Introduction

Autism spectrum disorder (ASD) is a neurological developmental disability that is characterized by impaired social communication and interaction as well as restricted, repetitive patterns of behavior, interests, or activities [1]. It has been clinically known that ASD is accompanied by other neurodevelopmental disorders such as intellectual disability or attention-deficit hyperactivity disorder (ADHD). Additionally, epilepsy and aggression are comorbidities in around 25% of ASD cases [2]. Currently, autism is receiving increased attention worldwide. The estimated prevalence of ASD among 8-year-old children in the US has risen from around 1.1% in 2008 to 2.3% in 2018 [3].

Researchers have identified one hundred potential genes that are significantly associated with autism [4], yet these genetic mutations cannot fully explain the pathogenic characteristics of autism. Another layer of complexity in “epigenetics” may make a significant contribution to the etiology of autism. Epigenetic modifications on DNA or histones can influence the transcriptional activation and silencing of genes. With the ongoing advancements in whole-exome and whole-genome sequencing data from autistic patients, mutations/dysregulation in genes that regulate DNA methylation mechanisms have been found to be closely linked to the pathogenesis of autism. Genetic studies have identified novel mutations in DNA methyltransferases and readers (DNMT3A, MECP2, MBD5) in autism pathogenesis [5,6,7]. These mutations can lead to loss or gain of function in epigenetic modifier enzymes. These genetic findings from patients provide substantial molecular evidence in support of the hypothesis that epigenetic dysfunction contributes to autism etiology. The most well-known Rett syndrome gene, MECP2, as a methylation-dependent transcriptional repressor, participates in regulating the expression of Dnmt1, TDP-43, and CREB genes [8, 9]. The dosage balance of MECP2 is crucial for the development of autism. Overexpression of MECP2 can lead to abnormal increases in excitatory neuronal dendritic spines, which in turn causes an imbalance between neuronal excitation and inhibition in neural circuits, triggering autism-like behavior [10]. Studies on transgenic (TG) mice with MECP2 duplication syndrome demonstrate that the medial prefrontal cortex (mPFC) plays a significant regulatory role in the social deficits of MECP2 duplication syndrome [11]. Functional MRI shows excessive excitability in the mPFC of Mecp2 TG mice, and specific deletion of Mecp2 in the mPFC can alleviate the social deficits in mice [12].

Drug-induced autism is also considered one of the major etiologies of autism. One of the major exposure risk factors for drug usage during pregnancy is antiseizure drugs, and a recent clinical study has confirmed that valproic acid, the most established drug, increased the incidence of autism in offspring, while other drugs have failed to establish the link [13]. VPA treatment in rats and mice can reproduce most symptoms of autism, including social deficiency, repetitive behaviors, and accompanied with seizures [14]. Interestingly, epigenetic dysregulation, including histone methylation/acylation dysregulation and DNA methylation abnormality, is recognized as the major underlying mechanism in these autism animal models. Importantly, VPA induced reduced MECP2 expression through miR-132-BDNF regulatory loop during E12.5 day treatment [15], and the mechanistic study showed that convergence of VPA induced alteration with major epigenetic regulators, including MECP2. However, the hub gene and the downstream pathway regulated by both MECP2 and VPA remain elusive, limiting the targeted intervention based on the pathway affected by these two most established monogenic and environmentally induced autism models.

GADD45A is a member of the GADD45 gene family, whose transcription levels increase under conditions of stress-induced growth arrest and treatment with DNA-damaging agents, responding to environmental stress. Recent research has found that in cell lines, the binding of GADD45A to DNA-RNA hybrids (R-loops) triggers the recruitment of TET1, mediating the demethylation of the tumor suppressor TCF21 promoter [16]. Genome analysis in embryonic stem cells has identified thousands of R-loop-dependent TET1 binding sites located on CpG islands, suggesting an important role in the regulation of methylation levels during neural development. Gadd45b, as a member of the same gene family as Gadd45a, has similar biological functions. Previous Studies, including our team, have found that the knockdown of Gadd45b leads to changes in adolescent social behavior and reduced expression of several genes associated with psychiatric disorders, including MECP2 and BDNF [17, 18]. These data suggest that Gadd45b plays an important role in the epigenetic regulation of complex social behavior, to some extent indicating the association of epigenetic-related genes with autism. A recent study found that the drug intervening Gadd45a-related pathway could rescue autism-like behaviors in BTBR mice [19]. However, whether dysregulation of GADD45A, the first reported member of the GADD45 family, is directly involved in the etiology of neurodevelopmental diseases, such as autism and epilepsy, and the underlying mechanism, remains unclear.

In the current study, for the first time, through transcriptome data from the VPA-treated animal models and MECP2 mutant models, we have identified that GADD45A is a hub gene dysregulated in both the VPA-induced autism model and MECP2-related animal models. We have shown that deletion of Gadd45a leads to autistic traits and accompanying symptoms in the mouse model, including epilepsy phenotype. GADD45A is preferentially expressed in the excitatory neurons, and social deficits can be reversed when the GADD45A level is rescued in the excitatory neurons of mPFC. Through transcriptome and ChIP-Sequencing studies, we have also shown that GADD45A could recruit TET1 to regulate the promoter region in the downstream genes, specifically enriched on the ion channels. Finally, we have confirmed that KCNQ5 is a critical target regulated by GADD45A through an R-loop-dependent pathway, which is essential for maintaining the normal function of excitatory neurons. Therefore, our study uncovered a critical role of GADD45A/TET1-KCNQ5 methylation and R-loop dependent regulatory axis, which may represent a key molecular pathway mediating the pathogenesis of neurodevelopmental disease, including autism and epilepsy, among others.

Results

GADD45A functions as a potential downstream effector gene of the etiology of autism in multiple autism models

