Relationship between Delta Rhythm, Seizure Occurrence and Allopregnanolone Hippocampal Levels in Epileptic Rats Exposed to the Rebound Effect

Abrupt withdrawal from antiepileptic drugs is followed by increased occurrence of epileptic seizures, a phenomenon known as the “rebound effect”. By stopping treatment with levetiracetam (LEV 300 mg/kg/day, n = 15; vs. saline, n = 15), we investigated the rebound effect in adult male Sprague-Dawley rats. LEV was continuously administered using osmotic minipumps, 7 weeks after the intraperitoneal administration of kainic acid (15 mg/kg). The effects of LEV were determined by comparing time intervals, treatments, and interactions between these main factors. Seizures were evaluated by video-electrocorticographic recordings and power band spectrum analysis. Furthermore, we assessed endogenous neurosteroid levels by liquid chromatography-electrospray-tandem mass spectrometry. LEV significantly reduced the percentage of rats experiencing seizures, reduced the seizure duration, and altered cerebral levels of neurosteroids. In the first week of LEV discontinuation, seizures increased abruptly up to 700% (p = 0.002, Tukey’s test). The power of delta band in the seizure postictal component was related to the seizure occurrence after LEV withdrawal (r2 = 0.73, p < 0.001). Notably, allopregnanolone hippocampal levels were positively related to the seizure occurrence (r2 = 0.51, p = 0.02) and to the power of delta band (r2 = 0.67, p = 0.004). These findings suggest a role for the seizure postictal component in the rebound effect, which involves an imbalance of hippocampal neurosteroid levels.


Introduction
Epilepsy is a chronic neurological disorder affecting over 50 million people worldwide [1]. Approximately 65% of patients remain seizure-free after undergoing treatment with antiepileptic drugs (AEDs) as a single administration or in combined therapy [2]. However, the only possibility to overcome multidrug resistance in patients with refractory seizures is surgical resection of the epileptic foci. For instance, temporal lobectomy is the most effective treatment for refractory temporal lobe epilepsy [3], which is the most common form of medically intractable epilepsy in adults [4].
In epileptic patients who are candidates for surgical treatment, the rapid discontinuation of AEDs is required during the presurgical evaluation of video-electroencephalogram (v-EEG) monitoring [5][6][7][8]. However, this procedure can trigger the so-called "rebound effect", which is characterized by transient seizure generalization or even prolonged increase in frequency of partial seizures [5]. Other adverse events observed during the presurgical evaluation of v-EEG monitoring could be the status epilepticus or sudden unexpected death in epilepsy [9,10]. These are dramatic events, but also uncontrolled seizures during the rebound effect could be damaging, because they were associated with development of brain hypoxia and neurodegeneration [11].
In general, the clinical effects of AED rapid discontinuation might depend on predicting factors of relapse, such as: (i) age, (ii) family history of epilepsy, (iii) etiology, (iv) EEG abnormalities, (v) number of seizures before discontinuation, vi) duration of seizurefree period on treatment, and vii) number of AEDs needed in the attempt to control the epilepsy [12]. In addition to this, the risk of adverse events and rebound phenomena can be different in relation to administered AEDs [6], but we still lack an easily quantifiable seizure prediction indicator to control the occurrence of spontaneous recurrent seizures (SRSs) during the presurgical evaluation of patients with refractory epilepsy.
Among other AEDs, (S)-α-ethyl-2-oxopyrrolidine acetamide (levetiracetam, LEV) [13] is commonly used in clinical practice. Moreover, LEV has already been tested in different animal models of seizures and epilepsy, in order to evaluate its anticonvulsant effects and tolerability [14][15][16]. Indeed, it has been demonstrated that LEV significantly reduces the seizure frequency during 2 weeks of treatment in pilocarpine-treated rats. However, the individual response of animals to LEV varies from complete seizure control to no effect, even if plasma drug levels are well within the therapeutic range [14].
LEV has the advantage of being water-soluble and thus can be administered without using any solvent potentially able to affect the central nervous system function. To this regard, LEV appears as an ideal tool to test the rebound effect in a post-status epilepticus model of chronic epilepsy [17]. Thus, we decided to use LEV and designed an experiment to fully characterize its therapeutic properties and the consequences of LEV rapid withdrawal. To this aim, kainic acid (KA)-treated rats were first monitored by video-electrocorticographic (v-ECoG) recordings coupled with power band spectrum analysis before, during LEV administration via osmotic minipumps for a week, and subsequently in the weeks following LEV withdrawal. In view of our previous finding indicating a selective reduction of allopregnanolone levels in the hippocampus of KA-treated rats, in which this neurosteroid was related to the seizure occurrence observed in epileptic animals [18], we determined whether allopregnanolone and other neurosteroids could be altered after LEV treatment, using liquid chromatography-electrospray-tandem mass spectrometry (LC-ESI-MS/MS). Indeed, neurosteroids have been implicated in a phenomenon similar to the rebound effect in epileptic female rats treated with progesterone for a week, in which a 2-fold increase in the seizure frequency occurred in 57% of animals in the week following progesterone withdrawal. Notably, this effect was independent of progesterone receptor activation and probably was due to the conversion of progesterone to other neurosteroids [19].

