If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Institute for Computational Molecular Science, Temple University, Philadelphia, Pennsylvania, USADepartment of Biology, Temple University, Philadelphia, Pennsylvania, USAInstitute for Genomics and Evolutionary Medicine, Temple University, Philadelphia, Pennsylvania, USA
Transient receptor potential melastatin 3 (TRPM3) is a heat-activated ion channel expressed in peripheral sensory neurons and the central nervous system. TRPM3 activity depends on the membrane phospholipid phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2), but the molecular mechanism of activation by PI(4,5)P2 is not known. As no experimental structure of TRPM3 is available, we built a homology model of the channel in complex with PI(4,5)P2via molecular modeling. We identified putative contact residues for PI(4,5)P2 in the pre-S1 segment, the S4–S5 linker, and the proximal C-terminal TRP domain. Mutating these residues increased sensitivity to inhibition of TRPM3 by decreasing PI(4,5)P2 levels. Changes in ligand-binding affinities via molecular mechanics/generalized Born surface area (MM/GBSA) showed reduced PI(4,5)P2 affinity for the mutants. Mutating PI(4,5)P2-interacting residues also reduced sensitivity for activation by the endogenous ligand pregnenolone sulfate, pointing to an allosteric interaction between PI(4,5)P2 and pregnenolone sulfate. Similarly, mutating residues in the PI(4,5)P2 binding site in TRPM8 resulted in increased sensitivity to PI(4,5)P2 depletion and reduced sensitivity to menthol. Mutations of most PI(4,5)P2-interacting residues in TRPM3 also increased sensitivity to inhibition by Gβγ, indicating allosteric interaction between Gβγ and PI(4,5)P2 regulation. Disease-associated gain-of-function TRPM3 mutations on the other hand resulted in no change of PI(4,5)P2 sensitivity, indicating that mutations did not increase channel activity via increasing PI(4,5)P2 interactions. Our data provide insight into the mechanism of regulation of TRPM3 by PI(4,5)P2, its relationship to endogenous activators and inhibitors, as well as identify similarities and differences between PI(4,5)P2 regulation of TRPM3 and TRPM8.
Transient receptor potential melastatin 3 (TRPM3) is a heat-activated, outwardly rectifying, and Ca2+-permeable nonselective cation channel expressed in a variety of tissues, including peripheral sensory neurons of the dorsal root ganglia, and the central nervous system (
), this regulation is complex, and sometimes controversial, with both negative and positive effects having been proposed. With the exception of TRPM1, which is very difficult to study in expression systems, all members of the TRPM subfamily have been shown to be positively regulated by PI(4,5)P2, and no negative regulation has been proposed for any TRPM subfamily member (
Currently, it is not known which residues in the TRPM3 protein PI(4,5)P2 binds to, and there is no experimentally determined structure available for TRPM3. To fill this key gap in knowledge, we generated a homology model of TRPM3, based on the experimental structure of TRPM4 in the ligand-free (apo) state (
). We then docked PI(4,5)P2 to our model of TRPM3 and identified putative PI(4,5)P2-interacting residues in the pre-S1 segment, the S4–S5 linker, and the proximal C-terminal TRP domain. We validated our results by docking PI(4,5)P2 to an apo structure of TRPM8 (
), experimentally determined recently. In silico mutations of the PI(4,5)P2 contact residues in TRPM3, followed by ligand-binding affinity changes via molecular MM/GBSA, showed reduced PI(4,5)P2 binding affinity to TRPM3. We experimentally validated the importance of these residues by demonstrating that their mutations increased sensitivity to inhibition by PI(4,5)P2 depletion in electrophysiology experiments. We also showed that mutating most of these residues increased sensitivity to Gβγ inhibition, and decreased sensitivity to agonist activation, indicating allosteric interaction between PI(4,5)P2 and endogenous activators and inhibitors. Furthermore, we demonstrated that gain-of-function disease–associated mutations did not change PI(4,5)P2 sensitivity, indicating that the mutations do not increase channel activity via promoting PI(4,5)P2 activation. Our data provide mechanistic insights into regulation of TRPM3 by its key endogenous cofactor PI(4,5)P2.
Our goal in this study was to identify the PI(4,5)P2 binding site of TRPM3. As there is currently no experimentally determined TRPM3 structure available, we generated a homology model of the human TRPM3 (hTRPM3) based on the cryo-EM structure of the mouse TRPM4 in the apo state (Protein Data Bank [PDB] ID: 6BCJ) (
) (Fig. 1, A and B). The template was selected as the closest homolog to TRPM3 in the TRPM family with an experimental structure available when the model was built (see the Experimental procedures section and Scheme S1 for details). Our model of TRPM3 aligns very well with the models of TRPM3 from different organisms generated recently by AlphaFold (
) in EDTA (PDB ID: 5ZX5) as the template (Fig. S2), all of which became available after our original homology model was built, providing a posteriori validation of our TRPM3 model.
Next, we identified putative residues interacting with PI(4,5)P2 in TRPM3 by using two complementary approaches. First, we scanned the surface of apo TRPM3 (model built on TRPM4) for putative binding sites using the program SiteMap (Schrödinger, LLC, 2018) (
). Second, we relied on sequence and structural information available on TRPM8 to detect, by homology, which residues are likely to interact with PI(4,5)P2. Specifically, (1) we generated a sequence alignment of TRP-domain residues, K995, R998, and R1108, in the rat TRPM8, which are conserved among TRPM family members (Fig. 1C), and were previously suggested to play a key role in PI(4,5)P2 interactions (
), we built a refined model of this channel bound to PI(4,5)P2 at a site that includes the conserved TRP-domain residues (Fig. 2, A and B). Comparing this complex with the apo-TRPM3 model showed that the most suitable site for lipid binding (i.e., the top-scoring binding spot combining SiteMap predictions and structural information) in TRPM3 corresponded to the PI(4,5)P2 site identified in TRPM8. We used this lipid-binding site to generate a model of TRPM3 in complex with a version of PI(4,5)P2 with truncated tails (similar to the synthetic diC8 PI(4,5)P2, which is fully functional in experiments) by molecular docking using the program Glide (
). We ranked the lipid binding modes by the standard precision scoring function. The best binding mode in TRPM3, defined as the best docking score (kilocalorie/mole) obtained at the binding site similar to that in TRPM8, is shown in Figure 1, A and B.
