Visualizing dynamics of membrane rafts on live cells

Abstract

Membrane rafts are cellular portals to external stimuli that trigger signaling cascades for sophisticated yet remarkable biochemical activities. Visualization of the topographic evolution of membrane rafts remains unreported on live cells due to the nanosized and dynamic nature. Here, an imaging strategy involving atomic force microscopy and Hadamard product is developed to unveil membrane-raft features. Michigan Cancer Foundation-7 (MCF-7) cells were subjected to fibrinogen or manganese(II) (Mn²⁺)/resveratrol, both of which are ligands of integrin α_Vβ₃ embedded within membrane rafts; the former promotes metastasis, and the latter enables apoptosis. MCF-7 cellular membranes responded to the two stimulants markedly different. The size, height, spatiotemporal trajectory, and persistent time of ligand-activated nanodomains are revealed. This approach opens up a visualized platform toward the understanding of activation-associated signaling cascades.

The size, height, trajectory, and duration of activated integrins—key portals in cell signaling—are visualized on live cells.

INTRODUCTION

The cell membrane gates extracellular stimuli, which drive membrane protein trafficking (1, 2) and trigger signaling cascades (3, 4), leading to subsequent cellular processes (5). The membrane raft hypothesis by Simons and Ikonen (6, 7) postulates the involvement of heterogeneous self-organized nano-sized membrane domains containing specific lipid-protein and protein-protein interactions in such gating (8, 9). Accordingly, understanding the nanoscale dynamics of membrane rafts is one of the first steps to delineate the corresponding behavioral patterns and to comprehend fundamental cellular mechanisms. The membrane raft nanodomains are abundant in cholesterol and saturated lipids (such as sphingolipids), ceramide (10, 11), and proteins, exhibiting a liquid crystalline bilayer [termed liquid-ordered phases (L_o)], whereas the surroundings contain less cholesterol to form liquid-disordered phases (L_d) (7). Many cellular mechanisms implicated in the raft hypothesis would be unveiled by direct observation of the lateral heterogeneity with spatiotemporal information. Membrane rafts are widely accepted to be 10 to 200 nm (12) based on extensive studies on reconstituted model membranes extracted from cells (13, 14). Conventional optical microscopic approaches to visualize the membrane dynamics remain a great challenge because the required resolution is smaller than the Abbe diffraction limit of ~200 nm. Because of the lack of topographic images demonstrating distinct nanoscopic features portrayed in the raft hypothesis, the very existence of rafts has been under active debate (15). Efforts to resolve this controversy have involved emerging spectroscopies or microscopies and advances in spatiotemporal resolution (16). For example, the use of extrinsic fluorescent molecules casts doubts on relevant measurements due to unexpected interactions of these molecules with the raft components. To this end, recent super-resolution fluorescence microscopic (SRFM) studies on live cells have enabled the direct observation of membrane rafts using two types of dyes independently. One type is associated with the localization of cholesterol and reveals the membrane heterogeneity of the nanodomains < 40 nm (17). Contreras and co-workers (17) developed a sterol-derivatized fluorescently labeled probe with a spacer between the sterol moiety and the dye to preserve membrane integrity. Kemmoku et al. (18) visualized heterogeneously distributed cholesterol-rich domains by iD4H, a cholesterol sensor produced from Clostridium perfringens theta-toxin (19). The other type is sensitive to the polarity of local environments, such as Laurdan (17) and di-4-ANEPPDHQ (18). Their emission wavelengths depend on the membrane lateral packing, validating no SRFM probe-induced interchange between L_o and L_d phases. Alternatively, deuterated lipids behave indistinguishably from the pristine ones, yet enable in vivo small-angle neutron scattering (SANS) measurements. Nickels et al. (20) studied a Gram-positive bacterium Bacillus subtilis by SANS and determined a domain size of <40 nm for membrane rafts, providing one of the first glimpses for the membrane raft size without any potential perturbation by fluorescent molecules. Currently, no single technique can escape its inherit methodological weakness, but the strengths of each approach can provide important aspects and collectively advance our understanding of membrane rafts (16, 21). Nonetheless, it is still highly desirable to develop imaging tools that require minimal experimental preparation and confer self-explanatory images with the morphological dynamics of membrane rafts on live cells. Currently, membrane raft height has only been estimated in cryo–electron microscopy (cryo-EM) experiments (22, 23), but membrane raft persistent time (i.e., lifetime) has yet to be estimated in any capacity based on experimental observations.

