Cryo-EM structure of the potassium-chloride cotransporter KCC4 in lipid nanodiscs

Cation-chloride-cotransporters (CCCs) catalyze transport of Cl- with K+ and/or Na+across cellular membranes. CCCs play roles in cellular volume regulation, neural development and function, audition, regulation of blood pressure, and renal function. CCCs are targets of clinically important drugs including loop diuretics and their disruption has been implicated in pathophysiology including epilepsy, hearing loss, and the genetic disorders Andermann, Gitelman, and Bartter syndromes. Here we present the structure of a CCC, the Mus musculus K+-Cl- cotransporter (KCC) KCC4, in lipid nanodiscs determined by cryo-EM. The structure, captured in an inside-open conformation, reveals the architecture of KCCs including an extracellular domain poised to regulate transport activity through an outer gate. We identify binding sites for substrate K+ and Cl- ions, demonstrate the importance of key coordinating residues for transporter activity, and provide a structural explanation for varied substrate specificity and ion transport ratio among CCCs. These results provide mechanistic insight into the function and regulation of a physiologically important transporter family.

KCCs are important for K + and Clhomeostasis, including in establishing low neuronal cytoplasmic Clconcentrations critical for inhibitory neurotransmission, and in volume regulation in many cell types Bergeron et al., 2003;Mount et al., 1999;Karadsheh et al., 2004). Among KCCs, KCC4 is most strongly activated by cell swelling and high internal [Cl -] and is uniquely active in acidic external environments Bergeron et al., 2003). KCC4 is expressed in tissues including the heart, nervous system, kidney, and inner ear and mice lacking KCC4 display progressive deafness and renal tubular acidosis Mount et al., 1999;Karadsheh et al., 2004;Boettger et al., 2002). Hearing loss in these animals is due to disrupted K + recycling by Dieter's cells in the cochlea and hair cell excitotoxicity, while renal tubular

Structure of KCC4 in lipid nanodiscs
Mus musculus KCC4 was heterologously expressed in Spodoptera frugiperda (Sf9) insect cells for purification and structure determination (Figure 2-figure supplement 2). To assess the activity of KCC4 in these cells, we utilized an assay that depends on the ability of KCCs to transport Tl + in addition to K + (Zhang et al., 2010). In cells loaded with the Tl + -sensitive fluorophore FluxOR red, Tl + uptake from the extracellular solution results in an increase in fluorescence signal ( Figure 1A). Cells infected with virus encoding KCC4, but not cells infected with a virus encoding an anion-selective volume-regulated ion channel SWELL1 (Kern et al., 2019) or uninfected Sf9 cells, displayed increased fluorescence over time consistent with KCC4 activity (Figure 1B,C). No significant difference in activity was observed between N-and C-terminally GFP-tagged mouse KCC4 ( Figure 1B,C), in contrast to a previous report for KCC2 (Agez et al., 2017), and C-terminally tagged KCC4 was used for subsequent study.
We reconstituted KCC4 into lipid nanodiscs in order to study the structure of the transporter in a native-like membrane environment. KCC4 was extracted, purified in detergent, and exchanged into nanodiscs formed by the membrane scaffold protein MSP1D1 and a mixture of phospholipids that approximates the composition of major species in neuronal membranes (2:1:1 molar ratio DOPE: POPC:POPS (2-dioleoyl-sn-glycero-3-phosphoethanolamine:1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine:1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine)) (  Ritchie et al., 2009). KCC4-MSP1D1 particles are similar in size and shape to KCC4 particles in detergent micelles by cryo-EM, but show improved distribution in thin ice which enabled reconstruction to high resolution ( Figure 2-figure supplement 3). Figure 2A contoured to highlight the position of the lipid belt surrounding the transmembrane region. To achieve the highest resolution reconstruction, the nanodisc density was subtracted and particles were subjected to focused classification and subsequent refinement ( Figure 2-figure supplement 4). The resulting map, at 3.65 Å overall resolution, enabled complete de novo modeling of the transmembrane and extracellular region of KCC4 and includes two partial extracellular glycosylation sites, a bound K + ion, and a bound Clion ( Figure  Overall architecture KCC4 is monomeric in the nanodisc structure. Density for the N-terminal region and C-terminal domain (CTD), which together comprise approximately half the expressed protein mass, is not observed in the cryo-EM maps ( Figure 2). The N-terminal region is weakly conserved, variable in length among CCCs (Figure 2-figure supplement 1), and is likewise unresolved in the structures of NKCC1 or KCC1 (Chew et al., 2019;Liu et al., 2019). We presume it is highly flexible in KCC4. The C-terminal domain, while similarly unresolved in KCC1 (Liu et al., 2019), is well conserved, has documented roles in regulation, expression, and trafficking (Arroyo et al., 2013;Marcoux et al., 2017;Payne, 2012;Hartmann and Nothwang, 2014;Rinehart et al., 2009), and mediates homodimerization of NKCC1 and the Archaean CCC (MaCCC) (Chew et al., 2019;Warmuth et al., 2009). We found no evidence of proteolytic cleavage of either region. Mass spectrometry of purified KCC4 showed high coverage (47%) and abundance (81% of all KCC4 peptides) for the CTD (Figure 2-figure supplement 7). We observe a progressive loss of detailed features and decrease in local resolution in TM11 and TM12 that connect the CTD to the core transmembrane region (Figure 2-figure supplements 5B and 6). Some two-dimensional class averages show a blurred cytoplasmic feature in the position we expect the CTD to emerge (Figure 2-figure supplement 3B,D), but attempts to classify distinct conformations of this feature were unsuccessful. We conclude that the monomeric structure reported here corresponds to full-length mouse KCC4 with flexible and/or disordered terminal regions.