The PI(4,5)P2 binding site in TRPM3 is formed by parts of the preS1 segment, the S4–S5 loop, and the proximal C-terminal TRP domain of the same subunit. The closest contact residues with PI(4,5)P2 are W761 in the preS1 segment, the N991 and K992 residues in segment connecting the voltage sensor–like domain (S1–S4) to the S4–S5 linker and the R1131 in the TRP domain (Fig. 1B). Figure 1B also shows the location of two additional residues, K1128 and R1141 in the TRP domain, which are not in close contact with PI(4,5)P2, but we experimentally characterized their mutations (see later). The numbering of these residues corresponds to the splice variant of hTRPM3 (hTRPM1325) (
), icilin (PDB ID: 6NR3), and with the menthol analog WS12 (PDB ID: 6NR2), offered a posteriori validation of our modeling (Table S1). Fig. S3 compares the PI(4,5)P2 binding pockets of the TRPM8–PI(4,5)P2–icilin structure, the TRPM8–PI(4,5)P2–WS12 structure, and our computational model. Our model superposes very well with both structures in the transmembrane domains, but it shows a better structural alignment of the PI(4,5)P2 binding site with the TRPM8–PI(4,5)P2–WS12 (PDB ID: 6NR2) structure than the TRPM8–PI(4,5)P2–icilin–calcium structure (PDB ID: 6NR3). In fact, the minimum RMSD values, calculated over amino-acid ranges facing the lipid-binding sites between any two aligned structures, were 1.49 and 2.34 Å, respectively (Table 1). Interestingly, the structural difference between the two experimental structures (minimum RMSD of 1.99 Å) is larger than that observed between our model and the closest experimental complex (minimum RMSD of 1.49 Å).
Table 1Structural comparison between TRPM8 model and experimental structures
RMSD was calculated independently on two different TRPM8-residue selections (details are provided in Experimental procedures section) using the TRPM8 model or the TRPM8 experimental structures 6NR2 or 6NR3 as reference (RMSD = 0.0 Å).
) listed five key residues in their cryo-EM costructures critical for PI(4,5)P2 interaction: R997 in the TRP domain, R850 in the S4–S5 loop, N692 and R688 in the pre-S1 segment (Fig. 1C), and K605 in the neighboring N-terminal cytoplasmic Melastatin Homology Region 4 (MHR4) domain. All these residues, with the exception of R850, are in contact with, or very close to PI(4,5)P2 in our TRPM8–PI(4,5)P2 model. R850 is in contact with the acyl chain of PI(4,5)P2 in our model, and only in contact with the PI(4,5)P2 headgroup in the 6NR3, but not in the 6NR2 structure, which is consistent with the better alignment of our model with the 6NR2 PI(4,5)P2-TRPM8 structure. Overall, our TRPM8–PI(4,5)P2 docking model validates our computational approach to identify the TRPM3 PI(4,5)P2 binding site and suggests that PI(4,5)P2 likely binds to a site that is similar in TRPM3 and TRPM8.
Furthermore, superimposition of our model to the experimental structure of TRPM7 in EDTA (PDB ID: 5ZX5) (
) revealed that the docked PI(4,5)P2 in our model of TRPM3 fits well in a cavity of the experimental structure of TRPM7 that accommodates a detergent cholesteryl hemisuccinate molecule (Fig. S4). Whether this binding site is occupied by PI(4,5)P2 in TRPM7 in a cellular environment, remains to be determined, nevertheless the presence of this lipid-binding pocket in TRPM7 suggests that the location of the PI(4,5)P2 binding site may be conserved in multiple members of the TRPM subfamily.
In our TRPM3-PI(4,5)P2 model, residues K992 and R1131 are equivalent to the experimentally determined PI(4,5)P2 contact sites R850 and R997 in TRPM8, located in the S4–S5 loop and the TRP domain (Fig. 1C). Specifically, N991 is adjacent to K992, and W761 in the pre-S1 segment of TRPM3 is shifted six residues from the R688 residue in TRPM8. The equivalent of W761 in TRPM8 (W682) is relatively close to PI(4,5)P2 in TRPM8, and so is the equivalent of R688 in TRPM3 (M767) highlighting the generally similar importance of the pre-S1 segment in PI(4,5)P2 binding in the two channels. The largest difference between the two binding sites is that the MHR4 region, which carries K605 in TRPM8, is not conserved in TRPM3, and the equivalent residue is far away from PI(4,5)P2 in our TRPM3 model. Overall, the two channels bind PI(4,5)P2 in a similar, yet not identical manner (Fig. S5).