Current understanding of the etiology of autism includes well-known factors such as prenatal exposure to VPA leading to autism in offspring [13], along with mutations in the MECP2 gene [20], and so on. Intriguingly, there were studies reporting that VPA exposure can suppress the expression of MECP2 [21,22,23]. In addition, we validated their conclusion by treating SH-SY5Y cells with VPA. We found that VPA treatment could attenuate the mRNA expression of MECP2 in vitro (Fig. 1A). This evidence indicated that there appears to be a relationship in the pathogenesis of autism between VPA-induced autism and MECP2-deficient autism, suggesting the potential shared downstream pathogenic molecular mechanisms. Therefore, in the GEO database, we collectively analyzed differential gene sets (GSE129241) in reprogrammed human neurons on post-induction day 1 treated with valproic acid to simulate the impact of VPA on early neural development, and differential gene sets (GSE230714) in reprogrammed human neurons of MECP2 knockout. We identified 385 genes that were commonly dysregulated between the two datasets. Moreover, the dosage balance of MECP2 is crucial for the development of autism. The gain-of-function of MECP2 represents the clinical causes of the MECP2 duplication patients. We analyzed the transcriptome data of transgenic MECP2 monkeys (GSE57974) and mice (GSE123372) in public databases, and identified 28 genes that were conserved and altered in the MECP2 overexpression models by intersecting the two different datasets. To identify the potential common pathogenic factors underlying seemingly different yet interconnected autism models, including the VPA-induced autism model and the autism model resulting from MECP2 dosage abnormalities, we reanalyzed the intersection genes from the previous two intersection sets and ultimately identified three shared differentially expressed candidate genes (Fig. 1B). Among them, GADD45A was of particular interest, as it has been reported to be epigenetically repressed by MECP2 in prostate cancer cell lines. [24]. Notably, GADD45A was the only gene that exhibited consistent downregulated alterations in both MECP2 transgenic monkey and mouse models compared with wild-type controls. In addition, dysregulation of GADD45B within the GADD45 family has been previously associated with autism-like behaviors [17]. Moreover, GADD45A expression was reported to be induced by valproic acid in vitro [25]. And our own data confirmed that VPA treatment could result in a time-dependent increase of GADD45A expression in SH-SY5Y cells (Fig. 1C). And we further utilized Western Blot to detect Mecp2 KO cortex in mice and found knockout of Mecp2 led to an increase of GADD45A expression (Fig. S1A), consistent with the result of GSE230714 in MECP2 knockout human neurons. Conversely, MECP2 duplication caused decreased expression of GADD45A in both transgenic monkey and mouse models, showing the opposite effect on GADD45A expression to that observed in MECP2 knockout and VPA exposure. Considering that MECP2 duplication syndrome (MDS) occurs predominantly in male patients [26], we next asked whether male autism patients display a similar dysregulated pattern of GADD45A. We recruited two independent cohorts of male children. Cohort 1 consisted of 9 children with ASD and 9 typically developing controls, and Cohort 2 included 7 children with ASD and 8 typically developing controls. Quantitative PCR analysis of peripheral blood samples revealed that GADD45A expression levels were significantly decreased in ASD patients in both cohorts (Fig. 1D, S1B). In Cohort 1, where multiple autism assessment scales (ABC, SRS, RBS-R, CBCL) were administered, we further examined the association between GADD45A expression and autism-related behavioral phenotypes. We found that two (ABC, SRS) of these scores were statistically negatively correlated with the relative expression levels of GADD45A, of which more SRS reflecting the more severity of social deficits in the autism spectrum (Fig. 1E, F, S1C, S1D). These results indicated that GADD45A dysregulation was closely relevant to the etiology of autism, at least in the social deficits module. In summary, through transcriptome mining in public databases, validation with our own patient samples, mouse samples, and cell experiments, as well as literature research, our findings suggested that GADD45A dysregulation is closely associated with autism-related molecular and behavioral phenotypes.

Fig. 1: GADD45A dysregulation is associated with autism in multiple models and male patients.
figure 1

A The RT-qPCR detected MECP2 mRNA expression of SH-SY5Y treated with 1 mM VPA after 0 h, 24 h, 48 h, 72 h, 96 h. Three individual biological replications per group. One-way ANOVA followed by Dunnett’s multiple-comparison test. B The Venn diagram in the upper left panel shows the intersection of the differential gene sets of MECP2 knockout and VPA-treated neurons. The Venn diagram in the upper right panel shows the intersection of differentially expressed gene sets between MECP2 transgenic monkeys and Mecp2 transgenic mice in the cortex. The Venn diagram below shows the intersection of the common differentially expressed genes from the two groups above to identify shared differentially expressed genes among multiple autism models. Whether the affected gene is upregulated or downregulated compared with control in each dataset is annotated by using arrows (↑/↓) and consistent color coding (red for upregulation, blue for downregulation). C The RT-qPCR detected GADD45A mRNA expression of SH-SY5Y treated with 1 mM VPA after 0 h, 24 h, 48 h, 72 h, and 96 h. Three individual biological replications per group. One-way ANOVA followed by Dunnett’s multiple-comparison test. D The RT-qPCR results manifest that the GADD45A mRNA level in autistic boys (Cohort 1) decreased compared with the control group using mRNA from peripheral blood. n = 9, each group. Unpaired Student’s t-test. E, F Both the scores of the Aberrant Behavior Checklist Scores (ABC) and Social Responsiveness Scale Scores (SRS) in all participants were utilized to analyze the correlation with GADD45A gene expression. Spearman’s rho (rs) and p values presented in the figures were calculated by Spearman’s rank correlation coefficient. *p < 0.05, ** p < 0.01. All data are presented as means ± SD.

Full size image

Gadd45a knockout mice display social deficits

To determine the role of Gadd45a in the brain, we generated Gadd45a KO mice using CRISPR-Cas9 technology. The successful establishment of the Gadd45a KO mice was confirmed by Sanger sequencing (Fig. 2A). The knockout efficiency was determined by quantitative PCR (qPCR) and western blot in brains (Fig. 2B, C). The KO mice were viable and had no physical dyskinesia. To determine whether and how the Gadd45a deletion influences neurological behavior. Until the mice grew to about 12–16 weeks, we conducted several behavior tests. WT mice and KO mice were tested using the three-chamber test to examine their sociability (preference for social interaction over non-social objects) and their ability to distinguish social novelty (Fig. 2D). Following the habituation period, an unfamiliar mouse (referred to as “stranger 1” or “S1”) was introduced into one of the chambers. Wildtype mice preferentially spent more time in the neighboring zone around S1, in comparison to the counterpart of the empty cage (EM), and interacted with S1 for a longer duration than with the empty cage. Whereas KO mice displayed almost no distinction in their interactions with S1 versus the EM (Fig. 2E, F). In the social novelty phase, another unfamiliar mouse (referred to as “stranger 2” or “S2”) was introduced in the empty cage. The instinct of normal mice was to have more contact with stranger 2. In contrast, mice with loss of Gadd45a exhibited comparable interaction times with both S1 and S2 (Fig. 2E, G). These results indicated that deletion of Gadd45a impaired both sociability and social novelty recognition, which is one of the representative symptoms of autism.

Fig. 2: Gadd45a KO mice exhibit typical ASD-like behaviors.
figure 2

A Schematic diagram of the Gadd45a knockout strategy with the Sanger Sequencing result of the genome DNA of KO mice. B RT-qPCR analysis of Gadd45a expression in the brains of KO mice, compared to WT mice. n = 3 mice for each group. Unpaired Student’s t-test. C Western Blot was used to validate the efficiency of knockout in the cortices of the brain at the protein level. D Schematics of the Three-chamber social test, consisting of two stages (Stage 1: Sociability test, Stage 2: Social novelty test). E Representative heat maps of the trajectory of the two groups in the two stages of the Three-chamber tests. F In the sociability test (Stage 1), Gadd45a KO mice do not show a preference for the stranger 1 (S1) compared with the Empty Cylinder (EM). WT mice, n = 14, KO mice, n = 11; S1 versus EM, paired Student’s t-test. G In the Social novelty test (Stage 2), there is no significant difference in the interaction time KO mice spent between stranger 2 and stranger 1. WT mice, n = 14, KO mice, n = 11; S1 versus S2, paired Student’s t-test. H Gadd45a KO mice exhibit stereotypical behaviors. KO mice spend significantly more time digging than WT mice. In addition, KO mice show a greater tendency to self-groom, though the grooming time of KO mice has no significant difference compared with WT mice. Each group, n = 8; WT versus KO, unpaired Student’s t-test. *p < 0.05, *** p < 0.001, and ns, no significant difference. All data are presented as means ± SD.

Full size image

We also noticed that there are some co-occurring conditions, including epilepsy, depression, anxiety, aggression, and stereotyped, repetitive behaviors, and so on in ASD patients and several mouse ASD models. Therefore, we kept Gadd45a knockout mice alone in the cages without cage lids and found that the mice exhibited significantly increased digging behaviors (Fig. 2H). Additionally, we observed increased but not significant repetitive grooming in Gadd45a knockout mice (Fig. 2H). The frequencies of jumping and rearing were also tested but showed no obvious difference (Fig. S2E). Intriguingly, there were spontaneous seizures observed in the knockout mice (Video S1), even though we assume the incidence rate is relatively low. Moreover, many male knockout mice displayed the tendency to fight with their littermates in their home cages (Video S2).