An unmasked reconstruction of KCC4 in nanodiscs is shown in
The monomeric structure of KCC4 contrasts with recent homodimeric structures of Danio rerio NKCC1 (Chew et al., 2019) and H. sapiens KCC1 (Liu et al., 2019), although dimerization of NKCC1 and KCC1 involve completely distinct interfaces ( Figure 3A,C). Disruption of putative tightly associated KCC4 homodimers during purification or sample preparation was excluded for the following reasons: (i) The portion of KCC4 in an early-eluting broad peak from a sizing column (           Figure 3A,B). Notably, this loop is poorly conserved in KCCs ( Figure 3E). In KCC4, the loop is incompatible with forming a dimer interface without substantial rearrangement due to steric clashes ( Figure 3B). A triple mutation designed to disrupt interaction between extracellular loops in KCC4 (S404E, R405E, G407E) has no effect on transport activity ( Figure 3B,F). This mutation (and those described later) did not substantially alter KCC4 folding or expression (Figure 3-figure supplement 1). We conclude dimerization as observed in the KCC1 structure (Liu et al., 2019) is not functionally relevant for KCC4. In NKCC1, dimerization is mediated predominantly through extensive protein-protein interactions in the CTD (Chew et al., 2019). These regions appear well conserved in KCCs ( Figure 3E, Figure 2figure supplement 1). Three pairs of mutations designed to disrupt CTD-CTD interactions in KCC4 (L678R, V680R; L670R, A673R; and S780R, L783R) resulted in a similar, but incomplete, reduction in KCC4 activity (by an average of 36, 41, and 44%, respectively) ( Figure 3C,D,F). This reduction is comparable to that observed in a truncated KCC4 construct missing the entire C-terminal region (KCC4DC, which includes amino acids 1-658) ( Figure 3F). These results suggest that monomeric KCC4 is active and that dimerization through the CTDs in a manner analogous to NKCC1 increases transport activity.