Theoretical modeling can be viewed as microscopy in silico (24, 25), enabling the visualization of plasma membranes with unparalleled spatiotemporal details compared to the finest available experimental techniques. Modeling is effective in characterizing raft size, lateral organization, and spatial distribution to explain raft related processes. For example, Fig. 1 displays simulation results summarized and illustrated by Fan et al. (26–28). Denoted above the snapshot panels are the physical models. These models were mainly based on a nonlinear diffusion equation with stochastic Gaussian noise for the lipids within the exoplasmic leaflet (28). Panel A was calculated for membranes under thermodynamic equilibrium near the miscibility critical point (T_critical), in which lateral heterogeneity was driven by thermal noise. Postulations for panels B and C were the presence of immobile proteins and attractive interactions between rafts, respectively. Panels D to F were calculated in which membrane components recycled to the cytoplasm. The distribution patterns were somewhat discernible, providing a rough blueprint for the different raft formation mechanisms. On the basis of these patterns, experimental topographic images may be used to determine the operating mechanism.

Fig. 1. Illustration of membrane rafts organized by theoretical mechanisms.

The snapshots were calculated by concisely denotative physical models of (A) transient compositions due to thermal fluctuations around the critical point (T = T_critical) (26, 63), (B) pinning by immobile membrane proteins (26), (C) strongly clustered proteins (27), (D) coupling of the membrane with cytoplasmic compositions (26, 64), stochastic recycling of lipids at (E) miscible (T > T_critical) (26), and (F) immiscible (T < T_critical) (26, 27) compositions. The membrane rafts (i.e., the L_o phase) are colored in black. The top row represents thermal equilibrium processes in which the thermodynamic forces dictated the formation and structures of membrane rafts (28). The snapshots in the bottom row were calculated by nonequilibrium cellular processes involving interactions of the membrane with the interior of the cell (26). For details, please see a review article by Fan et al. (26) from which Fig. 1 is adapted and modified.

Atomic force microscopy (AFM) can be used to obtain topographic information about cell membranes, enabling direct imaging under physiological environments (29–32) with nanometer-level spatial resolution. Typical AFM sample preparation is relatively simple without the need of fixing, staining, or labeling. However, the lack of chemical specificity makes the assignment and interpretation of the morphological details very challenging. Given the presence of thousands of lipid types and proteins in the cell membrane, the images exhibit overwhelming amounts of uncharacterizable and indistinguishable components. Hence, the heavily loaded information is unexplainable by analyzing the morphology alone. To reduce the complexity of membrane rafts, AFM imaging has been carried out on reconstituted cell membranes (33) or artificial model membranes (14, 33), composed of cholesterol and limited types of lipids, such as 1,2-dipalmitoyl-sn-glycero-3-phosphocholine, dioleoylphosphatidylcholine, and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphochline. These protocols allow unequivocal interpretation of the AFM images but at the loss of membrane proteins and in the absence of cellular bioactivities. Here, we present an AFM study of membrane rafts on live cells. AFM can simultaneously probe spatial information about the topography and mechanical stiffness of the cell surface (29, 30, 34). To better comprehend the images, the higher and stiffer characteristics of membrane rafts are accentuated by the operation of the Hadamard product (vide infra), which suppresses nonmembrane raft features (e.g., higher/softer and lower/stiffer). To remedy the chemical specificity shortcoming of AFM, we administered stimulants such as fibrinogen (35) and Mn²⁺/resveratrol (36, 37) to live Michigan Cancer Foundation-7 (MCF-7) cells to bind and activate integrins, transmembrane cell surface receptors that transduce biochemical and mechanical signals between the extracellular matrix and cytoskeleton (38). Fibrinogen binds to activated and unactivated integrins (α_Vβ₃ or α₅β₁) (38–45), whereas resveratrol binds to Mn²⁺-activated integrin α_Vβ₃, which most likely resides within the membrane raft (46–48). Hence, the downstream signaling pathways should be very different (36–38). Accordingly, the corresponding changes in size and height in the AFM images may be attributed to integrin-containing membrane raft dynamics. The AFM images resemble some of the patterns presented in Fig. 1, suggesting a strong correlation of the stimulant-triggered mechanisms to the simulation-based models.

