Stxbp1/Munc18-1 haploinsufficiency in mice recapitulates key features of STXBP1 encephalopathy and impairs cortical inhibition

Mutations in genes encoding synaptic proteins cause many neurodevelopmental disorders, but the underlying pathogeneses are poorly understood. Syntaxin-binding protein 1 (STXBP1) is an essential component of the neurotransmitter release machinery. Its de novo heterozygous mutations are among the most frequent causes of neurodevelopmental disorders including intellectual disabilities and epilepsies. These disorders, collectively referred to as STXBP1 encephalopathy, affect a broad spectrum of neurological and neuropsychiatric features common among neurodevelopmental disorders. To gain insight into STXBP1 encephalopathy pathogenesis, we generated new Stxbp1 null alleles in mice and found that Stxbp1 haploinsufficiency impaired cognitive, psychiatric, and motor functions and caused cortical hyperexcitability and seizures. Surprisingly, Stxbp1 haploinsufficiency reduced neurotransmission from cortical parvalbumin- and somatostatin-expressing GABAergic interneurons by differentially decreasing the synaptic strength and connectivity, respectively. These results demonstrate that Stxbp1 haploinsufficient mice recapitulate key features of STXBP1 encephalopathy and indicate that inhibitory dysfunction is likely a key contributor to the disease pathogenesis.


Introduction 38
Human genetic studies of neurodevelopmental disorders continue to uncover pathogenic variants 39 in genes encoding synaptic proteins (Deciphering Developmental Disorders Study, 2017;2015;40 mice. Furthermore, we used the partition test to examine the preference for social novelty, in 272 which a mouse is allowed to interact with a familiar or novel partner mouse. Both WT and 273 Stxbp1 tm1d/+ mice preferentially interacted more with the novel partner mice (Figure 2 To further assess the well-being and neuropsychiatric phenotypes of Stxbp1 haploinsufficient 279 mice, we performed the Nestlet shredding test and marble burying test to examine two innate 280 behaviors, nest building and digging, respectively. We provided a Nestlet (pressed cotton square) 281 to each mouse in the home cage and scored the degree of shredding and nest quality after 24,48,282 and 72 hours ( Figure 2O). Stxbp1 tm1d/+ and Stxbp1 tm1a/+ mice consistently scored lower than WT 283 mice at all time points (Figure 2P). In the marble burying test, the Stxbp1 tm1d/+ and Stxbp1 tm1a/+ 284 mice buried fewer marbles than WT mice ( Figure 2Q). The interpretation of marble burying 285 remains controversial, as it may measure anxiety, compulsive-like behavior, or simply digging 286 behavior (Deacon, 2006;Thomas et al., 2009;Wolmarans et al., 2016). Since Stxbp1 287 haploinsufficient mice show elevated anxiety and repetitive behaviors, the reduced marble 288 burying likely reflects an impairment of digging behavior, possibly due to the motor deficits. 289 Likewise, the motor deficits may also contribute to the reduced nest building behavior. 290 291 Cortical hyperexcitability and epileptic seizures in Stxbp1 haploinsufficient mice 292 Another core feature of STXBP1 encephalopathy is epilepsy with a broad spectrum of seizure 293 types, such as epileptic spasm, focal, tonic, clonic, myoclonic, and absence seizures (Stamberger 294 et al., 2016;Suri et al., 2017). To investigate if Stxbp1 haploinsufficient mice have abnormal 295 cortical activity and epileptic seizures, we performed chronic video-electroencephalography 296 (EEG) and electromyography (EMG) recordings in freely moving Stxbp1 tm1d/+ mice and their 297 sex-and age-matched WT littermates. We implanted three EEG electrodes in the frontal and 298 somatosensory cortices and an EMG electrode in the neck muscle to record intracranial EEG and 299 EMG, respectively, for at least 72 hours ( Figure 3A). The phenotypes of each mouse are 300 summarized in Supplementary Table 1. Stxbp1 tm1d/+ mice exhibited cortical hyperexcitability 301 and several epileptiform activities. First, they had numerous spike-wave discharges (SWDs) that 302 typically were 3-6 Hz and lasted 1-2 s ( Figure 3C,E,F). These oscillations showed similar 303 characteristics to those generalized spike-wave discharges observed in animal models of absence 304 seizures (Depaulis and Charpier, 2018;Maheshwari and Noebels, 2014). A much smaller 305 number of SWDs with similar characteristics were also observed in WT mice (Figure 3B), 306 consistent with previous studies (Arain et al., 2012;Letts et al., 2014). On average, the frequency 307 of SWD episodes in Stxbp1 tm1d/+ mice was more than 40 folds of that in WT mice (Figure 3E,F). 