Structures reveal gatekeeping of the mitochondrial Ca2+ uniporter by MICU1-MICU2

The mitochondrial calcium uniporter is a Ca2+-gated ion channel complex that controls mitochondrial Ca2+ entry and regulates cell metabolism. MCU and EMRE form the channel while Ca2+-dependent regulation is conferred by MICU1 and MICU2 through an enigmatic process. We present a cryo-EM structure of an MCU-EMRE-MICU1-MICU2 holocomplex comprising MCU and EMRE subunits from the beetle Tribolium castaneum in complex with a human MICU1-MICU2 heterodimer at 3.3 Å resolution. With analogy to how neuronal channels are blocked by protein toxins, a uniporter interaction domain on MICU1 binds to a channel receptor site comprising MCU and EMRE subunits to inhibit ion flow under resting Ca2+ conditions. A Ca2+-bound structure of MICU1-MICU2 at 3.1 Å resolution indicates how Ca2+-dependent changes enable dynamic response to cytosolic Ca2+ signals.


Introduction
The mitochondrial calcium uniporter is a highly selective multi-subunit Ca 2+ channel of the inner mitochondrial membrane that serves as the major conduit for uptake of Ca 2+ by mitochondria and controls ATP production in response to cytosolic Ca 2+ signals (Gunter and Pfeiffer, 1990;Kamer and Mootha, 2015;Pan et al., 2013;Pathak and Trebak, 2018). The uniporter is one of the most selective Ca 2+ channels known and it is only activated once cytosolic Ca 2+ levels reach a threshold (of approximately 0.4-3 mM) (Csordás et al., 2013;Kirichok et al., 2004;Liu et al., 2016;Mallilankaraman et al., 2012b;Payne et al., 2017). These properties, Ca 2+ selectivity and block of ion permeation under resting conditions, are fundamentally important for preventing mitochondrial Ca 2+ overload and for maintaining the electromotive force used for ATP production because potassium and other cations outnumber Ca 2+ by more than one million to one in the cytosol. The protein MCU constitutes the pore of the uniporter through which Ca 2+ ions flow and EMRE, a small transmembrane protein that associates with MCU, is necessary for Ca 2+ uptake in metazoan organisms (Baughman et al., 2011;Chaudhuri et al., 2013;De Stefani et al., 2011;Kovács-Bogdán et al., 2014;Sancak et al., 2013). Ca 2+ -dependent control is conferred by proteins that are unique to the uniporter À heterodimers of MICU1-MICU2 in most cells and of MICU1-MICU3 in neurons (Ashrafi et al., 2019;Kamer et al., 2017;Kamer and Mootha, 2014;Patron et al., 2014;Patron et al., 2019;Payne et al., 2017;Perocchi et al., 2010;Plovanich et al., 2013). The activity of the uniporter is further tuned by interactions with MCUR1 (Chaudhuri et al., 2016;Mallilankaraman et al., 2012a;Vais et al., 2015) and the MCU-like protein MCUb (Raffaello et al., 2013). The molecular mechanism by which the MICU1-MICU2 and MICU1-MICU3 heterodimers regulate the channel is a matter of intense study (Csordás et al., 2013;Hoffman et al., 2013;Kamer et al., 2017;Kamer and Mootha, 2014;Kamer et al., 2018;Paillard et al., 2018;Patron et al., 2014;Patron et al., 2019;Payne et al., 2017;Perocchi et al., 2010;Phillips et al., 2019;Plovanich et al., 2013). Multiple models for this regulation have been proposed but issues as fundamental as the stoichiometry between the heterodimers and the channel remain unclear.
Studies have resolved that MICU1-MICU2 heterodimers reside in the intermembrane space (IMS) where they regulate the uniporter by responding to [Ca 2+ ] IMS Patron et al., 2014;Plovanich et al., 2013). [Ca 2+ ] IMS , which closely follows the cytosolic Ca 2+ concentration due to the permeability of the outer mitochondrial membrane, is sensed by EF-hand motifs in each protein. At the~100 nM resting level of [Ca 2+ ] cytosol , MICU1-MICU2 and MICU1-MICU3 heterodimers prevent ion conduction through the channel, and they permit it when Ca 2+ levels rise (Csordás et al., 2013;Kamer et al., 2017;Kamer and Mootha, 2014;Liu et al., 2016;Mallilankaraman et al., 2012b;Paillard et al., 2017;Patron et al., 2014;Patron et al., 2019;Payne et al., 2017;Phillips et al., 2019). While structures of MICU1-3 and MCU-EMRE components have been determined (Kamer et al., 2019;Park et al., 2020;Wang et al., 2020;Wang et al., 2014;Wang et al., 2019;Wu et al., 2019;Xing et al., 2019), it is unclear how MICU proteins interact with the channel and the mechanism by which they confer Ca 2+ -dependent control to the channel is enigmatic. Further it is not known why MICU1, in particular, is necessary among MICU1-3 for regulating the channel Patron et al., 2014;Patron et al., 2019;Payne et al., 2017;Xing et al., 2019). Here, we present a 3.3 Å resolution cryo-EM structure of the MCU-EMRE-MICU1-MICU2 assembly, hereafter referred to as the holocomplex, under resting [Ca 2+ ] conditions. Together with functional data and a structure of a Ca 2+ -bound MICU1-MICU2 complex, the work reveals a molecular basis for Ca 2+ -dependent control of the uniporter and the unique role of MICU1 in this process.

