Piezos are pore-forming subunits of mechanically activated channels

Introduction

Mechanically-activated (MA) currents have been described in various mammalian cells, including inner ear hair cells¹, somatosensory dorsal root ganglion neurons², vascular smooth muscle cells³, and kidney primary epithelia⁴. The majority of these MA currents are cationic with Ca²⁺-permeability, leading to a search for cation channels able to convert mechanical forces into such currents. Few MA channels have been described to date^5-7; however, none of the candidates have been shown convincingly to mediate the physiological relevant non-selective cationic MA currents in mammals.

Mouse piezo1 (mpiezo1) was recently identified as a protein required for MA currents in a mammalian cell line. Expressing mpiezo1 or related mpiezo2 in a variety of mammalian cell lines induces large MA cationic currents⁸. mpiezo1-induced currents are inhibited by GsMTx4, a toxin widely used to study MA channels⁹. Piezo1 and piezo2 contain over 30 putative transmembrane domains and do not resemble known ion channels or other protein classes. Piezo proteins could be non-conducting subunits of cationic ion channels required for proper expression or for modulating channel properties^10-12. Alternatively, piezo proteins may define a novel class of ion channels involved in mechanotransduction.

Mechanosensitivity of dpiezo

Piezo sequences are present in the genomes of many animal, plant, and other eukaryotic species. Functional analysis of piezos from distant species could demonstrate a conserved role of these proteins in mechanotransduction; furthermore, a comparative analysis of MA currents could elucidate unique pore properties of channels induced by piezos from distinct species. We focused on the apparently single member of D. melanogaster piezo (dpiezo), as this distant invertebrate species is widely used to study mechanotransduction using genetic approaches^13-17. dpiezo is 24% identical to mammalian piezos, with sequence conservation throughout the length of the proteins (Supplementary Fig. 1). We cloned the full length dPiezo cDNA into pIRES2-EGFP vector. We recorded MA currents from fluorescent HEK293T cells expressing dPiezo-pIRES2-EGFP by applying force to the cell surface while monitoring transmembrane currents at constant voltage using patch-clamp recordings in the whole-cell configuration^2,18,19. dpiezo, but not mock-transfected cells showed large MA currents (Fig. 1a, b). These currents display a time constant of inactivation τ of 6.2 ± 0.3 ms (n = 32 cells) at -80 mV when fitted with mono-exponential function, which is faster than observed for mpiezo1 (~16 ms) and more comparable to mpiezo2 (~7 ms)⁸. Similar to its mammalian counterparts, dpiezo- MA currents are characterized by a linear current-voltage relationship with a reversal potential around 0 mV, consistent with it mediating a non-selective cationic conductance (Fig. 1c). We further characterized dpiezo-induced currents in HEK293T cells in response to negative pressure pulses applied through the recording pipette in the cell-attached mode, an alternative mechanosensitivity assay. Overexpression of dpiezo induced stretch-activated currents (Fig. 1d-e) with a pressure for half-maximal activation (P₅₀) of -31.8 ± 2.8 mm Hg (Fig. 1f), similar to the P₅₀ calculated for mpiezo1-induced currents (~30 mm Hg)⁸. Therefore, mechanosensitivity of the piezo family is conserved in invertebrates. Importantly, we demonstrate the physiological relevance of dpiezo in vivo in an accompanying paper²⁰.

Pore properties of piezos

We next compared fundamental permeation properties of mpiezo1- and dpiezo. Ruthenium red (RR), a polycationic pore blocker of TRP channels^21,22, blocks mpiezo1- and mpiezo2-induced MA currents⁸. We found that RR is a voltage-dependent blocker of mpiezo1, with an IC₅₀ value of 5.4 ± 0.9 μM at -80mV (Fig. 2a-c): At a concentration of 30 μM, extracellular RR inhibited inward MA currents without affecting outwards currents. Such voltage-dependence is a characteristic of open channel block. A high concentration of RR (50 μM) included in the pipette solution in the whole cell configuration showed no evidence of block, as large MA currents still displayed a linear current-voltage relationship (Supplementary Fig. 2). These results suggest RR blocks the pore of mpiezo1-induced MA channels from the extracellular side. Remarkably, dpiezo-induced MA currents were insensitive to RR concentrations that potently blocked mPiezo1-induced currents (Fig. 2d-e). Together, these results demonstrate that overexpression of dpiezo or mpiezo1 gives rise to MA channels with distinct channel properties.

