The Structural Basis for Low Conductance in the Membrane Protein VDAC upon b-NADH Binding and Voltage Gating

SUMMARY
The voltage-dependent anion channel (VDAC) forms the primary diffusion pore of the outer mitochondrial membrane. In its apo form, VDAC adopts an open conformation with high conductance. States of lower conductance can be induced by ligand bind- ing or the application of voltage. Here, we clarify at the atomic level how b-NADH binding leads to a low-conductance state and characterize the role of the VDAC N-terminal helix in voltage gating. A high-resolution NMR structure of human VDAC-1 with bound NADH, combined with molecular dy- namics simulation show that b-NADH binding re- duces the pore conductance sterically without triggering a structural change. Electrophysiology re- cordings of crosslinked protein variants and NMR relaxation experiments probing different time scales show that increased helix dynamics is present in the open state and that motions of the N-terminal heli- ces are involved in the VDAC voltage gating mechanism.

INTRODUCTION
The voltage-dependent anion channel (VDAC) is the main protein of the outer mitochondrial membrane. It forms a large diffusion pore allowing metabolite traffic between the cytosol and the mitochondrial intermembrane space. Most isoforms of VDAC re- constituted into planar lipid bilayers are sensitive to an applied electrical membrane potential, a phenomenon termed voltage gating (Benz, 1994; Colombini et al., 1996). Since the discovery of VDAC and its gating activity in the mid 1970s, resolving the underlying mechanism has been of ongoing interest but the mechanism has remained elusive. In the absence of a membrane potential, VDAC predominantly adopts a high conductance, oropen, state with a characteristic single-channel conductance of around 3.7 nS in 1 M KCl. In this open state, VDAC features a selectivity of around 2:1 for anions over cations (Colombini, 1989). At high positive or negative voltages, VDAC switches to low conductance, or closed, states, with transition midpoints of around ±30 mV. These states are characterized by conduc- tances of about 2 nS in 1 M KCl and a slight cation selectivity. In addition, while the open state of VDAC-1 conducts up to 2 3 106 adenosine triphosphate (ATP) molecules per second, this high ATP flux is essentially blocked in the closed state (Ros- tovtseva and Colombini, 1997). Apart from being triggered by a membrane potential, a closed state was also observed to be trig- gered by b-NADH (Zizi et al., 1994), low pH (Bowen et al., 1985), or hexokinase-I (Azoulay-Zohar et al., 2004) binding, as well as by interactions with pro- or anti-apoptotic proteins (Shimizu et al., 1999).

Several atomic resolution structures of the open state ofVDAC have so far been determined. The structure of human VDAC-1 (hVDAC-1) was determined in lauryldimethylamineox- ide (LDAO) detergent micelles by solution NMR spectroscopy (Hiller et al., 2008) and by a hybrid method approach combiningcrystallographic 4.1-A˚ -resolution data with a set of NMR re-straints in the detergent Cymal-5 (Bayrhuber et al., 2008). The structure of murine VDAC-1 (mVDAC-1) was obtained by X-ray crystallography crystals grown in lipidic medium and dif- fracting to a resolution of 2.3 A˚ (Ujwal et al., 2008). These threestructures all show the characteristic 19-stranded transmem- brane b barrel, a unique motive among the class of integral b barrel membrane proteins (Hiller et al., 2010a), while they resolve the N-terminal segment, which transverses the pore and aligns to the inside of the barrel wall, to different degrees (Bayrhuber et al., 2008; Hiller et al., 2008; Ujwal et al., 2008). In particular, the mVDAC-1 crystal structure was the first to resolve the break at Gly11 between a helices 1 and 2 (Ujwal et al., 2008). ATP passes the VDAC pore along basic residues of the N-terminal helices as revealed by combining molecular dynamics (MD) simulations to a Markov state model (Choudh- ary et al., 2014). Most recently, the structure of the hVDAC-1 mutant E73V was determined by NMR spectroscopy (Jaremkoet al., 2016).

