In fl uence of Gb 3 glycosphingolipids di ff ering in their fatty acid chain on the phase behaviour of solid supported membranes : chemical syntheses and impact of Shiga toxin binding †

Institute of Organic and Biomolecular Tammannstr. 2, 37077 Göttingen, Germany Institute of Organic Chemistry, Technische 38106 Braunschweig, Germany. E-mail: d.w BIOSS – Centre for Biological Signalling Stu Schänzlestrabe 18, 79104 Freiburg, Ger uni-freiburg.de Institute of Biology II, Albert-Ludwigs-Univ Freiburg, Germany † Electronic supplementary information the synthesis and analytical data of 27–30 the preparation and analysis of solid sup of natural Gb3 and bovine brain SM; ad of phase-separated lipid bilayers; calcu receptor coverage; ternary phase diagram DOI: 10.1039/c4sc01290a ‡ These authors contributed equally to th Cite this: Chem. Sci., 2014, 5, 3104


Introduction
Glycosphingolipids are the cellular receptors of a number of viruses and bacterial toxins.For example, simian virus 40 and other polyoma viruses as well as the bacterial AB 5 protein Cholera toxin tightly attach to the plasma membrane via the ganglioside G M1 resulting in plasma membrane invaginations. 1,2Shiga toxin, an AB 5 protein produced by Shigella dysenteriae and by enterohemorrhagic strains of Escherichia coli (EHEC) binds via its homopentameric B-subunits (STxB) to the neutral globoside Gb 3 . 3,4This process is the initial step for Shiga toxin to efficiently enter cells.Aer endocytosis, Shiga toxin leaves the endocytic pathway at the level of early endosomes following the retrograde transport route. 5Römer et al. 6 showed that the pentameric B-subunits, which can bind up to 15 Gb 3 molecules, induce a clathrin-and caveolin-independent tubule formation upon attachment to HeLa-cells and that invaginations were even found in articial membranes, when STxB binds to giant unilamellar vesicles (GUVs).These results suggest a signicant lipid contribution to the earliest steps of toxin uptake into cells and there are several lines of evidence that the cytotoxicity of Shiga toxin is strongly associated with Gb 3 density and binding to cholesterol-containing lipid ras. 5,7,8ollowing this hypothesis, it has been demonstrated that STxB binding leads to an increased lipid order in cellular membranes. 6Results obtained from articial membranes on solid substrates as well as from pore-spanning membranes further showed that STxB is capable of reorganizing lipids within these membranes concomitant with protein cluster formation. 9,10This process can induce a protein induced local compaction of particular lipids in one leaet leading to an asymmetric reduction in membrane area, which is prerequisite for membrane invaginations. 11n all these studies, mixtures of Gb 3 from natural sources have been used, which are composed of Gb 3 molecules with different fatty acid compositions.Even though the Gb 3 -concentration is only a few mol% in these membranes, there is strong evidence that the fatty acid composition of Gb 3 signicantly inuences the process of Shiga toxin binding, 12 membrane organization, 9,13 invagination, 6 and scission. 14Lingwood and coworkers found that Gb 3 s with long fatty acid chains (C 20 -C 24 ) bind Shiga toxin with higher affinity than those with short fatty acid chains (C 12 , C 14 ). 15Also, the a-hydroxylation of the fatty acid of Gb 3 gave rise to a larger binding affinity and larger binding capacity of STxB compared to the non-hydroxylated fatty acid. 137][18] However, as yet there is no systematic study available addressing the impact of the fatty acid composition of Gb 3 on the organization of lipid membranes that partially mimic the composition of the outer leaet of animal plasma membranes.This arises from the fact that, apart from some reports on semisynthetic approaches to generate Gb 3 molecules, 15,19 chemical access to these highly complex glycolipids is a challenging and time-consuming endeavour.Nevertheless, such a synthetic approach is the method of choice to gain chemically pure material which is needed for proper biophysical investigations.In 2007, Florent and co-workers already devised a synthetic route from the trisaccharidic subunit to Gb 3 -C22:0 and Gb 3 -C22:1 glycosphingolipids. 6n this paper, we describe a generally applicable and modular strategy to assemble the target glycosphingolipids starting from simple monosaccharide building blocks, sphingosine and fatty acids.Saturated, unsaturated, a-hydroxylated fatty acids and a combination thereof were employed.These molecules were reconstituted into membranes composed of sphingomyelin, cholesterol, and 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) resembling the composition of the outer leaet of animal plasma membranes and that are known to phase-separate. 20,21Solid supported membranes (SSMs) in conjunction with uorescence microscopy and atomic force microscopy were used to investigate the lateral organization of the membranes as a function of Gb 3 composition and STxB binding.Despite a reduced lateral lipid mobility found in SSMs, these membranes are sufficiently dynamic to allow for lipid rearrangement while offering the possibility to use uorescence microscopy and high resolution atomic force microscopy (AFM) to monitor nanometre-sized domains. 22,23We provide evidence that the fatty acid strongly inuences the phase behaviour of the membranes as well as the mode of STxB cluster formation.

