Intracellular Chloride Regulation in AVP+ and VIP+ Neurons of the Suprachiasmatic Nucleus

Several reports have described excitatory GABA transmission in the suprachiasmatic nucleus (SCN), the master pacemaker of circadian physiology. However, there is disagreement regarding the prevalence, timing, and neuronal location of excitatory GABA transmission in the SCN. Whether GABA is inhibitory or excitatory depends, in part, on the intracellular concentration of chloride ([Cl−]i). Here, using ratiometric Cl− imaging, we have investigated intracellular chloride regulation in AVP and VIP-expressing SCN neurons and found evidence suggesting that [Cl−]i is higher during the day than during the night in both AVP+ and VIP+ neurons. We then investigated the contribution of the cation chloride cotransporters to setting [Cl−]i in these SCN neurons and found that the chloride uptake transporter NKCC1 contributes to [Cl−]i regulation in SCN neurons, but that the KCCs are the primary regulators of [Cl−]i in SCN neurons. Interestingly, we observed that [Cl−]i is differentially regulated between AVP+ and VIP+ neurons-a low concentration of the loop diuretic bumetanide had differential effects on AVP+ and VIP+ neurons, while blocking the KCCs with VU0240551 had a larger effect on VIP+ neurons compared to AVP+ neurons.


Results
Vasoactive intestinal peptide (VIP) and vasopressin (AVP) mark the "core" and "shell" partition that has served as a useful anatomical model for dissecting SCN function. Generally, sensory inputs project to the ventrolateral SCN core, while core neurons synapse unto neurons in the dorsomedial shell. Several reports have suggested that excitatory GABA transmission may correlate with this anatomical feature 13-15, 17, 20 . Therefore, we performed Cl − imaging in VIP+ and AVP+ neurons of the SCN to look for differences in [Cl − ] i regulation between these two populations of neurons. To measure [Cl − ] i in SCN neurons, we used a newly-developed Cre-inducible mouse line, with a floxed Cl-Sensor allele inserted into the Rosa26 locus 33 . To obtain Cl-Sensor expression in the SCN, we crossed these mice with either VIP-IRES-Cre mice or AVP-IRES-Cre mice to give Cl-Sensor expression in either VIP+ or AVP+ neurons 34,35 . The resultant VIP::Cl-Sensor mice displayed Cl-Sensor expression in the ventrolateral SCN, while the AVP::Cl-Sensor mice displayed Cl-Sensor expression in the dorsomedial SCN, as expected for VIP and AVP expression (Fig. 1A,B 36 ).
Several studies have indicated that excitatory GABA transmission demonstrates circadian rhythmicity 13,14,16,17,20 . Therefore, we first compared baseline values of R Cl during the day (ZT 2 to 8) and night (ZT 12 to 18) in AVP+ and VIP+ SCN neurons. Interestingly, R Cl was higher during the day in both AVP+ (generalized estimating equations (GEE), p < 0.05) and VIP+ (GEE, p < 0.001) neurons (Fig. 1E) suggesting that [Cl − ] i is higher during the day in these neurons. Several reports have described regional variability of excitatory GABA transmission, with some agreement that it is more common in the dorsal SCN, suggesting that this phenomenon may be specific to AVP+ neurons. Indeed, in the nearby paraventricular nucleus, Haam et al. observed that AVP+ neurons had a more depolarized E GABA relative to their AVP− neighbors 37 . To determine whether AVP+ neurons demonstrate higher [Cl − ] i relative to VIP+ neurons, we compared baseline R Cl values between VIP+ and AVP+ neurons, but found no significant difference in R Cl during either the day or the night (Fig. 1E).
