Cryo-EM structure of a mitochondrial calcium uniporter

One Sentence Summary:

The structure of MCU reveals a tetrameric architecture and provides the molecular framework for understanding the mechanism of calcium selectivity.

Abstract

Calcium transport plays an important role in regulating mitochondrial physiology and pathophysiology. The mitochondrial calcium uniporter (MCU) is a calcium-selective ion channel that is the primary mediator for calcium uptake into the mitochondrial matrix. Here we present the cryo-electron microscopy structure of the full-length MCU from Neurospora crassa to an overall resolution of ~3.7 Å. Our structure reveals a tetrameric architecture, with the soluble and transmembrane domains adopting different symmetric arrangements within the channel. The conserved W-D-Φ-Φ-E-P-V-T-Y sequence motif of MCU pore forms a selectivity filter comprising two acidic rings separated by one helical turn along the central axis of the channel pore. The structure combined with mutagenesis gives insight into the basis of calcium recognition.

Mitochondrial Ca²⁺ transport is critical for shaping the dynamics of intracellular calcium signaling, regulating energy metabolism, generating reactive oxygen species, and modulating cell death (1, 2). Ca²⁺ uptake across the mitochondrial inner membrane was shown to occur via a “uniporter” (3) and electrophysiological studies of the mitoplast inwardly-rectifying Ca²⁺ current (I_MICA) showed that this “uniporter” is an ion channel that exhibits a remarkable selectivity for Ca²⁺ (4). Recent genomics studies have identified the key components of the uniporter holocomplex (uniplex) (5–9). In vertebrates, it comprises the mitochondrial calcium uniporter (MCU), an auxiliary transmembrane protein essential MCU regulator (EMRE) (8), and the auxiliary EF hand-containing proteins mitochondrial calcium uptake 1 (MICU1) and MICU2 (6, 10). EMRE plays a dual role in maintaining MCU in an open state and recruiting MICU1/2 (11) that regulate the activity of MCU in a Ca²⁺-dependent manner (6, 12–15). While MCU is widely distributed across all major branches of eukaryotes (16), EMRE is metazoan-specific (8). The MCU orthologue from Dictyostelium discoideum, an organism lacking EMRE, alone, is capable of reconstituting channel activity in yeast (17). In fungi, MCU is typically present without the MICU1/2 homologs (16). Recent studies have further shown that fungal MCU homologs are also able to reconstitute channel activity in vitro and in vivo on their own (18, 19). Taken together, these data establish that MCU is the Ca²⁺-conducting pore-forming unit of the uniplex.

A single protomer of MCU is predicted to possess two transmembrane helices (TM1 and TM2), two coiled-coils (CC1 and CC2), and a N-terminal domain (NTD) located on the matrix side. All MCU homologs contain a highly-conserved sequence motif W-D-Φ-Φ-E-P-V-T-Y (Φ denotes hydrophobic amino acids) located between TM1 and TM2. This motif is oriented facing the intermembrane space and is thought to form the selectivity filter in the oligomeric channel (5, 7, 9). The NTD is composed of ~100 residues and extends into the mitochondrial matrix. Crystallographic studies of the human MCU NTD revealed a distinct structural fold similar to a beta-grasp, and subsequent functional studies revealed its modulatory role in MCU function (20, 21). A structure of the NTD-deleted MCU homolog from C. elegans (cMCU-ΔNTD) was recently determined using NMR and negative-stain electron microscopy, revealing a pentameric architecture (22), although the absence of the NTD limited further structural insights into the full-length channel assembly. We therefore conducted structural studies of a full-length MCU homolog using cryo-electron microscopy (cryo-EM). After screening a number of MCU homologues based on phylogenetic analyses (16), we found that a homolog from Neurospora crassa (MCU_NC) was suitable for structural studies. To prevent proteolysis, we introduced a Tyr232Ala mutation into a flexible loop within the NTD (fig. S1). This Tyr232Ala mutant eluted at a similar volume as wild-type MCU_NC during size-exclusion chromatography, and exhibited a similar overall architecture, as determined by negative-stain EM (fig. S2). MCU_NC was prepared under two different Ca²⁺ conditions for EM studies, referred to as low and high Ca²⁺ (see Methods). In both conditions, MCU_NC was reconstituted in amphipol and subjected to single-particle 3D cryo-EM analysis. Interestingly, MCU_NC unambiguously showed a tetrameric arrangement by EM under both Ca²⁺ conditions (fig. S3). The attainable resolution of these reconstructions (4.7–7 Å, fig. S3, see Methods) appeared to be limited by flexibility of the NTD. Therefore, we attempted to cross-link the NTD of MCU_NC using the water-soluble cross-linker bis-sulfosuccinimidyl suberate (BS3), in the presence of high Ca²⁺, and reconstituted the BS3-crosslinked MCU_NC into nanodiscs for EM analyses. The 3D reconstruction of cross-linked MCU_NC in nanodiscs was determined to an overall resolution of ~3.7 Å (fig. S4). Comparison of the 3D reconstructions of cross-linked MCU_NC and native MCU_NC showed that cross-linking did not appreciably affect the structure of MCU_NC significantly (fig. S4). Thus, data of cross-linked and native MCU_NC were combined to yield an improved reconstruction at an overall resolution of ~3.7 Å (see Methods, fig. S4) that allowed for de novo model building (Methods and Table S1).

