Structure of the Lipid Nanodisc-reconstituted Vacuolar ATPase Proton Channel

Eukaryotic vacuolar H+-ATPase (V-ATPase) is a multisubunit enzyme complex that acidifies subcellular organelles and the extracellular space. V-ATPase consists of soluble V1-ATPase and membrane-integral Vo proton channel sectors. To investigate the mechanism of V-ATPase regulation by reversible disassembly, we recently determined a cryo-EM reconstruction of yeast Vo. The structure indicated that, when V1 is released from Vo, the N-terminal cytoplasmic domain of subunit a (aNT) changes conformation to bind rotor subunit d. However, insufficient resolution precluded a precise definition of the aNT-d interface. Here we reconstituted Vo into lipid nanodiscs for single-particle EM. 3D reconstructions calculated at ∼15-Å resolution revealed two sites of contact between aNT and d that are mediated by highly conserved charged residues. Alanine mutagenesis of some of these residues disrupted the aNT-d interaction, as shown by isothermal titration calorimetry and gel filtration of recombinant subunits. A recent cryo-EM study of holo V-ATPase revealed three major conformations corresponding to three rotational states of the central rotor of the enzyme. Comparison of the three V-ATPase conformations with the structure of nanodisc-bound Vo revealed that Vo is halted in rotational state 3. Combined with our prior work that showed autoinhibited V1-ATPase to be arrested in state 2, we propose a model in which the conformational mismatch between free V1 and Vo functions to prevent unintended reassembly of holo V-ATPase when activity is not needed.

olipid ring and involves membrane-embedded essential glutamate and arginine residues in the c subunits and a CT , respectively.
V-ATPase function is regulated in vivo by a unique mechanism referred to as reversible disassembly, a condition under which the enzyme dissociates into membrane-bound V o and cytoplasmic V 1 sectors (26,27) (Fig. 1). Reversible dissociation of V-ATPase is well characterized in the model organism S. cerevisiae (28), but more recent data suggest that the mammalian enzyme is regulated by a similar process in some cell types (5, 29 -31). Although the assembly status of yeast V-ATPase is mainly governed by nutrient availability (32), the situation in mammalian cells appears to be more complex. Besides glucose levels (30), V-ATPase assembly in animal cells can be induced by a variety of signals, including cell maturation (33) and stimulation by hormones (34) and growth factors (6). Upon enzyme dissociation, the activity of both sectors is silenced; that is, the V 1 no longer hydrolyzes MgATP (35,36), and the V o no longer translocates protons (37,38). Although studies in yeast suggest that V 1 activity silencing depends on the C-terminal domain of subunit H, possibly together with inhibitory MgADP (35,39,40), the mechanism by which passive proton transport across V o is blocked is less well understood.
The structure of eukaryotic V-ATPase has been analyzed by EM, and together with crystal structures of individual subunits and subcomplexes from yeast V-ATPase and related bacterial enzymes, the EM reconstructions have allowed generation of pseudoatomic models of the intact enzyme (22,41,42) and its functional V 1 (43) and V o (38,44) sectors. Although the resulting structural models together with biochemical data provide valuable information on the mechanism of ATP hydroly-sis-driven proton pumping, we only have a limited understanding of the mechanism of reversible enzyme dissociation and reassociation. We recently obtained a cryo-EM reconstruction of yeast V o (38), and although a comparison with EM models of holo V-ATPase showed that a NT undergoes a large structural change to bind the rotor subunit d in free V o , the resolution of the model was insufficient to precisely define the a NT -d interface.
Here we present a negative-stain 3D EM reconstruction of lipid nanodisc-reconstituted V o calculated at a resolution of ϳ15 Å. The model of nanodisc-bound V o suggests that the interaction between a NT and subunit d is mediated by charge complementation between acidic and basic residues on d and a NT , respectively. Site-directed mutagenesis and isothermal titration calorimetry experiments conducted with recombinant subunits identified acidic and basic patches on d and a NT that mediate the a NT -d interaction. A comparison with the recent EM reconstructions of yeast V-ATPase in three states (22) suggests that, upon enzyme dissociation, free V o is halted in state 3. We showed previously that autoinhibited, membrane-detached V 1 -ATPase is halted in state 2 (45), and we propose that this conformational mismatch to state 3 V o could function to prevent unintended reassembly of holo V-ATPase under conditions when the proton pumping activity of the enzyme is not needed.