We then used the Morris Water Maze to determine the cognitive performance during the 4-day training. There is no significant difference in latency to target, the time spent in the Target quadrant, and platform crossover number (Fig. S2A). Furthermore, in the elevated plus maze tests, Gadd45a knockout mice spent similar time in the closed arm, middle arm, and open arm as WT mice in counterparts, as well as the distance (Fig. S2B). And marble burying test proved that KO mice did not exhibit compulsive behavior (Fig. S2C). Additionally, we performed the open field test to examine the occurrence of anxiousness in knockout mice. The distance of movement, total fecal boli count, and the time in the center part had no obvious difference between KO and WT mice (Fig. S2D). These results demonstrated that loss of Gadd45a specifically led to impaired sociability, impaired social novelty, increased digging behaviors, but did not influence anxiety, memory, and compulsive behaviors.

Gadd45a is mainly expressed in the excitatory neurons, and social deficits can be reversed when Gadd45a expression level is rescued in the excitatory neurons of mPFC

A large body of research indicates that the frontal cortex governs many higher cognition behaviors, including social cognition, and is closely related to the etiology of autism [27, 28]. Firstly, an RNAscope probe targeted Gadd45a mRNA was utilized to clarify the location of Gadd45a expression due to the lack of high specificity of the anti-GADD45A antibody. The RNAscope fluorescence signals revealed that Gadd45a is abundantly expressed in the prefrontal cortex, including the medial prefrontal cortex (mPFC), which is thought to be of special importance for social cognition and behavior (Fig. 3A) [29]. We further attempted to figure out which types of brain cells Gadd45a exerts functions, especially in the prefrontal cortex. We extracted mRNA of PFC in different developmental stages and revealed that the transcriptional expression of Gadd45a rises from postnatal day 7 to postnatal day 60 (Fig. S3A). It suggested that the function of Gadd45a may be in keeping with neuron function maturing and cognitive development. We isolated and cultured primary neurons and mixed glia cells for the western blot to detect GADD45A expression level. We found that GADD45A tends to be expressed in neurons (Fig. S3B). Furthermore, we analyzed public single-cell RNA-sequencing data of mouse prefrontal cortex (GSE124952 contributed by Bhattacherjee A et al.) to extract Gadd45a expression profile [30]. The result showed that the excitatory neurons highly expressed Gadd45a compared with inhibitory neurons (Fig. S3D). Subsequently, to verify the result above, we used RNAscope probe-Gadd45a combined with antibodies for specific types of neurons to visualize the localization of Gadd45a mRNA. Gadd45a was expressed in almost all Ca2+ /calmodulin-dependent protein kinase II (CamKII)-positive neurons (75.5 ± 3.89%) throughout the frontal cortex and partially expressed in Gad67-positive neurons (20.56 ± 6.47%) (Fig. 3B, C). We also used a relatively specific antibody of GADD45A for immunofluorescence co-staining with different cell type markers in mouse brain sections. We observed no detectable colocalization of GADD45A with glial cell markers, including IBA1 (microglia), GFAP (astrocytes), and MBP (oligodendrocytes) (Fig. S3D). In addition, GADD45A was absent in NeuN⁺GAD67⁺ double-positive cells but was expressed in the NeuN⁺GAD67- cells (Fig. S3E). Moreover, consistent with our RNAscope results, GADD45A signals were prominently colocalized with CaMKII, but not with Gad67 (Fig. S3F), further confirming its selective expression in excitatory rather than inhibitory neuronal populations. To prove the role of Gadd45a in the prefrontal cortex to influence the social behavior, we chose the region, mPFC, highly correlated with the processing of social information, to rescue the expression level of Gadd45a by stereotaxic injection of adenoviruses (AAV9-CAMKIIa-Gadd45a-EGFP) which could selectively overexpress Gadd45a in excitatory neurons (Fig. 3D). After four weeks following the viral injection to Gadd45a knockout mice, the accuracy of virus infection in mPFC was confirmed by the EGFP signal in brain slices (Fig. 3E). The efficiency of Gadd45a expression in excitatory neurons was determined by most overlap between CamKII positive cells and EGFP positive cells (Fig. S3G). To examine the specificity of AAV-CaMKIIa-Gadd45a expression in KO mPFC, we revealed that restored GADD45A signals were confined to NeuN-positive, GAD67-negative, and GFAP-negative cells, confirming selective expression in excitatory neurons without detectable ectopic expression in inhibitory neurons or glia (Fig. S3H). Notably, the results of the Three-chamber test showed the rescue group preferentially spent more time on interaction with S1, in comparison to interaction with the empty cage (EM). And in the social novelty phase, the rescue group preferred to spend more time in the neighboring zone around S2, in comparison to the counterpart of S1(Fig. 3F, G). These results showed that the rescue of Gadd45a in excitatory neurons of mPFC could reverse the social deficits of knockout mice. It demonstrated that the lack of Gadd45a in excitatory neurons may contribute to the autistic phenotypes in our knockout mouse model.

Fig. 3: Gadd45a is abundantly expressed in the prefrontal cortex, including mPFC, and enriched in the excitatory neurons.
figure 3

A In-situ hybridization for detecting Gadd45a mRNA using RNAscope in the prefrontal cortex of adult WT mice and KO mice (From left to right. The first panel, the WT prefrontal cortex; The second panel, the magnified image for WT mPFC; The third panel, the negative probe for non-specific detection; The fourth panel, Gadd45a KO prefrontal cortex.) The dashed white box represents the enlarged view. Scale bars: 200μm. B RNAscope for Gadd45a (red) and immunofluorescence for CamKII (excitatory neurons, green) and Gad67 (inhibitory neurons, green) demonstrate major expression of Gadd45a in excitatory neurons. In the left panel, arrowheads mark CamKII and Gadd45a double-positive neurons. In the right panel, arrowheads mark Gad67-positive and Gadd45a-negative neurons. The white box represents the enlarged view. Scale bars: 20μm. C Relative quantification of Gadd45a-positive, CamKII-positive and Gad67-positive cells. D Schematic illustration of the viral injection site. E Representative photograph shows the accurate position of AAV infection (EGFP, green) in mPFC 4 weeks after AAV injection. F Representative heat maps of the trajectory of the two groups (KO mice subjected to control AAV or AAV-Gadd45a) in the two stages of the Three-Chamber tests. G The upper panel: In the sociability test (Stage 1), the group of KO mice injected by AAV-Gadd45a show a preference for the stranger 1 (S1) compared with Empty Cylinder (EM); S1 versus EM, paired Student’s t-test. The bottom panel: In the Social novelty test (Stage 2), KO mice injected by AAV-Gadd45a spend more interaction time with stranger 2 than stranger 1; S1 versus S2, paired Student’s t-test. AAV-Control group, n = 9, AAV-Gadd45a group mice, n = 8. *p < 0.05, *** p < 0.001, and ns, no significant difference. All data are presented as means ± SD.

Full size image

Gadd45a KO mice display no impairment in synaptic morphology and function

Next, we explored whether Gadd45a deletion in mice leads to defects in neurodevelopmental disorders. Considering the canonical function of Gadd45a in regulating the cell cycle, we performed Nissel staining in the prefrontal cortices and found no obvious abnormalities were detected in brain anatomy and cortical lamination (Fig. S4A). We then evaluated the synaptic development in Gadd45a knockout mice. To visualize dendritic spines, we sectioned the prefrontal cortices of WT mice and KO mice and performed Golgi staining. The density and morphology of spines were analyzed in each neuron. Gadd45a knockout mice had no difference in spine density (Fig. S4B, S4C). Furthermore, the Input-Output curve test of KO mice showed a similar slope to WT mice (Fig. S4D). The electrophysiological result proved the absence of obvious abnormalities in spine density and release probability.

