Altered GABAA,slow Inhibition and Network Oscillations in Mice Lacking the GABAA Receptor β3 Subunit

Harald Hentschke, Claudia Benkwitz, Matthew I. Banks, Mark G. Perkins, Gregg E. Homanics, Robert A. Pearce


Phasic GABAergic inhibition in hippocampus and neocortex falls into two kinetically distinct categories, GABAA,fast and GABAA,slow. In hippocampal area CA1, GABAA,fast is generally believed to underlie gamma oscillations, whereas the contribution of GABAA,slow to hippocampal rhythms has been speculative. Hypothesizing that GABAA receptors containing the β3 subunit contribute to GABAA,slow inhibition and that slow inhibitory synapses control excitability as well as contribute to network rhythms, we investigated the consequences of this subunit's absence on synaptic inhibition and network function. In pyramidal neurons of GABAA receptor β3 subunit-deficient (β3−/−) mice, spontaneous GABAA,slow inhibitory postsynaptic currents (IPSCs) were much less frequent, and evoked GABAA,slow currents were much smaller than in wild-type mice. Fittingly, long-lasting recurrent inhibition of population spikes was less powerful in the mutant, indicating that receptors containing β3 subunits contribute substantially to GABAA,slow currents in pyramidal neurons. By contrast, slow inhibitory control of GABAA,fast-producing interneurons was unaffected in β3−/− mice. In vivo hippocampal network activity was markedly different in the two genotypes. In β3−/− mice, epileptiform activity was observed, and theta oscillations were weaker, slower, less regular and less well coordinated across laminae compared with wild-type mice, whereas gamma oscillations were weaker and faster. The amplitude modulation of gamma oscillations at theta frequency (“nesting”) was preserved but was less well coordinated with theta oscillations. With the caveat that seizure-induced changes in inhibitory circuits might have contributed to the changes observed in the mutant animals, our results point to a strong contribution of β3 subunits to slow GABAergic inhibition onto pyramidal neurons but not onto GABAA,fast -producing interneurons and support different roles for these slow inhibitory synapses in the generation and coordination of hippocampal network rhythms.


Neuronal ensemble activity in the hippocampus of exploring rodents is dominated by network oscillations in the theta (5–12 Hz) and gamma (40–90 Hz) range. This activity hinges on the characteristics of inhibitory interneurons, many of which discharge during specific phases of theta and gamma oscillations (Fuentealba et al. 2008; Klausberger et al. 2003; Penttonen et al. 1998; Tukker et al. 2007), inhibiting other interneurons and pyramidal neurons for periods ranging from a few milliseconds up to several hundred milliseconds (Banks et al. 2000; Chapman and Lacaille 1999; Cobb et al. 1995). By virtue of this phase-locked inhibition, interneurons orchestrate the interwoven hippocampal rhythms, including “nesting,” the amplitude modulation of gamma oscillations at theta frequency (Bragin et al. 1995; Penttonen et al. 1998; White et al. 2000).

A specific GABAergic conductance in hippocampal CA1, termed GABAA,slow (Banks et al. 1998; Pearce 1993), has a decay time of 30–70 ms and is able, in vitro, to silence interneurons producing fast inhibitory postsynaptic currents (IPSCs; GABAA,fast) for approximately one theta period (Banks et al. 2000). It has therefore been postulated to contribute to the nested coordination of theta and gamma rhythms (Banks et al. 2000; White et al. 2000). There is accumulating evidence that GABAA,slow, which also occurs in neocortex and subiculum, is generated by distal dendrite-targeting interneurons: neurogliaform neurons (Price et al. 2005, 2008; Szabadics et al. 2007), LM neurons (Ouardouz and Lacaille 1997), O-LM neurons (Pouille and Scanziani 2004), and ivy cells (Fuentealba et al. 2008).

We hypothesized that the different kinetics of GABAA,fast and GABAA,slow would reflect different subunit compositions of their postsynaptic GABAA receptors. Of particular interest is the β3 subunit, which is prominent in dendritic regions of hippocampus and dentate gyrus (Sperk et al. 1997). In other brain regions, it has been associated with slow GABAergic current kinetics: its absence in GABAA receptor β3 subunit-deficient (β3−/−) mice confers fast decay kinetics on IPSCs in thalamic reticular (Huntsman et al. 1999) and neocortical neurons (Ramadan et al. 2003). Importantly, the β3 subunit plays a crucial role in the coordination of network rhythms as has been shown in thalamus (Huntsman et al. 1999) and olfactory bulb (Nusser et al. 2001) of β3−/− mice. These findings suggest that in addition to controlling hyperexcitability and preventing seizures, GABAA receptors with the β3 subunit may occupy strategic positions in neuronal networks generating GABAA receptor-dependent rhythms. We used β3−/− mice to determine the role these receptors play at GABAA,slow synapses and their contribution to the generation of hippocampal rhythms.

We show that GABAA,slow synapses containing β3 subunits provide inhibitory input into pyramidal cells and that the absence of these currents in β3−/− mice translates into weaker and less regular theta oscillations as well as weaker and faster gamma oscillations. By contrast, slow inhibitory control of GABAA,fast-generating interneurons and amplitude modulation of gamma oscillations at theta frequency is maintained in the mutant mice. These results support a model in which the interaction between relatively independent inhibitory subcircuits oscillating at different frequencies generates the nested rhythms observed in hippocampus and other cortical structures.


All experimental protocols conformed to American Physiological Society/National Institutes of Health guidelines and were approved by the Institutional Animal Care and Use Committee of the University of Wisconsin and the Institutional Animal Care and Use Committee of the University of Pittsburgh.

Generation of β3 knockouts

Global β3 knockouts and wild-type littermate controls were produced from heterozygous breeding pairs and genotyped by Southern blot analysis as described (Homanics et al. 1997). Mice were of a mixed C57BL/6J × Strain 129X1/S1 genetic background of the F10-14 generations.

Preparation of hippocampal slices

Mice aged 30–35 days (120–180 days for field recordings) were decapitated under deep halothane or isoflurane and ketamine anesthesia, and transverse hippocampal slices (400 μm thick) were prepared according to standard procedures (Banks et al. 1998). They were allowed to recover in dissection buffer at room temperature (field potential recordings) or 35°C (IPSC recordings) for ≥1 h before transfer to the recording chamber, which was perfused at 3 ml/min with artificial cerebrospinal fluid [ACSF, composition (in mM): 127 NaCl, 1.2 KH2PO4, 1.9 KCl, 26 NaHCO3, 2.2 CaCl2, 1.4 MgSO4, and 10 glucose] saturated with 95% O2-5% CO2. Dissection buffer was identical to ACSF except for the addition of ascorbic acid (2.5 mM) and kynurenic acid (5 mM) in the case of field potential recordings.

