P4–P7 mice were anesthetized using isoflurane. Anesthetized mice were then decapitated rapidly and their brains were quickly dissected out. The removed brains were transferred into an ice-cold cutting solution containing: 26 mm NaHCO₃, 210 mm sucrose, 10 mm glucose, 2.5 mm KCl, 1.25 mm NaH₂PO₄, 0.5 mm CaCl₂, and 7 mm MgCl₂. The brains were kept in a holding chamber for ~1 min prior to relocation to a vibratome (Leica, VS2000) for slicing. Slices, 300 μm in thickness, were obtained and transferred into an incubation chamber warmed to 35–37°C. Slices were kept at 35–37°C for 30 min. After that, the incubation container and the slices were placed on a benchtop at room temperature for at least 1 hour prior to experiments. The incubation chamber contained artificial cerebrospinal fluid (ACSF) with the following components: 125 mM NaCl, 26 mM NaHCO₃, 2.5 mM KCl, 1 mM NaH₂PO₄, 1.3 mM MgCl₂, 1.5 mM CaCl₂, and 12 mM d-glucose.

To image calcium activity, slices were placed in a Warner holding chamber superfused continuously with warm (33–35°C) ACSF. For recording calcium activity, we used an Axio Zoom V.16 (Zeiss) microscope with a PlanNeoFluar Z 1X, 0.25NA objective and a 56 mm working distance. Using multi-imaging acquisition software (Micromanager), 300 s long movies recorded using an sCMOS camera (pco.edge 4.2) at a 500 × 513 pixel resolution (50 ms exposure time; 6,000 frames).

TIFF files (6,000 frames per file) were imported into the Fiji-ImageJ¹ (41) software package. For each TIFF file, we selected a region of interest (ROI) such as the CA3 region of the hippocampus or the anterior, medial, or posterior cortex to measure changes in calcium fluorescence. Using the Fiji-ImageJ built-in functions, we plotted the z-axis profile for each ROI, and then obtained the fluorescent value per each frame for every ROI. The fluorescent values (F) for the 300 s-long recording period were then exported to Axograph analysis software (v.1.7.5, https://axograph.com/). Using the built-in functions of Axograph, we obtained the ΔF/F time series for each calcium event. The cut-off point for a recorded calcium event was 0.01 ΔF/F. For each calcium event, we measured its amplitude and duration (50% of the peak).

ROIs and signals ΔF/F for the 2D plots were determined according to a semi-automatic procedure as described in (23). Briefly, frames (500 × 500 pixels, inter-pixel distance: 6 μm) were inspected for the external edges of the area covered by each hemisphere, and only pixels within the area delineated by these lines were further considered. Pixels within this area were divided into ROIs by using the SLIC (simple linear iterative clustering) algorithm (42) and, for each ROI, the fluorescence signal F was estimated as the average value of the fluorescence intensity across all pixels in the ROI. The series of variations ΔF was then estimated for each ROI as ΔF = (F − m)/m, where m is the moving average value of F (moving window: 30 s). ROIs were numbered consecutively along the anteroposterior direction, with lower values assigned to ROIs towards the anterior cortex and higher values assigned to ROIs towards the posterior cortex and the CA3 area of the hippocampus.

In parallel to calcium fluorescence events, we recorded the local field potentials (LFP) from the CA3 region of the hippocampus using a microarray with eight bipolar electrodes (MicroProbes). Data were recorded through a RHD2000 multichannel amplifier with RHD2000 interface software (v1.5, Intan Technologies) at a 20 kHz sample rate and imported in MATLAB (R2020b, MathWorks, Inc.). The raw data were lowpass filtered (20^th order FIR filter, cutoff frequency 1.3 kHz), and the resultant signals were further filtered (cutoff frequency: 80 Hz, 8^th order Chebyshev Type I lowpass filter applied in both directions to avoid phase distortions) before being down-sampled to 200 Hz. For each recording, 30 s-long segments were selected for spectral analyses. LFP channels in each segment were standardized and processed via continuous wavelet transform (10 voices per octave, analytic bump wavelet) to perform a time-frequency analysis. All analyses were performed blind to the genotype.

GraphPad Prism 8 software (GraphPad) was used to graph the data for statistical analyses. We used Mann–Whitney U tests to determine significance between groups of data.

