Nav1.2 haplodeficiency in excitatory neurons causes absence-like seizures in mice

Mutations in the SCN2A gene encoding a voltage-gated sodium channel Nav1.2 are associated with epilepsies, intellectual disability, and autism. SCN2A gain-of-function mutations cause early-onset severe epilepsies, while loss-of-function mutations cause autism with milder and/or later-onset epilepsies. Here we show that both heterozygous Scn2a-knockout and knock-in mice harboring a patient-derived nonsense mutation exhibit ethosuximide-sensitive absence-like seizures associated with spike-and-wave discharges at adult stages. Unexpectedly, identical seizures are reproduced and even more prominent in mice with heterozygous Scn2a deletion specifically in dorsal-telencephalic (e.g., neocortical and hippocampal) excitatory neurons, but are undetected in mice with selective Scn2a deletion in inhibitory neurons. In adult cerebral cortex of wild-type mice, most Nav1.2 is expressed in excitatory neurons with a steady increase and redistribution from proximal (i.e., axon initial segments) to distal axons. These results indicate a pivotal role of Nav1.2 haplodeficiency in excitatory neurons in epilepsies of patients with SCN2A loss-of-function mutations.

V oltage-gated sodium channels in neurons play essential roles in the generation and propagation of action potentials. These channels consist of one pore-forming α subunit and one or two accessory β subunits. In the mammalian brain, four α subunits, namely, Nav1.1, 1.2, 1.3, and 1.6 encoded by SCN1A, 2A, 3A, and 8A, respectively, are expressed at high levels.
Mutations in voltage-gated sodium channel genes have been described in patients with a wide spectrum of neurological disorders including epilepsy. The first mutation in SCN2A was discovered in a patient with atypical generalized epilepsy with febrile seizures plus 1 . Subsequently, inherited SCN2A mutations were found in families with benign familial neonatal-infantile seizures 2,3 . We further reported a nonsense mutation SCN2A-R102* (RX) in a patient with epileptic encephalopathy, autism spectrum disorder (ASD) and intellectual disability 4 , which was the first report of a de novo SCN2A mutation in a patient with ASD or intellectual disability. Subsequently, we and others reported a number of de novo SCN2A mutations in patients with neurological disorders such as epileptic encephalopathy including Ohtahara syndrome, West syndrome, Lennox Gastaut syndrome [5][6][7][8][9] , ASD 10,11 , intellectual disability 12,13 , and schizophrenia 14,15 . Recent large-scale whole exome sequencing studies further revealed that SCN2A is the gene showing the most frequent and common de novo mutations among these patients [16][17][18][19][20] .
Although mutations of SCN1A have also been described in patients with epileptic encephalopathy, intellectual disability, and ASD 8,9,18,21,22 , the distributions of Nav1.1 and Nav1.2 are highly distinct from each other in brain. In neocortex, hippocampus and cerebellum, Nav1.1 is dominantly expressed in parvalbuminpositive GABAergic neurons such as fast-spiking (FS) basket cells and Purkinje cells in their axonal features [23][24][25] , while Nav1.2 is robustly expressed in glutamatergic neurons including most neocortical pyramidal cells at their axon initial segments [26][27][28][29] . Nav1.2 is densely expressed in unmyelinated axons of neurons in hippocampal dentate and cerebellar granule cells 30,31 , although in striatum Nav1.2 is present at unmyelinated axons of GABAergic medium spiny neurons 32 . We recently reported that Nav1.1 and Nav1.2 are expressed in a mutually exclusive manner not only in neocortex, hippocampus, and cerebellum, but also in striatum, where medium spiny neurons are Nav1.2-positive and presumed FS inhibitory interneurons are Nav1.1-positive 29 . In globus pallidus, all GABAergic neurons are Nav1.1-positive and the dense Nav1.2 signals are derived from axonal fibers of striatal medium spiny neurons 29 . Li and colleagues 33 reported that Nav1.2 is expressed in neocortical somatostatin-positive inhibitory neurons but not in parvalbumin-positive neurons. However, we found that Nav1.2 is expressed in caudal ganglionic eminence-derived vasoactive intestinal peptide-positive or reelin-positive/somatostatin-negative inhibitory neurons in neocortex and hippocampus, but is not expressed in parvalbumin or somatostatin-positive neurons, which are medial ganglionic eminence-derived inhibitory neurons 29 .
Contrary to SCN1A loss-of-function mutations in patients with severe epilepsies such as Dravet syndrome 22,34 , SCN2A gain-offunction (increased or accelerated, but not toxic) has recently been recognized as a cause of early infantile-onset severe epileptic encephalopathies such as Ohtahara syndrome, whereas loss-offunction SCN2A mutations underlie ASD or intellectual disability with later-onset mild epilepsy or without epilepsy 22,35,36 . Given that the predominant expression of Nav1.1 is in inhibitory neurons and that of Nav1.2 is in excitatory neocortical/hippocampal neurons, it seems reasonable that SCN1A loss-or SCN2A gain-of-function mutations lead to epilepsies. However, it still remains unclear why SCN2A loss-of-function mutations also cause epilepsies.
A mouse transgenic line Scn2a Q54 exhibiting partial seizures originating in hippocampus 37 has long been used as a model for diseases caused by SCN2A mutations. This mouse model harbors a GAL879-881QQQ Scn2a gain-of-function mutation and the mutant protein is ectopically expressed under the control of the rat promoter for a neuron-specific enolase gene, while intrinsic Scn2a genes remain intact. In mice with genuine Scn2a deficiency, no epileptic seizures have been described so far 38 .
In this study, we discovered that Scn2a haploinsufficient mice show a mild spontaneous epileptic phenotype of absence-like seizures. Contrary to the previous proposal that loss-of-function Scn2a mutations may reduce excitability of Nav1.2 expressing inhibitory neurons and thereby lead to epileptic seizures 33 , we show here that the epileptic phenotypes in mice with Scn2a deficiency depend on Nav1.2 deficiency in excitatory neurons, suggesting critical contributions of impaired functions of excitatory neurons to the pathophysiology of epileptic seizures associated with SCN2A mutations.

