Abnormal UP/DOWN Membrane Potential Dynamics Coupled with the Neocortical Slow Oscillation in Dentate Granule Cells during the Latent Phase of Temporal Lobe Epilepsy123

Visual Abstract


Introduction
During slow-wave sleep and anesthesia, neocortical networks are dominated by the slow oscillation [slowwave oscillation (SWO), 0.1-2 Hz; Steriade et al., 1993Steriade et al., , 2001Cowan and Wilson, 1994]. At the cellular level, the membrane potential (V m ) of nearly all neocortical neurons synchronously alternates between epochs of steady depolarization (the UP state), associated with firing, and epochs of long-lasting hyperpolarization and quiescence (the DOWN state), leading to a bimodal distribution of their V m Volgushev et al., 2006;Rudolph et al., 2007;Chauvette et al., 2011). The SWO is not confined to neocortical networks and propagates to other subcortical (Wilson and Kawaguchi, 1996;Collins et al., 2001;Loewenstein et al., 2005;Mahon et al., 2006;Ros et al., 2009) and paleocortical (Isomura et al., 2006;Hahn et al., 2012) structures, where the V m of principal cells also rhythmically alternates between UP and DOWN states. The SWO also propagates to the hippocampal formation (Isomura et al., 2006;Wolansky et al., 2006;Hahn et al., 2007). However, in contrast to cells in neocortical, subcortical, and paleocortical areas, the V m of hippocampal principal neurons does not display rhythmic alternations between UP and DOWN phases for prolonged periods of time and lacks bimodality (unlike hippocampal interneurons, see Hahn et al., 2006). This suggests a lower impact of neocortical SWO on the V m dynamics of hippocampal principal neurons than in other neocortical and subcortical structures. In line with this, the hippocampus can sometimes escape from the neocortical influence and generate specific activity patterns such as sharp-wave ripples during the neocortical silent (DOWN) state (Isomura et al., 2006).
The aforementioned relative uncoupling of the hippocampus with neocortical SWO is in agreement with the gating function of the dentate gyrus, a major entry point to the hippocampus (Acsády and Kali, 2007). Indeed, dentate granule cells (DGCs) are relatively silent [e.g., firing no or few action potentials (APs) spontaneously] in awake, anesthetized, and sleeping animals (Jung and Mc-Naughton, 1993;Leutgeb et al., 2007;Neunuebel and Knierim, 2012;Pernía-Andrade and Jonas, 2014;Diamantaki et al., 2016;Kowalski et al., 2016) and the proportion of active cells in a given environment is low (Chawla et al., 2005). In temporal lobe epilepsy (TLE), the dentate gyrus undergoes profound structural and functional network modifications (Dudek and Sutula, 2007). Several subtypes of inhibitory interneurons are lost (Sloviter, 1987;Houser and Esclapez, 1996), and the axons of DGCs (the mossy fibers) sprout to form a direct excitatory feedback circuit not present in controls (Tauck and Nadler, 1985;Represa et al., 1987;Scharfman et al., 2003;Epsztein et al., 2005). Altogether, these morphofunctional changes increase the excitability of DGCs (Artinian et al., 2011) and render the dentate "gate" more permissive to the propagation of neocortical pathological activities, such as epileptiform discharges (Nagao et al., 1996;Behr et al., 1998;Pathak et al., 2007;Bragin et al., 2012;Krook-Magnuson et al., 2015). These changes and associated gating abnormalities start to be observed very early during epileptogenesis (Wuarin and Dudek, 2001;Pathak et al., 2007), and notably before the recording of the first spontaneous seizures (the latent period). Spatial memory deficits associated with the disease can already be observed at this early stage (Chauvière et al., 2009). Here we asked whether the impact of the physiological SWO on the dynamics of the V m and firing of DGCs could be altered early during epileptogenesis.
To address this question, we combined whole-cell patch-clamp recordings of DGCs in vivo together with neocortical local field potentials (LFPs) recordings of the SWO in control and TLE rats under anesthesia during the latent phase of TLE [post-status epilepticus (SE) rats]. In contrast to control rats, the V m of DGCs in post-SE rats showed strong UP/DOWN state dynamics that were time locked to the neocortical SWO. Furthermore, the firing of DGCs was increased and more strongly paced by the SWO in post-SE rats. We conclude that the processing of SWO, an important physiological pattern of neocortical activity during sleep, by DGCs is impaired early during epileptogenesis.