Synthesis of Gb 3 glycosphingolipids
The power of a synthetic approach towards Gb 3 glycosphingolipids is most promising due to the highly dened nature of the obtained material.From a retrosynthetic point of view a glycosphingolipid can be divided into three major parts: (i) the carbohydrate head group as the hydrophilic and the ceramide as the lipophilic part that can be further subdivided into (ii) a sphingosine such as D-erythro-sphingosine and (iii) a fatty acid being attached to the amine of the sphingosine (Scheme 1).We assembled four different glycosphingolipids, which consist of the globotriaose head group, the sphingosine, but differ in the fatty acid chain.Saturated, unsaturated, a-hydroxylated derivatives and a combination thereof were prepared, all based on a C 24 fatty acid.C 24 fatty acids have been chosen as they are the major constituent (>50%, ESI †) found in natural Gb 3 mixtures such as toxin insensitive erythrocytes, 14 HeLa-cells, 24 and HEp-2 cells. 25he globotriaoside is traced back to three different monosaccharide building blocks (Scheme 1, 1-3).Glucose derivative 1 was obtained in four steps starting from 3,6-di-O-benzyl-D-glucal 26 being an ideal precursor for various glucose-derived building blocks.Fmoc protection of the 4-O-hydroxyl group followed by epoxidation with dimethyldioxirane (DMDO) 27 and epoxide opening using benzylic alcohol led to a glucose derivative with a free hydroxyl group in C-2 position.Acetylation and Fmoc removal provided glucosyl acceptor 1 (ESI †).Galactosyl building block 2 was synthesized starting from 3,6-di-O-benzyl-starting from 10 as well (Scheme 2).To facilitate the formation of the a-galactosidic linkage in the trisaccharide a remoteparticipating group (Bz) 30 was installed at the 4-hydroxyl of 10.DMDO-mediated epoxidation of galactal 11 and attack with p-methoxyphenol (MPOH) led to acetal 12 in 90% yield.Benzyl ether formation and removal of the anomeric protecting group under oxidative conditions furnished hemiacetal 13.
The latter was rst converted into a trichloroacetimidate 31 and subsequently transformed to phosphate 3 which proved to be more suitable in this glycosylation reaction than the trichloroacetimidate analogue.
With these monosaccharide building blocks in hand we started the assembly of the globotriaose.The sequence commences with the union of building blocks 1 and 2 using TMSOTf as Lewis acidic promoter (Scheme 3).The b-selectivity is ensured due to neighbouring group participation of the pivaloate of galactose building block 2. Subsequent removal of the Fmoc protecting group under mild basic conditions followed by the second glycosylation with building block 3 led to trisaccharide 14 in 79% yield (over three steps) and pure aselectivity as it was anticipated by the remote participation 30 of the benzoyl protecting group at C-4.In addition, a reaction temperature of À60 C proved to be benecial as well to achieve complete diastereoselectivity.In case of the more reactive perbenzylated galactosyl phosphate both, the yield and the selectivity, dramatically dropped down.
The commonly used technique to deprotect the permanent benzyl protecting groups by hydrogenolysis at the very end of an oligosaccharide synthesis is not a choice in the assembly of glycosphingolipids since their double bond would be destroyed during such a reduction step.Thus, the protecting groups of the globotriaose unit had to be exchanged before the union with the lipid part takes place.Taking these difficulties into account, benzoyl groups being easily cleaved under basic conditions were installed.In order to get perbenzoylated material, rstly, the three different acyl groups were removed under strong basic conditions, followed by a hydrogenolysis of the benzyl ethers using Pearlman's catalyst.Perbenzoylation of the naked trisaccharide was performed followed by removal of the anomeric benzoate using methylamine.Finally, a trichloroacetimidate moiety was installed as leaving group to yield globotriaoside derivative 17 in 99% yield.
