The Fluid Membrane determines Mechanics of Red Blood Cell Extracellular Vesicles and is Softened in Hereditary Spherocytosis

Extracellular vesicles (EVs) are widely studied regarding their role in cell-to-cell communication and disease, as well as for applications as biomarker or drug delivery vehicle. EVs contain both membrane and intraluminal proteins, affecting their structural properties and thereby likely their functioning. Here, we use atomic force microscopy for the mechanical characterization of red blood cell (RBC) EVs from healthy individuals as well as from a patient with hereditary spherocytosis (HS) due to ankyrin deficiency. We show that the EVs are packed with proteins, yet their response to indentation is similar to that of a fluid lipid vesicle lacking proteins. The bending modulus of RBC EVs of healthy donors is ~15 kbT, agreeing well with the bending modulus of the RBC membrane. Surprisingly, whereas RBCs become more rigid in HS, the excreted vesicles of a patient with this blood disorder have a significantly (~50%) lower bending modulus than donor EVs. These results shed new light on the mechanism and effects of EV budding and may underlie the reported increase in vesiculation and stiffening of RBCs in hereditary spherocytosis patients.


Introduction
Extracellular vesicles (EVs) are released by many cell types in vitro and in vivo and are present in most body fluids. They originate either from internal cellular organelles called multivesicular bodies (i.e. exosomes) or are shed directly from the plasma membrane (i.e. microvesicles). 1,2 They are suggested to play a prominent role in cell-to-cell communication as intercellular transport vehicles carrying proteins and RNAs. 3 They have also been suggested to play a role in immune responses 4 and cancer progression. 5,6 EVs from red blood cells (RBCs) are released both in vivo as well as under blood bank storage conditions. 7 A human red blood cell typically sheds 20% of its membrane area over its lifetime. 8 RBC EVs have been suggested to postpone the clearance of RBCs by the immune system 9 , to play a role in blood clotting 10 and they have been put forward as a potential biomarker for dengue virus infections. 11 In patients with RBC infections such as malaria, the number of EVs is often elevated. 12,13 During malarial infection, RBC EVs were demonstrated to facilitate communication between malaria parasites. 14,15 RBC vesiculation is also relevant in blood disorders, for example in Hereditary spherocytosis (HS).
HS is one of the most common hereditary RBC disorders in the western world and is accompanied by increased release of EVs. This vesiculation is caused by a reduced linkage between the membrane and the underlying cytoskeletal spectrin network leading to loss of RBC membrane. 12,[16][17][18] Reduced membrane surface area results in the formation of spherocytes.
Spherocytes are less deformable, which causes these cells to be retained and cleared by the spleen.
However, the relation between this blood disorder and EV release is still poorly understood as we are lacking structural and mechanical insights into EVs properties of both donors and patients.
These structural and mechanical properties of EVs influence their behavior, such as their interactions with cells. [19][20][21][22] Therefore, there has been a strong interest in quantifying EV mechanical properties 23,24 . At present, however, characterization of EVs is challenging due to their small size and a proper mechanical characterization of (RBC) EVs under physiological conditions has not yet been performed. Recently, we showed that the mechanics of small (< 200 nm) synthetic vesicles, i.e. liposomes, are accurately described by a quantitative model based on Canham-Helfrich theory. [25][26][27] However, the mechanical properties of natural vesicles will result from a combination of lipidic contributions as well as that of membrane proteins and intraluminal proteins. The presence of such proteins could lead to a nonzero shear modulus of the natural vesicles, which might therefore be better described as thin elastic shells, and hence potentially show typical behavior thereof, such as buckling. 28 Moreover, thin shell behavior could be caused by spectrin structures, which is known to provide shear resistance to the RBC membrane. [29][30][31] A high percentage of the RBC membrane is occupied by membrane proteins (~20 % at the hydrophobic core). 32 The effect of these proteins on membrane mechanics is not entirely clear; Nano-indentation studies with vesicles reconstructed from yeast membranes and influenza viruses both suggested that membrane proteins result in a large increase of the bending modulus of vesicles. 33,34 On the other hand, studies with model membranes have mostly reported membrane softening due to the presence of short peptides in the membrane. 35 Similar experimental studies with various larger membrane proteins showed either a neutral effect on membrane stiffness, 36 or a decrease in the membrane stiffness. 37 Recent simulations reconcile some of these results by reporting how integral membrane proteins can have diverse effects on membrane mechanics, either softening, neutral or stiffening. 38 Here, we use atomic force microscopy (AFM) nano-indentation for mechanical characterization of RBC EVs. We use quantitative image analysis and show that RBC EVs remain in a rather spherical shape upon adhesion to the sample surface. Protein analysis and imaging of collapsed vesicles show that vesicles contain significant amounts of membrane-associated proteins.