Animals
All experiments were approved by the Institut National de la Santé et de la Recherche Médicale (INSERM) animal care and use committee and in accordance with the European community council directives (2010/63/UE). Data were obtained from male Wistar rats between the ages of postnatal day 27 (P27) to P43 (weight range, 100 -130 g). We used the lithium/pilocarpine model of temporal lobe epilepsy with a ramp induction protocol. Rats were first injected with lithium chloride (3 mEq/kg). Fourteen to eighteen hours later, they were pretreated with methylscopolamine nitrate (1 mg/kg) to reduce the peripheral effects of pilocarpine. Thirty minutes after, they received a first low dose of pilocarpine (10 mg/kg). The same pilocarpine dose was repeatedly injected every 30 min until the behavioral manifestation of the first seizure was observed. The first seizure usually appeared after three to four doses of pilocarpine (30-40 mg/kg). This allowed us to take into account the individual differences in pilocarpine susceptibility between rats. One hour after the onset of status epilepticus, diazepam (8 mg/kg) was administered to stop the seizure. Recordings were performed 6-12 d after the status epilepticus. During this period, only interictal-like activities (ILAs), not seizures, could be recorded (mean frequency, 0.15 Ϯ 0.07 Hz; ϳ9 ILAs/min; n ϭ 19 rats).

Surgery
Animals were anesthetized with 1.5-2 g/kg urethane. The level of anesthesia was assessed using tail or foot pinching, body temperature, and recordings of the neocortical local field potentials. Body temperature was maintained at 37°C with a heating blanket (FHC). The animals were head fixed in a stereotaxic apparatus (SR-6, Narashige). A local analgesic (lidocaine) was applied as a gel on the ear bars to reduce pain during head fixation and was injected as a liquid below the skin before the first incision. An ophthalmic gel was applied to the eyes to prevent them from drying out during the surgery, and the eyes were covered with a piece of cardboard to protect them from the surgical light. The skull was exposed, and two small craniotomies (1.5 mm diameter) were drilled above the right hippocampus (Ϫ3.5 mm posterior to bregma; 2.5 mm lateral to bregma) to record in the dentate gyrus (ϳ3 mm deep from brain surface) and above the parietal cortex (Ϫ4 mm posterior to bregma; 4 mm lateral to bregma) to record in the parietal cortex (1 mm deep from brain surface with an anteroposterior angle of 20°).

In vivo patch-clamp recordings
The V m of DGCs was recorded in current-clamp mode using standard techniques for blind in vivo whole-cell recordings (Margrie et al., 2002). Borosilicate glass patch pipettes (resistance, 7-10 M⍀) were filled with a solution containing the following (in mM): 130 KMeSO4, 5 KCl, 5 NaCl, 10 HEPES-K, 2.5 MgATP, 0.3 NaGTP, 0.2 EGTA, and 0.1% biocytin, pH 7.25, adjusted with KOH. The V m was amplified by an Axoclamp-2B Amplifier (Molecular Devices), low-pass filtered at 3 kHz, and digitized with a Digidata 1440A Digitizer (Molecular Devices) at 20 kHz. Only cells with overshooting APs, resting V m less than Ϫ55 mV, and series resistance (R s ) Ͻ100 M⍀ were kept for further analysis. The V m was corrected for a 10 mV liquid junction potential.

LFPs and MUA recordings
LFPs and multiunit activity (MUA) were recorded from the parietal cortex and dentate granule cell layer using a glass electrode (ϳ5 M⍀) filled with Ringer's solution (135 mM NaCl, 5.4 mM KCl, 1 mM MgCl 2 , 1.8 mM CaCl 2 , and 5 mM HEPES, pH adjusted to 7.2 with NaOH, and target osmolarity of 290 mmol/kg). LFPs and MUA were amplified 1000 times using a DAM80 amplifier (World Precision Instruments), bandpass filtered (0.1-3 kHz), digitized with a Digidata 1440A Digitizer (Molecular Devices), and sampled at 20 kHz. At the end of the recording, the pipette was submerged in a solution containing the red fluorescent dye 1,1'-dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine perchlorate (DiI; Invitrogen), while the rat was still in the stereotaxic frame and reinserted at the recording depth for subsequent histological localization. Electrophysiological signals before and after DiI staining were similar.

Morphology of the recorded cells
At the end of the recordings, animals were injected with an overdose of urethane and transcardially perfused with PBS 1ϫ solution followed by 4% paraformaldehyde. The next day, 100-m-thick coronal slices were cut and processed with the avidin-biotin-CY3 method to visualize biocytin-filled neurons post hoc. The morphology of the recorded cells was reconstructed using Neurolucida software (MBF Bioscience).

