Disruption of membrane cholesterol organization impairs the activity of PIEZO1 channel clusters

Mechanosensitive ion channels are membrane proteins that sense mechanical stimuli, allowing cells to respond and adapt to physical forces. An essential family of eukaryotic mechanosensitive channels are PIEZO channels (Coste et al., 2010) comprised of two members, PIEZO1 and PIEZO2 (Wu et al., 2017a). These channels are associated with a number of physiological functions, such as the development of vascular architecture (Li et al., 2014; Rode et al., 2017). Human PIEZO1 also plays a key role in erythrocyte volume regulation (Cahalan et al., 2015), in which gain-of-function variants cause hereditary dehydrated stomatocytosis (xerocytosis; Zarychanski et al., 2012; Albuisson et al., 2013; Andolfo et al., 2013; Bae et al., 2013a). These mutations invariably slow channel inactivation, and this phenotype confers native resistance to malarial invasion in African populations (Ma et al., 2018).

Three recently published cryo-EM structures of purified mouse PIEZO1 provide detailed information of its trimeric, curved structure (Ge et al., 2015; Guo and MacKinnon, 2017; Saotome et al., 2018; Zhao et al., 2018). Three homomeric subunits form a central nonselective cation-permeable pore with a large-scale propeller-like assembly (Ge et al., 2015; Guo and MacKinnon, 2017). Large beams that extend from the central pore axis to the extremity of the protein have been proposed to underlie a lever-like mechanism to transfer force from the periphery to the channel gate (Guo and MacKinnon, 2017; Zhao et al., 2018). Activation of PIEZO1 is associated with a force-from-lipids mechanism (Martinac et al., 1990; Lewis and Grandl, 2015; Teng et al., 2015; Cox et al., 2016a; Syeda et al., 2016), in which lipid bilayer forces are sufficient to drive an as-yet-unknown conformational change in PIEZO1 and trigger gating. This suggests that both global effects on membrane physical properties and specific lipid interactions are likely to be essential for PIEZO1 function (Cox et al., 2016b; Cordero-Morales and Vásquez, 2018; Romero et al., 2019).

Cholesterol is a ubiquitous component of cellular membranes and a major modulator of membrane mechanical properties. Application of the cholesterol-depleting agent methyl-β-cyclodextrin (MBCD) reduced whole-cell indentation-induced currents in PIEZO1-expressing cells (Qi et al., 2015). This response was linked to the cholesterol-binding protein stomatin-like 3 (STOML3), previously shown to sensitize PIEZO1 channels (Poole et al., 2014). Furthermore, a cross-linking–based proteomics study reported a nonrandom degree of association between a cholesterol analogue and PIEZO1 (Hulce et al., 2013). Thus, there is mounting evidence indicating physiologically relevant interplay between cholesterol and PIEZO1-mediated mechanotransduction. In humans, the difference in the tissue expression pattern of STOML3 and the more ubiquitous PIEZO1 suggests that STOML3-mediated cholesterol sensitization of PIEZO1 may be tissue specific (Uhlén et al., 2015).

In this study, we characterized the effects of cholesterol on PIEZO1 activity in cells with undetectable levels of STOML3 (Sultan et al., 2008; Uhlén et al., 2015). Using superresolution stochastic optical reconstruction microscopy (STORM) imaging, we show that PIEZO1 channels in the plasma membrane associate as clusters. Cholesterol removal or disruption affected the dynamics of these PIEZO1 clusters but had only a minor effect on their size. Removal of cholesterol or disruption of cholesterol-rich domains reduced channel sensitivity, slowed down activation, and abolished inactivation in cell-attached patches. We imaged both cholesterol-rich membrane domains and PIEZO1 in HEK293T cells stably expressing the channel protein, using both standard and superresolution fluorescence microscopy to quantify the dynamic behavior of PIEZO1 and its relationship with cholesterol-rich membrane domains. Furthermore, we suggest a model of the PIEZO1 and cholesterol interaction and propose that this association is essential for the spatiotemporal activity of PIEZO1.

All recordings of human PIEZO1 were obtained from HEK293T cells stably expressing PIEZO1-1591–GFP (Syeda et al., 2016), a previously characterized fusion protein (Cox et al., 2016a). N2A cells were a kind gift from Dr. K. Poole (University of New South Wales, Kensington, Australia). PIEZO1 KO HEK293T cells were a kind gift from Prof. A. Patapoutian (Scripps Institute, La Jolla, CA). Transient transfection of PIEZO1-KO cells with plasmid expressing the R2456H Piezo1 mutant was performed using polyethyleneimine according to the manufacturer’s protocol. The same transfection protocol was used to prepare cells for transient versus stable cell line Piezo1 expression (see Fig. S6). Cells were plated onto 12-mm round glass coverslips and grown in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% FBS. Cells were treated with 0.1, 0.5, 1, 5, or 10 mM MBCD (C4555; Sigma-Aldrich), 100 µg/ml water-soluble cholesterol/MBCD complex (C4951; Sigma-Aldrich; product contains 40 mg cholesterol per gram of compound, and the compound molarity during the 100 µg/ml treatment is equivalent to 10 µM cholesterol and 75 µM MBCD) or 80 µM dynasore (D7693; Sigma-Aldrich) in serum-free DMEM for 30 min at 37°C. Untreated control samples were also incubated in serum-free DMEM at 37°C for 30 min before patch-clamping. Docosahexanoic acid (DHA) treatment was performed by culturing cells for 24 or 48 h in DMEM supplemented with 10% FBS and 25 µM DHA (D2534; Sigma-Aldrich). All recordings were performed in identical pipette/bath solutions having the following composition: 140 mM NaCl, 3 mM KCl, 1 mM MgCl₂, 1 mM CaCl₂, 10 mM glucose, and 10 mM HEPES, pH 7.4. Borosilicate glass pipettes (Drummond Scientific) were pulled using a Narishige puller (PP-83; Narishige) to achieve a bubble number of 6.0 (3-MΩ pipette resistance). The PIEZO1 currents were recorded at room temperature using an AxoPatch 200B amplifier (Axon Instruments) in the cell-attached patch configuration, at a 20-kHz sampling rate with 1-kHz filtration. Negative pressure was applied to the membrane patch using a computer-controlled high-speed pressure clamp-1 apparatus (HSPC-1; ALA Scientific Instruments).

