Architecture and rearrangements of a sperm-specific Na+/H+ exchanger

Abstract The sperm-specific sodium hydrogen exchanger, SLC9C1, underlies hyperpolarization and cyclic nucleotide stimulated proton fluxes across sperm membranes and regulates their hyperactivated motility. SLC9C1 is the first known instance of an ion transporter that uses a canonical voltage-sensing domain (VSD) and an evolutionarily conserved cyclic nucleotide binding domain (CNBD) to influence the dynamics of its ion-exchange domain (ED). The structural organization of this ‘tripartite transporter’ and the mechanisms whereby it integrates physical (membrane voltage) and chemical (cyclic nucleotide) cues are unknown. In this study, we use single particle cryo-electron microscopy to determine structures of a metazoan SLC9C1 in different conformational states. We find that the three structural domains are uniquely organized around a distinct ring-shaped scaffold that we call the ‘allosteric ring domain’ or ARD. The ARD undergoes coupled proton-dependent rearrangements with the ED and acts as a ‘signaling hub’ enabling allosteric communication between the key functional modules of sp9C1. We demonstrate that binding of cAMP causes large conformational changes in the cytoplasmic domains and disrupts key ARD-linked interfaces. We propose that these structural changes rescue the transmembrane domains from an auto-inhibited state and facilitate their functional dynamics. Our study provides a structural framework to understand and further probe electrochemical linkage in SLC9C1.


Introduction
Hyperactivation of sperm in female reproductive tracts is associated with dramatic changes in the agellar dynamics, which is pivotal for fertilization 1 .The process involves key signaling events at the sperm membranes, such as calcium in ux and membrane hyperpolarization, and in the cytoplasm, such as elevation of cAMP levels and pH.Several sperm-speci c biomolecular machineries are critical for temporal and spatial coordination of these signal cascade events.For instance, the CatSper channels 2,3 mediate calcium entry into the cell that trigger phosphorylation events accompanying sperm motility changes.Another unique protein which has coevolved with CatSper in metazoans and is a critical player in sperm capacitation is the protein product of the SLC9C gene [4][5][6] .
The SLC9 family of ion exchangers are membrane proteins that couple the uptake of Na + ions to proton extrusion and play key roles in salt and pH homeostasis in all kingdoms of life 7,8 .Mammals express thirteen different subtypes of SLC9s which are organized into three evolutionarily distinct sub-families -9A, 9B and 9C 9 .Some of them, such as SLC9A1, are ubiquitously expressed at the plasma membrane, where they ful ll critical house-keeping functions.Others, such as SLC9A9 exhibit tissue and organellespeci c expression and accomplish specialized functions centering around their abilities to maintain salt and pH balance in speci c cellular compartments.The role of the sperm-speci c SLC9C1 in sperm physiology was rst established using knock-out mouse models 10 which showed that SLC9C1-null mice were infertile and showed severely compromised sperm motility.More recently a genetic mutation in human SLC9C1 has been linked to asthenozoospermia 11 .SLC9C has been recognized to be a potential target for the treatment of male infertility as well as the design of non-hormonal male contraceptives 12 .
SLC9C exhibit characteristic differences from other SLC9s in terms of their structural and regulatory attributes.In SLC9As and 9Bs, and their prokaryotic orthologs of known structure, the membranedelimited ion-exchange domains (EDs) are tethered to relatively short or largely unstructured C-terminal soluble domains [13][14][15][16][17][18] .In contrast, SLC9Cs, feature a homologous ED, a canonical voltage-sensing domain (VSD) and a cyclic nucleotide binding domain (CNBD), interconnected via long, structured linkers, all within a single polypeptide.Foundational experiments recently performed with the sea urchin ortholog of SLC9C1 19 (sp9C1) have revealed that it is a bona-de electroneutral sodium-hydrogen exchanger.However, its ion exchange activity is under tight control of the VSD and is turned on only upon membrane hyperpolarization.cAMP binding tunes the voltage range over which it is functionally active, shifting it to less hyperpolarizing voltages.In terms of such regulatory mechanisms, sp9C1 is reminiscent of the wellstudied HCN channels 20 .However, the mechanisms underlying voltage and cyclic nucleotide modulation of SLC9C1 function are not understood.To this end, we elucidate several structures of sp9C1 in different conformational states, using single particle cryo-electron microscropy (cryoEM) and reconstruction methods.