Purification of V o Membrane Sector and Reconstitution into
Lipid Nanodiscs-We previously developed a procedure for purification of milligram amounts of yeast V-ATPase V o sector for functional and structural studies (38). Briefly, V o was solubilized from vacuolar membranes and affinity-captured via a calmodulin binding peptide fused to the C terminus of the vac- , and green (d). ATP hydrolysis in V 1 drives rotation of the c-ring, resulting in proton translocation across the interface of the c-ring and a CT . Upon reversible disassembly, subunit C is released into the cytoplasm, and the interactions between subunits of V 1 (DF and EG1-3) and V o (a NT , d) are broken. Disassembly of the enzyme results in a V 1 that does not hydrolyze MgATP and a V o that does not support passive proton translocation. Note that, upon release of V 1 from the membrane, a NT changes conformation to bind the central rotor subunit d (red asterisk) as reported previously (38).
uole-specific isoform of subunit a (Vph1p). For structural studies under more native-like conditions, V o was reconstituted into lipid nanodiscs as described under "Experimental Procedures" (Fig. 2). Purified detergent-solubilized V o (Fig. 2a) was mixed with Escherichia coli polar lipids and the recombinant membrane scaffold protein MSP1E3D1 (Fig. 2b), followed by detergent removal with polystyrene beads. In a final purification step, V o -containing nanodiscs (V o ND) were separated from "empty" discs by a second calmodulin affinity binding step followed by size exclusion chromatography on a Superdex S200 column (Fig. 2, c and d). Peak fractions of V o ND eluted from the gel filtration column were pooled, and the concentrated preparation was analyzed by SDS-PAGE and silver staining (Fig. 2e).
3D EM Reconstruction of V o ND-We initially generated a dataset of ϳ30,000 particles from EM images of negatively stained V o ND. However, attempts to reconstruct a 3D model of the complex using reference-free algorithms were unsuccessful, likely because of the limited size of the complex and the lack of characteristic features required for alignment. Models did not converge on a specific handedness, i.e. orientation of a NT with respect to the membrane sector, and density for a NT was smeared out over the membrane and did not allow us to distinguish its positioning relative to subunit d. We therefore decided to make use of the calmodulin binding peptide on the C terminus of subunit a to introduce an additional asymmetry to aid in alignments and angle determination in the 3D startup procedure. Purified calmodulin (CaM) was incubated with V o ND, followed by removal of excess CaM using glycerol gradient centrifugation (Fig. 2, f-h). Negative stain electron microscopy showed that the final V o ND-CaM preparation was monodis-FIGURE 2. Purification of V o and reconstitution into lipid nanodiscs. a, SDS-PAGE of yeast V o affinity-purified from solubilized yeast microsomal membranes via a calmodulin binding peptide fused to the C terminus of subunit a. b, SDS-PAGE of the membrane scaffold protein MSP1E3D1, purified by affinity chromatography via an N-terminal His 6 tag. c, SDS-PAGE of flow-through and elution fractions from the calmodulin column after nanodisc reconstitution to remove unfilled discs. Reconstitution of the V o into lipid nanodiscs is accomplished by mixing V o , MSP, and lipid. Upon removal of detergent, V o self-assembles into a nanodisc bilayer patch. d, size exclusion chromatography of V o ND after removal of unfilled discs. e, SDS-PAGE of the final preparation after gel filtration. f, glycerol gradient of V o ND-CaM. V o ND was mixed with a 5-fold excess of calmodulin, and the mixture was applied to a discontinuous 15-35% glycerol gradient and centrifuged at 200,000 ϫ g for 16 h at 4°C. Fractions were collected from the bottom of the gradient and analyzed by SDS-PAGE. Peak fractions (4,5) were pooled and used for negative stain electron microscopy. g and h, to verify binding of calmodulin to V o ND, calmodulin (A47C) was labeled with fluorescein maleimide and the mixture of V o ND, and labeled calmodulin was subjected to glycerol density centrifugation as in f. h, the fluorescence scan of the gel shown in g indicates co-migration of labeled calmodulin with V o ND. The gels in a-c, f, and g were stained with Coomassie blue; the gel in e was stained with silver. perse, with an average diameter of the particles of ϳ12 nm (Fig.  3a). A dataset of ϳ40,000 particles, generated from 380 micrographs such as shown in Fig. 3a, was subjected to reference-free alignment procedures as implemented in EMAN2. Fig. 3b shows class averages of side, top, and intermediate view projections (center row), representative raw particle images (top row), and the corresponding reprojections of the final 3D model (bottom row). Class averages, including those shown in Fig. 3b, were used for a 3D startup procedure in EMAN2, and the resulting 3D reconstruction was refined until stable. At this point, the reconstruction was strongly low pass-filtered and used as input for the 3D autorefinement procedure as implemented in the Relion 1.3 software package. The model was then refined until no further improvement was observed. The resolution of the final V o ND-CaM 3D reconstruction was estimated to be 14.9 Å (20.3 Å at 0.5 correlation) using the "gold standard" FSC protocol as implemented in Relion 1.3 (Fig. 3c).
