Gad1 knockout rats were generated using CRISPR/Cas9-mediated gene editing (29) in the Long-Evans background. Knockout rats were crossed with Long-Evans rats (Japan SLC, Inc., Japan) and bred in the Gad1 heterozygous (+/–) state in our laboratory. Gad1 (–/–) and Gad1 wild-type (+/+) rats were obtained from sibling mating between heterozygotes. All rats were reared under standard laboratory conditions under a 12-h/12-h light/dark cycle. All experimental procedures were approved by the Animal Care and Experimentation Committee of Gunma University and the Tohoku University Committee for Animal Research.

Implant surgery and EEG recording were performed as described previously (25). Briefly, rats were anesthetized with medetomidine, midazolam, and butorphanol mixture and placed on the stereotaxic apparatus. Heart rates and SpO₂ were continuously monitored. The scalp was excised, and the parietal skull bone was exposed. Small burr holes (~0.9 mm diameter) were made bilaterally over the skull, and twisted-pair wire electrodes (114 μm diameter, TFA-coated silver wire, Med Wire, USA) and/or silver-pleated screw electrodes (M1 × 3 mm, Unique Medical, Japan) were aseptically implanted. The tips of the twisted-pair wire electrodes were inserted into the cortex at a depth of 1 mm and firmly supported through a ceramic guide tube (KyoCera, Japan). After the implant surgery, the rats were returned to the home cage and were allowed to recover from the surgery for 1 to 2 days before the first recording was made. On the EEG recording, the rat with the implant was transferred from the home cage to a clear plexiglass cage (60 × 60 × 60 cm), on whose walls two piezoelectric speakers (Digi-Key Electronics, Thief River Falls, MN, USA) were pasted. A wired head-stage amplifier (HST/8o50-G1-GR; Plexon, Inc., Dallas, TX, USA) or wireless head-stage amplifier (TBSI Wireless Neural Recording System, Global Bio Inc., Mount Laurel, NJ, USA) were used for EEG recording. Raw electrical signals were further amplified and digitized using the Omniplex multichannel recording system (Plexon). In total, 45 rats of both sexes, including 20 Gad1 (–/–), 9 Gad1 (+/–), and 16 Gad1 (+/+) rats, were used in this study. The rats were separated into two age groups, younger (<2 months) and older (≥2 months), and their EEGs were examined for SWDs. Because SWDs were rare in younger Gad1 (+/–) and Gad1 (+/+) rats, only EEG data from older rats were further analyzed. Thus, in total, 100 h of EEG data were collected and analyzed from 10 Gad1 (–/–) rats (age: 2–11 months), 45 h from 6 Gad1 (+/–) rats (age: 3–12 months), and 70 h from 7 Gad1 (+/+) rats (age: 2–12 months). EEG recording was performed for up to 5 h per day, and recording lasted for up to 3 months in a single animal. Some EEG recording sessions were videotaped to record rat behavior.

In the acoustic stimulation experiment, a rat was placed in a clear plexiglass cage while wearing an EEG headset. The experimenter continuously monitored the EEG for SWDs and applied a series of acoustic stimulations from piezoelectric speakers (dominant frequency: 2 kHz; loudness: 82 dB [measured at the cage center]; ambient loudness: 44 dB) as soon as an SWD appeared on the monitor. The recording was started without acoustic stimulation for 1–2 h and continued for another 1–2 h with acoustic stimulation. In total, 22 h of EEG recording data were obtained from five Gad1 (–/–) rats (age: 2–10 months), as well as 28 h of EEG recording data from two Gad1 (+/–) rats and four Gad1 (+/+) rats (age: 3–9 months). Because Gad1 (+/–) and Gad1 (+/+) rats yielded similar results, the data from these genotype groups were merged and used as control group data.

All EEG recording data were analyzed offline for SWDs using custom-made programs written in MATLAB code. Briefly, raw voltage signals sampled at 1 kHz were subjected to high-pass filtering (cutoff: 0.9 Hz) and wavelet-transformed (30). SWDs were extracted from the total spectral power time series data by imposing frequency thresholds at 5 and 32 Hz; this interval contained most of the SWD oscillation power (Figure 1). The extracted SWDs were checked visually, and mechanical noise caused by rats tapping the cable or chewing food pellets was excluded from the data. By our definition, a spike-train waveform lasting <1 s was not regarded as an SWD.