Structure determination of the MCU-EMRE-MICU1-MICU2 holocomplex
The ability of human MICU1 to bind MCU channels from lower-eukaryote organisms (Phillips et al., 2019) and our ability to obtain high-resolution structures of the metazoan MCU-EMRE channel complex from the red flour beetle, Tribolium castaneum, (TcMCU-EMRE) (Wang et al., 2020)  A holocomplex sample that was suitable for structure determination required meticulous formulation. An MCU-EMRE subcomplex was prepared by coexpressing TcMCU and TcEMRE in mammalian cells, purifying the TcMCU-EMRE assembly, and reconstituting it into lipid nanodiscs that include the mitochondrial lipid cardiolipin. Cardiolipin was included because we recently identified that this lipid is important for the function of the uniporter (Wang et al., 2020). We previously observed that a symmetry mismatch between the fourfold symmetric IMS surface of the channel and the asymmetric N-terminal domain (NTDs) in metazoan MCU channels can be a hindrance for high-resolution cryo-EM studies (Baradaran et al., 2018;Wang et al., 2020). Because the NTD is dispensable for mitochondrial Ca 2+ uptake (Baradaran et al., 2018;Oxenoid et al., 2016;Wang et al., 2020;Wang et al., 2019), we removed this domain for high-resolution structural studies. We have recently shown that the TcMCU and TcEMRE constructs used for structural analysis (TcMCU DNTD and TcEMRE coexpressed) catalytze Ca 2+ uptake into human mitochondria (Wang et al., 2020).
In cells, a disulfide bond formed between flexible C-terminal regions of MICU1 and MICU2 helps stabilize the heterodimer (Kamer et al., 2019;Patron et al., 2014;Petrungaro et al., 2015). To mimic this condition and discourage the previously observed tendency of purified MICU1 and MICU2 to form homo-oligomers at the high protein concentrations needed for structural studies (Kamer et al., 2019;Patron et al., 2014;Wang et al., 2014;Wu et al., 2019;Xing et al., 2019), we expressed MICU1 and MICU2 as a single polypeptide with a linker connecting the two subunits. A further motivation for this approach was the expectation that a MICU1-MICU2 concatemer would facilitate interpretation of the cryo-EM data, as the sample might otherwise contain mixtures of homo-and heterodimers that could be difficult to distinguish computationally. By combining the purified MICU1-MICU2 and MCU-EMRE-nanodisc assemblies we obtained a monodisperse holocomplex sample that was used for cryo-EM analysis (Figure 1-figure supplement 3).
Cryo-EM data were collected using a low (<1 nM) concentration of Ca 2+ in order to investigate the resting state of the holocomplex. From a large dataset containing approximately 17 million particle images, single particle analysis indicated that most channels were associated with one MICU1-MICU2 heterodimer ( Figure 1A, Figure 1-figure supplement 4A). Images containing one MICU1-MICU2 heterodimer per channel yielded a 3D reconstruction of the holocomplex at 3.3 Å resolution ( Figure 1, Figure 1-figure supplement 4, Figure 1-figure supplement 5). Two MICU1-MICU2 complexes were observed in approximately 2% of channel assemblies ( Figure 1A, Figure 1-figure supplement 4A, as assessed from 2D class averages of side views). This subset of data yielded a~10 Å resolution 3D reconstruction (Figure 1-figure supplement 4C). In it, one MICU1-MICU2 heterodimer associates with the MCU/EMRE channel in the same manner as in the high-resolution structure. Density for the second MICU1-MICU2 heterodimer is considerably weaker and positioned to the side, next to the channel, where it might interact with the lipid nanodisc. This 3D reconstruction and the 2D class averages, which display blurring of the second heterodimer (Figure 1-figure supplement 4A), indicate that the second one does not have a discrete binding location. Except where noted, the discussion that follows pertains to the high-resolution structure of the holocomplex that contains one MICU1-MICU2 heterodimer. The cryo-EM density of this structure is well defined, especially within the TMD and MICU1 regions, which have the highest local resolution ( Table 1).
