Bilayer-Mediated Structural Transitions Control Mechanosensitivity of the TREK-2 K2P Channel

Abstract

The mechanosensitive two-pore domain (K2P) K⁺ channels (TREK-1, TREK-2, and TRAAK) are important for mechanical and thermal nociception. However, the mechanisms underlying their gating by membrane stretch remain controversial. Here we use molecular dynamics simulations to examine their behavior in a lipid bilayer. We show that TREK-2 moves from the "down" to "up" conformation in direct response to membrane stretch, and examine the role of the transmembrane pressure profile in this process. Furthermore, we show how state-dependent interactions with lipids affect the movement of TREK-2, and how stretch influences both the inner pore and selectivity filter. Finally, we present functional studies that demonstrate why direct pore block by lipid tails does not represent the principal mechanism of mechanogating. Overall, this study provides a dynamic structural insight into K2P channel mechanosensitivity and illustrates how the structure of a eukaryotic mechanosensitive ion channel responds to changes in forces within the bilayer.

Keywords: K(+) channel gating; K2P channel; KCNK10; KCNK2; KCNK4; Mechanosensitive; TREK-2.

Graphical abstract

Figure 1

Bilayer Stretch Induces Movement of TREK-2 from the Down State toward the Up State to Produce a Change in Cross-Sectional Area (A) Intrinsic mechanosensitivity of the purified core TM domains of TREK-2 reconstituted into a planar lipid bilayer. Example of currents recorded at +80 mV in response to a pressure jump of −80 mbar. (B) Top: a model for K2P channel mechanosensitivity based on comparison of the down and up state crystal structures of TREK-2. Bottom: to test this, we used MD simulations to examine whether membrane stretch can induce a down to up conformational transition. Simulation of membrane stretch (i.e., increased tension) involves an increase in the xy plane of the bilayer (red arrows) to increase the area/lipid (see also Figure S1A). (C) Top: comparison of Cα RMSD for the TM helices (M1-M4) of TREK-2 against the down state crystal structure during stretch and unstretched simulations. The structure moves away from the down state when stretched (red), but remains close to the down state when unstretched (black). Bottom: similar RMSD comparison against the up state showing rapid stretch-induced movement toward an up-state-like conformation. (D) Increase in cross-sectional area upon membrane stretch. The dotted box indicates the region of greatest structural change. The cross-sectional area of stretched structure is very similar to the up state crystal structure (PDB: 4BW5; pink line). Unstretched simulations remain close to the down state structure (PDB: 4XDJ; shaded gray area). The z axis is centered on the middle of the bilayer.

Figure 2

A Role for Changes in the Transmembrane Pressure Profile (A) Lateral pressure profiles calculated for the bilayers used in this study. Black: unstretched POPC bilayer. Red: POPC bilayer stretched at −50 bar. Note the reduction in positive pressures within the core of the bilayer. The stretched bilayer has the same thickness as an unstretched DLPC bilayer (∼31 Å), but the DLPC profile (green) retains positive pressures within the core. (B) RMSD comparison with the up state crystal structure. Hydrophobic mismatch in the thinner DLPC bilayer (green) is unable to drive movement toward the up state even after 500 ns, instead the structure begins to diverge from the up state (see also Figure S2B). By contrast, low levels of stretch at −20 bar for a similar period produce movement toward the upstate (blue). The changes produced by stretch at −50 bar and no stretch over 200 ns are shown for comparison.

Figure 3

Sequential Rearrangement of the TM Helices during Membrane Stretch (A) Movement of TMs measured as a change in three distances: (1) “Fenestration” between G324 and P198 of the adjacent subunit (blue), (2) “Zipper” between W326 and R237 (green), and (3) “Expansion” between M322 and G212 (red). These changes are shown at the start, during, and end of 100 ns stretch simulation. Note unzipping of interactions between M4 and M3 requires reorientation of W326 (green) on M4; also F244 (yellow) then rotates to fill the space created by the M2-M4 expansion. (B) Change in these distances in separate chains of TREK-2 during a stretch simulation. Note the sequence: an initial contraction of the fenestration followed by unzipping, expansion, and then full closure of the fenestration. (C) Correlation plot comparing the change in fenestration and zipper distances during an unstretched (gray) and stretched (red) simulation sampled every 1 ns. Distances in the TREK-2 crystal structures shown as black diamonds. Within the unstretched bilayer TREK-2 samples conformations close to the down state crystal structure (PDB: 4XDJ) and does not approach the up state conformation (PDB: 4BW5). When stretched (red), the structure samples conformations similar to the up state. In some cases incomplete closure of the fenestration occurs (orange) due to obstruction by lipid tails (see also Figures 6 and S5). The right-hand panel shows a similar comparison of fenestration and expansion distances for both stretch and unstretched simulations.

