Immediate versus late effects of vigabatrin on spike and wave discharges

Vigabatrin increases GABA concentrations by inhibiting GABA transaminase. In previous studies, it was shown that vigabatrin increases the incidence of Spike and Wave Discharges (SWD) in the WAG/Rij rat model for absence epilepsy. Since following a single dose of vigabatrin GABA concentrations are known to be increased for several days, the present study sheds light on how the previously described changes in SWD characteristics develop over a longer time frame. To achieve this, we injected adult WAG/Rij rats with 500 mg/kg and recorded their EEG for 48 h. SWD were quantified, and their peak frequencies were calculated. Our results showed three rapid onset effects: a sharp increase in SWD incidence, from 12.5 /hour to 133/hour), this increase lasted only 4.4 h, an increase in mean SWD duration, from 4.6 s to 8.1 s and a drop in peak frequency, from 8 to 6 Hz. Since it takes several hours before GABA concentrations are sufficiently increased, we propose that these immediate effects are caused by direct stimulation of both GABAA and GABAB receptors by the molecule vigabatrin. Next, the mean SWD duration decreased below baseline values after 4.4 h. Hazard rate analysis showed that this is caused by an increased probability of short SWD. We argue that these changes are caused by increased activation of both GABAA and GABAB receptors in the frontal cortex and the thalamus, and more specifically, in the Reticular Thalamic Nucleus (RTN). After approximately 34 h, the probability of short SWD returned to normal. This suggests the occurrence of downregulation of GABA receptors. The decrease in peak frequency was still present 48 h after injection. It has been argued that the balance between GABAA and GABAB receptor-mediated activity in the RTN is crucial for controlling this SWD characteristic. It can be concluded that a single dose of vigabatrin results in remarkable and opposite effects over time: an initial, proabsence effect is followed by an antiabsence effect.


Introduction
Absence seizures are stereotyped generalized seizures (Berg et al., 2010). In the electroencephalogram (EEG) of humans with absence epilepsy, bilaterally synchronized and generalized spike and wave discharges (SWD) are observed. They are usually accompanied by brief decreases in awareness and responsiveness. Absence epilepsy is associated with severe health risks, including other forms of epilepsy (Nabbout et al., 2017) and cognitive deficits (Cheng et al., 2017).
Gamma-aminobutyric acid (GABA) is recognized to be involved in the occurrence of SWD (Crunelli and Leresche, 2002;Gorji et al., 2011). Indeed, vigabatrin, a drug that increases GABA concentrations by irreversibly inhibiting GABA transaminase (Hammond and Wilder, reticular thalamic nucleus (RTN) is known to be associated with SWD termination (Luttjohann et al., 2014;Luttjohann and van Luijtelaar, 2015;Schofield et al., 2009). Other in vivo data suggest a prominent role for GABA in the thalamus in SWD characteristics as well: the GABA reuptake inhibitor tiagabine increases SWD incidence in WAG/Rij rats when injected into the ventral posterior medial nucleus of the thalamus (D'Amore et al., 2015).
As for the GABA receptors involved: GABA A receptors have been described to play an important role in the occurrence of SWD. In the GAERS model, as well as in other absence seizure models, GABA A receptor-dependent 'tonic' inhibition is increased in thalamocortical neurons (Cope et al., 2009). Additionally, WAG/Rij rats have a specific loss of alpha3 subunit immunoreactivity at inhibitory, GABA A receptors in the RTN (Liu et al., 2007). However, there is also evidence for the involvement of GABA B receptors: in WAG/Rij rats, a decreased expression and sensitivity of presynaptic GABA B receptors in the neocortex to the GABA B receptor agonist baclofen was found (Inaba et al., 2009;Merlo et al., 2007).
The acute effects of vigabatrin on SWD were described over a period of 6 h (Bouwman et al., 2007b). However, even though vigabatrin itself has a half-life of 2.2-3.3 hours in rat CSF (Tong et al., 2007), the drug has been shown to increase GABA levels in humans for more than a week after only a single dose (Ben-Menachem, 2011), and up to 3 days in rats (Tong et al., 2007). Since the alterations in SWD characteristics are believed to be caused by changes in GABA-ergic neurotransmission rather than by the molecule vigabatrin itself, we hypothesize that the SWD in rats remain changed for a much longer time than the 6 h reported previously.
In the present study, a single dose of vigabatrin was injected in adult WAG/Rij rats. Cortical EEG recordings were made over 48 h. We studied the time dependency of vigabatrin on four variables: the incidence of SWD, their duration, their hazard rates, and their peak frequency.