Transporter conformation
KCC4 adopts an inward-open conformation. The outer surface of the transporter is sealed from the extracellular solution, while a continuous cavity extends from the center of the transmembrane region to the cytoplasmic side ( Figure 4). The transmembrane region consists of twelve helices (TM1-TM12) with TM1-TM5 related to TM6-TM10 through an inverted repeat. TM2 and TM11 in KCC4 are linked by a membrane buried disulfide bond between amino acids C163 (TM2) and C626 (TM11) conserved between KCCs, but not other CCCs (Figure 2-figure supplement 1).
The position of the extracellular domain suggests its involvement in conformational changes during the KCC transport cycle. A segment of the extracellular domain close to the membrane forms a constriction that seals the internal vestibule from the extracellular solution. This is likely the extracellular gate based on comparison to other APC transporters ( Figure 4A,B; Shi, 2013). In KCC4, residues in EL4, TM1, TM3, and TM10 form an electrostatic and hydrophobic interaction network that seals the gate ( Figure 4A,B). R140 on TM1B extends towards the extracellular solution to interact with D575 on TM10 and E222 on TM3. The extracellular domain is positioned immediately above through an interaction between K485 on EL4 and D575. The outer portion of TM1B contributes W143 which, together with F486, surrounds K485 as it projects towards TM10. This is reminiscent of the extracellular gate in LeuT formed by an electrostatic interaction between TM1 and TM10 (R30 and D404) and capped by EL4 through an interaction with TM10 (D401 and A319) (Krishnamurthy and Gouaux, 2012).
We generated mutations at sites in this interaction network to assess its importance for KCC4 function. Disruption of the interaction from the extracellular domain (K485D, K485E, and F486A) or the TM region (R140E) significantly reduced KCC4 transport activity ( Figure 4C). The functional effects of these mutations are likely due to specific disruption of the interaction between the TM region and extracellular domain because a more subtle change (K485A) or mutation of a nearby, but less conformationally restricted, residue (E222R) had no effect on activity ( Figure 4C). By analogy to LeuT and other APC transporters, these data suggest opening of the KCC4 extracellular gate likely requires 'unzipping' of the electrostatic network and rotation of EL4 and the extracellular domain away from the surface of the TM region (Krishnamurthy and Gouaux, 2012;Penmatsa and Gouaux, 2014).
On the intracellular side of KCC4, a hydrophilic cavity is formed by TM1, TM3, TM6, and TM8 that exposes the inside of the transporter to the cytoplasm ( Figure 3C-E). At the top of this cavity are Cland K + binding sites. The cavity forms a bifurcated pathway for ion access to these sites, splitting into two routes approximately halfway through the tunnel due to the position of side chains of N439, R440, and R528. Both sides are open to an essentially equivalent degree ( Figure 4D). The only constriction outside of the local area surrounding the ions is formed at the position of N439 (from TM6B), G525, and L529 (from TM8) where the cavity narrows to~3.6 Å in diameter, still sufficiently large for passage of K + and Clions. Within~3 Å of each ion, the cavity narrows such that it would require at least partial ion dehydration.
The cavity surface is markedly electropositive ( Figure 4E). From the intracellular solution up to the position of the Cl -, charged and polar side chains (from R440, R528, R535, N131, N274, N439, and N521), backbone amides (from IL1), and a helical dipole (from TM6B) contribute electropositive character. Since intracellular Clions are typically present at lower concentrations than K + ions, this may serve to favor accumulation of the less abundant substrate near its binding site within transporter. Above the Clsite and around the K + site, the accessible surface becomes electronegative and would favor cation binding. The extracellular surface of the transporter outside of the sealed gate is markedly electronegative. How this relates to mechanisms for ion binding and release in outward-open states awaits additional structural information.

Ion binding sites
The central discontinuities in TM1 and TM6 result in protein backbone carbonyls and amides not involved in regular hydrogen bonding that are utilized in other APC transporters for substrate binding (Shi, 2013). Around this region, we observe two prominent non-protein density features ( Figure 5A,D). Based on structural, functional, and comparative analyses described below, we model these sites as bound K + and Clions.
The stronger of the two densities between TM1, TM6, and TM3 is modeled as a K + ion ( Figure 5A). It is surrounded by electronegative groups contributed by backbone carbonyls (N131 and I132 in TM1 and P429 and T432 in TM6) and a tyrosine hydroxyl from Y216 in TM3. The distances between electronegative groups and the ion are consistent with K + binding (2.8-2.9 Å ). The electronegative helix dipoles created by TM1A and TM6A may additionally contribute to a favorable electrostatic environment for cation binding. The coordinating tyrosine is conserved in all CCC family members that transport K + . In NCC, the position corresponding to Y216 is substituted by a histidine, which likely explains its K + -independence ( Figure 2-figure supplement 1).
To validate the assignment of the K + and Clsites and test the importance of coordinating residues in transporter activity, we mutated Y216 and Y589 that contribute to the binding sites to A, F, and H and assessed transporter activity. Mutations at both sites resulted in a marked loss in transport activity in the Tl + -flux assay: K + -coordinating Y216 mutations Y216A, Y216F, Y216H and Cl -coordinating Y589 mutations Y589F and Y589H reduced activity (by an average of 45, 78, 81, 65% and 59%, respectively) ( Figure 5G). Altering the electrostatic character around the Clsite with mutations N131A and N131D resulted in comparable transport defects. We conclude the observed K + and Clsites are critical for KCC4 activity.
How does substrate binding in the 1:1 K + :Clcotransporter KCC4 differ from the 1:1:2 Na + : K + :Clcotransporter NKCC1 (Warmuth et al., 2009;Wahlgren et al., 2018)? The KCC4 K + and Clsites correspond closely to sites for the same ions in NKCC1 ( Figure 5B,E). However, the proposed Na + site in NKCC1 is dramatically reorganized in KCC4 ( Figure 5C). In KCC4, TM8 is rotated farther away from TM1 and two consecutive Na + -coordinating serines in NKCC1 (conserved in all Na + -transporting CCCs) are substituted by glycine and alanine in KCC4 (and in KCC1-3) (Figure 2-figure supplement 1). The consequence is a loss of three of the five Na +coordinating positions, providing a structural explanation for Na + -independence in KCCs.
A second Clsite in NKCCs extracellular to the K + site is structurally conserved in KCC4 ( Figure 5F), but we do not observe evidence for ion occupancy at this site in the structure. Interestingly, this site is occupied in a recent structure of KCC1 in addition to the K + and Clsites observed in KCC4 (Liu et al., 2019). Mutagenesis of a nearby tyrosine Y466A and Y466H reduce activity of KCC4, suggesting that the integrity of both sites is important for transport ( Figure 5G). Understanding how stoichiometric transport of 1 Clwith 1 K + in KCC4 is accomplished will require structures in additional functional states.

