Gene-environment interaction counterbalances social impairment in mouse models of autism

Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by social communication deficits and repetitive/restricted behaviors. Although gene-environment interactions may explain the heterogeneous etiology of ASD, it is still largely unknown how the gene-environment interaction affects behavioral symptoms and pathophysiology in ASD. To address these questions, we used Cntnap2 knockout mice (genetic factor, G) exposed to valproic acid during embryonic development (environmental factor, E) as a gene-environment interaction (G × E) model. Paradoxically, the social deficits observed in the respective G and E models were improved in the G × E model; however, the high seizure susceptibility was more severe in the G × E -model than in the G and E models. Repetitive self-grooming and hyperactivity did not differ among the three models. The amplitudes of miniature excitatory postsynaptic currents in layer 2/3 pyramidal neurons of the medial prefrontal cortex were aberrant and similar in the G × E model when compared to the control group. Our findings suggest that the interaction of two risk factors does not always aggravate ASD symptoms but can also alleviate them, which may be key to understanding individual differences in behavioral phenotypes and symptom intensity.

Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by social communication deficits and restricted/repetitive behaviors before 3 years of age 1 . Etiological heterogeneity and complex behavioral symptoms hamper understanding of ASD. Over 1000 genetic factors 2 and various environmental factors such as valproic acid (VPA), maternal immune activation, and paternal age can induce ASD 3,4 . A multitude of genome-wide association studies have been conducted to identify a common target gene but have not been successful [5][6][7] , indicating that the etiology of ASD may not be explained by one common genetic factor. Instead, recent studies have shown that autistic symptoms can be caused or exacerbated by the interaction of genetic and environmental risk factors 8,9 , suggesting that gene-environment interaction may be a mechanism underlying the etiology of ASD. Yet, how the gene-environment interaction affects complex behavioral symptoms and pathophysiology in ASD remains largely unknown.
To date, one of the most important hypotheses to explain the pathophysiological mechanism of ASD is excitatory-inhibitory (E/I) imbalance 10,11 . E/I imbalance can result from abnormalities in excitatory or inhibitory neural structure and/or function 11,12 . The reduced or increased synaptic transmission may cause overprocessing (noise) 10 and misprocessing (disconnection) 11 of the transmission, respectively. The impaired transmission may cause connectivity problems in the neural circuits and thereby may lead to neurodevelopmental disorders such as ASD 10,11,13 . The impact of E/I imbalance may depend on the affected brain region. The prefrontal cortex (PFC; or medial prefrontal cortex, mPFC) is a very well-known brain region mediating social behavior 14 . Indeed, neural activity in the mPFC is increased during social behaviors 15,16 . An optogenetic study showed that over-activation of excitatory neurons in the PFC induces abnormal social behaviors 17 , recapitulating the effect of E/I imbalance in the mPFC as an inducer of social impairment.
In our previous study, contactin-associated-protein-like 2 knock-out mice (Cntnap2 KO, a genetic risk factor model) and mice exposed to VPA during embryonic development (VPA mice, an environmental risk factor model) displayed reduced and increased miniature excitatory postsynaptic currents (mEPSCs) in layer 2/3 pyramidal neurons of the mPFC, respectively, possibly because of aberrant expression of glutamate receptors 18 . Of note, although these two models show opposite neural transmission impairment in the mPFC, these mice show similar behavioral symptoms such as social deficits, repetitive behavior, hyperactivity, and seizure susceptibility [18][19][20] . Thus, it will be tantalizing to study how the interaction of these factors would affect behavioral abnormalities and the pathophysiology of the respective models.