In addition, two hydroxylated fatty acids being not commercially available had to be prepared.Based on a route developed by Kiso, Ando and co-workers, a-hydroxylated tetracosanoic acid 22 was generated (Scheme 4a). 32Alkyne 18, readily available from 1-eicosanol, was used to open epoxide 19.The secondary hydroxyl group was directly protected with a benzoyl group to afford 20 in 45% yield over two steps.Hydrogenation using palladium on charcoal furnished the saturated debenzylated alkane chain whereas Pearlman's catalyst induced an acyl group migration.Finally, the primary hydroxyl group of 21 was oxidized to fatty acid 22 in 73% yield using TEMPO/TCCA conditions.The protected a-hydroxylated nervonic acid 24 was synthesized starting from nervonic acid in two steps (Scheme 4b).Aer carboxylate enolate formation using freshly prepared LDA, a dry oxygen atmosphere attached to the reaction, derivatised the a-position with a hydroxyl group to yield 9 in 56%. 33inal benzoylation of this secondary hydroxyl gave the desired fatty acid 24.To complete total synthesis, globotriaosyl trichloroacetimidate 17 and the protected sphingosine 4 or 5, 34 respectively, were coupled in a glycosylation reaction using BF 3 $OEt 2 to afford 25 or 26, in moderate yields (Scheme 5).In case of the benzoyl protected sphingosine only a minor acyl group migration was observed.Staudinger reduction of the azide and direct coupling with the fatty acids 6-9 under peptide coupling conditions (e.g.HATU or HOBt/EDCI) 35 led to four protected glycosphingolipids, varying in their acyl chains.Global deprotection under Zemplén conditions resulted in the desired glycosphingolipids 27-30.

Phase behaviour of solid supported membranes doped with different Gb 3 s
SSMs that separate into a liquid-ordered (l o ) and liquid-disordered (l d ) phase at room temperature were obtained by spreading unilamellar vesicles composed of the well-established lipid mixture DOPC/sphingomyelin (SM)/cholesterol (Chol) in a molar ratio of 2 : 2 : 1 on mica surfaces. 36esicle spreading was achieved at elevated temperatures (55 C) above the phase-transition temperature of the lipid mixture (T m z 40 C). 37At room temperature, the membrane is phaseseparated into a l o and a l d phase, as can be revealed from uorescence micrographs (ESI, Fig. S1A †) using sterically demanding uorophors (Bodipy-PC, Texas Red DHPE, Oregon Green DHPE) that preferentially partition into the l d phase. 38he overall membrane integrity was veried for each experiment using the dye perylene, which distributes more equally between the l o and l d phase (ESI, Fig. S1B †).The percentage of the l o phase was determined by pixel analysis to be 66 AE 6% (n ¼ 35).To this lipid mixture, the globotriaosylceramide Gb 3 serving as the receptor lipid for STxB was added by replacing 5 mol% of the used bovine brain SM (ESI, Table S2 †).This exchange retains the principal l o /l d phase-separation of the membranes. 9To scrutinize the inuence of the fatty acid of Gb 3 on the phase behaviour of the SSMs, one of the four synthesized Gb 3 molecules (27-30) were incorporated into the phase-separated membranes.Besides the Gb 3 with the C24:0 fatty acid (27, Gb 3 -C24:0), we analysed the inuence of the a-hydroxylated fatty acid (29, Gb 3 -C24:0-2OH), an unsaturated fatty acid (28, Gb 3 -C24:1), as well as both, an unsaturated and a-hydroxylated fatty acid (30, Gb 3 -C24:1-2OH).
Characteristic uorescence images of a solid supported membrane composed of DOPC/SM/Chol/Gb 3 (40 : 35   with Gb 3 -C24:0 are shown in Fig. 1A and B. The uorescence of Texas Red DHPE (false coloured in green) indicates the l d phase, while the dark areas are attributed to the l o phase, which excludes the uorescent lipid.The area percentage of the l o phase was determined by pixel analysis and amounts to 74 AE 7% (n ¼ 84).By means of atomic force microscopy (AFM), the topography of the phase-separated membranes was analysed (Fig. 1C).Domains of lower height are visualized in the AFM image, which are assigned to the l d phase (darker colour coding) embedded in a higher l o phase.The height difference between the l o and l d phase was determined by histogram analysis (Fig. 1D) to be Dh(l o /l d ) ¼ 0.6 AE 0.2 nm (n ¼ 21) in agreement with height differences reported in literature. 39or membranes containing 5 mol% of Gb 3 -C24:0-2OH (ESI, Fig. S2 †), again a phase-separated lipid membrane was visualized in uorescence microscopy as well as in atomic force microscopy, from which the l o and l d phase could be unambiguously assigned with a typical height difference of Dh(l o /l d ) ¼ 0.8 AE 0.1 nm (n ¼ 87).However, the area percentage of the l o phase was, with 56 AE 5% (n ¼ 56), signicantly lower compared to the lipid mixture containing 5 mol% of Gb 3 -C24:0.Given the fact that only the 5 mol% component Gb 3 was replaced in the lipid mixture, while all other lipid components remained constant, a change of about 20% of the area percentage of the l o phase was rather unexpected suggesting a more global impact of the Gb 3 molecules on the phase-separation behaviour of the membrane as discussed in more detail below.
