Novel organelle anion channels formed by chromogranin B drive normal granule maturation in endocrine cells

All endocrine cells need an anion conductance for maturation of secretory granules. Identity of this family of anion channels has been elusive for forty years. We now show that a family of granule protein, CHGB, serves the long-sought conductance. CHGB interacts with membranes through two amphipathic helices, and forms a chloride channel with large conductance and high anion selectivity. Fast kinetics and high cooperativity suggest that CHGB tetramerizes to form a functional channel. Nonfunctional mutants separate CHGB’s function in granule maturation from that in granule biogenesis. In neuroendocrine cells, CHGB channel and a H+-ATPase drives normal insulin maturation inside or dopamine loading into secretory granules. CHGB’s tight membrane-association after exocytotic release of secretory granules separates its intracellular function from extracellular functions of its proteolytic peptides. CHGB-null mice show consistent impairment of granule acidification in pancreatic beta-cells. These findings together support that the phylogenetically conserved CHGB proteins constitute a new family of organelle chloride channels in the regulated secretory pathway among various endocrine cells.


INTRODUCTION
Cells rely on secretory pathways to send out specific bioactive molecules to their surroundings (Grimes and Kelly, 1992). Both constitutive and regulated secretory pathways have been studied extensively since early EM observations of intracellular vesicular trafficking. Many molecular players in these pathways are known, and conserved mechanisms for vesicle targeting and membrane fusion are established (Farquhar and Wellings, 1957  all three steps of the pathway. It was proposed that they interact with cargos, serve as lowaffinity, high-capacity Ca 2+ reserve. Their extracellular functions are executed by CHG-derived 5 peptides that are associated with various human diseases (Bartolomucci et al., 2011). However, the molecular mechanisms for all CHGs' intracellular functions remain to be elucidated.
Granin proteins of distinct subfamilies usually coexist in secretory granules and were hypothesized to work with different partners (Bartolomucci et al., 2011). Native CHGB forms high-order aggregates at low pH and with mM Ca 2+ (Yoo, 1995a, b). Partially purified native CHGB associates quite strongly with lipid vesicles (Yoo, 1995b). A "tightly membrane- In this work we provide the first elucidation of an important intracellular function of CHGB in the regulated secretory pathway. Starting from the contradiction in literature between the CHGB being tightly membrane-bound (Pimplikar and Huttner, 1992; Yoo, 1995b) and its isolation partially in heat-stable fractions or soluble protein complexes, we investigate the CHGBmembrane interaction in vitro, in culture cells, and in knockout mice to reveal a novel chloride 6 channel function of the CHGB that is needed for normal granule maturation in endocrine cells and thus in a perfect location to serve the long-sought anion conductance first proposed four decades ago.

CHGB inserts itself into membranes and induces nanotubules from bilayers
To study CHGB function, recombinant murine CHGB was purified from sf9 cells. During biochemical preparation, Triton X-100-like detergents were needed to keep CHGB soluble. In size-exclusion chromatography (SEC), purified CHGB was eluted as a single, symmetric peak ( Fig. 1A) with a size equivalent to an ~300 kDa globular protein. Due to posttranslational modifications, a high content of charged residues, or possibly detergent-binding, the recombinant CHGB ran at ~86 kDa in a reducing SDS-PAGE gel (Fig. 1B, 1C), which is larger than the 78 kDa calculated from its sequence, behaving similarly to mature human CHGB (Pimplikar and Huttner, 1992). Due to the detergent micelle (~100 kDa), the CHGB was likely a dimer, instead of a trimer. The detergent-solubilized CHGB treated with a bifunctional cross-linker, 4-(N-Maleimidomethyl) cyclohexane-1-carboxylic acid 3-sulfo-N-hydroxysuccinimide ester (sulfo-SMCC), showed cross-linked dimers, trimers, tetramers and high-order oligomers (U 2 , U 3 , U 4 , U n in Fig. 1C), indicating a dynamic equilibrium between dimers and oligomers (U n , n >= 4) and dominance of the dimers in detergents. Consistently, small amounts of oligomers were observed in an earlier SEC step during purification (Fig. S1A).
Calcium binding is a biochemical hallmark of CHGB due to high content of negatively-charged residues. We examined calcium-induced CHGB aggregates by light-scattering and negative-stain 8 Ficoll 400 gradient (Arrow in the top of Fig. 1E). Quantification of the protein bands revealed that > 98% of CHGB protein was in membrane (Fig. 1E, bottom panel). To visualize the effects of CHGB-membrane interaction on the bilayer structure, we examined CHGB vesicles by negative-stain EM (Fig. 1F, S2A-C). When CHGB : lipid molar ratio (PLR) > 1 : 1,000, equivalent to ~40 CHGB dimers per 100 nm vesicle, 25 nm nanospheres appeared on vesicles (red arrowheads in Fig S2B). More CHGB led to ~20 nm-thick nanotubules capped with 25 nmdiameter hemi-nanospheres (Fig. 1F, S2C). In some specimens, the nanospheres were severed into individual soluble nanoparticles. These results reveal CHGB-induced remodeling of membranes when local PLR is high, a condition likely being satisfied at the TGN sites for granule biogenesis, which might contribute to the heat-stable or soluble fraction used in many published studies (Benedum et al., 1986;Benedum et al., 1987). The nanotubules and nanospheres indicate strong positive curvature caused by CHGB. To avoid membrane remodeling, we intentionally limited the PLR < 1: 5,000 in most vesicle-based assays.
Lipid membranes drive CHGB oligomerization. When purified CHGB was first reconstituted into vesicles and then extracted with detergents plus ~0.1 mg/ml lipids, >70% of CHGB protein was eluted by SEC at a position equivalent to an ~0.8 MDa globular protein (Fig. S2D), suggesting that lipids may stabilize a high-order oligomer. Due to the elongated shape of the cryoEM map (Fig. 1D), biochemically cross-linked tetramers (Fig. 1C) and the detergent micelles, the smallest lipid-stabilized oligomers may be a tetramer (U 4 ; Fig. S1A). It may endow the function of CHGB in membrane. 9 To examine the biophysical nature of CHGB-induced membrane tubulation, we studied whether CHGB causes membrane leak. A fluorescein-release assay was implemented (Mukherjee et al., 2014) to monitor the increase in fluorescence when the fluorescein was released and unquenched. Our data showed that CHGB vesicles did not leak the 10Å fluorophore (Fig 2A).
Chloride leak raised the question of whether CHGB is deeply integrated in membrane.
Trypsinization of CHGB vesicles was performed to test this possibility. After 50 min, > 90% CHGB was digested, indicating its preferential insertion from extravesicular side. Trypsintreatment of Ca 2+ -loaded CHGB vesicles introduced no leak, meaning that trypsinization did not break the membrane (Fig. S3E). Mass-spectrometry and N-terminal sequencing of one membrane interacting fragment (MIF, Fig. S3D) identified CHGB 440-597, named as CHGB-MIF (Fig. 2D). Secondary structure analysis of CHGB-MIF revealed a shorter segment (Hex, CHGB 518-597) containing two -helices interspaced by a short random-coil loop ( Structural modeling of Helix3 (Fig. 2J) reveals a hydrophobic surface (yellow) that is >30 Å in length, enough to span a typical bilayer. Possible cooperativity within the Hex segment or between Hex and Helix 1 may contribute to the "tightly membrane-associated form" of CHGB (Pimplikar and Huttner, 1992). 11 Clleak from CHGB vesicles suggested either a transporter or a channel (Fig. 2C). We first tested if CHGB is a channel by bilayer recordings (Lee et al., 2013). The membrane-remodeling property made it difficult to record from many CHGB molecules. Instead, when diluted CHGB vesicles of low PLR (<1:10,000) were fused into planar lipid bilayers in the presence of 0.5 mM CaCl 2 , we observed multiple channel events (Fig. 3A). The channels were almost always open in a low voltage range (-50 to +50 mV), and switched off more frequently in a higher transmembrane electrostatic potential. These patches lasted for ~10 minutes, suggesting stable CHGB function. We also found that NaF and 0.5 mM CaCl 2 in the cis side helped minimize leak currents. With 150 mM For Cl -, the measured single channel conductance of CHGB is ~125 pS

