Abstract
The transduction of force into a biological signal is critical to all living organisms. Recently, disruption of ordered lipids has emerged as an ‘atypical’ force sensor in biological membranes; however, disruption has yet to link with canonical channel mechanosensation. Here we show that forceinduced disruption and lipid mixing activates TWIK-related K+ channel (TREK-1), and that this activation is dependent on phospholipase D2 (PLD2). PLD2 transduces the force into a chemical signal phosphatidic acid (PA) that is then sensed by TREK-1 with a latency of <3 ms. TREK-1 then produces a mechanically induced change in membrane potential. Hence, in a biological membrane, we show the ordered lipid is the force sensor, PLD2 is a chemical transducer, and the ‘mechanosensitive’ ion channel TREK-1 is a downstream effector of mechanical transduction. Confirming this central role for PA singling in force transduction, genetic deletion of PLD decreases mechanosensitivity and pain thresholds in D. melanogaster.
Introduction
All cells must sense force (mechanosensation), not just sensory pain nerves, including those that respond to the heartbeat, skeletal muscle stretch, and epithelial cell adhesion 1–3, Recent studies show that disruption of ordered proteins and lipids is a non-canonical sensor of mechanical force in biological membranes4,5. Force was shown to disassemble and flatten caveolae4 and force disrupts ordered lipids and lipid clusters5. The ordered phase, sometimes called a lipid raft, is enriched in saturated lipids and palmitoylated proteins6,7 and is subject to disruption (see Supplemental Fig. 1a-b). In C2C12 muscle cells, chemical disruption releases a palmitoylated protein phospholipase D (PLD2) from lipid rafts activating a mechano-transduction pathway (Supplemental Fig. 1c-d).5 This mechanical activation of PLD2 is shown to cause muscle cell differentiation5, miacropinocytosis8, and may act as a force senor upstream of mTOR9. These data point to lipid domains as bona fide sensors of force.
The downstream effector molecules of this non-canonical pathway have yet to be described for excitable systems. Here we hypothesize anionic signaling lipids are a transducer of mechanical force to ion channels. Many ion channels and transporters are regulated by lipids and could serve as effector molecules of lipid raft disruption. Additionally, in eukaryotic membranes10,11 excess lipids and cellular cytoskeletal structures can resist tension on and propagation through plasma membranes12–15, suggesting the non-canonical disruption mechanism likely plays a prominent role for at least some ion channel activation. Linking atypical raft mechanosensation directly to a canonical mechanosensitive channel would constitute a distinct mechanism separate from the traditional “force from lipid” model typical of sensory systems where the channel directly senses stretch16,17.
TWIK-related K+ channel type 1 (TREK-1) is a mechanosensitive member of the two-pore-domain potassium (K2P) family that protects against pain18, and it is one of many channels regulated by lipids suggesting it could be down stream of raft disruption. Here we show membrane disruption releases PLD2 and generates the anionic lipid-signal phosphatidic acid (PA) that directly activates TREK-1. Blocking the synthesis of PA completely blocks stretch activation of TREK-1 in whole-cell patch-clamp electrophysiology. We conclude that in biological membranes, raft disruption and lipid mixing is a dominant mechanosensor for TREK-1 channels and the anionic lipid PA is an essential intermediary.
Results
Mechanical disruption of lipid rafts in muscles cells
The best studied raft domains contain saturated lipids, cholesterol, sphingomyelin, and monosialotetrahexosylganglioside1 (GM1)7. Cells contain a second class of lateral heterogeneity comprised of anionic lipids PIP2 and PIP3, which cluster19–21 separate from GM1 domains and separate from each other5,22,23 (Supplemental Fig. S1a). Lipid domains are thought to be sub-100 nm diameter structures, may include protein interactions24, but even with super-resolution imaging techniques such as direct stochastic optical reconstruction microscopy (dSTORM) and stimulated emission depletion (STED)25–28 their exact structure remains controversial29,30. We previously used dSTORM to image livespontaneous and chemical disruption of GM1 domains in muscle cells5 leading us to asked if force induced disruption could directly be measured with imaging. Imaging direct disruption of lipid domains by force is important since GM1 and PIP2 lipids are part of all cellular membranes and force would be an endogenous disruptor of lipid order for most cell types.
To directly image force-induced disruption of lipid domains, we developed a process to chemically fix disrupted domains in the membrane of differentiated C2C12 muscle cells (mouse myocytes) using a pump and shear chambers (ibidi µ-Slide I0.4 parallel-plate) (Supplemental Fig. 2a). We fixed the cells with a combination of both paraformaldehyde and glutaraldehyde and confirmed by fluorescence recovery after photobleaching (FRAP) that are conditions inhibit extra-domain diffusion of GM1 lipids (Supplemental Fig. 2b). We tested muscle cells since they are known to respond to force and our previous experiments suggested force disrupts their GM1 domains5. Furthermore muscle cells endogenously express PLD and TREK-1 channels5,31 allowing for observation of their dynamics without overexpression artifacts.