SDS-PAGE analysis of BS3-crosslinked MCU_NC indicated a complete conversion of the monomeric MCU band to a single band with an apparent molecular weight most compatible with tetrameric MCU_NC (fig. S2). Both cross-linked and native MCU_NC in detergent micelles elute at similar volumes during size-exclusion chromatography suggesting that cross-linking did not alter the oligomeric state of MCU_NC. Finally, 2D classification and asymmetric 3D classification of both the native and cross-linked MCU_NC samples unambiguously showed a tetrameric organization of MCU_NC, with no evidence for a pentamer, strongly supporting the tetrameric architecture of MCU_NC in our final structures (figs. S3 and S4).

The overall shape of the MCU_NC homotetramer is a prolate spheroid with dimensions of approximately 40 Å x 48 Å x 130 Å (Fig. 1). The transmembrane domain (TMD) is formed by TM1 and TM2 helices and the matrix region comprises the coiled-coil domain (CCD) and the NTD. TM1 and CC1 form a long and continuous helix at the periphery of the channel, while the TM2 helices line the central symmetry axis (Fig. 1B). The TM helices are arranged such that TM1 from one protomer primarily interacts with TM2 from the adjacent protomer (Fig. 1E). TM2 is followed by ~15 amino acids near the matrix-membrane interface but could not be accurately modeled (fig. S6). Immediately C-terminal to this region, a short helix, which we termed the junctional helix (JH), is positioned nearly perpendicular to TM1 and TM2 and forms a junction between TM2 and CC2. In the CCD, CC1 and CC2 form a dimeric coiled-coil, resulting in four dimeric coiled-coils within the tetramer. Due to the uncertainty of the connection between TM2 and JH, we cannot rule out the possibility that CC2 interacts with CC1 from the neighboring protomer (fig. S6). Unlike human MCU (hMCU), CC2 of MCU_NC is followed by three putative helical regions at its C-terminus, which is not resolved in the 3D reconstruction (fig. S1). The NTD comprises six β-strands (β1−β6) and two α-helices (α1-α2) and is connected to CC1 of the CCD. α2, located between the CCD and the rest of the NTD, positions the NTD relative to the rest of the channel (Fig. 1, C to E). Despite the low sequence identity (~17% for the structured regions), the NTD is structurally very similar to the hMCU NTD fragment (PDB ID: 5KUJ, Cα RMSD. 1.8 Å, figs. S1 and S7) (20, 21).

Figure 1. Overall architecture of MCU_NC.

(A) Single-particle cryo-EM reconstruction and (B) model of MCU_NC viewed from the intermembrane space (left), within the inner mitochondrial membrane (middle), and from the mitochondrial matrix (right). (C) Cartoon diagram outlining the protein domains and secondary structures. (D) Detailed view of an MCU protomer as colored in (C). Dashed lines indicate parts that were not modeled in the structure. (E) Close-up views of interactions between TM2 and TM1 from neighboring protomers (upper), between CCD and α2 from NTD (lower).

