Cryo-EM Structures of the Magnesium Channel CorA Reveal Symmetry Break Upon Gating

Summary

CorA, the major Mg2+ uptake system in prokaryotes, is gated by intracellular Mg2+ (KD ~1–2 mM). X-ray crystallographic studies of CorA show similar conformations under Mg2+-bound and Mg2+-free conditions, but EPR spectroscopic studies reveal large Mg2+-driven quaternary conformational changes. Here, we determined cryo-EM structures of CorA in the Mg2+-bound “closed” conformation and in two “open” Mg2+-free states at resolutions of 3.8 A, 7.1 A and 7.1 A, respectively. In the absence of bound Mg2+, four of the five subunits are displaced to variable extents (~10 to ~25 A) by hinge-like motions at the stalk helix as large as ~35°. The transition between a single 5-fold symmetric closed state and an ensemble of low Mg2+, open, asymmetric conformational states, is thus the key structural signature of CorA gating. This mechanism is likely to apply to other structurally similar divalent ion channels.

Introduction

Magnesium ions (Mg²⁺) are essential in a variety of biochemical and structural processes among all living cells (Romani, 2011). In prokaryotes, the ~200 kDa Mg²⁺-dependent channel CorA is the key pathway for electrophoretic Mg²⁺ uptake (Dalmas et al., 2014b; Hmiel et al., 1986; Maguire, 2006). CorA can functionally complement the eukaryotic mitochondrial Mrs2 channel (Kolisek et al., 2003), which is essential for normal mitochondrial activity (Weghuber et al., 2006), and whose abnormal expression has been associated with diseases ranging from cancer to demyelination of neuronal tissue (Kuramoto et al., 2011; Leanza et al., 2013). Recent experimental evidence suggests that Mg²⁺ homeostasis through CorA takes place via a negative feedback loop, where Mg²⁺ binding at the inter-subunit interfaces leads to channel closure, whereas low Mg²⁺ concentrations tend to promote the transition to the open, conductive state (Dalmas et al., 2014b; Pfoh et al., 2012). Therefore, in CorA, Mg²⁺ acts as both the gating ligand and the selected charge carrier against other extracellular cations (Dalmas et al., 2014a).

Ligand-driven activation of multimeric ion channels is often a consequence of a shift in the interaction energies between subunits, typically arranged as symmetric (or pseudo-symmetric) oligomers (Auerbach, 2013). In most ligand-gated ion channels, ligand binding provides the energy required for a series of structural rearrangements that lead to the formation of a water-filled permeation pathway. In contrast, Mg²⁺-bound CorA forms a stable five-fold symmetric closed conformation, where five equivalent subunits associate to generate a long (~50 Å), narrow, presumably non-conductive pathway along the channel central axis (Dalmas et al., 2014a; Neale et al., 2015; Svidova et al., 2011). This finding suggests that the energetics and mechanism of CorA activation/deactivation differ from those of other ligand-gated channels.

Surprisingly, existing CorA crystal structures obtained in the presence (Eshaghi et al., 2006; Guskov et al., 2012; Lunin et al., 2006; Payandeh and Pai, 2006) and absence (Pfoh et al., 2012) of Mg²⁺ do not reveal significant changes in the channel conformation that could explain its gating mechanism (Fig. S1). However, site-directed spin labeling and electron paramagnetic resonance (EPR) spectroscopy studies of lipid-reconstituted CorA have demonstrated large changes in the quaternary structure of the channel associated with magnesium binding/unbinding (Dalmas et al., 2010; Dalmas et al., 2014b), suggesting that Mg²⁺ permeation may involve conformational rearrangements not captured in the crystal structures. The experiments we present here address these conflicting observations by determining the 3D structures of CorA in its Mg²⁺-bound closed and Mg²⁺-free open conformations in the absence of the constraints of a crystal lattice.