PIEZO1-GFP stable HEK293T cells were seeded in DMEM (Gibco) supplemented with 10% FBS inside 35-mm diameter Fluorodishes (FD35-100; World Precision Instruments) and grown to 70–85% confluency for 2–3 d. For colocalization studies, cells were incubated in 1:1,000 dilution of cholera toxin subunit B (CtxB)–Alexa Fluor 555 (Invitrogen; 5 mg/ml stock) and rinsed three times with PBS. After labeling, treatments were performed as described above, and cells were immediately imaged live on a Zeiss Elyra microscope.

Estimation of plasma membrane cholesterol using Filipin III was performed as previously described (Qin et al., 2006). Briefly, 50 mg/ml Filipin III (F4767; Sigma-Aldrich) stock in DMSO was diluted 1:1,000 in PBS and incubated for 1 h at room temperature with fixed, treated, or untreated cells. Afterward, cells were extensively washed in PBS and stored at 4°C before imaging experiments. TIRF imaging was performed as described below by exciting the sample with a 405-nm laser at a 63° angle and detection through an α Plan-Apochromat 100×/1.46 oil objective of the Zeiss Elyra microscope.

The plasmids shown in Table 1 were transfected using Lipofectamine 3000 (L3000001; Thermo Fisher Scientific) according to the manufacturer’s protocol and used for image cross-correlation spectroscopy (ICCS) control experiments (see Fig. S12). The D4Hα-mCherry (θ toxin domain 4) fluorescent marker of cholesterol domains (Johnson et al., 2012; Maekawa, 2017) was used to monitor cholesterol levels during MBCD treatments (see Fig. S2).

For STORM experiments, cells were treated with drugs, 5 mM MBCD, 50 µg/ml water-soluble cholesterol/MBCD complex, or 80 µM dynasore diluted in serum-free DMEM, or incubated in serum-free DMEM for control samples, for 30 min at 37°C. After treatment, cells were washed with warm PBS and fixed with 4% PFA for 15 min at room temperature followed by three rinses with PBS. Postfixation cells were permeabilized with Triton X-100 (0.01% vol/vol) at room temperature for 15 min. After permeabilization, cells were blocked with 5% BSA for 1 h at room temperature, followed by incubating the dishes with labeling reagents at room temperature for 1 h. PIEZO1-GFP was labeled with anti-GFP–Alexa Fluor 647 with 1:1,000 PBS dilution from stock (Bioss Antibodies; 1 µg/µl). Samples were stored at 4°C and imaged within 2 d from preparation on a Zeiss Elyra microscope.

STORM images were acquired on a TIRF microscope (Elyra; Zeiss) using a 100× α Plan APO oil-immersion objective lens (NA 1.46). Cells prepared and fixed as described above were covered with STORM buffer consisting of 25 mM HEPES, 5% glycerol (vol/vol), 25 µg/ml glucose oxidase, 25 µg/ml horseradish peroxidase, and 20 mM cysteamine (Dempsey et al., 2011). We ensured that out-of-focus drift was eliminated by setting the temperature of the heating enclosure at 26°C and using the definite focus of the Elyra microscope. Alexa Fluor 647 fluorophores were excited using a 638-nm laser, and emission was collected using a longpass filter at 640 nm. Alexa Fluor 647 fluorophores were photoconverted from dark to excitable state, ∼0.1 mW of continuous 405-nm laser radiation, and imaged with ≪1 mW of 488-nm light. Between 10,000 and 20,000 images were acquired per dataset, and single-molecule emissions were fitted using a 2D Gaussian distribution function in the algorithm provided in Zeiss’ Zen software. For image data acquisition, we used a cooled, electron-multiplying charge-coupled device camera (iXon DU-897D; Andor) with an exposure time of 30 ms and an imaging area containing cells of interest. The zoom lens before the camera was set to 1.6× such that the pixel size was 0.1 µm. The data were acquired and preprocessed using Zen software customized for this microscope, while further cluster decomposition and statistics were done using custom-built scripts in Matlab (The Mathworks). Preprocessing in Zen consisted of filtering out the features (molecules) detected with signal-to-noise parameters <6 and size ≤9 pixels (mask) diameter, which equaled 0.9 µm. No visible lateral (xy) or axial (z) drifts were observed in any datasets analyzed. No grouping of observation points was done, as we did not calibrate the blinking of Alexa Fluor 647 under imaging conditions in case they are uniformly randomly dispersed. Nevertheless, we do not claim that the number of PIEZO1-GFP entities observed per cluster are in absolute values. Rather, we quantify the extent (area and perimeter) of clusters and quantify the relative change in observations, number of emitting events, per cluster in each condition. Preprocessed point pattern data were exported to a TXT file and loaded into Matlab for further quantification. Point patterns of observations, per cell, were segmented into clusters using the approach called DBSCAN (Bar-On et al., 2012) using custom-built software in Matlab. Briefly, for each point detected, we calculated the local density within an ∼30-nm radius (average precision of detections in STORM using Elyra). Next, we started with the highest local density point and counted all the molecules within a circle of ∼30 nm. Then, the same operation was repeated for every point within the circle, and so on until all points of a cluster were processed, and all the points that passed the criteria were grouped into the first cluster. All the points from that cluster were removed from the original cluster data and stored. The operation was repeated until the next cluster was isolated. This iterative process was repeated until the last cluster was defined and points included were stored. We searched the objects with at least five detections (points) per cluster, and any object with fewer observations was not considered a cluster. For each cluster, we calculated the cluster area and perimeter using a combination of Matlab functions (boundary, alphaShape, and perimeter) and number of points per area (density). By defining the total cell surface using the thresholded image from the GFP channel, only the clusters within the cell boundaries were considered. This also allowed for counting the number of PIEZO1-GFP clusters per unit cell area. Clusters smaller than the approximate size of a single PIEZO1 trimer (∼200 nm²) were not considered for the statistics, as they could be the outcome of single-molecule blinking, which we did not account for.