Animals
This study was performed in accordance with European Community guidelines for the use of experimental animals. Approval of the local committee for animal studies was obtained (KUN-DEC 2001-34).
Sixteen male WAG/Rij rats, aged 12 months, were used for this experiment. The animals had a mean body weight of 337 g SD 16.5. Animals were housed in Makrolon IV cages with food and water available ad libitum, in a temperature-controlled room (20−21°C). The animals were subjected to a 12 h-12 h light dark cycle, with lights on at 8:00 p.m. Two animals were housed per cage until surgery.

Surgery
At 11 months of age the animals were surgically provided with a single tripolar EEG electrode set (Plastics One MS-332/2-A) under complete isoflurane anaesthesia. Electrodes were placed over frontal (in mm from bregma; AP + 2.0; L -3.5) and parietal (AP -6.0; L -4.0) cortex (Paxinos and Watson, 1998). A ground electrode was placed over the cerebellum on the contralateral side. The electrodes were fixated to the screws drilled into the skull with dental acrylic.
The analgesic Temgesic mix was injected systemically after surgery, in a dose of 0.05 mL/kg of 0.1 mg/mL. Animals were housed individually afterwards and were allowed to recover for at least two weeks.

Drug and treatment
The animals were randomly assigned to the treatment (Viga, n = 8) or the control group (Control, n = 8). In the Viga group, 500 mg/kg vigabatrin (Yamanouchi Pharma B.V., the Netherlands) was injected intraperitoneally. The drug was dissolved as 500 mg/12.5 mL water in order to avoid pain induced by excessively high osmolality. The control group was injected with 12.5 mL/kg of saline, also intraperitoneally. The injections took place in the second hour of the dark period.

EEG recording
Animals were placed in the recording cages 60 min before receiving the injection. Differential EEG recordings were made during this period and were used as baseline.
Subsequently the animals were injected with either vigabatrin or saline. EEG signals were recorded, amplified, filtered between 1 and 100 Hz and digitized at a sample rate of 512 Hz. A Windaq system (DATAQ Instruments, Akron, OH) was used for EEG data acquisition, monitoring, and storage on hard disk for offline analysis. Post-injection EEG recordings were made during 48 h, starting around 10:00 a.m.

EEG analysis
2.5.1. SWD incidence SWD were located and marked in the EEG using the SWComplex finder software (Electronic Research Group, Dr. P. van den Broek, Radboud University, Nijmegen), and subsequently verified by a trained expert. Mean SWD incidences per hour were plotted as cumulative points, for each group, over the course of the entire recording period. Two functions were subsequently fitted to these data using Graphpad Prism 6.0 for Windows (GraphPad Software, San Diego, California USA). A function to describe a linear increase in cumulative SWD incidence in the control group, Y = slope*X, and a function that adds to this linear function an additional increase according to a logistic (S) function in the treatment group: in which X is time, in hours, after injection of the drug, Y is the cumulative number of SWD at time X; slope is the linear increase of SWD (cumulative incidence); δ is the additional increase (total number) and T50 % is the sigmoid midpoint, the time point at which this extra increase reaches half of its value. Hill is a measure for the steepness of the S-curve.
The fits of the control and the treatment groups were used for further analysis. The same two types of fits were also used to describe the cumulative SWD incidence over time for all individual animals (see Appendix A, Table A1. for individual best fit values).

Incidence-based epochs
The recording period was divided into 5 epochs; the rationale was the outcomes of the fit of the SWD incidence in the treatment group, and the light-dark cycle. The five epochs were: 1) post-injection (p.i.; from the moment of injection to 10 % of the additional increase in SWD incidence, 0−0.75 h), 2) main SWD increase period (10-90 % of the additional increase, 0.75-4.4 h), 3) from 90 % of the additional increase (p.i. hour 4.4) to the end of light period1 (hour 22), 4) dark period2 (p.i. hour 22 to p.i. hour 34), 5) light period2 + the first 2 h of dark period 3 (p.i. hour 34 to p.i. hour 48). Epochs 1 and 2 took place in the first dark period, to allow precise monitoring of the early effects. SWD incidence and Mean SWD duration were subsequently calculated per epoch.