Whole cell recordings of IPSCs

Putative pyramidal cells in stratum pyramidale (SP) of CA1 were visualized using a video camera (VE-1000; DAGE MTI, Michigan City, IN) connected to an upright microscope (BX-50WI; Olympus America, Melville, NY) equipped with an infrared band-pass filter [775 ± 75 nm], a long working-distance water-immersion objective (×40, NA 0.7) and differential interference contrast optics. The microscope and recording pipette were under remote control using an integrated motorized control system (Luigs and Neumann, Ratingen, Germany).

Whole cell recordings were obtained at room temperature (22–24°C) using a MultiClamp 700A (Axon Instruments, Union City, CA) patch-clamp amplifier. All data were collected using pClamp software (Axon Instruments). Data were filtered at 2–5 kHz and digitized at 5–10 kHz (Digidata 1200, Axon Instruments). Patch pipettes were fabricated from borosilicate glass (KG-33, 1.7 mm OD, 1.1 mm ID; Garner Glass, Claremont, CA) using a Flaming-Brown two-stage puller (P-87; Sutter Instruments, Novato, CA), fire polished and coated with silicone elastomer (Sylgard, Dow Corning) to reduce electrode capacitance. Patch pipettes had open-tip resistances of 2–4 MΩ when filled with the recording solution (composition, in mM: 140 CsCl, 10 Na-HEPES, 10 EGTA, 2 MgATP, and 5 QX-314, pH 7.3). Access resistances were typically 10–20 MΩ and were then compensated 60–80%. Cells were held at −60 mV. GABAA IPSCS were isolated by bath application of 20 μM 6-cyano-7-nitroquinoxalene-2,3-dione (CNQX) and 40 μM d,ll-2-amino-5-phosphonovaleric acid (d,l-APV) to block AMPA and N-methyl-d-aspartate (NMDA)-mediated currents and by the inclusion of CsCl and QX-314 in the patch pipette, both of which block, among other potassium currents, GABAB receptor-mediated currents (Nathan et al. 1990).

Evoked IPSCs were elicited by applying stimuli (1–100 μA) to SP and s. lacunosum-moleculare (SLM) using patch electrodes filled with ACSF. SP stimulating electrodes were placed as close to the recorded cell body as possible, and stimuli were applied at a rate of 0.2 Hz. For SLM stimuli, a maximum stimulation rate of 0.05 Hz was used to minimize the previously observed rundown of GABAA,slow over time (Pearce 1993). SLM stimulating electrodes were consistently placed at ∼50 μm on the SLM side of the hippocampal fissure (HF), ∼100 μm deep in the tissue and at the same mediolateral position in CA1 as the apical dendrite of the cell being recorded. In those cells in which no evoked GABAA,slow IPSC was observed, the electrode was repositioned within SLM until a response could be elicited or (more typically) a region ∼100 × 100 μm in SLM had been searched to no avail. In all cases, maximal currents were determined by varying stimulus intensity until the response amplitude no longer increased.

IPSC analysis

Spontaneous GABAergic IPSCs were detected using an automated event detection algorithm based on a “pseudo-differentiation” as described previously (Banks et al. 2000). GABAA,slow IPSCs were well fit with the sum of single rising and decaying exponential components. Although GABAA,fast IPSCs decayed biexponentially (Banks et al. 1998), the decay was characterized using the weighted sum of these two exponential components (τDecay, wt). Spontaneous GABAA,slow IPSCs were defined as those events having 10–90% rise times >4 ms and decay times >40 ms. Individual spontaneous IPSCs were selected for averaging and exponential curve fitting when no other detected events occurred within ±100 ms (GABAA,fast) or ±250 ms (GABAA,slow) of the peak.

Stimulation in SLM inhibits interneurons producing GABAA,fast IPSCs for several hundred milliseconds poststimulus (Banks et al. 2000). We quantified this suppression of fast inhibition (SFI) in terms of changes in the frequency of GABAA,fast IPSCs recorded in pyramidal neurons. As action-potential-independent miniature IPSCs persist during SFI (Banks et al. 2000), the frequency of IPSCs underestimates the extent of SFI. Therefore we also quantified the average “instantaneous” poststimulus current in pyramidal neurons carried by GABAA,fast IPSCs. To this end, the amplitudes of all detected GABAA,fast IPSCs were determined by subtracting from the peak IPSC amplitude the mean of the current trace in an interval of [−2.5 to −1.5] ms relative to the peak. Note that this procedure was robust to changes in the base line or to the presence of GABAA,slow IPSCs but underestimated the amplitudes of IPSCs preceded by other GABAA,fast IPSCs. Each peristimulus current trace was reconstructed by adding artificial IPSCs of assessed amplitudes and uniform decay times (16 ms for wild-type animals, 14 ms for β3−/− animals, see Fig. 2) to a zero baseline. All sweeps of one experiment were averaged and then normalized to the average current in the recorded prestimulus interval, which was either [−400 to 0] or [−800 to 0] ms. The extent of SFI was quantified by the average normalized current in a poststimulus interval of 50–300 ms. Due to the long duration of SFI, the method was insensitive to the exact value of the decay time: variations of this parameter by ±2 ms (>10%) resulted in <1% change of the average normalized poststimulus current.

Conditioned depression and field recordings

Experiments were performed at room temperature (22–24°C). Pipettes were similar to those used for whole cell recordings but were filled with ACSF (resistance: 2–4 MΩ) to record field potentials. Bipolar stimulating electrodes were fabricated from tungsten (Microelectrodes Tungsten, World Precision Instruments, Sarasota, FL). For the conditioned depression paradigm (Fig. 3A) (Benkwitz et al. 2007; Pouille and Scanziani 2004), the recording electrode was placed in the CA1 pyramidal layer. One stimulating electrode was placed in alveus, lateral of the recording electrode, to activate recurrent (feedback) inhibition (conditioning pulse). The second electrode was placed in s. radiatum (SR) to activate Schaffer collateral inputs and thus evoke population responses (population spike) in pyramidal neurons. Conditioned responses were obtained at interstimulus intervals ranging from 5 to 2,000 ms and compared with the unconditioned response, which was the population spike evoked without prior alveus stimulus. Current pulses (0.1 ms duration) were delivered via constant current stimulus isolators (Model A365D, World Precision Instruments) at a stimulus rate of 0.05 Hz and were adjusted throughout the course of the experiment such that SR stimulation elicited half-maximal responses. Alveus stimuli were always supramaximal and thus most likely recruited, directly or via CA1 axon collaterals, a substantial number of “late persistent” (putative GABAA,slow-generating) (Maccaferri et al. 2000) O-LM interneurons in addition to the more excitable “onset transient” (putative GABAA,fast-generating) interneurons (Pouille and Scanziani 2004).

