Visualizing structural transitions of ligand-dependent gating of the TRPM2 channel

The transient receptor potential melastatin 2 (TRPM2) channel plays a key role in redox sensation in many cell types. Channel activation requires binding of both ADP-ribose (ADPR) and Ca2+. The recently published TRPM2 structures from Danio rerio in the ligand-free and the ADPR/Ca2+-bound conditions represent the channel in closed and open states, which uncovered substantial tertiary and quaternary conformational rearrangements. However, it is unclear how these rearrangements are achieved within the tetrameric channel during channel gating. Here we report the cryo-electron microscopy structures of Danio rerio TRPM2 in the absence of ligands, in complex with Ca2+ alone, and with both ADPR and Ca2+, resolved to ~4.3 Å, ~3.8 Å, and ~4.2 Å, respectively. In contrast to the published results, our studies capture ligand-bound TRPM2 structures in two-fold symmetric intermediate states, offering a glimpse of the structural transitions that bridge the closed and open conformations.


Abstract
The calcium-permeable transient receptor potential melastatin 2 (TRPM2) channel plays a key role in redox sensation in many cell types 1-3 . Channel activation requires binding of both ADP-ribose (ADPR) 2,4-6 and Ca 2+ 7 . The recently published TRPM2 structures from Danio rerio in the ligand-free and in the ADPR/Ca 2+ -bound conditions represent the channel in closed and open states, which uncover substantial tertiary and quaternary conformational rearrangements 8 .
However, it is unclear how these rearrangements occur within the tetrameric channel during channel gating. Here we report two cryo-electron microscopy structures of TRPM2 from the same species in complex with Ca 2+ alone, and with both ADPR and Ca 2+ , determined to an overall resolution of ~3.8 Å and ~4.2 Å respectively. In comparison with the published results, our studies capture TRPM2 in two-fold symmetric intermediate states, offering a glimpse of the structural transitions within the tetramer that bridge the closed and open conformations.

Main
The transient receptor potential melastatin (TRPM) ion channel family is part of the TRP channel superfamily and comprises eight members (TRPM1 to TRPM8) that carry out diverse functions in a variety of physiological pathways 9 . TRPM2 is a calcium-permeable non-selective cation channel that is widely expressed in the nervous, immune, and endocrine systems, and plays a crucial role in warmth and redox-dependent signaling 1-3 . Studies of TRPM2-deficient mice have shown that TRPM2 channels expressed in sensory and central neurons are responsible for sensation of warm temperatures and body temperature regulation 10,11 . TRPM2 has also been found to be activated by reactive oxygen species (ROS), consequently playing a key role in Ca 2+ signaling involved in chemokine production, insulin production, and cell death under oxidative stress [12][13][14][15][16] .
Extensive electrophysiological studies have shown that activation of TRPM2 by ROS is caused by an increase of intracellular ADP-ribose (ADPR), a TRPM2 agonist, and that both ADPR and Ca 2+ are required for channel activation 2,[4][5][6][7]17 . Notably, TRPM2 contains a putative enzyme domain, called the Nudix Hydrolase 9 Homology (NUDT9H) domain located at the C-terminus, which exhibits a high degree of homology to the mitochondrial ADP-ribose pyrophosphatase NUDT9 18,19 . As such, TRPM2 was initially classified as a channel-enzyme in which enzymatic activity was believed to be coupled to channel gating 2,18 . However, subsequent studies have since repudiated this mechanism, as the NUDT9H domain lacks enzymatic activity. Instead it has been suggested that the NUDT9H domain merely serves as a binding site for ADPR 20 .
Recently, structures of the zebrafish Danio rerio TRPM2 were reported in the ligand-free closed state and in the ADPR/Ca 2+ -bound open state (hereafter referred to as TRPM2 DR_closed and TRPM2 DR_open respectively) 8 . This study not only revealed the location of the NUDT9H domain, but also showed that the Ca 2+ ion binds in the cavity formed by the voltage-sensor like domain (VSLD) comprising transmembrane segments 1-4 (S1-S4). Unexpectedly, ADPR was found to bind in the cleft of the melastatin homology domain 1/2 (MHR1/2) and not, as had long been predicted, in the NUDT9H domain. Structural analyses of the closed and open conformations have provided a mechanism for ligand-dependent activation of the channel, wherein binding of ADPR in the MHR1/2 domain triggers substantial conformational rearrangements in the cytoplasmic domain (CD), which are sequentially transduced and propagated to the distal transmembrane channel domain (TMD) to induce pore opening.
In spite of this progress towards understanding ligand-dependent activation of the TRPM2 channel, it remains unclear whether these drastic quaternary structural rearrangements occur in a concerted, four-fold symmetric manner. We recently reported that the transient receptor potential vanilloid 2 (TRPV2) channel adopts two-fold symmetric conformations upon resiniferatoxin (RTx)-mediated activation 21 , and that two-fold symmetric states are associated with liganddependent gating of the TRPV3 channel 22 . It is presently unclear if reduced symmetry is associated with gating in other TRP channel families. To address these questions, we determined the cryoelectron microscopy (cryo-EM) structures of the full-length TRPM2 from Danio rerio (TRPM2 DR ) in the presence of Ca 2+ (referred to as TRPM2 DR_Ca2+ ) and in the presence of both ADPR and Ca 2+ (referred to as TRPM2 DR_ADPR/Ca2+ ). Our structural analyses identify unusual quaternary structural rearrangements in the channel assembly which likely represent intermediate gating states.
Moreover, a comparison with the published TRPM2 closed and TRPM2 open structures enabled us to speculate on the conformational pathway involving reduced symmetric rearrangements between the closed and open states of TRPM2.

