Cell Type-Specific Expression of Acid-Sensing Ion Channels in Hippocampal Interneurons

Acid-sensing ion channels (ASICs), a member of the degenerin/epithelial Na+ channel superfamily, are widely expressed in the mammalian CNS. Accumulating evidence suggests that ASIC current density is higher in GABAergic interneurons than that in glutamatergic pyramidal neurons (PNs) in the hippocampus. Such differential expression of ASICs in cortical networks is thought to be a key element for seizure termination. However, GABAergic interneurons are highly diverse; it is unclear whether the functional expression of ASICs differs in distinct GABAergic interneuron subtypes. Moreover, the subunit composition of ASICs in individual GABAergic interneurons remains unknown. By combining patch-clamp recording and single-cell reverse transcription (RT)-PCR analysis, we correlated ASIC currents with their gene expression in acute rat hippocampal slices. The results yielded several surprising findings. First, ASIC current density of oriens lacunosum-moleculare (O-LM) cells in the CA1 region, a classical type of dendrite-targeting interneuron, is 6 times greater than that of fast-spiking basket cells (BCs) in the dentate gyrus, a major class of soma-targeting interneuron. Second, the recovery of ASICs from desensitization is slowest in BCs, intermediate in PNs, and fastest in O-LM cells. Third, the tarantula venom psalmotoxin 1, the specific blocker for ASIC1a homomers, inhibits ASIC currents in BCs but not in O-LM cells. Finally, single-cell RT-PCR analysis reveals coexpression of ASIC1a and ASIC2 subunit transcripts in O-LM cells, whereas only ASIC1a subunit transcript is detected in most BCs. Thus, differential expression of ASICs in inhibitory microcircuits likely contributes to the distinct roles of GABAergic interneurons in normal physiology and pathophysiology.


Introduction
Acid-sensing ion channels (ASICs), a member of the degenerin/ epithelial Na ϩ channel (DEG/ENaC) cation channel superfamily, are widely expressed in the mammalian nervous system (Waldmann et al., 1997; for review, see Wemmie et al., 2006) and have been implicated in pain, ischemic stroke, seizures and many other neurological diseases (Xiong et al., 2004;Mazzuca et al., 2007;Ziemann et al., 2008;Coryell et al., 2009). Many studies showed that GABAergic inhibitory interneurons in the hippocampus have larger ASIC current densities than glutamatergic excitatory pyramidal neurons (PNs) (Bolshakov et al., 2002;Cho and Askwith, 2008;Ziemann et al., 2008). In line with this view, a recent report showed that ASIC1a null mice display delayed seizure termination after seizure induction and suggested that ASIC activation by falls in extracellular pH during seizures increases the cortical inhibition through preferential recruitment of GABAergic inhibitory interneurons (Ziemann et al., 2008).
In this study, we correlated the functional property with geneexpression profile of ASICs in the PNs in the CA1 region and two functionally distinct classes of GABAergic inhibitory interneurons-O-LM cells and fast-spiking BCs in rat hippocampal slices by combining patch-clamp techniques and single-cell reverse transcription (RT)-PCR (Lien et al., 2002;Liss et al., 2004). Using these approaches, we uncovered a previously unknown expression pattern of ASICs in GABAergic inhibitory microcircuits, which likely underlies the selective vulnerability of distinct types of GABAergic inhibitory interneurons under pathological conditions.

