Abstract
GABAA receptors containing δ subunits have been shown to mediate tonic/slow inhibition in the CNS. These receptors are typically found extrasynaptically and are activated by relatively low levels of ambient GABA in the extracellular space. In the mouse neocortex, δ subunits are expressed by some pyramidal cells as well as on parvalbumin-positive (PV+) interneurons. An important function of PV+ interneurons is the organization of coordinated network activity that can be measured by EEG. However, it remains unclear what role tonic/slow inhibitory control of PV+ neurons may play in shaping oscillatory activity. After validating expected functional loss of δ-associated current in cortex of PV δcKO mice of both sexes, we performed EEG recordings to survey network activity across wake and sleep states. PV δcKO mice showed altered spectral content of EEG during NREM and REM sleep that was a result of increased oscillatory activity in NREM and the emergence of transient high-amplitude bursts of theta-frequency activity during REM. Viral reintroduction of Gabrd to PV+ interneurons in PV δcKO mice rescued REM EEG phenotypes, supporting an important role for δ subunit-mediated inhibition of PV+ interneurons for maintaining normal REM cortical oscillations.
Significance Statement
The impact of the loss of slow ionotropic inhibition in parvalbumin-positive interneurons was evaluated with electroencephalographic recording. We discovered unexpected changes at low frequencies during sleep that were rescued by viral reintroduction.
Introduction
GABAA receptors (GABAARs) represent a key inhibitory signaling system in the CNS. GABAARs are heteropentameric chloride channels, and constituent subunits influence physiological and pharmacological receptor properties. Heteropentameric GABAARs containing α, β, and δ subunits, the latter of which are encoded by the Gabrd gene, are typically localized to extrasynaptic/perisynaptic sites of certain neuron types, where they participate in tonic and slow synaptic inhibition (Nusser et al., 1998; Cope et al., 2005). In contrast, other subunit combinations mediate fast, phasic inhibition (Essrich et al., 1998; Wulff et al., 2009). δ-containing GABAARs have been extensively studied in principal cell types that show high δ subunit expression, including hippocampal dentate gyrus granule cells (DGCs; Wei et al., 2003; Sun et al., 2004, 2018, 2020), cerebellar granule cells (CGCs; Nusser et al., 1998; Rossi et al., 2003; Rudolph et al., 2020), and thalamocortical cells (TCs; Belelli et al., 2005; Cope et al., 2005). In these cell types, receptors containing δ subunits also contain α4 (DGC and TC) or α6 (CGC) subunits, and their activities can modulate cell excitability and transitions between firing modes (Nusser et al., 1998; Sur et al., 1999; Cope et al., 2005; Glykys et al., 2008).
In addition to excitatory principal cell types, δ subunits are expressed by GABAergic interneurons (Glykys et al., 2007, 2008). In the hippocampus, parvalbumin-positive (PV+) interneurons have been shown to express receptors containing a unique partnership of α1/δ subunits that provide a source of tonic inhibition in these cells (Glykys et al., 2007). Additionally, single cell transcriptomic studies from mouse cortex and hippocampus indicate that among the interneuron cell types, PV+ cells have the strongest expression levels of Gabrd (Yao et al., 2021). However, because PV+ interneurons primarily serve to regulate and coordinate principal cell activity (Rudy et al., 2011; Tremblay et al., 2016; Pelkey et al., 2017), it is not immediately clear how the presence and modulation of tonic/slow inhibitory currents produced by δ-containing GABAARs in PV+ cells may influence coordinated network activity.
In the cortex and hippocampus, PV+ interneurons play an important role for synchronizing activity of pyramidal cells that give rise to oscillations seen in local field potential (LFP) and EEG recordings in vivo (Hu et al., 2014). Perisomatic inhibition from PV+ interneurons is critical for the generation of gamma oscillations (30–100 Hz), and selective ablation of δ subunits from these cells increases the peak frequency but not power of in vitro gamma oscillations generated in the CA3 region of the hippocampus (Ferando and Mody, 2015). Beyond the generation of gamma oscillations, PV+ interneuron activity in the cortex exhibits significant phase locking to theta oscillations and increases during sleep spindles (Hartwich et al., 2009; Peyrache et al., 2011; Averkin et al., 2016; Niethard et al., 2018; Brécier et al., 2022), but the impact of slow inhibition by δ-containing receptors on these population level activities has yet to be explored.
Here we report changes to spectral content and sleep-related oscillatory activity following the removal of Gabrd from PV+ neurons (PV δcKO). Removing a source of tonic inhibition of PV+ neurons resulted in changes to spectral content of EEG in both NREM and REM states without a change in the architecture of sleep observed in the PV δcKO mice. The oscillatory frequencies affected led us to examine sleep spindles. Total number of spindles detected across the 12 h light cycle was not altered, but individual sleep spindle events had larger amplitudes and longer durations than the those of WT littermates. Additionally, PV δcKO mice exhibited high-amplitude transient bursts of activity with a peak frequency of 7 Hz that occurred primarily during REM sleep. To distinguish whether the phenotype arose from altered network maturation following removal of Gabrd from PV+ neurons in early postnatal life or from acute lack of tonic/slow GABAA signaling in PV+ neurons during REM states, we reintroduced Gabrd to PV cells in δcKO mice and found rescue of the REM phenotype. We conclude that slow inhibition of PV neurons is important for the EEG structure of REM sleep.
Materials and Methods
Ethical approval
All animal procedures were performed according to NIH guidelines and approved by the Washington University Institutional Animal Care and Use Committee, protocol 22-0344. Pain and suffering were alleviated with appropriate anesthesia and analgesia during surgical procedures. Animals were reared under the care of the Washington University School of Medicine Division of Comparative Medicine. Animals had ad libitum access to food and water throughout. Mice were killed at the end of studies according to NIH guidelines for minimizing pain.
Mice
Gabrd floxed mice were a generous gift from Jamie Maguire (Lee and Maguire, 2013). Gabrdfl/fl mice were crossed with PVCre animals (Jax Strain #017320) to produce the ultimate breeding pairs consisting of Gabrdfl/flXPVCre+/− and Gabrdfl/flXPVCre−/− that generated litters containing both WT and PV δcKO mice used in this study. Animal sex and number used for EEG studies were as follows: Main Cohorts: WT, N = 8 (4M/4F), cKO, N = 8 (4M/4F); Viral Rescue Cohorts: PV δcKO + Gabrd, N = 8 (5M/3F), PV δcKO + GFP, N = 7 (5M/2F), PVCre+/− + Gabrd, N = 6 (3 M/3F), PVCre+/− + GFP, N = 6 (3M/3F).
Verification of genomic recombination
To confirm successful recombination of the Gabrd allele in PV δcKO mice, genomic DNA was extracted from cerebellum chosen for enrichment of PV+ cells. PCR primers designed to produce an amplification product only from the recombined Gabrd locus were used to confirm PV δcKO in the brain. Only mice with confirmed Gabrd locus genotypes from brain tissue were included in the study.
Forward 5′ – CTCCAGTTGCCAAGCCTTTA – 3′
Reverse 5′ – CCTGGCTAATCCAGAAGGAG – 3′
Slice electrophysiology
Mice (4–8 weeks old) were used for slice recordings. Coronal slices at 300 µm thickness were made from frontal cortex as described previously (Lu et al., 2023). PV interneurons were labeled for recording by fluorescent reporter tdTomato in Ai14×PV-Cre WT mice or virally administered GFP in δcKO mice as described below. Pyramidal neurons were morphologically identified through their pyramidal shape and dendritic projections toward layer 1. Recordings from both cell types were made in layer 2/3. Borosilicate glass pipettes (World Precision Instruments) with tip resistance of 3–7 MΩ were filled with internal solution containing the following (in mM): 130 CsCl, 10 HEPES, 5 EGTA, 2 MgATP, 0.5 NAGTP, and 4 QX-314; pH adjusted to 7.3 with CsOH; 290 mOsm. To record tonic currents, cells were voltage clamped at −70 mV and 1 µM THIP was bath applied. Tonic currents were analyzed as described previously (Lu et al., 2023).