Intrinsic intracellular properties
R S values were not compensated for during the recording, but were calculated and compensated for off-line (Crochet and Petersen, 2006). To calculate R S and membrane input resistance (R N ) values, we used voltage responses to multiple hyperpolarizing and depolarizing current steps (100 pA) injected shortly after breaking in. The voltage values used to calculate R S (V S ) were found by exponentially fitting the voltage response (avoiding the first 2 ms during which the time constants resulting from pipette access resistance will dominate) to each current step. The exponential curve was then back-extrapolated to the start time of the step. The intersection of the extrapolated curve with the current step onset time gave V S for each current step, which corresponds to the voltage drop across the R S . R S was then calculated as the slope of the linear fit of the I-V S curve and used to correct V m off-line by subtraction whenever current was injected through the patch pipette. After R S off-line compensation, the R N was calculated for each cell. The voltage values used to determine R N (V N ) were calculated for each step as the difference in V m between the baseline (before the current pulse) and steady-state voltage (100 ms after the start of the current pulse). The R N was also calculated not on a single step but by using all V N values, as the slope of the corresponding I-V N curve. The membrane time constant was given by the time constant of the exponential fit of the V N response to a 100 pA hyperpolarizing pulse. No holding current was injected during the recordings except when otherwise indicated. Resting membrane potential values were calculated as the mean of the Gaussian fit of V m distribution using Ͼ30-s-long epochs. If the V m distribution was best fitted by two Gaussian curves (bimodal cells), we used the voltage corresponding to the peak of the Gaussian curve fitting the lowest V m hump values (the DOWN state). For all V m analysis, the signal was detrended to remove ultraslow (Ͻ0.1 Hz) baseline fluctuations by low-pass filtering (0.1 Hz) then subtracting the low-pass-filtered data to the raw data. Threshold values were calculated using the first action potential in response to depolarizing steps of current as the voltage value at which the rate of depolarization (dV/dt) crossed 10 V/s.

Phase histograms
The times of intracellular APs were detected in highpass-filtered (300 Hz) V m as events higher than a predefined threshold. The times of MUA were determined in high-pass-filtered (1000 Hz) LFPs as events with amplitudes higher than five times the SD of the noise. The level of noise (and thus the detection threshold) was not significantly different between control and post-SE conditions (p ϭ 0.81; n ϭ 6 control rats, n ϭ 6 post-SE rats). The phase of APs and MUA relative to the SWO cycle was derived from a Hilbert transformation of the LFP using the signal-processing toolbox in MATLAB. The peaks of the UP states are defined as 0°after the Hilbert transformation, and the troughs of the DOWN states are 180°. Phase modulation was evaluated by applying Rayleigh circular statistics (Fisher, 1993) .

Slow oscillatory epoch detection
Epochs of significant slow oscillations were detected using temporal spectral analysis of V m in the SWO band (0.1-2 Hz; Fig. 1). Briefly, APs and hyperpolarizing steps used to probe the R S during the recording were removed by linearly interpolating the signal. The time-frequency power spectrogram of V m was computed using a 5 s sliding window advancing in steps of 0.2 s, and the timevarying power in the SWO frequency band (0.1-2 Hz) was extracted (Fig. 1A,B, bottom). For each control DGC, the value (Xi) corresponding to the highest 90% of all power values in the SWO frequency band was determined (Fig.  1C1). The average of all Xi values from control DGCs was then used as a unique threshold for SWO epoch detection in control and post-SE DGCs. The detection was performed on the time-varying power in the SWO frequency band ( Fig. 1C2; see also Sales-Carbonell et al., 2013). A candidate SWO epoch should be above threshold for at least 4 s (e.g., approximately three oscillation cycles) to be considered as significant. For DGCs that spontaneously fired in post-SE rats, the results were not significantly different if APs were not artificially removed from the data (power in SWO band: Student's t test, p ϭ 0.92 mm ; duration of SWO epochs: Student's t test, p ϭ 0.89 nn ; percentage of SWO: Mann-Whitney rank sum test, p ϭ 0.96 oo ; n ϭ 8).

V m distribution
To determine whether the V m of a DGC was unimodally distributed, the Hartigan dip test (Hartigan and Hartigan, 1985) was performed on the distribution of a Ͼ30 s lowpass-filtered (40 Hz) V m epoch centered on epochs of significant slow oscillatory content using the Matlab routine by Nicholas Price (Monash University; http:// www.nicprice.net/diptest/). Nonunimodal cells were classified as bimodal if their V m distribution was best fitted by the sum of two Gaussian curves. Unimodal cells were further qualified as skewed if their skewness, measured with the "skewness" function of the MATLAB statistics toolbox, was Ͼ0.5.
LFP UP-DOWN state detection UP and DOWN states were detected using the bimodal distribution of neocortical LFP values as previously described (Volgushev et al., 2006). Briefly, the level for DOWN state detection was set at the lower two-thirds of the distance between the peaks of the bimodal distribution of LFP values. The level for UP state detection was set at the higher two-thirds of the distance between the peaks of the bimodal distribution of LFP values (see Fig.  4A,B). States were defined as epochs above the threshold for UP states or below the threshold for DOWN states lasting Ͼ200 ms and were terminated when the threshold was crossed for Ͼ40 ms.