Interestingly, most EVs behave like empty liposomes with a fluid bilayer. We find that the bending modulus of healthy donor RBC EVs is ~15 kbT, which is similar to previously reported values for liposomes 27 and which also corresponds well to bending moduli found in studies of the RBC membrane. [39][40][41] These results were compared to EVs derived from RBCs from a patient with HS due to a mutation in the ANK1 gene. We reveal that these patient-derived EVs have an altered protein composition and a significantly softer membrane. This lower EV bending modulus could directly relate to the increased rate of vesiculation in HS patients.

Results
To study the mechanical properties of EVs, first the vesicles are imaged at high resolution to determine their geometry and center. RBCs derived from healthy volunteers were treated with a Ca 2+ ionophore to stimulate EV formation. RBC EVs were subsequently attached to poly-l-lysine coated surfaces and imaged using atomic force microscopy (Fig. 1A). Due to electrostatic interactions with the surface, vesicles spread onto the surface. High resolution imaging revealed that EVs stay in a fairly spherical shape and have a homogeneous appearance. We applied a correction for the tip shape (Fig. 1B) and a correction for deformation due to applied imaging forces (See materials & methods). Subsequently we quantified the shape of EVs by measuring the height over the radius of curvature. For the 2 donor samples this ratio was ~1.5 (Fig. 1C), indicating some spreading onto the surface. The difference in spreading between the samples can be attributed to either surface preparation or variation between the two donors. Next, we calculated the initial radius of the vesicles, assuming surface area conservation, finding very similar values 88 ± 2 nm (SEM, N = 72) and 88 ± 3 nm (SEM, N = 55) (Fig. 1D). This is slightly higher than the average radius found during nanoparticle tracking analysis (NTA), which gave R0 = 71 ± 1 nm (Fig. 1D, inset).
The mechanical behavior of the vesicles was studied by analyzing force indentation curves (FDCs), which were obtained during indentation experiments. The curves were obtained by moving to the center of the EV and applying a force of 2-10 nN. Typically, FDCs revealed linear behavior and a subsequent flattening. Finally, a significant increase in force and two discontinuities can be observed, which correspond to the compression and penetration of both lipid bilayers ( Fig.   2A). Some vesicles showed a much softer response, but with similar characteristics (Fig. 2B). A linear response is consistent with elastic behavior dominated by the physical properties of the membrane and indicates that the interior of the vesicle is not packed with polymerized proteins, in which case a Hertz-like behavior with a superlinear force-indentation response (exponent α of the FDC 1.5) would be expected.
Sometimes large irreversible break events could be observed (Fig. 2C), which usually led to collapsed EVs. Previous reports on liposomes and natural vesicles have described a complete recovery after deep nanoindentations, with no detectable change in the geometry of the vesicles 33,34 . Here, we observed EV collapse in ~40% of cases for both donors. The shape of the non-collapsed EVs was either similar to before indentation or more flattened. Typical collapsed EVs are shown in Figure 3. Some collapsed EVs appear as flat structures, with heights of 15-35 nm (Fig. 3A, Suppl. Fig. 1). Other EVs show elevated halo like edges (Suppl. Fig. 1), with similar maximum heights. Yet other EVs show partly elevated flat structures (Fig. 3B) or more complex structures (Fig. 3C,D). We compared the height of these structures with that of the lipid bilayer, which we measured when it was partly exposed at 4.1 ± 0.2 nm (SEM, N = 9), indicating that the observed structures are much larger. In fact, the recorded structures resemble proteins and aggregates thereof observed on the RBC inner cell membrane. 42,43 Their height corresponds well to previous observations of proteins sticking out up to 10 nm above the inner membrane. 42 This suggests that the vesicles in Figure 3B-D break at least partially open and expose their inner membrane, whereas the vesicle in Figure 3A likely stayed intact. The ruptured EVs show that the membrane of RBC EVs contains a significant amount of proteins.
To investigate protein content, RBC ghosts and EV proteins were subjected to gel electrophoresis.
RBC ghosts consist of the RBC membrane and proteins associated with the membrane. RBC ghosts can be isolated using hypotonic shock. In that way, hemoglobin levels are reduced dramatically and membrane associated proteins are enriched. In addition, RBC lysates were used to identify hemoglobin and other cytosolic proteins. RBC proteins have been studied extensively, which makes protein identification possible without immune staining. 44 From electrophoresis we see that hemoglobin and band 3 are present in the EVs, as well as small amounts of protein 4.1 and 4.2 (Fig. 3E). In contrast to previous observations, [45][46][47] we also find small amounts of spectrin.