Overall structure of the holocomplex MICU1 and MICU2 associate as a heterodimer and bind to the surface of the channel at the IMS entrance of its pore ( Figure 2). Interactions with the channel are mediated solely by MICU1; MICU2 makes no visible contacts with MCU or EMRE. As in cryo-EM and X-ray structures of fungal MCU and metazoan MCU-EMRE channel complexes without MICU1-MICU2 (Baradaran et al., 2018; Nguyen et al., 2018;Wang et al., 2020;Wang et al., 2019;Yoo et al., 2018), the pore comprises a single ion conduction pathway through the TMD that is surrounded by four MCU subunits. EMRE subunits, located at the periphery of the TMD, consist of a single transmembrane helix flanked by short N-and C-terminal disordered regions. The conformation of the pore is indistinguishable from that observed in a cryo-EM structure of TcMCU-EMRE without MICU1-MICU2 ( Figure 2C, Figure 2-figure supplement 1; Wang et al., 2020), which suggests that unlike gating conformational changes in many other ion channels, control by MICU1-MICU2 is not accomplished through motions of pore-lining helices. The structure of the holocomplex provides the most well-defined density to date for the ionselectivity filter, which is formed by a fourfold assembly of the WDXXEP signature sequences of MCU subunits (Figure 1-figure supplement 5E). The 'D' and 'E' residues of this motif (Asp 261 and Glu 264) form two rings of acidic amino acids that line the pore of the channel and can coordinate Ca 2+ through water-mediated ('D' ring) or direct interactions ('E' ring) (Baradaran et al., 2018;Fan et al., 2018;Nguyen et al., 2018;Wang et al., 2020;Wang et al., 2019;Yoo et al., 2018). Strong density that we assign as Ca 2+ on the basis of its coordination and analogy to other structures of MCU is present within the 'E' ring ( Figure 1-figure supplement 5E). Its presence under conditions of <1 nM Ca 2+ identifies this site as the high-affinity (K d 2 nM) Ca 2+ binding site previously postulated from electrical recordings made from mitoplasts that is thought to contribute to the channel's high selectivity for Ca 2+ (Kirichok et al., 2004). MICU1-MICU2 binding does not displace Ca 2+ from this site. The 'D' ring is directly coordinated by MICU1, and possibly because of this and/or the low concentration of Ca 2+ in the sample, density for Ca 2+ is not observed in this site. Aside from coordination of the 'D' ring, the ion conduction pathway is indistinguishable from the unliganded conformation (Figure 2-figure supplement 1; Figure 2C).
Three EMRE subunits are observed at the periphery of the TMD (EMRE 1-3). Density for a fourth, like that observed in structures without MICU1-MICU2 (Wang et al., 2020;Wang et al., 2019), is conspicuously absent ( Figure 1D and E). MICU1 interacts with only EMRE 1, which has the most well-defined density ( Figure 1B same conformation as EMRE 1, which is the same as the one observed without MICU1-MICU2, in which each EMRE transmembrane helix is positioned approximately 45˚from the membrane normal and interacts with TM1 of an adjacent MCU subunit ( Wang et al., 2020). MICU1-MICU2 is positioned~15 Å above where EMRE 4 would be located ( Figure 1D,E). It is possible that interactions between EMRE 4 and MICU1-MICU2, perhaps involving the disordered acidic C-terminal tail of EMRE ( Figure 1-figure supplement 1B), may destabilize binding of this subunit. In support of this hypothesis, nanodisc density is wider in the vicinity of where EMRE 4 would be positioned ( Figure 1D), which suggests that it may be present within the nanodisc but displaced from the surface of MCU.
A uniporter interaction domain and its receptor site MICU1 contains three domains : an N-lobe and a C-lobe, which contain its Ca 2+binding sites, and an N-terminal domain, which we name the uniporter interaction domain (UID) ( Figure 3A). The UID mediates all observed interactions with the channel. The UID comprises three a-helices (a1, a2, a3) and a small b-sheet formed by strands b1, b2 and b3 ( Figure 3A). Residues on a1 and a2 and a tight turn connecting these helices at Pro 124 directly contact the IMS surface of the channel (Figure 3D,F-G).
The binding of the UID identifies a receptor for MICU1 on the surface of the channel ( Figure 3B and C). The receptor is relatively flat and spans three MCU subunits and one EMRE subunit. The amino acids comprising it (MCU residues: Trp 255 B , Tyr 258 A&B , Ser 259 B , Asp 261 A,B&D , and Ile 262 A&B , and EMRE residue Leu 92, Figure 3C) are conserved among metazoan channels and are identical between human and beetle channels ( The a1 helix, which is the N-terminal end of the ordered region of MICU1, is positioned horizontally on the receptor, where it contacts one EMRE subunit (EMRE 1) and two MCU subunits through a collection of van der Waals interactions and hydrogen bonds ( Figure  A hydrophobic interaction between Phe 106 of a1 and the IMS end of EMRE's transmembrane helix at Leu 92, which is also highly conserved (Figure 1-figure supplement 1B), comprises the only contact observed between MICU1 and EMRE ( Figure 3G). Although not visible in the density, MICU1 and EMRE may also interact electrostatically -an acidic C-terminal region of EMRE and a basic region preceding a1, which are disordered, have been implicated in binding (Hoffman et al., 2013;Tsai et al., 2016) and would be in close proximity. A nearby conserved basic surface of MICU1 may be another point of contact for the acidic tail of EMRE (Figure 3-figure supplement 2).
The mouth of the pore is coordinated by Lys 126 and Arg 129 from a2, and Ser 122 of the a1-a2 turn through a network of hydrogen bonds that involves three of the four MCU-Asp 261 residues of the 'D' ring ( Figure 3D,F). This confirms the previous hypothesis that the 'D' ring is involved in MICU1 binding (Paillard et al., 2018;Phillips et al., 2019). The MICU1 residues participating in the hydrogen-bonding network (Tyr 114, Ser 122, Asp 125, Lys 126, Arg 129, and Asp 169) are conserved (Figure 3-figure supplement 3), but none were previously predicted to mediate interactions with the channel (Figure 3-figure supplement 4).