DOWN-UP transitions triggered V m averages
The amplitude of V m varies with time and across different cells. To compute the DOWN-UP transition (DUT)triggered V m averages, APs were removed from the signal by linear interpolation and subthreshold V m values were converted into dimensionless units of z-score as follows: , where x is the V m , mean is the average value of the variable, and SD is the standard deviation. Thus, the amplitude of V m is represented in the units of z, thereby allowing easy comparison across different cells. The times of neocortical DOWN state endings were used as transition points from the DOWN to the UP state. The values of z-scored V m , centered on the DOWN-UP transitions of neocortical SWO, were averaged across all detected neocortical DOWN-UP transitions.

Statistics
Statistical analyses were performed using SigmaStat (Systat Software), Statistica (Statsoft), and Prism (Graph-Pad) softwares (Table 1). For comparisons between groups with normal distribution and equal variance, the two-sample unpaired Student's t test was used. When data were not normally distributed or different variances were calculated between groups, the Mann-Whitney rank sum test was used. For comparisons before and after depolarization, the paired Student's t test was used when values were normally distributed; otherwise, the Wilcoxon signed rank test was used. The level of significance was set at p Ͻ 0.05. If not stated otherwise, n refers to the number of cells. The power of statistical tests used was calculated using GraphPad StateMat software (Graph-Pad). Values are given as the mean Ϯ SEM.

Pharmacological agents
All drugs (lithium chloride, scopolamine, pilocarpine, diazepam, and urethane) were diluted in NaCl 0.9% and administered intraperitoneally. Scopolamine methyl nitrate and pilocarpine hydrochloride were purchased from Sigma-Aldrich; diazepam was purchased from Roche; streptavidin conjugated to Cy3 was purchased from Jackson ImmunoResearch; and DiI was purchased from Life Technologies.

UP and DOWN state dynamics of dentate granule cell membrane potential in post-SE rats
During slow-wave sleep and anesthesia, the V m of neocortical cells rhythmically alternates between UP and DOWN states (Steriade et al., 1993Cowan and Wilson, 1994). We first asked whether long-lasting epochs of UP/DOWN states could be detected in the V m of DGCs from control and post-SE rats. We performed whole-cell patch-clamp recordings of DGCs in vivo in urethaneanesthetized control and post-SE rats (6 -12 days after induction of a status epilepticus). Visual inspection of the V m of DGCs from control rats did not reveal long-lasting epochs of UP/DOWN state modulation (Figs. 2A, 3A), as previously described (Isomura et al., 2006). In contrast, long-lasting epochs of UP/DOWN state modulation were observed in the V m of the majority of DGCs from post-SE rats (Fig. 2B), interrupted by epochs of reduced slow oscillatory content (Fig. 3B). To automatically detect and compare epochs of slow oscillation in the V m of DGCs, we developed a quantitative routine based on a single threshold set to the highest 90th percentile values of control DGCs' V m power in the SWO band (0.1-2 Hz; see also Sales-Carbonell et al., 2013; see Experimental procedure/ Data analysis/Slow oscillatory epoch detection; Fig. 1C). Epochs of significant SWO power were detected in 6 of 10 DGCs from control rats. The duration of each SWO epoch   3D). Over all recorded cells and recording time, the SWO band power of the Vm of DGCs was approximately three times higher in the post-SE versus the control condition (Mann-Whitney rank sum test, p ϭ 0.004 c ; Fig. 3E). Power spectral analysis of the V m of DGCs centered on these epochs revealed a high peak at ϳ0.8 Hz in post-SE rats, while a smaller peak (Student's t test, p ϭ 0.004 d ) was observed at a lower frequency (ϳ0.4 Hz) in controls (Fig.  2C,D). Autocorrelograms of the V m of DGCs during these epochs revealed a higher rhythmicity in the post-SE versus the control condition with larger negative peaks (minimum amplitude at trough, Ϫ0.26 vs Ϫ0.11; Mann-Whitney rank sum test, p ϭ 0.037 e ; n ϭ 8 and 10 cells, respectively; Fig. 2E,F) and second positive peaks in the post-SE condition, which were not present in controls. We conclude that the V m of DGCs in post-SE rats strongly oscillates in the slow-frequency range for prolonged periods of time, unlike what is observed in controls. We next wondered whether V m dynamics were similar during SWO epochs in DGCs from control and post-SE rats. Analysis of the V m distribution revealed in a majority of DGCs from post-SE rats a nonunimodal but bimodal distribution, a signature of strong UP/DOWN state dynamics (Hartigan's dip test of unimodality, p ϭ 0.017 Ϯ 0.008; see Materials and Methods; Fig. 2B3; n ϭ 5 of 8 cells). In the remaining cells, the V m distribution was unimodal, either symmetrical (n ϭ 1) or skewed (n ϭ 2; see Materials and Methods). In DGCs from control rats, however, the distribution was always unimodal (Hartigan' 4; Fig. 2A3). These differences in V m dynamics between DGCs from control and post-SE rats were not due to major differences in their intrinsic properties (Table 2).
We conclude that SWO and UP/DOWN state behavior are major components of V m dynamics in DGCs from post-SE rats, unlike what is observed in DGCs from control rats.