Next, we set out to determine the single channel conductance (γ) of MA channels induced by piezo proteins by using negative-pressure stimulations of membrane patches in cell-attached mode. Figure 3 illustrates the single MA channel properties of mpiezo1 or dpiezo. Openings of stretch-activated channels showed a striking difference in amplitude of single channel currents (Fig. 3a), as determined from the single channel current-voltage relationship for mpiezo1- and dpiezo (Fig. 3b, c). Linear regression of these I-V relationships resulted in slope-conductance values in these recording conditions of 29.9 ± 1.9 and 3.3 ± 0.3 pS for mpiezo1- and dpiezo-induced MA currents, respectively (n = 7 and 5 cells; mean ± s.e.m.). Therefore, dpiezo-dependent channels are 9-fold less conductive than mpiezo1-dependent channels.

mpiezo1 oligomerization

The pore of the majority of ion channels is formed by the assembly of transmembrane domains from distinct subunits (e.g., voltage-gated K⁺ channels, ligand-gated ion channels) or structurally repetitive domains within a large protein (e.g., voltage gated Na⁺ and Ca²⁺ channels). Since piezos lack repetitive transmembrane motifs presumably they oligomerize to form ion channels. To test this hypothesis, we determined the number of subunits in piezo complexes by expressing GFP-mpiezo1 fusion proteins in Xenopus oocytes, imaging individual spots with total internal reflection microscopy (TIRF), and counting discrete photobleaching steps (Fig. 4a-b and ref. 8). N-terminal GFP-mpiezo1 functionality was confirmed by overexpression in HEK293T cells (Supplementary Fig. 3). We used several GFP-fusion constructs of ion channels with known stoichiometry as controls: voltage-gated Ca²⁺ channel (α1E-GFP; monomer), NMDA receptor (NR1 co-expressed with NR3A-GFP; dimer of dimers), and cyclic nucleotide gated (CNG) channel (XfA4-GFP; tetramer)²³. We found that complexes of mpiezo1 frequently exhibited at most four photobleaching steps, consistent with the idea that piezos homo-multimerize. Fluorescent mpiezo1 (or CNG) complexes exhibiting bleaching in fewer than four steps can be explained by non-functional GFP or pre-bleached GFP²³ or general bias against noisier multi-step traces during data analysis (see Methods). Histograms of the number of photobleaching steps observed for mpiezo1 complexes were comparable to histograms obtained from tetrameric CNG channels (Fig. 4c). These results suggest that in living cells, piezos assemble as homo-multimers.

We further characterized piezo proteins biochemically by heterologously expressing and purifying mpiezo1 C-terminally fused with a glutathione S-transferase (mpiezo1-GST). Functionality of mpiezo1-GST was confirmed by overexpression in HEK293T cells (Supplementary Fig. 3). We observed a protein band at a position near the 260 kDa protein marker on a Coomassie blue-stained denaturing protein gel (Supplementary Fig. 4a). Western blot with a GST (S. japonicum form) antibody (Supplementary Fig. 4b) or a mpiezo1 specific antibody⁸ (Fig. 4) confirmed the presence of mpiezo1-GST in the mpiezo1-GST sample. Using native gel electrophoresis and Coomassie blue staining, we detected a prominent protein band at a position near the 1,236 kDa protein marker only in the mpiezo1-GST sample (Fig. 4d). Western blot using mpiezo1 antibody confirmed that this major band contains mpiezo1 (Fig. 4e). These data indicate that the purified mpiezo1-GST protein complex has a molecular weight of about 1.2 million Daltons, four times the predicted molecular weight of a single mpiezo1-GST polypeptide (318 kDa). Next, we asked whether any endogenous proteins are present in this mpiezo1-containing complex. Mass spectrometry of the ~1.2 million Dalton protein complex mainly detected peptides derived from mpiezo1-GST, but not from other endogenous membrane proteins. Although several non-transmembrane proteins were also detected, most of them were also present in the control sample, indicating an absence of specific interacting proteins in the complex (Supplementary Table 1). Moreover, mass spectrometry of the whole purified solution samples prior to gel electrophoresis confirmed that no other ion channel protein was detected (Supplementary Table 2). This argues that mpiezo1 is not tightly associated with any endogenous pore-forming protein.

