Structure of intact human MCU supercomplex with the auxiliary MICU subunits

The mitochondrial Ca2+ uniporter (MCU) supercomplex is essential for mitochondrial Ca2+ uptake. Here, we present high-resolution cryo-EM structures of human MCU-EMRE supercomplex (MES, 3.41 Å) and MCU-EMRE-MICU1-MICU2 supercomplex (MEMMS, 3.64 Å). MES adopts a V-shaped dimer architecture comprising two hetero-octamers, and a pair of MICU1-MICU2 hetero-dimers form a bridge across the two halves of MES to constitute an O-shaped architecture of MEMMS. The MES and MEMMS pore profiles are almost identical, with Ca2+ in the selectivity filters and no obstructions, indicating both channels are conductive. Contrary to the current model in which MICUs block the MCU pore, MICU1-MICU2 dimers are located on the periphery of the MCU pores and do not occlude them. However, MICU1-MICU2 dimers may modulate MCU gating by affecting the matrix gate through the EMRE lever.


/ 26
MEMMS are linked at the IMS side via MICU1 and MICU2, and at the matrix side via MCU NTDs 183 (Fig. 1A). Each MICU1 and MICU2 subunit contains four EF-hands, of which two are capable of 184 binding Ca 2+ ions (39). However, in our structure, no Ca 2+ is bound to MICU1 or MICU2 subunits, 185 which can be attributed to the deprivation of Ca 2+ by EGTA during expression and purification (Fig. 186 3A). Alignment of MICU1 and MICU2 shows that they have very similar core structures (N-lobe and 187 C-lobe), but their N-terminal domains and C-terminal helices are different (Fig. 3A). The N-terminal 188 domain and Ser 339 , Lys 340 , Lys 341 sequence of MICU1 can interact with EMRE as discussed above. 189 MICU1 and MICU2 form a hetero-dimer in a previously reported 'face-to-face' pattern, while the two 190 MICU2 subunits interact in a 'back-to-back' pattern (40). Consequently, the N and C lobes of MICU1 191 and MICU2 subunits are arranged in an alternative pattern to link two MCU channels (Fig. 3,B and 192 C). 193 Previous models suggested that MICU1/2 dimer occludes the MCU pore (22,23,39,40) at low 194 cytosolic [Ca 2+ ], which is obviously not the case as shown by the MEMMS structure (obtained in the 195 presence of EGTA). The accompanying manuscript also reports that MICU subunits do not occlude 196 the MCU channel in low cytosolic [Ca 2+ ], using direct patch-clamp analysis of MCU currents (29). 197 In addition to the EMRE/MICU1 interactions, the C-terminal helices of both MICU1 and MICU2 also 198 contribute to MICU localization onto the inner membrane ( fig. S7). In the MEMMS structure, although 199 it's difficult to analyze the detailed interactions between these two helices due to the vague local density, 200 one can still appreciate that the two helices are parallel to each other at the surface of inner 201 mitochondrial membrane (Fig. 2D and fig. S3G). The C-terminal helices of MICU2 have hydrophobic 202 residues partially buried in the inner membrane, while the positively charged residues point parallel to 203 the membrane, interacting with the negatively charged phosphates of the membrane ( fig. S7). This is 204 in agreement with the previous reports that MICU1 and MICU2 directly interact with the lipid 205 membrane (12,41,42). Previous reports also show that the C-terminal helix is important for the 206 interaction of MICU1 with MES (43,44). Accordingly, deletion of MICU1 C-terminal helix 207 significantly weakened the binding of MICU1 to MCU, and even lowered Ca 2+ uptake activity (43, 208 44). Although the density map of this area is not clear enough for deciphering detailed interactions, we 209 find that MICU1 C-terminal helix is in close vicinity of the EMRE helix (Fig. 2D). In a previous study, 210 9 / 26 Co-IP assay showed that MICU2 ΔC could not interact with MICU1 or MCU (39). These findings are 211 consistent with the MEMMS structure, in which the C-terminal helices act as an anchor to maintain 212 MICU1 and MICU2 near to each other at the surface of inner mitochondrial membrane. 213 In the MES structure solved under high [Ca 2+ ], no electron density can be associated with MICU1 or 214 MICU2. We consider that the loss of MICUs is an artifact of the purification procedure. Since the 215 membrane was solubilized in detergent, the C-terminal helix could lose its attachment with the 216 membrane. In addition, when EF hands are occupied by Ca 2+ , conformational change of MICUs likely 217 makes them more vulnerable to dissociation, leading to loss of MICU1 and MICU2 in the MES 218 structure. Ca 2+ uptake in high [Ca 2+ ] condition was impaired in MICU1ΔC cells (44), which also 219 suggests MICU1 is very likely attached to MCU in high [Ca 2+ ]. In fact, in a previous report, interaction 220 between MCU and MICU1 in high [Ca 2+ ] was detected through co-IP (44). 221 In the matrix, the four NTDs align in a bent "fish-tail" configuration ( fig. S8A). Note that this fish-tail 222 alignment is in agreement with the previously reported MCU NTD crystal structure from human (45, are not identical for all three pairs, and the only polar interaction that occurs commonly for all three 228 pairs is between Asp 123 and Arg 93 . We mutated Asp 123 to Arg, or Arg 93 to Asp, respectively, and found 229 that both mutations have negligible influence on the Ca 2+ uptake rate ( fig. S8D). Thus, the other 230 interactions between the three NTD pairs could still hold the two MCU channels together in vivo after 231 these mutations. 232 233