All MCU homologs contain the highly-conserved sequence motif W-D-Φ-Φ-E-P-V-T-Y, which has been proposed to form the selectivity filter (5, 7, 22). In our MCU_NC structure, this sequence motif is located in the N-terminal region of TM2 (Fig. 2) with the carboxylate side chains of conserved acidic residues Asp³⁵⁵ and Glu³⁵⁸ from each protomer directed towards the central symmetry axis, forming two acidic rings along the channel pore. The first acidic ring formed by Asp³⁵⁵, which we term Site 1 (S1), is located at the mouth of the pore and exposed to the intermembrane space. The distances between diagonally positioned Asp³⁵⁵ are approximately 8.8 Å, indicating that a hydrated Ca²⁺ is likely to bind to this ring (Fig. 2C, S1). The second acidic ring formed by Glu³⁵⁸ is termed Site 2 (S2). Notably, there is an extensive network of interactions surrounding Glu³⁵⁸ involving residues in the W-D-Φ-Φ-E-P-V-T-Y motif that position the carboxylate group of Glu³⁵⁸ toward the central symmetry axis. Specifically, Pro³⁵⁹ appears to make inter-protomer CH-π interactions with Trp³⁵⁴ from the adjacent subunit, the amide nitrogen in the indole ring of Trp³⁵⁴ hydrogen bonds to the carboxylate group of Glu³⁵⁸ from an adjacent protomer, and Tyr³⁶² is involved in both π-π interactions with Trp³⁵⁴ as well as hydrogen bonding interactions with Thr³⁶¹ from the adjacent protomer (Fig. 2, C and D, and fig. S4). Although the EM density for the MCU motif is sufficiently high to place side-chain atoms, the exact chemical nature of these interactions should be interpreted with caution. The distance between the carboxylate groups of Glu³⁵⁸ from diagonally opposing protomers is approximately 4.8 Å, suggesting that only dehydrated Ca²⁺ can be coordinated at S2. There is a strong EM density at the center of S2 (>17 σ, fig. S8A) that we tentatively assign to Ca²⁺ because: 1) the density peak is present in both an asymmetric reconstruction of MCU_NC as well as the corresponding half maps (fig. S8, B and C), and 2) calcium was present in high concentration during both protein purification and EM grid preparation (see Methods). The coordination of dehydrated Ca²⁺ by acidic residues has been observed in both TRPV6 and Orai1, and was proposed to be key to the Ca²⁺ selectivity of these channels (23, 24). We therefore suggest that, together, this S2 acidic ring plays an important role in the selective Ca²⁺ transport by MCU. Consistent with our structural observations, mutation of Glu³⁵⁸ has been shown to abolish hMCU activity (22). We also observed two polar amino acids, Thr³⁶¹ and Tyr³⁶², located one helical turn below S2 that line the central axis of the pore, which we tentatively term “Site 3” (S3). The diagonal distances between the side chains are large (13–14 Å), so S3 may play a role in hydrating Ca²⁺ ions exiting S2. We observed an EM density at the center of S3, the identity of which is unclear (figs. S8, A to C). The selectivity filter organization of MCU_NC is in contrast with other classical tetrameric cation channels, where the selectivity filter is formed by loops connecting a TM helix and a pore helix (25, 26).

Figure 2. Ion conduction pore and selectivity filter.