Results

We used single-particle cryo-EM (Bartesaghi et al., 2015; Cheng, 2015; Meyerson et al., 2014; Nogales and Scheres, 2015) to determine the structures of detergent-solubilized CorA from Thermotoga maritima (TmCorA) (Fig. S2) under conditions that stabilize its closed (Mg²⁺-bound) and open (Mg²⁺-free) states. Cryo-EM images obtained from plunge-frozen specimens of n-Dodecyl β-D-maltoside (DDM)-solubilized CorA in the presence of 40 mM Mg²⁺ (Fig. 1A) revealed clear density for the cytoplasmic domains, the transmembrane (TM) region of the homo-pentamer, and density for the lipid/detergent micelle surrounding the protein. We determined the structure of the five-fold symmetric, Mg²⁺-bound state at an overall resolution of ~3.8 Å (Figs. 1, 2, S3, Table S1 and Movie S1). Some areas of the periphery of the channel (Fig. 1D) are at lower resolution than the average, whereas other regions of the pentamer show higher resolution (~3.6 Å). A similar structure was obtained, albeit at lower resolution (~4.4 Å) in the absence of imposed five-fold symmetry. The resolution of the map was high enough to enable unambiguous tracing of the complete polypeptide chain (subunits A, B, C, D and E), guided by eight clear α-helical regions, seven well-separated β-sheets and strong densities for bulky side chains, such as Arg, Lys, His, Trp, Tyr and Phe (Figs. 1, 2 and S3). Furthermore, we were able to detect density for coordinated Mg²⁺ at the subunit interfaces (Fig. 2B), as well as strong peaks for putative Mg²⁺ along the permeation path near the GMN selectivity filter [Asn314 at the periplasmic loop (Dalmas et al., 2014a; Palombo et al., 2013; Payandeh and Pai, 2006)], and at the other end of the pathway near Asp277 (Fig. 2C). These putative permeating Mg²⁺ correlate well with the equivalent sites observed in the available crystal structures of Mg²⁺-bound CorA (Eshaghi et al., 2006; Nordin et al., 2013; Pfoh et al., 2012).

Fig. 1.

Cryo-EM of TmCorA in the presence of Mg²⁺. (A) Representative cryo-EM image at −1.6 μm defocus with CorA particles visible in different orientations. (B) Selected 2D class averages used to produce an initial model. (C) “Gold-standard” FSC plot for estimating resolution. (D) Final map viewed from the side colored according to local resolution calculated using the program RESMAP (Kucukelbir et al., 2014). Slices through the map at indicated positions are shown on the right.

Fig. 2.

Cryo-EM structure of T. maritima Mg²⁺ channel CorA in the presence of Mg²⁺ at 3.8 Å resolution. (A) Side view of density map and fitted coordinates in ribbon representation of the entire pentamer with each protomer shown in a different color: A in orange, B in magenta, C in blue, D in green and E in gold. A close-up view of one of the peripheral helices indicates that densities for side chains are clearly visible, with the exception of most negatively charged residues, as previously reported in other cryo-EM structures (Allegretti et al., 2014; Bartesaghi et al., 2014; Fromm et al., 2015). Further details are shown in Figure S3 and Movie S1. (B) Section through the pentamer at the level of the bound Mg²⁺ between protomers near Asp253^A/Asp89^B and Asp179^A/Pro14^B as viewed from the cytoplasm and indicated in (A). A close-up view of the region between protomers A and B is shown as an inset. (C) A slice along the five-fold symmetry-axis is shown to indicate the pore region of the channel with the narrowest region near Met291. Two of the strongest density peaks along the pore are highlighted (insets), one near Asn314 near the periplasmic loops, and another near Asp277 in the stalk helix at the end of the ion permeation pathway.

The present cryo-EM model of CorA in the closed state is similar to that of TmCorA determined by X-ray crystallography (PDB ID 4I0U), with minor differences in some of the TM segments, the C-terminal domain, and the periplasmic loops (Fig. S4). These regions are dynamic, as suggested from site-directed spin labeling experiments (Dalmas et al., 2010), and represent segments in the crystal structures with large crystallographic B-factors (Eshaghi et al., 2006; Lunin et al., 2006; Payandeh and Pai, 2006) (Fig. S4A). Regions in the X-ray structures that show some of the largest differences with the cryo-EM structure are in the periplasmic loop regions, spanning residues that are involved in crystal contacts (Fig. S4E and F).

