DNA was extracted from the tail or ear tissue, and PCR was performed using the KAPA Mouse Genotyping Standard Kit (KAPA Biosystems). The following primers were used for PCR. Fwd: CGAGCATTGAGATCCTCTTTCT; Rev: GTCATCATTGCTCCATCAACAG. The restriction enzyme HinfI was used to distinguish WT from Gnb1^K78R/+ mice.

These procedures are described in Supplementary material.

All behavioral experiments were performed in the Mouse Neurobehavioral Core facility at Columbia University Irving Medical Center.

ECT tests were performed as previously published with modifications (Frankel et al., 2001; Kapur et al., 2020). In brief, tests were performed on experimentally naive mice beginning at 6–8 weeks of age. A drop of anesthetic containing 0.5% tetracaine and 0.9% NaCl was placed onto each eye and a fixed electrical current was applied via silver transcorneal electrodes using a customized electroconvulsive stimulator (Ugo Basile model 57800-D01, Stoelting Co., Wood Dale, IL). The stimulator was set to produce rectangular wave pulses with the following parameters: 299 Hz, 0.2 s duration, and 1.6 ms pulse width.

To find the generalized tonic-clonic seizure threshold (also previously described as “minimal clonic forebrain seizure”), the seizure response was determined for a particular current, beginning with a current approximately 0.5–1.0 mA below the known average threshold for that genetic background and sex. Seizures were scored using a modified Racine scale (seizure scores: 0, no observable symptoms; 1, stun; 2, myoclonic jerk/twitch; 3, rearing, forelimb and jaw clonus; 4, forelimb tonic extension; 5, tonic hindlimb extension). The current was increased in 0.5 to 1.0 mA increments daily until the mouse had a first clear seizure response (a score of 3 or higher). As male and female mice have different thresholds, the sexes were analyzed separately. For analysis, the mean integrated root mean square (iRMS) current is reported for each genotype-sex group. Group means were calculated to determine the threshold.

For initial EEG screening, electrode implantation of adult mice approximately 6–8 weeks old and video-EEG was performed essentially as described (Asinof et al., 2016). Mice were anesthetized with tribromoethanol (250 mg/kg i.p., Sigma-Aldrich #T48402). Three small burr holes were drilled in the skull (1 mm rostral to the bregma on both sides and 2 mm caudal to the bregma on the left), 2 mm lateral to the midline. One hole was drilled over the cerebellum as a reference. Using four Teflon-coated silver wires soldered onto the pins of a microconnector (Mouser electronics #575- 501101), the wires were placed between the dura and the brain, and a dental cap was then applied. The mice were given a post-operative analgesic of carprofen (5 mg/kg subcutaneous Rimadyl injectable) and allowed a 48-h recovery period before recordings were taken. To record EEG signals, mice were connected with flexible recording cables to allow minimally restrictive movements within the cage. Signal (200 samples/s) was acquired either on a Grael II EEG amplifier (Compumedics) or on a Natus Quantum amplifier (Natus Neuro), and the data were analyzed either with the Profusion 5 (Compumedics) or NeuroWorks (Natus Neuro) software. Differential amplification recordings were recorded pair-wise between all three electrodes, as well as referential, providing a montage of six channels for each mouse. Mouse activity was captured simultaneously by video monitoring using a Sony IPELA EP550 model camera, with an infrared light to allow recordings in the dark. For initial screening and SWD event scoring, we recorded continuously for 24 to 48 h. For drug dosing studies, mice were recorded continuously for 48 h, injected with saline around 12 pm on day 1 and injected with the drug around 12 pm on day 2.

To analyze the effect of SWD events on movement, a custom-built MATLAB software was developed to perform real-time video-tracking while simultaneously conducting EEG recording on a Neuralynx Digital Lynx system. The software synced video-taping with EEG recording through a NetCom API provided by Neuralynx. An infrared camera was used to track the body position (the center of the whole body) of a mouse by subtracting each video frame from the background image, captured in the absence of the mouse. The animal's movement was calculated as the pixel distance between body positions divided by the time. Then, movement during SWD periods (detected by the algorithm described above) was averaged for each SWD duration or each inter-SWD interval, and further normalized for each animal to the average movement over the whole recording session.

All drugs used were purchased from commercial sources: ethosuximide (Sigma-Aldrich), valproate sodium (Sigma-Aldrich), phenytoin (Selleckchem), baclofen (Tocris), and bicuculline (Tocris or Hello Bio). For mouse injections, ethosuximide and valproate sodium were resuspended in 0.9% saline and phenytoin in 2% N,N-dimethylacetamide/10% propylene glycol/30% 2-hydroxypropyl-beta-cyclodextrin in H₂O, and administered via intra-peritoneal (IP) injections in a volume of 10 ml/kg. For MEA, drugs were either resuspended in H₂O for ethosuximide, valproate sodium, and baclofen or in DMSO for phenytoin. DMSO concentration on MEA never exceeded 0.1%. Concentrations used are reported in figure legends.