The final V o ND-CaM model was strongly low pass-filtered to serve as a reference for a new V o ND dataset (ϳ47,000 images) using the Relion 3D autorefinement procedure as described for the V o ND-CaM dataset. The 3D autorefinement converged to a final model of V o ND with an estimated gold standard resolu-tion of ϳ14.9 Å (16.3 Å at 0.5 correlation, Fig. 3d). As can be seen from Fig. 3, c and d, the final V o ND-CaM and V o ND models are very similar except for a small density in the V o ND-CaM map that is due to the CaM bound at the a subunit C terminus (Fig.  3c, top panel, red circle). However, because the V o ND-CaM map showed slightly more detail, we used this model to illustrate the features of nanodisc-bound V o as summarized in Fig. 4.
Side, top, and bottom views of the V o ND-CaM model are illustrated in Fig. 4, a-c, showing the characteristic features as seen in earlier reconstructions of bovine (44) and yeast (38) V o , including the densities above the membrane (a NT and subunit d), the cleft between the density for a CT and the c-ring (Fig. 4a,  arrowhead), which opens into a solvent (stain)-accessible pore as seen in the bottom view (Fig. 4c, arrowhead), and the large cavity on the cytoplasmic side of the c-ring (Fig. 4b, arrowhead). Semiautomatic fitting of homology models of the yeast V o subunits into the EM density is summarized in Fig. 4, d-f. As illustrated in Fig. 4, d and e, a NT (blue) was positioned with its proximal lobe (which is comprised by the N and C termini of subunit a, domain nomenclature as in Ref. 46) near the connection point to the membrane-bound a CT , placing its distal lobe near the central density corresponding to subunit d (green). Because of its pseudo-3-fold symmetry and the limited resolution of the EM reconstruction, the yeast subunit d homology model could be fit in three orientations corresponding to rotational states 1, 2, and 3, as seen in the recent cryo-EM reconstructions of intact yeast V-ATPase (22). However, only the orientation corresponding to state 3 preserved the contact between the subunit d N-terminal ␣ helix and the cytoplasmic face of the c subunit ring, as seen in the recent cryo-EM model of holo V-ATPase in state 3 (22) (Fig. 4g), and we therefore explored the contacts between a NT and d predicted by this configuration (Fig. 4h, see below). For filling the density corresponding to the yeast c-ring, we used the crystal structure of the K 10 ring from Enterococcus hirae (47) (Fig.  4h, magenta).
Interaction of a NT and Subunit d-We previously reported a 3D reconstruction from cryo-EM images of detergent-solubilized yeast V o that showed a NT and subunit d in close proximity (38). Although subsequent binding studies using recombinant a NT (residues 1-372, a NT(1-372) ) and d revealed a K d of the a NT(1-372) -d interaction of ϳ5 M (38), the resolution of the cryo-EM model, was insufficient to define the binding site(s) between a NT and d in detail. Fig. 4h illustrates the interface between a NT and subunit d based on the fitting of the homology models into the EM density of the V o ND-CaM model presented here. As can be seen, a NT and subunit d appear to contact each other via two distinct sites near the proximal and distal lobes of a NT (Fig. 4h). The contact near the distal lobe (Fig. 4h, bottom right) is mediated by two short ␣ helices, one from a NT (residues 242-256, yeast subunit a isoform Vph1p) and one from subunit d (residues 144 -154). The other site near the proximal lobe of a NT is mediated by a short helix-turn-helix motif in d (residues 38 -58) and a less well defined face in a NT (Fig. 4h,  bottom left). Many of the acidic and basic residues involved in these two contact sites are highly conserved from yeast to human (see isothermal titration calorimetry sections below). Considering the conserved nature of the charged residues, we reasoned that complex formation between a NT and d could be driven by electrostatic interactions involving the conserved residues. To test this hypothesis, we generated a NT double (R250A,K251A) and quadruple mutants (K247A,R250A, K251A,E254A) in the short ␣ helix in the distal lobe of a NT and triple (D144A,E146A,E150A) and quadruple (D37A,D40A,D41A, K43A) mutants in subunit d for in vitro binding experiments using isothermal titration calorimetry (ITC). As a negative control, we generated a triple mutant of d in an area predicted to be outside of the a NT-d contact area (E198A,E199A,E202A).
