Structure of the human K_2P13.1 channel reveals a hydrophilic pore restriction and lipid cofactor site

Abstract

Polyunsaturated fatty acid (PUFA) lipids modulate the neuronal and microglial leak potassium channel K_2P13.1 (THIK1) and other voltage-gated ion channel (VGIC) superfamily members through poorly understood mechanisms. Here we present cryo-electron microscopy structures of human THIK1 and mutants, revealing a unique two-chamber aqueous inner cavity obstructed by a hydrophilic barrier important for gating, the flow restrictor, and a P1–M4 intersubunit interface lipid at a site, the PUFA site, corresponding to the K_2P small-molecule modulator pocket. This overlap, together with functional studies, indicates that PUFA site lipids are THIK1 cofactors. Comparison with a PUFA-responsive VGIC, K_v7.1, reveals a shared modulatory role for the pore domain intersubunit interface, providing a framework for understanding PUFA action on the VGIC superfamily. Our findings reveal the distinct THIK1 architecture, highlight the importance of the P1–M4 interface for K_2P control by natural and synthetic ligands and should aid in the development of THIK subfamily modulators for neuroinflammation and autism.

K_2P channels comprise six subfamilies that conduct leak potassium currents essential for membrane potential stabilization and cellular excitability regulation^1,2 and influence diverse normal and pathophysiological processes^2,3, including action potential propagation^4,5, pain^6-8, sleep⁹, intraocular pressure¹⁰, retinal visual processing¹¹, migraine¹², depression¹³, pulmonary hypertension¹⁴, sleep apnea¹⁵ and inflammasome activation in macrophages^16-18 and microglia^18-20. Unlike other potassium channel classes, the selectivity filter (SF) ‘C-type’ gate forms the principal site of K_2P modulation^21-27. Structural studies have shown that the TREK^24,28,29, TWIK^30,31, TALK³² and TASK³³ subfamilies share a common architecture²⁷. However, the THIK subfamily, comprising K_2P13.1 (THIK1) and K_2P12.1 (THIK2), has remained structurally uncharacterized. This subfamily is thought to have a role in anesthesia³⁴, apoptosis³⁵ and ischemic responses³⁶, is inhibited by halothane^37-39, lidocaine³⁸, bupivacaine³⁸, quinidine³⁸ and the phosphodiesterase inhibitor 3-isobutyl-1-methyl-xanthine (IBMX)⁴⁰ and is activated by the signaling lipid arachidonic acid (AA)^38,39. Recently, THIK1 has emerged with a key role in microglial function affecting ramification and surveillance¹⁹, synapse pruning⁴¹ and NLRP3 inflammasome-mediated release of interleukin-1β (refs. 18,19,42), making it a target for the development of agents against neuroinflammation⁴³.

To understand THIK function, we determined the structure of human THIK1 and mutants in lipid nanodisc (ND) and detergent environments using cryo-electron microscopy (cryo-EM). The structures reveal a set of features unique to the THIK subfamily. A hydrophilic residue, Y273, the flow restrictor, forms a barrier that divides the central aqueous cavity into two parts: the cytoplasmic vestibule and a small aqueous chamber below the SF, termed the ‘pond’. Structures bearing the S136P (ref. 37) activatory substitution show that this change opens the flow restrictor by exploiting the natural flexibility of an M2 transmembrane helical break present in all K_2P channels. The structures also reveal a polyunsaturated fatty acid (PUFA) bound in the P1–M4 helix intersubunit interface, the PUFA site, coordinated by an unusual, stacked arginine pair, the arginine sandwich; in which mutations sensitize the channel to lipid modulation. Strikingly, the THIK PUFA site overlaps with the TREK subfamily K_2P modulator pocket where synthetic small-molecule activators bind^24,25,27,44. This structural and functional commonality establishes the role of the P1–M4 interface in natural and synthetic ligand control of K_2P function. Natural ligands such as fish and plant oil ω3 and ω6 PUFAs affect many voltage-gated ion channel (VGIC) superfamily members at binding sites that are thought to include those located in the pore domain (PD)⁴⁵. Comparison of the THIK1 PUFA site to the proposed PD PUFA site on the cardiac I_Ks channel K_v7.1 (KCNQ1)^46-49 uncovers structural commonalities that define a framework for understanding PUFA action in diverse VGIC superfamily members and highlight the potential of this shared PD intersubunit interface as a site for modulator development.

Results

Human THIK1 structure reveals key features

We identified a human THIK1 construct spanning residues 1–350 bearing N59Q and N65Q substitutions to remove glycosylation, denoted THIK1_EM, that maintained native-like function (Extended Data Fig. 1a-c) and that could be readily purified using heterologous expression in HEK293 cells. Incorporation of THIK1_EM into brain polar lipid NDs (Extended Data Fig. 1d,e) for single-particle cryo-EM (Extended Data Fig. 1f,j) yielded a high-quality density map with an overall resolution of 2.65 Å (Extended Data Fig. 2a,b and Table 1) and local resolution as good as 2.4 Å (Extended Data Fig. 2c) defining the THIK1_EM structure (Fig. 1a and Extended Data Fig. 2a-d). THIK1_EM shows the canonical K_2P structure (Fig. 1a and Extended Data Fig. 2a)²⁷, comprising a homodimer in which the M1 and M3 transmembrane helices from PD1 and PD2 face the bilayer, while M2 and M4 transmembrane helices from PD1 and PD2 line the central ion-conducting pore. As in other K_2P channels^24,28-33, the M1 helices are domain-swapped and connect to an extracellular domain, the cap, that sits above the extracellular mouth of the SF²⁷. Consistent with the determination of the structure under high-potassium conditions (200 mM KCl)^25,27, potassium ions occupy the S1–S4 SF ion-binding sites (Fig. 1a and Extended Data Fig. 3a). Similar to exemplar structures of members of the TREK^{5,24,25,28,29,50-56}, TWIK^30,31 and TALK³² K_2P subfamilies, the intracellular face of the channel is open to the cytoplasm and lacks an intracellular gate (Fig. 1a,b and Extended Data Fig. 3b,c).

Table 1 ∣.

Cryo-EM data collection, refinement and validation statistics

	THIK1EM ND (EMD-44870), (PDB 9BSN)	THIK1EM detergent * (EMD-45034), (PDB 9BYI)	THIK1EM-S136P ND (EMD-45077), (PDB 9C09)	THIK1EM-S136P detergent (EMD-45075), (PDB 9C07)	THIK1EM-S136P;Y273A ND (EMD-44978), (PDB 9BWS)
Data collection and processing
Magnification	105,000	105,000	105,000	105,000	105,000
Voltage (kV)	300	300	300	300	300
Electron exposure (e− per Å2)	50	50	50	50	46
Defocus range (μm)	−0.8 to −2.2	−0.8 to −2.2	−0.8 to −2.2	−0.8 to −2.2	−0.8 to −2.2
Pixel size (Å)	0.864	0.864	0.864	0.864	0.846
Symmetry imposed	C 2	C 2	C 2	C 2	C 2
Initial particle images (no.)	11,542,687	8,081,402	15,443,947	4,226,065	12,915,407
Final particle images (no.)	287,654	375,069	446,461	508,529	493,313
Map resolution (Å)	2.65	2.95	2.36	2.73	2.39
FSC threshold	0.143	0.143	0.143	0.143	0.143
Map resolution range (Å)	2.4–3.9	2.95–4.6	2.1–3.9	2.0–3.8	2.3–4.2
Refinement
Initial model used (PDB code)	9C07	9C07	9C07	6RV2	9C07
Model resolution (Å)	2.7	3.2	2.9	2.8	2.8
FSC threshold	0.5	0.5	0.5	0.5	0.5
Model resolution range (Å)
Map sharpening B factor (Å2)	−92.25	−113.15	−53.74	−85.51	−93.56
Model composition
Nonhydrogen atoms	4,346	4,505	4,300	4,512	4,286
Protein residues	514	524	514	524	514
Ligands	26	31	22	36	22
B factors (Å2)
Protein	49.00	80.71	41.29	43.05	41.28
Ligand	20.34	20.47	20.60	20.31	20.60
R.m.s.d.
Bond lengths (Å)	0.004	0.006	0.004	0.006	0.005
Bond angles (°)	0.620	0.567	0.634	0.687	0.663
Validation
MolProbity score	1.29	1.47	1.45	1.62	1.45
Clashscore	2.23	2.85	3.43	6.71	3.08
Poor rotamers (%)	0.92	0.00	0.69	0.23	0.93
Ramachandran plot
Favored (%)	95.85	94.19	95.45	95.93	95.26
Allowed (%)	4.15	5.81	4.55	4.07	4.74
Disallowed (%)	0.00	0.00	0.00	0.00	0.00

^*All structures were dephosphorylated forms, except THIK1_EM detergent. FSC, Fourier shell correlation.

Fig. 1 ∣. THIK1 structural features.

a, THIK1 cartoon diagram (salmon and deep teal) showing side (left and center) and intracellular (right) views. Flow restrictor Y273 and arginine sandwich R92 and R258 residues are shown as sticks. PUFA site EIC is shown in space-filling representation. Gray bars indicate the membrane. Potassium ions are shown as purple spheres. b, Slice through the THIK1 structure showing the locations of the intracellular vestibule, flow restrictor, pond and SF. c, THIK1 extracellular view. M1, M2, M3, M4, P1 and P2 helices are labeled. SF residues F113 and F240 are shown as sticks. Other elements are displayed as in a. The cap is not shown. d, Details of the THIK1 PUFA-binding site. M1 (teal), M2 (magenta), P1 (salmon) and M4 (violet) helices are shown as cylinders. PUFA site residues are shown as sticks and are labeled. Flow restrictor Y273 is in italics. EIC is shown in stick and transparent space-filling representation. e, Surface view of d. Select residues are labeled. EIC C8, C14 and C18 carbons are indicated. Colors are as in d. f, THIK1 PUFA-binding site hydrophilic interactions. The hydrogen-bond network (black) surrounding the EIC acidic headgroup is shown. The arginine stack distance (gray) is indicated. g,h, Structural superposition of THIK1 (salmon and deep teal) and the TREK1:ML335 complex (PDB 6CQ8) (yellow and light blue)²⁴. g, Extracellular view showing position of the TREK1 K_2P modulator pocket (orange oval). ML335 is shown as black sticks. EIC is shown in sticks and transparent space-filling representation. SF phenylalanine side chains are shown as sticks. M1, M2, M3, M4, P1 and P2 helices are labeled. h, Side view showing overlap regions of K_2P modulator pocket and PUFA site (orange). ML335 upper (pink) and lower (blue) rings are indicated. EIC C8 and C14 carbons are indicated. TREK1 residues are in italics. i, Close-up view of h.