RESULTS

Identification of membrane rafts on cell surface by AFM

Live MCF-7 cells were interrogated by FastScan AFM, which enabled the tracking of membrane raft motions with a temporal resolution of ~30 s per frame, about six- to 20-fold faster than that of typical AFM imaging with comparable quality. Limited by the specifications of our instrument, the largest imaging area was 36 μm by 36 μm, and hence, a complete MCF-7 cell could not fit into one single image. Other AFM microscopes can typically scan over a square of 90 μm by 90 μm, available for viewing whole cells (e.g., fig. S1) but cannot perform fast scan imaging. AFM imaging modes pertinent to this study are succinctly described in the Supplementary Materials. In fig. S2, the cytoskeleton went from upper left to lower right and was dynamically drifting and restructuring. For Fig. 2B (_i to _iii), multiple modes were operated simultaneously over the same cellular location, which was zoomed in for Fig. 2A. Both height mode (Fig. 2B_i) and amplitude error (Fig. 2B_ii) yielded topographic information. Amplitude error (in millivolts) showed higher spatial resolution, although the imaging mechanism granted height mode (in nanometers) straightforward interpretation of the surface morphology. The phase angles of phase mode images (Fig. 2B_iii) indicated different degrees of surface hardness. In principle, the advanced or lagged degrees in phase cannot be specifically assigned to a hard or soft surface. Nonetheless, the shape of the cytoskeletons could be readily recognized, and the response in phase would be the benchmark for a hard surface. In the case of Fig. 2B_iii, the cytoskeleton appeared phase advanced, and thus features with advanced and lagged phases were considered relatively hard and soft, respectively. We deposit in the Supplementary Materials a set of images (fig. S1) acquired by PeakForce QNM equipped with a 90-μm scanner. The results provided unambiguous quantitative information of surface hardness, although the image quality and scan rate were insufficient to track the dynamics on the cellular membrane.

Fig. 2. Raw images of a live MCF-7 cell and the operation of Hadamard product to manifest features higher and stiffer than their surroundings.

(A) Image at the cell edge. The gray semicircle at the top right was the petri dish substrate. (B_i to B_iii) Synchronically acquired FastScan AFM images at the same location by (B_i) height mode, (B_ii) amplitude error, and (B_iii) phase mode. (C) Operation scheme of Hadamard product. (D) Image after the operation of Hadamard product for raw data of (B_ii) and (B_iii). Red circles exemplify high and stiff features, while the yellow ones are either high but soft or stiff but low. Conditions: solution: phosphate-buffered saline (PBS) (10 mM phosphate, pH 7.4); scan rate: 51 s per frame; and drive frequency: 110 to 130 kHz. Image size: (A) 10 μm by 10 μm and [(B) and (D)] 5 μm by 5 μm.

Sterol- and sphingolipid-enriched domains (12) are higher and stiffer than the surrounding membrane. Accordingly, features that appeared higher in Fig. 2B_ii and stiffer in Fig. 2B_iii were considered to include membrane rafts, although contributions from other nanodomains are not excluded. Membrane rafts are elevated features compared to the surrounding membrane (49). Furthermore, membrane rafts correspond to L_o domains (5, 50), which are stiffer compared to surrounding L_d domains (51). Therefore, to pinpoint membrane rafts, the Hadamard product images were generated to accentuate only regions that are simultaneously higher and stiffer. The red circles mark examples that are both higher and stiffer (i.e., membrane rafts). The yellow circles mark examples where the features are higher but softer or stiffer but not elevated from their surroundings (Fig. 2D and fig. S1E). Although cytoskeletons are also higher and stiffer, their elongated shapes are discernible from rafts (fig. S2). The Hadamard image obtained by combining the amplitude error (height) and phase mode (hardness) enabled facile identification of features resembling membrane rafts. This assignment was further supported by the response toward administered fibrinogen and Mn²⁺/resveratrol (vide infra).

