Endogenous sphingomyelin segregates into submicrometric domains in the living erythrocyte membrane.

We recently reported that trace insertion of exogenous fluorescent (green BODIPY) analogs of sphingomyelin (SM) into living red blood cells (RBCs), partially spread onto coverslips, labels submicrometric domains, visible by confocal microscopy. We here extend this feature to endogenous SM, upon binding of a SM-specific nontoxic (NT) fragment of the earthworm toxin, lysenin, fused to the red monomeric fluorescent protein, mCherry [construct named His-mCherry-NT-lysenin (lysenin*)]. Specificity of lysenin* binding was verified with composition-defined liposomes and by loss of (125)I-lysenin* binding to erythrocytes upon SM depletion by SMase. The (125)I-lysenin* binding isotherm indicated saturation at 3.5 × 10(6) molecules/RBC, i.e., ∼3% of SM coverage. Nonsaturating lysenin* concentration also labeled sub-micrometric domains on the plasma membrane of partially spread erythrocytes, colocalizing with inserted green BODIPY-SM, and abrogated by SMase. Lysenin*-labeled domains were stable in time and space and were regulated by temperature and cholesterol. The abundance, size, positioning, and segregation of lysenin*-labeled domains from other lipids (BODIPY-phosphatidylcholine or -glycosphingolipids) depended on membrane tension. Similar lysenin*-labeled domains were evidenced in RBCs gently suspended in 3D-gel. Taken together, these data demonstrate submicrometric compartmentation of endogenous SM at the membrane of a living cell in vitro, and suggest it may be a genuine feature of erythrocytes in vivo.

The main aim of the present study was to assess the lateral organization of endogenous SM in the fl at membrane of featureless RBCs by high-resolution confocal imaging. To this aim, we swapped Dronpa, which is best suited to PALM, for the monomeric red fl uorescent protein mCherry, which is far more photostable, thus optimal for long-term imaging experiments. This new construct, referred to as His-mCherry-NT-lysenin (lysenin*), also allowed for colabeling with green BODIPYlipids, so as to address simultaneously the localization of endogenous (red signal) and fl uorescent exogenous SM or other polar lipids (green signal). Most crucial, our investigations were carried out in unfi xed cells, be cause conventional formaldehyde-based fi xatives do not crosslink lipids and do not arrest lateral protein diffusion ( 32 ).
We fi rst validated the binding specifi city and innocuity of lysenin* and verifi ed by radio-iodination that trace labeling yielded a visible signal by confocal microscopy. We saw that labeling of endogenous SM on RBCs partially spread onto coverslips revealed comparable submicrometric domains upon insertion of exogenous green BODIPY-SM. Combination of nonsaturating red lysenin* concentration with green BODIPY-SM yielded perfect colocalization and allowed for double labeling with other BODIPY-lipids as well. Lysenin* proved sensitive, convenient, and reliable to address the roles of temperature, cholesterol, and membrane tension for the biogenesis of endogenous SM domains. It also allowed labeling of RBCs gently suspended in a 3D-gel, so as to rule out physical stress imposed on the plasma membrane of partially spread RBCs, and revealed numerous domains . This indicated that submicrometric domains may also occur on circulating RBCs in vivo.