Materials and Methods
Patch-clamp techniques in hippocampal slices. Transverse hippocampal slices of 300 m thickness were cut from the brains of 15-to 23-d-old male Sprague Dawley rats using a vibratome (DSK-1000, Dosaka) as described previously (Lien et al., 2002;Liao and Lien, 2009). Animals were sacrificed by decapitation in agreement with national and institutional guidelines and all procedures were approved by the Animal Care and Use Committee of National Yang-Ming University. Slices were sectioned in the ice-cold cutting saline containing the following (in mM): 87 NaCl, 25 NaHCO 3 , 1.25 NaH 2 PO 4 , 2.5 KCl, 10 glucose, 75 sucrose, 0.5 CaCl 2 and 7 MgCl 2 . Following sectioning, slices were incubated in the cutting saline (oxygenated with 95% O 2 /5% CO 2 ) in a holding chamber at 34°C for 30 min, and then at room temperature until used. During experiments, an individual slice was transferred to a submersion recording chamber and was continuously superfused with oxygenated artificial CSF containing the following (in mM): 125 NaCl, 25 NaHCO 3 , 1.25 NaH 2 PO 4 , 2.5 KCl, 25 glucose, 2 CaCl 2 and 1 MgCl 2 .
Nucleated patch recordings. Nucleated patch recordings were made as described previously (Lien et al., 2002;Lien and Jonas, 2003) using an Axopatch 200B amplifier (Molecular Devices). Pipette capacitance was compensated. Signals were low-pass filtered at 5 kHz (four-pole Bessel), and sampled at 10 kHz using a digitizer Digidata 1322A (Molecular Devices). Pulse sequences were generated by Digidata 1322A via pClamp 9.2 (Molecular Devices). Minor and major axes of nucleated patches were measured. It was assumed that nucleated patches were approximately ellipsoid (Gentet et al., 2000) and the membrane surface area was calculated using the following formula: Surface area ϭ (major axis ϩ minor axis) 2 (/4) (1) Fast application of H ϩ . Fast application of H ϩ on nucleated patches isolated from identified neurons was performed as described previously (Koh et al., 1995). Fast application experiments were started 1-2 min after the patches were excised. Double-barreled application pipettes were fabricated from theta glass capillaries (2 mm outer diameter, 0.3 mm wall thickness, 0.12 mm septum, Hilgenberg GmbH), and mounted on a piezoelectric-based solution switching system (Burleigh LSS-3000, EXFO). The time necessary for complete exchange of solution was measured with an open patch pipette by switching between Na ϩ -rich and 20% Na ϩ -rich solutions. It was 227 Ϯ 50 s (n ϭ 3) by measuring 20 -80% rise time of junction potential change. ASIC currents evoked by 2 s pulses of H ϩ were applied every 15-30 s except in some pharmacological experiments (see Fig. 5), short pulses (50 ms) were used.
Focal H ϩ puffing to dendrites. To visualize dendrites, we filled neurons with red fluorescent dye sulforhodamine 101 (SR101; 10 M) via somatic recording. The dendrites were traced under epifluorescence microscope. To evoke ASIC currents along the dendrites, the puffing pipette was placed close to the dendrites (see Fig. 7). Voltage changes or ASIC currents were evoked by focal puffs of H ϩ (ϳ6 psi) using PicoSpritzer III (Parker Instrumentation) in whole-cell current or voltage-clamp recording. Acidic pH solution was injected via a patch pipette with an open tip (2 m). The PicoSpritzer was triggered by an external TTL pulse generated by Digidata 1322A. This approach approximates the ASIC current amplitudes at dendrites and the caveat should be pointed out here. It is clear that the current amplitude did not directly reflect channel density because the surface area exposed to the puff varied.
Morphological analysis. Morphological reconstruction was identical to those reported previously (Liao and Lien, 2009). A separate subset of CA1 pyramidal cells, O-LM cells and BCs were filled with biocytin (0.2%) during recordings. After ϳ30 min recording, slices were fixed overnight with 4% paraformaldehyde in phosphate-buffered solution (PB; 0.1 M, pH 7.3). Following wash with PB, slices were incubated with fluorescein isothiocyanate (FITC)-conjugated avidin-d (2 l/ml; Invitrogen) in PB and 0.3% Triton X-100 overnight at 4°C. After wash, slices were embedded in mounting medium Vectashield (Vector Laboratories). Labeled cells were examined by a two-photon microscope using a pulsed titanium/sapphire laser (Chameleon-Ultra II tuned to 800 nm; Coherent) attached to a Leica DM6000 CFS that was equipped with a 20ϫ/0.50 numerical aperture water-immersion objective (HCX APO L; Leica). The two dimensional morphologies of the cells were reconstructed from a stack of 38 -455 images (voxel size, 0.378 -1.514 m in the x-y plane; 0.4 -0.99 m along the z-axis) using ImageJ (v. 1.42q).
Solutions and drugs. The HEPES-buffered Na ϩ -rich external solution used for fast application in the control barrel contained (in mM): 135 NaCl, 5.4 KCl, 1.8 CaCl 2 , 1 MgCl 2 , 5 HEPES; pH adjusted to 7.4 with N-methyl-D-glucamine (NMDG). To evoke ASIC currents with various pH values, MES-buffered Na ϩ -rich external solution in the test barrel contained (in mM): 135 NaCl, 5.4 KCl, 1.8 CaCl 2 , 1 MgCl 2 , 10 MES, adjusted to the desired values with NMDG. The intracellular solution contained (in mM): 135 K-gluconate, 20 KCl, 0.1 EGTA, 2 MgCl 2 , 4 Na 2 ATP, 10 HEPES; pH adjusted to 7.3 with HCl. For low Na ϩ experiments, Na ϩ in the buffered Na ϩ -rich external solution was substituted by NMDG. For low Na ϩ /high Ca 2ϩ experiments, concentrations of NaCl and CaCl 2 were varied as indicated (see Fig. 3C). Bovine serum albumin (0.1%) was added in external solutions containing the spider toxin Psalmotoxin 1 (PcTX1; Peptides International) to prevent its absorption to tubing and containers. Amiloride from Tocris Bioscience was dissolved in water to give 10 mM stock solution and stored at Ϫ20°C. All other chemicals were from Sigma except where noted.
Data analysis and statistics. Data were analyzed and fitted with Clampfit 10.0 (Molecular Devices) and Mathematica 5.2 (Wolfram Research). Apparent input resistance was defined by the ratio of peak voltage change/1 s hyperpolarizing current (Ϫ100 pA). Desensitization and deactivation time constants of ASIC currents were obtained by fitting currents with the function:

I͑t͒ ϭ Ϫt
Ae decay ϩ C where A denotes the peak amplitude of current, decay represents the desensitization or deactivation time constant, and C denotes the amplitude of steady-state current. Concentration-response curves were fitted with the function: where A is the constant for the maximal effect, c denotes the concentration, EC 50 represents the half-maximal effective concentration, and n denotes the Hill coefficient.
For measuring reversal potential, data points of I-V relations (see Fig.  3B) were fitted with second order polynomials, from which the interpolated potentials were calculated. Theoretical reversal potential of sodium channels (E rev ) was plotted against external Na ϩ concentration according to the Nernst equation: where [Na ϩ ] o, [Na ϩ ] i are outer and inner Na ϩ concentrations and F, R, T have standard thermodynamic meanings (Hille, 2001). The single-channel conductance of ASIC was estimated by nonstationary fluctuation analysis (Hartveit and Veruki, 2007) of ASIC currents from 23 to 50 traces (see Fig. 6 A). To minimize errors due to rundown, the entire dataset was divided into nonoverlapping subsets of 7-10 traces (Engel and Jonas, 2005). Mean ensemble current and variances were determined for each subset and averaged. Mean variance ( 2 ) versus mean current ( I), for each sampling point, can then be obtained. To assign similar weights to all phases of the ensemble mean waveform, binning of mean I and corresponding 2 were used by dividing the mean current amplitude into a number of bins of equal amplitude (Hartveit and Veruki, 2007). The values of mean I and mean 2 within each bin were then further averaged. The averaged mean I was plotted against the averaged mean 2 and then fitted with the equation when they look like a parabola: developed from a binomial model, yielding values for apparent singlechannel current i, total number of ion channel N available for activation. b 2 is the variance of the background noise. In all BCs (n ϭ 5) and some PNs (3 of 7), the data can be adequately described with a straight line rather than a parabola. The slope of the variance-mean relation was plotted with the equation: The single-channel chord conductance ␥ was calculated as follows: The maximum open probability (P o, max ), corresponding to the fraction of available ion channels open at the time of the peak current (I peak ), can then be calculated from the following equation: Values indicate mean Ϯ SEM; error bars in figures also represent SEM.
Statistical significance among groups was tested using the nonparametric Kruskal-Wallis test. When it was significant, pairwise comparisons by the Wilcoxon rank-sum test were then carried out for each pair of groups. All tests were performed at the significance level ( P) indicated using GraphPad Prism 5.0. SEs of reversal potentials were obtained by analyzing data of individual experiments separately. SEs of parameters of doseresponse curves were calculated by a parametric bootstrap method (Efron and Tibshirani, 1998). A total of 500 artificial datasets were generated in which the original values were replaced by normally distributed random numbers with means and SEM identical to those of the original data points, and were fitted as the original dataset.
Single-cell RT-PCR. Single-cell gene expression profiles were analyzed using the single-cell RT-PCR approach as previously described (Martina et al., 1998;Lien et al., 2002;Liss and Roeper, 2004;Aponte et al., 2006). RNase-free patch-clamp buffer containing (in mM): 140 KCl, MgCl 2 , 5 HEPES, 5 EGTA (pH ϭ 7.3) was autoclaved before used. Patch-clamp capillaries were baked overnight at 220°C before use. The cytoplasm of a recorded neuron was harvested into the recording pipette, under visual control, without losing the gigaseal immediately after electrophysiological characterization. The contents of the patch pipette (ϳ2-5 l) were expelled into a 0.2 ml PCR tube (Axygen Scientific) containing 17 l of RT mix (SuperScript III platinum Two-Step qPCR Kit with SYBR green I, Invitrogen). The mix contained 5 l of diethylpyrocarbonate (DEPC)-treated water, 10 l of 2ϫ RT reaction mix (oligo(dT) 20 , random hexamers, MgCl 2 and dNTPs) and 2 l of RT enzyme mix (SuperScript III reverse transcriptase and recombinant ribonuclease inhibitor). Total volume was ϳ20 l. After a series of incubation at different temperatures according to the manufacturer's instructions, the cDNA-containing tube was stored at Ϫ70°C until used. The RNA strand in the RNA-DNA hybrid was then removed by adding 1 l (2 U/l) of E. coli RNase H and incubated at 37°C for 20 min before PCR process.
Following cDNA synthesis, the cDNA solution of a single cell was split into 3 aliquots (5 l) and each of them was used for gene amplification. A PCR approach with Rotor-Gene 3000 (Corbett Research) or StepOn-ePlus Real-Time PCR system (Applied Biosystems) was performed in a total volume of 25 l with 1 l of 10 M primers (each), 5.5 l of DEPC-treated water and 12.5 l of Platinum SYBR Green qPCR SuperMix-UDG (Invitrogen) containing Platinum TaqDNA polymerase, Mg 2ϩ , uracil DNA glycosylase, proprietary stabilizers, and dNTPs with dUTP. Primers were designed with Perlprimer v1.1.14 and selected for maximal specificity and intron-overspanning amplicons ( Table 1). The molecular weights of the ASIC1a and ASIC2 amplicons and other neuron-specific markers were subsequently examined on ethidium bromide-stained agarose gels; sizes were in close agreement with the expected lengths (see Fig. 4 A). For ASIC expression profiles, all analyzed neurons expressed neurofilament 3 (NF3), indicating selective harvesting from neurons. To exclude the possibility of contaminations, no template control was performed in parallel to every PCR, and the reverse transcriptase was omitted in a subset of cells. Additional controls to exclude nonspecific harvesting were performed by advancing pipettes into the slice and taking them out without seal formation and suction (Martina et al., 1998;Lien et al., 2002).