First, we initiated domain disruption by pumping phosphate buffered saline (PBS) across the C2C12 cells with a calibrated shear force of 3 dynes/cm2, a physiologically relevant force to muscle cells5,31. Immediately after applying shear, we infused fixative agents to the shear buffer which allowed the cells to be fixed in a mechanically-stimulated state. Non-sheared control cells were grown and treated similarly on static coverslips without shear. Fixed cells, with and without shear, were then prepared for dSTORM imaging using antibodies or cholera toxin B (CtxB) with high affinity corresponding to either PIP2 or GM1 domains respectively5. The amount of fluorescent labeling remained relatively constant between shear and static cells (Supplemental Fig. 2c).
We found shear forces robustly disrupted both GM1 and PIP2 domain size in fixed C2C12 cells (Fig. 1a-b). Interestingly, we observed that the shear specific disruption reduced GM1 domain size more than PIP2 domains. Prior to shear, the estimated size of GM1 domains were larger than PIP2 domains (167±3 vs. 154±1 nm in diameter). After shear, GM1 domains were smaller than PIP2 domains (131±3 vs. 139±1 nm in diameter, Fig. 1c-d). Domain size was determined using imagebased clustering with identical threshold parameters for all images (see methods). The absolute size has potential artifacts from the analysis32, however, the absolute size of rafts is inconsequential to our hypothesis. Rather, the change in raft size under otherwise identical conditions is important as it suggests force could directly affect palmitate-dependent protein localization to the GM1 domain and cause a translocation of PLD2 similar to chemical disruption5.
The cellular sensor of force in TREK-1 mechano-transduction
PLD2 is palmitoylated at cysteines near its PH domain, which is required to localize it to lipid rafts33. The PH domain also binds PIP2 which opposes the localization by palmitoylation (Supplemental Fig. 1c). We previously demonstrated that chemical disruption of GM1 domains activates PLD2 by substrate presentation 5 a result that is consistent with PLD regulation by caveolin34 and force induced flattening of caveolin4. After disruption, PLD2 rapidly translocates from GM1 to PIP2 domains where the enzyme mixes with its substrate phosphatidylcholine (PC) to produce PA. In addition to binding PIP2, PLD2 also binds directly to the C-terminus of TREK-1 and activates the channel through local production of phosphatidic acid35 presumably in the disordered region of the cell membrane5. Previous work also implicated TREK-1 C-terminus in mechanosensitivity36.
We reasoned mechanical PLD2 activation could transduce a component of force to TREK-1 in a biological membrane independent of direct TREK-1 stretch. If PLD2 and PA are intermediaries in TREK-1 activation, this mechanism would link atypical raft mechanosensation directly to a canonical mechanosensitive channel and support our proposed mixing and lipid-signaling mechanism for stretch activation. We estimated latency of PLD translocation to be very rapid5 (∼650 µsec) suggesting our proposed mixing mechanism is suitable for fast mechanical responses.
TREK-1 was previously shown to respond to mechanical force in cultured mammalian cells37. However, while TREK-1 is not endogenous to these cell types, PLD2 and GM1 domains are endogenous, and they are mechanosensitive5. However, prior to concluding TREK-1 must directly sense some type of curvature, tension, or force-from-lipid, a control experiment is needed that blocks PLD2 and shows the channel is still directly sensitive to force absent PLD2 activity, suggesting a control is needed to distinguish PLD2 mechanosensitivity from direct TREK-1 mechanosensitivity.
To distinguish the contribution of raft disruption from direct force acting on TREK-1, we tested TREK-1 currents with and without a catalytically dead K758R PLD2 mutant (xPLD2)38. The over expressed xPLD2 does not remove wildtype PLD from the membrane, but overexpressed xPLD2 is sufficiently high (>10 fold) 39 to outcompete wildtype binding to TREK-1 thus acting as a pseudo dominant negative. As such, the contribution of TREK-1 activation arising from raft disruption pathway should be xPLD2 sensitive, i.e., xPLD2 should block raft dependent TREK-1 activation. In contrast, the direct force from lipid (i.e. the membrane tension) acting on the channel should be PLD2 independent, resulting in TREK-1 stretch-activated current in the presence of xPLD2.
Consistent with previous experiments, overexpression of TREK-1 with wildtype or endogenous PLD2 led to a robust TREK-1 pressure-dependent current in HEK 293 cells (Fig. 2a). However, overexpression of xPLD2 inhibited all detectible stretch-activated TREK-1 current (Fig. 2b-c). This suggests that all detectable TREK-1 mechanosensitive current in HEK cells is PLD-dependent and an assumption that no endogenous proteins are contributing to the mechanotransduction pathway is likely incorrect. The PLD dependent current from TREK-1 is highly mechanosensitive responding to negative pressure (0 to −60 mmHg) with a half maximal pressure (P50) of ∼32 mmHg and yielding up to 200 pA of TREK-1 current (Fig. 2a and Supplemental Fig. S3a-b) consistent with previous studies40,41. The inhibition of current was highly significant with or without overexpression of PLD2 (p<0.002 and 0.007 respectively, Fig. 2c). Furthermore, the P50 significantly decreased from 37 to 31 mmHg when PLD2 was overexpressed (Supplemental Fig. 3b). Our stretch activated currents were measured using a pressure clamp in the inside-out patch configuration similar to previous experiments (see methods) allowing for direct comparison to these previous studies.