A hyperpolarizing action of GABA will elicit Cl − influx, while a depolarizing action will elicit Cl − efflux. To assess the polarity of GABA transmission in SCN neurons, we performed puff applications of the GABA A agonist isoguvacine in the presence of 2 µM TTX to block potential polysynaptic effects. In response, R Cl increased in AVP+ neurons during the day and night, indicative of Cl − influx and inhibitory GABA transmission (Fig. 2, top row). Cl − influx was also observed in VIP+ neurons during both the day and the night (Fig. 2, bottom row). These results indicate that GABA is inhibitory in both AVP+ and VIP+ SCN neurons. Interestingly, these GABA A -induced Cl − transients lasted for minutes, much longer than expected for GABA A -induced currents, but similar to the timecourse of Cl − transients reported in other cells [38][39][40] . The timecourse of these Cl − transients may represent the activity of chloride transporters. Indeed, it has been shown that transient shifts in GABA polarity can last for minutes and that chloride transporters mediate this equilibration process [41][42][43] .
Therefore, we next examined the degree to which the CCCs set [Cl − ] i in SCN neurons. Since our GABA A -induced Cl − transients indicated inhibitory Cl − influx upon GABA A receptor activation, we first investigated the contribution of KCC2, the neuron-specific CCC responsible for keeping [Cl − ] i levels low in neurons throughout the brain. To test for the activity of KCC2 in setting resting [Cl − ] i we used VU0240551, an antagonist that selectively targets the KCCs 44 . VU increased R Cl by 0.18 in AVP+ and by 0.27 in VIP+ neurons (Fig. 3A,B). Based on our calibration curve ( Supplementary Fig. 1), we estimate these changes in R Cl to reflect a 15 mM increase in [Cl − ] i in AVP+ neurons and a 29 mM increase in VIP+ neurons. VU had a significantly greater effect in VIP+ neurons compared to AVP+ neurons (GEE, p < 0.05), but there were no day/night differences within neuron type (Fig. 3C). Because several Cl − transporters are also transporters for the bicarbonate ion, we wondered if removing bicarbonate ions from the extracellular solution might alter [Cl − ] i and therefore the efficacy of VU. Therefore, we repeated VU application in a separate set of experiments using a HEPES-buffered solution (Fig. 3D). Blocking the KCCs with VU gave a pattern of results similar to that observed with solution containing bicarbonate. VU elicited an increase in R Cl in all conditions, indicative of [Cl − ] i increase. When comparing the effect of VU across solutions, we observed a difference in the amplitude of VU's effect in AVP+ neurons during the day (two-sample z-test, p < 0.05). Furthermore, experiments in HEPES solution revealed a day/night difference in the effect of VU that was not present in the bicarbonate solution (Fig. 3D, GEE, p < 0.05). In a separate set of experiments, we examined the effect of VU on the GABAergic reversal potential (E GABA ) using perforated-patch recording. In these experiments, VU elicited a depolarization of E GABA by approximately 23 mV (paired t-test, p < 0.001), and slowed the recovery of resting E GABA after Cl − loading (paired t-test, p < 0.05) (see Supplementary Fig. 2). Collectively, we interpret these findings to indicate that the KCC family of chloride cotransporters play a major role in [Cl − ] i regulation in SCN neurons.
Previous studies have implicated NKCC1 in [Cl − ] i regulation in SCN neurons [15][16][17][18]32 . To test for a contribution of NKCC1 to resting [Cl − ] i , we used the loop diuretic bumetanide which selectively targets NKCC1 when used at 10 µM 45 . Bumetanide increased R Cl in AVP+ neurons by approximately 0.04 (~4 mM) and decreased R Cl in VIP+ neurons by 0.04 (~3 mM) (Fig. 4). These changes were small but significantly different from baseline (GEE, p < 0.005). As with VU, we observed differences in the effect of bumetanide between AVP+ and VIP+ neurons (GEE, p < 0.001), but no day/night differences within neuron type. Surprisingly, bumetanide elicited a small increase in [Cl − ] i in AVP+ neurons, contrary to its expected role in blocking Cl − uptake. This result may be due to off-target effects of bumetanide, which is known to inhibit the KCCs at higher concentrations 10,44 . In a separate series of experiments, we tested the efficacy of bumetanide in a HEPES-buffered solution, which should diminish Cl − /HCO 3 − exchange. In HEPES, bumetanide reduced R Cl in VIP+ neurons but had little effect on AVP+ neurons (Fig. 4D). However, the effect of bumetanide on AVP+ neurons was significantly greater in the bicarbonate-buffered solution compared to the HEPES-buffered solution (two-sample z-test, p < 0.05). As in the bicarbonate-buffered experiments, bumetanide elicited a greater effect in VIP+ neurons compared to AVP+ neurons (GEE, p < 0.05). Using perforated-patch recording, we also investigated the effect of bumetanide on E GABA . In these experiments, bumetanide did not significantly alter E GABA or the timecourse for the recovery of E GABA following a Cl − depletion protocol ( Supplementary Fig. 3). Overall, we observed relatively small effects of bumetanide compared to VU, suggesting that the KCCs are the major regulators of resting [Cl − ] i in SCN neurons.