Different Responses to Treatment
During LEV treatment, seven out of 15 rats (47%) presented a complete response consisting of the suppression of SRSs during the overall period of observation. The remaining rats (53%, eight out of 15 rats) experienced an incomplete response to LEV. By comparing the occurrence of SRSs under LEV administration to that observed in the pre-treatment week of the same animals (during which all rats displayed at least one SRS), we observed a significant difference (p = 0.006, Fisher's exact test) suggesting that LEV efficaciously suppressed ictogenesis. In the same time interval, 93% of epileptic animals (14 out of 15 rats) receiving saline displayed SRSs (p = 0.035 vs. LEV).

Characterization of the Duration of SRSs
By visual inspection, SRSs recorded in LEV-treated rats which did not maintain an adequate drug response appeared to differ (i.e, duration and postictal ECoG) from those observed during the week preceding the implantation of minipumps (PRE-LEV and PRE-SALINE), and from those characterized in epileptic rats treated with saline in the same period, as well as in the following week (POST-SALINE). Moreover, SRSs recorded during the treatment with LEV also seemed to have different features, in terms of duration and postictal ECoG, in comparison to those that developed after the treatment (POST-LEV) ( Figure 1). This is in agreement with the literature suggesting that relapsing seizures are related to specific ictal and interictal activity [20].

Characterization of the Duration of SRSs
By visual inspection, SRSs recorded in LEV-treated rats which did not maintain an adequate drug response appeared to differ (i.e, duration and postictal ECoG) from those observed during the week preceding the implantation of minipumps (PRE-LEV and PRE-SALINE), and from those characterized in epileptic rats treated with saline in the same period, as well as in the following week (POST-SALINE). Moreover, SRSs recorded during the treatment with LEV also seemed to have different features, in terms of duration and postictal ECoG, in comparison to those that developed after the treatment (POST-LEV) ( Figure 1). This is in agreement with the literature suggesting that relapsing seizures are related to specific ictal and interictal activity [20].

Figure 1.
Spontaneous recurrent seizures (SRSs) after induction of status epilepticus in epileptic rats treated with levetiracetam (LEV) or saline. SRSs were recorded in epileptic rats during the treatment with LEV or saline, but also 1 week before and 1 week after LEV or saline administration by subcutaneous osmotic minipumps. Note that SRSs appeared in incomplete responders to LEV. Arrowheads point to onset and termination of SRSs. PRE-, before osmotic minipump implantation; POST-, after minipump implantation; SAL, saline.
After the Shapiro-Wilk test (p = 0.254), data were analyzed by 2-way (treatment x time) repeated measures analysis of variance (ANOVA). This analysis showed a statistically significant effect of treatment for daily duration of all SRSs (F(1279) = 14.10, p = 0.002), paralleled by a main effect of time for daily duration of all SRSs (F(20,279) = 2.13, p = 0.004). However, there was no statistically significant interaction between treatment and time for daily duration of all the different types of SRSs (nonconvulsive and convulsive) in rats (F(20,279) = 0.89, p = 0.595), indicating that LEV effects were not dependent on changes in this seizures' feature by time. Precisely, the total duration of SRSs was reduced by LEV in comparison to saline during the first two days of treatment (p = 0.018 for the 1st day, p = 0.004 for the 2nd day; Holm-Šidák test). LEV also decreased the total duration of SRSs at approximately 30% of control levels in the last day of treatment (SALINE vs LEV, p = 0.028) (Figure 2A). Interestingly, these differences were found on the 2nd day of treatment for nonconvulsive, stage 0-3 SRSs (p = 0.04, SALINE vs LEV) ( Figure 2B), and on the 1st (p = 0.048), 2nd (p = 0.016) and 7th day (p = 0.005) for convulsive, generalized tonic-clonic SRSs ( Figure 2C). However, there was no statistically significant interaction between treatment and time for daily duration of all the different types of SRSs (nonconvulsive and convulsive) in rats (F(20,279) = 0.89, p = 0.595), indicating that LEV effects were not dependent on changes in this seizures' feature by time. Precisely, the total duration of SRSs was reduced by LEV in comparison to saline during the first two days of treatment (p = 0.018 for the 1st day, p = 0.004 for the 2nd day; Holm-Šidák test). LEV also decreased the total duration of SRSs at approximately 30% of control levels in the last day of treatment (SALINE vs LEV, p = 0.028) ( Figure 2A). Interestingly, these differences were found on the 2nd day of treatment for nonconvulsive, stage 0-3 SRSs (p = 0.04, SALINE vs. LEV) ( Figure 2B), and on the 1st (p = 0.048), 2nd (p = 0.016) and 7th day (p = 0.005) for convulsive, generalized tonic-clonic SRSs ( Figure 2C).