Structural analysis of DDM-solubilized TmCorA in 1 mM EDTA (when Mg²⁺ is expected to be depleted) showed dramatic differences in the structure of the channel. Careful 3D classification revealed at least two distinct conformations (states I and II) in which the five-fold symmetry observed in the closed state is lost as a result of movements of the cytoplasmic domains (Figs. 3–5, S6 and Movies S2–S3). Iterative classification steps without applying symmetry were critical to avoid the pitfall of an incorrect structure for the open state; this is shown in more detail in Figure S5, which demonstrates that the appearance of extra densities that could have been wrongly interpreted as additional segments are in fact the result of incomplete separation of two closely related, yet distinct conformations. Cryo-EM maps of the two open states, each at an average resolution of ~7.1 Å, show an outward translation of subunit A of ~23 Å (measured at Pro249) and an inward translation of subunit B of ~21 Å (Fig. 4). These changes appear to reflect mostly rigid-body movements of the cytoplasmic domains, with smaller reorientations of subunits C and D (~11–18 Å in state II) and minimal rearrangements in subunit E. Structural changes in the TM segments are much more subtle, and surprisingly (in spite of the symmetry break observed in the cytoplasmic domains), this region retains an approximate five-fold symmetry (Figs. 4–5). Although clearly related, there are significant differences between the two resolved Mg²⁺-free conformations, with state II displaying greater movements in subunits C and D than state I (Figs. 3, 5 and S5), relative to the closed state.

Fig. 3.

Cryo-EM structures of T. maritima Mg²⁺ channel CorA in the absence of Mg²⁺ at 7.1 Å resolution. (A) Top (as viewed from the cytoplasmic side) and side view of the structure of two open states, labeled state I and II using the same color scheme as in Figure 1. A simplified, geometric representation of these movements is shown as a pair of pentagons (filled line, symmetric closed state; dashed line, asymmetric open states). The large outward movement of the cytoplasmic domain of subunit A and an inward movement of the cytoplasmic domain of subunit B can be seen in both states. The cytoplasmic domain of subunit C in state II shows a slight outward movement, whereas the cytoplasmic domain of subunit D shows a slight inward movement compared to state I (also see Movies S2–S3).

Fig. 5.

Local resolution of the two cryo-EM maps of TmCorA in the absence of Mg²⁺ viewed from the side (top) and from the cytoplasm (bottom) colored according to RESMAP (Kucukelbir et al., 2014). Slices through the map at indicated positions are shown on the bottom right. The lowest resolution is observed in the putative transmembrane region of the protein.

Fig. 4.

Comparison of the closed state with the open state II of TmCorA. Density map of the Mg²⁺-bound closed state (Fig. 1) filtered to 7 Å resolution in gray superimposed with the of asymmetric open state II shown in top (upper) and side views (lower). Sections through the superposed maps at the levels corresponding to Leu280, Leu294 and Gly309 illustrate the differences and similarities in tertiary and quaternary structures between the two states.

Is the asymmetric open CorA conformation physiologically relevant? There are examples of unusual membrane protein conformations resulting from detergent effects or other underlying causes (Jiang et al., 2003; Liu et al., 2009). Even though the cytoplasmic domain is not likely to be influenced by the presence of detergent, it is conceivable that the changes observed would not be seen if the channel was embedded in the environment of a lipid bilayer. We therefore carried out a series of measurements of lipid/nanodisc-reconstituted CorA (Fig. S6) to test whether equivalent structural changes can be observed upon removal of Mg²⁺. As shown in Figure 6, the distinctive asymmetric features detected in the class averages of detergent-solubilized Mg²⁺-free CorA are recapitulated in nanodisc-reconstituted CorA. Figure 6A shows selected side views of CorA in DDM micelles (left panels) and 1D1 nanodiscs (right panels). Open state asymmetry is evident from the different side views of 2D class averages. In DDM micelles, the Mg²⁺-bound state shows a single peak distribution for the diameter of the cytoplasmic domain (d0, gray bar histogram), while its asymmetric counterpart in 0 nominal Mg²⁺ displays a bimodal distribution (blue bar histogram), resulting form “narrow” (d1) and “wide” (d2) side views at the soluble domains (see density maps on the right). Large size variations in the dimensions of the nanodiscs and the various positions of the protein complex in larger nanodiscs (Fig. S6) presently pose challenges for the determination of a high resolution 3D structure from nanodisc-reconstituted CorA, but the side views of the 2D class averages, and the low resolution 3D maps establish that detergent-solubilized and nanodisc-reconstituted CorA undergo a similar symmetry break upon removal of Mg²⁺.

Fig. 6.