The previous results indicate that both NKCC1 and the KCC family of cotransporters contribute to resting [Cl − ] i in SCN neurons. We next investigated how these cotransporters interact to regulate [Cl − ] i . As discussed, blocking the KCCs with VU resulted in a substantial increase in R Cl (Fig. 3). However, because of the relatively minor effect of bumetanide, it was not clear what Cl − uptake pathways were mediating the effect of VU. In order to determine if NKCC1 mediates Cl − uptake in the absence of the KCCs, we applied VU after bumetanide treatment (Fig. 5A). The effect of VU was largely occluded in the presence of bumetanide (Fig. 5A,B). Therefore, although NKCC1 has a relatively minor role in setting resting [Cl − ] i , it does constitute a tonic Cl − influx pathway in SCN neurons. Conversely, we next examined whether blocking the KCCs could reveal a greater bumetanide effect. However, the amplitude of bumetanide's effect was similar in the presence or absence of VU (Fig. 5C,D), indicating that the KCCs are necessary for Cl − extrusion.

Discussion
We performed somatic Cl − imaging to investigate [Cl − ] i in two genetically-defined subpopulations of SCN neurons. We found that R Cl was higher in both AVP+ and VIP+ neurons during the day (ZT 2 to 8) compared to the night (ZT 12 to 18), suggesting that [Cl − ] i is elevated during the day. This observation is in agreement with several reports that have observed increased excitatory GABA transmission during the day and early night 6,13,[15][16][17] . However, we observed Cl − influx in response to GABA A receptor activation, indicative of an inhibitory effect of GABA. Similarly, we observed large changes in R Cl after application of the KCC antagonist VU, but small changes after blocking NKCC1 with bumetanide. VU increased [Cl − ] i dramatically, in the range of 15 to 30 mM, suggesting that the KCCs are the major determinants of [Cl − ] i in AVP+ and VIP+ SCN neurons. Therefore, our results add to a group of studies that have concluded that GABA is exclusively inhibitory in the mature SCN 5, 46-49 . Still, it should be remembered that the SCN is a very heterogeneous nuclei and that AVP and VIP-expressing neurons only constitute ~13% and ~9% of all SCN neurons respectively 50,51 , leaving open the possibility that our study did not address the SCN neurons demonstrating excitatory GABA.