Discussion
Our structural and mutational data are consistent with prior evidence for monomeric KCCs in cells in addition to homodimers and other heterooligomers (Blaesse et al., 2006;Puskarjov et al., 2012;Watanabe et al., 2009). Comparison of residues in KCC4 that correspond to those in the KCC1 homodimerization interface (Liu et al., 2019) suggest this interaction is not relevant for KCC4 function. However, comparison to the NKCC1 homodimerization interface (Chew et al., 2019) suggest KCC4 could similarly self-interact and mutational data suggests this interaction increases KCC4 activity. A recently posted report of KCC2 and KCC3 structures appears to show a homodimeric interaction through the CTDs very similar to that observed in NKCC1 (Chi, 2020). If monomeric and dimeric KCC4 are functionally distinct, regulated oligomerization in cells would be an opportunity to modulate transport activity. In addition to an increase in basal activity, dimerization involving a close juxtaposition of CTDs and transmembrane regions could enable regulation of transporter activity through CTD posttranslational modifications that may not be possible in monomeric KCC4 with a flexibly attached CTD (Hartmann and Nothwang, 2014;Blaesse et al., 2006;Puskarjov et al., 2012;Watanabe et al., 2009). In support of the in vivo relevance of this regulatory mode, a monomeric to dimeric transition in KCC2 has been correlated with an increase in transporter activity (Blaesse et al., 2006) and is influenced by phosphorylation of the CTD (Watanabe et al., 2009), while proteolytic cleavage of the KCC2 CTD correlates with a decrease in activity (Puskarjov et al., 2012). The ability of KCC4 to hetero-oligomerize with CCCs and other membrane proteins may be associated with weaker self-interaction compared to other CCCs (Simard et al., 2007;Mahadevan et al., 2014;Goutierre et al., 2019;Ivakine et al., 2013).
The structure of KCC4 provides insight into the architecture of KCCs and the mechanistic basis for coupled K + :Cltransport. KCC4 is observed in an inward-open conformation that exposes the inside of the transporter to the cytoplasm through a wide and electropositive tunnel that may serve to concentrate the less abundant intracellular substrate Cl -. This conformation is similar to that observed in NKCC1 and KCC1 and presumably represents the lowest energy state for CCCs in symmetrical salt concentrations and the absence of a transmembrane electrical gradient (Chew et al., 2019;Chi, 2020). We identify K + and Clions around central discontinuities in TM1 and TM6. A Na + site in the 1:1:2 Na + :K + :Clcotransporter NKCC1 is reorganized in KCC4 due to TM-TM displacement and loss of specific coordinating side chains, explaining Na + -independence in KCCs.

Cloning and protein expression
Cloning, expression, and purification were performed similarly to that described for LRRC8A (Kern et al., 2019). The sequence for KCC4 and KCC1 from Mus musculus were codon optimized for Spodoptera frugiperda and synthesized (Thermo Fisher, Cambridge, MA). Sequences were cloned into a custom vector based on the pACEBAC1 backbone (MultiBac; Geneva Biotech, Geneva, Switzerland) with an added C-terminal PreScission protease (PPX) cleavage site, linker sequence, superfolder GFP (sfGFP), and 7xHis tag, generating a construct for expression of mmKCC4-SNS-LEVLFQGP-SRGGSGAAAGSGSGS-sfGFP-GSS-7xHis. Mutations were introduced using standard PCR techniques with partially overlapping primers. MultiBac cells were used to generate bacmids according to manufacturer's instructions. Spodoptera frugiperda (Sf9) cells were cultured in ESF 921 medium (Expression Systems, Davis, CA) and P1 virus was generated from cells transfected with Escort IV Transfection Reagent (Sigma, Carlsbad, CA) according to manufacturer's instructions. P2 virus was generated by infecting cells at 2 Â 10 6 cells/mL with P1 virus at a MOI~0.1. Infection monitored by fluorescence of sfGFP-tagged protein and P2 virus was harvested at 72 hr post infection. P3 virus was generated in a similar manner to expand the viral stock. The P3 viral stock was then used to infect 1 L of Sf9 cells at 4 Â 10 6 cells/mL at a MOI~2-5. At 60 hr post-infection, cells were harvested by centrifugation at 2500 x g and frozen at À80˚C.