Architecture and assembly of SLC9C1
We expressed full-length sp9C1 in mammalian cells as an eGFP fusion construct.Puri cation of the protein in digitonin micelles yielded biochemically monodisperse protein which was used for single particle cryoEM imaging (Extended Fig. 1a-c).We will rst describe the overall architecture of sp9C1 obtained at pH 8 with K + as the dominant cation which yielded a nal reconstruction map at GS-FSC resolution of 3.8 Å (Extended Fig. 1d, e) The EDs of the two subunits pack against each other (Figs.1a-c, Supplementary Video 1) forming the core of the protein.Each ED comprises 13 transmembrane helices (TM1 through TM13), organized in a manner similar to other members of the SLC9 family (Extended Fig. 2a, b).Seven of the thirteen transmembrane helices (TMs 1-3 and 7-10) form the 'dimerization domain' and the remaining six (TMs 4-6 and 11-13) constitute the 'core domain' (Extended Fig. 2c).A wide, cytoplasmically-accessible funnel, lined with negatively charged residues, is observed between the dimerization and core domains of the ED (Extended Fig. 2c and d).The tip of the funnel features conserved residues poised to bind a Na + ion 21 .Thus, in our structure the ED is in an "inward open" state.Despite the structural similarity at an individual subunit level, the dimeric organization of the two EDs in sp9C1 is markedly different from other SLC9s (Extended Fig. 2e).TM10 of the two subunits of sp9C1, which line the inter-subunit cavity, tilt further away from each other by ~ 12º causing their extracellular ends to splay apart by an additional ~ 12Å, thus widening the crevice at the ED-dimer interface.In all our reconstructions, we were able to reliably identify densities for three lipid and four detergent molecules in this extracellularly accessible cavity (Fig. 1c).Lipid occupancy of this cavity might determine the stability of the sp9C1 dimers and act as clamps, holding the dimerization domain in place as the core helices move during the gating cycle of the ED 15,18,22,23 .
The VSDs (each comprising helices S1 through S4) are structurally estranged from the EDs, appearing as " oating buoys" on either side of the ED-dimer interface.At the closest, the VSD (extracellular end of S1) is ~ 22Å away from the ED (extracellular end of TM7) of the same subunit (Fig. 1c).This arrangement is in striking contrast to VGICs where the VSDs form intimate contacts with the ion translocating pore domain and the shared interfaces profoundly impact voltage-dependent channel opening 24 .The lack of any direct contact between the VSD and the ED in sp9C1 indicates that the allosteric mechanism underlying its voltage-regulation is different from that in VGICs.It is noteworthy that while the local resolution of our density map (FSC = 0.5) reaches ~ 3.2Å in the core of the protein, it drops to 4-6Å in the peripheral regions, including the VSD (Extended Fig. 1e).This is probably due to its lose packing with the protein core which leads to its higher structural exibility.Nevertheless we were able to reliably trace the backbone of most of the VSD helices (Extended Fig. 3a, b).The S4 helix of sp9C1-VSD, which is critical for its voltage-sensitivity, has six regularly spaced positively charged residues.The intracellular end of S4 forms a 9-residue-long 3 10 -helix harboring three positive charges (R809, R812, K815) which are highly conserved in different sp9C1 orthologs (Extended Fig. 3c).The remaining three positive charges (K800, R803 and R806) are arranged on the extracellular, α-helical end of S4.These three residues are relatively less conserved, particularly in the mammalian variants of SLC9C1, which could contribute to functional divergence in their voltage-sensitivities.The limited resolution of the VSD precludes unambiguous deduction of its conformational state.However, given the strong hyperpolarizing voltages necessary to drive the sp9C1-VSD into its resting (or down) state 19 , it is likely that in our structure (at 0mV) it is in the activated (or up) conformation.
