Structural mechanism of allosteric activation of TRPML1 by PI(3,5)P2 and rapamycin

Significance Rapamycin is a specific inhibitor of mammalian target of rapamycin (mTOR). Rapamycin can also activate transient receptor potential mucolipin 1 (TRPML1), a phosphatidylinositol 3,5-bisphosphate [PI(3,5)P2]–gated lysosomal cation channel whose loss-of-function mutations directly cause mucolipidosis type IV disease. We determined the high-resolution cryoelectron microscopy structures of TRPML1 in various ligand-bound states, including the open TRPML1 in complex with PI(3,5)P2 and a rapamycin analog at 2.1 Å. These structures reveal how rapamycin and PI(3,5)P2 bind at two distinct sites and allosterically activate the channel. Considering the high potency of TRPML1 activation by rapamycin and PI(3,5)P2, it is conceivable that some pharmacological effects from the therapeutic use of rapamycin may come from the TRPML1-dependent mechanism rather than mTOR inhibition.

used as the initial reference. Classes that showed clear features of the TRPML1 channel were combined and subjected to 3D auto-refinement and another round of 3D classification without performing particle alignment using a soft mask around the protein portion of the density. C4 symmetry was imposed in these steps. The best resolving classes were then re-extracted with the original pixel size and further refined. Beam tilt, anisotropic magnification, and per-particle CTF estimations, and Bayesian polishing were performed in Relion 3.1 to improve the resolution of the final reconstruction.
For the dataset of apo structure using the protein sample prepared in the presence of Tem alone, a total of 4,929 movies were collected and 4,745 were selected after motion correction and CTF estimation. A total number of 753,663 particles were extracted from the selected micrographs and were subjected to one round of 2D classification, from which 147,094 particles were selected. After the initial 3D classification, 47,179 particles were selected and subjected to a 3D auto-refinement job. Next, a soft mask excluding the micelle density was applied and particles were sorted into 4 classes without performing the alignment. From this, one class with 33,553 particles was selected and further refined, yielding a map at 2.6Å overall resolution ( Figure S2).
For the dataset of PI(3,5)P 2 -bound structure obtained using the protein sample prepared in the presence of PI(3,5)P 2 alone, a total of 6,498 movies were collected and 6,474 were selected after motion correction and CTF estimation. A total number of 1,150,065 particles were extracted from the selected micrographs and were subjected to one round of 2D classification, from which 469,210 particles were selected. After the initial 3D classification, 139,005 particles were selected and subjected to a 3D autorefinement job, yielding a map at 2.6Å overall resolution ( Figure S3).
For the dataset of protein sample prepared in the presence of PI(3,5)P 2 and Tem, a total of 5,071 movies were collected and 4,791 were selected after motion correction and CTF estimation. A total number of 1,262,471 particles were extracted from the selected micrographs and were subjected to one round of 2D classification, from which 638,713 particles were selected. After the initial 3D classification, 266,346 particles were selected and subjected to a 3D auto-refinement job. Next, a soft mask excluding the micelle density was applied and particles were sorted into 5 classes without performing the alignment. From this, two conformations of the channel with 67,182 and 142,902 particles for PI(3,5)P 2 -bound and PI(3,5)P 2 /Tem-bound states, respectively, were selected for further refinement. The particles of PI(3,5)P 2bound class were then classified with a soft mask around the PI(3,5)P 2 ligand density with regularisation parameter T=8, and 53,514 particles were selected from this classification. These particles were then refined and yielded a density map at an overall resolution of 2.4 Å. The particles of PI(3,5)P 2 /Tem-bound class were further refined in Relion and yielded a map with an overall resolution of 2.1 Å ( Figure S4).
For the dataset of ML-SA1-bound structure, a total of 5,163 movies were collected and 5,061 were selected after motion correction and CTF estimation. A total number of 831,409 particles were extracted from the selected micrographs and were subjected to one round of 2D classification, from which 309,621 particles were selected. After the initial 3D classification, 164,721 particles were selected and subjected to a 3D auto-refinement job. Next, a soft mask excluding the micelle density was applied and particles were sorted into 5 classes without performing the alignment. From this, two classes (with a total number of 41,980 particles) of the channel were selected and further refined, yielding a map at 2.3Å overall resolution ( Figure S5).