In this structure, the barrel of hVDAC-1(E73V) features a distinct ellipsoidal shape relative to the circular wild-type (WT) (Jaremko et al., 2016). An ellipsoidal barrel conformation was also observed in MD studies of membrane- embedded mVDAC-1 with deprotonated E73 at pH 7 (Villinger et al., 2010) and in cases when the N-terminal helix was removed entirely (Zachariae et al., 2012).On the background of these structural data, two classes of mechanistic models have been proposed for the voltage gating process, which, as a key difference, see the N-terminal helical segment of VDAC as either rigidly aligned to the b barrel or as be- ing dynamic relative to it. Arguments for a rigidly bound helical domain are provided by the observation that removing the N-ter- minal segment leads to structural instability of VDAC and impaired or even complete loss of voltage gating (Abu-Hamad et al., 2009; De Pinto et al., 2008; Mertins et al., 2012). In addition, chemical crosslinking of the N-terminal helix a1 to the inner bar- rel wall did not affect voltage gating, further suggesting that the N-terminal segment stays aligned to the barrel during gating (Tei- jido et al., 2012). In contrast, evidence for a mobile N-terminal part was given by experiments, in which the N-terminal segment was found to be exposed to the cytosol or accommodated outside the pore on the membrane surface (Blachly-Dyson et al., 1990; De Pinto et al., 2008; Shoshan-Barmatz et al., 2010; Song et al., 1998a, 1998b).

Consequently, different possi- bilities arise for the identity of the structurally mobile element un- derlying gating, the voltage sensor. With a mobile helical segment, gating might be achieved solely by movements of the helix relative to an otherwise rigid barrel. Alternatively, a reduced channel conductance could arise from a deformation and partially collapse of the b-barrel toward ellipsoid shapes (Za- chariae et al., 2012).Here, we investigate the mechanisms underlying the transi- tions of hVDAC-1 to low-conductance states upon ligandbinding, as well as in the voltage gating process. We determine the solution struc- ture of hVDAC-1 in complex with b-NADH and employ MD simulations to explain how b-NADH-binding blocks nucleotide flux through hVDAC-1. Finally, we employ NMR spin relaxation experiments andelectrophysiology measurements to characterize the role of dy- namics in helix a2 on the hVDAC-1 voltage gating.

RESULTS
A High-Resolution Structure of hVDAC-1 in Solution Towards an improved structural description of hVDAC-1 in so- lution, we refined the previously determined NMR structure of WT hVDAC-1 (PDB: 2K4T) (Hiller et al., 2008). Although most of the b-barrel residues and the N-terminal helix a1 had previ- ously been assigned, parts of the N-terminal segment, as well as many residues in loops and turns were previously not resolved. We made use of the single amino acid mutant hVDAC-1(E73V) to increase the amount of available assign- ments. In this mutant, the negatively charged side chain Glu73, which points toward the hydrophobic membrane envi- ronment, is replaced by a valine, with the effect that the b-bar- rel is stabilized in the first six strands (Villinger et al., 2010). Consequently, solution NMR spectra of hVDAC-1(E73V) in LDAO micelles had an approximately 2-fold increased experi- mental sensitivity compared with WT hVDAC-1, allowing the sequence-specific resonance assignment of 91% of the protein backbone (Figure S1A). On the basis of this assignment, 24 additional residues could be also identified in spectra of WT hVDAC-1, increasing the backbone assignment completeness of WT hVDAC-1 from 80% to 88%. Importantly, 12 out of 15 residues of the N-terminal helical polypeptide segment (T6 to K20) are now assigned in WT hVDAC-1 with the exception of G11, which divides this N-terminal segment into two short he- lices, a1 (residues 6–10) and a2 (residues 12–20). A direct com- parison of the amide proton chemical shift differences between hVDAC-1(E73V) and WT hVDAC-1 shows major differences for residues in the proximity of the mutation site, while the rest of the protein structure is not affected (Figures S1B and S1C).

The absence of chemical shift changes is of particular impor- tance for the N-terminal segment, which thus adopts the same structural arrangement in both WT hVDAC-1 and hVDAC-1(E73V). Using electrophysiology experiments, we compared the gating properties of the hVDAC-1 E73V mutant with WT hVDAC-1. hVDAC-1(E73V) does not show any signifi- cant different gating parameters compared with WT hVDAC-1 (Figures S1D–S1G), suggesting that the region of the barrel around residue E73, which is located on the periphery of the protein, is not involved in voltage gating. However, we found that spontaneous channel insertions of hVDAC-1(E73V) into the artificial bilayer was substantially more frequent than for WT hVDAC-1, allowing more observations per experiment time. To further improve the precision of the WT hVDAC-1 solution structure, we established stereospecific assignments of the 1Hd-Leu and 1Hg-Val methyl groups by transferring the avail- able stereo-unspecific assignments (Hiller et al., 2010b) to spectra of stereospecifically labeled samples (Gans et al., 2010) (Figure S2). These additional assignments then allowed to fix the otherwise dynamically ambiguous assignments in the structure calculation. Overall, the refined solution structure of WT hVDAC-1 has a backbone root-mean-square deviation of 1.5 A˚ , compared with 2.9 A˚ before the refinement (Figures 1A, 1B, and S2; Table 1). The structure is based on a network of 911 nuclear Overhauser effect (NOE) contacts, a >50% in- crease compared with the previous structure.

