Introduction

Magnesium (Mg2+), the most abundant divalent cation in living organisms, plays an essential role in many biological processes, including ATP synthesis and hydrolysis, DNA replication, protein synthesis, modulation of enzymatic activity, and protein stability1,2. It acts as a cofactor of more than 600 enzymes, including protein kinases, ATPases, exonucleases, and other nucleotide-related enzymes. It has been shown that Mg2+ is involved in various physiological functions such as muscle contraction, vasodilation, neuronal signaling, and immunity1,2,3. Intracellular Mg2+ concentrations are tightly regulated and dysregulation of Mg2+ homeostasis is associated with diseases including muscular dysfunction, bone wasting, immunodeficiency, cardiac syndromes, neuronal disorders, obesity, Parkinson’s disease, and cancer2,4,5,6.

The total Mg2+ content in cells amounts to between 17–30 mM, however, most Mg2+ is bound to ATP and other molecules, resulting in much lower free Mg2+ concentrations between 0.5–1.2 mM2,7. In mitochondria, where ATP synthesis and various metabolic processes occur, Mg2+ modulates enzymes involved in tricarboxylic acid cycle (TCA), oxidative metabolism and directly participates in mitochondrial metabolism in form of MgATP2-8,9. Although the high inner mitochondrial membrane potential (about −180 mV) can theoretically drive a large electrophoretic Mg2+ influx into the mitochondrial matrix, the free Mg2+ concentration in the matrix is found to be comparable to that in the cytosol (~0.8 mM), implying Mg2+ transport is tightly regulated in order to maintain normal mitochondrial physiology10. Most of the Mg2+ influx into mitochondria is mediated by the Mitochondrial RNA Splicing 2 (MRS2) channel, located in the inner mitochondrial membrane11,12,13,14. Knockdown of MRS2 in human cells leads to reduced uptake of Mg2+ into mitochondria, loss of respiratory complex I, disruption of mitochondrial metabolism, and cell death15,16. A loss of function mutation disrupting MRS2 is also associated with demyelination syndrome in rats17. Recent studies in mice show that MRS2 is required for lactate-mediated Mg2+-uptake in mitochondria18, and knockout of MRS2 causes reprogramming of the metabolism including upregulation of thermogenesis, oxidative phosphorylation and fatty acid catabolism via HIF1α transcriptional regulation19. Despite the physiological implications of MRS2, little is known about its Mg2+ translocation and regulatory mechanisms.

MRS2 belongs to the CorA protein superfamily characterized by a highly conserved Glycine-Methionine-Asparagine (GMN) motif in the loop between transmembrane helix 1 and 2. In prokaryotes, CorA regulates the intracellular Mg2+ concentration through a negative feedback mechanism, where at low Mg2+ concentrations, unbinding of Mg2+ ions from the CorA soluble domain favors a series of conductive states, allowing Mg2+ permeation20,21. MRS2 and CorA only share a sequence identity of ~14%, so insight into MRS2 Mg2+ translocation and regulatory mechanism based on CorA has been limited. Until very recently, the only MRS2 structure reported is the monomeric N-terminal soluble domain of yeast MRS222, lacking the transmembrane domain. A recent study of the N-terminal domain of human MRS2 revealed it forms a dimer, contrasting the pentameric assembly of CorA23. However, the missing structural information of the MRS2 full-length protein including the pore region hampers the understanding of how Mg2+ translocation occurs through the pore and how MRS2 is regulated.

Here, we use single particle cryo-EM to determine the structures of human MRS2 in the presence and absence of Mg2+. N-terminal sequencing of MRS2 revealed the mitochondrial transit peptide cleavage site at residue 71. From the homo-pentameric structures, M336 and R332 are identified as key gating residues, which are further tested by mutagenesis, and Mg2+-dependent growth and Ni2+-sensitivity assays. The structures also reveal that MRS2 possesses two Mg2+-binding motifs in the soluble domain between neighboring subunits, which are different from known bacterial CorA structures. Finally, an inter-subunit salt bridge between R116 and E291 in the soluble domain is found. Disruption of this salt bridge leads to an increase in channel activity. Altogether, this study provides insight into how translocation of Mg2+ is mediated via MRS2 and how it is regulated.

