Impaired KCC2 Function Triggers Interictal-Like Activity Driven by Parvalbumin-Expressing Interneurons in the Isolated Subiculum In Vitro

Abstract

The decreased expression of the KCC2 membrane transporter in subicular neurons has been proposed to be a key epileptogenic event in temporal lobe epilepsy (TLE). Here, we have addressed this question in a reduced model in vitro and have studied the properties and mechanistic involvement of a major class of interneurons, that is, parvalbumin-expressing cells (PVs). When exposed to the KCC2 blocker VU0463271, mouse subicular slices generated hypersynchronous discharges that could be recorded electrophysiologically and visualized as clusters of co-active neurons with calcium imaging. The pharmacological profile of these events resembled interictal-like discharges in human epileptic tissue because of their dependence on GABA_A and AMPA receptors. On average, PVs fired before pyramidal cells (PCs) and the area of co-active clusters was comparable to the individual axonal spread of PVs, suggesting their mechanistic involvement. Optogenetic experiments confirmed this hypothesis, as the flash-stimulation of PVs in the presence of VU0463271 initiated interictal-like discharges, whereas their optogenetic silencing suppressed network hyper-excitability. We conclude that reduced KCC2 activity in subicular networks in vitro is sufficient to induce interictal-like activity via altered GABAergic signaling from PVs without other epilepsy-related changes. This conclusion supports an epileptogenic role for impaired subicular KCC2 function during the progression of TLE.

Introduction

Accumulating evidence suggests a critical role of subicular circuits (Lévesque and Avoli 2020) in the initiation and propagation of interictal and ictal discharges in human patients (Cohen et al. 2002; Wozny et al. 2005; Huberfeld et al. 2007, 2011; Alvarado-Rojas et al. 2015) and animal models of temporal lobe epilepsy (TLE; Fujita et al. 2014; Toyoda et al. 2015; Wang et al. 2017).

In particular, the decreased expression of the K⁺/Cl⁻ cotransporter KCC2 (Payne et al. 1996), specifically in the subiculum and not in other regions of the surviving hippocampal formation (de Guzman et al. 2006; Palma et al. 2006; Huberfeld et al. 2007), has been highlighted as a potential key epileptogenic step for the progression of the disease (Miles et al. 2012). KCC2 is essential for maintaining physiological hyperpolarizing GABAergic signaling in mature circuits (Rivera et al. 1999), and its reduced function has been proposed to underlie the emergence of pathological network patterns, such as subicular-initiated interictal activity (Cohen et al. 2002; Buchin et al. 2016) interictal-ictal transitions (Huberfeld et al. 2011; Alvarado-Rojas et al. 2015) and generalized convulsions (Wang et al. 2017; Kelley et al. 2018).

To date, no study has directly examined whether pharmacologically reducing the activity of KCC2 in isolated subicular networks is, by itself, sufficient to trigger epileptiform activity or if other epilepsy-related changes are necessary for this increased excitability. This is an important question for assessing subicular KCC2 downregulation as a truly independent epileptogenic change. Although experiments using the selective KCC2 blocker VU0463271 (Delpire et al. 2012) have highlighted the importance of KCC2 as a regulator of hyperpolarizing GABAergic signaling (Sivakumaran et al. 2015; Otsu et al. 2020), the ability of this drug to generate epileptic-like activity in vitro remains unclear. VU0463271 failed to initiate epileptiform activity in naive slices (Sivakumaran et al. 2015; Kelley et al. 2016) despite being sufficient to alter it in tissue pre-exposed to a variety of convulsants (Hamidi and Avoli 2015; Sivakumaran et al. 2015; Kelley et al. 2016; Chen et al. 2020). Furthermore, the interpretation of these studies is also complicated because the effect of KCC2 blockade was not limited to the intrinsic subicular circuitry. Thus, this previous work did not mimic the subiculum-specific impairment of KCC2 function reported in epileptic tissue. Here, we have directly addressed this point by exposing isolated subicular slices of non-epileptic animals to VU0463271 and by studying the cellular mechanisms underlying the ensuing epileptiform activity.

In particular, we have focused on the role of a major class of GABAergic interneurons: parvalbumin-expressing fast spiking cells (PVs; Hu et al. 2014). We have quantified their morphofunctional characteristics, postsynaptic domain-specific connectivity, and network roles during VU0463271-induced epileptiform discharges. We have found that pharmacological inhibition of KCC2 profoundly alters synaptic signaling of subicular PVs, which becomes sufficient to initiate network bursts with physiological and pharmacological properties resembling interictal events in human epileptic tissue. In contrast, reducing their firing decreases the frequency of the observed interictal-like discharges.

Importantly, as we used slices from non-epileptic animals, no functional or structural epilepsy-related changes other than reduction of KCC2 function and PV firing were required to produce interictal-like events. Thus, our results from a reduced in-vitro model support the idea that the deafferentation-induced KCC2 downregulation described in tissue from patients suffering from TLE (Cohen et al. 2002; Huberfeld et al. 2007, 2015) can have causal epileptogenic effects, which are mediated by altered PV signaling at critical perisomatic postsynaptic target domains.

Materials and Methods

Ethical Approval

All experimental procedures used in this study were approved by the Institutional Animal Care and Use Committee of Northwestern University and complied with animal guidelines provided by the National Institutes of Health.

Animals

Mice were obtained from The Jackson Laboratory. In particular, the following strains (from Jackson Laboratory) were used: wild type C57BL/6J (stock #000664), PV-cre (stock #008069), Thy1-cre (stock #006143), tdTomato^flox (stock #007909), ChR2(H134R)-EYFP (Stock #012569), ArchT-EGFP^flox (stock #021188), and GCaMP6s^flox (stock # 024106). Appropriate crossing of the aforementioned Cre and floxed stop lines resulted in animals conditionally expressing tdTomato (PV-tdTomato), ChR2(H134R)-EYFP (PV-ChR2-EYFP), ArchT-EGFP (PV-ArchT-EGFP), and GCaMP6s (Thy1-GCaMP6s). Animals were housed on a 14/10 light/dark schedule with ad libitum access to food and water. Both male and female mice were used for the experiments.

Histology

PV-tdTomato (age P15-P60), PV-ChR2-EYFP (age P30), and PV-ArchT-EGFP (age P30) mice were anesthetized by intraperitoneal injection of Euthasol (calculated to yield a dose of pentobarbital of 300 mg/kg of bodyweight), and perfused with 0.9% NaCl saline followed by 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (PB), pH 7.4. After perfusion, brains were extracted from the skull and transferred in fixative solution at 4 °C for at least 24 h. Transverse hippocampal sections were cut serially at 50 μm thickness on a Leica VT 1000 vibratome and collected free-floating in 0.01 M phosphate buffered saline (PBS).

Immunohistochemistry

Slices were preincubated free-floating in a blocking solution containing 5% normal goat serum (NGS), 1% bovine serum albumin (BSA) and 0.2% Triton X-100 in PBS for 1 h at room-temperature (RT). Subsequently, they were incubated free-floating in the same solution containing the primary antibodies (all 1:500, see below) at 4 °C overnight. The primary antibodies used for the experiments of the main text were: mouse anti-NeuN (Abcam catalog #104224), rabbit anti-KCC2 (Thermo Fisher Scientific Cat# PA5-78544, RRID: AB_2735976), and mouse anti-Parvalbumin (Swant Cat# 235, RRID:AB_10000343). See Supplementary Methods for a list of the primary antibodies used for the experiments of Supplementary Figure 1. Next, slices were washed 3 × 15 min with fresh PBS and then incubated free-floating in a solution of 5% NGS, 1% BSA in PBS, containing secondary antibodies (all 1:500) at RT for 1 h. The secondary antibodies used for the experiments in the main text were: Alexa Fluor 488 goat anti-mouse IgG (Thermo Fisher Scientific Cat# A-11001, RRID:AB_2534069), Alexa Fluor 488 goat anti-rabbit IgG (Thermo Fisher Scientific Cat# A-11034, RRID:AB_2576217), and Alexa Fluor 568 goat anti-mouse IgG (Thermo Fisher Scientific Cat# A-11004, RRID:AB_2534072). See Supplementary Methods for a list of the secondary antibodies used for the experiments of Supplementary Figure 1. Finally, slices were washed 3 × 15 min with fresh PBS with the second washing step containing DAPI (1:100 000, LifeTechnologies, #62249) and then mounted and coverslipped using Mowiol mounting medium.