To further examine whether this piezo complex is indeed a tetramer, we treated the purified mpiezo1-GST protein with the crosslinker paraformaldehyde (PFA) and subjected the samples to denaturing gel electrophoresis and western blotting. PFA-treated samples contained three major additional higher-order piezo containing bands, with longer PFA treatments increasing the prominence of the higher bands (Fig. 4f). The distribution of the bands on the 3-8% gradient gel suggests that the four bands correspond to monomer, dimer, trimer and tetramer of mpiezo1-GST (Fig. 4f). The observation that mpiezo1 is crosslinked by formaldehyde, a crosslinker with a relative short spacer arm (2.3-2.7 Å), suggests that the subunits form a tetramer.

It is possible that mpiezo1 oligomers associate with other proteins; however such an association might not withstand the GST purification step. To probe this, we performed PFA crosslinking experiments on living cells prior to the purification procedure. On a native gel, the mpiezo1-GST complex purified from PFA-treated cells also migrated to the position near the 1236 kDa protein marker, similar to the sample from untreated cells (Fig. 4g). On a denaturing gel, on-cell PFA treatment resulted in four distinct Piezo1-specific bands, similar to results of PFA treatment on the purified complex (Fig. 4h). This suggests that mpiezo1 is not tightly associated with other proteins large enough to discernibly alter its size on denaturing gels, and confirms the results from mass spectrometry. However, our cross-linking studies with paraformaldehyde could miss weak interactors with mpiezo1. Regardless, together with the results obtained from single molecule photobleaching analysis in living cells, our biochemical data suggest that mpiezo1 forms a homomultimeric ion channel, most likely as a homotetramer.

mpiezo1 reconstitution in lipid bilayers

Finally, to assess if piezo proteins were sufficient to recapitulate the channel properties recorded from piezo-overexpressing cells, we reconstituted purified mpiezo1 proteins into lipid bilayers in two distinct configurations: droplet interface lipid bilayers (DIBs) assembled from two monolayers^24-26 (Fig. 5a-e, and 5l-q) and proteoliposomes²⁷ (Fig. 5f-i). In the first configuration, mpiezo1 was reconstituted into asymmetric bilayers that mimic the cellular environment: The extracellular facing lipid monolayer is predominantly neutral whereas the intracellular facing leaflet is negatively charged²⁸. In contrast, the lipid composition of the bilayer in the second configuration is uniform.

In the DIBs setting, representative segments from a 6 minute recording obtained at –100 mV show brief, discrete channel openings (Fig. 5a, b) blocked by addition of 50 μM RR to the neutral facing compartment (Fig. 5c). In contrast, no effect was observed when RR was introduced into the negative facing compartment (not shown). We detected efficient block of channel activity even at 5 μM RR (not shown). The asymmetric accessibility of RR block of reconstituted channels agrees with the data obtained from mpiezo1-overexpressing HEK293T cells (Fig. 2 and Supplementary Fig. 2), thereby establishing the fidelity of the assays and validating mpiezo1 protein as an authentic ion channel. The piezo currents exhibit ohmic behavior; records displayed at higher resolution (Fig. 5b) clearly demonstrate the occurrence of unitary events with γ values obtained from conductance histograms of 118 ± 15 pS and 80 ± 6 pS (n = 6) in symmetric 0.5 M KCl from the negative and positive branches of I-V plots, respectively (Fig. 5d, e).