To assess the function of the interacting residues and to gauge their importance and relevance for control of the full-length human MCU-EMRE channel, we evaluated mitochondrial Ca 2+ uptake when wild type and mutant MICU1 proteins were expressed in human MICU1 knockout cells. As expected, robust mitochondrial Ca 2+ uptake was observed when wild type MICU1 was expressed and uptake was suppressed by mutations within the Ca 2+ -binding EF-hands of MICU1 that cause it to constitutively inhibit the channel ( Figure 3E, Figure 3-figure supplement 5; Kamer and Mootha, 2014). Residues that were observed to interact with MCU or EMRE relieve this inhibition when mutated to alanine. This includes not only Lys 126 and Arg 129 of MICU1 that interact with the 'D' ring but also amino acids that interact with MCU and EMRE at a distance from the pore (e.g. Phe 106, Arg 117, and Tyr 121). Lys 126 seems particularly crucial -a single K126A mutation essentially   eliminates inhibition by MICU1. These studies verify the structurally observed interaction and indicate its importance for control of the full-length human channel.
The regions of the UID that interact with the channel are absent in MICU2 or MICU3 (Figure 3figure supplement 1). The a1 helix, the conserved Pro 124 in the bend between a1 and a2, and residues that interact with the 'D' ring are unique to MICU1. This explains why MICU1 is distinct from MICU2 and MICU3 in its ability to interact with MCU/EMRE, and it explains the observation that MICU2 and MICU3 have non-redundant roles in comparison to MICU1 in the regulation of the channel Patron et al., 2014;Patron et al., 2019;Payne et al., 2017;Xing et al., 2019).
The interaction with the pore is reminiscent of the interaction between m-conotoxin KIIIA, a small pore-blocking peptide toxin found in the venom of a cone snail, with the extracellular surface of the neuronal Na v 1.2 channel (Pan et al., 2019;Zhang et al., 2007; Figure 3-figure supplement 6). A short a-helical region of the toxin lies horizontally at the entrance of the pore of Na v 1.2, in a similar manner to the binding of the a2 helix of MICU1 with the uniporter. As in the UID-uniporter interface, an arginine and a lysine residue that are separated by one turn of the toxin's a-helix coordinate acidic amino acids at the entrance of the selectivity filter ( Figure 3D and Figure 3-figure supplement 6). The binding of the toxin is slightly off-center such that two of four analogous acidic amino acids are coordinated and the pore is not completely occluded -nevertheless this toxin blocks approximately 90% of Na + current though the channel (Zhang et al., 2007; Figure 3-figure supplement 6). The UID interacts with three of the four 'D'-ring amino acids at the mouth of the uniporter's pore and the occlusion of the pore is more extensive than observed for Na v 1.2 by mconotoxin KIIIA (Figure 2, Figure 3, Figure 3-figure supplement 6). The structure of the holocomplex and its similarity to a pore-blocking toxin complex indicate that MICU1 binding inhibits the flow of Ca 2+ ions by both obstructing the IMS entrance of the pore and by shielding the negative charge of the 'D' ring through coordination with basic amino acids.

MICU1-MICU2 and Ca 2+ -induced changes
The MICU1-MICU2 portion of the holocomplex is roughly parallelogram shaped, with overall dimensions of approximately 70 Â 65 Â 35 Å , under resting conditions in which its EF-hand domains are in Ca 2+ -free apo states ( Figure 4A). One of its large relatively-flat surfaces faces and curves slightly toward the membrane due to a slight bend between MICU1 and MICU2 ( Figure 2A). The heterodimer has pseudo twofold symmetry and two analogous interfaces between the subunits ( Figure  From biochemical analysis of the purified protein, and in acoord with a previous report that MICU1-MICU2 has an affinity for cardiolipin (Kamer et al., 2017), we found that the heterodimer associates with lipid membranes and lipid nanodiscs that contain cardiolipin (Figure 4-figure supplement 4, Figure 4F). This association is independent of Ca 2+ -both Ca 2+ -free and Ca 2+ -bound heterodimers bind to liposomes. To investigate the conformational change in MICU1-MICU2 upon Ca 2+ binding, we determined a cryo-EM structure of Ca 2+ -bound MICU1-MICU2 in complex with a lipid nanodisc containing cardiolipin at 3.1 Å resolution ( Figure 4F Table 1). The side of the heterodimer that associates with the nanodisc is the same as the one that faces the channel (Figures 1A and 4F). The structure reveals a substantial conformational change upon Ca 2+ binding that makes the heterodimer more compact and markedly more bent between MICU1 and MICU2 ( Figure 4A-D). Ca 2+ dependent rearrangements of the EF-hand motifs that involve their helix turn helix elements produce this bend ( Figure 4C-D, Figure 4-figure supplement 1). Ca 2+ binding also induces a rotation of the UID relative to MICU1 ( Figure 4E).