The THIK1 overall topology conforms with the general K_2P architecture but displays prominent variations from each of the other subfamilies (Extended Data Fig. 4). THIK1 M4 is near the ‘up’ position seen in some TREK structures²⁴ (Extended Data Fig. 4a,b) but THIK1 M2 and M3 are closer to the central axis, making for a narrower channel profile when viewed from the cytoplasmic side (Extended Data Fig. 4a,b). Similarly, THIK1 has a narrower cytoplasmic profile than the TALK representative K_2P5.1 (TASK2)³² because of differences in the positions and helical pitches of M1, M3 and M4 (Extended Data Fig. 4c). By contrast, the pitch and positions of all four THIK1 transmembrane helices also diverge from those in K_2P1.1 (TWIK1)³⁰ (Extended Data Fig. 4d), making for a wider inner pore in THIK1. THIK1 lacks the cytoplasmic X-gate seen in K_2P3.1 (TASK1)³³ and also diverges in the positions of its M3 and M4 (Extended Data Fig. 4e). Hence, the THIK1_EM structure highlights the structural diversity within the general K_2P architectural framework.

In addition to these overall K_2P architecture variations, the THIK1_EM structure reveals two unexpected features conserved throughout THIK1 homologs from vertebrates spanning from amphibians to humans (Supplementary Fig. 1). First, the THIK1 inner pore cavity is divided into two chambers by a barrier formed by the I139 and Y273 side chains in which the Y273 hydroxyl groups make a hydrogen bond across the narrowest part (2.5 Å) (Fig. 1a,b and Extended Data Figs. 3a and 5a,b). These tyrosines form the deepest part of this barrier with respect to the cytoplasmic vestibule (Extended Data Figs. 5a,b and 6a,b) and are unique to the THIK subfamily (Extended Data Fig. 5c). We denote Y273 as the ‘flow restrictor’ as, unlike classical hydrophobic gates in many VGIC superfamily members⁵⁷, the Y273 pair forms a barrier comprising hydroxyl moieties that have the potential to coordinate passing ions. The flow restrictor is much closer to the SF than the inner gates in other channels (~7 Å versus ~20 Å from the S4 ion to the THIK1 flow restrictor and TASK1 cytoplasmic X-gate³³, respectively) and divides the central aqueous pore into a small chamber between the flow restrictor and SF (denoted ‘the pond’) and large intracellular vestibule bounded by an asparagine ring (N143 and N277) found in both THIK subfamily members (Extended Data Fig. 5a-d). The pond has a volume of ~20 water molecules, has maximal dimensions of ~4.5 Å along the channel central axis by 20 Å across and is electrostatically negative in contrast to the electrostatically positive cytoplasmic vestibule (Extended Data Fig. 6). This two-chambered central cavity sets the THIK subfamily apart from other K_2P channels.

The second THIK1 unexpected feature is a well-defined banana-shaped density in the interdomain interface formed by the P1 and M4 helices of each monomer (Extended Data Fig. 3b,c). On the basis of its shape, we modeled this density as linoleic acid (EIC) (Extended Data Figs. 2b and 3c), a PUFA from a class of lipids that affects many K_2P channels^{26,27,38,39,58-60}; thus, we denoted this site as the ‘PUFA site’ (Fig. 1a). Because of steric considerations, the PUFA site lipid requires a largely straight chain between the carboxylate and C9 position and an unsaturated bond at C12 to accommodate the bend in the density. Accordingly, the density could be equally well fit by α-linolenic acid (α-LNL), which has an additional unsaturated bond at C15. Determination of the structure of THIK1_EM in detergent micelles at 2.95-Å resolution (Supplementary Figs. 2 and 3 and Table 1) revealed a structure that is not different from the THIK1_EM ND structure (root-mean-square deviation (r.m.s.d.) = 0.248 Å) except for having one more helical turn at the end of M2 (R163–R167) (Extended Data Fig. 7a-c) and a weaker density in the SF for the S2 ion (Extended Data Fig. 7d). Importantly, the same banana-shaped density was present in the PUFA site (Supplementary Fig. 3a), indicating that this lipid originated from the host cell and not from the lipids used for ND reconstitution. Both structures also show numerous common positions of hydrophobic chains in the outer membrane leaflet lipid belt that surrounds the channel (Extended Data Figs. 2a,c and 7e-f and Supplementary Fig. 3b) and inner leaflet site between the M1 and M2 helices (Extended Data Fig. 7e), suggesting that the outer surfaces of the channel strongly associate with lipids.

The PUFA is deeply embedded within the protein structure and contacts 18 amino acids from the M1, P1, M2 and M4 helices (Fig. 1c-e and Extended Data Fig. 5c-g), of which the majority are hydrophobic. Residues from the SF2–M4 loop (S246), M4 helix (N261) and loop connecting the cap and P1 helix (R92) coordinate the PUFA headgroup through a set of hydrogen bonds (S246 and N261) and a salt bridge (R92) (Fig. 1f and Extended Data Fig. 5e). R92 is part of an unusual structure in which its side-chain guanidinium group forms a 4.0-Å face-to-face stack (denoted ‘the arginine sandwich’) with the guanidinium of R258 from M4 (Fig. 1a,d,f). Such arginine–arginine stacks have been observed in a wide variety of proteins, where they are thought to affect molecular recognition, as well as protein–protein and protein–ligand interactions^61,62. This arginine sandwich pair occurs only in the THIK subfamily (Extended Data Fig. 5c,f), making it, together with the flow restrictor tyrosine, signature features of this K_2P subfamily.

The P1–M4 interface is central to function of the K_2P SF gate^{21,24,25,27,52,63} and houses the K_2P modulator pocket, a small-molecule-binding site in the TREK subfamily where the synthetic activators ML335 and ML402 bind^24,25,27,44. Superposition of THIK1 and the TREK1:ML335 complex²⁴ (Fig. 1g,h) shows that the THIK1 PUFA site and K_2P modulator pocket coincide. The positions of the PUFA carboxylate and C2–C7 carbons partially overlap with the portion of the K_2P modulator pocket occupied by the ML335 activator inter-ring linker and lower ring (Fig. 1h,i). A subset of these PUFA elements interact with the THIK1 P1 helix residue Y102, a residue corresponding to TREK1F134 that provides a buttress position for K_2P modulator pocket small-molecule activator binding^24,44 (Fig. 1h,i). R92 from the THIK1 arginine sandwich occupies the upper part of the K_2P modulator pocket near the position corresponding to the ML335 upper ring (Fig. 1h,i). The other arginine sandwich member, THIK1 R258, is stacked on top of R92 and corresponds to TREK1 K271 that makes cation–π interactions with the modulator upper ring^24,44. Hence, even though P1–M4 interface ligands from THIK1 and TREK1 have different physicochemical properties, the two binding sites bear similarities that support the idea that the THIK subfamily PUFA site is a variation of the TREK subfamily K_2P modulator pocket.

One key difference between the PUFA and K_2P modulator sites is the interactions made by the lower part of the PUFA, for which the TREK modulators have no corresponding element. The THIK1 PUFA C8–C14 methylene groups snake along the P1–M4 interface in a manner that allows the C15–C18 carbons to emerge through a gap between the THIK1 P1, M2 and M4 helices below residues F101, G105 and I265 at the middle level of the lipid bilayer (Fig. 1d,e). TREK1 lacks this gap because of the presence of a large amino acid, F170, at the THIK1 T138 position that forms the lower bound of the opening (Fig. 1d,e and Extended Data Fig. 5d,e). Together, these observations reveal how this interface pocket is tuned to accommodate diverse ligands in different K_2P subfamilies and show that both natural and synthetic ligands can target this site.

Activatory mutant opens the THIK1 flow restrictor

All K_2P channels except for the THIK subfamily have a proline at a position where a break occurs in the M2 helical structure^30,37,54 (Extended Data Fig. 5d). Placing a proline at this position in THIK1 and THIK2 (S136P and A155P, respectively) increased THIK channel currents in Xenopus oocytes and mammalian cells³⁷. To understand how this substitution acts, we introduced the S136P substitution in the THIK1_EM background, validated that this change increased channel function as expected³⁷ (Supplementary Figs. 4a-c) and used cryo-EM to determine its structure in lipid ND (2.36 Å) (Supplementary Figs. 4d-j and 5 and Table 1) and detergent (2.73 Å) (Supplementary Figs. 6 and 7 and Table 1) environments. Both are similar (all-atom r.m.s.d. = 0.486 Å) apart from an extra five-residue extension of the M2 helix (residues 163–167) (Extended Data Fig. 8a-c) and weaker density for the SF S2 ion (Extended Data Fig. 8d,e) in the detergent environment, recapitulating the THIK1_EM ND and detergent structural differences (Extended Data Figs. 3a and 7a,b,d). Overall, THIK1_EM S136P is also very similar to THIK1_EM (r.m.s.d._Cα = 0.352 Å) (Fig. 2a). There are ions at SF sites S1–S4 and density matching EIC coordinated by the arginine sandwich in both PUFA-binding sites (Fig. 2a). However, inspection of the flow restrictor revealed that Y273 tips away from the channel central axis, increasing the phenolic oxygen interatom distance from 2.5 to 5.7 Å (Fig. 2b,c). This local conformational change opens a gap providing an aqueous pathway between the pond and inner vestibule (Fig. 2d and Supplementary Videos 1 and 2) and suggests that Y273 movement is central to the mechanism by which the S136P substitution increases channel function.