Results
Absence-like seizures in Nav1.2 haploinsufficient mice. We have previously suggested that a truncated non-functional peptide (Nav1.2-RX) consisting of the N-terminal 101 amino acid residues might cause dominant negative Nav1.2 suppression leading to intractable seizures in a patient with the Scn2a-RX mutation 4 . In order to test this hypothesis, we generated knock-in mice carrying the RX mutation ( Fig. 1a and Supplementary Fig. 1a-c) and compared their phenotypes with those of previously reported Scn2a-knockout (KO) mice carrying the disrupted exon 1 with an insertion of neo cassette 38 . Western blot analyses of whole brain lysate at postnatal day (P) 0.5, using a newly-generated anti-Nterminal-Nav1.2 antibody (EM1) and an anti-internal (ASC-002) Nav1.2 antibody showed that wild-type Nav1.2 was highly expressed in wild-type (Scn2a +/+ ) mice, moderately expressed in heterozygote (Scn2a RX/+ ) mice and negligibly expressed in homozygote (Scn2a RX/RX ) mice ( Fig. 1b and Supplementary  Fig. 1d). Meanwhile, truncated Nav1.2-RX was undetected in Scn2a RX/+ and Scn2a RX/RX mice (Fig. 1c), suggesting that the mutated Scn2a allele was inactivated, presumably by nonsensemediated mRNA decay. Western blot analyses using the anti-pan Nav1 antibody (SP19) also showed reduced expression levels of Nav1 (total voltage-gated sodium channel alpha-subunits) in Scn2a RX/RX mice, compared with Scn2a +/+ mice ( Supplementary  Fig. 1e), consistent with inactivation of the Scn2a mutated allele. Like homozygous Scn2a KO (Scn2a KO/KO ) mice 38 , Scn2a RX/RX pups were born in the approximately expected Mendelian ratios, but all died within two days after birth (Fig. 1d). Scn2a RX/+ mice were viable and fertile, and had normal life spans, as observed in Scn2a KO/+ mice 38 .
We next investigated whether Scn2a KO/+ mice also have absence-like seizures. While visual inspections did not reveal any discernible behavioral seizures in Scn2a KO/+ mice, consistent with a previous study 38 , ECoG-EMG recordings from Scn2a KO/+ mice at 10-27 weeks of age detected absence-like seizures with SWDs associated with EMG suppression (Fig. 3a) as observed in Scn2a RX/+ mice. ECoG monitoring further revealed two prolonged non-convulsive seizures with duration of 30-45 s in 2 out of 6 Scn2a KO/+ mice examined (Fig. 3b). 3-h ECoG-multisite local field potential (LFP) recordings showed that Scn2a KO/+ mice had a greater incidence of ECoG SWDs than Scn2a +/+ mice (ECoG on somatosensory cortex, prevalence rate: Fisher's exact test, P = 0.1492, hourly incidence: Mann-Whitney test, U = 0, *P = 0.0286) (Fig. 3c, d). Moreover, ECoG-multisite LFP recordings revealed the predominant appearance of LFP epileptiform discharges in medial prefrontal cortex (mPFC) and caudate putamen (CPu) (Fig. 3c, d). While the incidence and duration of epileptiform discharges did not significantly differ between Scn2a RX/+ and Scn2a KO/+ mice (hourly incidence: Scn2a RX/+ , . Given that incidence and duration of SWD episodes were 1.018 ± 0.552 episodes per hour and 0.71 ± 0.04 sec in Scn2a RX/+ mice and 1.417 ± 0.417 episodes per hour and 0.58 ± 0.05 sec in Scn2a KO/+ mice, the percentage of total recording period displaying seizures was calculated to be around 0.02% for both lines. The susceptibility to hyperthermia-induced seizure was not altered in Scn2a KO/+ compared to Scn2a +/+ mice (Fig. 3e). Analysis of susceptibility to PTZ (50 mg per kg or 25 mg per kg) induced seizures showed a shorter latency to the appearance of absence seizure-like sudden immobility, myoclonus, and clonic convulsion in Scn2a KO/+ than in Scn2a +/+ mice (Fig. 3f, g) the basal expression levels of other sodium channel subunits (Fig. 4). Taken together, these results suggest that Nav1.2 haploinsufficiency is the pathological basis for the absence-like seizures in these mice.
