Postnatal Arx transcriptional activity regulates functional properties of PV interneurons

Summary

The transcription factor Aristaless-related X-linked gene (Arx) is a monogenic factor in early onset epileptic encephalopathies (EOEEs) and a fundamental regulator of early stages of brain development. However, Arx expression persists in mature GABAergic neurons with an unknown role. To address this issue, we generated a conditional knockout (CKO) mouse in which postnatal Arx was ablated in parvalbumin interneurons (PVIs). Electroencephalogram (EEG) recordings in CKO mice revealed an increase in theta oscillations and the occurrence of occasional seizures. Behavioral analysis uncovered an increase in anxiety. Genome-wide sequencing of fluorescence activated cell sorted (FACS) PVIs revealed that Arx impinged on network excitability via genes primarily associated with synaptic and extracellular matrix pathways. Whole-cell recordings revealed prominent hypoexcitability of various intrinsic and synaptic properties. These results revealed important roles for postnatal Arx expression in PVIs in the control of neural circuits and that dysfunction in those roles alone can cause EOEE-like network abnormalities.

Graphical Abstract

Highlights

• •Arx CKO mice displayed abnormal theta rhythms, occasional seizures, and anxiety
• •Loss of Arx in PV interneurons broadly altered their gene expression profiles
• •The perineuronal net around PV interneurons was altered in Arx CKO mice
• •Loss of Arx in PV interneurons altered their intrinsic and synaptic properties

Behavioral Neuroscience ; Molecular Neuroscience; Cellular Neuroscience; Transcriptomics

Introduction

Mutations in the Aristaless Related homeobox transcription factor (ARX) cause early onset epileptic encephalopathies (EOEEs), a wide spectrum of disorders characterized by pharmacoresistant seizures, persistent electroencephalography (EEG) abnormalities, and cognitive deficits (Shoubridge et al., 2010). ARX has multiple roles in development, including cell cycle regulation in dorsal excitatory precursors and migration/fate specification of ventral interneuron precursors. Interestingly, ARX is downregulated in dorsal precursors upon exiting the cell cycle but remains expressed in migrating and mature interneurons. This complex pattern of expression is highly suggestive of distinct roles for ARX in the different stages of brain development.

Arx mouse models generated from conditional embryonic gene deletion or mutations have generally recapitulated human EOEEs and exhibited a complex pattern of spontaneous seizures, motor dysfunction, and learning impairments (Colombo et al., 2007; Dubos et al., 2018; Kitamura et al., 2002; Lee et al., 2017; Marsh et al., 2009; Simonet et al., 2015). Although the exact mechanisms by which pathogenic variation of Arx results in the EOEEs is currently unclear, accumulating evidence has converged toward the concept of interneuonopathy as a causative event (Kato and Dobyns, 2005). The underlying implication of this concept is that Arx-mediated loss of specific GABAergic interneurons leads to emergent hyperexcitable networks. However, it is currently unknown whether the phenotypes in ARX-linked EOEEs are exclusively due to ARX dysfunction in both interneuron and excitatory neuron progenitors or whether its persistent dysfunction in all or a subpopulation of mature interneurons plays a significant role in the evolution of seizures, EEG, and other phenotypes in EOEE in patients (Nordli, 2012). Therefore, investigations into the role of postnatal Arx expression are needed for a more comprehensive understanding of the underlying pathologic mechanisms of EOEEs and the contribution of sustained expression of this transcription factor (TF) to the evolution and/or manifestation of certain clinical phenotypes of EOEEs.

The loss of ARX function in postnatal interneurons could impair postnatal neural networks and consequently influence the developmental trajectories and the pleiotropic phenotypes of EOEEs. Indeed, three lines of evidence in the mouse literature suggest an important role for impaired postnatal Arx transcriptional activity in the pathogenic mechanisms of Arx-related disorders. First, mild and disparate loss of GABAergic neurons have been noted in various Arx knock-in mice carrying polyalanine insertion mutations (Beguin et al., 2013; Dubos et al., 2018; Kitamura et al., 2002, 2009; Lee et al., 2017; Price et al., 2009), leaving a significant number of interneurons with dysfunctional Arx signaling intact. Second, although Arx^−/Y; Dlx5/6CIG mice exhibit a more robust interneuron loss phenotype, the male mice suffer from early postnatal lethality, whereas the females survive with variable Arx expression (Marsh et al., 2009). Finally, seizures noted in early postnatal Arx^−/Y; Dlx5/6CIG mice evolved from no spasms to full-blown human-like infantile spasms in adult males (Marsh et al., 2009). Taken together, these observations suggest that postnatal Arx transcriptional activity in GABAergic neurons may be necessary for broad neural circuit functions in the adult brain and that dysfunction in this postnatal role could contribute to the manifestation and/or evolution of EEOE phenotypes.

Cortical GABAergic interneurons constitute just 20% of the neuronal population, but they influence virtually all stages of circuit maturation due in part to their extensive heterogeneity and extended maturation trajectory (Klausberger and Somogyi, 2008; Le Magueresse and Monyer, 2013). Despite their heterogeneity, recent studies have parsed out interneurons into three major classes (PV, SST, and 5HT₃R) and have demonstrated that each subclass, for the most part, adopts a different maturation trajectory by virtue of their birth date (Le Magueresse and Monyer, 2013). The PV subclass has a particularly long window of maturation (Doischer et al., 2008) and their extended developmental trajectory is thought to provide several windows of vulnerability during which aberrant signaling factors can contribute to neurodevelopmental disorders (Morishita et al., 2015). Indeed, the function of PVIs is well known to be altered in several neurodevelopmental disorders, and emerging studies have now demonstrated that certain disease phenotypes can be replicated by selective ablation of a handful of genetic risk factors in these interneurons during postnatal development (Hu et al., 2014). Although Arx is highly expressed in PVIs compared with most other interneuron subtypes during early postnatal development (Fishell and Kepecs, 2019), the functional significance for this high and sustained expression has not been addressed. Thus, an understanding of how ongoing Arx transcriptional activity affects the postnatal maturation trajectory of PVIs is an important objective for the elucidation of the pathologic mechanisms of EOEEs. To that end, we generated a CKO mouse model in which Arx was postnatally ablated in PVIs and determined whether that could alter neural network function. We found a significant increase in hippocampal theta rhythms that was accompanied by occasional spontaneous seizures and anxiety-related behaviors. The disrupted genetic and molecular pathways underpinning these clinical phenotypes were primarily associated with a diverse array of extracellular-matrix-related genes. Finally, we found that the cellular basis for the hyperexcitable hippocampal networks in CKO mice was a profound reduction in the intrinsic and synaptic excitability of PVIs. Our data provide evidence that postnatal Arx transcriptional activity in PVIs is an important regulator of neural circuits and that loss of this function contributes to the pathologic mechanisms of EOEEs. Therefore, restoration of postnatal Arx transcriptional activity in PVIs in the postnatal brain following onset of EOEEs may be an effective strategy toward circuit-based therapies.