Results

Expression, purification, and biochemical characterization of human MRS2

MRS2, a mitochondrial membrane protein encoded in chromosome 6 of the nuclear genome, possesses a mitochondrial transit peptide (MTP) at its N-terminus, which is cleaved after translocation into the inner mitochondrial membrane. As such, we expressed human MRS2 conjugated with a C-terminal FLAG tag in Expi293F cells and purified it in the presence of Mg2+ using affinity and size-exclusion chromatography (Supplementary Fig. 1). Purified MRS2 shows a band slightly above 38 kDa based on SDS-PAGE (Supplementary Fig. 1b), which is smaller than the full-length size at 51 kDa, suggesting the N-terminal MTP has been cleaved in purified MRS2. Native-PAGE of purified MRS2 shows a major band between the native marker at 242 kDa and 480 kDa indicating MRS2 assembles into an oligomer (Supplementary Fig. 1c). Negative-staining EM followed by 2D classification of MRS2 particles shows a funnel-shaped structure, resembling that observed in CorA (Supplementary Fig. 1d). To identify the cleavage site of the MTP in human MRS2, online prediction webservers were initially used. The cleavage site was not detected using Target-2.024 and SignalP-6.025, while it was predicted at residue 21 by Mitofates26. These prediction results were contrary to the Uniport entry (Q9HD23) which suggested a cleavage at residue 50. Here, we used an experimental approach, N-terminal sequencing, of purified human MRS2 and the first amino acid detected is residue 71 starting with threonine (Supplementary Fig. 1e). To confirm the identity of the MTP, full-length MRS2 or truncated MRS2(71-443) conjugated with GFP were expressed in Expi293F cells and imaged using confocal microscopy. It shows that the full-length MRS2-GFP localizes in mitochondria (Supplementary Fig. 1f), while MRS2(71-443)-GFP can no longer be imported into mitochondria and largely localizes in the ER (Supplementary Fig. 1g).

Structural analysis of human MRS2 in the presence of Mg2+ using cryo-EM

Cryo-EM images of purified MRS2 were recorded using a 300 kV Titan Krios equipped with a K3 camera and energy filter (Supplementary Fig. 2a). Initial 2D class averages (Supplementary Fig. 2b) confirmed the by negative-staining EM observed pentameric arrangement of human MRS2, which has been shown for bacterial21,27,28,29,30,31 and archael32 members of this protein family and is contrary to the most recent prediction of human MRS2 to form a dimer23. The final 3D reconstruction of MRS2 in the presence of Mg2+ (referred to as MRS2-Mg2+) with C5 symmetry and without symmetry applied (C1) resulted in maps with average resolutions of 2.8 Å and 3.1 Å, respectively, with the highest local resolution estimated to be 2.3 Å (Fig. 1 and Supplementary Figs. 2c–f, 3–4).

Fig. 1: Structure of human MRS2 in the presence of Mg2+.
figure 1

a The 2.8 Å cryo-EM density of MRS2 shown in side and bottom views. Density corresponding to each subunit is delineated by different colors. b, c Fitted structural model of MRS2 shown as pentameric complex (b) and as single subunit (c). Magnesium ions bound to MRS2 are represented as green spheres. d Topology of MRS2 showing α-helices as rods and β-strands as slightly darker arrows.

Full size image

The human MRS2 pentamer exhibits a funnel-shaped structure with dimensions of approximately 130 Å by 100 Å by 100 Å (H x W x D) (Fig. 1). No ordered density representing detergent molecule or co-purified lipid is observed in the transmembrane region. The overall architecture of MRS2 resembles that of homologs in the CorA superfamily including TmCorA21,27,28,29,30, EcCorA31, MjCorA32, EcZntB33, and PaZntB34 (Supplementary Fig. 5), even though they share low sequence identity (Supplementary Fig. 6).

Each of the human MRS2 protomers consists of a large N-terminal soluble domain facing the mitochondrial matrix side. The N-terminal soluble domain is composed of six anti-parallel β-strands and seven α-helices arranged in a α/β/α fold manner (Fig. 1c, d). The arrangement of the α/β/α fold in MRS2 is different from that of TmCorA, which possess seven anti-parallel β-strands and six α-helices (Fig. 1d and Supplementary Fig. 5). The soluble domain connects to the long α8/TM1 helix spanning a total of 71 residues (residues 290–360). The α8/TM1 helix starts from the far matrix side of MRS2 while the C-terminal part of α8 enters the inner mitochondrial membrane, serving as the first transmembrane helix TM1. Following the long α8/TM1 helix, a loop in the intermembrane space containing the highly conserved GMN motif connects to the α9/TM2, which ends at the matrix side. A channel pore connecting intermembrane space and matrix is formed between the intertwined α8/TM1 helices from five subunits. Four Mg2+ ions (termed Mg2+−1, 2, 3, 4) were identified along the pore and two (termed Mg2+−5, 6) were found in between neighboring soluble domains in the MRS2-Mg2+ structures (Fig. 1b, c and Supplementary Fig. 4).

