The crystal structure of mouse VDAC1 at 2.3 Å resolution reveals mechanistic insights into metabolite gating

Structural Overview.

Optimized crystals of mVDAC1 belong to the monoclinic space group C2 with 1 molecule per asymmetric unit. The structure was solved by the single isomorphous replacement with anomalous scattering (SIRAS) technique by using an engineered cysteine to incorporate a single mercury atom, a method that was successfully used for LacY (16) and vSGLT (17). The model was refined from merged data to a resolution of 2.3 Å with R_free of 27.7% and R_work of 24.2% (Table S1). All 283 amino acids of mVDAC1 were built along with 47 water molecules. Thus, all structural features represented in this study, including the location of the N-terminal segment, were assigned on the basis of high-resolution data. Further details regarding the crystallization and structure determination can be found in Methods. Experimental and refined electron density maps are shown in Fig. S2.

Predictions regarding the number of strands contained in the VDAC β-barrel varied widely with 2 values in particular, 13 (18) and 16 (19, 20), emerging as the predicted number of β-strands. The structure of the mVDAC1 protomer forms a β-barrel composed of 19 β-strands with a shear number of 23 (Figs. 1 and 2). All of the strands exhibit an antiparallel pattern with the exception of strands 1 and 19 that associate in a parallel manner to close the barrel (Fig. 2A). The maximal height and width of the β-barrel is 35 and 40 Å, respectively. This report illustrates the high-resolution crystal structure of a eukaryotic β-barrel membrane protein with an odd number of strands (all known prokaryotic β-barrels have an even number of strands), confirming the architecture observed in the recent NMR study on hVDAC1 (21).

The correct orientation of the protein within the outer membrane of the mitochondria cannot be ascertained from the structure presented herein, because there have been multiple biochemical studies producing conflicting results on the endogenous position of the N and C termini (22–24) (i.e., whether they are facing the cytosol or intermembrane space). The hydrophilic N terminus of the protein forms a distorted α-helix that is nestled within the pore and tethered to the first β-strand. Both the N and C termini reside on the same side of the membrane and clearly demonstrate that the N terminus is not embedded in the membrane as has been suggested (18).

The orifice of the pore resembles a slightly elliptical cylinder with a maximal inner dimension of 27 × 24 Å (Fig. 2C). The pore forms a large pathway that transcends the entire length of the protein, likely representing the open conformation of the channel. These dimensions are in good agreement with those reported by electron microcopy (25), atomic force microscopy (26), and the recently published NMR structure of hVDAC1 (21). Structural alignments between the crystal structure of mVDAC1 and the 20 deposited NMR structures of hVDAC1 (Protein Data Bank ID code 2K4T) showed an rmsd that varied from 1.72 to 4.68 Å (the average rmsd among all 20 NMR structures is 4.79 Å) (Fig. S3). In the NMR study, only 5 residues were assigned in the voltage-sensing N-terminal segment (residues 6–10). These studies were performed on hVDAC1 solubilized in detergent micelles, which may be responsible for the high degree of mobility. The structure presented here at 2.3 Å is that of mVDAC1 crystallized in lipidic bicelles. This high-resolution image reveals structural insights into the conduction pathway and moreover the precise localization and structural features of the voltage-sensing N-terminal segment. The assignment of the N-terminal segment is essential, because this helix plays a crucial role in gating the pore.

Characteristics of the Pore.

The exterior of the β-barrel primarily consists of hydrophobic residues that are exposed to the lipid environment, whereas the interior is extensively hydrophilic, which is conducive for the efficient passage of ions and large metabolites (Fig. 3). Analysis of the interior walls of the pore reveals 15 positively-charged residues and 11 negatively-charged residues (Fig. 3). These charged residues are not uniformly distributed along the β-strands of the barrel, but are clustered primarily along strands 1–8. In contrast, there are fewer charged residues present on β-strands 9–19 (Fig. 3 B and D). The pore spanning N-terminal α-helical segment contains 3 positive (Lys-12, Lys-20, and Arg-15) and 2 negative (Asp-9 and Asp-16) charges facing the interior of the pore. Electrostatic calculations (Fig. 3 C and D) show that the interior of the pore has a higher density of positive versus negative charges, a fact that could account for the channel's preference for transporting anions over cations in the open conformation (2). The α-helix of the N-terminal segment is located approximately halfway down the pore, resulting in a partial narrowing of the cavity to 27 × 14 Å (Fig. 2C), yet there is still ample space for the passage of metabolites. Thus the structure presented here likely represents the open conformation. This charged α-helix is ideally positioned to regulate the conductance of ions and metabolites passing through the VDAC pore.