SWD peak frequencies
SWD peak frequencies were quantified as means of 10 SWD per individual animal in each of the aforementioned five epochs. The SWD were selected randomly using randomised sequences generated by www.random.org. Peak frequency was quantified in the selected SWD using the 'local extreme information' option in the SWComplex finder software.

Hazard rate
In order to examine the stopping mechanism of SWD in vigabatrintreated animals, hazard rate analyses (Maris et al., 2006) of SWD durations were made. The hazard function can be used to provide insight in the way mechanisms underlying brain state switches work. In this experiment the stopping of a seizure, which is a switch from the seizure state to normal state, was examined. This was done for each rat individually. In order to allow averaging of hazard rates over groups, bin lengths, based on SWD duration, were defined.

Statistical analysis
All statistical analyses were performed in Graphpad Prism 6.0 for Windows (GraphPad Software, San Diego, California USA) and IBM SPSS 21.0 for Windows (SPSS Inc., Chicago, Illinois USA).
Best fit values, df and R 2 of the group average are reported in the Results section. Best fit values, df and R 2 of the individual animals are reported in Table A1 (Appendix A). The best fit values of the slopes of linear increase of the individual animals of the treatment and the control groups were compared using an unpaired t-test (Appendix A) and they turned out not to be different.
The SWD incidence, their mean duration and peak frequencies of SWDs were analyzed per epoch, and compared with separate General Linear Model (GLM) Repeated Measures analyses with treatment (drug vs solvent) as between-subjects factor and epoch as within-subjects factor.
The hazard rate curves were analyzed per epoch. First, the curves were corrected for group differences, as a comparison of the absolute yvalues of the hazard rates is not a relevant variable. Rather, the question is whether the hazard rate curves are parallel. Subsequently, treatment-bin interactions were tested by means of a GLM analysis with treatment and bin as between-subjects factors. Within-subjects analysis was not possible due to the high individual variability of the hazard rates, resulting in some missing values, due to some animals having zero SWD in certain epochs.

SWD incidence
Cumulative mean SWD incidence can be seen in Fig. 1. A linear function was the best fit to the means per hour of the data of the controls. Best fit was Y = 16.8 X (goodness of fit: df = 48, R 2 = 0.98). Thus, the controls showed a mean SWD incidence of 16.8 (95 %CI: 16.5-17.1) per hour.
A nonlinear function was the best fit to the means per hour of the data of the vigabatrin group. Best fit was: Y = 18.1*X + (399)/ (1 + 10^((2.2-X)*0.6)), (goodness of fit: df = 45, R 2 = 0.99). The linear increase was 18.1 (95 %CI: 14.6-21.7) per hour on which a nonlinear increase was present. The total additional increase, was 399 SWD (95 %CI 285-494), which reached half of its maximum value after 2.2 h (95 %CI: 1.9-2.6 h). After 2 x this period, over 90 % of the additional increase was reached, and SWD incidence might be assumed to have returned to baseline values.
The fits for all individual animals are displayed in Fig. A1 (Appendix A), best fit values are given in Table A1 (Appendix A). All controls were fitted with the linear function. All animals from the treatment group were fitted with the nonlinear function.

Circadian rhythmicity
Even though a linear function was successfully fitted to the data of the cumulative SWD incidence of controls, some fluctuations in SWD incidence were observed ( Fig. 1). Similar fluctuations were observed in the treatment group, These fluctuations coincide with the light-dark cycle and follow a circadian rhythm.

SWD duration
Mean SWD duration can be seen in Fig. 2b. It varied between 4.4 and 6.4 s for controls throughout the recording period, which is in the normal range .The range of mean SWD durations was larger for animals treated with vigabatrin. A significant epochtreatment interaction was found in the GLM, F(5, 83) = 9.48, p < 0.0001. Multiple comparisons tests showed significant differences: mean SWD duration was 8.1 s SEM 1.3 for vigabatrin-treated animals, compared to 4.6 s SEM 0.22 for controls (p = 0.002) in epochs 1 and 2. In epochs 3 and 4, SWD durations were significantly lower for vigabatrin-treated animals, 4.0 s SEM 0.43, compared to controls, 5.8 s SEM 0.38 (p = 0.005).

