Addition of lysophospholipids with large head groups to cells inhibits Shiga toxin binding

Shiga toxin (Stx), an AB5 toxin, binds specifically to the neutral glycosphingolipid Gb3 at the cell surface before being transported into cells. We here demonstrate that addition of conical lysophospholipids (LPLs) with large head groups inhibit Stx binding to cells whereas LPLs with small head groups do not. Lysophosphatidylinositol (LPI 18:0), the most efficient LPL with the largest head group, was selected for in-depth investigations to study how the binding of Stx is regulated. We show that the inhibition of Stx binding by LPI is reversible and possibly regulated by cholesterol since addition of methyl-β-cyclodextrin (mβCD) reversed the ability of LPI to inhibit binding. LPI-induced inhibition of Stx binding is independent of signalling and membrane turnover as it occurs in fixed cells as well as after depletion of cellular ATP. Furthermore, data obtained with fluorescent membrane dyes suggest that LPI treatment has a direct effect on plasma membrane lipid packing with shift towards a liquid disordered phase in the outer leaflet, while lysophosphoethanolamine (LPE), which has a small head group, does not. In conclusion, our data show that cellular treatment with conical LPLs with large head groups changes intrinsic properties of the plasma membrane and modulates Stx binding to Gb3.

Membrane lipids are important regulators of basic cellular processes such as ligand binding, endocytosis and intracellular transport 1 . Glycerophospholipids and sphingolipids are common membrane lipids, with different classes and subtypes distributed asymmetrically between the two bilayers of the plasma membrane 2,3 . Cholesterol is an important regulator of plasma membrane fluidity and constitutes 30-50% of plasma membrane lipids 4,5 . Glycerophospholipids consist of a glycerol backbone and most often contain two hydrophobic fatty acid chains attached to sn-1 and sn-2 carbons and a polar head group attached to the sn-3 carbon through a phosphodiester linkage (Fig. 1). Lysophospholipids (LPLs) are glycerophospholipids with only one fatty acyl chain and are found in biological membranes at about 0.5-6% of total lipid weight 6 . Phospholipase A 2 (PLA 2 ) produces LPLs from glycerophospholipids by deacylating the fatty acyl at the sn-2 position (Fig. 1). The LPLs can be reacylated by lysophospho-acyltransferases. LPLs can have a conical shape when the surface area of the polar head group is larger than the corresponding area covered by the fatty acyl chain 7 . However, the conicity of different LPLs is strongly dependent on the size of the head group (Fig. 2) and, additionally, the saturation degree of the fatty acyl chain.
Despite their low abundancy, LPLs have several roles. Endogenous LPLs are important in membrane-shaping processes through the membrane phospholipid remodelling cycle, also known as Land's cycle 8 . Exogenously added LPLs can be directly incorporated into cellular membranes 9,10 . Changes in LPL content and membrane curvature are important for vesicle fusion and fission processes that require membrane bending 3,11,12 .
Besides the structural roles, intracellular LPLs are lipid mediators in signalling 13 . Endogenously synthesized LPLs are secreted into physiological fluids such as blood, saliva and tears 14,15 . Circulating LPLs have been identified as ligands for a variety of G-protein coupled receptors (GPCRs) 16 , and activated GPCRs induce signalling cascades important for cell growth and differentiation, migration and cell-cell attachment, inflammation and immunity. Receptor-dependent and -independent roles of LPLs have been summarized in several comprehensive reviews 15,17,18 . Figure 1. Phospholipid hydrolysis to lysophospholipid. Molecular structure and simplified drawing of a glycerophospholipid that is assumed to acquire a cylindrical shape and a lysophospholipid with a conical shape. The glycerol backbone is in green, the phosphodiester bond in blue and the head group in red. Structural nomenclature (sn) of carbons in the glycerol backbone is shown to the left and indicated by numbers from 1 to 3. Hydrolysis to lysophospholipid is catalysed by PLA 2 in the sn-2 position. Increased levels of LPLs have been found in cancer patient blood and ascites, and a role in cancer pathophysiology has been established [19][20][21] . LPLs are therefore currently under investigation as cancer biomarkers 22,23 . The LPC analogue edelfosine alters membrane structure and induces apoptosis in cancer cells, and it has been tested as an anticancer drug, but showed too low efficacy and high toxicity in clinical trials 24,25 .