Functional ASICs are differentially expressed in hippocampal microcircuits
We made whole-cell patch recordings from neurons in acute rat hippocampal slices under IR-DIC optics (Fig. 1 A). Neurons in the CA1 area and the DG were selected as previously described (Martina et al., 1998;Lien and Jonas, 2003;Aponte et al., 2008;Liao and Lien, 2009). Under current-clamp configuration at 23 Ϯ 1°C, CA1 PNs generated regular and accommodating AP trains upon injection of 1 s depolarizing current (ϩ600 pA) pulses and exhibited sag responses in response to 1 s hyperpolarizing current (Ϫ300 pA) pulses with apparent input resistance of 176 Ϯ 9.1 M⍀ (n ϭ 30, Fig. 1 A, middle). Under the same protocol, O-LM cells generated APs at a frequency Ն50 Hz and showed pronounced sag responses with apparent input resistance of 335.6 Ϯ 25 M⍀ (n ϭ 24, Fig. 1 A, left). In contrast to PNs and O-LM cells, BCs in the DG generated high-frequency AP trains (86.7 Ϯ 2.9 Hz, n ϭ 18) with negligible sag responses and exhibited relatively low input resistance (133.9 Ϯ 5 M⍀, n ϭ 18; Fig. 1 A, right). In a subset of experiments, the neuron types were further confirmed by post hoc morphological reconstruction of the recorded cells ( Fig. 1 B) and cell type-specific markers (supplemental Fig. 1, available at www. jneurosci.org as supplemental material). Consistent with previous reports (Sík et al., 1998;Maccaferri et al., 2000;Klausberger et al., 2003), analyses of single-cell RT-PCR showed that CA1 PNs expressed calcineurin, whereas O-LM cells and BCs expressed somatostatin and parvalbumin, respectively (supplemental Fig. 1, available at www.jneurosci.org as supplemental material). Moreover, nucleated patches subsequently isolated from these three types of neurons after whole-cell recordings exhibited cell type-specific voltage-gated K ϩ currents (Du et al., 1996;Martina et al., 1998;Lien et al., 2002). Total outward K ϩ currents in nucleated patches from CA1 PNs consisted of a large proportion of A-type currents, whereas patches from O-LM cells or BCs exhibited a predominant sustained component (supplemental Fig. 2, available at www.jneurosci.org as supplemental material).

Non-fast-spiking interneurons in the DG have relatively large ASIC currents
We next investigated whether the feature of low ASIC current is unique to fast-spiking (FS) cells. In the DG, we classified all re- corded non-fast-spiking interneurons into two functional groups (stuttering and accommodating) based on their firing patterns (Markram et al., 2004) (Fig.  2 A). Nucleated patches from both stuttering (STUT) and accommodating (AC) interneurons had significantly larger ASIC currents than FS BCs (Fig. 2 B, C). Overall, the ASIC currents in STUT and AC interneurons were 406 Ϯ 56 pA (n ϭ 10) and 399 Ϯ 36 pA (n ϭ 15), respectively (Fig.  2C). Also, both STUT and AC interneurons had higher input resistance than FS BCs. The scatter plot of the ASIC current against the input resistance revealed that FS BCs are featured by relatively low input resistance and low ASIC currents (Fig. 2 D).

