Structures of TRPM5 channel elucidate mechanisms of activation and inhibition

Abstract

The Ca²⁺-activated TRPM5 channel plays essential roles in taste perception and insulin secretion. However, the mechanism by which Ca²⁺ regulates TRPM5 activity remains elusive. We report cryo-EM structures of the zebrafish TRPM5 in an apo closed state, a Ca²⁺-bound open state, and an antagonist-bound inhibited state. We defined two novel ligand binding sites: a Ca²⁺ site (Ca_ICD) in the intracellular domain (ICD), and an antagonist site in the transmembrane domain (TMD). The Ca_ICD site is unique to TRPM5 and has two roles: modulating the voltage dependence and promoting Ca²⁺ binding to the Ca_TMD site that is conserved throughout TRPM channels. Conformational changes initialized from both Ca²⁺ sites cooperatively open the ion-conducting pore. The antagonist NDNA wedges into the space between the S1-S4 domain and pore domain, stabilizing the TMD in an apo-like closed state. Our results lay the foundation for understanding the voltage-dependent TRPM channels and developing new therapeutic agents.

Introduction

Taste perception is one of the fundamental chemosensations in mammals, detecting the availability and the quality of food by converting the signal from tastants into electrical signals that the brain can interpret. Highly expressed in type II taste bud cells, the TRPM5 channel has been considered a key player in sensing sweet, umami, and bitter stimuli^1,2. TRPM5 is activated upon the elevation of cytoplasmic Ca²⁺ concentration that is caused by the binding of tastants to the taste receptors³. Activated TRPM5 then depolarizes the membrane and causes the CALHM1 channel to release the neurotransmitter ATP, which binds to the downstream P2X receptors that trigger the action potential of gustatory neurons, thus conveying taste information to the brain^3,4. TRPM5 also participates in other physiological processes in diverse cell types in a similar manner. For example, it is involved in insulin secretion by pancreatic beta cells^5,6 and in the immune response of tuft cells⁷. Thus, TRPM5 has broad implications for metabolic syndromes and immune disorders, and it is a potential drug target for the treatment of metabolic disorders such as obesity and type 2 diabetes⁸.

The transient receptor potential superfamily, melastatin subfamily (TRPM) consists of eight family members (TRPM1-8) that have diverse functional properties⁹. While most TRP family members are nonselective cation channels that are permeable to Na⁺ and Ca²⁺, TRPM5 and TRPM4 are the only two that are monovalent cation–selective and impermeable to Ca²⁺ (Ref ^10–12). Moreover, TRPM5 and TRPM4 share substantial sequence similarity, and both are activated by intracellular Ca²⁺ in a voltage- and temperature-dependent manner; therefore, they have been classified as close homologs¹³. However, their biophysical properties vary with respect to Ca²⁺ sensitivity and ligand specificity, and the molecular basis for these differences is unknown¹⁴. For instance, TRPM5 is roughly 20-fold more sensitive to Ca²⁺ than TRPM4¹⁴. TRPM4 is inhibited by ATP, and its voltage dependence is modulated by decavanadate, while TRPM5 is insensitive to these ligands^14,15. By contrast, TRPM5, but not TRPM4, is modulated by the sweetener stevioside¹⁶. TRPM5 has also been an attractive pharmaceutical target for treating metabolic syndromes. For example, a family of small-molecule TRPM5 inhibitors, including N’-(3,4-dimethoxybenzylidene)-2-(naphthalen-1-yl)acetohydrazide (NDNA), has been invented. Of note, these inhibitors have been claimed to treat type II diabetes by enhancing insulin release and GLP-1 release¹⁷, which is contrary to the result of TRPM5 KO mice experiments^6,18. Recently, structures of the voltage-dependent TRPM4 and TRPM8 channels have been reported^19–24, but none have been captured in an active open state, preventing a detailed understanding of their gating mechanisms. To understand the molecular mechanisms by which Ca²⁺ activates and antagonist NDNA inhibits TRPM5, we performed electrophysiological and structural studies on TRPM5.

Results

Channel function, overall structures, and the ion-conducting pore.

The zebrafish TRPM5 is highly sensitive to Ca²⁺, and 1 μM Ca²⁺ elicited a robust outward rectifying current in an excised inside-out patch (Fig. 1a, Supplementary Fig. 6a). Interestingly, the outward rectification of the TRPM5 currents apparently depends on the Ca²⁺ concentration. That is, at low Ca²⁺ concentration, the activation of TRPM5 requires membrane depolarization, whereas, at high Ca²⁺ concentration, TRPM5 becomes markedly less voltage-dependent, having a nearly linear current–voltage relation (Fig. 1a–c). This unique property is conserved in human TRPM5 but not in its closest homolog TRPM4 (Extended Data Fig. 1a, g, j, p, s–u)²⁰. Moreover, this change in voltage dependence was previously observed with native TRPM5 currents in taste cells²⁵. Our data imply that besides being an agonist, Ca²⁺ may also serve as a modulator to tune the voltage dependence of TRPM5.

Figure 1: Zebrafish TRPM5 current.

a, Representative traces of Ca²⁺-activated currents from membrane patches excised from tsA201 cells overexpressing zebrafish TRPM5 recorded in the inside-out patch-clamp configuration. Ca²⁺ of 1, 30, 100, and 1000 μM were superfused and voltage clamps were imposed from +200 mV to −200 mV with a final tail current pulse at −140 mV. Background currents were subtracted by interleaved measurements with a calcium-free solution. b, c, Current amplitudes (at 50 ms) of experiments in (a) were plotted as a function of clamp voltage for unnormalized (b) and +200 mV normalized current values (c). Replicates consists of: 1 μM Ca²⁺ [n = 12 patches], 30 μM [n = 6], 100 μM [n = 4], 1000 μM [n = 3] from 12 transfections. Error bars represent SEM. Tail current analysis was also performed (see Supplementary Figure 6). Source data for b and c are available online.

We solved the structures of zebrafish TRPM5 in both glyco-diosgenin (GDN) and lipid nanodiscs (Extended Data Figs. 2–3; Table 1; Supplementary Figs. 1–4); the structures were virtually indistinguishable (Supplementary Figs. 4g). We focused on the TRPM5 in GDN because it produced cryo-EM maps of higher resolution. The apo, Ca²⁺-bound, and (Ca²⁺, NDNA)-bound structures were determined in the presence of 1 mM EDTA, 5 mM Ca²⁺, and 0.5 mM NDNA plus 5 mM Ca²⁺ and had estimated resolutions of 2.9, 2.3, and 2.8 Å, respectively (Extended Data Figs. 2b, 3b, 3e, 3i and Table 1). The maps were of excellent quality (Fig. 2a; Extended Data Fig. 4), which allowed us to de novo model nearly the entire protein (Fig. 2b), to identify two bound Ca²⁺ ions and an NDNA molecule in each subunit (Fig. 2c), two water molecules coordinating the Ca²⁺ in the transmembrane domain (TMD) and two water molecule in the pore loop region (Extended Data Fig. 5a–c), and to unambiguously define the channel gate and the selectivity filter (Extended Data Fig. 5c–g).

Table 1.

Cryo-EM data collection, refinement and validation statistics

	Apo–TRPM5 (EMD-23740, PDB 7MBP)	Ca2+–TRPM5 (EMD-23741, PDB 7MBQ)	Apo–TRPM5 (nanodisc) (EMD-23742)	Ca2+–TRPM5 (nanodisc) (EMD-23743)	Apo–TRPM5(E337A) (EMD-23746, PDB 7MBT)	Ca2+–TRPM5(E337A) (EMD-23747, PDB 7MBU)	(Ca2+, NDNA)–TRPM5 (EMD-23748, PDB 7MBV)	Apo–TRPM5(6μM Ca2+) (EMD-23744, PDB 7MBR)	Ca2+–TRPM5(6μM Ca2+) (EMD-23745, PDB 7MBS)
Data collection and processing
Magnification	130,000	105,000	130,000	130,000	105,000	105,000	105,000	130,000	130,000
Voltage (kV)	300	300	300	300	300	300	300	300	300
Electron exposure (e−/Å2)	49.6	47	49.6	49.6	47	47	47	49.6	49.6
Defocus range (μm)	−0.9 – −1.9	−0.9 – −1.9	−0.9 – −1.9	−0.9 – −1.9	−0.9 – −1.9	−0.9 – −1.9	−0.9 – −1.9	−0.9 – −1.9	1–2
Pixel size (Å)	1.042	0.826	1.042	1.042	0.826	0.826	0.826	1.042	1.00
Symmetry imposed	C4	C4	C4	C4	C1 for single subunit; C4 for tetramer	C1 for single subunit; C4 for tetramer	C4	C4	C4
Initial particle images (no.)	750k	~2,300k	~203k	~794k	1,600k	1,600k	~1,400k	~637k	~637k
Final particle images (no.)	84k	292k	13k	44k	208k	139k	109k	32k	23k
Map resolution (Å)	2.84	2.34	3.6	3.06	2.92	2.97	2.83	3.51	3.47
FSC threshold	0.143	0.143	0.143	0.143	0.143	0.143	0.143	0.143	0.143
Map resolution range (Å)	2.84–246.2	2.34–246.2	3.6–246.2	3.06–246.2	2.92–246.2	2.97–246.2	2.83–246.2	3.51–246.2	3.47–246.2
Refinement
Initial model used (PDB code)	de novo	de novo			de novo	de novo	de novo	7MBP	7MBQ
Model resolution (Å)	3.14	2.51			3.24	3.22	3.28	4.14	4.11
FSC threshold	0.5	0.5			0.5	0.5	0.5	0.5	0.5
Model resolution range (Å)	2.84–246.2	2.34–246.2			2.92–246.2	2.97–246.2	2.83–246.2	3.51–246.2	3.47–246.2
Map sharpening B factor (Å2)	−70.22	−43.99			−70.89	−70.72	−75.36	−117.99	−104.77
Model composition
Nonhydrogen atoms	30,710	30,618			30,762	30,782	30,806	30,710	30,618
Protein residues	3,984	3,984			39,84	3,984	3,984	3,984	3,984
Ligands	12	20			12	16	24	12	20
B factors (Å2)
Protein	NA	NA			NA	NA	NA	NA	NA
Ligand	NA	NA			NA	NA	NA	NA	NA
R.m.s. deviations
Bond lengths (Å)	0.186	0.19			0.189	0.190	0.286	0.186	0.19
Bond angles (°)	1.039	1.053			1.041	1.080	1.46	1.039	1.053
Validation
MolProbity score	1.51	1.61			1.72	1.85	1.45	1.51	1.61
Clashscore	6.32	6.23			9.33	11.6	6.90	6.32	6.23
Poor rotamers (%)	0.00	0.00			0.00	0.00	0.00	0.00	0.00
Ramachandran plot
Favored (%)	97.06	96.10			96.55	95.99	97.69	97.06	96.10
Allowed (%)	2.94	3.90			3.45	4.01	2.31	2.94	3.90
Disallowed (%)	0.00	0.00			0.00	0.00	0.00	0.00	0.00