SWD morphology
Before vigabatrin was injected, both groups of animals showed SWD Fig. 1. Cumulative mean SWD incidence for controls and vigabatrin-treated animals. The dotted lines mark the transitions of the 5 epochs. In epochs 1 and 2, the vigabatrin group (red line) showed an additional increase in SWD incidence compared to controls. A sigmoid curve, superposed on a linear function, was fitted to this increase. After epoch 2, at 4.4 h, SWD incidence was returned to values that were no longer different from those of the control animals. Controls (blue line) showed a constant SWD incidence over time, with some slight variation due to circadian rhythm (Smyk et al., 2011;Van Luijtelaar and Coenen, 1988). SWD incidence was at its lowest at the onset of the light period (daily minimum SWD incidence is marked with arrows) and gradually increased, reaching a peak in the early hours of the dark period (areas marked in light blue).
that are representative for the WAG/Rij model , as shown in Fig. 3a. Approximately 1 h after injection of vigabatrin (Fig. 3b), the morphology of the SWD started to change: they were, in average, longer, with a slight decrease in peak frequency. Also noteworthy is the slight drop in and increased variability of the amplitude of the spike component.
Approximately 2 h after injection (Fig. 3c), SWD had a markedly lower peak frequency and amplitude. Mean SWD duration was somewhat longer than in controls.
Approximately 24 h after injection (Fig. 3d), a large number of short SWD can be found in the EEG. Peak frequency is still decreased, as is the amplitude, making many of the SWD much more difficult to detect against the background EEG.

Peak frequency
There was no difference in mean peak frequency between controls, 8.0 Hz SEM 0.20, and vigabatrin-treated animals, 8.0 Hz SEM 0.14, during the baseline period. The SWD remained very similar over time In the control group, with an average peak frequency of 8 Hz SEM 0.14. In the vigabatrin group, SWD characteristics changed, with the peak frequency dropping significantly, reaching a minimum of 5.9 Hz SEM 0.17 around 3.5 h after injection. The peak frequency then gradually increased to 7.2 Hz SEM 0.23 after 48 h, but remained significantly lower, as illustrated in Fig. 2c.
A significant treatment effect was found, F(1, 14) = 46.5, p < 0.0001, next to an epoch-treatment interaction, F(5, 70) = 5.11, p = 0.0005. Post hoc tests showed that in all 5 epochs the peak frequency was significantly lower for vigabatrin-treated rats, with p-values ranging between p < 0.0001 and p = 0.008.

Hazard rates
Hazard rate analysis (Fig. 4) revealed no difference between the groups during baseline recordings, and no treatment-bin interaction was found, F (9, 98) = 0.23, p = 0.99. The hazard rates curves of vigabatrin-treated animals differed significantly from controls in all subsequent epochs. In epoch 1 a treatment-bin interaction was found, F(11, 115) = 2.38, p = 0.01: the hazard rate curve for the vigabatrin-treated rats was flatter and lower compared to the controls, and had a longer tail, meaning that vigabatrin-treated animals showed some long SWD. In epoch 2, the hazard rate curves were parallel again, no significant treatment or interaction effects were observed: F(10, 119) = 0.30, p = 0.98.
Treatment-bin interactions were found in all epochs, in the bins of durations of 3.5-16.8 s, all p < 0.0001. In all these curves the hazard rates were higher than controls for shorter durations, but lower than controls for higher durations, meaning that there is a higher probability of very short SWD. The point at which the curves for vigabatrin-treated animals and controls intersect gradually moves towards shorter SWD durations, from an SWD duration of 8 s for epoch 4, to 4 s for epoch 5.
Moreover, since the tails of the hazard rate curves of all epochs of the vigabatrin-treated animals are longer than that of the controls (40-50 seconds, compared to 16-30 seconds for controls), there is also a higher probability of some long SWD.