Surprisingly different phase behaviour was observed if the lipid mixture contained Gb 3 -C24:1.The uorescence micrographs of Texas Red DHPE show three uorescence intensities (Fig. 2A and B).While a bright green phase is still clearly visible and can be identied by the enrichment of the uorescent dye, a dark phase as well as a phase with an intermediate brightness is observed.We assign the membrane phase, from which Texas Red DHPE is mostly excluded to the l o phase, while the phase with the highest uorescence intensity is attributed to the l d phase.The phase of intermediate brightness is termed l i phase and was found independently of the l d marker dye used (ESI †).According to this assignment, one can determine the area percentages of all three phases (n ¼ 73) being for the l d phase 23 AE 6%, for the l i phase 36 AE 8% and for the l o phase 40 AE 6%.By atomic force microscopy, only in rare cases the three phases were resolved by their differences in height (Fig. 2C).In case of the detection of three different height levels we assigned the lowest one to the l d phase.As it is not possible to assign the l i phase and the l o phase to the different heights in the AFM images, we took the l d phase as the reference and determined two height differences with Dh 1 ¼ 1.2 AE 0.2 nm (n ¼ 4) and Dh 2 ¼ 1.6 AE 0.2 nm (n ¼ 5).In several AFM images, the two different heights could not be resolved resulting in an averaged value of Dh For Gb 3 -C24:1-2OH, which contains the cis-double bond as well as the hydroxylation in the 2-position, the membrane again only separated into two phases assigned to as l d and l o with an area percentage of 70 AE 7% (n ¼ 51) resembling the situation of Gb 3 -C24:0 more than the other Gb 3 species.Table 1 summarizes the area percentages found for the different Gb 3 molecules as well as the height differences between the observed lipid phases as deduced from AFM analysis in the otherwise identical lipid mixtures.
Our ndings demonstrate that the fatty acid composition of the glycosphingolipid Gb 3 strongly inuences the phase behaviour of the phase-separated lipid membranes, even though it amounts only to 5 mol% of the whole lipid   composition.This result implies that Gb 3 strongly interacts with the other lipid components thus altering the phase behaviour.
STxB binding to solid supported membranes doped with different Gb 3 s We thus elucidated the impact of STxB binding on the phase behaviour of the lipid mixture containing different Gb 3 species by using uorescently labelled STxB.In For Gb 3 -C24:0 containing membranes, the Oregon Green DHPE micrographs show phase-separated membranes with l o phase and l d phase aer STxB binding (Fig. 3A).The corresponding Cy3-uorescence image of the bound STxB shows inverse uorescence, indicating that the protein has been bound to the l o phase, in which the Gb 3 molecules are located (Fig. 3B).The area percentage of the l o phase does not change upon protein binding (76 AE 11%, n ¼ 52).From the STxB-Cy3 uorescence image as well as from the AFM images, it becomes obvious that the protein does not homogeneously bind to the l o phase but forms clusters.The height difference between the protein clusters on the l o phase and the overall height of protein-decorated l o -phase is Dh(l o+STxB_clusters /l o+STxB ) ¼ 0.8 AE 0.1 nm (n ¼ 6).The overall increase in height of the l o phase in the presence of bound protein is only very minor with an increase of Dh(l o+STxB /l d ) À Dh(l o /l d ) ¼ 0.1 AE 0.4 nm indicating that the protein surface coverage on the entire l o phase is, except for the dense protein domains, rather low.
Phase-separated membranes containing Gb 3 -C24:0-2OH with considerably reduced area percentage of the l o phase prior to protein binding of only 56 AE 5%, experience a signicant increase in l o phase aer STxB binding (77 AE 5%, n ¼ 62).The STxB-Cy3 uorescence is inhomogeneous (ESI, Fig. S3 †)  The disordered phase is assigned to the bright green uorescent phase, the intermediate phase to that with the lower uorescence intensity and the dark one to the ordered phase.
Fig. 4B shows the localization of Cy3-labeled STxB.Interestingly, the protein only binds to one of the three phases, namely to the assigned l o phase.This in turn would imply that aer STxB binding, the Gb 3 receptors are located only in the l o phase.The area percentages of the phases do not change upon STxB binding being 43 AE 7% and 31 AE 8% for the l o and l i phase, respectively.However, in contrast to the other Gb 3 containing membranes, the domain morphology changes considerably, i.e. all domains condense into larger ones.The height difference of Dh(l i /l d ) ¼ 1.3 AE 0.7 nm (n ¼ 12) corresponds to the height difference between the l d phase and the non-protein covered l i phase.The rather large difference between the l d and l i phase might be interpreted in a sense that the l i phase is highly ordered, composed of mostly saturated lipids.The protein bound to the Gb 3 containing l o phase results in a height difference of Dh(l o+STxB /l d ) ¼ 2.6 AE 0.6 nm (n ¼ 26).