Temporal relationship between the membrane potential of dentate granule cells and the neocortical slow oscillation in control and post-SE rats
Before investigating the temporal relationship between neocortical SWO and the V m of DGCs, we first wondered whether neocortical SWO was modified during the latent period in TLE (Fig. 4). LFP recordings in the parietal cortex (which projects to the medial entorhinal cortex; Burwell and Amaral, 1998) revealed no change in SWO power (Student's t test, p ϭ 0.40 l ) and rhythmicity (Student's t test, p ϭ 0.71 m ) between control (n ϭ 10) and post-SE (n ϭ 10) rats (Fig. 4A,C-F). The only difference was a small but significant increase in UP phase duration (control rats, 374 Ϯ 16 ms; post-SE rats, 450 Ϯ 28 ms; n ϭ 10; Student's t test, p ϭ 0.03 n ; Fig. 4G), while DOWN phase duration was not changed (Student's t test, p ϭ 0.74 o ; Fig.  4H). However, this difference did not result in a significant change of the parietal cortex SWO frequency (control rats, 0.75 Ϯ 0.07 Hz; post-SE rats, 0.74 Ϯ 0.06 Hz; n ϭ 10; Student's t test, p ϭ 0.88 p ; Fig. 4D,F).
We next looked for a temporal coupling between the neocortical SWO in the parietal cortex and the V m of DGCs. Cross-correlation between the V m of DGCs and LFPs of parietal cortex was small and peaks were highly temporally jittered for DGCs recorded in control rats (r ϭ 0.05 Ϯ 0.03; one-sample Student's t test, p ϭ 0.15 q ; n ϭ 10; Fig. 5A,C,E). In contrast, the correlation was high and highly significant in post-SE rats (r ϭ 0.45 Ϯ 0.04; onesample Wilcoxon signed rank test, p ϭ 0.008 r ; n ϭ 8;  DUT-triggered averaged V m values showed a positive bump at a fixed latency following neocortical DUT in all DGCs in post-SE rats (Fig. 5G). However, the individual responses in control rats were less consistent, often showing an absence of a clear bump after the neocortical DOWN-UP transition and in one case even a downward shift (Fig. 5F). In line with the results from individual cells, in post-SE rats, the mean DUT-triggered V m average showed a clear bump with a peak at 317 ms following neocortical DUTs ( Fig. 5H; n ϭ 3543 DUTs in 8 cells), while no positive bump was observed in controls ( Fig. 5H; n ϭ 4864 DUTs in 10 cells; Student's t test, p ϭ 0.0005 s vs post-SE at peak). We conclude that long-lasting UP/DOWN dynamics of the V m of DGCs in post-SE rats is temporally locked to the neocortical SWO.