(A) The ion conduction pathway of MCU_NC and the pore radius along the central axis (generated with HOLE software (30)) indicate three constrictions at Asp³⁵⁵, Glu³⁵⁸, and Tyr³⁶². Front and rear protomers were removed for clarity. (B) Putative Ca²⁺ coordination sites, among which a strong cryoEM density peak was observed and tentatively modeled as Ca²⁺ at site 2. (C) Top views of the putative Ca²⁺ coordination sites constituted by Asp³⁵⁵ at S1 (left), by Trp³⁵⁴, Glu³⁵⁸, and Pro³⁵⁹ at S2 (middle), and by Thr³⁶¹ and Tyr³⁶² at S3 (right). (D) Detailed view from the membrane plane showing the extensive network of interactions engaged by residues in the “WDΦΦEPVTY” motif. The CH-π (Pro³⁵⁹ and Trp³⁵⁴) and π-π (Trp³⁵⁴ and Tyr³⁶²) interactions are highlighted by blue dashed lines. (E) Mitochondrial calcium uptake of human MCU (hMCU) mutants from MCN_NC structure-based mutagenesis at the ion conduction pore. Representative traces of calcium uptake in digitonin-permeabilized cells after 10 μM CaCl₂ was added. Mutation of Trp²⁶⁰ to Ala or Phe suppresses hMCU channel function, whereas mutation of Thr²⁶⁷ or Tyr²⁶⁸ to Ala solely reduces the activity. Mutation of Tyr²⁶⁸ to Phe shows calcium uptake to the comparable extent as wildtype hMCU (hMCU) expressed in MCU knockout (KO) cells. (F) Bar graph showing the calcium uptake of hMCU mutants relative to the wildtype hMCU between 0.5 and 3-min time points (mean ± SEM, n ≥ 4 independent measurements). To detect the expression of hMCU mutants, cell lysates were analyzed by immunoblotting with anti-FLAG antibody. Wild-type HEK 293T cells and MCU knockout cells were confirmed by anti-hMCU antibody. β-actin was used as the loading control.

In order to investigate the role of the above-described interactions of the W-D-Φ-Φ-E-P-V-T-Y motif in Ca²⁺-permeation by MCU, we employed site-directed mutagenesis and Ca²⁺-uptake assays. Although functional assays are not currently available for MCU_NC, we exploited the high conservation of the selectivity filter within the MCU family and used an established mitochondrial Ca²⁺-uptake assay in HEK-293 cells lacking hMCU to test the activity of transfected hMCU mutants designed based on our MCU_NC structure (11). Specifically, we tested the importance of Trp²⁶⁰, Thr²⁶⁷, and Tyr²⁶⁸ (Trp³⁵⁴, Thr³⁶¹, and Tyr³⁶² in MCU_NC) for Ca²⁺ uptake by hMCU (Fig. 2, E and F). Importantly, mutation of Trp²⁶⁰ (Trp³⁵⁴ in MCU_NC) to either phenylalanine or alanine abolished hMCU-mediated Ca²⁺ uptake and mutation of Thr²⁶⁷ (Thr³⁶¹ in MCU_NC) or Tyr²⁶⁸ (Tyr³⁶² in MCU_NC) to alanine reduced the hMCU-mediated Ca²⁺ uptake substantially, indicating their importance for MCU function. Notably, Ca²⁺ uptake was restored in the Tyr²⁶⁸ (Tyr³⁶² in MCU_NC) to phenylalanine mutant, indicating that aromaticity is important for this position, in agreement with the above-described interactions. None of the tested mutations appreciably affected the expression level of hMCU compared to WT (Fig. 2F).

Viewing along the symmetry axis of the channel a symmetry mismatch occurs wherein the TMD adopts 4-fold symmetry while the CCD and the NTD show a 2-fold symmetric organization (Fig. 3). Importantly, this observed symmetry mismatch is apparent in the 3D reconstructions of both native MCU_NC and BS3-crosslinked MCU_NC (fig. S3). Local and global alignments of protomers A and B in the MCU_NC structure indicate the domain rearrangements between the TMD and the CCD-NTD account for the symmetry break (Fig. 3 and fig. S9) with the departure from 4-fold symmetry originating from two distinct interaction networks engaged by neighboring NTDs (Fig. 3). With respect to the NTDs and CCDs, the channel adopts a dimer-of-dimers assembly; within the NTD dimer comprising protomers A and B (the A/B dimer), there is a large dimer interface (~620 Ų) defined by both NTDs and CCDs, while there is a comparatively minimal dimer interface (~380 Ų) mediated by the NTD dimer comprising protomers B and C (the B/C dimer).