In each sp9C1 subunit, the CNBD hangs ~ 30Å below the inner lea et of the membrane.It has an evolutionarily conserved fold comprising an 8-stranded β-jelly roll (βR1) anked on the N-terminus by the αA helix and on the C-terminus by αB and αC helices.Unexpectedly, the CNBD is C-terminally linked to a second 8-stranded β-jelly roll (βR2) via an α-helical linker (αD helix) (Extended Fig. 3d).Extensive hydrophobic contacts and a possible cation-π interaction, mediated by conserved residues (W1091 and R1185), are observed at the interface between βR2 and αC helix (Extended Fig. 3d,e).The βR2s of the two subunits are structurally apposed against each other, and together with the CNBDs, form a "cytoplasmic cuff".CNBDs of cyclic nucleotide modulated ion channels do not feature a comparable βR2 domain [25][26][27] .
While the speci c functional role of βR2 is unclear, we will presently discuss a possible role of the βR2 in guiding cyclic nucleotide dependent conformational changes in sp9C1.
A distinct scaffolding domain, which we refer to as the 'allosteric ring domain' (ARD), directs the three dimensional arrangement of the ED, VSD and CNBD in the context of the sp9C1 dimers (Fig. 1d).The ARD lls the gap between the cytoplasmic surface of the transmembrane domains and the CNBDs.It is formed by the association of the polypeptide segments connecting the ED and VSD (the Exchanger-Voltage-sensor Linker or EVL) and the VSD and CNBD (the Voltage-sensor-CNBD linker or VCL).Each of the linkers are shaped like two paddles, annotated as PN < X> or PC < X> (where N and C designate the Nterminal or C-terminal paddle and X is either EVL or VCL) connected by a exible hinge.Inter-subunit interactions between PN EVL and PC VCL and intra-subunit contacts between PC EVL and PN VCL drive the assembly of the ARD.In addition, the ARD interacts intimately with the βR1 of CNBDs -PN EVL is nestled in a groove formed between PC VCL and βR1 of the neighboring subunit while PN VCL rests atop of the βR1 of the same subunit (Fig. 1d and Supplementary Video 1).The overall arrangement of the cytoplasmic domains and ARD of sp9C1 is similar to the SOS1 transporter 17 , which is a plant ortholog of SLC9s, although the latter does not feature a VSD and is not regulated by cAMP.
The characteristic arrangement of the ARD suggests that it may play an important role in the assembly of the sp9C1 dimers.To test this, we compared the stabilities of full length (FL) sp9C1 with 3 truncation mutants (named 950, 657 and 494 referring to the amino acid positions of the truncation sites) using Fluorescence Size Exclusion Chromatography (FSEC) 28,29 .All 4 proteins, expressed as eGFP fusion constructs and a nity puri ed in digitonin micelles, were largely monodisperse with retention volumes consistent with a dimeric assembly (Fig. 1e).However, when exchanged into mildly destabilizing conditions, a second peak, corresponding to a monomer, was observed.Constructs truncated at 657 and 494 were drastically more unstable than FL-sp9C1 as re ected by a much faster and greater extent of disassembly over a 10hr period.In both these constructs the ARD is partly or entirely deleted.Thus the intra-and inter-subunit interactions at the level of the ARD are critical for stability of the sp9C1 dimers.
The 950 construct, in which the CNBD, βR2 and the C-terminal tail was deleted, also exhibited elevated breakdown relative to the full-length protein.Thus the closed cuff arrangement of the cytoplasmic domains also contributes to dimeric stability, either via direct inter-subunit interactions between the βR2 domains or by facilitating a stable arrangement of the ARD via the ARD-CNBD interfaces.

Coupling of pH dependent dynamics of the ED and the ARD
The ARD structurally interweaves the three key dynamic elements of sp9C1 through covalent and noncovalent interactions.This raises a compelling hypothesis that the ARD energetically couples their intrinsic rearrangements and might itself exhibit non-trivial relaxations in response to conformational changes in the ED, VSD and CNBD.To explore this possibility we pursued single particle reconstructions of sp9C1 under acidic conditions (pH 6).We reasoned that since protons are co-substrates of the ED, by elevating its concentration, we could ostensibly drive the ED into a different conformation and visualize how the remainder of the protein structurally adapts to it.