Expression and purifi cation of lysenin*
The expression plasmid pET28/lysenin* encodes the monomeric C-terminal, nontoxic (NT) domain of the SM-specifi c toxin, NT-lysenin, that is expressed as a fusion protein with an N-terminal 6xHis-tag followed by the monomeric red fl uorescent protein mCherry. It was generated from pET28/His-Dronpa-NTlysenin ( 31 ) by swapping in-frame Dronpa for mCherry sequences using restriction enzymes KpnI (from Klebsiella pneumonia ) and Hin dIII (from Haemophilus infl uenzae) . The resulting plasmid was expanded into Escherichia coli strain BL21 (DE3) and recombinant protein, lysenin*, was expressed in lysogeny broth (LB) medium at 16°C for 72 h in the presence of 0.4 mM isopropyl ␤ -D-thiogalactoside. Bacterial extracts were prepared as previously described ( 33 ) and the recombinant protein was purifi ed using an Ni-NTA Superfl ow cartridge (Qiagen) and eluted with imidazole as in ( 34 ). Fraction analysis by SDS-PAGE revealed recombinant lysenin* at the expected size ( ‫ف‬ 45 kDa). Most enriched fractions were pooled, concentrated, and desalted as described ( 33 ), then aliquots were stored in 20 mM NaCl supplemented with 25 mM HEPES (pH 7.2) and 5% glycerol at Ϫ 80°C until use. Protein concentration was estimated by measuring the absorbance at 280 nm. ( 20 ), and further indicated by single-molecule tracking, based on discrete jumps between predicted mesoscale domains ( 6 ). Besides this dynamic evidence, we previously reported that insertion at trace level into the outer plasma membrane leafl et of exogenous SM, modifi ed by replacing part of its fatty acyl chain by the green fl uorophore BODIPY (BODIPY-SM), labels submicrometric assemblies visible by confocal microscopy on various cells, including cultured cell lines and red blood cells (RBCs) partially spread onto poly-L-lysine-coated coverslips (9)(10)(11). Other reported examples include: i ) activated blood platelets labeled with the lipidomimetic dialkylindocarbocyanine dye, DiIC18 ( 21 ); ii ) rabbit RBCs and macrophages labeled with Laurdan ( 13,22 ), an artifi cial membrane probe for "packing" of surrounding lipids; iii ) fi broblasts labeled with BODIPYlactosylceramide ( 23 ); and iv ) very specialized membranes, such as human skin stratum corneum ( 14 ) and lung surfactant ( 15 ), in which lipids are the main components and membrane cytoskeleton anchorage is lacking. Interestingly, domains of similar size have been reported for yeast plasma membrane proteins ( 24,25 ).
Recently, special attention has been given to the lateral organization of SM at the plasma membrane, in view of its abundance [ ‫ف‬ 35% of total phospholipid content in human RBCs ( 26,27 )], and its importance for the formation/maintenance of rafts and their signaling function. By confocal microscopy, fl uorescent patches labeled by trace insertion of BODIPY-SM into the plasma membrane of partially spread/stretched RBCs are ‫ف‬ 500 nm in diameter and show a 5-to 8-fold enrichment over the rest of the labeled surface. These domains appear stable in time and space and depend on temperature, membrane tension, cholesterol, and association to the membrane:spectrin anchorage complexes (9)(10)(11). Disappearance of BODIPY-SM domains upon inhibition of endogenous SM biosynthesis or surface depletion by SMase pointed to natural SM assemblies ( 9,10 ). However, acyl BODIPY substitution remained an important issue, calling for direct demonstration for endogenous SM lateral organization in living cells.
Using high-resolution imaging mass spectrometry, Kraft and colleagues ( 8 ) recently reported the existence of 15 N-SL stable submicrometric domains ( ‫ف‬ 200 nm average diameter) in fi xed fi broblasts, which depend on cholesterol and cytoskeleton. Another way to look at endogenous surface lipids, applicable to nonfi xed cells, is direct labeling via nontoxic (NT) binding proteins derived from lipidspecifi c toxins. SM is known to be subverted as a "receptor" for pore-forming toxins such as equinatoxin II [from the sea anemone Actinia equina ( 28 )] and lysenin [from the earthworm Eisenia foetida ( 29,30 )]. Molecular dissection of lysenin distinguished a pore-forming domain (amino acids 1-160) and a C-terminal SM-binding domain (amino acids 161-297). A chimeric protein made of the latter sequence, as NT lysenin fragment (NT-lysenin), and the fl uor escent protein, Dronpa, in the N-terminal position (Dronpa-NT-lysenin) preserved its ability to specifi cally bind to SM and allowed observation of submicrometric domains ( ‫ف‬ 250 nm in diameter) by super-resolution microscopy (PALM) on fi xed HeLa cells ( 31 ). isolated and washed RBCs ( ‫ف‬ 2 × 10 7 cells in 50 l) were incubated in suspension with a fi xed amount of 125 I-lysenin* (0.3 × 10 6 cpm) mixed with increased concentrations of cold lysenin* at 20°C for 20 min, then washed three times by centrifugation/ resuspension. Aliquots of the three SNs and the pellet were counted in a ␥ counter (2470 Wizard automatic ␥ counter; Perkin-Elmer). Radioactivity in pooled SNs and pellet was considered as free and bound lysenin*, respectively. When appropriate, RBCs were pretreated with m ␤ CD or SMase, as above, prior to labeling with 0.3 × 10 6 cpm 125 I-lysenin* mixed with 1 M cold lysenin* in the continuous presence of the pharmacological agents.

RBC labeling with lysenin* and immobilization on coated coverslips and IBIDI chambers or trapping in a gel
Before each experiment, lysenin* in 1 mg/ml DF-BSA was cleared of aggregates by centrifugation at 20,000 g for 10 min as above. Except when otherwise stated, freshly isolated erythrocytes were then incubated in suspension with 1.5 M lysenin* at 20°C or 37°C and immobilized onto 2 cm 2 poly-L-lysine-coated coverslips. Briefl y, coverslips were fi rst coated with 0.1 mg/ml poly-L-lysine (70-150 kDa; Sigma-Aldrich) at 20°C for 7 min, then DMEM was added for another 7 min. Poly-L-lysine:DMEM was then removed and coverslips were left to dry for at least 1 h. Washed RBCs ( ‫ف‬ 10 7 cells in 25 l) were plated onto a coated coverslip at 20°C for exactly 4 min, then the suspension was removed and replaced by fresh medium, and attached RBCs were allowed to spread for another 4 min (except for colocalization with BODIPY-lipids, see below). This led to a variable level of stretching, exploited in Figs