Structure determination and overall architecture of TRPM2 DR
TRPM2 DR shares approximately 50% sequence identity with human TRPM2 ( Supplementary Fig. 1), and when overexpressed in human embryonic kidney 293T (HEK293T) cells, wild-type TRPM2 DR channels exhibit large currents in response to direct application of ADPR and Ca 2+ to the cytosolic side of inside-out patches ( Supplementary Fig. 2) Table 1). In our study, we observed density corresponding to two unique structural features of the TRPM2 channel: a second coiled-coil and the NUDT9H domain at the C-terminus of the channel (Fig. 1). However, since these regions were the least well-defined in the EM maps, we facilitated model building and analysis by docking the homology model of the crystal structure of human ADP-ribose pyrophosphatase NUDT9 (PDB 1Q33) 19 into the NUDT9H density (see methods).
Our TRPM2 DR channel forms a homo-tetramer (Fig. 1a,b). Viewed orthogonally to the membrane plane, the channel can be divided into four layers. Like other published TRP channel structures, the TMD layer adopts a domain-swapped configuration between the VSLD and the pore 24,25 . Consistent with the previously reported TRPM2 structures 8,26,27 , we also identified a Ca 2+ binding site located in the cavity formed by the VSLD in the TRPM2 DR_Ca2+ and 6 TRPM2 DR_ADPR/Ca2+ structures ( Supplementary Fig. 6a-c). While the TMD and two membraneproximal CD layers resemble the architecture observed in the TRPM8 and TRPM4 structures 28-31 , the additional second coiled coil (CC2) and the NUDT9H domain together comprise a unique bottom CD layer in the TRPM2 structures (Fig. 1c). Similar to the previously reported TRPM2, TRPM4, and TRPM8 structures 8,26-31 , each TRPM2 DR protomer contains an N-terminal region composed of MHR1 to MHR4, a transmembrane channel region, and a C-terminal region. The CC2 serves as a link between the C-terminal NUDT9H domain and the rest of the channel (Fig.   1d).

TRPM2 DR structures in the intermediate states adopt two-fold symmetry
In contrast to the canonical four-fold symmetry reported in the TRPM2 closed and TRPM2 open structures, our TRPM2 DR_Ca2+ and TRPM2 DR_ADPR/Ca2+ structures adopt a two-fold symmetric arrangement when viewed along the central axis of the channel. This C2 symmetry was readily apparent upon reference-free 2D classification of the particle images (Supplementary Figs. 3e and 4e). To illustrate the key features associated with the observed two-fold symmetry, we focus first on the TRPM2 DR_Ca2+ structure, which exhibits different magnitudes of two-fold symmetry across the layers of the channel. This reduced symmetry is most pronounced in the middle layer of the CD, comprising the MHR1/2 and MHR3 domains ( Fig. 2a and Supplementary Fig. 7). Notably, comparisons with the published TRPM2 closed and TRPM2 open structures show that protomer A of our TRPM2 DR_Ca2+ structure resembles the closed conformation while protomer B approximates the open conformation (Fig. 2b,c). This observation suggests that our TRPM2 DR_Ca2+ structure may represent an intermediate state in which neighboring subunits adopt closed and the open conformations in an alternating manner, resulting in the two-fold symmetry between diagonally opposing subunits.