Mg2+-translocation pathway and gating mechanism

The five TM1 helices form the wall of the translocation pore, while mostly hydrophilic and charged side chains point to the center (Fig. 2a, b). Extra densities corresponding to four Mg2+ ions were found in the center along the pore (Fig. 2a, b and Supplementary Fig. 4). Although these extra densities are located at the center of the C5 symmetry axis and should therefore be interpreted carefully, they are also present in the non-symmetric C1 map and with that are unlikely to be artifacts generated by symmetry refinement (Supplementary Fig. 4b). The upper most Mg2+ (Mg2+-1), closest to the mitochondrial intermembrane space is coordinated by the highly conserved GMN motif. Mg2+-1 interacts with a ring of backbone carbonyl oxygens of G360 and amine group of N362 with a distance of 3.8 Å and 4.3 Å, respectively, indicating that the Mg2+-1 is in hydrated form35,36. Mg2+-2, Mg2+-3 and Mg2+-4 are coordinated by a ring of hydroxyl oxygens of T346, carbonyl oxygens of N339, and carboxylic oxygens of D329 with a distance of 5.2 Å, 4.3 Å and 3.5 Å, respectively. A highly negatively charged surface was observed in the loop located at the intermembrane space and the soluble domain in the matrix facing the pore (Fig. 2a). A similar surface has also been observed in CorA and is believed to provide a pulling force for the transport of Mg2+ ions28,30. The radius plot estimated by the program HOLE37 shows the narrowest region of the pore is formed by a ring of M336 side chains, with a pore radius of 1.7 Å, followed by R332, and N362/G360, which is part of GMN motif (Fig. 2c). Owing to (1) the hydrophobicity of M336, (2) the repulsive positive charge at R332, and (3) a pore size that is clearly too narrow to allow hydrated Mg2+ to pass through, M336 and R332 likely represent candidates to act as gating residues capable of interrupting the ion flow. M336 is conserved among members of the CorA family, for instance, it corresponds to M291 in TmCorA (Supplementary Figs. 5, 6). TmCorA relies on hydrophobic gates M291, L294, M302 (referred to as MM stretch) to control Mg2+ flux38,39,40. Interestingly, R332 is conserved among eukaryotes and for example also found in yeast Mrs2 but not in prokaryotic CorA homologs (Supplementary Fig. 6), suggesting the Mg2+-translocation mechanism may be different in eukaryotic MRS2 compared to prokaryotic CorA. In MRS2, the positive charge of R332 is partially masked by D329 from the neighboring subunit one helix turn below, lowering the energy required for Mg2+ to pass through the R332 gate (Fig. 2b, d, e). Furthermore, we found a network of hydrogen bonds from S224 in the loop between soluble α-helix 3–4 of an adjacent subunit to N333 and the potential gating residue R332 (Fig. 2d, e). This network might allosterically link the soluble domain and pore gating residue R332, opening the possibility of a regulation mechanism in which the soluble domain also plays a role.

Fig. 2: Mg2+-translocation pathway in MRS2.
figure 2

a Sliced side view of MRS2 illustrating the translocation pathway of Mg2+. The electrostatic potential of the surface is colored (from negative in red to positive in blue). Insets refer to regions highlighted in (b). b Enlarged view of the GMN motif and pore region highlighting residues coordinating Mg2+ ions. Potential gating residues M336 and R332 are labeled in red. Only helices from subunits B, C, and D are shown for clearer illustration. c Pore radius plot of MRS2. d, e Network of hydrogen bonds connecting gating residue R332 and the loop between soluble α-helix 3–4 of an adjacent subunit. Enlarged view of the interaction is shown in (e). f Mg2+-auxotrophic growth complementation assay using the empty vector and TmCorA (controls), MRS2 and its mutants. Serially diluted Mg2+-auxotrophic E. coli (BW25113 ΔmgtA, ΔcorA, ΔyhiD DE3) containing corresponding plasmids were spotted onto LB plates with and without MgSO4, and grown at 30 °C. g Ni2+-sensitivity assay of BL21-(DE3) expressing TmCorA (control), MRS2 and its mutants. Serially diluted E. coli containing corresponding plasmids were spotted onto LB plates with and without 1.8 mM NiCl2, and grown at 30 °C overnight.

Full size image

To evaluate the potential gating role of R332 and M336, we tested Mg2+-translocation activity of R332A and M336A mutants by using a Mg2+-dependent growth assay as described previously41,42. The Mg2+-auxotrophic E. coli strain BW25113 lacking the major Mg2+-transporters/channels (mgtA, corA, yhiD) can only grow either by supplementing the growth medium with high Mg2+ concentrations, or by complementation with a functional Mg2+-transporter/channel. Although expression of the WT HsMRS2 and single gating mutants R332A or M336A is not sufficient to support the growth of the strain without the supplementation of Mg2+, double mutant R332A/M336A complements the growth synergistically (Fig. 2f).

A Ni2+-sensitivity assay has been employed in characterizing magnesium channel MgtE and its mutants previously41,43. As coordination of Ni2+ is similar to that of Mg2+, and Mrs2 in yeast is able to translocate Ni2+ ions like CorA, although it has a higher selectivity for Mg2+ over Ni2+14,36, we utilized the toxicity of Ni2+ to E. coli to evaluate Ni2+-uptake and HsMRS2 channel activity. E. coli expressing WT HsMRS2 shows a higher Ni2+-sensitivity over the negative control of E. coli with empty vector only, indicating HsMRS2 does have some channel activity in E. coli (Fig. 2g). Notably, there is no significant increase in Ni2+-sensitivity in single gating mutants R332A or M336A. Double gating mutant R332A/M336A shows increased Ni2+-sensitivity indicating more uptake of Ni2+ compared to WT or single mutants. Together with the complementation growth assay using the Mg2+-auxotrophic E. coli strain, it suggests both R332 and M336 residues are involved in gating of Mg2+ and other divalent cations in HsMRS2, while both gates need to be opened to allow Mg2+ to pass through.