Transporter assay
The FluxOR-Red Potassium Ion Channel Assay (Thermo Fisher Scientific) was adapted for transport assays in Sf9 insect cells by adjusting the osmolarity of all buffers to 380 mOsm (by addition of sodium methylsulfonate). Cells were infected at a density of 1.5 Â 10 6 cells/ml and grown in suspension for 60-72 hr for robust KCC4-GFP expression. 100 uL of cells at 1 Â 10 6 cells/ml were plated and allowed to adhere for 1 hr before the assay. For experiments in Figures 3-5, 100 uL of cells at 1 Â 10 6 cells/ml were plated and allowed to adhere for 1 hr before viral infection. Cells were grown for 60-72 hr for robust transporter expression prior to assay. Growth media was replaced with 1X Loading Buffer and incubated at 27˚C away from light for 1 hr. The FluxOR Red reagent is a nonfluorescent indicator dye which is loaded into cells as a membrane-permeable acetoxymethyl (AM)ester. The non-fluorescent AM ester of the FluxOR Red reagent is cleaved by endogenous esterases into a fluorogenic Tl + -sensitive indicator. 1X Loading Buffer was subsequently removed and replaced with Dye-free Assay Buffer and FluxOR Background Suppressor. The assay was performed in 96-well, black-walled, clear-bottom plates (Costar). For data in Figure 1, fluorescence was measured on a Perkin-Elmer Envision Multilabel Plate Reader using bottom read fluorescence and a BODIPY TMR FP filter set (excitation 531 nm and 25 nm bandwidth, emission 595 nm and 60 nm bandwidth). For data in Figures 3-5, Fluorescence was measured on a Molecular Devices SpectraMax M3 using bottom read fluorescence and an excitation 560 nm with auto cutoff at 590 nm and emission 590 nm. The recordings were baseline corrected by subtracting the average fluorescence from 180 s prior to the addition of Basal Potassium Stimulus buffer and time zero is defined as the first data point recorded after the addition of stimulus. Global fits of all data to a one phase association model Y= (Plateau)*(1-e (-x/t) ) are displayed with 95% confidence interval bands ( Figure 1B). Alternatively, the final 50 counts were averaged as a measure of final fluorescence increase and normalized to wildtype KCC4 fluorescence increase from experiments performed on the same day. In some experiments, average GFP fluorescence was measured between FluxOR fluorescence measurements and used to normalize FluxOR fluorescence values to account for differences in expression level. This normalization did not change results in a significant way and so was not performed for the final data presented.

Protein purification
Cells from 1 L of culture (~7-12.5 mL of cell pellet) were thawed in 100 mL of Lysis Buffer (50 mM Tris, 150 mM KCl, 1 mM EDTA, pH 8.0). Protease inhibitors were added to the lysis buffer immediately before use (final concentrations: E64 (1 mM), Pepstatin A (1 mg/mL), Soy Trypsin Inhibitor (10 mg/mL), Benzimidine (1 mM), Aprotinin (1 mg/mL), Leupeptin (1 mg/mL), and PMSF (1 mM)). Benzonase (5 ml) was added after cells thawed. Cells were then lysed by sonication and centrifuged at 150,000 x g for 45 min. The supernatant was discarded, and residual nucleic acid was removed from the top of the membrane pellet by rinsing with DPBS. A 10%/2% and 10%/1% w/v solution of DDM/ CHS was clarified by bath sonication in 200 mM Tris pH 8.0 and subsequently added to buffers at the indicated final concentrations. Membrane pellets were transferred to a glass dounce homogenizer containing Extraction Buffer (50 mM Tris, 150 mM KCl, 1 mM EDTA, 1% w/v n-Dodecyl-b-D-Maltopyranoside (DDM, Anatrace, Maumee, OH), 0.2% w/v Cholesterol Hemisuccinate Tris Salt (CHS, Anatrace), pH 8.0). Membrane pellets were homogenized in Extraction Buffer and this mixture (100 mL final volume) was gently stirred at 4˚C for 1 hr. The extraction mixture was centrifuged at 33,000 x g for 45 min. The supernatant, containing solubilized KCC4-sfGFP, was bound to 5 mL of Sepharose resin coupled to anti-GFP nanobody for 1 hr at 4˚C. The resin was collected in a column and washed with 20 mL of Buffer 1 (20 mM Tris, 150 mM KCl, 1 mM EDTA, 0.025% DDM, 0.005% CHS, pH 8.0), 50 mL of Buffer 2 (20 mM Tris, 500 mM KCl, 1 mM EDTA, 0.025% DDM, 0.005% CHS, pH 8.0), and 20 mL of Buffer 1. Washed resin was resuspended in 6 mL of Buffer 1 with 0.5 mg of PPX and rocked gently in the capped column overnight. Cleaved KCC4 protein was eluted with an additional 25 mL of Buffer 1. The eluted pool was concentrated to~500 ml with an Amicon Ultra spin concentrator 100 kDa cutoff (MilliporeSigma, USA) and subjected to size exclusion chromatography using a Superose 6 Increase column (GE Healthcare, Chicago, IL) run in Buffer 3 (20 mM Tris pH 8.0, 150 mM KCl, 1 mM EDTA, 0.025% DDM, 0.0025% CHS) on a NGC system (Bio-Rad, Hercules, CA). Peak fractions containing KCC4 transporter were collected and concentrated.