To investigate the effect of gene-environment interaction, we injected VPA into pregnant Cntnap2 heterozygous female mice that had been mated with Cntnap2 heterozygous male mice on embryonic day 10.5 (E10.5) and investigated the autism-like behavioral phenotypes in the resulting offspring. We also investigated neural transmission changes using whole-cell patch clamp in layer 2/3 pyramidal neurons of the mPFC. We found that the interaction of Cntnap2 KO and prenatal exposure to VPA changed the phenotypes observed in the respective single-factor models. Remarkably, the interaction model showed improvement in social deficits in comparison with the single-factor models, whereas the levels of repetitive-grooming and locomotor-activity in the interaction group were comparable to those in the single-factor models. The interaction also increased seizure susceptibility in comparison with that of the single-factor models. The depressed and enhanced excitatory transmission in the mPFC of Cntnap2 KO and VPA-exposed mice, respectively, was rescued by the interaction of Cntnap2 KO and prenatal exposure to VPA. This study provides crucial experimental evidence on how the interplay of two risk factors contributes to phenotypic complexity and on the possible role of E/I imbalance in the modulation of social impairment in ASD.

Results
Social impairment is restored by the interplay of Cntnap2 KO and prenatal VPA exposure. To investigate gene-environment interaction, we injected phosphate-buffered-saline (PBS) or VPA (100 or 300 mg/ kg) into pregnant Cntnap2 heterozygous female mice that had been mated with Cntnap2 heterozygous male mice on E10.5 and obtained nine combination groups affected by different dosage levels of each factor. All mice had no considerable health problems but those exposed to 300 mg/kg of VPA had crooked tails (Supplementary Fig. 1) as we described previously 21 . Using these nine groups, we performed the three-chamber sociability interaction test. Vehicle-treated wild-type and Cntnap2 heterozygous mice, and mice exposed to 100 mg/kg VPA regardless of their genotypes preferred the compartment containing a novel conspecific or its vicinity, whereas Cntnap2 KO mice and wild-type (WT) mice exposed to 300 mg/kg VPA had no such preference. However, unexpectedly, the impaired social preference of Cntnap2 KO mice to a novel conspecific was ameliorated in the offspring of female mice treated with 100 or 300 mg/kg VPA (Fig. 1a-c). To confirm these unexpected findings, we performed the juvenile social play test in the following four groups: Cntnap2 WT treated with vehicle (WT × Veh), KO treated with vehicle (KO × Veh), WT exposed to 300 mg/kg VPA (WT × VPA), and KO exposed to 300 mg/kg VPA (KO × VPA). In this test, we measured the cumulative social interaction time, including allogrooming, following, pouncing, and sniffing (Fig. 1d). Again, the KO × Veh and WT × VPA groups exhibited significantly shorter social interaction time than the WT × Veh group. However, the social interaction time of the KO × VPA group was similar to that of the WT × Veh group ( Fig. 1e; two-way ANOVA, interaction: F(1, 30) = 28.04, p < 0.0001; genotype factor: F(1, 30) = 1.56, p = 0.2213; VPA factor: F(1, 30) = 0.08728, p = 0.7697). Thus, the interaction of Cntnap2 KO and prenatal VPA exposure alleviates social impairment observed in the respective single-factor models.