The descriptions of excitatory GABA transmission in the SCN have been riddled with discrepancies. Differences in methodology are likely to underlie some of these inconsistencies. Indeed, whole-cell, perforated-patch, cell-attached, and multi-unit recording techniques as well as Ca 2+ imaging have all been used to address the polarity of GABA transmission in the SCN. The timing of inhibitory post-synaptic currents within the interspike interval is critical in determining whether inhibitory currents will speed up or slow cell firing, further highlighting the nuance of GABA transmission in the SCN 4,9,52 . The issue may also be related to the complexity of intracellular Cl − regulation itself. Besides neurotransmission, [Cl − ] i is an important cellular feature linked to processes such as pH regulation, cell volume regulation, and even membrane potential 12,43,54 . Therefore, cell turgidity as well as the osmolarity and pH of solutions are all likely to influence measures of [Cl − ] i. Further, [Cl − ] i has been shown to change after neuronal damage, and in relation to the proximity of cells to the surface of a brain slice 55,56 . Additionally, previous studies have not adequately ruled out the possibility that disinhibition underlies the observed excitatory effects of GABA transmission in the SCN. The SCN network is known to have diffuse local connectivity 7,19,57 . Therefore, polysynaptic effects must be considered when applying GABA agonists to SCN neurons. Without the inclusion of TTX in the recording media, GABA-mediated inhibition of pre-synaptic inputs could be read out as excitation in the cell of interest. Further, some of the data in support of excitatory GABA transmission has been inferred by the effects of the GABA A receptor antagonist bicuculline 14,15,58 . Regrettably, these results are confounded by the off-target effects of bicuculline, which is known to antagonize SK channels at commonly used concentrations 59 . Indeed, SK channels have been shown to contribute to the resting membrane potential, afterhyperpolarization and spike-frequency adaptation of SCN neurons 60,61 .
We have successfully performed Cl − imaging techniques in SCN neurons, and have demonstrated their utility for monitoring Cl − flux and [Cl − ] i regulation. Cl − imaging offers several advantages compared to gramicidin perforated patch recording by leaving V m unperturbed and by allowing for the sampling of multiple neurons simultaneously. Further, the use of a genetically-encoded indicator allowed us to target specific populations of SCN neurons. Nevertheless, Cl-Sensor has room for improvement. Cl-Sensor's sensitivity to Cl − is not optimal for normal neuronal concentrations of chloride, and the intrinsic H + sensitivity of the indicator within a physiological pH range can be problematic.
Our methodology did not allow us to investigate subcellular differences in R Cl . Indeed, intracellular Cl − gradients have been reported in several types of neurons throughout the brain (see 62 for review). For example, a two-photon Cl − imaging study observed a somatodendritic chloride gradient in a class of retinal bipolar cells, concluding that Cl − is 20 mM higher in dendrites relative to the soma 63 . A somatodendritic Cl − gradient could explain why previous studies have shown GABA-evoked Ca 2+ transients in SCN neurons 15,16,18 , supporting an excitatory role of GABA, while we have observed inhibitory Cl − influx. While dendritic depolarization may be able to activate somatic voltage-gated calcium channels, Cl − efflux at dendrites may not be registered by somatic Cl − imaging. Similarly, a somatodendritic Cl − gradient could explain why bumetanide was able to diminish GABA-evoked Ca 2+ transients, but produced small changes in our measurements of somatic R Cl . Higher-resolution imaging techniques will be necessary to address whether somatodendritic Cl − gradients exist in SCN neurons.
Recently Although we did not observe a difference in resting R Cl between AVP+ and VIP+ neurons, we did observe differential regulation of [Cl − ] i between AVP+ and VIP+ neurons. We found that AVP+ and VIP+ neurons differed in their sensitivity to both VU and bumetanide. The increased sensitivity of VIP+ neurons to VU suggests that they may have lower resting [Cl − ] i relative to AVP+ neurons, consistent with studies that have observed a greater prevalence of excitatory GABA transmission in the dorsal SCN 6, 15, 16 as well as with recent data from two groups who, using the Cl − sensitive dye MQAE, concluded that [Cl − ] i is elevated in the dorsal SCN 9, 23 . Previous in situ hybridization 65 and immunocytochemical 66 studies have described regional expression of chloride transporters in the rat SCN. Therefore, these regional differences in expression may explain the differential effects of VU and bumetanide in AVP+ and VIP+ neurons. Specifically, KCC2 expression was limited to the ventrolateral SCN, and colocalized with neurons containing GRP or VIP. Markedly, KCC2 expression was absent from the dorsomedial SCN and did not colocalize with AVP-rather, KCC3 and KCC4 were found in the dorsomedial SCN 32 . This histology is in agreement with our observation that VU had a larger effect in VIP+ neurons compared to AVP+ neurons. Despite the paucity of KCC2 in the dorsomedial SCN, we observed that VU increased [Cl − ] i in AVP+ neurons, albeit less than it did in VIP+ neurons. The efficacy of VU in the AVP+ neurons may be explained by the non-specificity of VU for KCC2 44 . VU may have acted on KCC3 or KCC4 in AVP+ neurons.