Cross linking and mass spectrometry
Fractions corresponding to peaks 1 and 2 from size exclusion chromatography were separately pooled and concentrated to 0.5 mg/mL. Crosslinking was performed by adding 1 uL of glutaraldehyde from 10X stock solutions in water to 10 uL of KCC4 to achieve final glutaraldehyde concentrations of 0.02, 0.01, 0.005, 0.0025, and 0%. Samples were incubated for 30 min prior to quenching by addition of 1 uL 1M Tris-HCl and analysis by SDS-PAGE on 4-12% Tris-glycine gel (BioRad, USA). Deglycosylated samples were pretreated with 1:10 vol purified PNGase at 1 mg/mL (Addgene 114274) for 1 hr at 4˚C prior to the addition of glutaraldehyde.
For mass spectrometry, the band corresponding to purified KCC4 was excised from a 4-12% Trisglycine gel, digested with trypsin in situ, and the resulting peptides extracted and concentrated. Mass spectrometry was performed by the Vincent J. Coates Proteomics/Mass Spectrometry Laboratory at UC Berkeley. A nano LC column was packed in a 100 mm inner diameter glass capillary with an emitter tip. The column consisted of 10 cm of Polaris c18 5 mm packing material (Varian), followed by 4 cm of Partisphere 5 SCX (Whatman). The column was loaded by use of a pressure bomb and washed extensively with buffer A (5% acetonitrile/0.02% heptaflurobutyric acid (HBFA)). The column was then directly coupled to an electrospray ionization source mounted on a Thermo-Fisher LTQ XL linear ion trap mass spectrometer. An Agilent 1200 HPLC equipped with a split line so as to deliver a flow rate of 300 nl/min was used for chromatography. Peptides were eluted using a 4-step MudPIT procedure (Washburn et al., 2001). Buffer A was 5% acetonitrile/0.02% heptaflurobutyric acid (HBFA); buffer B was 80% acetonitrile/0.02% HBFA. Buffer C was 250 mM ammonium acetate/5% acetonitrile/0.02% HBFA; buffer D was same as buffer C, but with 500 mM ammonium acetate.
Protein identification was done with Integrated Proteomics Pipeline (IP2, Integrated Proteomics Applications, Inc San Diego, CA) using ProLuCID/Sequest, DTASelect2 and Census (Xu et al., 2015;Tabb et al., 2002;Park et al., 2008). Tandem mass spectra were extracted into ms1 and ms2 files from raw files using RawExtractor (McDonald et al., 2004). Data was searched against a Spodoptera frugiperda protein database with the purified mouse KCC4 sequence added, supplemented with sequences of common contaminants, and concatenated to form a decoy database (Peng et al., 2003). LTQ data was searched with 3000.0 milli-amu precursor tolerance and the fragment ions were restricted to a 600.0 ppm tolerance. All searches were parallelized and searched on the VJC proteomics cluster. Search space included all half tryptic peptide candidates with no missed cleavage restrictions. Carbamidomethylation (+57.02146) of cysteine was considered a static modification. In order to identify authentic termini, we required one tryptic terminus for each peptide identification. The ProLuCID search results were assembled and filtered using the DTASelect program with a peptide false discovery rate (FDR) of 0.001 for single peptides and a peptide FDR of 0.005 for additional peptides for the same protein. Under such filtering conditions, the estimated false discovery rate was less than 1%.