308 Importantly, SWDs frequently occurred in a cluster manner (i.e., ³ 5 episodes with an inter-309 episode-interval of £ 60 s) in Stxbp1 tm1d/+ mice, which never occurred in WT mice (Figure 3-310 supplement 1; Figure 3-supplement 2 Video S1). Furthermore, 56 episodes of SWDs from 10 311 out of 13 Stxbp1 tm1d/+ mice lasted more than 4 s, among which 54 episodes occurred during rapid 312 eye movement (REM) sleep ( Figure 3D; Figure 3-supplement 3 Video S2) and the other 2 313 episodes occurred when mice were awake. In contrast, only 1 out of 11 WT mice had 3 episodes 314 of such long SWDs, all of which occurred when mice were awake (Supplementary Table 1). In 315 Stxbp1 tm1d/+ mice, SWDs were most frequent during the nights, but occurred throughout the days 316 and nights (Figure 3F), indicating a general cortical hyperexcitability and abnormal synchrony 317 in Stxbp1 haploinsufficient mice. 318 319 Second, Stxbp1 tm1d/+ mice experienced frequent myoclonic seizures that were manifested as 320 sudden jumps or more subtle, involuntary muscle jerks associated with EEG discharges (Figure  321 3G,H). The large movement artifacts associated with the myoclonic jumps precluded proper 322 interpretation of EEG signals, but this type of myoclonic seizures was observed in all 13 323 recorded Stxbp1 tm1d/+ mice and the majority of episodes occurred during REM or non-rapid eye 324 movement (NREM) sleep ( Figure 3I; Figure 3-supplement 4 Video S3). There were 3 similar 325 jumps in 2 out of 11 WT mice that were indistinguishable from those in Stxbp1 tm1d/+ mice, but all 326 of them occurred when mice were awake ( Figure 3I). Moreover, the more subtle myoclonic 327 jerks occurred frequently and often in clusters in Stxbp1 tm1d/+ mice, whereas only isolated events 328 were observed in WT mice at a much lower frequency (Figure 3H,J; Figure 3

-supplement 5 329
Video S4). EEG and EMG recordings showed that the cortical EEG spikes associated with the 330 myoclonic jerks occurred before or simultaneously with the neck muscle EMG discharges 331 ( Figure 3H), consistent with the cortical or subcortical origins of myoclonuses, respectively 332 (Avanzini et al., 2016). 333 334 Stxbp1 haploinsufficiency reduces synaptic inhibition in a cell-type specific manner 335 To identify cellular mechanisms that may underlie the cortical hyperexcitability and neurological 336 deficits in Stxbp1 haploinsufficient mice, we examined neuronal excitability and synaptic 337 transmission in the somatosensory cortex. Whole-cell current clamp recordings of layer 2/3 338 pyramidal neurons in acute brain slices revealed only a small increase in the input resistances of 339 Stxbp1 tm1d/+ neurons as compared to WT neurons (Figure 4-supplement 1). Previous studies showed that synaptic transmission was reduced in the cultured hippocampal neurons from 341 heterozygous Stxbp1 knockout mice and human neurons derived from heterozygous STXBP1 342 knockout embryonic stem cells (Orock et al., 2018;Patzke et al., 2015;Toonen et al., 2006). 343 However, such a decrease in excitatory transmission is unlikely adequate to explain how Stxbp1 344 haploinsufficiency in vivo leads to cortical hyperexcitability. Thus, we focused on the inhibitory 345 synaptic transmission originating from two major classes of cortical inhibitory neurons, Pv and 346 Sst interneurons. A Cre-dependent tdTomato reporter line, Rosa26-CAG-LSL-tdTomato 347 (Madisen et al., 2010), and Pv-ires-Cre (Hippenmeyer et al., 2005) or Sst-ires-Cre (Taniguchi et 348 al., 2011) were used to identify Pv or Sst interneurons, respectively. We used whole-cell current 349 clamp to stimulate a single Pv or Sst interneuron in layer 2/3 with a brief train of action 350 potentials and whole-cell voltage clamp to record the resulting unitary inhibitory postsynaptic 351 currents (uIPSCs) in a nearby pyramidal neuron (Figure 4A,E). The connectivity rate of Pv 352 interneurons to pyramidal neurons was unaltered in Stxbp1 tm1d/+ ;Rosa26 tdTomato/+ ;Pv Cre/+ mice 353 ( Figure 4B), but the unitary connection strength was reduced by 45% as compared to 354 Stxbp1 +/+ ;Rosa26 tdTomato/+ ;Pv Cre/+ mice ( Figure 4C). In contrast, 355 Stxbp1 tm1d/+ ;Rosa26 tdTomato/+ ;Sst Cre/+ mice showed a 26% reduction in the connectivity rate of Sst 356 interneurons to pyramidal neurons ( Figure 4F), but the unitary connection strength was normal 357 ( Figure 4G). The short-term synaptic depression of both inhibitory connections during the train 358 of stimulations was normal (Figure 4D,H). Thus, cortical inhibition mediated by Pv and Sst 359 interneurons is impaired in Stxbp1 haploinsufficient mice, representing a likely cellular 360 mechanism for the cortical hyperexcitability and neurological deficits. 361 362