EEG surgery
Mice were anesthetized with isoflurane (5% for induction, 1.5–2% for surgery) and mounted in a stereotactic frame (Kopf). Bilateral holes were drilled in the skull for insertion of epidural screw electrodes for frontal (+0.7 AP, ±0.5 ML bregma) and parietal (−2.0 AP, ±1.5 ML bregma) electrodes. An additional screw over the cerebellum (−1.0 AP lambda) served as a common ground reference. To facilitate vigilance scoring of EEG, a single stainless steel wire was implanted in the nuchal muscle for EMG measurement. Animals were allowed to recover in their home cages for a minimum of 3 d before initiating EEG recordings.
EEG recording
For the duration of the experiment, mice were maintained on reverse lighting cycle. Mice were initially habituated to the EEG recording chamber with an hour-long exposure at least 24 h before recording during the dark phase to reduce environment novelty. Recording sessions were initiated 2 h prior to lights on to acclimate mice and minimize disturbance of natural sleep behaviors during 12 h light cycle used for analysis; we cannot exclude the possibility of some disturbance. Use of sleep data from the light phase was done mainly for practical reasons of having sufficient data from each sleep state. EEG was acquired with an Open Ephys system as previously described (Lambert et al., 2023). Up to four mice were recorded simultaneously with genotypes and sexes being distributed equally across recording chambers between recording sessions. Signals were digitized at 1,000 Hz and filtered from 1 to 250 Hz with a second-order Butterworth digital filter. Raw data were imported into MATLAB for further analysis.
Sleep scoring
EEG/EMG was scored for sleep/wake stages with the AccuSleep toolbox for Matlab (Barger et al., 2019). A subset of either 4 or 15 s epochs were scored as wake, NREM, or REM totaling a minimum of 5 min of each state. For initial analysis of sleep architecture, state transitions, and sleep stage power spectra, 4 s epochs were utilized to detect brief state transitions. For viral rescue experiments, 15 s epochs were chosen to balance labeling burden with accuracy of transitions (Yan et al., 2011). For instance, using briefer epochs revealed short bouts of wake interrupting assigned REM to NREM transitions. Given that these brief interruptions have little impact on averaged power spectra (Extended Data Fig. 2-1), and did not affect overall conclusions, we adopted the 15 s epochs for subsequent analyses. These manually scored epochs were used to calibrate a pretrained network, which subsequently scored the remainder of the epochs. A minimum of three successive epochs was required to classify a state change. Scored recordings were each further reviewed by an experienced rater to validate proper scoring of state transitions in both genotypes, and bouts demonstrating activity consistent with two states were assigned to the state that comprised the majority of the epoch window. A subset of six recordings (3 WT/3 PVcKO) were scored manually in entirety by two trained raters to assess the accuracy of AccuSleep scoring in both WT and PVcKO mice. For each animal, sleep stages were further characterized by determining the total number of bouts of wake, NREM, and REM and the median duration of each bout throughout the 12 h recording period. Additionally, the sleep fragmentation index for the 12 h period was determined by the number of awakenings divided by the total sleep time in hours.
Figure 2-1
Effect of epoch length on sleep spectra. A) Frontal NREM spectra with sleep scored in 4 sec epochs reproduced from Fig 1 (statistics in main figure 2 legend) B) Frontal NREM spectra with sleep scored in 15 sec epochs, significant Genotype X Frequency interaction from Two-Way RM ANOVA (F (255,3570) = 1856, p < 0.0001). C) Frontal REM spectra with sleep scored in 4 sec epochs reproduced from Fig 1 (statistics in main figure 2 legend) D) Frontal REM spectra with sleep scored in 15 sec epochs, significant Genotype X Frequency interaction from Two-Way RM ANOVA (F (255,3570) = 7.389, p < 0.0001). Horizontal bars correspond to significantly different frequencies from cluster-based permutation testing. Download Figure 2-1, TIF file.
EEG analysis
To generate average power spectra of behavioral states, EEG scored for each state was concatenated and spectra produced from 0 to 100 Hz with the Matlab pspectrum function. For NREM and REM, all EEG scored as these states across the 12 h recording was used to produce respective power spectra. For active wake spectra, only EEG segments identified by combined evidence of animal movement, EMG, and the presence of theta rhythm in the parietal electrodes were considered. Quiet wakeful periods were not included in spectral analysis because of the heterogeneity of behaviors (e.g., grooming, eating, resting, sleep transition) and the resulting highly variable EEG signatures. Spectra were consolidated into 256 frequency bins from 0 to 100 Hz prior to statistical analysis.
Sleep EEG was further analyzed with the Better OSCillation (BOSC) method (Whitten et al., 2011) applied to time–frequency spectrograms from a wavelet transform of the raw EEG signal combined from either 60 min of NREM or the entirety of REM combined from the entire recording session. Continuous wavelet transforms were produced utilizing a set of 100 complex Morlet wavelets centered from 1 to 100 Hz in 1 Hz steps with wavelet cycles increasing logarithmically from 3 to 30. For determination of Pepisode, oscillation amplitude threshold was set to the 95th percentile and duration threshold was set to three cycles. The peak of sigma frequency oscillations in NREM was calculated by determining the maximum Pepisode value of the BOSC spectra below 20 Hz for each animal.
State transition triggered spectrograms were produced with multitaper methods in the Chronux Matlab toolbox (Mitra, 2007). A sliding window of 2.5 s with a 500 ms step size was applied and multitaper parameters [TW, K] = [3 5] were used to calculate a sliding fast Fourier transform to each identified state transition. State transitions for analysis were limited to those that met the following criteria of continuous states on either side of the transition points: NREM to REM, wake to NREM, and NREM to wake required 2 min of consistent state prior to transition and 1 min following the transition, REM to wake required 1 min of consistent REM followed by a minimum of 16 s wake. Individual transition spectrograms were computed for each mouse and used to determine peak power and peak frequency data throughout transitions. For NREM to REM sigma frequency power transition, mean power of oscillations in frequencies 10–15 Hz were computed from transition spectra. These time series were smoothed with a moving average window of 15 s and were further consolidated by determining the mean of the signal in each time bin. Average spectrograms for all mice in each condition were combined for the group transition spectrograms.
Sleep spindles were detected with an automated algorithm validated for detection of spindles from rodent EEG (Uygun et al., 2019). Briefly, events were classified as spindles if the cubed root mean squared envelope of the sigma frequency bandpass filtered frontal EEG surpassed a low amplitude threshold for durations between 500 ms and 10 s and during that period had a peak rising above a secondary threshold. Spindle amplitude was calculated from maximum peak to trough of the z-score normalized EEG during each event. Durations were determined by the amount of time the transformed signal was above the minimum amplitude threshold. Interspindle intervals were defined as the time between each spindle start point. We opted to quantify spindle number rather than frequency given a lack of state changes (see below) and the variability of spindle frequency across behavioral state.
To generate event spectra of high-amplitude transient burst activity observed during REM sleep, EEG segments containing events were removed from the input to generate a power spectrum with event-free background REM. This background spectrum was used for normalization of the spectrum generated from the total REM EEG, leaving only the REM event-related spectral profile in the final spectrum.
Movement quantification
Data from three-axis accelerometer was combined to produce a single acceleration magnitude signal and further processed to determine the moving median absolute deviations of the signal with a 500 ms sliding window. Movement was defined as periods where this processed signal exceeded a threshold of 0.005 a.u., which was determined from periods of the recording where the animals were known to be still based on EEG and EMG signals and observed behavior. For high-amplitude EEG events occurring during wake in PVδ-cKO mice, average movement was determined as the mean of the transformed movement signal for the 10 s preceding events and across the duration of an event, and these periods were scored as containing movement if the mean exceeded the defined threshold.
Viral production and delivery
The expression plasmid pAAV-CBA-FLEX-GABRD-IRES-GFP was produced by substitution of synthesized DNA containing the coding sequence of Gabrd (CCDS19028.1), an internal ribosome entry site, and GFP into pAAV-CBA-FLEX-GFP (Addgene # 28304). The latter virus was as a control in rescue experiments as well as to label PV interneurons in δcKO slices for whole-cell recordings. PhP.eB viral particles containing expression plasmids were produced by the Washington University Hope Center Viral Vectors Core. Viral doses from 5.8 × 1010 to 1.0 × 1011 vg/mouse were delivered systemically with retro-orbital injections prior to EEG surgeries or slice recordings. Injection volumes were determined from viral titers and animals received either one or two injections over 72 h in alternating eyes to keep each injection volume below 150 μl. Virus was allowed to incubate for 4–6 weeks before ex vivo whole-cell recordings or in vivo rescue EEG recordings.