Results

Conditional deletion of Arx in PVIs

To examine the contribution of postnatal Arx dysfunction in the pathogenic mechanisms of EOEEs, we generated mice lacking Arx only in PVIs by first crossing mice with Cre driven by the PV promoter (Hippenmeyer et al., 2005) to TdTomato (TdTom) (Madisen et al., 2010) reporter mice to generate Arx; PV^CreTdTom mice (hereafter termed control mice; Figure 1A). Mice with Cre recombinase and TdTom expression in PVIs (Arx; PV^CreTdTom mice) were then crossed with mice carrying Arx flanked LoxP sites (Arx^LoxP/LoxP) (Fulp et al., 2008; Marsh et al., 2009) alleles to generate Arx^Flox/y; PV^CreTdTom CKO mice (hereafter termed Arx CKO mice; Figure 1A). Cre-recombination and mouse genotypes were confirmed by RT-PCR analysis in P21 mice (Figure 1B). Because Arx is an X-linked gene, we restricted our studies to male mice in both genotypes. We noted PV^Cre-driven recombination as measured by the appearance of TdTom + neurons in both control and Arx CKO mice as early as postnatal day 15 (P15) with increasing TdTom florescence at 35 days (P35) (Data not shown) and near-complete loss of Arx in TdTom-containing cells at 8 weeks (Figure 1C). Cell-type specificity of PV^Cre mice has been previously described (Fuchs et al., 2007; Hippenmeyer et al., 2005), and our timeline for PV^Cre recombination is consistent with a previous report by Carlen and colleagues (Carlen et al., 2012). We found very little overlap between TdTom and Arx immuno-histochemical signals from 6- to 8-week-old mice, suggesting that Arx was strictly ablated in PVIs (Figure 1C). Immunohistochemical staining of ARX and PV labeling of the same cohorts of mice revealed co-labeling of ARX and PV in 82.7 ± 4.6% of cortical and hippocampal PVIs in control mice but only in 11.7 ± 5.5% of PVIs in Arx CKO mice (Figure 1D). All Arx CKO mice were viable, fertile, and displayed no obvious brain abnormalities (data not shown). As PV^Cre-driven recombination was not detectable before the second postnatal week, early developmental processes in our Arx CKO mice were not affected. Thus, our PV^cre and Arx^LoxP/LoxP recombination approach resulted in a suitable Arx CKO mouse model to investigate the role of postnatal Arx transcriptional activity in the maturing brain.

Figure 1.

Conditional ablation of Arx in PVIs

(A) Schematic drawing of the breeding strategy for generation of Arx CKO mice.

(B) PCR-based analysis of mice genotypes. Lane L (1Kb) represents the DNA ladder. Arx CKO mice were identified by the presence of the TdTom allele and reduced recombinant Arx^Flox and Cre bands (−) compared with controls (+). Control mice also expressed the TdTom allele.

(C) Representative images from control and Arx CKO mice showing virtually no overlapping ARX immunoreactivity (green) in PVIs or TdTom + cells (red).

(D) Quantitative histochemical analysis of Arx expression in PVIs given as overlapping percent, showing very little overlap in Arx CKO mice (∗∗∗p < 0.001; unpaired, 2-tailed Student's t test; control & Arx CKO: n = 3). The lower Q1 and upper Q3 quartiles of box plots (D) represent data outside the 5–95 percentile, and dots outside this range denote outliers. Scale bars: 25 μm.

Loss of Arx led to increased theta oscillations and presence of occasional seizures

As the major phenotype in Arx-linked EOEEs are seizures and an abnormal EEG, we first examined EEG oscillatory activity and noted that power spectra calculated from hippocampal and cortical electrodes either during night or during day recordings did not differ between control and Arx CKO mice (Figures 2A–2C). However, analysis of hippocampal theta rhythms in epochs of high-theta oscillatory activities in those recordings revealed changes in theta power spectra (Figure 2E). Specifically, Arx CKO mice showed an increase in normalized peak power spectra in the theta frequency bands (Figure 2F). We also examined the EEG recordings for the presence of seizures and observed spontaneous electrographic seizures in some Arx CKO mice (2/12) but not in control mice (0/8; Figure 2G). Typically, seizure onset was marked by fast and high-amplitude spikes (100μV/5–10Hz), lasting more than 60 s and followed by a prolonged post-ictal flattening period (Figure 2G). These results demonstrate that the loss of Arx in PVIs alone is sufficient for the induction of epilepsy-like network hyperexcitability.

Figure 2.

Loss of Arx in PVIs resulted in increased theta rhythms and occasional seizures

(A and C) Representative traces of night (A) and day (C) time hippocampal and cortical EEG recordings.

(B and D) Grand average of normalized hippocampal (Left plots) and cortical (Right plots) power spectra from night (B) and day time (D) EEG recordings showed no differences between genotypes (p > 0.05; n ≥ 8 for both control and Arx CKO.

(E) Representative traces of EEG recordings dominated by theta oscillations.

(F) Power spectra of high theta epochs revealed a significant increase in theta rhythms in Arx CKO mice (∗∗p < 0.01; Control, Arx CKO: n = 7, 8).

(G) Representative traces of two electrographic seizures noted in two different Arx CKO mice (2/12). Expanded view shows that a buildup of rhythmic spikes preceded seizure onset. Data analyzed by two-way ANOVA (C & D) and by the unpaired, 2-tailed Student's t test (F). All data are given as mean ± SD.

Increased anxiety-like behaviors in Arx CKO mice

Arx-linked disorders are characterized by significant impairments in cognitive and adaptive behaviors (Dubos et al., 2018; Jackson et al., 2017; Kitamura et al., 2009; Price et al., 2009; Simonet et al., 2015). Therefore, we investigated if motor coordination, spatial learning and memory, fear conditioning, novel object recognition, and anxiety were abnormal in the Arx CKO mice. Motor coordination and strength were tested using the rotarod and grip strength tests. Motor coordination and learning rate measured in the rotarod test were indistinguishable from control mice (Figures S1A and S1B). Forelimb and hindlimb grip strengths and pulled force were also similar in both genotypes (Figures S1C and S1D). Thus, neuromuscular strength and coordination in mice are independent of postnatal Arx transcriptional activity in PVIs.

Next, we measured hippocampus-dependent cognitive function using the forward and reverse spatial learning and memory paradigms of the Morris water maze (MWM) test (Simonet et al., 2015). Control and Arx CKO mice showed similar performance in the forward and reverse learning phases of the MWM (Figures S2A–S2E). This lack of difference in performance was also noted during the probe or memory test in the form of equal number of crosses for the previous location of the platform during both the recall and purging phases of the probe test (Figures S2F and S2G).

We further probed learning performance using the contextual fear conditioning paradigm (Phillips and LeDoux, 1992) and novel object recognition test (Ennaceur and Delacour, 1988). Control and Arx CKO mice showed very low but similar initial freezing ratio prior to the context or foot shock (Figure S2H). Total freezing ratio measured 24 h after the context was also similar in both genotypes (Figure S2I). Similarly, there was no difference in performance between the Arx CKO and control mice in the novel object recognition test (Figures S2J and S2K).