The wild type and alanine mutants of a NT(1-372) and subunit d were expressed as N-terminal fusions with maltose binding protein (MBP) connected via a protease cleavage site as described previously (38). The proteins were purified away from MBP via anion exchange and size exclusion chromatography. Purity and proper folding of the resulting a NT(1-372) and subunit d constructs was confirmed by Coomassie-stained SDS-PAGE and CD spectroscopy, respectively (Fig. 5, a and b). As can be seen from the CD spectra, all a NT(1-372) and subunit d mutant proteins showed minima at 208 and 222 nm similar to the wild-type subunits and characteristic for highly ␣-helical proteins, indicating that the mutations did not interfere with proper folding of the recombinant subunits.
Isothermal Titration Calorimetry of Mutant a NT  and Wild-type d-We first conducted ITC titrations with double and quadruple mutants of the conserved charged residues in the short ␣ helix of the a NT distal domain that were facing subunit d (Fig. 6a). Titrating subunit d into wild-type a NT(1-372) revealed an exothermic binding reaction (Fig. 6b). Fitting the data with a single-site binding model revealed a ϳ1:1 stoichiometry with a K d of 6.7 M, similar to what we obtained earlier by titrating a NT(1-372) into subunit d (38). The ⌬H and ⌬S were Ϫ34.2 kJ/mol and Ϫ21.6 J⅐(mol⅐K) Ϫ1 , respectively, giving a ⌬G of the enthalpy-driven binding reaction of Ϫ28.1 kJ/mol. Titrating subunit d into the double a NT(1-372) mutant (R250A,K251A) produced significantly less heat (⌬H ϭ Ϫ7.4 kJ/mol), but, at the same time, the K d was similar compared with the wild-type proteins (K d ϭ 4 M, Fig. 6c) and appeared to be partly driven by entropy (⌬S ϭ 77.8 J⅐(mol⅐K) Ϫ1 , ⌬G ϭ Ϫ29.4 kJ/mol). On the other hand, titrating subunit d into the quadruple mutant of a NT(1-372) (K247A,R250A,K251A,E254A) resulted in a weak endothermic reaction that could not be fit to a single site model without fixing the stoichiometry at 1:1 (Fig. 6d). Under these conditions, the K d was ϳ1 mM, indicating that replacement of all four conserved charged residues by alanines disrupted the interaction between a NT(1-372) and d.
The cell contents of the ITC experiments (with titrated wildtype a NT(1-372) , double mutant, and quadruple mutant) were subjected to size exclusion chromatography as described under "Experimental Procedures." Subunit d and a NT(1-372) alone eluted at 65 and 60 ml, respectively (Fig. 6, e and f). Note that although subunit d alone runs as a monodisperse monomer (Fig. 6, e and j), a NT(1-372) exists in a concentration-dependent monomer-dimer equilibrium as reported earlier (38) and as evident from its elution profile and accompanying SDS-PAGE gel  FEBRUARY 3, 2017 • VOLUME 292 • NUMBER 5 (Fig. 6, f and k). The elution profile of the mixture of wildtype a NT(1-372) and d revealed two peaks at 58 and 65 ml (Fig.  6g), and analysis by SDS-PAGE showed that the peak around 58 ml (fraction 29) contained close to stoichiometric amounts of a NT(1-372) and d (Fig. 6l). The relatively small shift of 2 ml (one fraction) toward larger molecular size is consistent with the moderate K d of the a NT(1-372)-d complex formation of 6.7 M and in close agreement with our previous study (38).