Next, we mutated the predicted PI(4,5)P2-interacting residues in TRPM3 and tested the effects of the mutations on sensitivity to inhibition by decreasing PI(4,5)P2 levels. We expressed the WT and mutant channels in Xenopus oocytes and performed two-electrode voltage clamp (TEVC) experiments. We stimulated channel activity with 50 μM PregS and measured current amplitudes, then incubated the oocytes with 35 μM wortmannin for 2 h, and measured PregS-induced currents in the same oocytes (Fig. 3, A and B). Wortmannin at this concentration inhibits phosphatidylinositol 4-kinases and has been used to inhibit the activity of PI(4,5)P2-dependent ion channels (
), indicating that TRPM3 inhibition by 35 μM wortmannin is caused by inhibition of phosphatidylinositol 4-kinase, not PI3K. Mutating a PI(4,5)P2-interacting residue is expected to increase inhibition by high concentrations of wortmannin (
). The rest of the residues we mutated to A, but the W761A mutant was nonfunctional, thus we functionally characterized W761F instead. Mutations of all computationally predicted PI(4,5)P2-interacting residues (W761F, N991A, K992A, and R1131Q) showed significantly higher inhibition after wortmannin treatment than WT TRPM3 (Fig. 3, C–F), and their current amplitudes were also significantly lower than WT TRPM3 (Fig. 3, H–K). We also generated two additional mutations in the TRP domain in residues that are not in contact with PI(4,5)P2, K1128Q and R1141Q. Both mutants showed similar current amplitudes to WT TRPM3 (Fig. 3, K and L). The K1128Q mutant showed similar inhibition to WT (Fig. 3G), but the R1141Q mutant showed a small but significant increase in wortmannin inhibition compared with WT (Fig. 3F). This mutation is equivalent to R1008Q in the rat TRPM8, which reduced both PI(4,5)P2 and menthol sensitivity (
). Fig. S6, A–C shows that 100 nM rapamycin induced a significantly higher inhibition of the N993A mutant of the mouse TRPM3α2 (mTRPM3α2; equivalent of N991A in hTRPM3) than the WT mTRPM3α2 when the channels were stimulated with 25 μM PregS. Next, we stimulated the mTRPM3α2 with the combination of 25 μM PregS and 10 μM clotrimazole, which was shown to open an alternative pore, characterized by larger currents and less prominent outward rectification (
). Application of 100 nM rapamycin induced a significantly larger inhibition of currents induced by clotrimazole plus PregS in the N993A mutant compared with the WT mTRPM3α2 (Fig. S6, E–G). Current amplitudes for the N993A mutant were also lower than those in the WT TRPM3 (Fig. S6, D and H).
Mutation of the PI(4,5)P2 contact site R998Q resulted in a right shift in the diC8 PI(4,5)P2 dose response in excised patches (
). This and the low current amplitudes in the mutants prevented us from reliably comparing PI(4,5)P2 dose responses in our mutants. It was reported for TRPV1 that mutating a putative PI(4,5)P2-interacting residue increased the relative efficiency of PI(4)P to stimulate channel activity compared with PI(4,5)P2 (
). Therefore, we tested the relative effects of PI(4)P and PI(4,5)P2 on the N991A mutant. Fig. S7, A–C shows that the relative effect of PI(4)P compared with PI(4,5)P2 did not change.
Next, we used MM/GBSA calculations (Fig. 4) to predict the changes in the binding free energy (ΔΔG) of PI(4,5)P2 to the native (WT) TRPM3 versus the mutant channels that were characterized in Figure 3. In particular, we used the VSGB 2.0 model (
) for protein residues. We note that we did not include an implicit membrane model (i.e., a low-dielectric slab) and, therefore, the results should be taken as a qualitative indication.
As shown in Figure 4, the binding of PI(4,5)P2 is guided by a number of stabilizing interactions (Fig. 4A) established with key contact residues (Fig. 4, B–F and Table S2). Mutations of all these residues in our model resulted in a decreased PI(4,5)P2 binding affinity (for a native protein to bind better than the mutant, the calculated ΔΔG value is positive). Specifically, K992A had a more prominent effect than R1131Q, N991A, and W761F, respectively. This correlates well with K992A also having the most pronounced effect on inhibition by PI(4,5)P2 depletion (Fig. 3E). Regarding the binding modes, K992 engages in multiple interactions with PI(4,5)P2, including three hydrogen bonds and three salt bridges. Mutating K992 to alanine resulted in the loss of all these interactions with the exception of one hydrogen bond, the only interaction established by the amino acid backbone (Fig. 4C). Similar behavior was observed with the mutation R1131Q, the contact residue exerting the second largest effect on the binding affinity, which resulted in the loss of one hydrogen bond and two salt bridges (Fig. 4D), all established by the residue side chain. Next, mutating N991 to alanine and W761 to phenylalanine resulted in the loss of one hydrogen bond each (Fig. 4, E and F, respectively). To further corroborate our results, we performed additional sets of calculations of the binding affinity change upon mutation (Table S2 and Fig. S8) using, as the starting configurations, the docking poses of PI(4,5)P2 with even shorter tails than the ones included in the model, and with headgroups featuring different protonation states (hereinafter referred to as shortest-PI(4,5)P2). These calculations are clearly reproducible and agree with experimental observations. Although the trend is maintained overall (Figs. 4A and S8), W761F shows a reduction in the binding affinity change for shortest-PI(4,5)P2 (Fig. S8, light blue), because of headgroup protonation states that prevent interactions via hydrogen-bond formation. Hence, it appears from our calculations, that the PI(4,5)P2 protonation state featured in the proposed TRPM3 model (Fig. 4) is the one that favorably affects the binding of the phospholipid to the native protein. Interestingly, the protonation state of PI(4,5)P2 was suggested to critically impact the binding to related TRP channels (
). Furthermore, among the mutations leading to a decrease of binding affinity, W761 is located the furthest from PI(4,5)P2 (see structural model), and therefore, it is not unexpected that mutating this residue could affect to a lesser extent the binding of a smaller ligand (Table S2).
Of the remaining two mutations (Fig. 4, A and B), R1128Q had only a very small effect on both ΔΔG and the related binding mode, which correlates well with it not being in close contact with PI(4,5)P2 in our model, and the lack of effect on wortmannin inhibition. The R1141Q mutant, which is also not a PI(4,5)P2 contact site, also had only a minimal effect on both ΔΔG and the related binding mode, indicating that the small, but significant, effect on wortmannin inhibition was likely because of indirect effects. Overall, all Δ affinity calculations supported our computational docking and agreed with the experimental functional characterization of the PI(4,5)P2-interacting residues.
