ABC transporters control ATP release through cholesterol-dependent volume-regulated anion channel activity

Purinergic signaling by extracellular ATP regulates a variety of cellular events and is implicated in both normal physiology and pathophysiology. Several molecules have been associated with the release of ATP and other small molecules, but their precise contributions have been difficult to assess because of their complexity and heterogeneity. Here, we report on the results of a gain-of-function screen for modulators of hypotonicity-induced ATP release using HEK-293 cells and murine cerebellar granule neurons, along with bioluminescence, calcium FLIPR, and short hairpin RNA–based gene-silencing assays. This screen utilized the most extensive genome-wide ORF collection to date, covering 90% of human, nonredundant, protein-encoding genes. We identified two ABCG1 (ABC subfamily G member 1) variants, which regulate cellular cholesterol, as modulators of hypotonicity-induced ATP release. We found that cholesterol levels control volume-regulated anion channel–dependent ATP release. These findings reveal novel mechanisms for the regulation of ATP release and volume-regulated anion channel activity and provide critical links among cellular status, cholesterol, and purinergic signaling.

Intracellular molecules can mediate intercellular communications once secreted by cells. For example, ATP provides energy for numerous intracellular processes but also signals extracellularly through its cell surface receptors: the ionotropic purinergic P2X receptors and metabotropic P2Y receptors (P2YRs) 2 (1)(2)(3). P2 receptors have an affinity for ATP (ϳ0.1-100 M) corresponding to the low extracellular concentrations of ATP compared with intracellular levels (ϳ1-10 nM versus 1-10 mM, respectively) (2)(3)(4)(5)(6)(7). The extracellular release of ATP is thought to be controlled by both an ATP pore and fusion of ATP-containing vesicles with the plasma membrane. Indeed, P2 receptors are not localized only at vesicular fusion sites but also are present all along the plasma membrane, which supports a nonvesicular mechanism of ATP release (8).
Cell volume is tightly controlled to maintain normal cellular function, and cell swelling upon hypotonic stimulation releases ATP, as well as other compounds (9 -12), through ATP-permeable pores in the plasma membrane. Several molecules are proposed to mediate this stimulus-induced ATP release (5,6), including calcium homeostasis modulator (CALHM) (13), pannexin/connexin (14,15), P2X7 receptors (16), SLCO2A1 (17), and LRRC8 (18). However, the relative contributions of these channels and potential modulators of their activity are not clear. Systematic approaches, such as loss-of-function (LOF) and gain-of-function (GOF) screens, might identify other unknown factors involved in the regulation of ATP release.
The LOF approach to identify critical molecules typically involves the detection of phenotypes in genetically mutagenized model systems. For example, a genome-wide RNAibased LOF screen identified LRRC8 as a component of volumeregulated anion channel (VRAC) (19,20). However, this approach does not identify molecules with redundant functions, housekeeping genes that result in early lethality, or those with multiple functions that produce general phenotypes. By contrast, the GOF approach involves the detection of phenotypes via the overexpression of targeted genes. This approach benefits from its ability to identify molecules with functionally redundant homologs and from its high sensitivity based on high protein expression levels. Nevertheless, caution must be applied with this approach because abnormal gene function may be induced by artificially high expression. Furthermore, the cDNA library used in this approach can affect the outcome if the collection is biased toward certain cDNAs. To circumvent this issue, we prepared a collection of 17,284 nonredundant genes covering 90% of human protein-coding ORFs. We performed GOF analyses with this collection and identified ABCG1 as the most robust, specific modulator of purinergic signaling. Our studies further demonstrate that ABCG1 modulates hypotonicity-induced ATP release through LRRC8Acontaining VRACs in a cholesterol-dependent manner. These findings shed light on novel modulatory machinery for the release of ATP and neurotransmitters that act in cell autonomous and nonautonomous manners.