Response of membrane rafts to fibrinogen and Mn²⁺/resveratrol

The trajectory and height evolution of membrane rafts were monitored in the absence (Fig. 3A, movie S2, and fig. S3) and presence of stimulants (Figs. 3, B and C, and 4; movies S3 and S4; and figs. S4 and S5). In the absence of stimulus, the nominal diameter and height of membrane rafts were about 150 and 1.9 nm (fig. S6), respectively. This size is consistent with the current estimation of 10 to 200 nm described in the introductory section for membrane rafts (52). Several membrane rafts in Fig. 3 (A_i to A_v) were monitored (Fig. 3D_a), showing a variety of motions. Within 4 min, the rafts indicated by blue and yellow arrows traveled further than 1 μm in opposite directions. The two rafts indicated by purple and green arrows appeared stationary.

Fig. 3. Dynamics of membrane rafts on live MCF-7 cells.

Experiments were carried out in the environments of (A) blank phosphate buffer and additional treatment with (B) fibrinogen and (C) Mn²⁺/resveratrol. (D) Trajectories of membrane rafts. In (D_a) to (D_c), the circle diameters and arrows correspond to the sizes and moving directions of the rafts, respectively. The histograms and table S1 summarize the size distribution. In (D_b) the 1-μm circles (gray) indicate the fused aggregate. Solution: (A) PBS (10 mM phosphate, pH 7.4), (B) PBS with 70 μM fibrinogen, and (C) PBS with 50 μM Mn²⁺ and 10 μM resveratrol. Scan rate: 28 s per frame. Image size: [(A) and (B)] 5 μm by 5 μm and (C) 10 μm by 10 μm. Other imaging conditions were the same as those of Fig. 2.

Fig. 4. Progression of topographic heights of resveratrol-treated membrane rafts.

(A to D) Images were obtained from the same experiment as those in Fig. 3C. Data collection started immediately after the resveratrol solution was added into the AFM liquid cell. The images were taken after 60 min because of the ~1 hour required to activate integrins α_Vβ₃ by Mn²⁺ (fig. S8). (E) Examples of the height evolution for the protrusions. The height information was measured from the tapping mode deflection images. Some features did not appear at the beginning (e.g., 27 and 78) or in the end (e.g., 1, 3, 14, and 19) of the observation, and their data were plotted accordingly. Movie S5 presents the evolution in height for four features. (F) Histograms of height-dependent persistent time, which is defined as the duration during which the height remains above 3.7 nm (for detail, see fig. S7). Solution: PBS with 50 μM Mn²⁺ and 10 μM resveratrol. Scan rate: 28 s per frame. Image size: 10 μm by 10 μm. Other imaging conditions were the same as those of Fig. 2B.

The cells were imaged upon administering fibrinogen; images representing 23rd to 29th min after administering are presented in Fig. 3B. The ~0.8-μm–sized feature (in the top right corner of Fig. 3B_i) fluctuated in size and eventually dissipated. Small spots indicated by arrows (in the bottom right part of Fig. 3B_i) conglomerated and turned into ~1 μm in size and ~45 nm in height entity. The conglomerated features such as the ~0.8-μm and ~1-μm ones suggested that individual rafts (10 to 200 nm) in Fig. 3B_iii could not be resolved in the features (~0.8 and ~1 μm) of Fig. 3B_iv. This phenomenon was observed in 18 of 22 cells (fig. S11). The ligand-activated changes were absent in roughly 20% of the cells, most likely due to the lack of receptors in the section chosen to be monitored.