Discussion
Extensive biochemical and structural studies of Stxbp1/Munc18-1 have elucidated its crucial role 364 in synaptic vesicle exocytosis (Rizo and Xu, 2015), but provided little insight into its functional 365 role at the organism level. Hence, apart from being an essential gene, the significance of STXBP1 366 dysfunction in vivo was not appreciated until its de novo heterozygous mutations were 367 discovered first in epileptic encephalopathies (Saitsu et al., 2008) and later in other 368 neurodevelopmental disorders (Deciphering Developmental Disorders Study, 2015;Hamdan et 369 al., 2011;Rauch et al., 2012). In this study, we generated two lines of Stxbp1 370 haploinsufficient mice (Stxbp1 tm1d/+ and Stxbp1 tm1a/+ ) and systematically characterized them in all 371 of the neurological and neuropsychiatric domains affected by STXBP1 encephalopathy. These 372 mice exhibit reduced survival, hindlimb clasping, impaired motor coordination, learning and 373 memory deficits, hyperactivity, increased anxiety-like and repetitive behaviors, aggression, and 374 epileptic seizures. Sensory abnormality has not been documented in STXBP1 encephalopathy 375 patients (Stamberger et al., 2016) and we also did not observe any sensory dysfunctions in 376 Stxbp1 haploinsufficient mice. Thus, despite the large phenotypic spectrum of STXBP1 377 encephalopathy in humans, our Stxbp1 haploinsufficient mice recapitulate all key features of this 378 neurodevelopmental disorder and are construct and face valid models of STXBP1 379 encephalopathy. About 20% of the STXBP1 encephalopathy patients showed autistic traits 380 (Stamberger et al., 2016), but we and others (Kovačević et al., 2018;Miyamoto et al., 2017) did 381 not observe an impairment of social interaction in mutant mice using the three-chamber and 382 partition tests. Perhaps the elevated aggression in Stxbp1 haploinsufficient mice confounds these 383 tests, or new mouse models that more precisely mimic the genetic alterations in that subset of 384 STXBP1 encephalopathy patients are required to recapitulate this phenotype. 385 Prior studies using the other three lines of Stxbp1 heterozygous knockout mouse models reported 387 only a subset of the neurological and neuropsychiatric deficits that we observed here (Hager et 388 al., 2014;Kovačević et al., 2018;Miyamoto et al., 2017;Orock et al., 2018). For example, the 389 reduced survival, hindlimb clasping, motor dysfunction, and increased repetitive behavior were 390 not documented in the previous models. The previously reported cognitive phenotypes were 391 much milder than what we observed. Both Stxbp1 tm1d/+ and Stxbp1 tm1a/+ mice showed severe 392 impairments in the novel objection recognition and fear conditioning tests. In contrast, another 393 line of Stxbp1 heterozygous knockout mice showed normal spatial learning in the Morris water 394 maze and Barnes maze (a dry version of the spatial maze) in one study (Kovačević et al., 2018), 395 but reduced spatial learning and memory in the radial arm water maze in another study (Orock et 396 al., 2018). Different behavioral tests could have contributed to such differences among studies. 397 However, a subtle but perhaps key difference is the Stxbp1 protein levels in different lines of 398 heterozygous mutant mice. Stxbp1 is reduced by 50% in both of our Stxbp1 tm1d/+ and Stxbp1 tm1a/+ 399 mice, but only by 25-40% in other heterozygous knockout mice (Miyamoto et al., 2017;Orock 400 et al., 2018), which may lead to fewer or less severe phenotypes in the previous models. 401 402 Dysfunction of cortical GABAergic inhibition has been widely considered as a primary defect in 403 animal models of autism spectrum disorder, schizophrenia, Down syndrome, and epilepsy among 404 other neurological disorders (Contestabile et al., 2017;Lee et al., 2017;Marín, 2012;Nelson and 405 Valakh, 2015;Paz and Huguenard, 2015;Ramamoorthi and Lin, 2011). In many cases, the 406 origins of GABAergic dysfunction were either unidentified or attributed to Pv interneurons. Sst 407 interneurons have only been directly implicated in a few disease models (Ito-Ishida et al., 2015;408 distinct deficits at Pv and Sst interneuron synapses in Stxbp1 haploinsufficient mice, suggesting 410 that Stxbp1 may have diverse functions at distinct synapses. The reduction in the strength of Pv 411 interneuron synapses is consistent with the previous results that basal synaptic transmission is 412 reduced at the neuromuscular junctions of Stxbp1 heterozygous null flies and mice (Toonen et 413 al., 2006;Wu et al., 1998) and the glutamatergic synapses of human STXBP1 heterozygous 414 knockout neurons (Patzke et al., 2015). The reduced synaptic strength is likely due to a decrease 415 in the number of readily releasable vesicles or release probability given the crucial role of Stxbp1 416 in synaptic vesicle priming and fusion (Rizo and Xu, 2015). On the other hand, the reduction in 417 the connectivity of Sst interneuron synapses is unexpected, as Stxbp1 has not yet been implicated 418 in the formation or maintenance of synapses. Complete loss of Stxbp1 in mice does not appear to 419 affect the initial formation of neural circuits, but causes cell-autonomous neurodegeneration and 420 protein trafficking defects (Law et al., 2016;Verhage et al., 2000). Since Munc13-1/2 double 421 knockout mice also lack synaptic exocytosis, but do not show neurodegeneration (Varoqueaux et 422 al., 2002), the degeneration phenotype in Stxbp1 null mice is unlikely caused by the total arrest 423 of synaptic exocytosis. Thus, Stxbp1 may regulate other intracellular processes in addition to 424 presynaptic transmitter release, and we speculate that it may be involved in a protein trafficking 425 process important for the formation or maintenance of Sst interneuron synapses. Nevertheless, 426 the impairment of Pv and Sst interneuron-mediated inhibition likely constitutes a key mechanism 427 underlying the cortical hyperexcitability and neurobehavioral phenotypes of Stxbp1 428 haploinsufficient mice. Future studies using cell-type specific Stxbp1 haploinsufficient mouse 429 models will help determine the role of GABAergic interneurons in the disease pathogenesis. 430 431 There are over 100 developmental brain disorders that arise from mutations in postsynaptic 432 proteins, whereas mutations in much fewer presynaptic proteins have been identified to cause 433 neurodevelopmental disorders (Bayés et al., 2011;Deciphering Developmental Disorders Study, 434 2017). However, in addition to STXBP1, pathogenic variants in other key components of the 435 presynaptic neurotransmitter release machinery were recently discovered in neurodevelopmental 436 disorders. These include Ca 2+ -sensor synaptotagmin 1 (SYT1), vesicle priming factor unc-13 437 homolog A (UNC13A), and all three components of the neuronal SNAREs, syntaxin 1B 438 (STX1B), synaptosome associated protein 25 (SNAP25), and vesicle associated membrane 439 protein 2 (VAMP2) (Baker et al., 2015;Engel et al., 2016;Fukuda et al., 2018;Hamdan et 440 al., 2017;Lipstein et al., 2017;Rohena et al., 2013;Salpietro et al., 2019;Schubert et al., 2014;441 Shen et al., 2014;Wolking et al., 2019). Haploinsufficiency of these synaptic proteins is likely 442 the leading disease mechanism because the majority of the cases were caused by heterozygous Flpo mice (Raymond and Soriano, 2007) to remove the trapping cassette in the germline. The 459 resulting offspring were then crossed to Sox2-Cre mice (Hayashi et al., 2002)  ires-Cre (Hippenmeyer et al., 2005), Sst-ires-Cre (Taniguchi et al., 2011), and Rosa26-CAG-472 LSL-tdTomato (Madisen et al., 2010)  MfeI for the 3' probe (Figure 1-supplement 1A). 32 P-labeled probes were used to detect DNA 490 fragments. For Western blots, proteins were extracted from the brains at embryonic day 17.5 or 3 491 months of age using lysis buffer containing 50 mM Tris-HCl (pH 7.6), 150 mM NaCl, 1 mM 492 EDTA, 1% Triton X-100, 0.5% Na-deoxycholate, 0.1% SDS, and 1 tablet of cOmplete™, Mini, 493 EDTA-free Protease Inhibitor Cocktail (Roche) in 10 ml buffer. Stxbp1 was detected by a rabbit 494 antibody against the N terminal residues 58-70 (Abcam, catalog # ab3451, lot #GR79394-18, 495 1:2,000 or 1:5,000 dilution) or a rabbit antibody against the C terminal residues 580-594 496 (Synaptic Systems, catalog # 116002, lot # 116002/15, 1:2,000 or 1:5,000 dilution). Gapdh was 497 detected by a rabbit antibody (Santa Cruz Biotechnology, catalog # sc-25778, lot # A0515, 1:300 498 or 1:1,000 dilution). Primary antibodies were detected by a goat anti-rabbit antibody conjugated 499 with IRDye 680LT (LI-COR Biosciences, catalog # 925-68021, lot # C40917-01, 1:20,000 500 dilution). Proteins were visualized and quantified using an Odyssey CLx Imager and Image 501 Studio Lite version 5.0 (LI-COR Biosciences). 502 503

Behavioral Tests 504