ASIC gating is dependent on cell type
Steep pH dependence, Na ϩ selectivity, and blockade by amiloride and extracellular Ca 2ϩ ions are hallmark properties of ASIC gating (Waldmann et al., 1997;Bolshakov et al., 2002;Immke and McCleskey, 2003;Askwith et al., 2004;Wemmie et al., 2006;Cho and Askwith, 2008;Ziemann et al., 2008). To test pH dependence, we recorded ASIC currents upon different extracellular pH reductions. As represented by Figure 3A, the magnitude of ASIC current in an O-LM cell depended on the pH value. The doseresponse curve was fitted with the equation (Eq. 3), yielding the pH value of 6.0 Ϯ 0.0 for the half-maximal activation and the Hill coefficient of 1.2 Ϯ 0.0. We further determined the ionic permeability of ASICs by measuring the reversal potentials (E rev (Fig. 3C, gray line; also see Eq. 4). These results indicate that Na ϩ is the major permeant ion for ASICs. In addition to the sensitivity to amiloride (Fig. 1C), extracellular Ca 2ϩ was shown to compete with H ϩ for the ASIC activation site (Immke and McCleskey, 2003). Consistent with this notion, we found that high [Ca 2ϩ ] o (18 mM) inhibited acid (pH 5)-induced currents by 60 Ϯ 2.3% (supplemental Fig. 3, available at www.jneurosci.org as supplemental material).
ASIC1a is a principal subunit for a functional ASIC, whereas ASIC2 subunit plays a modulatory role in the gating of ASIC, such as desensitization and recovery from desensitization (Askwith et al., 2004). We thus explored the potential differences in ASIC subunit composition of these three neuron types by measuring the desensitization time constant ( desen ), deactivation time constant ( deact ) and the recovery from de-sensitization of ASIC currents. The desen values of native ASICs from these three cell types were all Ͻ1 s (Fig. 3D). On the other hand, deact values from three cell types were all Ͻ10 ms (Fig. 3F ). Notably, both desen and deact of ASICs in O-LM cells were significantly different from those of PNs and BCs. Interestingly, the recovery from desensitization as assessed by the recovery time constant ( recovery ) was fastest ( recovery ϭ 0.96 Ϯ 0.00 s) in O-LM cells, intermediate ( recovery ϭ 13.2 Ϯ 0.1 s) in CA1 PNs and slowest ( recovery ϭ 42.8 Ϯ 1.0 s) in BCs (Fig. 3F ). Consistent with the recovery time course, cumulative desensitization of ASIC currents evoked by 0.05 Hz repetitive acid (pH 5) pulses was fastest in BCs, intermediate in PNs and slowest in O-LM cells (Fig. 3G). Together, these results strongly suggest differences in ASIC subunit composition among these three neuron types. responses to 1 s depolarizing (ϩ700, ϩ500 and ϩ300 pA) or hyperpolarizing (Ϫ100 pA) current pulses in the whole-cell current-clamp configuration from a FS BC (left) and two non-FS interneurons: middle, STUT firing; right, AC firing. Membrane potentials before the pulses were Ϫ70 mV. Inset, stimulation protocol. B, Inward currents recorded at Ϫ60 mV after rapid pH changes from pH 7.4 to pH 5 in nucleated patches from a FS BC (left) and non-FS interneurons (middle, STUT; right, AC) in the DG. Traces are average of 11-75 sweeps. C, Bar plot summarizing the amplitude of ASIC currents from FS BCs (n ϭ 6), STUT (n ϭ 10) and AC (n ϭ 15) interneurons. **p Ͻ 0.005. D, Scatter plot of the input resistance against the ASIC current from all FS and non-FS cells.

ASIC2 subunit determines the recovery of native ASICs from desensitization
To further correlate the gating properties with gene-expression profile of individual neurons, we analyzed ASIC1a and ASIC2 subunit mRNAs of each neuron type at the single-cell level (Fig.  4 A). Single-cell RT-PCR analysis revealed that ASIC1a and ASIC2 mRNAs were coexpressed in CA1 PNs and O-LM cells, whereas only ASIC1a mRNA was detected in most BCs (9 of 11 ASIC-expressing BCs; Fig. 4 B). These results correlate well with the notion that the rapid recovery of ASIC currents in O-LM cells is mediated by ASIC1a/2a heteromers, while the prolonged re-covery of ASICs from desensitization in BCs is largely attributed to the expression of ASIC1a homomers (Askwith et al., 2004). To make this point clear, we illustrated the recovery time courses of ASICs from O-LM cells (blue line) and BCs (red line) together with those of various recombinant ASICs ( Fig. 4C; change to curves obtained by fitting data from Askwith et al., 2004).

PcTX1 preferentially inhibits ASIC currents in basket cells
Homomeric ASIC1a channel is potently inhibited by the PcTX1 (Baron et al., 2002;Chen et al., 2005;Xiong et al., 2008). We, therefore, examined the effects of PcTX1 on functionally distinct  Ϯ 33 ms, n ϭ 16). The desen was obtained by fitting the decay phase of the current with a monoexponential function. ***p Ͻ 0.0005. E, Left, Representative peak-scaled traces of ASIC currents showing deactivation. Right, Bar plot of deact of ASIC currents of O-LM cells (3.7 Ϯ 0.3 ms, n ϭ 10), PNs (9.3 Ϯ 1.2 ms, n ϭ 5) and BCs (9.8 Ϯ 1.0 ms, n ϭ 5). The deact was obtained by fitting the decay phase of the current with a monoexponential function. **p Ͻ 0.005. F, Left, Representative traces showing the recovery from desensitization was measured at Ϫ60 mV after rapid changes from pH 7.4 to 5 for 2 s and then changed back to pH 7.4 for 20 s, followed by a second pulse to pH 5 for 2 s. The peak current evoked by the second pH 5 pulse divided by the first pH 5 pulse was plotted against the pulse interval. Right, data points obtained from O-LM cells (n ϭ 6), PNs (n ϭ 16) and BCs (n ϭ 9) were fitted with either a mono-or a biexponential function, yielding the weighted recovery time constants of 0.96 s, 13.2 s and 42.8 s, respectively. G, Left, Cumulative desensitization of ASICs (at Ϫ60 mV) evoked by repetitive 2 s pulses (from pH 7.4 to 5) at 0.05 Hz in an O-LM, a PN and a BC, respectively. Only the first five traces were shown. Right, Time courses of onset of desensitization in O-LM cells (n ϭ 6), PNs (n ϭ 10) and BCs (n ϭ 7). The peak current amplitude, normalized to the first peak current amplitude, was plotted against the time of each pulse. Curves represent exponential functions fitted to the data points. Time constants for PNs and BCs were 23.6 Ϯ 1.8 and 8.3 Ϯ 0.3 s, respectively.