Mg2+-binding sites in the soluble domain and their role in channel activity regulation

Two additional Mg2+ ions (Mg2+-5 and Mg2+-6) have been identified in the interface of neighboring subunits between their soluble N-terminal domains termed soluble sites 1 and 2 (Figs. 1b, c and 3a–c). The Mg2+ ions in soluble sites 1 and 2 are coordinated largely by negatively charged residues. Mg2+-5 and Mg2+-6 are hydrated, with some water molecules around involved in coordination of Mg2+ with distances of 2.1 - 3.1 Å (Fig. 3b). Mg2+-5 in site 1 interacts with side chains of T246, D247, E138, carboxyl oxygen of E243, and E312 from the neighboring subunit (Fig. 3b, c). Mg2+-6 in site 2 is sandwiched between two negatively charged glutamate clusters from two adjacent subunits. E261, E297 and E293 contribute most in the coordination of Mg2+-6 with distances of 4.4–4.6 Å, while E263, E267, and E290 are about 6–7 Å from the Mg2+. The Mg2+-binding sites in the soluble domain of MRS2 are located at different locations compared to TmCorA and EcCorA structures (Fig. 3e and Supplementary Fig. 5). Sequence alignment also shows the residues involved in soluble site 1 and 2 of MRS2 are not conserved in prokaryotic homologs (Supplementary Fig. 6). When overlaying the HsMRS2 structure to the archaeal MjCorA X-ray structure32, in which 28 Mg2+ ions were modeled into the soluble domain in an asymmetric manner, the Mg2+ (atom spec: MgE406) between subunit E and A of MjCorA is close to soluble site 1 in MRS2; however, it is coordinated by E155, Y186, T90 and Q62, which is different to the groups of negatively charged residues (E138, E243, D247, E312) in MRS2 (Supplementary Fig. 5). Interestingly, adjacent to soluble site 2 of HsMRS2, a salt bridge is observed between R116 from one subunit and E291 from the neighboring subunit, with a distance of 2.7–2.8 Å (Fig. 3d, Supplementary Fig. 4c).

Fig. 3: Two Mg2+-binding sites in the MRS2 soluble domain and their role in channel activity regulation.
figure 3

a Two Mg2+-binding sites are observed in the MRS2 soluble domain between subunits. Insets refer to regions highlighted in (b, c). b Mg2+ in the soluble domain coordinated by negatively charged residues including E243, D247, and E138 from one subunit and E312 from the neighboring subunit in soluble binding site 1. In the soluble binding site 2, Mg2+ is coordinated by E261, E263, and E267 from one subunit and E290, E293, E297 from the neighboring subunit. Water molecules in soluble sites 1 and 2 are shown in red spheres. c Detailed coordination of Mg2+ ions in the soluble site 1 and 2. d Salt bridge formed between R116-E291 pair across subunits. Enlarged view of the R116-E291 is shown (lower panel). e Comparison of the Mg2+-binding sites in the soluble domain between HsMRS2 and TmCorA. Structure of TmCorA in closed state (PDB: 3JCF) in pink is overlaid with the MRS2 structure in blue. The Mg2+ ions in the soluble domain of TmCorA and MRS2 are represented as spheres in red and green, respectively. f Mg2+-auxotrophic growth complementation assay using MRS2 and its mutants. Serially diluted Mg2+-auxotrophic E. coli (BW25113 ΔmgtA, ΔcorA, ΔyhiD DE3) containing corresponding plasmids were spotted onto LB plates with and without MgSO4, and grown at 30 °C.

Full size image

It has been proposed that the Mg2+-binding sites in the soluble domain of TmCorA serve as a Mg2+-sensing motif, where unbinding of Mg2+ in those regions under low intracellular Mg2+ concentration facilitates channel opening and thereafter Mg2+ influx20,21. Cryo-EM study of TmCorA shows that it undergoes dramatic asymmetric conformational changes of the soluble domain under low-Mg2+ conditions27. To investigate whether soluble sites 1 and 2, along with the R116-E291 pair in HsMRS2 are involved in channel regulation, glutamate residues involved in corresponding interactions were mutated to lysine residues and tested using the Mg2+-dependent growth assay. Surprisingly, no noticeable change in growth was observed in the soluble Mg2+-binding site mutants when compared to WT, while the E291K mutation disrupting the salt bridge between adjacent subunits promotes Mg2+-dependent growth, suggesting higher MRS2 channel activity (Fig. 3f). We speculate that the increase in MRS2 channel activity of the E291K mutant is a result of higher flexibility of the soluble domain that allows the propagation of movement to the TM regions leading to pore opening.