For membranes containing Gb 3 -C24:1-2OH, two phases are identied with the l o phase being visualized by the red STxB-Cy3 uorescence (Fig. 5A and B).Of note, in accordance with the observation made for the saturated, a-hydroxylated compound Gb 3 -C24:0-2OH, the area fraction of the l o phase is signicantly increased from 70 AE 7% to 87 AE 7% (n ¼ 42).
This result implies that more DOPC molecules are recruited into the lipid dye excluding phase.In contrast to the other Gb 3 derivatives under investigation, the topography mapped by AFM (Fig. 5C) only shows two height levels with a height difference of Dh(l o+STxB /l d ) ¼ 2.7 AE 0.5 nm (n ¼ 44).
The higher phase is the l o phase with bound protein as deduced from the uorescence images.No individual proteins can be resolved, however the increase in height as a result of protein binding amounts to h STxB ¼ Dh(l o+STxB /l d ) À Dh(l o /l d ) ¼ 2.2 AE 0.7 nm, which indicates a high protein density on the surface resulting in a thickness that matches the crystallographic dimensions of STxB. 41able 2 summarizes the properties observed in uorescence microscopy and atomic force microscopy aer STxB binding to the membranes.

Discussion
There are several lines of evidence that membrane bending leading to the observed Shiga toxin invaginations in cells and GUVs are preceded by a local compaction of lipids in one leaet leading to an asymmetric reduction in membrane area. 9,11Such local lipid compaction is accompanied by protein cluster formation and lipid sorting. 9,40On the other hand, protein binding studies suggest that the binding capacity of a   (40 : 35 : 20 : 5) bilayers doped with different Gb 3 s on mica and height differences observed in AFM images after STxB binding.The ratios were determined by pixel analysis using fluorescence images.Height differences were extracted by fitting Gaussians to the height histograms (number of analysed fluorescence images and height histograms is given in brackets) membrane for STxB depends on the fatty acid of Gb 3 . 15,19,41hus, we asked the question whether protein cluster formation and lipid sorting leading to local lipid compaction is altered by the fatty acid composition of Gb 3 , which would then have great impact on the propensity of a membrane to invaginate as a result of STxB binding.We will rst discuss the individual observations made with the four synthetically derived Gb 3 molecules, which sum up to more than 50% of the Gb 3 s found in natural mixtures, 14,24,25 and then discuss the impact on STxB cluster formation and toxin invasion.
In case of Gb 3 -C24:0 with a long chain saturated fatty acid, the assigned l o phase amounts to 74% of the total membrane area.Incubation with STxB did not alter this percentage indicating that no gross lipid reorganization between the two phases takes place.We suggest that the Gb 3 molecules are preferentially localized in l o domains with the saturated fatty acid chain interacting with cholesterol resulting in a tightly packed l o phase before as well as aer STxB binding (Fig. 6).
The lipid mixture containing Gb 3 -C24:0-2OH exhibits about 20% less ordered phase (56% area fraction of l o phase) under the same experimental conditions before protein addition and thus suggests that the l o phase is mainly composed of the sphingolipids and cholesterol, which would sum up to a theoretical area fraction of about 50%, 42 if one takes only the geometry of the lipid molecules into account (ESI †).However, aer STxB binding the area percentage of the l o phase increases signicantly to 77%.It has been shown by others that acyl chain hydroxylation only slightly changes the transition temperature of ternary lipid mixtures but affects the interlipid interaction, e.g.attenuates the interaction with sterols leading to a uidization of the bilayer. 16This implies that part of the DOPC molecules has been relocated from the l d to the l o phase upon protein binding.In a similar fashion, the area percentage of membranes containing Gb 3 -C24:1-2OH increases by roughly 20% upon protein binding.Scheve et al. 43 reported that, owing to the sterical demand of a protein to bind to its receptor, it induces an increased lipid mixing eventually leading to the disruption of l o /l d phase separation, a phenomenon that could also explain the gross relocalization of DOPC from the l o to the l d phase upon STxB binding.