Primary cortical and hippocampal neurons were maintained on a monolayer of mitotically inactivated primary astrocytes. To generate primary astrocytes, we modified a previously described protocol (Schildge et al., 2013) as follows. Cortices from P0 WT mice were dissociated as described above. Cell suspensions were plated onto 75 cm² flasks pre-coated with poly-D-lysine (15 μg/mL, Sigma P6407) in astrocyte maintenance medium consisting of DMEM/F12 (Gibco), 10% heat-inactivated FBS (Gibco), 1x MEM-NEAA (Gibco), 1x glutamine (Gibco), and 1% penicillin/streptomycin (Gibco). Cultures were maintained in a humidified environment at 37°C in 5% CO₂. The medium was changed every other day. Once confluent, flasks were tapped vigorously for 1 min to detach microglia, oligodendrocytes, and neurons, and the medium was aspirated. The remaining adherent cells were washed with 1x PBS and passaged with 0.05% trypsin (Gibco) and re-plated in fresh flasks pre-coated with PDL. This process was repeated three times to generate a homogenous culture of astrocytes. The absence of neurons was confirmed by staining for MAP2 (Sigma #M4403). On the final passage, astrocytes were mitotically inactivated by treatment with 10 μM cytosine β-D-arabinofuranoside (Ara-C, Sigma #C6645) for 6 h, washed two times with 1x PBS, and cryopreserved in cold maintenance medium containing 10% DMSO (Gibco) at 500,000 cells per vial.

Mitotically inactive astrocytes were seeded in an astrocyte maintenance medium onto 15 mm coverslips (50,000 cells per coverslip) pre-coated with 50 μg/mL PDL. The medium was changed every other day. After 7 days (DIV7), cortical and hippocampal neurons were prepared from P0 WT and Gnb1^K78R/+ pups as described above. Neuronal preparations were plated at 75,000 cells per 15 mm coverslip containing primary astrocytes. Cultures were then maintained in Neurobasal-A, 1x B27 supplement, 1x GlutaMax, 1x MEM-NEAA, 1% HEPES, and 1% penicillin/streptomycin (NBA/B27 medium), and 50% medium was changed every other day with fresh NBA/B27 medium. On DIV8, cultures were transduced with 50,000 vg/cell of either AAV8-CaMKIIa-GFP (UNC Vector Core) or AAV9-mDlx-NLS-mRuby279 (AddGene #99130) to allow excitatory and inhibitory neurons to be identified by fluorescence at the time of recording.

Recordings were performed via conventional whole-cell current or voltage clamp methods at DIV13–16 using a Multiclamp 700B amplifier and a Digidata 1550 digital-to-analog converter. Signals were recorded at a 10-kHz sample rate and filtered at 3 KHz with a low-pass Bessel filter using pCLAMP 10 software (all equipment from Molecular Devices). Patch pipettes were fabricated with a P-97 pipette puller (Sutter Instruments) using 1.5 mm outer diameter and 1.28 mm inner diameter filamented capillary glass (World Precision Instruments). Pipette resistance was 3–6 MΩ. The external recording solution contained 145 mM NaCl, 5 mM KCl, 10 mM HEPES, 10 mM glucose, 2 mM CaCl₂, and 2 mM MgCl₂. The pH was adjusted to 7.3 using NaOH and the osmolality was adjusted to 325 mOsm with sucrose. The pipette solution contained 130 mM K⁺ methanesulphonate, 10 mM Na⁺ methanesulphonate, 1 mM CaCl₂, 10 mM EGTA, 10 mM HEPES, 5 mM MgATP, and 0.5 mM Na₂GTP (pH 7.3, 305 mOsm). The calculated free Ca²⁺ concentration (Schoenmakers et al., 1992) was about 22 nM. Experiments were performed at room temperature (21–23°C). A−14 mV liquid junction potential between the pipette and external solutions was calculated empirically, and the correction was applied before the experiment. Only cells with a stable series resistance of <15 MΩ were used for analysis.

Neuronal excitability was assessed using current clamp recording. For these experiments, resting membrane potential was measured immediately following the establishment of the whole-cell configuration. During recordings, current (<100 pA) was manually injected to hold the cells at approximately −60 mV. Membrane resistance and capacitance were calculated from the membrane potential changes in response to 1 s duration hyperpolarizing current steps that decreased in−5 pA increments. Action potentials were evoked and rheobase was obtained using 1 s duration depolarizing current steps that increased incrementally by 5–20 pA. An action potential was defined as a transient depolarization of the membrane which had a minimum rise rate of >10 mV/ms and reached a peak amplitude of >0 mV. Action potential characteristics were measured from the first action potential at rheobase. The threshold potential was measured at the point where the voltage increases at a rate >10 ms/mV. The duration was calculated from the full width at the half maximum voltage. For this calculation, the amplitude was measured from 0 mV to the peak potential. The maximum number of action potentials was measured from a 1-s current step.