Phospholipids and the matrix gate 234
In both MES and MEMMS structures, one CDL and one horizontal PLX molecules inserts into the 235 gap between TM1 and TM2 of each MCU subunit, and another vertical PLX molecule stands alongside 236 each TM2. Notably, most of the lipid chains of the CDL and vertical PLX in our structure are parallel 237 to the helices of TMD, while the lipid chains of the horizontal PLX are positioned horizontally in the 238 10 / 26 membrane ( Fig. 4A and fig. S6A). In a previous fungal structure (24), a horizontal PLX molecule was 239 also found in the wall of the MCU channel. However, when we compare these two structures, the 240 position of these two PLX molecules does not match. 241 In addition to interacting with TM1, TM2, and cc2 of one MCU subunit, each CDL molecule also 242 interacts tightly with the neighboring EMRE. Specifically, a highly conserved residue of MCU cc2, 243 Arg 297 , can form stable hydrogen bonds with both the phosphate group of CDL and the main chain 244 oxygen of Val 61 in EMRE (Fig. 4A), indicating Arg 297 could be a critical residue for MCU regulation. 245 This interaction (MCU-Arg 297 to EMRE-Val 61 ) is also observed in a previous human MES structure 246 (28), however, interacting phospholipids were not detected. In the same report, authors truncated the 247 N-terminal loop of EMRE, residue by residue until Lys 59 , and found that the Ca 2+ uptake activity of 248 MCU decreased gradually. Further truncation resulted in no EMRE expression, so the interaction 249 between Arg 297 and Val 61 was not tested (28). To supplement, we mutated Arg 297 to Asp, and strikingly 250 this mutation completely abolished the Ca 2+ uptake via MCU (Fig. 4B). Similarly, P60A mutation in 251 EMRE, just next to the MCU-Arg 297 , can also totally abolish MCU activity (47), adding importance 252 to correct interactions between cc2 and EMRE. 253 It has been proposed that cc2 and cc3 form a luminal gate near the matrix side of MCU that is 254 maintained in an open conformation via its interaction with EMRE (28, 48). The previous human MES 255 structure and our MES and MEMMS structures all detected stable hydrogen bonds between EMRE N-256 terminal loop and MCU cc2-cc3 (Fig. 4A), however, we also found several phospholipids filling the 257 gaps between helices from MCU and EMRE. These phospholipids could stabilize the gaps and provide 258 elasticity to this region, enabling the gate to be opened by EMRE (Fig. 4C). The MCU-R297D 259 mutation might dissociate the bound CDL and disrupt the attachment of EMRE on cc2. This would 260 leave cc2 free to roll aside and possibly push the negatively charged Glu 288 and Glu 293 residues of 261 MCU inward, thus making the channel non-conducting. We observed multiple hydrophobic 262 interactions between cc3 and cc1, which might help to achieve a correct position of the gate-forming 263 cc2 ( Fig. 4A and fig. S6A). The amino acid residues participating in these hydrophobic interactions 264 are highly conserved and were shown to be indispensable for MCU activity (48). Besides, the 265 11 / 26 negatively charged polar head of the horizontal PLX is also very likely involved in forming the gate, 266 because their conformation is quite stable and they protrude deeply into the channel (Fig. 4C). 267