RBC isolation and pharmacological treatments
RBCs were isolated from healthy volunteers. This study was approved by the Medical Ethics Institutional Committee of the Université catholique de Louvain; each donor gave written informed consent. Blood was collected by venipuncture into dry EDTA (K + salt)-coated tubes, diluted 1:10 in DMEM [containing 25 mM glucose and 25 mM HEPES (Invitrogen)], and washed twice by centrifugation at 133 g for 2 min and resuspension. For cholesterol or SM depletion, washed RBCs were respectively preincubated in suspension at 37°C with 0-0.25 mM methyl-␤ -cyclodextrin (m ␤ CD) (Sigma-Aldrich) for 30 min or at 20°C with 0-10 mU/ ml Bacillus cereus SMase (Sigma-Aldrich) for 10 min. RBCs were further pelleted as above and labeled with lysenin* (see below) in the continued presence of m ␤ CD or SMase (as appropriate), pelleted and resuspended in DMEM containing 5 mg/ml BSA then again in DMEM alone, and plated onto coverslips for imaging (see below).
Pellets containing MLVs or RBCs and supernatants (SNs) were analyzed by Western blotting. Proteins were resolved by SDS-PAGE (in 12% polyacrylamide gels) and transferred onto nitrocellulose membrane (Hybond™-C Extra; Amersham Biosciences, Roosendaal, The Netherlands). After migration and transfer, blots were blocked at room temperature for 1 h in TBS [20 mM Tris-HCl, 0.5M NaCl (pH 7.5)] supplemented with 5% nonfat dry milk and 0.05% Tween 20 and incubated at 4°C overnight with a mouse anti-N-terminal His antibody (1:4,000; GE Healthcare) in TBS containing 1% defatted (DF)-BSA (Sigma-Aldrich), followed by three washes in TBS plus 0.05% Tween 20. Blots were fi nally incubated at room temperature for 1 h with HRP-conjugated goat anti-mouse IgG (1:10,000; Life Technologies) in TBS supplemented with 5% nonfat dry milk and 0.05% Tween 20 and washed three times in TBS containing 0.05% Tween 20 and once in TBS. Immunoreactive bands were visualized using chemiluminescence (ECL SuperSignal® West Femto; Thermo Scientifi c) and acquired with a 4000MM Kodak Image Station (Eastman Kodak Co., Rochester, NY).

Lysenin* radio-iodination and binding to RBCs
Lysenin* radio-iodination was performed using precoated iodination tubes according to the manufacturer's protocol (Pierce). Briefl y, 50 g purifi ed lysenin* was labeled with 1 mCi Na 125 I to a specifi c radioactivity of ‫ف‬ 1,500 cpm/ng protein. More than 97% of the radioactivity was precipitable after incubation with 10% (v/v) trichloroacetic acid. Before binding to RBCs, lysenin* was mixed in DMEM containing 1 mg/ml DF-BSA, then cleared of aggregates by centrifugation at 20,000 g for 10 min at 4°C. Freshly was almost completely abrogated by 10 mU/ml SMase ( Fig. 1C ), supporting specifi c interaction with SM also in RBCs. This treatment decreased the SM level by ‫ف‬ 60%

Scanning electron microscopy
Erythrocytes spread onto poly-L-lysine-coated coverslips and labeled or not with lysenin* were extensively washed with DMEM, fi xed with 15 mM dimethylsuberimidate (pH 8.0) at room temperature for 20 min, rinsed twice in 0.14M cacodylate buffer (pH 7.4), and fi xed with 1% glutaraldehyde in cacodylate buffer at 4°C overnight. The next day, samples were extensively washed in 0.1M cacodylate buffer and postfi xed with 1% (w/v) osmium tetroxide at 4°C for 30 min. Samples were then dehydrated in graded ethanol series and critical-point dried (CPD 300, Leica, Austria). A 10 nm gold fi lm was sputter-coated, and specimens were observed in a CM12 electron microscope at 80 kV with the use of the secondary electron detector (Philips, Eindhoven, The Netherlands).

SM measurements
Lipids were extracted in chloroform/methanol and stirred. Samples were then centrifuged (3,320 g for 15 min) at room temperature and the organic phase was collected and washed with 0.05M NaCl. After stirring and centrifugation, the organic phase was washed again twice by addition of 0.36M CaCl 2 /methanol (1:1; v/v). After stirring and centrifugation, Triton X-100 in acetone was added to the organic phase, which was then dried under nitrogen fl ow at room temperature and resuspended in water. Extracted lipids were then analyzed with a fl uorometric SM assay kit (Abcam). Alternatively, lipids were separated by thin layer chromatography in chloroform:methanol:15 mM CaCl 2 (65:35: ( 36 ) and revealed by charring densitometry after staining with 10% cupric sulfate in 8% O -phosphoric acid ( 37 ). Band intensity of SM was quantifi ed and expressed by reference to the band corresponding to PC from the same sample and fi nally expressed as percentage of control.

Hemolysis
To evaluate the innocuity of lysenin*, RBCs were incubated in suspension with the indicated concentrations of lysenin* at 37°C for 30 min and then centrifuged (133 g for 2 min). To determine the innocuity of pharmacological agents, RBCs were either incubated with the indicated concentrations of SMase at 20°C for 30 min or with m ␤ CD as in ( 11 ). Hemoglobin released in SNs was then read at 560 nm in 96-well plates (SpectraCount TM , Packard BioScience Co.). For normalization as 100% value, full hemolysis was achieved by 0.2% Triton X-100.