Domain rearrangement at flexible junctions
To identify the origin of the two-fold symmetry, we performed global and local alignments of protomers A and B in the TRPM2 DR_Ca2+ structure. The MHR3, MHR4, and VSLD are conformationally similar between protomers, whereas significant differences in domain arrangement are observed at three junctions: NUDT9H-MHR1/2, MHR1/2-MHR3, and VSLDpore ( Fig. 3a,b). When the MHR3 and MHR4 domains of protomers A and B are aligned, the MHR1/2 and NUDT9H domains exhibit a drastic rotation along individual axes (Fig. 3c).
Moreover, this movement within the CD is propagated to the TM helices, resulting in two-fold symmetry within the TMD (Fig. 3d). The TMDs of the protomers A and B diverge at the S4b and the S4-S5 linker regions. In protomer A, the S4b adopts a 3 10 -helical structure, while the equivalent region in protomer B contains an a-helical structure and an unstructured loop. In addition, the S4-S5 linker in protomer A contains a p-helix, while the corresponding region in protomer B is ahelical and forms a continuous straight helix with S5. Due to the absence of a bend-inducing phelix at the junction between the S4-S5 linker and S5, the entire pore domain of protomer B is positioned differently from that of protomer A with respect to the central axis of the channel. Therefore, the flexible elements in the S4b and the S4-S5 linker lead to a two-fold symmetric configuration of the TMD. Notably, a similar two-fold symmetric TMD arrangement induced by a conformational change in the S4-S5 linker was recently observed in the TRPV2 channel 21 . The substantial conformational change at the S4b and S4-S5 linker between adjacent protomers leads to unexpectedly distinct configurations of the S6 gate. Although the distance between diagonally opposed gate residues in TRPM2 DR_Ca2+ are larger than those observed in the TRPM2 closed structure, the S6 gate in TRPM2 DR_Ca2+ is not as wide as that observed in the TRPM2 open structure ( Supplementary Fig. 8a). Taken together, the structural differences between protomers A and B in TRPM2 DR_Ca2+ originate at the flexible junctions located between NUDT9H-MHR1/2, MHR1/2-MHR3, and VSLD-pore. This flexibility allows the channel to assume a two-fold symmetric

Alternating quaternary structure rearrangement in the bottom CD layer
Based on these structural comparisons, we suggest that the four subunits of TRPM2 channel adopt a two-fold symmetric intermediate quaternary structure assembly before ultimately assuming the canonical four-fold symmetric arrangement in the open conformation. To visualize the conformational changes associated with reduced symmetry, we compared our structures with the published TRPM2 closed and TRPM2 open structures at the bottom layer of the CD, which is comprised of the mobile MHR1/2 and NUDT9H domains, and identified substantial rearrangements mediated by critical interfacial interactions (Fig. 5).
It was shown that in the apo conformation, the NUDT9H domain in the TRPM2 closed structure merely makes primary intra-subunit interactions with the MHR1/2 domain (Fig. 5a,d), while the binding of ADPR in the MHR1/2 domain relocates the NUDT9H domain to generate a secondary interface with the MHR1/2 domain from the neighboring protomer ( Fig. 5c,g).

Subunit-subunit interfaces in the middle CD layer
The interfaces between the MHR1/2 and MHR3 domains in adjacent subunits comprise the middle layer of the CD, which undergoes the largest conformational changes between the closed and the open state upon ADPR binding. Notably, the reduced symmetry is most pronounced in the middle layer of the CD in the TRPM2 DR_Ca2+ structure (Fig. 2a)  facilitating the channel gating (Fig. 6).