CHGB alone suffices to form an anion-selective channel
To test the Clselectivity of the CHGB channel, we recorded single channel currents with asymmetrical Cland symmetrical K + at pH 5.5 (15 vs 1.5 mM KCL, Fig. 3C). K-isethionate was added to maintain high osmolality and ionic strength. Under such conditions, the recorded channels had no measurable outward current in the positive voltage range, suggesting no detectable conduction of K + . Analysis of the average single channel currents at different negative voltages yielded an estimated chord conductance of ~58 pS (black line in Fig. 3D, for 15 mM Cl -) and by extrapolation a reversal potential of ~+65 mV, close to the calculated Nernst potential (+59 mV) of Cl -, far away from that (0 mV) of K + , H + or OH -. The detected channel is thus Cl -selective with no K + conduction. As negative controls, vesicles prepared with BSA, CHGA ( Fig   S4I), a CHGB deletion mutant lacking the CHGB-MIF (CHGBΔMIF) and CHGB-MIF all failed to generate any channel activity (data not shown), suggesting that the observed activity is probably genuine to CHGB. When we measured reversal potentials of the channels in bi-ionic 12 conditions, the estimated permeation ratio between Fand Cl - (Fig. 3E, 3F) is ~1.2. Because of the small number of channels in these recordings, we made sure that in higher voltage ranges, the channel closure reached zero current level briefly (as demonstrated by the red trace in Fig. S5A) such that the reversal potential determined from the current-voltage curves (Fig. 3F)  Because only a small number of channels were measured in lipid bilayers, we questioned if a trace amount of contaminating channels might have contributed to the recorded activities. We separated 20 micrograms of purified CHGB that had been kept in cold room for ~4 days, and detected two faint smaller bands (1 and 2 in Fig. S5C). When the CHGB bands and the two shorter bands were cut out and digested for HPLC/MS analysis and Proteomic Identification (Supplementary Tables S1 and S2), all three bands were heavily dominated by CHGB peptides.
The two shorter bands shared 5 peptides and were therefore CHGB degradation products. Other candidates had much fewer matched peptides, less than 6 for the CHGB band and only one for two smaller bands. Among them only KvQT member 5 (accession number E9Q9F in Table S1) is a K + channel, but no anion channel. When we compared the scanned density of the bands (Fig.   S5C, right), the two degradation bands, usually absent in fresh samples (Fig 1B), were ~1.5% of total CHGB. Results from diluted proteins in the same gel suggested that any contaminant that is more than 0.12% of the total mass would have been detected. The purified CHGB thus has > 99.8% purity, making it very unlikely (<0.2%) for a contaminating anion channel to account for the observed channel activity in lipid bilayers.