Redistribution of Nav1.2 during brain development. Western blot analyses showed that Nav1.2 expression was already detectable on embryonic day 14.5 and steeply up-regulated during postnatal development ( Supplementary Fig. 4). At 6-8 weeks of age, Nav1.2 expression in 6-week-old neocortex and hippocampus showed 50% reduction in Scn2a fl/+ /Emx1-Cre and no significant alterations in Scn2a fl/+ /Vgat-Cre, compared with Scn2a fl/+ mice (Fig. 6), indicating that most Nav1.2 is expressed in excitatory neurons in neocortex and hippocampus at adult stages. We further investigated the temporal and spatial expression pattern of Nav1.2 by immunohistochemistry. Although two commercial (ASC-002, G-20) and one original (EM-1) Nav1.2 antibodies displayed similar staining patterns ( Supplementary Fig. 5) and their immunosignals disappeared in Scn2a KO/KO when compared with Scn2a +/+ mice ( Supplementary Fig. 6), the goat antibody G-20 gave the clear signal and allowed double staining with rabbit or mouse antibodies, and was selected for subsequent immunohistochemical analyses. At P0.5, immunosignals for ankyrinG were detected in both Scn2a KO/ KO and Scn2a +/+ mice, suggesting that AIS structures in Scn2a KO/ KO mice remain largely intact ( Supplementary Fig. 7). Immunohistochemistry of Nav1.2 in wild-type mice detected immunosignals throughout the central nervous system with drastic changes in signal intensity and subcellular distribution from neonatal period to adulthood ( Fig. 7 and Supplementary Figs. 4b and 8). During this period, the intensity of Nav1.2-immunosignals grew in unmyelinated fibers, including mossy fibers of hippocampal dentate granule cells ( Fig. 7p-t) and cerebellar parallel fibers (Supplementary Fig. 8a-d). Nav1.2 signals also appeared and grew more intense in unmyelinated AIS segments of neocortical and hippocampal pyramidal cell axons ( Fig. 7f-j, p-t) and nodes of Ranvier ( Fig. 7k-o). At P15.5 in neocortex and hippocampus, Nav1.2 was strongly observed at the ankyrinG-positive AISs of excitatory neurons (Fig. 8). Consistent with previous studies 26,33 , the AISs of P15.5 neocortical pyramidal cells expressed Nav1.2-immunoreactivity in their proximal part and Nav1.6-immunoreactivity in 150 ***  Fig. 4 Scn2a haplodeficiency did not alter expression levels of other sodium channel subunits. Quantitative RT-PCR analyses of brain mRNAs prepared from P14.5 Scn2a +/+ and Scn2a KO/+ mice (N = 11, each genotype). Scn2a mRNA expression in Scn2a KO/+ whole brain was reduced to about 50% level of that in Scn2a +/+ mice while there were no significant changes in Scn1a, Scn3a, Scn5a, Scn8a, Scn1b, Scn2b, Scn3b, and Scn4b mRNA expression levels in  Fig. 9). It is noteworthy that diffuse Nav1.2-immunoreactivity assumed to be distal axonal features throughout neocortex and hippocampus continued to become stronger until adulthood (Fig. 7), while such diffuse signals of Nav1.6 were not observed ( Supplementary Fig. 10). These observations suggest that Nav1.2, rather than Nav1.6, is the major voltage-gated sodium channel at more distal axonal sites of excitatory neurons. To localize Nav1.2 in the cerebral cortex at adult stage, we employed pre-embedding silver-enhanced immunogold electron microscopy ( Supplementary Fig. 11). Most metal particles binding Nav1.2 were detected on the cell membrane of thin processes, presumably distal unmyelinated portions of preterminal axon. By contrast, only a few particles were detected in axon terminals, and none at myelinated portions of axons.
Reduced action potential amplitude in excitatory cells. Our previous study of voltage-clamp analysis of cultured Scn2a KO/+ hippocampal pyramidal neurons after 5-9 days in vitro showed ã 45% reduction in maximum sodium conductance density 38 . Here we further investigated the properties of action potentials of Scn2a KO/+ excitatory and inhibitory neurons in neocortical layer II/III and hippocampal CA1 region using current-clamp recordings of acute mouse brain slices. To discern between excitatory and inhibitory neurons, we crossed Scn2a KO/+ with a Vgat-Venus line that expresses green fluorescent proteins selectively in inhibitory neurons 44 .
In cortical pyramidal neurons at early postnatal ages (P7-8), peak amplitudes in single action potentials and spike trains were lower in Scn2a KO/+ than in Scn2a +/+ mice (Fig. 9a, b,  Supplementary Fig. 12a, b and Supplementary Table 1). Half widths in single action potentials and spike trains were broader in Scn2a KO/+ than in Scn2a +/+ mice at P7-8 (Fig. 9a, c and Supplementary Table 1). Maximum rates of rise of action potential upstroke at threshold in response to different holding membrane potentials were lower in Scn2a KO/+ than in Scn2a +/+ mice at P7-8 (Fig. 9d), consistent with the lowered voltage-gated sodium channel current density in hippocampal cultured neurons from Scn2a KO/+ mice 38,45 . At P15-22, peak amplitudes in cortical pyramidal neurons were similar between the genotypes (Fig. 9e, f, Supplementary Fig. 12c, d and Supplementary Table 1), whereas half width was again broader in Scn2a KO/+ than in Scn2a +/+ (Fig. 9e, g and Supplementary Table 1). In hippocampal pyramidal cells, peak amplitudes were lower in Scn2a KO/+ than Scn2a +/+ mice at P7-8 but not at P15-20 (Supplementary Table 1). Half widths in single action potentials and spike trains were similar among the genotypes at both age groups (Supplementary Table 1). In contrast to these excitatory neurons, neocortical and hippocampal FS inhibitory neurons showed no significant differences in the electrophysiological properties between the genotypes at any age groups examined (Fig. 9h-m,  Supplementary Fig. 13 and Supplementary Table 1). These results suggest that excitatory neural activity is primarily impaired whereas the FS inhibitory activity remains unchanged in Scn2a KO/+ mice.

Discussion
Here we demonstrated ethosuximide-sensitive absence-like seizures with bilateral SWDs in mice heterozygous for a patientderived Scn2a-RX nonsense knock-in mutation originally described in a patient with epileptic encephalopathy, intellectual disability, and ASD 4 . These same phenotypes were also observed in this study in a previously generated KO mutant mouse 38 . We further revealed that the RX mutation did not produce a truncated Nav1.2-RX peptide, but rather inactivated the mutated Scn2a allele, leading to Nav1.2 haplodeficiency in Scn2a RX/+ mice and presumably in the patient. Although we previously suggested a dominant-negative effect of Nav1.2-RX truncated protein 4 and actually SWDs were rather prominent in Scn2a RX/+ mice compared to Scn2a KO/+ mice and itself may suggest a modifying effect of a minor amount of the Nav1.2 truncated protein which was not detectable in our western blot analysis, the incidence and duration of epileptiform discharges still did not significantly differ between Scn2a RX/+ and Scn2a KO/+ mice. This indicates that Nav1.2 haploinsufficiency is the major underlying basis for epileptic seizures. We further showed that a selective Scn2a deletion in dorsal-telencephalic excitatory neurons in mice (Scn2a fl/ + /Emx1-Cre) reproduced the absence-like seizures, whereas mice with a global Scn2a deletion in inhibitory neurons (Scn2a fl/ + /Vgat-Cre) showed no discernable epileptic abnormalities. Contrary to the suggestion that loss of functional Nav1.2 in inhibitory neurons may contribute to the pathogenesis of epileptic seizures in patients with SCN2A mutations 33 , our findings indicate that Nav1.2 haplodeficiency in excitatory neurons causes epilepsy.