Immunohistochemistry
Mice were perfused with phosphate-buffered saline prior to dissection. Fresh cerebellar tissue was harvested to isolate genomic DNA for knock-out validation. After removal of the cerebellum, brain hemispheres were separated with a midline cut, and one hemisphere was immersion fixed in 4% paraformaldehyde overnight at 4°C and cryoprotected for 48 h in 30% sucrose at 4°C. Brains were frozen on dry ice, and sagittal sections of 45 μm were cut on a freezing microtome. Free-floating sections were blocked in 3% normal donkey serum, 0.3% Triton-X detergent, and 1% BSA in PBS for 1 h at room temperature. Sections were incubated in primary goat anti-GFP antibody (Abcam, ab6673) and primary rabbit anti-PV (Swant, PV28) diluted 1:1,000 in block solution at 4°C, overnight with shaking. Subsequently, sections were washed three times with PBS and incubated in secondary Alexa Fluor 555 conjugated donkey anti-rabbit (Invitrogen, A31572) and secondary Alexa Fluor 488 conjugated donkey anti-goat (Invitrogen, A11055) diluted 1:500 in PBS for 2 h at room temperature, in the dark with shaking. Finally, sections were washed three times with PBS and nuclei stained with Hoechst. Regions of the cortex were assessed for colabeling of viral GFP and PV immunostaining to assess cell type specificity of viral expression. For validation of rescue virus expression, GFP antibody staining (primary incubation as above) was performed with goat anti-GFP antibody (AbCam ab6673, 1:5,000) followed by visualization using a nickel-intensified chromogenic avidin-biotin amplification strategy (Vector Elite kit, VectorLabs) performed according to manufacturer's instructions.
Behavioral tests
Locomotor activity was measured during the light cycle by a female experimenter unaware of experimental group. Locomotion was tested over a 48 h period in transparent enclosures (47.6 × 25.4 × 20.6 cm) constructed of Plexiglas and surrounded by computerized photobeam instrumentation (Kinder Scientific). Total ambulations, rearing, time at rest, and distance traveled were analyzed. All equipment was cleaned with 2% chlorhexidine diacetate or 70% ethanol between animals.
Alternation behavior Y-maze was measured by placing mice in the center of the maze containing three arms at 120 angles, each measuring 10.5 cm wide, 40 cm long, and 20.5 cm deep. Mice explored the maze for 10 min, with complete entry defined when the hindlimbs had completely entered the arm. Alternation was defined three consecutive choices of different arms with no re-exploration. We measured the number of alternations and arm entries, from which the percentage of alternations was calculated.
Accelerating rotarod was tested by training mice on the 6 cm rod (5–40 rpm) for 30 min over 2 d. On the third day, mice were tested for 1 h.
Sensorimotor gating was tested with the acoustic startle/prepulse inhibition (PPI) assay. The responses to 120 dB auditory stimulus (40 ms broadband burst) and to a prepulse anticipating the startle pulse were measured. At stimulus initiation, 1 ms force readings were measured and averaged for the startle amplitude. Startle trials were preceded by 5 min white noise (65 dB). The first five and last five trials were startle pulse alone. Twenty prepulses were presented at 4, 8, or 16 dB above background. Prepulse inhibition scores were calculated as %PPI = 100 * (ASRstartle pulse alone − ASRprepulse + startle pulse) / ASRstartle pulse alone.
In-cage running wheel tasks
For normal running wheel, male and female mice of both genotypes were tested blind to genotype for baseline voluntary wheel running activity on a standard mouse activity wheel (Mouse Motor Skill Sequences Activity Wheel, Lafayette Instrument) for 24 h. Each activity wheel chamber contained a conventional anodized aluminum running wheel suspended from the chamber top. A thin layer of bedding covered the chamber floor, and mice had access to food and water ad libitum. An optical rotation sensor mounted 0.508 cm from the rungs transmitted wheel rotation information to the Activity Wheel Monitoring System Software (Lafayette Instrument) on a nearby computer. The normal activity wheel contained 38 consecutive rungs (0.4 cm in diameter) spaced 0.614 cm apart. One revolution of the wheel equaled 0.40 m in distance traveled. During voluntarily running, the number of wheel rotations (counts) and distance traveled were quantified across the 24 h testing session. Data were collected in 10 s bins to calculate running speeds. Total distance traveled each hour and cumulatively over the 24 h session, as well as maximum running speed, was recorded.
To provide a more sensitive test for fine motor coordination and motor learning, the mice were tested on the complex activity wheel over two sessions. This task was favored because of previous evidence that sleep disturbances affect performance (Nagai et al., 2017). Customized running wheels were constructed by removing 18 rungs, creating irregularly spaced intervals between rungs that were 0.614, 1.6, or 2.6 cm apart randomly patterned. All other parameters of the activity wheel apparatus remained the same as in the normal activity wheel test. Mice were tested on the complex wheel during two 24 h sessions, the first immediately following the normal wheel, and the second session following 24 h off with no wheel access. Voluntary running was assessed in 10 s bins as described for the normal wheel condition, and total distance each hour as well as cumulatively over the two 24 h sessions were recorded.
Experimental design and statistical analysis
Group sizes and statistical approaches were chosen a priori based on effect sizes and similar design of past experiments (Lu et al., 2023; Salvatore et al., 2024). Experimenters were blind to genotype for EEG and behavioral experiments. For EEGs, blinding was broken following sleep scoring and initial power spectra generation. For behavioral tests, blinding was broken when preparing for statistical comparisons. Comparisons of tonic current induced by THIP were performed with unpaired t tests.
Changes observed in EEG power spectra and BOSC spectra were assessed with a two-way repeated-measures ANOVA with factors for genotype/treatment and frequency. Of a significant Genotype × Frequency or Genotype × Treatment interaction was identified, differences in frequency ranges were determined by nonparametric cluster-based permutation testing. Student's t-scores were first computed for group differences at each frequency, a null distribution of t-scores for each frequency was determined by 1,000 permutations of group assignments, and contiguous frequencies with t-scores above a critical value corresponding to an unadjusted alpha threshold of 0.01 were identified as candidate significant clusters. Multiple-comparison correction was performed for candidate frequency clusters based on frequency cluster length compared with the null distribution of frequency cluster lengths obtained during the 1,000 permutations. Frequency clusters that exceeded the 95th percentile of the null distribution for cluster lengths were considered statistically significant between groups. In the case of the event spectra for the viral conditions, Dunnett's multiple comparisons was used to compare each treatment condition to PV δcKO + GFP virus. EEG results from each electrode were treated independently for analysis due to the qualitative differences across electrodes observed in the recordings.
Comparisons of summary statistics are presented as mean ± SEM, and simple comparisons between two groups were performed with unpaired t tests as indicated, except in cases when visual inspection of data distributions did not appear normal and nonparametric tests were performed as indicated. An alpha level of 0.05 was used throughout to denote statistical differences. Further details of design are given in figure legends.