CHGB conducts Fand Clbetter than other anions
Because it was critically important to rule out stringently the possibility of trace contaminant channels yielding the measured channel activity in Fig. 3, we quantified anion flux from a large number (> 1.0E11) of CHGB vesicles. The Ag/AgCl measurement in Fig. 2C was limited to Cl -, not other anions; nor was it highly stable due to slow mixing of valinomycin and drifts of electrode potential. We instead implemented a light scattering-based flux assay (Fig. S3G) (Stockbridge et al., 2013) by both steady-state and stopped-flow fluorimetry. In both systems, valinomycin triggers K + efflux and anion release from vesicles, resulting in a sudden drop of intravesicular osmolality and collapse of vesicles into ellipsoids that scatter more light (Fig. S3G, right side). Extrusion was used to control vesicle dimensions so that our measurements would not be dominated by large-sized vesicles. As a positive control, the bacterial EriC Cl -/H + cotransporter in vesicles yielded a robust increase in light scattering (Fig. S4H) (Stockbridge et al., 2013). The CHGB vesicles prepared in parallel delivered a strong signal (Fig. 4A). The increase in steady-state light scattering saturated within the 10-s break after valinomycin addition, faster than the 20-s duration of Clefflux in Fig. 2C. A non-specific Clchannel blocker, DIDS was able to block the strong signal with an apparent K d = 0.5 M (Fig. 4B).
As negative controls, recombinant CHGA in vesicles produced no signal ( Fig. S4I-J); nor did CHGB-MIF or CHGBMIF (Fig. 4C-D). Hence, CHGB-MIF alone is insufficient to form the anion channel although it is indispensible to a functional channel. Considering the negatively charged residues contributing to anion selectivity of other Clchannels (Maduke et al., 2000), we deleted a short loop (residues 540-551; CHGB-L1) right before Helix 3 (Fig. S4C) and mutated 14 three negatively-charged residues in this loop to Ala (marked with arrows in Fig. S4C-D; Fig.   4E). CHGB-L1 had significantly lower flux than wild-type CHGB. Mutation of the wellconserved E558 blocked approximately 50% of flux while E545A showed no effect and E552A had mild impairment (Fig. 4E). These data further demonstrate Helix 3 and the loop right ahead of it are important for ion conduction of the CHGB channel (Fig. S4D).
We next compared the relative ion selectivity of the CHGB channel by loading different anions into vesicles. Our data show that the CHGB channel conducts Cland Fmuch better than Br -, I -, NO 3 -, SCN -, formate or citrate (Fig. 4F), suggesting that the channel is much more selective than other known Clchannels or transporters (Fahlke, 2001;Maduke et al., 2000). Relative flux of Fand Clis consistent with the measured permeation ratio P F /P Cl =1.2 (Fig. 3F). An apparent physiological significance for higher anion selectivity of the CHGB channel is to prevent small intracellular organic anions (metabolites) from being passively concentrated into secretory granules and dumped as a waste.
The ion selectivity data (Fig. 4F) make two predictions. One is that extravesicular 300 mM Bror Ishould not affect the light-scattering signal, while Fshould completely stop it, precisely what our experiments showed (Fig. 4G). The other is that a low concentration of extravesicular Clshould shift the initial steady-state Nernst potential across vesicle membranes and significantly inhibit Clflux because of a voltage-dependent factor affecting the rate of valinomycin-mediated potassium flux ( Fig. 4H; Section 18 in Supplementary information). Indeed, extravesicular [Cl -] of 0.1 to 2 mM significantly impaired the light-scattering signal (Fig. 4I).

High-cooperativity among CHGB subunits in forming functional channels
Steady-state fluorimetry measurement cannot detect fast channel-opening events because of slow and uneven equilibration of valinomycin among vesicles. A stopped-flow fluorometer (Hayner et al., 2014) with a dead time of 2 ms was used to overcome this limit. We compared the lightscattering signals by both steady-state and stopped-flow fluorimetry, and titrated PLRs of CHGB vesicles with ~0.4 mg/ml lipids. When PLR < 1:50,000, no signal was detected in either system ( Fig. 5A-B). When PLR reached 1:1,000, the signal approached maximum. All stopped-flow traces showed instantaneous jump, suggesting a fast efflux of K + /Clwithin 2 ms. Because of fast and even partitioning of valinomycin into vesicles, the instantaneous jump in light scattering reflected a sudden change in vesicle shape after KCl efflux. When measurements from the first 5 seconds from the steady-state fluorimetry and the first 40 ms of the stopped-flow data (Fig. 5C &5D) were plotted against [CHGB] and fitted with a Hill equation (Fig. 5E), the estimated Hill coefficients were ~1.2 from the steady-state measurement and ~4.2 from the stopped-flow data, suggesting high positive cooperativity among CHGB subunits in forming a functional channel.
The steady-state data revealed less cooperativity probably because of uneven and slow mixing of valinomycin with vesicles and the complication from water movement in response to osmolality change. When the data in Fig. 5E were plotted against the average number of CHGB subunits per 100 nm vesicle (Fig. 5F), the threshold for strong signals occurred at ~4 CHGB subunits per vesicle, and nearly 80% of the maximum signal was achieved with ~8 CHGB subunits per vesicle. Assuming a Poisson distribution of CHGB subunits in vesicles, our data suggest that most probably 4 CHGB subunits form one channel (Section 19 in Supplementary Information). 16 Valinomycin might form a K + channel. The light-scattering assay requires fast K + flux. Varying [valinomycin] per vesicle is hence expected to affect the signal. Titration of [valinomycin] in the stopped-flow measurement (Fig. 5G) found that when [valinomycin] <0.25 M, the flux signal was almost non-detectable, whereas 2-5 M delivered the maximal signal. Fitting of the dose response data with a Hill equation (Fig. 5H) identified a Hill-coefficient of n=2.6, indicating high cooperativity among valinomycin molecules in moving K + across membrane. These data accord with our observation that 0.1 M valinomycin triggered no detectable flux signal. The cooperativity suggests that valinomycin function as trimers or high-order oligomers in membrane, reminiscent of the dimeric valinomycin channel in ultrathin membranes (Gliozzi et al., 1996). To match the Clflux, each valinomycin oligomer must have a conductance > 3 pS, a lower limit for an uncharacterized valinomycin channel.