Scn2a fl/fl /Emx1-Cre and Scn2a fl/fl /Vgat-Cre mice both died prematurely, suggesting that expression of Nav1.2 in either excitatory or inhibitory neurons is essential for postnatal viability. Nav1.2 expression in Emx1-Cre-positive excitatory neurons is robust in brain subregions, such as cerebral cortex, olfactory bulb and hippocampus 29 , and is estimated to explain~30% of the total Nav1.2 amount in whole brain at P0.5. Similarly, the amount of Nav1.2 expressed in inhibitory neurons at P0.5 is estimated to bẽ 60% of the whole-brain Nav1.2 amount. We have shown that Nav1.2 is localized at AISs in vasoactive intestinal peptidepositive or reelin-positive/somatostatin-negative inhibitory neurons in neocortex and hippocampus, and distributed in unmyelinated axons of GABAergic medium spiny neurons in striatum 29,32 . Although the reason(s) for the premature death in these mice is unknown, given that perinatal death in Scn2a KO/KO seems to be associated with severe hypoxia and massive neuronal apoptosis in brainstem 38 , Nav1.2 deficiency in cortico-brainstem projections or local inhibitory circuits in cerebral cortex and brainstem presumably contributes to premature death.
SCN2A mutations have been described in patients with a wide spectrum of epilepsies, intellectual disability and ASD. SCN2A mutations in patients with the severe end of epilepsies such as early-infantile epileptic encephalopathy, Ohtahara syndrome and West syndrome are almost exclusively missense, while nonsense, frameshift and splice site mutations are dominant in patients with ASD and intellectual disability associated with milder, later-onset epilepsy or without epilepsy (reviewed in Yamakawa 22 ). Recent patch-clamp analyses confirmed that Nav1.2 channels with missense mutations found in patients with early-infantile severe epilepsies had gain-of-function effects, while mutations found in patients with ASDs or late-onset epilepsies had loss-of-function effects 35,36 . Although the epileptic phenotype of the patient with SCN2A-R102* mutation who showed intellectual disability and ASD was rather severe 4 , it was milder and later-onset compared to those of early-infantile epileptic encephalopathy. These observations suggest that Scn2a-deficient mice are models for ASD and intellectual disability with milder epilepsies rather than early-infantile epileptic encephalopathy, and that Scn2a knock-in mice with gain-of-function missense mutations are models for early-infantile epileptic encephalopathy. In fact, Scn2a RX/+ and Scn2a KO/+ mice did not show spontaneous convulsive seizures but only mild absence-like seizures with SWDs. Similarly, the patient carrying the SCN2A-R102* nonsense mutation showed absence and atonic seizures 4 . The patient with ASD and intellectual disability harboring the splice site mutation, putatively SCN2A-K90Vfs*9, was also reported to have behavioral episodes characterized by a stone faced expression and limp posture, suggestive of absence seizures 11 . Absence or absence-like epilepsies in patients with ASD and intellectual disability is consistent with absence-like seizures in mice with Scn2ahaploinsufficiencies. The Scn2a Q54 mouse, long considered a model of diseases with SCN2A gain-of-function mutations, showed partial seizures originating from hippocampus 37 . In contrast, our ECoG recordings of Scn2a RX/+ , Scn2a KO/+ , and Scn2a fl/+ /Emx1-Cre mice all showed absence-like seizures with SWDs. It is plausible that lossof-function and gain-of-function Scn2a mutations culminate in different seizure phenotypes with specific patterns of epileptiform discharges. However, Nav1.2 expression in the Scn2a Q54 mouse was driven by the neuron-specific enolase promoter and it may lead to ectopic expression and other epistatic effects. Direct comparisons with knock-in mice bearing Scn2a missense  46,47 . We showed that Scn2a KO/+ mice exhibit normal susceptibility to hyperthermia-induced seizures that seems reminiscent of the temperature-independent nature of the epilepsies in a major proportion of the patient population. This contrasts with the temperature-sensitivity of epilepsies in patients with SCN1A mutations 48,49 and Scn1a-deficient mice 50,51, and rats 52 . Such disparate temperature-sensitivities of seizures in patients and animal models harboring SCN1A and SCN2A mutations may be accounted for by the distinct cellular and regional distributions for Nav1.1 and Nav1.2 rather than by distinct kinetics of the proteins 22,29 .
Our immunohistochemistry data showed that at unmyelinated fibers such as hippocampal mossy fibers, axons of striatal medium spiny neurons and parallel fibers of cerebellar granule cells, Nav1.2 is expressed and incremented through development. In contrast, Nav1.2 at AISs and nodes of Ranvier in myelinated fibers in neocortex and hippocampus became intense at~P15 where Nav1.2 and Nav1.6 were co-expressed at proximal and distal AISs respectively. Nav1.2 density in these areas gradually decreased in later stages and was replaced with Nav1.6, consistent with previous studies [26][27][28][30][31][32]53 . However, we found that diffuse Nav1.2 signals in neocortex and hippocampus (presumably in synaptic terminals) continued to increase through development. Of note, Nav1.2 was suggested to be abundantly expressed in synaptic terminals of cerebellar parallel fibers 54 . Contrary to the increase of the diffuse Nav1.2 signals, Nav1.6 did not show such increase in the corresponding brain regions, suggesting that Nav1.2 is a major voltage-gated sodium channel in axon terminals responsible for synaptic transmission at adult stages. Because of rather late onset of epilepsies in patients with SCN2A loss-of-function mutations and the reproduction of absence seizures in mice with dorsal-telencephalic excitatory neuron-specific Scn2a deletion, it would be now of interest whether the seizures are caused by Nav1.2 deficiency at the distal axons of neocortical or other dorsal-telencephalic excitatory neurons.