All recordings were obtained in current-clamp mode using an Axopatch 200B patch clamp amplifier (Axon Instruments, Union City, CA) and pClamp 8.0 software (Axon Instruments). Field potentials were low-pass filtered at 5 kHz and digitized at 10 kHz (Digidata 1200, Axon Instruments). For data analysis ClampFit 8.0 (Axon Instruments), Origin 6.1 (MicroCal, Northampton, MA), MS Excel (Microsoft, Redmond, WA), Prism 4.0 (GraphPad, San Diego, CA) and custom-written Matlab 6.5.1 (The MathWorks, Natick, MA) routines were used.

In vivo surgery and electrophysiology

Animals used for in vivo recordings were aged 10–56 wk (β3+/+) and 32–57 wk (β3−/−).

Field potentials from area CA1 of the dorsal hippocampus were recorded as described previously (Hentschke et al. 2007). Briefly, under isoflurane anesthesia, the animals were chronically implanted with 16-channel linear microwire arrays (Jellema and Weijnen 1991). With 16 electrode sites separated by 100 μm, the electrodes spanned the hippocampus from the granule cell layer of the ventral leaf of dentate gyrus to the alveus in CA1.

The animals were allowed to recover for 1 wk before their first recording session. They were placed in an open plastic tray (20 × 30 cm) and allowed to move freely throughout the experiment. An observer classified the animal's behavior as “exploring,” “immobile,” and “grooming.” Local field potentials were recorded at a bandwidth of 1–300 Hz using a small headstage preamplifier (HS-16, Neuralynx, Tucson, AZ) and two 8-channel amplifiers (Lynx-8, Neuralynx). A screw in the occipital bone served as the electrical reference (“animal ground”). Data were digitized at 1 kHz (Digidata 1322A, Molecular Devices, Union City, CA) using pClamp v8.0 (Molecular Devices).

When data collection for a particular animal was complete, the tissue was fixed by transcardial perfusion with 0.1 M PBS followed by 4% paraformaldehyde. The brain was then removed, placed in 4% paraformaldehyde for ≥24 h. Slices of 100 μm thickness, mounted in saline with a coverglass (“wetmount”), were used to verify the location of the electrode array.

In vivo data analysis

All data analysis was performed with custom-written routines in Matlab v7.2 (Mathworks) as previously described (Hentschke et al. 2007). Briefly, in a preprocessing step, the raw data were passed through digital band-pass filters designed for the extraction of signals in different frequency bands. The −3-dB (corner) frequencies were 5–12 Hz for theta and 40–90 Hz for gamma. Gamma signals were additionally passed through a bandstop (notch) filter (corner frequencies: 59 and 61 Hz) to eliminate A/C line frequency noise. The raw and filtered field potential data were divided into variably sized chunks according to the animal's behavior. These chunks were subdivided into segments of 4,096 points each, corresponding to 4,096 ms, overlapping by a third (1,365 points). Remaining data segments shorter than this duration were discarded. All spectral and cross-correlation parameters were computed for each segment, averaged within each animal, and then averaged across the population. The gamma centroid (Fig. 5D), a measure of the average gamma frequency, was obtained by weighting frequency bins in the range [40–90 Hz] by their power and calculating the mean.

Statistical analysis

For statistical evaluation of the differences between genotypes in in vitro experiments, we used unpaired, two-tailed t-test or analyses of variance with Bonferroni-corrected post hoc tests. Statistical analyses of in vivo data were based on analytical descriptions of the depth profiles of the various parameters analyzed (Hentschke et al. 2007). For each parameter, we chose a function (see following text) with recording depth as the independent variable. The functions have no theoretical underpinnings related to hippocampal physiology but were chosen to fit the data well with a minimum of free parameters. P values <0.05 computed from an F-test were interpreted to indicate a difference between the complete depth profiles of the genotypes. In cases of significant effects, we employed Bonferroni-corrected t-test for unpaired data to identify individual recording sites with significant differences between genotypes.

The functions fitted to the depth profiles were as follows [x, the recording depth, ranged from 0 mm (hippocampal fissure) to 0.6 mm (alveus)]: a + be−cx: theta and gamma power (Fig. 5, C and D); a+bx: theta peak frequency (Fig. 5C); a+bx+cx2+dx3: gamma centroid (Fig. 5D); a+be−c(x−d)2: CVamplitude and CVIPI of theta (Fig. 6A) and CVIPI of the gamma envelope (Fig. 7E); a+bx+cx2: CVamplitude of the gamma envelope (Fig. 7E); a/(1+be−cx): lags of theta cross-correlation (Fig. 6C); 1−a+bx2+ae−cx2: peaks of theta cross-correlation (Fig. 6C); a+bx+[c/(1+e−d(x+e))]: lags of Ctheta,γEnv (Fig. 8C); and a+bx+cx2: peaks of Ctheta,γEnv (Fig. 8C) and |γEnv(theta)|2/|γ|2 (Fig. 7D).

Additionally, to obtain a statistical measure of the difference between genotypes independent of this analytical approach and hypothesis testing, we also report an effect statistic, Hedges' d (denoted d), which is a standardized difference between two means, including 95% confidence intervals (CIs) (Nakagawa and Cuthill 2007) for all recording sites and parameters. d can attain any value. The higher the absolute value of d, the stronger the effect, and in cases in which the 95% CI do not encompass zero the effect can be considered significant (P < 0.05) within the framework of hypothesis testing (Nakagawa and Cuthill 2007).

All results except for d are presented as means ± SD.


GabaA,slow is largely absent from CA1 pyramidal neurons of β3−/− mice

Spontaneous IPSCs were recorded from CA1 pyramidal neurons in hippocampal slices of wild-type and β3−/− mice at room temperature. In wild-type animals, IPSCs with fast decay kinetics (∼15 ms, here termed GABAA,fast) were interspersed with events of much slower decay, corresponding to GABAA,slow (Pearce 1993) (Fig. 1A). As has been described before, these events were three to four orders of magnitude rarer than fast IPSCs but could be readily identified by their slower kinetics (Banks et al. 1998) (Fig. 1, B and C). Current traces from pyramidal neurons in β3−/− slices also contained numerous fast IPSCs but, in contrast to traces from wild-type animals, were largely devoid of slow IPSCs (Fig. 1, A–C). Analysis of current traces from 11 wild-type and 14 β3−/− cells confirmed that spontaneous GABAA,slow IPSCs were extremely rare in the latter (Fig. 2A). The few spontaneous events that were detected did not differ significantly from those in wild-type animals in amplitude or kinetics (Fig. 2A).