Excitatory neurons in Gadd45a knockout mice exhibit abnormal excitement at rest and task states

To elucidate precisely the causative pathological features of neurons in Gadd45a knockout mice exhibiting autistic behaviors, we used chronic electrophysiological recordings in adult male mice. In order to record and recognize the spiking activities of excitatory neurons in mPFC of mice, a tetrode electrode array (2*2) has been chosen (Fig. 4A, B). The mice were first moved into a field. After 1-hour habituation, we started to record electrophysiology data. According to the width of the half-wave of the spike waveform, the spike was divided into wide-spiking (WS, trough-to-peak duration> 350 μs) and narrow-spiking (NS, trough-to-peak duration< 300 μs) (Fig. S4E). The WS was considered as excitatory neurons, following Liu’s method [31]. The frequency of WS across all mice was examined. WS from Gadd45a knockout mice was significantly higher than the wildtype group in the rest state (n = 6, p < 0.05) (Fig. 4C, S4F). In addition, we also tested the WS of rescued AAV mice and revealed that after injection of rescued AAV the WS was distinctly reduced compared with the counterpart of KO mice subjected to control AAV injection (Fig. 4D). To explore the correlation between mPFC neuronal activities and real-time social interaction, after recording spikes in the field, we moved mice to the Three-chamber and synchronously recorded electrophysiology data during mice under the whole process of the Three-chamber experiment. The process of the Three-chamber experiment was described in the method section (Fig. 4E, H). In stage one, the social situation was defined as mice in the region of caged mice (electrophysiology data were labeled social). While the mice stayed in the middle region, we defined them as mice in the rest situation (electrophysiology data were labeled as baseline). In stage two, the social situation was defined as mice near the stranger mice, while the rest/baseline stage was the mice stayed in the middle field of the Three-chamber. The frequency of spikes in different situations from neurons was analyzed. In stage one, nearly all WS neurons from wildtype mice exhibited an increase in spike firing rate in social conditions, compared to under baseline conditions (Fig. 4F, G). Although WS neurons from Gadd45a knockout mice might exhibit hyperexcitability under baseline conditions, knockout mice did not show a significant firing-rate alternation when mice switched into the social stage (Fig. 4F, G). Not surprisingly, in stage two of the Three-chamber experiment, only WS neurons from wildtype mice exhibited a firing-rate increase while mice were close to the stranger. The knockout mice neurons remained at a nearly baseline level under social situations (Fig. 4I, J). These results indicate that Gadd45a knockout resulted in aberrant excitability of excitatory neurons.

Fig. 4: KO mice display abnormal excitability of excitatory neurons.
figure 4

A Diagram showing electrophysiological recording of mPFC in mice. B A representative photograph shows the location of the electrophysiological recording site in PrL of mPFC in all the following electrophysiological experiments. The green signals (EGFP) represent the position of AAV infection in (D) experiments. C The excitatory neurons in the mPFC of KO mice at the resting state show higher firing frequency than their counterpart of WT mice. D The excitatory neurons in the mPFC of KO mice injected with AAV-Gadd45a at resting state show reduced firing frequency than the counterparts of KO mice injected with AAV-Control. Data in (C, D) were calculated with the average firing frequency of units from 6 mice per group. Unpaired Student’s t-test. E, H Schematic illustration of the electrophysiological recording paradigm in two stages of the Three-chamber test. F, I Raster plots of spikes of the example WS neuron in three individual mice of WT group and KO group are shown during Stage 1 and Stage 2. The social state is marked as a gray zone. The whole period of a raster plot lasted 20 seconds. G, J In stage 1 and stage 2, WS neurons in WT mice show enhanced firing rates during the social state. However, WS neurons in KO mice displayed no difference between social state and baseline. Paired Student’s t-test. * p < 0.05, ** p < 0.01, and ns, no significant difference. All data are presented as means ± SD.

Full size image

GADD45A interacts with TET1 and modulates Kcnq5 expression to maintain neural function

We try to clarify how Gadd45a maintains the normal function of neurons. A previous study has shown that GADD45A can collaborate with TET1 to regulate DNA demethylation as an epigenetic-related factor [16]. In addition, it has been shown that Tet1 deletion is reported to be closely associated with autism through methylation-dependent regulation of Oxtr [32]. We first verified the interaction of GADD45A and TET1 in the prefrontal cortex through co-immunoprecipitation and co-localization of GADD45A and TET1 immunofluorescence (Fig. 5A, S5A), which indicated that the loss of GADD45A may influence the TET1 normal function. The dot blot of 5-mc revealed that the DNA methylation in the prefrontal cortices had a certain degree of elevation (Fig. S5B). To explore the regulatory mechanism of GADD45A in neuron excitability, we first used bulk RNA-sequencing to find the differentially expressed genes in the prefrontal cortices between wildtype mice and knockout mice. The differentially expressed genes from transcriptome sequencing were subjected to gene function enrichment analysis. The KEGG results showed that the differential genes in KO mice’s prefrontal cortices were mainly enriched in the pathway: Neuroactive ligand-receptor interaction (Fig. 5B). We also performed KEGG enrichment of upregulated and downregulated genes, respectively (Fig. S5C). Especially GSEA analysis and KEGG enrichment of upregulated genes showed the gene pathway Neuroactive ligand-receptor interaction was elevated in KO prefrontal cortex (Fig. S5C, D). These results are highly correlated with neuron excitability, which may account for the phenotype of KO mice. Then, we performed the TET1 ChIP-seq in the prefrontal cortex of knockout mice and WT mice to investigate how Gadd45a deletion influences the function of TET1 in the prefrontal cortex. We intersected the differentially expressed genes identified by RNA-seq with the differential binding targets obtained from ChIP-seq analysis (Fig. 5C). Given the DNA demethylation function of TET1, we further analyzed genes that showed concordant changes in both differentially expressed genes in RNA-seq and differential ChIP-seq target genes. Intriguingly, we found a potential downstream gene, Kcnq5, simultaneously present in differentially expressed genes in both the transcriptome and ChIP-seq data (Fig. 5C, D, S5E, S5F). This gene is a member of the KCNQ potassium channel gene family that is differentially expressed in subregions of the brain and skeletal muscle. The protein encoded by this gene yields currents that activate slowly with depolarization and can form heteromeric channels with the proteins encoded by the KCNQ2 and KCNQ3 genes [33, 34]. The ChIP qPCR and Methylation-specific PCR in mice prefrontal cortices results showed that Gadd45a knockout caused the reduced binding strength of TET1 in the promoter region of Kcnq5, and the increasing methylation level of the CpG island (CGI) in the promoter region of Kcnq5 (Fig. 5E, F), which could explain the decrease of Kcnq5 mRNA level (Fig. 5G). Moreover, we examined the sharp decline in KCNQ5 protein level in Gadd45a knockout prefrontal cortices (Fig. 5H). Due to the study reporting that KCNQ2 and KCNQ5 could form heteromers [33], we also found that KCNQ2 protein level decreased in the prefrontal cortices of KO mice (Fig. S5G, H), which was matched with the previous study reporting Kcnq2 knockout mice have decreased KCNQ5 protein level [35]. It might result from the instability of KCNQ potassium channels caused by the loss of KCNQ5. To determine whether reduced Kcnq5 expression in excitatory neurons of the mPFC is sufficient to induce autism-like phenotypes, we used an AAV-mediated approach to selectively knock down Kcnq5 in excitatory neurons of mPFC in wild-type mice (Fig. 5I). Strikingly, the Three-chamber test revealed that these mice exhibited significantly impaired sociability and social novelty compared with control mice (Fig. 5J, K), which closely resembled the phenotype observed in Gadd45a knockout animals. Subsequently, we tested whether the FDA-approved KCNQ2-5 agonist, Retigabine, could rescue the autistic behaviors of Gadd45a KO mice. Notably, after acute Retigabine treatment for KO mice, in the sociability phase, the interaction time of the rescued group with S1 became evidently more than the interaction time with EM, while in the social novelty phase, the interaction time of the rescued group with S2 increased but had no significance compared to the time with S1 (Fig. 5L, M). We also recorded electrophysiology data of KO mice after the retigabine treatment. Retigabine could partly reduce the firing frequency of excitatory neurons in the mPFC of KO mice at the rest state (Fig. S5I). Furthermore, we used an AAV vector to selectively restore KCNQ5 expression in excitatory neurons of mPFC in Gadd45a knockout mice (Fig. S5J). The Three-chamber experiment showed that KCNQ5 restoration significantly rescued sociability (In Stage 1) (Fig. S5K). Although the improvement in social novelty (In Stage 2) did not reach statistical significance, there was an increasing trend in interaction time with Stranger 2 (Fig. S5L). These results supported the conclusion that impaired KCNQ5 function, resulting from reduced Kcnq5 transcription, is a critical contributor to the autism-like phenotypes observed in Gadd45a KO mice.