Modulation of the firing of dentate granule cells by the neocortical slow oscillation in control and post-SE rats
We next aimed at determining whether the change in the dynamics of the V m of DGCs observed in post-SE rats had any impact on their firing. We did this both at the singlecell level using intracellular recordings of the spiking activity of DGCs (Fig. 6) and at the network level using extracellular recordings of MUA in the dentate granule cell layer (see Fig. 8).
The V m of DGCs from control rats was hyperpolarized (Ϫ79.1 Ϯ 2.47 mV; n ϭ 10), and most of them (80%) were silent (i.e. not firing any spontaneous action potential) during the recording time (mean recording time of silent cells, 892 Ϯ 225 s; range, 410 -2273 s; n ϭ 8). The mean firing rate was low (active cells: 0.14 Ϯ 0.08 Hz, n ϭ 2; all cells: 0.03 Ϯ 0.02 Hz, n ϭ 10). Accordingly, few neocortical UP phases were associated with the firing of intracellularly recorded DGCs in controls (mean, 1.97 Ϯ 1.81%; n ϭ 10). In contrast, 75% of DGCs were sponta- neously firing in post-SE rats (Fig. 6A,B; active cells: 0.84 Ϯ 0.47 Hz, n ϭ 6; all cells: 0.63 Ϯ 0.37 Hz, n ϭ 8; Mann-Whitney rank sum test, p ϭ 0.029 t for all cells), and the number of neocortical UP phase associated with the firing of DGCs was significantly higher (Fig. 6C; mean ϭ 13.4 Ϯ 6.81%; Mann-Whitney rank sum test, p ϭ 0.045 u ; n ϭ 8). However, when spiking was observed during a neocortical UP phase, the mean number of spikes was    Fig. 6D). We conclude that the firing of DGCs in post-SE rats is increased and highly modulated by the neocortical SWO. Because DGCs from control rats rarely fire or fire at a very low rate, it is difficult to determine whether their firing can be modulated by the neocortical SWO. To circumvent this problem, we artificially depolarized control cells to induce a firing rate comparable to that observed in DGCs from post-SE rats (mean Vm after depolarization, Ϫ56.5 Ϯ 2.55 mV; n ϭ 8 of 8 cells firing; mean firing rate, 0.29 Ϯ 0.06 Hz; Mann-Whitney rank sum test, p ϭ 0.02 w vs control rats; p ϭ 0.72 x vs post-SE rats; n ϭ 10 control DGCs; n ϭ 8 control DGCs depolarized; n ϭ 8 post-SE DGCs; Fig. 6E,F). Under depolarization, the number of neocortical UP phases associated with intracellular spiking was also significantly increased (paired Student's t test, p ϭ 0.004 y before vs after depolarization; Fig. 6G; Mann-Whitney rank sum test, p ϭ 0.64 pp vs post-SE; n ϭ 8). However, in this condition, spikes were significantly less strongly and less precisely modulated by the neocortical SWO than spikes from DGCs in post-SE rats (mean length of Rayleigh vector, 0.29 Ϯ 0.07; Student's t test, p ϭ 0.0007 z vs post-SE; mean phase dispersion, 67.7 Ϯ 3.47°; Student's t test, p ϭ 0.0005 aa vs post-SE; n ϭ 8 control DGCs depolarized; n ϭ 6 post-SE DGCs; compare Fig. 6D,H). Furthermore, the modulation of the V m of control DGCs by the neocortical SWO was not significantly increased after depolarization (Fig. 7). Thus, when the spiking of control DGCs is artificially increased through direct depolarization, the modulation of their firing by the neocortical SWO is much weaker and less precise than that of DGCs from post-SE rats. Accordingly, their V m does not show strong SWO modulation. To see whether the increased spiking rate and SWO modulation of DGCs from post-SE rats could be observed at the network level, we next recorded MUA in the dentate granule cell layer of control and post-SE rats. MUA was significantly increased in post-SE (n ϭ 17) versus control (n ϭ 7) rats (Fig. 8A,B,G; 7.80 Ϯ 1.44 Hz in post-SE rats; 1.61 Ϯ 0.41 Hz in control rats; Mann-Whitney rank sum test, p ϭ 0.01 bb ). We compared the strength of the modulation of MUA by the neocortical SWO in control and post-SE conditions (Fig. 8C,E). MUA in post-SE rats was more strongly modulated by the neocortical SWO ( Fig.  8H; mean length of Rayleigh vector, 0.39 Ϯ 0.04 in post-SE rats vs 0.19 Ϯ 0.03 in control rats; Mann-Whitney rank sum test, p ϭ 0.031 cc ), the mean phase at which MUA occurred was not different (Fig. 8I; Mann-Whitney rank sum test, p ϭ 0.95 dd ), but the dispersion around the mean phase was significantly smaller in post-SE rats ( Fig.  8J; mean phase dispersion, 61.6 Ϯ 2.26°in post-SE rats vs 72.7 Ϯ 1.47°in controls; Mann-Whitney rank sum test, p ϭ 0.01 ee ). In sharp contrast, MUA recorded in the parietal cortex was not significantly increased in post-SE (n ϭ 17) versus control (n ϭ 14) rats (Mann-Whitney rank sum test, p ϭ 0.86 ff ; Fig. 9A-C), nor was the strength of MUA modulation by the SWO (Student's t test, p ϭ 0.96 gg ; Figs. 8D,F, 9D), the mean phase (Student's t test p ϭ 0.48 hh ; Fig. 9E) or the phase dispersion around the mean (Student's t test, p ϭ 0.86 ii ; Fig. 9F).

Discussion
In the present study, we investigated whether the impact of the neocortical SWO on DGCs, at the gate of the hippocampus, could be altered in an animal model of TLE. In line with this hypothesis, while the V m of control DGCs weakly oscillated in the SWO frequency range for short periods of time, the V m of most DGCs from post-SE rats showed strong UP/DOWN state dynamics for extended periods, time locked to neocortical SWO. Furthermore, the firing of DGCs was increased and strongly modulated by the neocortical SWO in post-SE rats, but not in control rats. Altogether, these results point to a major alteration in the processing of SWO, an important physiological pattern of activity, by DGCs during the early phase of TLE.