To further confirm that PLD2 is responsible for stretch activation of TREK-1 currents in cellular membranes, we truncated the C-terminus of TREK-1 (truncated TREK-1) and compared its pressure activation to wildtype—PLD2 binds to the C-terminus of TREK35 (Supplemental Fig. S3c). As expected, C-terminally truncated TREK-1 was almost completely insensitive to stretch activation by pressure with no significant difference between it and current measured in non-transfected cells (Fig. 2c and raw traces, Supplemental Fig. S3a).
For lipid mixing to act as a relevant mechanosensor for an ion channel, latency (response time) needs to be fast, (i.e. µs to low ms time frame). To determine the latency for PLD2 based lipid mixing we measured the delay time for PLD2 specific stretch activation of TREK-1 with pressure clamp. This is possible since all the observed stretch activated TREK-1 current was PLD2 dependent (Fig. 2c). Our instrument setup has a 6 ms delay (4 ms in the instrument and 2.9 ms in the connectors; see methods). After instrument delay we see initial TREK-1 currents almost immediately and the current reaches statistical significance within 2.1 ms at 60 mmHg (Fig. 2d), demonstrating PLD2 lipid mixing is capable of latencies suitable for fast activation of a mechanosensitive channel.
We also found that purified TREK-1 in liposomes show major regulation by the thickness of the membrane in which TREK-1 resides (Fig. 2e) as well as by anionic lipid42. These results show in a purified system how PLD2 is the primary force transducer for TREK-1 activation in HEK cell membranes (see Fig. 2f). Since domain disruption can lead to a change in thickness and PLD activation by raft disruption is known to occur5, we looked to observe if TREK-1 responded to shear by translocating similar to PLD2.
The TREK-1 force transduction pathway
We previously showed PLD2 responds directly to shear force using a live PLD2 shear assay. 3 dynes/cm2 shear force robustly increased PLD activity measured in real time5. Here we investigate protein translocation in response to mechanical force. We directly imaged lipid rafts, TREK-1, and PLD2 localization using dSTORM in C2C12 cells. PLD2 and TREK are endogenously expressed in C2C12 cells. We applied a force of 3 dynes/cm2 mechanical shear to cells and determined lipid mixing by monitoring fluorescently labeled PLD2 and TREK-1 localization with GM1 and PIP2 and each other. Pair correlation was used to determine localization. Pair correlation is not affected by oversampling, thus making it appropriate for stochastic methods such as dSTORM32,43.
Prior to shear, PLD2 was robustly localized with GM1 domains and moderately with PIP2 domains as shown previously 5 while TREK-1 was most strongly associated with PIP2 domains (Fig. 3a-c). TREK-1 also localized with PLD2 (Fig. 3d) as expected since PLD is known to bind directly to TREK-135. TREK-1 localized minimally with GM1 domains, a result consistent with a proportion of TREK-1 preferring the disordered, bulk-like lipids when not bound to PLD42. Images of each individual channel for the abovementioned localizations are included in Supplemental Fig. 4.
After mechanical shear, PLD2 robustly translocated from GM1 to PIP2 domains (Fig. 3a-b). Shear-induced ejection of PLD2 from GM1 domains was almost complete—more robust than chemical disruption5. PLD2 and TREK-1 colocalization remained roughly constant after shear (Fig. 3d) suggesting the two proteins shift as a complex towards PIP2 domains (unsaturated lipids). We co-labeled TREK-1 and PLD2— consistent with this model, TREK-1 also shifted toward PIP2 domains after shear (Fig. 3c). Localization between TREK-1 and GM1 domains does not change (Supplemental Fig 5) unless cholesterol is present. In a companion study we show cholesterol increases TREK-1 localization to GM1 domains in primary neurons and shear disrupts that interaction (Neyabosadri et al, Figure 4A-B). Together this suggests a model whereby the PLD2-TREK-1 complex translocates from GM1 domains towards PIP2 domains where PLD2 is active, allowing for activation of TREK-135 (Fig. 2, 3e). Since PIP2 directly antagonizes TREK-144,45, the localization of TREK-1 to PIP2 before and after shear (Fig 3c,e) explains the strong inhibition of TREK-1 by xPLD2 (Fig. 2b-c).
Mechanical disruption of lipid rafts in neurons
To establish our mechanical mechanism of domain disruption in a neuronal system, we tested mechanical shear in neuroblastoma 2a (N2A) cells, a generic mammalian neuronal line. Among cells, nerves can be particularly sensitive to mechanical force and play an important role generating pain and touch sensation in animals. PLD and lipid partitioning are found in all types of cells including neurons46.
N2a cells were found to recapitulate both the domain disruption and PLD2 activation in response to mechanical shear as was observed in muscle cells. Application of 3 dynes/cm2 lead to a significant decrease in the relative size of the labeled domains in N2a cells of ∼50 nm in diameter (Fig. 4a-b). This decrease also lead to an increase in the activity of PLD2 (Supplemental Fig. 6a) as would be predicted by our previous findings in muscle cells. These findings show that the mechanism is not constrained to muscle cells and likely is activated in any cells which contain both lipid domains and express PLD2.