Figure 2: The overall architecture.

a, The cryo-EM map of Ca²⁺–TRPM5 viewed parallel to the membrane. One subunit is highlighted in red. The unsharpened reconstruction is shown as transparent envelope in. b, The atomic model of Ca²⁺–TRPM5 condition in the same view as (a). The Ca²⁺ ions are shown as green spheres. c, The atomic model of (Ca²⁺, NDNA)–TRPM5 viewed parallel to the membrane. One subunit is highlight in cyan. The NDNA molecule is colored in orange. d–f, The ion-conducting pore in apo–TRPM5 (d), Ca²⁺–TRPM5 (e), and (Ca²⁺, NDNA)–TRPM5 (f) viewed parallel to the membrane. Purple, green, and red spheres define radii of >2.3, 1.2–2.3, and <1.2 Å, respectively. The pore region (shown in cartoon), residues (shown in sticks) forming the gate, and the selectivity filter in two subunits are depicted. Lower right panel: the channel gate viewed from the intracellular side; the distance between the Cα atoms of adjacent I966 residues is labeled. Upper right box: a cartoon representing two subunits of the apo state. The unoccupied/occupied Ca²⁺ sites are shown as unfilled/filled circles, respectively. The membrane area is shown as a gray background. g, Plot of pore radius along the pore axis.

The tetrameric TRPM5 is assembled with a TMD formed by six transmembrane helices and four characteristic intracellular melastatin homology regions (MHR1/2 and MHR3/4) (Extended Data Fig. 2d–e). Despite being the closest homolog to TRPM4, TRPM5 has a distinct monomeric structure, with the MHR1/2 domain tilting toward the TMD, resulting in a more compact tetrameric assembly and a different intersubunit interface (Extended Data Fig. 6a). Besides the conserved Ca²⁺ site in the TMD (Ca_TMD), as observed in the structures of TRPM2, TRPM4, and TRPM8, we observed a novel Ca²⁺ binding site (Ca_ICD) in the intracellular cytosolic domain at the interface between MHR1/2 and MHR3/4 domains (Fig. 2b).

The most remarkable difference between the apo and Ca²⁺-bound structures was at the ion-conducting pore (Fig. 2d–e, g). The intracellular half of the pore, which constitutes the channel gate, is restricted by I966 in the apo structure, giving a smallest radius of 0.8 Å (Fig. 2d, g); this represents an apo (resting) closed state (apo–TRPM5). By contrast, the Ca²⁺-bound structure has an enlarged pore with a smallest radius of 2.7 Å (Fig. 2e, g), which allows the passage of partially dehydrated monovalent cations, thus representing an agonist-bound active open state (Ca²⁺–TRPM5).

The extracellular half of the pore is confined by the pore loop, which shows little conformational change during channel gating and is generally considered to be responsible for ionic selectivity (Fig. 2d–e; Extended Data Fig. 6b). Here, we identified two ordered water molecules in each subunit (Extended Data Fig. 5c, e–f). Notably, the water molecules form a tight oxygen ring along with the backbone oxygen atoms of G905, constituting the narrowest site in the pore loop region (Fig. 2d). Here, the pore radius is approximately 2.5 Å, about the Na⁺–O distance in a six-coordinate hydrated sodium (2.4 Å)²⁶. We suggest this oxygen ring acts as the selectivity filter, and the four water molecules provide a favorable hydration layer for sodium ions to permeate²⁷. A similar selectivity filter likely also exists in TRPM4, given by the conserved sequences (Supplementary Fig. 5) and structures of their pore loop (Extended Data Fig. 6b). The replacement of Q977 (equivalent to Q906 in zebrafish TRPM5) by a glutamate gives human TRPM4 a moderate permeability to Ca²⁺ (Ref ²⁸). A glutamate at this position may attract Ca²⁺ ions by creating a binding site near the selectivity filter²⁹. In addition, the glutamate might no longer form the same hydrogen bonding network, resulting in an altered conformation and size of the selectivity filter.

In the presence of NDNA and Ca²⁺, we observe a well-defined density in each subunit wedging into a cleft between the S1-S4 domain and the pore domain of the TMD (Fig. 2c). This density unambiguously hews to the shape of an NDNA molecule (Extended Data Fig. 7a–b). Of note, this binding site is known as the “vanilloid binding site” in the structures of TRPV and TRPA channels^30–32, but has not been reported in members of the TRPM family. Despite Ca²⁺ binding to both Ca_TMD and Ca_ICD, the TMD of (Ca²⁺, NDNA)–TRPM5 shows an apo-like conformation with a closed pore (Fig. 2c, f–g). We thus define it as an antagonist-bound inhibited state.

Two calcium binding sites.

The Ca_TMD site is located within the S1-S4 domain and is surrounded by four amino acids and two water molecules in an octahedral geometry (Fig. 3a). The key residues coordinating the Ca_TMD are absolutely conserved across the TRPM family members (Fig. 3c upper panel). Their replacement by an alanine largely abolished Ca²⁺-invoked currents (Extended Data Fig. 8), which indicates that binding of Ca²⁺ to Ca_TMD is indispensable for the activation of zebrafish TRPM5. The integrity of Ca_TMD site is also important for the activity of rat TRPM5 and other Ca²⁺-dependent TRPM channels^33,34. We propose that the octahedral geometry of the Ca_TMD site is likely conserved among Ca²⁺-sensitive TRPM channels, in light of the high sequence conservation of the Ca_TMD site and the similar spatial organization of the Ca²⁺-coordinating residues (Fig. 3c; Extended Data Fig. 6c).

Figure 3: Ca²⁺ binding sites.

a, b, The Ca_TMD (a) and Ca_ICD (b) sites. Ca²⁺ is shown as a green sphere. The coordinating residues and water molecules are shown in sticks and spheres, respectively. Polar interactions are indicated by yellow bars. c, Sequence alignment of the Ca_TMD (top) and Ca_ICD site (bottom) among zebrafish TRPM5 (drM5), human TRPM5 (hsM5), human TRPM4 (hsM4), human TRPM2 (hsM2), and human TRPM8 (hsM8). The Ca²⁺ coordinating residues in zebrafish TRPM5 are indicated by asterisks, and conserved coordinating residues are in red. The residue numbers are according to zebrafish TRPM5 (UniProtID: S5UH55). Sequence segments are separated by vertical bars. d, Remodeling of the Ca_ICD site upon Ca²⁺ binding. Apo–TRPM5 and Ca²⁺–TRPM5 are in blue and red, respectively. Black arrows indicate the movement of E212 and D336. The MHR1/2 domains in apo–TRPM5 and Ca²⁺–TRPM5 are represented by blue and red surfaces, respectively, showing the movement of MHR1/2. e, Voltage-clamped current amplitudes were measured (as performed in Fig. 1) from inside-out patches pulled from tsA cells overexpressing E337A mutant channels. The number of patches are (1 μM Ca²⁺ [n = 7 patches], 30 μM [n = 7], 100 μM [n = 5], 1000 μM [n = 5] from 8 transfections). Error bars represent SEM. See Extended Data Fig. 1b for representative traces. f, The superimposition of Ca_ICD site in TRPM5 (red) and TRPM4 (yellow, PDBID: 6BQR)²¹, by aligning the helix α11 and its equivalent in human TRPM4 (residues 396–403). The coordinating residues are shown as sticks. Equivalent structural elements and residues in TRPM4 are labeled with a prime symbol. The orientation of helix α12 and its equivalent (α12’) in human TRPM4 (residues 372–386) are indicated by colored 3D arrows. The differences between α11 and α11’, and between E337 and E396’, are indicated by black arrows. Source data for e are available online.