Baclofen-evoked currents were recorded using the voltage clamp technique at a holding potential of−65 mV (corrected for junction potential). Series resistance was compensated at 60%. The external recording solution was supplemented with 300 nM tetrodotoxin, 25 μM bicuculline, 10 μM CNQX, and 100 μM AP-5 (all purchased from Hello Bio). Cells were perfused with an external recording solution at a rate of 0.25 mL/min. 0.1, 1.0, and 10 μM baclofen solutions were applied sequentially to neurons using a custom-built perfusion system with a 250-μm aperture located approximately 500 μm from the soma of the recorded cells. The solution exchange time at the recorded neuron was <0.5 s.

Quantification was carried out using custom-written scripts for Igor Pro v. 6 (Wavemetrics, USA) and R v. 3 (www.R-project.org). Statistical comparisons were made using the tests indicated in the Results section. P-values <0.05 were considered significant.

The DNAs encoding the proteins expressed in the oocytes, and RNA preparation, were as described (Reddy et al., 2021). GIRK1 (rat), GIRK2A (mouse), Gβ₁ (bovine), and Gγ₂ (bovine) cDNAs used in this study were cloned into high-expression oocyte vectors pGEM-HE or pGEM-HJ (Rishal et al., 2005). To express human GIRK2 (GeneBank: P48051), we used DNA synthesized on our request by ThermoFisher. To ensure robust activation by Gβγ, the GIRK1/2 channels were expressed at a low surface density of ~3–5 channels/μm² (Yakubovich et al., 2015) by using low doses of GIRK1 and GIRK2 RNAs (0.025–0.5 ng RNA/oocyte), which resulted in GIRK basal current (I_basal) of ~2–3 μA and maximal Gβγ-evoked currents of ~8–12 μA with saturating doses of Gβγ for heteromeric GIRK1/2. In contrast, to obtain comparable surface levels and Gβγ-evoked currents of homotetrameric GIRK2 or GIRK2-YFP (Kahanovitch et al., 2014), we had to inject 2 ng of GIRK2 RNA/oocyte. Note that I_basal for GIRK2 was very low, ~0.05–0.1 μA. The amounts of Gβγ RNAs injected (per oocyte) were 5 ng Gβ₁ and 2 ng Gγ₂.

Oocyte defolliculation, incubation, and RNA injection were performed as described previously (Rishal et al., 2005; Rubinstein et al., 2009). The standard extracellular ND96 solution contained (in mM) 96 NaCl, 2 KCl, 1 MgCl₂, 1 CaCl₂, and 5 HEPES, and was titrated with NaOH to a pH of 7.6–7.8. Oocytes were defolliculated with collagenase (Type 1A, Sigma) in Ca²⁺-free ND96 solution and injected with 50 nL of RNA, and incubated for 2–4 days in NDE solution (ND96 solution supplemented with 2.5 mM pyruvate and 50 μg/ml gentamicin) at 20°C before testing. CaCl₂ was omitted in Ca²⁺-free ND96. Whole-cell GIRK and Ca_v3.2 currents in oocytes were measured using a two-electrode voltage clamp (TEVC) with Geneclamp 500 (Molecular Devices) and using agarose cushion electrodes filled with 3M KCl, with resistances of 0.1–0.8 MΩ for current electrode and 0.2–1.5 MΩ for voltage electrode. To measure GIRK currents, we used ND96 solution or high-K⁺ solution (HK24), in mM 24 KCl, 72 NaCl, 1 CaCl₂, 1 MgCl₂, and 5 HEPES. The pH of all solutions was 7.4–7.6.

ETX was dissolved in double-distilled water to obtain a 1-M stock solution. During the experimental procedure, we used eight concentrations of ETX solutions (in mM): 0.01, 0.03, 0.1, 0.3, 1, 3, 10, and 30. To prepare the experimental solutions, we diluted the ETX stock solution in HK24 to 30 mM, and the lower concentrations have been prepared by sequential dilutions.

The effect of ETX on GIRK was measured as follows: GIRK currents were measured at a holding potential of−80 mV while the solutions were changed from one to another. First, current was measured in ND96 solution for 5 s. Then, the solution was switched to HK24 for 30 s to measure the basal current. The next solution was HK24 with 0.01 mM ethosuximide for another 30 s, and the same step was repeated seven times for each of the ethosuximide solutions. Finally, HK24 + 2.5 mM Ba²⁺ solution was used to block all GIRK current and to reveal non-GIRK basal current.

Dose–response data were fitted to the different models using SigmaPlot 11 or 13 (Systat Software Inc.). The data were fitted to equations 1–3, representing three standard models: the standard one-component binding isotherm is

where x is drug concentration and K_d,app is the apparent dissociation constant; the one-site Hill equation is

where n is Hill coefficient; and the two-component binding isotherm is

where K_d1,app and K_d2,app are apparent dissociation constants for the 1st and 2nd binding sites, respectively, and c is the fraction of inhibition obtained by the binding of ETX to the high-affinity site.