Fig. 5: The deletion of Gadd45a caused insufficient binding of TET1 to regulate the expression of Kcnq5.
figure 5

A Co-IP experiments show the interaction of GADD45A and TET1 in the prefrontal cortex of WT mice. B Differentially expressed genes for KEGG enrichment (Qvalue < =0.05) in the prefrontal cortex between KO and WT groups. C Venn plot for the overlap of significantly dysregulated genes from RNA-seq with dysregulated genes from TET1 ChIP-seq in Gadd45a KO prefrontal cortices. D ChIP-seq profiling of TET1 in wild-type and Gadd45a knockout mice at the Kcnq5 locus. The TET1 peak is around the transcriptional start site of Kcnq5. The bottom two tracks show H3K4me3 signals around the Kcnq5 transcription start site, based on public ChIP-seq data from neuronal nuclei of mouse forebrains (GSE190102), marking active promoter regions. The dashed black box represents the significantly different peak of TET1 binding to the promoter of Kcnq5. Log2 fold-change of binding within the region is shown relative to wild-type with the p-value of the comparison. Two biological replicates per genotype were used for ChIP-seq. E ChIP-qPCR analysis for TET1 targeting at the target region of the Kcnq5 promoter was performed in the prefrontal cortex of WT and KO groups. The fold enrichment of TET1 was normalized by ChIP of IgG. n = 4, per group, unpaired Student’s t-test. F MS-PCR analysis of methylation (M) and unmethylation (U) levels in the prefrontal cortex between WT and KO groups (upper panel). The relative intensities of methylation (M) and unmethylation (U) bands were quantified and plotted (bottom panel). Each group n = 4, unpaired Student’s t-test. G qPCR data reveals the downregulation of Kcnq5 expression in the prefrontal cortex of KO mice. n = 4, per group, unpaired Student’s t-test. H The Western Blot result shows an intense reduction in KCNQ5 expression in the prefrontal cortex of KO mice. The grayscale analysis of the protein bands is shown in the right panel. n = 4, each group, unpaired Student’s t-test. I The upper left panel: Schematic illustration of the viral injection site. The upper right panel: Representative photograph shows the accurate position of AAV infection (mCherry, red) in mPFC 4 weeks after AAV injection. Scale bars: 100μm. The bottom panel: 4 weeks after injection of AAV-shRNA-Kcnq5, the mouse brain sections of mPFC were double-stained for KCNQ5 and CamKII, which shows selective knockdown of the Kcnq5 expression in the excitatory neurons. The white box represents the enlarged view. Scale bars: 10μm. J In the sociability test (Stage 1), the group of WT mice injected by AAV-shRNA-Kcnq5 shows no preference for the stranger 1 (S1) compared with Empty Cylinder (EM); S1 versus EM, paired Student’s t-test. K In the Social novelty test (Stage 2), WT mice injected by AAV-shRNA-Kcnq5 spend comparable interaction time between stranger 2 and stranger 1; S1 versus S2, paired Student’s t-test. AAV-Control group, n = 8, AAV-shRNA-Kcnq5 group mice, n = 10. L In the sociability test, KO mice treated with retigabine (2.5 mg/kg) (Retigabine group) preferred to interact with the stranger 1 (S1) compared with the Empty Cylinder (EM) but KO mice injected with vehicle (Control group) showed no preference; S1 versus EM, paired Student’s t-test. Each group, n = 9. M In the Social novelty test, Retigabine group seemed to spend more interaction time with stranger 2 than with stranger 1, though the difference was not significant; S1 versus S2, paired Student’s t-test. Each group, n = 9. *p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001 and ns, no significant difference. All data are presented as means ± SD.

Full size image

R-loop structure comprising lncRNA KCNQ5-DT mediates GADD45A/TET1 to regulate KCNQ5 in vitro