CHGB channel is required for granular maturation in neuroendocrine INS-1 cells
Because CHGB is a housekeeping protein in secretory granules (Brunner et al., 2007), we tested whether its channel can serve the anion conductance first recognized in chromaffin granules as essential for granule acidification (Johnson et al., 1982). Because the secretory granules are too small (mostly <500 nm) for direct patch-clamp recording, no specific inhibitor for the CHGB channel is available yet, and prior attempted recordings from purified granules all suffered from To perform ratiometric pH measurements, we first examined whether a Lysosensor DND-160 preferentially stained secretory granules in INS-1 cells. A fluorescently labeled secretory granule protein, syncollin-pHluorin , was expressed ( Fig. S6A) and imaged together with DND-160-stained acidic compartments in live cells. DND-160 overlapped very well with syncollin, meaning that nearly all DND-160-stained compartments were secretory granules, which is consistent with published observations (Stiernet et al., 2006).
To alter the CHGB proteins, siRNAs were used to suppress the endogenous CHGB, which allowed transient overexpression of CHGB or its mutants that supersede the siRNA effects. We blue; Fig. S7B) found that wild-type CHGB restored granular acidification, close to cells treated with CTL siRNAs (black, Fig 6B). Contrastingly, CHGBMIF failed to restore granule acidification (blue in Fig. 6B & Fig. S7B). Because CHGBMIF supported granule biogenesis ( Fig. 6A), but had no channel function (Fig. 4D), these findings demonstrated directly that granule acidification in live cells needs CHGB channel function.
De-acidification of secretory granules might alternatively result from decreased activity of vesicular H + -ATPase. To test this possibility, we examined the expression of a key vATPase subunit, ATP6V0A2, and found no change among the four differentially treated cells (Fig. 6C).
This result was expected because our CHGB siRNAs were on target (Fig. S6D) and did not affect the vesicular H + -ATPase.
When proinsulin and insulin were detected by western blot from the four differentially treated cells (Fig. 6D, E). CHGB knockdown decreased total insulin (Ins + ProIns; top in Fig. 6E) and impaired insulin maturation (Ins / ProIns; bottom in Fig. 6E). Overexpressed CHGB corrected both defects. But CHGBMIF overexpression restored only total insulin, not insulin maturation ( Fig. 6E) because it lacked channel function and failed to acidify the granules for normal insulin maturation.

CHGB channel affects dopamine loading in PC-12 cells
To evaluate whether the CHGB conductance is generic to endocrine cells, we tested it in PC-12 cells, which contain the same machineries for regulated secretion as adrenal chromaffin cells.  Fig. 6H). The CHGB channel in PC-12 cells is therefore required for proper loading of dopamine into secretory granules.

CHGB remains on cell surface after exocytotic release of insulin-granules
Because all known CHGB extracellular functions are mediated by its processed peptides, we questioned whether the full-length CHGB protein was retained after granule release. In PC-12 cells, the "tightly membrane-associated" CHGB stayed on cell surface for 2-4 hours and recycled labeling with fluorescent antibodies yielded much stronger signal on the labeled than control cells (Fig. S7E). The surface retention of full-length CHGB is a good way to separate it from its processed peptide hormones.

Granule acidification in pancreatic beta-cells in CHGB knockout mice is impaired.
If the mechanism for CHGB channel is native to secretory granules, granule acidification should be affected in the CHGB knockout mice. We raised a knockout mouse strain produced by the Wellcome Trust Sanger Institute and distributed by EMMA (#10088; URL: https://www.infrafrontier.eu/). Wild-type and homozygous littermates were raised for experiments. Female and male mice were analyzed separately in parallel. Fig. 7D showed the isolated islets and the detection of CHGB from liver tissues of wild-type (CHGB+/+) and knockout (CHGB-/-) mice. Genotype confirmation was done for every mouse (Fig. S8A).
Western blot detection was performed for different types of tissues in order to confirm the penetrance of the knockout mutation. Freshly isolated islets were stained with DND-160 and imaged. To avoid possible variation in staining, we focused on the cells closer to the surface of each islet. The wild-type beta-cells (top row of Fig. 7E) have much more well-stained granules than the cells from the knockout mice (bottom row of Fig. 7E; contrast of these images was enhanced to show the otherwise faintly-stained granules with increased background; raw data were used for ratiometric measurements). Ratiometric measurements from hundreds of granules found that CHGB knockout caused significant deacidification in the insulin-secretory granules in pancreatic beta-cells from both male and female mice ( Fig. 7F; S8B).

CHGB-membrane interaction and channel activity in secretory granules
Our data collectively support a working model that CHGB inserts itself into lipid membranes from the luminal side, which promotes CHGB oligomerization by shifting a dynamic equilibrium between the dimers and the predicted tetramers, and generates a Clchannel ( was attributed to the saturation of an unknown intracellular mechanism for dopamine concentration. Our model now depicts a concrete mechanism that without CHGB the chromaffin granules de-acidify and pump much less dopamine from cytosol to granules.