Assay development for genome-wide GOF screen
Hypotonicity induces ATP release (5,6), which we observed by performing a luciferin-luciferase bioluminescence assay with cerebellar granule neurons treated for 30 s with a hypotonic solution (final concentration, 250 mmol/kg) (Fig. 1A). By contrast, HEK cell lines from frozen stocks responded differently to the hypotonic stimulation, and we identified a subline of HEK cells in which the hypotonic solution did not show a significant increase in extracellular ATP relative to isotonic solution (Fig. 1B). Therefore, this HEK subline was used for GOF screening.
ATP release in response to hypotonicity can stimulate G qcoupled P2YRs, which subsequently activate PLC and inositol 1,4,5-trisphosphate receptors to induce the release of calcium from the endoplasmic reticulum into the cytosol (Fig. 1C). Thus, we evaluated changes in cytosolic calcium concentrations to indirectly evaluate hypotonicity-induced ATP release from the HEK cell subline. We performed calcium 4 dye-based FLIPR assays in a 384-well plate format and monitored realtime responses as changes in fluorescence analyzed as the peak fluorescence value normalized to each baseline value (⌬F/F) (Fig. 1D). With this method, a calcium response was observed in HEK cells in response to the application of the hypotonic solution (Fig. 1E). This response was significantly attenuated (p Ͻ 0.001) by an inhibitor of P2 receptors, 300 M suramin, suggesting that ATP-activated P2 receptors mediate the hypotonicity-induced calcium response. Importantly, these results demonstrate that the calcium FLIPR assay can be used as a sensitive and real-time detector of ATP release.
To identify the machinery responsible for ATP release in a GOF screen, we utilized a nonredundant genome-wide ORF collection that included 3,896 transmembrane ORFs from Ori-Gene and 15,743 ORFs from the Broad Institute (through Thermo Fisher Scientific). After comparison with the HUGO database (21), we cloned an additional 3,274 ORFs from the ORFeome Collaboration (22) into mammalian expression vectors. The final ORF collection contained 17,284 nonredundant ORFs (Fig. 1F), covering 90% of the 19,224 nonredundant human ORFs in the HUGO database, some of which were tagged with a V5 epitope at the C terminus. The HEK cells were transfected with these expression plasmids and some redundant ORFs (for a total 20,680 ORFs) in a 384-well plate format (80 -90% transfection efficiency estimated by the red fluorescence from cells transfected with RFP using the same procedure; Fig. 1F, inset). Two days after transfection, stimulus-induced calcium responses were assayed using the FLIPR Tetra system. For this screen, we used a combined stimulus with a minimal difference in hypotonicity (i.e. 320 mmol/kg stimulant and 340 mmol/kg assay solution) to widen the range of screening and 100 M glutamate to activate glutamate receptors as a control. We then calculated averages and standard deviations for the peak calcium responses (⌬F/F) for each 384-well plate. With this initial screen, we identified 45 ORFs that induced a calcium response Ͼ3 times the standard deviation above the plate mean. ). Hypotonic stimulation increased ATP released from primary cerebellar neurons (n ϭ 4) but not HEK cells (n ϭ 8) relative to isotonic stimulation. RLU, relative luminescent unit. C, schematic diagram of hypotonic activation of G q -coupled P2YR signaling in HEK cells. D, calcium responses using the calcium FLIPR assay. Each trace was analyzed by measuring a peak (1) response normalized to its baseline (2) response (⌬F/F). E, hypotonicity-induced calcium responses were measured with the calcium FLIPR assay in HEK cells preincubated without (vehicle) or with a P2 receptor antagonist, 300 M suramin, for 15 min; traces and quantification of peak calcium response (⌬F/F) with error bars are shown (n ϭ 4). F, overview of the GOF screen with a genome-wide ORF collection to identify modulators of hypotonicity-induced ATP release. The genome-wide ORF collection, including 17,284 nonredundant ORFs, was generated (see details under "Experimental procedures"). Each ORF was transfected transiently into HEK cells, followed by measurements of fluorescence intensity from each well using the calcium FLIPR assay. The data are means Ϯ standard deviation; unpaired t test (A and E). ***, p Ͻ 0.001.