Regulation of mechano/pain thresholds in vivo
We then reasoned that if domain regulation of lipids were a common mechanism of mechanosensation than evidence of this mechanism could be observed in diverse animal species. We thus chose D. melanogaster (fruit flies) due to both its possession of a PLD2 ortholog as well as its evolutionary distance from the mammalian cell lines used previously. We found that GM1 domains exist in the brain of flies (Fig. 4c) and that they can be disrupted39. Observation of the activity of the PLD ortholog in neuronally-erived cell lines (BG2-c2) also showed sensitivity to mechanical shear forces (Fig. 4d), consistent with regulation through raft disruption. We wanted to observe the behavior of these animals in order to determine if there was a physiological contribution from PLD to the mechanosensitivity of the whole animal. One behavior which could show this contribution is an arousal-assay which measures the amount of mechanical stimulation needed to excite a fly into movement from rest. If PLD-synthesized PA is also upstream of mechanosensitive potassium channels in flies (as in mammalian cells) we would predict a decrease in the amount of force needed to cause fly arousal.
Using single-animal measurements of arousal threshold47,48 (see Supplemental Fig. 6b), we examined whether PLD could regulate fly mechanosensation. Flies without functional PLD (PLDnull) were subjected to a series of incremental vibrational stimuli every 30 minutes for 24 hours. For each series, the level of stimulation required to arouse the fly, indicated by motion, was recorded using automated machine vision tracking. Averages over the 24 hours were compared to genetically matched controls. PLDnull flies showed a significantly lower arousal threshold than their control strains, indicating increased sensitivity to mechanical force (Fig. 4e). We further used a neuronal-specific driver, Nsyb GAL4, combined with a PLD RNAi line (PLD-KD) and obtained a similar result, indicating that the phenotype is neuronal-specific (Fig. 4f). PLD also regulated responses to noxious pain stimuli (Supplemental Fig. 6c-e).
Discussion
Our data demonstrate, in a biological membrane, lipid domain disruption is sensor that communicates force to the canonical mechanosensitive channel TREK-1 (Fig. 4g). The necessity of functional PLD2 for TREK-1 mechanosensitivity (Fig. 2) in HEK cells suggests the enzyme can dominate as the primary transducer and TREK-1 as a secondary transducer in the membranes of cultured cells. Hence the membrane first elicits an enzyme-mediated chemical signal and then the chemical signal is converted into an electrical signal by the channel. In theory, PLD2 could be activated by mechanical force and raise the global PA levels sufficient to activate TREK-1, however, truncated TREK-1 lacking the PLD localization domain (Fig. 2c) failed to respond and suggests that PLD2 localization is a significant contributor to TREK-1 activation. Since the PLD-TREK-1 complex was shown above to translocate to PIP2 domains and PIP2 is a TREK-1 antagonist44, it is likely that without co-localization the local concentration of PA does not reach sufficient levels to compete with PIP2 binding11. It may be possible if PLD2 dependent PA production could reach activating concentration if the membrane and or cytoskeleton were sufficiently disrupted.
Conservative estimates of the latency of TREK-1 activation by chemical transduction through PLD2 is less than 2.1 ms. In reality, the latency could be much faster, as the 2.1 ms latency is likely an upper limit due to the error in our instrument delay, not PLD2 mixing—TREK-1 current appeared to rise immediately after the instrument delay (Fig. 2e). Regardless, the 2.1 ms upper limit is safely within previously established mechanosensitive current latency49, suggesting that domain disruption is suitable mechanism for most mechano-transduction pathways. This direct measurement in HEK293 cells roughly agree with the 650 µs theoretical latency previously estimated based on spatial separation of GM1 and PIP2 domains in C2C12 cells5.
In artificial membranes TREK-1 responds directly to membrane thinning42 and stretch41,50, i.e. FFL appears to directly open and close the channel17,51. In contrast, in a biological membrane, our electrophysiology experiments, imaging, and enzyme assay clearly show that anionic lipid production by PLD2 is necessary for mechanical TREK activity in HEK cells (Fig. 4g)—no amount of stretch was able to substantially activate the channel absent PLD2 (Fig. 2b-d and Supplemental Fig. S3a).
In a physiological system the two variables, thickness and PA, likely act as coincidence detectors, allowing for careful control over the hyperpolarization effect of TREK-1. It is likely that the translocation of TREK-1 from GM1 domains to the bulk membrane facilitates the thickness change required for activation of the channel since the thickness differential is ∼30% (12 Å), while pressure-mediated thinning can only change the thickness <4% (1.5 Å) before resulting in lysis 15. The hydrophobic mismatch between the protein and lipid after translocation likely provides the curvature speculated to activate the channel by FFL. PA production could also directly alter the curvature, thickness, or additional signaling enzymes near TREK-1, but these putative pathways would not significantly change the finding that PLD2 is a primary transducer for TREK-1 activation in a cellular membrane. The hydrophobic mismatch component of TREK-1 activation is also seen in prokaryotic mechanosensors 52. The effect of translocation and lipid presentation is not known for prokaryotes.