Specifi c binding of lysenin* to SM in artifi cial and natural membranes
By swapping Dronpa for mCherry, we generated a useful probe consisting of an N-terminal His tag fused to mCherry and the NT C-terminal domain of lysenin*, as the minimal fragment able to recognize SM ( 38 ). By Western blotting anti-His tag, two bands can be observed, one at ‫ف‬ 45 kDa, corresponding to full lysenin*, and an N-terminal spurious fragment at ‫ف‬ 30 kDa (lanes 5 and 3 in Fig. 1A, B , respectively). As expected, the latter was not detected in the pellet, neither from MLVs (lane 1 in Fig. 1A ) nor from RBCs (lane 1 in Fig. 1B ), indicating binding specifi city of the full lysenin *. In addition, lysenin* only bound to and sedimented with liposomes when containing SM, indicating specifi c interaction with SM in model membranes ( Fig. 1A , compare lanes 1 and 3). Moreover, 125 I-lysenin* binding to RBCs I-lysenin* bound to RBCs were measured and expressed as percentage of control cells. SM depletion decreases lysenin* binding by ‫ف‬ 90%. D: Binding isotherm of 125 I-lysenin* to RBCs. RBCs were incubated in suspension with 125 I-lysenin* (0.3 × 10 6 cpm) together with the indicated concentrations of cold lysenin*. Bound lysenin* is expressed as (molecules/RBC) × 10 6 and is presented as the mean ± SEM of three samples. Representative of two independent experiments. E: Scatchard plot of data in (D). Bound values represent (molecules/RBC) × 10 6 and bound/free ratios are in nl /RBC. It has to be noticed that free values represent both the free "RBCbindable" full lysenin* ( ‫ف‬ 45 kDa) and the "non-bindable" fragment ( ‫ف‬ 30 kDa). fl uorescence intensity was normalized to the average signal obtained at the common 2.5 M concentration. Increasing lysenin* concentrations changed neither the abundance nor the size of domains ( ‫ف‬ 0.5 m diameter on average; data not shown), but their average fl uorescence intensity increased proportionally with concentration up to 5 M (profi les in Fig. 3A, B and quantifi cation in suffi cient for lysenin* binding to membranes and validated its use as a specifi c reporter of endogenous SM at the outer leafl et of RBCs. Moreover, up to 7.5 M lysenin* showed no toxicity, as indicated by hemolysis assay (supplementary Fig. II).

Quantifi cation of accessible lysenin* binding sites
To determine binding parameters of the interaction between lysenin* and SM, RBCs were incubated with increasing 125 I-lysenin* concentrations. Binding roughly increased in proportion with concentration until ‫ف‬ 2.5 M, then leveled off to reach saturation at ‫ف‬ 3.5 × 10 6 lysenin* molecules bound per RBC. Analysis by Scatchard plot revealed a single class of binding sites ( Fig. 1D, E ). Unless otherwise stated, a nonsaturating lysenin* concentration was used hereafter ( р 1.5 M). Three conclusions were derived. First, in this assay, lysenin* specifi cally and exclusively interacted with SM at the outer RBC membrane leafl et. Second, the estimated level of occupancy at saturated lysenin* binding was ‫ف‬ 3% of SM abundance in this leafl et, estimated at ‫ف‬ 10 8 per RBC ( 26 ). Third, the single component in the Scatchard plot was compatible with monovalent lysenin*:SM interaction.

Like exogenous BODIPY-SM, lysenin* labels submicrometric domains on RBCs partially spread onto poly-L-lysine coverslips
These conclusions encouraged us to use lysenin* as a SM-specifi c probe for vital confocal imaging. When RBCs were labeled in suspension with 1.25 M lysenin*, then partially spread onto poly-L-lysine-treated coverslips, a limited number of well-defi ned round submicrometric domains were observed on most cells ( Fig. 2Aa ). Such fl uorescent patches were undistinguishable in size and shape from those observed upon exogenous BODIPY-SM insertion into the outer plasma membrane leafl et of partially spread/stretched RBCs [compare Fig. 2Ab with Fig.  2Aa ; see also ( 9,10 )]. Lysenin*-labeled submicrometric domains vanished upon moderate SM digestion ( ‫ف‬ 25%) by 3 mU/ml SMase, whereas diffuse labeling was largely preserved. At higher SM digestion ( ‫ف‬ 60% by 10 mU/ml SMase), both submicrometric domains and diffuse lysenin* labeling were strongly decreased (supplementary Fig. IC). Domains were also observed upon spreading onto coverslips precoated with poly-D-lysine instead of poly-Llysine (supplementary Fig. IIIa-c) and on loosely attached stomatocytes/discocytes in plastic IBIDI chambers (supplementary Fig. IIId-f), ruling out specifi c artifact due to poly-L-lysine-coverslip spreading. These experiments indicated that endogenous SM formed submicrometric domains, readily decorated by lysenin*.
Next, to address independently whether lysenin* was an "innocent" probe or might affect domain formation, we looked at the effect of an ‫ف‬ 10-fold lysenin* concentration range on the number, size, and fl uorescence intensity of labeled domains. In practice, to avoid saturation of detectors by emitted fl uorescence, we analyzed two series of concentrations (from 2.5 to 5 M and from 0.6 to 2.5 M) using identical settings for image acquisition in each, then Colocalization between lysenin* and BODIPY-SM domains is independent of the order of labeling but depends on lysenin* enrichment Because lysenin*-labeled submicrometric domains enriched in endogenous SM that were seen on partially spread RBCs appeared undistinguishable from those revealed upon insertion of exogenous BODIPY-SM, locally delivered by the bulky protein BSA as carrier, we performed double labeling experiments with non-or sub-saturating lysenin* concentrations (see Fig. 1D ), using two orders of probe addition. When erythrocytes were fi rst labeled in suspension with lysenin* at low (non-saturating) concentration (1.25 M) then incubated with BODIPY-SM in the absence of toxin, all domains were double labeled by lysenin* and BODIPY-SM ( Fig. 2Ba ). At higher lysenin* concentration (2.5 M), moderately lysenin*-labeled domains remained equally labeled by BODIPY-SM (yellow arrowheads in Fig. 2Bb ), but accessibility of BODIPY-SM to lysenin*-saturated domains was strongly reduced (red arrowheads in Fig. 2Bb ; intensity profi les in Fig. 2Bb' ). However, when the order was reversed (i.e., RBCs sequentially labeled in suspension with BODIPY-SM then with 2.5 M lysenin* in the continued presence of BODIPY-SM), SM domains showed perfect colocalization (supplementary Fig. IV), suggesting that lysenin* is able to recognize both endogenous SM and BODIPY-SM. The identity between endogenous SM domains decorated by nonsaturating lysenin* concentrations or revealed upon insertion of the exogenous analog, BODIPY-SM, validated a posteriori the latter tracer as a bona fi de qualitative probe. The partial loss of accessibility to BODIPY-SM:BSA complexes ( ‫ف‬ 68 kDa) upon saturating lysenin* labeling ( ‫ف‬ 45 kDa) can be readily explained by a steric hindrance and is consistent with the observed 4-to 12-fold enrichment of lysenin* in the submicrometric SM domains.