As a final measure of adaptive behaviors, we evaluated anxiety using the open field (OF) test (Simonet et al., 2015). Control and Arx CKO mice covered equal amount of distance in the OF arena (Figures 3A and 3B), but the Arx CKO mice spent significantly less time in the center (Figures 3A and 3C). This reduction in time spent in the center of the arena was noted only for locomotor activity during the second of three 5-min bins, which resulted in a significant reduction in the acclimatization ratio (Figure 3C). This enhancement in anxiety-like behavior was further indicated by a significant increase in fecal counts during the open field test (Figure 3D). To further test anxiety, we performed the elevated plus-maze (EPM) test (Dawson and Tricklebank, 1995) and noted that control and Arx CKO mice made similar number of entries into the open and closed arms of the EPM (Figures 3E and 3F). However, the Arx CKO mice spent significantly less time in the open arm of the test apparatus (Figures 3E and 3G). Taken together, these results suggest that loss of postnatal Arx in PVIs contributes to the expression of anxiety-like behaviors but not cognitive nor motor coordination in Arx-linked disorders.

Figure 3.

Arx CKO mice displayed anxiety-like behaviors

(A) Schematic representation of the zones and a mouse movement in the open field arena. The middle square was taken as the center or anxiety zone.

(B) Total distance traveled was similar in control and Arx CKO mice.

(C) Three five-minute bins, showing reduced time in the center of the arena by Arx CKO mice (∗p < 0.05). The inset plot of reduced acclimatization ratio measured as the ratio of center time in the second bin over the first bin (∗p < 0.05; Control, Arx CKO: n = 23, 19).

(D) Reduced fecal boli number in Arx CKO mice (∗p < 0.05, Control & Arx CKO: n = 13).

(E) Schematic representation of the EPM used for anxiety test and example traces of a mouse movement in each arm.

(F and G)(F) Arx CKO mice made equal amount of entries in the EPM open and close arms but spent less time in the open arms (G, ∗p < 0.05, control, Arx CKO: n = 23, 19). The lower Q1 and upper Q3 quartiles of box plots (C & F) represent data outside the 5–95 percentile, and dots outside this range denote outliers. Data analyzed by Mann Whitney U test or two-way ANOVA followed by Bonferroni post hoc tests (B, C, & F) and by the unpaired, 2-tailed Student's t test (D & G). All data are given as mean ± SEM.

Postnatal loss of Arx in PVIs altered their transcriptomic profiles

Loss of function of Arx during early embryonic development disrupts genes expression of molecular and cellular processes such as cell migration, axonal guidance, neurogenesis, and regulation of transcription (Fulp et al., 2008). Given the developmental tilt of these pathways, we reasoned that the disrupted pathways underpinning the network and behavioral abnormalities in adult Arx CKO mice might be different than the previously described developmentally expressed genes. Therefore, we performed genome-wide expression profiling of FACS-sorted PVIs (PND 35–40). RNA sequencing (RNA-Seq) from cortical and hippocampal TdTom + PVIs from five control and four Arx CKO mice resulted in an extensive library of paired reads, which were then mapped to 25K + genes and normalized in DEseq2 (Love et al., 2014). We first assessed the quality of the sequencing data using hierarchical clustering and principal component analysis (PCA) analyses and found a modest separation between control and Arx CKO samples (Figures 4A and S3A), reflective of their distinct transcriptomic profiles. Using a modified Bland-Altman (MA) plot, we noted that ablation of Arx in PVIs changed the expression of many genes, but only 144 were significantly dysregulated (Figure 4B). Among the differentially expressed genes (DEGs), 28 were upregulated and 116 genes were downregulated (Figure 4B). The epilepsy-related genes LAMB1, COL18A1, Nid1, and DOCK6 (Pisciotta et al., 2018; Radmanesh et al., 2013; Suzuki et al., 2002; Vasudevan et al., 2010) were among the downregulated DEGs (See Table S1 for the full list of DEGs). As hypothesized, we found virtually no overlap between our DEGs and those from embryonic Arx deletion (Fulp et al., 2008), proving that the postnatal Arx transcriptional regulatory program is distinct from the embryonic brain. To validate the RNA-Seq data, we performed quantitative real-time PCR (qRT-PCR) analysis on Arx and 4 DEGs (Gria3, Olmf1, Cck, and Nid1) using different biological replicates from the RNAseq analysis. Arx was chosen for validation, as it was expected to be reduced in expression in spite of not being a DEG, an observation that is likely due to its low abundance (mean normalized read count = 188) or differential expression of its intronic and coding sequences as previously shown (Fulp et al., 2008). Nonetheless, we observed a statistically significant increase in Gria3 and Cck and a decrease in Arx as well as Nid1 (Figure S3B). Although not statistically significant, Olfm1 mRNA expression also increased more than 2-fold in Arx CKO tissues (Figure S3B).

Figure 4.

Loss of Arx in PVIs altered their transcriptomic profiles

(A) PCA plot showing separation between control and Arx CKO biological replicates.

(B) MA plot of DEGs and Venn diagram with the number of up- and downregulated DEGs. Red dots indicate up- and downregulated genes, whereas the black dots indicate non-significant genes based on cut-off criteria (|log2FC| ≥ 1, FDR < 0.05).

(C and D) Functional networks formed by (C) upregulated and downregulated (D) DEGs. Black nodes represent the queried DEGs, and the gray nodes are other genes related to the DEGs.

(E) Percent weight table for each of the functional association networks made by the up- and downregulated DEGs.

(F) Top 3 significantly enriched Gene ontology [−log₁₀ (FDR)] terms associated with functional networks made by DEGs.

(G and H) Open chromatin regions with Arx-binding sites were more likely to be found in up- (G) and downregulated (H) DEGs than randomly generated sites (∗∗∗∗p < 0.0001).

(I and J) Upregulated (I) and downregulated (J) DEGs binding motifs were mostly found in distal regions relative to the TSSs. Values on bar graphs represent number of genes associated with peaks in each genomic bin. Transcriptomic data were from 5 control and 4 Arx CKO mice. Data analyzed by Benjamini-Hochberg FDR method (B & F) and Chi-square test (G & H).

Given the multiplicity of disrupted genes, we choose a biological pathway-based analytical approach to identify the most important molecular and cellular processes underlying the network and behavioral abnormalities in Arx CKO mice. To that end, we applied gene ontology (GO) analysis using the Database for Annotation, Visualization and Integrated Discovery (DAVID) (Huang da et al., 2009). Separate GO analysis for the upregulated and downregulated DEGs was performed for the enrichment of cellular component. Notably, synapse, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) glutamate receptor complex, and terminal bouton were the top three significant cellular component GO terms associated with the upregulated DEGs (Figure S3C with full table pathways). By contrast, proteinaceous extracellular matrix, extracellular matrix, and collagen fibril organization were the three most significant cellular component GO terms associated with the downregulated genes (Figure S3D). Hence, the prominent biological pathways underpinning network and behavioral abnormalities in Arx CKO mice are distinct from those dysregulated following embryonic loss of Arx.