The first a helix (a1) connects to a small hydrophobic patch of residues V143 and L150 on the barrel wall, while the second a helix (a2) fea- tures NOE contacts between the side chains of residue V17 and the backbone amide of residues A205 and A222 (Fig- ure 1C). The two helices are separated by residue G11, which is line-broadened beyond the detection limit by exchange pro- cesses on the intermediate NMR time scale. Overall, the N-ter- minal segment of hVDAC-1 is positioned inside the pore at the same site as in the available crystal structures, describing the open conformation of hVDAC-1 (Figure 1D). hVDAC-1 contains a binding site for the metabolite b-NADH (Hiller et al., 2008). The interaction substantially reduces ADP/ATP flux through the channel (Lee et al., 1994), and since ATP flux is also reduced in the low-conductance state (Ros- tovtseva and Colombini, 1997), it has thus been suggested that b-NADH binding triggers a conformational switch to the low-conductance state (Zizi et al., 1994). b-NADH-dependent closure was thus proposed as one mechanism by which the local b-NADH concentration might regulate mitochondrial metabolism (Zizi et al., 1994). We have previously identified the interaction surface of b-NADH with WT hVDAC-1 at strands 16–18 based on backbone amide chemical shift per- turbations (Hiller et al., 2008). Notably, the dissociation con- stant of the VDAC–b-NADH complex is 16 mM in lipid bilayers and thus well below physiological b-NADH concentrations, while in detergent micelles the dissociation is increased to 9 mM (Zizi et al., 1994; Hiller et al., 2008). Here, we further characterize the location of this binding site by chemical shift perturbation data of the ILV side-chain methyl groups, as well as by intermolecular NOEs. Strong chemical shift perturba- tions were observed for the methyl groups of ILV residues V237, L242, L262, and L263, which fully agree with the loca- tion of the changes at the protein backbone (Figures S3A and S3B).

A single-point mutant in the NADH binding site, VDAC(L242A), showed a 15% reduction in chemical shift perturbation compared with WT, corresponding to a slightly reduced affinity, while another single-point mutant, VDAC(G244A), could not be refolded (data not shown). Impor- tantly, no significant chemical shift changes occur in the N-terminal segment upon b-NADH binding, neither for the backbone amide nor in ILV methyl groups. b-NADH binding does therefore not induce a conformational change at the N-terminal a helices.Structural information of the hVDAC-1-b-NADH complex at the atomic level was obtained from 13C-resolved and 15N-resolved [1H,1H]-NOESY spectra. From these experiments, ten intermolecular NOEs could be unambiguously assigned, which connect the protein with the nicotinamide and the adja- cent ribose moiety of b-NADH. No intermolecular NOEs were observed between the protein and the adenosine moiety of b-NADH. The intensities of the NOEs to the nicotinamide moiety were on average 2.8 times higher compared with the NOEs to the ribose moiety, identifying the nicotinamide moiety as the primary binding epitope (Figure 2A). The intramolecular NOE intensitypattern of hVDAC-1 remained essentially unchanged upon b-NADH binding and the structure of the hVDAC-1-b-NADH complex was thus obtained in a structure calculation by adding the intermolecular NOEs to the restraints established for apo hVDAC-1 (Figures 2B, 2C, and S3C). Superimposition of the b-NADH conformers from the calculated structural ensemble re- veals a small binding pocket formed by residues of strands 16 and 17 (K236, V237, N238, L242, I243, and G244) of hVDAC-1that accommodates the nicotinamide moiety of b-NADH, while the rest of the b-NADH molecule is flexible. The predominant interaction of b-NADH via its nicotinamide moiety is also confirmed by chemical shift perturbation experiments showing that nicotinamide alone, but not b-NAD+, interacts with hVDAC-1 (Figure 2D).