Arrangement and Coordination of the N-Terminal Helix.

Numerous studies have focused on the importance of the N-terminal α-helical segment's role in channel function. There has been a wide range of predictions as to the functional disposition of this domain, ranging from forming a segment of the channel wall (22, 27) to acting as the voltage sensor (2, 11). In 2007, De Pinto et al. (28) used NMR to resolve a peptide fragment of the N terminus (Prn2–20) corresponding to amino acids 2–20 of hVDAC1. Their findings show a complex behavior of the N-terminal domain where the peptide is unstructured in aqueous solvent and forms a well-ordered α-helix from residues 5–16 in SDS. In the recently reported NMR structure of hVDAC1, an accurate analysis of the structure of the N-terminal segment and its localization with respect to the rest of the protein was not possible because of the incomplete assignment of this region (residues 1–5 and 11–20 unassigned) (21).

However, our high-resolution structure presented herein shows the N-terminal segment (amino acids 1–26) forming a hydrogen-bonding pattern that facilitates its orientation against the interior wall of the pore (Fig. 4A) adjacent to β-strands 8–19. The first 5 N-terminal amino acids extend down the pore where 2 hydrogen bonds are observed between the carbonyl oxygens of Ala-2 and Pro-4 and the main-chain nitrogen of His-122 and ND2 of Asn-124, which are located on the wall of the pore. These interactions stabilize the N terminus to β-strand 8. Continuing in a C-terminal direction down the protein, a distorted helix is formed from amino acids 6–20 where the hydrogen-bonding pattern is broken at Leu-10 and Gly-11, separating the helix into 2 segments. Although the stable helical hydrogen-bonding pattern is disrupted, these 2 segments are capable of maintaining a rigid structure because Arg-15 forms bridging bidentate hydrogen bonds with the carbonyl oxygens from Ala-8 and Leu-10 (Fig. 4B). The helical portion of the N-terminal segment has 2 hydrogen bonds to β-strands 12 and 16, stabilizing its interaction with the wall of the pore. Connecting the soluble N-terminal domain to the first β-strand of the barrel is a highly conserved sequence (among mammals) of Gly-21–Tyr-22–Gly-23–Phe-24–Gly-25 (Fig. 1). This portion of the N-terminal segment is hydrogen-bonded to β-strands 16 and 18.

Physiological Modulators of VDAC.

As predicted, VDAC has a large hydrophilic pore capable of facilitating the passage of ions and large metabolites such as ATP. There have been a number of reports suggesting the channel may have specific nucleotide binding sites (2, 30). To address this issue, we attempted to identify an ATP binding site in the structure of mVDAC1. We resolved the structures of mVDAC1 in the presence of 5, 10, 50, and 100 mM ATP, but the corresponding structures revealed no additional electron density that would correspond to ATP. Similarly, we have crystallized mVDAC1 in the presence of other known nucleotide modulators (ADP and AMP-PNP) and observed no changes in the mVDAC1 structure, as determined by inspection of F_{o(nucleotide)} − F_o(native) maps. We therefore conclude that, at least in this conformation, there is likely no stable or specific binding site for these nucleotides.