Results
Validation of selective loss of δ-containing GABAAR activity in cortex
Although mice harboring conditional deletion of Gabrd in PV+ neurons have been previously characterized (Ferando and Mody, 2013, 2015; Lee and Maguire, 2013), neocortical circuits have not previously been investigated, and problems with the Cre/lox approach in other contexts have been documented (Kobayashi and Hensch, 2013; Song and Palmiter, 2018; Luo et al., 2020). In addition to genomic validation (see Materials and Methods), we performed functional validation by recording two classes of neurons in prefrontal cortex that express δ subunits and that are expected to be differentially affected by the conditional Gabrd deletion approach: PV+ interneurons and layer 2/3 pyramidal neurons (Drasbek and Jensen, 2006). We identified WT PV+ cells by crossing PV-Cre mice with the Ai14 reporter line (RRID:IMSR_JAX:007914; Madisen et al., 2010). WT PV+ cells challenged with the δ-preferring agonist THIP (1 µM) exhibited tonic currents that were blocked by the GABAA receptor antagonist gabazine (50 µM tested in 7 cells; current density −0.84 ± 0.17 pA/pF to THIP, reduced to 0.09 ± 0.13 pA/pF with gabazine). A low concentration of THIP was used to help ensure selectivity (Drasbek and Jensen, 2006). THIP currents were reduced in δcKO PV+ cells (Fig. 1A–C). In layer 2/3 pyramidal cells, THIP current density was smaller in WT layer 2/3 pyramidal neurons than in PV+ interneurons, resulting mainly from larger membrane capacitance of pyramidal neurons, but the THIP current density was similar between WT and δcKO mice (Fig. 1D–F). Taken together, the results support the importance/prominence of δ-containing receptors in PV+ interneurons and the selective deletion of Gabrd from the targeted cell type.
Reduced THIP currents in PV+ interneurons but not layer 2/3 pyramidal neurons (PN) in PV δcKO mice. A, Tonic currents induced by bath application of 1 µM THIP recorded in PV+ interneurons in layer 2/3 of acute cortical slices from adult WT (top) and PV δcKO (bottom) animals (WT N = 10 slices/4 mice, PV δcKO N = 11 slices/4 mice). Holding potential was −70 mV. B, THIP current density was significantly reduced in cells from PV δcKO mice (p = 0.0004, unpaired t test). C, Increase in standard deviation of current induced by THIP was significantly reduced in PV+ cells from cKO slices (p = 0.0001, unpaired t test). D, Tonic currents induced by 1 µM THIP in cortical L2/3 pyramidal neurons from WT (top) and PV δcKO (bottom) mice (WT N = 10 slices/4 mice, PV δcKO N = 11 slices/4 mice). E, No difference in THIP current density between L2/3 pyramidal neurons from WT or PV δcKO slices (p = 0.5156, unpaired t test). F, Increase in standard deviation (SD) of current induced by THIP application was not significantly altered in L2/3 pyramidal neurons in PV δcKO (p = 0.2225, unpaired t test).
Deletion of Gabrd from PV+ cells spares sleep/wake behaviors but alters EEG spectra across sleep states
We assessed sleep/wake behaviors and network oscillations in PV δcKO mice and WT littermates with EEG across the 12 h light phase. Although genotype validation in slices was performed in layer 2/3 (Fig. 1), it is possible that PV interneurons from other layers contribute to the EEG signal. We validated the use of AccuSleep for automated scoring of our PVcKO mice by comparing its performance to manually scored data in a subset of mice (3 WT/3 Pv δcKO). Inter-rater agreement of two trained scorers or one scorer and the AccuSleep output was assessed with Cohen's kappa (Kaufman and Rosenthal, 2009). Comparison between manual scores revealed Cohen's kappa values of 0.95 ± 0.03 for WT and 0.93 ± 0.01 for PV δcKO. Comparison between AccuSleep output and manual scores revealed Cohen's kappa values of 0.94 ± 0.004 for WT and 0.94 ± 0.01 for PV δcKO. Overall, we interpret this to indicate good agreement between scoring methods and accurate scoring of the Pv δcKO mice, which have not previously been scored with the AccuSleep trained models.
We scored sleep/wake behaviors of the 12 h light cycle recording session based on parietal EEG, EMG, and movement data, categorizing states as wake, NREM sleep, and REM sleep. PV δcKO mice had similar sleep behavior patterns to WT mice (Fig. 2A), with similar distributions of wake, NREM, and REM sleep across the full recording session. Sleep fragmentation index, measuring the number of awakenings compared with total sleep time, did not differ between genotypes (Fig. 2B). Additionally, both genotypes showed a similar number of bouts of each stage with comparable median durations of stages (Fig. 2C,D). Taken together, the deletion of δ-containing receptors from PV+ neurons did not alter the behavioral composition of sleep observed in the 12 h light cycle.
PV δcKO mice have altered EEG spectral content during sleep but no difference in sleep architecture. A, Proportion of 12 h lights-on cycle spent in wake, NREM sleep, and REM sleep by WT and PV δcKO mice was not significantly different. B, Sleep fragmentation index did not differ between genotypes. C, WT and PV δcKO mice exhibited similar numbers of wake (left), NREM (center), and REM (right) bouts across the recording. D, WT and PV δcKO mice exhibited similar median durations for wake (left), NREM (center), and REM (right) bouts. E, 5 s of representative active wake, NREM, and REM used for sleep scoring and initial spectral analysis from WT (black) and PV δcKO (red) mice (parietal EEG, nuchal EMG, and 3-axis accelerometer data are shown). F, Frontal active wake EEG power spectra and (G) parietal EEG power spectra showed no significant changes in spectral content between WT and PV δcKO mice. H, Significant increase in the power of a broad range of low- to mid-frequency activity in both frontal and (I) parietal EEG during NREM sleep [two-way repeated-measures ANOVA revealed significant Frequency × Genotype interaction in both frontal (F(255,3570) = 18.56, p < 0.0001) and parietal (F(255,3570) = 7.641, p < 0.0001) electrodes]. J, Multipeaked profile observed in average power spectrum of REM sleep in frontal EEG of PV δcKO mice, with significant increase in power in both frontal and (K) parietal EEG during REM sleep [two-way repeated-measures ANOVA showed significant Frequency × Genotype interaction in both frontal (F(255,3570) = 7.389, p < 0.0001) and parietal (F(255,3570) = 3.759, p < 0.0001) electrodes]. Horizontal bars in G–J indicate all frequencies with significant differences from post hoc cluster-based permutation testing [N = 8 (4 M/4F) mice/genotype].
Initial EEG analysis produced average power spectra from behaviorally similar periods (Fig. 2E) across the recording session to detect any consistent changes to oscillatory activity during each behavioral state. During active wake, PV δcKO mice exhibited normal theta oscillations with no statistical difference to broad gamma frequency power in cortical EEG recordings (Fig. 2F,G). Although we observed a small increase in power in the 20–40 Hz domain, there was no Genotype × Frequency interaction detected in either frontal or parietal EEG for this behavioral state. In NREM sleep, PV δcKO mice had reliably elevated power across broad frequencies spanning theta through beta ranges (Fig. 2H,I). Spectra produced from combining REM stages across the recording session showed changes more prominent in frontal EEG (Fig. 2J) but still detected in parietal EEG (Fig. 2K). Frontal EEG REM spectra in PV δcKO mice contained multiple peaks likely produced as harmonics of peak at a fundamental frequency in the theta range (Fig. 2J). In parietal EEG the REM spectrum was dominated by theta rhythms generated from dorsal hippocampus in both genotypes but still showed a significant Genotype × Frequency interaction across the entire spectrum. These findings suggest that signaling through δ-containing GABAAR in PV+ cells may regulate oscillatory activity generated in the cortex during sleep behaviors. Because we found no significant effects on active waking EEG (Fig. 2F,G), we focused on sleep EEG phenotypes because of the statistically robust findings and unexpected effect in lower frequency bands (Fig. 2H–K), as opposed to the gamma band, where PV+ interneurons have been most strongly implicated (Ferando and Mody, 2015; Averkin et al., 2016; Antonoudiou et al., 2020; Hadler et al., 2024).
Changes to oscillatory activity in both NREM and REM sleep in PV δcKO mice
Although initial analysis of sleep spectra showed altered power in PV δcKO mice, these results represent the average network activity present in these behavioral states across the entire 12 h recording session. To assess alterations to oscillatory activity during sleep stages with more temporal precision, we utilized the BOSC method that detects oscillations from a time–frequency representation of EEG and allows for both power and duration thresholds to separate oscillatory activities from background EEG. Based on observations in average spectra above, we limited BOSC analysis to frontal EEG which showed altered activity in both NREM and REM stages (Fig. 3). Additionally, focus on frontal recordings allowed us to probe transient sleep spindle activity with higher resolution during NREM (Kim et al., 2015; see below).
