All animals received humane care in compliance with the European Communities Council Directive of 2010 (2010/63/EU), and the study was approved by the institutional and regional committees for animal care. Experiments were performed on 25 adult (4–10 weeks old) male mice from the GAD67-GFP (delta neo) line (Tamamaki et al., 2003). The mice expressed eGFP protein under control of the GAD67 promoter, allowing fluorescent labeling of GABAergic interneurons in live tissue. Male animals were used similarly to earlier studies in order to restrain variability (Bouilleret et al., 1999) and recent analysis has confirmed the rationale for this selection due to differences related to sex in this model (Twele et al., 2016).

Experiment design required a delay between kainic acid lesion and in vitro tissue recording (Figure 1A). Mice 3–4 weeks old, 3–5 days after weaning, were anesthetized by isoflurane inhalation (1–1.5%) and placed in a stereotaxic frame. A droplet of 50 nl of KA solution was injected in the right dorsal hippocampus (from the bregma: anteroposterior -1.8 mm, mediolateral -1.8 mm, dorsoventral -1.8 mm) through a fine needle of outer diameter 0.1 mm (Hamilton, NV, United States), connected to a 10 μl syringe, mounted on an electric pusher (WPI, Sarasota, FL, United States). The KA solution contained: KA (20 mM, Sigma–Aldrich), NaCl (0,9%) and Ruby Red dye (M = 3000, 10 mg/ml, Invitrogen), so that the site of injection could be identified. The injection procedure required 30–40 min of isoflurane anesthesia. In this model, the injection site is central between Dentate Gyrus, CA3 and CA1, as kainate diffuses from this point to lesion the dorsal hippocampus. As expected the lesion was observed in the dorsal tissue ipsilateral to the injection (Figure 1B), and the fluorescence distribution was altered by the lesion. Since mice in the control group did not receive saline injection, it cannot be excluded that some of the changes observed could be related to the injection procedure rather than to the epileptogenic effect of kainate. In those animals, GAD67-GFP interneurons spread across all areas and layers of the hippocampus. In KA-treated animals the injection site was sclerotic, and the anatomical layers became less visible in the dorsal hippocampus, while in the ventral hippocampus the tissue was less affected.

Slices were prepared 2–3 weeks post-injection, or at the corresponding age of 5–7 weeks in untreated animals. In the kainate lesion model, spontaneous seizures appear 2 weeks after the lesion. We did not check whether each animal presented seizure at the time of tissue preparation, because seizure time can be as low as a few minutes per 24 h. However, we performed the routine check for this model to observe tissue lesion at the injection site, as slices were cut. Animals were anesthetized by brief inhalation of halothane and were then decapitated and their brain removed. A tissue bloc containing left and right hippocampi was cut in horizontal slices of 350 μm thickness with a vibratome (Microm HM 650V, Walldorf, Germany) in a cold solution (2–4°C) containing (mM): choline chloride (110), KCl (2,5), NaH2PO4 (1,25), NaHCO3- (25), D-Glucose (7), CaCL2 (0,5) and MgCl2 (7). Slices were then transferred to a storage interface-type chamber (35°C), where they were placed in an artificial cerebro-spinal fluid (ACSF) solution containing (mM): NaCl (125), KCl (2,5), NaHCO3- (26), D-Glucose (10), MgCl2 (2) and CaCl2 (2), while their upper surface was exposed to a humidified O2/CO2 95/5 atmosphere (Maier et al., 2009). Cutting and ACSF solutions were oxygenated with O2/CO2 95/5 gas mixture (pH 7,4). One slice was transferred to the submerged recording chamber and immobilized with a nylon mesh. Continuous perfusion was performed with oxygenated ACSF, 10 ml/min at 32°C. Perfusion flow reached both the lower and the upper sides of the slice (Hájos et al., 2009). This recording condition preserved spontaneous activity. The chamber was placed on an Axioscope 2 FS microscopy (Zeiss) platform, equipped with a 40× DIC objective. Green fluorescent GAD67 interneurons were visualized using selective filters (Ex: 485/20 nm, Em: 530/25 nm).