Discussion
There are three main findings in this study, as a consequence of a single injection of vigabatrin, all with a different duration of occurrence. Firstly, an immediate increase in SWD incidence was found, lasting 4.4 h. Secondly, mean SWD duration was increased as well in the same period. However, while SWD incidence returned to baseline values after 4.4 h, mean SWD duration decreased to values well below baselines, and gradually normalized until approximately 34 h after injection. The hazard rates demonstrated that, while initially the likelihood of short SWD decreases and long ones appear after injection of vigabatrin, later the likelihood of short SWD is increased, leading to a markedly different distribution of SWD lengths. The likelihood of long SWD is stable in all epochs.
Thirdly, the treatment resulted in a change in the morphology of SWD, quantified by a lower SWD peak frequency. Again, this effect was time-dependent: the peak frequency dropped, quickly, from 8 to approximately 6 Hz following injection. Subsequently, the peak frequency remained low and only slowly increased, and after 48 h it was still significantly lower than in controls. Earlier studies did not acknowledge Baselines (BL) are not significantly different in any of the panels. Panel a shows the SWD incidence per hour. A significant epoch-treatment interaction was found. Multiple comparisons tests showed that SWD incidence being higher in vigabatrin-treated rats than in controls during epochs 1 and 2. Panel b shows the mean SWD duration. Overall, an epoch-treatment interaction was found. For vigabatrin-treated animals, multiple comparisons test showed that mean SWD duration was significantly increased during epochs 1 and 2. During epochs 3 and 4, SWD duration was decreased in the vigabatrin group. Panel c shows the mean peak frequencies of SWD (and SEM) over time. The peak frequencies remained unchanged in control animals over time. In the treatment group, SWD peak frequencies dropped to 5.9 Hz in the first hours after injection with vigabatrin. After this, they remained significantly lower throughout the remainder of the recording period, although a tendency towards baseline levels was noticed.
this long-lasting, large effect on SWD morphology.
The three effects have markedly different time lines, which suggests that the underlying mechanisms of these effects are different as well. Because the EEG was recorded for only 6 h in earlier studies (Bouwman et al., 2007b(Bouwman et al., , 2007a(Bouwman et al., , 2005Liu et al., 1990;Maris et al., 2006), the time-dependency of these effects had remained undetected so far. We propose that the immediate effects on SWD incidence, mean SWD duration, and SWD peak frequency can be explained by a direct effect of the molecule vigabatrin on both GABA A and GABA B receptors, whereas the later effects are the consequence of increased concentrations of GABA in cortex and thalamus.