Unsaturated Gb 3 induces a visible demixing of the ordered phase, which we termed l i and l o phase according to their uorescence intensities in the uorescence images.Just recently, Sodt et al. 44 reported on substructures within the liquid-ordered phase of lipid bilayers as obtained by molecular dynamics simulations.These substructures, identied in a 10 ms all-atom trajectory of l o /l d coexistence were composed of saturated hydrocarbon chains packed with local hexagonal order and separated by interstitial regions enriched in cholesterol and unsaturated chains.In our case, the unsaturated Gb 3 molecules might foster such demixing of the l o phase interacting preferentially with the unsaturated species of bovine brain sphingomyelin (ESI, Table S2 †) and cholesterol and separate from the saturated species of bovine brain sphingomyelin. 45ven though the molecules are unsaturated, it is expected that they still exclude the bulky uorophore, as it has been shown in monolayer experiments that ceramides with an unsaturated fatty acid tightly pack despite their large steric demand. 46As there is not only a difference in uorophore partition between the two observed phases but also a height difference as obtained by AFM images, we suggest that Gb 3 -C24:1 is localized only in one of the two phases thus increasing the height owing to the size of the Gb 3 head group. 47We cannot unambiguously decide whether the l i or l o phase in the uorescence micrographs contains the Gb 3 , as the area percentage of the domains are similar but given the percentage of l d phase (24% for Gb 3 -C24:1 compared to 26% for Gb 3 -C24:0) we assume the two phases to be the result of segregation of the l o phase.
Using the phase diagram and tie-lines gathered from uorescence micrographs of GUVs reported by Bezlyepkina et al., 42 one can estimate the compositions and area fractions of the l o and l d phases.In general, the area fraction we found for the l d phase in our SSMs composed of DOPC/SM/cholesterol (2 : 2 : 1) is lower than that found for GUVs (63%), which probably reects the inuence of the solid support (ESI †).Despite this general trend, we found that independent of whether we theoretically replace either 5 mol% of SM or DOPC with 5 mol% of unsaturated Gb 3 in the DOPC/SM/cholesterol (2 : 2 : 1) lipid mixture, this does not explain the observed changes in l d area fraction suggesting that the Gb 3 inuences also the surrounding lipids (ESI, Fig. S5 †).We expect to observe the same behaviour when using 1-palmitoyl-2-oleoyl-sn-glycero-3phosphocholine bearing one saturated and one unsaturated fatty acid chain instead of DOPC as the general phase behaviour is very similar, 37,48 even though POPC can stabilize nanoscopic domains. 49er STxB binding, a change in lateral membrane organization as well as protein cluster formation was expected.However, in contrast to our previous study, 9 where we used a Gb 3 mixture, protein cluster formation as well as membrane reorganization strongly depended on the used Gb 3 molecule.We found that STxB binds to the l o phase demonstrating the localization of the different Gb 3 s with the sphingomyelin/ cholesterol enriched phases regardless of the fatty acid attached to the molecule. 40,50Membranes containing Gb 3 with a saturated fatty acid show densely packed protein clusters surrounded by less compact bound STxB.In contrast, STxB binds rather homogeneously and with high density to the l o phase of membranes containing Gb 3 with an unsaturated fatty acid.This result might be correlated with the nding that the membrane binding capacity for STxB is larger in case of unsaturated Gb 3 s. 15,19Given the affinity constants of STxB to Gb 3 , all binding sites on the membrane are expected to be occupied. 13However, the number of Gb 3 molecules bound to a STxB pentamer might vary for the different Gb 3 -doped membranes.From the crystal structure of STxB co-crystallized with a soluble receptor analogue, it is known that up to 15 Gb 3 molecules can bind to the pentamer resulting in dense packing of the receptor lipids. 41,51If a large number of Gb 3 s is bound to the STxB pentamer, the membrane gets depleted in Gb 3 resulting in localized protein clusters rather than in a homogeneously covered membrane surface.As a lower limiting case, we assumed that all 15 binding sites of STxB are occupied with Gb 3 , which would result in a theoretical protein surface coverage of roughly 16% (ESI †).From the AFM images (Fig. 3C), we estimated a 15% surface coverage for the STxB clusters on Gb 3 -C24:0 containing membranes, which is in a similar range as the theoretically estimated value.However, as there are also proteins bound to the membrane but not in clusters, one can state that the number of Gb 3 molecules bound per STxB is less than 15 but probably very high.Receptor clustering as a result of toxin binding was also described for Cholera toxin subunit-B resulting in less protein surface coverage at higher G M1 concentrations. 52In contrast, the unsaturation of the fatty acid in Gb 3 increases the area demand of this molecule, which might decrease the number of Gb 3 s bound to a pentamer.In this case, we indeed observed a more homogeneous but still high protein surface coverage on the l o phase.Taking full protein coverage into account would theoretically lead to about 6-10 Gb 3 -C24:1 molecules and about 3-5 Gb 3 -C24:1-2OH molecules bound per pentamer (ESI †), which is still a sufficiently large number to strongly bind the protein onto the membrane taking into account that the high affinity binding sites of the protein are occupied rst. 51Of note, there are indications that Gb 3 might be buried as a result of glycosphingolipid specic interactions with cholesterol, an aspect that we cannot exclude as all membranes contained cholesterol. 12,53,54ur rough calculations suggest that the number of recruited Gb 3 molecules to the STxB pentamer is greatly inuenced by the fatty acid of the Gb 3 molecule.This difference in lipid recruitment also inuences the membrane organization more globally.In case of membranes containing Gb 3 with an a-hydroxylated fatty acid, a signicant protein-induced lipid resorting is observed resulting in a signicantly increased l o phase area fraction (21% for Gb 3 -C24:0-2OH, 17% for Gb 3 -C24:1-2OH).However, neither for Gb 3 -C24:1 nor for Gb 3 -C24:0 containing membranes a substantial increase in l o phase area fraction was observed.