MRS2 structure under EDTA condition

To investigate whether MRS2 undergoes conformational changes in response to environmental Mg2+ concentrations, we depleted most of the free Mg2+ in purified MRS2 using dialysis and addition of EDTA and performed structural analysis by cryo-EM. The resulting structures of MRS2 in the presence of 1 mM EDTA (referred to as MRS2-EDTA) with C5 symmetry and without symmetry applied (C1) were determined at an average resolution of 3.3 Å and 3.6 Å, respectively, with the highest local resolution estimated to be 2.7 Å (Fig. 4a, b). Unlike TmCorA, no significant structural changes can be observed when EDTA was included in the sample buffer. The Cα RMSD between MRS2-Mg2+ and MRS2-EDTA is 0.481 Å (Supplementary Fig. 7a). While inspecting the Mg2+-binding sites found in the HsMRS2 structure in the presence of Mg2+, we do observe scattered densities in soluble site 2 but not soluble site 1 (Supplementary Fig. 7b). However, due to (1) the lower overall and local resolution of the MRS2-EDTA structure and (2) under EDTA conditions Mg2+ is not expected to be found in the position that is exposed to solvent, we speculate that these extra densities may represent water molecules coordinated by surrounding negatively charged residues. We have not modeled anything into these extra densities due to the high uncertainty. In the central pore regions, densities for Mg2+−1, 2, 3 but not for Mg2+-4 were observed along the pore of MRS2 even after dialysis and addition of EDTA which should remove most of the free Mg2+ in the solution (Fig. 4c). The GMN motif is known to have a high affinity towards Mg2+ with an estimated KD of 1.3 µM36, which may be able to trap Mg2+ in the pore in the presence of EDTA. The conservation of Mg2+−1, 2, 3 and the loss of Mg2+-4, which is just below the R332/M336-rings, further suggests the gating properties of these residues and that MRS2 remains at low channel open probability without activation.

Fig. 4: Structure of MRS2 under EDTA condition.
figure 4

a Cryo-EM density map of MRS2 in the presence of 1 mM EDTA at 3.3 Å. b Model of the MRS2-EDTA structure. c Comparison of MRS2-Mg2+ and MRS2-EDTA showing the loss of Mg2+ beneath the R332-ring.

Full size image

Discussion

In this study, we determined the structures of the human mitochondrial magnesium channel MRS2 in the presence of Mg2+ and under EDTA condition. MRS2 assembles into a homo-pentamer and displays an overall architecture similar to structures in the CorA family, with slight differences in the secondary structure arrangement (Supplementary Fig. 5). MRS2 possesses a central pore for ion permeation. Near the intermembrane side of the pore, a conserved GMN motif is located to capture Mg2+ ions. Besides the naturally present high membrane potential across the inner mitochondrial membrane, a negative surface potential at the pore-facing side of the soluble domain provides a pulling force for Mg2+ influx. Along the pore, R332 and M336 have been identified as the main gating residues. Double mutations of these residues significantly increase Mg2+- and Ni2+-uptake, suggesting a higher channel activity. Single mutants did not change the channel activity compared to the wildtype. Loss of the Mg2+ ion below the M336/R332 gate towards to mitochondrial matrix in the MRS2-EDTA structure further supports the role of M336 and R332 in gating. Strikingly, the R332 gate does not seem to be conserved among prokaryotic CorA homologs, suggesting that MRS2 may adopt a different Mg2+-translocation and regulatory mechanism.

In TmCorA, two Mg2+-binding sites (M1 and M2) have been identified in the soluble domains of adjacent subunits28,29,30. Mg2+ ions are coordinated by D89 and D253 at the M1 site, and D253, E88, L12, D175 at the M2 site. It has been proposed that these M1 and M2 sites serve as divalent cation sensors responding to intracellular ion concentrations and modulating channel activation. Unbinding of Mg2+ in the low affinity M1 and M2 sites under low intracellular Mg2+ concentration facilitates channel opening. TmCorA D253K mutation abolished the Mg2+-dependent protease susceptibility28 and Mg2+-dependent channel inhibition20. Previous EPR, cryo-EM and MD-simulation studies of CorA show dramatic movements of the soluble domain under EDTA condition and suggest that the inter-subunit binding of Mg2+ in the soluble domain holds the five subunits in the closed, less flexible conformation20,27,44. In the present human MRS2-Mg2+ structure, two inter-subunit Mg2+ ions were also identified but these are in different positions when compared to those in CorA (Fig. 3e and Supplementary Fig. 5), and employ different sets of residues in Mg2+ coordination (Supplementary Figs. 5 and 6). Mutating residues in soluble sites 1 and 2 from glutamate (E) to lysine (K) shows no observable differences when compared to WT in the Mg2+-dependent growth assay. One possibility is that in MRS2 the E to K mutation facilitates salt bridge formation to the glutamate residues of the adjacent subunit, promoting the closed conformation of the channel. This idea is supported by the loss of function phenotype of the D253K TmCorA mutant using cellular and fluorescence based Mg2+-uptake assays38. On the other hand, introducing a E291K mutation in HsMRS2, which disrupts a salt bridge between adjacent subunits and likely mimics low-Mg2+ conditions, shows a gain of function phenotype (Fig. 3f), potentially due to increased flexibility in the soluble domain which might be associated with the regulation of MRS2 channel activity.