Recently, we published that exogenous addition of the conical lysophosphatidylinositol (LPI), but not phosphatidylinositol, protects cells against Shiga toxin (Stx) by inhibiting the binding of the toxin 26 . Whether this was due to a specific effect of LPI or if a similar effect could be obtained with other LPLs, was not tested at that time.
In this study we have investigated whether various LPLs affect binding of protein toxins and explored the mechanistic details of how LPI affects Stx binding.
Stx belongs to an AB 5 toxin family and consists of one enzymatically active moiety (A) and a binding pentamer (B 5 ). Stx binds to the glycosphingolipid Gb3 at the surface of the plasma membrane, is internalized and sorted retrogradely towards the ER, from where the cytotoxic A moiety is translocated to the cytosol and inhibits protein synthesis, leading to cell death 1 . Each B unit has three binding sites with varying affinities for Gb3, providing up to 15 possible interactions between Stx and the carbohydrate moiety of the receptor 27,28 . The affinity of one interaction is at a millimolar scale, whereas multivalency leads to a very strong interaction between Stx and Gb3 receptor analogues, measured in vitro 29,30 . Multivalent binding at the plasma membrane is possible due to the mobility of glycolipids in the lipid bilayer, and clustering of receptors is important for optimally spaced multivalent binding 31 . Figure 2 summarizes major classes of LPLs that were investigated in this study. The polar head group defines the class of the lysophospholipid while the fatty acyl chain length and saturation degree expands the molecular variety within each class. Here we demonstrate that insertion of LPLs with a large head group and a saturated fatty acyl chain affects membrane properties and the ability of Stx to bind to cells.

Results
Treatment with LPLs with large head groups can prevent cell binding and toxicity of Stx. Our earlier studies have revealed that LPI is able to protect cells against Stx by inhibiting binding of Stx to cells 26 . To investigate whether this inhibition was specific for LPI or whether also other LPLs could affect Stx binding, we treated cells with different classes of LPLs and measured Stx binding to HEp-2 cells. All substances tested are listed in Table 1. As shown in Fig. 3a, LPLs with large head groups (LPI, LPS, LPC) and edelfosine have an inhibitory effect on Stx binding in a concentration-dependent manner, while LPLs with small head groups (LPE and LPA) do not have such an effect. The LPC analogue edelfosine was included in this study to investigate if any of the effects observed here are due to enzymatic conversion of LPLs. Edelfosine has a similar shape as LPC, but edelfosine contains an ether-linked methyl group in the sn-2 position instead of a hydroxyl group, and an ether linked C18:0 in the sn-1 position, making it metabolically stable. LPC 18:0 and edelfosine C18:0 gave very similar effect on Stx binding, as shown in Fig. 3a.
To investigate a possible importance of the fatty acyl chain length in LPI-mediated effects, we compared the inhibitory effects of synthetic pure LPI 18:0, LPI 16:0 and LPI from bovine liver containing mostly 18:0 fatty acyl chains (further on called LPI) on Stx binding. Indeed we found that LPI and LPI 18:0 give a stronger reduction in Stx binding than LPI 16:0, suggesting that the inhibitory effect of LPI also depends on the length of the fatty acyl chain (Fig. 3b,c).
Data presented in Fig. 3a-c indicated that the geometry of the lipids might determine the potency of various LPLs to inhibit the binding of Stx. An unsaturated carbon bond in the fatty acyl chain induces a kink and changes the shape and packing abilities of the lipid molecules. We therefore compared the effects of saturated LPC 18:0 with unsaturated LPC 18:1 (Δ 9) and found that LPC 18:1 has no effect on Stx binding, whereas LPC 18:0 strongly inhibits binding (Fig. 3d).