We sought to understand how Ca 2+ -dependent conformational changes in MICU1-MICU2 might confer Ca 2+ -dependent control to the channel ( Figure 5). Superimposing the Ca 2+ -bound MICU1- CU CU CU CU2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2       MICU2 structure on the UID in the resting holocomplex reveals that the Ca 2+ -induced conformational changes in the heterodimer would place hydrophilic regions of MICU2 within the membrane, which would be highly unfavorable thermodynamically ( Figure 5, middle). Rather than inserting into the membrane, we hypothesize that the bending motion of MICU1-MICU2 acts like a lever against the membrane to pry MICU1 from the mouth of the pore when [Ca 2+ ] IMS is elevated. Additionally, the Ca 2+ -dependent conformational changes within MICU1 cause a rotation of the UID that could disfavor binding to the channel ( Figure 4E). The conformational changes would dislodge the UID from the blocking conformation, which would allow Ca 2+ ions to permeate though the channel while [Ca 2+ ] IMS remains elevated. A dislodged Ca 2+ -bound MICU1-MICU2 complex would be free to bind to the surface of the inner mitochondrial membrane, with which it would associate due to its inherent affintity for membranes, and it may remain proximal the channel as a result of electrostatic interactions with the acidic tail of EMRE ( Figure 5, right) . When resting [Ca 2+ ] IMS is restored, a MICU1-MICU2 complex, again devoid of its Ca 2+ ligands, would resume the blocking configuration ( Figure 5, left, Figure 5-figure supplement 1). This mechanism of MICU1-MICU2 control is compatible with the observed dimerization of human MCU-EMRE channels that is due to association of their NTDs -there is sufficient room for MICU1-MICU2 complexes to associate with both channels of the dimer and to control Ca 2+ influx through their pores (  . The UID blocks the pore (gray tube). Following elevation of [Ca 2+ ] IMS , Ca 2+ binding causes the MICU1-MICU2 heterodimer to bend (the Ca 2+ -bound MICU1-MICU2 structure is depicted with the channel in the center and right panels). The center panel indicates that bending would dislodge the UID from its receptor site in order to avoid thermodynamically unfavorable interactions of MICU2 with the membrane ('X'). The dislodged Ca 2+ -bound MICU1-MICU2 heterodimer would no longer block the pore, thereby allowing Ca 2+ permeation through the channel, and it would be free to interact with the membrane (right). Upon the return of resting [Ca 2+ ], the heterodimer would resume its blocking conformation (left). One MICU1-MICU2 heterodimer is depicted and only one can bind to the receptor on the channel at a time but multiple MICU1-MICU2 heterodimers may be associated with the channel. A brown line depicts a hypothetical interaction of the acidic C-terminus of EMRE with MICU1. The online version of this article includes the following figure supplement(s) for figure 5:

Discussion
With analogy to classical toxin-channel interactions (MacKinnon and Miller, 1988), the structure of the holocomplex indicates that MICU1-MICU2 prevents mitochondrial Ca 2+ overload under resting conditions  and confers Ca 2+ -dependent control by a pore-blocking mechanism. Operating like a regulatable toxin, resting levels of Ca 2+ permit the blocking conformation while elevated levels disrupt it. A single MICU1-MICU2 heterodimer appears capable of conferring these properties for the simple reason that the receptor site on the channel can accommodate only one UID. However, more than one MICU1-MICU2 heterodimer may be tethered to the channel at any given time, as is indicated by the cryo-EM analysis. Nucleation of more than one heterodimer in the vicinity of the channel would increase the local concentration of MICU1-MICU2 and facilitate a rapid response following restoration of resting cytosolic Ca 2+ levels. While MICU1-MICU2 visibly interacts with only one EMRE subunit, the displacement of an EMRE subunit (EMRE 4) from the TMD might serve as an additional mechanism to control ion flow since the expression of EMRE is necessary for Ca 2+ uptake by mitochondria Sancak et al., 2013;Tsai et al., 2016). Additionally, the displacement of an EMRE subunit suggests that the number of EMRE subunits associated with an MCU tetramer can vary, as has been inferred from a recent study (Payne et al., 2020).