Generally, resting membrane potential in SCN neurons is approximately −45 mV during the night and exhibits oscillations of roughly 10 to 15 mV throughout the day 61 15 . This Cl − uptake may be mediated by the anion exchangers (AEs), which exchange intracellular bicarbonate for extracellular chloride, or could be mediated by a Cl − channel 12,43 . Indeed, our results in HEPES-buffered solutions, which minimize the presence of bicarbonate transport, imply the presence of bicarbonate-dependent Cl − regulation in SCN neurons. Although generally similar, the experiments done in HEPES-buffered solution revealed several interesting differences. The effect of bumetanide in AVP+ neurons differed between solutions, and experiments done in HEPES-buffered solution revealed a day/night difference in VU's effect on VIP+ neurons (Figs 3 and 4). These findings implicate the activity of the Na + -driven Cl-HCO 3  Fig. 3). However, the effect of VU was occluded in the presence of bumetanide (GEE, p < 0.001), suggesting that NKCC1 mediates the Cl − accumulation elicited by VU. Conversely, the effect of bumetanide in the presence of VU was similar to the effect of bumetanide alone (Fig. C and D), indicating that the KCCs are necessary to mediate Cl − extrusion in these neurons.
ScIeNtIfIc REPORtS | 7: 10226 | DOI:10.1038/s41598-017-09778-x exchanger (NDCBE) or the AE family of cotransporters in SCN neurons. Further research will be necessary to address the role of these transporters in SCN physiology.
Overall, our results demonstrate day/night and regional differences in [Cl − ] i regulation and highlight the KCC family of chloride co-transporters as regulators of [Cl − ] i in SCN neurons. Therefore, our results add to a growing number of studies that point to the importance of [Cl − ] i in SCN function.

Methods
Animal strains and housing. Cl − imaging experiments were performed with C57BL/6 mice in which a floxed Cl-Sensor allele was inserted into the Rosa26 locus 33 . We crossed these mice with either AVP-IRES-Cre 34 or VIP-IRES-Cre mice 34,35 to yield AVP::Cl-Sensor or VIP::Cl-Sensor mice. Tail snips were sent to an external facility for genotyping (Transnetyx, Inc). Mice were heterozygous for both the Cl-Sensor and Cre transgenes. Tissue was prepared from adult male and female mice between two and six months old. Electrophysiological experiments were performed on wild type male Wistar rats aged 3 to 9 months.
All animals were entrained to 12:12 light-dark (LD) cycles, with the time of lights on represented as ZT 0. All procedures were approved in advance by the Institutional Animal Care and Use Committee of Oregon Health and Science University, and all experiments were performed in accordance with the approved animal protocol.
Confocal microscopy of fixed tissue. Each mouse was deeply anaesthetized with isoflurane and transcardially perfused with 10 mL of phosphate buffered saline (PBS) followed by 10 mL of 4% paraformaldehyde solution in PBS (pH 7.4). The brain was removed and post-fixed for 1-2 hours at 4 °C in the same solution. After repeated washes in 0.1 M PB, the brain was blocked and secured to a vibratome insert with cyanoacrylate adhesive and agarose supports. Coronal (40 µm thick) sections were cut with a Leica vibratome in 0.1 M PB and subsequently washed in the same buffer. For optical clearing, we treated the tissue with a glycerol/0.1 M PB gradient (25% to 90%). The tissue was incubated in each solution at 4 °C with light agitation until equilibrated. After clearing, sections were transferred into a 10% glycerol/0.1 M PB solution and counterstained with DAPI. Tissue sections were transferred onto glass slides in 10% glycerol solution and the cover glass was mounted with ProLong Diamond media after removing the excess buffer. Images were taken with a Zeiss Axioskop 2 TM fluorescent microscope using AxioVision 4.8 software (Carl Zeiss MicroImaging, Inc.). Confocal micrographs consisted of several 0.4 μm thick optical sections adjusted for optimal brightness and contrast using FIJI software.