Characterization of SRSs Occurrence
After the significant result (p < 0.05) of the Shapiro-Wilk test, data were analyzed by Friedman repeated measures ANOVA on ranks. To this aim, we considered all seizures occurring in the different treatment conditions (i.e., preceding the minipump implantation, during saline or LEV delivery, and after removal of minipumps), as illustrated in Figure 3. This analysis revealed that differences in the median values among the treatment groups were statistically different (p = 0.01), when considering the weekly occurrence of all SRSs types. Then, the multiple comparison procedure (Tukey's test) showed that the rapid LEV discontinuation induced an abrupt increase in the weekly occurrence of all

Characterization of SRSs Occurrence
After the significant result (p < 0.05) of the Shapiro-Wilk test, data were analyzed by Friedman repeated measures ANOVA on ranks. To this aim, we considered all seizures occurring in the different treatment conditions (i.e., preceding the minipump implantation, during saline or LEV delivery, and after removal of minipumps), as illustrated in Figure 3. This analysis revealed that differences in the median values among the treatment groups were statistically different (p = 0.01), when considering the weekly occurrence of all SRSs types. Then, the multiple comparison procedure (Tukey's test) showed that the rapid LEV discontinuation induced an abrupt increase in the weekly occurrence of all generalized SRSs, up to 700% of previous median levels in the LEV-treated group (LEV vs POST-LEV, p = 0.002, Figure 3A). Interestingly, these changes did not occur for stage 0-3 SRSs ( Figure 3B) but involved only tonic-clonic SRSs (LEV vs POST-LEV, p < 0.001, Figure 3C). generalized SRSs, up to 700% of previous median levels in the LEV-treated group (LEV vs POST-LEV, p = 0.002, Figure 3A). Interestingly, these changes did not occur for stage 0-3 SRSs ( Figure

Relationship between Frontal Power Band Spectrum and SRSs in the Week Following the Treatment
A total of 26 out of 30 epileptic animals (SALINE = 13, LEV = 13) were analyzed to determine a linear relationship between the frontal power band spectrum of postictal ECoG before and after the treatment (the independent variable), and the total number of SRSs after the treatment (the dependent variable). Two animals (SALINE = 1, LEV = 1) were excluded because their frontal electrodes were damaged (only occipital electrodes were useful to detect v-ECoG signals), whereas two other animals (SALINE = 1, LEV = 1) were excluded because they did not develop SRSs in the week following the treatment.
We did not find any linear relationship between: (i) the frontal power band spectrum in the pre-treatment interval, and the number of SRSs in the week following saline administration ( Figure 4A-F); (ii) the frontal power band spectrum in the pre-treatment interval, and the number of SRSs in the week following LEV administration ( Figure 4G-L); and (iii) the frontal power band spectrum in the post-treatment interval, and the number of SRSs in the week following saline administration ( Figure 4M-R). Interestingly, the power of delta was linearly related to the number of SRSs after LEV administration (r 2 = 0.73, p < 0.001) ( Figure 4S). At variance, a weaker, inverse relationship was found between the power of fast-alpha, and the number of SRSs after LEV administration (r 2 = 0.33, p = 0.041) ( Figure 4V). Finally, no relationship was found between the frontal power of theta, slowalpha, beta, and gamma in the post-treatment interval, and the number of SRSs in the week following LEV administration ( Figure 4T,U,W,X).