ASIC currents in nucleated patches from either O-LM cells or
BCs. Because of the very slow recovery time course ( recovery ϭ 42.8 s) of ASICs in BCs, we thus obtained stable ASIC current amplitudes by applying short (50 ms) acid pulses to nucleated patches (Fig. 5). In control conditions, ASIC currents of O-LM cells and BCs showed little desensitization and were stable over time (Fig. 5, left traces). PcTX1 (30 nM) had only minimal effects on ASIC currents in nucleated patches from O-LM cells (Fig. 5,  top, right). The peak current in the presence of PcTX1 was 102 Ϯ 1% of the control (n ϭ 4, p Ͼ 0.05; Fig. 5, top, right). In contrast, PcTX1 (30 nM) significantly blocked 76 Ϯ 5% of the ASIC currents in nucleated patches from BCs (n ϭ 6, p Ͻ 0.005; Wilcoxon signed-rank test; Fig. 5, bottom, right).

ASIC channel number accounts for differences in current density
Finally, we determined whether the remarkable differences in current densities among the neuron types were derived from differences in channel number (N), maximum open probability (P o, max ) or single-channel conductance (␥). We retrieved N, P o, max and ␥ from the ensembles of macroscopic currents (Fig. 6 A) using nonstationary fluctuation analysis. The ensemble variance was calculated from the fluctuation of ASIC currents around the mean (Fig. 6 B), plotted against the ensemble mean, and fitted with a parabolic function (Fig. 6C). Analysis from patches of O-LM cells yielded the ␥ of 5.05 Ϯ 0.74 pS (n ϭ 7; Fig. 6 D) and the channel number N of 427 Ϯ 82 (n ϭ 7). For CA1 PNs, though the ␥ (6.57 Ϯ 1.05 pS, n ϭ 7) was comparable to that of O-LM cells, the estimated N was 69 Ϯ 45 (n ϭ 4; 3 of 7 cells were fitted with a linear function; see below), significantly less than that of O-LM cells ( p Ͻ 0.05, Wilcoxon rank-sum test). Notably, the variancemean relation of ASIC currents from all BCs (n ϭ 5) displayed a straight line rather than a parabola, indicating low open-channel probability (Alvarez et al., 2002). In this case, we can determine the ␥, but not the N and P o, max in the membrane (Alvarez et al., 2002). By fitting the data points with a linear function, we obtained the ␥ of 6.29 Ϯ 1.75 pS (n ϭ 5; Fig. 6 D) from the slope. Collectively, the ␥ values were not significantly different among the three groups ( p ϭ 0.72, Kruskal-Wallis test). Similarly, the P o, max values at pH 5 were not significantly different between O-LM cells and CA1 PNs (O-LM cells, 0.54 Ϯ 0.06, n ϭ 7 versus PNs, 0.48 Ϯ 0.08, n ϭ 4; p ϭ 0.65, Wilcoxon rank-sum test). In sum, ASIC channel number greatly contributes to the cell typespecific difference in current density.

Dendritic ASIC currents in O-LM cells are higher than those of PNs and BCs
ASICs are distributed at somatodendrites of neurons (Wemmie et al., 2002;Alvarez de la Rosa et al., 2003;Zha et al., 2006Zha et al., , 2009. After filling neurons with red fluorescent dye SR101 (10 M) via somatic recording, dendrites were traced under epifluorescent microscope. Similar to somatic ASIC currents, dendritic ASIC currents of O-LM cells (V-clamp at Ϫ70 mV) evoked by local acid puffing were transient and reversibly inhibited by amiloride (100 M) (Fig. 7A). Then we determined whether dendritic ASIC currents differed among three distinct types of neurons by local acid puffing along the apical dendrites of CA1 PNs (Fig. 7B), bitufted dendrites of O-LM cells (Fig. 7C) and apical dendrites of  BCs (Fig. 7D). ASIC currents in dendrites with the puffing site at distances of 30 -100 m from the soma (280 Ϯ 65 pA, n ϭ 8) of O-LM cells are much larger than those of CA1 PNs (132 Ϯ 28 pA, n ϭ 9) and BCs (52 Ϯ 8 pA, n ϭ 6) (Fig. 7E). Consistently, ASIC currents effectively generated AP trains in O-LM cells when neurons were recorded in the current-clamp configuration but only caused subthreshold membrane depolarization in PNs and BCs under the same conditions (Fig. 7F ).