Our nal maps of sp9C1 at pH6, with K + or Na + as the primary cation, reached GS-FSC resolutions of 3.7 Å and 4.2 Å respectively (Extended Fig. 4a, b).However the density for the VSDs worsened relative to the pH 8 reconstructions.We speculate that protonation of titratable residues in the VSD results in their increased structural heterogeneity, probably similar to VSD-mediated proton dependent inhibition of many VGICs [30][31][32] .Beyond this, very modest changes were observed in the remainder of the protein, such as 2-3Å displacements in the βR2 (Extended Fig. 4c-f).While it is possible that in digitonin micelles the ED is conformationally arrested, it is noteworthy that reconstructions of SLC9A1/CHP1 in lipid nanodiscs also show limited pH dependent rearrangements in the ED 16 , raising the possibility that the inward open state of the ED might be intrinsically favored in different SLC9s, particular under in vitro conditions.An alternate possibility is that the pKa-s of the titratable residues in the ED, which underlie its proton dependent regulation, is signi cantly lower than pH 6.
Constant pH MD simulations of a prokaryotic electrogenic ortholog of SLC9s have suggested that a highly conserved positively charged residue (equivalent to R399 in sp9C1) plays a critical role in ion exchange, at least in part, by in uencing the pKa of neighboring Asp/Glu residues which constitute the principle Na + binding site (Extended Fig. 5c) 33,34 .Applying the same logic to sp9C1, we reasoned that neutralization of R399 might increase the pKa of nearby acidic residues (E233, D209, D238) and thereby facilitate capturing proton dependent conformational changes in the protein.Accordingly, we imaged the R399A mutant of sp9C1 at pH 8 and 6, with Na + as the dominant cation.At pH8, reconstruction of the R399A mutant identi ed a single structural class with an overall GS-FSC resolution of ~ 3.1Å (Extended Fig. 5a).However at pH 6 our reconstruction analyses revealed two distinct structural classes at GS-FSC resolutions of 3.5 Å and 3.1 Å which we refer to as the Relaxed (R) and Compressed (C) forms respectively (Extended Fig. 5b).While the R399A reconstructions recapitulated all key architectural features observed with WT sp9C1 (Extended Fig. 5d), they exhibit clear pH dependent rearrangements (Fig. 2a, Supplementary Video 2).Alignment of the pH8 and pH6 structures of the R399A revealed that upon acidi cation, in both the R and C classes, the TM5b moved upward and laterally by ~ 2Å each (Fig. 2b).This movement causes the side-chain of P210 to be inserted into the substrate binding site, possibly altering protein-substrate interaction, although the sodium coordinating residues remain largely static (Fig. 2b).The last 2-3 turns of TM13 also undergoes a ~ 12° change in tilt (Fig. 2b).The resultant RMSDs of these two regions of the protein between pH 8 and 6, is considerably larger with the R399A mutant than with WT sp9C1 (with K + or Na + as the dominant cation) (Extended Fig. 5e).At the level of the ARD however, dramatic changes were observed in Class C but not in Class R (Fig. 2c).In Class C, there is a clear displacement of the 6-residue long linker between TM13 and PN EVL which culminates in an ~ 8°i nward rotation of the PN EVL .PC VCL of the opposite subunit, tightly packed against the PN EVL , moves along with it as a rigid body (Extended Fig. 5f).As a result of this rotation, the PC VCL of the two subunits which are 15.9Å apart (at level of residues H914 and Q917) in pH 8, are only 4.3Å apart in the pH 6 Compressed Class.These structural observations support our hypothesis that the ARD undergoes coupled rearrangements together with the ED.The ARD-βR1 interfaces however remain largely preserved between the pH8 and the pH6-compressed form (Extended Fig. 5g), indicating that the compact cytoplasmic-cuff moves with the ARD pointing towards the stability of the ARD-βR1 interfaces.
Functional measurements of sp9C1 indicate that the R399A mutation causes a dramatic decrease in ion turnover rates 19 .R399 is engaged with E233 via a salt-bridge interaction.This interaction likely couples the core helices of the ED as it toggles between the inward-open and the outward open states as observed in the NHE1-CHP1 complex 16 .The loss of function in R399A might arise from stabilization of an intermediate state of the transporter.Without the salt-bridge, the core helices get trapped in this state and are unable to isomerize into the outward open state.In addition, the R399A mutation in sp9C1 has also been shown to perturb VSD dynamics, causing the gating-charge displacement vs voltage curve to shift rightward along the voltage-axis and become shallower 19 .The lack of clear density for the VSD at pH 6 (much like what we observed with WT sp9C1) does not allow us structurally de ne such an intermediate state of the VSD.However, the coupled rearrangement of the ARD with the ED enables us to propose that the allosteric communication between the VSD and ED is mediated, at least in part, by the ARD.