Validation of asymmetric CorA structure. (A) Mg²⁺-dependent asymmetric transitions are maintained in nanodisc-reconstituted CorA. Side views of selected 2D class averages of CorA in the presence (top rows) and absence of Mg²⁺ (bottom rows) in detergent (left) and in lipid nanodiscs (right). In the presence of Mg²⁺, the main features of the cytoplasmic domain show an average diameter of ~87 Å (d0) whereas two side view classes can be observed in the absence of Mg²⁺, with either two or three distinct density features (narrow, d1 and wide, d2). Detergent-solubilized and nanodisc-reconstituted particles show similar distance distribution profiles. (B) Biochemical demonstration of Mg²⁺-dependent asymmetry by crosslinking studies using the double cysteine mutant V183C/A8C in lipid bilayers (left). In the presence of Mg²⁺ and Copper Phenanthroline (CuPhe), oligomer symmetry is evident as ladder (with five steps) of cross-linked CorA monomers as all subunits are involved in cysteine bridge formation. In the absence of Mg²⁺, subunit asymmetry is reflected as a reduction in the number of bands in the ladder. PAGE gel analysis shows that this process is Mg²⁺ dependent, since low [Mg²⁺] results in a change in intensity of each band in the ladder, especially that from tetramers and pentamers (top right). Densitometric quantification of band intensity (bottom right) shows that as [Mg²⁺] increases, the intensities of the cross-linked bands increases while that from the monomer band decreases.

To further validate the hypothesis that Mg²⁺ removal triggers a transition to this asymmetric conformation under physiological conditions, we carried out crosslinking experiments on lipid-reconstituted CorA using variants carrying cysteine substitutions. The experiment is based on the assumption that under conditions where Mg²⁺ is bound and CorA displays five-fold symmetry, double cysteine mutants designed to crosslink at the subunit interface should show a ladder of five bands under Polyacrylamide gel electrophoresis (PAGE). Any break in the symmetric arrangement of the subunits would be expected to lead to a reduction in the number of bands of the crosslinking ladder. Figure 6B shows that the double mutant 183C-8C located at the interface between the cytoplasmic domains indeed displays a five-band ladder only in the presence of Mg²⁺ and the catalyst Cu²⁺ phenanthroline. However, in the absence of Mg²⁺ (1 mM EDTA), the top band (pentamer) disappears and the intensity of the following band (tetramer) is significantly reduced. Furthermore, this change in the crosslinking pattern is dependent on the Mg²⁺ concentration (Fig. 6B, right panels). Evaluation of the relative intensity of the bands by densitometry suggests an apparent Mg²⁺ K_D of ~0.5 mM.

An additional test of the functional relevance of the asymmetric Mg²⁺-free structures revealed by cryo-EM was based on double electron-electron resonance (DEER) distance determinations obtained for spin-labeled CorA in the open and closed states. Spin labels attached at the intracellular end of the stalk helix (residues 247, 250 and 252) were shown previously to move closer to each other (between 4 and 8 Å), for both next-neighbor and diagonally related distances, as measured by DEER (Dalmas et al., 2014b). Using a model traced on the cryo-EM map obtained with no Mg²⁺, we used the molecular dynamics simulations with dummy spin-labels (MDDS) method (Islam et al., 2013) to calculate the theoretical distance distributions between spin-labels corresponding to that model. The MDDS-calculated distance distributions were internally consistent with the DEER experiments. The overall reduction in inter-subunit distances seen experimentally is a consequence of subunit A moving away (beyond DEER detection range) and subunit B moving towards the permeation path (and subunits C, D and E). Taken together, these biophysical and biochemical experiments provide strong evidence to support our claim that the asymmetric Mg²⁺-dependent gating of CorA observed by cryo-EM is physiologically relevant.

Discussion

The structures we report for CorA here suggest an unexpected mechanism for ligand-driven gating. We propose that inter-subunit forces mediated by Mg²⁺ stabilize a five-fold symmetric closed state, but that a reduction in intracellular Mg²⁺ leads to interfacial Mg²⁺ release and subsequent large range cytoplasmic domain rearrangements. These conformational changes would then drive gating transitions along the transmembrane segments. Our findings thus potentially integrate the functional mechanisms of gating in other ligand-gated channels such as Cys-loop receptors (Du et al., 2015; Miyazawa et al., 2003; Unwin and Fujiyoshi, 2012; Wilson and Karlin, 2001) and P2X receptor family (Baconguis et al., 2013; Hattori and Gouaux, 2012; Heymann et al., 2013), where interfacial ligand binding energies have been surmised to play an important role in driving gating transitions. The quaternary structural changes in CorA appear to be an interesting variant of what is observed in the vast majority of ligand-gated channels, where ligand binding energies are typically used to transition from an unliganded (apo) closed state, to the ligand-bound activated (conducting) open state (Auerbach, 2013; Hille, 2001; Unwin, 2013). In CorA, the changes are in the opposite direction, and it is likely that the reduction in inter-subunit stabilization energies due to the release of Mg²⁺ leads to an increase in overall dynamics and conformational flexibility, thus leading to opening of the channel. It is important to emphasize that the gating model we present does not imply the existence of a fixed set of open states. Rather, we suggest that the open state of CorA consists of an ensemble of multiple asymmetric conformations that interconvert to generate transiently conductive states (Fig. 7B).