The newly defined Ca_ICD site is at the interface between the MHR1/2 and MHR3/4 domains (Fig. 2b), in a negatively charged pocket. Part of the pocket is formed by a twisted helical segment, α12, of the MHR3 domain (Fig. 3b). The unique conformation of α12 allows the side chains of D336 and E337 on one side of the twist, and backbone oxygen of D333 on the other side, to face toward each other, thus creating a binding pocket to accommodate Ca²⁺.

To investigate the mechanism underlying Ca_ICD binding, we carried out structural comparisons and electrophysiological experiments. The overlay of Ca²⁺–TRPM5 and apo–TRPM5 structures revealed that E337, C324, and the backbone oxygen of D333 form a rigid notch surrounding the binding site and show little movement upon Ca²⁺ binding; by contrast, D336 and E212 act as a flexible cap to enclose the Ca_ICD site (Fig. 3d). Specifically, E212 from MHR1/2 approaches the Ca_ICD site by moving approximately 3 Å, thus pulling the MHR1/2 domain closer to the MHR3 domain; meanwhile, the side-chain of D336 flips approximately 90° toward the Ca²⁺. We propose that E337, the only negatively charged residue in the rigid notch, plays a key role in the binding of Ca²⁺. Indeed, replacement of E337 with an alanine (E337A) rendered TRPM5 voltage-sensitive even at high Ca²⁺ concentration (up to 1000 μM), distinct from the wild-type TRPM5, which becomes nearly voltage-independent at high Ca²⁺ concentration (Figs. 1a–c, 3e; Extended Data Fig. 1b, k, s–u; Supplementary Fig. 6a–b). Mutations of other coordinating residues to alanine only moderately altered the voltage sensitivity (Extended Data Fig. 1c–f, l–o, s–u; Supplementary Fig. 6c–f).

The Ca_ICD site is unique for TRPM5 because the residues coordinating Ca_ICD are conserved among TRPM5 orthologues, but not in other TRPM channels, except for TRPM4 (Fig. 3c lower panel). To understand why a site like the TRPM5 Ca_ICD was not observed in the published TRPM4 structures^19–22 despite the conserved sequence, we compared their intracellular domains (ICDs). We found that the key residues in TRPM4 are not close enough to each other to form a binding site because of two major structural differences. First, the structural element in TRPM4, which corresponds to the twisted helical segment α12 in TRPM5, is an intact α-helix, so that E396 (equivalent to E337 in TRPM5) cannot face the backbone oxygen of A392 (equivalent to D333 in TRPM5) (Fig. 3f). Second, the interface of MHR1/2 and MHR3, where the Ca_ICD site is located, has a markedly different arrangement than in TRPM5, manifested by different angles between helices α11 (on MHR1/2) and α12 (on MHR3) (Fig. 3f).

Antagonist binding site.

The antagonist NDNA is highly potent and inhibits Ca²⁺- induced TRPM5 currents with an IC₅₀ of approximately 2.4 nM (Fig. 4a–b). The molecular structure of NDNA consists of two rings, a naphtalen and a dimethoxybenzylidene, which are linked by an acetohydrazide group (Extended Data Fig. 7a). In the (Ca²⁺, NDNA)–TRPM5 structure, the NDNA molecule is located at the interface between the S1-S4 domain and the pore domain (S5 and S6), near the Ca_TMD site (Fig. 4c). The two rings of NDNA are perpendicular to each other, forming a wedge shape. The naphtalen ring forms the base of the wedge, bracing on the S3 helix; the dimethoxybenzylidene ring forms the tip that penetrates through the cleft between S4 and S5, pressing against the pore domain of the adjacent subunit (Fig. 4d, e). The interaction between NDNA and TRPM5 is further enhanced by a hydrogen bond between the acetohydrazide linker and E853 on S5, and the proximity of the same linker to W793 on S3 (Fig. 4e). Within the binding site, while most residues preserve their conformations in the apo state, the side chain of W869 flips to accommodate NDNA, forming a hydrogen bond with one of the methoxyl moieties on the dimethoxybenzylidene ring (Extended Data Fig. 7c).

Figure 4: Effect and binding site of antagonist NDNA.

a, Voltage-clamped (+200 mV to −200 mV) calcium activated whole-cell currents from tsA201 cells over-expressing zebrafish WT TRPM5 were suppressed upon super-fusion of 10 μM NDNA. b, IC₅₀ of NDNA, 2.4 nM, was determined by plotting (I_{+200 mV}, _NDNA / I_{+200 mV}, _bath) using various NDNA concentrations (1 fM, 10 pM, 100 pM, 1 nM, 100 nM, 0.5 μM, 10 μM). Concentration is log (M). Each point represents the mean, and bars indicate SEM. Number of cells is indicated in brackets. From non-linear fitting, the Hill Slope is −0.5, and the 95% CI is 0.5 – 23 nM. c, The pore domain of (Ca²⁺, NDNA)–TRPM5 viewed from the extracellular side. The four bound NDNA molecule is shown in orange. Transmembrane helices surrounding a copy of NDNA is labeled. Prime symbol indicates the adjacent subunit. Ca_TMD is shown in green sphere. d and e, Two close-up views for the detailed interactions mediated by the NDNA. One TRPM5 subunit is colored in cyan, whereas the adjacent subunit is colored in light cyan. Polar interactions between NDNA and residues are indicated by black lines. g, h, Ratio of whole-cell current amplitudes in the presence and absence of NDNA (10 μM) for various TRPM5 mutants. Each point represents a single cell and bars denote mean value. The number of cells analyzed were: tsA (n = 3 cells, 2 transfections), WT (n = 4, 4 transfections), C796A (n = 3 cells, 3 transfections), I849A (n = 3 cells, 1 transfection), E853A (n = 5 cells, 4 transfections), I836A (n = 5 cells, 5 transfections), W793A (n = 3 cells, 2 transfections), V852A (n = 3 cells, 2 transfections), L833A (n = 6 cells, 4 transfections), W869A (n = 3 cells, 2 transfections). Source data for b, f and g are available online.

NDNA binds to a region critical for channel gating. To investigate how alterations in the binding site affect the response of TRPM5 to NDNA, we performed mutagenesis and electrophysiology studies. NDNA failed to suppress the current of mutants E853A, I836A, and W793A, but inhibited the currents of C796A and I849A (Fig. 4f–g; Extended Data Fig. 9). Interestingly, NDNA seemed to potentiate the current of mutants L833A and W793A (Fig. 4f–g; Extended Data Fig. 9). While the mechanism by which the NDNA potentiated these mutants is unclear, the specific NDNA binding mode appears to be critical, and even subtle variations could lead to opposite effects on channel function. This is reminiscent of previous studies of TRPV and TRPA1 channels in which a number of small-molecule compounds, including both agonists and antagonists, bind to a similar location as NDNA^30–32,35. Mutants L833A, W869A, and V852A showed hardly any basal currents precluding evaluation of NDNA inhibition (Extended Data Fig. 9). Overall, most of the mutants altered the sensitivity of TRPM5 to NDNA, thus supporting that the binding site is critical for NDNA-mediated channel inhibition.

In the (Ca²⁺, NDNA)–TRPM5 structure, although both Ca_TMD and Ca_ICD sites are occupied by Ca²⁺, we observed major differences relative to the Ca²⁺–TRPM5 open state. First, the ICD showed the same trend of motion relative to the apo–TRPM5 structure but to a lesser extent (Extended Data Fig. 7d). Second, the S1-S4 domain retained apo-like conformation and did not show marked conformational changes as observed in the Ca²⁺-bound open state, resulting in a different coordination of Ca_TMD, with Q771 on the S2 helix not involved in the binding (Extended Data Fig. 7e, g). Lastly, the pore domain is closed, similar to the apo state (Extended Data Fig. 7h–i; Fig. 2f–g). Together, our data suggest that NDNA inhibits Ca²⁺-induced TRPM5 activation in a non-competitive manner. Despite Ca²⁺ binding, NDNA limits the movement of the S1-S4 domain and the pore domain, thus stabilizing the TMD in an apo-like closed state.

The two roles of the Ca_ICD site.

The alanine mutants of the key residues in the Ca_ICD site had the same Ca²⁺-induced channel activation as the wild type, but they shifted the voltage dependence toward a positive membrane potential to different degrees, with E337A having the strongest phenotype (Fig. 3e; Extended Data Fig. 1; Supplementary Fig. 6). E337A remained voltage-dependent regardless of Ca²⁺ concentration, which was in sharp contrast to the wild type, in which current changed from voltage-dependent at low Ca²⁺ concentration to nearly voltage-independent at high Ca²⁺ concentration. We further looked into the same mutant on human TRPM5 (E351A) and observed the same phenotype (Extended Data Fig. 1g, h, p, q, s–u).

These results indicate that Ca_ICD modulates the voltage dependence of TRPM5. To ground our interpretations of the electrophysiological data, we determined the structure of E337A in the presence of 5 mM Ca²⁺ at 2.9 Å resolution (Fig. 5a; Extended Data Fig. 10; Supplementary Fig. 7). As expected, the ICD showed a wild type apo-like conformation and the Ca_ICD site was unoccupied (Fig. 5b, c). This supports the idea that the replacement of E337 by an alanine indeed impaired Ca²⁺ binding to the Ca_ICD site and that the altered voltage dependence of E337A was caused by abolished Ca²⁺ binding to the Ca_ICD site.