Functional gene networks were constructed to identify dysregulated pathways based on co-expression, co-localization, physical interaction, predicted interaction, shared protein domain, and other functional relations using the GeneMANIA cytoscape app (Montojo et al., 2014). In the generated networks, genes/proteins are depicted as dark (Query genes) and gray (Predicted genes) circles and the detected functional relationships between genes/proteins as connecting lines or edges (Figure 4C). Eight genes (2 unregulated, 6 downregulated) among the DEGs were omitted from the constructed networks because they were not recognized by the GeneMANIA-Cytoscape app. Among the rest, the upregulated genes were connected primarily via co-expression networks (69.68%) (Figure 4E) and to a lesser extent via co-localization (18.73%), interaction with predicted genes/proteins (2.28%), shared protein domains (0.09%), and other functional relations (9.21%) (Figure 4E). The downregulated DEGs were also primarily co-expressed together, yielding a co-expression network weight of 76.47% (Figure 4E). The remaining downregulated DEGs were connected via co-localization (9.65%), interaction with predicted genes/proteins (6.86%), physical interactions (0.94%), shared protein domains (0.24%), and other functional relations (5.83%) (Figure 4E). GO analysis of the generated networks that were clustered around synaptic and ECM processes are noted in DAVID (Figures 4F and Tables S2 and S3). These results indicate that postnatal Arx expression in PVIs regulates a network of genes widely connected by co-expression and that the upregulated and downregulated genes within its regulatory controls differ in functional network weight, physical interactions, and function. More importantly, the level of significance for the top three GO categories in the network analysis was more profound for the downregulated DEGs compared with the upregulated genes, highlighting the importance of dysregulation of ECM-related gene networks in the production of network and behavioral abnormalities in Arx CKO mice.

Gene expression changes identified in our RNAseq could be a consequence, rather than a cause, of the noted network and behavioral abnormalities. To determine whether the expression changes were directly due to loss of Arx, we utilized in silico analysis. This approach was chosen, as the low Arx transcript levels limit direct assessment of binding through cell-specific chromatin immunoprecipitation. Using previously published open chromatin sites in PVIs derived by assay of transposase-accessible chromatin (ATAC), we performed transcription-factor-binding motif (TFBM) enrichment analysis in the 100 kb upstream and downstream of the transcription start site (TSS) of DEGs (Mo et al., 2015). ATAC sequencing is a validated method for identifying the open chromatin signature (Buenrostro et al., 2013). Using Cistrome in combination with the UCSC Table Browser (Karolchik et al., 2004; Liu et al., 2011), we identified a total of 23 of 28 (82.1%) upregulated DEGs (Figure S3E) and 95 of 116 (81.9%) downregulated DEGs (Figure S3F) with at least one overlapping PV-ATAC site and Arx-binding motif. A list of DEGs with at least one overlapping PV-ATAC site and Arx-binding motif is given in Table S4 (FDR <0.05). The synapse-associated genes Gria3 and Sv2b were among the upregulated DEGs with overlapping PV-ATAC sites and Arx-binding motifs (Table S4). The downregulated DEGs with overlapping PV-ATAC sites and Arx-binding motifs included the extracellular matrix gene Col18a1 and Nid1 (Table S4). To determine if Arx-binding motifs were more likely to be present in PV-ATAC sites near DEGs, we first compared the probability of finding an Arx-binding motif in PV-ATAC sites in such DEG with randomly generated genomic regions matched for length and percent GC content in RSAT (Nguyen et al., 2018). Our results showed the presence of Arx-binding sites in PV-ATAC sites were more likely in the upregulated DEGs than in randomly generated PV-ATAC sites (Figure 4G). Arx-binding motifs were also more likely to be found in PV-ATAC sites from downregulated DEGs than in randomly generated matched sites (Figure 4H). We also measured the region-to-gene associations of overlapped PV-ATAC sites and Arx-binding motifs using the Genomics Region of Enrichment Analysis Tool (GREAT) (McLean et al., 2010) and noted that less than 10% of such association for upregulated (Figure 4I) and downregulated (Figure 4J) DEGs occurred within 5 Kb of the annotated TSSs. By contrast, there was a clear tendency for this PV-ATAC sites/Arx motifs association for upregulated and downregulated DEGs to be confined to regions distant from the annotated TSSs (Figures 4I and 4J). These results indicate that Arx-binding motifs in the postnatal brain are enriched at putative enhancers of both up- and downregulated DEGs. Thus, Arx regulatory roles in postnatal gene regulation might be as a secondary modulator of gene expression through binding at distal enhancers instead of a primary and necessary driver of transcription mediated through binding at proximal enhancers (Lenhard et al., 2012).

Arx ablation disrupted the perineuronal nets around PVIs

ECM-related genes are important in the development and maturation of neural circuits (Stranahan et al., 2013). Therefore, we asked whether the profound downregulation of ECM-regulated genes in PVIs resulted in structural alterations. We first assessed the density of PVIs using conventional immunolabeling techniques. We found that the density of PVIs was not altered in Arx CKO mice (Figures 5A and 5B). To rule out non-cell autonomous effects on the other GABAergic interneurons, we measured the density of hippocampal SST, CB, and CR GABAergic interneuron subtypes and found no differences (Figure 5B). To further assess structural alterations, we investigated whether the loss of Arx in PVIs led to molecular changes in the perineuronal nets (PNNs), an ECM specialization that consists of proteoglycans such as aggrecan (ACAN) and generally wraps PVIs (Giamanco and Matthews, 2012). Using immunolabeling techniques, we observed a marked reduction in the intensity of the aggregating proteoglycans marker ACAN (Figures 5C and 5E) and a substantial decrease in the number of the glycosylated ACAN marker Cat-315 (Figures 5D and 5F) throughout the hippocampus of Arx CKO mice. As expected, co-labeling of TdTom and Cat-315 showed that PVIs are preferentially wrapped by the PNNs in control mice, but this structural relationship appears to be disrupted in Arx CKO mice as evidenced by the profound reduction in the number of TdTom + cells surrounded by Cat-315 (Figures 5D and 5G).These results further confirm the RNA-Seq GO analysis and establish an important role for postnatal Arx signaling in the structural maintenance of PVIs in neural circuits.

Figure 5.

Perineuronal nets are disrupted around PVIs in Arx CKO mice without altering the density of GABAergic interneurons

(A) Representative images of PV immunolabeling in the hippocampus of control (Top) and Arx CKO (Bottom) mice.

(B) Density of PV, CB, CR, and SST GABAergic interneurons per mm² in the CA1 hippocampus was comparable in both genotypes (p > 0.05; two-way ANOVA; control, Arx CKO: N = 3).