cAMP dependent re-arrangements in sp9C1 cAMP exerts two main effects on sp9C1 function.It reduces the extent of hyperpolarization required to activate the ion-exchange activity and stabilizes the resting or down state of the VSD 19 .We investigated the structural mechanisms underlying cAMP dependent modulation of sp9C1 via single-particle reconstructions of sp9C1 in presence of cAMP.Initial reconstruction analyses indicated that, in presence of cAMP, sp9C1 was structurally heterogenous and the heterogeneity was worse at pH 8 than at pH 6.Thus we pursued our reconstruction efforts of sp9C1, in presence of cAMP, only at pH 6.We were able to identify three distinct, almost equally populated, structural classes in the presence of cAMP (Extended Fig. 6a).The classes, which we call G (Grip), GnT (Grip and Twist) and GnTL (Grip and Twist Like) (Fig. 3a and Extended Fig. 6b) were resolved at GS-FSC resolutions of 4.1Å, 3.9Å and 4.2Å respectively.Much like our pH 6 reconstructions in cAMP free conditions, densities for the VSDs were not clear in the G and GnTL classes.In the GnT class, several intracellular helical turns of the four VSD helices could be resolved, which de ned the position of the VSD within the plane of the membrane.The EDs of all three structural classes was practically identical to that observed in cAMP free state (Extended Fig. 6c).The cytoplasmic domains however exhibited large rearrangements.
When compared to the cAMP free structures, we nd that in class G, the cytoplasmic cuff retains its compact/closed form but moves downward by about 9Å (Fig. 3a).In the GnT and GnTL classes however, the CNBD and βR2 domains of each subunit twist outward by ~ 58°, opening the cytoplasmic cuff (Fig. 3a, b).As a result of this movement, the separation between the βR2 domains in the GnT and GnTL conformations increases dramatically by about ~ 60Å, relative to the Apo and G conformations.For the GnT class, the density map was of su cient quality to identify a cAMP molecule, bound in an anticon guration (Extended Fig. 6d), within the CNBD.cAMP binding causes the βR1 and αC helix of the CNBD to move in closer to each other by ~ 5Å.A similar conformational change in also observed in the CNBDs of classes G and GnTL (Extended Fig. 6c), although the cAMP density is relatively less clear.Thus, G, GnT and GnTL all represent distinct cAMP bound quaternary conformations of sp9C1.
Structural alignment of the cytosolic domains (CNBD and βR2) of the cAMP bound conformations with that of the apo form shows that the interface between the αC and βR2 remains relatively invariant during ligand binding (Extended Fig. 6e).Thus it is more likely that cAMP is gripped by the CNBD as a result of the movement of βR1 towards αC, as opposed to a movement of the αC-βR2 unit towards βR1.The movement of αC with respect to βR2 is possibly restricted by the hydrophobic interface between them.Thus βR2 might be acting as a structural rudder guiding the speci c rearrangement of the CNBD.As βR1 moves inward to grip cAMP, the intra-subunit interface between βR1 and ARD interface is ruptured.This causes the cytoplasmic domain of each subunit to become relatively untethered to the rest of the protein and is free to move, which leads to the large conformational changes observed the GnT and GnTL states.
cAMP binding induced structural changes in the cytoplasmic cuff and rearranges the ARD.In the GnT conformation, the average plane of the ARD is displaced downward by ~ 2Å with respect to the Apo state (Fig. 3f).Interestingly the PC EVL and PN VCL paddles rotate counter-clockwise about the membrane normal by ~ 7°.The latter movement is particularly signi cant, since it leads to a robust, 9Å lateral displacement of the VSD helices within the plane of the membrane (Fig. 3f, Supplementary Video 3).