Discussion
Earlier studies reported that GABAergic inhibitory interneurons possess larger ASIC current density than pyramidal cells in the hippocampus (Bolshakov et al., 2002;Cho and Askwith, 2008;Ziemann et al., 2008). In contrast to them, our study shows that fast-spiking, parvalbumin-expressing BCs have comparable ASIC current density to CA1 PNs. Moreover, ASIC current densities and gating greatly differ between two functionally distinct GABAergic (perisomatic versus dendritic) inhibitory interneurons. Furthermore, single-cell RT-PCR analysis reveals coexpression of ASIC1a and ASIC2 subunits in O-LM cells, whereas only ASIC1a is detected in most BCs. The lack of the ASIC2 subunit likely accounts for the prolonged recovery of native ASICs from desensitization in BCs.

Comparisons between native and recombinant ASICs
Heterologously expressed mouse ASIC1a homomers and ASIC1a/2 heteromers in CHO cells desensitize rapidly ( decay Ͻ 1 s) (Askwith et al., 2004). In comparison with them, the values of desen of native ASICs from all three cell types are Ͻ1 s (Fig. 3D), suggesting that either ASIC1a homomers or heteromeric channels assembled by ASIC1a/2a or ASIC1a/2b combination likely mediate the majority of native channels. Further analyses of ASIC gating implicate differential expression of ASIC subunits in different cell types. Most obviously, the recovery from desensitization of ASICs is fastest in O-LM cells, intermediate in CA1 PNs and slowest in BCs (Fig. 3F ). These suggest that ASIC1a/2a heteromers likely mediate ASICs of O-LM cells because the recovery from desensitization of heterologously expressed mouse ASIC1a/2a heteromers as estimated by the recovery is Ͻ1 s, whereas the values of recovery of either recombinant ASIC1a homomers or ASIC1a/2b heteromers are relatively slow (Fig. 4C). Conversely, the very slow recovery of ASICs from desensitization in BCs strongly suggests that the majority of ASICs in BCs is assembled as ASIC1a homomers. This notion is strongly supported by the results of single-cell RT-PCR and pharmacological sensitivity to PcTX1. However, differences between native and recombinant ASICs still exist, despite that the recovery time course of native ASICs qualitatively recapitulates those of heterologously expressed ASICs in CHO cells. First, there are quantitative differences in the recovery time course of ASICs, particularly between the native ASICs in BCs ( recovery ϳ42.8 s) and the recombinant ASIC1a homomers ( recovery ϳ7 s) (Fig. 4C). Second, recombinant mouse ASIC1a/2a heteromers generate currents that desensitize faster than ASIC1a homomers (Askwith et al., 2004). In contrast, ASICs in O-LM cells (putative "ASIC1a/2a" heteromers) significantly desensitize slower than those (putative "ASIC1a" homomers) in BCs (Fig. 3D). These could be attributed to the differences in expression systems.
The estimated values of ␥ are not significantly different among the three groups. Notably, single-channel conductances of native ASICs compare well with those measured directly from unitary currents (from 4.2 pS under 10 mM [Ca 2ϩ ] o to 11.2 pS under 1 mM [Ca 2ϩ ] o ) (Immke and McCleskey, 2003). Also, the values of P o, max at pH 5 are not significantly different between O-LM cells and CA1 PNs. As for BCs, analysis of cumulative desensitization reveals that desensitization of ASICs in nucleated patches of BCs develops more rapidly than those of PNs and O-LM cells during repetitive application of acid pulses (Fig. 3G). This is in agreement with the suggestion that the open probability of ASICs in BCs is relatively low by the nonstationary fluctuation analysis.

