Abnormal patterns of sleep and waking behaviors are accompanied by neocortical oscillation disturbances in an Ank3 mouse model of epilepsy-bipolar disorder comorbidity

ANK3 is a leading bipolar disorder (BD) candidate gene in humans and provides a unique opportunity for studying epilepsy-BD comorbidity. Previous studies showed that deletion of Ank3-1b, a BD-associated variant of Ank3 in mice leads to increased firing threshold and diminished action potential dynamic range of parvalbumin (PV) interneurons and absence epilepsy, thus providing a biological mechanism linking epilepsy and BD. To explore the behavioral overlap of these disorders, we characterized behavioral patterns of Ank3-1b KO mice during overnight home-cage activity and examined network activity during these behaviors using paired video and EEG recordings. Since PV interneurons contribute to the generation of high-frequency gamma oscillations, we anticipated changes in the power of neocortical EEG signals in the gamma frequency range (> 25 Hz) during behavioral states related to human BD symptoms, including abnormal sleep, hyperactivity, and repetitive behaviors. Ank3-1b KO mice exhibited an overall increase in slow gamma (~25-45 Hz) power compared to controls, and slow gamma power correlated with seizure phenotype severity across behaviors. During sleep, increased slow gamma power correlated with decreased time spent in the rapid eye movement (REM) stage of sleep. Seizures were more common during REM sleep compared to non-REM (NREM) sleep. We also found that Ank3-1b KO mice were hyperactive and exhibited a repetitive behavior phenotype that co-occurred with increased slow gamma power. Our results identify a novel EEG biomarker associating Ank3 genetic variation with BD and epilepsy and suggest modulation of gamma oscillations as a potential therapeutic target.


INTRODUCTION
Epilepsy and bipolar disorder (BD) are chronic, episodic, and debilitating disorders that often lead to suicide.Epidemiological studies suggest that approximately 10% of people with epilepsy experience symptoms of BD, including mania, depression, and disrupted patterns of sleep [1].Reciprocally, people with BD are at a four-fold increased risk of developing epilepsy [2][3][4].In addition to these shared features and bidirectional comorbidity, anticonvulsant medications are often used to effectively treat BD, and kindling effects seen in BD are hypothesized to be related to kindling in epilepsy [5].These data suggest that these two disorders share mechanistic links.However, identifying such mechanisms has remained a challenge [6].
The ANK3 gene provides a singular opportunity to explore the nexus of these disorders.ANK3 is a leading BD candidate gene with links to epilepsy [7].Rare variants of significant effect in coding regions of ANK3 have been identified in families with BD and mood-related psychiatric symptoms [8][9][10], and common variants in noncoding regions of ANK3 are more widely associated with BD through multiple, independent genome-wide association studies (GWAS) [11][12][13][14][15][16].These GWAS show a hotspot for single nucleotide polymorphisms (SNPs) in an intronic, candidate cisregulatory region upstream of ANK3's third alternative-first-exon (exon-1b).Exon-1b and its upstream cis-regulatory sequence is highly conserved among vertebrates, and ChIP-seq data shows similar transcription factor binding affinities for these regions between rodents and humans (https://genome.ucsc.edu/index.html)[17,18].Furthermore, human studies suggest that SNPs in this region are associated with reduced expression of ANK3 transcripts in patients with BD and accompanied by altered neuropathology [19][20][21].A recent study identified a female patient with a novel homozygous missense variant (c.178 T > C; p.Tyr60His) in exon-1b of the ANK3 gene who exhibited seizures and mood disturbance [22].
ANK3 is a very large gene that undergoes significant alternative splicing to produce various isoforms of the scaffolding protein, ankyrin-G (ankG).ANK3 has three alternative-first-exons, two of which are utilized by neurons and glia in the brain (exon-1e and exon-1b) to transcribe three main groups of ankG isoforms with different functions.Members of the 480 kDa group of isoforms are found at GABAergic synapses and the axon initial segment (AIS) of neurons [23].At the AIS, exon-1e and exon-1b code for alternative N-termini (NT2 and NT3, respectively) of AnkG, which serve to regulate the binding of sodium and potassium channels to the adjacent membrane binding domain [24].