Our whole-cell current-clamp recordings showed that, at early postnatal stage (P7-8), neocortical and hippocampal Scn2a KO/+ excitatory pyramidal neurons displayed decreased action potential peak amplitudes by~10-15%. Nevertheless, our previous study using voltage-clamp recordings showed a~45% reduction in maximum sodium conductance density in dissociated hippocampal neurons from Scn2a KO/+ newborn mice after 5-9 days culture in vitro 38 . Similarly, acutely dissociated guinea-pig hippocampal neurons treated with a moderate dose of tetrodotoxin, a sodium channel blocker, reduced sodium current to less than half, whereas action potential amplitude was only slightly affected, suggesting a surplus of sodium channels in neurons for action potential firing 55 . However, the study showed that such surplus was required for repetitive action potential firings, and this could also be the case in Scn2a KO/+ mice. The present current-clamp recordings further showed that action potential peak amplitude recorded from neocortical and hippocampal Scn2a KO/+ excitatory neurons reached normal levels at P15-22, raising a possibility that Nav1.2 may not be needed to generate action potentials at later postnatal stage, consistent with the developmental changes in subcellular localization and distribution of Nav1.2. Alternatively, Nav1.2 haploinsufficiency may be compensated by other subtypes of voltage-gated sodium channel. No significant changes in mRNA expression levels of the other sodium channel subunit genes were observed in Scn2a KO/+ brains, excluding dosage compensation of loss of Nav1.2. Instead, Nav1.6 co-expression with Nav1.2 in the AISs of pyramidal neurons may compensate for loss of Nav1.2. Steep up-regulation of Nav1.6 expression in the AISs of pyramidal neuron between P7.5 and P15.5 could account for the significant decrement of action potential peak amplitudes in Scn2a KO/+ pyramidal neurons at P7-8, but not at P15-22. Despite the possible functional compensation for loss of Nav1.2 at AISs and nodes of Ranvier, Nav1.6 may not have sufficient compensatory effects on absence-like seizures at adult stages, presumably due to the limited expression of Nav1.6 at synaptic terminals. It has been reported that impairing voltagegated sodium channels function in FS parvalbumin-positive inhibitory neurons in mice led to epileptic seizures 25 . However, no differences were detected in the responses of neocortical FS interneurons to current injections between Scn2a KO/+ and wild-  type mice, consistent with our previous observation that Nav1.2 is not expressed in parvalbumin-positive interneurons 29 . How does the Nav1.2 haploinsufficiency in dorsaltelencephalic excitatory neurons cause epileptic seizures? One possible mechanism is abnormal activity in thalamocortical circuits, which has long been proposed as a basis for absence epilepsy 56,57 . Impaired firing properties in cortical excitatory neurons may impact their downstream input onto inhibitory neurons in the thalamic reticular nucleus, which in turn fails to suppress excitatory thalamocortical relay neurons. The excited thalamocortical neurons may then provide feedback inhibition through the thalamic reticular nucleus inhibitory neurons and generate thalamocortical hyper-synchronous oscillations that result in absence seizures 57 . P/Q-type calcium channel gene CACNA1A mutations have been described in patients with absence epilepsy 58 , and Cacna1a KO mice have absence epilepsies 41 . Recently, we reported that a selective Cacna1a gene deletion in cortical layer VI pyramidal cells, which innervate thalamic relay neurons and reticular thalamic neurons, caused upregulation of T-type calcium current in thalamic relay neurons and resulted in absence epilepsy in mice 59 , suggesting that impaired cortical excitatory input to thalamic regions may cause rebound burst of thalamocortical relay neurons and leads to thalamocortical hyper-synchronous oscillations as may be the case for ethosuximide-sensitive absence-like seizures in Scn2adeficient mice. A second plausible mechanism considers a rebound hyper-excitability within neocortex. In neocortex, excitatory, and inhibitory neurons form reciprocal and highly complex networks. Impaired excitatory inputs into inhibitory neurons may even disinhibit the downstream excitatory neurons and the consequent epileptic discharges widely spread to the corticothalamic circuit. Meeren and colleagues 60 actually reported that somatosensory cortex is the initiation site of epileptic activity in a rat model of absence epilepsy. Thirdly, the predominant appearance of epileptiform discharges in CPu of Scn2a KO/+ mice and Scn2a fl/+ /Emx1-Cre mice may implicate CPu in the pathology of absence-like seizures. CPu is the largest compartment of Black filled circles and gray filled circles represent WT and Scn2a KO/+ neurons, respectively. Data represent means ± SEM, *P < 0.01, ***P < 0.001. Scale bars: (a, e, h, k) vertical 10 mV; horizontal 10 milliseconds basal ganglia that receive excitatory inputs from cortex and thalamus and send output back to the thalamus, and cortex via thalamus. Basal ganglia were suggested to modulate the occurrence of SWDs generated in the thalamocortical circuits 61 . The present ECoG recordings also revealed an increased incidence of SWDs in Scn2a fl/+ /Emx1-Cre mice, compared to Scn2a-haploinsufficient mice. We surmise that the remaining Nav1.2 expression in non-dorsal telencephalic regions such as CPu 32 of Scn2a fl/+ /Emx1-Cre mice may have aggravating effects on absence-like seizures. Alternatively, given Nav1.2 expression in caudal ganglionic eminence-derived vasoactive intestinal peptideor reelin-positive inhibitory neurons of neocortex and hippocampus 29 , selective Nav1.2 elimination in excitatory neurons may shift an excitation/inhibition balance toward inhibition, which enhances hyperpolarization of neurons and low threshold rebound burst-firing via de-inactivation of T-type calcium currents, increasing the risk of SWD generation. Furthermore, it is possible that the decreased neuronal activity caused by the Nav1.2 deficit affects network formation during development. Congenital Scn2a mutations may impair or affect maturation, migration, or innervations of inhibitory or excitatory neurons and alter global brain network excitability. Thus, the circuit basis for the development of absence-like seizures in mice with Scn2a-deficiency requires further study.