Genome-wide ORF-based GOF screen identifies ABCG1 as a purinergic modulator
We next validated the 45 ORFs identified from the initial screen with three replicates ( Fig. 2A). Calcium responses induced by stimulation with the minimum hypotonic assay solution containing 100 M glutamate were notably high from HEK cells transfected with ORFs encoding GRM1, GRM5, ABCG1 variant 2, and ABCG1 variant 4 ( Fig. 2A). GRM1 and GRM5 encode mGluR1 and mGluR5, respectively, which are G q -coupled metabotropic glutamate receptors that modulate intracellular calcium signaling (23). ORFs encoding ionotropic glutamate receptors were not identified in this screen because their responses to 100 M glutamate were much smaller in the absence of chemical potentiators or their auxiliary subunits. Of the two ATP-binding cassette (ABC) transporter ABCG1 variants, variant 2 is shorter and was used for subsequent experiments, because both variants induced similar responses. HEK cells harboring the ABCG1 ORF were treated with an isotonic solution (final concentration, 340 mmol/kg) or stimulated with a hypotonic solution (final concentration, 250 mmol/kg) or glutamate (final concentration, 100 M) in the isotonic solution. The cells responded robustly to only the hypotonic solution (Fig. 2, B and C). These results indicate that ABCG1 is a novel enhancer of the hypotonicity-induced calcium response.

ABCG1 enhances hypotonicity-induced ATP release
We next investigated the mechanism by which ABCG1 expression enhances hypotonicity-induced calcium responses in the FLIPR assay. To exclude extracellular calcium as a potential source, we added 5 mM EDTA, a calcium chelator, to the media and observed a modest increase rather than a decrease in the hypotonicity-induced intracellular calcium responses (Fig.  3A). This suggests that the calcium source for the hypotonicityinduced calcium response is unlikely to be extracellular. Calcium can also be released intracellularly from the endoplasmic reticulum in an inositol 1,4,5-trisphosphate-dependent manner (Fig. 1C). Indeed, preincubation of ABCG1-transfected HEK cells with a PLC blocker, 10 M U73122, as well as the P2R blocker, 300 M suramin, for 15 min reduced the hypotonicityinduced calcium response significantly (p Ͻ 0.001). Similarly, the addition of an ATP diphosphatase, 10 unis/ml apyrase, to hydrolyze ATP to AMP in the medium reduced the hypotonicity-induced calcium responses. These results suggest that ABCG1 modulates G q -coupled P2YR signaling.
To determine whether ABCG1 modulates other G q -coupled receptors, we measured calcium responses in response to acetylcholine, because HEK cells endogenously express muscarinic-type acetylcholine receptors (mAChRs). We found that ABCG1 expression did not influence the ACh-induced calcium response but increased hypotonicity-induced calcium responses (Fig. 3, B and C). These results establish ABCG1 as a specific enhancer of hypotonicity-induced calcium responses through purinergic signaling. This could be due to an increase in extracellular ATP upon hypotonic stimulation. Thus, we measured extracellular ATP amounts from HEK cells transfected with ABCG1 or RFP (control) before and 30 s after hypotonic stimulation. The medium EDITORS' PICK: Transporters, cholesterol, and VRAC-mediated ATP release was collected using a 96-well liquid handler, and ATP was measured using a luciferin-luciferase bioluminescence assay. ATP amounts were similar between RFP (control)-and ABCG1-expressing cells before stimulation but were significantly higher in ABCG1-expressing cells after hypotonic stimulation (p Ͻ 0.001) (Fig. 3D). Elevated ATP in the extracellular media could result from impaired ATP degradation extracellularly or increased ATP release from the cytosol. We verified that the ATP release and calcium response induced by hypotonicity were not due to impaired exonuclease activity. Hypotonicity-induced calcium responses and ATP release were similar between treatments with vehicle and 100 M ARL67156 to block extracellular nuclease activity in both ABCG1-and RFP-transfected HEK cells (Fig. S1). From these results, we concluded that ABCG1 increases ATP release upon hypotonic stimulation.