To compare the effect of the head group size on Stx binding, only LPLs with chain length 18:0 or mostly 18:0 were presented in a separate figure. The LPLs were ordered according to the inhibitory effect on Stx binding (Fig. 3e). The degree of inhibition corresponds nicely with the size of the LPL head groups, as shown in Fig. 2. There is almost no inhibitory effect of the LPLs with the smallest head groups and approximate 80% inhibition with the largest head group. An equivalent treatment of HeLa cells with LPLs gave very similar results to that shown for HEp-2 cells ( Supplementary Fig. S1).

LPL effects on toxin cell association, transport and toxicity. Shiga toxin 2 (Stx2) is produced by
Escherichia coli (E. coli) and is genetically and immunologically distinct from Stx (produced by Shigella dysenteriae) and Shiga toxin 1 (produced by E. coli), which is almost identical to Stx, but all these toxins share similar structure and enzymatic function 32,33 . However, Stx2 is more often associated with serious disease than Stx1 33 . We decided to investigate whether treatment with LPLs could inhibit binding and cytotoxicity also of Stx2. As shown in Fig. 4a, treatment with conical LPLs (10 μ M) with large head groups (LPI with mostly C18:0 and LPS with mostly C18:0) strongly inhibited cell association of Stx2, while LPLs with small headgroups (LPA 18:0 and LPE with mostly C18:0) did not. Very similar results were obtained when Stx2 binding was performed on ice (not shown). We also investigated the effect of LPL treatment on intracellular transport of Stx and toxic activity on cells, and our data show that the effect of different LPLs on these events corresponds to the LPL effect on binding. As shown in Fig. 4b, Stx transport to the Golgi was completely inhibited by LPI and strongly reduced by LPS and LPC, measured as sulfation of a modified Stx binding moiety with two sulfation sites attached (StxBSulf2). LPI provided a strong protection against Stx and Stx2 (Fig. 4c,d).
The LPL-induced inhibition of Stx binding is independent of intracellular signalling and membrane turnover. LPLs are bioactive signalling molecules that can bind to specific GPCRs at the plasma membrane and activate cell signalling cascades. GPR55 is a receptor for LPI and O-1602 is an agonist 16,34 . To investigate if the effects observed here are dependent on GPR55 signalling, we tested whether the binding of Stx could be inhibited by treating cells with O-1602. This was not the case since confocal microscopy images showed persistent Stx uptake upon treatment with the agonist, but strong inhibition of binding with LPI ( Fig. 5a). In addition, other lysophospholipids (LPC and LPS) that are known to bind to their own specific GPCR 34 also inhibit binding of Stx, as shown in Fig. 3a.
To test further whether cell signalling is required for the inhibition of Stx binding, we depleted cellular ATP prior to the LPI treatment. The effects of LPI on Stx binding on ice were not affected after inhibition of ATP synthesis ( Fig. 5b), supporting the idea that the effects are signalling independent and also not due to a change in membrane turnover.
Finally, we also tested if LPI would prevent Stx binding on formalin-fixed cells. We performed a chemical cell fixation in 10% formalin solution that crosslinks amine and amide groups of proteins in cells 35 , while membrane lipids stay less affected and retain some mobility 36 . Cell fixation itself reduced Stx binding to cells by 2-fold, but we could still see the same degree of inhibition upon LPI treatment (Fig. 5c) as in live cells, and no inhibition of Stx binding upon treatment with 10 μ M LPA (not shown). In summary, these results clearly show that LPI-induced inhibition of Stx binding is independent of receptor signalling and ATP, indicating that LPI directly affects Gb3 behaviour at the plasma membrane.

Characterization of LPI-induced inhibition of Stx binding to cells.