In summary, we showed that Scn2a-haploinsufficiency in mice gives rise to absence-like seizures in a dorsal telencephalic excitatory neuron-dependent manner. Our findings should contribute to understanding of the pathomechanisms of epilepsies in patients with SCN2A mutations and absence epilepsy itself, the mechanism of which is still not fully elucidated.

Methods
Animals. All animal experimental protocols were approved by the Animal Experiment Committees of RIKEN Institute and Shizuoka Institute of Epilepsy and Neurological Disorders.
Generation of knock-in Scn2a mice with the R102* mutation. We isolated the PAC clones 348A2 and 386F4 by screening a pooled mouse genomic PAC library (BACPAC Resource Center, Oakland, CA, USA) with dot blot hybridization using [α-32P] dCTP-labeled DNA corresponding to the genomic fragment containing exon 2 of the mouse Scn2a gene as the probe. A 400 bp PmaCI fragment of a PAC clone was inserted into the blunt-end filled BglII site of pEGFP-C2 (TaKaRa Bio, Shiga, Japan) to obtain pL1. Next, a 5 kb Eco105I-AflIII fragment of a PAC clone was inserted into the AflIII and blunt-end filled SacI sites of pL1 to generate pL2. Then, a SalI fragment of pMCDTApA (a generous gift from Dr. Yagi, Osaka University) was inserted into the XhoI site of pL2 to generate pL3. In order to obtain pR1, a 3.2 kb PmaCI-EcoRI fragment of a PAC clone was inserted into the EcoRI and blunt-end filled HindIII sites of pEGFP-C2. Using the QuikChange Site-Directed Mutagenesis kit (Agilent Technologies, Santa Clara, CA, USA), the nucleotide substitution (CGG to TGA) leading to the R102* (RX) mutation was introduced into pR1. Following that, a 2.0 kb EcoRI-EcoRV fragment of a PAC clone was inserted into the EcoRI and blunt-end filled SalI sites of pR1mutated to obtain pR2. A 1.2 kb EcoRI fragment of a PAC clone was subsequently inserted into the EcoRI site of pR2 to generate pR3. In order to inactivate the SacII site in pR3, pR3 was digested with SacII, filled with T4 DNA polymerase and self-ligated. Then, an XhoI-ApaI fragment of the resulted plasmid vector was inserted into the XhoI and ApaI sites of ploxPfrtPGKneofrt to generate pR4. Finally, pR4 was digested with SacII and NotI and was inserted with a SacII-Eco52I fragment of pL3 to create the targeting vector. All constructs were verified by sequencing.
The targeting vector was digested with SacII for linearization and transfected into E14 ES cells with a Gene-Pulser (Bio-Rad, Hercules, CA, USA) at 3 µF and 800 V. Transfected ES cells were placed on neomycin-resistant, mitomycin Ctreated mouse embryonic feeder cells, and neomycin-resistant ES colons were selected in the presence of 150 µg mL −1 of Geneticin (G418, Thermo Fisher Scientific, Waltham, MA, USA). EcoRI-digested genomic DNA was isolated from individual clones and analyzed by Southern blotting using the 5′ and 3′ probes that corresponded to the genomic sequence upstream and downstream of the targeting vector. PCR analysis was performed for verifying the presence of the RX mutation. The ES cells from two correctly targeted clones (7A1 and 9H1) were injected into C57BL/6J blastocysts to produce male chimeras with greater than 50% agouti coat color, which were then bred to C57BL/6J females to obtain F1 mice heterozygous for the RX mutation. F1 heterozygotes were subsequently crossed with C57BL6 mice to generate F2 Scn2a RX+neo/+ mice that were crossed with CAG-Flpe transgenic mice in a C57BL/6J background 62 to remove the neo cassette. The absence of the neo cassette in F3 Scn2a RX/+ /CAG-Flpe mice was verified by PCR. F3 Scn2a RX/+ /CAG-Flpe mice were then crossed with C57BL/6J mice to obtain Flpe-lacking F4 Scn2a RX/+ mice. Homozygous mice were obtained by interbreeding F4 Scn2a RX/+ mice. No phenotypic differences were observed among the mice derived from the two ES cell clones. Scn2a fl/+ mice were thereafter maintained by crossing with C57BL/6J mice. Scn2a RX/+ mice on a congenic C57BL/6J background for more than 10 backcross generations were subjected to ECoG-EMG recordings. Genotyping of Scn2a knock-in mice was performed by PCR with the specific primers (forward: 5′-TGT CTC AGA TCC CCT ATT GCT -3′, reverse: 5′-CTT GGT AAC TTT GCC GAG TC -3′) that detect the wild-type allele (269 bp) and the targeted knock-in allele (484 bp) (Supplementary Fig. 1c). Unprocessed original scans of blots/gels are shown in Supplementary Fig. 14. Scn2a knockout mice. Scn2a KO mouse line has been generated by an insertion of neo cassette into exon 1 of the Scn2a gene 38 . Scn2a KO/+ mice were maintained on a congenic C57BL/6J background for more than 10 backcross generations.