During our final manuscript preparation, additional structures of human MRS2 (hMrs2) under various conditions have been reported45. As in the present study, hMrs2 assembles into a homo-pentamer and displays a similar structure in the presence of Mg2+, low EDTA, high EDTA and without Mg2+ and EDTA (termed hMrs2-rest). In general, the reported structures45 are in agreement with ours, with 0.652 Å Cα RMSD between structures in the presence of Mg2+. Two Mg2+ ions are identified along the pore of the reported structures, corresponding to Mg2+−1, 4 in our study. Two additional Mg2+ ions (Mg2+−2, 3) have been found in both of our structures, coordinated by a ring of hydroxyl oxygens of T346 and carbonyl oxygens of N339 with a distance of 5.2 Å and 4.3 Å, respectively. We reason that the additional two Mg2+ ions identified is likely due to higher Mg2+ concentration (40 mM) in our sample, compared to 20 mM used in the previous publication45. Interestingly, a tentative Cl− has been placed in the R332-ring which is proposed to facilitate Mg2+-permeation by masking the electrostatic repulsion caused by the R332-ring45. Besides, MD-simulation demonstrates that the presence of Cl− around the R332-ring lowers the energy barrier for Mg2+ permeation. When we built the MRS2-Mg2+ and MRS2-EDTA models based on our cryo-EM maps, we also see a consistent density above the R332-ring in both structures (Supplementary Fig. 8). Based on the coordination by the R332 residues, we speculate the density may be contributed by a water molecule or possibly an anion such as a chloride ion. Although it is tempting to speculate a chloride ion around the positively charged R332-ring, the presence of a chloride ion in the pore is challenged by the negatively charged pore entry near the GMN motif and the D329-ring, which create a barrier for a chloride ion to enter the pore from either side (Fig. 2a, b). Of note, a previous MD-simulation study on CorA has observed that hydration events occur along the pore including the hydrophobic MM stretch, lowering the free energy barrier for Mg2+ permeation40. Indeed, water molecules near polar residues T343, N362, E368, and carbonyl oxygens of G356 and V357 along the pore can be found in our MRS2-Mg2+ structure. Together, we support an assignment of a water molecule in the density instead of a chloride ion. However, due to the resolution limit in both studies, the identity of the density near R332 remains to be confirmed by additional experimental approaches, such as X-ray anomalous scattering.

Besides a pore dilation and increase in gate flexibility40, other proposed mechanisms of CorA channel opening include helix rotation of TMs displacing the hydrophobic gating residues46, as well as relaxation/bending motion of the soluble domains coupled with TM movement leading to gate opening38,39 (Supplementary Fig. 9). Here, we observed a connection between gating residue R332 and the soluble domain of an adjacent subunit via hydrogen bonds (Fig. 2d, e). We speculate that movement of the soluble domain has a role in the opening of the pore gates in MRS2 (Supplementary Fig. 9e), which is further supported by the increased channel activity upon disrupting the inter-subunit salt bridge between R116 and E291 as mentioned above.

In addition to the gating residue R332, which shows no significant increase in mitochondrial Mg2+-uptake when mutated to R332A, R332K, or R332E in the recent study45, we identified M336 as another gating residue. We found that a single mutation of R332 showed no significant change in channel activity unless M336 was mutated as well, suggesting both gates need to be opened simultaneously or sequentially in order to allow for Mg2+ permeation. A previous electrophysiological study shows that matrix Mg2+ reduces the open probability of yeast Mrs2, suggesting a negative feedback mechanism14. In our study, we observed the loss of Mg2+-4 around the D329-ring in the MRS2-EDTA structure (Fig. 4c). It is tempting to propose that the D329-ring plays a role in the negative feedback mechanism of MRS2. Under low matrix Mg2+ concentration, Mg2+-unbinding from D329 may allow the interaction between D329 and R332, masking the positive charge of R332 and lowering the energy barrier of Mg2+ to pass through the R332 gate (Supplementary Fig. 9d). When the Mg2+ concentration is high in the mitochondrial matrix, Mg2+ is captured by the D329-ring, decoupling the interaction between D329 and R332, re-establishing a barrier by the positive R332-ring. However, the detailed mechanism of how the gating residues are regulated remains to be investigated. Elucidation of the MRS2 structure in an open state will help to address this question.

In summary, our study identifies N-terminal residue 71 as the first amino acid of human MRS2 after protein translocation into the inner mitochondrial membrane and cleavage of its MTP, and reports structures of human MRS2 in the presence and absence of Mg2+ at 2.8 Å and 3.3 Å, respectively. We identified R332 and M336 as gating residues and a for stability and channel activity important salt bridge between R116 and E291, which provides further insight into the Mg2+ permeation and regulatory mechanism in MRS2.