From our results we conclude that the unsaturated Gb 3 species are the most important ones for the rst step of Shiga toxin internalization.Even though they have a larger area demand than the saturated species, they still get clustered underneath the protein.This in turn would lead to an asymmetric reduction in membrane area, which is proposed to be prerequisite for membrane invaginations.This idea can be supported by the observation that in case of Gb 3 -C24:1 the l o phase gets reorganized and further phase-separated leading to a STxB decorated phase that contains more unsaturated lipid species, which would imply a lower bending modulus. 55Reorganization of l o phase lipid components have also been described by Safouane et al., who found a protein-induced cosorting of l o phase lipids into membrane tubes. 40In contrast, the saturated species form very compact protein clusters, which recruit most of the Gb 3 molecules.However, these structures might be too rigid to allow for an energetically favourable bending of the membrane.The hydroxylated Gb 3 species foster large membrane reorganization, which might be of importance in the following processes once Shiga toxin has generated invaginations such as scission of the membrane 6 and lipid sorting upon transport along the retrograde route.

Conclusions
Prerequisite for understanding the impact of individual Gb 3 species on the phase behaviour of lipid membranes and their ability to bind STxB is a chemical access to pure Gb 3 s.We developed a modular chemical approach that can be adapted to other glycosphingolipids, starting with three monosaccharide building blocks, sphingosine and different fatty acids.Chemical synthesis resulted in four Gb 3 species, which constitute the major fraction of Gb 3 in different cell types in nature.Reconstituted into phase-separated lipid membranes, these different Gb 3 molecules turned out to signicantly inuence the overall membrane organization even though the bilayers contained only 5 mol% of the globoside.
Protein cluster formation and lipid redistribution are discussed to be of major importance in the rst step of the internalization process of Shiga toxin and Shiga-like toxins giving rise to the infection of cells.From our results we conclude that the unsaturated Gb 3 species are the most important ones for the rst step of Shiga toxin internalization as binding of STxB leads to protein clusters concomitant with a large area demand of the lipids and probably a lower bending modulus owing to the unsaturation compared to saturated ones.This would lead to an asymmetric reduction in membrane area, which is proposed to be prerequisite for membrane invaginations induced by Shiga toxin.

Scheme 2 Scheme 3
Scheme 2 Preparation of a-galactosyl building block 3.

Fig. 2
Fig. 2 SSMs composed of DOPC/SM/Chol/Gb 3 -C24:1 (40 : 35 : 20 : 5) doped with perylene (0.5 mol%) and Texas Red DHPE (0.1 mol%).(A) Fluorescence micrograph of Texas Red DHPE (false coloured in green).Membranes segregate into three phases as indicated by three brightness levels.Scale bar: 10 mm.(B) Magnification of the region marked in (A).Additionally to liquid-ordered (l o , dark) and liquid-disordered domains (l d , bright green) a third lipid phase of intermediate brightness (l i ) is identified.Scale bar: 4 mm.(C) AFM image showing the three phases.While the lowest phase is assigned to the l d phase, the two other phases differing in height cannot be unambiguously assigned to the l o and l i phase.Scale bar: 2 mm.(D) Histogram analysis showing the heights of the three phases.Taking the l d phase as the lowest phase, two height differences of Dh 1 ¼ 0.9 AE 0.4 nm and Dh 2 ¼ 1.8 AE 0.4 nm are calculated for the image shown in (C).