Nanodisc reconstitution
Freshly purified and concentrated KCC4 in Buffer three was reconstituted into MSP1D1 nanodiscs with a mixture of lipids (DOPE:POPS:POPC at 2:1:1 molar ratio, Avanti, Alabaster, Alabama) at a final molar ratio of KCC4:MSP1D1:lipids of 0.2:1:50. Lipids in chloroform were prepared by mixing, drying under argon, washing with pentane, drying under argon, and placing under vacuum overnight. The dried lipid mixture was rehydrated in Buffer 4 (20 mM Tris, 150 mM KCl, 1 mM EDTA pH 8.0) and clarified by bath sonication. DDM was added to a final concentration of 8 mM and the detergent solubilized lipids were sonicated until clear. Lipids, Buffer 4 containing 8 mM DDM, and KCC4 protein were mixed and incubated at 4˚C for 30 min before addition of purified MSP1D1. After addition of MSP1D1, the nanodisc formation solution was 47.5 mM KCC4, 104 mM MSP1D1, 13 mM DOPE:POPS:POPC, and 4 mM DDM in Buffer 4 (final concentrations). After mixing at 4˚C for 30 mins, 60 mg of Biobeads SM2 (Bio-Rad, USA) (prepared by sequential washing in methanol, water, and Buffer four and weighed damp following bulk liquid removal) were added and the mixture was rotated at 4˚C overnight (~12 hr). Nanodisc-containing supernatant was collected and spun for 10 min at 21,000 x g before loading onto a Superose 6 Increase column in Buffer 4. Peak fractions corresponding to KCC4-MSP1D1 were collected and spin concentrated using a 100 kDa cutoff for grid preparation.

Grid preparation
The KCC4-MSP1D1 nanodisc sample was concentrated to~1 mg/mL and centrifuged at 21,000 x g for 10 min at 4˚C prior to grid preparation. A 3 uL drop of protein was applied to a freshly glow discharged Holey Carbon, 400 mesh R 1.2/1.3 gold grid (Quantifoil, Großlö bichau, Germany). A Vitrobot Mark IV (FEI/Thermo Scientific, USA) was utilized for plunge freezing in liquid ethane with the following settings: 4˚C, 100% humidity, one blot force, 3 s blot time, 5 s wait time. The KCC4 detergent sample was frozen at 4.5 mg/mL and centrifuged at 21,000 x g for 10 min at 4˚C prior to grid preparation. A 3 mL drop of protein was applied to a freshly glow discharged Holey Carbon, 400 mesh R 1.2/1.3 gold grid. A Vitrobot Mark IV (FEI/Thermo Scientific, USA) was utilized for plunge freezing in liquid ethane with the following settings: 4˚C, 100% humidity, one blot force, 4 s blot time, 1 s wait time. Grids were clipped in autoloader cartridges for data collection.

Data collection
KCC4-MSP1D1 grids were transferred to a Talos Arctica cryo-electron microscope (FEI/Thermo Scientific, USA) operated at an acceleration voltage of 200 kV. Images were recorded in an automated fashion with SerialEM (Mastronarde, 2005) using image shift with a target defocus range of À0.7À 2.2 mm over 5 s as 50 subframes with a K3 direct electron detector (Gatan, USA) in super-resolution mode with a super-resolution pixel size of 0.5685 Å . The electron dose was 9.333 e -/ Å )/s (0.9333 e -/ Å 2/frame) at the detector level and total accumulated dose was 46.665 e-/Å 2. KCC4-detergent grids were transferred to a Titan Krios cryo-electron microscope (FEI/ Thermo Scientific, USA) operated at an acceleration voltage of 300 kV. Images were recorded in an automated fashion with SerialEM (Mastronarde, 2005) with a target defocus range of À0.7 to À2.2 mm over 9.6 s as 48 subframes with a K2 direct electron detector (Gatan, USA) in super-resolution mode with a super-resolution pixel size of 0.5746 Å . The electron dose was 6.092 e -/ Å )/s (1.2184 e -/ Å /frame) at the detector level and total accumulated dose was 58.4832 e -/Å . See also Table 1 for data collection statistics.