Fig. 2 ∣. The S136P substitution widens the THIK1 flow restrictor.

a, Ribbon diagram superposition of the structures of THIK1 (salmon) and THIK1-S136P (cyan). Flow restrictor Y273 and arginine sandwich R92 and R258 residues are shown as sticks. PUFA site EIC is shown in space-filling representation. Inset, close-up view of flow restrictor. Right, cytoplasmic view. Potassium ions are shown as purple spheres. b, Cryo-EM density (8σ) for flow restrictor regions of THIK1 (salmon) and THIK1-S136P (cyan). c, Cytoplasmic view of flow restrictor conformational changes. d, Slice through the THIK1-S136P structure showing the locations of the intracellular vestibule, flow restrictor, pond and SF. The THIK1 Y273 conformation (salmon) is shown in stick representation. e, Structural comparison of S136P substitution site. Inset, local conformation and hydrogen-bond distances for THIK1 (salmon) and THIK1-S136P (cyan). Common distances are in black. f, Zoomed-in view of S136P site, rotated ~90° relative to the view in the inset of a.

Remarkably, despite the side-chain conformational changes at Y273, the backbones and Cβ atoms of S136 and S136P from the THIK1_EM and THIK1_EM-S136P structures overlay extremely well (Fig. 2e,f). There is a small expansion of the M2 kink because of changes in the L132–G134 backbone carbonyl positions (~0.5 Å) (Fig. 2e and Supplementary Videos 1 and 2). Superposition of THIK1 with TREK, TWIK, TASK and TALK subfamily representatives, where the corresponding S136 site is naturally a proline (Extended Data Fig. 5d), shows that the SF1, SF1–M2 linker and upper part of S2 before the kink align. By contrast, the lower part of M2 following the S136 site adopts a range of positions that span a ~15° arc (Extended Data Fig. 5h), indicating that this section of M2 is mobile. THIK1_EM and THIK1_EM-S136P occupy the middle of this distribution. The THIK1 S136 site occurs at a break at the N terminus of a helical segment in all K_2P structures. Such sites accommodate prolines⁶⁴. As a result, subtle changes in M2 caused by the S136P substitution appear to exploit the flexibility at this junction and allow M2 to flex in a way that enables the flow restrictor to open (Supplementary Videos 1 and 2) and enhance function³⁷.

Flow restrictor shapes the permeant ion energetic landscape

The THIK1 ion conduction pathway has a very different shape than the archetypal K_2P channel TREK1 (Fig. 3a-c). Whereas the pore cavity below the TREK1 SF is wide (Fig. 3c), the THIK1 flow restrictor and asparagine ring respectively block and narrow the central pathway below the SF (Fig. 3a). Although this general cavity shape is maintained in the S136P mutant where the flow restrictor Y273 has been opened, the movement of Y273 not only connects but reshapes the sizes of the pond and vestibule (Fig. 3b,d). To test the functional importance of the flow restrictor constriction, we made a mutant, Y273F, that removed the Y273 hydroxyl that blocks the pathway (Figs. 1a,b and 2a). This change increased THIK1 activity to a level similar to the S136P change (I/I_{WT (wild type)} = 2.16 ± 0.15 and 2.36 ± 0.17, for Y273F and S136P, respectively) (Fig. 3e,f and Table 2). Reducing the flow restrictor size further with Y273L and Y273A yielded even larger current enhancements (I/I_WT = 4.17 ± 0.20 and 4.85 ± 0.23, respectively) (Fig. 3e,f and Table 2), supporting the idea that the Y273 flow restrictor is a key controller of THIK1 activity. The M2 pore-lining residue I139 supports Y273 (Extended Data Fig. 5a,b). Changing this position to alanine enhanced currents similar to Y273A (I/I_WT = 4.77 ± 0.25 and 4.85 ± 0.23, respectively) (Fig. 3e,f), consistent with the increased function of a glycine change at this site⁶⁰. Rather than producing additive effects, combining I139A and Y273A was less stimulatory than the individual mutants (I/I_WT = 3.78 ± 0.10) (Fig. 3e,f and Table 2), consistent with the structural interactions between these two residues. Similar trends were observed in the S136P background (Table 2). Y273F/L/A and I139A all increased currents more than S136P alone (I/IS136P = 2.45 ± 0.16, 3.25 ± 0.26, 3.30 ± 0.28 and 3.00 ± 0.26 for Y273F, Y273L, Y273A and I139A, respectively) (Supplementary Fig. 8a,b and Table 2) and there was no additive effect of I139A;Y273A (I/I_S136P = 3.12 ± 0.16) (Fig. 3e,f and Table 2). Thus, regardless of context, reducing the size of the flow restrictor residue or perturbing its nearest neighbor dramatically increased channel function. Together, these data indicate that the flow restrictor position Y273 limits ion flux through the channel.

Fig. 3 ∣. THIK1 flow restrictor controls inner cavity shape and function.

a–c, Pore profile comparisons for THIK1 (a; deep salmon), THIK1-S136P (b; cyan) and TREK1 (c; PDB 6CQ6, light blue)²⁴ calculated using HOLE⁷⁶. SF2, flow restrictor and asparagine ring residues are shown as sticks. Select channel elements are labeled. d, THIK1 model and pore profiles. e, Representative TEVC current–voltage responses for THIK1 (black) and THIK1 mutants: S136P (cyan), I139A (purple), I136A;Y273A (orange), Y273A (magenta), Y273F (green) and Y273L (red). Inset, protocol. f, Average currents $I∕I_{<\mathrm{FL}>}$ at 0 mV for THIK1 (black) and THIK1 mutants: S136P (cyan), I139A (purple), I136A;Y273A (orange), Y273A (magenta), Y273F (green) and Y273L (red), where $I_{<\mathrm{FL}>}$ is the average current for full-length THIK1. ****P < 0.0001, ***P < 0.001, **P < 0.01 and *P = 0.01–0.05. NS (not significant), P > 0.05. Error bars are the s.e.m. g, Pore profile of THIK1-S136P;Y273A (magenta). h, PMF for THIK1 (black), THIK1-Y273F (green), THIK1-Y273A (magenta), THIK1-S136P (cyan), THIK1-S136P;Y273F (dashed green) and THIK1-S136P;Y273A (dashed dark magenta). The z axis corresponds to that shown in d with z = 0 shifted from the bottom of the SF to the flow restrictor. Statistical analysis was performed using the Kruskal–Wallis test (nonparametric ANOVA) followed by Dunn’s multiple-comparisons test. In d,h, locations of the flow restrictor (light blue) and SF (orange) are indicated.

Table 2 ∣.

Relative current amplitudes at 0 mV

	THIK1			THIK1EM
		$I∕I_{\mathrm{WT}}$	n	$I∕I_{\mathrm{WT}}$	$I∕I_{S136P}$	n
Arginine sandwich	WT	1.00 ± 0.02	107	0.93 ± 0.13		15
	S136P	2.36 ± 0.17	16	1.39 ± 0.09	1.00 ± 0.06	28
	R92A	0.74 ± 0.06	27
	R258A	1.18 ± 0.07	30
Flow restrictor	I139A	4.77 ± 0.25	14	4.35 ± 0.38*	3.00 ± 0.26	3
	I139A;Y273A	3.78 ± 0.16	15	4.52 ± 0.37*	3.12 ± 0.26	3
	Y273A	4.85 ± 0.23	15	4.85 ± 0.43*	3.30 ± 0.28	14
	Y273F	2.16 ± 0.15	17	3.55 ± 0.23*	2.45 ± 0.16	13
	Y273L	4.17 ± 0.31	15	4.72 ± 0.38*	3.25 ± 0.26	10
PUFA site	F101A	1.40 ± 0.11	25
	Y102A	0.97 ± 0.08	25
	G105F	0.57 ± 0.04	11
	G105I	1.54 ± 0.24	9
	S246A	1.32 ± 0.10	18
	N261A	0.11 ± 0.01	14
	F262A	1.39 ± 0.11	16
	R92A;N261A	0.05 ± 0.01	6
	F101A;F262A	1.52 ± 0.12	3

^*S136P background.

To understand the consequences of flow restrictor disruption further, we determined the lipid ND-embedded structure of a mutant in which the flow restrictor residue was substituted to alanine, THIK1_EM-S136P;Y273A, at 2.39-Å resolution (Supplementary Figs. 9 and 10 and Extended Data Fig. 9). This structure is essentially identical to THIK1_EM-S136P ND (r.m.s.d. = 0.378 Å) (Extended Data Fig. 9a-d) and preserves the arginine sandwich, PUFA site lipid features (Extended Data Fig. 9a) and four SF ions (Extended Data Fig. 9e) of THIK1_EM and THIK1_EM-S136P. The absence of the flow restrictor tyrosine (Extended Data Fig. 9c,d) creates a continuous inner cavity (Fig. 3g) with a gap between Y273A β-carbons of 13.8 Å (Extended Data Fig. 9d) that is much larger than the 2.3-Å and 5.7-Å gaps between Y273 hydroxyls in the THIK1_EM and THIK1_EM-S136P structures (Fig. 2a and Extended Data Fig. 9d). As a result, the narrowest point is I139 (9.6 Å) (Fig. 3d and Extended Data Fig. 9c). This unobstructed path from the cytoplasm to SF, together with the functional data for Y273A mutants (Supplementary Fig. 8a,b and Table 2), further supports the idea that the flow restrictor and its movements control THIK1 function.

To examine how the flow restrictor affects energetic barriers traversed by permeant ions, we computed the potential of mean force (PMF) for potassium ion entry from the cytoplasm to the pond for THIK1 and Y273F, Y273A, S136P, S136P;Y273F and S136P;Y273A mutants. These calculations show that ions entering from the cytoplasm must surmount a ~6-kcal mol⁻¹ barrier at the flow restrictor and, following entry into the pond, are stabilized by a ~2-kcal mol⁻¹ exit barrier (Fig. 3h). The barrier height is similar for THIK1 and S136P, Y273F and S136P;Y273F mutants because of the fact that these simulations rapidly relax to the Y273 open conformation seen in THIK1-S136P (Supplementary Fig. 11 and Supplementary Videos 3 and 4). This rapid relaxation supports the idea that the S136P substitution biases the Y273 conformation toward a state that is naturally sampled by the channel. Y273F also eliminates a minimum before the barrier that occurs because of coordination of potassium by Y273 hydroxyl groups (Fig. 3h), in line with the idea that this residue directly interacts with the permeant ion. Furthermore, consistent with the wide intracellular opening in THIK1_EM-S136P;Y273A (Fig. 3g and Extended Data Fig. 9c,d), Y273A reduces the barrier between the cytoplasm and pond by ~4 kcal mol⁻¹ (Fig. 3h), explaining the stimulatory effects of Y273A versus Y273F (Fig. 3f). Taken together, these data support the idea that Y273 forms a barrier that limits ion passage from the cytoplasmic vestibule into the pond and can coordinate the permeant ion. Perturbation of this barrier affects channel function.