Relationship between Frontal Power Band Spectrum and SRSs in the Week Following the Treatment
A total of 26 out of 30 epileptic animals (SALINE = 13, LEV = 13) were analyzed to determine a linear relationship between the frontal power band spectrum of postictal ECoG before and after the treatment (the independent variable), and the total number of SRSs after the treatment (the dependent variable). Two animals (SALINE = 1, LEV = 1) were excluded because their frontal electrodes were damaged (only occipital electrodes were useful to detect v-ECoG signals), whereas two other animals (SALINE = 1, LEV = 1) were excluded because they did not develop SRSs in the week following the treatment.
We did not find any linear relationship between: (i) the frontal power band spectrum in the pre-treatment interval, and the number of SRSs in the week following saline administration ( Figure 4A-F); (ii) the frontal power band spectrum in the pre-treatment interval, and the number of SRSs in the week following LEV administration ( Figure 4G-L); and (iii) the frontal power band spectrum in the post-treatment interval, and the number of SRSs in the week following saline administration ( Figure 4M-R). Interestingly, the power of delta was linearly related to the number of SRSs after LEV administration (r 2 = 0.73, p < 0.001) ( Figure 4S). At variance, a weaker, inverse relationship was found between the power of fast-alpha, and the number of SRSs after LEV administration (r 2 = 0.33, p = 0.041) ( Figure 4V). Finally, no relationship was found between the frontal power of theta, slowalpha, beta, and gamma in the post-treatment interval, and the number of SRSs in the week following LEV administration ( Figure 4T,U,W,X).  and epileptic rats treated with saline (n = 13) were used to evaluate whether a relationship existed between changes in the power band spectrum during the postictal events, and the number of SRSs in the week following the treatment. The relationship (r 2 ) between the frontal power band spectrum in the pre-treatment interval and the number of SRSs in the week following saline administration were determined as in (A-F). Relationship between the frontal power band spectrum in the pre-treatment interval and the number of SRSs in the week following LEV treatment were shown in (G-L). Similarly, relationship between the frontal power band spectrum in the posttreatment interval and the number of SRSs in the week following saline administration were illustrated in (M-R). Finally, relationship between the frontal power band spectrum in the post-treatment interval and the number of SRSs in the week following LEV administration were illustrated in (S-X). Statistical analysis was performed using a linear regression analysis with number of seizures being the dependent and the frontal power band spectrum as the independent variable. PRE-, before osmotic minipump implantation; POST-, after minipump implantation; SAL, saline.

Levels of Neurosteroids in the Hippocampus and Neocortex of Epileptic Rats, Measured 2 Weeks Following the Treatment
The levels of pregnenolone sulfate, pregnenolone, progesterone, 5α-dihydroprogesterone, allopregnanolone, and pregnanolone were determined in both the hippocampus (SALINE = 8, LEV = 11) and neocortex (SALINE = 8, LEV = 13) of 21 epileptic rats, 2 weeks . Relationship between power band spectrum and spontaneous recurrent seizures (SRSs) following treatment with levetiracetam (LEV) or saline. Epileptic animals treated with LEV (n = 13) and epileptic rats treated with saline (n = 13) were used to evaluate whether a relationship existed between changes in the power band spectrum during the postictal events, and the number of SRSs in the week following the treatment. The relationship (r 2 ) between the frontal power band spectrum in the pre-treatment interval and the number of SRSs in the week following saline administration were determined as in (A-F). Relationship between the frontal power band spectrum in the pre-treatment interval and the number of SRSs in the week following LEV treatment were shown in (G-L). Similarly, relationship between the frontal power band spectrum in the post-treatment interval and the number of SRSs in the week following saline administration were illustrated in (M-R). Finally, relationship between the frontal power band spectrum in the post-treatment interval and the number of SRSs in the week following LEV administration were illustrated in (S-X). Statistical analysis was performed using a linear regression analysis with number of seizures being the dependent and the frontal power band spectrum as the independent variable. PRE-, before osmotic minipump implantation; POST-, after minipump implantation; SAL, saline.

Levels of Neurosteroids in the Hippocampus and Neocortex of Epileptic Rats, Measured 2 Weeks Following the Treatment
The levels of pregnenolone sulfate, pregnenolone, progesterone, 5α-dihydroprogesterone, allopregnanolone, and pregnanolone were determined in both the hippocampus (SALINE = 8, LEV = 11) and neocortex (SALINE = 8, LEV = 13) of 21 epileptic rats, 2 weeks after treatment (64 days after KA-induced status epilepticus). Particularly, two hippocampi were excluded Briefly, pregnenolone sulfate was significantly reduced in the hippocampus (p = 0.001) and neocortex (p = 0.009) of epileptic rats treated with LEV, in comparison to the epileptic animals treated with saline. The levels of pregnenolone and progesterone did not differ significantly in both hippocampus and neocortex. Furthermore, 5α-dihydroprogesterone was significantly reduced in the hippocampus (p = 0.018), but not in the neocortex of LEV-treated rats, compared to saline-treated rats. At variance, allopregnanolone was significantly increased in the neocortex (p = 0.034), but not in the hippocampus of LEV-treated rats. Additionally, pregnanolone was significantly reduced in the neocortex (p ≤ 0.001), but not in the hippocampus, of LEV-treated rats ( Table 1). In summary, in comparison to epileptic animals treated with saline, pregnenolone sulfate was the only analyte reduced both in the hippocampus and neocortex of epileptic animals treated with LEV. At variance, levels of pregnanolone and allopregnanolone were 3α-HSD, 3α-hydroxysteroid dehydrogenase; 3β-HSD, 3β-hydroxysteroid dehydrogenase/Δ 5-4 isomerase; 5α-R, 5α-reductase; 5β-R, 5β-reductase.