Our previous study discovered that GABAergic parvalbumin (PV) interneurons exclusively express NT3-AnkG isoforms at the AIS, while excitatory pyramidal cells in many brain regions express NT2-ankG.This work showed that PV interneurons in Ank3-exon1b (Ank3-1b) knockout (KO) mice have a reduced number of voltage-gated sodium channels at the AIS, increased firing threshold, and diminished action potential dynamic firing range at frequencies above 100 Hz [7].Additionally, this work showed that Ank3-1b deletion results in reduced cortical network inhibition and an absence epilepsy phenotype that was recapitulated by knocking out all ankG isoforms selectively in PV interneurons.This makes Ank3-1b mice a valuable and so far unique model for studying monogenic epilepsy-BD comorbidity.These findings led to the hypothesis that imbalanced excitation and inhibition due to PV interneuron dysfunction may underlie the genetic association between ANK3 and BD and serve as a shared mechanistic link connecting thalamocortical epilepsy and mood disorder.
PV interneuron deficiencies and aberrant gamma rhythms have been implicated in BD through human studies [25][26][27][28][29], and a familial mutation of ANK3 (W1989R) caused disrupted cortical and hippocampal gamma oscillations in a knock-in mouse model [9].The fast-spiking properties of PV interneurons are important for generating and modulating high-frequency gamma oscillations [30].Thus, we hypothesized changes in the power of gamma oscillations in Ank3-1b KO mice.Additionally, we hypothesized that changes in gamma power in this model would be accompanied by changes in behaviors relevant to epilepsy-BD comorbidity, such as sleep disturbance and altered activity levels.
Previous behavioral studies of Ank3-1b heterozygous (Ank3-1b +/KO ) KO mice have been used to model BD.In one study, a battery of behavioral assays was used to phenotype Ank3-1b +/KO mice and found decreased anxiety, increased motivation for reward at baseline, and a transition to depression-related features after chronic stress via isolation [31].However, the effect sizes were relatively small, making reproducibility a challenge for in vivo electrophysiological studies that are often limited in cohort size due to technical and time constraints.Also, brief behavioral assays may not capture mood oscillations congruent with those of BD patients, which require study over longer periods of time.
One way to address these issues is to utilize recent highthroughput methods of monitoring rodent behavior over prolonged uninterrupted periods of home cage activity, allowing the study of some common endophenotypes of BD that have not yet been explored in this model, such as sleep disturbances and repetitive behaviors.In the present study, we adopted an automated approach to behaviorally phenotype Ank3-1b KO mice using the open-source learning algorithm, DeepLabCut [7], while simultaneously examining gamma oscillations using paired video-EEG recordings taken from the Ank3-1b KO mice and their littermate controls (Supplementary Fig. 1).

MATERIALS AND METHODS
All experimental procedures and data, including Supplementary Figs.1-14 are included in the article and in Supplementary information.Materials and protocols are available upon request from the corresponding author (A.Y.L.).

RESULTS
Increased slow gamma power in Ank3-1b KO mice correlated with epilepsy phenotype severity Previous characterization of Ank3-1b mice showed that both Ank3-1b KO/KO and Ank3-1b KO/+ KOs have frequent seizure episodes, with SWDs occupying approximately 4% of total EEG activity from Ank3-1b KO/KO mice [7].Since SWD spiking occurs at 6-8 Hz, creating pronounced peaks in power spectra, and because these seizures are accompanied by behavioral arrest, it was important to remove these events from our analyses.To do this, we utilized a supervised learning algorithm to automatically identify seizures in the EEG data [32].This algorithm identifies SWDs by first identifying consecutive spikes that are at least 2.5 x the average baseline voltage of the overall EEG and then based on the weighed scores of three additional properties: 1.) ~6 Hz spiking frequency (set range from 5 to 10 Hz), 2.) ~16 to 32 Hz harmonic commonly seen in power spectra of SWDs, and 3.) sharpness of spikes as estimated using the D4 wavelet transform.We trained this algorithm on EEG from the Ank3-1b model to identify tentative SWDs (Fig. 1A & B) according to scores most resembling a subset of hand-scored data.Though the learning algorithm identified minimal false-positive SWD events (example WT trace seen in Fig. 1B), we chose to manually verify every automated call made by 'detect_SWD' (Fig. 1C).This data was then used to index EEGs and remove all SWDs before spectral analysis.