Figure 5: The structures of the Ca_ICD-deficient mutant E337A.

a, The upper panel shows the consensus map obtained from the Ca²⁺–TRPM5(E337A) data. The cartoons in the lower panel represent the two conformations that have distinct occupancies at the Ca_TMD site, obtained by single subunit analysis of the same data: apo–TRPM5(E337A) in magenta and Ca²⁺–TRPM5(E337A) in cyan. The unoccupied Ca²⁺ sites are shown as unfilled circles; occupied Ca²⁺ sites are shown as green circles. The cell membrane is represented in gray. b–e, the Ca_ICD site (b) and Ca_TMD site (d) in apo–TRPM5(E337A), and the Ca_ICD site (c), and Ca_TMD site (e) in Ca²⁺–TRPM5(E337A). The cryo-EM densities are shown in black mesh. The Ca²⁺ density is shown as a green sphere. Unoccupied sites are indicated by a dashed gray circle. f, The superimpositions of the S1-S4 domain of the apo–TRPM5(E337A) (magenta) and the Ca²⁺–TRPM5(E337A) (cyan) structures. g, The superimpositions of the S1-S4 domain of the apo–TRPM5(E337A) (magenta) and the apo–TRPM5 (blue) structures. h, The superimpositions of the S1-S4 domain of the Ca²⁺–TRPM5(E337A) (cyan) and the Ca²⁺–TRPM5 (red) structures. i, Pore radius plot along the pore axis. j, k, Remodeling of the Ca_ICD site upon Ca²⁺ binding in TRPM5 (j) and TRPM4 (k). Apo–TRPM5 and Ca²⁺–TRPM5 are in blue and red, respectively (j). Apo–TRPM4 (PDBID: 6BQR) and Ca²⁺–TRPM4 (PDBID: 6BQV) are in gray and yellow, respectively²¹. (k). The Cα atoms of key residues and the Ca²⁺ are shown as spheres. Shown in parentheses are the distances of the root-mean-square-deviation (RMSD) between S2 (residues 767–772 in TRPM5 and 827–832 in TRPM4) and S3 (residues 793–798 in TRPM5 and 864–869 in TRPM4), and the distances of the Cα movements of E994 in TRPM5 and E1068 in TRPM4.

Although our data analysis workflow involved a focused classification of the TMD (Supplementary Fig. 7), part of the TMD is still not unambiguously defined, and the densities for Ca_TMD were markedly weaker than those in the structure of Ca²⁺-bound wild-type TRPM5, indicating the structural heterogeneity of the TMD. We, therefore, performed structural analysis on a single subunit and obtained two conformations that had the same ICD but distinct S1-S4 domains in the TMD (Fig. 5). One had an empty Ca_TMD, termed apo–TRPM5(E337A) (Fig. 5b, d), and the other had an occupied Ca_TMD, termed Ca²⁺–TRPM5(E337A) (Fig. 5c, e). This suggests that an unoccupied Ca_ICD site lowers the binding affinity of Ca²⁺ for the Ca_TMD site. The cooperativity between these two Ca²⁺ binding sites agrees with the observation that upon activation with 1 μM Ca²⁺, the current amplitudes of the E337A mutant at a clamp of +200 mV were substantially smaller (75%) than those of the wild type (Extended Data Fig. 1t).

Within the TMD, a closed ion-conducting pore was observed in the Ca²⁺–TRPM5(E337A) structure (Fig. 5i). This further supports the role of Ca_ICD as a voltage-modulating site, as its absence renders TRPM5(E337A) inactive due to the lack of membrane depolarization under the conditions for structural determination (Fig. 3e). Interestingly, the cytosolic vestibule, the part underneath the channel gate in Ca²⁺–TRPM5(E337A), is similar to that of Ca²⁺–TRPM5 (Fig. 5i), suggesting that the pore in Ca²⁺–TRPM5(E337A) may represent an intermediate state prior to channel opening.

The cooperativity between Ca_TMD and Ca_ICD sites implies that Ca_ICD might be physiologically relevant. To test this hypothesis, we determined the structure of wild-type TRPM5 in the presence of 6 μM Ca²⁺, a concentration similar to the Ca²⁺ EC₅₀ of TRPM5 channels excised from native taste receptor cells (8 μM at −80 mV)²⁵. Interestingly, this condition yielded both the apo conformation and the Ca²⁺-bound open conformation (Extended Data Fig. 3d–g; Supplementary Fig. 3b). The ratio of protein particles belonging to the apo and open conformations, respectively, is 1.4:1. This data thus quantitatively correlates agonist-induced conformational changes to the EC₅₀ determined by excised patch recordings. Furthermore, Ca_TMD and Ca_ICD are clearly occupied in the open state, which indicates that Ca²⁺ at EC₅₀ concentration binds equally well to both sites, thus supporting the physiological relevance of Ca_ICD (Extended Data Fig. 3g).

To understand why the occupation of the Ca_ICD site is required for a high-affinity Ca_TMD site in TRPM5 but not in TRPM4, we compared the conformational changes in their Ca_TMD sites upon binding of Ca²⁺. In TRPM5, the helices S2 and S3—containing the coordinating residues and the TRP helix, which are key elements in transducing signals from the ICD to TMD—undergo substantial movement (Fig. 5j). By contrast, in TRPM4, the Ca_TMD site and the TRP helix mostly showed only minor sidechain rearrangement (Fig. 5k). This difference suggests that the Ca_TMD site in TRPM4 may be primed for Ca²⁺ binding, whereas in TRPM5, a high-affinity Ca_TMD site likely requires extensive rearrangement of the S2 and S3, assisted by Ca²⁺ binding to the Ca_ICD site.

Taken together, our data suggest the Ca_ICD site is physiologically relevant and has two important roles. First, it acts as a voltage modulator that shifts the voltage dependence toward negative potential, reminiscent of the effect of decavanadate and PIP2 on TRPM4^15,36. Second, it promotes Ca²⁺ binding to the Ca_TMD site and facilitates channel activation.

Signal transduction from Ca_ICD to the TMD.

The structural comparison of TRPM5 in the absence versus the presence of Ca²⁺ showed conformational rearrangement throughout the protein, with individual domains mostly showing rigid body movement (Fig. 6a). To understand the respective contributions of the two Ca²⁺ binding sites and how they cooperatively open the channel, we traced the conformational changes from the ICD and the S1-S4 domain to the ion-conducting pore.

Figure 6: The signal transduction from ICD to TMD.

a, The superimposition of apo–TRPM5 (blue) and Ca²⁺–TRPM5 (red) structures by aligning the coiled-coil poles in the C-terminal domain (CTD), viewed parallel to the membrane. One subunit is also shown in cartoon representation. Ca²⁺ is shown as green spheres. b, The superimposition of the MHR1-4 domains of apo–TRPM5 (blue) and Ca²⁺–TRPM5 (red) by aligning the MHR3/4 domain, viewed parallel to the membrane. The rotation of the MHR1/2 relative to the MHR3/4 domain upon Ca²⁺ binding is indicated. The surfaces are outlined in blue for apo–TRPM5 and filled with red for Ca²⁺–TRPM5. c, Superimposition of the MHR3/4 domain and the CTD rib and pole helices of apo–TRPM5 (blue) and Ca²⁺–TRPM5 (red) by aligning the CTD coiled-coil poles, viewed from the intracellular side. One subunit in both structures are shown in blue (apo–TRPM5) or in red (Ca²⁺–TRPM5). The rotation of the rib helices is indicated. d, The superimposition of the ICD–TMD interface of apo–TRPM5 (blue) and Ca²⁺–TRPM5 (red) by aligning the CTD coiled-coil poles (not shown), viewed from the intracellular side. The rotations of helices square_N and square_C are indicated. The green circle highlights the location of the intersubunit interface and ICD–TMD interface, as detailed in panels (e, f). e, f, The conformational rearrangement at the intersubunit interface and the ICD–TMD interface in apo–TRPM5 (e) to Ca²⁺–TRPM5 (f), viewed parallel to the membrane. Prime symbol indicates residues or structural elements from the adjacent subunit. Interactions are shown in yellow bars. The single headed arrows indicate the movement of the square and TRP helices. The double headed arrow indicates the angle between S6 and TRP. The positions of the intersubunit interface are shown by green circles.

Clamped by two lobes, i.e., the MHR1/2 and MHR3/4 domains, the Ca_ICD site underwent an opening of 5.8° upon binding of Ca²⁺ (Fig. 6b). As a result, the rib helix, which penetrates into the interface between the MHR3/4 domain and the adjacent MHR1/2 domain, showed a clockwise rotation of 9.4° as viewed from the intracellular side (Fig. 6c). Meanwhile, just above the rib helix, four helices that form a square shape also rotated clockwise (Fig. 6d). Because these “square” helices mediate the contact between adjacent subunits, their rotation altered the intersubunit interface. Specifically, in the absence of Ca²⁺, the N- and C-termini of adjacent square helices are in close contact (Fig. 6e). Upon binding of Ca²⁺, this interface is disrupted as the adjacent termini rotate away from each other, resulting in a new interface between the N-terminus of the square helix and helix α28, where R552 and D610 form a salt bridge (Fig. 6f). Notably, the square helix in TRPM5 is broken into two short segments in the middle where E560 interacts with the N-terminus of helix α27 (Fig. 6e, f), a feature that is unique to TRPM5. By contrast, all the other TRPM channels have (or are predicted to have) a continuous square helix (Extended Data Fig. 6d–e).