Model building, refinement, and validation
The structure of mouse TRPML1 (PDB code: 5WPV) was used as the initial model and was manually adjusted in Coot 9 and refined against the map by using the real-space refinement module with secondary structure and non-crystallographic symmetry restraints in the Phenix package 10 . The final structure models include residues 40-198, 215-285 (or 215-291), and 296-527. About 40 residues at the amino terminus and 50 residues at the carboxy terminus are disordered and not modeled. The statistics of the geometries of the models were generated using MolProbity 11 . All the figures were prepared in PyMol (Schrödinger, LLC.), UCSF Chimera 12 . Pore radii were calculated using the HOLE program 13 .

Electrophysiology
The N-terminal GFP tagged, plasma membrane-targeting TRPML1 mutant (TRPML1-4A) 14,15 was overexpressed in HEK293 cells and the channel activities were directly measured by patching the plasma membrane. In this setting, the extracellular side is equivalent to the luminal side of TRPML1 in endosomes or lysosomes. 48 h after transfection, cells were dissociated by trypsin treatment and kept in serumcontaining complete medium; the cells were re-plated onto 35 mm tissue culture dishes and kept in a tissue culture incubator until recording. Patch clamp in the whole-cell or inside-out configuration was used to measure TRPML1 activity on the HEK plasma membrane. The standard bath solution for whole-cell current recording contained (in mM): 145 sodium methanesulfonate, 5 NaCl, 1 MgCl 2 , 10 HEPES buffered with Tris, pH 7.4; and the pipette solution contained (in mM): 140 cesium methanesulfonate, 5 NaCl, 5 MgCl 2 , 10 EGTA, 10 HEPES buffered with Tris, pH 7.4. The bath solution for inside-out configuration contained (in mM): 140 potassium methanesulfonate, 5 NaCl, 2 MgCl 2 , 0.4 CaCl 2 , 1 EGTA, 10 HEPES buffered with Tris, pH 7.4; and the pipette solution contained (in mM): 145 sodium methanesulfonate, 5 NaCl, 1 MgCl 2 , 0.5 EGTA, 10 HEPES buffered with Tris, pH 7.4. For the whole-cell recording of PI(3,5)P 2 -activated channel, we had to include a high concentration of PI(3,5)P 2 (100 μM) in the pipette solution (cytosolic side) to quickly obtain a stable PI(3,5)P 2 -evoked current, likely because of the slow diffusion of this lipid ligand. The patch pipettes were pulled from Borosilicate glass and heat polished to a resistance of 2-5 MΩ (2-3 MΩ for inside-out patch, and 3-5 MΩ for whole-cell current recording). Data were acquired using an AxoPatch 200B amplifier (Molecular Devices) and a low-pass analog filter set to 1 kHz. The current signal was sampled at a rate of 20 kHz using a Digidata 1550B digitizer (Molecular Devices) and further analyzed with pClamp 11 software (Molecular Devices). After the patch pipette was attached to the cell membrane, the giga seal (>10 GΩ) was formed by gentle suction. The inside-out configuration was formed by pulling the pipette away from the cell, and the pipette tip was exposed to the air for 2 seconds. The whole-cell configuration was formed by a short zap or suction to rupture the patch. The holding potential was set to 0 mV. The whole-cell and inside-out macroscopic current recordings were obtained using voltage pulses ramped from −140 mV to +50 mV over 800 ms. The sample traces for the I-V curves of macroscopic currents shown in each figure were obtained from recordings on the same patch. All data points are mean ± s.e.m. (n ≥ 5).