Methods

Protein expression and purification

Full-length human MRS2 (Uniprot Q9HD23) conjugated with a FLAG tag at the C-terminus was expressed in Expi293F cells (Thermo Fisher Scientific, A14527). Cells at a density of 2.0−3.0 × 106 cells/ml were transfected with the MRS2-encoding pCMV plasmid using polyethylenimine (PEI) following the protocol described previously47. Cells were harvested 48 h after transfection, and resuspended in Lysis buffer (20 mM HEPES, 150 mM NaCl, 10% [v:v] glycerol, 40 mM MgCl2, pH 7.3) supplemented with cOmplete, EDTA-free protease inhibitor cocktail (Roche). After cell disruption by sonication on ice, the membrane fraction was collected by ultracentrifugation at 100,000 × g for 1 h at 4 °C. The pelleted membranes were resuspended with Lysis buffer and pelleted again at 100,000 × g for 1 h. The membrane pellets were flash frozen in liquid nitrogen and stored at −80 °C until use. For solubilization, membranes at a protein concentration of ~4 mg/ml were solubilized with 1% (w:v) n-dodecyl-β-d-maltopyranoside (DDM) (Anatrace) and 0.1% cholesteryl hemisuccinate Tris salt (CHS) (Anatrace) for 2 h at 4 °C, followed by another round of ultracentrifugation to remove insoluble material. The supernatant was incubated with ANTI-FLAG M2 Affinity Gel (Millipore) on an end-over-end rotator for 2 h at 4 °C. The MRS2-bound beads were incubated in Lysis buffer containing 1% lauryl maltose-neopentyl glycol (LMNG) for 30 min at 4 °C, followed by extensive washing with Washing buffer (Lysis buffer supplemented with 0.01% LMNG). Proteins were eluted with 0.2 mg/ml 3× FLAG peptide. Eluted fractions were concentrated to ~3 mg/ml using a 50 kDa cut-off centrifugal filter and loaded onto a Superdex 200 Increase 3.2/300 size-exclusion column (Cytiva) with SEC buffer (20 mM HEPES, 150 mM NaCl, 40 mM MgCl2, 0.003% LMNG, pH 7.3). Peak fractions (~0.5 mg/ml) were collected for cryo-EM grid preparation. For the MRS2 sample in 1 mM EDTA, the sample was prepared in the same procedure except that the sample was dialyzed overnight to remove MgCl2, followed by size-exclusion chromatography with 1 mM EDTA SEC buffer (20 mM HEPES, 150 mM NaCl, 1 mM EDTA, 0.003% LMNG, pH 7.3). The protein fractions were resolved and analyzed using 4–12% Bis-Tris gel (Invitrogen) for SDS-PAGE and 4–16% Bis-Tris gel (Invitrogen) for BN-PAGE.

N-terminal sequencing

Purified MRS2 was resolved by SDS-PAGE, followed by transferring to methanol activated PVDF membrane in a XCell Mini-Cell Blot Module (Thermo Fisher) at 30 V for 2 h. The MRS2 band on PVDF was stained with Ponceau S, excised and shipped to Creative Proteomics for N-terminal sequencing via Edman degradation.

Negative staining electron microscopy

3 µL of MRS2 (0.02 mg/mL) was applied on a glow discharged carbon-coated 400 square mesh copper grid (EMS) and incubate for 1 min. The grid was then blotted by filter paper (Whatman) and washed once with 3 µL of Nano-W negative staining solution (Nanoprobes) followed by incubation with 3 µL of Nano-W for 1 min. The grid was blotted to remove excess staining solution and air-dried by waving. Images were recorded using an FEI Tecnai T20 TEM operated at 200 kV with a direct electron detector K2 Summit (Gatan Inc). Data was collected using SerialEM48 at a nominal magnification of 25,000x, a pixel size of 3.04 Å/px, and defocus range between −1.5 µm and −2.5 µm. A total of 459 images was collected. Image processing was performed using cisTEM v1.0.049. 215,448 particles were picked from the micrographs after CTF estimation, extracted with a box of 108 px (~328 Å) and analyzed by 2D classification.

Cryo-EM grid preparation and data collection

3 μl purified MRS2 at approximately 0.5 mg/ml was applied to a glow-discharged 400-mesh R 1.2/1.3 Cu grid (Quantifoil). The cryo grids were blotted for 6 s at 4 °C and 95%, and plunge-frozen into liquid ethane using a Leica EM GP2 (Leica) and stored in liquid nitrogen. The grids were screened on an FEI Tecnai T20 TEM before data collection.

Cryo-EM datasets were acquired with SerialEM48 using a Titan Krios (FEI, now ThermoFisher Scientific) operated at 300 kV and equipped with an energy filter and K3 camera (Gatan Inc.). Movies of 50 frames with a dose of 1 e−/Å2 per frame (50 e−/Å2 total dose) were recorded at a nominal magnification of 105,000x, corresponding to a physical pixel size of 0.83 Å/px (super-resolution pixel size 0.415 Å/px) in CDS mode at a dose rate of 10 e-/px/s and a defocus range of −0.7 to −2.0 µm. In total, 3991 and 9656 movies were collected for MRS2-Mg2+ and MRS2-EDTA, respectively (Supplementary Table 1).

Cryo-EM data processing

The overall workflow of image processing is illustrated in Supplementary Fig. 2. All processing was performed within cryoSPARC v3.3.250. Movies were processed with patch motion correction and patch CTF estimation. Good 2D class averages generated from ~1000 manually picked particles served as templates for automatic particle picking.