268
The MCU pore and its regulation by MICU1/2 269 We detected three Ca 2+ ions in each pore of the MES structure (using buffers containing Ca 2+ ), and 270 only a single Ca 2+ ion in each pore in the MEMMS structure (using buffers deprived of Ca 2+  was below 100 nM. This is expected, as the sample was washed multiple times with buffers containing 283 0.1 mM EGTA (see Methods). The estimated Kd for Ca 2+ binding of MICU1/2 dimer is ~600 nM (51), 284 therefore it is also expected that the EF hands of MICU subunits in our MEMMS structure are Ca 2+ -285

free. 286
In MEMMS, MICU1/2 drag the two MCU tetramers closer to each other as compared to MES (fig. 287 S4E). However, the pores of the MES and MEMMS structures are similar at both the selectivity filter 288 and the putative matrix gate and show no possible obstructions for Ca 2+ permeation (Fig. 5,A to C). 289 This indicates that the binding of MICU1/2, does not occlude or obstruct the MCU pore. This 290 conclusion is in a striking contrast to the currently accepted model of MICU1/MICU2 dimer (22, 23, 291 39, 40 In conclusion, here we report the first structure of intact MCU supercomplex as a 20-subunit O-shaped 305 dimer of hetero-decamers, with auxiliary MICU1 and MICU2 subunits attached. We discovered that a 306 pair of MICU1-MICU2 hetero-dimers link the two MCU channels, which is obviously different from 307 previous models that assume MICU1/2 oligomers to ride across the MCU pore and occlude it in low 308  (Fig. 5E). 324 We compared our MES structure with previous fungal MCU structures to find that human MES has a 325 swollen CCD enlarged by EMRE ( fig. S10). This curvature is very likely facilitated by Pro 216 of cc1, 326 which is conserved in mammals but absent in fungi ( fig. S10E). The cc2s in reported fungal MCU 327 structures are not well resolved, indicating that their position is flexible possibly due to lack of EMRE. 328 The curvature of cc1 and the tight cc1-cc3 interaction could probably elevate the position of cc2 and      on the left and bar graph in the right (mean ± SEM, n ≥ 3 independent measurements). Western blot 520 of cell lysates from the different groups were performed to make sure that EMRE expression was 521 similar by using antibody to strep. β-actin was used as the loading control. 522 supercomplex. Two sets of imagined levers are shown. EMRE is the first lever, with its pivot on TM1, 570 its C-terminal loop attached to MICU1, and its N-terminal loop attached to CCD. cc2 is the second 571 lever, with its pivot on the loop linking cc2 and cc3, its N-terminal attached to TM2, and its Arg 297 572 attached to EMRE. Arg 297 functions as the point of contact between the first and the second levers. 573 Pivot and movement of the first lever is indicated by black triangle and arrows, respectively. Pivot and 574 movement of the second lever is indicated by gray triangle and arrows, respectively. The movement 575 of TM2 is marked by a red arrow. MICU1/2 conformational change is represented by a shape change.
After extensively biochemical studies, we obtained the high quality and quantity protein complexes with the combination of the EMRE C-terminus tagged and the other four subunits without tags viruses.