Fig. 1.

Spontaneous GABAergic inhibitory postsynaptic currents (IPSCs) recorded from pyramidal neurons in CA1 of β3+/+ and β3−/− mice. A: raw data traces from β3+/+ (left) and β3−/− (right) animals. GABAA,slow IPSCs, as identified by slower rise and decay times, are marked (★). B: overlaid normalized slow and fast IPSCs extracted from the raw data traces of β3+/+ (left) and β3−/− (right) animals. C: scatter plot of 10–90% rise time vs. decay time of IPSCs extracted from all experiments (β3+/+, n = 11 and β3−/−, n = 14).

Fig. 2.

Characteristics of spontaneous and evoked IPSCs in β3+/+ and β3−/− mice. A: GABAA,slow IPSCs. B: GABAA,fast IPSCs. Kinetics of both classes of IPSC were determined from sIPSCs only. Stars indicate, P values of unpaired t-test comparing wild-type animals and mutants (*, P < 0.05). β3+/+, n = 11 and β3−/−, n = 14.

The paucity of spontaneous GABAA,slow IPSCs has been attributed to very low spontaneous firing rates of the underlying interneuron classes (Banks et al. 1998). The even lower incidence of GABAA,slow sIPSCs in β3−/− mice might be attributed to accordingly lower firing rates of the underlying interneurons in this genotype. Therefore to assess the prevalence of GABAA,slow independently of spontaneous activity, we recorded evoked GABAA,slow IPSCs by electrically stimulating in SLM. In slices from mutant mice, GABAA,slow currents thus evoked had less than a third of the amplitude of the wild-type counterparts (Fig. 2A), which was consistent with a substantial reduction of the number of postsynaptic receptors mediating GABAA,slow currents in β3−/− mice.

Our findings thus far demonstrate that elimination of the β3 subunit in fact severely diminished GABAA,slow in pyramidal neurons but did not completely abolish it. Interestingly, differences between genotypes were not restricted to GABAA,slow. GABAA,fast IPSCs decayed slightly faster in mutant mice than in wild-type animals (Fig. 2B) as can also be seen in the example traces in Fig. 1. Other IPSC parameters were not significantly different between the genotypes.

Divergent roles of β3-containing GABAA receptors in recurrent inhibition and suppression of fast inhibition

GABAA,slow IPSCs exert a powerful and long-lasting (up to hundreds of milliseconds) inhibition of interneurons and pyramidal neurons when evoked by electrical stimulation (Banks et al. 2000). Therefore alterations of this current should affect the excitability of both neuron types within the duration of the evoked IPSC. We tested this hypothesis with two different sets of experiments. First, to assess the inhibitory impact of GABAA,slow on pyramidal neurons, we applied the conditioning pulse paradigm (Fig. 3A) (Pearce 1996). One stimulation electrode in the alveus delivered the first (conditioning) pulse. The pulse excited CA1 pyramidal axons antidromically and always produced a population spike at a very short latency (Fig. 3B). Strong electrical stimuli, employed here, also produced a field excitatory postsynaptic potential (EPSP) and, in pyramidal neurons, long-lasting inhibition, which is largely independent of GABAA,fast (Pearce 1993). With a variable time delay, the second stimulation electrode, positioned in SR, excited Schaffer collateral/commissural input and gave rise to a second, orthodromically evoked population spike. The long-lasting inhibition of pyramidal neurons due to the first (conditioning) pulse caused the population spike evoked by the second pulse to be reduced in amplitude compared with an unconditioned response, evoked by the same stimulus in SR but without prior conditioning stimulus (Fig. 3B, left). This effect has been attributed to GABAA,slow (Pearce 1993; Pearce et al. 1995). In wild-type animals, the conditioned response recovered with time in an almost logarithmic fashion, approaching amplitudes of the unconditioned pulse with interstimulus intervals of 500 ms (Fig. 3C). In β3−/− hippocampus, by contrast, conditioned responses were stronger already at the shortest interstimulus interval tested (5 ms). At an interstimulus interval of 20 ms responses reached a plateau of ∼80% recovery, and from 200 ms onward roughly corresponded to those seen in wild-type animals. Statistically significant differences between the genotypes were found for interstimulus intervals of 10–40 ms (2-way ANOVA followed by Bonferroni-corrected post hoc tests), a range that is compatible with both dendritic shunting and somatic hyperpolarization of pyramidal cells by GABAA,slow currents. Over these intervals, the conditioned response in β3−/− mice was about twice that of the wild-type animals. We conclude that in β3−/− mice, stimulus-evoked, long-lasting inhibition in pyramidal neurons is severely impaired, and that the substantial reduction of GABAA,slow currents (Fig. 2A) is the likely cause.

Fig. 3.

Conditioned depression is weaker in β3−/− mice. A: simplified schematic of CA1 circuitry and illustration of the approximate locations of the recording electrode in stratum pyramidale (SP) and stimulation sites in alveus (StimALV, conditioning stimulus) and stratum radiatum (StimSR). Also shown are pyramidal cells (PYR), interneurons (IN), and the major putative axonal pathways activated by the stimuli. B: field potential responses elicited with a single stimulus in SR (left) and paired stimuli (conditioning stimulus in ALV followed by a single stimulus in SR) with different inter stimulus intervals (ISIs; 3 separate sweeps overlaid; stimuli in SR marked by arrows on top and population spikes marked by thin arrows at the bottom). Note the antidromically evoked, short-latency population spike (PS) following the conditioning stimulus in alveus. C: plots of the amplitude of the conditioned population spike, normalized to the amplitude as obtained with a stimulus in SR without prior conditioning pulse. Interstimulus intervals ranged from 5 to 2,000 ms. Stars indicate P < 0.05 as computed by a 2-way ANOVA followed by Bonferroni-corrected post hoc tests comparing wild-type animals (n = 7) and mutants (n = 4).

To assess the inhibitory impact of GABAA,slow on GABAA,fast–producing interneurons, an electrical stimulus was delivered in SLM and the ensuing temporary suppression of fast inhibition (SFI) quantified (Banks et al. 2000) (Fig. 4). In the recording from a wild-type hippocampus shown in Fig. 4, A–C, fast IPSCs were diminished in frequency and amplitude for a period of several hundred milliseconds poststimulus as has been reported previously (Banks et al. 2000). The resultant drop in the “instantaneous” average current (Fig. 4C, see methods) was quantified in a poststimulus time window of 50–300 ms. Surprisingly, fast IPSCs were suppressed to similar levels in both genotypes: in wild-type animals to 42 ± 20% and in β3−/− animals to 31 ± 17% of the prestimulus level (t-test, P = 0.32; Fig. 4D). There was only a weak negative correlation (correlation coefficient, −0.26) between spontaneous IPSC frequency and the degree of SFI (Fig. 4E), indicating only a weak contribution of (nonsuppressible) miniature IPSCs to the net inhibition of pyramidal cells.