BOSC analysis of sleep stages in frontal EEG highlights increased oscillatory activities in PV δcKO mice. A, Continuous wavelet transform of 10 min of concatenated NREM sleep from frontal EEG in a WT mouse (black overlaid signals throughout), lower panel shows expanded timescale (corresponding to time marked with white bar in upper panel) of 30 s NREM containing oscillatory events consistent with sleep spindles. B, Representative wavelet transform of NREM sleep from a PV δcKO mouse showing larger periodic increases in power, expanded timescale shows longer durations of high power oscillatory activity in sigma frequency ranges. C, Average BOSC spectra from periods of NREM sleep show increased periodic σ frequency oscillatory activity with a peak frequency of 14 Hz; inset: significant increase of Pepidode σ range peak in PV δcKO mice (p < 0.0001, unpaired t test). Two-way repeated-measures ANOVA revealed significant Genotype × Frequency interaction (F(99,1386) = 31.07, p < 0.0001). D, Wavelet transform of 10 min combined REM sleep from a representative WT mouse showing theta-frequency oscillations, expanded timescale in bottom panel. E, Wavelet transform of frontal EEG during REM sleep scored based on parietal signals from a representative PV δcKO mouse. Transient, high-amplitude oscillatory bursts override theta-frequency oscillatory activity typical of REM sleep. Expanded time scale of bottom panel reveals similarity of spectral profile of transient events to periodic oscillatory activity observed in NREM sleep in both genotypes. F, Average BOSC spectra from REM sleep across the full recording session contains strong theta rhythm associated peak with increased oscillatory activity detected in PV δcKO mice centered ∼20 Hz Two-way repeated-measures ANOVA indicated a significant Frequency × Genotype interaction (F(99,1386) = 10.33, p < 0.0001). Horizontal bars in panels C and F indicate all frequencies with significant differences after post hoc cluster-based permutation testing.
Continuous wavelet transforms of frontal EEG during NREM showed typical periodic increases in power spanning frequencies ranging from 5 to 25 Hz in both WT and PV δcKO mice (Fig. 3A,B). BOSC spectra produced from these time–frequency transforms revealed changes to NREM oscillations in PV δcKO mice with an increase in the peak Pepisode of oscillatory activity detected in the sigma frequency range (10–15 Hz) typically associated with sleep spindle activity (Fig. 3C). Wavelet spectrograms of frontal EEG during REM states displayed concentrated power within the theta range in both genotypes (Fig. 3D–E). In PV δcKO mice, transient increases in power expanding into frequencies in the beta range were associated with instances of high-amplitude bursts of activity in the raw EEG (Fig. 3E, bottom panel). BOSC spectra produced from these transforms showed a prominent theta-frequency peak indicative of REM state with increased oscillatory activity detected in frequencies between 13 and 30 Hz (Fig. 3F). The changes observed with BOSC-based analysis confirmed that spectral alterations in PV δcKO mice are at least in part due to altered transient activities rather than to persistent changes to background EEG or stationary aspects of the EEG signals.
NREM sleep of PV δcKO mice contains altered sleep spindles
Following observations of altered sigma frequency oscillations during NREM states in PV δcKO mice combined with previous studies showing the recruitment of cortical PV interneuron activity by ascending thalamic inputs during spindle oscillations (Hartwich et al., 2009; Niethard et al., 2018; Fernandez and Lüthi, 2020), we assessed potential alterations to sleep spindles, characterized by activity in the sigma band, in PV δcKO mice. We first examined the transition between NREM and REM sleep, which has been previously shown to be sensitive to manipulations affecting sleep spindle activity (Franken et al., 1998; Astori et al., 2011; Bandarabadi et al., 2020). Average NREM to REM transition spectrograms of frontal EEG from WT and PV δcKO groups (Fig. 4A) show persistent low-frequency power in NREM shifting to theta-dominated REM sleep. A surge in spindle activity during the transition corresponds with a transient increase in sigma frequency power in frontal EEG in both genotypes. In PV δcKO mice, baseline NREM sigma power was elevated (Fig. 4B,C), and the peak in sigma power preceding the REM transition was higher. Additionally, there was a small but significant difference in the relative increase in sigma power (Fig. 4C, right). We interpret these changes to suggest that PV δcKO mice exhibit changes to spindle-like activity.
PV δcKO mice have enhanced spindle-related power at state transitions and larger sleep spindle oscillations in frontal EEG. A, Grand average NREM to REM state transition spectrogram from all WT (top) and PV δcKO mice showing relative increase in spindle-associated power ∼10–12 Hz as the animals approach state transition, with additional evidence of altered activity after the transition to REM in PV δcKO mice. B, Average sigma power throughout NREM to REM transition calculated from the average transition spectrogram for each mouse. C, Increased baseline (before −1 min to transition) sigma power (p = 0.0004, unpaired t test), higher power sigma frequency peak during transition state (p < 0.0001, unpaired t test), and in the relative increase above ongoing sigma frequency activity (p = 0.0253, unpaired t test) in PV δcKO mice during NREM to REM state transition. D, No difference in total spindle incidence from automated detection across the full recording. E, Spindle amplitude distributions shifted toward higher amplitudes for PV δcKO mice with inset showing significantly increased average spindle amplitude (p = 0.0056, unpaired t test). F, Spindle event duration distributions exhibit reduced probability of shorter duration events and increased probability density in long event duration (>2 s) tail region; inset shows significantly increased average event duration for PV δcKO mice (p < 0.0001, unpaired t test). G, Start to start interval distributions show little difference between genotypes, with both groups possessing two main peaks corresponding to the periodic occurrence of temporally clustered spindle events; inset shows now significant difference to average interspindle interval (p = 0.6077, unpaired t test).
To test this hypothesis, we used a previously validated algorithm for automated detection of sleep spindles from rodent EEG (Uygun et al., 2019). From the 12 h recording session, we found no difference in total spindle incidence between genotypes (Fig. 4D). Following detection of individual spindle events, we analyzed spindle amplitudes, spindle durations, and interspindle intervals. We found that removal of δ-containing receptors from PV+ cells resulted in higher spindle amplitudes and durations (Fig. 4E,F); however, spindle frequency measured by interspindle intervals was unchanged in PV δcKO mice (Fig. 4G). Although the vast majority of events detected as spindles occur during NREM states (93.0 ± 1.1% WT vs 90.9 ± 1.2% PV δcKO), PV δcKO mice exhibited a small increase in number of events detected in REM sleep [1.7 ± 0.2% (39 ± 4 events) versus 3.5 ± 0.6% (83 ± 16 events), p = 0.0073, Mann–Whitney test]. Because of the relative purity of abnormal events in REM of PV δcKO mice, we focused on high-amplitude events seen above in the REM wavelet spectra (Fig. 3E) for further characterization.
High-amplitude transient bursts of theta-frequency activity observed in PV δcKO mice enriched during REM
Throughout the recording sessions, PV δcKO mice can be distinguished by the presence of periodic high-amplitude bursts of activity primarily during sleep states and preceding transitions into NREM sleep. Although these events resemble some features of sleep spindles (frontal prominence, waxing and waning envelope, clustering of incidence), they are distinguished by their persistence during REM sleep when spindles typically subside to theta oscillations. Here, we present three examples of these events occurring in a PV δcKO mouse. When observed during wakefulness (Fig. 5A, left), a portion of high-amplitude bursts of activity (20.4 ± 12.8% of events detected during wake) are associated with behavioral arrest, indicated by the cessation of preceding movement for the duration of the event (accelerometer records). Additionally, we show repetitive occurrence of these events during a complete REM bout (Fig. 5A, center) containing many high-amplitude events. While repetitive bursts of activity were evident in frontal and parietal EEG, relative silence persisted in EMG and accelerometer records. An expanded timescale of one of these events demonstrates the background persisting theta oscillations of REM present in the parietal EEG and highlights the frontal prominence of event incidence and amplitude. A representative event occurring during the termination of a REM bout (Fig. 5A, right) shows individual event durations can last many seconds and that events typically cease upon state transition.