Spectral analysis of all EEG recordings, indiscriminate of behavior, was first conducted to look for large overall differences across genotypes (Fig. 1D).Ank3-1b deletion differentially affected the power of the different rhythm types across the entire recording time (Fig. 1D; generalized linear mixed model: significant genotype by rhythm type interaction: F(8,48) = 5.392, p < 0.001).The power of delta and slow gamma rhythms, but not theta, significantly differed across genotypes (main effect of genotype on delta power: F(2,16) = 4.085, p = 0.037; Fig. 1E, main effect of genotype on gamma power: F(2,16) = 4.038, p = 0.038).The slow gamma power that we measured in the ~25-45 Hz range did not appear to be an integer multiple of lower frequency peaks, and the peak was not in the range of typical theta harmonics.This suggests that the gamma that we measured did not merely reflect theta harmonics.
To explore the extent to which slow gamma power was modulated by the phase of lower-frequency rhythms [33], we performed a cross-frequency coupling analysis.Coupling of slow gamma amplitude to the phase of lower frequency rhythms (<10 Hz) was apparent in most of the mice (Supplementary Figs.2-3).Comparison of theta-gamma phase-amplitude coupling (PAC), but not theta-delta PAC, was stronger in Ank3-1b KO/KO mice than in wild-type mice (main effect of genotype on theta-delta PAC: F(2,11) = 0.759 p = 0.491; Fig. 1F, the main effect of genotype on theta-gamma PAC: F(2,11) = 6.655, p = 0.013; post hoc pairwise comparisons: Ank3-1b KO/KO vs wildtype theta-gamma PAC: p = 0.029; Ank3-1b KO/KO vs ; Ank3-1b KO/+ theta-gamma PAC: p = 0.051; Ank3-1b KO/+ vs wildtype: p = 1).These results support the conclusion that the slow gamma rhythms recorded were physiologically relevant and not simply reflective of noise (EEG traces and gamma filtered traces see Supplementary Fig. 4).We also found a significant relationship between slow gamma power and seizure classification scores (Supplementary Fig. 5).
The presence of these overall differences across genotypes encouraged us to next investigate effects of Ank3-1b deletion during specific behavioral and sleep states.We used DeepLabCut (DLC) [34] to obtain estimates of animals' positions, and a previously published method to classify awake behaviors and sleep [35], from home cage videos (Supplementary Fig. 6).
We next examined whether disturbances in REM and NREM sleep patterns were accompanied by disturbances in EEG rhythms during sleep.We found that Ank3-1b deletion affected oscillatory power differently during REM and NREM sleep (generalized linear mixed model: significant genotype by sleep stage interaction: F(3,342) = 92.3,p < 0.001; significant genotype by sleep stage by rhythm type interaction: F(12,342) = 1178.1,p < 0.001), with stronger effects of Ank3-1b deletion on power spectra observed during REM than NREM (Fig. 2B, E).Therefore, we next analyzed EEG rhythms during NREM and REM separately.