We speculate that the square helix plays a role in the signal transduction from the Ca_ICD site to the TMD because it links the remodeling of the intersubunit interface upon Ca²⁺ binding to the TRP helix—a key element involved in channel gating—through helices α27 and α28. Indeed, the replacement of E560 by an alanine, which presumably weakens the interaction between the square helix and helix α27, led to slower channel activation and deactivation kinetics (Extended Data Fig. 1i, r). Moreover, the R578Q polymorphism in human TRPM5 (which corresponds to position 561 on the square helix in zebrafish TRPM5) has been associated with obesity-related metabolic syndrome³⁷.

Channel opening by synergistic action of Ca_ICD and Ca_TMD.

Accompanying the conformational changes of the ICD induced by Ca²⁺ binding, the TRP helix is pushed toward the TMD where the Ca_TMD binding site is located (Fig. 7a). Here, surrounded by the TRP helix, S4-S5 linker, S2, and S3, is the region where the conformational changes induced by Ca²⁺ binding to the Ca_ICD and Ca_TMD sites meet, and where complex remodeling occurs (Fig. 7a, b). We performed a detailed structural analysis, and propose a mechanism by which the motion of the TRP helix promotes Ca²⁺ binding at the Ca_TMD site, ultimately leading to channel opening.

Figure 7: The channel opening.

a, The superimposition of the TMD of a single subunit in apo–TRPM5 (blue) and Ca²⁺–TRPM5 (red) by aligning their S1-S4 domain, viewed parallel to the membrane. The center-of-mass movement of the pore domain is indicated. b, The superimposition of the pore domain in apo–TRPM5 (blue) and Ca²⁺–TRPM5 (red) by aligning their S1-S4 domain, viewed from the intracellular side. The pore domain of Ca²⁺–TRPM5 is shown in surface representation and the S5-S6 domain of one subunit from each structure is shown as a cartoon. The relative movements of helices S5 and S6 are indicated. c, d, Close-ups of the circled area in (a), viewed from the intracellular side. The remodeling of the Ca_TMD site from apo–TRPM5 (c) to Ca²⁺–TRPM5 (d). The movements of S2, S3, S4, and TRP helices are indicated by arrows. Interactions are shown in yellow bars. e, f, Close-ups of the boxed area in (a). W984 on the TRP helix switches its interaction partner from P847 and G846 in apo–TRPM5 (e) to I841 in Ca²⁺–TRPM5 (f). Interactions are shown in yellow bars. The movement of W984 is indicated. The contact area between the TRP helix and the S4-S5 linker is highlighted in grey. The segment between I836 and I841 turns into a 3₁₀-helix in Ca²⁺–TRPM5. The inset shows the view along the axis of the S4 helix. g, A cartoon scheme of the activation and inhibition mechanism of TRPM5. Conformational changes initialized from both Ca²⁺ sites cooperatively open the ion-conducting pore. The antagonist NDNA wedges into the space between the S1-S4 domain and pore domain, stabilizing the TMD in an apo-like closed state. The movements of individual structural elements are indicated by arrows.

In the absence of Ca²⁺, i.e., when the Ca_ICD site is unoccupied, the Ca_TMD site is in a configuration that is difficult to access by Ca²⁺ because two crucial residues, E768 and D797, are locked by R834 in a triangular hydrogen bond network (Fig. 7c). This conformation is stabilized by the interaction between helices S3 and S4, with W793 and H837 stacking with each other (Fig. 7c). This explains why only a subset of particles from the Ca_ICD site-deficient mutant E337A showed an occupied Ca_TMD site even at high (5 mM) Ca²⁺ concentration (Supplementary Fig. 7).

When the Ca_ICD site is occupied, the TRP helix tilts toward the TMD, leading to three consequences. First, it pushes the S4 helix away from S2 and S3 (Fig. 7c), allowing E768 and D797 to be readily released from R834 to coordinate Ca²⁺ together with Q771, N794, and two water molecules (Fig. 7d). Second, E994 on the TRP helix approaches the Ca_TMD site to coordinate one of the two water molecules, thus helping the Ca_TMD site bind Ca²⁺ (Fig. 7d). Third, Y995 on the TRP helix accommodates the flipped H837 by forming a hydrogen bond, thereby assisting in the decoupling of S4 from S3 by breaking the π-stacking between H837 and W793 (Fig. 7d). The decoupling of S4 from S3 is important, because it allows a relative movement between the S4-S5 linker and W984, a residue on the TRP helix that is absolutely conserved in the TRP superfamily and is crucial for channel gating^38–40. As a result, the W984 switches its interaction partners from P847 and G846 on the N-terminus of S5 to the backbone oxygen of I841on S4, forcing the last turn of the S4 helix and part of the S4-S5 linker to stretch into a 3₁₀-helix (Fig. 7e–f). The movement of W984 breaks the major interaction between the TRP helix and S5, eventually enabling the pore domain (helices S5 and S6) to relocate. Indeed, the structural comparison between the apo and open states of TRPM5 showed a movement of the pore domain by half an α-helical turn toward the extracellular side with an outward expansion, thus opening the ion-conducting pore (Fig. 7a–b).

Discussion

Our TRPM5 structures define two Ca²⁺ binding sites, Ca_ICD and Ca_TMD, which are 70 Å apart. Comparison of the structures in the apo and open states illustrates a molecular mechanism by which Ca²⁺ binding to the two locations synergistically leads to a complex conformational rearrangement at the interface between the ICD and the TMD (Fig. 7g). This rearrangement eventually gives rise to the decoupling between the TRP helix and the S5 helix and the opening of the ion-conducting pore. We conclude that Ca_TMD functions as an orthosteric binding site for channel activation, while Ca_ICD modulates the voltage dependence and the accessibility of Ca_TMD. The molecular basis by which the Ca_ICD site modulates the voltage dependence is still unclear; that will require the identification of the voltage sensor(s) and its working mechanism.

We also defined a novel antagonist binding site in the TMD and elaborated a non-competitive inhibition mechanism by which the antagonist NDNA stabilizes the ion-conducting pore in an apo-like closed conformation (Fig. 7g). Because NDNA is a potent TRPM5-selective antagonist, our work not only will facilitate the characterization of TRPM5 currents in many physiological processes, but is also important for the ongoing development of drugs targeting TRPM5.

TRPM4 and TRPM5 share substantial sequence similarity (55.7% between human TRPM4 and TRPM5, 58.1% between zebrafish TRPM4 and TRPM5), and both depolarize the cell membrane by sensing cytosolic Ca²⁺, but they are involved in different physiological processes. The unique Ca_ICD site endows TRPM5 with a complex gating and modulation mechanism by Ca²⁺, which may link to the physiological roles of TRPM5 in taste signaling and the Ca²⁺ oscillation during insulin secretion by the pancreatic beta cells^5,6. We have elaborated a Ca²⁺-induced gating mechanism of a voltage-sensitive TRPM channel, which differs from that of the voltage-insensitive TRPM2^34,41–43. Our study highlights the important role of the ICD as a ligand-sensing domain in TRPM channels and lays a solid foundation for the development of novel therapeutic drugs that distinguish between TRPM4 and TRPM5.

Methods

TRPM5 expression and purification

Genes encoding full-length human and zebrafish TRPM5 (UniProtKB accession numbers Q9NZQ8, and S5UH5, respectively) were synthesized by Bio Basic and were sub-cloned into a pEG BacMam vector with an His8 tag, GFP, and a thrombin cleavage site at the N terminus⁴⁴. Site-directed mutagenesis is performed by using QuikChange II Site-directed mutagenesis (Qiagen) or Q5 Site-Directed Mutagenesis (NEB) protocol and confirmed via Sanger sequencing (Eurofins). For baculovirus production, each TRPM5 ortholog in a BacMam vector is transformed into DH10Bac cells, followed by P1 and P2 baculovirus generated in Sf9 cells. P2 viruses (8%) were used to infect tsA201 cells grown in Freestyle 293 Expression Medium in suspension culture (ThermoFisher). Infected cells were incubated for an initial 12 h at 37 °C before 10 mM sodium butyrate was added. Cells were then moved to a 30 °C incubator and allowed to grow for another 60 h with vigorous shaking. At 72 h post-infection, cells were harvested by centrifugation at 5000 rpm, 4 °C for 30 min. Cell pellets were washed with buffer containing 150 mM NaCl and 20 mM Tris pH 8.0 (TBS buffer) and stored at −80 °C.

Cell pellets from 200 ml culture were thawed on ice and resuspended in TBS buffer containing 1 mM PMSF, 0.8 μM aprotinin, 2 μg mL⁻¹ leupeptin, 2 mM pepstatin A (which are all protease inhibitors) plus 1% GDN detergent (Anatrace). Protein was extracted from the membrane by whole-cell solubilization for 1 h at 4 °C with rotation. The solubilized protein was incubated with 2 mL TALON cobalt metal-affinity resin (Takara Bio) for 1 h. The TALON resin was then washed with 20 ml TBS buffer supplemented with 0.02% GDN and 15 mM imidazole. Protein was eluted with TBS buffer supplemented with 0.02% GDN and 250 mM imidazole. The eluent was concentrated to 500 μL and further purified by size-exclusion chromatography in TBS buffer containing 0.02% GDN. Peak fractions containing TRPM5 were pooled and concentrated to 4–5 mg/mL for grid freezing.