PUFA site lipid is important for channel function

All THIK1 structures have a lipid in the PUFA-binding site (Figs. 1d and 2a and Extended Data Figs. 7a-9a) that overlaps with the point of ligand control in the TREK family (Fig. 1g-i)^24,25,27,44. These structural observations, together with a prior report that the PUFA AA could stimulate rat THIK1 (refs. 38,39) with a stoichiometry indicating two binding sites³⁹, prompted us to investigate the roles of PUFAs in THIK1 modulation by first testing the functional consequences of mutations impacting the PUFA-binding site. Alanine substitution of the arginine sandwich residue that directly coordinates the PUFA carboxylate, R92A (Fig. 1f), reduced channel function (I/I_WT = 0.74 ± 0.06) (Fig. 4a,b and Table 2). Change of its partner, R258A (Fig. 1f), was mildly stimulating (I/I_WT = 1.18 ± 0.07) (Fig. 4a,b and Table 2). Substitutions at the other two PUFA carboxylate coordination sites, S246A and N261A (Fig. 1f), identified a particular importance for N261 as its removal eliminated channel function (I/IWT = 1.32 ± 0.10, 0.11 ± 0.01, 0.05 ± 0.01 for S246A, N261A and R92A;N261A, respectively) (Fig. 4a,b and Table 2). Substitution of the buttress residue from the P1 helix, Y102A, was neutral (I/IWT = 0.97 ± 0.08) (Table 2), whereas changes to residues framing the lateral membrane-facing portion of the PUFA site either enhanced (F101A, F262A, G105I and F101A;F262A) or decreased (G105F) function (Fig. 4a,b and Table 2). Placed in structural context, the data show that PUFA interactions with the residues that directly coordinate the lipid carboxylate (R92 and N261) are key for function, whereas the residues above (S246 and R258) or at the lateral opening to the membrane (F101, G105 and F262) tune activity (Fig. 4c). Taken together, these data support the idea that the PUFA site is critical for THIK1 activity.

Fig. 4 ∣. The THIK1 PUFA site is important for function.

a, Representative TEVC current–voltage responses for THIK1 (black) and THIK1 mutants R92A (cyan), F101A (yellow green), Y102A (dark green), G105F (red), G105I (orange), S246A (green), R258A (maroon), N261A (blue) and F262A (dark blue). Inset, protocol. b, Average currents $I∕I_{<\mathrm{FL}>}$ at 0 mV for THIK1 (black) and THIK1 mutants. Colors are as in a. Residues coordinating the lipid headgroup and lining the pocket are indicated. ****P < 0.0001, ***P < 0.001, **P < 0.01 and *P = 0.01–0.05. Statistical analysis was performed using a two-sided Mann–Whitney test. Error bars are the s.e.m. c, Effects of mutants on THIK1 basal activity. Colors indicate an increase (cyan), a decrease (red), both (purple) or no effect (white). d–g, Representative TEVC current–voltage responses (d,f) and time course responses (e,g) for THIK1 (purple) and THIK1-R92A (red) to EIC (d,e) and AA (f,g). d,f, Lipid structure. Unsaturated bond locations are marked by numbers. The lipid class (ω6) is indicated. Current–voltage curves show the baseline (black), lipid application (purple or red) and after 2-min washout (purple or orange dashed lines). h, Lipid modulation of THIK1 and THIK1-R92A $I∕I_0$ at 0 mV, where $I_0$ is the current before lipid application. Light purple and orange show the response after 2-min washout. Lipids are AA, AAMe, EPA, DGLA, α-LNL, EIC, OA and SA. $X:Y$ indicates the number of carbons and unsaturated bonds. ****P < 0.0001, ***P < 0.001, **P < 0.01 and *P = 0.01–0.05; NS, P > 0.05. Statistical analysis was performed using the Kruskal–Wallis test (nonparametric ANOVA) followed by Dunn’s multiple-comparisons test. Error bars are the s.e.m. i, TEVC dose–response curve for THIK1-R92A activation by EIC. The curve shows the fit to the Hill equation. EC₅₀ = 0.7 μM, Hill coefficient ‘n’ = 2.0. Error bars are the s.e.m.

As the PUFA site lipid is deeply buried in the P1–M4 interface and must be tightly bound to survive the purification required for structure determination, we decided to compare the effects of PUFA application to WT channels and the R92A mutant, as this change reduced basal activity (Fig. 4a,b and Table 2) and is expected to affect lipid binding. In line with the idea that the PUFA is strongly bound to the WT channel, we observed modest responses for THIK1 to the C18 lipid that best fits the cryo-EM density (EIC) (Fig. 4d,e) and AA (Fig. 4f,g), a C20 lipid reported to simulate rat THIK1 (refs. 38,39). By contrast, these two ω6 PUFAs produced a robust and lasting activation of R92A (Fig. 4d-h). Furthermore, AA methylester (AAMe), the neutral version of AA, was ineffective (Fig. 4h and Supplementary Fig. 12a,b), underscoring the importance of the negatively charged carboxylate. α-LNL, eicosapentanoic acid (EPA), dihomo-γ-EIC (DGLA) and oleic acid (OA) also activated R92A, contrasting their minimal or slightly inhibitory effects on THIK1 (Fig. 4h and Supplementary Figs. 12c-j). These results demonstrate that unsaturated fatty acids of different lengths from the ω3 (α-LNL and EPA), ω6 (EIC, AA and DLGA) and ω9 (OA) classes can activate the channel. Importantly, stearic acid (SA), a C18 saturated fatty acid, had no effect (Fig. 4h and Supplementary Fig. 12k,l). Measurement of the dose–response curve for THIK1-R92A to EIC revealed a saturating response (half-maximal effective concentration (EC₅₀) = 0.7 μM) having a stoichiometry that matches the number of observed PUFA sites (Figs. 1a and 4i). Together with the mutational data, these R92A effects on lipid modulation suggest that the PUFA is largely a structural element rather than a regulatory lipid in the WT channel and that mutations impacting the PUFA site can sensitize the channel to lipid modulation.

To understand how the PUFA site might accommodate a range of PUFAs, we used molecular dynamics simulations in which we placed different PUFAs at the EIC position. In all cases, the helices forming the PUFA site accommodated the different lipids with minimal movement (Extended Data Fig. 10a). By contrast, we saw varied behavior for the lipids. The C18 ω6 PUFA that fit the density best, EIC, was very stable, as was its C18 ω3 counterpart α-LNL (Extended Data Fig. 10b). DGLA and AA, two C20 ω6 PUFAs, rose through the binding site to place the carboxylate headgroup closer to the R92–R258 pair (Extended Data Fig. 10a,c,f). These movements decreased the PUFA carboxylate:R92 guanido group distance from the starting position to a constant distance of ~2.8 Å (Extended Data Fig. 10c-g) and illustrate that differences in lipid length and position of the final unsaturated bond can be accommodated by the PUFA site, consistent with the lateral exit gap around the lipid tail. We also noticed that PUFAs that moved the most (for example, AA) could make interactions with R258 (Extended Data Fig. 10f), suggesting that plasticity in the arginine sandwich (Extended Data Fig. 10h,i) may allow the channel to respond to varied PUFAs because of the enhanced electrostatic potential that surrounds this cation–cation pair⁶². As changes in the P1–M4 interface and its surrounding loops that comprise the PUFA site have diverse effects on the K_2P SF C-type gate^{21,24,25,27,44,63,65}, such plasticity likely underlies the contrasting abilities of various PUFAs to inhibit or activate THIK1. Together, these combined studies of the PUFA site lend further support to the importance of the K_2P P1–M4 interface in K_2P function and modulation ^{21,24,25,44,52,63}.

Discussion

Structural characterization of THIK1 reveals two functionally important elements, the flow restrictor and PUFA site, that embody the THIK subfamily and that are built upon the K_2P architecture shared with TREK^24,28,29, TWIK^30,31, TALK³² and TASK³³ channels. Distinct from other K_2P channels^24,27-33, the THIK1 aqueous inner cavity is divided into two chambers, the vestibule and pond, by the flow restrictor Y273 from the M4 helix. The hydrophilic nature of the flow restrictor, a barrier in which the component amino acids are hydrogen-bonded (Figs. 1b and 2a), sets it apart from classical ion channel gates where permeation barriers comprise hydrophobic residues⁵⁷ and provides a rationale for the observation that THIK1 has one of the smallest single-channel conductances among K_2P channels³⁸. Movement of the hydroxyls forming this barrier enable it to affect permeant ion passage from the cytoplasmic vestibule into the pond and control ion flux (Figs. 2a and 3 and Supplementary Videos 1 and 2). The unusual physicochemical nature of the flow restrictor provides an important example of the diverse ways that an ion channel family can be modified to affect function.

All K_2P channels, except for THIK subfamily members (Extended Data Fig. 5d and Supplementary Fig. 1)³⁷, have a proline at a site where there is a natural kink in the M2 transmembrane helix (Extended Data Fig. 5h). The structure of the S136P mutant shows that this M2 break accommodates the activating proline substitution with minimal local helical backbone changes (Fig. 2e). This local structural similarity between WT and S136P channels (Fig. 2e) is consistent with ideas regarding a proline origin for nonproline kinks in transmembrane protein helices⁶⁶ and suggests that the THIK subfamily originated from a predecessor having a proline at this site. Linkage of flow restrictor movement to subtle M2 structure changes (Fig. 2a-d), together with the rapid relaxation of the flow restrictor residue to S136P-like states in simulations (Supplementary Fig. 11), suggests that the S136P substitution exploits M2 helix flexibility at the helical kink to influence Y273 flow restrictor conformation (Extended Data Fig. 5h and Supplementary Videos 1 and 2). The resulting widening of the flow restrictor barrier shows how proline substitution at this M2 site increases the activity of both THIK subfamily members, THIK1 (ref. 37) and THIK2 (ref. 37), and emphasizes the role of the M2 helix in shaping K_2P function^27,52,67-69. Together, our observations strongly support the idea that Y273 acts as a barrier that limits passage of ions from the cytoplasmic cavity into the pond and that the perturbation of this barrier is central to THIK channel function (Fig. 5).