Fig. 7.

Proposed schematic model for conformational changes. (A) Reduced model of CorA in ribbon representation displaying only the C-terminal part of the complex with the two transmembrane helices and the stalk helix (residues 249–351), viewed from the cytoplasm (left) and from the side, parallel to the membrane (right). The closed Mg²⁺-bound state is displayed in gray and the asymmetric open Mg²⁺-free state II (color coding as in Figure 1). Superposition of the stalk helix and TM segments from the closed and open states reveals that relative to subunit E (showing the least movement in the pentamer) the stalk helix of subunit A can bend as much as ~35° away from the permeation pathway, while subunit B moves ~24° in the opposite direction. (B) Proposed schematic model for the asymmetric opening of pentameric magnesium channel CorA. When Mg²⁺ is bound, it stabilizes neighboring subunits in the closed state (gray). When Mg²⁺ concentration drops, Mg²⁺ ions between subunits are progressively and randomly removed leading to an ensemble of open states in a step-wise manner. In this transition, all five protomers are equivalent initially, but once the first inter-protomer interface is disrupted, it is likely that the movements of the rest of the protomers are in a predictable sequence (see Movie S4) either counterclockwise as indicated here or clockwise.

Recently Cleverley et al (Cleverley et al., 2015) studied Mg²⁺-free CorA from Methanocaldococcus jannaschii using cryo-EM methods and deposited a density map at a reported resolution of ~22 Å, noting that the cytoplasmic and transmembrane domain was arranged asymmetrically under these conditions. Because the density map reported by Cleverley and co-workers is at low resolution and does not enable definition of structural features such as the transmembrane domain or the outlines of the individual protomers, the map bears little resemblance in detail to the structure of the open conformation we report here (Fig. S7). Nevertheless, the loss of symmetry suggested in these maps is broadly consistent with the results we report here. As noted earlier, the magnitude of asymmetry observed in crystal structures of CorA in the Mg²⁺-free form is minimal, but further evidence for the potentially important role played by divalent Mg²⁺ in stabilizing the symmetric pentamer comes from studies of a mutant form of CorA (TmCorA-ΔNcc) crystallized in the presence of only monovalent ions (Pfoh et al., 2012). While the resulting structure shows a conformation broadly similar to the closed divalent-bound structures (Fig. S1), there are clear asymmetries found in regards to Cs⁺ binding along the permeation path (near the GMN selectivity filter). Slight asymmetry is also found in the conformation of the cytoplasmic domains. While instances of asymmetric conformational changes are anticipated for channels such as the eukaryotic Na⁺ and Ca²⁺ channels that are composed of non-identical subunits (Catterall and Zheng, 2015), it is somewhat unexpected in homo-oligomeric channels such as CorA.

Our results imply that the stabilization energy of Mg²⁺ binding at the inter-subunit interfaces of the cytoplasmic domains probably drives the key transition in and out of symmetry (between non-conductive and conductive conformations). The major intra-subunit conformational change originates from a hinge-bending movement in the “stalk” helix, at the transition between cytoplasmic and membrane domains (residues 280–291). This movement can be as much as 35° (Fig. 7A; subunit A in our nomenclature). Thus, we propose that Mg²⁺ release driven by a decrease in the concentration of intracellular Mg²⁺ results in destabilization of one of the five equivalent subunit interfaces (in orange, Fig. 7B; Movie S4). Destabilization of this interface nucleates a local increase in the conformational dynamics of one subunit, which, in turn, sequentially destabilizes its complementary interface in one the neighboring subunits (Fig. 7B), concomitant with release of additional Mg²⁺ and allowing for large domain excursions away from the symmetry axis. This would, as a consequence, randomly trigger the inward movement of subunit B (or, alternatively, subunit E). Our proposed mechanism is therefore incompatible with previous proposals that have assumed concerted movements of the five subunits, leading to a symmetric open state (Chakrabarti et al., 2010; Dalmas et al., 2014b; Nordin et al., 2013).