Membrane-insertion induces CHGB channel formation
In neuroendocrine cells, CHGB is co-translationally transferred into ER, where it should be inserted in membrane. Similarly, our data suggest that CHGA is fully membrane-attached ( Fig.   23 S4I). Our data show that both CHGB and CHGA are fully reconstituted in the liposomes so we propose that the full-length CHGB is always in membrane in the regulated secretory pathway.
Because both ER and Golgi membranes have their own Clchannels, the CHGB channel is nonessential for ER and Golgi. The Cys-loop domain of the CHGB guides its delivery to its indispensible role in the secretory granules. Further, partial processing of a fraction of CHGB proteins into short peptides in secretory granules should render a significant number of anion channels defective before granule release. Without studying whether the "tightly membraneassociated" CHGB on cell surface is functional (Pimplikar and Huttner, 1992), we propose that CHGB channel's main function is intragranular and may be tightly regulated. After granule release, the full-length CHGB is retained in membrane and is recycled back into the granules, which separates the intracellular functions of the CHGB channel from the extracellular functions of CHGB peptides, especially those that act as peptide hormones.
Membrane insertion of a protein without canonical hydrophobic transmembrane segments to form ion channels has multiple precedents. Hemolysin, C-type lectin, VopQ, etc. are a few Similarly, structures of CHGB dimer, tetramer, and higher-order oligomers will be needed to reveal their allosteric changes during channel formation. Further, the sidedness of CHGB insertion in membrane endows special functions such as membrane remodeling and pH-or Ca 2+ -dependent regulation, which await more investigations.
A "tightly membrane-associated form" of the CHGB is resistant to treatment of basic pH, but soluble by Triton X-114 (Pimplikar and Huttner, 1992). Beside a good mechanism to retain the full-length protein and separate it from the CHGB peptides, such a mechanism remains consistent with our working model because calcium-induced CHGB aggregates ( Considering the highly conserved helices (Fig. S4B, S4D) and the key residues (e.g. E558 in Fig.   4E, S4D), we propose that the Clconduction represents a universal property within the CHGB subfamily in regulated secretory granules. Verification of a causative relation between altered CHGB-channel function and these diseases will make the CHGB channel a druggable target in the future.

Separation of CHGB function in granule biogenesis, maturation and release
Without firmly verified physiological functions, CHGB was proposed to be involved in

Relations between CHGB and other proposed granular channels
Besides IP 3  . All these conflicting results may be partially due to difficulty in avoiding contaminating intracellular or plasma membranes. Only in a well-controlled clean system can high-level certainty be achieved in order to resolve these controversies. Patch-clamp methods do not allow direct recordings from secretory granules due to their small size (mostly < 400 nm).
We will need highly specific CHGB channel blockers in order to separate the CHGB channels 28 from other Clchannels in cell membranes. Needless to say, specific CHGB channel inhibitors will have high clinical value when the connections between altered CHGB function and human diseases are fully solidified.
To conclude, our data for multiple angles converge to the conclusion that CHGB in membrane forms a chloride channel that is essential to normal granule acidification and cargo maturation in both cultured endocrine cells and primary beta-cells. Two conserved amphipathic helical segments mediate the membrane-induced channel formation. The channel function probably is a generic property for the CHGB subfamily. The anion selectivity of the CHGB channel is high and unique within this subfamily and its structural basis awaits further investigations.
29 Methods: Details available in the supplemental information.
CHGB preparation and reconstitution: Recombinant CHGB were purified and inserted into lipid vesicles by removing detergents with BioBeads.
EM examination of CHGB-containing vesicles: Glow-discharged carbon-coated grids were used to present CHGB vesicles for negative-staining and for imaging in an electron microscope.
Single particle EM of CHGB dimers: Individual dimers were images by negative stain and cryo-EM. Datasets of individual molecules were built for image analysis and 3D reconstruction.

Purification of the recombinant CHGB and its mutants expressed in sf9 cells
We used baculovirus to over-express the protein (Invitrogen). Preparation of other constructs followed the same general procedure as what is described here unless separately stated. The cDNA of the mouse CHGB gene was cloned from a pcDNA3.1 plasmid (a gift from Dr. Barbara The CHGB protein concentration was estimated by OD 280 using a calculated extinction coefficient of 82,405 M -1 cm -1 . Purified CHGB was subjected to 10% SDS-PAGE to confirm its purity. For western blot, the protein bands in the SDS-PAGE gel were transferred to a PVDF membrane, immunostained with monoclonal antibodies, and visualized by chemiluminescence (Thermo scientific; SuperSignal™ West Pico).

Preparation of proteoliposomes
Reconstitution follows closely a published protocol(Lee S. et. al, 2013). 5 or 10 mg of Egg PC lipids (Avanti polar lipids were dried in a glass with Argon gas, vacuum-treated for 1.0 hour, hydrated with autoclaved MilliQ water, and vortexed before being sonicated in an iced waterbath sonicator to make small unilamellar vesicles. N-decyl β-D maltopyranoside (DM; Affymetrix --Anatrace) was added to 40 mM. After 5-6 hours at RT, the detergent/lipid solution became almost completely transparent. Lipid/detergents and proteins in detergents were mixed in a desired protein: lipid molar ratio (PLR), and rocked overnight in a cold room. Next day, Bio-Beads were added to gradually remove detergents. After reconstitution, the vesicle solution became cloudy, and was aliquoted, flash-frozen in liquid nitrogen and stored at -80⁰C until experimental use. Control vesicles were prepared similarly without protein. Other proteins were treated in the same procedure even though some of them were soluble and did not get into the vesicles at all.
For vesicle floatation assay, reconstituted vesicles were mixed with 20% Ficoll 400 by 1:3 volume ratio. The mixture (~0.3 ml) was the bottom layer, above which 10% and 5% Ficoll were loaded sequentially. The gradient was centrifuged at 250,000 x g for 3.0 hours at 4 degrees C.
The vesicles migrated to the top 5% layer, and were clearly visible as a narrow band in the gradient. Non-reconstituted proteins stayed at the bottom. The gradients were fractionated for protein detection.