BODIPY-SM-and lysenin*-labeled endogenous SM submicrometric domains are regulated by temperature and cholesterol
Despite our evidence indicating that BODIPY-SM can be used as a qualitative vital tracer of SM submicrometric domains, the spatial identity between BODIPY-SM-and lysenin*-labeled patches did not yet exclude an effect of the inserted BODIPY moiety on the functional properties of the SM domains. Because we had previously shown that temperature and cholesterol regulated the abundance of BODIPY-SM domains ( 9-11 ), we next investigated whether they likewise affected endogenous SM submicrometric domains labeled by lysenin*. As shown in Fig. 4A , confocal imaging of erythrocytes labeled with lysenin* revealed well-defi ned submicrometric patches at all examined temperatures (from 10°C to 42°C), but the number of domains varied with temperature, peaking at 20°C ( Fig. 4B ), as previously reported upon BODIPY-SM insertion ( 9 ).
The role of cholesterol was next examined in RBCs at 37°C using partial depletion by m ␤ CD. At 0.25 mM, this treatment led to an ‫ف‬ 25% cholesterol depletion without causing any hemolysis ( 11 ) or affecting the endogenous SM content (data not shown). This moderate cholesterol depletion strongly decreased lysenin* domain abundance Fig. 3C ; provided light-saturated profi les were excluded). A 4-to 12-fold enrichment was observed in this concentration range. We concluded that lysenin* did not induce SM domain formation or expansion, and could instead be used as a reliable probe to detect variations in local SM concentration. with retention by the spectrin cytoskeleton, at least in partially spread RBCs. By FRAP, ‫ف‬ 60% of fl uorescence recovery was obtained in bleached areas, with a t 1/2 of ‫ف‬ 30 s ( Fig. 6B ). For reference, BODIPY-SM domains showed an ‫ف‬ 75% recovery, with a t 1/2 of ‫ف‬ 10 s ( 11 ). These results indicated that SM submicrometric domains are stable assemblies, in time and space, of highly dynamic individual lipids/clusters. The slower restoration of lysenin* signal in submicrometric domains by nonbleached lipids, as compared with BODIPY-SM, could be attributed to the much larger individual size and/or steric hindrance of the conjugated toxin fragment.

Endogenous SM submicrometric domain abundance, size, and positioning depend on RBC spreading
Based on similar properties of lysenin* and BODIPY-SM domains, we next took advantage of mCherry photostability to analyze in detail the relationship between the presence of domains and membrane tension, as fi rst indicated by the extent of spreading. In our previous report on BODIPY-SM domains, we simply distinguished two stages: partially versus extensively spread RBCs, the former bearing domains and the latter not ( 11 ). After careful analysis of a larger number of RBCs labeled by low lysenin* concentration, three different patterns, related to increasing stage of spreading, could be distinguished: i ) stage #1, with minimal spreading (as suggested by combination of smaller average total area with strongest diffuse label at the cell margin, indicating projection of micrometer-thick membrane curvature), showed mostly large domains with central location (arrowhead in Fig. 7 , inset #1); ii ) stage #2,  Fig. V). However, total binding of 125 I-lysenin* to RBCs was unaffected ( Fig. 5B ), indicating that loss of SM domains was not due to impaired lysenin* binding, but refl ected a crucial role of cholesterol in their stabilization. Taken together, these results showed that, as observed after exogenous BODIPY-SM insertion, endogenous SM submicrometric domain abundance depends on both temperature and membrane cholesterol content. Moreover, they confi rmed a role for lipid phase behavior and cholesterol in SM domain formation and/or maintenance.