Actin however, seems to be absent in the EVs. Furthermore, we find that the membrane protein stomatin is enriched in EVs compared to RBCs, which agrees with previous findings. 47 This indicates that the EVs do have a distinct protein content compared to their donor cells, and that they do contain cytoskeletal elements and membrane proteins that might play a role in their mechanical properties. We also identified and quantified the lipids in RBC EVs, with thin layer chromatography (TLC). Using the method developed by Yao and Rastetter, 48 we were able to separate and visualize seven (phospho)lipid classes. Analysis revealed that lipid composition of EVs was very similar to the native RBC membrane, with only a modest increase in phosphatidylserine (PS) in the vesicles (Suppl. Fig. 2).
Recently, we investigated the mechanical behavior upon indentation of liposomes and found excellent correspondence with a Canham-Helfrich theory 25,26 based indentation model. 27 A signature of this behavior is a flattening of the FDC, marking the onset of formation of an inward directed lipid tether, and occurring at an indentation of 0.35 -0.40 Rc, where Rc is the radius of curvature of the vesicle. In contrast, in a thin elastic shell model, buckling is predicted to soften the response, which is expected to occur at smaller (~ 0.05 Rc) indentations for shells with the geometry of a vesicle. 33 We determined the inflection point of the FDCs of the donor samples from the peak in the derivative of smoothed FDCs (Fig. 4B, inset). The obtained distributions for the donor samples were centered at ~0.4 Rc, close to the previously predicted value by our quantitative model based on Canham-Helfrich theory (0.35 -0.4 Rc) 27 (Fig. 4B). The good agreement with our model suggests that the bending behavior of the RBC EVs is dominated by a fluid membrane and that the membrane skeleton and membrane proteins are not resulting in a significant membrane shear modulus.
Next, we set out to estimate the membrane bending modulus, an important intrinsic mechanical property of the membrane. We previously showed that the mechanical properties of adherent vesicles can be understood in terms of membrane bending and internal osmotic pressure, and that the bending modulus can be assessed using the vesicle stiffness, radius and tether force. 27 The EV stiffness was determined by fitting the linear response for indentations up to 0.1 Rc (Fig. 4C). The two donor samples had different stiffness; respectively 10.9 ± 0.5 mN/m (SEM, N = 72) and 5.8 ± 0.4 mN/m (SEM, N = 55). To estimate the pressure over the membrane, the retrace of indentation curves was analyzed. A tether, marked by a force plateau with force Ft, was detected during the retrace in ~60% of FDCs (donor 1: Ft = 130 ± 10 pN, SEM, N = 49; donor 2: Ft = 100 ± 10 pN, SEM, N = 25) (Suppl. Fig. 4). The tether force equals = 2 √2 , with σ the tension in the membrane and the bending modulus of the membrane. [49][50][51] We can subsequently estimate the pressure in the membrane using the Young-Laplace equation: = 2 / , with the osmotic pressure difference over the membrane. With these measurements, we used our recent model to fit the bending modulus of the vesicles. 27 There was no significant difference in bending modulus between donor 1 and 2 (Suppl. truncation is expected to result in a disturbed cytoskeletal network and its connection to the plasma membrane. In turn this leads to increased vesiculation and hence a reduced ratio of surface area to volume, resulting in spherocytic cells (Fig. 5A). Such RBCs are poorly deformable and will be cleared prematurely from the blood circulation, leading to hemolytic anemia. Reduced RBC deformability was confirmed by laser diffraction ektacytometry (Fig. 5B).
RBCs from the HS patient were stimulated with Ca 2+ ionophore to induce EV formation. The lipid and protein composition of patient EVs were compared to that of EVs from the healthy donor RBCs. Analysis of the phospholipid classes using TLC revealed a similar pattern in the donor and patient derived EVs, with the patient EVs having slightly more cholesterol and PS, and less phosphatidylethanolamine (PE) (Suppl. Fig. 2). To detect potential differences of specific lipid species we used matrix-assisted laser desorption and ionization mass spectrometry (MALDI-TOF MS). We found a mostly similar lipid composition between donor and patient derived EVs, with the only notable difference being a small shift in sphingomyelins (SM) towards species with a longer chain length in patient EVs (Suppl. Fig. 7). Interestingly, we found reduced levels of α1spectrin, ankyrin and actin in patient EVs, while expression of these proteins is equal between healthy and patient RBCs (Fig. 5C). Tubulin levels, however, were increased in patient EVs, which reflects the increased expression of tubulin in the RBCs. The increased level of tubulin in the RBCs from the HS patient could be due to high reticulocyte numbers in this patient. Increased numbers of reticulocytes are a hallmark of HS indicating increased activity of the hematopoetic system trying to compensate for the hemolytic anemia. During maturation to erythrocytes, reticulocytes lose their tubulin. 52 Overall, the protein content of the patient EVs had a distinct pattern when compared with the healthy donor sample (Suppl. Fig. 6).