We initially examined the effect of intraperitoneal (i.p.) injections of VPA (150–300 mg/kg, dissolved in 1 ml of saline) using three Gad1 (–/–), two Gad1 (+/–), and two Gad1 (+/+) rats. In this pilot experiment, we observed that all became drowsy and inactive after the injection, except for one Gad1 (+/+) rat. To avoid acute poisoning and the need for repeated i.p. administration, we next tried a milder oral formulation of the drug, using another set of rats (two Gad1 (–/–), four Gad1 (+/–), and two Gad1 (+/+) rats). VPA (10 mg/ml) was dissolved in water and made available to the rats ad libitum for 7 days. At the end of this period, we collected the blood serum samples from these animals and outsourced them to a local biochemistry lab (Nagahama Life Science Laboratory, Shiga, Japan) for the VPA blood concentration measurement. VPA concentration was measured based on the enzyme immunoassay method (Emit2000 valproate assay, Siemens Healthcare, Japan). The average blood concentration of VPA was 40.4 ± 10.4 μg/ml (24.1–73.7; n = 8 animals). The effective blood concentration of VPA for the treatment of epilepsy in humans is 50–100 μg/ml (31). Based on these observations, we decided to adopt the oral administration of the drug for the EEG recording experiment and the GABA immunohistochemistry. For the recording experiment, a pre-medication EEG recording was performed 2–3 days before starting drug administration [control group: VPA (–)]. EEG recording was resumed on the 3rd day after the initiation of drug administration and performed for a total of 2–3 days until the 7th day of the VPA treatment [experimental group: VPA (+)] (Table 1). In total, we obtained 70 h of EEG recording data from four Gad1 (–/–) rats (age: 5–11 months) and 68 h from four Gad1 (+/+) rats (age: 5–12 months). The water intake and body weight were monitored during the experiment (Table 2).

Four Gad1 (+/+) VPA (–), five Gad1 (+/+) VPA (+), four Gad1 (–/–) VPA (–), and four Gad1 (–/–) VPA (+) rats were euthanized by intraperitoneal injection of pentobarbital (150 mg/kg animal) and perfused with one-half Karnovsky fixative (2% paraformaldehyde and 2% glutaraldehyde in phosphate-buffered saline). The fixed brain was immersed in 30% sucrose in 0.1 M phosphate buffer for cryoprotection and embedded in Tissue-Tek OCT compound (Sakura Finetek Japan, Tokyo, Japan). The frozen brain was cut into 10-μm-thick sections using a cryostat (CM1950, Leica, Wetzlar, Germany) and mounted onto coated slide glasses (MAS-01, Matsunami Glass, Kishiwada, Japan). Coronal sections (ca. bregma −2.7 mm) were selected to examine the RTN and neighboring thalamic nuclei. Immunostaining was performed using a STAINperfect Immunostaining Kit (ImmuSmol, Bordeaux, France), following the manufacturer's instructions. The primary antibodies used in these experiments included chicken polyclonal anti-GABA (1:500 dilution; ImmuSmol), rabbit polyclonal anti-parvalbumin (PV) (1:2,000, GTX134110, Genetex, Hsinchu, Taiwan), chicken normal serum (for negative control, 1:2,000 dilution; Jackson ImmunoResearch), and the secondary antibodies were preabsorbed Alexa Fluor-488 conjugated goat anti-chicken IgY (1:500 dilution; ab150173, Abcam, Cambridge, UK) and preabsorbed Alexa Fluor-568 conjugated goat anti-rabbit IgG (1:500 dilution; ab175695, Abcam, Cambridge, UK). For fluorescent nucleus staining, we used a CellStain−4′,6-diamidino-2-phenylindol (DAPI) solution (1:500 dilution, Dojindo Kumamoto, Mashiki, Japan). Fluorescence images were acquired using an LSM 780 laser confocal microscope (Zeiss, Oberkochen, Germany) and a 63 × oil objective lens (numerical aperture: 1.4) and analyzed using the ZEN confocal image analysis software (Zeiss) and ImageJ software (National Institutes of Health, Bethesda, MD, USA), respectively (32–34) (Supplementary Figures S1, S2).

Median values were compared using the Wilcoxon rank-sum, Wilcoxon signed-rank, and Kruskal–Wallis tests using the MATLAB Statistics and Machine Learning Toolbox. Multiple comparisons were performed using the multicompare function with Fisher's least significant difference option in the same toolbox. All plots were drawn using Excel software (Microsoft Corp., Redmond, WA, USA). In the box-whisker plot, the upper (lower) whisker was drawn from the top (bottom) of the box, up (down), to the largest (smallest) data point within 1.5-fold of the interquartile range. Outliers outside the upper and lower whiskers are not shown in the plot.