(C and D) Representative images of aggrecan (C, ACAN) and Cat-315/TdTomato co-labeling in the hippocampus of control (Left) and Arx CKO mice (Right) (D).

(E–G)ACAN intensity was reduced in Arx CKO mice (E, ∗p < 0.05; unpaired 2-tailed Student's t test; control = 5 and Arx CKO = 7). Quantitative analysis of Cat-315 + neurons per mm² revealed a significant reduction in the density (F, ∗∗p < 0.01; inpaired 2-tailed Student's t test; control = 6 and Arx CKO = 4) and in the number of neurons with overlapping expression of this marker with TdTom (G, ∗∗∗∗p < 0.0001; unpaired, 2-tailed Student's t test; Control = 6 and Arx CKO = 4). Data are given as mean ± SEM (B). Scale bars: (A) 25 μm, (C) 50 μm, (D) 100 μm.

Reduced membrane excitability of PVIs in Arx CKO mice

The PNN has been shown to be a potent regulator of neuronal excitability (Balmer, 2016; Dityatev et al., 2007; Favuzzi et al., 2017; Lensjo et al., 2017), raising the possibility that the loss of PNN around PVIs from Arx CKO mice may have altered the intrinsic excitability of these interneurons. To test this possibility, we recorded intrinsic membrane properties of PVIs from the first Cornu Ammonis (CA1) region of the hippocampus in whole-cell current clamp mode. Passive intrinsic membrane properties were not significantly affected by the loss of Arx (see Table S5 for values of resting membrane potential, capacitance, input resistance, and membrane time constant). Similarly, action potential properties measured from the initial spike evoked by the rheobase current (Figure 6A) were indistinguishable between genotypes with regard to amplitude, duration at half amplitude, and fast afterhyperpolarization potential (AHP) (Table S5). However, the rheobase current amplitude required to evoke this initial action potential (AP) was significantly higher in PVIs from Arx CKO mice (Figure 6B). In addition, the membrane threshold for the initial AP firing was more depolarized in these PVIs (Figure 6C). These hypo-excitability measures were associated with much slower onset latencies for the initial APs (Figure 6D). Trains of APs triggered by 500 ms depolarizing current steps in both Arx CKO and control mice were relatively brief in duration and occurred at high frequency with very low adaptation ratio (Figure 6E), suggesting that PVIs that underwent Cre-recombination in both genotypes still exhibited major physiological hallmarks of wild-type fast-spiking PVIs (Kuhlman and Huang, 2008; Rudy and McBain, 2001). However, the number of APs generated by a given 500ms current step was significantly reduced in PVIs from Arx CKO mice (Figure 6F), indicating that ablation of Arx in PVIs alters their steady state firing properties. These results suggest that postnatal Arx transcriptional activity is an important regulator of intrinsic membrane properties of PVIs.

Figure 6.

Intrinsic membrane properties of PVIs are altered in Arx CKO mice

(A) Representative rheobase spikes recorded on PVIs.

(B–D) PVIs in Arx CKO mice were hypoexcitable based on higher rheobase current (B, ∗∗∗p < 0.001; control & Arx CKO: n = 20/12, cells/mice), more depolarized AP threshold (C, ∗∗p < 0.01; control & Arx CKO: n = 20/12, cells/mice), and delayed latency to rheobase spikes (D, ∗∗p < 0.01; control: n = 23/12, Arx CKO: n = 20/12, cells/mice).

(E) Representative spikes at maximal depolarizing current injection.

(F) PVI steady state spike frequency was reduced in Arx CKO mice (p < 0.0001; control: n = 16/10, Arx CKO: n = 20/10, cells/mice). ∗Symbols in F denote post hoc test significance levels: ∗p < 0.05, ∗∗p < 0.01. Data are given as mean ± SEM and analyzed by the unpaired, 2-tailed Student's t test (B & C), Mann Whitney U test (D), or two-way ANOVA followed by Bonferroni post hoc tests (F).

Dendritic complexity of PVIs is intact in Arx CKO mice

The hypo-excitability measures in PVIs from Arx CKO mice could depend on the types and density of voltage-gated ion channels over the neuronal membrane (Migliore and Shepherd, 2005) or altered complexity of their dendrites (van der Velden et al., 2012). Given that our RNA-Seq data revealed a lack of prominent dysregulation of voltage-gated ion channels, we measured dendritic complexity using Sholl analysis (Sholl, 1953). Quantitative analysis of biotin-labeled PVIs in the hippocampal CA1 region did not reveal any significant differences in dendritic complexity as measured by number of intersections, branch number, and maximum branch length in Sholl concentric circles near or distant from the soma (Figures S4A–S4D). Thus, these results suggest that dendritic morphology did not substantially contribute to PVI or overall CA1 network excitability.

Reduced spontaneous synaptic excitability in PVIs

Brain oscillations are thought to emerge through an interactive interplay between intrinsic membrane and synaptic properties (Benayoun et al., 2010). Thus, we reasoned that the hypo-excitability of PVIs in Arx CKO mice could be accompanied by alterations in synaptic excitability. To measure synaptic excitability, we simultaneously recorded spontaneous excitatory postsynaptic currents (sEPSCs) and inhibitory postsynaptic currents (sIPSCs) on PVIs at a holding potential of −40mV in aCSF free of glutamate and GABA receptor blockers (Zhou et al., 2009). The inward sEPSC onto PVIs appeared as rapidly rising and slowly decaying currents, whereas the rise and decay kinetics of the outward IPSCs currents were slower as previously described (Figure 7A) (Wuarin and Dudek, 1993). There was a reduction in the mean frequency of spontaneous EPSCs onto Arx CKO PVIs (Figure 7B) as well as a rightward shift of the cumulative probability curve for inter-event intervals (Figure 7B). Although the mean sEPSC amplitude was similar in both genotypes, there was a shift toward larger sEPSC events in their cumulative distribution in Arx CKO mice (Figure 7C).

Figure 7.

Impaired spontaneous synaptic transmission

(A) Example traces of sEPSCs and sIPSC recorded on PVIs in the CA1 hippocampus of control and Arx CKO mice.

(B) Reduced sEPSC frequency onto PVIs in Arx CKO mice (Cumulative and box plots: ∗∗∗∗p < 0.0001, ∗p < 0.05; control: n = 16/10, Arx CKO: n = 20/12, cells/mice).

(C) Cumulative amplitude shifted toward larger events without affecting the mean (cumulative and box plots: ∗∗∗∗p < 0.0001, p > 0.05; control: n = 16/10, Arx CKO: n = 20/12, cells/mice).

(D) The cumulative frequency of sIPSCs shifted toward less events without any effects on their mean frequency (cumulative and box plots: ∗∗∗∗p < 0.0001, p > 0.05; control: n = 17/11, Arx CKO: n = 21/12, cells/mice).

(E) Cumulative distribution and mean of sIPSC amplitude are normal in Arx CKO mice (cumulative and box plots: p > 0.05 for both graphs; control: n = 14/10, Arx CKO: n = 19/12, cells/mice).