PIEZO1-GFP diffusion inside the plasma membrane of stably transfected HEK293T cells was imaged with a TIRF Elyra (Zeiss) microscope using a 100×/1.46 lens at an imaging rate of 29 Hz. The microscope stage was equipped with an enclosed heating module (37°C). To quantify the diffusion of PIEZO1-GFP entities and its confinement by cholesterol domains, we applied kICS confinement analysis as previously reported (Abu-Arish et al., 2015). By assuming the isotropic diffusion of fluorescent particles, we circularly averaged the correlation function at each temporal k² and temporal lags τ, as shown in examples of Fig. S11, A–C. Subsequently, we fitted the resulting correlation function to extract the diffusion of PIEZO1-GFP outside of cholesterol clusters and the lateral unbinding rate of PIEZO1-GFP from cholesterol rich domains, as described in Eq. 2 of Brandão et al. (2014). The kICS correlation functions from late temporal lags and for large spatial frequencies was fitted using Eq. 9 of Brandão et al. (2014).

The same TIRF modality was used to image cholesterol labeling D4H-mCherry expressed in HEK293T cells, before and after MBCD (see Fig. S2). To monitor the cholesterol removal from the same cell, we first imaged several fields of view of cells before MBCD addition. Cells were imaged for 2.5 s (33 ms/frame) before (control), and then every 5–10 min a series of images (2.5 s) were recorded after addition of 5 mM MBCD. The average intensity per cell per time frame was extracted using a custom-built program in Matlab (see Fig. S2).

Finally, TIRF imaging was used to image fixed transiently transfected PIEZO1-KO cells expressing PIEZO1-GFP from plasmid or Piezo1-GFP stable cell line (see Fig. S6). TIRF imaging ensured that PIEZO1-GFP signal was predominantly from plasma membrane.

To further investigate the effect of MBCD on PIEZO1 colocalization with CtxB, we performed experiments in which the same cells were imaged before and after treatment. We imaged cells before treatment as described above while recording their positions and added drugs to the dish, ensuring that the dish could not move to preserve the saved cells’ coordinates. Live-cell imaging acquisition began 20 min after the treatment was applied.

Cells plated on dishes were labeled with a 1:200 dilution of 5 mM Laurdan (D250; Molecular Probes), incubated at 37°C for 30 min, washed with warm PBS, fixed and permeabilized, and then the PIEZO1-GFP was labeled with anti-GFP–Alexa Fluor 647. Laurdan data were acquired on a Zeiss LSM 880 microscope with excitation from a mode-locked two-photon Insight DeepSea operating at 80-MHz repetition rate and a wavelength of 780 nm. Excitation and photon collection were performed with a Plan-Apochromat 63× 1.4 oil objective (Zeiss). Laurdan data were acquired in two channels with wavelengths of 400–460 and 470–530 nm.

ICCS was applied to quantify the level of colocalization between PIEZO1 and ordered/disordered phases of Laurdan (Comeau et al., 2008). Each image pair was preprocessed to minimize the artifacts due to channel misalignment, noise, and cell edges (Comeau et al., 2008). First, a correction matrix for shift of far red (Alexa Fluor 647) images, to align with Laurdan images, was calculated using Matlab function imregtform. The corrected images were then noise filtered using a 2D Gaussian filter with 1-pixel σ. Then the minimum intensity from images in each channel was subtracted, and a final image was normalized to its maximum. Finally, the resulting images were used to define a mask, from thresholding above the average intensity, and all the pixels not belonging to the mask (equal to 0) were set to have the value of average intensity of pixels within the mask (equal to 1). This process ensures that pixels not belonging to cell areas are not contributing to the fluctuations and inherent correlation functions in ICCS (Comeau et al., 2008). We extracted the number density of particles in red (PIEZO1), green (either ordered or disordered Laurdan phase), and cross-correlation channels.

An MTS assay (Ab197010, Abcam) was used to interrogate the effect of MBCD pretreatment of cells in the presence of serum-free medium according to the manufacturer’s protocol. Briefly, cells were seeded at a density of 30 × 10⁴ cells per well in a 96-well plate and pretreated with MBCD or CoCl₂, as a positive control to induce cell death, for 30 min and then incubated in the presence of MTS. The subsequent colorimetric change was measured using a VersaMaxTunable microplate reader (Molecular Devices) and normalized to untreated cells.

Fig. S1 shows the dose-dependent effects of MBCD on PIEZO1-GFP in stable cell line. Fig. S2 shows the effect of MBCD on cell viability and cholesterol levels. Fig. S3 shows the cholesterol labeling experiment using Filipin III staining. Fig. S4 shows the quantitation of the first event latency of PIEZO1-GFP response. Fig. S5 shows the effect of cholesterol depletion via MBCD on N2A cells. Fig. S6 shows the effect of cholesterol depletion via MBCD on transiently transfected WT human PIEZO1. Fig. S7 shows the effect of DHA treatment on PIEZO1-GFP. Fig. S8 shows the effect of cytoskeleton manipulation on the membrane diffusion of PIEZO1-GFP. Fig. S9 shows additional parameters calculated from ICCS for assessment of PIEZO1-GFP/CtxB-Alexa Fluor 555 colocalization in the plasma membrane. Fig. S10 shows time-lapse imaging of single cells exposed to treatments. Fig. S11 shows examples of kICS correlation functions. Fig. S12 shows ICCS control experiments. Video 1 shows examples of STICCS correlation function fluctuations over time. Videos 2, 3, and 4 show live TIRF imaging of hP1-1591–GFP stable cell line in control condition, after MBCD treatment, and after water-soluble cholesterol/MBCD complex treatment, respectively.