To determine whether KCNQ2 channels alter interneuron GABAergic network excitability, we first recorded interneuron activity under basal conditions (2.5 mM Ko). For these experiments, we used horizontal slices, preserving the connections between different network subregions. Figure 1 compares the fluorescence activity across different hemispheres. We note that the observed calcium fluorescence changes reflect activity across a population of neurons, as mesoscale imaging does not allow for detection of calcium events at a single-cell resolution. In our control slices, the regions with the highest basal activity were the hippocampal formation and in particular the CA3 region of the hippocampus and the posterior cortex, likely reflecting the entorhinal cortex (Figures 1A,B). This was consistent with earlier reports showing that the hippocampal formation shows spontaneous activity in slices from neonatal mice (43, 44).

In the absence of KCNQ2 channels, we found an even greater activity in these subregions reflected as changes in the amplitude, duration, or frequency of calcium events (Figure 1). For instance, in the posterior cortex we found a two-fold increase in the amplitude of the calcium signals (control: 0.053 ± 0.052 ΔF/F, n = 24; IN:Kcnq2: 0.14 ± 0.095 ΔF/F, n = 25; p < 0.0001, Mann–Whitney U test), whereas in the CA3 region of the hippocampus we observed changes both in the duration of the calcium signal (control: 0.75 ± 0.28 s, n = 24,; IN:Kcnq2: 0.81 ± 0.26 s, n = 26; p = 0.0006, Mann–Whitney U test) and its frequency (events/s; control: 0.012 ± 0.012 Hz, n = 28; IN:Kcnq2 0.019 ± 0.015 Hz, n = 26; p = 0.0021, Mann–Whitney U test) (Figure 1B). Thus, deletion of Kcnq2 from GABAergic neurons leads to increases in activity primarily in the hippocampus and posterior cortex, which is comparable to our earlier findings of deleting KCNQ2 channels from excitatory neurons (23).

The low basal activity in the neocortex in 2.5 mM extracellular potassium prevented us from determining whether KCNQ2 channels also play a role in the neocortex, a region that highly expresses KCNQ2 channels across development. This is because spontaneous neocortical activity in slices at these ages is sparse (45). Thus, we examined the effects of ablating KCNQ2 channels from GABAergic cells in slices superfused with 8 mM Ko. Increasing the extracellular potassium depolarizes neurons as it shifts the Nernst equilibrium potassium potential to more positive values, leading to increased neuronal excitability. Despite the change in the extracellular potassium concentration, calcium activity was still restricted to the hippocampal formation in our control slices (Figure 2A). However, in slices deficient of KCNQ2 channels, we detected activity across the hemisphere with prominent activity both at the hippocampal formation and the neocortex. This is clearly illustrated in the two-dimensional plot in Figure 2A. Notice that in the absence of KCNQ2 channels fluorescence signal is apparent across the hemisphere.

To quantify the impact of KCNQ2 deletion from GABAergic interneurons, we divided our hemispheres into different subregions and determined the amplitude, duration, and frequency of the calcium events (Figure 2B). With respect to the amplitude, we found that Kcnq2 ablation increased calcium activity primarily in the posterior and anterior cortex, whereas the duration of the calcium signals was increased in all regions except the posterior cortex. The amplitude and duration of the calcium signals were a function of the number and duration of neurons active at any given moment. Thus, the aforementioned changes suggest that Kcnq2 deletion increases the population activity of interneurons across the neocortex. This would be consistent with an increased excitability of interneurons, making them more prone to firing action potentials following incoming activity and further explaining the increase in the frequency of calcium events in the neocortex (Figures 2B,C).

In parallel to our calcium measurements, we recorded the LFP from the CA3 region of the hippocampus. With an elevated potassium concentration (8 mM Ko), we routinely observed robust LFP activity independent of the presence or absence of KCNQ2 channels. We also noticed that the LFP activity was comprised of continuous events with intermittent bursts lasting ~30 s (Figure 2B), with this bursting being independent of the presence of KCNQ2 channels. Next, we assessed the spectral content of these bursts (Figure 3A) and we found that, while bursts had power spectra spanning across the theta (3–7 Hz), alpha (8–14 Hz), and beta (15–30 Hz) frequencies (Figure 3B), oscillations in the range 5–10 Hz (i.e., upper theta/low alpha frequencies) showed a significant increase (Figure 3C). The 5–10 Hz frequency band likely reflects oscillations in the alpha frequency range, which have been previously associated with increased fast spiking interneuron activity.