Discussion
In this study, we observe a two-fold symmetric arrangement in the homo-tetrameric channel TRPM2 in complex with only Ca 2+ (TRPM2 DR_Ca2+ ), and with both ADPR and Ca 2+ (TRPM2 DR_ADPR/Ca2+ ), a feature previously observed in the TRPV2 and TRPV3 channels 21,22 .
Interestingly, comparing our structures with the recently published four-fold symmetric We consider the possible reason for the unique two-fold symmetric arrangement observed in our TRPM2 DR structures may lie in the biochemical preparation of our samples. While the construct (full-length wild type TRPM2 DR ), expression system (HEK293 cells), and detergent (digitonin) we used were very similar to those of the previously published structures 8 , our biochemical preparation also included Ca 2+ . Notably, our TRPM2 DR_Ca2+ structure is the only structure of TRPM2 to date that was determined in the presence of Ca 2+ only, thus enabling us to dissect the role of Ca 2+ in TRPM2 gating. It is possible that Ca 2+ binding in the VSLD confers flexibility to the junction at the S4-S5 linker, thus priming the channel for opening. Consistent with this idea, the density for Ca 2+ in the TRPM2 DR_Ca2+ reconstruction is stronger in protomer B compared to that of protomer A (10.8 s versus 9.5 s, respectively) ( Supplementary Fig. 6a,b). In addition, we cannot exclude another possibility that endogenous ADPR was captured in the Ca 2+bound TRPM2 during sample preparation, as we also observe weak densities in the cleft of the MHR1/2 domain in protomers B and D of the TRPM2 DR_Ca2+ reconstruction, which may correspond to ADPR ( Supplementary Fig. 6d). However, the quality of the density in these regions was insufficient to unambiguously assign these detailed structural components.

Declaration of Interests
The authors declare no competing interests.        Individual pore domains are highlighted in black frames. The protomer configurations are indicated by type "A" or "B" and also depicted as cartoon diagrams in insets. Ca atoms of the P839 residues are shown as black dots, and distances between Ca atoms are indicated (Å).

References
i, Cartoon diagram of alternating quaternary structural rearrangements in TRPM2 channel gating.

Protein expression and purification
Zebrafish TRPM2 showed optimal biochemical stability based on a screen of eight orthologues including human and rat TRPM2. A codon-optimized and full-length gene for zebrafish (Danio rerio, XP_009303266.1) TRPM2 (TRPM2 DR ) was synthesized and cloned into a modified pEG BacMam vector 32 in frame with a C-terminal FLAG affinity tag (Bio Basic Inc.).
The full length wild-type protein was expressed by baculovirus-mediated transduction of To solve the TRPM2 DR_ADPR/Ca2+ structure in complex with both ADPR and Ca 2+ , after purifying the protein by size-exclusion chromatography as described above, peak fractions were combined and mixed with Amphipol PMAL-C8 (Avanti polar lipid) at 1:10 (wt/wt) ratio and incubated overnight at 4°C with gentle agitation. To remove detergent, 25 mg ml -1 Bio-Beads SM-reconstituted protein was mixed with 50 µM ADP-Ribose (Sigma-Aldrich) and further purified on a Superose 6 10/300 GL column (GE Healthcare) equilibrated with buffer C (20 mM Tris pH8, 150 mM NaCl, 0.5 mM CaCl 2 , 50 µM ADPR). The peak fractions were collected and concentrated to ~1.25 mg mL -1 and incubated with ADPR to a final concentration of ~1 mM for 30-45 min before sample freezing for cryo-EM study.

Cryo-EM sample preparation
For cryo-EM study of the TRPM2 DR_Ca2+ structure, a thin amorphous carbon film was floated onto UltrAuFoil® R1.2/1.3 300-mesh grids (Quantifoil). 3 µL of TRPM2 (0.5 mg/ml) was applied to freshly plasma cleaned grids, and the grids were manually blotted 33 using a custombuilt manual plunger in a cold room (≥95% relative humidity, 4 °C). Sample was blotted for ~4 s with Whatman No.1 filter paper immediately prior to plunge freezing in liquid ethane cooled by liquid nitrogen.
The cryo-EM grids for the analysis of the TRPM2 DR_ADPR/Ca2+ structure were prepared by applying 3 µL PMAL-C8 reconstituted TRPM2 (~1.25 mg mL -1 ) in the presence of ADPR and Ca 2+ to freshly glow-discharged UltrAuFoil® R1.2/1.3 300-mesh grids (Quantifoil). The grids were blotted for 2 s at 25 °C under 95% humidity before being plunged into liquid ethane using a Leica EM GP2 (Leica Microsystems). The grids were stored in liquid nitrogen before data acquisition.