Fig. 5 ∣. Model of THIK1 function.

Changes in channel conformational equilibrium at the flow restrictor affect ion passage.

The THIK1 PUFA site is occupied by a single chain lipid buried in the P1–M4 interface (Fig. 1c,d) and uses a signature element of the THIK subfamily, the arginine sandwich formed by the stacking of R92 and R258 guanidium groups, to interact with the PUFA carboxylate (Extended Data Fig. 5c,f). This stacked cation–cation pair creates an enhanced electrostatic potential that influences interactions with oppositely charged partners⁶² and appears to enable the PUFA site to accommodate PUFAs from both ω3 and ω6 classes (Fig. 4c-i, Extended Data Fig. 10 and Supplementary Fig. 12,). Remarkably, the THIK PUFA site structure encapsulates the essence of the TREK subfamily K_2P modulator pocket where synthetic small-molecule activators bind^24,25,27,44 (Fig. 1g-i). This structural coincidence of ligand-binding sites shows that, as anticipated^24,27, this interdomain modulator binding site exists in diverse K_2P subfamilies and can interact with natural ligands.

The THIK1 PUFA appears to act as a structural cofactor, as evidenced by its copurification with the channel (Extended Data Figs. 2b and 3b,c) and the modest responses of the WT channel to application of a variety of PUFAs (Fig. 4d-h and Supplementary Fig. 12). Notably, weakening the interactions with the PUFA carboxylate headgroup through the R92A substitution sensitizes the channel to these same lipids. THIK1-R92A can be activated by EIC, a PUFA that fits the observed experimental density (Extended Data Figs. 2b and 3c), with a stoichiometry that matches the number of observed PUFA sites (Fig. 4i) and the response of rat THIK1 to AA³⁹. Although THIK channel pharmacology remains limited⁴³, it is noteworthy that the R92A change also attenuates THIK1 inhibition by IBMX⁴⁰, suggesting that the PUFA site participates in the response of THIK channels to exogenous agents. Together, these observations expand the role for the P1–M4 interface in the function of diverse K_2P subfamilies^{21,24,25,44,52,63} and raise the possibility that PUFAs or other endogenous factors modulate other K_2P channels^26,27,58,59 at the equivalent interface.

SF C-type gating is the main K_2P gating mechanism^21-27 and is operative in other potassium channel classes^27,70. Notably, functional and simulation studies indicate that PUFAs such as EIC activate the voltage-gated potassium channel KCNQ1 by stabilizing its SF gate through binding to a site, termed ‘site II’, framed by the PD P and S6 helices^46-49. Comparison of the KCNQ1 and THIK1 PDs shows not only that K_v and K_2P channel pores share a common architecture (r.m.s.d._Cα = 0.257 for SF P helix) (Fig. 6a) but also that the KCNQ1 site II coincides with the THIK1 P1–M4 interface PUFA site (Fig. 6a,b). This convergence strongly suggests that the ligand-binding properties of the K_2P PUFA site are shared with other VGIC superfamily members and that PUFA occupation of such interdomain sites influences SF gate function. Identification of this PD interdomain site that is used by small-molecule activators and PUFAs in K_2P channels^24,25,44 and PUFAs in K_v channels^46-49 provides a framework for addressing the interplay between modulator binding and channel elements that control ion flux in the VGIC superfamily. Moreover, this common regulatory framework highlights the potential of targeting this class of interdomain interfaces for modulator development.

Fig. 6 ∣. A common PUFA site in VGIC PDs.

a, Extracellular view showing superposition of THIK1 (deep salmon) and KCNQ1 (marine, light blue, cyan and light cyan; PDB 6UZZ)⁷⁷. K326 and D301 residues framing the putative KCNQ1 PUFA site, site II⁴⁷ and EIC (yellow orange) are shown in space-filling representation. P1–M4 and corresponding P–S6 interfaces are indicated. The circle denotes PD. Voltage sensor domains (VSDs) are labeled. KCNQ1 residues are in italics. b, Comparison of THIK1 (deep salmon) and KCNQ1 (marine and light blue; PDB 6UZZ)⁷⁷ P1–M4 and P–S6 interfaces. Residues framing the putative KCNQ1 PUFA site⁴⁷, K326 and D301, and EIC (yellow orange) are shown in space-filling representation. The arginine sandwich pair is indicated.

Some K_2P channels, including the THIK1–THIK2 pair⁷¹, can function as heterodimers^71-75. Although the structural consequences of K_2P heterodimerization remain unknown²⁷, the THIK1 signature features, the flow restrictor and PUFA site, are present in THIK2 (Extended Data Fig. 5) and are, thus, likely to be important in both THIK2 homodimers and THIK1–THIK2 heterodimers. These elements may also contribute to THIK modulation by anesthetics^38,39, cold temperatures³⁸ and small-molecule modulators^40,43. The THIK subfamily is expressed in a wide range of tissues³⁹ and is thought to affect anesthesia³⁴, apoptosis³⁵ and ischemic responses³⁶. There is growing evidence that THIK1 is important for microglial function^18,19,41,42 in neuroinflammation and synapse development^18,19,42. The THIK subfamily structural features defined here should aid development of agents to affect Alzheimer disease⁴³, amyotrophic lateral sclerosis⁴³ and autism¹⁹, where such processes are thought to affect disease progression.

Methods

Expression and purification of THIK1_EM and mutants

The gene for a construct of human THIK1 (UniProt Q9HB14), THIK1_EM, spanning residues 1–350 and bearing N59Q and N65Q substitutions to remove glycosylation sites in the cap domain, followed by a 3C protease cleavage site, monomeric enhanced green fluorescent protein (mEGFP) and a 8×His-tag, was cloned into a modified pFastBac expression vector in which the polyhedrin promotor was replaced by a mammalian cell active cytomegalovirus promoter⁷⁸. Expression vectors for S136P and S136P;Y273A mutants were made on this background. These vectors were used for recombinant bacmid DNA generation using chemically competent DH10EmBacY (Geneva Biotech) cells. These bacmids were then used to transfect Spodopetera frugipdera Sf9 cells to make baculoviruses for the constructs of interest⁷⁹.

Flasks of Expi293F cells (Gibco, A14528) were grown to 2.5–3 × 10⁶ cells per ml at 37 °C supplemented with 8% CO₂ and shaken at 120 rpm before transduction with 10% (v/v) baculovirus stock of THIK1_EM or mutants. Then, 16–24 h after transduction, 10 mM sodium butyrate was added to the culture to enhance protein expression⁸⁰. Flasks were then moved to 30 °C for 24 h before harvesting the cells using centrifugation at 5,000g for 20 min. The pellet was gently washed with Dulbecco’s PBS (Gibco, 14190144) to remove leftover culture medium, centrifuged (3,500g for 20 min) to recover the cell pellet and then flash-frozen with liquid nitrogen for storage at −80 °C.

A pellet from ~5 L of culture was resuspended in 250 ml of hypotonic buffer (20 mM KCl, 1 mM, 10 mM Tris-HCl pH 7.8, PMSF and 0.1 mg ml⁻¹ DNase 1, supplemented with Pierce protease inhibitor (Thermo Fischer Scientific)) and was gently stirred on a MONO DIRECT Variomag magnetic stirrer (Thermo Fischer Scientific) at 4 °C for 30 min. The membranes were isolated by ultracentrifugation at 185,500g for 30 mins. Using a 16-gauge Precision Glide needle and 30-ml syringe, the membrane pellet was homogenized in 300 ml of solubilization buffer (buffer S) (500 mM KCl and 50 mM Tris-HCl pH 7.8, supplemented with 1 mM PMSF, 0.1 mg mll⁻¹ DNase 1, 1% (w/v) β-DDM, 0.2% (w/v) cholesteryl hemisuccinate Tris salt (CHS) and Pierce protease inhibitor). Protein extraction was performed by gently stirring at 4 °C for 2 h and ultracentrifugation at 185,500g for 1 h to remove cell debris. The supernatant was diluted 1:1 with buffer S, supplemented with 0.1% β-DDM and 0.02% CHS to reduce the detergent concentration and incubated for 3 h with 10 ml of Talon metal affinity resin (Takara Biotech) pre-equilibrated with buffer S supplemented with 0.1% β-DDM and 0.02% CHS. Resin was then collected in an Econo-Pac chromatography column (Bio-Rad) and washed with ten column volumes (CV) of buffer S supplemented with 0.5% β-DDM and 0.1% CHS, followed by ten CVs supplemented with 0.2% β-DDM and 0.04% CHS, ten CVs of buffer S supplemented with 0.1% β-DDM and 0.02% CHS and ten CVs of wash buffer (500 mM KCl, 0.05% β-DDM, 0.01% CHS and 50 mM Tris-HCl pH 7.8). The resin was then washed using ten CVs of wash buffer supplemented with 10 mM imidazole and ten CVs of wash buffer supplemented with 30 mM imidazole. THIK1_EM–GFP was eluted with two CVs of wash buffer supplemented with 500 mM imidazole and the eluent was concentrated using an Amicon Ultra-15 centrifugal unit with a 100-kDa cutoff to a final volume of 6 ml. Two Econo-Pac 10DG desalting columns (Bio-Rad) were equilibrated using 20 ml of desalting buffer (200 mM KCl, 0.05% β-DDM, 0.01% CHS and 20 mM Tris-HCl pH 7.8) and 3 ml of concentrated THIK1_EM–GFP was applied to each desalting column and eluted using 4 ml of desalting buffer before concentrating to 5 mg ml⁻¹ using an Amicon Ultra-15 filter (100-kDa cutoff).