PV δcKO mice exhibit increased incidence of transient bursts of high-amplitude theta-frequency activity during state transitions and REM sleep. A, Three representative examples of high-amplitude synchronous EEG activity in a PV δcKO mouse. Records from top: hypnogram, spindle detection result, frontal (F) and parietal (P) EEG, nuchal EMG, and three-axis accelerometer. Examples include a singular event during quiet wake (left panel) in which the animal undergoes behavioral arrest for the duration of the event. The central panel displays an entire REM bout with high-amplitude events increasing toward the transition from REM back to wake. EEG for event indicated with red bracket is expanded below with strong parietal theta rhythm supporting scoring of REM brain state. The right panel presents a protracted (>4 s) high-amplitude event immediately preceding the transition from REM to wake. An expanded view of Figure 5A panels is given as Extended Data Figure 5-1. B, Average spectrograms for wake to NREM state transitions, both genotypes display a slow ramping of increased power while transitioning to NREM, with PV δcKO mice exhibiting a higher frequency peak at the time of transition into NREM sleep. Two-way repeated-measures ANOVA reveals significant Time × Genotype interaction for the evolution of peak frequency during wake to NREM transition (F(22,308) = 1.730, p = 0.0234). C, REM to NREM transitions show high incidence of high-amplitude events in PV δcKO mice, with increase in theta-frequency peak power during REM leading up to the state transition. Two-way repeated-measures ANOVA shows significant Time × Genotype interaction for peak power (F(22,308) = 4.118, p < 0.0001). D, NREM to wake state transitions are spared from high-amplitude oscillatory bursts. Two-way repeated-measures ANOVA shows significant Time × Genotype interaction with respect to peak frequency during state transition to waking (F(22,308) = 3.538, p < 0.0001). E, Spectral profile of events occurring during REM revealed following normalization of total REM spectra to event-free REM background activity. Event frequency spectra power is more prominent in frontal (top) than parietal (bottom) electrodes. Two-way repeated-measures ANOVAs reveal significant Frequency × Genotype for both frontal (F(255,3570) = 5.900, p < 0.0001) and parietal (F(255, 3570) = 5.229, p < 0.0001) electrodes.
Figure 5-1
Enlarged panels from Figure 5A. 5A-1) Left panel of Figure 5A showing behavioral arrest during abnormal EEG event. 5A-2) Center panel of Figure 5A showing REM sleep. 5A-3) Right panel of Figure 5A showing event cessation after transitioning out of REM. Download Figure 5-1, TIF file.
To determine if events were constrained by brain state, we identified state transitions and characterized the peak frequencies observed through the transitions. In wake to NREM state transitions (Fig. 5B), we observed an emergence of some transient bursts around the transition evidenced by a transient increase in the frequency of spectral peaks in PV δcKO mice. Transitions out of REM where animals briefly transition through wake like EEG (Fig. 5C) were dominated by the presence of high-amplitude events in REM that subsided upon state transition. The presence of high-amplitude events resulted in an increase in the peak power during REM preceding the transition in PV δcKO mice. NREM to wake transitions (Fig. 5D) exemplified a transition occurring in the absence of high-amplitude events, but a Frequency × Genotype interaction was observed for peak frequency of the transition.
Because of the relatively high event prevalence in REM states, uniformity of REM EEG in rodents, and lack of sleep spindles contributing to periodic increases in spectral power, we focused further characterization to REM states. Removal of periods of REM EEG with events allowed the generation of a baseline REM spectrum for each animal. We then could use these baseline spectra to normalize the spectra previously generated from the entirety of REM states across the full recording session. This revealed the theoretical spectral profile of the events alone (Fig. 5E) since ongoing normal REM spectral content would be subtracted from the spectra. The spectral profile of REM events in PV δcKO showed multiple evenly spaced peaks consistent with harmonics produced from a fundamental frequency of 7 Hz. The power of event spectra was greater in frontal EEG (Fig. 5E, top) than parietal EEG (Fig. 5E, bottom), but event-related spectral power was still larger than any persisting spectra from WT mice.
PV δcKO mice exhibit mild behavioral abnormalities compared with their WT littermates
To further evaluate whether findings of altered EEG during sleep states could represent changes in brain activity relevant for cognitive functions, we performed an initial battery of behavioral assays in PV δcKO mice and compared results with littermate controls. EEG signals of similar characteristics to those identified here (i.e., sleep spindles) have impact on motor and cognitive learning, so we focused on screens associated with these behaviors (Kam et al., 2019; Peyrache and Seibt, 2020). A 48 h spontaneous activity monitoring identified no differences in amount of ambulatory activity or rearing events between PV δcKO mice and their WT littermates (Fig. 6A,B). A potential motor coordination and/or learning deficit was followed up by testing a multiday accelerating rotarod test (Fig. 6C) and an inverted screen test (Fig. 6D). Neither revealed a deficit in PV δcKO mice. To screen for potential deficits in working memory, spontaneous alterations were measured in a Y-maze task. The task revealed no difference in percent alterations across genotypes and no difference in the distance traveled (Fig. 6E,F). Altogether, the loss of Gabrd in PV neurons and subsequent changes to EEG patterns observed in sleep states did not correlate with gross changes in any behavioral parameters we assessed.
Lack of behavioral phenotype in initial battery. A, Locomotor activity and (B) rearing were measured for 48 h as described in the Materials and Methods (n = 9 WT and 9 PV δcKO mice). A trend toward a rearing difference between genotypes was detected (F(1,16) = 3.422, p = 0.0829). C, The possible motor coordination phenotype was further evaluated with a multitrial accelerating rotarod test. No genotype difference was discerned (n = 9 WT and 9 PV δcKO mice; F(1,16) = 1.703, p = 0.2103). D, Similarly, mice performed indistinguishably on an inverted screen test (n = 9 WT and 9 PV δcKO mice, p > 0.99, Mann–Whitney test). E, F, Cognitive screening was performed with an alternating Y-maze test. No difference in alternation rate of distance traveled was detected (n = 9 WT and 9 PV δcKO mice, unpaired two-tailed t tests, p = 0.66 and 0.70 respectively). G, H, Running wheel performance of WT and δcKO mice. Based on effect sizes of previous studies, an n of 20 mice per group (equal males and females) was chosen a priori. The results revealed a mild phenotype in Day 2 of complex wheel performance (main effect of genotype p = 0.0389; F(1,45) = 4.524).
Sleep deprivation has led to selective deficits in complex running wheel, attributable to lack of motor learning consolidation (Nagai et al., 2017). Somewhat consistent with this, we found mild deficits in wheel running that were selective to Day 2 in a complex wheel running task (Fig. 6G,H). These results suggest that there could be mild alterations in motor learning consolidation because of the sleep disturbances observed.
Viral rescue of Gabrd expression in PV δcKO mice reduces presence of high-amplitude bursts in REM
Because our recordings are performed in adult mice, we cannot differentiate phenotypic changes arising from an absence of δ-mediated GABA signaling during the recording session from either a developmental alteration of network organization or a homeostatic compensation following Gabrd loss in PV+ cells. To test whether the high-amplitude events observed during REM resulted from lack of δ-containing receptors during the recordings, we utilized a viral strategy to reintroduce Gabrd expression in a PV+-specific manner in adult mice (Fig. 7A). We produced AAV particles packaged in the PHP.eB serotype containing an expression construct to allow for the widespread cre-dependent expression of Gabrd in PV+ cells following systemic administration of virus. Retro-orbital injection of either Gabrd-IRES-GFP expressing “Rescue” virus or GFP expressing “Control” virus into adult PV δcKO littermates served as the primary rescue arm of the experiment. We included an additional group of PVCre+/− mice with WT Gabrd alleles to serve as controls for possible overexpression of Gabrd in PV+ interneurons and potential ectopic Gabrd expression as not all PV+ cells in the CNS normally express Gabrd (Belelli and Lambert, 2005). Cell type-selective viral expression in the cortex was confirmed histologically in cortical tissue through colocalization of virally expressed control GFP and parvalbumin in cortical interneurons (Fig. 7B). In five animals examined, transduction efficiency, measured as the percentage of PV-immunopositive cells that were positive for GFP, was 79.6 ± 4.6%. Selectivity, measured as the percentage of GFP-positive cells that were PV-immunopositive, was 95.7 ± 2.2%. We did observe expression of GFP in thalamic nuclei other than the PV+ reticular nucleus, but we attribute this expression to low PV expression in this population, as indicated by the Allen Brain Atlas (Yao et al., 2021). For the rescue virus, IRES coupled expression was apparently insufficient to allow detection of native GFP fluorescence. Therefore, functional validation of rescue was not possible. Instead, we verified expression with amplified antibody-based detection of GFP, which confirmed cortical expression and weak thalamic expression at levels lower than the control virus (Extended Data Fig. 7-1).