Immediate effects of the molecule vigabatrin
Vigabatrin can be detected in rat CSF within one hour after an i.p. injection, and the molecule has a half-life of 2.2-3.3 hours (Tong et al.,  The curves of the baselines (BL) were not significantly different. In epoch 1 the distributions show an overall increase in SWD duration. The hazard rate curve for vigabatrin-treated rats is flatter than that of the controls, meaning that there is a lower probability of stopping at a certain length point, leading to longer SWD. In epoch 2 the distribution returns to baseline levels. Also, the hazard rate curve of the Viga group is parallel to that of controls, with the exception of a longer tail for vigabatrin-treated rats, meaning that the Viga animals have some long SWD. Between epochs 3-4, the distributions and hazard rates for the vigabatrin group show very different patterns compared to previous epochs: the distributions show a larger number of short SWD. However, the long right-side tail of the histogram shows that there are also some very long SWD. The steep start of the hazard rate curve means a high probability of short SWD, and the long tail means that these animals still have long SWD. In epoch 5, the distribution of the vigabatrin group seems to be comparable to that of controls. However, the hazard rate curve suggests that there is still an increased likelihood of short SWD. Also, the long tail remains. 2007). After injection, GABA transaminase activity decreases rapidly, reaching a minimum within 3-4 hours (Hammond and Wilder, 1985). GABA concentrations in the rat brain are at its maximum level around 4 h post injection (Hammond and Wilder, 1985;Valdizan et al., 1999) and stay elevated for up to 3 days (Tong et al., 2007). It is therefore very unlikely that all early onset effects are caused by alterations in GABA levels. Rather, it seems to be a direct and immediate effect of the molecule vigabatrin on GABA receptors.
The question is which of the GABA receptors are involved in the early onset effects. Stimulating the GABA A receptor with other drugs seems to have similar effects on SWD: intracerebroventricular injections of the GABA mimetic muscimol are known to cause an increase SWD incidence in WAG/Rij rats, which can be counteracted with the GABA A antagonist bicuculline (Peeters et al., 1989). Similarly, the inhibitory neurosteroid Org-20599, which acts as a positive allosteric modulator of the GABA A receptor, when injected systemically, causes increases in both SWD incidence and duration (Bouwman, 2006).
Systemic injection of the GABA B agonist R-Baclofen has similar effects on SWD as described in the present study: increases in incidence and mean duration have been observed in WAG/Rij (Bouwman, 2006) rats and GAERS (Liu et al., 1992). In line, the GABA B antagonist CGP35348 is known to reduce SWD activity in rat models for absence seizures (Chan et al., 2006;Cortez and Snead, 2006;Puigcerver et al., 1996;Smith and Fisher, 1996).
The immediate effects in the present study could therefore be the consequence of interaction with GABA A receptors or with GABA B receptors, or with both. It is not known whether vigabatrin can interact directly with the GABA A receptor complex, although our results suggest it can, albeit with a low binding affinity (results are presented in appendix B). The high concentration of vigabatrin that was administered in the present study, which is commonly used preclinically, is likely to be sufficient to trigger this. In contrast, there is only very limited evidence for direct effects of vigabatrin on the GABA B receptor: a case report describes that vigabatrin was used successfully against focal seizures in a patient with a partial GABA B receptor subunit deletion (Chin et al., 2019). However, the decreased SWD peak frequency suggests that both GABA A and GABA B receptors are involved. It has been argued that the balance between GABA A and GABA B is crucial for controlling SWD characteristics; it is suggested that the thalamocortical circuit can generate two types of spike-and-wave oscillations: 2-4 Hz for strong GABA B conductance, and 5-10 Hz when GABA A conductance is dominant. Their frequency is determined by the receptor type that mediates inhibition in thalamic relay cells, with increased relative GABA B concentrations resulting in a decreased peak frequency Destexhe, 1999).
In vivo and in vitro data seem to confirm this. In GAERS, it has been shown that GABA B receptors in the ventrolateral thalamus and RTN are responsible for the production of oscillatory activity underlying SWD and for de novo oscillations (Liu et al., 1992). In tissue slices of the ferret geniculate nucleus, activation of GABA A receptors leads to oscillatory activity of around 10 Hz. However, concomitant GABA B receptor activation leads to activity of 3−4 Hz (Bal et al., 1995;von Krosigk et al., 1993). The frequency of the latter type of oscillatory activity is similar to human spike-wave discharges, but may very well be of a different nature, such as delta waves representing slow wave sleep. Nevertheless, these data confirm that the frequency of oscillatory activity depends on both GABA A -and GABA B -mediated inhibition. The decrease in SWD peak frequency following injection of vigabatrin may therefore be the result of increases of GABAergic activity, and more specifically, mainly of an increase in GABA B -ergic, and to a lesser extent in GABA A -ergic activity.
All in all, it seems likely that the rapid effects observed in the present study can be explained if we assume a direct stimulation of both GABA A and GABA B receptors by vigabatrin. However, while direct effect of vigabatrin on GABA A receptors are possibly present (Appendix B), its effects on GABA B receptors are less clear. Other, off-target effects of vigabatrin could be present.