During identified NREM sleep epochs, Ank3-1b deletion did not produce generalized effects on power across all frequencies (generalized linear mixed model: significant genotype by rhythm type interaction: F(6,168) = 53.7,p < 0.001).Instead, effects of Ank3-1b deletion on rhythmic power were most pronounced for slow gamma rhythms (Fig. 2B, Supplementary Fig. 1 Seizure activity and neocortical rhythms in Ank3-1b knockout mice.A Boxplots of all positive and negative classification scores of SWD-like activity in EEG recordings from WT and Ank3-1b KO/+ and Ank3-1b KO/KO mice.B Example EEG trace from automated SWD prediction with highest classification score from each genotype.Note: the trace for the WT mouse does not meet the requirements for classification as an SWD, demonstrating the need for C manual validation.C Boxplots of classification scores after manual verification of SWDs.Note: WT mice had no confirmed SWDs.D Average power spectra for all recordings per genotype with 95% confidence intervals.E Plots of estimated gamma power (25-45 Hz) for all recordings per genotype with 95% confidence intervals.Significant increases in gamma power are seen between Ank3-1b KO/KO and WT mice (*p = 0.024) and between Ank3-1b KO/KO and Ank3-1b KO/+ (*p = 0.015).F Average Theta:Gamma PAC (*p = 0.029).
Previous characterization of these mice showed that Ank3-1b KO/KO mice have premature mortality rates [7], and meet the criteria for sudden death in epilepsy (SUDEP), so we also looked at seizure activity during REM and NREM sleep.Our data indicates that seizures are approximately 10X more common during REM (1.24 seizures/hr) than during NREM (0.13 seizures/hr) sleep, and this increased seizure frequency was statistically significant in REM compared to NREM sleep (Supplementary Fig. 10A, B).This result was striking because seizures triggered during REM sleep are more likely to cause sudden death compared to waking and NREM sleep [51].We then looked at the effect of seizure occurrence on slow gamma power in Ank3-1b KO/+ and Ank3-1b KO/KO mice and found that seizure occurrence affected gamma power differently between genotypes during REM sleep.We found that seizure occurrence had no significant effect on gamma power in Ank3-1b KO/+ , but seizure occurrence was associated with significantly increased gamma power in Ank3-1b KO/KO mice (Supplementary Fig. 10C).This highlights again a positive correlation between seizure phenotype severity and increased slow gamma power and suggests that seizure activity and slow gamma rhythms may be associated.When we looked at the effect of seizure occurrence on theta power in Ank3-1b KO/+ and Ank3-1b KO/KO mice during REM sleep, we found that seizure occurrence affected theta power differently between genotypes.We found that seizure occurrence had no significant effect on theta power in Ank3-1b KO/+ , but seizure occurrence was associated with significantly decreased theta power in Ank3-1b KO/KO mice (Supplementary Fig. 10C).Seizures were extremely rare during NREM; only a single mouse exhibited seizures during NREM.Thus, we did not examine gamma power for seizures occurring during NREM.
Again, it is important to note that seizures may be unaccounted for due to potentially unidentified sleep bouts, since thorough analysis of sleep requires simultaneous EMG analysis.may be having seizures during REM and NREM periods that our algorithm was unable to identify due to disrupted theta-to-delta ratios.Thus, it is critical for future studies to further explore these effects in order to better understand the risk and underlying mechanisms leading to SWDs during sleep in this model.

DISCUSSION
Human genetic studies have advanced our understanding of psychiatric illness through the identification of rare and common variants associated with complex neurodevelopmental disorders.However, understanding the role that such variants play in polygenic traits like mood disturbances seen in BD and epilepsy has remained a challenge.One such challenge is the difficulty of modeling subjective human traits like mood in rodents.Even more difficult is modeling rhythmic disorders like BD and epilepsy in which shifts in mood and activity levels occur on various time scales as long as weeks to months.
Many studies attempting to utilize Ank3 as a model of BD in mice have shown promise.For example, one study showed that Ank3-1b KO/+ mice model aspects of BD such as shifts from maniclike features, including reduced anxiety and increased motivation for reward, to depressive-like features after chronic stress that were attenuated by lithium [31].However, these behaviors were of small effect size and difficult to reproduce in small-scale studies.Alternatively, a pyramidal cell forebrain conditional KO of Ank3 exhibited similar behavioral phenotypes modeling BD but with greater effect size [64].In this model, behaviors at baseline were manic-like, including hyperactivity, and changed to depressive-like features after social defeat stress.This phenotype was also attenuated by lithium and valproate.However, the genetic manipulation was more severe than the diminished ANK3 expression reported in human studies, and neither study modeled the full range of BD symptoms and endophenotypes.Furthermore, the seizure phenotype was unknown at the time of these studies, so attention was not given to the relevance of these behaviors to epilepsy.