During the preparation of this manuscript, cryo-EM structures of a holocomplex comprising human MCU, EMRE, MICU1 and MICU2 subunits in resting and activated [Ca 2+ ] conditions were reported (Fan et al., 2020). The conclusions of that work are congruous with the present study and validate the proposed mechanism of Ca 2+ -dependent regulation of the uniporter by MICU1-MICU2. Unlike ours, which used lipid nanodiscs, the structures were determined using detergent-solubilized protein, but the absence of a lipid membrane-like environment does not seem to have dramatically influenced the binding of the MICU1-MICU2 regulatory complex to the channel under resting Ca 2+ conditions ( Figure 5-figure supplement 3). Interestingly, an additional interaction with the 'D'-ring at the mouth of the pore, involving a loop region of MICU1 (amino acids 258-274) that is disordered in our structure, was observed ( Figure 5-figure supplement 3). Weak density for this loop in our structure and its disorder in X-ray structures of isolated MICU1 proteins are indications of its flexibility . A structure of an intact holocomplex at elevated levels of [Ca 2+ ], which was included in the recent work, indicates that the Ca 2+ -bound MICU1-MICU2 regulatory complex binds adjacent to the pore in essentially the same manner that we propose ( Figure 5, right, and

Conclusion
This work reveals a mechanism for Ca 2+ -dependent control of the mitochondrial calcium uniporter by the MICU1-MICU2 regulatory complex. The mechanism is distinct from the conformational changes that underlie gating, the process by which ion channels permit or prevent ion condution, in many other ion channels -these often involve changes in the dimensions of their ion conduction pores due to motions of transmembrane a-helices that line these pores, as was first exemplified by structural studies of potassium channels (MacKinnon, 2003). Our current understanding of how changes in [Ca 2+ ] IMS regulate the uniporter comes from the integration of insights from the present work and the large body of preceeding studies. The MICU1-MICU2 complex exists as a heterodimer that is located in the IMS. Under resting cytosolic [Ca 2+ ] conditions, when the concentration of Ca 2+ is less than~0.2 mM, Ca 2+ -binding EF-hand domains within MICU1 and MICU2 are devoid of Ca 2+ and the heterodimer adopts a Ca 2+ -free conformation that is relatively flat. An inherent affinity of MICU1-MICU2 for lipid membranes and an affinity for a flexible acidic region on the IMS portion of EMRE increase the local concentration of the heterodimer near the MCU/EMRE channel. MCU channels that have multiple EMRE subunits associated with them would tend to have more MICU1-MICU2 complexes nearby, as was inferred from a recent study (Payne et al., 2020). At these low resting concentrations of Ca 2+ , a domain that is specific to MICU1, the uniporter interaction domain (UID), binds to a receptor site at the mouth of the channel's pore and blocks it, with analogy to how protein toxins from venomous organisms block neuronal cation channels. In this blocking conformation, the UID inhibits ion permeation that could otherwise dissipate the mitochondrial electromotive force. Although MICU2 interacts extensively with MICU1, it does not appear to contact the channel directly. When the Ca 2+ concentration in the IMS becomes elevated, the binding of Ca 2+ to the EFhand domains within MICU1 and MICU2 induce conformational changes in the MICU1-MICU2 complex that bend it and cause the UID to rotate. These conformational changes relieve blockage of the pore and thereby permit Ca 2+ ions to permeate through the channel while [Ca 2+ ] IMS remains elevated. These processes are reversed upon the restoration of resting [Ca 2+ ] IMS levels. MICU1-MICU2 heterodimers, and by analogy MICU1-MICU3 heterodimers in neurons, thus confer Ca 2+ -dependent gating to the uniporter by operating like endogenous pore-blocking toxins, the binding of which is governed by Ca 2+ -dependent conformational changes within them. This mechanism represents one of the ways that the mitochondrial Ca 2+ uniporter, an unusual and intricate ion channel complex, responds to physiological signals to regulate mitochondrial respiration and other cellular processes.

Materials and methods
Key resources
MICU1 and MICU2 were expressed as a single polypeptide with a linker connecting the two proteins. The construct consists of human MICU1 (residues 94-476) connected to human MICU2 (residues 51-434). Unstructured regions of the polypeptide and a Ser-Asn peptide comprise a 44 amino acid linker between the structured regions of MICU1 and MICU2 (spanning~32 Å in the structure). The expression construct was obtained in a stepwise manner. cDNA for human MICU1 (encoding residues 94-476) was amplified from a normal human brain cDNA library (BioChain, Inc) and ligated into the XhoI and EcoRI sites of an expression plasmid (Goehring et al., 2014) to contain an N-terminal Venus tag and PreScission protease cleavage site (Venus-PreScission-MICU1). cDNA encoding human MICU2 (amino acids 51-434) was amplified in the same manner and inserted into that plasmid, using MfeI/EcoRI and SalI restriction sites, to yield the final expression plasmid (Venus-PreScission-MICU1-MICU2). The expression and purification of MICU1-MICU2 followed a similar procedure to that for MCU and EMRE. The pellet from 0.3 L of cell culture (prepared using the same growth conditions as described above) was resuspended in 20 ml lysis buffer, DDM was added to a final concentration of 1%, and the sample was agitated at 4˚C for 1 hr. The sample was clarified by centrifugation (60,000 g for 1 hr, 4˚C) and the supernatant was filtered through a 0.22 mm polystyrene membrane (Millipore). 2 ml GFP nanobody resin was added and the sample was rocked at 4˚C for 1 hr. The beads were washed with 100 ml buffer (20 mM HEPES pH 7.5, 500 mM NaCl and 1 mM DDM), the MICU1-MICU2 protein was eluted using 0.1 mg PreScission protease (16 hr, 4˚C, supplemented with 1 mM DTT), and the sample was further purified by SEC (Superose 6 Increase, 10/300 GL) in 20 mM HEPES pH 7.5, 150 mM NaCl, 5 mM EDTA, 5 mM EGTA and 1 mM DDM. The peak fractions corresponding to the MICU1-MICU2 protein were pooled and concentrated to 1 mg/ml (Vivaspin 2, 100 kDa cutoff). To remove DDM,~300 mg wet Bio-Beads SM2 (Bio-Rad) were added and the sample was rotated (4˚C, 3 hr). Purified MICU1-MICU2 protein was then combined with the purified with MCU-EMRE-nanodisc sample (using a molar ratio of 2 MICU1-MICU2 : one channel, 4˚C, 30 min.) and the complex was purified by SEC (Superose 6 Increase, 10/300 GL column, GE Healthcare) in 20 mM HEPES pH 7.5, 150 mM NaCl, 5 mM EGTA. The peak fractions were collected, concentrated to 1 mg/ml (Vivaspin 2, 100 kDa cutoff), and used immediately for cryo-EM grid preparation. From the Maxchelator software (Schoenmakers et al., 1992), and assuming a typical trace concentration of Ca 2+ in buffer components (~2-5 mM), we estimate [Ca 2+ ] free~1 00 pM in the sample.