Possible functions of ASICs in perisomatic and dendritic inhibitory interneurons
Although our understanding of interneuron heterogeneity is still far from being complete, a major dichotomy in the inhibitory control of pyramidal cells has already been established (Freund and Katona, 2007). Dendritic inhibitory interneurons (for instance, O-LM cells) are responsible for the control of the efficacy and plasticity of glutamatergic inputs from specific sources that terminate in the same dendritic domain (Miles et al., 1996). On the other hand, perisomatic inhibitory interneurons (for instance, BCs) control the output of pyramidal cells (Miles et al., 1996). In addition, several intrinsic properties (see Introduction of this study) of these two major types of interneurons are found to be very different. In this study, we uncover the expression of ASICs as a novel cell type-specific property which might underlie their differences in synaptic transmission and relative susceptibility to brain injury.
Several lines of evidence suggest that ASICs play crucial roles in synaptic transmission. First, long-term potentiation, a form of synaptic plasticity underlying learning and memory, is abrogated at Schaffer collateral-CA1 pyramidal cell synapses in ASIC1a knock-out mice (Wemmie et al., 2002). Second, presynaptic re-lease probability is increased in hippocampal neurons from ASIC1a knock-out mice (Cho and Askwith, 2008). Finally, ASICs interact with several postsynaptic scaffolding proteins such as PSD-95, PICK1 and CaMKII (for review, see Wemmie et al., 2006). More recently, ASIC2 subunits, but not ASIC1a subunits, were shown to target ASICs to the synapse via an association with PSD-95 and loss of ASIC2 subunits decreases the percentage of spines responding to acid (Zha et al., 2009). Interestingly, our results demonstrate that ASIC2 subunit mRNAs were not detected in most BCs and relatively small ASIC currents were detected in BC dendrites. Thus, subunit composition of ASICs in different types of GABAergic inhibitory interneurons might also contribute to cell type-specific long-term plasticity at glutamatergic synapses onto hippocampal interneurons (Nissen et al., 2010).
Deletion of ASIC1a gene was shown to increase the presynaptic release probability and led to paired-pulse depression at the hippocampal glutamatergic synapses (Cho and Askwith, 2008). Conversely, overexpression of ASIC1a subunits in ASIC1a knock-out neurons increases paired-pulse ratio. Thus, relative abundance of ASICs might underlie the heterogeneity of GABAergic inhibitory outputs (Maccaferri et al., 2000). Intrigu-  Ϫ70 mV), respectively. C, Left, A PN filled with SR101 and the puff pipette was placed near the apical dendrite. Right, Brief acid puffs (1 s, pH 4) at different distances along the PN dendrite evoked relatively smaller inward currents (V-clamp ϭ Ϫ70 mV) and subthreshold membrane depolarization in the current-clamp recording (holding potential Ϫ70 mV), respectively. D, Left, A BC filled with SR101 and the puff pipette was placed near the apical dendrite. Right, Local acid puffs (2 s, pH 4) evoked relatively smaller ASIC currents (V-clamp ϭ Ϫ70 mV) and subthreshold membrane depolarization in the current-clamp recording (holding potential Ϫ70 mV), respectively. E, ASIC currents as a function of distance from the soma. Lines represent exponential functions fit to the data points. Data are from 8 O-LM cells, 9 PNs and 6 BCs. F, Summary plot of spike number plotted against distance from the soma. Lines represent exponential functions fit to data points. Data are from 8 O-LM cells, 9 PNs and 6 BCs. All experiments were performed in the presence of 1 mM kynurenic acid and 1 M SR95531.
ingly, consistent with this notion, we found a similar correlation between the functional ASIC current and the release probability of interneurons. In this study, BCs with relatively low ASIC expression have high release probability and exhibit paired-pulse depression at short interspike intervals (Kraushaar and Jonas, 2000;Maccaferri et al., 2000). On the contrary, O-LM cells with relatively high ASIC expression have low release probability and show no paired-pulse modulation. (Maccaferri et al., 2000).

Functional implications
Interneurons exhibit differential susceptibility to injury by epileptic insults that are accomplished by transient acidification (Morin et al., 1998;Cossart et al., 2001;Wittner et al., 2001Wittner et al., , 2005. In experimental epilepsy, dendrite-targeting interneurons, such as O-LM cells, selectively degenerate, leading to "disinhibition" in dendritic compartments of PNs. In contrast, soma-targeting interneurons, mainly BCs, are relatively spared, resulting in enhanced somatic inhibition (Cossart et al., 2001;Wittner et al., 2005). Increasing studies showed that neurons lacking ASICs are resistant to ischemic/acid injury (Xiong et al., 2004;Gao et al., 2005;Wemmie et al., 2006). Our data show that BCs, in contrast to O-LM cells and other non-fast-spiking interneurons, have strikingly low-density and rapidly desensitizing ASICs. Furthermore, BCs have relatively low input resistance (ϳ134 M⍀, see Results) compared with O-LM cells (ϳ336 M⍀). Consequently, relatively large ASIC currents in O-LM cells during transient acidification together with the high input resistance will lead to membrane depolarization and firing of O-LM cells (Fig. 7). Thus, acidification can preferentially induce firing of O-LM cells accompanied by massive Ca 2ϩ influx via voltagegated Ca 2ϩ channels and thereby might have a greater impact on O-LM cells than BCs. Interestingly, in addition to the unique ASIC expression feature, fast-spiking BCs have very efficient Ca 2ϩ buffer capacity (Aponte et al., 2008). Quantitatively, the magnitude of Ca 2ϩ binding ratio ( s ) of BCs ( s ϳ200; Aponte et al., 2008) is tenfold larger than that of O-LM cells ( s ϳ20; Liao and Lien, 2009). Together, all these factors likely contribute to selective degeneration of O-LM cells and disinhibition of pyramidal cell dendrites induced by experimental epilepsy.

Perspectives
Our results point to the cell type-specific expression of ASICs in hippocampal GABAergic inhibitory microcircuits and the findings demonstrate how the distribution and proportions of ASIC subunits vary in different cells under physiological conditions. Our work provides a better understanding of how acidosis in the hippocampus could influence the network and how hippocampal acidosis might suppress seizure activity. A challenge of future studies will be to delete ASIC gene in a cell type-restricted manner to address the meaning of differential expression of ASICs in the brain.