Giant excised plasma membrane patches were prepared, stained with antibodies, and imaged as described (Peleg et al., 2002). Oocytes were mechanically devitellinized using fine forceps in a hypertonic solution (in mM: NaCl 6, KCl 150, MgCl₂ 4, HEPES 10, and pH 7.6). The devitellinized oocytes were transferred onto a Thermanox™ coverslip (Nunc, Roskilde, Denmark) immersed in a Ca²⁺-free ND96 solution, with their animal pole facing the coverslip, for 10–20 min. The oocytes were then suctioned using a Pasteur pipette, leaving a giant membrane patch attached to the coverslip, with the cytosolic face toward the medium. The coverslip was washed thoroughly with fresh ND96 solution and fixed using 4% formaldehyde for 30 min. Fixed giant plasma membrane patches were immunostained in 5% milk in phosphate buffer solution (PBS). Non-specific binding was blocked with Donkey IgG 1:200 (Jackson ImmunoResearch #017-000-003). Anti-Gβ rabbit polyclonal antibody (T-20, Santa Cruz Biotechnology #SC-378) was applied at 1:200 dilution, for 45 min at 37°C. Anti-rabbit IgG DyLight 649-labeled secondary antibody (1:400; SeraCare Life Sciences #072-05-15-06) was then applied for 30 min at 37°C, washed with PBS, and mounted on a slide for visualization. Imaging was done with the LSM 510 META confocal microscope (Zeiss) with a 20× -air objective or a 40× -water immersion objective using the 633 nm laser in spectral mode, and emission was measured at 673 nm. Imaging of proteins in giant plasma membrane patches was performed using the confocal microscope in λ-mode. Images were centered on the edges of the membrane patches, so that background fluorescence from the coverslip could be seen and subtracted.

Sample sizes and statistical tests are detailed in the figure legends. Data are presented as mean ± SEM. Statistical analysis was performed on raw data, except Figure 4A (see legend). Each dataset was tested for normality (Shapira–Wilk test) and equal variance (Brown–Forsyth test). Data that did not show normal distribution were analyzed using non-parametric methods. Statistical analyses for adult behavioral tests were performed using Prism (GraphPad). Two-group comparisons were performed using unpaired two-tailed t-test with correction using the Holm–Sidak method or one-way repeated measures ANOVA with Bonferroni correction. Multiple group comparisons were done using two-way repeated measures ANOVA followed by the Sidak test. The correlation between movement and SWD duration or inter-SWD interval was determined by a Pearson test. All other statistics were performed in R using a Mann–Whitney U-test with 1,000 or 10,000 permutations, as indicated in legends (see the MEA methods section for more details). For oocyte electrophysiology experiments, statistical analysis was performed on raw data with SigmaPlot 13 (Systat Software Inc.) or Prism (GraphPad). Two-group comparisons were performed using a two-tailed t-test, if the data passed the Shapiro–Wilk normality test and the equal variance test, otherwise, a Mann–Whitney rank sum test was used. Multiple group comparisons were done using one-way ANOVA on ranks followed by Dunn's test.

We generated an inbred mouse line carrying the first mutation identified in a patient, K78R, using CRISPR/Cas9. The K78R knockin mutation (c.233 A>G; p.K78R human NM_002074, mouse NM_008142) was verified by Sanger sequencing and PCR (Supplementary Figures 1A–C). As a model for GNB1 haploinsufficiency, we also obtained a knockout line with a 71-bp deletion (Del71) (Supplementary Figure 1D). Both Gnb1^Del71 and Gnb1^K78R lines showed homozygous embryonic lethality (Supplementary Figure 1E). However, heterozygous Gnb1^Del71 pups (Gnb1^Del71/+) were viable, while heterozygous K78R pups (Gnb1^K78R/+) showed reduced viability on an isogenic C57BL/6NJ (B6NJ) background. To minimize premature lethality that was observed as a result of deficient maternal behavior of B6NJ dams, we also generated a mixed background with FVB.129. Indeed, F1 hybrid (F1H) litters were generally larger and healthier, as observed by increased survival (Supplementary Figure 1E). We observed normal Gβ₁ protein levels in postnatal day 0 (P0) Gnb1^K78R/+ cortex and, as expected, a ~50% reduction of Gβ₁ levels in Gnb1^Del71/+ mice compared to WT (Supplementary Figure 1F). We also observed high Gβ₁ membrane expression and low cytosolic expression in Gnb1^K78R/+ mice (Supplementary Figure 1G). Taken together, these data demonstrate that the K78R mutation does not confer a null allele and that the protein is properly targeted to the membrane.