Raft disruption may directly affect many ion channels since palmitoylation alone is sufficient to target proteins to GM1 domains6 and many ion channels are palmitoylated53 and respond to lipid binding or changes in the lipid environment. In theory, mechanical disruption of GM1 domains could cause translocation of any palmitoylated channel in a single step, directly exposing the channel to an activating lipid. For example, the voltage-activated sodium channel (Nav)1.9 is localized to lipid domains and domain disruption (by chemically removing cholesterol) induces channel translocation out of GM1 domains and a corresponding pain response54. In such cases, entering membrane domains containing disordered lipids may be sufficient to activate the channel.
Disruption of palmitate-mediated localization may be a common mechanism for activating transduction pathways; many important signaling molecules are palmitoylated including tyrosine kinases, GTPases, CD4/8, and almost all G-protein alpha subunits55. Translocation of these proteins, or a palmitoylated binding partner, from lipid domains by mechanical force could alter their available substrates and affect downstream signaling. Since many mechanosensitive channels are either regulated or directly gated by lipids, it is likely that other channels are downstream effectors of domain disruption in biological membranes. For example, some transient receptor potential (TRP) channels may be activated by domain disruption since they are regulated by lipid rafts56 and function downstream of anionic lipid signaling57.
While fly orthologs of TREK-1 exist, it is not clear which of these may bind PLD, greatly hampering research into the downstream pathway of PLD in the fly. More work will need to be done to determine the native binding partner of PLD, and thus PA, in D. melanogaster. This said, since leak potassium channels, such as TREK-1, are typically protective against pain i.e. they hyperpolarize the cell58, we expect PA synthesis should decrease sensitivity to pain in flies, which is consistent with the result we observed in the PLD deficient flies.
Lastly, PA’s regulation of D. melanogaster mechanosensation and pain (Fig. 4) adds in vivo support to a growing list of anionic lipids that set force sensing thresholds. For example, PIP2 sets the threshold for mechanical B-cell activation59. Likewise, sphingosine-1-phosphate (S1P), an anionic lipid similar to PA, regulates pain in mice60. PLD’s activation by mechanical force and substrate presentation helps explain how anionic lipids could directly set pain thresholds and mechanosensitivity via canonical mechanosensitive ion channels. A single lipid regulating multiple channels simultaneously is much more likely to affect a whole animal than a single channel.
Author Contributions
Conceptualization, E.N.P. and S.B.H.; Methodology, E.N.P., M.G., K.R.M, and S.B.H; Investigation, E.N.P., M.G., and K.R.M; Resources, W.W.J., E.M.J., and S.B.H.; Writing – Original Draft, E.N.P. and S.B.H.; Writing – Review and Editing, E.N.P., M.G., M.A.P, E.M.J., and S.B.H.; Supervision, W.W.J. and S.B.H.; Funding Acquisition, W.W.J. and S.B.H.
Materials and Methods
Expression and Purification of TREK-1
Expression and purification of TREK-1 was performed as previously explained44. Briefly, Pichia pastoris (SMD1163H) transformed with the TREK-1 gene in pPICZ-B vector was grown in 2.8L baffled flasks. Overnight cultures were grown overnight at 30C, 250rpm to an OD600 of ∼16. Cells are then harvested and resuspended in minimal methanol media and incubation temp is reduced to 25C. Induction was maintained by the addition of 0.5% methanol every 12 hours. Expression was continued for ∼48-60 hours. Cells were pelleted, frozen in N2l, and stored at −80C.
Cells were milled and powder was added to lysis buffer (50 mM Tris pH 8.0, 150 mM KCl, 60 mM dodecyl-β-D-maltoside (DDM), 0.1 mg/mL DNAse 1, 0.1 μg/ml pepstatin, 1 μg/ml leupeptin, 1 μg/ml aprotinin, 0.1 mg/ml soy trypsin inhibitor, 1 mM benzamidine, and 0.1 mg/ml AEBSF, with 1 mM phenylmethylsulfonyl fluoride added immediately before use) at a ratio of 1g pellet/4mL lysis buffer. After stirring for 4 hours membranes were extracted at 35,000 xg for 45 min and supernatant was applied to cobalt resin (Clontech) column by gravity flow. Column was serially washed and eluted in IMAC buffer(50 mM Tris pH 8.0, 150 mM KCl, 4 mM DDM, pH 8) with 30 mM and 300 mM imidazole respectively. Eluted protein was concentrated and applied to Superdex 200 column (GE Healthcare) equilibrated in buffer (20 mM Tris pH 8.0, 150 mM KCl, 1 mM EDTA, 2 mM DDM).