The cells were imaged upon exposure to resveratrol in the presence of Mn²⁺ (Fig. 3C and fig. S12). Images were acquired after 1 hour to ensure activation of α_Vβ₃ by Mn²⁺ (please see confocal images in figs. S8 and S13, which includes representative images of 27 cells in total). Moreover, confocal microscope characterization (fig. S10) demonstrates that α_Vβ₃ localizes to C-Laurdan–defined liquid-ordered domains (membrane rafts). The protrusions increased in number and grew slightly larger and higher (Fig. 4). In the top right and bottom left corners of the images, there were elevated terraces higher than the background membrane. In the terraces, individual protrusions were distinctly delineated, remaining discrete rather than merging, unlike the conglomerated features observed upon exposure to fibrinogen (e.g., Fig. 3, B_iv and B_v). To include the features in the top right and bottom left corners in one frame, the image size was twice of those in Fig. 3 (A and B).

The height of the integrin membrane rafts upon Mn²⁺ activation in the presence of resveratrol was monitored (Fig. 4 and movie S5). In the absence of either Mn²⁺ activation or resveratrol, the rafts were ~1.9 nm higher than the background membrane (fig. S6), consistent with theoretical modeling (53) and AFM studies on reassembled rafts that were isolated and purified from cell membranes (33). After the introduction of both Mn²⁺ and resveratrol, the protruded features were tracked. In a representative experiment (Fig. 4), about 28% of the protrusions did not respond, and the duration for their presence lasted 10 to 20 min (i.e., the heights returned to the background level; fig. S7). Another ~31% of the protrusions showed heights of more than 10 nm (Fig. 4E), consistent with a fully extended α_Vβ₃ integrin based on studies by simulations (54, 55), biomembrane force probe (56), and cryo-EM (23). The rest ~41% was activated, but the maximal height was not higher than 10 nm. For the membrane rafts with heights corresponding to fully extended integrins, about 65% of the stimulated events completed in the 90th to 120th min, and about 35% continued after the 120th min (Fig. 4E). The persistent time (defined on page 3 of Supplementary Text and fig. S7A) broadly spread from 10 to 60 min (Fig. 4F), providing the information about the lifetime of a resveratrol-bound, Mn²⁺ activated integrin.

DISCUSSION

Biochemical implications

An important prospect of visualizing stimulant-triggered membrane dynamics is to guide, with the support of theoretical modeling, the exploration of downstream signaling pathways beneath the outer exoplasmic leaflet. The AFM images of MCF-7 cells upon administering fibrinogen and Mn²⁺/resveratrol (Fig. 3, B and C) resembled the simulated patterns in Fig. 1 (C and D), respectively, suggesting mechanisms associated with attractive protein-raft interactions (28) for the former and membrane components recycling to the cytoplasm (28) for the latter. In these cases, the pure physical models appeared to be able to correspond to specific biochemical processes. The unbound bent integrin β₃ chain reorients and stretches upon binding of the β₃ cytoplasmic tail to talin and kindlin (57). The subsequent binding to fibrinogen converts the integrin to a fully activated extended open conformation. Alternatively, ligands (e.g., fibrinogen) in the extracellular matrix can activate unbound bent integrins directly without the assistance of talin binding (58). The fibrinogen-bound activated integrins can then recruit signaling regulators, such as vinculin, which reinforces the β subunit–talin–actomyosin interaction and facilitates integrin clustering (59), equivalent to the attractive protein-raft interactions in Fig. 1C.

The AFM images of MCF-7 cells in the presence of Mn²⁺/resveratrol bear resemblance to Fig. 1D. The corresponding physical model suggests recycling of small molecules between the cell membrane and cytosol. The added Mn²⁺/resveratrol interacts with the exposed β₃ domain (36), which activates the integrin and downstream molecules including Rac–guanosine triphosphatase (60). This recruits phospholipase C from the cytosol to cell membrane (61) and elicits the breakdown of phosphatidylinositol 4,5-bisphosphate (PIP₂) to inositol triphosphate (IP₃) and diacylglycerol. The IP₃ is released to the cytosol in exchange with phosphatidylinositol, which is synthesized in the endoplasmic reticulum and then delivered to the cell membrane to reform PIP₂ (62). In either the fibrinogen or the Mn²⁺/resveratrol case, we cannot completely rule out the possibility that the activated integrins may be bound to the actin cytoskeleton.