Confocal Microscopy

Confocal microscopy image stacks were captured using a Nikon A1R confocal microscope. Multichannel fluorescence images were saved individually for analysis and merged together for co-localization studies using the ImageJ software suite (Schneider et al. 2012).

Recovery of Biocytin-Filled Cells and Reconstructions

Biocytin-filled neurons were fixed in 4% PFA in 0.1 M PB at 4 °C for at least 24 h. Endogenous peroxidase activity was quenched with a 3% H₂O₂ solution for 15 min. Sections were incubated overnight at 4 °C in avidin-biotinylated-HRP complex (Vectastain ABC Elite kit) with 0.1% Triton X-100 in PB, followed by a peroxidase reaction with DAB tetrahydrochloride as a chromogen. Cells were revealed by adding 0.025% H₂O₂, and the reaction was stopped when dendritic and axonal processes were clearly visible under light microscopy examination. After several washing steps in 0.1M PB, slices were postfixed with 0.1% OsO₄ in PB (1–2 min), and then mounted on slides with Mowiol (Hoechst AG). Cells were reconstructed using a NEUROLUCIDA-based station and software.

Dendritic and Axonal Density Plots

Dendritic and axonal density-plots were generated using previously described methods (Anstötz et al. 2016).

Acute Slice Preparation

Mice (P21–P30) were deeply anesthetized with isoflurane in an induction chamber containing a separating platform to avoid direct contact between the animals and the anesthetic. After decapitation, the brain was carefully removed and glued to a specimen block in a chamber filled with cooled artificial cerebrospinal fluid (ACSF) with the following composition (in mM): 130 NaCl, 24 NaHCO₃, 3.5 KCl, 1.25 NaH₂PO4, 1 CaCl₂, 2 MgSO₄, and 10 glucose, saturated with 95% O₂, 5% CO₂ at pH = 7.4. Transverse sections (400 μm) of an entire brain hemisphere containing the hippocampal formation were cut using a vibrating microtome (Leica VT 1200 S, Leica Biosystems). Slices recovered at 30–32 °C for at least 30 min and were then stored at room temperature until use. Subicular mini slices were prepared from regular slices by severing the connections between the subiculum and CA1 hippocampal subfield/entorhinal cortex (Fiske et al. 2020). Cuts were performed at locations with clear scatter of PC bodies and with unambiguous widening of the pyramidal cell layer (pcl).

Electrophysiological Recordings and Analysis

General

Slices were transferred to a recording chamber positioned under a direct microscope (Scientifica) equipped with oblique illumination optics (Olympus) and an infrared camera system (VX-45, TILL Photonics). Cells were visualized using a 60× infrared water-immersion objective. Slices/mini slices were superfused with preheated ACSF (31–33 °C, TC-324B, Warner Instruments) with the following composition (in mM): 130 NaCl, 24 NaHCO₃, 3.5 KCl, 1.25 NaH₂PO₄, 2 CaCl₂, 1 MgCl₂, 10 glucose, saturated with 95% O₂, 5% CO₂ at pH 7.4.

Pipettes were pulled from thin borosilicate capillaries (Prism FLG15, Dagan Corporation) with a resistance of 3–5 MΩ when filled with an internal solution containing the following for current clamp (in mM): 115 K-methylsulfate, 10 KCl, 5 creatine phosphate, 5000 units creatine kinase, 4 ATP-Mg₂, 0.3 GTP-Na₂, 16 KHCO₃, 0.25% biocytin equilibrated with 95% O₂–5% CO₂ at pH 7.3. Voltage clamp recordings were performed with an internal solution containing the following (in mM): 125 CH₃CsO₃S, 16 KHCO₃, 4 ATP-Mg₂, 0.3 GTP-Na₂, 10 QX-314, 0.25% biocytin equilibrated with 95% O₂–5% CO₂ at pH 7.3 Recordings were performed using a Multiclamp 700 amplifier (Molecular Devices). Series resistances were balanced via a bridge circuit in current-clamp configuration and measured in voltage-clamp recordings (17.4 ± 8.5 MΩ, n = 18). Signals were filtered at 3 KHz and digitized at a minimum of 20 KHz using a Digidata 1550A and the Clampex 10 program suite (Molecular Devices) was used for the analysis of membrane properties and action potentials.

Paired Recordings

Simultaneous double and triple whole-cell recordings were performed with the postsynaptic PC held in voltage clamp at +10 mV and the presynaptic PV in current clamp. The distance between the soma of the 2 neurons was 70 ± 23 μm (n = 17 pairs). The interneuron was stimulated every 10 s with a 1 ms suprathreshold current injection to elicit a single action potential. Unitary inhibitory postsynaptic currents (uIPSCs) were identified by the presence of short-latency responses to the presynaptic action potential. For each connection identified (n = 28), several uIPSCs (n = 10) were collected and averaged, and the resulting average was measured and analyzed. Amplitude, half-width, and 20–80% risetime were estimated using the Clampfit 10 program (Molecular Devices). Latency was measured manually from the peak of the action potential in the interneuron to the onset of the uIPSC in the PC. Putative failures were identified visually.

Optogenetic Stimulation

Control of cellular population expressing opsins (either ChR2[H134R]-EYFP or ArchT-EGFP) was achieved by exposing the slice to flashes of blue light of appropriate duration (either 1 ms or 15 s) generated by a collimated LED (460 nm, Prizmatix) attached to the epifluorescence port of the direct microscope. Light flashes were directed to the microscopic field via a mirror coupled to a 60× objective (1.0 numerical aperture).

Analysis of VU0463271-Dependent Epileptiform Activity

These experiments were performed on surgically isolated subicular slices. For time course experiments, large amplitude events were selectively detected using the Strathclyde Electrophysiology Software (courtesy of Dr John Dempster, University of Strathclyde, Glasgow, United Kingdom), setting a threshold of 20X the standard deviation of a 1 min control period in the absence of the drug. All other VU0463271-related experiments (without time course) were performed on slices showing spontaneous activity after preincubation in 10 μM VU0463271 at 30 °C for at least 30 min. Electrical activity recorded in cell-attached or whole-cell configuration was measured using the Clampex 10 program suite (Molecular Devices).

Linear and Nonlinear Fits

Linear and allometric fits were performed by Origin 2020b using the following equations: y = a + b^*x (linear) and y = a^*x^b (allometric).

Calcium Imaging and Analysis

Images were acquired with a Scientifica Slice Scope equipped with an Olympus XLUMPlanF L N 20x/NA 1.0 lens. A collimated LED (460 nm, Prizmatix) was used as an excitation source coupled to a GFP filter set (excitation band: 469 ± 17.5 nm; emission band: 525 ± 19.5 nm; dichroic reflection/transmission band 452–490 nm/505–800 nm; Thor labs). LED-generated flashes (50 ms duration) at 10 Hz were synchronized with the acquisition of emitted fluorescence by a Zyla sCMOS 4.2 camera (Andor). Data files were imported and analyzed using the μManager (Edelstein et al. 2014) and imageJ software packages. First, background and bleaching were corrected by subtracting a duplicate with a Gaussian blur filter from the recording. Then, individual cells as regions of interests were outlined and their intensity over time measured. The intensity profile of each cell in a slice was normalized from 0 (minimal intensity) to 1 (maximal intensity). Peaks were detected and analyzed using the Peak Analyzer function of Origin 2020b. Cross-correlations were performed by comparing a cell’s intensity profile with the average intensity profile of all the cells of interest in the slice.