Acute slice preparation. During their light phase (ZT 1-3 for day experiments and ZT 10-12 for night experiments), animals were removed from their housing chambers, anaesthetized with isoflurane, and decapitated. The brain was quickly removed and submerged in an ice-cold slicing solution consisting of (in mM): 111 NaCl, 26 NaHCO 3 , 11 dextrose, 6 Na-gluconate, 4 MgCl 2 , 3 KCl, 1 NaH 2 PO 4 , and 0.5 CaCl 2 , saturated with 95% O 2 , 5% CO 2 . The brain was blocked and 175 μm thick coronal slices were prepared with a Leica VT1000S vibrating blade microtome. Slices were incubated in slicing solution for 1-4 hours at 34 °C before recording.
Experiments were performed during ZT 2-8 for 'day' mice and ZT 12-18 for 'night' mice. Cl-Sensor fluorescent imaging was performed using epifluorescent methodology similar to that described by Friedel et al. 69 . Excitation light was supplied by a monochrometer (Polychrome IV; Till Photonics) providing 10 nm bandwidth output. Excitation at 500 nm preceded excitation at 436 nm to promote Cl-Sensor photostability 69 . Excitation duration ranged from 20 to 200 ms, with the excitation at 500 nm 1.5 times longer than that at 436 nm in order to obtain similar intensity values. The fluorescent signal passed through a double bandpass emission filter [470(24) + 535(30) nm] (Chroma Technology Corporation). Imaging was performed with an upright fluorescent microscope and a 63x water-immersion objective, NA 0.90 (Leica). Images were acquired with a charge-coupled device camera (CCD, ORCA-ER 12 bit level; Hamamatsu). Camera gain was set to 100, and binning was 4 × 4. Equipment control and image processing were performed with Metafluor software (Molecular Devices). Regions of interest (ROI) were defined around neuronal soma, and a dim region of the field of view was selected as background. Background values were subtracted for each wavelength independently. Because the YFP moiety of Cl-Sensor is quenched by Cl − ions, we choose to plot the emission following 436 nm excitation over that at 500 nm excitation so that R Cl would be a proxy for [Cl − ] i , with a higher ratio indicating higher [Cl − ] i .
When sampling at 2 and 5 second intervals, an exposure-dependent increase of R Cl was observed, most likely due to an accumulation of Cl-Sensor's YFP moiety in an inactivated state 69 . To correct for this instability, we fit our data with a single-exponential function, and subtracted the time-dependent component. Despite this correction, we observed that steady-state R Cl remained sensitive to exposure duration. We corrected for this exposure-dependent trend in each condition (AVP+ day, AVP+ night, VIP+ day, VIP+ night) independently in order to avoid any assumptions about the similarities of baseline R Cl across conditions. All values were adjusted to a 500 nm exposure of 100 ms.