EM sample preparation and data acquisition
For all samples, 4 ml of purified protein was applied to glow-discharged (10 s) Quantifoil R 1.2/1.3 grids (Au 400; Electron Microscopy Sciences) and plunge-frozen in liquid nitrogen-cooled liquid ethane, using a Vitrobot Mark IV (FEI) operated at 4˚C with a blotting time of 2-3 s (blot force 0) with 100% humidity. Grids were clipped and loaded into a 300 keV Titan Krios microscope (FEI) equipped with a Gatan K3 direct electron detector. Micrographs were collected in super-resolution mode (pixel size of 0.532 Å ) with a nominal defocus range of À1.0 to À3.0 mm. The dose rate was 20 electrons/pixel/s. Images were recorded for 4 s with 0.1 s subframes (40 total frames), corresponding to a total dose of 71 electrons per Å 2 .

Image processing and model building
For the holocomplex, 21,115 movie stacks were gain-corrected, twofold binned (using a pixel size of 1.1 Å ), motion corrected, and dose weighted using MotionCor2 (Zheng et al., 2017). Contrast transfer function (CTF) estimates were performed in CTFFIND4 using non-dose weighted micrographs (Rohou and Grigorieff, 2015). 19,972 micrographs with CtfMaxResolution values better than 4 Å were selected for further processing. 17,440,131 particles were picked automatically and extracted (using a binned pixel size of 4.4 Å ) using RELION 3.0 (Zivanov et al., 2018). These particles were then imported into cryoSPARC v.2 (Punjani et al., 2017). The particles were cleaned-up by one round of Ab initio reconstruction and two rounds of heterogeneous refinement. 2,388,041 particles from the best classes were selected and subjected to non-uniform refinement in cryoSPARC v.2, which yielded a reconstruction at~9.4 Å overall resolution. The refined particles were then centered and re-extracted in RELION 3.0 with a pixel size of 2.2 Å , imported into cryoSPARC v.2 for another two rounds of heterogeneous refinement. 761,813 particles from the best classes were chosen and used for non-uniform refinement in cryoSPARC v.2, which yielded a reconstruction at~4.4 Å overall resolution. The refined particles were centered and re-extracted in RELION 3.0 with a pixel size of 1.1 Å , and imported into cryoSPARC v.2 for another two rounds of heterogeneous refinement. 510,869 particles from the best class were subjected into non-uniform refinement in cryoSPARC v.2, and this yielded a reconstruction at 3.9 Å overall resolution. One round of Bayesian polishing in Relion 3.0 improved the resolution to 3.6 Å . After local CTF refinement in cryoSPARC v.2 and another round of Bayesian polishing in RELION 3.0, the particles were further classified by heterogeneous refinement in cryoSPARC v.2 and 3D classification (focused classification on the TMD) in RELION 3.0. 350,160 particles from the classes with a more well-defined TMD region were chosen and subjected to 3D auto refinement in RELION 3.0. Postprocessing in RELION 3.0 (using a calibrated pixel size of 1.064 Å ) yielded the final reconstruction at 3.3 Å . All resolution estimates are based on gold-standard Fourier shell correlation (FSC) calculations. Estimation of the local resolution of the map was performed using Resmap (Kucukelbir et al., 2014).