Repetitive self-grooming and hyperactive behaviors are not affected by the interplay of Cntnap2 KO and prenatal VPA exposure. Next, we investigated the effect of the interaction of Cntnap2 KO and prenatal exposure to VPA on repetitive behavior, another core symptom of ASD. To this end, we measured cumulative time spent on self-grooming. The Cntnap2 KO × Veh, WT × VPA, and KO × VPA groups spent significantly more time on self-grooming than did the WT × Veh group ( Hyperactivity is another important symptom observed in the Cntnap2 KO 19 and prenatal VPA exposure models 20,22 . To determine the effect of the interaction of these two factors on hyperactivity, we performed the open-field test. The distance moved was significantly longer in the Cntnap2 KO × Veh and WT × VPA groups than in the WT × Veh group ( Fig. 2b; unpaired t-test: WT × Veh vs. WT × VPA, p = 0.0067; Bonferroni's post hoc test: WT × Veh vs. KO × Veh, p < 0.01). The combination of VPA and Cntnap2 KO also significantly increased locomotor-activity in comparison with the WT × Veh group. However, the increased locomotor-activity was similar to that of the Cntnap2 KO × Veh and WT × VPA groups (two-way ANOVA, interaction: F(1, 45) = 1.302, p = 0.2599; genotype factor: F(1, 45) = 15.40, p = 0.0003; VPA factor: F(1, 45) = 6.542, p = 0.0140; Bonferroni's post hoc test: WT × Veh vs. KO × VPA, p < 0.001). Taken together, the above data show that the interaction of Cntnap2 KO and prenatal VPA exposure does not affect the repetitive and hyperactive behaviors observed in the respective single-factor models. electric shock-induced seizure susceptibility is increased by the interplay of Cntnap2 Ko and prenatal VPA exposure. In a previous study, Cntnap2 KO mice exhibited spontaneous seizures following mild handling stress 19 . VPA-exposed rats also showed increased seizure susceptibility following electric stimulation relative to control rats 20 Two-way ANOVA followed by Bonferroni multiple comparison test was performed to determine betweengroup differences. WT × Veh, n = 9; KO × Veh, n = 8; WT × VPA (300 mg/kg), n = 9; KO × VPA (300 mg/kg), n = 9. In e, asterisk (*) shows the results of statistical comparison between WT × Veh and the indicated group; *, p < 0.05, **, p < 0.01. Sharp (#) shows the results of statistical comparison between the indicated groups; ##, p < 0.01, ###, p < 0.001. In all bar graphs, data are the mean ± S.E.M. Abbreviations: S, compartment or cage www.nature.com/scientificreports www.nature.com/scientificreports/ to VPA on seizure susceptibility, we measured the convulsive threshold determined by the convulsive current 50 (CC50; Fig. 3). The CC50 values were lower in the Cntnap2 KO × Veh and the WT × VPA groups than in the WT × Veh. Interestingly, the KO × VPA group had lower CC50 values than Cntnap2 KO × Veh and the WT × VPA groups, suggesting that seizure susceptibility is increased by the interaction of Cntnap2 KO and prenatal VPA exposure. impaired excitatory transmission in each single-factor model is counterbalanced by the interplay of Cntnap2 KO and prenatal VPA exposure. Proper regulation of neural activity in the mPFC is very important for social behavior [15][16][17] . In our previous study, Cntnap2 KO mice and mice prenatally exposed to VPA showed decreased and increased amplitudes of mEPSCs in layer 2/3 pyramidal neurons of the mPFC, respectively, compared to those in the respective control mice 18 . We also observed that the counterbalancing of impaired excitatory transmission using α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) modulators specifically improved social behavior in Cntnap2 KO mice and mice prenatally exposed to VPA 18 . Thus, the improved sociability in the G × E group may be due to neural transmission in the mPFC corrected by the interaction. To examine this possibility, we measured mEPSCs and miniature inhibitory postsynaptic currents

Discussion
In this study, we investigated how the interplay between genetic and environmental risk factors affects phenotypic complexity using the Cntnap2 KO and prenatal exposure to VPA as genetic and environmental factors, respectively. We observed that the interplay between these two factors improved social deficits but increased seizure susceptibility, with no further changes in repetitive self-grooming behaviors or hyperactivity compared to the respective phenotypes of Cntnap2 KO and VPA mice (Supplementary Table 1). Finally, we also found that the aberrant excitatory neural transmission in the mPFC of the single-factor models was corrected by the interaction of the two factors. Our study provides an important clue to understanding the potential role of gene-environment interaction in the behavioral and pathophysiological spectrum in ASD.