To describe how b-NADH can affect ATP flux without inducing a structural change of the protein, we characterized hVDAC-1 using MD simulation. MD trajectories were run in a 1,2-dio- leoyl-sn-glycero-3-phosphocholine lipid bilayer, taking the hVDAC-1-b-NADH structures as initial conditions, with b-NADH removed from the structure to generate the apo state, or replacing b-NADH by b-NAD+ to generate a hypothetical hVDAC-1–b-NAD+ complex. In our simulation of hVDAC–1- b-NADH, the b-NADH molecule remained bound to the recep- tor throughout the simulation, with the location of its center ofmass along the pore axis (z) shifting only by 1–2 A˚ toward the pore center (Figures 2A and 2C). The nicotinamide moiety ex- hibits a lower root-mean-square fluctuation than the adenosine moiety, reflecting tight docking of the nicotinamide moiety against the VDAC wall and free diffusion of the adenosine moi- ety within the pore (Figure S3D). These results are consistent with the observed disorder of the adenosine moiety in the structure calculations (Figure 2B). When b-NADH is bound to the hVDAC-1 pore, chloride ions encounter a larger energy bar- rier for translocation, as indicated by their spatial probability distribution along the pore axis compared with apo hVDAC-1(Figure 3B). Thereby, Cl— ions are especially unlikely to occupylocations along the pore axis where b-NADH density is maxi- mized; these same regions have a likelihood of Cl— occupancy in the absence of b-NADH. Such anti-correlation between theion density and ligand density within the pore is consistent with steric blockage. b-NAD+ on the other hand did not main- tain stable contacts with hVDAC-1. b-NAD+ unbound in the very initial phase of the simulation (5–10 ns) and diffused to a peripheral region of the protein (Figures 3A and 3D). We did not observe b-NAD+ completely dissociate from hVDAC-1, as the limited size of the simulation box mandates that b-NAD+ interact at least weakly with hVDAC-1.

Interestingly, the valuesfor z of b-NAD+ (which has a total charge of —1) and K20 in theN-terminal segment are strongly correlated during the unbind- ing process due to favorable electrostatic interactions(Figure 3D). Overall, the simulations indicate the significance of K20 in conferring affinity to b-NADH, due to even stronger inter- actions between the positively charged K20 and b-NADH (charge of —2) than with b-NAD+ (charge of —1). Overall, our NMR and MD data thus show that b-NADH does not trigger aconformational change in hVDAC-1 but blocks anion flux by occlusion.To probe the role of the N-terminal helices in voltage gating, we used single-channel electrophysiology experiments of a cross- linked VDAC-1 mutant. Previously reported data of a mVDAC-1 mutant, chemically crosslinked between residues 10 and 170, had shown that affixing the N-terminal a1 helix to the barrel did not have an impact on the voltage gating properties (Teijido et al., 2012), while electrophysiology measurements on two other crosslinks, V3C-K119C and A14C-S193C, between the N-terminal segment and the barrel wall did show an effect on voltage gating (Mertins et al., 2012). Based on a gating model in which helix a2 can move independently of helix a1, we designed a set of four different double cysteine mutants (A14C,S193C), (A14C,A205C), (V17C,A205C), or(G21C,G246C) on the background of cysteine-free hVDAC- 1(C127S,C232S) (Craig and Dombkowski, 2013). The biochem- ical characterization showed that, for all four mutants, the crosslink can be formed reversibly, thus further confirming the position of the N-terminal segment inside the VDAC pore (Figure S4A).