Structure of V o in Lipid Nanodiscs Defines a NT -d Interface
In the size exclusion profile of the double mutant titration (subunit d into a NT(1-372) R250A,K251A), the interacting peak is shifted slightly to a larger volume at ϳ61 ml as a shoulder of the subunit d peak (ϳ66 ml) (Fig. 6h), indicating a weakening of the a NT(1-372) -d interaction and consistent with the reduced binding enthalpy and the SDS-PAGE gel of the peak fractions (Fig. 6m). This trend is continued for the titration of the quadruple mutant (subunit d into a NT(1-372) K247A,R250A,K251A, E254A), which only shows a shoulder for a NT unresolved from the subunit d peak (Fig. 6i) with elution profiles similar to the individual subunits (Fig. 6n). Taken together, the ITC and gel filtration data showed that mutation of the conserved charged residues on the short ␣ helix in the distal domain of a NT disrupts the a NT -d interaction.
Isothermal Titration Calorimetry of Mutant d and Wild-type a NT(1-372) -As mentioned above, subunit d appears to contact a NT at two sites, the short ␣ helix in the distal domain and at a second site near the proximal domain (Fig. 4h). To verify the fit of subunit d in the EM model and to test whether the contact site near the proximal lobe of a NT contributes to the interaction between the two subunits, we generated triple and quadruple mutants of subunit d by replacing conserved charged residues that are facing a NT from the two sites on d (Fig. 7, a and e). ITC titrations of both triple (Fig. 7b) and quadruple (Fig. 7f) alanine mutants of d with wild-type a NT(1-372) revealed weak endothermic reactions that could not be fit to a single site binding model without fixing n ϭ 1. Under these conditions, the K d s for the two titrations of the triple and quadruple mutants of d were ϳ0.25 mM and ϳ1.7 mM, respectively. Consistent with the ITC titrations, gel filtration profiles (Fig. 7, c and g) and SDS-PAGE of the peak fractions (Fig. 7, d and h) indicated elution of noninteracting subunits. Contrary to the alanine mutations of residues predicted to be in the a NT -d binding interface (Fig. 7,  a-h), mutagenesis of a patch of acidic residues outside the interface (E198A,E199A,E202A) did not interfere with complex formation (Fig. 7, i-l). Taken together, the ITC and gel filtration experiments showed that both contact sites between a NT and d as seen in the EM fit contribute to the binding interaction between the two subunits. However, because disrupting either of the two sites weakened the interaction beyond detection by ITC or gel filtration, this suggests that the individual interactions are weak and that only the combined avidity of the two interactions results in a measurable affinity.

Discussion
We have developed a protocol to reconstitute purified V-ATPase V o membrane sector into lipid nanodiscs. When reconstituted into nanodiscs, V o is stable, as evident from the lack of subunit a degradation products (Fig. 2e) sometimes seen in the detergent-solubilized complex (38). Negative stain EM showed that the preparation is monodisperse, and we were able to reconstruct 3D models of the nanodisc-bound V o with and without calmodulin bound to the calmodulin binding peptide at the C terminus of subunit a. Fitting of a NT and subunit d homology models into the EM density revealed that the two subunits are in contact, as described previously for the negative stain and cryo-EM models of detergent solubilized bovine and yeast V o , respectively (38,44). However, the slightly better resolution of the V o ND models allowed us to identify two sites of contact between a NT and d, both involving charged and highly conserved residues. ITC analysis of triple and quadruple alanine mutants of a NT and d confirmed the involvement of the charged residues in the interaction, but the analysis also showed that the individual interactions are weak and that only the combined avidity of both binding sites leads to a measurable K d of ϳ6 M (Ref. 38 and the data presented here). As mentioned under "Introduction," V-ATPase is regulated by a reversible disassembly mechanism that results in membranedetached V 1 and membrane-bound free V o . A relatively moderate affinity between d and a NT can be rationalized by the fact that this interaction has to be broken when reassembly of holo V-ATPase is initiated.
A comparison of the V o ND 3D models with the recent cryo-EM reconstructions of holo V-ATPase in three states (22) revealed that free V o appears to be halted in rotary state 3 based on the orientation of subunit d relative to a NT . As mentioned under "Results," because of the limited resolution of the V o ND 3D models and because of the pseudo-3-fold symmetry of subunit d, d could also be fit in the state 2 orientation with a comparable model-map correlation compared with the state 3 fit (the fit of state 1 is of much lower quality, see Fig. 4). The ITC data, however, are only consistent with the state 3 orientation. Furthermore, only the state 3 orientation preserves the contact between the N-terminal ␣ helix of d and the cytoplasmic loops of the proteolipid ring, as seen in the models of all three states of the holo enzyme (22). Taken together, the data therefore indicate that free V o is halted in a single conformation corresponding to state 3 of holo V-ATPase.