Of all the LPLs tested, both the bovine liver LPI (mostly 18:0) and the synthetic pure LPI 18:0 gave the strongest reduction in Stx binding. We decided to study the ability of LPI to inhibit Stx in more detail and used LPI from bovine liver (called LPI in the text and figures) for this purpose. First we tested for how long cells had to be incubated with LPI (10 μ M) to obtain inhibition of binding. We observed a strong inhibition already after 10 min and maximal inhibition after 30 min (not shown). Next we tested whether LPI had to be incubated with the cells at 37 °C to inhibit binding of Stx, and as shown in Fig. 6a, this was indeed the case. Pretreatment with LPI on ice gave only a slight effect on Stx binding on ice compared to more than 50% reduction in Stx binding on ice when lipid treatment was performed at 37 °C (Fig. 6a). This is in agreement with the idea that LPI has to be inserted into the membrane for its action. for 20 min at 37 °C. Cells were washed, lysed and the radioactivity was measured. Cell-associated 125 I-Stx2 is expressed as % of control (mean ± deviation; n = 2). (b) Sulfation assay to measure the amounts of StxBSulf2 transported to the Golgi apparatus where the modified toxin is sulfated. HEp-2 cells were pre-incubated with 35 SO 4 2− for 1.5 h and then incubated with 10 μ M LPLs with acyl chains consisting of pure or mostly C18:0 for 30 min, followed by addition of 2 μ g/ml StxBS2 for 1.5 h. Shiga toxin was immunoprecipitated from cell lysates and radioactivity was counted. Results are shown as % of ctrl (mean ± SEM; n = 3). N/A stands for not applicable (below detection limit). (c,d) Toxicity assay with Shiga holotoxin (Stx) and Stx2. Cells were pretreated with 5 μ M LPLs with acyl chains consisting of pure or mostly C18:0 for 30 min, followed by incubation with tenfold serial toxin dilutions for 4 h in leucine-free medium. Protein synthesis was measured and results are shown as % of no toxin (mean ± deviation; n = 2 for each of the toxins). We extended our studies of the concentration dependent effect of LPI on Stx cell association at 37 °C, and the data reveal that even 2 μ M LPI has an inhibitory effect (Fig. 6b). It should be mentioned that LPI concentrations higher than 20 μ M were avoided as it can lead to cell rounding and detachment, and is close to the critical micelle forming concentration of 30 μ M for LPI 18:0 37 .
LPI is transported in blood, bound to albumin 34 . We therefore pretreated cells with 10 μ M LPI in the presence of 10% fetal bovine serum (FBS) to see if we could abolish the LPI-mediated effect on Stx binding. And indeed, there was no reduction in Stx binding when cells were pretreated with LPI in the presence of FBS (Fig. 6c). In agreement with this finding, addition of FBS-containing medium to LPI-treated cells rapidly restored the ability of the cells to bind Stx (Fig. 6d). Even just washing the cells with HEPES-buffered cell growth medium without FBS was sufficient to diminish the effect of LPI on Stx binding, indicating that there is an equilibrium between the aqueous LPI and the LPI inserted into the membrane (Fig. 6d).
Interestingly, LPI treatment not only inhibited binding of Stx, but also binding of an IgM antibody against Gb3 (Fig. 6e). The fluorescence intensity signals of Stx and Gb3 antibody staining on confocal images were quantified, and we observed a significant reduction for both ligands (Fig. 6f).
To investigate whether LPI can release Stx already bound to cells, we prebound Stx to cells on ice and then added LPI at 37 °C. Remarkably, LPI was able to induce dissociation of prebound Stx from the cell surface. Cell-associated Stx was reduced by more than 50% after 20 min incubation with LPI at 37 °C (Fig. 7a).

LPI effect on Stx binding is reversed by methyl-β-cyclodextrin (mβCD).