To make ND complexes, brain extract polar lipids (Avanti polar lipids) were prepared by dissolving in chloroform, drying under vacuum and storing at −80 °C. For reconstitution, the lipids were dissolved in an 80 mM stock solution by sonication in 200 mM KCl, 0.5% β-DDM, 0.1% CHS and 20 mM Tris pH 7.8. THIK1_EM from the metal affinity chromatography step, purified membrane scaffold protein (MSP) 1E1 (ref. 81) and prepared lipid stock were incubated together at a 1:3:300 molar ratio for 1 h in a microcentrifuge tube gently rotating at 4 °C on a Boekel 260200 Orbiton platform rotator. Adsorbent SM-2 biobeads (Bio-Rad, 1528920) were prepared by washing first with methanol, followed by water and then size-exclusion chromatography (SEC) buffer (200 mM KCl and 20 mM Tris-HCl pH 7.8). For detergent removal, 120 mg of prepared biobeads were added per 1 ml of reaction mixture and incubated for 1 h gently rotating at 4 °C. The reaction mixture was moved to another tube using a 27-gauge Precision Glide needle and incubated with another 120 mg aliquot of prepared biobeads per ml rotating gently at 4 °C overnight.

Next, 200 μl of anti-GFP nanobody Sepharose resin⁸² pre-equilibrated with SEC buffer was combined with 1 ml of the recovered sample and incubated in 2-ml microcentrifuge tubes gently rotating at 4 °C for 2 h. The resin was collected in a column and washed using ten CVs of SEC buffer to remove unbound components including empty NDs. On-column phosphatase treatment was performed by incubating the resin with 10 ml of SEC buffer supplemented with 1 mM MnCl₂ and 4,000 U (10 μl) of lambda protein phosphatase (LPP; New England BioLabs, P0753L) standing at 4 °C overnight. The resin was then washed using five CVs of SEC buffer and the mEGFP affinity tag was cleaved using 10 ml of SEC buffer supplemented with 400 mM KCl, 1 mM EDTA and 3C protease⁸³ at a 50:1 ratio of resin to protease volume, treated standing at 4 °C overnight. The cleaved ND sample was eluted using 10 ml of SEC buffer and concentrated to 0.5 ml using an Amicon Ultra-15 filter (100-kDa cutoff) before purification by SEC on a Superose 6 increase 10/300 column pre-equilibrated with SEC buffer. Peaks obtained were validated by SDS–PAGE before finally concentrating to 0.6–1.0 mg ml⁻¹ for cryo-EM sample preparation. Protein concentrations were determined by absorbance⁸⁴ using extinction coefficients of 104,740 and 171,225 M⁻¹ cm⁻¹ for detergent and ND samples, respectively.

For the preparation of detergent samples, hypotonic wash and solubilization were performed as described above except that both buffers were supplemented with 1 mM TCEP. The supernatant acquired after ultracentrifugation following solubilization was incubated with the anti-GFP nanobody Sepharose resin (1 ml of resin per 10 g of Expi293F cells) gently rotating at 4 °C for 2 h. Resin wash, on-column treatment with LPP and 3C protease application were performed as described above. The cleaved sample was eluted in SEC buffer supplemented with 0.05% β-DDM, 0.01% CHS and 1 mM TCEP and concentrated to 1 ml using an Amicon Ultra-15 filter (100-kDa cutoff) before purification by SEC using a Superdex 200 increase 10/300 GL column equilibrated with 200 mM KCl, 1 mM TCEP, 0.01% glyco-diosgenin (GDN) and 20 mM Tris-HCl pH 7.8. Peaks obtained were validated by SDS–PAGE before final concentration using an Amicon Ultra 0.5-ml filter (100-kDa cutoff) to 2–2.5 mg ml⁻¹ for cryo-EM sample preparation.

Cryo-EM sample preparation and data collection

First, 3 μl of concentrated protein sample was applied to a glow-discharged (30 s at 15 mA) grid (Quantifoil Au or Cu R1.2/1.3 300 mesh); after a wait time of 5 s, grids were blotted once for 2.5–6 s (4 °C and 100% humidity) using an FEI Vitrobot Mark IV (Thermo Fisher Scientific) and plunge-frozen in liquid ethane. The grids were screened using Serial EM⁸⁵ on 200-kV Talos Arctica cryo-transmission EM instrument (Thermo Fisher Scientific) equipped with a K3 direct detector camera (Gatan) at the University of California, San Francisco (UCSF) EM facility. Micrographs were collected using EPU data acquisition software on FEI Titan Krios (Thermo Fisher Scientific) (Stanford Linear Accelerator Center (SLAC) National Accelerator Laboratory) at 300 keV, equipped with a K3 direct-electron detector and post-BioQuantum GIF energy filter (Gatan), set to a slit width of 20 eV. Super-resolution counting mode at a nominal magnification of ×105,000 with a pixel size of 0.43 Å (physical pixel size of 0.864 Å) and a nominal defocus range between −0.8 and −2.2 μm was used (exposure time: 2.5 s) and 50 frame video stacks were collected with a total dose of ~50 e⁻ per Ų.

Structure determination (image processing, model building and refinement)

A total of 14,254, 13,491, 13,544, 6,830 and 16,053 video stacks were collected for THIK1_EM ND, THIK1_EM detergent, THIK1_EM-S136P ND, THIK1_EM-S136P detergent and THIK1_EM-S136P;Y273A ND samples, respectively. Cryo-EM data processing was performed using cryoSPARC (version 3.2.0)⁸⁶. Patch motion correction (cropped by a factor of 2) and patch contrast transfer function (CTF) estimation (default parameters) were performed before manually curating the exposures by ice thickness, CTF-fit resolution, total full-frame motion and motion curvature. Selected micrographs were used for reference-free circular blob picker (diameter: 80–200 Å) for particle picking followed by extraction at a box size of 288 Å. Several rounds of two-dimensional classification were performed and further cleaning was performed by categorizing particles by three-dimensional ab initio reconstruction and heterogeneous refinement. Finally, nonuniform and local refinement was performed to improve resolution and map quality.

The optimal sharpening for each density map was performed using the phenix.auto_sharpen tool in PHENIX^87,88. A preliminary model for THIK1_EM was built using TASK1 (Protein Data Bank (PDB) 6RV2)³³, as this K_2P is closest in sequence to THIK1, and docked in the final map using phenix.dock_in_map. The model then was manually adjusted to fit the density using Coot (version 0.9.8.7)⁸⁹ and disordered residues from the N terminus, C terminus and M2–M3 loop were removed. Iterative structure refinement and model building were performed using phenix.real_space_refine⁹⁰ before adding cofactors and further refinement. The final geometry validation statistics were calculated by MolProbity⁹¹ and refinement statistics were generated using the comprehensive validation (cryo-EM) function in Phenix. Figures were prepared and model comparisons were performed using UCSF Chimera (version 1.15)⁹² and the PyMol package (http://www.pymol.org/pymol). Contact analysis was performed using LIGPLOT^93,94.

Two-electrode voltage-clamp (TEVC) electrophysiology

Xenopus laevis oocytes were isolated according to the UCSF Institutional Animal Care and Use Committee (protocol AN193390) and digested using collagenase (Worthington Biochemical Corporation, LS004183; 0.8–1.0 mg ml⁻¹) in Ca²⁺-free ND96 (96 mM NaCl, 2 mM KCl, 3.8 mM MgCl₂ and 5 mM HEPES pH 7.4). Oocytes were maintained at 18 °C in ND96 (96 mM NaCl, 2 mM KCl, 1.8 mM CaCl₂, 2 mM MgCl₂ and 5 mM HEPES pH 7.4) supplemented with antibiotics (100 U per ml penicillin, 100 μg ml⁻¹ streptomycin and 50 μg ml⁻¹ gentamicin) and used for experiments within 1 week of isolation.

K_2P channels were subcloned into a previously reported pGEMHE/pMO vector²¹. mRNA for oocyte injections was prepared from linearized plasmid DNA (linearized with AflII) using the mMessage Machine T7 transcription kit (Thermo Fisher Scientific). RNA was purified using RNEasy kit (Qiagen) and stored as stocks and dilutions in RNase-free water at −80 °C. Defolliculated stage V–VI oocytes were microinjected with 0.5–38 ng of mRNA and currents were recorded within 18–96 h after microinjection. For amplitude comparisons, currents were compared at the same time of expression for sample injected with equivalent amounts of mRNA. For PUFA response studies, cells expressing the equivalent amplitudes of basal current were compared.

For recordings, oocytes were impaled with borosilicate recording microelectrodes (0.3–2.0 MΩ resistance) backfilled with 3 M KCl and were subjected to constant perfusion of ND96 at a rate of 3–5 ml min⁻¹. Except where otherwise indicated, the recording solution was 96 mM NaCl, 2 mM KCl, 1.8 mM CaCl₂ and 1.0 mM MgCl₂, buffered with 5 mM HEPES pH 7.4. AA (Sigma-Aldrich, A3611; CAS number: 506-32-1), AAMe (Sigma-Aldrich, A9298; CAS number: 2566-89-4), EPA (Sigma-Aldrich, E2011; CAS number: 10417-94-4), DGLA (Sigma-Aldrich, E4504; CAS number: 1783-84-2), α-LNL (Sigma-Aldrich, L2376; CAS number: 463-40-1) and EIC (Sigma-Aldrich, L1376; CAS number: 60-33-3) were prepared immediately before use from DMSO stocks (100 mM), with final DMSO concentrations of 0.1%. For each recording, control solution (ND96) was perfused over a single oocyte until current was stable before switching to solutions containing the test compounds at various concentrations. Currents were evoked from a holding potential of −80 mV followed by a 500-ms ramp from −140 to +50 mV. Data were recorded using an Axoclamp 900A amplifier (Molecular Devices) controlled by pCLAMP10.7 software (Molecular Devices) and digitized at 1 kHz using a Digidata 1550B digitizer (Molecular Devices). Representative traces and dose–response plots were generated in GraphPad Prism 9 (GraphPad Software). The dose response was fitted with the Hill equation $(I∕I_0=\text{minimum}+(X^{\text{Hill slope}})\times (\max -\min )∕(X^n+{{\mathrm{EC}}_{50}}^n))$ to determine the EC₅₀ and Hill coefficient ‘n’.