All behavioral experiments except the tube test were performed and analyzed blind to the 505 genotypes. Approximately equal numbers of Stxbp1 mutant mice and their sex-and age-matched 506 WT littermates of both sexes were tested in parallel in each experiment except the resident-507 intruder test where only male mice were used. In each cage, two mutant and two WT mice were 508 housed together. Before all behavioral tests, mice were habituated in the behavioral test facility 509 for at least 30 minutes. Gender effect was inspected by two-way or three-way ANOVA. If both 510 sexes showed similar phenotypes, the data were aggregated together to simplify the presentation; 511 otherwise they were presented separately. The ages of the tested mice were indicated in figures. the latency for the mouse to fall from the rod were recorded for each trial. Mice were tested in 4 525 trials per day for 2 consecutive days or in 3 trials per day for 4 consecutive days. There was a 526 30-60-minute resting interval between trials. 527 528 Dowel test: A mouse was placed in the center of a horizontal dowel (6.5-mm or 9.5-mm 529 diameter) and the latency to fall was measured with a maximal cutoff time of 120 seconds. 530 531 Inverted screen test: A mouse was placed onto a wire grid, and the grid was carefully picked up 532 and shaken a couple of times to ensure that the mouse was holding on. The grid was then 533 inverted such that the mouse was handing upside down from the grid. The latency to fall was 534 measured with a maximal cutoff time of 60 seconds. 535 536 Wire hang test: A mouse was suspended by its forepaws on a 1.5-mm wire and the latency to fall 537 was recorded with a maximal cutoff time of 60 seconds. 538 539 Foot slip test: A mouse was placed onto an elevated 40 ´ 25 cm wire grid (1 ´ 1 cm spacing) and 540 allowed to freely move for 5 minutes. The number of foot slips was manually counted, and the 541 moving distance was measured through a video camera (ANY-maze, Stoelting). The number of 542 foot slips were normalized by the moving distance for each mouse. Instruments). A mouse was held by the tail and allowed to grasp a trapeze bar with its forepaws. 550 Once the mouse grasped the bar with both paws, the mouse was pulled away from the bar until 551 the bar was released. The digital meter displayed the level of tension exerted on the bar in gram-552 force (gf). 553 554 Acoustic startle response test: A mouse was placed in a well-ventilated, clear plastic cylinder and 555 acclimated to the 70-dB background white noise for 5 minutes. The mouse was then tested with 556 4 blocks of 52 trials. Each block consisted of 13 trials, during which each of 13 different levels 557 of sound (70, 74, 78, 82, 86, 90, 94, 98, 102, 106, 110, 114, or 118 dB, 40 milliseconds, inter-558 trial interval of 15 seconds on average) was presented in a pseudorandom order. The startle 559 response was recorded for 40 milliseconds after the onset of the sound. The rapid force changes 560 due to the startles were measured by an accelerometer (SR-LAB, San Diego Instruments). 561 562 Pre-pulse inhibition test: A mouse was placed in a well-ventilated, clear plastic cylinder and 563 acclimated to the 70-dB background noise for 5 minutes. The mouse was then tested with 6 564 blocks of 48 trials. Each block consisted of 8 trials in a pseudorandom order: a "no stimulus" trial 565 (40 milliseconds, only 70-dB background noise present), a test pulse trial (40 milliseconds, 120 566 dB), 3 different pre-pulse trials (20 milliseconds, 74, 78, or 82 dB), and 3 different pre-pulse 567 inhibition trials (a 20-millisecond, 74, 78, or 82 dB pre-pulse preceding a 40-millisecond, 120-568 dB test pulse by 100 milliseconds). The startle response was recorded for 40 milliseconds after the onset of the 120-dB test pulse. The inter-trial interval is 15 seconds on average. The rapid 570 force changes due to the startles were measured by an accelerometer (SR-LAB, San Diego 571 Instruments). Pre-pulse inhibition of the startle responses was calculated as "1 -(pre-pulse 572 inhibition trial/test pulse trial)". 573 574 Hot plate test: A mouse was placed on a hot plate (Columbus Instruments) with a constant 575 temperature of 55 °C. The latency for the mouse to first respond with either a hind paw lick, 576 hind paw flick, or jump was measured. If the mouse did not respond within 45 seconds, then the 577 test was terminated, and the latency was considered to be 45 seconds. 578 579 Novel object recognition test: A mouse was first habituated in an empty arena (24 ´ 45 ´ 20 cm) 580 for 5 minutes before every trial. The habituated mouse was then placed into the testing arena 581 with two identical objects for the first three trials. In the fourth trial, the mouse was exposed to 582 the familiar object that it interacted during the previous three trials and a novel object. In the fifth 583 trial, the mouse was presented with the two original, identical objects. Each trial lasted 5 584 minutes. The inter-trial interval was 24 hours or 5 minutes. In the modified version, Stxbp1 tm1d/+ 585 and WT mice were exposed to the objects for 10 and 5 minutes during each trial, respectively. 