Our results show that behavioral and physiological phenotypes caused by a genetic mutation can be changed by an environmental input. VPA upregulates target gene expression by inhibiting histone deacetylase 24,25 . Thus, the altered gene expression may lead to phenotypic changes in the Cntnap2 KO mice. For example, our previous studies have shown that prenatal exposure to VPA upregulates Pax6, which may induce differentiation of glutamatergic neurons and thereby enhance synaptic maturation in glutamatergic neurons 21,23 . These changes in glutamatergic neurons caused by prenatal exposure to VPA may alter excitatory neural transmission and behavioral phenotypes in Cntnap2 KO mice.
Perhaps the most interesting finding in our study is that the interaction of Cntnap2 KO and prenatal exposure to VPA resulted in normal social behaviors comparable to those of the control group in the three-chamber sociability and juvenile social play tests. Currently, we are not sure how the interaction of the two factors paradoxically recovers the social impairments observed in the respective single-factor models. Our current and previous findings suggest two possibilities. First, we found that the aberrant mEPSC amplitudes in the mPFC region of the single-factor groups were corrected in the KO × VPA group, suggesting that the recovered excitatory neural transmission in the mPFC may be associated with the improved social behaviors in the interaction group. Indeed, neural transmission in the mPFC is very important in regulating social behaviors [15][16][17] . In our previous study, Cntnap2 KO mice exhibited an underexpression of glutamatergic receptors and decreased mEPSCs amplitudes in layer 2/3 pyramidal neurons of the mPFC, whereas VPA mice exhibited the opposite pattern 18 . Additionally, a positive allosteric modulator and an antagonist of the AMPA receptor recovered impaired social behaviors without affecting repetitive self-grooming or hyperactivity in Cntnap2 KO mice and VPA mice, respectively 18 . Thus, the corrected excitatory neural transmission in the mPFC of the KO × VPA group might underlie the improved social behaviors. However, this possibility should be tested further, since a recent report suggested that social impairments in adult Cntnap2 KO mice may be induced from abnormal activity of parvalbumin-positive interneurons in the mPFC 26 . This discrepancy may be due to the difference in experimental conditions such as the age of mice. Since we used 4-to 5-week-old mice, whose synaptic structures undergo dynamic changes 27 , the synaptic response might be different from that of adult mice. Thus, further study is required to address this with a novel conspecific; E, compartment or cage with no conspecific, WT, Cntnap2 WT mice; KO, Cntnap2 KO mice; Het, heterozygous mice; 100 mpk, 100 mg/kg of VPA; 300 mpk, 300 mg/kg of VPA; VPA, treatment with valproic acid at E10.5; Veh, treatment with phosphate-buffered saline at E10.5.

Scientific RepoRtS |
(2019) 9:11490 | https://doi.org/10.1038/s41598-019-47680-w www.nature.com/scientificreports www.nature.com/scientificreports/ discrepancy. Second, there might be a compensation mechanism in the neural structure of the KO × VPA group. Indeed, a functional magnetic resonance imaging (fMRI) study showed that long-range connectivity in the frontal cortex region was reduced in CNTNAP2 risk allele carriers 28 . Another fMRI study showed that Cntnap2 KO mice had reduced local connectivity in the PFC, and the reduction in connectivity showed a significant correlation with impaired social behaviors 29 . On the other hand, increased local connectivity in the mPFC have been suggseted in prenatal VPA exposure model of autism 13,30 . Given that the CNTNAP2 protein has been implicated in synaptic stability 31,32 , the reduced synaptic contacts in Cntnap2 KO mice may be compensated by the effects of prenatal VPA exposure, which may enhance synaptic maturation in glutamatergic neurons 21 . Thus, the compensation mechanism might contribute to relieving social impairments by recovering abnormal functional connectivity in the PFC. All these possibilities should be tested in further study using more in-depth electrophysiological and imaging approaches.