Cholesterol is an important regulator of membrane organization in general, and it can also interact with Gb3 38,39 thereby changing the orientation of the carbohydrate moiety relative to the membrane. Mβ CD is commonly used to extract membrane cholesterol and is reported to be cholesterol-specific at concentrations below 10 mM, with only negligible binding of other lipids at such low concentrations 40 . We tested whether the reduced binding caused by LPI could be restored after cellular treatment with mβ CD. In control cells, mβ CD treatment for 15 min at concentrations 0.3-10 mM did not affect Stx binding on ice or Stx cell association after 20 min incubation at 37 °C (Fig. 7b,c). In contrast, addition of mβ CD after cell pretreatment with LPI rescued both the Stx binding on ice and cell-associated Stx at 37 °C (Fig. 7b,c). LPI has almost no effect on binding of ricin. To address whether binding of ligands to other receptors is affected by LPI treatment, we investigated the binding of ricin which is an AB class toxin. Ricin has only one binding subunit that binds to various membrane glycolipids and glycoproteins with a terminal galactose. The cell association of ricin was only slightly affected by 10 μ M LPI treatment (reduction by 10%), as shown in Fig. 7d. Other substances tested gave similar results (LPC, LPS, LPE, lysoplasmalogen LPE and edelfosine; data not shown).
LPL treatment and membrane lipid packing. Lipid packing is one of the key physicochemical properties of biological membranes. Multiple environment-sensitive dyes, which change their spectroscopic properties in response to environment parameters, such as polarity, hydration, viscosity, etc., have been developed and used to study lipid packing in both artificial and biological membranes 41 . Quantitative assessment of lipid packing is achieved by measuring the shift in the dye emission profile between liquid ordered (Lo) and liquid disordered (Ld) phases, giving a relative index called generalized polarization (GP) 42,43 . The shift in emission is due to the polarization state of the dye which is affected by accessibility of the dye to water molecules.
To study the effects of LPI and other LPLs on the plasma membrane lipid packing, we used a recently developed environment-sensitive probe NR12S based on the Nile Red fluorophore used for lipophilic staining. In line with other environment-sensitive probes, such as Laurdan and di-ANEPPDHQ 43 , NR12S shows a blue-shifted emission in Lo phase vesicles and a red-shifted emission in Ld phase vesicles 44 . Important key features of NR12S are its ratiometric response to lipid packing (Lo vs. Ld phase) and insensitivity to surface charge, selective localization to the outer leaflet of the plasma membrane with low internalization rate and a high fluorescence quantum yield 44 , which makes it suitable for studying plasma membrane properties in live cells. Cholesterol extraction has been shown to red-shift the emission spectrum of NR12S, indicating a decrease in lipid order at the outer plasma membrane leaflet 44,45 and thus cholesterol depletion by mβ CD was used as a positive control for a shift towards Ld phase in our study. Cell treatment with 0.2% ethanol was used as a reference and gave no notable effect on lipid packing when compared to medium only.
First, we performed spectral imaging with a laser scanning microscope, where the whole emission spectrum of the NR12S was registered, to look for any effects on lipid packing in the plasma membrane caused by various treatments in live HEp-2 cells. We observed a clear red-shift in the emission spectrum of the dye when cells were treated with 10 μ M LPI for 20 min prior to staining with NR12S (Fig. 8a). The analysis of the spectrum was performed on selected regions of the plasma membrane free for filopodia and cell-cell contacts. This was done in order to make sure that the effects we measure are directly mediated by LPI, and not due to changes in cell morphology. Quantitative analysis of lipid packing was performed by calculating the change in generalized polarization value (Δ GP) compared to ethanol. Treatment with 10 μ M LPI gave a significant decrease in the GP, while 10 μ M LPE had no effect on the GP value (Fig. 8b).
Since we had found that treatment with LPC 18:1 in contrast to LPC 18:0 did not affect binding of Stx, we investigated the effect of both LPC 18:1 and LPC 18:0 on lipid packing. We show that cell treatment with LPC 18:1 gives a strong red-shift within the outer leaflet, demonstrating that, even thought it does not inhibit Stx binding (Fig. 3d), it becomes incorporated in the plasma membrane (Fig. 8a,b). This further strengthens the idea that it is membrane insertion of conical LPLs with a large head group and a saturated fatty acyl chain that inhibits binding of Stx.
To study the dynamics of lipid packing upon LPI treatment, we performed TIRF time-lapse imaging where the basal part of the plasma membrane was imaged every 5 min for 20 min. Cells were pre-stained with NR12S before addition of LPI or ethanol, and we observed a clear reduction in GP value (Fig. 8c). Furthermore, we investigated the effect of addition of 10 μ M LPI on the whole cell morphology by live cell imaging using wide-field microscopy and observed that HEp-2 cells rapidly obtained a more rounded surface and had less filopodia upon LPI treatment (Fig. 8d).