Purification of the human mitochondrial Ca 2+ uniporter supercomplex
All procedures were carried out at 4°C . To purify the MES complex, four liters of transfected cells were harvested, washed with 1×PBS and resuspended in 10 mM Tris pH 7.4, 225 mM sorbitol, 60 mM sucrose, 2 mM CaCl2 and 0.1% BSA, 1 mM PMSF. The suspension was homogenized by a soft blender for 150 s and the homogenate was centrifuged at 3,000 g for 10 min. Supernatant was further centrifuged at 20,000 g for 45 min to obtain the crude mitochondria. The pellet was suspended and extracted in 25 mM Tris pH 7.8, 150 mM NaCl, 2 mM CaCl2 with 1% digitonin. After incubation for an hour, the extraction was centrifuged at 20,000 g for 20 min at 4°C and the supernatant was applied to Strep-Tactin Sepharose by gravity at 4°C . The resin was washed three times with W buffer, which contained 25 mM Tris pH 7.8, 150 mM NaCl, 2 mM CaCl2 with 0.1% digitonin. The target proteins were eluted with W buffer plus 5 mM desthiobiotin, concentrated to 100 μL by 100 kDa cut-off centrifugal filter (Millipore) and further purified by Superdex200 increase 5/150 GL also in W buffer.
To obtain the MEMMS complex, the same protocol was used with some modification: 2 mM CaCl2 in all the buffer was replaced by 0.1 mM EGTA. The peak fractions were collected for EM sample 2 / 21 preparation, the presence of the complex was verified by Blue Native-PAGE and confirmed by mass spectrometry.
Grids were blotted for 5 s and flash-frozen in liquid ethane using an FEI Mark IV Vitrobot operated at 8°C and 100% humidity. The grids were transferred to a Titan Krios (FEI) electron microscope equipped with a Cs corrector, operating at a voltage of 300 kV. Images were recorded by a K2 Summit direct electron detector (Gatan, Inc.) equipped with a GIF Quantum energy filter (slit width 20 eV) in the super-resolution counting mode. Data acquisition was performed using AutoEMation II with a nominal magnification of 105,000 times, which yields a super-resolution pixel size of 0.5455 Å on image plane, and with defocus ranging from -1.5 μm to -2.0 μm. The dose rate on the detector was ~8.0 counts per pixel per second with a frame exposure time of 0.175 second and a total exposure time of 5.6 seconds. Each micrograph stack contains 32 frames. The total dose rate was approximately 50 e -/Å 2 for each micrograph.

Image processing
A simplified flowchart of the procedure for image processing of MES is presented in fig. S2A. A total of 4,997 cryo-EM movie stacks were automated collected. The motion correction was performed using MotionCor2 (52) with 2×2 binning, resulting in a pixel size of 1.091 Å, and meanwhile, dose weighting was performed, yielding motion-corrected integrated images for further processing. After whole image CTF estimation using CTFFIND3, 4,706 good micrographs were manually selected from the dataset (53). A total of 471,401 particles were auto-picked using RELION-3.0 (30). After several rounds of two-dimensional (2D) classification, 327,716 particles were selected for further three-dimensional (3D) analysis. A total of 19,739 particles from the first 500 micrographs were used to generate initial models for the first round of 3D classification using RELION-3.0. Multi-reference 3D classification was performed for the 327,716 particles and 89,734 particles were selected then subjected to 3D refinement.
Each particle was recentered using the in-plane translations measured in 3D refinement and reextracted from the motion-corrected integrated micrographs. Gctf (54) was used to refine the local 3 / 21 defocus parameters. The well centered particles with more accurate defocus parameters were subjected to 3D refinement without symmetry, which resulted in an electron density map at 3.67 Å resolution.
The 3.67 Å map was fitted into a copy of itself rotated by 180° using Chimera (55), confirming a C2 symmetry in this map. Another round of 3D refinement with C2 symmetry was performed and yield a map at 3.51 Å resolution.
The 89,734 particles were further classified in to three classes using masked local angular search 3D classification with a step size of 0.9° and a local search range of 5°. Particles from two classes, which were in relatively good quality but slightly varied in the separation angle of the two MCU-EMRE hetero-octamer, were further refined and yield two maps at 4.18 Å and 3.41 Å resolution, respectively.
To improve the density of N-terminal domains (NTDs), 3D refinement with a focused mask was performed for the 46,879 particles in the class of higher resolution, resulting in a focused map with a resolution of 3.27 Å after post processing. Since the density of the hetero-octamer transmembrane domain (TMD) and coiled-coil domain (CCD) of all the three classes could perfectly fit in each other, the 89,734 particles before the 3D classification were expanded according to C2 symmetry using relion_particle_symmetry_expand, then subtracted by the density of one NTD and the other whole hetero-octamer using RELION-3.0. The 179,468 subtracted particles were subjected to 3D refinement with a soft mask without symmetry, resulting in a focused map with a resolution of 3.27 Å after post processing. The focused map of NTDs and two copies of the focused map of TMD+CCD were fit into the 3.41 Å map then combined using PHENIX (32) Combine Focused Maps, resulting in the final map.
The reported resolutions are based on the gold-standard Fourier shell correlation 0.143 criterion (56).
All density maps were sharpened by applying a negative B-factor that was estimated using automated procedures (57). Local resolution variations were estimated using Resmap (58).
A simplified flowchart of the procedure for image processing of MEMMS is presented in fig. S2B. A total of 9,899 cryo-EM movie stacks were motion corrected, 2×2 binned and dose weighted using MotionCor2. After whole image CTF estimation using CTFFIND3, 9,113 good micrographs were manually selected from the dataset. A total of 986,805 particles were autopicked using RELION-3.0.
After several rounds of 2D classification, 699,244 particles were selected and subjected to 3D classification, using the MES 3.41 Å map low-pass filtered to 50 Å as the initial model. After 3D 4 / 21 classification, 250,977 particles of the class with an additional "cap" comparing to MES were selected then subjected to 3D refinement. Each particle was recentered and re-extracted from the motioncorrected integrated micrographs. Gctf was used to refine the local defocus parameters. The reextracted particles were subjected to 3D refinement without symmetry, which resulted in a map at 4.09 Å resolution.
The 250,977 particles were further classified into three classes using masked skip-alignment 3D classification. A total of 45,864 particles of the class with clear "cap" were subjected to 3D refinement with a soft mask then yield a map at 3.64 Å resolution. Since the density of the MICUs (i.e. the "cap") is still fragmentary, these particles were subtracted by the density of MES in the 3.64 Å map of MEMMS using RELION-3.0 and subjected to 3D refinement. Then 33,930 particles were selected after a final round of 3D classification and subjected to 3D refinement with a soft mask and C2 symmetry, leading to a reconstruction of MICUs at 3.71 Å resolution with much better density. To improve the density of NTDs, the 45,864 particles were subtracted by the density of CCDs, TMDs and MICUs, then subjected to 3D refinement with a soft mask and C2 symmetry, resulting in a density map of NTDs at 3.39 Å resolution. In order to improve the density of TMD+CCD, the 45,864 particles were expanded according to C2 symmetry using relion_particle_symmetry_expand, subtracted by the density of the rest part beside TMD+CCD, then subjected to 3D refinement with a soft mask.
Eventually, the resolution of TMD+CCD was improved to 3.30 Å. The focused map of NTDs, MICUs and two copies of the focused map of TMD+CCD were fit into the 3.64 Å map then combined using PHENIX Combine Focused Maps, resulting in the final map.
The reported resolutions are based on the gold-standard Fourier shell correlation 0.143 criterion. All density maps were sharpened by applying a negative B-factor that was estimated using automated procedures. Local resolution variations were estimated using Resmap.