In summary, this study manifests the strengths of AFM in the synchronous acquisition of topographic and hardness information to generate the corresponding Hadamard image. These Hadamard images enable the facile identification of membrane rafts, providing temporal information about location, movement, size, and height. Specifically, this study provides an estimation for the persistent time of an activated integrin. Equipped with these Hadamard images with the distribution and outlines of membrane rafts, it is therefore feasible to examine the computational snapshots and identify the interrelated physical models to guide subsequent studies on downstream signal transduction pathways.

MATERIALS AND METHODS

Chemicals and solutions

A phosphate-buffered saline (PBS) solution containing 1.0 mM Hepes (C₈H₁₈N₂O₄S, Sigma-Aldrich) and 10% (v/v) fetal bovine serum (FBS; Gibco) was used for the AFM imaging of live MCF-7 cells in control experiments without stimulants. For experiments involving the activated integrins, the solutions of the stimulants were prepared in PBS. Specifically, the solutions were 0.14 mM fibrinogen (Sigma-Aldrich) or 20 μM resveratrol (≥ 99%; high-performance liquid chromatography grade, Sigma-Aldrich) in PBS composed of 1.0 mM Hepes and 10% FBS. The resveratrol solution also contained 100 μM MnSO₄ (BioReagent grade, Sigma-Aldrich). For the control experiments, the AFM liquid cell (a custom-made petri dish; Alpha Plus Scientific Corp., Taoyuan City, Taiwan) contained cultured MCF-7 cells and 5.0 ml of PBS. For the subsequent stimulation, another 5 ml of the stimulant solution was added to make a 10-ml solution of 70 μM fibrinogen or 10 μM resveratrol with 50 μM MnSO₄.

Cell culture

The MCF-7 human breast cancer cell line (HTB-22TM, American Type Culture Collection) was purchased from the Bioresource Collection and Research Center (Hsinchu, Taiwan). Cells were cultured in Dulbecco’s modified Eagle’s medium (12800017, Gibco, Waltham, MA, USA) supplemented with 10% FBS, 1% penicillin-streptomycin (SV30010, HyClone, Logan, UT, USA), and 10 mM Hepes. The culture was maintained in a humidified incubation system with 5% CO₂ at 37°C. The growth medium was refreshed every 48 hours, and cells were trypsinized using 0.1% trypsin (SH30042.01, HyClone, Logan, UT, USA) and subcultured upon reaching ~90% confluence. For AFM imaging, 8 × 10⁴ MCF-7 cells per well were seeded and incubated in a custom-made petri dish in a 5% CO₂ incubator at 37°C for 48 hours before studying.

AFM imaging on live MCF-7 cells

AFM imaging was performed using a Dimension AFM equipped with FastScan and Icon scanners (Bruker). To preserve the cell viability during AFM observation, the imaging stage was hosted in a vibration-isolation chamber in which the PBS solution was maintained at 37°C (Model 335 temperature controller, Lakeshore). The suitable spring constants of AFM tips for imaging cell membranes by tapping mode (FastScan scanner) and PeakForce QNM Mode (Icon scanner) were found to be 0.25 N/m (FastScan-D, Bruker) and 0.6 N/m (HHQ:NSC36, MikroMasch), respectively. Typical imaging parameters for tapping mode were amplitude setpoints of 60 to 120 mV and scan rates of 10.0 to 18.6 Hz, corresponding to 51.2 to 27.5 s per image frame. For PeakForce QNM Mode, the parameters for amplitude, setpoint force, and scan rates were 100 to 200 nm, 0.10 nN, and 0.4 to 0.6 Hz, respectively.

Supplementary Materials

The PDF file includes:

Supplementary Text

Figs. S1 to S14

Data S1

Legends for movies S1 to S5

Tables S1 and S2

References

Other Supplementary Material for this manuscript includes the following:

Movies S1 to S5