The reasons why bound Mg²⁺ stabilize the symmetric pentamer but render it non-conductive are likely due to the conformation of the long and narrow permeation path, which is potentially incompatible with the occupancy of a hydrated (or partially hydrated) Mg²⁺ (Neale et al., 2015). Still undefined, however, is the mechanism by which the large reorientations of the cytoplasmic domains are allosterically propagated to the TM domains so as to generate a conductive pathway for Mg²⁺. Propagation of a conformational “wave” from the cytoplasmic domain to the TM segments can be anticipated from the wide range of the hinge-bending motion observed in the stalk helix (Fig. 7A). Superposition of the stalk helix and TM segments from the closed and open states reveals that relative to subunit E (showing the least movement in the pentamer), the stalk helix of the “nucleating” subunit A can bend as much as ~35° away from the permeation pathway, while subunit B moves ~24° in the opposite direction. We hypothesize that the large conformational changes in the cytoplasmic domain are not the only effects of the loss of inter-subunit Mg²⁺. Instead, we propose that the loss of the interactions, stabilized by Mg²⁺, results in greater overall flexibility of the protein complex, including the hydrophobic gate region of the narrow pore. The higher flexibility likely increases the frequency of hydration events (Neale et al., 2015) that enable Mg²⁺ translocation. The dramatic difference between the structure of CorA in solution, as compared to the structure in a crystalline lattice, underscores the value of structure determination by cryo-EM methods, especially in situations where large-scale quaternary structural changes are potentially inhibited by crystal contacts or by crystallization of selected subpopulations present in a protein sample.

Experimental Procedures

Sample Preparation

CorA expression and purification

CorA from T. maritima was expressed and purified as previously described (Dalmas et al., 2010). Briefly, after elution from IMAC TALON (Clontech Laboratories Inc., Mountain View, CA) with 250 mM imidazole, 1 mM EDTA was added and the protein was concentrated using AMICON 100 kDa cut-off membrane filters (EMD Millipore, Billerica, MA). The concentrated protein was then purified to homogeneity by gel filtration (Fig. S2) on a Superdex 200 10/300 GL column (GE Healthcare Bio-Sciences, Pittsburgh, PA) equilibrated in a Mg²⁺-containing buffer (buffer A: 50 mM HEPES, pH 7.3; 150 mM NaCl; 40 mM MgCl₂; 0.5 mM DDM) or a Mg²⁺-free buffer (buffer B: 50 mM HEPES, pH 7.3; 150 mM NaCl; 1 mM EDTA; 0.5 mM DDM) before preparation of EM grids.

Preparation of CorA in nanodiscs

Plasmid harboring the N-terminal His-tagged construct of the membrane scaffold protein 1D1 (pMSP1D1) for nanodisc preparation was obtained from (#20061, Addgene, Cambridge, MA). MSP1D1 was expressed and purified according to protocols described previously (Alvarez et al., 2010). Protein was cleaved with TEV protease to remove the His-tag in buffer containing 50 mM Tris, pH 8.0; 0.5 mM EDTA; 100 mM NaCl and 1 mM DTT; and residual uncleaved protein was captured on Superflow Ni-NTA Resin (Qiagen Inc., Valencia, CA), while the flow-through containing cleaved MSP1D1 was desalted on PD-10 columns (GE Healthcare Bio-Sciences) against buffer B without DDM. TmCorA was reconstituted into 1D1 nanodiscs following the protocols established by Sligar laboratory (Ritchie et al., 2009). In brief, purified His-tagged CorA, cleaved MSP1D1, and lipid-detergent micelles were mixed in a Mg²⁺-containing buffer (buffer C: 50 mM HEPES, pH 7.5; 200 mM NaCl; 40 mM MgCl₂) at 5:10:550 molar ratio. Lipid detergent micelles were prepared from POPC, POPG (Avanti Polar Lipids Inc., Alabaster, AL) and DDM at 8:2:35 molar ratio by solubilization of dried lipids with 200 mM DDM in water using sonicator with microtip (Branson Ultrasonics, Danbury, CT). Excess detergent in reconstitution mix was removed by overnight incubation at 4°C with Bio-Beads SM-2 Adsorbents (Bio-Rad Laboratories Inc., Philadelphia, PA). CorA-loaded nanodiscs were purified from empty nanodiscs using SuperFlow Resin followed by size exclusion chromatography (Superdex 200 10/300 GL; Fig. S6) in either buffer C or a Mg²⁺-free buffer (buffer D: 50 mM HEPES, pH 7.5; 200 mM NaCl; 1 mM EDTA). As a control, empty 1D1 nanodisc, without CorA, prepared using cleaved MSP1D1 and lipid-detergent micelles at 10:650 molar ratio, were separated by size exclusion chromatography against buffer C.