Recent studies have indicated that GADD45A could modulate the expression of downstream genes by engaging in interactions with TET1, which are mediated through R-loop formation [16]. Therefore, we wonder whether the GADD45A-TET1 complex is mediated by the R-loop structure to regulate Kcnq5. Firstly, we used SiRNA targeted to GADD45A to construct the GADD45A knockdown model in the human neuroblastoma SH-SY5Y cell line. We validated through qPCR that the knockdown of GADD45A in SH-SY5Y cells resulted in a decrease in the transcription levels of KCNQ5 (Fig. 6A), yielding consistent findings with in vivo experiments. What is more, we verified the co-localization of GADD45A and TET1 in the nucleus of SH-SY5Y cells using immunofluorescence (Fig. S6A, B). Additionally, following the knockdown of the cell line and treatment with 5-Azacyidine (5-AZA), which is a DNA methyltransferase (DNMT) inhibitor, for 48 h, qPCR results revealed a restoration in the transcription levels of KCNQ5 (Fig. 6B). This indicated that the regulation of KCNQ5 by GADD45A in humans is dependent on DNA methylation, suggesting a conservative regulatory mechanism in human and mouse. Subsequently, to monitor the impact of the R-loop structure on the regulation of GADD45A and TET1 on KCNQ5, we utilized ribonuclease H1 (RNH1), which functions to cleave the RNA portion of RNA/DNA hybrids, to remove R-loops. We then used the S9.6 antibody, which can specifically recognize R-loops, for immunoprecipitation in the lysates of SH-SY5Y cells transfected with the RNH1 plasmid or empty vector. We found that the intensity of the GADD45A protein band was reduced in the lane of immunoprecipitation using S9.6 antibody when the R-loops were partially eliminated. It proved that the GADD45A protein could bind to R-loop structures in the SH-SY5Y cell line. (Fig. 6C). We also examined the expression of GADD45A and KCNQ5 when we overexpressed RNH1 in SH-SY5Y. The qPCR and western blot results showed that deprivation of R-loop did not influence the expression level of GADD45A, but lowered the expression of KCNQ5 protein level (Fig. 6D, S6C, S6D). Furthermore, it is noted that the TET1 occupancy at the CGI of KCNQ5 promoter was also decreased after removing R-loop by RNH1 plasmid transfection (Fig. 6E). KCNQ5 has an antisense orientation transcript, lncRNA KCNQ5-DT, overlapping with a CGI around the transcription start site (TSS) of KCNQ5. There is a GC skew within exon 1-2 of KCNQ5-DT (Fig. 6F). It was reported that nearby lncRNAs and GC skew could favor R-loop formation [16, 36, 37]. To evaluate the binding of GADD45A to R-loops near the TSS of KCNQ5, we first designed three amplification regions in the KCNQ5-DT exon region. Amplicon 1 exhibits GC skew on exon 2, amplicon 2 exhibits GC skew on exon 1, and amplicon 3 overlaps with the KCNQ5 CGI region on exon 1(Fig. 6F). Subsequently, exogenous Flag-GADD45A was overexpressed in SH-SY5Y cells, and ChIP-qPCR revealed that GADD45A exclusively binds to amplicon 1(Exon 2 of KCNQ5-DT), but not to amplicons 2 and 3 (Fig. 6G). Furthermore, we co-overexpressed RNH1 and Flag-GADD45A. Overexpression of RNH1 reduced GADD45A binding to amplicon 1(Fig. 6H), emphasizing that the association of GADD45A with KCNQ5 requires R-loop formation. It indicated amplicon 1(Exon 2 of KCNQ5-DT) preferentially formed R-loop. Furthermore, due to the potential role of KCNQ5-DT in nearby R-loop formation, we used KCNQ5-DT SiRNA to knock down the lncRNA, which would lead to less R-loop formation. When we knocked down KCNQ5-DT, the qPCR results showed that the expression level of KCNQ5 was also accordingly reduced (Fig. 6I). In addition, through ChIP-qPCR in Flag-GADD45A-expressed cells, the knockdown of KCNQ5-DT led to less occupancy of GADD45A on the R-loop site (Amplicon 1) (Fig. 6J). Furthermore, we pretreated freshly dissected mPFC tissue with RNase H1 and S1 nuclease to digest R-loop structures in situ, and then performed ChIP-qPCR to assess TET1 binding at the Kcnq5 promoter CpG island. We found that the disruption of R-loops significantly reduced TET1 enrichment at the Kcnq5 promoter (Fig. S6E), supporting the notion that R-loop formation is required for TET1 recruitment of the Kcnq5 promoter in mouse mPFC tissue. Above all, these results suggested that the R-loop is involved in the process of GADD45A interacting with TET1 to bind to the CGI region and exert gene regulation through an anti-sense KCNQ5-DT mediated transcriptional regulation.

Fig. 6: The process of GADD45A/TET1 regulating the transcription of KCNQ5 requires the R-loop-mediated regulation.
figure 6

A The RT-qPCR result shows that the knockdown of GADD45A in vitro also leads to a decline in the expression of KCNQ5 mRNA. Three individual biological replications per group. One-way ANOVA followed by Dunnett’s multiple-comparison test. B SH-SY5Y cells were treated with 5 μM 5-AZA for 24 h after transfection of GADD45A siRNAs. RT-qPCR reveals the restoration of KCNQ5 mRNA level. Three individual biological replications per group. Two-way ANOVA with Sidak’s multiple comparisons test. C By using the S9.6 antibody to immunoprecipitate R-loop/Protein complexes in SH-SY5Y cell lysates with or without overexpression of RNase H1(RNH1), the Western Blot result shows that GADD45A can bind to the R-loop structure. ACTB is used for negative control, which does not bind to R-loop. Asterisk indicates the heavy-chain staining. D The RT-qPCR experiments were used to detect the expression of KCNQ5 and GADD45A in SH-SY5Y cells with or without expression of RNH1. Each group n = 3, unpaired Student’s t-test. E ChIP-qPCR of TET1 at the CGI region of the KCNQ5 promoter in SH-SY5Y cells overexpressing RNH1 or control plasmid. Data are normalized to the ChIP of IgG. Each group n = 3, unpaired Student’s t-test. F Top, the KCNQ5 and KCNQ5-DT loci are shown with the plot of GC skew. Bottom, GC% (green line) and GC skew (red line) in the exon 2 of KCNQ5-DT (in the Amplicon 1). Grey shading indicates the high GC skew score. The Amplicon 1 is located in the region (from the 200th to the 400th) G After exogenous expression of Flag-GADD45A, Flag ChIP-qPCR analysis shows the occupancy of GADD45A in Amplicon 1 in exon 2 rather than Amplicon 2 and 3 in exon 1 of KCNQ5-DT. Both Flag and IgG are assessed in three regions, as presented in (F), and normalized by input. Each group, n = 3. Two-way ANOVA with Sidak’s multiple comparisons test. H ChIP-qPCR of Flag-GADD45A at the exon 2(Amplicon 1) region of KCNQ5-DT in SH-SY5Y cells overexpressing RNH1 or control plasmid. Data are normalized to the ChIP of IgG. Each group, n = 3, unpaired Student’s t-test. I The RT-qPCR shows that the knockdown of KCNQ5-DT in vitro also leads to a decline in the expression of KCNQ5 mRNA. Each group, n = 3, unpaired Student’s t-test. J In the exogenously Flag-GADD45A-expressed SH-SY5Y cells, the ChIP qPCR proves that knockdown of KCNQ5-DT can attenuate the binding of GADD45A to amplicon 1. Data are normalized to the ChIP of IgG. Each group, n = 3, unpaired Student’s t-test. *p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001 and ns, no significant difference. All data are presented as means ± SD.

Full size image

Discussion

The potential mechanism that the loss of Gadd45a leads to autistic behaviors is displayed in the schematic diagram (Fig. 7). VPA exposure and MECP2 gene alteration represent the most established etiology of drug and monogenic cause of autism in both patients and rodent models. Here, using transcriptome analysis from VPA-treated and MECP2 models of biological samples, we have identified a key gene, GADD45A, which is co-regulated by both environmental and genetic perturbations. Through detailed characterization of Gadd45a KO mice, for the first time, we have found that deletion of Gadd45a specifically led to impaired sociability, social novelty, significantly increased digging behaviors, and spontaneous seizure but did not obviously influence anxiety, memory, and compulsive behaviors, which is related to the etiology of neurodevelopmental disease, including autism, as well as epilepsy. Specifically, we have found that Gadd45a expression in excitatory neurons of the prefrontal cortex seems to play a critical role in the etiology of the phenotype. Furthermore, we used AAV-Gadd45a to restore the insufficiency of Gadd45a in mPFC and found that the damaged sociability and social novelty of KO mice have been mostly rescued. Mechanistically, we have found that deletion of Gadd45a leads to the abnormal excitability of excitatory neurons, accompanied by major gene pathway changes, which include neuroactive ligand-receptor interaction, as well as ion channels. Specifically, through analysis of multi-omics, including RNA-seq and ChIP-seq assays, we have found that GADD45A can bind to promoter-adjacent R-loops and recruit TET1 to the promoter of the ion channel genes, KCNQ5, which plays a key role in the pathogenesis of autism-related phenotypes of Gadd45a KO mice.