GABAergic interneurons release GABA and activate GABAA receptors, which in turn inhibit or facilitate activity based on the equilibrium membrane potential of chloride ions. At our developmental time point (P4–P7), GABA is primarily inhibitory in the neocortex (45, 46) whereas in the hippocampus it still transitions from excitatory to inhibitory as recently reported by Murata et al., i.e., excitatory at P3 and inhibitory by P7 (46). Additionally, multiple groups have shown that in addition to phasic GABA release, GABA is found tonically activating extracellular GABAA receptors (47). We previously found that application of picrotoxin, a GABAA receptor inhibitor that blocks both synaptic and extrasynaptic GABAA receptors, increases excitability in brain slices of 1 week-old animals (23). Thus, we hypothesized that eliminating GABAA receptor activity would also allow for greater interneuron excitability due to unabated excitatory network activity.

To investigate the effects of GABAA receptor inhibition on the interneuron network activity, we first applied 50 μM picrotoxin in slices from control animals (Figure 4A). Blocking GABAA receptors had several effects. First, calcium activity was no longer restricted to the hippocampal formation, but instead was present across multiple regions. This was especially evident in the anterior cortex, where the barely detectable initial activity levels increased to large calcium events in the presence of picrotoxin. Second, the activity became more periodic and synchronous, with bursting activity followed by quiescence periods (Figures 4A,B). Third, the amplitude and duration of the calcium events became larger and longer, likely reflecting greater recruitment of interneurons in the network (Figures 4B,C).

Next, we investigated the effect of blocking inhibition in the IN:Kcnq2 slices. Similar to the slices from control animals, we found that activity was synchronous and present across all regions in the IN:Kcnq2 slices (Figures 4A,B). We observed robust events even in the medial cortex, a region typically showing very low-level activity in slices from control animals (Figures 4B,C). Additionally, we observed large calcium events independent of whether their duration was brief or long lasting (Figure 4C). Overall, the loss of KCNQ2 channels from interneurons further amplified the pro-excitatory effects of picrotoxin. Together, our data suggest that the main role of GABAergic interneuron activity is inhibitory at the network population level.

Multiple groups have shown that network population events in slices depend on ongoing excitatory network activity (43, 44, 48, 49). In neonatal mice, NMDA receptors have been proposed to have a strong excitatory role (43, 50). As our recording conditions included elevated potassium concentrations, which would make it easier to relieve the magnesium block, the role of the NMDA receptors might have been enhanced. Indeed, we found that in slices from control animals, application of the competitive NMDA receptor antagonist APV (25 μM) substantially reduced the LFP, the calcium event frequency, and on many occasions the amplitude and duration of the calcium events (Figure 5). We observed similar effects in slices from IN:Kcnq2 mice (Figure 5A). However, one notable exception was in the anterior cortex, where application of APV in IN:Kcnq2 slices reduced the frequency of the events but had very weak effects on the amplitude and duration of the calcium events (Figures 5B,C).

Not surprisingly, simultaneous application of synaptic blockers for AMPA, NMDA, and GABAA receptors almost entirely eliminated calcium activity in control slices (Figure 6A). Similarly, we found a very strong reduction in interneuron activity in slices from IN:Kcnq2 mice upon application of all three synaptic blockers. We did notice some remaining low-level activity in the medial and anterior cortex (Figures 6B,C; Table 1); however, this activity was barely above background levels. Thus, GABAergic interneuron activity in neonatal slices is driven primarily by fast excitatory transmission, and loss of KCNQ2 channels from interneurons increases their population activity.