Since the closed cuff stabilizes the sp9C1 dimers, we gauged the effect of cAMP binding on the biochemical stability of sp9C1 dimers using a thermal disassembly assay.Puri ed sp9C1-eGFP was heated to different temperatures in absence or presence of cAMP (at different concentrations) and the monodispersity of the sample was analyzed using FSEC.A distinct lower molecular weight peak, likely corresponding to a disassembled monomer, became progressively more dominant with increasing temperature, saturating by ~ 46°C.The ratio of the two species of distinct sizes (ρ) was plotted against the temperature to obtain a temperature-dependent disassembly curve (TDC) at each cAMP concentration.With increasing concentrations of cAMP, the saturating level of ρ (ρ sat ) increased and T 1/2 of the TDC decreased (Fig. 3d).Between cAMP-free and 5mM cAMP, ρ sat increases from ~ 0.9 to 2.4 and T 1/2 reduces from 38°C to 28°C, indicating a large loss of dimer stability upon cAMP binding.A doseresponse relationship for cAMP, based on ρ sat or T 1/2 , indicates that cAMP interacts with puri ed sp9C1 with an EC 50 of ~ 12-15uM.
To determine the effect of CNBD-ARD interface on cAMP induced loss of protein stability, we mutated two conserved residues localized at this interface and determined their TDCs in the absence and presence (5mM) of cAMP.Y1035C and K880A destabilize the sp9C1 dimers in cAMP free-conditions, as re ected by a signi cantly lower T 1/2 with respect to the WT.Both mutations also robustly decrease cAMP dependent effects (Fig. 3e) -Δρ sat (the change in ρ sat between cAMP-free and 5mM cAMP conditions) for the R880A and Y1035C mutants are ~ 70% lower than that for the WT sp9C1 and there is a pronounced decrease in the cAMP dependent shift (ΔT 1/2 ) of the TDCs.A likely interpretation of these observations is that Y1035C and K880A destabilizes the intra-subunit interface between the ARD (PN VCL ) and βR1, thereby perturbing the compact closed-cuff arrangement of the cytoplasmic domains in the cAMP-free state.As a result, the cAMP dependent loss of stability caused by the opening of the cytoplasmic cuff is diminished in the mutants.Overall, the results of these thermostability experiments are consistent with our structural observations and points to an important role for the CNBD-ARD interface in governing the stability of the compact form of the cytoplasmic cuff, which in turn affects the cAMP dependent dynamics of sp9C1.

Mechanism of electrochemical linkage in SLC9C1 and conclusion
Our reconstructions of sp9C1 in multiple conformations enables us to propose a biophysical basis underlying cAMP regulation of sp9C1 function (Fig. 4).We hypothesize that in the absence of cAMP and at depolarizing voltages, sp9C1 is in an auto-inhibited state.The inhibition of the ED arises from the speci c arrangement of the ARD, which is in turn spatially constrained by the activated VSD (due to the rigid link between the S4 and PN VCL ) and the apo CNBD (via the intra-and inter-subunit interfaces).Upon binding cAMP, the cytoplasmic cuff opens and relieves the structural constraint imposed by the cytoplasmic domains on the ARD.This enables the ARD to rearrange more easily as the voltage-sensor deactivates (S4 moves down), thereby decreasing the hyperpolarizing voltages required to drive the voltage-sensor into its down state.Concurrently, it decreases the energy barrier for the ARD to undergo coupled rearrangements with the ED, as it toggles between its inward-open and outward-open states, catalyzing ion exchange.Overall, cAMP induced potentiation of sp9C1 ion turnover rates and facilitation of voltage-sensor deactivation proceeds via relieving the transporter from an auto-inhibited state.
While further studies will be necessary to etch the energetic and structural landscape underlying SLC9C1 function it is noteworthy that our model of cAMP dependent facilitation of SLC9C1 activity bears striking similarities to that proposed for HCN channels 35,36 .In the case of the latter, the VSDs are at best weakly coupled to cAMP binding 37 and the scale of cAMP dependent rearrangements is small 27,38 as compared to what we observe here with SLC9C1.Yet it is remarkable how, despite architectural and dynamic differences, the underlying biophysical themes underlying cAMP regulation are convergent between the HCN channel and SLC9C1, arguably an HCN transporter.