Possible mechanisms of the switch to strong UP/ DOWN state V m dynamics in post-SE DGCs
A great deal of experimental and computational work has been conducted to understand the mechanisms leading to neocortical SWO and associated neuronal UP/DOWN state dynamics in the cortex Bazhenov et al., 2002;Compte et al., 2003;Cossart et al., 2003;Shu et al., 2003;Haider et al., 2006;Holcman and Tsodyks, 2006;Rudolph et al., 2007;Chauvette et al., 2010). SWO has been proposed to result either from synaptic or intrinsic mechanisms. In the synaptic hypothesis, stochastic EPSPs could summate during the DOWN state and reach the activation threshold of intrinsic conductances, such as I NaP , initiating the UP state could result from the spontaneous activation of intrinsically bursting neurons in layer V of cortex (Le Bon-Jego and Yuste, 2007;Ló´rincz et al., 2015). Once an UP state is triggered, a barrage of synaptic excitatory and inhibitory inputs, occurring during the full duration of the UP state, contributes to its steady nature (Shu et al., 2003;Haider et al., 2006;Rudolph et al., 2007). This supposes a network with strong recurrent excitation (Sanchez-Vives and McCormick, 2000;Cossart et al., 2003) as well as feedback inhibition resulting from fast spiking basket cells activated during UP states (Shu et al., 2003;Fanselow and Connors, 2010;Massi et al., 2012;Tahvildari et al., 2012). The end of the UP state has been proposed to result from synaptic depression (Bazhenov et al., 2002;Holcman and Tsodyks, 2006), the buildup of activitydependent potassium conductances (Sanchez-Vives and McCormick, 2000;Compte et al., 2003), and the activation of inhibitory interneurons (Shu et al., 2003) acting through GABA A (Lemieux et al., 2015) and/or GABA B (Parga and Abbott, 2007;Mann et al., 2009) receptors.
The lack of strong UP/DOWN state dynamics in control DGCs could result from the strong local inhibitory circuit of the dentate gyrus (Acsády et al., 1998;Nitz and Mc-Naughton, 2004;Ewell and Jones, 2010;Sambandan et al., 2010) together with strong dendritic filtering (Soltesz et al., 1995;Schmidt-Hieber et al., 2007;Krueppel et al., 2011) that dampen the impact of excitatory inputs coming from layer II cells of the entorhinal cortex. In post-SE rats, however, the level of inhibition in DGCs is reduced during the latent period of TLE, notably at dendritic sites controlling inputs from the entorhinal cortex , presumably as a result of the early loss of several types of inhibitory interneurons in the dentate gyrus (Sloviter, 1987;Houser and Esclapez, 1996). In addition, excitatory inputs from layer II cells of the entorhinal cortex are increased 5 d post-SE as a result of decreased inhibition within the entorhinal cortex . This imbalance in the excitatory/ inhibitory drive could contribute to the strong UP/DOWN state V m dynamics and firing observed in DGCs post-SE. Furthermore, at this early stage of epileptogenesis a switch of E(GABA) toward excitation has been reported (Pathak et al., 2007).
Alternatively, the absence of strong UP/DOWN state dynamics in control DGCs could result from the absence of a direct recurrent excitatory circuit in the dentate gyrus. Indeed, control DGCs only make indirect excitatory connections onto their peers through activation of mossy cells and need to fire at a sufficiently high rate to be able to discharge their targets (Henze et al., 2002). However, the axons of DGCs (the mossy fibers) sprout after status epilepticus to form a recurrent excitatory feedback circuit within the den- tate gyrus not present in controls (Tauck and Nadler, 1985;Represa et al., 1987;Scharfman et al., 2003;Epsztein et al., 2005), which increases the excitability of DGCs (Artinian et al., 2011). These recurrent fibers result in part from the aberrant development of adult-born DGCs induced by the status epilepticus (Kron et al., 2010). This process is likely to be increased in our conditions because we use relatively young animals with a high level of adult neurogenesis and increased plasticity (Kuhn et al., 1996). Mossy fiber sprouting increases with time post-SE. It can already be observed during the latent period, but in a smaller proportion compared with the chronic stage (Wuarin and Dudek, 2001). Interestingly, computational work showed that even a low degree of mossy fiber sprouting can have a strong effect on network dynamics in a context of reduced inhibition (Santhakumar et al., 2005).