Flux Assay
The flux assay was performed similarly as previously published61. Briefly, 5 μL of sonicated proteoliposomes was added to 195 μL of flux assay buffer (150mM NaCl, 20mM HEPES pH 7.4, 2μM ACMA) in duplicates in a black 96-well plate (Costar 3915). A protocol was set up on a Tecan M200 Pro to initially read the fluorescence (excitation 410, emission 490) every twenty seconds for one minute as a baseline. The temperature inside the plate reader was set at 25°C. Then, using the protonophore CCCP (1μM final concentration), we collapsed the electrical potential allowing protons into the vesicles. Fluorescence was read every twenty seconds for seven minutes. Next, the potassium-selective ionophore valinomycin (20nM final concentration) was added to terminate the chemical gradient, and fluorescence was read every twenty seconds for five to ten minutes. An average of the duplicates was taken, and then the data was normalized similar to that published in Su et al 62. Briefly, at each time point (F-Fstart)/(Fstart – Fend) was calculated, where F is the fluorescent value at that time, Fstart is the average of the first four initial readings before the addition of CCCP, and Fend is the final fluorescent value after 5 minutes of valinomycin. Next, control proteoliposome flux was normalized to 1 throughout CCCP, and the TREK flux was normalized to the control accordingly. In flux assay figures, addition of CCCP occurs at t=80 seconds and ends at t=520 seconds before valinomycin was added.
Cell Culture and Gene Expression
HEK293t cells (ATCC Cat# CRL-3216, RRID:CVCL_0063) were maintained in the solution consisting of the DMEM (Corning cellgro) culture media, 10% FBS, 100 units/mL penicillin, and 100 µg/mL streptomycin. Cells were plated on poly-D-lysine-coated 12 mm microscope cover glass ∼12h, ∼36h, or ∼60 h before transfection in low confluence (5%, 2.5%, or 1.25%). Genes for target proteins were transiently co-transfected to HEK293t cells with X-tremeGENE 9 DNA transfection agent (Roche Diagnostics). Full-length human TREK-1(hTREK-1) with C-terminus GFP tag in pCEH vector was a gift from Dr. Stephen Long. Mouse PLD2 constructs(mPLD2) without GFP tag in pCGN vector were gifts from Dr. Michael Frohman. Both functional (mPLD2) and inactive mutant (mPLD2-K758R, single mutation) form of mPLD2 were used together blindly to test mPLD2 effect on hTREK-1. hTREK-1 was co-transfected with mPLD2 or K758R with the 1(0.5g of hTREK-1):4(2g of PLD2) ratio35. All the salts for internal/external solutions were purchased from either Sigma or Fisher Scientific.
Electrophysiology
The transfected HEK293t cells were used in 24∼36 hr. after transfection. Standard excised inside-out patch-clamp recording procedure for TREK-1 was performed following the lead of others (Brohawn et al., 2014; Honore et al, 2006; Coste et al., 2010). Currents were recorded at room temperature with Axopatch 200B amplifier and Digidata 1440A (Molecular Devices). Borosilicate glass electrode pipettes (B150-86-10, Sutter Instrument) were pulled with the Flaming/Brown micropipette puller (Model P-1000, Sutter instrument) resulting in 3∼6 MΩ resistances with the pipette solution (in mM): 140 NaCl, 5 KCl, 1 CaCl2, 3 MgCl2, 10 TEA-Cl, 10 HEPES, pH 7.4 (adjusted with NaOH). Bath solution consists of (in mM): 140 KCl, 3 MgCl2, 5 EGTA, 10 TEA-Cl, 10 HEPES, pH 7.2 (adjusted with KOH). Low concentration of TEA (10 mM), which has been known to be insensitive to TREK-1 current (Piechotta et al., 2011), was added into both pipette/bath solutions to block the endogenous potassium channels in HEK293 cells (Thomas and Smart, 2005). Patch electrodes were wrapped with parafilm to reduce capacitance. Currents measured using Clampex 10.3(Molecular Devices) were filtered at 1 kHz, sampled at 20 kHz, and stored on a hard disk for later analysis. Pressure clamping on the patch was performed using high speed pressure clamping system (ALA Scientific) through the Clampex control. Data was analyzed off-line by a homemade procedure using IgorPro 6.34A (WaveMetrics).
hTREK-1 current either co-expressed with mPLD2 or K758R was activated by negative pressure steps from 60 to 0 mmHg in 10 mmHg decrements at +30 mV membrane potential, and 5 traces for each case were recorded and averaged for the analysis. Inside-out patch has generally less patch size variability than cell-attached recording when pressure clamped (Suchyna et al., 2009), but in other to further minimize the patch size variability in inside-out patches, patch size was estimated using a method described by Sakman and Neher (Sakman and Neher,1995), and the current density (I_density; pA/µm2) was calculated for the further analysis. Then, a Boltzman equation, I_density = base +{max/[1+exp((P50-P)/slope)]} was used to fit the data with a constraint of base=1 due to poor saturation of the current at high pressure. P is the applied pressure, P50 is the pressure that activates 50% of maximum current, and slope shows the sensitivity of current activation by pressure. In some experiments with hTREK-1+K758R co-expression where the activated currents were too small to fit to the Boltzman equation, the current amplitude at P=-30 mmHg (I_m30) was compared with its 5x standard deviation(I_5xSD). If I_m30 < I_5xSD, the experiment was excluded from the Boltzman equation fitting and corresponding P50-slope analysis. This empirical rule (we call it 5xSD rule) can discern 4 out of 5 wilt type cell-attached recording cases as null experiments suggesting that it could be a usable/useful empirical criterion for our experiment. Then, the current density at −60 mmHg and P50-slope data were used for statistical analysis. Mann Whitney test was done to assess statistical significance using Prism6 (GraphPad software), and outliers were eliminated using a built-in function in Prism with Q = 1 %. The values represented are Mean +/-SEM.