We also assessed Gβ₁ expression in primary cortical neurons from WT and Gnb1^K78R/+ pups. Gβ₁ is expressed in both excitatory and inhibitory neurons (Supplementary Figures 2A–C), but is not expressed in astrocytes (Supplementary Figure 2D). Gβ₁ preferentially localizes in the soma, near the plasma membrane, and is not present in the axon initial segment (Supplementary Figures 2E, E'). Gβ₁ is expressed in both deep and upper layer cortical neurons (Supplementary Figure 2F). Immunocytochemistry did not detect distinct differences in Gβ₁ expression levels and localization patterns in WT and Gnb1^K78R/+ neurons (Supplementary Figure 2E' and data not shown); however, the resolution of this method is not sufficient to detect moderate changes in total or plasma membrane expression levels.

Both male and female Gnb1^K78R/+ pups had a significantly lower body weight than WT littermates at birth (Figures 1A, B). To assess behavioral phenotypes pertinent to key clinical features of GNB1 encephalopathy, we tested Gnb1^K78R/+ pups for physical and sensorimotor developmental milestones between P4 and P11 and Gnb1^K78R/+ adult mice for motor and cognitive functions. Early developmental milestones including body weight, surface righting reflex, negative geotaxis, vertical screen holding, and separation-induced ultrasonic vocalizations (USV) were assayed as previously described (Yang et al., 2012). We performed tests on both B6NJ and F1H backgrounds to assess phenotype stability. Gnb1^K78R/+ pups weighed significantly less than WT littermates across early postnatal days (Figure 1C) and remained significantly smaller as young adults at 6 weeks (Supplementary Figures 3A, B). Gnb1^K78R/+ pups exhibited deficits in the surface righting reflex test, being initially faster than WT littermates at P4 but not progressing afterward, and had a significant delay in both 90° and 180° negative geotaxis tests. They did not, however, show deficits in vertical screen holding, suggesting normal grip strength (Figure 1C). Finally, Gnb1^K78R/+ pups emitted few vocalizations indicating abnormal development (Figure 1C). Pups on an F1H background were less extensively studied. As noted above, they appeared healthier than on a B6NJ background but presented similar, yet milder, developmental delays compared to WT littermates (Supplementary Figure 3B).

We assessed spontaneous locomotor activity in the open field, and observed that adult Gnb1^K78R/+ mice were significantly less active compared to WT littermates (Figure 1D). Vertical exploration (rearing) and center time were not different (Supplementary Figure 3C), suggesting that reduced activity is not attributable to heightened anxiety-like behaviors in this test. Gnb1^K78R/+ mice did not show deficits in the RotaRod test, suggesting normal balance and grip strength (Supplementary Figure 3D).

We tested learning and memory in adult mice. While Gnb1^K78R/+ mice did not exhibit deficits in the classic contextual and cued fear conditioning tests (Supplementary Figure 3E), they displayed spatial learning and memory deficits in the Morris Water Maze test. These data suggest that, while Gnb1^K78R/+ mice might recognize contextual cues, they are impaired in using spatial cues to navigate a complex environment. In the water maze, Gnb1^K78R/+ mice were slower at acquiring the spatial learning task and the subsequent reversal task (Figure 1E). Importantly, these deficits were observed with normal swimming speed in Gnb1^K78R/+ mice (Supplementary Figure 3F), indicating that swimming ability did not confound performance in this test. Gnb1^K78R/+ mice also displayed memory deficits. During the probe test of both acquisition and reversal, WT mice spent significantly more time in the trained quadrant than in the other three quadrants, whereas Gnb1^K78R/+ mice did not show a preference for the quadrant where the platform used to be (Figure 1E and Supplementary Figure 3G). In conclusion, behavioral phenotypes in Gnb1^K78R/+ mice on the NJ background might be relevant to understanding GNB1 encephalopathy, which includes global developmental delay, ambulatory deficits, and intellectual disability. More comprehensive behavioral analysis is needed to understand the roles of sex and genetic background in the manifestation of impairments.

Seizures, in a variety of types, including absence-type epilepsy, are common among patients with GNB1 encephalopathy. We did not observe spontaneous convulsive seizures in Gnb1^K78R/+ mice, and consequently assessed seizure susceptibility by electroconvulsive threshold (ECT) testing in Gnb1^K78R/+ mice using stimulus settings for the minimal generalized (clonic forebrain) seizure endpoint compared to both Gnb1^+/+ (WT) and Gnb1^Del71/+. Female and male Gnb1^Del71/+ mice showed ECT values equivalent to WT controls (Figure 2A), demonstrating that Gnb1 haploinsufficiency does not lead to enhanced seizure susceptibility. Gnb1^K78R/+ mice, however, showed a significantly lower seizure threshold compared to WT (Figure 2A), indicating that the K78R mutation may result in hyperexcitability.