Fig. 4.

Suppression of fast inhibition. A: 10 current sweeps with stimulation in s. lacunosum-moleculare (SLM) at t = 0 (arrowhead) recorded from a pyramidal cell of a wild-type animal. Stimulation artifact was clipped. B: peristimulus time histogram of GABAA,fast IPSCs. The amplitude range was clipped to 400 pA to enhance detail in the lower range. Bin widths were 25 ms and 25 pA. C: instantaneous average current, normalized to the average value in [−800 0] ms. Gray shaded area demarcates poststimulus analysis window of [50 300] ms. D: population averages of average current in poststimulus window (β3+/+, n = 6; β3−/−, n = 7). E, scatterplot of poststimulus current vs. IPSC frequency.

In summary, these in vitro experiments established that there is a substantial loss of GABAA,slow/long-lasting inhibition in pyramidal cells but not in fast IPSC-generating interneurons, in the CA1 region in the β3−/− genotype.

Absence of the β3 subunit alters hippocampal theta and gamma oscillations but does not interfere with nesting

To investigate how the differences between genotypes in terms of slow inhibition in hippocampal subnetworks translated into differences of rhythms involving the whole hippocampal network, we performed multi-site field potential recordings from CA1 of freely moving β3−/− and wild-type animals. In agreement with previous work (Buzsáki et al. 2003; Hentschke et al. 2007), field potentials in wild-type mice had strong components in the theta (5–12 Hz) and gamma (40–90 Hz) bands that were more prominent when the animals explored their environment than when they were immobile (Fig. 5, A and B). As the behavioral dependence of hippocampal rhythms in mice has been investigated in detail (Hentschke et al. 2007) and due to the occurrence of epileptiform activity during immobility in β3−/− mice (Supplementary Methods and Supplementary Fig. S1),1 we restricted the following analyses to exploring animals. Theta and gamma oscillations were altered in several respects in the mutant mice. Theta oscillations were significantly weaker and slower than in wild-type animals (Fig. 5C; P < 0.01 for theta power and P < 0.0001 for peak frequency. Also note the high values of d in SR/SLM). Gamma oscillations were also much weaker in β3−/− mice (Fig. 5D, left, P < 0.0001). However, in contrast to theta oscillations, gamma oscillations in β3−/− animals were faster than in wild-type animals (Fig. 5D, right, P < 0.0001), possibly reflecting the faster decay of IPSCs in these animals (Fig. 2B).

Fig. 5.

Local field potential activity in hippocampal area CA1 in awake, behaving β3+/+ and β3−/− mice. A: local field potentials (LFPs) in β3+/+ (left) and β3−/− (right) mice during immobility (top blocks of traces) and exploration (bottom blocks). The laminae recorded from are indicated to the left (ALV, alveus; HF, hippocampal fissure; SO, s. oriens; SR, s. radiatum). B: corresponding power spectra of signals recorded in HF (top, immobility; bottom, exploration; averages of spectra computed from ∼4-s segments from ∼30 min of data). C: laminar profile of power in the theta band (left bar graph) and theta peak frequency as obtained from power spectra (right bar graph). The bars show averages across animals for both genotypes (β3+/+ black, β3−/− gray) during the exploring state. The schematic adjoining the left bar graph depicts the approximate recording location. The thin plots labeled d to the right of the bar graphs depict the values of Hedges' d including 95% confidence intervals (see methods). Wild-type and β3−/− animals differed significantly from each other in both parameters (F-test; P < 0.01 for theta power; P < 0.0001 for peak frequency). Small stars to the right of a pair of bars indicate the P values of Bonferroni-corrected t-test (*, P < 0.05; **, P < 0.01). This scheme applies to all plots depicting depth profiles. n = 7 for both β3+/+ and β3−/− genotype in this and all following depth profiles. D: gamma power (left) and gamma centroid (right). P < 0.0001 for both gamma power and gamma centroid (F-test).

Next we investigated the regularity and coordination across laminae of theta rhythms. The regularity of theta oscillations at each recording site was expressed as the coefficient of variation of the time intervals between troughs (CVIPI, Fig. 6A, left) and the coefficient of variation of trough amplitudes (CVamplitude, Fig. 6A, right). In both genotypes, CVamplitude was approximately twice as large as CVIPI (Fig. 6A). The laminar profiles of both parameters had a peak at the border of SP and SR, highlighting the sites of least regular theta rhythms. In β3−/− mice, theta oscillations were significantly less regular (CVIPI and CVamplitude, P < 0.001), particularly in the dendritic laminae.

Fig. 6.

Characterization of theta oscillations in β3+/+ and β3−/− mice. A: regularity of theta oscillations at each recording site. Left: the coefficient of variation (CV) of the interpeak interval between troughs (CVIPI); right: the CV of trough amplitudes (CVamplitude). Troughs were chosen instead of peaks because in the distal dendritic recording sites, which show the highest theta amplitudes, trough amplitudes are higher than peak amplitudes. In both respects, theta oscillations were significantly more variable in β3−/− animals than in wild-type animals (CVIPI and CVamplitude, P < 0.001; F-test). Quantitatively similar results were obtained for theta peaks. B: cross-correlation functions (Cθ,θRef) of theta oscillations in HF, the reference site, with theta oscillations from all other sites (including HF, see abbreviations to the left of the curves). The lowermost curve is the auto correlation Cθ. The lags and amplitudes of the positive peaks marked by black dots were used as measures of the phase shift and similarity, respectively, of theta signals in the different laminae with respect to the theta signal in HF. C: theta phase shifts across laminae (left) and peak amplitudes of Cθ,θRef (right) of both genotypes. Lags of Cθ,θRef had a near-sigmoidal laminar profile and were not significantly different between genotypes (P = 0.15, F-test). However, wild-type mice had significantly higher peak amplitudes of Cθ,θRef (P < 0.01, F-test).

Cross-correlations of theta signals between the hippocampal fissure (HF), which served as the reference site, and all other sites, revealed the well-known gradual shift of theta phase across laminae reaching half a theta period lag (50 ms) in the alveus (Fig. 6B). This average laminar profile of theta phase lags was indistinguishable between genotypes (Fig. 6C, left, P = 0.15). However, β3−/− mice had significantly lower peak cross-correlation values (Fig. 6C, right, P < 0.01). This impaired coordination of theta oscillations at different sites, together with the finding that theta was less powerful and more variable in the apical dendritic region, is consistent with the idea that β3−/− mice lack an important element shaping theta oscillations.