For MRS2-Mg2+, 1,568,444 particles were picked and extracted at 1x binned pixel size of 0.83 Å with a box size of 320 px (~266 Å). Particles were then subjected to 2D classification to remove junk particles. 573,010 good particles selected from one round of 2D classification was subjected to ab-initio reconstruction (K = 2) and heterogenous refinement (K = 2) with C1 symmetry to further sort particles. Non-uniform refinement was performed with 450,554 selected particles, followed by local motion correction and CTF refinement to correct for beam-tilt, spherical aberrations, and per-particle defocus parameters. Non-uniform refinement with polished particles resulted in maps at 2.8 Å (with C5 symmetry applied) and 3.1 Å (with C1 symmetry applied), according to gold-standard FSC = 0.143 criterion (Supplementary Fig. 2).

For MRS2-EDTA, 4,802,706 particles were picked and extracted at 4x binned pixel size of 3.32 Å and sorted by one round of 2D classification, ending up with 1,755,556 particles. Ab-initio reconstruction (K = 3) was performed with the subset of particles after 2D classification, followed by heterogenous refinement (K = 3) using all particles with C1 symmetry was applied to remove junk particles. Particles from one class (2,237,316 particles) were re-extracted at 1x binned pixel size of 0.83 Å and subjected to non-uniform refinement followed by another round of heterogenous refinement (K = 2) to further sort particles. Final non-uniform refinement by using 1,744,117 particles yielded maps at 3.3 Å (with C5 symmetry applied) and 3.6 Å (with C1 symmetry applied) according to gold-standard FSC = 0.143 criterion.

The summary of data collection and imaging processing parameters are shown in Supplementary Tables 1 and 2. The cryo-EM maps of MRS2-Mg2+ (C1 and C5) and MRS2-EDTA (C1 and C5) have been made available and have the following accession codes: EMD−41624, EMD-41629, EMD-41628 and EMD-41630, respectively.

Model building and refinement

For the atomic model of MRS2-Mg2+, predicted MRS2 AlphaFold51 structure (AF-Q9HD23-F1) was used as an initial model. The initial model was rigid-body fitted into the local resolution filtered map using UCSF Chimera v.1.1652. The model was then manually rebuilt in COOT v.0.9.753 using the local resolution filtered map, which was generated from refinement with C5 applied. The Mg2+ ions assigned in the pore regions were confirmed in the C1 map. Loop regions (residues 174–181, 273–287) were built with the unsharpened map. Iterative rounds of manual refinement in COOT and real-space refinement in Phenix v.1.20.1-448754 were performed. For the MRS2-EDTA structures, models were generated from multiple rounds of manual refinement and real-space refinements with the final model of MRS2-Mg2+ as initial model. The quality of the model and fit to the density was assessed using MolProbity55 and Phenix54. All structural figures were prepared using UCSF Chimera v.1.1652 or UCSF ChimeraX v.1.456. PDBs have been made available together with the EM maps with the following PDB IDs: 8TUL (MRS2-Mg2+) and 8TUP (MRS2-EDTA).

Subcellular localization of MRS2

Expi293F cells at a density of 2.0–3.0 × 106 cells/ml were transfected with the MRS2-GFP or MRS2(71-443)-GFP encoding pCMV plasmid using polyethylenimine (PEI) following the protocol described previously47. 24 h after transfection, the cells were stained with 100 nM MitoTracker Red (Thermo Fisher Scientific) or 1 µM ER-Tracker Red (Thermo Fisher Scientific) for 30 min at 37°C. The cells were washed twice with PBS and imaged using a Zeiss LSM 880 confocal laser scanning microscope.

Mg2+-dependent E. coli growth assay

The Mg2+-auxotrophic E. coli strain (BW25113 ΔmgtA, ΔcorA, ΔyhiD DE3), which has been used previously41,42, was transformed with His-TmCorA, MRS2(71-443)-FLAG, its mutants or without insert in a pET28a backbone. The transformants were inoculated and grown in kanamycin (50 µg/mL) containing LB medium (LBK) supplemented with 100 mM MgSO4, shaking at 250 rpm and 37 °C overnight. The overnight cultures were used to inoculate fresh LBK in a 1:100 ratio and grown until they reached an OD600 of 0.6. The bacterial cultures were then diluted 10-fold serially with LBK medium, spotted onto LBK agar plates containing 0.1 mM IPTG and 0 or 100 mM MgSO4, and incubated at 30 °C for 2 days. Plates were imaged using ChemiDoc MP Imaging System (Bio-rad).

Ni2+-sensitivity assay

BL21(DE3) (Sigma, CMC0014) was transformed with His-TmCorA, MRS2(71-443)-FLAG, its mutants or without insert in a pET28a backbone. The transformants were inoculated and grown in LBK medium, shaking at 250 rpm and 37 °C overnight. The overnight cultures were used to inoculate fresh LBK in a 1:100 ratio and grown until they reached an OD600 of 0.6. The bacterial cultures were then diluted 10-fold serially with LBK medium, spotted onto LBK agar plates containing 0.1 mM IPTG, and 0 or 1.8 mM NiCl2, and incubated at 30 °C overnight. Plates were imaged using ChemiDoc MP Imaging System (Bio-rad).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.