ABCG1-dependent ATP release is mediated by ionic strength and VRAC
To identify the component in the hypotonic solution inducing ATP/P2YR-dependent calcium responses in ABCG1-transfected cells, we evaluated the contributions of ionic strength and osmolality, which are lower under hypotonic conditions. We again used the calcium FLIPR assay to measure calcium responses from ABCG1-transfected HEK cells stimulated with varying ionic strength with a constant osmolality (340 mmol/ kg) maintained by adding mannitol. Solutions with a lower ionic strength induced larger calcium responses ( Fig. 4A; see the dose-response plot on the right). However, ionic strength did not influence ACh-evoked responses from endogenous mAChRs in HEK cells transfected with RFP ( Fig. 4A, control). To examine the contribution of osmolality, we stimulated ABCG1-transfected HEK cells with varying osmolality with a constant ionic strength. The peak amplitudes of the calcium responses were relatively insensitive to a change in the osmolality, and the response to ACh stimulation showed similar responses (Fig. 4B). These results suggest that a low ionic strength is the primary factor driving hypotonicity-induced calcium responses from ABCG1-transfected HEK cells through P2YRs, but not mAChRs.

ABC transporters regulate hypotonicity-induced ATP release through cholesterol modulation
We next examined how ABCG1 modulates VRAC activity. ABCG1 is one of 48 ABC transporters, which typically use their ATPase activities to transport substrates (27). The ATPase activity of ABCG1 can be disrupted by a point mutation replacing the lysine residue at position 124 with methionine (K124M) (28). We measured hypotonicity-induced calcium responses in HEK cells expressing the ABCG1.K124M mutant using the calcium FLIPR assay. The hypotonicity-induced calcium response was abolished in HEK cells harboring the ABCG1.K124M mutant (Fig. 5A), suggesting that ATPase activity is required for VRAC modulation.
To determine whether any other ABC transporters contribute to hypotonicity-induced ATP release, we re-evaluated 39 ABC transporter family proteins in the ORF collection with four replicates. Three ABC transporters (ABCG1, ABCG4, EDITORS' PICK: Transporters, cholesterol, and VRAC-mediated ATP release and ABCA1) induced significantly higher hypotonicity-induced calcium responses than RFP-transfected HEK cells (control) (Fig. 5B), supporting redundant modulatory functions among several ABC transporters in the hypotonicityinduced responses.
Because ABCG1 as well as ABCG4 and ABCA1 export cholesterol from intracellular compartments or the plasma membrane (29,30), we postulated that these ABC transporters regulate ATP release and VRAC activity through cholesterol modulation. To test this, we treated RFP-expressing HEK cells with 5 mM methyl-␤-cyclodextrin (M␤CD) for 1 h, which is an established method to reduce cellular cholesterol (31). We confirmed that M␤CD treatment reduced the fluorescent signal of filipin that recognizes cholesterol (Fig. S2, A and B). Then we measured hypotonicity-induced calcium responses with the calcium FLIPR assay (Fig. 6A). M␤CD treatment enhanced hypotonicity-induced calcium responses (Fig. 6B) robustly compared with its effects on ACh-evoked calcium responses (Fig. 6C), suggesting that cholesterol depletion enhanced hypotonicity-induced responses more than G q -coupled receptor signaling. Furthermore, we quantified ATP release from these HEK cells 30 s after hypotonic stimulation by using the luciferin-luciferase bioluminescence assay and found that cholesterol depletion via M␤CD treatment enhanced the amount of ATP released into the medium (Fig. 6D). Moreover, this increased release was abolished by inhibiting VRAC activity with 20 M DCPIB (Fig. 6D). The hypotonicity-induced ATP release was observed in a M␤CD, presumably cholesterol dosedependent manner (Fig. S3). These results demonstrated that hypotonicity-induced ATP release depends on cholesterol content and VRAC activity.
Cholesterol depletion and ABCG1 expression enhanced hypotonicity-induced ATP release and P2YR-mediated calcium signaling, supporting a model in which ABCG1 transports cholesterol from cells, which induces ATP release upon hypotonic stimulation through VRAC. If this is the case, cholesterol repletion should reduce ABCG1-dependent ATP release. To examine this, we assayed calcium responses and ATP release from ABCG1-expressing HEK cells incubated with a mixture of cholesterol and M␤CD (Fig. 7A). Hypotonicity-induced calcium responses were significantly reduced in cells supplied with 10 mM cholesterol/M␤CD (p Ͻ 0.001) (Fig. 7B), with no change in ACh-induced responses (Fig. 7C), suggesting that cholesterol repletion specifically inhibits hypotonicity-induced purinergic signaling. Accordingly, ABCG1 transfection increased hypotonicity-induced ATP release, which was abolished by cholesterol repletion (Fig. 7D). Further addition of cholesterol did not change the hypotonicity-induced ATP release (Fig. S3). Altogether, we concluded that hypotonicityinduced responses are sensitive to cellular cholesterol level, such that cholesterol depletion enhances hypotonicity-induced ATP release through VRAC activity.