We then selected the mutant hVDAC- 1(A14C,S193C) for our subsequent studies as it was the most stable mutant under oxidizing conditions. 2D [13C,1H]-HMQC NMR spectra of stereospecifically labeled LV methyl groups in the reduced and oxidized state indicate by complete conver-sion of the peak pattern that the a2 helix crosslink is efficiently formed, while leav- ing the global fold of the protein intact (Figure S4B).Single-channel electrophysiology mea- surements for WT hVDAC-1, and for themutant hVDAC-1(A14C,S193C) in either its reduced or oxidized state were recorded at an applied membrane potential of +50 mV (Figure 4). The data validated the previous work by Mertins et al. (2012) showing that the oxidized mutant, hVDAC- 1(A14C,S193C), still features gating, but its closed state is signif- icantly less stable than for WT and reduced crosslink mutant, with the consequence that the average lifetime of the closed state is substantially shorter (Figures 4D, S4C, and S4D). Thus, the chemical crosslink forces hVDAC-1 into a new type of closed state, clearly demonstrating that movements of helix a2 have a central role in the WT hVDAC-1 gating mechanism.Toward a further understanding of the VDAC N-terminal helix in voltage gating processes, we assessed the dynamics of this segment in the open state of hVDAC-1 on different time scales using NMR spectroscopy. A first indication for increased dy- namic behavior of the N-terminal segment results from the peak amplitudes in 2D [13C,1H]-HMQC spectra of ILV methyl groups (Figure 5A). The methyl groups located within or in spatial proximity to the N-terminal segment of hVDAC-1 have on average 33% lower intensity ((1.43 ± 0.32) 3 106) than the other ILV methyl groups of the barrel ((2.12 ± 0.68) 3 106).

This difference in amplitude is indicative of increased transverse relaxation and/or conformational ex- change dynamics. Hence, we quantified the transverse relax- ation rate constants of the hVDAC-1 ILV methyl resonances as well as the exchange contributions of milli- to microsecond dynamics.The contribution of conformational exchange to the R2 rates on the ms–ms time scale were determined by a [13C,1H]-multiple- quantum Carr-Purcell-Meiboom-Gill (CPMG) relaxation dispersionexperiment (Korzhnev et al., 2004). This experiment measures the effective transverse relaxation rate constants R2,eff as a function of the frequency of the refocusing pulses (nCPMG). The data show that for all LV side-chain methyl groups in hVDAC-1 in LDAO micelles, flat dispersion curves are obtained, indicating the absence of substantial contributions to relaxation from processes on the ms–ms time scale (Figures 5B and S5). In contrast, however, measurements of the transverse relaxation rate constants of the ILV methyl resonances reveal a prominent difference between residues in or close to the N-terminal segment, and those in the remainder of the barrel, consistent with increased dynamics on the ps–ms time scale (Figures 5C, 5D, and S5; Table S1). Overall, the relaxation data demonstrate that the N-ter- minal segment features increased dynamics relative to the rest of the protein already in the ground state, further pointing to a possible role of the N-terminal helices in VDAC voltage gating.

DISCUSSION
The data presented in this work have elucidated how b-NADH binding leads to reduced VDAC single-channel conductance. b-NADH binding occludes the channel conductance sterically and by electrostatic repulsion. Based on a refined atomic resolu- tion structure of hVDAC-1, we could show that the mechanism of channel conductance reduction upon b-NADH binding is different to the voltage gating process. In contrast to our starting hypothesis to find hVDAC-1 in a new conformational state, we observed no major structural changes of hVDAC-1 upon addition of b-NADH. MD simulations confirmed that the blockage effect of b-NADH binding on ADP/ATP flux can be solely explained by steric hindrance and electrostatic repulsion. Similar occlusion mechanisms and/or masking of positively charged residues in a2 might underlie other reported channel closures, such as hexoki- nase-I or the hexavalent dye RuR binding, which both induce a low-conductance state in WT, but not in E73Q-mutated mVDAC-1 (Zaid et al., 2005). The E73V mutant and the E73Q mutant each show WT gating, implying that E73 is not part of the voltage gating mechanism but rather a receptor site in agree- ment with recent electrophysiology studies (Queralt-Martin et al., 2019). Steric occlusion effects that are independent of the voltage gating mechanism might thus contribute to these re- ported channel closures and careful experiments will be required to distinguish the effects.

In addition, we have confirmed here that dynamics of the N-terminal helix underly the voltage gating process of hVDAC-1. Since chemical crosslinking of this part of the protein to the barrel wall resulted in a distinct change in voltage gating parameters, the mechanism of voltage gating must include mo- tions of the N terminus. These could be either the key gating step, reducing the conductance of VDAC directly, or as a pre- requisite to allow for subsequent conformational changes, such as distortions of the barrel shape. Relaxation VBIT-4 experiments showed that the N-terminal segment has already increased dynamics in the ground state compared with the remaining bar- rel. On the basis of the data shown here, gating models that assume a permanently rigid helix annealed to the barrel wall are ruled out.