Morphological parameters (maximum cells distance, nearest neighbor distance, and convex hull area) were analyzed using a custom Visual Basic (Microsoft) script. Probability maps of convex hull contours were obtained by aligning them to their center of gravity and rotation. A 25 μm by 25 μm grid was used to measure different levels of enclosing probabilities and was displayed as a contour-plot.

Experimental Design and Statistical Analysis

Data are presented in the text as mean ± standard deviation (SD). Box plots in the illustrations show the median as middle dash, average as empty circle, lower and upper quartiles as upper and lower box borders, and minimum and maximum values as whiskers. Bands in time course plots figures are mean ± standard error (SE). Statistical comparisons used 2-sided nonparametric tests (one sample [null hypothesis median = 0]: Wilcoxon signed rank, 2 independent samples: Mann–Whitney; 2-paired samples: Wilcoxon signed rank; >2 groups paired samples: one-way repeated measures analysis of variance [ANOVA] on ranks with Bonferroni post-hoc adjustment). Significance was accepted at the level of P < 0.05. Probability values in the text are rounded to the third decimal place. Analysis was performed using the following software: Origin 2020b (OriginLabs), MS Excel (Microsoft), pClamp (Molecular Devices LLC), and the Strathclyde Electrophysiology Software (courtesy of Dr John Dempster, University of Strathclyde, United Kingdom).

Drugs

VU0463271, gabazine, and NBQX were purchased from Tocris. VU0463271 was prepared as a 10 mM stock solution in DMSO and used in experiments at 10 μM. Gabazine was prepared as a 25 mM stock solution in water and used in experiments at 12.5 μm. NBQX was prepared as a 100 mM stock solution in DMSO and used in experiments at 20 μM. Aliquots were stored at −20 °C prior to use. DMSO (0.1%) was also included in control solutions when experiments involved VU0463271.

Results

Blockade of KCC2 expressed by subicular neurons (Fig. 1A) affects intracellular chloride concentrations and the reversal potential of GABA_A receptor-mediated currents (Otsu et al. 2020, Fig. 1B). In order to mimic the effect of a subiculum-specific reduction in function of the KCC2 transporter, we applied the selective inhibitor VU0463271 (10 μM) to slices with the subiculum surgically isolated from its hippocampal and cortical afferents (Fiske et al. 2020). We recorded simultaneous triplets (n = 11) or pairs (n = 1) of subicular pyramidal cells (PCs; total = 35 neurons) under whole-cell conditions. As shown in Fig. 1C and D, drug-treated slices slowly developed spontaneous network-driven, large-amplitude (threshold >20× SD of baseline) synaptic events, which were not observed in tissue not exposed to the blocker (n = 4 pairs and 8 triple recordings for a total of 32 cells).

Figure 1 .

Effects of pharmacological block of KCC2 on population network activity in isolated mouse subicular tissue. (A) Confocal image showing the perisomatic KCC2-immunoreactivity (green) of a subicular PC. A higher magnification of the area delimited by the box in the left panel is shown to the right. (B) Simplified cartoons illustrating the electrophysiological experiment and its proposed mechanistic explanation (i.e., increase of intracellular chloride concentrations following KCC2 block and change in GABA_A receptor mediated signaling from hyper- to depolarizing). The presynaptic GABAergic terminal of an interneuron (IN) and its postsynaptic membrane containing a GABA_A receptor (GABA_A) and the K-Cl cotransporter KCC2 (KCC2) are shown. (C) Selective time-dependent development of large-amplitude (>20× SD, see Materials and Methods for details) compound excitatory postsynaptic potentials in pyramidal cells (PCs) of slices exposed to VU0463271 (10 μM, top 3 traces shown from left to right at different time scales) compared with slices exposed to vehicle (bottom 3 traces shown from left to right at different time scales). PC 1, PC 2, and PC 3 indicate the activity from different PCs of the triple recordings. One and 2 indicate the expanded sections of the leftmost traces. Notice the presence of large events only in the presence of the drug. (D1) Summary time course plots comparing the frequency of large (>20× baseline SD) events in the presence (blue circles and light blue band, mean ± SE) versus absence of the blocker (black circles, and light gray band, mean ± SE). VU0463271 was applied after 1 min in drug-free solution and maintained for the rest of the experiment. (D2) Box plots comparing the frequency of large events occurring during the same time windows (t1, t2: black bars in C) in slices exposed (VU) or not exposed to the drug (ctrl). n.s. = not significant, ^***P < 0.001. Individual data points are shown at the right of the box charts. All experiments were performed on slices from wild type C57BL/6J mice.

The frequency of the events detected during the first 5 min following VU0463271 application (0.0 ± 0.1 events/min) was not different from that measured during the same time window in control slices (0.0 ± 0.1 events/min, P = 0.90, Mann–Whitney test, 2-sided). In contrast, when we compared values at a later time (5 min period following 30 min of drug application) we found a significant difference in frequency of events (3.2 ± 3.6 events/min in treated tissue versus 0.1 ± 0.1 events/min in control slices, P < 0.001, Mann–Whitney test, 2-sided).

Although this indicates that inhibiting KCC2 activity in the isolated subiculum is sufficient to trigger hyper-synchronization, most of these large-amplitude events remained subthreshold and did not trigger action potentials. We interpreted this apparently puzzling observation as the consequence of our recording conditions (whole-cell configuration), based on low-chloride containing pipettes (10 mM, see Material and Methods for details). We reasoned that the large volume of intracellular solution contained in the patch electrode would dialyze the cytoplasm of the recorded neuron. Therefore, despite the pharmacological blockade of KCC2, dialysis would set a low intracellular chloride concentration. However, this effect would not impact the nonrecorded neurons of the slice, in which the endogenous intracellular chloride concentration would be unperturbed and thus readily affected by KCC2 inhibition. Therefore, we predicted that application of VU0463271 would generate suprathreshold events in neurons studied under cell-attached configuration, where dialysis of the cytoplasm does not occur (Fig. 2A). As expected, when we recorded from n = 28 neurons (n = 4 triples, n = 6 pairs, and n = 4 singles) in slices preincubated with VU0463271 for 40 min, we observed synchronous suprathreshold epileptiform activity (i.e., with action currents, Fig. 2B), reminiscent of the type originally described in human epileptic subicular slices (Cohen et al. 2002). The number of spikes/event was 2.5 ± 1.0, with an average interspike interval within every epileptiform event of 7.7 ± 2.0 ms, an event duration of 14.3 ± 8.2 ms, and an interevent interval of 2.9 ± 1.8 s. In a subset of recordings (n = 20 cells), we were able to compare the occurrence of suprathreshold events in cell-attached 30 ± 13 events/min) versus whole-cell configuration (0 ± 1 events/min) in the same recorded neurons (P < 0.001, Wilcoxon signed rank test, 2-sided) and thus confirm our hypothesis that recording in whole-cell configuration caused the loss of firing. When recorded under whole-cell conditions from slices similarly pre-exposed to VU0463271 (Fig. 2C), the peak of the collected events was 7.8 ± 3.2 mV, their risetime was 10.3 ± 3.1 ms, their half width was 41.3 ± 8.1 ms, and their interevent interval was 2.1 ± 1.4 s (n = 4 triple, n = 4 double, and n = 6 single recordings for a total of n = 26 neurons). These properties (large amplitude and slow kinetics) further corroborate their network-driven origin.

Figure 2 .