586 The movement of mice was recorded by a video camera placed above the test arena. The amount 587 of time that the mouse interacted with the objects (T) was recorded using a wireless keyboard 588 (ANY-maze, Stoelting). The preference index of interaction was calculated as T novel object /(T familiar 589 object + T novel object ). 590 591 Fear conditioning test: Pavlovian fear conditioning was conducted in a chamber (30 ´ 25 ´ 29 cm) that has a grid floor for delivering electrical shocks (Coulbourn Instruments). A camera 593 above the chamber was used to monitor the mouse. During the 5-minute training phase, a mouse 594 was placed in the chamber for 2 minutes, and then a sound (85 dB, white noise) was turned on 595 for 30 seconds immediately followed by a mild foot shock (2 sec, 0.72 mA). The same sound and 596 foot shock were repeated one more time 2 minutes after the first foot shock. After the second 597 foot shock, the mouse stayed in the training chamber for at least 18 seconds before returning to 598 its home cage. After 1 or 24 hours, the mouse was tested for the contextual and cued fear 599 memories. In the contextual fear test, the mouse was placed in the same training chamber and its 600 freezing behavior was monitored for 5 minutes without the sound stimulus. The mouse was then 601 returned to its home cage. One to two hours later, the mouse was transferred to the chamber after 602 it has been altered using plexiglass inserts and a different odor to create a new context for the 603 cued fear test. After 3 minutes in the chamber, the same sound cue that was used in the training 604 phase was turned on for 3 minutes without foot shocks while the freezing behavior was 605 monitored. The freezing behavior was scored using an automated video-based system 606 (FreezeFrame, Actimetrics). The freeze time (%) during the first 2 minutes of the training phase 607 (i.e., before the first sound) was subtracted from the freeze time (%) during the contextual fear 608 test. The freeze time (%) during the first 3 minutes of the cued fear test (i.e., without sound) was 609 subtracted from the freeze time (%) during the last 3 minutes of the cued fear test (i.e., with 610 sound). 611 612 Y maze spontaneous alternation test: A mouse was placed in the center of a Y-shaped maze 613 consisting of three walled arms (35 ´ 5 ´ 10 cm) and allowed to freely explore the different arms 614 for 10 minutes. The sequence of the arms that the mouse entered was recorded using a video 615 camera (ANY-maze, Stoelting). The correct choice refers to when the mouse entered an alternate 616 arm after it came out of one arm. 617 618 Elevated plus maze test: A mouse was placed in the center of an elevated maze consisting of two 619 open arms (25 ´ 8 cm) and two closed arms with high black walls (25 ´ 8 ´ 15 cm) and allowed 620 to freely explore for 10 minutes in the presence of 700-750-lux illumination and 65-dB 621 background white noise. The mouse activity was recorded using a video camera (ANY-maze, 622 Stoelting). 623 624 Light-dark chamber test: A mouse was placed in a rectangular light-dark chamber (44 ´ 21 ´ 21 625 cm) and allowed to freely explore for 10 minutes in the presence of 700-750 lux illumination 626 and 65-dB background white noise. One third of the chamber is made of black plexiglass (dark) 627 and two thirds is made of clear plexiglass (light) with a small opening between the two areas. 628 The movement of the mouse was tracked by the Open Field Locomotor system (OmniTech). 629 630 Hole-board test: A mouse was placed at the center of a clear chamber (40 ´ 40 ´ 30 cm) that 631 contains a black floor with 16 evenly spaced holes (5/8-inch diameter) arranged in a 4 ´ 4 array. 632 The mouse was allowed to freely explore for 10 minutes. Its open-field activity above the 633 floorboard and nose pokes into the holes Yes,were detected by infrared beams above and below 634 the hole board using the VersaMax system (OmniTech and right) of equal size with 10 ´ 5 cm openings between the chambers. WT C57BL/6J mice 651 were used as partner mice. A test mouse was placed in the apparatus with a mesh pencil cup in 652 each of the left and right chambers and allowed to freely explore for 10 minutes. A novel object 653 was then placed under one mesh pencil cup and an age-and sex-matched partner mouse under 654 the other mesh pencil cup. The test mouse was allowed to freely explore for another 10 minutes. 