For the structure of MICU1-MICU2 bound to nanodiscs and in 1 mM Ca 2+ , 4675 movie stacks were gain-corrected, twofold binned (using a pixel size of 1.064 Å ), motion corrected, and dose weighted using MotionCor2 (Zheng et al., 2017). Contrast transfer function (CTF) estimates were performed in CTFFIND4 using non-dose weighted micrographs (Rohou and Grigorieff, 2015). 4280 micrographs with CtfMaxResolution values better than 5 Å were selected for further processing. Particles were picked using Gaussian blob or 2D template-based autopicking and extracted (3,957,711 or 4,397,598 particles, respectively, with a binned pixel size of 2.128 Å ) using RELION 3.0 (Zivanov et al., 2018). The two sets of particles were then imported into cryoSPARC v.2 to do processing separately (Punjani et al., 2017). For the particles from Gaussian autopicking, the particles were cleaned-up by one round of Ab initio reconstruction and three rounds of heterogeneous refinement (using seven classes), and 135,522 particles from the best classes were selected and subjected to non-uniform refinement in cryoSPARC v.2, which yielded a reconstruction at~4.3 Å overall resolution. For the particles picked using 2D templates, particles were cleaned-up by three rounds of heterogeneous refinement and 152,512 particles from the best classes were selected and subjected to non-uniform refinement in cryoSPARC v.2, which yielded a reconstruction at~4.3 Å overall resolution. The two sets of particles were then merged, centered and re-extracted in RELION 3.0 with a pixel size of 1.064 Å , and duplicate particles or particles with CtfMaxResolution values worse than 4 Å were removed. The resulting 181,934 particles were imported into cryoSPARC v.2 for additional rounds of heterogeneous refinement. 142,055 particles from the best classes were chosen and used for non-uniform refinement in cryoSPARC v.2, which yielded a reconstruction at~3.7 Å overall resolution. One round of Bayesian polishing in Relion 3.0 improved the resolution to 3.5 Å . After another round of Bayesian polishing in RELION 3.0, followed by CTF refinements (global and local) and several rounds of heterogeneous refinement in cryoSPARC v.2 (139,974 particles were selected), the resolution was improved to 3.2 Å . The selected particles were subjected to additional Bayesian polishing in RELION 3.0, followed by CTF refinement (global and local) and heterogeneous refinement in cryoSPARC v.2 (yielding 115,687 particles), which yielded the final reconstruction at 3.1 Å resolution. All resolution estimates are based on gold-standard Fourier shell correlation (FSC) calculations. Estimation of the local resolution of the map was performed using Resmap (Kucukelbir et al., 2014).
Mitochondrial Ca 2+ uptake experiments cDNA encoding full-length wild type human MICU1 was amplified from a normal human brain cDNA library (BioChain, Inc), whereas an EF-hand mutant MICU1 gene with mutations abolishing Ca 2+ binding (D231A, E244K, D421A, E432K) was chemically synthesized (Genewiz, Inc). Both genes were sub-cloned into the XhoI and EcoRI sites of a mammalian expression vector that includes a C-terminal Rho-1D4 antibody tag (Baradaran et al., 2018;Molday and MacKenzie, 1983). Mutations F106A, R117A, Y121A, K126A and R129A were generated by PCR using mutagenic primers on the background of the EF-hand mutant and verified by sequencing. MICU1 knockout cells, derived from human HEK-293T cells and generously provided by V. Mootha ) (authenticated using mitochondrial Ca 2+ uptake assays (Figure 3-figure supplement 5); tested negative for mycoplasma), were grown in DMEM media (Invitrogen). 3 mg of each plasmid was transfected into~1.5Â10 6 of these cells using Lipofectamine 3000 (Invitrogen), and the cells were grown for an additional 24 hr at 37˚C before use.
The mitochondrial Ca 2+ uptake assay using these cells was performed as described (Baradaran et al., 2018), with slight modifications. Briefly, after addition of 1 mM Calcium Green-5N (Life Technologies) and 0.05 mM digitonin, 20 mM CaCl 2 (from a 4 mM stock in water) was added at the 100 s time point to initiate mitochondrial Ca 2+ uptake. The data plotted in Figure 3E were obtained from Ca 2+ uptake curves (e.g. Figure 3-figure supplement 5) in the following manner. A rate of Ca 2+ uptake (uptake_rate) was defined as the slope of a linear fit of the Ca 2+ uptake curve between 111 and 135 s. Relative uptake was then defined as: (uptake_rate UID-mutant -uptake_rate EFhand-mutant ) / (uptake_rate WT -uptake_rate EF-hand-mutant ), where 'WT' represents wild type MICU1, 'EF-hand-mutant' represents MICU1 bearing the EF-hand mutations, and 'UID-mutant' represents a given mutation of the UID on the background of the EF-hand-mutant. From this equation, a relative uptake of 1 indicates a mitochondrial Ca 2+ uptake rate equivalent to that observed for wild type MICU1 and a relative uptake of 0 indicates a mitochondrial Ca 2+ uptake rate equivalent to that observed for the EF-hand-mutant.

Assessing the association of MICU1-MICU2 with liposomes
To prepare liposomes, 5 mg of dried lipids (POPE: POPG: cardiolipin [18:1] with a 2:2:1 wt ratio) was mixed with 1 ml reconstitution buffer (20 mM HEPES pH 7.5, 150 mM NaCl) and the sample was sonicated until homogeneous (~2 min). 20 ml of purified MICU1-MICU2 protein (0.5 mg/ml) was added to 80 ml of liposomes or 80 ml of reconstitution buffer, each supplemented with a final concentration of 1 mM CaCl 2 or with 1 mM EGTA and 1 mM EDTA. As a negative control, 20 mg of purified green fluorescent protein was combined with liposomes or buffer in the same manner. After incubation for 30 min at 4˚C, the samples were centrifuged at 140,000 g (30 min at 4˚C). The supernatants were removed and the pellets were resuspended for SDS-PAGE analysis (by resuspending the pellets in 40 ml reconstitution buffer, adding 40 ml SDS-PAGE loading buffer containing 100 mM DTT, and analyzing 20 ml by Coomassie-stained SDS-PAGE).