Discussion
We here report that insertion of conical LPLs with large head groups inhibits binding of Stx and Stx2 to cells and that LPI even can induce dissociation of cell-bound Stx. A number of studies have revealed that binding of Stx to its receptor Gb3 is dependent not only on the structure of the receptor itself, but also on the membrane environment 38,39,46 . Physical properties of the plasma membrane are regulated by the lipid composition and temperature, and in this study we have investigated effects of adding LPLs with different head groups and fatty acyl chains. Earlier studies on HeLa cells have revealed that radioactively labelled LPI rapidly gets inserted into the outer leaflet of the plasma membrane 10 and that the rate of translocation to the inner leaflet is slow 3 . Similar observations have been made with radioactively labelled LPC on human erythrocytes 9 . The rapid insertion, lack of transmembrane translocation and large head group to acyl chain ratio forms the basis for the ability of LPLs with large head groups to induce the formation of positive curvature 47 . Such a change in membrane curvature is in agreement with the morphological alterations we observe in HEp-2 cells incubated with LPI. Similar morphological cell-shape changes have been reported in CHO-K1 cells upon incubation with LPI 48 . In contrast, we observed that LPE, which has a small head group, does not induce any changes in cell morphology, possibly due to the smaller head group to fatty acyl chain area ratio that gives less impact on lipid packing in the outer leaflet of the plasma membrane. An alternative explanation is a more uniform distribution of LPLs with small head groups between the two leaflets due to a more rapid translocation to the inner leaflet. However, this seems less likely due to short term experiments here used and experiments performed on ATP depleted and also fixed cells.
As demonstrated here, LPLs can affect lipid packing of the outer leaflet of the plasma membrane with a change towards a more liquid disordered phase, shown by the shift in the fluorescence of the NR12S probe which is predicted to be localized in the outer leaflet of the plasma membrane 44 . Since the fluorescence of the NR12S has been shown to be independent of the surface charge of the membrane 44 , it should specifically respond to changed packing of the fatty acyl chains rather than to alterations at the head groups of the lipids. It is therefore tempting to hypothesize that insertion of the conical LPI and LPC 18:0 reduces the packing of the fatty acid groups by introducing more space between the adjacent acyl chains, while there might be an opposite effect at the surface of the lipid layer with increased packing of the head groups. Consequently, lipid packing may contribute to regulate the presentation of the carbohydrate epitope and/or the lateral diffusion of Gb3 in the membrane, as such dependency has been shown for other receptors 49 .
Importantly, although both LPI and mβ CD give a shift towards liquid disordered phase in the membrane, only LPI reduced Stx binding to the cells, suggesting that it is not the reduction in the lipid packing as such (at least not at the level of the fatty acyl chains) that leads to reduced Gb3 availability for binding. An interesting possibility is that the treatment of cells with LPI or LPC 18:0 and the observed increase in the disordered phase might facilitate interaction of Gb3 with cholesterol, bend the carbohydrate part and thereby inhibit Stx binding 38,39 . The reduced binding could then be restored by extraction of cholesterol. Furthermore, it has been reported that increased membrane disorder in erythrocyte membranes obtained by extraction of cholesterol with mβ CD can increase flip-flop of exogenously added LPC from the outer to the inner leaflet 50 , whereas addition of cholesterol will decrease the movement of LPC 51 . Thus, an alternative or additional explanation for the ability of mβ CD to restore binding of Stx is that it promotes movement of LPI to the inner leaflet, thereby reducing the amount of LPI in the outer leaflet. The finding that LPC 18:1 gives a strong increase in the lipid disordered phase, but has no effect on binding of Stx strengthens the idea that it is the shape of the lipid that is important for inhibition of toxin binding. Also, if conical LPLs increase the crowding of head groups, this might lead to reduced Gb3 diffusion and thus inhibit multivalent binding of Stx to the Gb3. The requirement for multivalent binding and also for specific carbohydrate conformations to fit with the binding sites of Stx might explain why Stx binding is inhibited by LPI treatment, whereas there is essentially no effect on binding of the AB toxin ricin.