EEG patterns in Gad1 (–/–) rats during the awake or sleep (rapid eye movement and non-rapid eye movement) states appeared to be similar to those in normal animals (32) (Figure 1A). However, on visual inspection of the EEG data, frequent large-amplitude spiky waveforms caught the attention (Figure 1A, SWD), which lasted a few seconds and oscillated at approximately 8 Hz (Figure 1B). These features are suggestive of spike-wave discharges in rodents (5, 7). These SWDs were observed while the rats were awake but quiet or at rest after active exploration. SWDs were observed in Gad1 (–/–) rats as early as 2 months of age, whereas they were rarely detected in Gad1 (+/–) or Gad1 (+/+) rats at similar ages (Figures 1C, D, Table 3). Only one out of seven rats showed SWDs (29 SWDs were detected from a total of 10 h of recording). The other six control rats did not show SWD in their EEG (a total of 175 h of recording). However, even in Gad1 (+/+) rats, SWDs eventually prevailed in the EEG at older ages. No overt convulsive seizures or other epileptic activity were observed in Gad1 (–/–) rats (data not shown).

As SWDs were rare in young Gad1 (+/–) and Gad1 (+/+) rats, only SWDs from older rats were analyzed and compared among the three genotypes (Figure 2). SWDs were significantly longer in Gad1 (–/–) rats (median: 2.3 s; n = 7,305 SWDs from 10 rats) than in Gad1 (+/–) rats (2.0 s; n = 1,749 SWDs from 6 rats) or Gad1 (+/+) rats (2.1 s; n = 3,766 SWDs from 7 rats) (Figure 2A; P = 4.2 × 10⁻¹³; Kruskal–Wallis test). The difference between Gad1 (+/+) and Gad1 (+/–) rats was not statistically significant (P = 0.41; Wilcoxon rank-sum test). The SWD length shows long-tailed distributions (Figure 2B), with the longest SWD lasting longer than 10 s. Most SWDs were shorter than 5 s in duration, regardless of genotype. The median SWD count in Gad1 (–/–) rats (61/h) was the largest among the three genotypes (P = 8.3 × 10⁻⁷; Kruskal–Wallis test) (Figure 2C), while the median SWD in Gad1 (+/–) rats (28/h) was significantly smaller than in Gad1 (+/+) rats (39/h) (P = 0.011; Wilcoxon rank-sum test). Therefore, the total SWD time per hour was greatest in Gad1 (–/–) rats (median: 155 s/h), which was significantly different from Gad1 (+/–) and Gad1 (+/+) rats (median: 86 and 102 s/h, respectively, P = 5.7 × 10⁻⁶; Kruskal–Wallis test) (Figure 2D). The total SWD time in Gad1 (+/–) was shorter than Gad1 (+/+) rats, however the difference was insignificant (P = 0.09; Wilcoxon rank-sum test). These results indicate that Gad1 (–/–) rats had the most severe SWD phenotype and that GAD1 plays a role in suppressing SWD generation in rats.

In human patients with absence seizures, SWD appears abruptly in EEG, while the patients temporally lose consciousness and are often unaware of external sensory stimuli (1–3). Therefore, we investigated whether the rats were responsive to external acoustic stimulation when SWDs appeared in the EEG. We presented buzzer tones through speakers to awake-behaving animals during live EEG monitoring. The experimenter presented the buzzer tone as soon as an SWD was detected in the EEG displayed on a monitor. Gad1 (–/–) and control rats responded overtly to these buzzer tones (Table 4). Most SWDs were interrupted readily by a single volley of buzzer tones (Case 1 SWD, Figure 3A), whereas the remaining SWDs were not (Case 2 SWD, Figure 3B). The appearance of SWDs in EEG was highly correlated with the inactivity of animals caused by reward withdrawal or a loss of motivation to achieve the reward during behavioral tasks (33). Because buzzer stimulations were not associated with reward and did not require a voluntary response, the SWDs observed in the rats, including in Case 2, likely reflect the animal's gradual loss of interest in, or sensory adaptation to, repeated buzzer tones (11, 33). If this is true, the number of SWDs should increase over time. To test this possibility, the SWDs were categorized into the first and second halves of the 1-h recording session. The total SWD count, including that of Cases 1 and 2, was not significantly different between the first and second halves of the session in Gad1 (–/–) rats (median: 63 and 48 per 30 min, respectively; n = 9; p = 0.82, Wilcoxon signed-rank test) and control rats (median: 37 and 9 per 30 min, respectively; n = 11; p = 0.10) (Figure 3C). Similarly, the SWD counts of Case 2 were not significantly different between the first and second halves of the session in Gad1 (–/–) rats (median: 6 and 16 per 30 min, respectively; p = 0.51) or control rats (median: 2 and 0 per 30 min, respectively; p = 0.11). We extended three recording sessions for an additional hour and observed no significant increase in SWD counts in any group (data not shown). Therefore, our results do not suggest that the SWDs were produced by sensory adaptation or a loss of motivation over time. Interestingly, the SWD count in Case 2 was significantly higher in Gad1 (–/–) rats (median Case2/(Case1 + Case2) ratio: 18%; n = 18) than in the control group (median ratio: 5%; n = 20) (p = 0.015, Wilcoxon rank-sum test) (Figure 3D). These results indicate that rats responded to sensory stimuli when SWDs appeared in EEG, and Gad1 (–/–) rats had more SWDs resistant to sensory interruption.