Figure 4

Specificity of the Stretch Simulations (A) Overlaying TWIK-1 structures simulated for 200 ns in the presence or absence of membrane stretch (−50 bar) reveals no change in structure. The starting structure is shown in gray. (B) No change in cross-sectional area in response to membrane stretch. The shaded gray area represents starting structure. Inset: comparison of Cα RMSD for M1-M4 helices against the TWIK-1 crystal structure (3UKM) with or without stretch. Overall, there is no change in response to stretch, even at −50 bar. (C) Overlay of different up and down state crystal structures available for TWIK-1, TREK-1, TREK-2, and TRAAK demonstrating the wide variety of conformational states. Colors are as shown in (D). (D). Correlation plots of fenestration versus expansion distances for TREK-2 simulations with or without stretch. Equivalent distances in the known crystal structures are plotted as colored diamonds. In the absence of stretch (gray dots) the structure samples conformational states similar to the down state crystal structures of both TREK and TRAAK. But when stretched (red dots), the structure quickly samples conformations similar to the many up state crystal structures. Dots represent sampling every 1 ns during the simulation. (E) Structure of TREK-2 in the down state showing location of norfluoxetine (NFx; cyan) within its binding site in the upper fenestration. G324 on M4 and P198 on M2 highlighted blue. (F) RMSD comparison against the up state crystal structure for unstretched (purple) and stretched simulations (cyan), both had NFx located within its binding sites at the start of the simulation. The control stretch simulation is identical but with the NFx deleted (Apo, red). The presence of NFx within the fenestration binding site prevents rapid movement toward the up state conformation upon stretch.

Figure 5

State-Dependent Binding of Lipids Influences Movement of the TM Helices (A) Left: without stretch, lipids interact with the groove between M4 and M2. Examples of interacting lipids shown as yellow spheres. Right: when stretched, the fenestration closes (P198 and G324 in blue). However, in rare cases the head group can move toward the fenestration (orange arrow) allowing lipid tails to clog the fenestration and prevent closure. Middle: Fenestration distances (P198-G324) during a stretched simulation are plotted against the minimum distance of lipid tails from P198 within the fenestration. During stretch (red), lipids initially occupy this site (“Bound,” i.e., come within 4 Å of P198), but upon closure of the fenestration this site is no longer available. Without stretch (gray) the fenestration remains open and lipids are bound within this groove. In some stretch simulations (orange dots) full closure does not occur due to “lipid clog.” Deletion of the obstructing lipid (purple) allows rapid closure of the fenestration. (B) Left: a second stretch-dependent lipid binding site between the tips of M2/M3. Without stretch, a hydrophobic groove is present between F226 on M2 and L243 on M3 (yellow). A snapshot is shown of lipid tails (green) within this groove during a non-stretch simulation. The lipid moves away (gray arrow) upon stretch. Right: after stretch, lipid occupancy decreases and the tips rotate allowing the groove to close. Middle: plot of the tip groove size (i.e., distance between F226 and L243) against the minimum distance of a lipid tail from S240 (purple) within this groove. Without stretch, lipids bind freely with occasional exchange (gray dots), but when stretched lipid tails move away from this site (gray arrow), the tips rearrange and the groove closes.

Figure 6

Effects of Membrane Stretch on the Selectivity Filter and the Inner Pore Cavity (A) Snapshot of the filter and inner cavity during an unstretched simulation. Locations visited by K⁺ ions within the S0-S4 binding sites are shown as small purple spheres. Also shown are water molecules (cyan) and an example of a POPC lipid tail penetrating the lower part of the fenestration. Note this type of lipid penetration does not occur in all unstretched simulations (see also Figure S5). P198 on M2 is shown in blue. (B) Effect of membrane stretch on ion occupancy within the filter. Left: occupancy along the pore axis of ions (purple) and water in the filter for an unstretched simulation. Transitions occur between the S0 and S1 sites in the unstretched simulations (see Figure S5). Right: overall probability of ions in the filter ± stretch. The non-stretched simulations (black line) show increased K⁺ occupancy at S0 and reduced occupancy at S1 suggesting how stretch may directly influence the filter gate. (C) Left: rare example where lipid penetration, via route shown in (A), produces transient dewetting. Presence of water (cyan) and lipid tails (yellow) along the axis of the pore is shown against time. Dewetting correlates with the entry of lipid tails, but does not occur in all unstretched simulations. Right: stretch results in an increase in water density deep within the cavity (red line), but overall the inner pore remains hydrated during the unstretched simulations (see also Figure S5).

Figure 7

Direct Access of Permeant Ions to the Selectivity Filter in the Absence of Membrane Stretch Example of a giant excised patch recording for mechanosensitive TREK channels expressed in Xenopus oocytes. Currents show alternating activation by membrane stretch (−15 mmHg applied to the patch pipette; red bar) or replacement of intracellular K⁺ (gray bar) with Rb⁺ (blue bar). Rb⁺ as the permeant ion produces direct activation of a voltage-dependent gate within the filter, indicating free access from the cytoplasmic side. Membrane stretch (red bar) also activates channels to a similar extent. When applied together with Rb⁺ the effects are synergistic. If lipids directly prevented K⁺ permeation (Brohawn et al., 2014a), then Rb⁺ would not be able to access the filter in the absence of membrane stretch. Instead, Rb⁺ activates the filter gate both before and after membrane stretch, indicating free access to the filter. Inhibition of channel activity by 100 μM tetrapentylammonium (TPA) is shown in green as a control. This result clearly demonstrates that direct lipid occlusion of the inner pore does not represent the primary mechanism of mechanogating.