Generation of Scn2a conditional knockout mice. The targeting vector harboring the floxed allele was generated from the plasmid carrying the RX allele by correcting the gene mutation in coding exon 2 and inserting two loxP cassettes on either side of exon 2 with the Red/ET recombination system (Gene Bridges, Dresden, Germany). Briefly, a genomic fragment containing wild-type exon 2 was amplified by PCR with primers having HindIII sites at their 5′ ends, whose nucleotide sequences were as follows: 5′-CCC AAG CTT CGT GTA AGG GGA AAA AGT TCT A -3′ and, 5′-CCC AAG CTT AAG TGT TGA AGG GAG TGA GTG A -3′. The resultant amplicons were digested with HindIII and inserted into the HindIII site of pfrt-PGK/gb2-neo/kan-frtloxP to generate pfrt-PGK/gb2-neo/ kan-frtloxP+exon2. Next, a genomic fragment containing downstream of exon 2 was amplified by PCR with primers having XhoI sites at their 5′ ends, whose nucleotide sequences were as follows: 5′-TTT TCT CGA GCA TTC ACT TTA GTG AGA TGG C -3′ and, 5′-TTT TCT CGA GAA AGT CCA GTG CAT GTA TG -3′. The resultant PCR products were digested with XhoI and inserted into the XhoI site of pfrt-PGK/gb2-neo/kan-frtloxP+exon2. The resulting plasmid DNA was then amplified by PCR with primers: 5′-CAT TCT GCA CGC TTC AAA G -3′ and, 5′-CAC CAT AAA GCT CAA AGG CA -3′. The PCR amplicons were used for transformation of an Escherichia coli strain, containing the targeting vector carrying the R102* allele and pSC101-BAD-gbaA (Gene Bridges). The resulting targeting vector harboring the floxed allele was verified by restriction enzyme digestion and DNA sequencing.
The targeting vector harboring the floxed allele was digested with SacII for linearization and transfected into E14 ES cells with a Gene-Pulser (Bio-Rad). The ES cells from two correctly targeted clones (2B2 and 3H12) were injected into C57BL/6J blastocysts to obtain male chimeras that were subsequently bred to C57BL/6J females to generate F1 mice heterozygous for the targeted allele. F1 heterozygotes were then crossed with CAG-Flpe transgenic mice on a C57BL/6J background 62 to generate N2 mice heterozygous (Scn2a fl/+ /CAG-Flpe) for the floxed allele and lacking the neomycin cassette. Absence of the neo-cassette was verified by PCR analysis. Scn2a fl/+ /CAG-Flpe mice were subsequently crossed with C57BL/6J mice to obtain N3 Scn2a fl/+ mice without the CAG-Flpe transgene. Absence of the CAG-Flpe transgene was verified by PCR analysis. Scn2a fl/+ mice were thereafter maintained by crossing with C57BL/6J mice. Homozygous (Scn2a fl/fl ) mice were obtained by interbreeding Scn2a fl/+ mice. No phenotypic differences were observed among the mice derived from the two ES cell clones. Genotyping of Scn2a conditional KO mice was performed by PCR with the specific primers (forward: 5′-TGT CTC AGA TCC CCT ATT GCT -3′, reverse: 5′-CCA GTA GAA CAC CAT AAA GCT CA -3′) that detect the wild-type allele (925 bp), the targeted floxed allele (1,162 bp) and the deleted allele (284 bp) ( Supplementary  Figs. 2c and 3a). Unprocessed original scans of blots/gels are shown in Supplementary Fig. 14.
EIIa-Cre, Emx1-Cre, Vgat-Cre, and Vgat-Venus mouse lines. The EIIa-Cre transgenic line, in which the Cre-loxP recombination occurs in germline cells, was previously generated by injection of the Cre cassette under the control of the adenovirus EIIa promoter into mouse zygotes 42 , and was maintained on a C57BL/ 6J background. EIIa-Cre mice were cross-mated with Scn2a fl/fl mice, and heterozygous (Scn2a fl/+ /EIIa-Cre) offspring were subsequently backcrossed with C57BL/ 6J mice to obtain Scn2a del/+ mice lacking the EIIa-Cre transgene. Absence of the EIIa-Cre transgene was verified by PCR analysis. Homozygous (Scn2a del/del ) mice were obtained by interbreeding Scn2a del/+ mice.
Empty spiracles homolog 1 (Emx1)-Cre knock-in line without a neomycin cassette was previously generated by targeted insertion of the Cre cassette into the Emx1 gene 43 , and was maintained on a C57BL/6J background. Heterozygous (Scn2a fl/+ /Emx1-Cre) mice were obtained by cross-mating Scn2a fl/fl mice and Emx1-Cre mice, and subsequently backcrossed with Scn2a fl/fl mice to obtain homozygous (Scn2a fl/fl /Emx1-Cre) mice.
Vesicular GABA transporter (Vgat, also known as vesicular inhibitory amino acid transporter, Viaat)-Cre BAC transgenic line was previously generated by pronuclear injection of the Cre cassette under the control of the mouse Vgat promoter 25 , and maintained on a C57BL/6J background. Heterozygous (Scn2a fl/ + /Vgat-Cre) mice were obtained by cross-mating Scn2a fl/fl mice and Vgat-Cre mice. Because the Scn2a and Vgat-Cre transgene alleles were mapped close together on the same chromosome 25 , the floxed Scn2a and Vgat-Cre alleles in Scn2a fl/ + /Vgat-Cre obtained by cross-mating Scn2a fl/fl mice and Vgat-Cre mice were presumed to be arranged in trans-configuration. Scn2a fl/+ /Vgat-Cre mice were then bred with C57BL/6J mice to obtain Scn2a fl/+ /Vgat-Cre mice, in which the floxed Scn2a and Vgat-Cre transgene alleles were located in cis-configuration as a result of germ-line recombination between the alleles in Scn2a fl/+ /Vgat-Cre parents. The resultant Scn2a fl/+ /Vgat-Cre offspring were subsequently crossed with Scn2a fl/fl mice to obtain homozygous (Scn2a fl/fl /Vgat-Cre) mice, whereas Scn2a fl/+ offspring were crossed with Scn2a fl/fl mice to obtain Scn2a fl/fl controls Genotyping of EIIa-Cre, Emx1-Cre and Vgat-Cre mice was performed by PCR with the Cre-specific primers (forward: 5′-AGG TTC GTT CAC TCA TGG A-3′, reverse: 5′-TCG ACC AGT TTA GTT ACC C -3′) that yield a 235-bp amplicon.
Vgat-Venus transgenic mouse line expressing a fluorescent protein specifically in inhibitory cells has been generated by pronuclear injection of the Venus cassette under the control of the mouse Vgat promoter 44 .