Another interesting finding in this study is the increased seizure susceptibility of Cntnap2 KO × VPA mice in comparison with the respective single-factor groups. Seizure susceptibility can be induced by increased excitation or decreased inhibition in any brain region 33 . Since the KO × VPA group exhibited mEPSCs and mIPSCs in the mPFC similar to those of the WT × Veh group, increased excitability in other brain regions may be the reason for the increased seizure susceptibility of the KO × VPA group. For example, in a previous report, Cntnap2 KO mice had decreased expression of inhibitory neuronal marker proteins such as glutamate decarboxylase (GAD) 67, parvalbumin, calbindin 2, and neuropeptide Y in the somatosensory cortex and striatum, suggesting abnormal inhibitory neuronal function 19 . A whole-cell recording study showed that perisomatic inhibitory neural inputs were decreased in the hippocampal CA1 pyramidal neurons of Cntnap2 KO mice 34 , suggesting increased excitability in the hippocampus. Another whole-cell recording study showed that Cntnap2 KO mice displayed reduced excitatory and inhibitory transmission, but the sum of these transmissions resulted in an increased E/I ratio in layer 2/3 of the somatosensory cortex 35    www.nature.com/scientificreports www.nature.com/scientificreports/ over-excitation caused not only by increased glutamate receptor expression and transmission 36 but also by decreased levels of GAD65/67 in the somatosensory cortex 37 . Thus, the interaction of Cntnap2 KO and prenatal exposure to VPA may increase over-excitability in some brain regions such as the somatosensory cortex, which might contribute to the increased seizure susceptibility in the interaction group.
One confounding result in our study is that the Cntnap2 KO × VPA interaction did not induce any significant changes on repetitive self-grooming or hyperactivity behaviors, although each factor independently induced these behavioral symptoms. The pathophysiological mechanisms of increased repetitive self-grooming and hyperactivity behaviors in Cntnap2 KO mice and VPA-exposed models are not clear, yet. In previous studies, the repetitive self-grooming and hyperactivity behaviors were rescued by a dopamine receptor antagonist in Cntnap2 KO mice 19 and by N-methyl-D-aspartate (NMDA) receptor blockers in prenatal VPA exposure models 38,39 . These results suggest the involvement of the dopaminergic and glutamatergic systems in these behavioral alterations in Cntnap2 KO mice and prenatal VPA exposure models, respectively. However, since these two systems are involved in the cortico-basal ganglia-thalamic pathway, which is important for motor coordination and pathophysiology of repetitive and hyperactive behaviors 40,41 , Cntnap2 KO or prenatal exposure to VPA may be sufficient to fully induce repetitive and hyperactive behaviors, or one of these two factors may be dominant. This may presumably explain why the interplay of Cntnap2 KO and VPA resulted in no further increase in those behavioral symptoms. Obviously, these possibilities await experimental validation.
Our results suggest several points to consider. First, the interaction of two different risk factors does not always aggravate ASD symptoms but can also alleviate them. Second, an environmental factor can change phenotypes induced by a genetic factor. In addition to prenatal exposure to VPA, other environmental factors such as maternal immune activation 8,42 , advanced paternal age 43,44 , and sex difference 9,45 may increase the phenotypic  www.nature.com/scientificreports www.nature.com/scientificreports/ complexity of genetic predispositions. Furthermore, an environmental input may rescue impaired neural physiology and autism-like symptoms. Second, although we did not test this here, the effects of an environmental input may depend on its timing. Indeed, VPA is a very well-known teratogen and its effects can be different depending on exposure period 46 . If it is taken during the first trimester period, it greatly increases the risk of craniofacial and limb malformation and neuropsychiatric symptoms such as impaired cognition and autistic behaviors in humans 47 . We have also previously characterized such exposure-period-dependent phenotypic change in animals. We found that animals exposed to VPA during neural tube closure (E12.5 48 in rats and E10.5 22 in mice) showed an increased incidence of autism-like symptoms 22,48 . If VPA is given earlier than this critical time point, it causes embryonic lethality. Conversely, if it is given later, VPA does not cause embryonic lethality or autism-like phenotypes 48 . Thus, the timing of exposure may be an important variable in gene-environment interaction studies.