We next used video-electroencephalogram (EEG) to examine Gnb1^K78R/+ mice for spontaneous seizure events. Gnb1^K78R/+ mice exhibited highly frequent generalized bilateral spike-and-wave discharges (SWDs) (Figures 2B, C; mean ± SEM Gnb1^K78R/+: 1723 ± 108.9), while WT mice exhibited extremely rare SWDs (Figures 2B, C; mean ± SEM WT: 11.7 ± 6.5) in a 24 h period. The mean duration of SWDs ranged from 2.2 to 6.7 s (Figure 2D), with some events lasting longer than 30 s, and the fundamental frequency shows that the peak spectral power was 7.3 Hz (Figure 2E). We observed similarly frequent SWDs on both the F1H and B6NJ backgrounds. Together, these data show that Gnb1^K78R/+ mice possess an enhanced susceptibility to seizures and display a high number of SWDs, an electrographic signature reminiscent of absence (blank-staring) seizures that are not convulsive, and usually do not affect the behaviors measured in standard tests such as those examined in this study (Frankel et al., 2005). This seizure type is often found in patients with GNB1 encephalopathy (Petrovski et al., 2016; Hemati et al., 2018).

Multi-electrode arrays (MEA) measure the local field potentials of cultured neurons within a larger neuronal network and are a popular means to assess neuronal network activity and synchrony. We used MEA to characterize the spontaneous activity of ex vivo neuronal networks from WT and Gnb1^K78R/+ P0 cortices. Examination of MEA raster plots from WT and Gnb1^K78R/+ networks (examples are shown in Figure 3A) revealed visible differences in bursting patterns. Using our custom MEA analysis package (Gelfman et al., 2018), we analyzed a variety of spiking and bursting features (see Methods). We show the combination of nine plates corresponding to eight biological replicates (Figure 3B). For each feature, we performed a Mann–Whitney U (MWU)-test followed by 1,000 permutations to compare activity in WT vs. mutant wells across the time points of interest. We also combined the permutated MWU p-values from each plate using the Fisher's method, which gave similar results (see Supplementary Table 1 for p-values from both statistical methods).

When comparing the activity (mean firing rate, MFR) of WT neurons with respect to time (days in vitro, DIVs), we observed an increase in MFR until DIV13, followed by stabilization of MFR. We thus present data from DIV15 to DIV27, when network activity has stabilized indicating functionally mature networks. Both WT and Gnb1^K78R/+ networks developed at the same rate, as observed by similar increases in time to reach the number of active electrodes (nAE) per well, reaching a maximum by DIV15 (Supplementary Figure 4A). We observed striking bursting phenotypes in Gnb1^K78R/+ networks relative to WT. Mutant neurons showed a decrease in burst frequency (decreased number of bursts per minute) as they fired very long bursts followed by very long inter-burst intervals (IBIs), leading to an increase in the mean number of spikes within a burst (Figures 3A, B). We observed only a small and delayed decrease in MFR, suggesting that MFR is not driving the bursting phenotype; however, there was a decreased spike frequency within bursts (Figure 3B).

Primary cultured neural networks become highly synchronized over time (Wagenaar et al., 2006; McSweeney et al., 2016). Therefore, we evaluated key synchrony parameters including the percentage of spikes in network spikes (NS) (Supplementary Figure 4B), which identifies highly synchronized short events (10 ms), as well as mutual information (Supplementary Figure 4C) and Spike Train Tiling Coefficient (Supplementary Figure 4D), which assess pairwise electrode correlations between nearby electrodes and at the well-level, respectively (Gelfman et al., 2018). For each feature, we show raw as well as normalized data recorded on each plate by dividing the activity in each well by the mean activity of WT wells. We did not observe significant differences between WT and Gnb1^K78R/+ networks using any of the synchrony measures suggesting that altered synchrony does not underlie the bursting phenotypes of Gnb1^K78R/+ networks.

Ethosuximide (ETX) and valproic acid (VPA) are two antiepileptic drugs (AED) commonly used to treat absence seizures in humans, which also interrupt SWDs in rodent models (Marescaux et al., 1992; Manning et al., 2004; Vrielynck, 2013; Kim et al., 2015). Acute treatment of Gnb1^K78R/+ mice with either ETX or VPA nearly abolished SWDs (Figure 4A). Owing to the short half-life of both drugs, SWDs progressively recovered an hour after treatment. Notably, while the ETX showed a short half-life in this mouse model, it is expected to be much longer in large mammals and humans (Patel et al., 1975; el Sayed et al., 1978). On the other hand, phenytoin, a voltage-gated Na⁺ channel blocker and an AED with a probable worsening effect on absence seizures in humans and rodents (Marescaux et al., 1992; Tokuda et al., 2007; Kim et al., 2015), did not affect SWD occurrence (Figure 4A).