In summary, our study reveals that it is the geometrical shape of the LPL that determines its efficiency in preventing cellular binding of Stx. The potency depends on both the size of the head group and the length and saturation status of the fatty acid chain. Our data suggest that treatment with conical LPLs with large head groups both changes the physicochemical properties of the plasma membrane and leads to an altered Gb3 receptor conformation and/or distribution.

Materials and Methods
Reagents and antibodies. All  reduction in protein synthesis due to the lipid treatment with the most potent lipids at 10 μ M when incubated for 4.5 h so we show the toxicity data from the experiment with 5 μ M LPL.
Immunofluorescence staining and confocal microscopy. HEp-2 cells were seeded out on coverslips in 4-well plates and grown for one day. First, the cells were pretreated with 10 μ M LPI for 30 min at 37 °C and then incubated with 100 ng/ml Stx1m or Gb3 antibody (5 μ g/ml) for 20 min at 37 °C. Cells were then washed and fixed with 3% methanol free EM grade PFA (Gb3 samples) or 10% formalin solution (Stx samples) for 5 min on ice and then 15 min at room temperature (RT), followed by permeabilization in 0.1% TritonX-100 for 2 min and blocking in 5% FBS in PBS for 1 h at RT. The Stx samples were incubated with primary antibody (4 μ g/ml STX1-3C10) in 5% FBS in PBS for 1 h at RT followed by incubation with secondary antibodies IgG Alexa Fluor 488 1:500 (Stx samples) or IgM Alexa Fluor 488, 1:500 (Gb3 samples). Cells were then washed, fixed in 10% formalin solution and permeabilized in 0.1% Triton X-100 for 2 min. The samples were mounted in ProLong Gold with DAPI and investigated using a Zeiss LSM 710 or 780 laser scanning confocal microscope (Carl Zeiss MicroImaging, Jena, Germany) equipped with an Ar-Laser Multiline (458/488/514 nm), a DPSS-561 10 (561 nm), and a Laser diode 405-30 CW (405 nm). The objective used was a Zeiss Plan-Apochromat 63x/1.40 Oil DIC M27. Images were acquired using the ZEN 2010 software (Carl Zeiss MicroImaging). Fiji software 54 was used for quantification of signal intensities and for image preparation.
Spectral GP imaging on confocal microscope. HEp-2 cells were seeded in 8-well Lab-Tek chambered coverglass dishes at a density of 3 × 10 4 cells/well one day before the experiment. The cells were washed twice with warm HEPES-buffered Live Cell Imaging Solution (cat. no. A14291DJ, ThermoFisher Scientific) and then incubated with 10 μ M LPI, LPC 18:0, LPC 18:1, LPE or 0.2% ethanol or with Live Cell Imaging Solution alone for 20 min at 37 °C. As a positive control for an increase in membrane fluidity, cells were treated with 5 mM mβ CD for 45 min at 37 °C, and then medium was replaced with fresh Live Cell Imaging Solution. To stain cell membranes with NR12S probe, NR12S was diluted in Live Cell Imaging Solution and immediately added to cells. After 7 min, the imaging was started and continued for 10-15 min resulting in 8-10 images taken per condition. The images were acquired using Zeiss LSM 780 confocal microscope equipped with a 63x objective, NA 1.4 and a 32-channeled GaAsP detector array. Laser light at 514 nm was used for the excitation of NR12S, and the fluorescence detection range was set between 521 and 687 nm with 8.7 nm intervals. The focus was set to equatorial plane of the cells slightly above the cover glass surface. The sample chamber was controlled for temperature (37 °C) and CO 2 . Spectra for each image pixel were obtained from the intensity values of the different detection channels by using the ImageJ software function "Plot Z-axis profile" 41 . The analysis was performed on manually selected regions of the plasma membrane which were not in contact with other cells (to avoid any effects of cell-to-cell contacts). The background signal was determined by applying selections of the same size on a dark region of the image, which was subsequently subtracted from the values obtained from the regions of interest. At least 24 regions of interest were analysed per condition for each independent experiment.