Neuronal population activity was recorded with a 16 channels silicon extracellular electrode (NeuroNexus, Ann Arbor, MI, United States), mounted on a micromanipulator. The electrode was composed of two branches spaced by 150 μm, each branch comprising two tetrodes (Figure 1C). Each electrode channel could detect spontaneous local field potentials, including extracellular multiunit action potentials (Cohen and Miles, 2000) and extracellular synaptic potentials (Bazelot et al., 2010). The silicon electrode was placed in CA3 with one tetrode in Stratum Oriens, two tetrodes in Stratum Pyramidale and one tetrode in Stratum Radiatum. Signals were amplified 1000 times and filtered between 1 Hz and 20 KHz, by a 16 channels amplifier with high input impedance (Xcell, Dipsi, Châtillon). Noise amplitude ranged from 5 to 10 μV, while action potentials, observed in Stratum Pyramidale, ranged from 40 to 200 μV.

Interneurons were studied with the patch-clamp technique in the whole-cell configuration. Borosilicate pipettes were filled with a solution containing (mM): K-Gluconate (130), KCl 5), HEPES (10), EGTA (10), MgCl2 (2), Mg-ATP (4), Tris-GTP (0,4), Na2-phosphocreatinin (10), and biocytin (0,1%). The pipette resistance ranged from 2 to 6 MΩ in ACSF. This solution generated a non-corrected liquid junction potential of -15 mV. All chemicals were purchased from Sigma–Aldrich. An Axopatch 200A amplifier (Molecular Devices) was used in current-clamp mode to establish whole-cell configuration. Data were digitalized by a 16-bit analogic-digital conversion interface with 16 channels (Digidata 1322A, Molecular Devices), and displayed with the accompanying software Axoscope.

Interneuron intrinsic properties were derived from current-clamp recording. Resting potential was obtained by averaging membrane potential for 10 s when no current was injected. Membrane resistance and capacitance were measured from hyperpolarization steps of 50 pA. We computed the average of 10 traces without synaptic events. An exponential curve was adjusted to obtain values for time constant, resistance and capacitance. Then, membrane potential was adjusted to -65 mV and a series of current steps lasting 1000 ms were applied. We counted the number of elicited action potentials generated and identified the threshold values as the intersection of the linear fit of the frequency-injected current (F-I) curve with the axis of current intensities. Input-output gain was computed as the slope of the linear fit.

Extracellular signals recorded from each electrode were filtered to isolate local field potentials (SEP, 5–300 Hz) characteristic of synchronous activities. They were automatically detected by an algorithm previously presented in (Cohen and Miles, 2000) that gave a proportion of false positives and false negatives that did not exceed 3%. We compared the characteristics of SEP measured in Stratum pyramidale, using a single electrode site from each multichannel recording. SEP duration was estimated as the time of the rising phase of the spike, which is less sensitive to noise than the full spike-wave duration.

We distinguished SEP of amplitude less than 400 μV that had similar distribution in control and KA-treated tissue and those of larger amplitude that were at least an order of magnitude more abundant in epileptic tissue. Small SEP events correspond to synchronization of only a fraction of neurons, while large SEP events correspond to massive synchronization (Cohen et al., 2006). The detection of SEP signals from recording sites located 200 μm apart implied that both small and large SEP synchronous activities were taking place across spatially extended neuronal populations. In order to compare the SEPs that we observed with reported spectral signatures of sharp waves (Foffani et al., 2007), we computed the spectrum of SEP of small and large amplitude (Supplementary Figure 1). Spectra were computed by averaging STFTs of individual SEPs as in Foffani et al. (2007), with additional high-pass filtering at 50 Hz (Bessel 4th order in forward and reverse time directions) to attenuate the dominant low-frequency component and better reveal ripple or potential fast-ripple components.

Signal analysis was performed with routines implemented using Labview (National Instruments, Austin, TX, United States). Since the kainate lesion was unilateral, we distinguished three groups: control tissue, tissue ipsilateral and ventral relative to the lesion, and contralateral tissue. Combined patch-clamp and extracellular recordings analysis was performed on a total of 25 cells from 22 slices, 15 animals (8 control and 10 with kainate lesion). Numbers per group are: Control (10 cells from 9 slices, 8 animals), Contralateral (6 cells from 6 slices, 5 animals) and Ipsilateral (9 cells from 5 slices, 5 animals). One animal was prepared per day, yielding recordings from 1 or 2 slices. For kainate lesioned animals, when two slices were recorded the same day, they could be either from same or opposite sides, which was chosen before preparing the tissue and aimed at ensuring groups had similar sizes. Recording lengths were of comparable size, ranging from 30 to 45 min. For intracellular properties in control animals, we pooled another 12 patch-clamp recordings from another 10 slices, 9 animals.