Pressure clamp instrument delay was calculated as follows: the pressure device requires ∼4ms as measured within the device itself. The distance the pressure must travel from the device to the pipette tip is ∼100 cm at a propagation speed of 34,300 cm/s (the speed of sound in air). This results in a delay of ∼2.9 ms for propagation to the patch. The distance and speed of pressure propagation through the pipette tip and solution was considered negligible. The resulting current was compared to a baseline current of 1 ms before pressure clamp began. To determine the first significant time point, a 1ms running average was taken of the post-clamped current and compared to baseline using a one-way ANOVA followed by the Fisher’s LSD test.
Cell Culture
All cells were grown in Dulbecco’s Modified Eagle Medium (DMEM) containing 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin unless otherwise noted. C2C12 cells (ATCC Cat# CRL-1772, RRID:CVCL_0188) were changed to a serum-free media containing no FBS or antibiotics 24 hours prior to experimentation unless otherwise noted. For the in vivo assay, PBS-glucose buffer contained D-glucose (20mM) in PBS (VWR, 45000-446).
Fixed cell preparation
C2C12 cells were grown to 80% confluence and then allowed to differentiate overnight in serum free media. Cells were rinsed, treated as needed, and then fixed with 3% paraformaldehyde and 0.1% glutaraldehyde for 10 min to fix both protein and lipids. Glutaraldehyde was reduced with 0.1% NaBH4 for 7 min followed by three-10 min washes with PBS. Cells were permeabilized for 15 min with 0.2% Triton X-100, blocked with 10% BSA/0.05% Triton/PBS at rt. for 90 min. Primary antibody (PLD2, Cell Signalling #13891; TREK-1, Santa Cruz #sc-50412; PIP2, Echelon Biosciences #z-P045) was added to a solution of 5% BSA/0.05% Triton/PBS for 60 min at rt at a concentration of 1:100 followed by 5 washes with 1%BSA/0.05% Triton/PBS for 15 min each. Secondary antibody (Life Technologies #A21244 and A21235; cy3B antibodies were produced as described previously5) was added in the same buffer as primary for 30 min at rt followed by 5 washes as above. A single 5 min wash with PBS was followed by a post-fix with fixing mixture, as above, for 10 min w/o shaking. This was followed by three-5 min. washes with PBS and two-3 min. washes with dH2O. Cells only receiving CTxB treatment were not permeabilized.
The dual-fixation protocol is used to minimize any effects from post-fixation aberrations. While always good practice for super-resolution in general, this dual-fixation also ensures that the movement of any molecule of interest which may not have been immobilized by the initial fixation can be fully immobilized after labeling since the antibodies or toxins used for labeling will be efficiently cross-linked during this post-labeling fixation step. While some have proposed that this problem should be solved by adding the label before the initial fixation63 we believe that in the absence of easily-attainable monomeric labeling molecules it would have likely led to clustering artifacts due to the (often) multimer nature of the labeling proteins. Shear force was applied to cells in ibidi µ-Slide I0.4 Luer chambers with a flow rate calibrated to apply 3.0 dynes/cm2. Fixation media (see above) was applied to cells using a syringe pump (Harvard Apparatus PHD ULTRA) and kept at 37°C using an in-line heater (Warner SH-27B).
Imaging Protocols
Super-resolution dSTORM imaging
Images were recorded with a Vutara 352 super-resolution microscope (Bruker Nano Surfaces, Salt Lake City, UT) which is based on the 3D Biplane approach38. Super-resolution images were captured using a Hamamatsu ORCA Flash4.0 sCMOS camera and a 60x water objective with numerical aperture 1.2. Data were analyzed by the Vutara SRX software (version 5.21.13). Single molecules were identified by their brightness frame by frame after removing the background. Identified particles were then localized in three dimensions by fitting the raw data in a customizable region of interest (typically 16 X16 pixels) centered on each particle in each plane with a 3D model function that was obtained from recorded bead data sets. Fit results were stored as data lists for further analysis. Fixed samples were imaged using a 647 nm and 561 nm excitation lasers, respectively, and 405 nm activation laser in photoswitching buffer comprising of 20 mM cysteamine (Sigma, #30070), 1% betamercaptoethanol (Sigma, #63689) and oxygen scavengers (glucose oxidase (Sigma #G2133) and catalase (Sigma #C40)) in 50mM Tris (Affymetrix, #22638100)+10 mM NaCl (Sigma, #S7653) +10% glucose (Sigma, #G8270) at pH 8.0 at 50 Hz and maximal powers of 647 nm, 561 nm and 405 lasers set to 8, 10, and 0.05 kW cm-2 respectively. Live cell imaging was performed in DMEM supplemented with oxygen scavengers and 0.1% betamercaptoethanol in 50mM Tris+10mM buffer +2% glucose. An autocorrelative algorithm38 was used to correct for drift correction.