(F) Ratio of excitatory to inhibitory amplitudes was not altered (p > 0.05; Control: n = 17/10, Arx CKO: n = 18/12, cells/mice).

(G) A shift toward less excitatory events (mEPSCs) was still evident in presence of TTX with no effects on the mean frequency (cumulative and box plots: ∗∗∗∗p < 0.0001, p > 0.05; control: n = 20/12, Arx CKO: n = 17/12, cells/mice).

(H) Cumulative distribution of mEPSC amplitudes shifted toward smaller events as reflected in the mean amplitude (cumulative and box plots: ∗∗∗∗p < 0.0001, ∗p < 0.05; Control & Arx CKO: n = 17/12, cells/mice).

(I) Biphasic cumulative distribution mIPSC frequency with no effects on the mean (cumulative and box plots: ∗∗∗∗p < 0.0001, p > 0.05; Control: n = 15/10, Arx CKO: n = 11/10, cells/mice).

(J) Cumulative distribution of mIPSC amplitudes shifted toward larger events with no effects on the mean amplitude (cumulative and box plots: ∗∗∗∗p < 0.0001, p > 0.05; control: n = 19/12, Arx CKO: n = 17/12, cells/mice). Data are given as mean ± SEM and analyzed by the K-S (cumulative plots) or the unpaired, 2-tailed Student's t (Box plots) tests.

Analysis of sIPSCs also revealed a significant rightward shift in the cumulative probability of inhibitory inter-event intervals, reflecting a reduction in the frequency of sIPSCs (Figure 7D). However, the mean frequency of these sIPSC did not differ between genotypes (Figure 7D). Analysis of sIPSCs also showed no significant differences in mean events amplitude nor in their cumulative distribution in Arx CKO and control mice (Figure 7E). As expected from the lack of changes in EPSC and IPSC amplitudes, their mean amplitude ratio was not altered in Arx CKO mice (Figure 7F). These results support a role for transcriptional activity of postnatal Arx in the regulation of some spontaneous excitatory and inhibitory synaptic properties in PVIs.

Alterations in miniature synaptic transmission onto PVIs

To measure the contribution of network activity to the changes in spontaneous synaptic transmission onto PVIs in the CA1 hippocampal region, we recorded miniature EPSCs (mEPSCs) and IPSCs (mIPSCs) in the presence of 0.5μM tetrodotoxin (TTX). In contrast to sEPSC, quantitative analysis of mEPSCs showed no significant differences in mean frequency of excitatory events between Arx CKO and control mice (Figure 7G). However, similar to the sEPSCs, the cumulative probability curve of mEPSCs inter-event intervals shifted toward less frequent events in Arx CKO mice compared with controls (Figure 7G). In contrast to sEPSCs, the mean amplitude of mEPSCs recorded on PVIs from Arx CKO mice was significantly reduced compared with controls, and this observation was consistent with a leftward shift toward smaller events in their cumulative distribution (Figure 7H).

The mean frequency of the mIPSCs in Arx CKO mice was not statistically different from controls (Figure 7I). Analysis of the cumulative distribution of mIPSCs inter-event intervals showed a biphasic curve, with a cluster of events shifted leftward toward more frequent events and another shifted rightward toward less frequent events (Figure 7I). The mean amplitude of mIPSC events revealed that Arx CKO mice were indistinguishable from control mice, but the cumulative distribution of those events shifted toward larger mIPSCs (Figure 7J). These results suggest that loss of Arx in PVIs was primarily associated with changes to excitatory synaptic properties, including a reduction in EPSC mean frequency as well as significant changes in the cumulative distribution of sEPSC/sIPSC IEI and amplitude. Many of these changes were either blunted or reversed by the addition of TTX, suggesting Arx-mediated alterations in synaptic drives onto PVIs might be influenced by intrinsic properties of presynaptic cells and/or network excitability.

Loss of Arx reduced evoked excitatory drive onto PVIs

The quantitative analysis of mEPSCs and mIPSCs on PVIs are mostly consistent with a reduction in synaptic excitability. To further strengthen this conclusion, we measured the input-output (I-O) relationship of whole-cell evoked EPSCs (eEPSCs) and IPSCs (eIPSCs) on PVIs in response to Shaffer collateral (SC) stimulation (Figure 8A). We found that the I-O relationship of the eEPSC amplitude was reduced in Arx CKO mice (Figure 8B). This reduction in synaptic transmission was further manifested in a reduction in charge transfer and decay kinetics (Figures 8C and 8D). By contrast, eIPSC I/O relationship for amplitude, charge transfer, and decay kinetics revealed no differences between genotypes (Figures 8E–8G). However, the ratio of eEPSC/eIPSC amplitudes onto PVIs in Arx CKO mice was reduced (Figure 8H), indicating a shift toward less excitation.

Figure 8.

PVIs in Arx CKO mice are synaptically hypoexcitable

(A) Representative traces of EPSC and IPSC evoked by SC stimulation in the CA1 hippocampal region.

(B–D) I/O curves for eEPSC amplitude (B), area under the curve (C), and decay kinetics (D) revealed a profound reduction in excitatory drives onto PVIs (p < 0.0001; control: n = 19 cells/10 mice, Arx CKO: n = 18 cells/12 mice).

(E–G) Evoked inhibitory drive is normal as shown in I/O curves for eIPSC amplitude (E), area under the curve (F), and decay kinetics (G) (p < 0.0001; control: n = 24/12, Arx CKO: n = 21/12, cells/mice).

(H) The ratio of excitatory to inhibitory of the I/O curves shifted toward less excitation in Arx CKO mice (p < 0.0001; control: n = 11/8, Arx CKO: n = 10/8 cells/mice).

(I and J) The paired-pulse ratio for both eEPSC (I) and eIPSC (J) was normal in Arx CKO mice (p > 0.05; control: n = 17/12, Arx CKO: n = 16/10, cells/mice for both EPSC and IPSC PPR). Representative paired-pulse traces above graphs were recorded at interval of 50 ms. Data are given as mean ± SEM and analyzed by two-way ANOVA followed by Bonferroni post hoc tests. ∗Symbols denote post hoc test significance levels: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

The reduction in synaptic excitability of CA1 PVIs following the postnatal loss of Arx could arise from alterations in the pre-synaptic machinery. Therefore, we recorded whole-cell patch paired pulse responses (PPR) in PVIs at SC-CA1 synapses (Figures 8I and 8J). We found that the ratio of eEPSC PPR recorded on Arx null PVIs was unchanged relative to controls (Figure 8I), suggesting that Arx-mediated impairments in evoked excitatory transmission were independent of presynaptic defects. The eIPSC PPR in Arx CKO was indistinguishable from control mice (Figure 8J), a consistent observation with the lack of change in inhibitory synaptic drive onto in PVIs. Taken together, these results indicate that network hyperexcitability in Arx CKO mice was driven primarily by a reduction in SC drive onto PVIs that was independent of the presynaptic release machinery.