Expression and puri cation
The cDNA fragment encoding full-length Strongylocentrotus purpuratus SLC9C1 was synthesized (Genscript Inc.) and cloned into a modi ed pEG BacMam vector 39 , with a PreScission protease cleavage site followed by a C-terminal mEGFP-Twinstrep tag.All mutants of sp9C1 (R399A and the constructs truncated at amino acid positions 494, 657 and 950) were generated in the background of this parent construct using standard molecular biology techniques (Genscript Inc.).For protein expression, HEK293F cells, cultured in suspension in Freestyle 293 media (supplemented with 2% Heat Inactivated FBS), were transfected with plasmid DNA, isolated from large volumes of bacterial cultures using Endotoxin free Plasmid Puri cation kits (Qiagen).Linear PEI (25kDa) was used for transient transfections at a ratio of 1:3 (plasmid:PEI mass ratio).Post-transfection, cells were grown for 12-14 hrs at 37°C, following which sodium butyrate was added to the transfected cells to a nal concentration of 10mM and the cultures were transferred to 30°C and grown for another 48-54 hrs.Cells were collected by centrifugation at 3,000g for 30mins and washed with PBS.Pelleted cells were ash frozen in liquid nitrogen and stored at -80°C until use.
For puri cation, frozen cell pellets were resuspended in chilled lysis buffer (300mM NaCl, 50mM Tris, 10mM DTT, 20% glycerol, 1% digitonin, pH 8), brie y sonicated on ice and gently agitated at 4°C for 1-1.5 hrs.The whole-cell extracts were spun at 100,000g for 1 hr and the supernatant was incubated anti-GFPnanobody resin (generated by PCF, University of Iowa) for 8-10 hrs.Protein bound resin was washed 4 times in batch mode, each time with 5 resin volumes of wash buffer (300mM NaCl, 50mM Tris, 10mM DTT, 10% glycerol, 0.1% digitonin, pH 8).After the last wash, the resin was resuspended in 2x resin volume of wash buffer and incubated with Precision protease (ThermoScienti c) for 12-14hrs.The protein released by protease cleavage was concentrated to ~ 500ul using 100kDa centrifugation lters and the resultant protein was further puri ed by gel ltration chromatography.The gel ltration buffer used was modi ed according to desired condition for single-particle imaging.For pH 8 samples, the buffer was 300mM NaCl, 25mM HEPES, 1mM TCEP, 0.05% digitonin, buffered to pH 8 using NaOH and for pH 6 samples, the buffer was 300mM NaCl, 25mM MES, 1mM TCEP, 0.05% digitonin, buffered to pH 6 using NaOH.For conditions where K + was used instead of Na + , NaCl was replaced with KCl and KOH was used to bring the nal pH of the solution to the desired value.1-1.25ml of the peak fractions of the protein was collected and concentrated to ~ 2.5-3.5mg/ml for preparing cryoEM grids.
For all thermostability experiments, the constructs were similarly expressed in HEK293F cells.After extraction in digitonin buffer, the protein was puri ed using Streptactin a nity resin (IBA Life Sciences).
Resin bound protein was washed as before and eluted using wash buffer, supplemented with 10mM Desthiobiotin.Eluted protein was further puri ed by gel ltration chromatography using the SEC buffer: 300mM NaCl, 50mM Tris, 0.05% digitonin, buffered to pH 8. Peak protein fractions were combined and concentrated to ~ 1.5mg/ml, ash frozen in liquid nitrogen (in 5µl aliquots) and stored in -80°C until use.All steps of protein puri cation were performed at 4°C.

Thermostability assessment using Fluorescence Size Exclusion Chromatography
For all thermostability tests, the destabilizing buffer (TSB) was: 300mM NaCl, 50mM HEPES, 50mM MES, 1mM TCEP, 2mM DDM, 0.4mM CHS, buffered to pH 6.For the time-dependent disassembly experiments, a frozen aliquot of protein was thawed and diluted 200-fold (by volume) into the TSB and immediately injected (in under 1min) into the FSEC instrument (Shimadzu).Periodic injections (every 50min) of the sample were performed using the SIL40C autosampler and over the entire period the samples (diluted into TSB) were held at 4°C.For experiments probing cAMP dependent change in thermostability, thawed protein was diluted 200-fold into TSB, supplemented with an appropriate concentration of cAMP, incubated for 1min and heated to the desired temperature for 10min (on a PCR thermocycler) and the sample was subjected to FSEC analysis within 5mins of the heat step.In all cases, Superose 6 10/300 Column was used for separation of species and the FSEC buffer was: 300mM NaCl, 20mM Tris, 100µM GDN, pH 8.The column and buffers were at room temperature for all experiments.The chromatography pro le of the different species was monitored by RF20Axs detector with Ex./Em.wavelength settings of 488nm/510nm.Each chromatography pro le (between 10 and 16ml) was tted to a sum of 2 Gaussians: . For all ts, B 1 was constrained to be < B 2 implying that the rst gaussian term accounts for the higher molecular weight (dimer) species and the second term accounts for the disassembled monomer.The relative population of the 2 species was calculated as: In all cases, experiments were performed in duplicates with different batches of puri ed protein.All error bars are the standard deviation of measurements.