BODIPY-SM and endogenous SM submicrometric domains are stable in time and space
Because the plasma membrane of RBCs is stabilized by a strong tangential spectrin network connected by two nonredundant perpendicular anchorage complexes ( 39 ), we looked at SM submicrometric domain stability using FRAP on lysenin*-labeled domains. Despite high mCherry photoresistance, up to 50% photobleaching could be reached. As shown in Fig. 6A , domains in unbleached and bleached areas (white square) retained a fi xed position and a comparable size over an interval of ‫ف‬ 2 min, compatible

Segregation between endogenous SM domains and other lipids depends on RBC spreading
Using a combination of green BODIPY (BODIPY 505 ) derivatives of GSLs (GlcCer and GM1) and PC, and a red BODIPY (BODIPY 589 ) derivative of GlcCer, we recently reported discrimination between PC and GlcCer domains contrasting with large colocalization of GlcCer with GM1 ( 10 ). However, because red BODIPY is a more bulky fl uorophore than green BODIPY, we remained concerned by perturbations of fl uorophore-substituted fatty acyl chain lateral interactions within the outer membrane leafl et. To further investigate this issue, we took advantage of red lysenin* to perform double-vital imaging under more reliable conditions. As shown in Fig. 8 , we found that the relation between endogenous SM labeled by low lysenin* concentration (1.25 M) and other lipids depends on RBC membrane stretching and the lipid class. On stage #1 RBCs, all green BODIPY-PC and -GM1 domains coincided with lysenin* domains (yellow arrowheads in Fig. 8a, b ). BODIPY-GlcCer also partially colocalized with lysenin* domains, but additional spots were only labeled by BODIPY-GlcCer (white arrow in Fig. 8c ), which can be reasonably attributed to the higher melting temperature of GlcCer than GM1, SM, and PC. However, further stretching (stage #2 RBCs) promoted segregation between red lysenin* domains (red arrowheads in Fig. 8d-f ) and the three BODIPY-lipids (white arrowheads in Fig.  8d-f ). The question was irrelevant for stage #3 where SM domains were absent. These results raised the hypothesis that SM differentially associates with PC, GM1, and GlcCer, according to the balance between intrinsic packing (refl ected by melting temperature) and membrane tension. with more spreading (increased average total area and lesser marginal signal), displayed not only large central domains as above (arrowhead) but also smaller domains that extended toward the cell periphery (arrows in Fig. 7 , inset #2); and iii ) stage #3, with most stretched RBCs (largest average cell area and little or no peripheral signal), displayed no detectable enriched domain over the diffuse labeling ( Fig. 7 , inset #3). Irrespectively of the RBC spreading level, domains did not refl ect "anatomical" features, such as surface projection or vesiculation. Indeed, in contrast to lysenin*, TMA-DPH, a "bulk" membrane tracer ( 40 ), did not cluster at the RBC plasma membrane into round domains (supplementary Fig. VI). RBC surface was further analyzed by high-resolution scanning electron microscopy, fi rst without labeling to verify the preservation of its smooth membrane under control conditions, then after labeling with lysenin* to exclude formation of membrane protrusion. As shown in supplementary Fig. VIIa, untreated poly-L-lysine-coated RBCs fi xed with dimethylsuberimidate showed a smooth surface devoid of spicules and microvesicles. Upon labeling with lysenin* and subsequent fi xation with dimethylsuberimidate, domains appeared undistinguishable as observed in living cells (supplementary Fig. VIIb' ) and the RBC surface remained  ( 11 ) for comparison purposes]. Fluorescence recovery is expressed as percentage of signal before photobleaching, after correction of residual fl uorescence immediately after bleaching. Curves derived by monoexponential fi tting. As compared with BODIPY-SM domains (green squares, t 1/2 ‫ف‬ 10 s; ‫ف‬ 75% maximal recovery), notice the expected slower (t 1/2 ‫ف‬ 33 s) and decreased fl uorescence recovery ( ‫ف‬ 60% at infi nite time) of endogenous SM domains decorated by lysenin* (red circles). Fig. 7. Abundance, size, and localization of endogenous SM submicrometric domains depend on RBC spreading stage. RBCs were labeled in suspension with lysenin* at room temperature, allowed to attach onto poly-L-lysine-coverslips, placed upside-down in Lab-Tek chambers, and directly analyzed by confocal microscopy. This led to three stages of RBC spreading corresponding to differential domain abundance, size, and localization: i ) stage #1, notice minimal spreading and strongest peripheral labeling, associated with large central domains (arrowhead); ii ) stage #2, intermediate spreading associated with both large central (arrowhead) and small peripheral domains (arrows); and iii ) stage #3, most spread RBCs, with only diffuse homogenous lysenin* labeling but no domain. Scale bars for general view, 5 m; insets, 2 m. Representative images from >20 experiments.

Lysenin* also labels SM submicrometric domains on RBCs in suspension in a gel
Finally, because spreading onto coverslips induces an artifi cially fl at conformation, different from the biconcave ovoid shape of circulating RBCs, we developed an alternative technique to visualize erythrocytes by high-resolution confocal vital imaging after suspension in a 3D-gel. Gentle suspension of lysenin*-labeled RBCs into the gel, which better preserved their ovoid shape, revealed more SM submicrometric domains than after stretching ( Fig. 9 ). This key observation demonstrated that domains were not an experimental artifact limited to stretched RBCs, but were probably a relevant feature of erythrocytes circulating in the blood. However, variation in size and close proximity between domains precluded rigorous analysis at this preliminary stage.