Previous studies have shown that Ca²⁺ binds and regulates VDAC (31, 32), a result with clear implications for the role of the protein in mitochondrial-dependent cell death. Under initial crystallization conditions (where trace Ca²⁺ may be present), we were unable to identify electron density or a coordination pattern that resembles Ca²⁺ ligation. To further address this issue we solved the structure of mVDAC1 in the presence of 25 mM EGTA (a calcium-chelating agent that should bind all Ca²⁺ ions in solution and from the protein) and separately in the presence of 25 mM CaCl₂. Crystals formed in the presence of EGTA, and separate successful crystallization trials in the presence of AlCl₃ or InCl₃ were indistinguishable from the initial conditions with trace Ca²⁺. When Ca²⁺ is present in abundance, we observe electron density near Glu-73 that is indicative of metal ion coordination between 2 antiparallel mVDAC1 monomers (Fig. S4). De Pinto et al. (33) had shown that Glu-73 is buried in a hydrophobic environment, an observation that is explained by the current structure, which demonstrates its localization within the membrane. Furthermore, mutagenesis and functional analyses have implicated this residue in Ca²⁺ binding (32) and hexokinase-mediated protection against mitochondrial-dependent cell death (34). Based on the data presented here, Ca²⁺ must be present at excessive concentrations for stable binding to the protein. The only way this could happen endogenously would be with enormous levels of Ca²⁺ present and/or other gross disruption of the lipid interface with the outside of the pore, events that may occur during mitochondrial-dependent cell death.

Interestingly, crystallization in the presence of Mg²⁺–ATP results in a very similar localization pattern of the Mg²⁺ as was observed for Ca²⁺ at position Glu-73 of the crystal antiparallel dimer (Fig. S4). As was the case in the presence of ATP alone, this crystallization scenario does not reveal stable or specific binding site for the nucleotide. As mentioned above, Glu-73 is essential for the binding of the antiapoptotic protein hexokinase to VDAC (34). It is well established that the hydrophobic N terminus of hexokinase is required for its interactions with mitochondria (35), presumably through insertion into the membrane. Furthermore, high concentrations of Mg²⁺ are required for hexokinase to associate with the mitochondria (36). Thus, Mg²⁺ coordination to Glu-73 may reveal a structural feature that facilitates association of the N-terminal segment of hexokinase with the external surface of the VDAC pore, a process likely to be concomitant with inhibition of mitochondrial-dependent cell death.

Regulation of VDAC Permeability.

It is well established that VDAC is the principal governor for the exchange of metabolites between the intermembrane space of the mitochondria and the cytosol. The structure presented here reveals a number of features that are consistent with the open conformation of the channel and provides a possible mechanism for regulating the flow of metabolites. The arrangement of charges in the wall of the β-barrel are clustered primarily on the first half of the pore (strands 1–8), whereas the second half is less hydrophilic (β-strands 9–19). In this second half of the pore, the soluble N-terminal segment descends into the cavity forming a distorted α-helix (amino acids 6–20) containing 5 charged residues that face toward the aqueous environment. In this configuration the helix, rather than the β-barrel, contributes to a symmetrical charge distribution (Fig. 3).

There is a general consensus that the N-terminal segment of VDAC is involved in voltage gating and therefore it may adopt different conformations depending on factors external to the protein (e.g., protein–protein interactions, ion interactions, lipid environment). In our crystals, the α-helix portion of the N-terminal segment is positioned halfway down the pore, causing a narrowing of the cavity (27 × 14 Å) (Fig. 2C). This constriction site is ideally situated to regulate the flow of metabolites. The connection from the α-helix to the first β-strand is a flexible and highly conserved sequence (Gly-21–Tyr-22–Gly-23–Phe-24–Gly-25). The alternating glycine residues are well suited to perform a hinge function, facilitating rearrangement of the helix in response to stimuli. Specifically, rotation at this hinge region to move the helix away from the wall of the pore would place the N-terminal domain into a position of greater obstruction within the cavity where it would block the transport of metabolites while still permitting ion flux (Fig. 5).

What Are the Structural Rearrangements Underlying Such a Movement of the N-Terminal Gating Helix?