Fig. 7: The schematic diagram of the potential mechanism that the loss of Gadd45a leads to autistic behaviors.
figure 7

GADD45A plays a critical role in regulating the expression of Kcnq5, which encodes a potassium channel, through its interaction with TET1 in an R-loop-dependent manner. In excitatory neurons, GADD45A binds to R-loop loci and recruits TET1 to the Kcnq5 promoter to regulate transcription. Loss of Gadd45a disrupts the normal expression and function of KCNQ5 and other ion channels during neurogenesis, thereby contributing to autism-like behaviors. Consistently, AAV-mediated knockdown of Kcnq5 in excitatory neurons of the mPFC (AAV-shKcnq5) can impair social behaviors in wild-type mice. Conversely, restoring Gadd45a expression (AAV-Gadd45a), re-expressing Kcnq5 (AAV-Kcnq5), or enhancing KCNQ5 channel activity with the Kv7 channel activator retigabine can rescue the autism-like phenotypes in Gadd45a KO mice. Together, these findings highlight the GADD45A-TET1-KCNQ5 axis as a potential therapeutic target for ASD, particularly in the context of MECP2 dosage abnormalities and VPA exposure. The figure was created in BioRender. Song, Y. (2025) https://BioRender.com/ntz3fix.

Full size image

Sex differences in the prevalence and clinical presentation of ASD are well established, with males being disproportionately affected [38, 39]. This raises the possibility that certain molecular mechanisms may exert sex-specific effects in ASD pathogenesis. Although the main experiments and mechanistic investigations in this study were performed in male Gadd45a knockout mice, we also examined whether Gadd45a disruption would lead to autism-related phenotypes in females. We performed Three-chamber behavioral tests in female Gadd45a KO mice and found that their social behaviors (sociability and social novelty) were not impaired (Fig. S7A, B). Furthermore, we examined the protein levels of KCNQ5, the key downstream molecule that we identified as mediating the autism-like phenotypes in male KO mice. In female KO mice, KCNQ5 expression in the prefrontal cortex was comparable to that of wild-type mice (Fig. S7C, D). These results suggest that the disruption of Gadd45a does not significantly affect social behaviors or KCNQ5 expression in females, implying that GADD45A may play a more critical role in the molecular mechanisms underlying autism pathogenesis in males. In addition, our preliminary analysis of peripheral blood from a small cohort of female participants (3 ASD patients and 3 controls) did not reveal a marked reduction of GADD45A expression in ASD (Fig. S7E). Future studies with larger cohorts of both sexes and with broader molecular analyses will be needed to fully validate whether certain functions of GADD45A may be compensated in females and to further elucidate the sex-dependent functions of GADD45A. Such efforts may ultimately inform more precise and personalized therapeutic strategies that take into account sex differences in ASD.

GADD45A has been traditionally regarded as a stress-responsive gene critical for cell cycle regulation, such as Cyclin-D, in response to stress and UV radiation. However, one of the major novel discoveries in our study, here we have found a novel regulatory axis that, in excitatory neurons, GADD45A can regulate the expression of the K+ ion channel through recruiting TET1 to the promoter of KCNQ5. As the downstream gene KCNQ5 is a member of the KCNQ (Kv7) potassium channel family, it was recently reported that the variants of Kv7 in humans were linked to autism spectrum disorders [40,41,42]. Consistent with our finding that KCNQ5 levels were reduced in prefrontal excitatory neurons in our model, recent single-nucleus RNA sequencing data [43] from a large ASD patient cohort also revealed decreased KCNQ5 mRNA expression in certain excitatory neuronal subtypes compared with controls (Fig. S7F). In our study, we proved that the knockdown of Kcnq5 in the excitatory neurons of mPFC could compromise the sociability of mice. Furthermore, we found that KCNQ2-5 agonist Retigabine and AAV-mediated restoration of KCNQ5 expression level could partially reverse the defects of social ability in Gadd45a KO mice, which indicated the impaired function of K+ channels may be the major contributor to the autistic phenotype in our mouse model. In line with our findings of retigabine rescue in KO mice, several recent studies found that this Kv7-targeting agonist showed efficacy in ASD-relevant models [44, 45]: in Ank2 cKO mice, retigabine rescued cortical hyperexcitability and juvenile seizure-related mortality, and in Fmr1-KO hippocampus, retigabine normalized CA1 excitability and abolished seizure-like events. However, according to FDA information, the clinical utility of retigabine is limited by boxed-warning toxicities (retinal abnormalities/vision risk, skin or mucosal discoloration, and urinary retention). To better enable translational progress toward ASD therapeutics, next-generation Kv7 activators should prioritize: improving chemical stability; maximizing selectivity for neuronal KCNQ2-5 over cardiac KCNQ1; optimizing brain exposure and metabolism; and minimizing pigmentation risk while widening the safety margin [46]. Notably, structurally optimized Kv7 activators (e.g., pynegabine/HN37) are already being tested in clinical trials among people with epilepsy [47, 48], which may contribute to therapeutic development for ASD in the future. Therefore, the GADD45A–TET1–KCNQ5 regulatory axis we proposed in this study may serve as an important basis for understanding ASD etiology and a potential direction for future therapeutic exploration.

This study further revealed that the special genomic structure R-loop is involved in regulating the normal function of neurons to maintain neuronal excitability. R-loop structure has been recently shown to play a critical role in neural stem cell differentiation [49]. In vitro, we have demonstrated that GADD45A, as a reader of R-loops, can recognize the R-loop near the upstream of the KCNQ5 promoter (exon 2 of KCNQ5-DT) formed by ssDNA and lncRNA (KCNQ5-DT), mediating the epigenetic regulation of KCNQ5 by guiding TET1. We have identified that KCNQ5-DT plays a crucial role in forming R-loop near KCNQ5 in the human genome. Although the exact details of lncRNA involved in this R-loop in the mouse genome are still elusive due to the absence of known adjacent lncRNA by now, we still proved the in vivo existence of an R-loop structure that mediates TET1 binding to the CpG island of Kcnq5 by enzymatically digesting R-loop structures in the mouse cortex. Currently, there are few reports on the role of R-loops in the pathogenesis of autism. Recently, Hun-Goo Lee proposed that utilizing R-loop formation to promote endogenous repair mechanisms can drive the excision of the long CGG repeat to treat fragile X syndrome [50]. Our study suggests that the reader factor of R-loops, like GADD45A, may also mediate the etiology of autism, laying a theoretical foundation for future research on epigenetic therapy for autism spectrum disease.

The expression of GADD45A is also altered in autism patients, indicating that the dysregulation of GADD45A is a core characteristic of autism patients. Interestingly, we have found that GADD45A is a potential gene regulated by MECP2. Mecp2 TG mice cortex showed reduced Gadd45a expression. Interestingly, Mecp2 TG mice shared some common phenotypes with our Gadd45a knockout mice. The mPFC in Mecp2 TG mice exhibited abnormal overexcitation similar to our electrophysiological recordings in Gadd45a KO mice [11, 12]. It indicates that MECP2-GADD45A/TET1-KCNQ5 could form a critical circuit, mediating the phenotype of autism, as well as epilepsy. Moreover, epilepsy in MECP2 duplication syndrome is common, partly consistent with spontaneous seizure in Gadd45a KO mice [51]. Therefore, GADD45A may be a potent target gene of the therapy for MECP2 duplication syndrome. Since an increase of GADD45A is observed after VPA exposure in neurodevelopment and MECP2 loss of function, whether this elevation directly contributes to the onset of autism requires further investigation. In future studies, it will be important to explore whether reducing GADD45A expression to physiological levels—either following VPA exposure or in Rett syndrome models—can ameliorate autism-related phenotypes, thereby assessing GADD45A as a potential therapeutic target that requires precise dosage control.

In conclusion, this study reports the novel autism candidate gene GADD45A and sheds light on the poorly understood biological role of GADD45A-TET1/R-loop/KCNQ5 in neurons to maintain normal neuronal excitability. Therefore, the GADD45A-TET1/R-loop/KCNQ5 regulatory axis we proposed in this study may serve as an important basis for understanding ASD etiology and a potential direction for future therapeutic exploration.