The average GP value for each selected region was calculated according to equation: , where I(573) and I(599) are the fluorescence intensities at wavelengths 573 nm and 599 nm, respectively. The intensity values were directly obtained from the Ld membrane phases is arbitrary, the GP values are consistent only within one experimental study, and can not be directly compared to values obtained with other dyes or imaging settings. Therefore, a difference in GP value between different treatments and the ethanol control rather than the absolute GP value was used for the final quantification: ∆ = − GP GP treatment GP ethanol ( ) ( ) .
TIRF imaging. HEp-2 cells were seeded as for spectral GP imaging. The cells were washed twice with warm Live Cell Imaging Solution and then incubated with freshly prepared 10 nM NR12S solution in Live Cell Imaging medium for 7 min at 37 °C. The NR12S solution was replaced by fresh Live Cell Imaging medium and the first set of TIRF images was taken (time point 0 min). Then, 10 μ M LPI or ethanol was added to the cells and a time-series of TIRF images was acquired starting at 5 min after the addition of the drugs. TIRF images were acquired using a DeltaVision OMX V4 system (Applied Precision, GE Healthcare) with a ring TIRF design (licensed to GE Healthcare from Yale University) rotating the laser beam at the back focal plane of the objective lens. The microscope was equipped with a 60x ApoN NA 1.49 objective (Olympus), 514 nm (100 mW diode) laser and three cooled sCMOS cameras (PCO). The stage and the objective were temperature controlled at 37 °C and the sample holder was enclosed by a humidified 5% CO 2 incubation chamber. The system was controlled using the OMX Acquisition control software running under a Windows 7 operating system. Image registration was performed using the SoftWoRx (Applied Precision, GE Healthcare) software package running under a CentOS v6 (Linux) operating system. Two emission filters used, 541/22 and 609/37, corresponded to the regions of Lo and Ld phase of the NR12S spectrum, respectively. To visualise changes in GP values, pseudo-coloured GP images were generated using ImageJ macro adapted from Owen et al. 55 .
Live cell imaging. HEp-2 cells were seeded in 35 mm MatTek glass bottom dishes at a density of 1 × 10 5 cells/well one day before the experiment. At the day of the experiment, cells were washed once and incubated in serum-free HEPES-buffered cell growth medium and images were acquired on a DeltaVision microscope (Applied Precision, Issaquah, WA, USA) equipped with an Elite TruLight Illumination System, a CoolSNAP HQ2 camera and a 60x objective (Olympus, NA 1.42, Plan Apo N, UIS2, 1-U2B933). Images were captured under controlled CO 2 conditions at 37 °C in a temperature-controlled incubation chamber. Image acquisition was started and 10 μ M LPI or LPE was added after 10 min, and the imaging continued for 10 min. Brightfield illumination system with polarizer allowed us to obtain differential interference contrast images for whole cell morphology studies. Images were acquired every 60 sec for 20 min using softWoRx software (Applied Precision, GE Healthcare). Images illustrating the cell morphology before and after treatment were processed using Fiji software. To test potential toxic effects of LPI treatment, cells were left in the microscope chamber for 2 hours and control images were taken to ensure that cells had not become apoptotic or detached.

Statistics.
All experiments were performed in duplicates or triplicates. The experimental results are presented as mean values ± standard error of the mean (SEM) of n independent experiments, where n is indicated in each figure legend, if not otherwise stated. The data for presentation were normalized to control, with control set as 100% and two-tailed unequal variance t-test was used when treatment values were compared to normalized control. Two-tailed paired or unpaired t-tests were used to determine difference between means of two different treatments when appropriate. The levels of significance were set at * p ≤ 0.05, * * p ≤ 0.01 and * * * p ≤ 0.005.