655 The position of the test mouse was tracked through a video camera (ANY-maze, Stoelting), and 656 the approaches of the test mouse to the object or partner mouse were scored manually using a 657 wireless keyboard. Partner mice were habituated to the mesh pencil cups in the apparatus for 1 658 hour per day for 2 days prior to testing. A partner mouse was used only in one test per day. 659 660 Partition test: The partitioned cage is a standard mouse cage (28.5 ´ 17.5 ´12 cm) divided in half with a clear perforated partition (a hole of 0.6-cm diameter). WT C57BL/6J mice were used as 662 partner mice. A test mouse was housed in one side of the partitioned cage for overnight. In the 663 afternoon before testing, an age-and sex-matched partner mouse was placed in the opposite half 664 of the partitioned cage. On the next day, the time and number of approaches of the test mouse to 665 the partition were scored using a handheld Psion event recorder (Observer, Noldus) in three 5-666 minute tests. The first test measured the approaches with the familiar overnight partner. The 667 second measured the approaches with a novel partner mouse. The third test measured the 668 approaches with the returned original partner mouse. 669 670 Nestlet shredding test: A mouse was individually housed in its home cage, and an autoclaved 671 Nestlet was given to the mouse. The quality of the nest was assessed every 24 hours for 3 672 consecutive days. 673 674 Marble burying test: A clean standard housing cage was filled with approximately 8-cm deep 675 bedding material. 20 marbles were arranged on top of the bedding in a 4 ´ 5 array. A mouse was 676 placed into this cage and remained undisturbed for 30 minutes before returning to its home cage. 677 The number of buried marbles (i.e., at least 2/3 of the marble covered by the bedding) was 678 recorded. 679 680

Video-EEG/EMG 681
Mice at 3-4 weeks of age were anesthetized with 2.5% isoflurane in oxygen, and the body 682 temperature was maintained by a feedback based DC temperature control system at 37°C. The 683 head was secured in a stereotaxic apparatus, and an incision was made along the midline to expose the skull. Craniotomies (approximate diameter of 0.25 mm) were performed with a round 685 bur (0.25-mm diameter) and a high-speed rotary micromotor at coordinates (see below) that were 686 normalized by the distance between Bregma and Lambda (DBL). Perfluoroalkoxy polymer 687 (PFA)-coated silver wire electrodes (127-µm bare diameter, 177.8-µm coated diameter, A-M 688 Systems) were used for grounding, referencing, and recording. A grounding electrode was placed 689 on the right frontal cortex. An EEG reference electrode was placed on the cerebellum. Three EMG reference electrode were inserted into the neck muscle. All the electrodes were connected 694 to an adapter that was secured on the skull by dental acrylic. The skin around the wound was 695 sutured, and mice were returned to the home cage to recover for at least one week. Before 696 recording, mice were individually habituated in the recording chambers (10-inch diameter of 697 plexiglass cylinder with bedding and access to food and water) for 24 hours. EEG/EMG signals 698 were sampled at 5000 Hz with a 0.5-Hz high-pass filter, and synchronous videos were recorded 699 at 30 frames per second from freely moving mice for continuous 72 hours using a 4-channel 700 EEG/EMG tethered system (Pinnacle Technology). 701

702
To detect spike-wave discharges (SWDs), EEG signals of each channel were divided into 10-703 minute segments, and each segment was filtered by a third order Butterworth bandpass filter with 704 0.5-400 Hz cutoffs. The filtered data was divided into 250-millisecond non-overlapping epochs. ; f1 = 100; f2 = 300; 709 f3 = 0.5; f4 = 80) where the power of the upper band (100-300 Hz) was contrasted with that of 710 the lower band (0.5-80 Hz). The above features were computed in MATLAB. An EEG segment 711 that exceeded thresholds for all of the above features was identified as a SWD candidate. The 712 candidates were further classified by a convolutional neural network in Python that was trained 713 with manually labeled EEG segments. The first layer of the network contained 32 filters that 714 returned their matches with 10-millisecond (kernel size) non-overlapping (stride) candidate 715 segments across the three EEG channels. Successive convolutional layers were stacked 716 sequentially. For every two consecutive convolutional layers, there was a pooling layer that 717 down-sampled the outputs by a factor of 5 to reduce computation. The overall network consisted 718 of two layers of 32 filters, one layer of pooling, two layers of 64 filters, one layer of pooling, two 719 layers of 128 filters, and one layer of pooling. The network was trained through an iterative 720 approach. In each training iteration, the optimizer (Adadelta) updated the