It is worth mentioning that, although our binding model likely captures a highly represented conformational state sampled by the TRPM3 channel when bound to PI(4,5)P2, it is expected that other states may exist featuring alternative networks of interactions yet compatible with the proposed phospholipids-binding site model.
Next, we mutated two residues in the rTRPM8 that are equivalent to PI(4,5)P2-interacting residues in our TRPM3 model. The R851 residue in TRPM8 corresponds to the K992 residue in the S4–S5 linker in TRPM3 (Fig. 1C), and it was in direct contact with PI(4,5)P2 in the cryo-EM structure of the fcTRPM8 (R850) (
). The W682 residue is the equivalent of W761 in TRPM3 (Fig. 1C), and while is not in a direct contact with PI(4,5)P2 in the fcTRPM8 cryo-EM structure (R850), it is located relatively close. Since the W682A mutant was nonfunctional, we characterized the W682Q, which displayed small, yet measurable, menthol-induced currents. Figure 5, A–G shows that both the R851Q and the W682Q mutants showed significantly higher level of inhibition by wortmannin, with W682Q having a larger effect. Current amplitudes showed a similar pattern; both mutants were significantly lower than WT TRPM8, and the W682Q having a larger effect (Fig. 5H). The decrease in amplitudes was even more pronounced at negative voltages for inward currents (Fig. 5I), in agreement with earlier results with the R995Q PI(4,5)P2 mutant (
). This is likely caused by the allosteric interaction between PI(4,5)P2 and voltage in modulating TRPM8. Wortmannin treatment substantially accelerated deactivation after cessation of menthol stimulation (Fig. 5, A–D), which is in contrast to TRPM3, where the deactivation kinetics after washing out PregS was not affected by wortmannin (Fig. 3, A and B).
Stimulation with menthol, or cold, was shown to increase the apparent affinity of TRPM8 for PI(4,5)P2 (
) indicating an allosteric interaction between menthol and PI(4,5)P2 activation. Next, we asked if this allosteric interaction also happens in the opposite direction and tested if mutations of the PI(4,5)P2-interacting residues in TRPM8 have an effect on agonist sensitivity. Figure 6, A–D shows that both the R851A and the W682Q mutant had right shifted menthol dose response. Similar to the effect on current amplitudes and wortmannin inhibition, the effect of the W682Q mutant (Fig. 6, C and D) was more pronounced than that of R851A (Fig. 6, B and D).
We also tested if a similar allosteric effect also exists in TRPM3. Figure 7, A–D shows that both the N991A and the K992A mutant shifted the PregS dose response to the right.
TRPM3 activity is inhibited by direct binding of Gβγ to the channel (
). To test if an allosteric interaction between Gβγ inhibition and PI(4,5)P2 activation is present, we expressed WT and mutant TRPM3 channels with or without Gβ1γ2 in Xenopus oocytes and measured PregS-induced currents. The N991A and K992A mutants were inhibited significantly more by Gβγ than WT TRPM3 (Fig. 8, A–E). The K1128Q mutant, which did not affect PI(4,5)P2 sensitivity, had no effect on Gβγ inhibition either (Fig. 8E). Interestingly, the R1131Q mutant was not inhibited, rather potentiated by coexpressing Gβγ (Fig. 8E). Consistently with the lack of inhibition by Gβγ, the R1131Q mutant was also not inhibited by stimulating Gi-coupled M2 muscarinic acetylcholine receptors (Fig. 8, F–J). These data indicate that while there is an allosteric interaction between PI(4,5)P2 and Gβγ, the R1131 residue in the TRP domain also plays some role in transmitting the inhibitory effect of Gβγ.
Gain-of-function mutations in TRPM3 have recently been shown to cause intellectual disability and seizures (
). The two disease-associated mutations, V990M and P1090Q, were shown to increase basal channel activity, as well as increase in agonist sensitivity and increase in heat sensitivity, with V990M affecting agonist sensitivity more prominently, whereas P1090Q predominantly affecting heat sensitivity (
). Next, we tested if the increased basal activity and agonist sensitivity also translated into higher sensitivity to PI(4,5)P2. When WT (Fig. 9, A and B) and V990M (Fig. 9, C and D) and P1090Q mutant channels (Fig. 9, E and F) were treated with 35 μM wortmannin for 2 h, currents evoked by 50 μM PregS were inhibited to a similar extent (Fig. 9G). PregS-induced average current amplitudes were not significantly different in the mutant and WT channels (not shown), similar to our earlier data (
), presumably because the overactive channels tend to damage the cells expressing them and thus in the surviving oocytes are selected for lower expression levels of the mutants. The mutants were also inhibited to a similar extent to WT channels by wortmannin when currents were evoked by PregS corresponding to the respective EC50 (
) of the mutant and WT channels (Fig. 9H). These data indicate that the disease mutants do not increase channel activity by increasing their apparent affinity for PI(4,5)P2.
Our work aims to understand the molecular mechanism of PI(4,5)P2 regulation of TRPM3. We used computational docking, changes in the binding affinity estimated by computational mutagenesis, site-directed mutagenesis, and electrophysiology to identify PI(4,5)P2-interacting residues in the channel protein. Our data indicate that residues in three regions, the pre-S1 segment, the S4–S5 loop, and the TRP domain, play important roles in forming the PI(4,5)P2 binding site in TRPM3. Mutations of PI(4,5)P2-interacting residues decreased the binding affinity in silico (positive ΔΔG values in Figure 4) indicating that the native protein binds better than the mutant) and increased sensitivity to inhibition by decreasing PI(4,5)P2 levels in electrophysiology experiments (Fig. 3). Mutating PI(4,5)P2-interacting residues also decreased sensitivity to PregS activation and increased sensitivity to Gβγ inhibition indicating allosteric interaction between PI(4,5)P2 and agonists as well as a physiological inhibitor. On the other hand, disease-associated gain-of-function mutations did not change PI(4,5)P2 sensitivity, indicating that the mutations did not increase channel activity by enhancing PI(4,5)P2 activation.