Recordings of single channel events in bilayer lipid membranes (BLMs)
The CHGB in egg PC vesicles at 0.50 mg/ml in varying PLR of 1:10,000 to 1:2,000 were first tested in 150 to 300-micron bilayer membranes. We observed that immediately after vesicle fusion the membranes were quiet and stable, but in some cases there were sudden changes in a short while (within a few minutes) with multiple high-conductance events, which usually became quiet again after a few tens of seconds. These observations have defied our efforts to record macroscopic currents from many CHGB channels.
Instead vesicles with PLR of 25,000 to 1:10,000 led to small currents (usually < 100 pA at 100 mV) from a handful of channels. The channels remained mostly open in low Vm ([-50, +50 mv]), but frequently closed in higher voltages (Vm >80 or <-80 mV). These channel events often occurred for tens of minutes and then disappeared. We suspected that certain lipid effects might cause such behavior in our recording and will need to use a bSUM to study it (Zheng H was used to measure currents and events and future data analysis was done in IGOR Pro (WaveMetrics, Inc.). Current recordings were filtered at 1 kHz using a Bessel filter and sampled at 2-5 kHz. The liquid junction potential between the solutions we used was < 0.1 mV. With balanced salt solutions and under 0 mV holding potential, the recording system had a -0.8 to -1.0 pA leak current across a thinned membrane in the absence of any channels. This small negative current was corrected when reversal potential was read out from fitting the I-V curves.
For the recordings made from bilayers under asymmetric Cland symmetric K + , the cis side had 1.5 mM KCl, 150 mM K-isethionate, 10 mM MES-HCl, pH 5.5. The trans side started with 15 mM KCl and 10 mM MES, pH5.5, and was balanced with 135 mM K-isethionate after the appearance of channel activities. The liquid junction potential between these two solutions was measured to be close to zero. In these solutions, we recorded no significant outward currents.
We also tested the patching of blebbed membranes fused from reconstituted vesicles without much success. Recordings from giant unilamellar vesicles (GUVs) need a completely different set up, and will be tested in the future for a separate publication.

Measurement of chloride efflux by a Ag/AgCl electrode
Direct Clefflux from vesicles was measured as described previously (Stockbridge et al., 2012).
We followed closely what was described in the paper and discussed with Dr. To titrate the channels per vesicle, PLR varied from 1: 100,000 to 1:1,000. Signals from three independent experiments (different batched of vesicles) were averaged and normalized against the maximal signals. Data were with a Hill-equation: The estimated Hill co-efficient, n, for the steady-state measurement is ~1.4, suggesting that a functional channel needs more than one CHGB subunits. The steady-state measurement suffers from uneven mixing of valinomycin with vesicles due to uneven partitioning at the starting point, and from the slower water diffusion during the relaxation step after a change in vesicle shape.

Preparation of bacterial EriC transporter
The expression construct was obtained from Dr. Christopher Miller at Brandeis University. The protein expression and purification followed a published procedure (Maduke M et al., 1999).
Eric protein was purified using Ni-NTA affinity chromatography and size-exclusion chromatography in a Superdex 200 column (GE healthcare). Protein concentration was measure and reconstituted in egg PC lipids in the same way as described above for the CHGB reconstitution.

A stopped-flow system to observe the fast kinetics of anion flux
To overcome the shortcomings of the steady-state experiments, we modified the light-scattering

Negative-stain Electron microscopy of reconstituted CHGB vesicles
Copper grids coated with a thin layer of carbon film were baked at 70°C overnight the day before experiments. After glow discharge, three microliters of reconstituted CHGB vesicles with specific PLRs were loaded. The sample was incubated on a carbon-coated grid at room temperature for 30 seconds before being blotted. After that, the grid was stained with 2.0% phosphotungstic acid (PTA), pH ~8.0 for 30 seconds. The grid was air-dried before being observed in a JEOL JEM2200FS microscope. Images were taken with a Gatan K2 Direct Electron Detector at 25,000 x with a defocus level of -2.0 microns and a calibrated pixel size of 1.92 Å.
For 3D reconstruction from negative-stain EM images, images were recorded with an electron dose of 20 e − /Å 2 on a 4K × 4K Gatan K2 Summit Direct Electron Detector (Gatan, Pleasanton, CA) in counting mode. 180 images were selected based on the power spectra determined by CTFFIND3 (Mindell JA et al., 2003). 140 images with minimal astigmatism were selected for particle picking. The particles were selected using the Boxer module in EMAN 2 (Tang G, et al. (2007). A total of ~5,400 particle images were manually selected. An initial model was generated by angular reconstitution in IMAGIC 5 ( van Heel M et al., 1996) and finally refined in SPIDER (Frank J, et al., 1996& Jiang QX et al., 2003. The final map was calculated from ~3,000 particle images at a nominal resolution of 30Å. The handedness of the map was tested from images from +15 and -15 degrees as we did for the C3PO negative stain map (Llaguno MC, et al., 2014). The small, compact size of the dimer made us less confident in the handedness at this point. A high-resolution map will be used to further examine chirality.