We next investigated the effects of ETX on neural networks on MEA. Interestingly, even an acute application of ETX appeared to have a rescue effect on the aberrant bursting of Gnb1^K78R/+ networks, decreasing the mean duration of bursts and IBIs (data not shown). Therefore, we wanted to determine whether chronic ETX exposure could have a lasting effect on the phenotype. ETX distributes equally among blood plasma, cerebrospinal fluid, and saliva (Löscher and Frey, 1984). The therapeutic range is typically between 300 and 750 μM (Patsalos et al., 2018), and can go as high as 1.1 mM (Kobayashi et al., 2009). We applied ETX on cultured networks at three different concentrations (250 μM, 750 μM, and 2 mM), every other day from DIV3 until DIV31. We observed that ETX reverted Gnb1^K78R/+ aberrant bursting in a concentration-dependent manner (Figures 4B, C). At 250 μM, ETX did not affect the bursting phenotype, while at 750 μM, ETX did restore close-to-normal firing to the mutant network. ETX increased number of bursts per minute, decreased mean duration of bursts, and decreased IBIs, compared to the vehicle or 250 μM-treated Gnb1^K78R/+ networks. Interestingly, 2 mM ETX seemed to produce an “overshoot” effect and greatly reduced the network activity. Overall, these data show that chronic ETX treatment, at therapeutic doses, leads to a cumulative and prolonged rescue of the Gnb1^K78R/+ bursting phenotypes. The beneficial effects of ETX in Gnb1^K78R/+ mouse and neuronal models implicate Gβ₁ signaling in the mechanism of action of ETX.

To determine whether the K78R mutation is associated with changes in intrinsic neuronal excitability, conventional whole-cell current clamp recording was performed on cultured cortical excitatory and inhibitory neurons obtained from WT and Gnb1^K78R/+ mice. The neuronal cultures were transduced by viral vectors containing reporter genes under the control of the appropriate promoters (see Methods section), and the excitatory or inhibitory identity of the neurons was determined by fluorescence at the time of recording. As shown in Supplementary Table 2, and illustrated in Supplementary Figure 5, while there were small changes to the membrane resistance, capacitance, and action potential duration in Gnb1^K78R/+ inhibitory neurons, there was no consistent pattern of changes to membrane characteristics that would indicate a clear change in basal excitability parameters.

Given the prominence of GIRK channels as Gβ₁ effectors that regulate the neuronal activity, we tested the hypothesis that there is aberrant activation of GIRK channels in neurons from Gnb1^K78R/+ mice. Whole-cell voltage-clamp recording of GIRK currents was carried out following activation of GABA_B receptors using a 20-s baclofen application at 0.1, 1, and 10 μM. In neurons, baclofen-evoked outward currents are almost entirely mediated by GIRK channels (Luscher et al., 1997; Takigawa and Alzheimer, 1999; Koyrakh et al., 2005), with a half-maximal effect (EC₅₀) at ~3 μM (Sodickson and Bean, 1996). Recordings were made from WT and Gnb1^K78R/+ cultured excitatory and inhibitory neuronal subtypes obtained from both the cortex and the hippocampus. As shown in Figure 5A, the amplitude of baclofen-evoked current in cortical Gnb1^K78R/+ inhibitory neurons was larger than in WT neurons at the highest (10 μM) baclofen concentration. Notably, no significant difference was observed in cortical excitatory neurons. Similarly, in hippocampal neurons (Figure 5B), the baclofen-evoked GIRK current was larger in both subtypes of Gnb1^K78R/+ neurons with the most pronounced increase occurring in inhibitory neurons at the highest baclofen concentration. These results are consistent with the idea that changes in GIRK channel modulation contribute to the neuronal activity changes observed in Gnb1^K78R/+ mice.

To study the effects of the K78R mutation on GIRK channel activation by Gβγ, GIRK1/2 or GIRK2 channel was expressed in Xenopus oocytes together with Gβ_WTγ or Gβ_K78Rγ (Rubinstein et al., 2009). Exchange from physiological low-K⁺ ND96 to high-K⁺ HK24 solution resulted in the development of inward currents carried mostly by GIRKs (Figures 6A, F). GIRK1/2 and GIRK2 show substantial differences in gating: GIRK1/2 shows high, Gβγ-dependent basal current (I_basal) and a modest 1.5–4-fold activation by Gβγ (Figure 6A; compare black and dark green traces), whereas GIRK2 has low, Gβγ-independent I_basal and shows strong ~30–50-fold activation by Gβγ (Rubinstein et al., 2009) (Figure 6F; compare black and dark green traces).

To capture the dose dependency of Gβγ activation on GIRK channels, we titrated the amount of expressed Gβγ protein by injecting increasing amounts of Gβ_WT or Gβ_K78R and Gγ RNA (5/1 RNA amounts, w/w). For GIRK1/2, small activation by Gβ_WTγ was observed with 0.1 ng Gβ RNA, and saturation was reached at 0.5–1 ng Gβ RNA (Figure 6B and Supplementary Figure 6C). Interestingly, at low doses of Gβ RNA (0.05-0.2 ng), Gβ_K78Rγ activated GIRK1/2 better than Gβ_WTγ (Figures 6A, B and Supplementary Figure 6C), suggesting gain-of-function (GoF). However, at higher Gβ doses, Gβ_K78Rγ ability to activate GIRK1/2 progressively decreased (Figure 6B and Supplementary Figure 6C). Notably, similar dose-dependent effects for Gβγ were observed in the follow-up study with GIRK4 and GIRK1/4 (Reddy et al., 2021), which are homologous to GIRK2 and GIRK1/2, respectively.