Relationship between the Number of SRSs after Treatment and Levels of Hippocampal Neurosteroids
In view of our previous findings on the relationship between allopregnanolone hippocampal levels and seizure occurrence in epileptic rats [18], we also analyzed the linear relationship between the number of SRSs after treatment with LEV (the independent variable) and hippocampal levels of allopregnanolone in LEV-treated rats (the dependent variable). Particularly, we evaluated the possible relationship between the number of SRSs after treatment with LEV (8 weeks after treatment with KA) and levels of various neurosteroids in the hippocampus.

Relationship between the Number of SRSs after Treatment and Levels of Hippocampal Neurosteroids
In view of our previous findings on the relationship between allopregnanolone hippocampal levels and seizure occurrence in epileptic rats [18], we also analyzed the linear relationship between the number of SRSs after treatment with LEV (the independent variable) and hippocampal levels of allopregnanolone in LEV-treated rats (the dependent variable). Particularly, we evaluated the possible relationship between the number of SRSs after treatment with LEV (8 weeks after treatment with KA) and levels of various neurosteroids in the hippocampus.

Relationship between the Power of Delta Band in the Postictal Component of Seizures and Levels of Hippocampal Neurosteroids
As we found a relationship between hippocampal allopregnanolone levels and SRSs, we also evaluated the possible linear relationship between the percentage of delta power of postictal ECoG after treatment with LEV (the independent variable) and levels of various neurosteroids in the hippocampus (the dependent variable). No linear relationship was found between the power of delta in postictal traces and hippocampal levels of pregnenolone sulfate (r 2 = 0.0, p = 0.898) ( Figure 7A), pregnenolone (r 2 = 0.06, p = 0.486) ( Figure  7B), progesterone (r 2 = 0.0, p = 0.885) ( Figure 7C), 5α-dihydroprogesterone (r 2 = 0.14, p = 0.253) ( Figure 7D), or pregnanolone (r 2 = 0.0, p = 0.958) ( Figure 7F). At variance, the power of delta was related to hippocampal levels of allopregnanolone (r 2 = 0.67, p = 0.004) ( Figure  7E).

Relationship between the Power of Delta Band in the Postictal Component of Seizures and Levels of Hippocampal Neurosteroids
As we found a relationship between hippocampal allopregnanolone levels and SRSs, we also evaluated the possible linear relationship between the percentage of delta power of postictal ECoG after treatment with LEV (the independent variable) and levels of various neurosteroids in the hippocampus (the dependent variable). No linear relationship was found between the power of delta in postictal traces and hippocampal levels of pregnenolone sulfate (r 2 = 0.0, p = 0.898) ( Figure 7A), pregnenolone (r 2 = 0.06, p = 0.486) ( Figure 7B), progesterone (r 2 = 0.0, p = 0.885) ( Figure 7C), 5α-dihydroprogesterone (r 2 = 0.14, p = 0.253) ( Figure 7D), or pregnanolone (r 2 = 0.0, p = 0.958) ( Figure 7F). At variance, the power of delta was related to hippocampal levels of allopregnanolone (r 2 = 0.67, p = 0.004) ( Figure 7E). Pharmaceuticals 2021, 14, x FOR PEER REVIEW 10 of 17