We measured the effects of cholesterol depletion, supplementation, and dynasore (a compound known to disrupt cholesterol-rich domains) on PIEZO1-GFP channels stably expressed in HEK293T cells. These cells express exceedingly low levels of STOML3 (Sultan et al., 2008; Uhlén et al., 2015). We observed a significant modulation of channel sensitivity and kinetics when cholesterol was removed by MBCD or dynasore was added compared with control cells (Fig. 1, A–D). High doses (100 µg/ml) of water-soluble MBCD/cholesterol complex (Fig. 1 B) did not affect PIEZO1-GFP gating. However, PIEZO1-GFP pressure sensitivity was significantly decreased after treatment with 5 mM MBCD (Fig. 1, E and F), but not at lower dosages (Fig. S1), without significant impact on cell viability (Fig. S2, A and B). The treatment of HEK293T cells with 5 mM MBCD was verified to effectively decrease the level of cholesterol by measuring the fluorescence intensity of a genetically encoded DH4a-mCherry cholesterol probe (Johnson et al., 2012; Maekawa and Fairn, 2015; Fig. S2, C and D). The efficacy of the MBCD and water-soluble cholesterol treatment was also verified by staining cells with the cholesterol probe Filipin III (Fig. S3).

Desensitization to applied pressure was accompanied by a substantial increase in latency of the response, at both the level of the first opening (Fig. S4) and onset of the maximal current (Fig. 1 G). None of the compounds altered the average maximal PIEZO1-GFP current (Fig. 1 H). Both MBCD and dynasore caused similar delays in PIEZO1-GFP activation, but only MBCD caused a rightward shift in the midpoint activation pressure of the channel by ∼40%. Interestingly, both MBCD and dynasore treatment substantially reduced channel inactivation, as quantified using normalized steady-state current (Fig. 1 I). Similar effects of MBCD were observed on endogenous mouse PIEZO1 channels in N2A cells (Fig. S5).

The inactivation of PIEZO1 is currently described as an intrinsic process in the channel gating cycle, which can be altered by introducing mutations at specific sites in the protein (Poole et al., 2014; Wu et al., 2017b; Moroni et al., 2018; Zheng et al., 2019; Lewis and Grandl, 2020). The R2456H mutant, a xerocytosis-causing variant of human PIEZO1, displays a significant loss of inactivation with respect to WT PIEZO1 (Fig. 2 A), as previously reported (Bae et al., 2013a). We tested whether the effects of MBCD treatment on the channel inactivation were specific to the WT channel or could affect the R2456H channel as well (Fig. 2, A and B). The degree of inactivation was quantified using normalized steady-state current (Fig. 2, A–C). Interestingly, the MBCD treatment appeared to slow the inactivation of R2456H PIEZO1 (Fig. 2, B and C), increasing the steady-state current while also increasing the latency of the response (Fig. 2 D and Fig. S4), when the mutant was transfected into PIEZO1 KO HEK293T cells. MBCD did not change the sensitivity of R2456H to applied pressure (Fig. 2, E and F).

Interestingly, cholesterol removal via MBCD did not significantly change the midpoint of activation of the transiently transfected WT PIEZO1 to applied pressure either (Fig. S6, A and B). We hypothesized that the shift in pressure sensitivity observed in our stable cell line after MBCD treatment was dependent on the mode of channel expression (i.e., stable versus transient). Indeed, when observed in TIRF mode, transiently expressed hP1-1591-GFP was detected in higher amounts in the basal plasma membrane compared with cells stably expressing the same PIEZO1 clone (Fig. S5, D–F).

The transiently expressed R2456H mutant displayed a higher sensitivity to pressure (Fig. 2 F) compared with transiently expressed WT PIEZO1 (Fig. S6 B) and stably expressed PIEZO1-GFP (Fig. 1 F). Nontransfected PIEZO1 KO HEK293T cells did not produce mechanosensitive currents when challenged with suction pulses (Fig. 2 G). The P₅₀ values and slopes of Boltzmann fits are summarized in Table 2. Only the stable cell line and the PIEZO1 native to N2A cells (Fig. S5) appeared to right-shift their sensitivity after MBCD treatment. All the PIEZO1 variants studied (PIEZO1-GFP, transiently transfected WT, transiently transfected R2456H mutant, and N2A native) displayed slower inactivation after cholesterol removal. These results suggest that membrane cholesterol, like other lipids (Romero et al., 2019), contributes to PIEZO1 inactivation. The sensitivity of PIEZO1-GFP to its lipid environment was also observed when cells where cultured in presence of DHA (Fig. S7), which caused an increase in pressure sensitivity, latency of the response, and loss of inactivation.

To study the dynamics of PIEZO1-GFP in a live cell, we performed spatiotemporal kICS analysis of pixel fluorescence fluctuations in the basal membrane of our stable cell line (Fig. 3 A, Video 2, Video 3, and Video 4). The PIEZO1-GFP fluorescent signal in TIRF mode was detected as distinct puncta of varying sizes. The smaller fluorescent entities diffusing away from large puncta were defined as clusters, and puncta larger than the point spread function (∼200-nm radius) were defined as groups of clusters (Fig. 3 B). The term “cluster” was used to indicate a diffraction-limited fluorescent entity composed of an unknown amount of PIEZO1-GFP proteins. The same definition of a PIEZO1-GFP cluster is used in all diffraction-limited imaging experiments, except the superresolution imaging section (STORM), in which individual proteins were detected with an average localization precision of ∼30 nm. PIEZO1-GFP appears to experience a limited diffusion, schematically described in the cartoon as the motion from A to B (Fig. 3 B), owing to its clustering and heterogeneity of the plasma membrane. Before the addition of MBCD, the channel’s average diffusion rate was ∼0.005 µm²/s (Fig. 3 D). Cholesterol removal by MBCD doubled the diffusion rate, whereas addition of 50 µg/ml of cholesterol had no effect.