Table 1
Ratios of the different membrane phases (l o : liquid-ordered, l i : liquid-intermediate) in DOPC/SM/Chol/Gb 3 (40 : 35 : 20 : 5) bilayers doped with different Gb 3 s on mica and height differences observed in AFM images.The ratios were determined by pixel analysis using fluorescence images.Height differences were extracted by fitting Gaussians to the height histograms.For compound 27, 29, and 30 Dh 2 corresponds to the height difference between l o and l d phase (Dh(l o /l d )) (number of analysed fluorescence images and height histograms is given in brackets) l o /l d phase-separated membranes containing a mixture of different Gb 3 species, it has been shown that the Gb 3 s are located in the l o phase aer STxB binding.9The position of the protein uorescence therefore allows us to localize the Gb 3 molecules aer protein binding.Aer preparation of SSMs composed of DOPC/SM/Chol/Gb 3 (40 : 35 : 20 : 5) on mica containing one of the four different Gb 3 species, they were incubated with 60 nM of STxB for 1 h, rinsed with buffer and inspected by uorescence microscopy.The uorescent lipid Texas Red DHPE was exchanged against Oregon Green DHPE or Bodipy-PC, which also label the l d phase, to be able to monitor the l d phase (green) at the same time as the uorescence of the Cy3-labeled STxB (red).Control experiments reveal that non-specic binding of STxB to DOPC/SM/Chol (2 : 2 : 1) membranes lacking Gb 3 can be neglected.

Fig. 3
Fig. 3 SSMs composed of DOPC/SM/Chol/Gb 3 -C24:0 (40 : 35 : 20 : 5) doped with perylene (0.5 mol%) and Oregon Green DHPE (0.1 mol%) after incubation with 60 nM Cy3-labeled STxB.(A) Fluorescence micrograph of Oregon Green DHPE.Membranes segregate into l o (dark) and l d (green) domains.Scale bar: 15 mm.(B) Fluorescence micrograph of STxB-Cy3.The inhomogeneous red fluorescence of the labeled protein is inverse to the Oregon Green DHPE labeled l d phase.Scale bar: 15 mm.(C) AFM image showing a lower l d phase (dark colour coding), higher l o phase and STxB clusters on top of the ordered phase.Scale bar: 2 mm.(D) Height difference between STxB clusters and protein decorated l o phase is determined to Dh(l o+STxB_clusters /l o+STxB ) ¼ 0.6 AE 0.2 nm for the image shown in (C).

Fig. 4
Fig. 4 SSMs composed of DOPC/SM/Chol/Gb 3 -C24:1 (40 : 35 : 20 : 5) doped with perylene (0.5 mol%) and Bodipy-PC (0.1 mol%) after incubation with 60 nM Cy3-labeled STxB.(A) Fluorescence micrograph of Bodipy-PC.The membrane segregates into three phases as indicated by three brightness levels.Scale bar: 15 mm.(B) Fluorescence of Cy3labeled STxB shows protein binding to the l o phase.Top right corner: overlay with Bodipy-PC fluorescence.Scale bar: 15 mm.(C) AFM image of the bilayer after STxB incubation.Three different height levels are observed.Scale bar: 2 mm.(D) Histogram analysis of the heights of the three phases.Taking the l d phase as the lowest phase, where no protein has been bound to, two height differences of Dh(l i /l d ) ¼ 1.2 AE 0.8 nm and Dh(l o+STxB /l d ) ¼ 2.4 AE 0.8 nm are calculated.

Fig. 6
Fig. 6 Schematic drawing of the phase behaviour of lipid membranes induced by different Gb 3 species in absence and presence of STxB.For simplicity, Gb 3 molecules have only been drawn in the upper leaflet of the bilayer.The localization of the Gb 3 s prior to protein binding is only hypothesized.(27) Membranes doped with Gb 3 -C24:0 segregate into l o and l d phase with an l o phase area percentage of more than 70%.Binding of STxB does not lead to lipid redistribution between the phases.After STxB binding Gb 3 is located in the l o phase and STxB demixes the l o phase by clustering the Gb 3 leading to STxB clusters.(29) Membranes doped with Gb 3 -C24:0-2OH exhibit a low l o area percentage due to a presumably tight lipid packing excluding DOPC.STxB binding recruits and clusters Gb 3 in the ordered phase leading to protein clusters.Additionally, the l o phase area percentage is increased indicating that Gb 3 recruitment allows for inclusion of DOPC in the l o phase.(30) Prior to protein binding membranes containing Gb 3 -C24:1-2OH show a similar behaviour as described for 27.Interestingly, protein binding again leads to a strong increase of the l o phase area fraction.The high protein density atop the l o phase indicates steric pressure between proteins bound to the receptor lipids inducing a mixing of the phases. 43(28) Gb 3 -C24:1 induces a demixing of the membrane in absence of protein resulting in a third intermediate phase differing in height and fluorophore partition.STxB selectively binds only to one of the three phases showing that Gb 3 -C24:1 is solely located in this phase.

Table 2
Ratios of the different membrane phases (l o : liquid-ordered, l i : liquid-intermediate) in DOPC/SM/Chol/Gb 3