Preserving versus manipulating the physiological dynamics of intracellular chloride concentration determines whether exposure of slices to VU0463271 triggers supra- or subthreshold activity in the recorded neurons. (A) Schematic cartoons highlighting the differences imposed by cell-attached (left) versus whole-cell recording configuration (right). Under cell-attached configuration, pharmacological inhibition (KCC2 block) of the KCC2 membrane transporter affects intracellular chloride dynamics leading to an increased internal concentration [Cl⁻] in every neuron of the network (including the cells monitored by the electrodes, a triple recording in the example). Physiological dynamics are indicated by the green/yellow color of the cytoplasm. The magnified view of the experimental configuration highlights that, following the elevation of the intracellular [Cl⁻], the net current via GABA_A receptors becomes inward (i.e., net movement of chloride ions to the outside, arrow) in resting neurons. In contrast, during whole-cell recordings, the large volume of solution contained in the recording pipette clamps the intracellular [Cl⁻] to the same level of the pipette. Under these conditions, the impact of KCC2 inhibition on the intracellular chloride dynamics does not occur in the recorded neurons and the net current via GABA_A receptors remains outward at rest (i.e., net movement of chloride ions to the outside, arrow). Thus, currents generated by GABA_A receptors during network events have opposite directions in the recorded neurons compared with the remaining cells of the slice. (B1) Example of the spontaneous activity observed under cell-attached recording configuration in a triple recording from a slice pre-exposed (at least 30 min) and still in the presence of VU0463271 (10 μM). PC 1, PC 2, and PC 3 indicate the traces from the different pyramidal neurons. Notice the presence of time locked rhythmical suprathreshold burst events in all the recorded cells (shown at higher temporal magnification in the right inset). (B2) Quantification of VU0463271-induced activity. Left, schematic illustrating the measured parameters. Right, summary box plots and individual data points for (left to right) the number of spikes/event, event average interspike interval, event length and interevent interval. (C1) As in (B1), but in whole-cell configuration. Notice the presence of sub-threshold large amplitude events (shown at higher temporal magnification in the right inset). Notice their complex shape indicative of compound synaptic transmission. (C2) Left, measured parameters. Right, summary graphs for peak amplitude, risetime, half width and interevent interval. All experiments were performed on slices from wild type C57BL/6J mice.

We confirmed and expanded this finding by taking advantage of calcium imaging, which allows the simultaneous monitoring of activity in a larger number of cells compared with patch-clamp electrophysiology and does not interfere with intracellular chloride concentrations. Figure 3A shows an example of the local activity of a cluster of 11 PCs in an isolated subicular slice prepared from a Thy1-GCaMP6s mouse exposed to 10 μM VU0463271. The peak of the mean cross-correlation of each individual neuron with the average of all the signals was 0.988 ± 0.007 (lag 0.0 ± 0.0 s, n = 11 cells), indicating their tight synchrony. Figure 3B shows values obtained from the analysis of several slices (n = 16). The interevent interval was 7.7 ± 2.8 s and the overall peak cross-correlation and lags were 0.960 ± 0.054 and 0.0 ± 0.0 s, respectively. We defined the spatial properties of these clusters of coactive subicular PCs (Fig. 3C) by calculating their projected convex hull areas (62 567 ± 32 114 μm²), nearest neighbor (59 ± 17 μm), and maximal cell distances (454 ± 117 μm).

Figure 3 .

Synchronous calcium transients in clusters of neurons from isolated subicular slices exposed to VU0463271 (10 μM). (A1) Left, pseudo-colored images collected before (t1, left), at the peak (t2, middle) and after return to baseline (t3, right) of the calcium transient. Notice the simultaneous activation of several cells. Insets show 2 individual cells (cell 1 and cell 2) at increased magnification. (A2) Left, spontaneous rhythmic calcium signal shown as mean ± SE of all cells in the cluster of interest. The transient identified by the red bar (between t1 and t2) is shown in more detail to the right for all cells of the measured cluster (cell1 to cell11). Signals from cell 1 and cell 2 (insets of panel A1) are shown in red and blue, respectively. Signals from the remaining cells are in gray. (A3) Mean ± SE of all the cross-correlation of each individual cell in the cluster with the average signal. The inset shows an expanded view at shorter time lags. Notice the high synchronicity of the events. (B1) Properties of the calcium transients. Left, schematic of the measured indexes. Right, summary plots for interevent intervals and half widths. (B2) Summary graph (mean ± SE) of all the averaged cross-correlations for all the imaged slices. (C1) Left, cartoon showing the parameters measured in the cell clusters. Right, population graphs and individual data points for all the quantified indexes in the imaged clusters (area, maximum cell distance and nearest neighbor distance). (C2) Density probability of the convex hull area of the imaged cluster (green contour plots). The green areas and their borders indicate the probability of including clusters (see Materials and Methods for details). Purple circles represent the positions of all the imaged cells. All experiments were performed on slices from Thy1-GCaMP6s mice.

Thus, the functional inhibition of KCC2 activity in isolated subicular networks is sufficient to generate synchronized excitatory events resembling interictal activity. When studied in subicular human tissue from epileptic patients, interictal-like events have a specific pharmacological profile requiring both GABAergic and glutamatergic transmission (Cohen et al. 2002). For comparison, we decided to explore the nature of the synaptic activity underlying VU0463271-induced events revealed by our experimental conditions. Blocking either GABA_A-type (Fig. 4A and B) or AMPA-type receptors (Fig. 4C and D) almost completely suppressed their occurrence. In fact, the addition of the GABA_A receptor antagonist gabazine reduced the frequency of large spontaneous events from 2.8 ± 4.3 events/min to 0.3 ± 0.4 events/min (n = 15 triples and n = 1 pair for a total of 47 neurons, P < 0.001, Wilcoxon signed rank test, 2-sided), and the inclusion of the AMPA-type glutamate receptor antagonist NBQX decreased it from 2.7 ± 2.8 events/min to 0.1 ± 0.2 events/min (n = 3 triples and n = 4 pairs for a total of 17 cells, P = 0.001, Wilcoxon signed rank test, 2-sided). Interestingly, in the presence of VU0463271 and gabazine, the very rare residual events became suprathreshold paroxysmal depolarizing shifts, similar to what has been described in isolated subicular preparations exposed to GABA_A and GABA_B receptor antagonists in the absence of VU0463271 (Fiske et al. 2020). In contrast, this transformation was not observed in the presence of NBQX.

Figure 4 .

VU0463271-induced network events require both GABAergic and glutamatergic synaptic transmission. (A) Traces from a triple recording under whole-cell recording conditions in the presence of VU0463271 (10 μM, left) and after the addition of gabazine (12.5 μM). PC 1, PC 2, and PC 3 indicate the different PCs of the triple. Events indicated by the black arrowhead are shown at magnified temporal scale in the insets. Notice that gabazine dramatically reduced event frequency and transformed the few residual events from sub-threshold EPSPs into bursts. (B1) Summary plot of the time course of several experiments (mean ± SE). Notice the slow development of large events similar to Figure 5 and the dramatic decrease in frequency following GABA_A receptor inhibition. (B2) Summary box plots and individual data points comparing the frequency of the events in VU0463271 (VU) and after the addition of gabazine (VU + gbz). ^***P < 0.001. (C), (D1), and (D2) Same organization as in (A), (B1), and (B2), but for experiments testing the effect of blocking glutamatergic synaptic transmission via NBQX (20 μM). Notice the almost complete disappearance of the events. All experiments were performed on slices from wild type C57BL/6J mice.

Taken together, these results suggested that the application of VU0463271 to naïve isolated subicular circuits in vitro is capable of producing spontaneous synchronous events with physiological and pharmacological hallmarks resembling the interictal-like events reported in human epileptic tissue studied in vitro. Moreover, the fact that we were able to observe epileptiform activity following exposure of non-epileptic slices to VU0463271 suggests the possibility of taking advantage of this model in vitro to test the role played by specific neuronal types.