For nanodisc reconstitution, the eluent after immobilized metal affinity chromatography was mixed with MSP2N2 and soybean lipid extract at a molar ratio of 1:1:200 (TRPM5:MSP2N2:lipid). Three rounds of Bio-Beads (BIO-RAD) incubation at 4 °C was performed to facilitate nanodisc reconstitution. The Bio-Beads were then removed, and the sample was concentrated to 500 μL using an Amicon 100 kDa concentrator (MilliporeSigma). Size-exclusion chromatography was done in TBS buffer to further purify TRPM5-nanodisc complex. Peak fractions of TRPM5-nanodisc were collected and concentrated to 5 mg/mL for freezing grid.

EM sample preparation and data acquisition

Freshly purified TRPM5 protein in GDN detergent was mixed with 1 mM EDTA (apo–TRPM5 and apo–TRPM5(E337A)), 5 mM Ca²⁺ (Ca²⁺–TRPM5 and Ca²⁺–TRPM5(E337A)) or 6 μM Ca²⁺ before grid preparation. For TRPM5-nanodisc sample, we added 0.05 mM digitonin to improve particle distribution on the grid. The apo–TRPM5(nanodisc) condition contains 1 mM EDTA, and the Ca²⁺–TRPM5(nanodisc) contains 1 mM Ca²⁺ and 0.5 mM steviol (Sigma). After mixing with the designated additives, a 2.5 μL aliquot of the sample was applied to a glow-discharged Quantifoil holey carbon grid (gold, 1.2/1.3 μm size/hole space, 300 mesh or gold, 2/1 μm size/hole space, 300 mesh), blotted for 1.5 s at 100% humidity using a Vitrobot Mark III, and then plunge-frozen in liquid ethane cooled by liquid nitrogen. The grids were loaded into a FEI Titan Krios transmission electron microscope operating at 300 kV with a nominal magnification of 130,000× and an energy filter (20 eV slit width). The apo–TRPM5, Ca²⁺–TRPM5(6 μM)(GDN), apo–TRPM5(nanodisc), and Ca²⁺–TRPM5(nanodisc) dataset was recorded by a Gatan K2 Summit direct electron detector in super-resolution mode with a binned pixel size of 0.521 Å. Each K2 movie was dose-fractionated to 40 frames for 8 s with a total dose of 49.6 e⁻/Ų. The Ca²⁺–TRPM5, Ca²⁺–TRPM5(E337A), NDNA/Ca²⁺–TRPM5 datasets were collected by a K3 direct electron detector in super-resolution mode with a binned pixel size of 0.413 Å (K3). Each K3 movie was dose-fractionated to 75 frames for 1.5 s with a total dose of 47 e⁻/Ų. The automated image acquisition was facilitated using SerialEM⁴⁵. The nominal defocus range was set from −0.9 μm to −1.9 μm.

Cryo-EM data analysis procedure

The detailed workflow of data processing procedure is summarized in Supplementary Fig. 2–4. In general, the raw super-resolution tif movie files for each dataset were motion-corrected and 2x binned using MotionCor2 v1.1.0 or RELION 3.0 or RELION3.1 (Ref ^46,47). The per-micrograph defocus values were estimated using Gctf 1.06 or ctffind 4.1.10 (Ref ^48,49). Particle picking was performed using gautomatch v0.56 (https://www2.mrc-lmb.cam.ac.uk/research/locally-developed-software/zhang-software/) or topaz v0.2.4 (Ref ⁵⁰). Junk particles were removed by 2D classification and heterogeneous refinement using CryoSPARC (v0.65 or v2.0.9 or v3.0.0)⁵¹. Selected good particles were then used to generate an initial 3D model by ab initio reconstruction followed by homogeneous refinement with C4 symmetry⁵¹. Multiple rounds of CTF refinement and Bayesian polishing were performed in relion to further improve the map resolution⁵².

At this stage, conformational heterogeneity was observed in the transmembrane domain (TMD) of the consensus refinement, indicating significant flexibility is present for TRPM5, especially in the extracellular pore loop area. To further improve the map quality, we performed focused classification by subtracting TMD signals from the particles⁵³. After TMD-focused classification, we focused on the map with the highest nominal resolution and well-defined extracellular region for atomic model building.

For the Ca²⁺–TRPM5(E337A) dataset, conformational heterogeneity is still present in the transmembrane domain after TMD-focused classification. To overcome this issue, we performed symmetry expansion to the best particle set obtained from the TMD-focused classification and subtracted the single-subunit signals. Focused classification was conducted at the single-subunit level followed by 3D refinement. Two distinct conformations of the single subunit are identified for the Ca²⁺–TRPM5(E337A) dataset. The two conformation differ by the Ca²⁺ occupancy in the transmembrane domain, i.e., apo–TRPM5(E337A) and Ca²⁺–TRPM5(E337A). The single subunit maps were used for model building. To obtain tetrameric map for apo–TRPM5(E337A) and Ca²⁺–TRPM5(E337A), we further identified the homotetrameric TRPM5 particles that were solely composed of particles from each of the single subunit class. Although homo-tetrameric particles obtained after this procedure were very less, the refinement map still allowed us to generate a tetrameric model based on the single subunit map (see the model building section).

For the (Ca²⁺, NDNA)–TRPM5 dataset, conformational heterogeneity is observed in the ICD after TMD-focused classification. We then performed symmetry expansion (C4) and subtracted the ICD for each single subunit of TRPM5 particles. Subsequent 3D classification allowed us to obtain a homogeneous set of single TRPM5 subunit. We then identified homo-tetramer of TRPM5 that are consists of the homogeneous TRPM5 single subunit and refined the structure. This allowed us to obtain a map with better defined ICD to assist model building, despite with slightly worse nominal resolution compared to the consensus refinement before ICD classification.

For all dataset, the Gold standard Fourier shell correlation (FSC) 0.143 criteria were used to provide the map resolution estimate⁵⁴. The cryo-EM maps were visualized using UCSF ChimeraX⁵⁵. PDB structures are visualized using UCSF ChimeraX or PyMOL^55,56.

Model building

The atomic model for apo–TRPM5 was built into the cryo-EM density manually using Coot v0.89 and subjected to real-space refinement in Phenix^57,58. The apo–TRPM5 model contained residues 16–429, 446–473, 489–653, 698–1020, and 1027–1092. One GDN molecule lacking one of the two maltose groups (GDP), and one diosgenin molecule (DIO) were modeled into the lipid- or detergent-like densities for each chain. The geometrical restraints for DIO, GDP and NDNA were generated using the Grade Web Server (http://grade.globalphasing.org). Glycosylation at N921 was modeled as N-acetyl-beta-D-glucosamine (NAG). The Ca²⁺–TRPM5 open models were built by first docking the apo-TRPM5 closed model into the corresponding cryo-EM map density and adjusted manually in Coot. Two Ca²⁺ atoms were added to the TMD and ICD Ca²⁺ binding sites of the Ca²⁺–TRPM5 model. The Ca²⁺–TRPM5 open model was of sufficient resolution to allow us further place two ordered water molecules in the TMD Ca²⁺ binding site, and two water molecules in the selectivity filter for each chain. The apo–TRPM5(E337A) and Ca²⁺–TRPM5(E337A) model were first built based on the cryo-EM maps of the single-subunit refinement result. The single-subunit model was then rigid-body-fitted into the tetramer cryo-EM maps reconstructed from homo-tetrameric TRPM5 particles. We did not build atomic model for Apo–TRPM5(nanodisc) and Ca²⁺–TRPM5(nanodisc) dataset because these maps are identical to the corresponding maps from the GDN detergent conditions (Extended Data Fig. 2j). It is worth mentioning that although we included 0.5 mM steviol when preparing the Ca²⁺–TRPM5(nanodisc) grid, we are not able to identify density that corresponds to the steviol molecule.