Next we examined whether the amplitude modulation of gamma at theta frequency (nesting, Fig. 7A) was affected by the elimination of the β3 subunit. For this purpose, we computed the power of the gamma envelope (γEnv) in a narrow theta frequency band [|γEnv(theta)|2, Fig. 7B] and compared it to gamma power (|γ|2) as quantified in Fig. 5d. Within each genotype, there was an excellent correlation between both parameters at HF (Fig. 7C), SLM, distal SR, SO and alveus (data not shown, R2 ranging from 0.66 to 0.98). Furthermore, the ratio of |γEnv(theta)|2 and |γ|2 was not significantly different between genotypes (Fig. 7D, P = 0.26). The fact that both parameters scaled linearly and to very similar degrees in both genotypes illustrates that the absence of the β3 subunit did not impair the amplitude modulation of gamma appreciably despite a strong effect on gamma power. This was further underlined by minor or no differences in the amplitude and timing variability of γEnv, respectively (Fig. 7E): only CVIPI was significantly higher in β3−/− animals than in wild-type animals (P < 0.0001), and the difference between the genotypes was small (maximum: 7.8% at HF).

Fig. 7.

Modulation of gamma oscillation amplitude at theta frequency (nesting). A: field potential from a single recording site (HF) in a wild-type animal, split up into signals of different frequency content: gamma oscillations and gamma envelope (γEnv; top gray and black traces, respectively), raw data (1–300 Hz, bottom gray trace) and theta component (bottom black trace). B: power spectral density (PSD) of raw data and of γEnv. Both spectra feature a prominent peak at the same frequency (8.9 Hz). Checkered area below the PSD of γEnv represents the power of γEnv in a narrow frequency band centered at the theta peak (7.9 – 9.9 Hz, |γEnv (theta)|2), whereas the gray area below the PSD of the raw data corresponds to the gamma power (|γ|2) as illustrated in Fig. 4A. C: |γEnv(theta)|2 vs. |γ|2. Each point is a datum from 1 animal recorded at HF. Lines represent linear fits (β3+/+: slope = 0.108, 95% confidence interval = [0.090 0.125], R2 = 0.98; β3−/−: slope = 0.089, 95% confidence interval = [0.057 0.119], R2 = 0.92). D: laminar profile of the ratio |γEnv(theta)|2/|γ|2. The difference between genotypes was not significant (P = 0.26; F-test). Note that for both genotypes the ratio of powers increased from a minimum in SR toward alveus although gamma power did not (Fig. 4D). E: variability of γEnv, expressed as the coefficient of variation of the interpeak interval (CVIPI, left) and the coefficient of variation of peak amplitudes (CVamplitude, right). Only CVIPI was significantly higher in β3−/− animals than in wild-type animals (P < 0.0001; F-test).

Finally, we inspected the relation of γEnv to theta oscillations (Fig. 8). For each recording site, the cross-correlation between the two signals was computed. As γEnv did not shift across laminae (Fig. 8A), the resulting cross-correlation functions essentially recapitulated the laminar phase shift of theta signals (Fig. 8B) (Hentschke et al. 2007). As in the case of theta, laminar phase shift profiles did not differ between the genotypes (Fig. 8C, left, P = 0.27). However, the peak cross-correlation, which quantifies the similarity between theta and γEnv, was significantly lower in β3−/− mice (Fig. 8C, right, P < 0.001). This impaired coordination between the two signals likely reflects degraded theta oscillations (Fig. 6) because gamma nesting was largely preserved (Fig. 7).

Fig. 8.

Coordination of gamma with theta rhythms. A: overview of temporal relations of gamma with theta rhythms across the laminae of CA1 (wild-type animal). Theta signals from all recording sites were rendered as an amplitude-coded contour plot in the background and overplotted by the envelope of gamma signals (gray lines) from the same recording sites. Note the shift of theta phase across laminae by almost half a theta period and the near absence of a laminar phase shift of γEnv. Theta troughs (dark contours) coincide with peaks of γEnv at the fissure, whereas an almost inverse relation is observed at the alveus. B: Cθ,γEnv, the cross-correlation function of theta with γEnv. Note that in contrast to Cθ,θRef (Fig. 6, B and C), Cθ,γEnv describes the local, within-electrode relationship of theta to γEnv. In HF, the coincidence of bouts of gamma with the troughs of theta as visible in A is reflected in a prominent negative central peak of Cθ,γEnv (black dot, lowermost curve). Matching peaks (also marked by black dots) were identified in the increasingly right-shifted cross-correlation functions from the recording sites dorsal of HF, and the lags and amplitudes of the peaks were determined. C: cross-correlation lags (left) and peak amplitudes (right). Peak amplitudes, but not lags, were significantly different between wild-type and β3−/− mice (P < 0.001 and P = 0.27, respectively; F-test).


In the present study, we show that in the hippocampal CA1 region, GABAA receptors with the β3 subunit contribute substantially to the generation of GABAA,slow in pyramidal neurons, participate in long-lasting recurrent inhibition in hippocampal CA1 pyramidal neurons, and contribute to the generation or expression of gamma and theta rhythms. By contrast, these receptors are not required for the suppression of GABAA,fast-producing interneurons and the nesting of gamma oscillations at theta frequencies. Before expanding on these points we discuss the suitability of the animal model we used.

β3−/− phenotype

The β3 subunit is found in the neonatal and postnatal rodent brain (Fritschy et al. 1994; Laurie et al. 1992) and is involved in the activity-dependent expression of GABAA receptors (Saliba et al. 2007). Its unavailability in β3−/− mice halves the expression of GABAA receptors in the brain and leads to 90% neonatal mortality and behavioral and neurological deficits in the surviving animals (DeLorey et al. 1998; Homanics et al. 1997; Krasowski et al. 1998). A detailed analysis of receptor composition in neocortex revealed that in β3−/− mice the density of GABAA receptors containing α2 and α3 subunits was greatly reduced (Ramadan et al. 2003), as these subunits preferentially combine with the β3 subunit (Benke et al. 1994). By contrast, expression of the α1 subunit, preferentially combining with β2 and associated with fast IPSCs, was unchanged in absolute terms. No evidence of compensatory upregulation of other subunit combinations was found. Thus the β3−/− genotype exposes a lack of a specific subset of GABAA receptors, namely those containing the β3 subunit and its preferred association partners. Therefore we consider it a very useful model for the study of subunit-dependent GABAergic transmission with the caveat of possible side effects stemming from the occurrence of epileptic activity in the animals, such as selective cell loss (see following text).