In particular, we decided to investigate the involvement of PVs because of their special role in the control of cellular excitability and synchronization (Hu et al. 2014). In order to manipulate PVs, we took advantage of the PV-cre mouse line established by Hippenmeyer et al. (2005) to label this class of interneurons with tdTomato (tdT) and/or control them via excitatory (ChR2[H134R]-EYFP) and inhibitory (ArchT) opsins in knockin mice (see Materials and Methods section for their detailed descriptions). However, as off-target recombination has been reported to occur to various degrees in different brain regions (in several commonly used cre-lines: Hu et al. 2013; Müller-Komorowska et al. 2020), we directly validated our experimental conditions and confirmed the specificity and selectivity of our approach for subicular networks. (see Supplementary Fig. 1). The morphology, firing pattern and basic membrane properties of tdT-expressing cells (tdTs) of PV-tdTomato mice was examined by performing whole-cell patch-clamp recordings with biocytin-filled pipettes followed by the structural analysis of the recovered neurons. As shown in Fig. 5A–C, the vast majority of the recorded interneurons displayed a fast-spiking phenotype with typical membrane characteristics (McCormick et al. 1985; Kawaguchi et al. 1987) such as low input resistance, short-duration action potentials, large amplitude after-hyperpolarizations (AHPs) and high frequency firing (>50 Hz) with little to no accommodation in response to depolarizing current steps (Ascoli et al. 2008). When quantitatively analyzed, the estimated input resistance was 137 ± 50 MΩ (n = 65 cells), the average membrane resting potential was −67 ± 3 mV (n = 62 cells), and the threshold current step required to observe the first action potential was 185 ± 63 pA (n = 47 cells). The amplitude of the first spike generated at current threshold was 64 ± 7 mV (n = 47 cells) and its half width was 0.34 ± 0.04 ms (n = 47 cells). The most negative peak of the AHP and its half width were − 23 ± 4 mV and 9.4 ± 4.3 ms (n = 47 cells), respectively. When we built current/frequency plots (n = 47 cells), saturation was not reached at higher than 200 Hz, and intermediate levels of current injection were often associated with a “stuttering” phenotype, as typically observed in fast-spiking interneurons (Ascoli et al. 2008). Post-hoc reconstructions (Fig. 5D and E) revealed a roughly multipolar, stellate-like dendritic arbor, quite distinct from the classically described basket and axoaxonic cells of the CA3–CA1 hippocampal region with their well-defined apical and basal dendrites (Freund and Buzsáki 1996, see also Supplementary Fig. 2). Total dendritic length was 2790 ± 1191 μm (n = 22 reconstructions) and total axonal length measured 10 039 ± 4234 μm (n = 14 reconstructions, Fig. 5F). On average, as illustrated in Figure 5G and H, the dendritic arborization showed a directional bias (62.1% vs., 37.9% in the deep vs., superficial half relative to the soma, n = 22 dendritic trees). This preference was opposite to what was found when the same analysis was applied to axons (28.6% vs., 71.4%, for deeper vs., superficial half, n = 14 axons). This structural organization indicates that PVs offer a larger available postsynaptic surface to terminals targeting their deeper dendrites, and that their synaptic output is then preferentially channeled toward more superficial areas.

Figure 5 .

Firing patterns, membrane properties and structural characteristics of subicular tdT⁺ interneurons of the PV-tdTomato mouse. (A1) and (A2) Typical action potential trains in in response to long (1 s) depolarizing current pulses. Notice the single action potential at threshold current levels and the membrane deflection following a hyperpolarizing current step (top), the stuttering phenotype at intermediate levels (middle) and the rapid firing triggered by larger current steps (bottom). The right insets show the initial part of the traces at a magnified temporal scale. Notice the minimal spike frequency adaptation. (B) Current/frequency plot. The blue and red curves indicate the maximum (blue symbols) and average (red symbols) frequency measured within the train. Bands are ±SE. Notice that in both cases frequencies > 200 Hz can be reached. (C) Summary plots of measured basic membrane properties. Left to right: membrane input resistance, resting potential, threshold current, action potential (AP) peak amplitude and half width and AHP peak and half width. Box plots and individual data points are illustrated on the left and right side of the graphs, respectively. (D) Micrograph of a recovered biocytin-filled tdT⁺ cell. Notice the high density of synaptic terminals. Dotted lines indicate the border between the pcl and the molecular layer (ml). (E) Reconstructed tdT⁺ neurons from the subiculum and its adjacent regions, that is, CA1 hippocampus and pre subiculum. Dendrites in red and axons in blue. Notice the dense axonal arborization in all cases and the larger tangential spread in the case of the CA1 cell. Notice the multipolar dendritic arborization of the subicular cell (see also Supplementary Fig. 2) (F) Overlap of all the reconstructed interneurons aligned on their soma (green circle) and deep/superficial position within the subiculum. (G) and (H) Summary plots of density probability for dendrites and axons, respectively. Notice the opposite orientation biases with respect to the deeper versus most superficial parts of the region. All experiments were performed on slices from PV-tdTomato mice.

Given their impact on network function, we paid particular attention to evaluating the synaptic output of subicular PVs onto postsynaptic PCs by performing simultaneous paired recordings (Fig. 6A and B). This is important to establish the strength of their functional connections and their specific postsynaptic domain targeting. The overall probability of finding a functionally detectable unitary connection was approximately 50%. As shown in Figure 6C, uIPSCs recorded in voltage-clamp configuration at a holding potential of +10 mV (n = 28 pairs) had a peak amplitude of 252 ± 303 pA, a short latency (0.5 ± 0.3 ms), and kinetics described by risetime and half width of 1.1 ± 0.6 ms and 12.0 ± 2.8, respectively. Failures were rare (4 ± 7%) demonstrating reliability of transmission. Interestingly (Fig. 6D), latency was correlated with uIPSC risetime (R² = 0.56, P < 0.001, and n = 28 pairs), indicating that the faster events were occurring on a target domain close to the presynaptic cell of origin. Furthermore, the larger amplitude events were also associated with faster risetimes, and originated a typical L-shaped curve, similar to what was reported for uIPSCs in the CA1 hippocampal region (Maccaferri et al. 2000). These results suggest that the larger events originate close to the postsynaptic somatic recording site. We decided to investigate this issue by identifying points of close apposition between the pre- and postsynaptic neurons (Fig. 7A and B), as previously reported for excitatory connections between subicular PCs (Fiske et al. 2020). In comparison to that study (Fig. 7C), putative contact sites from PVs onto PCs (total n = 64 from n = 11 reconstructed pairs) appeared mostly restricted to the perisomatic compartment with a substantial fraction (~38%) located within 50 μm from the soma, ~ 53% between 50 and 150 μm from it, and a minimal fraction further away. Furthermore, larger events were associated with the presence of at least a single contact site close to the soma. Plotting the peak uIPSC current versus the minimal distance of the putative contact site(s) from the postsynaptic soma resulted in an L-shaped plot (Fig. 7D) similar to that obtained in Figure 6D, and consistent with a dendritic filtering effect that is more prominent at more distal dendritic sites (Rall 1977). As expected, the distance from the soma of the putative contact site(s) was correlated to the risetime kinetics of the uIPSC (Fig. 7E, R² = 0.762, P = 0.017, n = 11 pairs). The presynaptic interneurons of the connected pairs were located deeper within the stratum pyramidale relative to the postsynaptic PC (Fig. 7F, n = 17 pairs). This is in agreement with the expectation suggested by the axonal density distribution shown in Figure 5H. Thus, our data indicate that subicular PVs predominantly innervate the perisomatic area of postsynaptic PCs. As this area is in close proximity to the action potential initiation site of subicular PCs (Colbert and Johnston 1996), PVs would be predicted to influence action potential generation and synchronization in the target population (Cobb et al. 1995; Miles et al. 1996). Furthermore, the average projected convex hull area of the axonal arborization of a single subicular PV (61 529 ± 24 000 μm², respectively, n = 14 reconstructed axons) was very similar to what was calculated for the clusters observed in our calcium-imaging experiments. Therefore, assuming spike-to-spike transmission between pairs of cells, it is possible that a single PV could drive population activity in clusters. More realistically, we think that, because of the variability in the amplitude of uIPSCs observed in pairs (see Fig. 6C), convergent activation of several PVs may be required.