Data quantification and statistical analysis

For Fig. 3f, the results are from 24 independent oocyte batches for THIK1, 6 independent oocyte batches for THIK1-S136P, 3 independent oocyte batches for THIK1-I139A;Y273A and 4 independent oocyte batches for THIK1-I139A, THIK1-Y273A, THIK1-Y273F and THIK1-Y273L. P values for Fig. 3f are as follows: P < 0.0001 (THIK1 versus THIK1-S136P), P < 0.0001 (THIK1 versus THIK1-I139A), P < 0.0001 (THIK1 versus THIK1-I139A;Y273A), P < 0.0001 (THIK1 versus THIK1-Y273A), P = 0.0005 (THIK1 versus THIK1-Y273F), P < 0.0001 (THIK1 versus THIK1-Y273L), P > 0.9999 (THIK1-I139A versus THIK1-I139A;Y273A), P > 0.9999 (THIK1-Y273A versus THIK1-I139A;Y273A) and P = 0.0508 (THIK1-Y273A versus THIK1-Y273F). Statistical significance was assessed using a Kruskal–Wallis test (nonparametric analysis of variance (ANOVA)) followed by Dunn’s multiple-comparisons test. Full detailed statistical information is available in the Source Data.

For Fig. 4b, the results are from 26 independent oocyte batches for THIK1, 5 independent oocyte batches for THIK1-R92A, 6 independent oocyte batches for THIK1-R258A, 3 independent oocyte batches for THIK1-N261A and THIK1-Y102A, 2 independent oocyte batches for THIK1-G105F, THIK1-G105I and THIK1-F262A and 4 independent oocyte batches for THIK1-S246A and THIK1-F101A. P values for Fig. 4b are as follows: P < 0.0001 (THIK1 versus THIK1-R92A), P = 0.0158 (THIK1 versus THIK1-R258A), P = 0.0001 (THIK1 versus THIK1-S246A), P < 0.0001 (THIK1 versus THIK1-N261A), P = 0.0012 (THIK1 versus THIK1-F101A), P = 0.5993 (THIK1 versus THIK1-Y102A), P < 0.0001 (THIK1 versus THIK1-G105F), P = 0.0178 (THIK1 versus THIK1-G105I) and P = 0.0005 (THIK1 versus THIK1-F262A). Statistical significance was assessed using a two-sided Mann–Whitney test. Full detailed statistical information is available in the Source Data.

For Fig. 4h, the results are from five independent oocyte batches for AA, two independent oocyte batches for EIC, α-LNL, OA and DGLA, one independent oocyte batch for SA and three independent oocyte batches for AAMe and EPA on THIK1 and THIK1-R92A. P values for Fig. 4h are as follows: P = 0.0135 (α-LNL, THIK1 versus THIK1-R92A), P < 0.0001 (EIC, THIK1 versus THIK1-R92A), P < 0.0001 (OA, THIK1 versus THIK1-R92A), P > 0.9999 (SA, THIK1 versus THIK1-R92A), P < 0.0001 (EPA, THIK1 versus THIK1-R92A), P < 0.0001 (AA, THIK1 versus THIK1-R92A), P = 0.7920 (AAMe, THIK1 versus THIK1-R92A) and P = 0.0027 (DGLA, THIK1 versus THIK1-R92A). Statistical significance was assessed using an ordinary one-way ANOVA followed by Šídák’s multiple-comparisons test. Full detailed statistical information is available in the Source Data.

For Fig. 4i, the results are from two independent oocyte batches for EIC on THIK1-R92A.

For Extended Data Fig. 1, the results are from 19 independent oocyte batches for THIK1 and 8 independent oocyte batches for THIK1_EM. The P value is as follows: P = 0.2466 (THIK1 versus THIK1_EM). Statistical significance was assessed using a two-sided Mann–Whitney test. Full detailed statistical information is available in the Source Data.

Molecular dynamics and computational analysis

Simulation methods.

General setup.

Simulation systems were prepared using CHARMM-GUI^95,96. Experimental structures of THIK1, THIK1-S136P and THIK1-S136P;Y273A (residues 14–167 and 188–295) were embedded in a pure 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine bilayer. For S136, missing residues 163–167 were transferred from the S136P mutant structure after superposing the channels. Neutral terminal capping was used and substitution of Y273 to phenylalanine or alanine was performed as appropriate to generate Y273F, Y273A and S136P;Y273A mutant structures. The system was solvated with neutralizing 0.18 M KCl in a box size of 130 × 130 × 130 ų. Fatty acids (AA, DGLA, α-LNL or EIC) were placed in the lipid-binding site as appropriate. System builds had ~100,000 atoms. A list of all simulations is found in Supplementary Table 1.

All simulations were performed using GROMACS 2020.6 (ref. 97) with the CHARMM-36m force field⁹⁸ and TIP3P water. Potassium ions were manually placed in the filter, as described below. We first performed energy minimization followed by a multistep equilibration gradually reducing restraints on protein and lipid atoms over 5 ns. For AA, an additional initial energy minimization step was performed to ensure all double-bonds adopted the cis conformation, using small (<0.5 Å) manual movement of fatty acid carbons and a modified force field penalizing the trans conformation. The original force field was used in all subsequent steps.

A simulation timestep of 2 fs was used. Coordinates were saved every 100 ps. Temperature was maintained at 303.15 K using v-rescale (time constant: 1 ps) and pressure at 1 atm using a Parinello–Rahman semi-isotropic barostat⁹⁹ (time constant: 5 ps, compressibility: 4.5 × 10⁻⁵ bar⁻¹). Hydrogens were constrained using the LINCS¹⁰⁰ algorithm. Nonbonding interactions were cut off at 1.2 nm, with force-switching from 1.0 nm. The particle mesh Ewald¹⁰¹ method was used to treat long-range electrostatic interactions. The nearest neighbor list was updated every 20 steps with a cutoff at 1.2 nm and a Verlet cutoff scheme.

Unbiased simulations.

For unbiased molecular dynamics simulations, four K⁺ ions were placed in the filter at locations observed in the structures (S1, S2, S3 and S4). For each build system, we performed three independent repeats starting from the same energy-minimized structure, followed by equilibration and then a 200-ns production run.

Umbrella sampling simulations.

The PMF for a K⁺ ion entering the channel from the cytoplasm was determined along the z axis through the central pore, parallel to the membrane normal. Specifically, the reaction coordinate was taken as the z distance from the K⁺ to the center of mass of the Cα atoms of residues 138–144 and 272–278 (corresponding to the location of the flow restrictor). Windows were created at 1-Å intervals between +7 and −13 Å and then at −14.5, −16, −17.5, −19 and −20.5 Å, for a total of 26 windows. The reaction coordinate value was restrained with a harmonic potential with force constant 1,000 kJ mol⁻¹ nm⁻². An additional flat-bottom restraint with a 15-Å radius was applied in the xy plane to prevent the ion from moving away from the protein as the windows approach the bulk. Reaction coordinate values were saved every 1,000 steps.

For all umbrella windows, three K⁺ ions were restrained in the S0, S2 and S4 sites (based on the most common ion configuration observed in our unbiased simulations) using a flat-bottom potential in the z direction with a width of 2 Å and force constant of 1,000 kJ mol⁻¹ nm⁻², centered on the center of mass of the O atoms of residues 114, 115, 241 and 242, residues 112, 113, 239 and 240 and residues 110, 111, 237 and 238, respectively.

Equilibration was performed separately for each umbrella window with the umbrella restraints, before a production run of 100 ns. For THIK1 and Y273F mutant, selected windows when the potassium ion passed the flow restrictor (z = −1 to 2 Å) were run in duplicate.

Analysis.

Analysis was performed using MDAnalysis^102,103. To investigate the behavior and effect of bound FA, protein snapshots were first aligned to the respective initial experimental structure using the Cα atoms of the transmembrane portion (residues 14–42, 96–162 and 188–295) and the r.m.s.d. of the fatty acid (C atoms) or protein residues comprising the fatty-acid-binding site (Cα of residues 97–110 and 254–270) relative to the initial structure measured. We excluded the first 50 ns in all calculations to allow for equilibration.

PMFs were obtained from umbrella sampling simulations using the Grossfield weighted histogram analysis method (WHAM)¹⁰⁴. Convergence was assessed by running WHAM on blocks of 20 ns, starting at 5-ns intervals; times before the profile settles were discarded as equilibration. Profiles were aligned by setting ‘bulk’ to 0 kcal mol⁻¹, using an average value between −15 and −20 Å.

Extended Data

Extended Data Fig. 1 ∣. Functional characterization of THIK1_EM and cryo-EM analysis in MSP1E1 nanodiscs.

a, Exemplar Two Electrode Voltage Clamp (TEVC) recordings of K_2P13.1(THIK-1) (grey) and THIK1_EM (orange) in Xenopus oocytes. 0 mV trace is indicated. Inset shows protocol. b, Representative TEVC current-voltage responses for K_2P13.1(THIK-1) (black) and THIK1_EM (orange). Inset shows protocol. c, Average currents $\mathrm{I}∕{\mathrm{I}}_{<\mathrm{FL}>}$ at 0 mV where ${\mathrm{I}}_{<\mathrm{FL}>}$ = average currents for full length K_2P13.1 (THIK-1). n.s. p > 0.05. Statistical analysis was performed using Kruskal-Wallis test (nonparametric ANOVA) followed by Dunn’s multiple comparisons test. Error bars are S.E.M. d, Exemplar SEC (Superose 6 Increase 10/300 GL) for THIK1_EM:MSP1E1 nanodiscs. e, peak fraction SDS-PAGE from ‘d’. More than three purifications were carried out for consistency. f, electromicrograph (~105,000x magnification) from 14,254 movies, and g, 2D class averages. h, Workflow for cryo-EM data processing for the THIK1_EM:MSP1E1 nanodisc complex in cryoSPARC-3.2⁸⁶. Red arrow indicates the class of particles extracted without Fourier cropping after the initial cleanup. These were further subjected to heterogeneous, non-uniform, and local refinement jobs that resulted in the final map at 2.65 Å (red box). i, Particle distribution plot and j, gold-standard Fourier Shell Correlation (FSC) curves for the THIK1_EM:MSP1E1 nanodisc complex.

Extended Data Fig. 2 ∣. THIK1_EM:MSP1E1 nanodisc (ND) cryo-EM map and model quality.

a, Cryo-EM map of the THIK1_EM:ND complex. THIK1_EM and associated lipids are colored magenta and turquoise. Locations of select channel elements are indicated. Nanodisc density is transparent. b, Cryo-EM maps for indicated THIK1_EM elements. Select residues are indicated. Channel elements are deep salmon. Linoleic acid (EIC) is yelloworange. Maps are rendered at 4-5σ. c, THIK1_EM:ND local resolution. Select channel elements and lipid belt are labeled. d, THIK1_EM:ND local B-factor.