Postsynaptic glutamate receptors hypofunction contributes to reduced synaptic excitability

Given that PPR of eEPSCs and dendritic complexity were unaltered in PVIs from Arx CKO mice, we reasoned that the reduction in synaptic excitability could arise from alterations in the function of postsynaptic glutamate receptors (GluRs). We measured mixed AMPA and N-Methyl-D-aspartic acid or N-Methyl-D-aspartate (NMDA) eEPSCs in response to SC stimulation and characterized the AMPA and NMDA components of those EPSCs with respect to their conductance, rectification patterns, and functional ratio. Examination of the I-V relationship for both AMPA- and NMDA-receptor-mediated currents at different holding potentials revealed considerable inward rectification in both genotypes, but this rectification behavior appears to be more moderate in Arx CKO mice (Figures 9A–9C). Accordingly, the rectification index (RI), defined as the ratio of current amplitude at +50 and −50 mV, for both AMPA and NMDA receptor currents was significantly reduced in PVIs from Arx CKO mice (Figures 9D and 9E). This suggests that excitatory synapses on PVIs were more likely to be activated via Ca²⁺ impermeable GluRs. A caveat, however, is that the recorded reversal potentials of AMPA and NMDA receptor currents were different from their expected 0 mV value (Suyama et al., 2017). Although several factors could account for this discrepancy, poor space clamp error may have largely contributed to this error in voltage. Because this error could potentially affect the accuracy of the RI of AMPA and NMDA receptors, we limited our analysis to cells that did not substantially differ from their reversal potentials to ensure that space clamp errors were evenly distributed across genotypes. Of note, the RI values were consistently lower in Arx CKO PVIs compared with controls irrespective of differences in reversal potentials, confirming the accuracy of our RI data.

Figure 9.

Impaired functional properties of AMPA and NMDA receptors in Arx CKO mice

(A) Representative traces of mixed AMPA- and NMDA-receptor-mediated currents on PVIs evoked by SC stimulation in the CA1 hippocampal region.

(B–E) (B and C) I–V curves of AMPA- (B) and NMDA-receptor (C)-mediated conductance normalized to highest respective current. (C and D) Reduced AMPA (D; ∗∗p < 0.01; control: n = 12/8, Arx CKO: n = 10/8, cells/mice) and NMDA (E; ∗∗p < 0.01; ∗∗p < 0.01; control: n = 12/8, Arx CKO: n = 10/8, cells/mice) receptor RI onto PVIs.

(F) Representative traces of AMPA- and NMDA-receptor-mediated EPSCs recorded at −50 mV and +50 mV, respectively.

(G) Reduced ratio of AMPAR/NMDAR currents onto PVIs in Arx CKO mice (∗∗p < 0.01; ∗∗p < 0.01; control: n = 12/8, Arx CKO: n = 10/8, cells/mice). Data are given as mean ± SEM and analyzed by the unpaired, 2-tailed Student's t test.

To further elucidate the synaptic mechanisms of reduced excitatory drive to PVIs, we measured the ratio of AMPA/NMDA receptors. As expected, based on the RI measurements, the AMPAR/NMDA receptor current ratio was significantly reduced in PVIs from Arx CKO mice compared with controls (Figures 9F and 9G), further supporting the postsynaptic basis of eEPSC impairments noted in PVIs following the loss of Arx. Taken together, these results suggest that the synaptic basis of network and behavioral abnormalities in Arx CKO mice can be attributed primarily to a reduction in excitatory synaptic drive onto PVIs as a result of dysregulation of postsynaptic GluRs.

Discussion

Loss of function of Arx during embryonic development leads to defects in migration of interneurons, epilepsy, and cognitive deficits. ARX remains expressed in mature interneurons, with an unknown function and relationship to disease progression. We addressed this important question by ablating postnatal Arx expression exclusively in the PVIs and noted an increase in EEG theta oscillations, occasional spontaneous seizures, and behavioral abnormalities. RNAseq analysis suggests that these network effects were likely due to dysregulation of coherent set of genes, with those related to the ECM as the central hub of dysregulated pathways underlying the network abnormalities. The cellular and synaptic mechanisms mediating these network effects appear to involve a profound reduction in excitability of intrinsic and synaptic properties onto PVIs. Therefore, our results place Arx in PVIs at nexus of a broad set of genes essential for sculpting as well as maintaining mature neural networks and dysfunction in those roles contributes to the pathologic mechanisms of Arx-linked EEOEs.

Seizures and other network abnormalities are important clinical manifestations of EEOEs (Nordli, 2012; Shoubridge et al., 2010). We found a selected increase in theta oscillations and rare electrographic seizures following the postnatal loss of Arx in PVIs. An alteration in theta oscillations was expected, given previous reports of tight coupling of PVI excitability with both theta and gamma oscillations (Amilhon et al., 2015; Carlen et al., 2012; Gonzalez-Burgos and Lewis, 2008). Our findings were unexpected, as they are in the opposite direction of previous studies that demonstrated reduced theta rhythms following optogenetic silencing of activity of PVIs or hypoexcitability induced by selective reduction of NMDAR function in these cells (Amilhon et al., 2015; Carlen et al., 2012; Korotkova et al., 2010). This discrepancy could be due to a variety of experimental and biological factors. First, the level of PVI dysfunction due to Arx dysregulation and optogenetic stimulation is very different resulting in different effects on theta oscillations. Second, the subtype of PVIs altered could lead to activation of distinct oscillatory generators (Colgin, 2013) shifting theta rhythms distinctly. Third, the lack of effect on gamma and enhanced theta oscillations may represent functional features of hippocampal circuits disposed to spontaneous, albeit sporadic, seizures in Arx CKO mice as a result of selective disruption of an important postnatal transcriptional regulator of neural circuits. Future studies are needed to determine whether the regulation of gamma rhythms and the direction of theta modulation in hypoexcitable PVIs are contingent on whether signaling dysfunctions mediating hypoexcitability of these cells lead to neural circuits disposed to spontaneous seizures.

The presence of occasional spontaneous seizures (2/12) compared with the regularity of seizures in the Arx^−/Y; Dlx5/6CIG mouse line (100%) where deletion occurs embryonically and includes all interneurons (Marsh et al., 2009) could be due to the following: (1) our 3-day EEG recordings being an insufficient time to capture infrequent seizures; (2) variable Cre recombination in the Arx CKO mice could have influenced seizure threshold and, thus, reduced the frequency of spontaneous seizures; and (3) complete penetrance of reduced threshold for spontaneous seizures may require postnatal Arx dysfunction in all interneurons or in combination with different interneuron subtypes. Interestingly, occasional rather than regular seizures were also noted in another CKO mouse model in which Scn1a (NaV1.1) was ablated in PVIs (Ogiwara et al., 2013), suggesting that the threshold for the development of more regular seizures may require both pre- and postnatal effects on PVIs. Thus, it will be important to determine how dysfunction in postnatal Arx signaling in other interneurons, either alone or in concert, contribute to spontaneous seizures in future studies. Nonetheless, our results confirm the essential role of PVIs in network oscillations (Amilhon et al., 2015) and show that postnatal dysfunction in Arx transcriptional activity in PVIs alone can contribute to the pathophysiology of EEOEs.