Figure 6 .

Putative contact sites and functional properties of uIPSCs generated by tdT⁺ cells onto pyramidal neurons. (A) and (B) Examples of reconstructed pairs between a tdT⁺ presynaptic interneuron (IN, dendrites, gray; axon, red) and a postsynaptic pyramidal cell (PC, dendrites, black, axon, blue). Points of close apposition between the presynaptic axon and postsynaptic dendrites are highlighted by green circles. Notice the presence of a putative somatic contact in (A). pml, polymorphic layer; pcl, pyramidal cell layer; ml, molecular layer; presub, presubiculum; CA1, CA1 subfield of the hippocampus. Insets show the uIPSCs recorded from the illustrated pairs. The presynaptic action potentials (top) and postsynaptic responses (bottom) are shown in gray and their averages in red, respectively. Notice that the uIPSC from pair (B) is of smaller amplitude and slower kinetics, and the lack of putative somatic contact sites in the reconstructed pair. (C) Summary plots of the uIPSC properties measured for all the recorded pairs. (D) Top, linear correlation between risetime and latency suggesting somatic targeting of spatially close postsynaptic target cells. Bottom, relationship between the uIPSC risetime and amplitude. Values included in the graph were measured from averaged uIPSCs. Notice the L-shape of the scatterplot. Fits are indicated by the continuous line (top, linear and bottom, allometric). All experiments were performed on slices from PV-tdTomato mice.

Figure 7 .

Structural analysis of synaptic connections between presynaptic tdT⁺ interneurons and postsynaptic pyramidal cells. (A) Summary plot containing all the reconstructed pyramidal neurons. The putative contact sites originated by tdT⁺ cells are indicated by the green circles. Red circles indicate, for comparison, the location (within the reference provided by the blue concentric dotted circles) of pyramidal to pyramidal cell connections previously reported (Fiske et al. 2020). (B) Example of a confocal image showing basket-like structures formed by tdT⁺ terminals (red) onto a subicular neuron (NeuN, green). The separate channels are shown to the left and the merged image to the right. (C) Comparison of the somatic distances of putative contact sites established by either presynaptic tdT⁺ cells (green bars) or pyramidal neurons (red bar, data from Fiske et al. 2020) with postsynaptic pyramidal cells. Dotted lines connect the centers of the bars. Notice the absence of putative synapses at the more distal dendritic sites in the case of connections originating from tdT⁺ cells. (D) Summary plot relating the shortest distance of putative contact sites to peak uIPSC amplitude. Notice the L-shape of the scatter plot, similar to what is shown in Figure 3D. The fit is indicated by the dotted line. (E) Relationship between uIPSC risetime and average distance of putative contact sites from the soma. Linear regression indicated by the dotted line. (F) Summary graph marking the position of the presynaptic tdT⁺ cell (dark small circles) relative to the postsynaptic target (large circle, dendrites are overlapped, gray). Notice the preferential deeper layer location of the presynaptic soma compared with its target. The inset shows a box chart of the intersomatic distances measured in n = 17 pairs. All experiments were performed on slices from PV-tdTomato mice.

If the epileptiform events observed in the presence of VU0463271 truly depended on the activity of a few PVs, we reasoned that even their limited optogenetic activation should successfully generate similar discharges, whereas their optogenetic inhibition should reduce spontaneous network synchronous activity.

We began by verifying the effectiveness of optogenetic stimulation on PVs in subicular slices not yet exposed to VU0463271. Under these experimental conditions (Fig. 8A and B), 1 ms-long optical stimulation reliably triggered individual action potentials in PVs as measured either in cell-attached or whole-cell configuration (n = 18 cells). As shown in Figure 8C, no effects were detected in PCs in cell-attached mode (n = 17 cells), whereas inhibitory postsynaptic potentials (0.4 ± 1.0 mV, n = 17 cells) could be measured in whole-cell configuration. When the same experiment was repeated on slices pre-exposed to VU0463271 (for at least 30 min) and showing the typical epileptiform activity, the same light stimulation was capable of evoking suprathreshold burst events in PCs, which were reminiscent of spontaneous discharges (Fig. 8D). A comparison of the properties of spontaneous versus light-triggered events (Fig. 8E) revealed differences in the number of action potential/event (2.5 ± 1.1 vs., 2.7 ± 1.2, n = 7 triples, n = 6 pairs and n = 4 single cells, for a total of 37 neurons, P = 0.011, Wilcoxon signed rank test, 2-sided) and in the event durations (14.4 ± 7.5 ms vs., 16.8 ± 7.8 ms, n = 5 triples, n = 8 pairs and 4 single cells, for a total of 35 neurons, P = 0.007, Wilcoxon signed rank test, 2-sided, 2 cells of 2 separate triple recordings produced only single-spike events, so durations could not be measured). No changes, however, were detected in the interspike intervals within the event (8.9 ± 5.2 ms vs., 9.4 ± 5.2 ms, n = 5 triples, n = 7 pairs and 4 single cells, for a total of 35 neurons, P = 0.075, Wilcoxon signed rank test, 2-sided). We interpret these differences as resulting from the artificial nature of channelrhodopsin stimulation. Optogenetic stimulation is likely to drive highly synchronous firing in several PVs, probably in excess of how many are needed to generate spontaneous events, hence the stronger events (longer and with more spikes). However, most importantly, the average spontaneous baseline interevent interval (spont, 2.6 ± 2.3 s, n = 7 triples, n = 6 pairs, and n = 4 single cells, for a total of 37 neurons) measured before delivering the light flash was indistinguishable from the specific interevent interval measured between the light-evoked event and the next spontaneous discharge in the same cells (post evoked, 2.6 ± 2.2 s, P = 1.0, repeated measures ANOVA on ranks with Bonferroni post-hoc adjustment). This argues for a “reset” of the network timing. In fact, light stimulation was timed randomly relative to the spontaneous events, as indicated by the difference between the baseline interevent interval (spont) compared with that calculated between the last spontaneous discharge and the event evoked by the flash (pre evoked, 1.5 ± 1.0 s, n = 7 triples, n = 6 pairs, and n = 4 single cells, for a total of 37 neurons, P < 0.001, repeated measures ANOVA on ranks with Bonferroni post-hoc adjustment). The “pre evoked” interevent interval was also different from that measured after the reset of network timing (post evoked, P < 0.001, n = 7 triples, n = 6 pairs and n = 4 single cells, for a total of 37 neurons, P < 0.001, repeated measures ANOVA on ranks with Bonferroni post-hoc adjustment). Based on this, the optogenetic-triggered reset of network timing is strong evidence for a mechanistic involvement of PVs in the generation of epileptiform events.

Figure 8 .