GABAA,slow and GABAA,fast in pyramidal neurons are mediated in part by GABAA receptors containing the β3−/− subunit

The β3 subunit is prominent in dendritic compartments of hippocampal pyramidal cells (Sperk et al. 1997). It has previously been found to confer slow kinetics on GABAA receptors in other brain regions (Huntsman et al. 1999; Ramadan et al. 2003). As GABAA,slow IPSCs originate in the dendrites of pyramidal neurons (Banks et al. 1998; Pearce 1993), we considered receptors containing the β3 subunit to be likely candidates mediating GABAA,slow in these neurons. Barring the possibility of a drastic and selective reduction of the number of GABAA,slow-generating neurons in β3−/− mice, our results confirm this hypothesis. In addition, the slight but significant acceleration of GABAA,fast IPSC decay in β3−/− mice betrayed the presence of the β3 subunit in (synaptic) GABAA receptors mediating GABAA,fast IPSCs.

A likely explanation for the alteration of both GABAA,slow and GABAA,fast IPSCs is the presence and prevalence in hippocampus of two GABAA receptor subtypes containing the β3 subunit with different kinetics and little spatial overlap: α5β3γ2 and α2β3γ2. As both α5 (Brünig et al. 2002) and β3 (Scotti and Reuter 2001) subunits are constituents of peri- and extrasynaptic GABAA receptors of pyramidal neurons, and given the likely contribution of spillover to GABAA,slow IPSCs (Banks et al. 2000), it appears likely that receptors incorporating these two subunits contribute substantially to GABAA,slow. The recent findings that in hippocampus the α5 subunit contributes to slow phasic currents (Prenosil et al. 2006; Zarnowska et al. 2008) as well as tonic GABAergic currents (Caraiscos et al. 2004; Prenosil et al. 2006) is in accord with this idea.

The β3 subunit also associates with α2 subunits (Fritschy and Möhler 1995), forming synaptic GABAA receptors on pyramidal cells (Brünig et al. 2002; Fritschy et al. 1998). Given the strongly reduced expression of the α2 subunit in β3−/− mice (Ramadan et al. 2003), we suggest that the significantly faster IPSC decay of GABAA,fast in this genotype as compared with wild-type animals stems from a relative shift toward receptors containing α1 and β2 subunits (Okada et al. 2000; Vicini et al. 2001). The preserved high frequency and amplitude of fast IPSCs in pyramidal neurons of β3−/− mice suggest that either the other interneurons (such as parvalbumin-positive basket cells) that utilize other receptor types (Klausberger et al. 2002) increase their firing rate or that subunit replacement occurs.

Slow phasic GABAergic currents meeting most of the characteristics of GABAA,slow have recently been described in neocortex (Sceniak and MacIver 2008; Szabadics et al. 2007) and subiculum (Prenosil et al. 2006). In the subiculum, in contrast to hippocampus, GABAA,slow does not depend on the α5 subunit (Prenosil et al. 2006), demonstrating that GABAergic IPSCs with slow kinetics may be mediated by several different receptor subtypes. Similarly the remaining (albeit very sparse) GABAA,slow in pyramidal neurons of β3−/− mice in our study point to the existence of subtypes of the GABAA receptor that are devoid of β3 subunits but nonetheless possess slow kinetics, such as the population of small-amplitude GABAA,slow IPSCs that lack α5 subunits (and likely β3 subunits) identified recently in CA1 cells (Zarnowska et al. 2008). Rather than depending on the slow intrinsic kinetics of their constituent receptors, these synapses may reflect the presence of a specific architecture and/or a higher proportion of perisynaptic versus synaptic receptors that could lead to slower current kinetics independent of subunit composition (Szabadics et al. 2007).

Inhibition in local CA1 circuitry

Due to the large charges they transfer, IPSCs with slow kinetics are expected to have a strong and long-lasting impact on their neuronal targets. The conditioning pulse paradigm was designed to assess this inhibitory impact by engaging slow inhibition prior to orthodromic excitation of pyramidal neurons. It was important to choose a stimulation site for the preconditioning pulse that would elicit long-lasting dendritic inhibition and yet minimize (interfering) glutamatergic currents in pyramidal neurons. SLM, optimal for eliciting GABAA,slow under conditions of blocked glutamatergic receptors, was less suited as a site for the conditioning pulse due to the presence of entorhinal afferents. Therefore we chose to stimulate in the alveus. Stimulation in alveus excited—directly or via recurrent CA1 axons—various interneuron types, among them basket, bistratified, trilaminar, ivy, and O-LM cells (Blasco-Ibànez and Freund 1995; Fuentealba et al. 2008; Pouille and Scanziani 2004; Wierenga and Wadman 2003). Dendritically projecting O-LM cells are among the putative presynaptic neuron types mediating GABAA,slow, and, accordingly, stimulation in alveus has been shown to elicit long-lasting dendritic inhibition (Benkwitz et al. 2007; Maccaferri et al. 2000; Pouille and Scanziani 2004) (see also methods) and to “veto” integration of Schaffer and temporoammonic inputs (Ang et al. 2005). This preconditioning pulse depressed subsequent, orthodromically elicited population spikes less in β3−/− mice than in wild-type animals. This finding demonstrates that inhibition mediated by GABAA receptors harboring the β3 subunit has the potential to curb pyramidal action potential activity for long intervals (on the order of 1 theta period) via local dendritic shunting and/or somatic hyperpolarization. Specifically, in view of the largely reduced prevalence of GABAA,slow currents in pyramidal neurons of β3−/− mice, we surmise that GABAA,slow-generating synapses form part of this powerful inhibition of pyramidal dendrites. The preconditioning pulse in alveus most likely did not activate a subset of dendrite-targeting interneurons, lacunosum-moleculare, and neurogliaform cells, which are confined to distal dendritic laminae (Price et al. 2005). As these are also among the putative interneuron types producing GABAA,slow we presume that our results are a conservative estimate of the action potential-depressing impact of GABAA,slow.