Neurons release ATP by regulating VRAC activity through cholesterol modulation
To validate the identified ATP release machinery, we returned to cerebellar granule neurons that exhibit hypotonicityinduced ATP release (Fig. 1A). For these experiments, we performed the luciferin-luciferase bioluminescence assay to measure ATP release from primary cerebellar granule neurons cultured for 9 days in vitro (Fig. 8A). The hypotonicity-induced ATP release was blocked by preincubating the neurons with 20 M DCPIB, the VRAC inhibitor (Fig. 8B). Furthermore, lentivirus infection with shRNA for LRRC8A at day 1 in vitro similarly blocked the ATP release 30 s after stimulation with the hypotonic solution (Fig. 8C). These results indicate that, like HEK cells, cerebellar neurons utilize LRRC8A-containing VRACs for hypotonicity-induced ATP release.
Hypotonicity-induced ATP release was magnified by depleting cholesterol from the cultured neurons via treatment for 1 h with 5 mM M␤CD (Fig. 8D). Moreover, this enhancement was attenuated by incubating the neurons for 1 h with 20 M DCPIB, confirming the contribution of VRAC activity to cholesterol-dependent ATP release. The hypotonicity-induced

EDITORS' PICK: Transporters, cholesterol, and VRAC-mediated ATP release
ATP release was also reduced in neurons preincubated with a mixture of 10 mM cholesterol and M␤CD for 3 h (Fig. 8E). All of these results in neurons are consistent with what we observed in HEK cells, supporting the roles of cholesterol and VRAC in hypotonicity-induced ATP release.

Discussion
We utilized a genome-wide ORF collection for a GOF screen that identified ABC transporters as modulators of hypotonicity-induced ATP release. Further experiments demonstrated that ABC transporter activity modulates hypotonicity/ionic strength-induced ATP release through LRRC8A-containing VRACs by regulating cellular cholesterol levels. These findings reveal a novel ATP-dependent ATP release pathway for the modulation of extracellular ATP that triggers purinergic cellular signaling.