There are currently five channels in the TRPM family for which structural data are available (
). When we compare the PI(4,5)P2 binding site in TRPM8 revealed by the structural study with our computationally identified binding site in TRPM3, the two overlap, sharing some of the interacting residues, with some differences (Figs. 1C and S5). Overall, the preS1 segment, the S4–S5 loop, and the TRP domain are involved in both channels in forming the PI(4,5)P2 binding site. The R1131 residue in the TRP domain in TRPM3 is equivalent to the R997 PI(4,5)P2 contact residue in the fcTRPM8 structure (
), and to the R998 residue in the rat TRPM8, which was proposed as a PI(4,5)P2-interacting residue and experimentally shown to exhibit decreased PI(4,5)P2 sensitivity before structures became available (
). The K992 residue in the S4–S5 loop of TRPM3 is equivalent to the R850 PI(4,5)P2 contact residue in the fcTRPM8 structure and to the R851 residue in the rat TRPM8 that we characterized in this study (Figs. 5 and 6). The pre-S1 segment of the fcTRPM8 has two PI(4,5)P2 contact residues R688 and N692 (Fig. 1C). These residues are not conserved in TRPM3 (Fig. 1C); yet the equivalent residues in our TRPM3 model, that is, M767 and G672, are both located within 5 Å of the PI(4,5)P2 headgroup (Fig. S5D), with M767 engaging hydrophobic interactions that stabilize the overall complex. The W761 PI(4,5)P2 contact residue in the pre-S1 of TRPM3, is equivalent to the W682 residue in TRPM8, which is close to the R688 residue as well as to PI(4,5)P2, but it was not close enough to designate it as a PI(4,5)P2 contact site in TRPM8 (
). Interestingly, when we mutated this residue to a glutamine (W682Q) in the rat TRPM8, it behaved similar to the W761F mutation in TRPM3, that is, it increased sensitivity to PI(4,5)P2 depletion (Fig. 5). Whether this residue is in a closer contact with PI(4,5)P2 in a cellular environment in the rat TRPM8, or its mutation affected PI(4,5)P2 interactions indirectly, or both, it is difficult to tell. Finally, the K605 residue from an adjacent cytoplasmic MHR4 domain was also in contact with PI(4,5)P2 in TRPM8. This residue is not conserved in TRPM3 and was not close to PI(4,5)P2 in our homology model.
It is well established that channel agonists can increase PI(4,5)P2 sensitivity (apparent affinity) for various PI(4,5)P2-sensitive ion channels. For example, the apparent affinity of the G protein–activated inwardly rectifying K+ channel GIRK4 (Kir3.4) for PI(4,5)P2 is increased by factors that stimulate channel activity, such as Gβγ and intracellular Na+ (
). The opposite was also proposed, as a mutation in the putative PI(4,5)P2-interacting residue R1008 in TRPM8 not only decreased apparent affinity for PI(4,5)P2 but also induced a marked right shift in the menthol dose response (
). In the view of the structures of TRPM8 however, this residue is likely to be a menthol-interacting residue, as it was in close contact in the TRPM8 structure with the menthol analog WS12, but not with PI(4,5)P2 (
), therefore, it most likely primarily affected menthol sensitivity, and the effect on PI(4,5)P2 was a secondary allosteric effect. Our data indicate that in both TRPM8 and TRPM3, mutating PI(4,5)P2 contact residues also decrease agonist sensitivity. Similarly, mutating most PI(4,5)P2-interacting residues also made it easier for TRPM3 to be inhibited by Gβγ. This is likely to be an allosteric effect, as the Gβγ binding peptide in TRPM3 (
) is located far away from the PI(4,5)P2 binding site (Fig. S9). Interestingly, the R1131Q mutant did not display any Gβγ inhibition, pointing to the complex role of this residue in channel regulation.
In contrast to the apparent allosteric interaction between PI(4,5)P2 and agonist or Gβγ, disease-associated gain-of-function mutations in TRPM3 that increased both heat and agonist sensitivity (
) did not decrease sensitivity for inhibition by PI(4,5)P2 depletion (Figure 9), indicating that the mechanism of increased channel activity is not the consequence of increased sensitivity to PI(4,5)P2.
In an earlier work, well before structures became available, three residues in the TRP domain of TRPM8 were proposed to act as PI(4,5)P2-interacting residues (
). Even though mutations in this segment behaved in a way compatible with reduced PI(4,5)P2 interactions, this segment was far away from the plasma membrane in the subsequently determined TRPM4 structures, which is incompatible with acting as a PI(4,5)P2-interacting domain (
). It was also proposed that similar, nonconserved, and short charged amino acid segments are responsible for the effects of PI(4,5)P2 on other TRP channels, including TRPM3 and TRPM8, but for channels other than TRPM4, no experimental testing was performed with mutants in those segments (
). The proposed segments are at different locations in different TRPM channels, suggesting that the activation mechanism by PI(4,5)P2 is not conserved between different TRPM channels. Our data showing that the PI(4,5)P2 binding site in TRPM3 is similar, yet not identical to that in TRPM8 suggests that the PI(4,5)P2 binding site in TRPM channels shows substantial level of conservation.