Materials and methods

Study participants

The boys aged less than 8 years old with ASD were recruited as autism groups from muti-centers in Sichuan province in China. The control groups at the same age range were also obtained from boys attending outpatient clinics. The study obtained informed consent from the parents or legal guardians of the boys’ participants. Inclusion criteria for the control group comprised being under 8 years old and devoid of secondary sexual characteristics. Exclusion criteria encompassed sexual dysplasia, tumors, developmental delay, mental retardation, epilepsy, and other neurological and psychiatric disorders. The inclusion criteria for boys with ASD included being diagnosed with ASD according to the DSM-5 criteria, validated by the Autism Diagnostic Observation Schedule 2nd revision (ADOS-2) and the provision of informed consent. Exclusion criteria for this group involved organic heart disease and abnormal sexual development. In Cohort 1, those participants were further assessed by the CBCL scale, ABC scale, SRS scale, and RBS-R scale, which are different standardized questionnaires used to score autistic features with the aim of screening children with autism. In addition, Whole-blood samples from the ASD and control individuals were collected for further RNA extraction. It followed the principles outlined in the Declaration of Helsinki and received approval from the Human Ethics Committee of West China Second University Hospital of Sichuan University (2020096) and the Third People’s Hospital under Ethics ([2020]s-112).

Gadd45a-knockout (−/−) (KO) mice

All animal experiments in this study were conducted following the approved ethical guidelines and were approved by the Animal Care and Research Committee of Sichuan University (K2019038). C57BL/6 Gadd45a-heterozygous (+/−) mice were constructed by means of the CRISPR/Cas9 method (View solid-biotech, Beijing, China) with deletion of a 2007-bp fragment encompassing exons 1–4 (Fig. 2A). Details are available in Supplementary Methods.

Western blotting

40 μg of protein lysate was added into each lane and was separated by 12% SDS-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to polyvinylidene difluoride (PVDF) membranes. Following a 1-hour blocking step at room temperature in 5% milk in 1×TBST, the membranes were incubated overnight at 4 °C with specific primary antibodies. After extensive washing with TBST, the primary antibodies were detected using HRP-conjugated anti-rabbit or anti-mouse secondary antibodies. Finally, an enhanced ECL chemical substrate was utilized for visualizing the developed immunoblot. Details are available in Supplementary Methods.

Immunofluorescence and RNAscope detection

The cells were fixed for immunofluorescent staining and cultured on confocal dishes. Paraffin-embedded and frozen brain sections were used for immunofluorescence staining and RNAscope assays. Localizing the GADD45A protein in the mouse brain is challenging due to the non-specific labeling of commercially available antibodies. Therefore, we used the enhanced in situ hybridization technique, RNAscope, which can visualize the individual RNA molecules and has low background noise [52], to label Gadd45a mRNA using the Gadd45a RNAscope probe (Cat. No. 460571, ACD). Details of immunofluorescence and RNAscope detection are available in Supplementary Methods.

RNA extraction, cDNA synthesis, and quantitative real-time PCR

Samples’ total RNA was extracted utilizing TRIzol (Thermo Fisher, 10296010), chloroform, and isopropyl alcohol. Subsequently, cDNAs were synthesized and amplified using the HiScript 1st strand cDNA Synthesis Kit (Vazyme, R111). RT-PCR was conducted on an Applied Biosystems HT7900 using the Power SYBR Green PCR Master Mix (Applied Biosystems, 4368706) in accordance with the manufacturer’s protocols. The primers used are listed in the supplementary materials (Table S1).

Stereotaxic surgery

4-week-old mice were anesthetized with 1% pentobarbital sodium (5 mL/kg body weight) and positioned on a stereotaxic frame (RWD Life Science, China) for viral microinjection. Details are available in Supplementary Methods.

Cell culture and transfection

Human neuroblastoma SH-SY5Y cells were obtained from the Precella company. Cells were cultured in DMEM/F12 with 15% FBS (Thermo Fisher) and 1% penicillin and streptomycin at 37 °C. In the drug treatment assay, cells were treated with or without 1 mM VPA (MedChemExpress, S3944) or 5 µM 5' AZA (MedChemExpress, HY-10586). Primary cortical neurons and mixed glia were isolated from embryos and P0 pups as previously described [53, 54]. Expression plasmids encoding Flag-GADD45A, GADD45A SiRNAs and KCNQ5-DT SiRNA were synthesized by Gene Universal (Chuzhou, China). The RNH1 plasmid was synthesized by Youbio (Changsha, China). The detailed SiRNAs and plasmid information are listed in Tables S2 and S3. The plasmids and SiRNAs were transiently transfected using the jetPRIME Transfection Reagent (101000046, Polyplus) following the manufacturer’s protocol.

RNA-seq and Chromatin immunoprecipitation (ChIP) assays

Full methods are available in Supplementary Methods.

Co-immunoprecipitation (Co-IP) assays

Proteins extracted from mouse prefrontal cortex tissues were lysed in RIPA lysis buffer for CO-IP assays. Full methods are available in Supplementary Methods.

Methylation-specific PCR (MSP) and dot blot

Full methods are available in Supplementary Methods.

Golgi staining and Nissl staining

The density of dendritic spines in the cortex was quantified through Golgi staining using the Golgi–Cox Kit from HEPENGBIO (Shanghai). The images were captured by the Pannoramic MIDI II scanner. Then the spines were counted and analyzed as the number per 10 µm with the assistance of ImageJ software. Nissl staining is a useful tool to indicate neuron state, which was utilized for nucleic acid visualization. 40-mm-thick cryostat sections were stained with Nissl Staining Solution (Beyotime, C0117) following the manufacturer’s protocol.

Behavioral tests

Behavioral tests were conducted during the light-on period of the day using age-matched male littermates aged 8–12 weeks, except that in the Three-chamber test of Figure S7, female mice were used. Prior to testing, mice were acclimated to the handler through daily handling sessions of 2-3 min each for 5 days. A minimum of 2 days of rest was provided between each test session. Full methods of each test involved are available in Supplementary Methods.

In vivo electrophysiological recording

During the experiment, the mice were first anesthetized with 4% isoflurane (i.r) in the chamber. Then the mouse was moved to the stereotaxic instrument (68030, RWD) to fix the head. In this process, the dose of isoflurane was kept at 2% to 1%. The scalp was removed after sterilization with iodophors and 75% ethanol. For the input-output experiment, A craniotomy (1.5-2 mm in diameter) was made in the region of CA1 (AP-1.2 mm; ML1.0 mm; DV-1.8 mm) for the stimulation electrode and the Prelimbic cortex (PrL) (AP1.5; ML0.4; DV-1.8) for the recording electrode. Details are available in Supplementary Methods.

Quantification and statistical analysis

Data were reported as the mean ± SD. GraphPad Prism 8.0 was used for analyzing the data and figure generation. Levene’s test was used to evaluate the homogeneity of variance of the data. Two-tailed Student’s t-test was used to compare differences between the two groups. ANOVA and Fisher’s least significant difference (LSD) test were used to compare differences among multiple groups. The correlation analysis was performed by the Spearman correlation coefficient(r). p < 0.05 was considered statistically significant. Significance was reported as ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. No significant values are not denoted except for emphasis. In addition, for Fig. 6F, the GC skew score was calculated using the formula (nG–nC) / (nG + nC), where n represents the number of guanine (G) or cytosine (C) nucleotides within a sliding window of 100 bases. The GC skew distribution across the genome was then visualized using the Proksee platform (https://proksee.ca/).