Genome-wide ORF collection
Our GOF screen with the genome-wide ORF collection covering 90% of nonredundant, human ORFs identified ABCG1, mGluR1, and mGluR5 as modulators of hypotonicity-and glutamate-induced purinergic calcium signaling (Fig. 2). Glutamate-induced calcium responses are observed in cells expressing ionotropic glutamate receptors or G q -coupled mGluRs. However, fluorescence signals from calcium-sensitive dyes are easier to detect for mGluR activity, because calcium responses   test (B and C) and two-way ANOVA followed by Tukey's test (D)). ***, p Ͻ 0.001. RLU, relative luminescent unit. EDITORS' PICK: Transporters, cholesterol, and VRAC-mediated ATP release are transient from ionotropic receptors, which desensitize quickly upon glutamate application. There are only two known G q -coupled mGluRs (23) that we identified in our screen, supporting its quality and sensitivity. With regard to the hypotonicity-induced response, the screen identified only ABCG1 as a robust modulator of ATP release.
The results from our GOF screen with the genome-wide ORF collection are compelling. However, there are caveats with this method. One such caveat involves the properties of the encoded proteins. In this screen, we transfected one plasmid/ ORF in each well. Thus, genes that code for functional proteins are detectable. However, many proteins function as a complex with other proteins, which would be missed in this type of screen. For example, this study revealed that LRRC8A is an essential component for the hypotonicity-induced ATP release, but it was not detected by the GOF screen. One potential reason is that LRRC8A functions as a heteromer with LRRC8B-F (19,20), and LRRC8A overexpression by itself does not increase overall VRAC activity (19,20). Another example pertains to the cholesterol transporters, ABCG5 and ABCG8. We tested 39 ABC transporters and found 3 that enhanced hypotonicity-induced purinergic signaling (Fig. 5B). However, it remains unclear whether the negative 36 transporters play any role in regulating ATP release in vivo. Because ABCG5 and ABCG8 require heteromerization to be a functional transporter (32,33), it is unlikely that expression of either transporter alone reconstitutes its transporter activity in transfected cells. Thus, a multigene expression system could be helpful in the future, although it is currently infeasible, because a combination of only 2 genes would involve 361 million pairs (19,224 genes times 19,224 genes). Nevertheless, these potential combinations could be narrowed down by incorporating expression pat-terns. Another caveat involves the quality of the ORFs. We have been collecting ORFs from the OriGene transmembrane protein collection and the ORF collection from the Broad Institute (22) since 2009 and custom cloning in the laboratory. These ORFs are cloned into vectors with a cytomegalovirus promoter and likely to be expressed in transfected HEK cells. However, the expression of all ORFs has not been confirmed. It will be cost-effective to generate one human ORF collection with validated expression and to expand the collection with various splice isoforms.

Molecular signaling for ATP release
Our screen identified ABC transporters as modulators of ATP release through LRRC8A-containing VRAC in both nonpolarized cells and polarized neurons. Some ABC transporters are expressed in polarized epithelial cells, so we also examined hypotonicity-induced ATP release from MDCK cells, a frequently used model cell line for polarized epithelial cells. We did not detect significant ATP release from untreated or cholesterol-depleted MDCK cells in response to hypotonicity (Fig.  S4). Because our proposed machinery of hypotonicity-induced ATP releases requires LRRC8A downstream of ABC transporters (ABCG1/ABCG4/ABCA1) and cholesterol, this result suggests the lower activity of LRRC8A in MDCK cells. ABC transporter-like ion channel, cystic fibrosis transmembrane conductance regulator/ABCC7, p-glycoprotein/multidrug-resistant protein 1/ABCB1, and other ABC transporters were previously shown to regulate ATP release (34 -36), but there is evidence against the ABC transporter as a pore for ATP release (5,6). There are currently two models: (i) ATP transporters by themselves excrete ATP from cells and (ii) ATP transporters modulate some factors that regulate ATP release.  (2,9) ϭ 55.95, p Ͻ 0.001). The data are means Ϯ standard deviation (one-way ANOVA followed by Tukey's test (B-E)). ***, p Ͻ 0.001. RLU, relative luminescent unit.