We next examined VPA, a medication commonly used to treat both the absence and the myoclonic seizure, which inhibits GABA degradative enzymes such as GABA transaminase (34, 35), thereby increasing the GABA level in the brain. As SWD was a common symptom in both older Gad1 (+/+) and Gad1 (–/–) rats, we expected that an increase in GABA caused by VPA would alleviate the SWD phenotype in both genotypes. Consistent with this expectation, the SWD phenotype in Gad1 (+/+) rats (n = 4) was significantly alleviated by VPA treatment (Figure 4). The median length of individual SWDs was decreased by VPA treatment, from 2.2 s (without VPA treatment; n = 2,516 SWDs) to 1.9 s (with VPA treatment; n = 1,625 SWDs) (Figure 4A). A decrease in SWD length was observed over the entire range of SWD sizes (Figure 4B). The median count of SWD appearances per hour decreased from 73 to 51 (Figure 4C); thus, the accumulated SWD length per hour decreased from 200 to 127 s following VPA treatment (Figure 4D).

Counterintuitively, the SWD phenotype in the Gad1 (–/–) rats (n = 4) was worsened by VPA treatment (Figure 5). The median SWD length without treatment (median: 2.4 s; n = 3,113 SWDs) increased to 3.0 s following VPA treatment (n = 3,132 SWDs) (Figure 5A). Notably, SWDs shorter than 4 s in duration decreased in number, whereas those lasting longer than 5 s became more frequent (Figure 5B). The number of SWD occurrences, irrespective of size, did not differ significantly between treatment groups (Figure 5C). Overall, accumulated SWD length per hour increased from 289 to 307 s; however, this increase was not significant (Figure 5D). In summary, VPA treatment exacerbated the SWD symptom in Gad1 (–/–) rats, unlike in control rats, suggesting that a mere increase in GABA level by the drug cannot compensate for GAD1 deficiency.

GAD1 is abundantly expressed in GABAergic neurons in the RTN, which plays an important role in SWD generation (15–19). We finally examined GABA immunoreactivity in the RTN (Figure 6). We identified GABAergic RTN neurons and their axons through co-immunostaining with an antibody against parvalbumin (PV), a peptide marker for the GABAergic RTN neurons (36–39). In Gad1 (+/+) rats without VPA treatment [VPA(–)], a strong GABA immunoreactivity was detected in the somata of RTN neurons (Figure 6A, left). A further enhanced GABA immunoreactivity was observed after VPA treatment [VPA(+)] (Figure 6B, left). As expected, GAD1 deficiency reduced GABA immunoreactivity in the somata (Figure 6C, left). VPA treatment restored GABA immunoreactivity in the somata of Gad1 (–/–) rats (Figure 6D, left). These results are quantified and summarized in Figure 6E. GABA immunoreactivity was significantly lower in Gad1 (–/–) rats (P = 7.4 × 10⁻⁹; Kruskal–Wallis test). The GABA immunoreactivity was restored in Gad1 (–/–) after VPA treatment, comparable to that of Gad1 (+/+) rats (P = 0.69; Wilcoxon rank-sum test).

By contrast, the patterns of GABA immunoreactivity in the RTN axons targeting the thalamus were different from those found in the RTN somata. The GABA immunoreactivity in the RTN axons was similar between Gad1 (+/+) (Figure 6A, right) and Gad1 (–/–) rats (Figure 6C, right). VPA treatment led to a marked increase in GABA immunoreactivity in the axons of both Gad1 (+/+) rats (Figure 6B, right) and Gad1 (–/–) rats (Figure 6D, right) to a similar degree. These results are summarized in Figure 6F. The difference in GABA immunoreactivity was not significantly different between the two genotypes without VPA treatment [VPA(–)] (P = 0.54; Wilcoxon rank-sum test) or with VPA treatment [VPA(+)] (P = 0.73). These results suggest that GAD1 has a less influential role in axonal GABA production. In summary, we detected a significant reduction in GABA immunoreactivity in the RTN neuronal somata in Gad1 (–/–) rats, consistent with the notion that GAD1 is essential for GABA synthesis in the RTN. The residual GABA in Gad1 (–/–) rats and excess GABA after VPA treatment are presumably produced by another GABA-synthesizing enzyme, GAD2, which is also expressed in RTN neurons and predominantly localized in the axons (28). The absence of functional GAD1 in the RTN and the excess GABA production in the axons after VPA treatment may account for the contradictory VPA treatment effect observed in Gad1 (–/–) rats (see the “Discussion” section).