Antibody generation. Rabbit polyclonal anti-N-terminus-Nav1.2 antibody (EM-1) was raised against oligopeptides corresponding to the amino acids MAQSVLVPPGPDSFRFF of mouse Nav1.2 plus C at its C-terminus for coupling. Antibody was then affinity-purified using the SulfoLink kit (Thermo Fisher Scientific). Nav1.2-immunoreactivity was abolished when the antibody was preabsorbed with the immunogenic oligopeptides.
Electrocorticography-electromyography recordings. Adult mice (6-11-week-old Scn2a RX/+ and Scn2a +/+ littermate mice: 10-27-week-olds Scn2a KO/+ and Scn2a +/+ littermate mice: 6-8-week-old Scn2a fl/+ /Emx1-Cre, Scn2a fl+ /Vgat-Cre and control littermates: both sex, C57BL/6J congenic background) were used in this study. The ECoG electrodes were implanted using 1.5% halothane anesthesia with N 2 O:O 2 (1:1) ventilation. Stainless-steel screws (1.1 mm diameter), and in some experiments, Teflon-insulated silver wires served as ECoG electrodes and were secured to the skull and dura over the somatosensory cortex (1.5 mm lateral to midline, 1.0 mm posterior to bregma) and the cerebellum (at midline, 2.0 mm posterior to lambda) as a reference electrode. The EMG electrodes were placed 2-3 mm apart from the cervical region of the trapezius muscle. Recordings were obtained at least 1 week after the electrodes were implanted, to allow for full recovery. Video monitoring was simultaneously performed to detect behavioral seizures. Recordings were performed for a week, sampled at 500 Hz and then analyzed off-line using a software, SleepSign (Kissei Comtec, Nagano, Japan), blind to the genotype. Ethosuximide (33.3 mg mL −1 in saline, 200 mg per kg body weight, Toronto Research Chemicals), or vehicle (the same volume of saline) was intraperitoneally injected.
ECoG and LFP signals with 3 or more high-amplitude-spikes that were over twice more than that of the background signal in a 1-s window, were considered to be epileptiform discharges and included in the analysis.
Measurement of pentylenetetrazol-induced seizures. Adult mice (10-week-old Scn2a RX/+ , Scn2a KO/+ and Scn2a +/+ littermates, both sex, C57BL/6J congenic background) were used. Experiments of PTZ injection were performed between 9:00 a.m. and 6:00 p.m. A freshly prepared PTZ (Sigma-Aldrich) solution was administered intraperitoneally at a dosage of 25 or 50 mg per kg body weight in a total volume of 100-250 μL. After PTZ injection, the mice were placed in a clear plastic cage and observed for 30 min or earlier if the animal died before 30 min. Video monitoring system was used to determine the latency to the first behavioral seizures from the time of PTZ injection. The behavioral seizure events were determined based on the behavioral and motor activities. We classified the seizures as absence seizure-like immediate behavioral arrest, myoclonic, clonic, and generalized tonic-clonic seizures. Absence-like seizures consisted of > 2 sec behavioral arrest with or without vibrissal and facial myoclonus. Myoclonic seizures consisted of body twitching. Clonic seizures consisted of a whole body jerk with or without irregular, bilateral forelimb movements. Generalized tonic-clonic seizures consisted of fully generalized clonic seizures sometimes followed by a tonic phase involving entire body and result in death. In calculations, 1800 s were assigned for mice without seizures 30 min after PTZ administration. Latency of behavioral seizures by PTZ injection was determined in a genotype-blinded manner.
Measurement of hyperthermia-induced seizures. Mice (4-week-old Scn2a KO/+ and Scn2a +/+ littermates, male, C57BL/6J congenic background) were put on the perforated horizontal partition with 9 × 9 holes with diameter of 2.5 mm in a hermetically-enclosed plexiglass box (30 × 30 × 30 cm) and heated by blowing hot air from below using Air-Therm (World Precision Instruments, Sarasota, FL, USA). Rectal temperature was continually monitored at baseline and at seizure onset by temperature probe (IT-18; Physitemp Instruments, Clifton, NJ, USA). Before increment of air temperature, mouse was kept at least 3 min at 37°C. We gradually elevated the body temperature by 0.5°C per 2 min. When a seizure appeared, the mouse was promptly rescued by cooling with ice cubes. Measurement of the body temperature threshold in which seizure was induced was made.
Quantitative RT-PCR. Brains were obtained from mice (P14.5 Scn2a KO/+ mice and Scn2a +/+ littermates, both sex), and total RNA was isolated and purified from brain using the RNeasy Midi Kit (Qiagen, Valencia, CA, USA), and reverse transcribed using the Super Script III First Strand Synthesis System (Thermo Fisher Scientific). The resultant cDNA was then amplified using the Applied Biosystems 7500 Real-Time PCR System (Thermo Fisher Scientific) with pre-designed Taq-Man assays for mouse Scn1a, 2a, 3a, 5a, 8a, and 1b-4b as well as the internal standard 18 S rRNA or GAPD (Thermo Fisher Scientific). Relative gene expression was determined using the 2 −ΔΔCt method 64 . Briefly, the cycle threshold value (Ct) on each PCR amplicon curve was determined, and the ΔCt was calculated by subtracting the Ct value of the internal standard from the Ct value of sodium channel mRNA. The ΔΔCt was then calculated by subtracting the ΔCt value for each individual from the mean ΔCt value of controls. Relative gene expression was calculated as 2 −ΔΔCt and represented as percentages, relative to mean expression level of Scn2a +/+ littermate controls (100%).
Statistical analyses. Statistical analyses were conducted using the Prism 6 statistical package (GraphPad Software, La Jolla, CA, USA) and Excel (Microsoft, Redmond, WA, USA). Comparisons between two genotype groups were performed using Mann-Whitney U test and unpaired t-test, unless otherwise described. Dependent variables were analyzed using analysis of variance with Tukey's multiple comparison test. Values with P < 0.05 were considered to be significant.
Data availability. All data supporting the findings of this study are available within the article and its supplementary information file or are available from the corresponding author upon reasonable request.