Electrophysiology

In the inside-out patch clamp configuration, voltage-clamped membrane currents were measured from tsA201 cells overexpressing plasmids encoding N-terminal GFP tagged WT and mutant TRPM5 channels from zebrafish and human. Following 1 d post-transfection with Lipofectamine 2000, cells were trypsinized and replated onto poly-L-lysine-coated (Sigma) glass coverslips. After cell adherence, the coverslips were transferred to a low-volume recording chamber with a pH 7.4 bath solution containing (in mM) 150 NaCl, 3 KCl, 10 HEPES, 2 CaCl₂, 1 MgCl₂, and 12 mannitol. Cells with fluorescence at the plasma membrane were patched with pipettes containing a pH 7.4 solution of (in mM) 150 NaCl, 10 HEPES, and 5 EGTA. Upon tight-seal formation, the bath solution was superfused with the calcium-free EGTA solution. Following excision, patches were exposed using a manifold to super-fused bath solutions containing various free calcium concentrations. For preparing 1, 30, 100, or 1000 μM of free calcium, 4.46, 5.01, 5.1, or 6 mM CaCl₂ was added to a pH 7.4 solution of 150 mM NaCl, 10 mM HEPES, 5 mM EGTA. Free calcium concentrations were calculated with https://somapp.ucdmc.ucdavis.edu/pharmacology/bers/maxchelator/CaEGTA-TS.htm. At room temperature (21–23 °C), patches from a holding voltage of 0 mV were clamped (Clampex 11.0.3, Multiclamp 700 B) using 50-ms steps from +200 mV to −200 mV (intracellular side relative to extracellular) with a final tail pulse at −140 mV. Electrical signals were digitized at 10 kHz and filtered at 2 kHz. Typically, measurements of TRPM5 activation by individual bath solutions containing various calcium concentration were interleaved with measurements where the bath solution was superfused with calcium-free EGTA solution. Using offline analysis (ClampFit 11.0.3) the currents in the absence of calcium were then subtracted from currents measured in the presence of calcium to acquire specific calcium-activated currents. Current amplitudes were measured at the end of the pulse. For normalizing current, the clamp at + 200 mV was chosen. Whole-cell measurements were performed in tsA201 cells following 1 d transfection of zebrafish TRPM5 Ca_TMD mutant channels. Patch pipettes were filled with a 1 μM free calcium concentration solution (pH 7.4) composed of (in mM): 150 NaCl, 1 MgCl₂, 10 Hepes, 5 EGTA, 4.45 CaCl₂. The bath solution (pH 7.4) contained (in mM) 150 NaCl, 3 KCl, 10 Hepes, 2 CaCl₂, 1 MgCl₂, and 12 mannitol. Voltage clamps (50 ms steps from +200 to −200 mV) were imposed approximately 1 minute after the whole cell configuration was acquired. Analysis was performed with GraphPad.

The NDNA patch clamp experiments (Fig 4; Extended Data Fig. 9) were performed in the same way as experiments for Ca_TMD mutants (Extended Data Fig. 8). For determining the IC₅₀ of NDNA, whole cell current analysis was performed where TRPM5 currents were evoked with 1 μM calcium in the patch pipette (as described above). Upon whole cell acquisition, currents were first measured in bath solution and then re-measured 30–60 s following super-fusion of bath solution containing various NDNA concentrations (1 fM, 10 pM, 100 pM, 1 nM, 100 nM, 0.5 μM, 10 μM). NDNA was stored at 50 mM (DMSO) and serially diluted using bath solution. For each cell measured, only one concentration of NDNA was tested. Inhibition kinetics was monitored using a step protocol (+100 mV) and typically complete (steady-state current) within a minute. Inhibited current was plotted as a function of NDNA concentration and fitted using Prism software (inhibitor versus response, variable slope equation).

Preparation of N’-(3,4-dimethoxybenzylidene)-2-(naphthalen-1-yl)acetohydrazide (compound NDNA)

N’-(3,4-dimethoxybenzylidene)-2-(naphthalen-1-yl)acetohydrazide (NDNA) was synthesized according to the US Patent US8193168 (Ref¹⁷) (Supplementary Fig. 8a). Briefly, A solution of commercial available ethyl 2-(naphthalen-1-yl)acetate (compound 1) (20 g, 93.34 mmol, 1 eq), NH₂NH₂.H₂O (9.54 g, 186.69 mmol, 9.26 mL, 2 eq) in EtOH (100 mL) was stirred at 80°C for 16 h. TLC (Petroleum ether/Ethyl acetate = 2/1) showed that most of the compound 1 (R_f = 0.5) was consumed and a new spot (R_f = 0.05) was given. The reaction mixture was concentrated under vacuum to give white solid. The white solid was triturated with Petroleum ether/Ethyl acetate = 4:1 (100 mL) for 10 min. The mixture was filtered, and the filter cake was dried under vacuum to give the tittle compound 2 (12.8 g, 59.58 mmol, 63.8% yield, 93.2% purity) as a white solid. ¹H NMR (400 MHz, DMSO-d₆, see Supplementary Fig. 8b upper panel) δ ppm 9.34 (s, 1H), 8.13 (d, J = 8.4 Hz, 1H), 7.92 (d, J = 7.6 Hz, 1H), 7.81 (d, J = 5.6 Hz, 1H), 7.54–7.51 (m, 2H), 7.45–7.44 (m, 2H), 4.25 (s, 2H), 3.84 (s, 2H). LCMS 0–60% ACN–H₂O, ESI + APCI: Rt = 0.767 min, m/z = 201.1 (M+H)⁺.

A solution of 2-(naphthalen-1-yl)acetohydrazide (compound 2) (10.8 g, 50.27 mmol, 1 eq) and 3,4-dimethoxybenzaldehyde (compound 3) (8.35 g, 50.27 mmol, 1 eq) in EtOH (50 mL) was stirred at 80°C for 30 min. A white solid separated out. TLC (Petroleum ether/Ethyl acetate = 1/1) showed that compound 2 (R_f = 0.1) was consumed and a major spot (R_f = 0.3) formed. The reaction mixture was cooled to 20°C. 100 mL EtOH was added to above solution and stirred for 10 min. The mixture was filtered and the filter cake was dried under vacuum to give NDNA (9.20 g, 26.33 mmol, 52.3% yield, 99.7% purity) as a white solid. ¹H NMR (400 MHz, DMSO-d₆, see Supplementary Fig. 8b lower panel) δ ppm 11.06–11.38 (m, 1H), 8.18 (m, 1.4H), 7.96 (m, 1.6H), 7.93 (m, 1H), 7.55–7.48 (m, 4H), 7.31–7.30 (m, 1H), 7.18 (m, 1H), 7.01–6.99 (m, 1H), 4.53–4.23 (m, 2H), 3.80–3.65 (m, 6H). LCMS 5–95% ACN–H₂O, ESI + APCI Rt = 0.826 min, m/z = 349.0 (M+H)⁺.

Data availability

The cryo-EM density map and coordinates and atomic models were deposited in the EMDB (Electron Microscopy Data Bank) and the Research Collaboratory for Structural Bioinformatics Protein Data Bank (RCS-PDB), respectively, under the accession numbers EMD-23740 and PDB 7MBP (apo–TRPM5), EMD-23741 and PDB 7MBQ (Ca²⁺–TRPM5), EMD-23742 (apo–TRPM5(nanodisc)), EMD-23743 (Ca²⁺–TRPM5(nanodisc)), EMD-23744 and PDB 7MBR (apo–TRPM5(6μM Ca²⁺)), EMD-23745 and PDB 7MBS (Ca²⁺–TRPM5(6μM Ca²⁺)), EMD-23746 and PDB 7MBT (apo–TRPM5(E337A) single subunit and tetramer), EMD-23747 and PDB 7MBU (Ca²⁺–TRPM5(E337A) single subunit and tetramer), and EMD-23748 and PDB 7MBV ((Ca²⁺, NDNA)–TRPM5). Source data are provided with this paper.

Extended Data

Extended Data Figure 1: Patch-clamp analysis of Ca_ICD mutants.

Representative traces of inside-out voltage-clamp measurements (+200 mV to −200 mV) from tsA201 cells overexpressing human TRPM4 (hsTRPM4), human TRPM5 (hsTRPM5), and zebrafish TRPM5 (drTRPM5) channels. Patches were stimulated with 1, 30, 100, or 1000 μM Ca²⁺. The number of patches analyzed: a, hsTRPM4(WT): 1 μM Ca²⁺ [4], 100 μM [4], 1000 μM [4] from 3 transfections; b, drTRPM5(E337A): 1 μM Ca²⁺ [6], 30 μM [6], 100 μM [4], 1000 μM [4] from 8 transfections; c, drTRPM5(C324A): 1 μM Ca²⁺ [6], 30 μM [6], 100 μM [6], 1000 μM [6] from 4 transfections; d, drTRPM5(D333A): 1 μM Ca²⁺ [5], 30 μM [4], 100 μM [5], 1000 μM [3] from 3 transfections; e, drTRPM5(E212A): 1 μM Ca²⁺ [4], 30 μM [3], 100 μM [4], 1000 μM [3] from 3 transfections; f, drTRPM5(D336A): 1 μM Ca²⁺ [3], 30 μM [3], 100 μM [3], 1000 μM [3] from 2 transfections; g, hsTRPM5(WT): 1 μM Ca²⁺ [5], 30 μM [5] from 3 transfections; h, hsTRPM5(E351A): 1 μM Ca²⁺ [3], 30 μM [3] from 1 transfection; and i, hsTRPM5(E560A): 1 μM Ca²⁺ [5], 30 μM [3], 100 μM [3], 1000 μM [3] from 5 transfections. j–r, Mean current (50 ms) of experiments were plotted versus indicated voltage. The +200 mV clamp was chosen for normalization. Horizontal bars represent SEM. The current-voltage relation plots of drTRPM5(E337A) are identical to those presented in Fig. 3e. s–u, Individual patch clamp measurements I_−200mV / I_+200mV, I_+200mV, I_−200mV, of experiments are shown as individual points, where bars represent mean values. v, w, Currents of drTRPM5 WT and Ca_ICD mutants plotted as a function of Ca²⁺ concentration for voltage clamps of −200 mV and +200 mV. Symbols represent mean current and horizontal bar is SEM.