Extended Data Fig. 3 ∣. THIK1_EM:ND cryo-EM maps of key features.

a, Selectivity filter and flow restrictor densities (6σ) for THIK1_EM:ND (deep teal and salmon). Select residues are labeled. Potassium ions are shown as purple spheres. b, Cyro-EM map of the THIK1_EM:ND complex showing the channel cartoon (magenta and turquoise) and location of PUFA density (red and yelloworange). Turquoise subunit transmembrane helices are indicated. Nanodisc density is transparent. c, Slice through the THIK1_EM:ND complex cyro-EM map (7σ) showing location of PUFA density (red and yelloworange). Inset shows local density. PUFA site shows linoleic acid (EIC). Grey bars denote the membrane.

Extended Data Fig. 4 ∣. K_2P13.1 (THIK-1) structure comparisons.

Superpositions of K_2P13.1(THIK-1) (salmon) (PDB:9BSN) with: a, K_2P2.1(TREK-1):ML335 (light blue) (PDB:6CQ6)²⁴ (RMSD_Ca = 1.1613 Å) b, K_2P10.1(TREK-2):norfluoxetine (Nfx) (hot pink) (PDB:4XDK)²⁹ (RMSD_Ca = 1.580 Å). c, K_2P5.1(TASK-2) pH 6.5 (orange) (PDB:6WLV)³² (RMSD_Ca = 1.816) and pH 8.5 (teal) (PDB:6WM0) (RMSD_Ca = 1.639 Å)³². d, K_2P1.1(TWIK-1) (deep olive) (PDB:3UKM)³⁰ (RMSD_Ca = 2.277 Å). e, K_2P3.1(TASK-1) (grey) (PDB:6RV2)³³ (RMSD_Ca = 1.323 Å). For simplicity, only one K_2P13.1(THIK-1) Arg sandwich position and PUFA site are labeled. Site of the K_2P3.1 (TASK-1) X-gate is labeled in ‘e’. Each panel shows side (left) and cytoplasmic (right) views. Grey bars denote the membrane.

Extended Data Fig. 5 ∣. K_2P13.1(THIK-1) structural and sequence features.

a, Cytoplasmic view of the K_2P13.1(THIK-1) Asn ring (Asn134 and Asn227) and flow restrictor barrier (Ile139 and Tyr273). Ser136 is also shown. Linoleic acid (EIC) is shown in space filling. b, Lateral view of the K_2P13.1(THIK-1) central cavity. Asn ring (Asn134 and Asn227) and flow restrictor barrier (Ile139 and Tyr273) are labeled. EIC and S4 potassium ion (violet) are shown in space filling. Pond and inner vestibule regions are shown in blue. c, M4 sequence comparison. Labels indicate: K_2P13.1(THIK-1) EIC contact residues (orange), flow restrictor (red), Asn ring (green), and K_2P2.1(TREK-1) K_2P modulator pocket residues that contact the ML335 activator (light blue)²⁴. Black asterisk denotes Arg sandwich residue. d, M2 sequence comparison. Labels indicate K_2P13.1(THIK-1) EIC contact residues (orange) and Asn ring (green). Red ‡ indicates S136P mutation site. e, PUFA site showing hydrogen bond and salt bridge interactions (dashed lines) and van der Waals contacts ≤ 5 Å. Colors indicate residues from the M1 (teal), M2 (magenta), P1 (salmon), and M4 (violet) helices. EIC is yelloworange and black. f, P1 sequence comparison. Labels indicate K_2P13.1(THIK-1) EIC contact residues (orange) and K_2P2.1(TREK-1) K_2P modulator pocket residues that contact the ML335 activator (light blue)²⁴. Black asterisk denotes Arg sandwich residue. g, M1 sequence comparison. Labels indicate K_2P13.1(THIK-1) EIC contact residues (orange). h, Superposition of the SF1-M2 regions of K_2P13.1(THIK-1) (salmon), K_2P13.1(THIK-1) S136P (cyan), and K_2P13.1(THIK-1) S136P/Y273A (magenta) with K_2P2.1(TREK-1): ML335 (light blue) (PDB:6CQ6)²⁴, K_2P10.1(TREK-2):norfluoxetine (Nfx) (hot pink) (PDB:4XDK)²⁹, K_2P5.1(TASK-2) pH 6.5 (orange) (PDB:6WLV)³² and pH 8.5 (teal) (PDB:6WM0)³², K_2P1.1(TWIK-1) (deep olive) (PDB:3UKM)³⁰, pH 7.4 (7SK0)³¹, and pH 5.5 (7SK1)³¹, and K_2P3.1(TASK-1) (grey) (PDB:6RV2)³³. Cyan lines show the range of M2 movements indicated by the 15° double headed arc. Potassium ions are shown as purple spheres. GenBank Sequences in c, d, f, and g are for human: K_2P13.1(THIK-1) 16306555; K_2P13.2(THIK-2) 11545761; K_2P2.1(TREK-1) 14589851; K_2P10.1(TREK-2) 20143944; K_2P4.1(TRAAK) 15718767; K_2P3.1(TASK-1) 4504849; K_2P9.1 (TASK-3) 542133161; K_2P15.1(TASK-5) 333440483; K_2P5.1(TASK-2) 333440483; K_2P16.1(TALK-1) 14149764; K_2P17.1(TALK-2) 17025230; K_2P1.1(TWIK-1) 4504847; K_2P6.1(TWIK-2) 4758624; and K_2P18.1(TRESK) 32469495.

Extended Data Fig. 6 ∣. THIK1_EM electrostatic surface potentials.

a, Electrostatic surface potentials calculated using APBS¹⁰⁵. Side, cytoplasmic, and extracellular views are shown. b, Slice through the center of K_2P13.1_EM showing the inner cavity. Selectivity filter ions are shown as purple spheres. Location of Tyr273 is indicated. One subunit is shown as transparent. Select channel elements are labeled. Grey bars denote the membrane.

Extended Data Fig. 7 ∣. Structural analysis of THIK1_EM in nanodiscs and detergent micelles.

Superposition of THIK1_EM:DDM:CHS:GDN, THIK1_EM:Det (orange) and THIK1_EM:Nanodisc, THIK1_EM:ND (deep teal) structures showing a, side and b, cytoplasmic views. c, Comparison of THIK1_EM:Det (orange) and THIK1_EM:ND (deep teal) flow restrictor regions. Linoleic acid (EIC) is shown in space filling. d, THIK1_EM:DDM:CHS:GDN (deep teal and salmon) selectivity filter and flow restrictor densities (9σ). Select residues are labeled. Potassium ions are purple spheres. e, Comparison of modeled lipid positions for THIK1_EM:Det (orange) THIK1_EM:ND (deep teal). Common lipid positions are shown in space filling. Lipid belt, inner site lipid, and PUFA are indicated. Select channel elements are labelled. f, Extracellular view of ‘e’. Circle indicates lipid belt boundary. Potassium ions shown as purple spheres.

Extended Data Fig. 8 ∣. Structural comparison of THIK1_EM S136P in nanodiscs and detergent micelles.

Superposition of THIK1_EM S136P:DDM:CHS:GDN, THIK1_EM S136P:Det (deep blue) and THIK1_EM S136P:Nanodisc, THIK1_EM S136P:ND (cyan) structures showing a, side and b, cytoplasmic views. c, Comparison of THIK1_EM S136P:ND (cyan) and THIK1_EM S136P:Det (deep blue) flow restrictor regions. Distances between Tyr273 hydroxyl oxygens are indicated. Linoleic acid (EIC) is shown in space filling. d, and e, Selectivity filter and flow restrictor densities (7-7.5σ) for d, THIK1_EM S136P:ND (cyan) and e, THIK1_EM S136P:Det. Select residues are labeled.

Extended Data Fig. 9 ∣. THIK1_EM S136P;Y273A:ND structural comparisons.

a, and b, Superposition of THIK1_EM S136P;Y273A:ND (magenta), THIK1_EM S136P:ND (cyan) and THIK1_EM:ND (salmon) a, side and b, cytoplasmic views. c, Comparison of THIK1_EM S136P;Y273A:ND (magenta), THIK1_EM S136P:ND (cyan). Distance between Ile139 closest approach is indicated. d, Comparison of THIK1_EM S136P:ND (cyan) and THIK1_EM S136P;Y273A:ND (magenta) flow restrictor regions. Distances between Ala273 methyl groups and Tyr273 hydroxyls are indicated. EIC is shown in space filling. e, Selectivity filter and flow restrictor densities (8σ) for THIK1_EM S136P;Y273A:ND (magenta). Select residues are labeled. Potassium ions are shown as purple spheres.

Extended Data Fig. 10 ∣. K_2P13.1(THIK-1) PUFA site simulations show binding plasticity.

a, Cα RMSD of residues lining the lipid binding pocket for each PUFA tested. b, PUFA RMSD with respect to the initial build (Cα residues 97-110 and 254-270). c-f, representative PUFA site structures from the distributions in ‘a’ for: c, EIC (yellow), d, α-LNL (firebrick), e, DGLA (orange), and f, AA (purple). In c-f Distances for AA carboxylate and Arg258 are indicated by the arrows. Starting structures are shown as transparent. Lipid carboxylate displacements are indicated by the red arrows. g-i, Observed distance distributions for: g, PUFA_C1-R92_CZ, h, PUFA_C1-R92_CZ, and i, R92_CZ-R258_CZ atoms.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41594-024-01476-3.

Data availability

Coordinates and maps were deposited to the PDB and EM Data Bank under the following accession numbers: THIK1 ND, 9BSN and EMD-44870; THIK1 detergent, 9BYI and EMD-45034; THIK1-S136P ND, 9C09 and EMD-45077; THIK1-S136P detergent, 9C07 and EMD-45075; THIK1-S136P;Y273A ND, 9BWS and EMD-44978. The human THIK1 sequence was obtained from UniProt (Q9HB14). Data or materials will be provided on request from the corresponding author. Source data are provided with this paper.