During intracellular recordings, the cell was filled with biocytin (0,1%) by diffusion from the pipette. After the recordings, each slice was fixed in a 4% Paraformaldehyde solution, dissolved in a 0,12 M phosphate Buffer (PBS) solution, and left under continuous agitation at 4°C during 24 h. Slices were then transferred to a 0,12 M PBS solution, sucrose (30%) at 4°C for cryoprotection and each slice was left for at least a night under agitation. Freezing-unfreezing (x3) on dry ice was then used to permeabilize the membranes, followed by three washes (PBS 0,12 M at ambient temperature). Saturation and permeabilization were performed with PBS 0,12 M, milk (0.2%), Triton 100X (0.3%), for 3 h at ambient temperature. Slices were then incubated overnight with 1/500 Streptadivin-Cy3 (Invitrogen) and 1/1000 DAPI in PBS 0,12 M, milk (0.2%), Triton 100X (0.1%) at 4°C. The Streptavidin revealed biocytin (avidin-biotin complex) whilst DAPI revealed cellular nuclei. After three washes with PBS 0,12 M, the slices were mounted in a Prolong Gold Antifade medium (Molecular probes, Invitrogen) that preserved fluorescence. Cy3, GFP, and DAPI were visualized with a fluorescent microscope equipped for structured light imaging (Optigrid system, Qioptiq). Acquisition was made with 4× or 20× lens with a high-resolution camera (Qimaging Retiga EXi 1392 pixels × 1042 pixels). The images were scanned and visualized using Volocity software (Improvision, Perkin Elmer).

Cell counts and soma size estimation were performed on cumulative stacks of 350 μm thickness, using the GFP fluorescence in this mouse line. CA3 sectors were divided according to Cornu Ammonis layers. We pooled Lacunosum Moleculare and Radiatum as the exact border between the two was not always sharp. Counts were divided by the volume examined to obtain a cellular density, in order to compensate for possible hippocampal KA-induced shrinkage (Wolf et al., 2002).

Statistical analysis was performed with Labview and R language and environment for statistical computing (R Core Team, 2017).

Spontaneous extracellular potential features are normalized for recording length, so that number of event expected for a given amount of time, for a range of the variable, corresponds to the area under the curve multiplied by the time measured in seconds. For distributions relative to firing we used Kolmogorov–Smirnov two-sample test and Bonferroni correction for multiple testing.

Tissue and cellular properties, may have non-Normal distributions, which require non-parametric test. We used Anderson–Darling two-sample test, which offers good power even for limited sample size, and Benjamini–Hochberg procedure for multiple testing.

Spontaneous neuronal activity was assessed in acute slice preparation by recording spontaneous extracellular SEPs, in control tissue, and ipsilateral or contralateral side of KA lesion (Figure 2 and Table 1). Slices exhibited stable spontaneous activity for prolonged recording sessions (>1 h). Both control and KA slices exhibited spontaneous SEP discharges (Figure 2A). These discharges spanning a wide range of amplitudes were similar to those induced pharmacologically in acute experiment, for which amplitude is indicative of partial or extensive synchrony of neuronal populations (Cohen et al., 2006).

Activity patterns were different in the three groups (Figures 2B–D). Control slices exhibited discharges of small amplitudes (Figure 2B). Epileptic slices further exhibited large amplitude discharges together with short episodes of ramp up activity (Figures 2C,D). We pooled the SEPs detected in our recordings to estimate the distribution of their characteristic shape and interval. Distributions were significantly different for amplitude, interval between discharges and duration (Table 1). While discharges below 400 μV had similar distribution pattern for all groups, the most prominent signature of epileptic tissue was an increased amount of discharges above 500–700 μV, rising to 10-fold, and more, above control tissue for larger amplitudes (Figure 2E). Thus, we distinguished small SEPs below 400 μV and larger SEPs that contribute to a distinct event distribution in epileptogenic transformation. Changes in interval distribution were more pronounced in tissue ipsilateral to the KA injection, with a ten-fold increase for intervals above 4 s (Figure 2F). Duration of SEP fell sharply in control tissue at around 20 ms, while both ipsi- and contralateral tissue distributed to larger values, with stronger difference on the contralateral side (Figure 2G). In order to compare SEPs with sharp waves reported earlier (Foffani et al., 2007), we computed the spectral signature of small and large SEPs (Supplementary Figure 1). In our protocole frequency spread of the low frequency component overlapped with potential ripple frequency (around 200 Hz), and would thus make further inference rely on modeling of the spectrum decomposition. A single peak around 300 Hz in the ipsilateral group fell below the range of fast-ripple (above 400 Hz).