Late effects on SWD duration
To our knowledge, the late decrease of mean SWD duration between approximately 4.4 h and 34 h after injection, has not been reported before. However, a biphasic effect of a single dose of vigabatrin on hippocampal epileptic afterdischarges has been described in vivo: an early (4 h) proconvulsant effect was followed by a late 24−48 h anticonvulsant effect (Stuchlik et al., 2001).
Analysis of the hazard rates provides insight in how SWD stopping probability has changed over time. Initially, the hazard rates, i.e. the probability of an SWD stopping, given that it has not stopped yet, are decreased compared to controls. This is in line with the initial increase in mean SWD duration, since a lower hazard rate implies that an SWD is less likely to be terminated, resulting in long SWD. However, from epoch 3 onwards, there is a greater likelihood of short SWD. At the same time, however, the long tails of the hazard curves show a high probability for long SWD as well. In this period, it turned out that the influence of short SWD outweighed the effect of the long SWD on the mean SWD duration.
These results together show that treatment with vigabatrin did not so much alter overall SWD lengths, but rather it affected the distribution of SWD durations. We have previously demonstrated that vigabatrin affects the SWD stopping mechanism (Bouwman, 2006;Bouwman et al., 2007a;Maris et al., 2006). In WAG/Rij rats, the likelihood of observing both very short and very long SWD is higher in vigabatrintreated animals than in controls (Maris et al., 2006). However, this does not explain the decrease in mean SWD duration after 4.4 h lasting approximately 30 h, since the duration of this effect is far beyond the halflife of the molecule vigabatrin (Tong et al., 2007).This suggests that other factors, rather than the molecule vigabatrin itself, are the cause of these late effects. Considering the steady increase of GABA caused by vigabatrin, we assume that the late effects are caused by GABA.
In order to explain how elevated GABA levels contribute to the late effects, brain region should be taken into account, since regional differences in the response to the GABA-mimetic tiagabine have been reported. Injection of the drug in the ventral basal thalamus led to an increase in SWD, whereas injection into the somatosensory cortex (D'Amore et al., 2015) and the hippocampus (Tolmacheva and van Luijtelaar, 2007) reduced SWD incidence.
Vigabatrin distribution is also known to be region-specific: frontal cortex concentrations of vigabatrin have been reported to be substantially higher than hippocampal concentrations (Tong et al., 2009). This may in turn lead to increased GABA concentrations in the cortex.
The vigabatrin concentrations in the thalamus, however, are not known. Given the rapid effects on mean SWD duration, and assuming that the thalamus is crucial for the duration of the discharges (Schofield et al., 2009), it seems likely that vigabatrin concentrations in the thalamus are high as well. It has been proposed that RTN is involved in SWD termination. More specifically, both cortico-thalamic and intrathalamic coupling changes are associated with SWD termination and include the rostral part of the RTN (Luttjohann et al., 2014;Luttjohann and van Luijtelaar, 2015). Moreover, injections of vigabatrin directly into the thalamus are known to increase total SWD duration in GAERS (Liu et al., 1991). Therefore it is safe to assume that the increase in duration of SWD is due to elevated concentrations of GABA in the thalamus.
Again, the question arises which type of GABA receptor is involved. Short SWD have been observed in rats systemically treated with GABA A agonists diazepam (Bouwman et al., 2004) and loreclezole (Ates et al., 1992). WAG/Rij rats have a specific loss of alpha3 subunit immunoreactivity at inhibitory, GABA A -ergic synapses in the RTN (Liu et al., 2007). Whether this property plays a role in the effects of GABA on the duration of SWD remains to be established. However, GABA A -mediated inhibition in the RTN is known to be important in controlling thalamic excitability (Huguenard and Prince, 1994).
However, GABA B may be involved as well: GABA B activation is known to suppress burst firing in the RTN (Cain et al., 2017). Therefore, a vigabatrin-induced increase in GABA B activation may lead to shorter SWD as well. Moreover, the fact that SWD peak frequency remains lowered, suggests that GABA B -ergic activity is still increased because, as mentioned before, GABA B receptors in the ventrolateral thalamus and RTN are responsible for the production of oscillatory activity underlying SWD (Liu et al., 1992).
Approximately 34 h after injection, the mean SWD duration returned to baseline levels. Since GABA concentrations are still known to be elevated by then (Tong et al., 2007), it is possible that adaptation of GABA receptors occurs. In a rat model for temporal lobe epilepsy, GABA A receptors were demonstrated to lose their sensitivity to diazepam in 24-48 h (Leroy et al., 2004). However, since the peak frequency remains decreased even after 48 h, GABA B receptors might desensitize more slowly.

Additional finding -circadian rhythmicity
An additional finding was the observation of circadian fluctuations in the SWD incidence. Circadian rhythmicity in SWD incidence has been described in the WAG/Rij model (Smyk et al., 2011;van Luijtelaar and Coenen, 1988) as well as in humans (Halasz et al., 2002). Therefore, this additional finding was not unexpected. In WAG/Rij rats, SWD are most common at the beginning of the dark phase. The minimum number of SWD is reached during the first 2 h of the light phase (Mamalyga, 2014;Smyk et al., 2011;van Luijtelaar and Coenen, 1988). Our data correspond well with this pattern.