These limitations illustrate the trade-off that exists between models with BD-related phenotypes of large effect size and those that more accurately model BD variable expressivity but have phenotypes of smaller effect size.This trade-off affects overall reproducibility in rodent experiments and their translational relevance to BD treatment in human patients.Our approach was able to identify new BD-related behaviors in Ank3-1b mice and provides advanced methodology that can be tailored towards more reliable long-term behavioral analysis of mice.We developed algorithms suitable for analyzing long periods of undisturbed home cage activity that more faithfully capture behavioral rhythmicity seen in BD and epilepsy.
Furthermore, our results illustrate a robust behavioral and electrophysiological phenotype that can be used for future in vivo single-unit electrophysiological experiments conducted over periods of weeks to months with the intention of gaining mechanistic insights into behaviors relevant to mood symptoms seen in human BD and epilepsy patients.For example, the sleep disturbance phenotype that we have identified in this model has direct relevance to patient treatment of BD and epilepsy because sleep disturbances are present across all states of both disorders and highly consequential for treatment [65].While manic states of BD are characterized by decreased need for sleep and depressive states of BD are characterized by increased sleep, even euthymic states of BD show characteristics of sleep disturbance [40][41][42][43][44][45][46][47].Sleep disturbance is common among patients with epilepsy, and seizures during sleep, particularly during REM sleep, can be fatal [51,66].
Recently, a quantitative phosphoproteomic study has implicated Ank3 in sleep need, lending support to our findings of dysregulated sleep in Ank3-1b mice [67].Future studies utilizing EMG will be helpful for studying specific types of REM disturbances in Ank3-1b mice and further understanding the role of Ank3-1b in sleep need.Interestingly, there is growing evidence for a role of REM sleep in emotional memory processing in the hippocampus [68][69][70][71], and gamma is thought to promote memory consolidation during REM sleep [72].Additionally, our data is consistent with phenotypes seen in the Kv3.1/Kv3.3 and Stargazer mouse models.Like Ank3, Kv3-type channels are expressed in interneurons, and Kv3.1 is expressed only in PV interneurons.Deletion of Kv3.1/Kv3.3 in PV interneurons leads to decreased action potential dynamic range, hyperactivity, and increased slow gamma power   [73][74][75][76].The Stargazer mouse model also has many overlapping features with the Ank3-1b mouse model.For example, both models exhibit the absence epilepsy and ataxia, respond to ethosuximide, have PV-interneuron dysfunction, and show an increase in interictal slow gamma power (Maheshwari et al., 2015).A recent study characterized the behaviors of the Stargazer model during home-cage activity and found overlapping phenotypes with Ank3-1b mice, including hyperactivity, increased repetitive behaviors, sleep deficits (Schirmer et al., 2022).
Given studies showed reduced PV interneuron firing rates in ranges above 100 Hz in Ank3-1b KO/KO mice [7], we hypothesize that the increased slow gamma power seen here was due to an inability of PV interneurons in KO mice to fire at frequencies high enough to produce fast gamma oscillations.Unfortunately, we were unable to measure fast gamma in this study due to the low sampling frequency of the pre-existing dataset.PV interneurons in the cortex and hippocampus have been shown to be more active during REM than NREM [77]; the PV interneuron dysfunction in Ank3-1b mice could explain the larger effect of Ank3-1b loss on REM sleep compared to NREM sleep, and why seizures are more common during REM sleep, in this model.Furthermore, we hypothesize that the increased slow gamma seen during other behavioral states may relate to changes in behavior that we were unable to identify here due to limitations of the existing dataset, including limited recording durations, sample size, and camera angles for distinguishing more complex behaviors.