By the same reasoning, electrical stimulation in SLM can be expected to tap the full power of slow GABAergic dendritic inhibition of pyramidal cells as it stimulates the axons of both local and dendritically projecting interneurons. Stimuli in this lamina are also very efficient at inducing long-lasting inhibition of interneurons: if administered at a strength sufficient to induce GABAA,slow IPSCs in pyramidal cells, they temporarily interrupt the stream of GABAA,fast IPSCs impinging on the same cells. This suppression of fast inhibition (SFI) (Banks et al. 2000) is not mediated by GABAB receptors; it is best explained by a long-lasting inhibition of interneurons producing GABAA,fast IPSCs by GABAA,slow currents (Banks et al. 2000; White et al. 2000). Previous modeling studies indicated that pauses ≤600 ms in duration reflect the influence of IPSCs with decay times of ∼100 ms, i.e., GABAA,slow, as they delay the spontaneous firing of the interneurons that generate GABAA,fast IPSCs (Banks et al. 2000). The fact that in wild-type hippocampi, the depression of population spikes was much briefer (Fig. 3), and indeed similar in duration to GABAA,slow IPSC decay (Fig. 2), may reflect the local shunting influence of the inhibitory conductance on excitatory current rather than the longer-lasting membrane hyperpolarization that underlies SFI. In addition, activation of different sets of presynaptic elements via alveus versus SLM stimulation may contribute to differences in duration of SFI and suppression of evoked firing.

Surprisingly, SFI was as strong in β3−/− mice as in wild-type animals (Fig. 4). Therefore we conclude that long-lasting inhibition of GABAA,fast-producing interneurons does not require GABAA receptors harboring the β3 subunit. This type of inhibition, and the remaining slow inhibition in pyramidal neurons, must then utilize β1 or β2 subunits. Although IPSC duration is thought to reflect primarily intrinsic receptor gating properties at some synapses (Jones and Westbrook 1996; Mozrzymas 2004), and the high affinity and slow kinetics imparted by the β3 subunit may indeed serve as the primary determinant of slow IPSC decay in some cases (Burgard et al. 1996; Schofield and Huguenard 2007), other factors, such as the spatiotemporal profile of transmitter and spillover onto perisynaptic receptors, clearly play important roles at others. These include slow inhibitory synapses in dentate gyrus (Wei et al. 2003) and neocortex (Szabadics et al. 2007). As noted in the preceding text, the similarity between the duration of spontaneous GABAA,slow IPSCs in wild-type and β3−/− animals would indicate that the decay of some classes of IPSCs in hippocampus largely reflects factors other than intrinsic receptor kinetics.

In vivo rhythms affected by the absence of the β3 subunit

The absence of the β3 subunit leads to dramatically increased oscillatory synchrony and power in olfactory bulb (Nusser et al. 2001) and thalamus (Huntsman et al. 1999). In both brain regions, a single inhibitory neuron type, disinhibited in β3−/− mice, inhibits its targets more powerfully and thus enhances the structures' propensity to oscillate. By contrast, in hippocampus, the absence of the β3 subunit weakened and perturbed rhythms. Theta oscillations were weaker, slower, less regular (particularly in distal dendritic regions), and less coordinated across laminae in the β3−/− genotype than in wild-type animals. As pyramidal neurons are the major substrates of theta field potential-generating current dipoles, these findings argue that pyramidal dendritic slow inhibition via β3 subunit-containing GABAA receptors constitutes one of the several, spatially segregated sources of theta rhythms (Banks et al. 2000; Leung 1984; Montgomery et al. 2009) The spike rate and timing of O-LM neurons, which are among potential presynaptic interneuron types mediating GABAA,slow in pyramidal neurons, are in good agreement with this idea (Klausberger et al. 2003). Additionally, neurogliaform and LM cells, likely activated by entorhinal (temporoammonic) input, could contribute to regular slow dendritic inhibition. Because GABAA,slow IPSCs generated by neurogliaform cells do show strong, GABAB receptor-sensitive short-term depression (Price et al. 2008), their influence may be partially attenuated during repetitive activation. However, this characteristic would also endow them with the capacity to increase, as well as decrease, their contribution to network oscillations, if this attenuation was itself subject to modulation.

Gamma field activity in CA1 depends on fast inhibitory currents in pyramidal neurons, produced by perisomatic fast-spiking cells that receive rhythmic excitatory drive from CA3 pyramidal neurons (Csicsvari et al. 2003; Fisahn et al. 1998). In dendritic regions of CA1, a large proportion of gamma power appears to be volume-conducted from dentate gyrus (Bragin et al. 1995; Csicsvari et al. 2003). As discussed in the preceding text, the absence of the β3 subunit results in a net loss of GABAergic inhibition and likely in a preponderance of GABAA receptors with faster kinetics. These alterations of fast synaptic inhibition may be particularly dramatic in the molecular layer of dentate gyrus in which both α2 and β3 subunits are even more prevalent than in hippocampus proper (Sperk et al. 1997). As the decay time of IPSCs is a major determinant of gamma field oscillations, fewer and faster IPSCs should translate into weaker and faster gamma oscillations in β3−/− mice compared with wild-type animals. This is indeed what we observed. Yet less oscillatory power does not necessarily imply lower firing rates of the contributing neurons; it could also mirror their desynchronization (Robbe et al. 2006).

In addition to changes in the individual theta and gamma rhythms, the coordination between the two oscillations was substantially impaired in β3−/− compared with WT mice (Fig. 8). However, when examined independent of the underlying theta rhythm, the degree of amplitude modulation of the gamma oscillation at theta frequency was unchanged (Fig. 7C), and the regularity of the amplitude variation of gamma oscillations was only slightly reduced (Fig. 7E). These findings suggest that slow inhibitory currents in GABAA,fast-producing interneurons that do not utilize the β3 subunit—such as those that underlie the suppression of fast inhibition that remains intact in β3−/− mice (Fig. 4)—helped maintain the degree of gamma nesting that we observed. Taken together, these results point to the disruption in the theta rhythm as the primary factor in the impaired theta-gamma coordination. Another contributing factor may have been the depression of GABAA,slow currents in pyramidal cells, and the ensuing repercussion on interneuronal firing precision (Tort et al. 2007): desynchronized spike timing of gamma-generating neurons within the theta cycle would cause gamma bouts to be less precisely coordinated with theta (Fig. 8C). In summary, our observations are consistent with a model in which theta and gamma oscillations are produced by two relatively independent subcircuits that oscillate at different frequencies with the coordination of gamma oscillations at theta frequency resulting in part from the periodic inhibition of gamma oscillators by theta oscillators (White et al. 2000).


This work was supported by NIH Grants NS-056411 (R.A. Pearce) and GM47818 (R.A. Pearce and G.E. Homanics), the DFG (C. Benkwitz), and the UW Department of Anesthesiology.


We thank M. Stüttgen for helpful advice on statistics.

Present address of C. Benkwitz: Dept. of Anesthesia and Critical Care, Massachusetts General Hospital, Harvard Medical School, Boston, MA 02114.


  • 1 The online version of this article contains supplemental data.


View Abstract