Discussion
In the present study, LEV administration resulted in the following major effects: (i) SRSs were efficaciously suppressed in the majority of rats during the week of drug treatment, and the duration of SRSs was reduced in rats partially responding to LEV, being this result significant in both nonconvulsive and convulsive SRSs; ii) we observed a remarkable rebound effect after abrupt LEV withdrawal; iii) the delta power in postictal recordings was linearly related to the seizure occurrence in the previously LEV-treated rats, thus suggesting the involvement of this electrographic component in the rebound effect. Additionally, we found that hippocampal levels of allopregnanolone were related to delta power in postictal recordings of epileptic rats.
A required premise for our experiment was the confirmation of the anticonvulsant effects of LEV. To this aim, we adopted the systemic KA model of chronic epilepsy [17], by introducing substantial differences with respect to most of the published research on testing AEDs in the same animal model. With few exceptions [21], the previous experiments did not address the frequency of nonconvulsive SRSs or the overall seizure duration, but only the frequency of convulsive SRSs was monitored by behavioral analysis [22]. Moreover, topiramate, carbamazepine, and carisbamate, but not LEV were the examined AEDs after the systemic injection of KA [23][24][25]. Particularly, repeated intraperitoneal injections of topiramate (0.3-100 mg/kg) or carisbamate (0.3-30 mg) resulted in a dose-dependent effect on convulsive SRSs, though carisbamate appeared more efficacious than