Pair correlation and cluster analysis was performed using the Statistical Analysis package in the Vutara SRX software. Pair correlation analysis is a statistical method used to determine the strength of correlation between two objects by counting the number of points of probe 2 within a certain donut-radius of each point of probe 1. This allows for localization to be determined without overlapping pixels as done in traditional diffraction-limited microscopy. Localization at super resolution is beyond techniques appropriate for diffraction-limited microscopy such as Pearson’s correlation coefficient.
Fluorescence Recovery After Photobleaching (FRAP)
For fixation studies N2a and C2C12 cells were grown in DMEM with 10% FBS until 16 hours before use in which they were switched into serum free DMEM. On the day of the experiment, DMEM in live cells was replaced with DMEM w/o phenol red. Fixed cells were rinsed once with PBS and then put into a mixture of PBS with 3% PFA and 0.1% glutaraldehyde for 20 min at 37C. Fixed cells were then rinsed with PBS 5 × 5 min and then placed back into phenol-free DMEM. CTxB (ThermoFisher C34778, 100 ug/ml) was then applied 1:200 into each plate and allowed to incubate for >30 min before imaging. Imaging and data collection was performed on a Leica SP8 confocal microscope with the Application Suite X v.1.1.0.12420. 5 images were taken as baseline after which a selection of 1 or more ROI were bleached at 100% laser power for 6-8 frames. Recovery was measured out to 5 min and fluorescence of the ROI(s) were quantified. The fluorescence before bleaching was normalized to 1 and after the bleaching step was normalized to 0.
Drosophila Assays
For behavior experiments, 1 to 5-day old flies were collected in vials containing ∼50 flies at least 12 hours before the experiment. Flies were allowed to acclimate to behavior room conditions for >30 min (dim red light, ∼75% humidity) before each assay. Shock avoidance was tested by placing flies in a T-maze where they could choose between an arm shocking at the indicated voltage every 2 seconds and an arm without shock. Flies were given 2 min to choose which arm, after which flies were collected and counted to determine the shock avoidance index for each voltage and genotype. Control and knockout flies were alternated to avoid any preference and the arm used for shock was also alternated to control for any non-shock preference in the T-maze itself.
Arousal threshold protocol has been described in detail previously48. Briefly, animals were exposed hourly to a series of vibrations of increasing intensity ranging from 0.8 to 3.2 g, in steps of 0.6 g. Stimuli trains were composed of 200 ms vibration with 800 ms inter-vibration interval and 15 s inter-stimuli train interval. Stimulation intensity and timing were controlled using pulse-width modulation via an Arduino UNO and shaft-less vibrating motors (Precision Microdrives, model 312–110). Arousal to a given stimulus was assigned when an animal (1) was inactive at the time of the stimulus, (2) satisfied a given inactivity criteria at the time of the stimulus, and (3) moved within the inter-stimuli train period (15 s) of that stimulus.
Statistics
All statistical calculations were performed using a Student’s t-test unless otherwise noted. Significance is noted as follows: ns: p>0.05; *:p<0.05; **:p<0.01; ***:p<0.001.
Supplemental Information
Supplemental Discussion
Curvature and hydrophobic mismatch are popular mechanisms to describe direct opening of ion channels. Unlike a channel, PLD2 is a soluble protein that associates with the membrane through palmitoylation. Hence, the PLD2 specific TREK-1 mechanosensitivity is unlikely through changes in bilayer thickness or membrane curvature. Rather bilayer thickness and membrane curvature are likely mechanisms reserved for regulation of TREK-1.
Also, great care was taken to ensure that conclusions presented here from the imaging were based not on single conditions, but on relative changes between two states. While there are likely conditions that cannot be fully accounted for, since all samples are treated identically (save for the treatment) any artifacts resulting from the protocol used will be largely negated, allowing for an accurate conclusion based on the comparison between the two states observed. Cross-correlation analysis already compensates for many of these variables such as changes in labeling density and overcounting68. These corrections apply both to the labeling protocol, as well as the size and correlation analysis which, while not claiming to be absolute, exhibit changes relative to one another. This is the data upon which our conclusions are based. It is worth mentioning that due to the closeness of the resolution ability of dSTORM with the reported sizes of the observed lipid domains that the estimates of relative size change are likely underestimated, increasing the significance for our conclusions on both size and protein association to these domains.
Acknowledgements
We thank Tamara Boto and Seth Tomchik for their assistance in the Drosophila shock experiments, Michael Frohman from Stony Brook for the mouse PLD and mutant PLD cDNA, Steven Long from Memorial Sloan Kettering for human TREK-1-GFP, Padinjat Raghu for the PLD mutant Drosophila, Andrew S. Hansen for PLD experiments, multiple aspects of experimental design and discussion, Yul Young Park for the electrophysiology experimentation, and Carl Ebeling for his help and discussion on the imaging analysis. This work was supported by a Director’s New Innovator Award to S.B.H. (1DP2NS087943-01) from the National Institutes of Health, an R01 to W.W.J. (R01AG045036) from the National Institute on Aging, and a graduate fellowship from the Joseph B. Scheller & Rita P. Scheller Charitable Foundation to E.N.P. The authors declare no conflict of interest.