kICS analysis measures the lateral unbinding rate from the PIEZO1-GFP clusters, k_unbind. This measurement is defined as the rate at which fluorescent entities (particles or clusters) move away from locally confined areas or docking areas. This parameter can also be interpreted as rate at which a measured protein escapes a local confining environment. MBCD-treated cells caused PIEZO1-GFP molecules to dissociate at a higher rate from associated local confinement, whereas cholesterol addition did not affect this affinity (Fig. 3 C). This result implies that MBCD treatments break the underlying local cholesterol scaffold by removing the cholesterol and changing the affinity of PIEZO1-GFP to its local environment, whereas cholesterol addition does not change the confinement barrier, but only the number of barriers. Perturbing the actin (cytochalasin D, jasplakinolide) or the microtubule cytoskeleton (colchicine) did not affect the local lateral unbinding of the PIEZO1-GFP from clusters (Fig. S8 A). Interestingly, among these compounds, only jasplakinolide appeared to reduce the effective diffusion in the basal membrane, consistent with its effect on the organization of the cortical actin meshwork (Fig. S8 B).

To confirm our observations on PIEZO1-GFP clustering dynamics, we performed STORM in TIRF mode. By applying STORM, we were able to detect the boundaries of each fluorescent entity with nanometric precision and therefore describe the biophysical characteristics of large and small PIEZO1-GFP clusters.

Our stable cell line carrying a PIEZO1-GFP fused construct permitted us to visualize stochastic blinking events originating from an Alexa Fluor 647–labeled anti-GFP antibody. This allowed us to avoid issues with the specificity of commercially available primary PIEZO1 antibodies. Fig. 4 A shows the PIEZO1-GFP stable cell line and the resolution improvement due to STORM analysis. The insets in the figure focus on subcellular areas in which the GFP signal (in green) was superimposed with the STORM signal (in red) inferred from the optical reconstruction of Alexa Fluor 647 blinking events. Software cluster decomposition and classification allowed us to characterize PIEZO1-GFP clusters by defining their area (Fig. 4 B), perimeter (Fig. 4 C), and cluster density/µm² (Fig. 4 D).

In control experiments, PIEZO1-GFP clusters appeared to distribute nonuniformly around the cell’s basal membrane and to pack into a variety of cluster areas and perimeters. Cluster areas ranged from few hundred to several thousand square nanometers. Less than 10% of total clusters in relative frequency histograms appeared to be ∼1,000–2,000 nm² in area (Fig. 4 B) and ∼200–300 nm in perimeter (Fig. 4 C). All tested conditions appeared to increase the relative frequency of this fraction of clusters, indicating breakdown of larger order clusters into smaller ones. Despite the statistical difference detected between the median values for cluster sizes between conditions, the distribution of these cluster populations largely overlapped. In addition, the stably expressing HEK293T cells displayed variable levels of cluster densities in their basal membrane (Fig. 4 D), and we identified three major populations: low-density PIEZO1-GFP cells (∼2–3 clusters/µm²), medium-density PIEZO1-GFP cells (∼6 clusters/µm²), and high-density PIEZO1-GFP cells (∼8–9 clusters/µm²). As suggested by the size histograms, each treatment appeared to impact the density of PIEZO1-GFP clusters in the basal membrane by increasing the number of cells displaying reduced cluster densities. Of all tested conditions, dynasore appeared to uniformly reduce the cluster density in all three observed groups. MBCD and cholesterol treatments, on the other hand, greatly reduced the fraction of cells displaying high densities of PIEZO1-GFP, enriching the fraction of cells displaying low-density PIEZO1-GFP clusters. Because of the filtering process applied to STORM data, it is important to mention here that the smallest cluster population of <30-nm radius would likely have been excluded from the analysis if not found to produce sufficient blinking events and thus preventing us from estimating exact cluster densities per square micrometer. Our results nevertheless offer reliable estimates for densities of larger PIEZO1 clusters above that size.

These results suggest that PIEZO1-GFP assembles into clusters in the plasma membrane of cells. Cholesterol levels in the plasma membrane appear to regulate the degree of clustering by mostly affecting the density of clusters in the basal membrane.

To visualize association of PIEZO1-GFP and cholesterol-rich domains in stably expressing HEK293T cells, we used fluorescently labeled (Alexa Fluor 555) CtxB, a well-known marker of lipid rafts (Day and Kenworthy, 2015). Because of the heterogeneous and dynamic behavior of PIEZO1-GFP entities in the membrane, we collected imaging data as time-lapse series in two channels and examined the degree of cross-correlation between GFP- and Alexa Fluor 555–tagged entities using STICCS (see Materials and methods). This approach allows for quantitation of density of fluorescent entities per square micrometer (amplitudes of correlation functions) and to calculate the number of overlapping (colocalized) entities in a series as well as their dynamics. In control experiments, PIEZO1-GFP appeared as punctate entities consistent with channel clustering that often colocalized with CtxB–Alexa Fluor 555 (Fig. 5, top row). Quantification of PIEZO1-GFP/CtxB colocalization revealed that PIEZO1-GFP colocalizes with CtxB at ∼40% (Fig. S9 A). These colocalized entities dissociated upon addition of MBCD (Fig. 5 A, second row). This resulted in a larger fraction of smaller PIEZO1-GFP entities detected in the membrane (Fig. 5 B) as well as an increase in colocalized entities (Fig. 5 C). The density of fluorescent CtxB did not change across conditions, indicating that GM1 ganglioside levels were not affected (Fig. S9 B).