Discussion
In the present study, LEV administration resulted in the following major effects: (i) SRSs were efficaciously suppressed in the majority of rats during the week of drug treatment, and the duration of SRSs was reduced in rats partially responding to LEV, being this result significant in both nonconvulsive and convulsive SRSs; (ii) we observed a remarkable rebound effect after abrupt LEV withdrawal; (iii) the delta power in postictal recordings was linearly related to the seizure occurrence in the previously LEV-treated rats, thus suggesting the involvement of this electrographic component in the rebound effect. Additionally, we found that hippocampal levels of allopregnanolone were related to delta power in postictal recordings of epileptic rats.
A required premise for our experiment was the confirmation of the anticonvulsant effects of LEV. To this aim, we adopted the systemic KA model of chronic epilepsy [17], by introducing substantial differences with respect to most of the published research on testing AEDs in the same animal model. With few exceptions [21], the previous experiments did not address the frequency of nonconvulsive SRSs or the overall seizure duration, but only the frequency of convulsive SRSs was monitored by behavioral analysis [22]. Moreover, topiramate, carbamazepine, and carisbamate, but not LEV were the examined AEDs after the systemic injection of KA [23][24][25]. Particularly, repeated intraperitoneal injections of topiramate (0.3-100 mg/kg) or carisbamate (0.3-30 mg) resulted in a dose-dependent effect on convulsive SRSs, though carisbamate appeared more efficacious than topiramate in suppressing convulsive SRSs [23,25]. Also intraperitoneal and oral administration of carbamazepine (respectively, 30 and 100 mg/kg) effectively reduced the frequency of motor SRSs [24]. In additional experiments, carbamazepine (30 mg/kg) significantly reduced the frequency of convulsive SRSs without affecting nonconvulsive SRSs recorded in the dentate gyrus. Carbamazepine adequately suppressed the frequency of all SRSs only at 100 mg/kg, while duration remained unchanged also at this dose [21]. Thus, LEV in our animals produced effects comparable to those of the other AEDs.
Similarly, an important achievement of our study was the possibility to reproduce the rebound effect due to rapid AED withdrawal, in order to provide a ground on which to characterize the frequently observed consequences of presurgical evaluation of patients screened for surgical resection of epileptic foci [5]. In clinical practice, AED discontinuation is required during a long EEG monitoring period to enhance the rate of successful diagnostics [7] and to precisely defining the epileptogenic zone [26,27]. However, an important concern is that few guidelines on AED withdrawal during a lengthy stay in EEG monitoring have been published [8], so the outcome might be doubly uncertain. Indeed, this procedure may trigger secondarily generalized tonic-clonic SRSs, SRS clustering, and status epilepticus in patients [9,28]. Different dynamics could be responsible for these consequences. For instance, a previous poor response to drug treatment seemed to induce severe SRSs in patients, after carbamazepine withdrawal [5,29]. In monotherapy, one explanation for the increased seizure frequency above pre-AED values upon discontinuation of treatment could be the development of tolerance to the antiseizure effect of LEV, which was a mechanism described by several groups [15,30,31]. Particularly, tolerance to LEV could be a result of functional adaptation of LEV's target(s) to the presence of the AED, which might result in increased seizure frequency after AED discontinuation. However, in the attempt to define a risk index for the rebound effect after rapid AED withdrawal, our animal model could be useful [17,18,32], since our animals developed an abrupt increase of generalized tonic-clonic SRSs, after LEV discontinuation. Alternatively, other mechanisms could contribute to the rebound effect, in particular the time window in which specific brain areas might be interested by neurogenesis [33]. In this regard, it was demonstrated that kindling with pentylenetetrazole administration for approximately 25 days led to a remarkable increase in seizure susceptibility which was coincident with the timing required for newborn neurons to differentiate and integrate into the hippocampal network [34], a phenomenon which could contribute to the time-dependent increase in SRSs of KA-treated rats in a manner independent of AED administration.
In our study, we found the delta power band spectrum of the postictal ECoG increased in presence of a higher occurrence of generalized SRSs, supporting the view that delta oscillations are important for ictogenesis. In clinical studies, power changes in specific frequency bands demonstrated their potentiality as seizure prediction indicators [35]. LEV, in turn, appeared to be effective in decreasing epileptiform EEG abnormalities. In drugnaïve epilepsy patients, LEV monotherapy was reported to have a region-specific spectral effect on the background EEG activity by decreasing slow frequency power and increasing fast frequency power [36]. In addition to this, the relevance of the immediate postictal EEG was long-debated and became of interest when it was found to be useful to localize the epileptic zone, at least in some type of epilepsies [37,38]. In comparison with the ictal EEG, the postictal EEG is less modified by muscle and movement artifacts, and allows the analysis of a longer period of EEG recording [37]. Interestingly, SRSs were followed by postictal EEG changes in more than 70% of patients affected by temporal lobe epilepsy, and regionally accentuated delta slowing was the most frequent postictal change reported after SRSs [39]. Furthermore, in a cohort study of patients suffering from temporal lobe epilepsy, it was demonstrated that the relative spectral delta power during postictal EEG increased more in secondary generalized seizures than in simple partial seizures, so to appear strictly related to the increased severity of seizures [40]. In patients with a history of uncontrollable generalized tonic-clonic SRSs, and mainly those at greatest risk for sudden unexpected death in epilepsy, it was observed that low delta (0.5-1.5 Hz) and gamma  coupled signals were remarkably increased in the postictal generalized EEG suppression state, as compared to baseline. In general, these changes in neuronal oscillations were consistent in the postictal suppression state, regardless of whether the suppression was generalized or not across channels. Then, it was also suggested that very high frequency oscillations (600-2000 Hz) could be combined with low delta as an additional biomarker for postictal states [41].
We explored neurosteroid levels in the brain as a possible mechanism involved in the rebound effect, since it has been reported that administration of progesterone is followed by a rebound effect with the 2-fold increase in seizure frequency after abrupt interruption of treatment [19]. This suggested that changes in brain neurosteroid levels could be accompanied by corresponding alterations in the seizure occurrence. Indeed, in a previous study [18] we found a significant relationship between the hippocampal levels of allopregnanolone and seizures occurring in chronic epileptic rats, a phenomenon that we interpreted as the activation of an anticonvulsant compensatory mechanism [18]. Here, we confirmed the significant relationship between allopregnanolone levels and seizure occurrence in the hippocampus of LEV-treated rats, but we found an even more remarkable relationship between the power of delta oscillations in postictal traces and hippocampal levels of allopregnanolone. Since delta activity was also related to the seizure frequency in LEV-treated rats, allopregnanolone could have played a role in the rebound effect, which should be further investigated with specific pharmacological approaches aimed at inhibiting allopregnanolone production (i.e., finasteride) [42].
Additionally, we found various changes in neurosteroid levels in both neocortex and hippocampus by comparing saline and LEV-treated rats. Especially, the most consistent alteration was the remarkable reduction in pregnenolone sulfate levels found in both brain regions of LEV-treated rats. However, pregnenolone sulfate is a proconvulsant [42] and its reduction could be hardly related to the observed rebound effect. The anticonvulsant neurosteroid pregnanolone also was dramatically reduced in the neocortex of LEV-treated rats but not in the hippocampus. Moreover, allopregnanolone, which acts similarly to pregnanolone as a positive modulator of inhibitory currents [43], was moderately but significantly increased in the neocortex of LEV-treated rats, maybe playing a compensatory role for the reduction in pregnanolone levels. Overall, these changes in neocortical neurosteroid levels appear to be less relevant to understand the complex dynamics of the rebound effect, for which the hippocampal neurosteroid levels were apparently more critical.

Animals
The study protocol was authorized by the Italian Ministry of Health (323/2015-PR), after approval by the university Animal Welfare Body. All experiments were performed according to the European Directive 2010/63/EU and the consequent Italian act (DM 26/2014). Adult male Sprague-Dawley rats (Charles River, Calco, Italy) were housed in a specific pathogen-free facility with a controlled environment and ad libitum access to water and food. A total of 30 rats, with initial weights of 175-200 g, were used in this study. All efforts were done to refine procedures and protect the animals' welfare.