A similar pattern of activity was obtained from mpiezo1 reconstituted in asolectin liposomes²⁷ (Fig. 5f-k). A selection of recordings shows the presence of two channels in the membrane which reside predominantly in the open state (Fig.5f, g), as discerned in a higher time resolution display (Fig. 5k) . These recordings were obtained in the presence of 50 μM RR inside the recording pipette, to ensure functional selection of a single population of mpiezo1 channels facing the RR-free compartment. mpiezo1 in asolectin proteoliposomes under these conditions (symmetric 0.2 M KCl) exhibits a γ = 110 ± 10 pS at V = –100 mV and 80 ± 5 pS at V = 100 mV (Fig. 5h-j) (n=8). Finally, reconstitution of control samples purified from nontransfected cells as well as heat-denatured purified mpiezo1-GST into either bilayer systems under otherwise identical conditions failed to reproduce this pattern of channel activity (not shown).

We then tested the ability of the reconstituted mpiezo1 to conduct sodium (Fig. 5 l-q). Initially, single channel currents were recorded from asymmetric bilayers in symmetric 0.2 M KCl; γ = 58 ± 5 pS (Fig. 5 l, o). Subsequent addition of 0.2M NaCl in presence of 0.2M KCl increased the unitary conductance of reconstituted channels to 95 ± 5 pS (Fig. 5 m, p) while retaining sensitivity to RR block (Fig. 5n, q). These results confirm that these channels conduct both sodium and potassium as would be expected from a cationic non-selective channel. This assertion was further substantiated by recording mpiezo1 currents from proteoliposomes under bi-ionic conditions (0.2 M KCl/0.2M NaCl) (Supplementary Fig. 5a-h). A summary of the current–voltage relation for the mpiezo1 channel, extracted from 204,088 events obtained in three experiments, shows that the single channel current is ohmic between –100 and 200 mV with a slope conductance of 102 ± 2 pS (Supplementary Fig. 5i). The current reversed direction at 0.0± 0.3 mV demonstrating that the channel does not select between K⁺ and Na⁺, and importantly, displays open channel block by RR (Supplementary Fig. 5j-l).

The difference in γ between overexpressed mpiezo1 in cells and reconstituted mpiezo1 in lipid bilayers may be attributed to many variables, including the distinct lipid environments which are known to strongly influence conductance measurements^29-33. Moreover the ionic conditions used in the two systems are different, as divalent cations present in HEK293T cell-attached experiments also affect the conductance values. Indeed, when divalent cations are excluded from the recording pipette, γ of mpiezo1-induced currents in HEK293T cells is 58.0 pS ± 1.5 pS (150 mM NaCl solution, Supplementary Fig. 6), compared to 29.9 ± 1.5 pS in the presence of divalent ions (Fig. 3). The near equivalence of γ values together with the similar pattern of channel activity demonstrates that reconstitution of mpiezo1 into two distinct bilayer systems produces channels with identical functional properties (Supplementary Table 3).

Future reconstitution and recording of dpiezo in lipid bilayers will show whether the difference in conductance between mpiezo1 and dpiezo arises from intrinsic properties. The membrane milieu and lipid composition are known to modulate the activity of the embedded channel proteins in a drastic and deterministic manner [e.g., see^29-33]. It is not entirely surprising that the conditions to emulate the cellular environment in the reconstituted system in terms of the mechanical state of the membrane or its lipid composition have thus far been inadequate to retrieve the activation features of MA ion channels. Furthermore, the complexity of protein clusters and dynamic cytoskeletal interacting partners at the cell membrane³⁴ introduce regulatory constraints on channel activity. Further investigation may clarify whether piezo ion channel subunits are intrinsically mechanosensitive or use unknown interacting partners to sense membrane tension.

Conclusion

Here, we provide compelling evidence to support the hypothesis that piezo proteins are indeed ion channels. First, overexpression of dpiezo or mpiezo1 in a human cell line gives rise to MA channels with distinct biophysical and pore-related properties. Second, isolated piezo complexes do not contain detectable amounts of other channel-like proteins. Finally, purified mpiezo1 protein reconstituted into proteoliposomes and planar lipid bilayers in the absence of any other cellular components gives rise to RR-sensitive cationic ion channel activity. The mouse piezo1 complex is estimated to weigh ~1.2 million Daltons with 120-160 transmembrane segments, being, to our knowledge, the largest plasma membrane ion channel complex identified to date.

METHODS SUMMARY

Supplementary Material