Since interneurons exert a strong control on hippocampal network activity through inhibitory feedback, we compared interneurons without and with epileptogenic lesion, starting with their anatomical distribution (Figures 3A,B). To quantify these changes, we first counted the number of cells in stacks of images taken from our 350 μm slices (Figure 3C). Significant cell loss was found in the ipsilateral side, in all layers (Table 2). As expected, the damage was observed in the tissue closer to the lesion site, yet the contralateral side also presented changes, except in Stratum Oriens. Soma size (Figures 3D,E and Table 3) increased in all layers on both sides, relative to control. In Lucidum, both density and soma size were significantly different between ipsilateral and contralateral sides. Areas of greater cell loss coincided with stronger soma enlargement, suggesting that cellular hypertrophy took place to compensate for the functional deficit of inhibition induced by cell loss and synaptic reorganization.

We then asked if the changes in interneuron size could impact their intrinsic and firing properties. In order to limit the effect of interneuron diversity, we focused on cells with soma in Stratum Pyramidale, where cell density shows a strong decrease. This population includes basket cells that exert fast feedback inhibition on pyramidal cells (Miles et al., 1996). Intrinsic membrane properties were characterized using intracellular recordings of visually identified fluorescent cells (Figures 4, 5 and Tables 4, 5).

The passive properties were estimated from the voltage response to hyperpolarizing current step. We estimated the resistance from the maximum amplitude of the voltage step and the capacitance by dividing the time constant by the resistance. Each group of cells showed distinct combination of resistance and capacitance (Figure 4A). Resistance (Figure 4B) was different between control and contralateral, control and ipsilateral and contra- and ipsilateral. Capacitance (Figure 4C) showed the same pattern of statistical differences.

Since the change in time constant could impact the firing property of the interneuron, by filtering input and intrinsic currents, we tested how the resting membrane potential and the firing threshold varied. Resting membrane potentials, around -60 mV, did not present a significant difference between control, ipsi- or contralateral tissue. We estimated the threshold current that could elicit firing by applying series of current steps and counting the number of action potentials evoked (Figures 5A,B). We found a significant increase in the threshold current required to elicit action potentials in contralateral tissue, where membrane time constant was the most strongly altered during epileptogenesis.

Although changes in cellular properties varied in the direction of decreased excitability in epileptic tissue, cellular contribution to population activity also depends on the numbers of cells and their connectivity. In order to address directly the question of the contribution of interneurons to synchronous discharge, we combined intracellular recording with extracellular monitoring of population discharges (Figure 6). In control tissue, SEPs occured mostly in association with either subthreshold excitatory synaptic potential (Figure 6A) or action potentials (Figure 6B) in the intracellular recordings. In contrast, the typical behavior of interneurons in epileptic tissue was to exhibit little synaptic potential or firing associated with SEPs of equivalent amplitude (Figures 6C,E). However, these neurons strongly fired when SEPs of larger amplitude were generated in the tissue. These findings suggest that interneurons in KA treated mice do not lack synaptic input, yet they are functionally tuned to selectively respond to larger population discharges.

To study the tuning of interneurons to the largest population discharges, we measured across our recordings how an interneuron responded to SEP in two distinct ranges of amplitude, 30–300 μV and 400–1200 μV (Figure 6E). Distributions were significantly different across groups of tissue and amplitude ranges (Table 6). In control tissue interneuron firing was time-locked to small SEPs, and was stronger during large SEPs but not time-locked to them. On the contrary, in epileptic tissue time-locking of interneurons was strong for large SEPs, with less firing than during small SEPs at intervals larger than 0.1–0.2 s, with a subsequent refractory period. In order to confirm this pattern, we measured, for SEP event of a given amplitude, the occurrence of intracellular firing in a time window of ±0.1 s (Figure 6F), and we counted the number of action potentials (Figure 6G). Control and epileptic tissue showed very distinct behavior of interneurons. In control tissue, interneuron firing was relatively stable over the range of amplitudes, with a decay from around 1/4 chance of firing and 0.4 action potential per SEP below 300 μV down to 1/5 and 0.2 for large SEPs. On the contrary, in epileptic tissue firing probability started below 1/10 probability with less than 0.1 action potential per SEP and increased up to one-to-one firing of around 2 action potentials per SEP above 600–700 μV. Thus, epileptic tissue interneurons appear to be irresponsive to synchrony of sparse populations but contribute to all the largest population discharges.