Further studies examining the relationship between the increases in slow gamma power observed here, and changes in power of other frequency bands, with changes in Ank3-1b mice behavior will be important for understanding the human relevance of these mice as a model of BD.For example, thetagamma coupling is highly implicated in processes of memory and navigation in both humans and rodents [78], and gamma coherence is disrupted during various tasks in human subjects with BD [27,79].Studies of gamma in BD have shown increased power in the 30-50 Hz range in BD patients during visual tasks and resting states [28,29,80].Additionally, the global increase we see in slow gamma across behaviors may reflect volume-conducted signals from specific brain regions involved in different behaviors, so in vivo recordings using implanted probes will be critical for identifying the sources of these signals.This increased slow gamma phenotype in Ank3-1b mice may represent another way in which EEG disturbances can be used in translational studies to develop treatments for patients with BD.

Fig. 2
Fig. 2 Ank3-1b knockout mice showed disturbed sleep patterns and increased slow gamma during sleep, with stronger effects observed during REM sleep than NREM.A Plots of time spent in NREM (left) and frequency of NREM bouts (right) per hour in which each point represents an hour of measurement.Ank3-1b KO/KO mice have increased numbers of NREM bouts compared to WT mice (*p = 0.004).B Individual plots of average power spectra during NREM sleep in which each plot represents power measured from an individual mouse with bootstrapped 95% confidence intervals showing variability in power between bouts.C Plots of estimated gamma power (25-45 Hz) in which each point is a gamma power measurement during a NREM bout.Ank3-1b mice have increased slow gamma power compared to WT during NREM sleep (Ank3-1b KO/KO vs. WT: *p < 0.001; Ank3-1b KO/+ vs. WT: *p = 0.001; Ank3-1b KO/KO vs. Ank3-1b KO/+ : *p < 0.001).D Plots of time spent in REM (left) and frequency of REM bouts (right) per hour in which each point represents an hour of measurement.Ank3-1b KO/+ and Ank3-1b KO/KO mice have decreased hourly REM sleep duration compared to WT (*p = 0.002 and *p = 0.03, respectively), and Ank3-1b KO/KO and Ank3-1b KO/+ mice have decreased REM bouts per hour compared to WT (*p = 0.002 and *p < 0.001, respectively).E Individual plots of average power spectra during REM sleep in which each plot represents an individual mouse with bootstrapped 95% confidence intervals showing variability in power between bouts.F Plots of estimated gamma power (25-45 Hz) in which each point is a gamma power measurement during a REM bout.Ank3-1b mice have increased slow gamma compared to WT during REM sleep (Ank3-1b KO/KO vs. WT: *p < 0.001; Ank3-1b KO/+ vs. WT: *p = 0.02; Ank3-1b KO/KO vs. Ank3-1b KO/+ : *p < 0.001).

Fig. 3
Fig. 3 Ank3-1b knockout mice exhibited hyperactivity and increased slow gamma power during awake behaviors.A Plots of time spent in awake rest per hour (left) and bouts of awake rest per hour (right) in which each point represents an hour of measurement.Ank3-1b KO/KO mice spent decreased time per hour in awake rest (*p = 0.006) and had decreased bouts of awake rest (*p = 0.04) compared to WT mice.B Individual plots of average power spectra during awake rest in which each plot represents an individual mouse with bootstrapped 95% confidence intervals showing variability in power between bouts.C Plots of estimated gamma power (25-45 Hz) in which each point is a gamma power measurement during a bout of awake rest.Ank3-1b mice have increased slow gamma compared to WT mice during awake rest (*p < 0.001, for all comparisons).D Plots of time spent walking per hour (left) and bouts of walking per hour (right) in which each point represents an hour of measurement.Ank3-1b KO/KO and Ank3-1b KO/+ mice spent increased time (*p < 0.001 and *p = 0.04, respectively) and had increased bouts of walking per hour (*p = 0.002 and *p < 0.001, respectively) compared to WT mice.E Individual plots of average power spectra during walking in which each plot represents an individual mouse with bootstrapped 95% confidence intervals showing variability in power between bouts.F Plots of estimated gamma power (25-45 Hz) in which each point is a gamma power measurement during a bout of walking.Ank3-1b mice had increased slow gamma power compared to WT mice during walking (*p < 0.001, for all comparisons).