Recently, we determined the 6.2-to 6.5-Å crystal structure of autoinhibited yeast V 1 -ATPase (40). Interestingly, a comparison of the structure of autoinhibited V 1 with the structure of V 1 as part of holo V-ATPase (22) revealed that the membranedetached V 1 is halted in state 2 based on the rotational position of the DF rotor relative to the inhibitory H subunit (40). The observation that autoinhibited V 1 is halted in state 2 together with the findings presented here that free V o is halted in state 3 indicates that there is a conformational mismatch between the two complexes as a result of regulated enzyme disassembly. We speculate that this mismatch may serve to prevent unintended reassembly of the enzyme when the disassembled state is required. How this mismatch is relieved when reassembly is required is not known, but it is possible that V 1 binding to the assembly chaperone regulator of the H ϩ -ATPase of vacuolar and endosomal membranes (RAVE) (48) changes V 1 conformation to enable V o binding. Another possibility is that the V o conformation is altered by interac-tion with specific phosphoinositides that have been shown to bind to V o and promote enzyme assembly on the vacuolar membrane (49).
Here we have shown that yeast V o can be reconstituted into lipid nanodiscs, resulting in a highly monodisperse preparation that is amenable to structure determination by single-molecule EM. Future studies using cryo-EM will allow high-resolution structural studies of the complex in a more native environment compared with the detergent-solubilized state, allowing, for example, an examination of the interaction with specific lipid molecules that have been shown to be either essential for V-ATPase function (50) or  (1-372) . i-l, as a negative control, a subunit d mutant with acidic residues (Glu-198, Glu-199, and Glu-202) outside of the predicted interface with a NT changed to alanines was titrated with wild-type a NT  . Fitting the data revealed a K a of ϳ4 ϫ10 5 Ϯ 8 ϫ 10 4 M (K d ϳ 2.5 M; N ϳ 1.2; ⌬H ϭ Ϫ12.4 Ϯ 0.49 kJ/mol; ⌬S ϳ 63 J⅐(mol⅐K) Ϫ1 ). Shown are representative ITC titrations of at least two repeats for each mutant. Note that the gel filtration column was repacked after the experiments in Fig. 6, resulting in a slightly different elution volume for the recombinant subunits for the two sets of titrations shown in Figs. 6 and 7. FEBRUARY 3, 2017 • VOLUME 292 • NUMBER 5 involved in the mechanism of reversible enzyme disassembly (49). These studies are ongoing in our laboratory.

Experimental Procedures
Reagents-Undecyl-␤-D-maltoside (UnDM) was from Anatrace. E. coli polar lipid extract was obtained from Avanti. Calmodulin-Sepharose beads were from GE Healthcare or Agilent. CDTA was from Fisher Scientific. All other reagents were of analytical grade.
Purification of Yeast V o -Cell growth of a yeast strain expressing subunit a isoform Vph1p with a C-terminal fusion of calmodulin binding peptide, membrane preparation, and V o extraction and purification were performed as described previously (38) with the following modifications. Yeast cells were harvested during the second log phase by centrifugation at 2600 ϫ g, washed in water, and resuspended in lysis buffer (25 mM Tris-HCl (pH 7.4), 500 mM sorbitol, and 2 mM EGTA), and broken in a Bead Beater (Omni International) using zirconium beads (BioSpec). After removing cell debris and mitochondria by low-speed (2500 ϫ g, 10 min) and medium-speed (12,000 ϫ g, 20 min) centrifugation, membranes were collected by ultracentrifugation (370,000 ϫ g, 2 h), washed once in buffer (25 mM Tris-HCl (pH 7.4) and 500 mM sorbitol), and pelleted again (370,000 ϫ g, 1 h). The total protein concentration of the membrane samples was determined by BCA protein assay (Thermo Scientific) of trichloroacetic acid-precipitated membranes. Isolated membranes were diluted to 10 mg/ml and stored at Ϫ80°C until use.