Electron microscopy

Sample preparation and Data Acquisition

Vitrified specimens were prepared by adding 3 μl of CorA (4.0 mg/ml and 3.7 mg/ml in the presence and absence of Mg²⁺, respectively) to plasma cleaned Cu R1.2/1.3 holey carbon grids (Quantifoil Micro Tools GmbH, Groβlobichau, Germany) or Cu R1.2/1.3 C-flat holey carbon grids (Protochips Inc., Morrisville, NC) for the nanodisc samples. Grids were blotted for 5–7 s after a 10 s pre-blotting time, then plunge-frozen in liquid ethane using a Leica EM GP instrument (Leica Microsystems Inc., Buffalo Grove, IL), with the chamber maintained at 4°C and 86% humidity. Similar conditions were used to prepare nanodisc specimens, except that we used a protein concentration of 1.45 mg/ml and 0.85 mg/ml for the Mg²⁺ and the Mg²⁺-free samples, respectively. The nanodisc data were collected on a second microscope equipped with a Falcon II detector (FEI Company, Hillsboro, OR) at a pixel size of 1.4 Å (Fig. S6C). Following vitrification, grids were post-mounted into autoloader cartridges and transferred to the microscope. Cryo-EM imaging was done on an FEI Titan Krios microscope (FEI Company) operated at 300 kV, aligned for parallel illumination. Projection images for the detergent solubilized samples were acquired with a K2 Summit camera operated in super-resolution counting mode with a physical pixel size of 1.275 Å or 1.352 Å and a super-resolution pixel size of 0.6375 Å or 0.676 Å (due to slightly different detector set-up during the acquisition of the data for the closed Mg²⁺-bound and the Mg²⁺-free open-states) with a defocus range between −1.0 and −2.5 μm (Figs. 1A and S5A). The detector was placed at the end of a GIF Quantum energy filter (Gatan Inc., Pleasanton, CA), operated in zero-energy-loss mode with a slit width of 20 eV. The dose rate used was ~4.5 e⁻/pixel·s (equivalent to ~2.8 or 2.5 e⁻/Ų·s at the specimen plane) to ensure operation in the linear range of the detector (Li et al., 2013). The total exposure time was 15.2 s, and intermediate frames were recorded every 0.4 s giving an accumulated dose of ~40 e⁻/Ų and a total of 38 frames per image.

Image Processing

Movie frame alignment, CTF determination and particle picking were done as described previously (Bartesaghi et al., 2014). For the closed state of CorA in the presence of Mg²⁺ (Table S1 and Fig. 1A), 70,596 particles were extracted from 945 micrographs using a binning factor of 2 (with respect to the super-resolution image size) and a box size of 256×256 pixels. 2D class averages (Fig. 1B) were generated with EMAN2 (Tang et al., 2007) and selected views used to generate a C5-symmetric initial model using EMAN2’s e2initialmodel.py. This low-resolution initial model was used as reference to run 3D classification in RELION 1.3 (Scheres, 2012). 3D classification, refinement and post-processing using RELION produced a final map at 3.8 Å resolution from 46,206 particles with C5 symmetry imposed (Figs. 1, 2, S3). For the open state of CorA in the absence of Mg²⁺ (Table S1 and Fig. S5A), 173,653 particles were extracted from 2,495 micrographs and subjected to iterative 3D classification and refinement in RELION using the same initial model derived above. Two different states at ~7.1 Å resolution each (Figs. S5C–E) were obtained from 26,271 and 27,416 particles, respectively (no symmetry imposed). Map post-processing and FSC plots were done in RELION 1.3 and local resolution was determined using RESMAP 1.1.4 (Kucukelbir et al., 2014).