Optogenetic stimulation of PVs in the absence and in the presence of VU0463271 generates different responses. (A) Left, confocal images showing separate signals for ChR2-EYFP expression (top), PV immunoreactivity (middle) and DAPI counterstain (bottom) in a subicular interneuron. Right, same neuron with all signals merged. (B) Control experiment showing that optogenetic stimulation (blue arrowed waves and plus sign) triggers single spikes in fast-spiking PV subicular interneurons (PV) in slices not exposed to the drug. Top, typical fast-spiking firing pattern and membrane response to a hyperpolarizing current step of a cell selected for optogenetic stimulation. Bottom, a 1 ms blue light flash triggers a single action potential both in cell-attached (c-a, left) and whole-cell (w-c, right) configuration. (C) No suprathreshold activity is observed (same experimental conditions as in B) when recording from a pyramidal cell (PC). Top, Notice the different firing pattern and membrane properties compared with (B) and the lack of firing under either cell-attached or whole-cell recording conditions (bottom). (D) Optogenetic stimulation of PV subicular interneurons in slices exposed to VU0463271 (10 μM) and cell-attached recording from 3 pyramidal neurons (PC1, PC2, and PC3). The black and blue arrowheads indicate the spontaneous (s) and light-evoked (e) events, which are displayed to the right at higher temporal resolution to the right. Notice their similarity. (E) Summary plots comparing the properties of spontaneous versus light evoked (spikes/event, interspike interval and event length). Evoked events show minor but significant differences in spikes/event and event length. ^*P <<  0.05. The last plot compares the average baseline interevent interval before light stimulation (spont) with the intervals between the light-evoked event and the one immediately preceding (pre evoked) or following it (post evoked). Notice the difference between the spontaneous versus “pre evoked” intervals indicating the temporal randomness of stimulation and the return to similar values (spont vs., post evoked) showing the reset of the timing of the network event generator. ^***P << 0.001. All experiments were performed on slices from PV-ChR2-EYFP mice.

To reinforce this conclusion we tested the effects of optogenetic inhibition of PVs (Fig. 9A and B) on VU0463271-dependent activity. Before performing this experiment, we confirmed that, in the absence of the drug, activation of ArchT effectively hyperpolarized fast-spiking interneurons (−2.9 ± 1.5 mV, n = 11 cells) in slices prepared from PV-ArchT-EGFP mice (Fig. 9C and D). In the presence of VU0463271, when a 15 s illumination was delivered to the slice (Fig. 9E and F), the most prominent observation was a marked decrease in the frequency of spontaneously occurring discharges (from 2.3 ± 0.9 events/15 s to 1.0 ± 0.9 events/15 s, n = 3 triples, n = 6 pairs, and n = 4 single cells, for a total of 25 cells, P < 0.001, Wilcoxon signed rank test, 2-sided). In agreement with our previous interpretation, this result confirms that the activity of PVs is a crucial determinant of epileptiform activity triggered by functional inhibition of KCC2. Additional changes (Fig. 9F), albeit minor, were detected in the number of spikes per event (from 4.3 ± 2.3 to 4.4 ± 2.3, n = 3 triples, n = 5 pairs, and n = 2 single cells, for a total of 21 neurons, P < 0.028, Wilcoxon signed rank test, 2-sided, the overall number of recordings is lower than what was reported for event frequency because in n = 4 cells no events could be observed in the presence of the light) and in the event duration (from 30.1 ± 22.5 ms to 28.7 ± 20.8 ms, n = 3 triples, n = 5 pairs, and n = 2 single cells, for a total of 21 neurons, P < 0.037, Wilcoxon signed rank test, 2-sided). No differences were found in the average interspike interval within an event (from 8.6 ± 3.3 ms to 8.3 ± 3.1 ms, n = 3 triples, n = 5 pairs, and n = 2 single cells, for a total of 21 neurons, P < 0.082, Wilcoxon signed rank test, 2-sided). The changes measured in the properties of the events (number of spikes and event duration) were probably due to reduced inactivation of voltage-gated sodium/calcium channels because of both membrane hyperpolarization and the reduced occurrence of events during the light stimulation.

Figure 9 .

Optogenetic inhibition of PV subicular interneurons reduces VU0463271-dependent network bursts. (A) Left, confocal images of ArchT-EGFP fluorescence (top, EGFP), PV immunoreactivity (middle, PV) and DAPI nuclear counterstain (bottom, DAPI) of a subicular interneuron. Right, same cell with all signals superimposed. (B) Cartoon showing the stimulation experimental setup. A flash of light (blue arrowed wave and minus sign) inhibits PVs (PV) connected to pyramidal cells (PC). (C) Light activation of ArchT in PVs. Notice the typical electrophysiological signature of fast-spiking cells (top) and the hyperpolarization of the membrane potential caused by the light flash (blue bar). (D) Summary box plot for ArchT-induced hyperpolarization in n = 11 cells. (E) Example of a simultaneous cell-attached recording from 3 pyramidal neurons (PC 1, PC 2, and PC 3) in the presence of VU0463271 (10 μM). Three sweeps are superimposed, and events from a single sweep are highlighted in red. Notice the synchronous activity and the reduced occurrence of events in the presence of the light (blue bar). An expanded view of selected bursts (red arrowhead) is shown to the right (pre-light). The blue arrowheads indicate the region of the trace in the presence of the light expanded to the right (light). (F) Summary graphs and individual data points of the properties of the events in the absence versus presence of light stimulation. Left to right: event frequency (events/15 s), spikes/event, interspike interval and event length. Notice the large suppression in the occurrence of network bursts during the flash and the minor changes in spikes/event and event length. ^*P <<  0.05, ^***P <<  0.001. All experiments were performed on slices from PV-ArchT-EGFP mice.

Lastly, we decided to compare the cell-type specificity of the properties of epileptiform events in double recordings from PCs and PVs (identified by tdT expression in slices prepared from PV-tdTomato mice, Fig. 10A). As shown in Figure 10B, epileptiform discharges in PVs were associated with a larger number of action potentials (11.6 ± 8.3 spikes/event, in n = 12 PVs vs., 3.9 ± 1.0 spikes/event, in n = 12 PCs, P = 0.003, Wilcoxon signed rank test, 2-sided) and longer durations (70.5 ± 36.7 ms, in n = 12 PVs vs., 25.1 ± 10.2 ms, in n = 12 PCs, P = 0.006, Wilcoxon signed rank test, 2-sided), but with similar average interevent intervals (7.8 ± 2.9 ms, in n = 12 PVs vs., 8.9 ± 2.0 ms, in n = 12 PCs, P = 0.456, Wilcoxon signed rank test, 2-sided). Firing in PVs was observed to precede action potentials in the vast majority of the events recorded in the same pair (83 ± 21%, n = 12 double recordings) by 5.3 ± 6.4 ms (n = 12 double recordings, P = 0.007, one sample Wilcoxon signed rank test, 2-sided). These results revealed another striking resemblance to what was described in subicular slices from human epileptic tissue (Cohen et al. 2002), further highlighting the value of our model in capturing some essential alterations that occur in epileptic tissue.

Figure 10 .

Cell type-specific spike timing during epileptiform discharges in the presence of VU0463271. (A) Simultaneous recording under cell attached conditions from a PV (PV, top) and a pyramidal cell (PC, bottom). Notice the synchronous activity in the 2 cells. The event indicated by the black arrow is shown at increased temporal resolution in the right inset. Notice the difference in shape of the action potential waveforms in the PV and PC. The schematic cartoon indicates the source of the traces. (B) Summary boxplots of the properties of the events in the 2 distinct cell types. Notice in the 3 leftmost charts the presence of a larger number of spikes/event, the similar average interspike interval and the longer duration in PVs compared with PCs. Two rightmost summary plots show that PVs fire before PCs in the vast majority of events (light red area: PV leading, light green area: PC leading), and the overall quantification of the delay between the first spike in PVs versus PCs. Positive and negative values indicate earlier occurrence of the first spike in PVs versus PCs, respectively. ^**P << 0.01. All experiments were performed on slices from PV-tdTomato mice.

In conclusion, our data provide compelling evidence that excitatory signaling from PVs is mechanistically involved in subicular epileptiform activity induced by KCC2 inhibition and may act as a cellular trigger for epileptiform activity.

Discussion

Our work is a novel and original attempt to model and study a proposed critical epileptogenic step of TLE in a reduced preparation in vitro, and provides compelling mechanistic evidence for the involvement of PVs in the generation of epileptiform activity similar to what has been described in human epileptic tissue.

A “Model of Subicular-Specific Epileptogenesis” In Vitro?