Figure 3. Hinge rotation enables the mismatch in channel symmetry.

(A) Top view comparison of channel symmetry sliced through the three layers depicted in the tetrameric MCU_NC channel. The TMD exhibits 4-fold symmetry, while both the CCD and NTD display 2-fold symmetry. The lines are drawn between Cα atoms of Trp³⁵⁴, Asp⁴⁰⁷, and Ser²⁵⁵. (B and C) Alignment of protomer A (red) and protomer B (blue) at NTD (B) and at TMD (C), showing the rigid body rotations of TMD and NTD around the hinge point JH. (D) Distinct interfacial networks between each NTD and its neighboring partner.

The dimer-of-dimers assembly of soluble domains and the tetrameric assembly of the TMD in MCU_NC is analogous to that of ionotropic glutamate receptors (iGluRs) (27). In iGluRs, the ligand binding domain (LBD) and amino terminal domain (ATD) assume 2-fold symmetric arrangements, while the TMD adopts a 4-fold symmetric arrangement. The LBD transitions into various conformations, including a pseudo-four-fold arrangement during the gating cycle of iGluR (28, 29), indicating the NTD dimer-of-dimers assembly might play a comparable role in MCU gating. This notion is further supported by recent studies suggesting that phosphorylation of the NTD or divalent cation binding to the NTD modulate MCU function (fig. S7) (20, 21).

Many structural and architectural features observed in our MCU_NC structure contrast with those of cMCU-ΔNTD (Fig. 4). First, cMCU-ΔNTD and MCU_NC adopt distinct pentameric and tetrameric stoichiometries, respectively, which could result from construct design (truncation of the NTD versus full-length protein), choice of detergents (zwitterionic Fos-choline-14 versus dodecylmaltoside), and/or protein preparation (extraction from inclusion body versus the membrane). The Ca²⁺ channel function has not been shown in either cMCU (22) nor MCU_NC. Therefore, the in vivo oligomeric status of MCU has not been established and we cannot exclude the possibility that fungal and cMCUs may adopt different oligomeric arrangements. These discrepancies await further validation. Second, the arrangement of TM1 and TM2 in MCU_NC establishes an extensive inter-protomer interface, whereas the TM helices in cMCU-ΔNTD only form intra-protomer interactions. Third, while the selectivity filter sequence is located on TM2 in MCU_NC, the corresponding residues in cMCU-ΔNTD are positioned within a loop, leading to different selectivity filter structures (fig. S10). Finally, the CCD in MCU_NC consists of four dimeric coiled coils, formed by CC1 and CC2, whereas the CC2 from each protomer in cMCU-ΔNTD forms a pentameric coiled coil in the CCD (Fig. 4).

Figure 4. Comparison of cryo-EM structure MCU_NC and NMR structure cMCU-ΔNTD.

(A and B) Side and top views showing the tetrameric configuration of MCU_NC. (C) The selectivity filter sequence (yellow) of MCU_NC is located at the beginning of TM2. (D) Close-up view of the hydrophobic interactions (silver spheres) between CC1 and CC2. Hydrophilic residues are colored in teal. (E) Viewed from the intermembrane space, CC1 (orange) and CC2 (red) of MCU_NC form dimeric coiled coils within each protomer via extensive hydrophobic interactions (highlighted by silver spheres). (F and G) Side and top views showing cMCU-ΔNTD (PDB ID: 5ID3) forms a pentamer. (H) The “DΦΦE” motif (yellow) in cMCU is located at the loop connecting TM1 and TM2. (I) Close-up view of the hydrophobic residues (silver spheres) located on CC1 and CC2. Hydrophilic residues are colored in teal. (J) Viewed from the intermembrane space, CC2 (red) forms a pentameric helical bundle via hydrophobic interactions (silver spheres) pointing towards the central axis. Hydrophobic residues (silver spheres) on CC1 (orange) are exposed to the mitochondrial matrix.

Our studies provide structural insights into the design principle of the MCU_NC selectivity filter, which will serve as a platform to understand the mechanism of selective calcium permeation by this channel family.