There are a number of stimuli known to favor the closed state of the channel: voltage potentials in excess of ± 30 mV (37), protein interactions (38), and chemical modulators like NADH (2). These stimuli may disrupt the hydrogen bonds that position the N-terminal segment along the interior wall of the β-barrel. Once interactions with the wall of the pore are disrupted, the N-terminal segment would be free to reorient within the channel lumen aided by the flexibility in the hinge region. At this point, the repositioning of the α-helix may be facilitated by the protein's inherent chemical potential: electrostatic calculations reveal a cluster of positive charges along the α-helix and a patch of negative charges located on the opposing wall (β-strands 2, 3, 5, 6, and 8) in the region where the helix could possibly interact if VDAC were to adopt the more occlusive conformation of the pore. Mutational data from the Colombini group (39) support this hypothesis, wherein substitutions of residues forming the N-terminal α-helix alter voltage-dependent gating in a charge-dependent manner. Alternatively, the 2 segments of the α-helix portion are joined by bridging bidentate hydrogen bonds from Arg-15 (Fig. 4B). Perturbation of this interaction could also impair metabolite flux by repositioning of amino acids 1–10.

Further evidence in support of alterations in the position of the α-helix came from the recent NMR structure of hVDAC1. Hiller and colleagues (21) report an NADH interaction site on β-strands 17 and 18 involving residues at position Leu-242, Ile-243, Gly-244, Ala-261, Leu-263, and Asp-264 (residues from hVDAC1) (Fig. 5). NADH has been shown to favor the closed state of the channel (2), and the interaction site described above flanks the hinge region of the N-terminal segment as determined from the data in the present study. Our structure clearly demonstrates that there is insufficient space for NADH binding in this region. Hence, the N-terminal segment would have to be repositioned to accommodate NADH and in doing so would close the channel.

As is the case with most high-resolution structures, many critical features of the protein's anatomy have been identified, explaining several key features of its physiological behavior. However, many new questions also arise that will require additional structure and functional studies to answer. The structure and hypothesized mechanism of mVDAC1 presented herein establish a template to be tested by biochemical/biophysical techniques and provides critical information for our understanding of mitochondrial biology.

Expression, Purification, and Refolding.

The expression and purification protocol was adapted from Koppel et al. (11) with extensive modifications. Briefly, histidine-tagged mVDAC1 was expressed in Escherichia coli and the inclusion bodies (IB) were isolated. Solubilized IB were purified by using a Talon affinity column and refolded by using n-dodecyl-N,N-dimethylamine-N-oxide. Refolded protein was applied to a Superdex 75 column for isolation of a homogenous protein population. More details can be found in SI Text.

Single-Channel Conductance Measurements.

The single-channel conductance across black lipid bilayers was recorded as described (13). More details can be found in the SI Text.

Crystallization and Structure Determination.

Bicelles were prepared as described (15). Purified mVDAC1 at a concentration of 15 mg/ml was mixed in a 4:1 protein/bicelle ratio with a 35% (2.8:1) DMPC/CHAPSO bicellar solution, giving 12 mg/ml mVDAC1 in 7% bicelles. Our best crystals grew at 20°C in 18–20% MPD, 0.1 M Tris·HCl (pH 8.5) with 10% PEG400 added to the protein drop only. Heavy atom labeling of single cysteine constructs was done by using methyl mercuric acetate. Native data were collected at the Advanced Light Source (Berkeley, CA) at beamline 5.0.2, and Hg-derivative data were collected at the Advanced Photon Source (Chicago) at the microfocal beamline 24-ID-E. Both native and derivative data were processed by using the program DENZO (40).

One heavy atom site was found for Cys-127, and SIRAS phases were calculated in PHENIX and MLPHARE (41). Density modification was done by using DM from the CCP4 package (42). Diffraction data from 20–2.3 Å were used for refinement and electron-density map calculations. Graphic operations and model building were performed with COOT (43). Refinement and map calculations were performed by using PHENIX version 1.3 (44) and CNS version 1.2 (45). At the end of the refinement, the F_o − F_c map was virtually featureless at ± 3 sigma. Figures were produced with PyMOL (www.pymol.org). Molecular topologies and parameters of DMPC were created by using ELBOW in PHENIX. The quality of the structure was checked and validated by using PROCHECK. Refinement statistics are shown in Table S1.