Overview
We here addressed the lateral organization of endogenous SM decorated by lysenin* in RBCs. Three major conclusions were derived. First, lysenin* is a new, reliable, versatile, and quantitative probe to directly study SM localization by confocal imaging, and may have broad applications in biology. In retrospect, this probe of endogenous SM validates our previous observations on SM submicrometric domains using insertion of exogenous BODIPY-SM into partially spread nonfi xed RBCs. This main conclusion, derived from two vital imaging tools, is also entirely consistent with recent evidence by highresolution methods in fi xed cells ( 8,31,38,41 ). Second, we provide basic mechanistic insights on domain composition and regulation, based on lipid phase transition (temperature), cholesterol, and membrane tension. Third, the abundance of SM domains in gently suspended (ovoid shape), nonstretched, and nonfi xed erythrocytes strongly supports their relevance for circulating erythrocytes in vivo.  ( 9,10 ). This procedure led to three patterns of RBCs, as in Fig. 7 ; only stage #1 (a-c) and stage #2 (d-f) RBCs are presented. Notice large colocalization between lysenin* and the three BODIPY-lipids (yellow arrowheads) on stage #1 RBCs, contrasting with larger dissociation on stage #2 RBCs (white arrowheads, domains decorated by BODIPYlipids only; red arrowheads, domains labeled by lysenin* only). Representative of at least three independent experiments for each BODIPY-lipid. Scale bars, 2 m. perturbation, if any. Indeed, higher lysenin* concentrations changed neither domain number nor area, but proportionally increased their labeling intensity, indicating that domains were not induced, but rather revealed, by lysenin*. Moreover, the single component in the Scatchard plot was compatible with monovalent lysenin*:SM interaction onto RBC plasma membrane. Furthermore, up to 7.5 M lysenin*, no detectable hemolysis was induced.

Estimation of the SM surface coverage by lysenin* and of the size of SM submicrometric domains
Comparing the surface projection of lysenin* [ ‫ف‬ 45 kDa; 10 nm 2 ( 47 )] with that of SM (0.7 nm 2 ), values above imply that <20% of the RBC surface occupied by SM was covered by lysenin*. The diameter of individual submicrometric domains we evidenced here by confocal microscopy in unfi xed, partially spread, featureless RBCs using BODIPY-SM and lysenin* is ‫ف‬ 500 nm (range: 350-950 nm at 1.25 M, 96 domains), close to the resolution limit of the method. More precise values have been recently provided by PALM, high-resolution ion mass spectrometry, or electron microscopy, but cells were not featureless and were formaldehyde-fi xed. PALM with Dronpa-NT-lysenin revealed SM domains of ‫ف‬ 250 nm in diameter on fi xed HeLa cells ( 31 ), but of ‫ف‬ 500 nm on fi xed LLC-PK1 cells ( 41 ). By electron microscopy, SM-rich domains labeled with lysenin on fi xed Jurkat cells were ‫ف‬ 150 nm in diameter ( 38 ). High-resolution ion mass spectrometry based on 15 N-SLs revealed domains of ‫ف‬ 200 nm in diameter in fi xed fi broblasts, interestingly regulated by cholesterol and actin cytoskeleton ( 8 ). While it is unclear whether these different values result from alternative methods or cells, all previous researches, combined with this study, have concurred to establish the concept of rather stable submicrometric SM domains on eukaryotic cells, thus conceptually distinct from transient nanometric rafts. However, it cannot be excluded that the larger level of organization of submicrometric domains could result from raft coalescence.

Control of SM submicrometric domains by temperature, cholesterol, and membrane tension
As previously shown for BODIPY-SM domains (9)(10)(11), the abundance of endogenous SM domains peaked at 20°C and decreased upon cholesterol depletion, suggesting liquidphase separation. This issue was already discussed for BODIPY-SM in ( 9 ). We will now focus on the role of cholesterol. The strong decrease of lysenin* domain abundance Selection of human RBCs as the simplest and most characterized cell system We selected human RBCs as the simplest and most robust eukaryotic cell system, for which membrane lipid composition ( 42 ), proteome ( 43 ), and cytoskeletal control ( 39 ) are extensively characterized. In contrast to other eukaryotic cells, the absence of vesicular traffi cking avoids any confusion with peripheral endosomes. Most importantly, scanning electron microscopy confi rmed the fl at RBC surface, i.e., lack of structural features such as membrane projections that are typical of most other cells. These membrane projections would artifi cially increase membrane area in confocal image pixels, because of superposition of perpendicular clustered objects, in the nonresolved thickness of the optical sections. Normal RBCs resist severe deformation (including the >10,000 "quality controls" across tiny splenic pores in their 120 day lifetime). Robustness and deformability of RBCs are attributed to high cholesterol content, that regulates lipid ordering and membrane fl uidity, as well as to strong plasma membrane anchorage to the underlying spectrin network via two nonredundant anchorage complexes, ankyrin and 4.1R complexes ( 39 ).