CryoEM grid preparation and imaging For all samples imaged under cAMP free conditions, 3.5ul of concentrated puri ed protein was applied to glow-discharged copper holey carbon grids.For the cAMP dataset, SEC puri ed protein was rst concentrated to ~ 1.5mg/ml, diluted 100-fold (volumetrically) into SEC buffer, supplemented with 2.5mM cAMP and incubated for 2hrs on ice.The protein was subsequently concentrated to ~ 3mg/ml and an additional amount of cAMP was spiked into it so that the nal protein and cAMP concentrations were ~ 2.5mg/ml and 5mM respectively.The sample was incubated overnight on ice before preparing cryoEM grids.R 1.2/1.3grids were used for all datasets, except for the pH 6/KCl dataset where R 2/2 grids were used.In all cases, grids were blotted for 5.5s at 100% humidity with a blot force of 0 and then plunge frozen in liquid ethane using a Vitrobot Mark IV (Thermo sher Scienti c).All data were collected on a Titan Krios (electron microscope operating at an accelerating voltage of 300kV), equipped with a K3 Detector (Gatan).Images were recorded with EPU software (ThermoFisher Scienti c) in super-resolution mode with a pixel size of 0.54 Å or 0.69 Å, and a nominal defocus of -0.9 to -1.9µm.For all datasets, a total dose of 65 electrons per Å 2 was used, which was fractionated over 40 frames.

Image Processing and Map Calculation
Image processing and map calculations were performed using CryoSPARC 40 and Relion 41 .The recorded movies were motion corrected (Patch motion correction) and then subjected to contrast transfer function (CTF) estimation (Patch CTF).Micrographs with CTF resolution > 6Å or total pixel drift > 60 pixels were discarded.Blob picker was used to pick particles from 500-1000 images, which after several rounds of 2D class averaging, yielded reasonable 2D projections.These 2D classes were used for template based particle picking from curated movies.Furthermore, these (relatively few) particles were also subjected to Ab initio reconstruction (1 class) and subsequent homogenous re nement to generate a representative initial 3D map of the protein.The template picked particles were subject to several rounds of heterogenous re nements using the initial 3D map as reference.At each step of heterogenous re nement, the particles which get classi ed to low resolution classes were discarded.The set of particles thus retained were re ned using Non-uniform 42 and local re nement techniques on CryoSPARC (with C2 symmetry imposed), exported to Relion and further classi ed using 3D classi cation, without image alignment.The optimal set of particles were imported back into CryoSPARC and subject to CTF re nements 43 to obtain the nal maps.For processing of all datasets, except for R399A, particle stacks were binned by 2.

Model Building and Re nement
An initial model of sp9C1 was generated using AlphaFold 44 using its Colab platform.This initial model was divided into segments which were tted into density maps and subsequently modi ed on COOT 45 .To generate the nal models, several iterations of manual model building and real space re nement against full maps were performed in Phenix 46 .NAMDINATOR 47 was used improve model building, particularly in instances where there are large domain level rearrangements.In 3 of the 10 cases, (namely 8K, 8Na and GnT) where the VSD densities are featured, the initial model building and re nement for the VSDs were done with the unsharpened map (because sharpening causes fragmentation of the densities in these regions) while the remainder of the model was generated using the sharpened maps.In all instances, the nal map-to-model validation were performed against the sharpened maps.The nal re ned atomic models were validated using MolProbity 48 .All structural analyses were performed on UCSF Chimera 49 or MATLAB (Mathworks).Structural gures were generated using UCSF ChimeraX or Pymol.

Figure 1 Architecture
Figure 1

Figure 3 cAMP
Figure 3