We repeated the mechanical measurements for the HS derived EVs. These EVs show similar size (R0 = 93 ± 3 nm, SEM, N = 74), however they appear somewhat more flattened than the two donor samples with H/Rc = 1.32 ± 0.03 nm (SEM, N = 74). In total 74 particles were indented (Suppl.  Fig. 4). The tether force is significantly lower than for the two healthy donor samples, whereas the spread-out shape would suggest high pressurization and hence high tension in the membrane. This suggests a lower bending modulus of the membrane. Indeed, the bending modulus of patient derived EVs is significantly lower than of EVs derived from both healthy donors with 7 (5 -8, 68% bootstrapping c.i.) kbT ( Fig. 5D and Suppl. Fig. 7). Hence, the mechanical properties of the HS patient derived EVs are significantly different compared to healthy donor derived EVs.

Discussion
In this study, we investigated the mechanical properties of EVs from RBCs. The large percentage of membrane area occupied by membrane proteins (~20% at the hydrophobic core), recently raised questions regarding the fluidity of natural membranes, both for synaptic vesicles 53 and the RBC membrane 32 . Perhaps surprisingly, we show that the mechanical behavior of RBC EVs agrees well with the theoretical behavior of a vesicle consisting of a fluid lipid bilayer without proteins. We show this by comparison of our experimental data with our recent model, based on Canham-Helfrich theory, 25,26 for indentation of small (<200 nm) vesicles. 27 In particular, the measured inflection point in the FDCs corresponds well to the value predicted by our model (Fig. 4B), and the indentation data collapse on the predicted reduced pressure vs. stiffness curve (Fig. 4D).
Furthermore, even the bending modulus of the EVs of the two donors in this study is similar to the bending modulus of liposomes with complex lipid mixture (designed to mimic the RBC lipid composition) obtained previously. 27 This suggests that the net effect of the various membraneassociated proteins in RBC EVs on the bending behavior of the membrane is neutral. This is different from the conclusions by Calo et al., 33  For two healthy donor RBC EVs, we found that the bending modulus is ~15 kbT. Our estimate is comparable to the bending modulus found for the RBC membrane using flicker spectroscopy. [39][40][41] However the bending modulus of the RBC membrane has been estimated to be higher using micropipette aspiration (~40 kbT) 54 and optical tweezers assisted measurements of membrane fluctuations (~60 kbT). 55 The variability in the observed values for the RBC membrane makes it hard to directly compare the RBC EV bending modulus with that of the RBC membrane, but since the membrane modulus does not appear to be strongly affected by membrane proteins and the lipid composition of the EVs is comparable to the RBC membrane, the RBC membrane and RBC EV bending moduli are likely similar.
Interestingly, we find that EVs from a patient with HS due to ankyrin deficiency show a ~2 times lower bending modulus than both donor derived samples. This is unlikely to be caused by differences in lipid composition, because we found only subtle differences in lipid composition between patient and donor derived EVs (Suppl. Fig. S7, Suppl. Table S1). Moreover, the small differences that were detected, include higher cholesterol content and a shift to longer chain length sphingomyelins in patient EVs. Both these observationslong chain length SM through thickening of the lipid bilayer and increasing the coupling between the two bilayer leaflets 56would be more compatible with stiffening of the membrane than the observed softening. 57 We do, however, observe significant differences in donor and patient EV protein content. This makes it conceivable that EV proteins, rather than phospholipids, are likely to cause the 2-fold decrease in the EV bending modulus in HS.
RBCs from patients with HS are less deformable (Fig. 5B) and stiffer than healthy RBCs. 18 It was previously found that the rate of vesiculation is increased in HS patients. 12 This is thought to be the primary reason of the rounding and stiffening of HS RBCs, which ultimately leads to clearance from the circulation, resulting in hemolytic anemia. 12,[16][17][18] Excretion of the softer part of the membrane would leave behind the part of the membrane with a higher bending modulus, contributing to the stiffening of the RBCs, which is fully consistent with higher bending modulus estimates from the membrane of spherocytes. 41 In this way, the excretion of EVs with low bending modulus as reported here seems to be a mechanism that contributes to stiffening of spherocytic

RBCs.