To investigate whether the GoF at low Gβγ levels and loss-of-function (LoF) at high Gβ_K78Rγ levels (i.e., RNA doses) are due to altered surface expression of Gβ_K78R, we monitored the levels of membrane-attached Gβγ in giant excised membrane patches of oocytes (Figure 6E) (Peleg et al., 2002; Rubinstein et al., 2009). We found that Gβ_K78Rγ consistently showed higher surface density than Gβ_WTγ at all RNA doses (Figure 6C). To directly compare the actual dose-dependence of GIRK1/2 activation for Gβ_WTγ and Gβ_K78Rγ, we plotted GIRK1/2 currents as a function of Gβγ surface levels as measured in oocytes of the same groups (Figure 6D). Per equal amount of expressed Gβγ, Gβ_K78Rγ activated GIRK1/2 similarly, or even less, compared to Gβ_WTγ. The highest dose of Gβ_K78Rγ caused a large decrease in GIRK1/2 current, but we did not reach a similar expression level of Gβ_WTγ to allow a direct comparison. The tendency for higher activation of GIRK1/2 at low doses of Gβ_K78Rγ RNA, and a smaller activation with high Gβ_K78Rγ doses, was consistently observed in additional experiments (Supplementary Figure 6C). Overall, these results suggest that K78R is a gain-of-expression mutation, and the apparent GoF is due exclusively to higher surface expression compared to Gβ_WTγ. Further investigation of this phenomenon showed suppression of GIRK1/2 channel expression by high doses of Gβ_K78Rγ RNA and a reduction in channel open probability (Reddy et al., 2021), explaining the attenuated GIRK1/2 currents at high expression levels of Gβ_K78Rγ.

The activation of GIRK2 by coexpressed Gβγ showed a similar pattern, with an apparent strong GoF for Gβ_K78Rγ at lower Gβ RNA doses which subsided at saturating RNA doses, 2–5 ng (Figures 6F, G, and Supplementary Figure 6C). With 0.5 ng Gβ RNA, GIRK2 current evoked by Gβ_K78Rγ was more than 6-fold larger than with Gβ_WTγ (Figure 6G). Similar to GIRK1/2, the surface expression of Gβ_K78Rγ was higher than Gβ_WTγ at all RNA doses (Figures 6H, J). Re-plotting GIRK2 activation as a function of Gβγ surface levels showed that K78R activated GIRK2 similarly to Gβ_WTγ in the range of surface expression studied (Figure 6I). Thus, similar to GIRK1/2, the physiological GoF of Gβ_K78Rγ on GIRK2 is due to gain of expression.

We further tested the possible link between ETX-produced amelioration of the MEA bursting phenotypes and GIRK channels. The primary mode of action for ETX is thought to be through inhibition of low voltage-activated T-type Ca²⁺ channels (Ca_V3) (see Discussion), but ETX has also been shown to inhibit GIRK channels in vitro (Kobayashi et al., 2009). We assessed ETX inhibition of heterotetrameric GIRK1/2 and homotetrameric GIRK2 in Xenopus oocytes (Figure 7 and Supplementary Figure 6). Both basal and Gβγ-evoked GIRK1/2 currents were inhibited in a dose-dependent manner by ETX, with an EC₅₀ of ~1.5–2 mM (Figures 7A, B and Supplementary Figures 6A, B), similar to a previous report (Kobayashi et al., 2009). I_basal of the mouse GIRK2, used throughout this study, was too small to reliably monitor inhibition. Using human GIRK2 that gave a larger I_basal, we observed ETX inhibition with an EC₅₀ of 3.7 ± 0.4 mM (n = 22), similar to the previous report (Kobayashi et al., 2009). Remarkably, Gβγ-activated GIRK2 was inhibited at much lower doses (Figures 7A–C), with an EC₅₀ of ~55 μM (Figure 7 and Supplementary Figures 6A, B). A milder increase in ETX potency by Gβγ was also observed for GIRK1/2 (Figure 7B, Supplementary Figure 6B). Enhanced inhibition of GIRK by ETX following activation by Gβγ is an interesting, and potentially clinically important, phenomenon deserving further study.

The dose–response curves for ETX inhibition of both GIRK1/2 and GIRK2 were well fitted by a Hill equation with apparent K_d (K_d,app) values close to the observed EC₅₀ (Figure 7 and Supplementary Figures 6A, B). However, for GIRK1/2, the low Hill coefficient (n_H ~0.5–0.6) could indicate two binding sites with different affinity—satisfactory fits were also obtained assuming two affinity components (Supplementary Figures 6A, B)—or two populations of channels with different sensitivity to block. Importantly, ETX inhibited both GIRK2 and GIRK1/2 channels activated by Gβ_K78Rγ with similar or even a slightly higher apparent affinity compared to Gβ_WTγ (Figure 7B and Supplementary Figures 6A, B), in line with the strong effect of ETX on the Gnb1^K78R/+ neurons.