Extended Data Figure 2: Cyro-EM analysis of apo–TRPM5.

a, The representative 2D class average of apo–TRPM5. b, The Fourier shell correlation (FSC) curves for the apo–TRPM5. The cryo-EM map FSC is shown in black and the model vs. map FSC is shown in red. The map resolution was determined by the gold-standard FSC at 0.143 criterion, whereas the model vs. map resolution was determined by a threshold of 0.5. c, The angular distribution of particles that gave rise to the apo–TRPM5 cryo-EM map reconstruction. d, A schematic domain organization of a single TRPM5 subunit. Secondary structures and important domains are labeled. e, The atomic model of a single TRPM5 subunit in cartoon representation. The domains are colored as in (d). The left and right panels are two different views of the same subunit rotated 180° along the central axis.

Extended Data Figure 3: Cyro-EM analysis of TRPM5 in the presence of different concentrations of Ca²⁺, or in the presence of Ca²⁺ and NDNA.

a and d, The representative 2D class average the 5mM Ca²⁺ dataset (a) and 6 μM Ca²⁺ dataset (d), respectively. b and e, The Fourier shell correlation (FSC) curves for the 5mM Ca²⁺ dataset (b) and 6 μM Ca²⁺ dataset (e). The cryo-EM map FSC is shown in black and the model vs. map FSC is shown in red. The map resolution is determined by the gold-standard FSC at 0.143 criterion, whereas the model vs. map resolution is determined by a threshold of 0.5. c and f, The angular distribution of particles that give rise to the cryo-EM map reconstruction for 5mM Ca²⁺ dataset (c) and 6 μM Ca²⁺ dataset (f). g, The close up view of the Ca_TMD and Ca_ICD of the 6 μM Ca²⁺ dataset. From left to right, Ca_TMD of apo–TRPM5(6 μM Ca²⁺), Ca_TMD of Ca²⁺–TRPM5(6 μM Ca²⁺), Ca_ICD of apo–TRPM5(6 μM Ca²⁺), and Ca_ICD of Ca²⁺–TRPM5(6 μM Ca²⁺). The cryo-EM densities are shown in mesh representation. The expected Ca²⁺ density is indicated by a circle. h, The representative 2D class average the (Ca²⁺, NDNA)–TRPM5 dataset. i, The Fourier shell correlation (FSC) curves for the (Ca²⁺, NDNA)–TRPM5. The cryo-EM map FSC is shown in black and the model vs. map FSC is shown in red. The map resolution was determined by the gold-standard FSC at 0.143 criterion, whereas the model vs. map resolution was determined by a threshold of 0.5. j, The angular distribution of particles that gave rise to the (Ca²⁺, NDNA)–TRPM5 cryo-EM map reconstruction.

Extended Data Figure 4: Local resolution estimation of TRPM5 structures and representative densities.

a-d, The local resolution estimation for apo–TRPM5(GDN) (a), Ca²⁺–TRPM5(GDN) (b), Ca²⁺–TRPM5(E337A)(GDN) consensus (c), (Ca²⁺, NDNA)–TRPM5(GDN) (d). For each map, a side view, a top-down view of the TMD from the extracellular side, and a focused side view of the S6 and pore helix are shown. The color bar unit is in Ångstroms. e, Representative densities from Ca²⁺–TRPM5(GDN) map. For the GDN density, one maltose group of the molecule is not resolved in the cryo-EM density map.

Extended Data Figure 5: The gate and the selectivity filter of TRPM5.

a, Cryo-EM densities of the Ca_ICD site, contoured at 0.018. b, Cryo-EM densities of the Ca_TMD site, contoured at 0.022. c, Cryo-EM densities of the water molecule and residues in the selectivity filter, contoured at 0.023. Hydrogen bonds are shown as solid yellow lines. The “lower” water molecule is surrounded by the sidechain of Q906 and the backbone oxygen atoms of F904 and G905, forming three hydrogen bonds. d, Cryo-EM densities of I966, which forms the channel gate, contoured at 0.03. e, The selectivity filter formed by two layers of ordered water molecules (blue spheres) and backbone oxygen atoms (pink spheres) of G905. f and g, The two hydration layers the selectivity filter viewed from the extracellular side. Upper layer in (f) and lower layer in (g).

Extended Data Figure 6: Comparison of TRPM5 with other TRPM channels.

a, A structural comparison between Ca²⁺–TRPM5 and TRPM4 (PDBID: 6BQV)²¹. A single subunit is in color and shown as a cartoon. The TRPM5 channel is more compact, but wider than, the TRPM4 channel. b, An overlap of the selectivity filter of Ca²⁺–TRPM5 (red) and TRPM4 (yellow). c, A comparison of the Ca_TMD site for the available TRPM members. From left to right, drTRPM5, hsTRPM4 (6BQV), drTRPM2 (6DRJ), nvTRPM2 (6CO7), hsTRPM2 (6PUS), and pmTRPM8 (6O77)^{21,23,34,41,42}; Shown in parentheses are the PDBIDs. d, Comparison of the “square” helices in TRPM2 (6PUO), TRPM4 (6BQR), TRPM5, TRPM7 (5ZX5), TRPM8 (6O6A); Shown in parentheses are the PDBIDs^21,23,42,44. Only TRPM5 has a broken square helix. e, A sequence alignment of the square helix across different TRPM5 orthologs and TRPM family members. Red indicates that the α-helix is observed in structures. For TRPM1, TRPM3, and TRPM6, in which no structures are currently available, the helical annotation is based on the secondary structure prediction from PSIPRED server⁴⁵.

Extended Data Figure 7: Comparison of (Ca²⁺, NDNA)–TRPM5 with apo–TRPM5 and Ca²⁺–TRPM5.

a, the chemical structure of N’-(3,4-dimethoxybenzylidene)-2-(naphthalen-1-yl)acetohydrazide (NDNA). b, Two close-up views of the cryo-EM densities of NDNA molecule. The surrounding protein structural element is shown in cartoon representation. c, Comparison of the NDNA binding site between (Ca²⁺, NDNA)–TRPM5 and apo–TRPM5 structures. The W869 is flipped in the (Ca²⁺, NDNA)–TRPM5 structure (cyan) compared to that in apo–TRPM5 structure (blue). d, Overlay of (Ca²⁺, NDNA)–TRPM5 (cyan) with apo–TRPM5 (blue) and Ca²⁺–TRPM5 (red) structures view from the intracellular side. One subunit is shown in cartoon representation and the other three subunits are in surface representation. The ICD of (Ca²⁺, NDNA)–TRPM5 adopts an intermediate state compared to the apo–TRPM5 and Ca²⁺–TRPM5 structures. e, The superimposition of the S1-S4 domain between (Ca²⁺, NDNA)–TRPM5 (cyan) and apo–TRPM5 (blue). f, The superimposition of the S1-S4 domain between (Ca²⁺, NDNA)–TRPM5 (cyan) and Ca²⁺–TRPM5 (red). g, A close-up view of the Ca_TMD site in (Ca²⁺, NDNA)–TRPM5 structure. The Q771 moved away from Ca_TMD. h and i, An overlay of the pore domain between (Ca²⁺, NDNA)–TRPM5 (cyan) with apo–TRPM5 (blue) (h) and Ca²⁺–TRPM5 (red) (i) structures viewed from the extracellular side.

Extended Data Figure 8: Patch-clamp electrophysiology experiments of Ca_TMD mutants of zebrafish TRPM5.

a, Representative whole-cell current traces of tsA overexpressing WT and Ca_TMD mutant TRPM5 channels. Clamps were imposed from +200 mV to −200 mV. The number of cells measured were tsA201 [n = 4 cells], TRPM5(WT) [5], TRPM5(E768A) [5], TRPM5(Q771A) [4], TRPM5(N794A) [4], TRPM5(D797A) [4], and TRPM5(E994A) [4] from 2–3 transfections. b, Mean current amplitudes of experiments in (a) were measured at 50 ms and plotted as a function of clamp voltage. Horizontal bars represent SEM. c, Individual measurements at clamps of +200 mV (I_+200mV) and −200 mV (I_−200mV) of experiments in (a) are shown as individual points, with bars representing mean values.

Extended Data Figure 9: Patch-clamp electrophysiology experiments on the NDNA binding site mutants.

a–j, Current voltage relations of whole-cell measurements in tsA cells over-expressing WT and mutant TRPM5 channels. Symbols represents mean current and horizontal bars are SEM. k–n, Individual measurements (of experiments in a–j) for clamps of +100 mV and −100 mV. Each point represents an individual cell and bars represent mean current. Cells were first measured (clamps from −100 mV to +100 mV) in bath solution and then re-measured following bath perfusion of 10 μM NDNA. See the legend of Fig 4 for the number of cells used.

Extended Data Figure 10: Ca²⁺–TRPM5(E337A) in GDN detergent.

a, The representative 2D class average of Ca²⁺–TRPM5(E337A). b, The FSC curve for the consensus map of Ca²⁺–TRPM5(E337A). The map resolution was determined by the gold-standard FSC at 0.143 criterion. c, The angular distribution of particles that give rise to the consensus map of Ca²⁺–TRPM5(E337A). d, The FSC curve for apo–TRPM5(E337A) (left) and Ca²⁺–TRPM5(E337A) (right). For each panel, the cryo-EM map FSC curve is shown in black and the model vs. map FSC is shown in red. The map resolution was determined by the gold-standard FSC at 0.143 criterion, whereas the model vs. map resolution was determined by a threshold of 0.5.