weights of the filters, 721 and the loss function (binary cross entropy) evaluated how well the network predicted SWDs. 722 This iteration process continued until the loss function was minimized. Methods implemented to 723 reduce overfitting included dropout (i.e., 50% of the neurons were randomly dropped out from 724 calculation for each iteration) and early stopping (i.e., training process was stopped when the loss 725 function on validation set did not decrease for 3 iterations). The trained neural network removed 726 99% of the false-positive candidates and the remaining candidates were further confirmed by 727 visual inspection. For each SWD, the duration (the time difference between the first and last 728 peaks) and spike rate were quantified. The SWD cluster was defined as a cluster of 5 or more 729 SWD episodes that occurred with inter-episode-interval of maximal 60 s. 730 To identify myoclonic seizures, we visually inspected the EEG/EMG traces and videos to 732 identify sudden jumps and myoclonic jerks. When the mouse suddenly and quickly move the 733 body in less than one second, if one or more limbs leave the cage floor, then this is classified as a 734 sudden jump. If all limbs stay on the cage floor, then this is classified as a myoclonic jerk. The 735 state of the mouse right before the myoclonic seizure was classified as REM sleep,NREM sleep,736 or awake based on the EEG/EMG. Data were analyzed offline using AxoGraph X (AxoGraph Scientific). 768 769 Neuronal intrinsic excitability was examined with the K + -based pipette solution. The resting 770 membrane potential was recorded in the whole-cell current clamp mode within the first minute 771 after break-in. After balancing the bridge, the input resistance was measured by injecting a 500-772 ms hyperpolarizing current pulse (10-100 pA) to generate a small membrane potential 773 hyperpolarization (2-10 mV) from the resting membrane potential. Depolarizing currents were 774 increased in 5-or 10-pA steps to identify rheobase currents. 775 Postsynaptic currents were recorded in the whole-cell voltage clamp mode with the Cs + -based 777 patch pipette solution. Only recordings with series resistance below 20 MΩ were included. 778 IPSCs were recorded at the reversal potential for excitation (+10 mV). To record unitary 779 connections between inhibitory interneurons and pyramidal neurons, Pv and Sst interneurons 780 were identified by the Cre-dependent expression of tdTomato. Pyramidal neurons were first 781 recorded in whole-cell voltage clamp mode (+10 mV) with the Cs + -based patch pipette solution, 782 and a nearby Pv or Sst interneuron was subsequently recorded in the whole-cell current clamp 783 mode with the K + -based patch pipette solution. Action potentials were elicited in Pv or Sst 784 interneurons by a 2-ms depolarizing current step (1-2 nA) with 10 s inter-stimulus intervals. 785 Unitary IPSC (uIPSC) amplitudes were measured from the average of 30-50 sweeps. We 786 considered a Pvalb or Sst interneuron to be connected with a pyramidal neuron when the average 787 uIPSC amplitude was at least three times the baseline standard deviation. 788

789
Statistics 790 All reported sample numbers (n) represent biological replicates that are the numbers of tested 791 mice or recorded neurons. Statistical analyses were performed with Prism 6 (GraphPad 792 Software). D'Agostino-Pearson, Shapiro-Wilk, and Kolmogorov-Smirnov tests were used to 793 determine if data were normally distributed. If all data within one experiment passed all three 794 normality tests, then the statistical test that assumes a Gaussian distribution was used. Otherwise, 795 the statistical test that assumes a non-Gaussian distribution was used. All statistical tests were 796 two-tailed with an alpha of 0.05. The details of all statistical tests, numbers of replicates and 797 mice, and P values were reported in Supplementary Table 2 graphs are mean ± s.e.m. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001. 843 genotypes of the offspring at weaning (i.e., around the age of 3 weeks) are shown in the pie 866 charts. The male, female, and total Stxbp1 tm1d/+ and Stxbp1 tm1a/+ mice were significantly less than 867 Mendelian expectations. Note that the genotypes of some female mice were not determined and 868 therefore, they were not included in this analysis. (B) Survival curves of a subset of Stxbp1 tm1d/+ , 869 Stxbp1 tm1a/+ , and WT mice that were monitored for 80 weeks. The numbers of observed mice are 870 indicated in the figures. n.s. P > 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001. 871 872 Figure 1-supplement 3. Normal performance of Stxbp1 tm1d/+ mice in rotarod, dowel,