Calibration of Cl-Sensor and estimation of [Cl
− ] i . In order to relate R Cl to [Cl − ] i , it was necessary to construct a calibration curve. Cl-Sensor was calibrated using a 0 mM Cl − solution consisting of (in mM): 120 Na-gluconate, 26 NaHCO 3 , 11 dextrose, 2.7 K-gluconate, 2 Ca-gluconate, 1 MgSO 4 , and 1 NaH 2 PO 4 , saturated with 95% O 2 , 5% CO 2 . This solution was mixed with aCSF to produce solutions of 0, 4, 20, 40, 60, and 80, and 123 mM Cl − . 50 μM β-escin was added to all calibration solutions to permeabilize cells. AVP::Cl-Sensor and VIP::Cl-Sensor day and night-entrained mice were used for analysis. R Cl values were corrected for exposure as discussed above. Average steady-state R Cl was plotted against Cl − concentration ( Supplementary Fig. 1) and calibration data was fit with the following logistic dose-response sigmoidal curve: Where R Cl is the fluorescence intensity ratio (F 436 /F 500 ) for chloride, K d is the dissociation constant for Cl − binding, R min and R max are the minimum and maximum asymptotic values of R Cl , and p is the Hill coefficient 33 . To obtain estimates of [Cl − ] i , we re-arranged the equation for [Cl − ] i :  Supplementary Fig. 1). Our K d is considerably higher than previously-reported values 33,70,71 . We also observed substantial variability in R Cl at specific Cl − concentrations between calibration experiments. Furthermore, R Cl was fairly non-linear in our range of operation (see Supplementary Fig. 1; our R Cl values were generally between 1.0 and 1.3). For these reasons, we elected to leave fluorescent measurements in values of R Cl instead of converting them into estimates of [Cl − ] i in subsequent analysis.
Perforated-patch electrophysiology. Brain slices were prepared as above. Recordings were performed with an Axopatch-1D amplifier (Axon Instruments), filtered at 2 kHz, digitized at 5 kHz, and acquired with Patchmaster v5.3 (HEKA Elektronik). Pipette solution contained (in mM): 120 KCl, 20 K-gluconate, 15 HEPES, 2 NaCl, 1 EGTA and either 0.2 of Lucifer Yellow or Texas Red. pH was adjusted to 7.26 with KOH. Gramicidin (Sigma) was dissolved in DMSO to a concentration of 50 mg/mL, aliquoted, and frozen. Before an experiment, this stock solution was diluted to a final pipette concentration of 30-100 μg/mL. A drop of gramicidin-free pipette solution was first applied to the backend of the pipette. After capillary action filled the pipette tip, the pipette was back-filled with the gramicidin-containing solution moments before submersion into the recording chamber. After gigaseal formation, series resistance (R s ) was monitored with a −5 mV voltage step to monitor the progress of perforation. Only recordings with a R s less than 100 MΩ were used for experiments. Cells were voltage clamped at −60 mV and cells with holding currents less than −30 pA were discarded. E GABA was determined using voltage ramp protocols. Every 10 seconds, a 400 ms voltage ramp protocol (ΔV ≅ 60 mV) was executed 100 ms after puff-application of 1 mM GABA. A current trace recorded in the absence of GABA was subtracted from currents obtained in the presence of GABA. The subtracted current trace was then plotted against the ramp command potential, and E GABA was recorded as the x-intercept. E GABA was not corrected for R s or liquid junction potentials.
Drugs. All drugs used in this study were acquired from Tocris Bioscience. Bumetanide and VU0240551 were dissolved in DMSO, stored as 10 mM stock solutions, and applied through the bath at 10 μM. A 100 mM stock of isoguvacine in water was diluted in aCSF to 1 mM and focally applied (5 psi) through a micropipette connected to a Picospritzer (General Valve Corporation).

Statistics and analysis.
Igor Pro (Version 6.22 A; Wavemetrics) was used for plotting, curve-fitting and data analysis. Data are presented as the mean ± standard error. For imaging experiments, generalized estimating equations (GEE) were used to determine statistical significance 72 . GEE models are similar to ANOVA and general linear models in that they estimate a mean response, except standard errors are adjusted for clustered or correlated measurements that originate from multiple observations made from the same brain slice. Therefore, each brain slice is treated as an independent measure, while the ROIs influence the average and standard error of each experiment. GEE test statistics are based on chi-square or z-statistics rather than F-and t-distributions. When comparing drug effects between HEPES-buffered and bicarbonate-buffered solutions, we formed z-statistics equal to the difference between the estimated effects from each separate GEE model (one for each solution) divided by the standard error of the difference. Significance level was Bonferroni-adjusted for these comparisons. For electrophysiology experiments, we used the Student's t-test. For all tests, a p-value less than 0.05 was considered to be statistically significant.
Data availability. The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.