Model building and refinement and validation
The atomic model of MES was manually built and adjusted in COOT (31). And then, the model with the ligands were subjected to global refinement and minimization in real space refinement using PHENIX with secondary structure and NCS restraints. The crystal structure of MICU1 (PDB 4NSC) and MICU2 (PDB 6AGH) were used as the initial models for MICUs in MEMMS. The model was 5 / 21 refined in real space using PHENIX with secondary structure and NCS restraints. The final atomic models were evaluated using MolProbity (59). Pore radii were calculated using the HOLE program (60). All the figures were prepared in PyMol (61).

Gene knockout by CRISPR/Cas9
Gene knockout by CRISPR/Cas9 was performed using a previously described protocol (62). Two sets of guide RNA sequences were designed. Guide sequences used for gene knockout were as follows:

Co-immunoprecipitation and western blot
All co-immunoprecipitation experiments were performed at 4°C. In brief, related HEK 293T knockout cells at 80%-90% confluence were transfected with 15 μg corresponding plasmids using lipofectamine 2000 (Thermo Fisher Scientific) and grown in a 37℃ CO2 incubator for 24 hours. Transfected cells were lysed in 1 mL lysis buffer (25 mM Tris pH 7.8, 150 mM NaCl, 1 mM EGTA, cOmplete protease inhibitors) with 1% digitonin. The cell lysate was incubated for 30 min on ice and centrifuged for 10 min at 4°C at 20000 g. A small portion of the sample was used for whole cell lysate analysis and the rest was collected and incubated with anti-FLAG magnetic agarose (Thermo Fisher Scientific) for two hours in 4℃. The beads were collected on a magnet, washed three times with 1 mL lysis buffer which contained 0.1% digitonin, and eluted with 150 μL SDS-gel loading buffer for western blot.  Western blot of cell lysates from the different groups were performed to make sure the MCU expression was similar, using antibody to MCU. β-actin was used as the loading control.