Conclusions
In summary, our results show three different time lines of changes in SWD characteristics following a single dose of vigabatrin. SWD peak frequencies quickly decrease and remain lowered long after vigabatrin itself has been eliminated, supporting the idea that enhancement of GABAergic neurotransmission at the level of the thalamus affects SWD characteristics (Bouwman et al., 2007a). However, the relatively short duration of the increase in SWD incidence suggests another underlying mechanism, which is possibly a direct effect of vigabatrin on GABA A and GABA B receptors. Moreover, rapid changes in the distribution of SWD durations follow a similar time line as the effects on SWD incidence, but subsequently undergo another change in distribution after several hours. This points towards a crucial role of GABAergic neurotransmission in the thalamus.
The balance between concentrations of vigabatrin, GABA A , and GABA B receptor activation seems critical to explain these time lines. There is some, albeit limited, evidence that vigabatrin directly stimulates GABA A and GABA B receptors, which would explain the immediate effects on SWD, all of which occur well before GABA concentrations have accumulated after injection of vigabatrin. Later, when GABA concentrations have risen, stimulation of both GABA A and GABA B receptors in the frontal cortex and the thalamus induce different changes of SWD duration. This is in line with previous findings of the involvement of both GABA A (Huguenard and Prince, 1994;Liu et al., 2007) and GABA B (Inaba et al., 2009;Merlo et al., 2007) receptors in the pathophysiology of SWD.
The present study offers some insight in the direct and the indirect effects of vigabatrin on absence seizures. We hypothesize that GABA A and GABA B receptors are critical in explaining the differential effects on SWD over time. However, the exact role of both receptors remains to be elucidated. Other, off-target effects of vigabatrin cannot be fully excluded. Studies in which vigabatrin will be combined with various specific GABA A and GABA B receptor agonists and antagonists could be used to further unravel the underlying mechanism.
Introduction It is assumed that the anti-epileptic activity of vigabatrin is mainly due to an irreversible inhibition of the γ-aminobutyric acid transaminase (GABA-T) activity, thus to an increase of the availability of the inhibitory neurotransmitter GABA (γ-aminobutyric acid) for its receptor sites (Hammond and Wilder, 1985). The aim of the present study was to investigate whether, besides this GABA-T inhibition, vigabatrin also directly interacts with the GABA A receptor complex and if so how the interaction with the complex should be interpreted, in terms of the agonist-antagonist spectrum.
Methods Receptor-binding assays were conducted, using rat brain membranes. [ 3 H]TBOB ([ 3 H]-t-butyl-bicyclo-ortho-benzoate) was used as a tracer ligand for the GABA A receptor complex (Lawrence et al., 1985). The method used in the present experiments is described in van Rijn et al. (1990).
On the binding of [ 3 H]TBOB we determined the dose-effect curves of GABA, of vigabatrin and of Org 20,549; of each substance alone and of several combinations of them. Org 20,549, a water-soluble neuro-active steroid, is a positive allosteric modulator of the GABA A receptor complex (Van Rijn et al., 1999). The interactions between the three substances was described qualitatively using isobolic analysis.
Results and discussion All parameter estimates are given in Isobolic analysis (see Fig. B1.) revealed that in displacing [ 3 H]TBOB the effect of the combination vigabatrin and GABA is not different from additivity. This outcome is compatible with the suggestion that these two compounds compete for the same binding site on the GABA A receptor complex, although the affinity of vigabatrin for this site is lower than that of GABA itself.
Earlier we showed that according to the isobole, GABA and Org act in synergy (Van Rijn et al., 1999). Now we show that the isobole of vigabatrin and Org 20,549 also shows a synergistic interaction. Therefore, we conclude that vigabatrin not only might interact with the GABA site of the GABA A receptor complex as GABA does, but that the positive allosteric interaction between binding of simultaneously GABA and the neurosteroid also holds for the combination vigabatrin and the neurosteroid. This conclusion points to vigabatrin being a low affinity competitive agonist of GABA.  A three-parameter-sigmoid-E max equation was fitted to the experimental data and the B max , EC 50 and slope parameters were determined. The parameter estimates and the 95 % confidence intervals (CI) of the fits are given. The number of experiments in triplicate is indicated by n.