CryoEM study of CHGB dimers
Quantifoil R2/2 grids (Quantifoil Micro Tools GmbH, Jena, Germany) were coated with a thin carbon film (~2-3 nm). The ChemiC (Ni-NTA) grids were prepared as described before (Anonymous, 1986). 3.0 μl of purified CHGB in detergents was loaded. After incubation for ~15 minutes in a wet-chamber of >90% humidity, the grid was blotted inside a Vitrobot and plungefrozen into liquid ethane bathed in liquid nitrogen (FEI, Hillsboro, OR). After screening in a JOEL2200, good specimens were imaged at HHMI Janelia Farm Research Campus. A Titan Krios microscope equipped with a 4K × 4K Falcon 2 Director (no movie function at the time) was used. The scope was operated at 300 kV and was equipped with a Cs corrector. Automatic data collection was run by a proprietary software package, EPU (FEI, Hillsboro, OR). Images were taken under a defocus of −2.5 to −4.0 microns at a magnification of 37,000 ×, which gave rise to a calibrated pixel size of 1.89 Å at the specimen level.
Because the CHGB dimers were quite small, we scanned many areas for good recognition of the particles. Only a small dataset was successfully built, which came from ~300 of 4K x 4K images.
These images all displayed good Thon rings to a resolution of ~ 6.0 Å with minimal astigmatism and defocus values ranging from -1.0 to -4.0 µm and showed visible particles. 24,086 particles were picked manually and extracted in 196x196 Å 2 boxes. The low-resolution negative-stain map was used as the reference for 3D refinement. Five rounds of 2D classification into 50 distinct classes were done with the program RELION 1.3 ( Scheres SH., 2012). The classes with well-defined particles were selected. 3D classification was performed with these selected particles into five classes. Two classes showing higher resolution features were selected for further refinement. The 12,123 particles that were assigned to these two classes were subjected to 5 additional rounds of refinement using a high-resolution frequency limit of 6 Å. A soft mask was introduced to redo the 3D classification and remove ~45% of the particles. The final map was calculated from ~6,900 particles and the estimated resolution was 9.8 Å by using a threshold

Measuring Ca 2+ efflux from reconstituted vesicles
Proteins (CHGB, its mutants or IP 3 R) were reconstituted into vesicles in a PLR of ~1:5000 in a buffer made of 20 mM HEPES, pH7.5, 100 mM NaCl, 1.0 mM EDTA and 2.0 mM β-ME (high pH) or a buffer made of 20 mM MES, pH5.5 (low pH). To load CaCl 2 , 1.0 mM CaCl 2 was added to the buffer. Right before each experiment, freshly prepared lipid vesicles 10 mg/ml) was extruded by 20 stokes through a membrane filter with an average pore size of 100 nm (Avanti polar lipids). A PD-10 desalting column (Sigma) was used to change the vesicles into a buffer containing 100 mM NaCl and 20 mM HEPES, pH 7.4, 1.0 mM EDTA and 2.0 mM β-ME.
Chelex 100 was used to treat the buffers in order to remove residual calcium ions. Vesicles out of the column (~3.0 mg/ml lipids) were diluted to 0.2 mg/ml into the external buffer inside a quartz cuvette with constant stirring. A Ca 2+ -sensitive fluorophore, Indo-1, was added to 1.0 M. The efflux, if any, was initiated by adding 0.5 µM valinomycin. Indo-1 fluorescence at 410 nm was measured in a Horiba fluoroLog spectrophotometer (HORIBA Scientific Inc.) using the Fluorolog-2 module and an excitation wavelength of 330 nm.
For the IP 3 R-containing vesicles, the protein was purified from rat cerebellum as reported before (Jiang QX et al., 2002), and reconstituted in egg PC in the presence of 2.0 mM CaCl 2 . The efflux of calcium was initiated by adding 1.0 μM IP 3 at different time points. At the end of the experiments, 30 mM β-octylglucoside was added to disrupt the liposomes and release all calcium ions to determine the maximum signal.
To examine the knockdown effect, we lysed the cells for western-blot. 96 hours after transfection, cells were collected using 10 mM EDTA in 1 x PBS. Extracellular EDTA was

Detection of insulin and proinsulin in INS-1 cells
INS-1 cells were transfected with 100 nM control siRNAs or CHGB siRNAs in a 6-well plate and fed with fresh medium after 48 hours. 96 hours after transfection, cells were collected using EDTA and lysed by freeze-thaw cycles as described above. An anti-insulin antibody (L6B10 from Cell Signaling) was used to detect insulin and proinsulin. The band intensity in western blot images was measured in ImageJ (Schneider CA. et al., 2012).
For the rescue experiments, cells transfected with CHGB siRNAs were incubated for 48 hours, and then transfected with a pcDNA3.1 plasmid carrying the genes for the wild-type CHGB or its deletion mutant. After two more days the cells were collected for western blotting analysis. A 12% tricine-SDS-PAGE gel was used for separating insulin and proinsulin. The internal loading control was also used to correct for small variations.

Western blot detection of V-ATPase subunit A2
Cells were treated the same as above. The collected cells were lysed by freeze-thaw cycles.
Roughly 20 mg of lysate protein were separated by SDS-PAGE and detected by western blot using an anti-ATP6V0A2 antibody (Abcam).