As previously observed (Fig. 3 D), the MBCD-treated PIEZO1-GFP entities also appeared to diffuse faster in the basal membrane than in controls, consistent with the idea that cholesterol-rich domains spatially confine PIEZO1-GFP. Interestingly, the diffusion speed of colocalized PIEZO1-GFP/CtxB entities was faster after MBCD treatment compared with the diffusion speed of colocalized entities in the control condition (Fig. 5 D). The MBCD treatment approximately doubled the diffusion speed of both individual CtxB entities (Fig. S9 C) and the PIEZO1-GFP/CtxB colocalized entities (Fig. 5 D). The effects of MBCD treatment were also observed in time-course experiments at the single-cell level (Fig. S10). Dynasore- and cholesterol-treated cells showed no significant difference in either PIEZO1-GFP cluster numbers (Fig. 5 B) or their colocalization with CtxB (Fig. 5 C) compared with control. In addition, CtxB did not significantly affect the diffusive dynamics of PIEZO1-GFP. The validity of ICCS as a tool to assess colocalization was verified and confirmed in a series of control experiments (Fig. S11 and Fig. S12).

Collectively, these results suggest that PIEZO1-GFP associates into larger clusters in a cholesterol-dependent manner. Disruption of cholesterol organization in the membrane facilitates diffusion of PIEZO1-GFP, suggesting a confinement by cholesterol-rich fences.

Using two-photon microscopy (excitation 780 nm), we were able to distinguish the fluorescent signals emitted from Laurdan associated with lipid-ordered membrane domains (emission 400–460 nm) and lipid-disordered ones (emission 470–530 nm) and independently assess the degree of colocalization between each lipid phase and PIEZO1-GFP using ICCS. As previously observed, the density of PIEZO1-GFP fluorescent entities appeared to increase upon cholesterol domain disruption after MBCD and dynasore treatments (Fig. 6 B). The density of “ordered” Laurdan (fraction of Laurdan associated with ordered lipids) was significantly increased upon cholesterol enrichment (Fig. 6 C). Conversely, Laurdan phase distribution appeared to be sensitive to cholesterol disruption, displaying a significant increase in detectable “disordered” Laurdan entities (fraction of Laurdan associated with disordered lipids) after exposure to dynasore (Fig. 6 D), but surprisingly not after MBCD treatment. In the control samples PIEZO1-GFP was found to colocalize with both the ordered Laurdan entities (Fig. 6 E) and the disordered counterpart (Fig. 6 F), and both populations appeared to increase their respective density of colocalized entities only after being treated with dynasore.

Dynasore treatment slightly increased the average density of PIEZO1-GFP colocalized with the ordered membranes (Fig. 6 E; dynasore, 19.1 colocalized entities/µm², +1% versus control) but had a dramatic effect on its association with the disordered membranes (Fig. 6 F; dynasore, 52.8 colocalized entities/µm², +285% versus control) membrane regions detected by Laurdan. Cholesterol enrichment did not affect the density of PIEZO1-GFP colocalized with ordered lipids, despite increasing the fraction of ordered Laurdan.

These results suggest that dynasore treatment increases the density of detectable disordered membrane domains, causing a large fraction of PIEZO1-GFP to associate with disordered lipids. MBCD treatment did not appear to significantly increase the fraction of disordered membrane domains while increasing the density of PIEZO1-GFP entities, highlighting the different mechanism of action of the two drugs on HEK293T membranes.

In this study, we examined the effect of cholesterol depletion on PIEZO1’s response to mechanical stimuli in the form of negative pressure applied to cell-attached membrane patches. Our patch-clamp results indicate that the manipulation of membrane cholesterol alters PIEZO1-mediated mechanotransduction. We observed changes to collective (synchronized) inactivation and a latency of activation of PIEZO1 channel clusters when MBCD or dynasore was added to stably expressing HEK293T cells (Fig. 1).

Several possible explanations describe the observed phenomena. The impact of these drugs on channel activity could arise from effects on PIEZO1 organization, disrupted lipid-channel direct interactions, altered organization of cytoskeletal structures, and/or changes to plasma membrane organization and its global mechanical properties such as tension and fluidity. How these drugs impact the plasma membrane is still a matter of debate.

MBCD has been extensively used as a chelator of cholesterol (Zidovetzki and Levitan, 2007) for the removal and delivery of the lipid to membranes in vitro, and its mechanisms of action have been explored in detail (López et al., 2011; Chun et al., 2013). In contrast, the exact mechanism of action on the plasma membrane of dynasore (a known inhibitor of dynamin and endocytosis; Preta et al., 2015) is currently unknown and is speculated to interfere with calcium release by acting on the PLCγ-ERK1/2 cascade (Basagiannis et al., 2017), known to be involved in lipid remodeling. The induction of phosphorylation of PLCγ via dynasore, leading to enhanced lipase activity (Gresset et al., 2010), might explain its effect on the integrity of lipid domains. By changing the organization of membrane lipids, we have found that Piezo1 sensitivity, inactivation, and latency of the response can be modulated.

The rapid inactivation of PIEZO1 by membrane stretch is a key feature of this mechanosensitive ion channel (Coste et al., 2010; Gottlieb et al., 2012). Slowing inactivation in mutant PIEZO1 channels has been linked to several mechanopathologies, the first being hereditary xerocytosis, a familial anemia (Zarychanski et al., 2012; Bae et al., 2013b). The effect of cholesterol depletion on PIEZO1 has similar characteristics observed for these mutations (Bae et al., 2013b). For example, two mutations introduced to the ion-conducting pore of PIEZO1 resulted in a channel devoid of inactivation (Bae et al., 2013b). Interestingly, the slow-inactivation phenotype of the PIEZO1 R2456H mutant could be further exacerbated by cholesterol depletion via MBCD (Fig. 2, B and C). PIEZO1 R2456H displays a lower activation threshold (Fig. 2, E and F) besides its reduced inactivation (Fig. 2 C), in contrast with previous observations in whole-cell patch-clamp experiments (Glogowska et al., 2017; Romero et al., 2019). Furthermore, whenever we observed loss of inactivation, we also detected a delay in the latency of response, suggesting that the two parameters are functionally linked. Taken together, these data highlight that both lipids and mutations can affect PIEZO1 inactivation, with synergistic results.