Our results also provide new insights into the vesiculation process in HS. Suggested mechanisms for membrane shedding by RBCs are related to clustering of membrane proteins driving curvature generation 46,47,58 and to the balance between membrane bending and stretching of the spectrin cytoskeleton. [58][59][60] The weaker linkage of the membrane with the underlying cytoskeleton in HS likely causes loss of organization in the membrane. This is supported by the observation that diffusion of membrane proteins is faster in RBCs from patients with HS. 61 We speculate that loss of organization of the membrane gives room to accumulation of specific membrane proteins that lower the bending modulus locally and hence the energy barrier for vesicle formation. In this way, the loss of membrane linkage in HS could lead to increased vesiculation.
To conclude, we have presented an innovative approach to characterize the structural and material Electrophoresis and immunoblotting. Proteins were quantified using BCA (bicinchoninic acid) analysis (Thermo Scientific) and equal protein amounts were subjected to gel electrophoresis, as indicated. Proteins were either blotted onto PVDF membranes (Merck Millipore) or stained in the gel using PageBlue (Life Technologies). Antibodies for western blot were purchased from Sigma and Abcam. Blots and gels were imaged using an Odyssey imager (LI-COR).
Lipid extraction and quantification. Lipids were extracted from the samples using the Bligh and Dyer method. 63 Samples were diluted 7 times in a 2:1 methanol:chloroform (v/v) mixture. Samples were vortexed, after which chloroform and distilled water were added till a final ratio of 1   Fig. 8). For calculation of R0 a minimum radius of curvature of 5 nm was assumed at the contact between the vesicle and the underlying surface, since a sharper contact angle would be non-physical 65 .
AFM FDC analysis. Cantilever response was measured on the sample surface and fitted linearly.
The resulting fit was subtracted from the measured response when indenting vesicles, to obtain FDCs. Contact point was found by using a change point algorithm 66  Confidence intervals were estimated using the bias corrected percentile method with 500 bootstrapping repetitions, for which a set of observed value combinations equal in size to the original data set was randomly drawn and fitted.
Nanoparticle Tracking Analysis (NTA). EVs were sized by recording 5 videos of 60 s using the NanoSight LM10 system (Malvern Instruments). A camera level of 11 was used and videos were recorded at 22°C. Analysis of the videos was performed using the NTA 2.0 software, using default settings. Threshold was set at 5.

Mass Spectrometric Analysis
All samples were extracted prior to mass spectrometric characterization according to the method developed by Bligh & Dyer. 63 All MALDI mass spectra were recorded with DHB (0.5 M in methanol) as the matrix. 67 All lipid samples of interest were mixed 1:1 (v/v) with the matrix solution and subsequently applied onto the MALDI target under slight heating with a conventional hairdryer that helped to improve the homogeneity of crystallization.
All mass spectra were acquired on an Autoflex I MALDI mass spectrometer (Bruker Daltonics, Bremen, Germany) with ion reflector. The system utilizes a pulsed 50 Hz nitrogen laser, emitting at 337 nm. The extraction voltage was 20 kV and gated matrix suppression was applied to prevent the saturation of the detector by matrix ions. All spectra were acquired in the reflector mode using delayed extraction conditions.
Spectral mass resolutions, signal-to-noise (S/N) ratios, and peak intensities were determined by the instrument software Flex Analysis 3.0 (Bruker Daltonics). The mass spectrometer was calibrated using a lipid mixture of known composition.
Due to the small lipid amounts in the vesicle samples and their largely unknown compositions, no internal standards were used and this is the reason why only relative changes can be given, i.e. a lipid "profiling" approach is taken. Although the negative ion mass spectra were also recorded, changes of the (phospho)lipid compositions are most obviously reflected by the positive ion spectra.  and the first one is typically larger than the second. B) Similar behavior, but a much softer response. Inset shows the smoothed data zoomed on the initial regime, where we can see that the initial linear response softens (black arrow). Note that this response is fully reversible. C) A FDC with a large discontinuity, after which the particle is ruptured (in blue). All FDCs were obtained with EVs from donor 1 RBCs.   shear stress than RBCs from a healthy donor. C) 30 µg RBC or EV protein was subjected to electrophoresis. After blotting the proteins, α1-spectin, ankyrin, band3, β-tubulin and β-actin were detected using immunoblotting. D) Comparison of the bending modulus of EVs from two healthy donors and the patient. Error bars mark 68% confidence intervals determined by bootstrapping.