In conclusion, we demonstrated that the interplay of Cntnap2 KO and prenatal VPA exposure increases the phenotypic complexity by changing the original phenotypes induced by the respective factors. Thus, the complex interaction of multiple factors may explain the behavioral and pathophysiological spectrum in ASD.

Methods
Animals. Cntnap2 KO mice were kindly provided by Dr. Daniel H. Geschwind 19 . The day of checking the vaginal plug was designated as E0 and the day of birth as P0. VPA or PBS was injected subcutaneously at E10.5 22 . From the obtained offspring, 3-week-old male mice were transferred to new cages according to their genotype and treatment (3-6 mice per cage). Genomic DNA was extracted from the tail at P14 and the genotype of each animal was identified using PCR.
Animals were maintained on a standard light-dark cycle (on: 2:00 am; off: 2:00 pm) at 20 °C to 24 °C and relative humidity between 30% and 70%. Animals were given free access to food and water. Animal handling, housing, and treatments, including anesthesia, euthanasia, and administration, were performed in accordance   Behavioral studies. Behavioral studies were conducted during the dark cycle. To minimize the unpredictable effects of handling stress, animals were habituated for 3 days to the experimenter's handling. For this purpose, a mouse was placed on experimenter's cupped hands and allowed to freely explore for 1 min, once a day. On the day of behavioral tests, mice were acclimated to the behavioral room for 1 h before the tests.
Three-chamber sociability test. The three-chamber sociability test was conducted as previously reported 18,49 . The task consisted of two sessions. In the first session, the subject mouse was introduced to the middle compartment and habituated for 5 min. After habituation, a new animal (same age, strain, and no previous contact with the subject) was introduced inside a wired cage randomly to either the left or right compartment, while the other wired cage was empty during the 10-min sociability test. Time spent in each compartment and around the cage (within 5 cm from the cage) was measured as stay duration and approach duration, respectively 18 .
Self-grooming test. The test was conducted as previously reported 18,50 . Before the experiment, each mouse was placed in a polycarbonate cage (20 × 26 × 13 cm) and habituated for 10 min. The cumulative time spent grooming was measured over the next 10 min with observers blinded to the mouse's group from at least 2 m away from the cage.
Juvenile social play. The subject mouse was isolated for 30 min in a cage with new bedding (3 cm deep) to stimulate social interaction before the test. A stranger mouse was subsequently introduced to the subject's cage. The cumulative social interaction time was measured for 10 min. The social interactions included nose-to-nose sniffing, nose-to-anus sniffing, following, allogrooming, and crawling under the partner 51 .
Electric shock seizure threshold test. Electric shock seizure threshold was measured according to previously reported methods 23 . Briefly, electric shock was applied using an ECT apparatus (Electro Convulsive Therapy, Ugo Basile, Varese, Italy) through ear clips (frequency: 100 pulses/s, pulse width: 0.5 ms, shock duration: 1 s). To minimize the number of animals sacrificed, electric current was increased or decreased stepwise by 2 mA according to the animal's response to electric shock. A seizure was defined as overt hind limb extension. The convulsive current 50 (CC50) was defined as the current at which 50% of all animals showed seizure and was determined by log current vs. response calculation.
Open-field test. Exploratory activity in a novel environment was assessed in an open-field box (40 × 40 × 30 cm). Mice were introduced into the center, and the total distance moved in the whole arena and velocity were measured for 20 min using a CCD camera-assisted motion-tracking apparatus and software (EthoVision 3.1, Noldus Information Technology, Leesburg, VA, USA). Slice preparation. Slices were prepared from mice aged 4 to 5 weeks. Mice were anesthetized using isoflurane, and the brains were quickly removed and transferred into ice-cold sucrose solution (in mM, 212 sucrose,