solidity and thus to refractiveness through cholesterol depletion. In this context, it is noteworthy that an association of Gb3Cer with other GSLs or functionally different pools of Gb3Cer may be involved in lipid raft dynamics ( 48,49 ). However, it should be mentioned that these studies, including the impressive investigations on Stx-mediated induction of