Model fitting, refinement and validation

The map “*.mrc” was cropped to 120 ų, its origin changed to 0, renamed to “*.map” and then converted to “*.mtz” using PHENIX 1.9 (Adams et al., 2010) (phenix.map_to_structure_factor). For the closed state, one pentamer (PDB ID 4I0U) was fitted by rigid body fitting into the map using UCSF Chimera (Pettersen et al., 2004) and its position saved relative to the map. Coordinates were refined in COOT (Emsley et al., 2010) using the “*.mtz” file while using the “Calculate/Map Sharpening” option in COOT, which allows one to apply different B-factors to different areas of the map. The structure was first refined manually in COOT, then refined using PHENIX real space refinement. Geometry was minimized during multiple rounds in PHENIX and final outliers were corrected in COOT. Side chains were removed in areas of the map with insufficient densities for side chains. Molprobity (Chen et al., 2010) was used to check the geometry of the atomic model. For the open states, the closed state atomic model obtained above was fragmented into seven overlapping pieces per subunit (residues 6–182, 182–204, 204–247, 245–279, 278–311, 311–327, 327–351) and fitted by rigid body fitting in UCSF Chimera followed by COOT, then connected and fitted in COOT and used for further refinement and geometry minimization in PHENIX. The N-terminal 16–18 residues and C-terminal 2–4 residues were removed due to lack of density. For accuracy of fits, besides the to ~7.1 Å filtered final maps, also maps filtered to 4.5 Å were used (based on the by ResMap determined local resolution) which displayed density features along helical densities in local higher resolution regions for larger side chains such as tyrosines and separated β-sheets best seen in subunits B and E. Finally, all side chains were removed in the open state models due to the lack of sufficient density to assign their position. Figures and movies were produced using UCSF Chimera, PyMOL (The PyMOL Molecular Graphics System, Version 1.3. Schrödinger, LLC) or Cinema4D (MAXON).

Molecular Dynamics simulations

All molecular dynamic (MD) simulations of the spin-labeled CorA were carried out with the CHARMM (Brooks, 1983) package, using the all-atom CHARMM36 protein force field with the CMAP corrections and “dummy” nitroxide (ON) spin-label force field parameters (Islam et al., 2013), which were found to provide accurate representation of the dynamics of the original MTSSL spin-label. An open state model and closed (4I0U) structure were used to construct the geometries of the CorA systems for simulation. The dummy ON spin-labels were linked directly to the C_α atoms in residues 247, 250 and 252 of each of the subunits. Since the ON spin labels do not interact with each other, it is possible to introduce all ON dummy spin labels simultaneously into a single protein for a long molecular dynamics simulation. By doing so, it was also possible to keep the original wild type residues of CorA in the simulation. However, to avoid extensive clashes between the ON atoms and the relatively long side-chain residues, any side-chain atom starting from the C_γ is removed. Truncating long side chains will not cause any drastic change in the distance distribution since the dummy spin-label force field has a non-bond term that accounts for the influence of nearby side chains. All the simulations were performed in vacuum under NVT at 300 K using the Langevin thermostat with a collision frequency of 10.0 ps⁻¹. To begin, an Adopted Basis Newton-Raphson (ABNR) energy minimization (100 steps) and a short (10 ps) molecular dynamics simulation of the dummy ON spin labels were performed with a time step of 0.5 fs by fixing the coordinates of all other atoms of the protein to its initial starting structure. Finally, a 1 ns equilibration simulation and a 4 ns production MD simulation were performed by fixing the protein and using a time step of 1 fs which provided spin-pair distances between various subunits with time. All the distances between the subunits of a given residue are then combined and used to calculate the distance distributions of the spin-pair.

Crosslinking experiments

The present crosslinking experimental protocol was adapted from (Reyes et al., 2009). In brief, POPC and POPG (Avanti Polar Lipids Inc.) were mixed in a 3:1 (w/w) ratio and dried on a rotatory evaporator followed by an additional drying step under N₂ gas. The dried lipids were resuspended in Mg²⁺-free buffer (buffer E: 50 mM HEPES, pH 7.5 and 200 mM NaCl) using sonication to a final concentration of 10 mg/ml. Triton X-100 was added to the formed liposomes to a final concentration of 20% (Detergent:Liposome w/w) and rocked for an hour. In a 1:1000 protein to lipid concentration, CorA (in 0.5 mM DDM) along with 50 mM tris(2-carboxyethyl)phosphine (TCEP) and 50 mM EDTA were added to the liposomes and rocked for 30 minutes. The proteoliposomes were diluted in buffer E by 10-fold and rocked with Bio-Beads SM-2 Adsorbents (Bio-rad Laboratories Inc.) for 1 hour to overnight. The proteoliposomes were pelleted at 100,000 × g and resuspended in buffer E. This was repeated once more before the crosslinking experiments to remove residual TCEP and EDTA. If required, Mg²⁺ was added a few minutes prior to the crosslinking reaction. Crosslinking was triggered by adding 100 μM of a mixture of Cu²⁺ bis-1,10-phenanthroline in a 1:2 molar ratio and incubated for 10 minutes. The crosslinking reactions were quenched with 100 mM N-ethyl maleimide before running on SDS–PAGE (Fig. 6B).