Membranes were solubilized by addition of UnDM from a 20% stock solution in water to a final concentration of 0.6 mg of detergent/mg of membrane protein for 1 h with gentle agitation. Extracted membranes were supplemented with 4 mM CaCl 2 and centrifuged at 180,000 ϫ g for 1 h to remove the insoluble fraction. The supernatant was then applied to a 5-ml calmodulin-Sepharose column pre-equilibrated in calmodulin washing buffer (10 mM Tris-HCl (pH 8), 150 mM NaCl, 2 mM CaCl 2 , 10 mM ␤-mercaptoethanol, and 0.1% UnDM). The column was washed with 5 column volumes each of washing buffer and washing buffer without NaCl and eluted with elution buffer (10 mM Tris-HCl (pH 8), 10 mM CDTA, 10 mM ␤-mercaptoethanol, and 0.1% UnDM).
Lipid Nanodisc Reconstitution of V o -E. coli total lipid extract (Avanti Polar Lipids) was suspended by sonication in disc-forming buffer (20 mM Tris-HCl (pH 7.4), 100 mM NaCl, and 0.5 mM EDTA) with the addition of 1 mM DTT (EMD Millipore). Detergent-solubilized V o , purified MSP, and lipid were combined at a molar ratio of 0.02:1:25 with the addition of protease inhibitors, 1 mM PMSF, 1 mM leupeptin, 1 mM pepstatin, and 1 mM chymostatin (EMD Biosciences) and incubated at room temperature for 1 h with mixing. Prewashed Bio-Beads SM-2 (Bio-Rad) were added at 0.4 g/ml and incubated with mixing for 2 h at room temperature. The self-assembled nanodisc sample was recovered from the Bio-Bead mixture with a syringe. To remove unfilled (empty) nanodiscs (ND) from V ocontaining discs (V o ND), the reconstituted sample was supplemented with 10 mM CaCl 2 and applied to a 1-ml calmodulin resin column, washed with disc-forming buffer and eluted with the same buffer without CaCl 2 and supplemented with 10 mM CDTA. As a final polishing step, the eluted V o ND sample was concentrated to 2 ml, applied to a Superdex 200 HR 16/500 column on an AKTA FPLC (GE Healthcare) equilibrated with disc-forming buffer and eluted at 0.5 ml/min.
Preparation of Calmodulin-The gene for human calmodulin 1 was synthesized (BioBasic, Markham, ON, Canada) and cloned into a modified pMAL-c2E expression vector with a Prescission protease cleavage site between MBP and the N terminus of calmodulin. Briefly, E. coli Rosetta 2 harboring the calmodulin expression plasmid was grown to mid-log phase in rich broth (Luria broth supplemented with 0.2% glucose), and expression was induced with 1 mM isopropyl 1-thio-␤-D-galactopyranoside for 18 h at 18°C. Cells were harvested by centrifugation, resuspended in lysis buffer (20 mM Tris-HCl (pH 7.4), 200 mM NaCl, and 1 mM EDTA and lysed by sonication (Hielscher Ultrasonics). The lysate was cleared by centrifugation at 20,000 ϫ g and passed over a pre-equilibrated amylose column (New England Biolabs). Bound protein was washed using the same buffer and eluted with the buffer supplemented with 10 mM maltose. Protein was cleaved using Prescission protease to remove the MBP tag and dialyzed into anion exchange buffer (20 mM Tris-HCl (pH 7.4), 1 mM EDTA, and 1 mM DTT). The sample was passed over a MonoQ anion exchange column attached to an FPLC and pre-equilibrated in buffer and eluted using a 30-column volume linear gradient of buffer to buffer plus 500 mM NaCl. Calmodulin-containing fractions were pooled, concentrated, and subjected to size exclusion chromatography (Superdex 75 16/500 column). For fluorescence detection of calmodulin, residue Ala-47 was changed to cysteine using QuikChange site-directed mutagenesis with the following primers: A47_fwd, GCC AGA ATC CAA CCG AAT GTG AAC TGC AAG ATA TGA TTA ACG; A47_rev, CGT TAA TCA TAT CTT GCA GTT CAC ATT CGG TTG GAT TCT GGC. For fluorescence detection, calmodulin (A47C) was reacted with fluorescein maleimide for 1 h in the dark. Excess label was removed by a Sephadex G25 spin column.
Labeling of V o ND with Calmodulin-Purified calmodulin was added in a 5:1 molar ratio to V o ND, and the sample was loaded onto a discontinuous glycerol gradient (15-35% (v/v), 10 mM MOPS (pH 7), and 4 mM CaCl 2 ) for separation of unbound calmodulin. The gradient was subjected to centrifugation at 285,000 ϫ g for 16 h. Afterward, 1-ml fractions were collected by fractionation from the bottom of the gradient and analyzed by SDS-PAGE.