EDITORS' PICK: Transporters, cholesterol, and VRAC-mediated ATP release
With regard to the first model, the high-resolution structures of ABC transporters and cystic fibrosis transmembrane conductance regulator (37,38) suggest that these proteins do not support ATP permeation, because the pockets are too small for ATP. However, there may be accessory proteins that change the structures of the transporters substantially. As evidence supporting the second model, we showed that ATP transporter ABCG1 modulates ATP release via the regulation of the cellular cholesterol level and VRAC activity. Recent evidence suggests that VRAC itself is capable of gating its channel activity in response to hypotonicity/lowered ionic strength (37,38). Further, as stated above, several ABC transporters, including ABCG1, can export cholesterol from the cell (29,30). Thus, it is reasonable to consider this machinery as a general modulator of ATP release that may be implicated in disease. For example, human patients or mice with elevated cholesterol levels might have altered VRAC-dependent ATP release and purinergic signaling and consequently may have defects in pain sensation or other disorders. Future analyses of ABC transporters and cellular cholesterol levels in association with human disease status will help determine this.
We showed that hypotonicity-induced ATP release requires the ATPase activity of ABCG1. This result supports the critical role of ABCG1 transporter activity in modulating ATP release through VRAC, whose activity depends on intracellular ATP (39,40). However, other studies showed that low-hydrolyzable ATP is sufficient to permit VRAC activity (41,42). Because ATP hydrolysis is required for the ATPase activity of ABCG1, additional studies are needed to clarify how nonhydrolyzable ATP modulates VRAC activity.

ATP-permeable pore and its regulation
Our results suggest that, under hypotonic conditions, ATP is released through LRRC8-containing VRACs. Moreover, this release is enhanced by cholesterol depletion. Typically, the osmolality of hypotonic solution is ϳ200 mmol/kg that is substantially lower than isotonic solution ϳ330 mmol/kg. It remains unclear whether such a robust reduction in osmolality occurs in the brain in vivo. However, under the ABCG1-expressing condition, just a 6% osmolality difference between the hypotonic stimulus and the assay solution induced VRAC activity ( Fig. 2A). Furthermore, we found that altering the ionic strength of the extracellular solution induced VRAC activation (Fig. 4A). At physiological ionic strength (170 mM), VRAC shows low activity. With the expression of ABCG1 or cholesterol depletion, a 5 mM difference in ionic strength is sufficient to induce a 20% difference in VRAC activity. Importantly, cholesterol depletion by itself does not activate VRAC. Perhaps cholesterol plays a critical role in regulating VRAC activity under physiological conditions with small changes in ion concentration.
The mechanism(s) by which cholesterol depletion induces the release of ATP through LRRC8A-containing VRACs is not known. Interestingly, the structure of the LRRC8A homomer shows electron density at the pore, which may indicate some type of lipid (43)(44)(45)(46). Our findings suggest that this lipid could be cholesterol. Under normal conditions, cholesterol could plug or narrow the pore diameter, and only small molecules, such as ions, can pass through, such as observed upon ionic activation of LRRC8A channels reconstituted in liposomes (25). However, under cholesterol-depleted conditions, the cholesterol may be removed from the pore, enabling larger molecules to permeate. Consistent with this, we showed that restoration of the depleted cholesterol substantially reduced ATP release through LRRC8A-containing VRACs. It is possible that VRAC activity is dysregulated under high-cholesterol conditions, inducing an imbalance in the homeostasis of VRAC-permeable molecules in human disorders. Further studies of the relationships between cholesterol and VRAC substrates may provide insights into such human disorders.
In summary, the results of this study show that ABCG1 activity modulates hypotonicity-induced ATP release through VRACs in a cholesterol level-dependent manner. Further studies are needed to confirm these data and to clarify the mechanism by which ABCG1 regulates VRAC permeability.