Experimental Design
Rats were implanted with electrodes, treated with an intraperitoneal injection of KA (15 mg/kg; Sigma-Aldrich, Milan, Italy) to induce a status epilepticus one week after surgery, and given a subcutaneous injection of Ringer's lactate solution (3-5 mL) along with softened rat chow at the end of status epilepticus to minimize discomfort.
After 6 weeks from status epilepticus induction, rats were anesthetized with volatile isoflurane in order to implant subcutaneously a minipump delivering a continuous dosing over one week (2ML1 ALZET, flow rate: 10 µL/h, DURECT Corporation, Cupertino, CA, USA). Rats were randomly divided into two groups: the first group consisted of 15 epileptic rats that had a minipump able to deliver LEV (generously provided by UCB Pharma, Brussels, Belgium) dissolved in saline at 300 mg/kg/day [44]; the second group, control group, consisted of 15 epileptic rats with a minipump delivering saline. The duration of the status epilepticus was considered in the randomization approach to guarantee that important baseline levels did not differ between the LEV group and saline control group. Particularly, status epilepticus lasted for 9.51 ± 29.12 h in the saline-treated group and 10.31 ± 57.56 h in the LEV-treated group (p < 0.602; SALINE vs. LEV), respectively. In all animals, SRSs were continuously monitored. Specifically, a blind analysis of all video-ECoG recordings belonging to saline-and LEV-treated groups were analyzed by three different expert raters. The postictal events were then collected and evaluated by a blind power spectrum analysis. The levetiracetam withdrawal was defined as "rapid" because we completely discontinued the levetiracetam without a gradual reduction in dosage.

v-ECoG Recordings
As described previously [45], rats were implanted with epidural electrodes in order to continuously record ECoG data from frontal and occipital cortices. More precisely, guiding holes were drilled and epidural electrodes (stainless steel Ø = 1 mm; PlasticsOne, Roanoke, VA, USA) were implanted in frontal (bregma 0 mm, 3.5 mm lateral from midline) and occipital cortices (bregma −6.5 mm, 3.5 mm lateral from midline) of both hemispheres. One electrode was implanted below the lambda on the midline in all rats and used as a reference. Electrodes were connected through steel wire to terminal gold pins (Bilaney Consultant GmbH, Düsseldorf, Germany) inserted in a plastic pedestal (PlasticsOne) cemented on heads. The four recording electrodes were used for video-ECoG monitoring, whereas only frontal electrodes were used for power band analysis. Volatile isoflurane was used during electrode implantation to induce a deep anesthesia, assessed by deep breath, loss of tail and eye reflexes. Gel containing 2.5 g lidocaine chloride, 0.5 g neomycin sulfate and 0.025 g fluocinolone acetonide (Neuflan ® gel; Molteni Farmaceutici, Scandicci, Italy) was applied at the end of the surgery to reduce acute pain and risk of infection. All rats were monitored until complete recovery from anesthesia and housed in single cages with no grids or environmental enrichments to avoid risk of headset loss. The EcoG was recorded via cable connection between headset and preamplifiers. Electrical activity was digitally filtered (0.3 Hz high-pass, 500 Hz low-pass), acquired at 1 kHz per channel, and stored on a personal computer after the mathematical subtraction of traces of recording electrodes from trace of reference electrode, using a PowerLab8/30 amplifier connected to 4 BioAmp preamplifiers (AD Instruments; Dunedin, Otago, New Zealand). Videos were digitally captured through a camera connected to the computer and synchronized to the ECoG traces through LabChart 8 PRO internal trigger.

Behavioral and ECoG Analysis
ECoG traces were digitally filtered offline (band-pass: high 50 Hz, low 1 Hz) and manually analyzed using LabChart 8 PRO software (AD Instruments) by expert raters. Status epilepticus was defined as the period of time in which rats either did not recover normal behavior between a seizure and the other, or in which they displayed continuous shaking for more than 5 min. The end of status epilepticus was characterized by a progressive reduction in frequency of the continuous electrographic spikes, preceding a silent period. Moreover, the termination of status epilepticus was accompanied by recovery of normal behavior. In our animal model, status epilepticus was allowed to self-terminate [17].
Seizures were defined as ECoG segments with a minimum duration of 10 s, continuous synchronous high-frequency activity, and an amplitude of at least twice the previous baseline [17]. They were also screened for the presence of a postictal depression, that was the period of time in which the brain recovered from the seizure before regaining normal function. Seizures and their durations were determined in the ECoG traces, and then investigated for related behavior [46] by using the synchronized video recordings. In particular, all seizures were scored as stage 0 (or subclinical) if a clear epileptiform ECoG signal was observed without corresponding evident behavior in the video; stage 1-2 in