The relationship between loss of channel inactivation and shift in sensitivity remains unclear. Slowly inactivating double mutants displayed a leftward shift in pressure dependence (Bae et al., 2013b), whereas MBCD-treated WT proteins experienced a rightward shift. Conversely, enrichment of membrane fatty acids with DHA resulted in the sensitization of PIEZO1-GFP but also in loss of inactivation. The repeated stimulation of membrane patches from stable PIEZO1-GFP cells treated with DHA caused a delay in the channel latency and the progressive loss of inactivation at high pressure regimes (Fig. S7 E), a finding in contrast with a recent study using whole-cell patch-clamping and activation of PIEZO1 by cell indentation (Romero et al., 2019). This suggests that the magnitude and mode of application of the mechanical stimulus also play a role in defining the inactivation behavior of PIEZO1. The activation thresholds quantified in Table 2 are in good agreement with previously published estimates (Cox et al., 2016a; Bae et al., 2013b).

We also observed that the mode of expression of PIEZO1 and its mutant plays a role in determining the channel mechanosensitive properties. The sensitivity of all transiently expressed channels did not change after MBCD treatment, as opposed to what we observed in PIEZO1-GFP stable cells and N2A, but all nevertheless experienced a significant loss of inactivation after cholesterol depletion. We speculate that transient (24–48-h) overexpression of the channel was not sufficiently long to allow proper localization of the PIEZO1 channel with respect to cholesterol domains. In contrast, we can imagine that a stable cell line would have regulated its PIEZO1 levels and subcellular distribution over many generations to avoid long-term toxicity due to acute overexpression, allowing for a significant MBCD-driven modulation of both inactivation and sensitivity to be detected. This conclusion is supported by MBCD’s effect on N2A cells (Fig. S5). It is worth noting that plasmid-borne PIEZO1 levels were found to be higher than in stable cells (Fig. S6 E) and N2A cells (Fig. S5 F). Together, these results suggest that cholesterol plays an important role in regulating PIEZO1 inactivation, while PIEZO1 sensitivity is dependent on the mechanical properties of its host membrane rather than on the presence of cholesterol specifically.

Because of the possibility that cholesterol regulates inactivation by affecting the clustering of channels, we performed STORM superresolution microscopy to investigate the behavior of PIEZO1-GFP in the basal membrane of our stable cell line. Currently, the latency of channels could only be explained by fracture kinetics that predicted the presence of clusters (Bae et al., 2013a). The association of PIEZO1 with cholesterol-rich membrane domains appears to be a crucial determinant of its mechanotransductive properties (Qi et al., 2015). Possible evidence of cluster formation was also observed previously using structure illuminated microscopy (Cox et al., 2016a; Maneshi et al., 2018).

Using superresolution microscopy, we demonstrated that PIEZO1 indeed forms membrane clusters at the nanoscale. PIEZO1-GFP clusters in the cell membrane have a broad range of sizes (Fig. 4), from presumably isolated single channels up to assemblies of the protein in the tens of thousands of square nanometers. The stable cell line probed here displays a range of PIEZO1-GFP expression levels, as well as a nonhomogeneous distribution of the clusters on the cellular basal membrane (Fig. 4 D). We identified cluster subpopulations that are disrupted and are more sensitive to cholesterol modulation. Approximate estimates of the PIEZO1 protein area based on the most detailed available structure (PDB accession no. 6B3R) suggest that a single trimer projects a ∼280-nm² area when observed perpendicularly to the membrane plane (Ge et al., 2015). Estimating channel numbers in the cluster is complicated by the dome-like structure that is adopted by PIEZO1 and the need to assess how they would pack (Guo and MacKinnon, 2017). Nevertheless, all the treatments used appear to reduce the median density of PIEZO1-GFP clusters and significantly enrich a subset of smaller clusters (∼1,000–1,500 nm² in area; Fig. 4 B, histogram). Correlative experiments on live cells allowed us to follow the cluster biophysical properties during treatment (Fig. S10).

Removal of membrane cholesterol via MBCD appeared to generate more PIEZO1-GFP entities in the membrane (Fig. 5 B) and to cause a faster diffusion of channels in the basal membrane (Figs. 3 D and 5 D), suggesting that cholesterol domains confine PIEZO1-GFP clusters within lipid boundaries, rather than cytoskeletal fences (Fig. S8). Interestingly, MBCD treatment increased the colocalization between PIEZO1-GFP and the raft marker CtxB (Fig. 5 C), suggesting that upon cholesterol disruption, PIEZO1-GFP clusters retained interactions with lipid raft components that are enriched in GM1 ganglioside, the target of CtxB. The dynasore treatment, on the other hand, failed to increase the diffusion of PIEZO1-GFP clusters in the membrane (Fig. 5 D) and did not change the degree of colocalization with the marker CtxB compared with control (Fig. 5 C).

The association of PIEZO1 and lipid rafts has been suggested, since depletion of STOML3 affects PIEZO1 activation (Poole et al., 2014). These colocalized PIEZO1-GFP/CtxB entities also appeared to diffuse faster than in the control condition (Fig. 5 D), similarly to what was observed in single-color PIEZO1-GFP in situ imaging (Fig. 3 D). Our most accurate estimates for the diffusion of PIEZO1-GFP clusters were in the order of 0.01 µm²/s, in the same order of magnitude as previously reported by the Pathak laboratory (0.067 µm²/s; Ellefsen et al., 2019). This is in support of the idea that channel inactivation could be regulated by raft-mediated clustering of the protein.

We used Laurdan imaging to investigate the localization of PIEZO1-GFP in the membrane with respect to liquid disordered and more ordered regions. In control conditions, ∼40% of detected PIEZO1-GFP entities appeared to be colocalized with ordered domains (Fig. 6 E and Fig. S9), whereas the remaining fraction colocalized with disordered lipid entities (Fig. 6 F). MBCD treatment did not increase the colocalization of PIEZO1-GFP with disordered lipids (Fig. 6 F), supporting the idea that an interaction between ordered, raft-like membrane fractions and PIEZO1-GFP clusters is maintained after cholesterol depletion.