Lysenin* binding specifi city and validation as nonperturbing probe
We fi rst confi rmed the specifi city of lysenin* for SM, based on: i ) selective binding to pelletable liposomes when containing SM, but not when containing PC (which bears the same phosphocholine headgroup); and ii ) suppression by SMase of 125 I-lysenin* binding and lysenin* labeling of submicrometric domains. Taken together with previous observations ( 44 ), our data are consistent with recognition by the C-terminal domain of lysenin* of the linkage between the phosphocholine headgroup and the ceramide backbone of SM, as suggested for equinatoxin II, another SMspecifi c toxin ( 28 ).
In contrast to insertion of exogenous fl uorescent lipid analogs, toxins offer several advantages: i ) targeting of nonmodifi ed endogenous lipids; ii ) versatile coupling with fl uorescent proteins or fl uorophores; and iii ) successful labeling of cells in 3D-gels, which we failed to achieve using BODIPY-SM:BSA. However, only a few lipid-specifi c fl uorescent toxins are validated so far ( 45 ); and admittedly, their size is much greater than the targeted lipid (e.g., projected diameter of lysenin* is ‫ف‬ 15 times larger than endogenous SM). This size discrepancy predicts steric hindrance but does not preclude specifi city, as perhaps best exemplifi ed by EGFferritin conjugates for which grafting a ‫ف‬ 450 kDa ferritin moiety allowed us to faithfully follow the fate of the small EGF molecule [ ‫ف‬ 6 kDa; ( 46 )]. Complementarity between lipid analogs and lipid-specifi c toxins, when available, is illustrated by this paper.
Binding isotherms based on isotopic dilution of 125 I-lysenin* allowed us to derive the fractional level of SM occupancy by comparison to total SM, estimated at ‫ف‬ 10 8 per RBC ( 26 ). At low lysenin* concentration (1.25 M), sufficient for adequate labeling of SM domains, ‫ف‬ 1.3 × 10 6 lysenin* molecules were bound per RBC, i.e., ‫ف‬ 1% of total SM, demonstrating tracer occupancy and suggesting minimal

Biological relevance of SM submicrometric domains
Admittedly, most of the observations reported here have been obtained using RBCs spread onto poly-L-lysine coverslips, up to almost a fl at two-dimensional rigid system which is far from the biconcave shape and plasticity of RBCs in the circulation and in particular during splenic fi ltration. However, imaging of living RBCs gently suspended in a 3D-gel, thus without artifi cial stretching, not only confi rmed the existence of micrometric domains but revealed that their abundance is underestimated in the classical coverslip assay, even though this approach yielded essential results. If abundant submicrometric domains turn out to be the rule in circulating RBCs, and perhaps many other cell types, what could be their function? Two opposite roles may be considered. First, submicrometric lipid domains could promote lipid resilience necessary to membrane deformability by providing stretchable membrane reservoirs, by analogy to caveolae in endothelial cells ( 50 ). This function would be essential for RBC squeezing into the narrow pores of spleen sinusoids. Conversely, boundaries of submicrometric lipid domains might refl ect high-tension fragility sites, i.e., propensity to fragmentation and hemolysis during splenic fi ltration of senescent or diseased RBCs. These opposite potential properties might even coexist, and one or the other might prevail, depending on the type of lipid domains and the intensity of stretching/deformation. Testing these hypotheses would also depend on biophysical studies, which could not only exploit, but also possibly explain and help in managing RBC membrane fragility diseases. upon moderate cholesterol depletion, which was also found with 15 N-SLs on fi broblasts ( 8 ), can be explained neither by a modifi cation of SM level (unchanged as checked by thin layer chromatography; data not shown), nor by a decrease of lysenin* binding to SM (unchanged as shown by 125 I-lysenin*), and pointed instead to a modifi cation of SM lateral organization. Our approach did not allow us to distinguish whether moderate cholesterol depletion fragmented SM domains into smaller assemblies below the detection limit of conventional confocal microscopy or caused SM randomization over the plasma membrane. We can however conclude that cholesterol is a major stabilization factor for SM submicrometric domains. We suggest that cholesterol abundance and dynamic distribution along the plasma membrane must play a crucial role in the dynamic regulation of domain biogenesis (nucleation) and/ or expansion (size), as for transient nanometric rafts.
A third regulator of submicrometric domain is membrane tension, known to be itself regulated by cholesterol ( 11,49 ). In this study, membrane tension was manipulated in two ways: i ) increase by stretching on coverslips up to an almost two-dimensional object; versus ii ) decrease by gentle suspension on a gel, so as to better preserve their 3D ovoid shape. In full agreement with our previous observations based on exogenous BODIPY-SM insertion ( 11 ), endogenous SM labeling by lysenin* disclosed a decrease then disappearance of submicrometric domains on coverslips, from the loosely attached stage #1 to maximally stretched stage #3. Conversely, endogenous SM domains were most abundant on more relaxed RBCs suspended in 3D-gel. We confi rm that endogenous SM domains were stable in time and space, as shown by FRAP, and were also restricted by strong membrane spectrin anchorage (data not shown). Biophysical studies are clearly required to measure the relation between elastic membrane tension and submicrometric domains. These could benefi t from acute dissociation of 4.1R complexes (pharmacological) or genetic defi ciency of ankyrin complexes in spherocytosis.

Relation with other polar lipids
We previously suggested the segregation of at least three different classes of domains, respectively enriched in GSLs, SM, and PC ( 10 ). However, red BODIPY-SM was not available, precluding colocalization experiments between SM and other lipid classes (PC, GM1, GlcCer), and the bulky red BODIPY may be more prone to artifacts. These two major limitations could now be circumvented by the combination of red lysenin*, validated above, with green BODIPY-lipids. We here found that lysenin* domains showed large coenrichment with green BODIPY-PC and -GSL (GM1 and GlcCer) domains on poorly stretched RBCs (stage #1), but large segregation upon increased stretching (stage #2). This observation suggests that the erythrocyte membrane is not organized in three different classes of submicrometric domains as previously suggested ( 10 ), but is rather composed of domains that can be spatially related, depending on intrinsic cohesiveness of individual lipids (melting temperature) and the extent of membrane stretching.