Animals
All animal handling was in accordance with protocols approved by the Institutional Animal Care and Use Committee of Yale University. Animal care and housing were provided by the Yale Animal Resource Center, in compliance with the Guide for the Care and Use of Laboratory Animals (National Academy Press, Washington, D.C., 1996). The animals were maintained in a 12-h light/dark cycle with ad libitum access to food and water. WT (C57BL/6J, stock no. 000664) mice were obtained from the Jackson Laboratory. Sample sizes were estimated by previous studies. Both male and female animals were used.

Primary cultured cerebellar granule neurons
Primary cultured cerebellar granule neurons were prepared as described previously (47). Briefly, cerebella were dissected from P7 mice and treated with trypsin, and the cells were seeded on poly-D-lysine-coated 384-well plates at a density of 5 ϫ 10 4 cells/well and grown in medium with cytosine arabinoside to suppress growth of non-neuronal cells in a humidified incubator at 37°C with 5% CO 2 . All experiments were performed at day 9 in vitro.

Plasmid transfection
Plasmid was introduced into cells by reverse transfection, unless otherwise noted. Briefly, the cells were plated in a well with a mixture of transfection reagent (FuGENE 6; Promega) and plasmid. The cells were incubated for 48 h at 37°C, 5% CO 2 to allow for gene expression.

Calcium FLIPR assay
The cells were washed with an assay buffer (1ϫ Hanks' balanced salt solution with calcium, magnesium, 2.5 mM probenecid, and 20 mM HEPES, pH 7.4) using a Versette liquid handler (Thermo Scientific). The cells were then loaded with calcium-4 FLIPR dye (Molecular Devices) in the assay buffer and incubated for 1 h at 37°C, 5% CO 2 . The calcium FLIPR assay was performed using a FLIPR TETRA System (Molecular Devices). Solution osmolality was confirmed with a vapor pressure osmometer (Vapro, model5520).

ATP detection
The cells were washed once and then incubated in the assay buffer for 1 h at 37°C with 5% CO 2 . Stimulation and media collection before and after stimulation were performed using a Versette liquid handler (ThermoScientific). Extracellular ATP levels were measured using a luciferin-luciferase bioluminescence assay (ATP determination kit; Thermo Fisher Scientific) in a bioluminescence plate reader (GloMax; Promega).

Cholesterol modification
Cholesterol free modification was performed using established, cyclodextrin-based methods (31). The cells were washed with serum-free medium and then incubated with either serum-free medium containing 5 mM methyl-␤-cyclodextrin (Sigma) for 1 h or serum-free medium containing cholesterol (Sigma) as 10 mM cholesterol mixed with methyl-␤-cyclodextrin for 3 h.

Western blotting
HEK cells were washed with a TE buffer of 20 mM Tris, 1 mM EDTA once, incubated with 1% SDS in the TE buffer, and sonicated briefly. Protein concentration was measured using a BCA assay (Pierce) and adjusted to 0.5 mg/ml. Same protein amount was loaded on SDS-PAGE, transferred to PVDF membranes, and then detected with anti-LRRC8a (Bethyl Labs, rabbit, 1:5000) or anti-␤-tubulin (Sigma, rabbit, 1:10,000) antibodies.

ABC-family phylogenetic tree
Reference human protein sequences for each member of the ABC family were obtained from the UniProt database and then used to generate a protein alignment (ClustalW) and phylogenetic tree (Jukes-Cantor, UPGMA, no outgroup) with Geneious6 software (Biomatters).

Quantification and statistical analysis
Quantification and statistics are detailed in the figure legends. Statistical tests were performed using GraphPad Prism 8.0 software (GraphPad Software). All data are given as means Ϯ standard deviation. Data normality was confirmed using the Shapiro-Wilkes test. Then statistical analysis was performed using one-way ANOVA followed by Tukey's post test, two-way ANOVA followed by Tukey's post, or unpaired t test as required by experimental design. In two-way ANOVA analyses, p values from Tukey's post hoc tests were shown for comparisons between data groups with differences in one experimental factor.