Viral delivery of Gabrd to PV δcKO mice reduces high-amplitude bursts of theta-frequency oscillations during REM sleep. A, Experimental design for viral reintroduction of Gabrd into PV δcKO mice. Mice of either PV δcKO (Gabrdfl/fl × PVCre+/−) or PVCre+/− background were injected systemically with AAV.PhP.eB particles containing either cre-dependent Gabrd-IRES-GFP or GFP expression cassette. B, Coexpression of viral GFP with parvalbumin (PV) antibody staining in cortex indicating cell type specificity of viral expression pattern. Additional data on rescue expression is given in Extended Data Figure 7-1. C, Five minutes of combined frontal REM EEG from a group of viral treated mice. Horizontal brackets on records for each mouse indicate regions shown on expanded timescale. Fewer and lower-amplitude events are observed in PVδ cKO mice that received Gabrd (purple) than those that were treated with GFP control (green). All records are on the same amplitude scale to demonstrate difference in the size of events during REM. D, REM power spectra in viral treated PV δcKO mice show significant Treatment × Frequency interaction for both frontal (left, F(255,3328) = 8.288, p < 0.0001) and parietal (right, F(255,3328) = 1.806, p < 0.0001) spectra. E, REM power spectra in PVCre+/− control mice show significant Treatment × Frequency interaction for both frontal (left, F(510,3328) = 2.839, p < 0.0001) and parietal (right, F(510,3328) = 1.876, p < 0.0001) EEG. F, REM event spectra from frontal (left) and parietal (right) electrodes indicate a significant reduction in the prominence of high-amplitude transient activity in PV δcKO mice treated with Gabrd expressing virus compared with GFP (G) control. Two-way ANOVA revealed significant Treatment × Frequency interaction for both frontal (F(765,5888) = 5.910, p < 0.0001) and parietal (F(765,5888) = 4.056, p < 0.0001) EEG. Inset: Total number of events detected during REM, significant interaction from one-way ANOVA (F(3,23) = 10.03, p = 0.0002). Horizontal bars in panels D and E indicate frequencies with significant group differences following post hoc cluster-based permutation testing; bars and asterisks in panel F represent frequencies that are different from the PV δcKO mice treated with GFP virus following Dunnett's multiple comparisons.
Figure 7-1
Amplified GFP immunolabeling from rescue virus in 5 brain areas. The panels show A) sagittal view of mouse brain, B) parietal cortex C) hippocampus D) thalamus E) striatum and F) cerebellum. G) Quantification of cortical GFP labeling in seven rescue animals and two control-GFP virus animals. A region of interest of equivalent to panel B was selected, and cell density was calculated as the total number of labeled cells divided by the area. Calibration bar in F is for panels B-F. Download Figure 7-1, TIF file.
Following a 4 week incubation period after injection to allow for adequate expression of viral constructs, we again performed 12 h light cycle EEG recordings to assess changes to the transient high-amplitude events in REM following the reintroduction of Gabrd into PV+ cells of PV δcKO mice. In pooled REM states across the recording session, abnormal events were reduced in PV δcKO mice that received rescue virus yet persisted in their littermates that received control viral injections (Fig. 7C, left records). These results support the primary hypothesis that abnormalities arise from changes to acute GABA signaling through δ-containing receptors in PV neurons rather than from a developmental consequence of early gene deletion.
We did observe some unexpected results in other limbs of the experiment. Typical REM EEG was observed in PVCre+/− mice that received the rescue viral construct, suggesting no strong phenotype was associated with δ overexpression. When compared with PV+/− mice that received rescue virus, those that were treated with control GFP expressing virus showed some unexpected differences in REM spectra (Fig. 7E, purple), but they remained free of the transient high-amplitude, long-duration events observed in PV δcKO mice (Fig. 7C, right records).
In PV δcKO mice, average power spectra from REM states showed a loss of multipeak signature in frontal EEG in the rescue condition (Fig. 7D, left). However, in both frontal and parietal EEG, a broader separation between rescue and GFP conditions was present in beta and low gamma frequencies that was not previously observed between WT and PV δcKO mice. Interestingly, similar differences were present between the two viral conditions in PVCre+/− mice (Fig. 7E). We interpret these results to suggest a common unanticipated effect of the GFP viral condition across both genetic backgrounds. Additionally, PV+/− mice in the rescue condition showed no gross differences from naive PVCre+/+ mice (Fig. 7E) supporting a lack of over/ectopic expression effects by the rescue construct. Finally, event spectra, generated as described above, showed multipeaked signature remained in PV δcKO mice injected with control virus, and the power of these peaks was reduced in all other conditions in both frontal and parietal EEG (Fig. 7F). These results support the conclusion that GFP-associated effects described above were not associated with high-amplitude REM events in PVCre+/− δcKO mice.
Overall, the findings demonstrate the effective rescue of normal REM EEG signatures by viral reintroduction of Gabrd through attenuation of high-amplitude transient events in PV δcKO mice, providing valuable insights into the role of Gabrd in PV+ neurons for regulating oscillatory activity during REM states, an effect not anticipated by previous work on the role of PV neurons in brain oscillations.
Discussion
GABAARs containing δ subunits have been primarily studied in select principal cell types (Nusser et al., 1998; Stell and Mody, 2002; Cope et al., 2005; Brickley and Mody, 2012). PV+ interneurons also express functional δ subunit containing GABAARs, yet their role in controlling PV+-related brain activity remains unclear (Glykys et al., 2007). After confirming selective functional loss Gabrd in PV+ cells, we surveyed the impact on cortical EEG. PV δcKO mice exhibited similar sleep and waking behavior to WT littermates, but PV δcKO mice exhibited altered power spectra of parietal and especially frontal EEG during both NREM and REM sleep states. Increased sigma frequency oscillations in PV δcKO mice corresponded to higher amplitude and longer sleep spindles with no increase in spindle number. During REM sleep, PV δcKO mice displayed transient bursts of oscillations with a principal frequency near 7 Hz. REM spectral analysis revealed a multipeak signature in PV δcKO mice with higher power in frontal than parietal EEG. Finally, viral reintroduction of Gabrd expression into PV+ cells of PV δcKO mice rescued the altered REM EEG. Our results demonstrate an important acute role for slow inhibition in PV+ neurons for the maintenance of normal EEG structures of REM sleep.
Previous studies of PV δcKO mice examined hippocampal gamma oscillations in vitro and in vivo (Ferando and Mody, 2013, 2015; Barth et al., 2014). Constitutive Gabrd KO mice show a higher frequency peak of kainate-induced in vitro CA3 gamma oscillations, and both heterozygous and homozygous PV δcKO mice showed a higher peak frequency than WT animals (Ferando and Mody, 2013, 2015). An in vivo study of CA1 LFPs demonstrated that ovarian cycle-linked fluctuation of gamma power was abolished in female PV δcKO mice (Barth et al., 2014). Our study provides a broader EEG survey of cortical network activity. The design included frontal EEG, which exhibited the largest sleep-associated phenotypic differences in oscillatory activities. Increased spindle amplitudes and durations in PV δcKO mice could indicate more cells responding to ascending thalamic inputs during spindles, as PV+ cortical interneuron activity increases during spindle events (Niethard et al., 2018; Brécier et al., 2022), and absence of δ-mediated inhibition of these cells should increase their recruitment by excitatory inputs.