This also suggested that, after dynasore treatment, the cell membrane retained intact cholesterol-rich domains able to confine PIEZO1-GFP diffusion observed for the control. In some experiments, the treatment also caused an increase in detectable PIEZO1-GFP entities suggestive of cluster breakdown (Fig. 6 B). Interestingly, dynasore, but not MBCD, increased the number of disordered Laurdan entities (Fig. 6 D) while promoting the colocalization of PIEZO1-GFP clusters with disordered lipid entities (Fig. 6 F). In this case, PIEZO1 showed no increased retention of CtxB (Fig. 5 C), suggesting that an increased fraction of channel clusters became associated with disordered lipid domains while being separated from cholesterol-rich membrane fractions.

This might explain the difference observed in electrophysiology between MBCD, which maintains CtxB/PIEZO1 colocalization, and dynasore, which does not, and link to the effect on channel inactivation. Possibly, as PIEZO1 clusters become separated from rafts, they lose their ability to collectively inactivate and localize to a different mechanical environment. The effect of MBCD on the mechanical properties of the membrane could produce stiffer environments able to desensitize the channels of the cluster (Byfield et al., 2004; Khatibzadeh et al., 2013). This suggest that during the channel gating cycle, a lipid-dependent component plays a major role in regulating the kinetics of inactivation. This fits well with recent extensive data from the Vasquez laboratory (Romero et al., 2019) and extends the effects to cholesterol. These lipid-mediated effects may also explain why native PIEZO1 shows ameliorated inactivation in some cell types (Rode et al., 2017; Del Mármol et al., 2018). Cholesterol enrichment did not appear to increase the fraction of PIEZO1-GFP in proximity of ordered lipids (Fig. 6 E) or CtxB (Fig. 5 C), but significantly increased the fraction of ordered lipids (Fig. 6 C), indicating that cholesterol enrichment induced an enlargement of existing cholesterol domains. Also, cholesterol enrichment did not sensitize the PIEZO1-GFP response or affect its kinetics (Fig. 1 B), indicating that the additional membrane stiffening due to ordered lipids did not affect the channel response. In contrast, PIEZO1 has been reported to be desensitized following membrane stiffening by supplementation of saturated fatty acids (Romero et al., 2019), which become incorporated into membrane phospholipids. It appears therefore that PIEZO1 sensitivity is mostly affected by the stiffness of nonraft membranes, in which the higher line tension (Uline et al., 2012; Markin and Sachs, 2015) of the disordered membrane phase lowers the threshold for channel activation.

We envision a scenario based on the “picket-fence model” (Fujiwara et al., 2016) in which cholesterol fences (rather than cytoskeletal fences) organized as a dynamic network on the cell surface delimit the 2D diffusion of PIEZO1 clusters. In our view, stiffened lipid structures connecting several clusters would allow fast propagation of membrane tension between them, facilitating the synchronized gating behavior of the channels.

Further work should aim to probe whether these effects are generated by specific lipid–protein interactions, global effects on the mechanics of the bilayer, or both. Cholesterol-mediated effects on PIEZO1 may be relevant to cell types frequently exposed to cholesterol in the circulation such as red blood cells, vascular endothelial cells (Park et al., 1998), leukocytes (Zhang et al., 2011), or cells displaying high levels of membrane cholesterol such as myelinating oligodendrocytes and Schwann cells (Belin et al., 2017).

Our electrophysiology results suggest that the inactivation of PIEZO1 currents could be mediated by cholesterol domains bridging PIEZO1 clusters and not only an intrinsic effect of the conformational changes of the PIEZO1 protein during its gating cycle, as shown by previous mutation studies (Bae et al., 2013b; Glogowska et al., 2017; Wu et al., 2017b; Moroni et al., 2018; Zheng et al., 2019). This ability of lipids to modulate or even reverse the slow-inactivating phenotype of PIEZO1 mutants has recently been suggested (Romero et al., 2019). Along with this study, our work highlights the critical importance of studying PIEZO1 and its disease-causing variants in native PIEZO1-expressing cells where possible.

David A. Eisner served as editor.

We acknowledge the access to instruments at the University of New South Wales Biomedical Imaging Facility and Mark Wainwright Analytical Centre, as well as Cancer Institute NSW for purchase of the Zeiss 880 multiphoton microscope.

P. Ridone has been supported by the University of New South Wales (University International Tuition Award); C.D. Cox was supported by an Early and Mid-Career Fellowship from New South Wales Health; and M. Vassalli was a recipient of the Australian Government Endeavor Fellowship 5504_2016. This project was also supported by Australian Research Council Grant DP160103993 and the National Health and Medical Research Council of Australia (Principal Research Fellowship APP1135974) to B. Martinac. We also acknowledge funds from the Office of Health and Medical Research, New South Wales State Government.

The authors declare no competing financial interests.

Author contributions: P. Ridone designed and performed electrophysiology experiments on all samples except N2A cells; analyzed all electrophysiology data; sample preparation and data acquisition in microscopy experiments; designed and produced the figures in the manuscript; and wrote the manuscript. E. Pandzic was responsible for sample preparation and data acquisition for microscopy experiments; analysis of all microscopy data; was involved in the design and production of figures and videos; and wrote the manuscript. M. Vassalli initialized the collaboration between B. Martinac, P. Ridone, and E. Pandzic; performed the pilot microscopy experiments (TIRF and STORM); suggested Laurdan and Filipin experiments; and critically reviewed the manuscript. C.D. Cox performed electrophysiology experiments on N2A cells; performed cell viability assay; and critically reviewed the manuscript. A. Macmillan involved in the experimental design of microscopy experiments and critically reviewed the manuscript. P.A. Gottlieb provided stable PIEZO1-GFP cell lines and critically reviewed the manuscript. B. Martinac was head of laboratory and PhD supervisor of P. Ridone; provided equipment and funds; and critically reviewed the manuscript.