Sleep slow oscillation, UP/DOWN states, and epilepsy
Epileptic discharges in focal epilepsy are frequently observed during non-rapid eye movement sleep (Malow et al., 1998). Most previous studies in vivo have focused on the impact of SWO on pathological activities, such as interictal epileptiform discharges or "UP spikes" (de Guzman et al., 2010;Bragin et al., 2012;Frauscher et al., 2015). No studies have analyzed the impact of the SWO itself on the V m of principal cells in this pathological condition. Bragin et al. (2012) reported a decrease in the frequency of the field SWO recorded extracellularly in the dentate gyrus of chronically epileptic mice, due to an increased duration of both active and silent network phases. In our study, we tested the impact of the field neocortical SWO directly on the V m of DGCs. We observed an increased duration and power of the V m slowoscillation of DGCs in post-SE versus control rats. Furthermore, the peak frequency of intracellular V m slow oscillation was higher in the post-SE than the control condition (Fig. 2D). This frequency difference probably results from the fact that in controls, the V m slowoscillation of DGCs is poorly temporally coupled with the neocortical field activity (Fig. 5E). On the other hand, the V m slow-oscillation of DGCs in post-SE rats is highly temporally coupled to the neocortical field SWO (Fig. 5E) and oscillates at the same frequency (compare Figs. 2D, 4D, red traces). Thus, the difference we observed between the V m slow-oscillation frequencies likely reflects an increased temporal coupling in the post-SE condition rather than an absolute change in the frequency of the neocortical field SWO, which is not different (Fig. 4D). The only study on the consequences of a status epilepticus on intracellular UP/DOWN state V m dynamics was performed in vitro using neocortical slices (Gerkin et al., 2010). In that work, the frequency of UP states recorded in layer 2/3 neocortical pyramidal neurons was increased 24 h after a picrotoxin-induced seizure. This resulted from shorter DOWN states, associated with increased single-cell intrinsic excitability. Indeed, a higher number of action potentials was generated per UP state following seizure induction. However, in our study, the frequency of the SWO was not modified in the parietal cortex, and neocortical MUA activity was not increased, suggesting no change of neocortical excitability during the latent period in our model of TLE. In contrast to neocortical neurons in controls, which display strong UP/ DOWN state dynamics, the Vm of control DGCs was only weakly modulated by the neocortical SWO and clear UP/ DOWN dynamics were rarely observed. However, several days after the induction of the status epilepticus, clear UP/DOWN dynamics could be seen. The scarcity of UP/ DOWN V m dynamics in controls and the abundance fol-lowing the SE points to a morphofunctional change in the network, rather than a change in the intrinsic excitability of DGCs, as a possible explanation. In line with this hypothesis, increasing single-cell excitability through direct current injection via the recording pipette was not enough to induce SWO in the V m of control DGCs. Furthermore, the number of spikes per neocortical UP phase was not modified in DGCs from post-SE rats, arguing against an increased intrinsic excitability.

Functional consequences of higher UP/DOWN state V m dynamics in DGCs from post-SE rats
At the entrance of the hippocampus, the dentate gyrus has been proposed to transform dense neocortical inputs into sparse and specific neuronal representations (Chawla et al., 2005;Acsády and Kali, 2007;Leutgeb et al., 2007), which can then be imposed onto their downstream CA3 pyramidal cells target for subsequent storage (Mc-Naughton and Morris, 1987;Treves and Rolls, 1992). In accordance with this function, DGCs are particularly reluctant to fire action potentials both in vitro (Scharfman, 1991;Ewell and Jones, 2010) and in vivo (Jung and McNaughton, 1993;Penttonen et al., 1997;Leutgeb et al., 2007;Neunuebel and Knierim, 2012;Pernía-Andrade and Jonas, 2014;Diamantaki et al., 2016;Kowalski et al., 2016). By bringing DGCs closer to firing threshold, UP states likely facilitated the generation of action potentials. Indeed, 75% of DGCs were spontaneously firing in post-SE rats against 20% in controls, and the firing rate of individual cells was strongly increased. This increase likely results from the increased number of UP states rather than increased intrinsic excitability of DGCs, because the number of spikes generated by individual UP states was not different between control and post-SE rats. The MUA recorded in the dentate granule cell layer was not only increased, but also was more strongly and precisely paced by the neocortical SWO in post-SE versus control rats. Previous studies (Mori et al., 2004) have shown that DGCs can have either an inhibitory (at low firing rate) or an excitatory (at high firing rate) impact on downstream targets. In our recordings, the increase in the firing rate of individual cells, although substantial, was still too low to allow faithful excitation of CA3 pyramidal cells during the UP state by individual DGCs (Henze et al., 2002). The increased frequency could instead enhance their inhibitory impact on CA3 activity during the neocortical UP state already observed in controls (Isomura et al., 2006). On the other hand, because more DGCs are firing spontaneously and because their firing is more strongly synchronized by the SWO, the likelihood of having an ensemble of DGCs coactive will be increased. Altogether, these cells could have a significant impact on single-CA3 pyramidal cell firing behavior. The net impact of these changes on downstream targets remains to be investigated but is difficult to anticipate because CA3 and CA1 networks are themselves profoundly modified in TLE. The slow cortical oscillation is able to bias the occurrence of hippocampal sharp-wave/ripple events (Sirota et al., 2003;Battaglia et al., 2004;Mölle et al., 2006;Logothetis et al., 2012) and is important for hippocampusdependent declarative memory consolidation in humans (Marshall et al., 2006). Future studies will tell whether the increased modulation of the activity of DGCs by the slow cortical oscillation that we report has an incidence on the replay and consolidation of information during sleep in patients with TLE.