TRPM2 DR_Ca2+ structure
The Cryo-EM data used for the TRPM2 DR_Ca2+ structure were acquired using the Leginon automated data acquisition program 34 . All image pre-processing (frame alignment, CTF estimation, particle picking) were performed in real-time using the Appion image processing pipeline 35 during data collection. Images of frozen hydrated TRPM2 were collected on a Talos Arctica (Thermo Fisher) TEM operating at 200 keV. Movies were collected using a K2 Summit direct electron detector (Gatan) in counting mode at a nominal magnification of 36,000x corresponding to a physical pixel size of 1.15 Å/pixel. A total of 3,039 movies (64 frames/movie) of TRPM2 were collected by navigating to the center of four holes and image shifting ~2 µm to each exposure target. Movies were collected using a 16 second exposure with an exposure rate of 5.2 e -/pixel/sec, resulting in a total exposure of ~63 e -/Å 2 (1.17 e -/Å 2 /frame) and a nominal defocus range from -1.2 µm to -2 µm. The MotionCor2 frame alignment program 36 was used to perform motion correction and dose-weighting as part of the Appion pre-processing workflow. Frame alignment was performed on 5 x 5 tiled frames with a B-factor of 100 applied. Unweighted summed images were used for CTF determination using CTFFIND4 37 . Difference of Gaussians (DoG) picker 38 was used to automatically pick particles from the first 636 dose-weighted micrographs yielding a stack of 176,961 particles that were binned 4 × 4 (4.6 Å/pixel, 80 pixel box size) and subjected to reference-free 2D classification using RELION 2.1 39 . The best nine classes were then used for template-based automated particle picking against the whole dataset using RELION. A total of 1,791,114 particles were extracted from these micrographs and binned 4 x 4 (4.6 Å/pixel, 80 pixel box size). Reference-free 2D classification in RELION was then used to sort out non-particles and poor-quality picks in the data. A total of 435,692 particles corresponding to 2D class averages that displayed strong secondary-structural elements were input to 3D autorefinement in RELION without symmetry imposed. EMD-7127 was low-pass filtered to 30 Å and used as an initial model. The refined particle coordinates were then used for re-centering and reextraction of particles binned 2 x 2 (2.3 Å/pixel, 160 pixel box size). The resulting stack was subjected to 3D auto-refinement using the map obtained from the previous refinement as an initial model, and with a soft mask (5 pixel extension, 5 pixel soft cosine edge) generated from a volume contoured to display the full density. These particles were then subjected to 3D classification (k=6, tau fudge=12) without angular or translational searches using the same soft mask. Particles contributing to the classes that possessed the best resolved densities around the transmembrane domain of the channel were 3D auto-refined and then re-centered and re-extracted without binning (135,215 particles, 1.15 Å/pixel, 320 pixel box size). These particles were then 3D auto-refined and subjected to 3D classification (k=3, tau fudge=12) without angular or translational searches.
Two classes, comprising 94,028 particles, displayed the best-resolved density around the transmembrane region and C2 symmetry. 3D auto-refinement of these particles with C2 symmetry imposed yielded a ~3.9 Å reconstruction as determined by gold-standard 0.143 Fourier shell correlation (FSC) 40 , using phase-randomization to account for the convolution effects of a solvent mask on the FSC between the two independently refined half maps 41 . To improve map quality, all particles collected at greater than 2 µm defocus were removed. 3D auto-refinement of the final particle stack (93,573 particles) with C2 symmetry imposed yielded a ~3.8 Å reconstruction as determined by gold-standard 0.143 FSC.