In our earlier work, we used a homology model, based on the structure of TRPV1 combined with mutagenesis, to predict PI(4,5)P2-interacting residues in the epithelial Ca2+ channel TRPV6 (
). TRPV5 and TRPV6 are products of a relatively recent gene duplication, and they share 75% identity, and they are functionally far more similar to each other than to other members of the TRPV subfamily. This gives us confidence that our computationally determined PI(4,5)P2 binding site in TRPV6 likely reflects the actual PI(4,5)P2 binding site with a reasonable accuracy, even if there is no TRPV6 PI(4,5)P2 costructure available currently. Similarly, in the current work, we docked PI(4,5)P2 to the apo structure of TRPM8 (
). This makes us confident that our experimentally tested computational prediction of the PI(4,5)P2 binding site in TRPM3 reflects the functionally relevant PI(4,5)P2 binding site with reasonable accuracy.
In conclusion, our data provide mechanistic insight into regulation of TRPM3 by its key physiological cofactor, PI(4,5)P2. We identify its binding site on the channel, characterize the interaction between PI(4,5)P2 and other physiological regulators of TRPM3, and compare its regulation by PI(4,5)P2 to that of TRPM8.
Xenopus laevis oocyte preparation and RNA injection
All procedures of preparing X. laevis oocytes were approved by the Institutional Animal Care and Use Committee at Rutgers New Jersey Medical School. Frogs were anesthetized in 0.25% ethyl 3-aminobenzoate methanesulfonate solution (pH 7.4) (MS222; Sigma–Aldrich), then bags of ovaries were surgically collected, and rotated with 0.1 to 0.3 mg/ml type 1A collagenase (Sigma–Aldrich) at 16 °C overnight in OR2 buffer containing 82.5 mM NaCl, 2 mM KCl, 1 mM MgCl2, and 5 mM Hepes (pH 7.4). Afterward, oocytes were washed with OR2 several times and then kept in OR2 solution supplemented with 1.8 mM CaCl2, 100 IU/ml penicillin, and 100 μg/ml streptomycin at 16 °C.
To express exogenous proteins, RNA was microinjected into oocytes using a nanoliter-injector system (Warner Instruments). RNA was in vitro transcribed from the linearized pGEMSH vectors, which contained the complementary DNA clones for hTRPM3 (
) rat TRPM8, human M2 muscarinic (hM2) receptor, or Gβγ subunits by using the mMessage mMachine T7 Transcription Kit (Thermo Fisher Scientific). TRPM3 and TRPM8 mutants, which were used in this article, were generated by the QuikChange II XL Site-Directed Mutagenesis Kit from Agilent, and the mutated DNA constructs were confirmed by DNA sequencing. For coexpression of TRPM3 constructs and Gβ1γ2 subunits, 40 ng of TRPM3 was coinjected with 5 ng Gβ1 and 5 ng Gγ2. In the case of coexpressing TRPM3 and hM2 receptors, these two were injected at 1:1 ratio, 40 ng each. Oocytes were used for electrophysiological experiments after 48 to 72 h incubation at 16 °C after RNA injection.
Oocytes were placed in extracellular solution, which contained 97 mM NaCl, 2 mM KCl, 1 mM MgCl2, and 5 mM Hepes, pH 7.4. Currents were measured with a protocol consisting a voltage step from the 0 mV holding potential to −100 mV, followed by a ramp to 100 mV once every 0.5 s with a GeneClamp 500B amplifier and analyzed with the pClamp 9.0 software (Molecular Devices). Currents were recorded by thin wall glass pipettes that contained inner filament and were filled with 1% agarose in 3 M KCl. In all TEVC experiments, different concentrations of PregS were applied to activate TRPM3 channels, and various concentrations of menthol were used to trigger responses of TRPM8 channels. The hM2 receptor was activated by 5 μM acetylcholine. For wortmannin experiments specifically, PregS, or menthol-induced currents were measured, then the same oocyte was incubated with 35 μM wortmannin for 2 h, and currents were measured again using the same protocol. In the bar graphs in Figure 3, the individual panels show experiments that were performed on the same day.
Excised inside–out patch clamp electrophysiology
Oocytes were placed in a recording chamber filled with bath solution, which contained 97 mM KCl, 5 mM EGTA, 10 mM Hepes, pH 7.4. Before starting measurements, the vitelline layer was carefully removed with forceps without damaging the oocyte. Then a giga-ohm seal was formed using a borosilicate glass pipette (World Precision Instruments) with resistance from 0.8 to 1 MΩ. The pipette was filled with a solution containing 97 mM NaCl, 2 mM KCl, 1 mM MgCl2, 5 mM Hepes, and 100 μM PregS at pH 7.4. Currents were measured by an Axopatch 200B amplifier and analyzed with the pClamp 9.0 software. Compounds were dissolved in the bath solution and delivered to the inner side of cell membrane by a custom-made gravity-driven perfusion system. Either 25 μM PI(4,5)P2, 25 μM PI(4)P, or 10 μM AASt PI(4,5)P2 was applied in these experiments to reactivate TRPM3. At the end of every recording, 30 μg/ml Poly-Lys (Poly-K) was applied.