ELISA assay for detecting dopamine released from PC-12 cells
On day 0, 3 x10 5 PC-12 cells were seeded into each well of a 6-well culture plate. CHGB Eagle Biosciences). We performed these experiments four times and did not observe obvious toxicity to the PC-12 cells due to the siRNA or cDNA transfection. The measured dopamine was thus normalized against that from the cells transfected with CTL siRNAs. The amount of total protein contained in the cell lysates was measured to confirm that the measured dopamine was released from approximately the same number of cells treated in four different conditions. Protein probabilities were assigned by the Protein Prophet algorithm (Nesvizhskii Al, et al., 2003). Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. Proteins sharing the identified peptides were grouped into clusters.

Estimated ion flow rate through the CHGB conductance requires a fast-conducting channel, not a slow-acting transporter.
The main difference between a transporter and a channel is the flux rate. Even for the fastest known ion transporter (Cl -/HCO 3 transporter), its turnover rate (up to 10 5 per second) would still be several orders of magnitude slower than that (~10 7 per second) of a channel. The stoppedflow-based flux assays can provide a good estimate of the flux rate that is limited by the maximum flow through valinomycin molecules. Titration of valinomycin concentration in Fig   5H suggested that ~200 valinomycin molecules per vesicle were needed to generate a significant signal within the 2-ms fast mixing. We assume that almost all valinomycin molecules partition to lipid membranes due to their hydrophobicity. Valinomycin shows a turnover rate of ~10 4 per second at room temperature in the absence of a transmembrane electrostatic potential. Thus, the estimated K + flux per 100 nm vesicle is ~ 2 x 10 6 per second, or ~4 x 10 3 in 2 ms, which alone is not enough to release a significant fraction of ~10 5 K + ions from each 100-nm vesicle without a transmembrane potential.
In the very beginning of mixing two solutions the mixture had a lipid concentration of ~0 Please note that the Nernst potential became stable once shifted away from the initial infinity.
From these estimates, a strong Nernst potential inside would drive the valinomycin transport of K + ions out. An electrostatic driving force for K + efflux at the first 2 ms can be expressed as a Correction Factor ---exp(zFV/RT), which is ~8.9 x 10 4 at +295 mV, where  is a factor for effective conversion of the electrostatic energy into the mechanic motion and is assumed to be unity here under the consideration of no significant energy loss due to either partial charge loss or charge delocalization during the movement of the K + /valinomycin across the bilayer. Under this consideration, the initial K + flux rate through ~200 valinomycin molecule in each vesicle could be as high as ~1.8x10 8 per ms at +295 mV, which is the peak rate. Once the K + and Clions started to flow out, correction factors quickly decays when the vesicular potential drops to below 180 mV. Such a high initial rate would be enough to quickly dump most of the ~10 5  We consider the following assumptions. 1) When N monomers are available in a vesicle and sufficient to join together to form a channel, the assembly is nearly 100% efficient. Based on the biochemical data, the CHGB protein was fairly stable and our reconstitution procedure was able to incorporate all protein into vesicles in a preferred orientation; 2) One channel suffices to conduct enough Clfrom the interior of a vesicle within 1-2 ms. The conductance of CHGB with 300 mM Clwould be even larger than 125 pS; 3) With assumption 1, all vesicles with less than N copies of CHGB monomers will have no channel and will not contribute to the light-scattering change. These together will form a non-functional fraction with a probability of P(k<N); 4) Those vesicles with more than N monomers will form at least one functional channel, and the surplus monomers (very likely dimers as the basic units) will not interfere with the function of the assembled channels. Because we are using a large number of vesicles for our experiments, the statistical average will likely overcome the experimental variations in vesicle size, completeness in the CHGB insertion from the extravesicular side into individual vesicles, freedom of CHGB dimers to diffuse and interact with each other in vesicles, relative ratio of individual lipid types in the egg PC mixture, diffusion time for valinomycin molecules to arrive at membranes, and degree of reaching complete mixing of two equal volumes within the 2 ms dead time. We think that it is relatively reasonable that these assumptions will be satisfied.
Our experimental data in the stopped-flow measure are listed to the last column. When a least squares analysis was used to compare the differences between the experimental data and the theoretical predictions, the best fit is N=4, which is better than N=5, but much better than N=2, 3, 6, 7 or 8. The agreement of the predictions from Poisson statistics and the experimental data appears not co-incidental, especially at the lower range of average CHGB monomer per vesicle ( <= 4), where we expect that the assumption of random distribution is better satisfied.
Considering the dimers being the dominant species in detergents and the tetrameric form observed in the chemically cross-linked fractions, our data support the tetramers as functional channels, not the pentamers or other oligomers.

NOTE: Multiple CHGB-related peptides that are identified and shared between both bands.
Four shared peptides suggest that the two bands have the same origin. All other candidates only had one peptide detected, very unlikely to be real candidates. Among all the contaminants there is not a single one that can server as an anion channel.  Table 1 for the main band   and Supplementary Table 2 for the two contaminating bands. Right: The scanned densities of these three bands were compared. The two contaminating bands account for only ~1.4% of the density as the CHGB band, suggesting that the contaminants make a very small fraction of the total protein mass. From the data in table 3, the two contaminating bands are chiefly the partial degradation products of CHGB. These data together show that the amount of other proteins, if any, must be significantly less than 0.5% (the density of the contaminating band 2), and below 0.12%, which is the detection level in the gel. The purified CHGB is thus at least 99.8% pure with a small amount (< ~1.5%) of degradation. In most cases, the degradation bands in freshly purified CHGB were not detectable as shown in