The emergence of transient high-amplitude events during REM states was unexpected based on previously reported subtle REM phenotypes in hippocampal LFP from PV δcKO mice (Barth et al., 2014). The events in PV δcKO mice have similar frequencies and topographic distributions to spike-wave discharges (SWDs) in models of absence epilepsy (van Luijtelaar and Coenen, 1986; Buzsáki et al., 1990; Coenen et al., 1991; Frankel et al., 2005). Although the events were relatively suppressed during active wake, when they occurred during wake a subset were associated with behavioral arrest, possibly consistent with an absence seizure. Gabrd mutations are associated with human epilepsies (Dibbens et al., 2004; Macdonald et al., 2012). Therefore, future study of waking behavior in these mice could be investigated as a novel model of absence epilepsy. In contrast to our observations of Gabrd deletion restricted to PV cells that seems to induce potential epileptiform discharges, an interneuron-wide deletion of Gabrd decreased seizure susceptibility in response to kainic acid administration (Lee and Maguire, 2013). However, the events in PV δcKO mice diverge from typical discharges seen in absence epilepsy models due to occurrence in REM. REM sleep in both rodent models and patients with absence epilepsy is typically void of SWD (Strohl et al., 2007; Ng and Pavlova, 2013). Furthermore, SWD epileptiform activity that develops in APP/PS1 mice spares REM (Jin et al., 2018).
The phenotype also shares similarity to narcolepsy. Hypocretin (HCRT)/orexin-deficient narcoleptic mice exhibit remarkably similar transient events of high-amplitude 7 Hz bursts of activity during REM sleep (Bastianini et al., 2012; Vassalli et al., 2013). Events in HCRT KO mice are exaggerated in frontal EEG and occur in medial prefrontal cortex but not hippocampus (Vassalli et al., 2013). If high-amplitude REM events in PV δcKO mice result from similar circuit changes as REM events in HCRT KO mice, then previous PV δcKO studies utilizing hippocampal LFP would likely not detect these events (Vassalli et al., 2013; Barth et al., 2014). To our knowledge, roles for tonic inhibition, cortical PV interneurons, or δ subunits have not been found in human narcolepsy. However, the δ preferring agonist gaboxadol (THIP) is in trials to increase slow-wave sleep in primary insomnia patients (Wafford and Ebert, 2006). Interestingly, the EEG phenotype in our studies was associated with only mild deficits in the behavioral studies we conducted (Fig. 6). The observation of motor learning consolidation deficits could be followed up with additional days of testing or more challenging tasks to determine whether effect size might increase with higher demands.
Cortical PV+ interneurons exhibit high levels of activity during REM sleep (Niethard et al., 2018; Aime et al., 2022; Brécier et al., 2022). Although EEG analysis prevents determining whether the high-amplitude events observed during REM correlate with aberrant PV+ interneuron activity in vivo, our rescue experiments support a role for δ mediated inhibitory tone in PV+ cells for preventing this abnormal activity. It is unclear why the loss of slow/tonic inhibition of PV+ neurons would manifest selectively in REM sleep. Changes to the ambient GABA concentrations across brain states could activate extrasynaptic δ-subunit containing GABAARs. Tonic current in the hippocampus and cortex is regulated by synaptic activity (Glykys et al., 2007; Trujeque-Ramos et al., 2018); however, dynamic regulation of GABA transporters (Gaspary et al., 1998; Richerson and Wu, 2003) and direct release by astrocytes (Liu et al., 2000; Kozlov et al., 2006) are also potential mechanisms that could regulate ambient GABA. A better understanding of the dynamics of extracellular GABA across brain states will help clarify how the loss of tonic inhibition in PV+ neurons may result in phenotypes limited to specific brain states.
While Gabrd expression in PV+ cells has only been reported in interneurons of the cortex, hippocampus, and amygdala (Glykys et al., 2007; Ferando and Mody, 2015; Yao et al., 2021; Antonoudiou et al., 2022), our cre/lox approach ablates Gabrd in all PV cells across the brain. Additionally, our choice of widespread systemic delivery for viral rescue does not allow for regional or PV subtype specificity. Therefore, we cannot exclude potential effects of PV δcKO in cell populations beyond those that motivated our study. Noncortical PV neurons relevant to oscillations herein include those found in the reticular nucleus of thalamus (nRT) whose activity drives the transition of thalamocortical cell firing mode to NREM burst firing and spindle oscillations (Fernandez and Lüthi, 2020). Considerable δ-associated tonic inhibition occurs in the thalamus, yet nRT PV+ cells do not respond to delta selective agonists or express δ subunits (Pirker et al., 2000; Cope et al., 2005). Therefore, it is unlikely that ablation of Gabrd in nRT neurons alone drives the oscillatory phenotypes presented here.
Additionally, Gabrd expressing thalamic relay nuclei of the ventrobasal complex and the lateral dorsal nucleus transcribe PV RNA, yet protein expression in these regions is below the threshold of detection by immunohistochemical methods (Tanahira et al., 2009). Moreover, PV cre driver lines drive reporter genes in these regions (Tanahira et al., 2009). Although these nuclei are not classically considered PV+, the activity of the PV locus in these cells would delete Gabrd. We observe evidence of cre activity in these cells demonstrated by expression of viral GFP in our study, raising concern for off target cKO of Gabrd in a portion of thalamocortical cells. Although signaling through δ-containing receptors in thalamic relay nuclei has been implicated in electrocortical features of slow-wave sleep (Vyazovskiy et al., 2005; Mesbah-Oskui et al., 2014), the thalamic nuclei affected by unintended cre activity project to cortical regions distant from the frontal EEG electrodes that show the largest phenotype (Vertes et al., 2015). Further, constitutive KO of Gabrd, which removes tonic inhibitory currents from all thalamocortical cells, does not cause abnormal high-amplitude events in EEG during REM sleep (Mesbah-Oskui et al., 2014).
Although our work does not pinpoint how δ subunits in PV+ neurons are important for spindles, there are some plausible scenarios. Dynamics of cortical spindles are thought to contain important contributions from cortical interneurons (Hartwich et al., 2009; Peyrache et al., 2011; Averkin et al., 2016; Niethard et al., 2018; Brécier et al., 2022). Tonic inhibition in these interneurons during REM and other states devoid of spindles may reduce responsiveness of cortical network. With loss of tonic inhibition, abnormal spindle-like activity encroaches. Although speculative, this scenario presents several testable ideas for future work. For instance, this possibility predicts that in WT animals during REM, interneurons are under strong inhibitory control.
Our interpretation of the extent of rescue following viral reintroduction of Gabrd into PV δcKO mice is complicated by additional EEG findings in animals receiving GFP virus. However, despite a common increase in power of beta and low gamma ranges in GFP treated mice, high-amplitude REM events remained reliably elevated and detectable only in PV δcKO and not PVCre+/− mice in this viral condition. The basis for GFP-induced changes is not clear, but we note that GFP expression levels, judged by fluorescence, were much higher in control conditions than in the δ subunit rescue condition. Thus, the unexpected effects of overexpression of GFP likely did not influence the δ-rescue arm of experiments (Fig. 7). Despite these limitations, the present study demonstrates a novel role for δ subunit containing GABAARs on PV+ neurons in the regulation of normal oscillations during sleep states. Future studies to determine what cellular and circuit activities underlie the high-amplitude REM events observed in PV δcKO will be a crucial step to elucidate the mechanism by which tonic inhibition of PV neurons contributes to the maintenance of normal REM oscillations in the cortex.
Footnotes
We thank members of the Taylor Family Institute for Innovative Psychiatric Research for discussion and input and Dr. Jamie Maguire for the floxed δ mice, Dr. Jin-Moo Lee for the PV-Cre mice, Jinli Wang for statistical advice, and Sara Conyers for assistance with behavior data collection. The work was funded by NIMH grants MH123748 (S.M), MH122379 (C.F.Z, S.M), MH126548 (P.M.L), NICHD grant P50 HD103525 (Washington University Intellectual and Developmental Disability Research Center), the Taylor Family Institute for Innovative Psychiatric Research (S.M, C.F.Z), and the Bantly Foundation (C.F.Z).
C.F.Z. is a member of the Scientific Advisory Board for Sage Therapeutics and holds equity in Sage Therapeutics. Sage Therapeutics had no role in the design or interpretation of the experiments herein. The remaining authors declare no competing financial interests.
- Correspondence should be addressed to Steven Mennerick at menneris{at}wustl.edu.