Maintenance and transfection of human embryonic kidney 293 cells
Human embryonic kidney 293 cells were purchased from American Type Culture Collection (catalog number: CRL-1573). Human embryonic kidney 293 cells were cultured in minimum essential medium supplemented with 10% fetal bovine serum and 100 IU/ml penicillin plus 100 μg/ml streptomycin. Cells were incubated in 5% CO2 at 37 °C. Cells were tested to confirm that they were not infected by mycoplasma. Cells were used up to 25 passages and then discarded. Cells were transiently transfected with complementary DNA encoding different TRPM3 constructs (200–400 ng) using the Effectene reagent (Qiagen). mTRPM3α2 and its mutant were cloned into the bicistronic pCAGGS/IRES-GFP vector. The components of rapamycin-inducible pseudojanin phosphatases (
After 24 h of transfection, cells were plated on poly-d-lysine–coated 12 mm cover slips. Experiments were performed 48 to 72 h after transfection. Coverslips were placed in recording chamber filled with extracellular solution (137 mM NaCl, 5 mM KCl, 1 mM MgCl2, 10 mM Hepes, and 10 mM glucose, pH 7.4). Since mTRPM3 constructs were in the background of bicistronic pCAGGS/IRES-GFP vector and rapamycin-inducible phosphatases were labeled with red florescent protein, cells that showed both GFP and red florescent protein fluorescence were selected for the whole-cell patch-clamp experiments. Patch pipettes were prepared from borosilicate glass capillaries (Sutter Instruments) using a P-97 pipette puller (Sutter Instrument) with a resistance of 2 to 4 MΩ. Those recording pipettes were filled with intracellular solution containing 140 mM potassium gluconate, 5 mM EGTA, 1 mM MgCl2, 10 mM Hepes, and 2 mM Na-ATP, pH 7.4. After formation of gigaohm-resistance seals, the whole cell configuration was established, and currents were recorded by applying a ramp protocol once every 1 s. The holding potential was 0 mV; followed by a −100 mV step for 100 ms; plus a ramp protocol from −100 mV to +100 mV over the period of 500 ms. All recordings were made with an Axopatch 200B amplifier, filtered at 5 kHz, and digitized through a Digidata 1440A interface. Data were collected and analyzed with the pClamp10.6 (Clampex) acquisition software (Molecular Devices) and further analyzed and plotted with Prism 9 (GraphPad by Dotmatics). TRPM3 channels were activated by PregS, and 100 nM of rapamycin was applied to activate phosphatases.
Statistical analysis was performed with Origin 2021 and GraphPad Prism 9. Data were plotted as mean ± SEM and scatter plots or mean ± SD when scatter plots are not provided. Sample sizes were not predetermined by any statistical method; however, they were similar to what is generally used in the field. All recordings were performed in random order. Statistical significance was evaluated with t test, or ANOVA with Bonferroni’s post hoc test, or the Kolmogorov–Smirnov nonparametric test, using GraphPad Prism 9, as described in the figure legends. p Values are reported in figures or figure legends.
TRPM8 experimental structure refinement and molecular docking of full-length PI(4,5)P2
The cryo-EM structure of full-length apo TRPM8 from Ficedula albicollis (PDB ID: 6BPQ) (
), which contains several unresolved amino acid ranges (∼4.1 Å resolution) as well as protein residues with missing atoms, was used as the starting configuration to generate a refined structural model of the TRPM8 channel. The Prime Loop Prediction (
) (both distributed by Schrödinger, LLC, 2018) were used to perform the following tasks: (1) loop refinement by serial loop sampling, at the ultraextended accuracy level. In particular, four unresolved amino acid ranges in the transmembrane region were sampled, including 714 to 722, 819 to 822, 889 to 895, and 976 to 990 (sequence numbering as in F. albicollis); (2) side-chain prediction of protein residues with missing atoms, performed with no backbone sampling; (3) pKa prediction of protein residues at pH 7, followed by analysis and optimization of hydrogen-bond networks; (3) structure refinement via restrained minimization of heavy atoms (hydrogens not restrained) using the OPLS (
) force field. The minimization convergence criterion was set to 0.30 Å RMSD for heavy atom displacement. The resulting apo TRPM8 structure was then searched for putative ligand-binding sites using SiteMap (
) (Schrödinger, LLC, 2018) was used to dock PI(4,5)P2 against TRPM8, using a rigid-receptor and flexible-ligand protocol. The ligand was prepared by using the default protocol of LigPrep (Schrödinger, LLC, 2018). Binding modes were ranked using the Glide standard precision scoring function. The best binding mode of PI(4,5)P2 against TRPM8 is shown in Figure 2. After our refined TRPM8–PI(4,5)P2 complex was generated and used for subsequent modeling of the TRPM3 channel as in a complex with PI(4,5)P2, seven additional experimental structures of TRPM8 became available (Table S1). Three of these structures report the TRPM8 channel in complex with PI(4,5)P2 as well as Ca2+ ions and/or small-molecule ligands (
TRPM3 homology model and molecular docking of a truncated PI(4,5)P2 molecule
No experimental structure of the TRPM3 channel is currently available. The cryo-EM structure of the TRPM4 channel (3.1 Å resolution) in the apo state with short coiled coil from Mus musculus (PDB ID: 6BCJ) (
) (https://swissmodel.expasy.org/), based on the human sequence UniProtKB: Q9HCF6. The choice of the template is exemplified in Scheme S1. Essentially, the closest relative to TRPM3 in the TRPM family (cladogram) with an available structural template was selected, that is, TRPM4 (
). Hence, the TRPM3 protein residues facing the most “druggable” binding spot were selected by homology and used to center the docking grid for subsequent docking of PI(4,5)P2. The best binding mode of a truncated version of PI(4,5)P2 against TRPM3 is shown in Figure 1. As a matter of fact, because of the extreme flexibility of the lipid tail, the docking algorithm failed in generating binding poses for the full-length PI(4,5)P2 lipid. Instead, starting from the PI(4,5)P2 headgroup, a series of truncated versions of a growing lipid were docked successfully against the binding site on TRPM3 until a maximum tail length was reached (our truncated lipid is similar to the synthetic diC8 PI(4,5)P2 molecule, which is experimentally functional in activating TRPM3 (