TRPM2 DR_ADPR/Ca2+ structure
For the TRPM2 DR_ADPR/Ca2+ structure, the cryo-EM data were acquired using the EPU automated data-acquisition program. Images were collected on a Titan Krios (FEI) operating at 300 keV equipped with a Falcon III direct electron detector operating in counting mode. 2496 movies were collected at a nominal magnification of 59,000x with a physical pixel size of 1.39 Å/pixel using a nominal defocus range of -0.5 to -2.25 µm. Each movie (45 frames) was acquired using a dose rate of ~0.91 e -/pixel/s and a total exposure of ~40 e -/Å 2 .
Motion correction and dose-weighting was performed using the MotionCor2 frame alignment program on 5 x 5 tiled frames with a B-factor of 150 applied 36 . Gctf 42 was used for CTF estimation of unweighted summed images and 2426 good micrographs were selected. 1719 particles were manually picked and subject to reference-free 2D classification (k=10, tau=2) in RELION 3.0 43 . The best seven classes were used as templates for auto-picking of a total of 1,949,622 particles. The dataset was extracted unbinned and subjected to reference-free 2D classification, construction of ab-initio model, heterogeneous refinement, and homogeneous refinement in CryoSPARC 44 , yielding a 3D reconstruction of ~4.3 Å resolution. In parallel, the 1,949,622 particles were extracted, Fourier binned 4 x 4 (5.56 Å/pixel, 64 pixel box size) and subjected to reference-free 2D classification. Good 2D classes showing secondary structural features were selected. After further removing micrographs with a figure of merit (FOM) below 0.1 and with astigmatism above 500 nm, a total of 736,565 particles from 2366 good micrographs were combined and input to 3D auto-refinement in RELION with C1 symmetry. The model of ~4.3 Å resolution generated by CryoSPARC was low-pass filtered to 30 Å and used as an initial model without a reference mask. The refined particles were re-centered, re-extracted, Fourier binned 2 x 2 (2.78 Å/pixel, 128 pixel box size), and subject to 3D auto-refinement with C1 symmetry, using the model from the previous auto-refinement as the reference mask and with a soft mask (5 pixel extension, 5 pixel soft cosine edge). The refined particles were input to 3D classification (k=4, tau=12) without alignment using the same mask. 135,120 particles comprising the best 3D class which shows the most well-defined cytoplasmic domain (CD) were re-centered, re-extracted, unbinned, and subject to 3D auto-refinement with C2 symmetry with a mask around the full density, yielding a final construction of ~4.2 Å resolution determined by gold-standard 0.143 Fourier shell correlation (FSC) using RELION 3.0 43 . Per-particle CTF refinement and Bayesian polishing 45 were attempted, but the resolution was not improved.
Cells between passage 10 -30 grown in 40-mm wells were transiently transfected at 30 -50% confluency with plasmids encoding for TRPM2 DR and green fluorescent protein (GFP) using Microscopy Data Bank with the accession number EMD-### and EMD-###.  are shown (e). Particles comprising the "best" class averages were 3D auto-refined without symmetry to yield a ~4.5 Å resolution reconstruction. Refined particle coordinates were used for re-centering and extraction of particles. Particles were Fourier binned by 2 x 2, followed by 3D

Supplementary
auto-refinement and no-alignment 3D classification. For each class, the number of contributing particles and percentage relative to total particles input to classification are listed below, respectively. 135,215 particles corresponding to the best-resolved classes were combined and 3D auto-refined to yield a ~4.6 Å resolution reconstruction. Refined particle coordinates were recentered and extracted unbinned, auto-refined, and subjected to no-alignment classification to obtain a subset of 94,028 particles. Particles collected at greater than 2 µm defocus were removed to obtain a final particle stack of 93,573 particles, which was auto-refined with C2 symmetry enforced to yield a final reconstruction at ~3.8 Å resolution (f). e, Particles comprising the good 2D classes were 3D auto-refined to a reconstruction of ~11.5 Å with C1 symmetry using the 3D reconstruction generated by cryoSPARC as the reference model.

Supplementary
Refined 736,565 particles were re-extracted, re-centered, Fourier binned 2 x 2, 3D auto-refined with C1 symmetry and subjected to no-alignment 3D classification. For individual class, the number of particles and the percentage relative the total number of particles input to the classification are listed. 135,120 particles comprising the 3D class, in which the cytoplasmic domain (CD) is most well-resolved, were re-centered, re-extracted, and unbinned, and were subjected to 3D auto-refinement with C2 symmetry, yielding a final reconstruction of ~4.2 Å.