Epileptogenesis refers to the process that converts the brain from a healthy state to a chronic condition suffering from spontaneous seizures, that is, epilepsy (Pitkänen et al. 2015). This, of course, is not exactly what occurs in a reduced preparation in vitro. Hence, by “model of epileptogenesis in vitro” we refer to a reduced experimental setup that can generate epileptiform activity when challenged by what are proposed to be key epileptogenic steps in in-vivo studies. In our case, this is the decreased expression of KCC2 specifically in the deafferented subiculum (Huberfeld et al. 2007). The strength and usefulness of our simplified model rests on its ability to disentangle causal factors from mechanistically unrelated epiphenomena. This goal may be difficult to achieve when working with tissue obtained from epileptic animals or human patients. For example, mossy fiber sprouting in the dentate gyrus, one of the most compelling structural alterations long assumed to be epileptogenic in TLE, may actually turn out to be just an epiphenomenon (Heng et al. 2013; Buckmaster 2014).

An important comparison between our model and human tissue is that altered chloride homeostasis in human epileptic tissue seems to occur only in fraction of subicular PCs (Huberfeld et al. 2007). Interestingly, bath application of VU0463271 in hippocampal slices was reported to generate GABA_A receptor-mediated excitation in a roughly similar proportion of PCs (Otsu et al. 2020). This would explain the similarities between the spontaneous activity recorded here and the interictal discharges observed in the subiculum of the sclerotic hippocampal formation of patients suffering from TLE (Cohen et al. 2002; Reyes-Garcia et al. 2018).

KCC2 Model of “Epileptogenesis In Vitro”: Interictal- and Ictal-Like Electrical Activity

Under our experimental conditions, we observed interictal-like discharges but never detected events reminiscent of long-lasting ictal-like activity. We offer the following potential explanations.

First, under our experimental conditions in vitro, subicular networks have a decreased number of neurons and reduced connectivity compared with the intact situation in vivo. This may explain why we never observed full-blown ictal-like events, which may require a larger, more intact neuronal network. However, other mechanisms related to synaptic or intrinsic plasticity may also be relevant. In fact, a previous study (Huberfeld et al. 2011) suggested that ictal discharges are driven by pre-ictal events that depend critically on NMDA receptor-dependent synaptic plasticity. NMDA receptor-dependent long-term potentiation was directly observed in cultured rodent slices when epileptiform activity was maintained for several hours (Abegg et al. 2004). Therefore, it is possible that our acute experimental preparation did not allow for sufficient development of this type (or of other forms) of synaptic plasticity. A third factor may be the loss of subcortical dynamic neuromodulatory inputs, which could be required to generate ictal-like events (for example, acetylcholine, see: D'Antuono et al. 2007) or indirectly favor plasticity (Palacios-Filardo and Mellor 2019). Additionally, human epileptic tissue shows hyper-innervation of PCs by PV terminals (Muñoz et al. 2007), which may increase the impact of PVs on the network, but this change is not present under our experimental conditions. Lastly, the reported deafferentation of the subiculum regards primarily CA1 inputs (Huberfeld et al. 2015), and the combination of intrinsically-generated rhythms with direct external inputs from the entorhinal cortex may be required to detonate an ictal-like event (Herrington et al. 2015). To limit the effect of VU exclusively to local subicular circuits, we were forced to use a preparation disconnected from any external input, including the entorhinal cortex. The actual relevance of some of these hypotheses can be tested experimentally. For example: 1) longer exposure to VU0463271 may be attempted to trigger various forms of plasticity, 2) neuro-modulation may be simulated by exogenous application of specific pharmacological agonists, and 3) the impact of entorhinal afferents by optogenetic stimulation (Suh et al. 2011) can be tested even in a reduced subicular slice.

Contribution of PVs to Epileptiform Activity

The role(s) played by cortical GABAergic interneurons in the epileptic brain are increasingly being recognized as complex (Jiang et al. 2016; Cǎlin et al. 2018; Lévesque et al. 2019; Magloire et al. 2019). In addition to their remarkable morphofunctional diversity (reviewed by Tremblay et al. 2016 in the cortex and by Pelkey et al. 2017 in the hippocampus), it is important to consider the specific type of activity investigated (inter-ictal, pre-ictal, or ictal), the area of interest (inside/outside epileptic focus) and the temporal stage considered during the natural history of the disease. After neuronal loss in the CA1 subfield (Huberfeld et al. 2015), a fraction of the PCs of the subiculum of TLE patients regress to a state lacking KCC2 expression, which perturbs intracellular chloride homeostasis (Cohen et al. 2002; Huberfeld et al. 2007). Our study suggests that, at this stage, reduced KCC2 function in subicular circuits may trigger epileptiform discharges that are dependent on PVs. The most parsimonious explanation is that the natural tendency of local subicular excitatory circuits to generate epileptiform activity (Fiske et al. 2020) cannot be suppressed by GABA_A receptor-mediated signaling under conditions of impaired KCC2 functions. Comparing the epileptiform activity recorded in the complete absence of inhibition (Fiske et al. 2020) with the data presented here in the presence of VU0463271 reveals remarkable differences, suggesting a more complex scenario. In the complete absence of inhibition, spontaneous epileptiform events occurred more infrequently than events resulting from KCC2 inhibition, by about 2 orders of magnitude. Therefore, we suggest that subicular population bursts in the absence of GABAergic transmission (Fiske et al. 2020) are likely to be initiated by the random firing of PCs interconnected by a positive synaptic feedback system (Böhm et al. 2015; Fiske et al. 2020) similar to what has been proposed to occur in the hippocampus (Traub and Miles 1991). In contrast, the more frequent occurrence of epileptiform events observed in the presence of VU0463271 might be predominantly initiated by fast signaling devices such as PVs themselves (Jonas et al. 2004). This is also supported by our direct observation that PVs initiate firing during synchronous activity. The possibility of PVs driving PCs appears especially relevant given the proximity of their postsynaptic target domains to the action potential initiation site of PCs (Colbert and Johnston 1996). In addition, PVs are endowed with a dense axonal arborization, which controls a large number of target cells (Cobb et al. 1995; Miles et al. 1996) as also suggested here by the higher probability of finding connections from PV to PC compared with PC to PC connections (about one order of magnitude higher, see: Böhm et al. 2015 and Fiske et al. 2020).

Conclusion

In summary, our work has taken a direct and quantitative approach to study the mechanistic role of subicular interneurons in the generation of epileptiform network activity in the subiculum, a key region involved in TLE. Our results are consistent with the “KCC2 downregulation” theory of epileptogenesis and suggest that the powerful perisomatic input of PVs (Freund and Katona 2007), when altered by reductions in KCC2 activity, is sufficient to produce network hyper-excitability and hyper-synchronization that is reminiscent of epileptic interictal-like activity. Our model may provide the opportunity to study additional factors potentially involved in interictal-ictal transition, such as synaptic and/or intrinsic plasticity, the strength of neuro-modulation and, finally, the role of entorhinal afferents.

Lastly, these findings are also of general interests as mutations of phosphorylation sites of the KCC2 transporter are associated with epilepsy (Duy et al. 2019), which raise the possibility of exploiting its intrinsic modulation (Moore et al. 2018) or increasing its expression levels (Tang et al. 2019; Lorenzo et al. 2020) as future therapeutic strategies.

Notes

We would like to thank Drs Anis Contractor and Gordon M.G. Shepherd, (Northwestern University) and Dr Alfonso Apicella (University of Texas at San Antonio) for reading the manuscript. We are grateful to Dr John Dempster (University of Strathclyde) for the use of the Strathclyde Electrophysiology Software package. We also indebted to Ms Sun Kyong Lee for technical assistance. Conflict of Interest: None declared.

Funding

This work was supported by the National Institute of neurological Disorders and Stroke (grant NS096092 to G.M.). Confocal images were generated at the Northwestern University Center for Advanced Microscopy generously supported by the National Cancer Institute CCSG P30 CA060553 awarded to the Robert H Lurie Comprehensive Cancer Center.