KCNQ2 channels are a core component of the M-current (51, 52), a slowly activating non-inactivating voltage-gated potassium channel. The M-current’s main function is to prevent excessive firing (53). Thus, following strong activity, KCNQ2 channels and the M-current are activated, leading to spike frequency adaptation. This quiescence period is most typically associated with excitatory cells, and in particular pyramidal neurons (54). Consequently, for many years, the roles of the M-current and KCNQ2 channels in interneurons have been ignored. Despite this, a few studies using pharmacology have shown that a KCNQ-mediated current is present in interneurons, including somatostatin and parvalbumin GABAergic cells of the hippocampus (33, 35, 55). We and others have also shown that deletion of KCNQ2 channels from parvalbumin interneurons increases their firing activity, leading to elevated spontaneous inhibitory transmission (35, 36). Our study reveals that deletion of KCNQ2 channels from GABAergic neurons has a significant effect on the posterior cortex, which includes the entorhinal cortex, as well as on the CA3 area of the hippocampus. Previous studies have shown that the entorhinal cortex displays prominent spontaneous activity early in life (43, 44), potentially due to its high level of interconnectivity (56), although the exact mechanism remains unclear. The absence of KCNQ2 channels from GABAergic neurons is likely to amplify the propensity for spontaneous activity in these two regions, suggesting that the loss of KCNQ2 activity may have a more pronounced effect on these regions than on other cortical regions such as the medial and anterior cortex. Although the activity in these regions also increased in the absence of KCNQ2 channels from GABAergic neurons, the effect was much weaker than in the posterior cortex and hippocampal CA3 area.

A limitation of previous studies was the age of the animals. Most prior studies on KCNQ2 channels and interneuron activity were performed in juvenile or adult mice (33, 35, 36, 55). However, KCNQ2 channels are already expressed prior to birth, with their levels increasing as the animals further develop (18). Considering the key role of GABAergic transmission in the development of neuronal networks (48, 57), we sought to examine whether KCNQ2 channels also alter GABAergic interneuron activity in immature neurons. We found that, indeed, loss of KCNQ2 channel activity leads to elevated interneuron activity across the forebrain. Therefore, we suggest that KCNQ2 channels act as a brake early in development, limiting the recruitment of interneurons. This was evident in experiments in which we blocked GABAergic transmission.

Consistent with some recent papers (58, 59), we found that application of the GABAA receptor antagonist picrotoxin increases interneuron network excitability in both control and Kcnq2 GABAergic null neurons. This result is consistent with earlier and more recent findings that GABAA receptors are inhibitory at the network level within the first week of rodent life (46, 58–60). Although several studies have shown that GABA acts to depolarize neurons, at the network level its effects are likely inhibitory (59). This could be because tonic GABAA receptor activity likely acts as a shunt, clamping the resting potential of neurons to values that are depolarized yet subthreshold to action potential generation.

A major development over the last 10 years has been the recognition that KCNQ2 loss-of-function variants can lead to severe DEE (4, 5). Our current and prior work suggest that KCNQ2 loss of function would impair the function of both excitatory and GABAergic neurons. An unanswered question is whether loss of KCNQ2 channel activity from interneurons will further promote seizures or will dampen overall activity. In our previous study, deletion of Kcnq2 solely from interneurons led to increased excitatory transmission due to homeostatic changes (36). However, in constitutive Kcnq2 loss-of-function mice, pyramidal neuron excitability would already be elevated; thus, additional homeostatic changes should not lead to any further increase in neuronal excitability. We predict that loss of Kcnq2 from interneurons might instead have a protective role. Indeed, when we crossed our IN:Kcnq2 mice to Pyr:Kcnq2 mice (forebrain excitatory neuron Kcnq2 null mice), we found that the mice survived longer than the Pyr:Kcnq2 mice (data not shown). In contrast, pharmacological activation of KCNQ2 channels using KCNQ activators such as retigabine (61) will dampen excitability of GABAergic interneurons, which might decrease their ability to dampen overall network excitability. Indeed, a recent study showed the deleting KCNQ2 channels from parvalbumin interneurons allowed retigabine to have stronger anti-convulsant effects (35).

A limitation of this study is the lack of determination of the impact of KCNQ2 channels on interneuron neuronal physiology in different subregions at a single-cell level. For instance, we do not know whether all types of interneurons at this early stage of life depend on KCNQ2 channel activity or whether only a subset depends on them. We also do not know whether ablation of KCNQ2 channels from interneurons alters the properties of the excitatory cells or whether the synaptic drive to interneurons is affected. Lastly, it is possible that some of our effects might not solely be due to loss of KCNQ2 channel activity. For instance, KCNQ2 channels form heteromers with KCNQ3 and KCNQ5 channels (62). Thus, deletion of KCNQ2 might also alter the protein levels of KCNQ channels in general (37), further amplifying the observed phenotype.