Electron Microscopy-Carbon-coated copper grids were subjected to glow discharge in air for 45 s. Samples of V o ND and calmodulin-labeled V o ND (V o ND-CaM) at ϳ1 mg/ml were diluted 1:100 in 20 mM Tris-HCl (pH 7.4) and 150 mM NaCl supplemented with 0.5 mM EDTA or 4 mM CaCl 2 , respectively, and applied to glow-discharged grids for 1 min, washed with water, and stained with 2% (w/v) uranyl formate (Electron Microscopy Science). Grids were observed in a JEOL JEM-2100 transmission electron microscope operating at 200 kV. Images were acquired on a charge-coupled device (F415MP, Tietz Video and Image Processing Systems GmbH) at a nominal magnification of ϫ60,000 and an underfocus of between ϳ1-2 m. The calibrated pixel size on the specimen level was 1.75 Å.
Image Analysis and 3D Reconstruction-Single particles were selected (192 ϫ 192 pixels) using e2boxer from EMAN2 with a semiautomated monitored picking procedure. Particles were phase flip-corrected for their contrast transfer functions by micrograph within the EMAN2 package. Datasets of 47,422 and 40,092 particles were collected for V o ND and V o ND-CaM, respectively. The datasets were normalized, bandpass-filtered, binned by 2, and circularly masked. Datasets were analyzed using reference-free alignment as implemented in EMAN2.
Twenty-seven class averages from the CaM-V o ND dataset were selected for use as initial references in a startup procedure for 3D reconstruction. Initial models were generated using the standard EMAN2 procedure and subjected to seven rounds of refinement. The resulting model was low pass-filtered to 40 Å and used to start up automated 3D refinement in Relion 1.3 (52). Half-datasets were refined independently for six iterations, and the unfiltered final maps served as inputs to the "relion_postprocess" protocol for automatic masking and a resolution estimate from calculation of the corrected gold standard Fourier shell correlation (FSC). The resolution was estimated at 14.9 Å using the 0.143 gold standard FSC cutoff (20.3 Å at 0.5 FSC). The final model was filtered to 40 Å and used as a reference for starting up the 3D reconstruction of the V o ND dataset using the Relion 1.3 autorefinement procedure as described above for the V o ND-CaM dataset. The resolution of the V o ND reconstruction was estimated at 14.9 Å using the 0.143 FSC cutoff (16.3 Å at 0.5 FSC).
Fitting of Atomic Models of Subunits-We used the atomic structure of the K 10 ring from E. hirae (PDB code 2BL2) to fit into the corresponding density for the yeast V o c-ring. Homology models were calculated using Phyre2 (53) for a NT and subunit d against template bacterial subunit structures of Meiothermus ruber I NT (PDB code 3RRK) and Thermus thermophilus C subunit (PDB code 1V9M). Semiautomated and manual docking of the structures was performed in Chimera (54).

Structure of V o in Lipid Nanodiscs Defines a NT -d Interface
Circular Dichroism Spectroscopy-Far UV CD spectra of a NT(1-372) and subunit d constructs were collected on an Aviv 420 spectrometer in 25 mM sodium phosphate (pH 7.4) in a 1-mm path length cuvette at 10°C. For the subunit d scans, 0.1 mM tris(2-carboxyethyl)phosphine (TCEP) was included in the buffer. Scans of buffer only were subtracted from the spectra.
Isothermal Titration Calorimetry-The interactions between subunit d and a NT(1-372) constructs were determined using a VP-ITC (MicroCal). Proteins were prepared in 20 mM Tris (pH 7), 0.5 mM EDTA, and 1 mM TCEP at 10°C. Ligand (subunit d, 370 -400 M) was titrated into solutions of a NT(1-372) constructs (25 M). The titrations were corrected for ligand heat of dilution by subtraction of blank experiments of ligand into buffer. Analysis of data was performed using MicroCal VP-ITC Origin software. After completion of the titrations, the cell contents were subjected to size exclusion chromatography over a Superdex 200 H/R 16/500 column. At least two titrations were performed for each wild type and mutant construct combinations.
Other Methods-Protein concentrations were determined using the BCA assay (Pierce) after TCA precipitation of proteins as described previously (38). Site-directed mutagenesis was verified by DNA sequencing (Eurofins).
Author Contributions-N. J. S. and S. W. designed the study. All experiments were carried out by N. J. S. N. J. S. and S. W. wrote the manuscript.