)). For simplicity, in this work, the PI(4,5)P2 lipid with truncated tails, which was modeled in complex with TRPM3, is referred to as PI(4,5)P2. Note that in for TRPM8, the PI(4,5)P2 molecule was modeled as a full-length lipid. As for the molecular docking, we used the same protocol implemented for TRPM8. Related figures were generated using the Visual Molecular Dynamics (VMD) molecular visualization program (
A number of structural alignments were performed to compare TRPM3 and TRPM8 structures, including models (TRPM3 and TRPM8) and experimental structures (TRPM8). Superposition of the atomic coordinates was all performed based on sequence alignments (using the algorithm Needelman–Wunsch with BLOSUM-62 matrix). Alignments were generated using the Match Maker tool in UCSF Chimera (
), version 1.15, and analyzed in VMD. A number of structural alignments were performed, described as follows. (1) The model of TRPM8 in complex with (full length) PI(4,5)P2 and that of TRPM3 in complex with (truncated) PI(4,5)P2 was aligned. (2) The TRPM8/PI(4,5)P2 model and the experimentally determined structure of TRPM8 in complex with the menthol analog WS-12 and PI(4,5)P2 (PDB ID: 6NR2), and the complex of TRPM8 with icilin (PDB ID: 6NR3), PI(4,5)P2, and calcium (
). Pairwise backbone RMSD values were calculated for two separate selections (Table 1), including amino acid ranges facing the lipid binding sites, using the VMD RMSD Trajectory Tool. Before RMSD was calculated, structures were aligned on each selection. The first selection (sel-1 in Table 1) included residues 670 to 685 (on pre-S1), residues 724 to 735 (on pre-S1), and residues 851 to 865 (on linker). The second selection (Sel-2 in Table 1) included residues 670 to 685 (on pre-S1), residues 724 to 735 (on pre-S1), residues 851 to 865 (on linker), and residues 997 to 1009 (on TRP domain). (3) The following structures were aligned to the TRPM3–PI(4,5)P2 model: the experimental structure of TRPM4 (
). At the time of writing, four AlphaFold models were available of TRPM3, each from a different organism (UniProt sequence ID: Q9HCF6 [human; Fig. S1], J9S314 [M. musculus], F1QYX6 [Danio rerio], and F1LN45 [Rattus norvegicus]). All four structures were superimposed (not shown), revealing striking structural similarities. All figures related to (1) to (3) were generated using VMD.
Changes in the binding affinity (or Gibbs free energy of binding, ΔΔG in kcal/mol) of PI(4,5)P2 to TRPM3 were calculated upon mutating key binding residues in the putative PI(4,5)P2 binding site. These residues were also mutated experimentally. To do so, a physics-based scoring was employed (
). Essentially, residue mutations and ΔΔG calculations were performed on the TRPM3 model bound to truncated PI(4,5)P2 molecules as generated from molecular docking, that is, the native structural complexes or WT. A total of three binding modes were used as native configurations, including the TRPM3 model bound to the truncated PI(4,5)P2 presented in this study, and two additional poses with different protonation states of the phospholipid headgroup and even shorter tails. Then, the “Residue-Scanning and Mutation” tool from BioLuminate (
) (Schrödinger, LLC, 2018) was used to perform calculations upon mutating the native system, as described in Table S2. For each of the three WT proteins, six additional mutants were generated, reaching a total of 21 systems. For each system, and for both the WT and the mutant, an MM/GBSA refinement of the bound and unbound states was performed using Prime (Schrodinger, LLC, 2018), via the VSGB 2.0 implicit continuum solvation model (
). No implicit membrane model (i.e., a low-dielectric slab) was used, and therefore, the results should be regarded as an approximation to the electrostatic energy.
The structural complexes were refined by side-chain prediction with backbone sampling/minimization of the mutated residue, before a minimization in the region around the mutation site was performed to relax and optimize the side-chain interactions with the lipid. Systems were prepared for the calculations using the Protein Preparation Wizard (
The change in the binding affinity (the net free energy difference) was calculated by addressing the free energy changes in vertical lines, easier to simulate than the experimental observables (horizontal lines).
A positive value of indicates that the WT binds better than the mutant. Affinity changes were plotted using Microsoft Excel (https://www.microsoft.com/). Related figures were generated using VMD.
All data are contained in the article and supporting information. The structural model of TRPM3 in complex with a PI(4,5)P2 phospholipid with short tails is also available as a supporting information file. The authors request that any published work derived from the use of such data include a reference to this publication.
The authors declare that they have no conflicts of interest with the contents of this article.
The hTRPM31325 clone in a mammalian expression vector was provided by C. Harteneck (Eberhard Karls University Tubingen, Tubingen, Germany). The mTRPM3α2 clone was a kind gift from Drs Veit Flockerzi and Stephan Phillipp, and the rat TRPM8 clone was provided by Dr David Julius. The human muscarinic M2 receptor and the Gβ1 and Gγ2 clones were provided by Dr Diomedes Logothetis. This research includes calculations carried out on High Performance Computing resources at Temple University supported in part by the National Science Foundation through major research instrumentation grant number 1625061 and by the US Army Research Laboratory under contract number W911NF-16-2-0189 . Molecular graphics and analyses performed with UCSF Chimera, developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, with support from the National Institutes of Health P41-GM103311 .
S. Z., V. C., E. G., and T. R. conceptualization; S. Z. and E. G. formal analysis; S. Z., M. G., and E. G. investigation; S. Z., V. C., E. G., and T. R. writing–original draft; S. Z., V. C., M. G., E. G., and T. R. writing–review & editing; S. Z. and E. G. visualization; V. C. and T. R. supervision; V. C. and T. R. funding acquisition.
Funding and additional information
This study was supported by grants NSNS055159 (to T. R.) and GM131048 (to V. C. and T. R.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Note added in proof
After our manuscript was accepted, a paper describing structures of TRPM3 determined by cryo-EM, was published, showing PI(4,5)P2 in essentially the same location as predicted in our study: Zhao, C., and MacKinnon, R. (2022) Structural and functional analyses of a GPCR-inhibited ion channel TRPM3 Neuron, online ahead of press.