Data processing
The processing pipeline is shown in Figure 2-figure supplement 4A-C. We used Cryosparc2 (Punjani et al., 2017) for initial model generation and refinement until reconstructions reached 4-5 Å resolution. Bayesian polishing and nanodisc subtraction in Relion 3.0.7 (Zivanov et al., 2019;Zivanov et al., 2018) were used to achieve highest resolution reconstructions. While the contribution of disordered or flexible N-and C-terminal regions to alignments is unknown, the remaining 55 kDa asymmetric membrane protein is among the smallest in terms of resolved mass resolved by cryo-EM to date.
A total of 1572 movie stacks were collected, motion-corrected and binned to 1.137 Å /pixel using MotionCor2 (Zheng et al., 2017), and CTF-corrected using Ctffind 4.1.13 (Rohou and Grigorieff, 2015; Figure 2-figure supplement 4A). Micrographs with a Ctffind reported resolution estimate worse than 5 Å were discarded. A small number of particles (~1000) were picked manually and subjected to two-dimensional classification to generate references for autopicking in Relion. 1,826,000 particles were autopicked and extracted at 2.274 Å /pixel (2x binned) for initial cleanup. Non-particle picks and apparent junk particles were removed by several rounds of two-dimensional class averaging. The remaining 887,132 particles were extracted at 1.137 Å /pixel and imported into Cryosparc. An additional round of 2D classification generated a particle set of 491,111. These particles were the input of an ab initio reconstruction (non-default values: four classes, 0.1 class similarity, 4 Å max resolution, per-image optimal scales). 2D classification of particles (160,868) that contributed to the most featured volume resulted in a set of 125,593 particles which were the input of an ab initio reconstruction. Alignments were iteratively improved using non-uniform (NU) refinement (0.89 window inner radius, 120 voxel box size, 10 extra final passes, 10 Å low-pass filter, 0.01 batch epsilon, minimize over per-particle scale, 1-4 Å dynamic mask near, 3-8 Å dynamic mask far, 6-10 Å dynamic mask start resolution, 4-6 Å local processing start resolution). Two separate NU refinement output volumes were input into a heterogeneous refinement job of the 886,528 particle set (forced hard classification, 10 Å initial resolution, five final full iterations). The more featured class (538,280 particles) was heterogeneously refined (forced hard classification, 10 Å initial resolution, five final full iterations) (Figure 2-figure supplement 4B). The particles and volume from one output (354,234 particles) were input into the first of three iterative NU refinements (10 extra final passes for the second and third iteration).
Particle positions and angles from the final cryoSPARC2 refinement job were input into Relion (using csparc2relion.py from the UCSF PyEM [Asarnow, 2016]) and 3D refined to generate a 4.18 Å map (6 Å low-pass filter, 0.9 degrees initial sampling, 0.9 degrees local searches) (Figure 2-figure  supplement 4C). A second 3D refinement following Bayesian particle polishing improved the map and reported resolution (4.01 Å ) (6 Å low-pass filter, 0.9 degrees initial sampling, 0.9 degrees local searches). CTF refinement with beam tilt group estimation and per-particle defocus was performed, although subsequent 3D refinement did not markedly improve the map. Particle subtraction was performed to remove the contribution of the nanodisc density from alignments and subsequent 3D refinement markedly improved the map (reported resolution 3.86 Å or 3.72 Å after postprocessing) (6 Å low-pass filter, 0.9 degrees initial sampling, 0.9 degrees local searches). A final improvement in map quality and reported resolution and was obtained by removing poor particles with a 3D classification job (two classes, 10 Å initial low-pass filter, 16 tau fudge, no angular sampling). The final particle set (110,143) was subjected to 3D refinement to generate a final map at 3.72 Å resolution (3.65 Å after postprocessing) (6 Å low-pass filter, 0.9 degrees initial sampling, 0.9 degrees local searches). Particle distribution and local resolution was calculated using Relion (Figure 2-figure supplement  5A,B). FSCs reported in Figure 2-figure supplement 5 were calculated using Phenix.mtriage.

Modeling, refinement, and structure analysis
The final cryo-EM maps were sharpened using Phenix.autosharpen (Adams et al., 2010). The structure was modeled de novo in Coot and refined in real space using Phenix.real_space_refine with Ramachandran and NCS restraints. Validation tools in Phenix, EMRinger (Barad et al., 2015), and Molprobity  were used to guide iterative rounds of model adjustment in Coot and refinement in Phenix. Cavity measurements were made with HOLE implemented in Coot (Emsley et al., 2010). Electrostatic potential was calculated using APBS-PDB2PQR (Dolinsky et al., 2004) Pymol plugin. Figures were prepared using PyMOL, Chimera, ChimeraX, Fiji, Prism, Adobe Photoshop, and Adobe Illustrator software.