Diminished cognitive and adaptive behaviors are core features of the Intellectual disability described in EEOEs (Verma et al., 2019). To that end, we assessed the performance of Arx CKO and control mice in behaviors within the cognitive, motor, and anxiety domains. Interestingly, we only found a behavioral phenotype within the anxiety domain in the Arx CKO mice. More specifically, the mutant mice exhibited an increase in anxiety-related behaviors, an observation that is well correlated with the enhanced theta oscillations in these mice, given the direct relationship between the magnitude of these rhythms and anxiety-provoking environments (Gordon et al., 2005). Although the expression of anxiety-like behaviors was consistent across the OF and EPM tests used in this study, this phenotype was apparent in the time spent in the open arm of the EPM but not in the number of entries into the open and closed arms, suggesting that Arx-dependent activity in PVIs mediates the expression of anxiety-like behaviors as opposed to a general effect on activity levels. Indeed, the anxiety phenotype in the OF was noted without any discernible differences in activity levels of control and Arx CKO mice. Although control and Arx CKO mice display similar levels of spontaneous activity in the OF, our quantification of spontaneous activity levels are lower than others have reported (Bolivar et al., 2000; Bothe et al., 2005; Miller et al., 2010; Moy et al., 2007; Sik et al., 2003). Similarly, both genotypes displayed low exploration activity in the NOR test (Bolivar et al., 2000; Bothe et al., 2005; Miller et al., 2010; Moy et al., 2007; Sik et al., 2003). Although we cannot fully explain these differences, environmental factors such as noise, temperature changes, and housing are known to impact levels of exploration and spontaneous activity in mice (Van Meer and Raber, 2005). Alternatively, the difference in these behaviors may be explained by the 129P2/OlaHsd x 129S6/SvEvTac x C57BL/6J hybrid background of our mice. Genetic variability across inbred strains exists in most behavioral tests, including those for cognitive and emotional disorders (Van Meer and Raber, 2005). In particular, the 129S1 strain, an important contributor to the hybrid background strain in this study, has consistently been shown to display low levels of spontaneous and exploratory activity in OF and NOR tests, respectively (Bolivar et al., 2000; Bothe et al., 2005; Miller et al., 2010; Moy et al., 2007; Sik et al., 2003).

Previous reports have stated that anxiety or emotional stress can impair spatial learning and memory (Goodman and McIntyre, 2017; Packard and Wingard, 2004). Thus, we were surprised by the lack of effects on spatial learning and memory in the Arx CKO mice in light of the increased anxiety-like behaviors in these mice. This is particularly interesting given the association of spatial performance in the early stages of the MWM to anxiety or thigmotaxic behavior in mice. Although increase in thigmotaxic behavior in Arx CKO mice was evident in the OF test, analysis of path efficiency and distance traveled to the escape platform did not reveal any differences in the expression of such behavior at any point during the MWM test. These differences in the two paradigms are likely due to context-dependent expression of fear (Van Meer and Raber, 2005).

The selective impairments of anxiety in the Arx CKO mice among the three behavioral domains tested in this study is consistent with the previously described relationship between alterations in the functional properties of PVIs and manifestation of cognitive and motor deficits relevant to neurodevelopmental disorders (Hu et al., 2014). Notably, alterations in the neurotransmitter release properties of PVIs by deletion of the Glutamic acid decarboxylase 67 gene or in their excitability by genetic deletion of ion channels such as NMDARs, GluR1, metabotropic GluR5, and the sodium voltage-gated channel alpha subunit 1 (SCN1A) have led to variable cognitive and spontaneous activity phenotypes from significantly affected to indistinguishable from controls (Billingslea et al., 2014; Carlen et al., 2012; Fuchs et al., 2007; Korotkova et al., 2010; Saunders et al., 2013). Our results suggest that anxiety-like behaviors are among the emergent behavioral phenotypes from mice with dysfunctional PVIs and that their manifestation may be primarily related to postnatal alterations in the functional properties of these cells. Although this selective anxiety phenotype in our postnatal Arx CKO model is interesting, it is in contrast to the diverse adaptive behavioral deficits in various Arx mutant mice (Dubos et al., 2018; Jackson et al., 2017; Kitamura et al., 2009; Price et al., 2009; Simonet et al., 2015). These discrepant observations are likely due to differences in prenatal and postnatal Arx roles and/or the number/type of interneuron subtypes where Arx is dysregulated. Nonetheless, our present results demonstrate that loss of Arx transcriptional activity in maturing PVIs alone can lead to anxiety-related behaviors commonly noted in EOEEs.

To understand the molecular determinants of the network and behavioral abnormalities in Arx CKO mice, we assessed the transcriptomic profiles of PVIs and found dysregulation of a diverse and broad set of genes. Except for Calb1, these deregulated genes did not overlap with DEGs following embryonic deletion of Arx (Fulp et al., 2008), suggesting that the postnatal Arx transcriptional program mostly diverged from its embryonic counterpart. Interestingly, Arx was not among our DEGs, but qPCR analysis of its expression revealed a significant reduction as expected. The basis for the incoherence between the DEG and qPCR analyses with respect to Arx could be due to a number of factors. First, low Arx read counts in the RNAseq could have diminished the accuracy of quantifying Arx expression (Lozoya et al., 2018). Second, P35-40 total RNA from FACS sorted Arx null PVIs used for genome-wide expression profiling may have been contaminated with slowly degrading mRNA transcribed prior to recombination at ∼ P15. Thus, it may be advantageous in the future to isolate RNA from Arx CKO models in older mice to mitigate the possible impact of slowly degrading Arx mRNA on transcriptomic profiling studies. Third, high GC content of Arx could have severely limited the efficiency of Arx sequencing (Dohm et al., 2008). Fourth, our flox/flox excision approach to generate nonfunctional ARX could have excluded and/or included regulatory elements that differentially affected the stability of the generated Arx RNA. In that context, it is worth noting that conditional ablation of Arx in the embryonic brain resulted in both upregulated and downregulated Arx expression in microarray analysis (Fulp et al., 2008). Interestingly, the upregulated signal was associated with the coding sequence or cDNA, whereas the downregulated signal was associated with a small segment of Intron 1 of the Arx genome rather than spliced RNA. Future studies are needed to sort out the regulatory mechanism that ultimately controls Arx RNA stability. Nonetheless, our qPCR results generally support the RNAseq analysis and demonstrated a clear reduction in Arx expression in PVIs.

Given the diversity and number of altered DEGs in Arx CKO mice, we used bioinformatics approaches to identify the biological pathways that DEGs converged upon and those most essential for the production of Arx CKO network abnormalities. The low Arx expression in the mature brain necessitated this approach, as its low abundance would be a major obstacle to ChIP-Seq and other biochemical functional analyses. We found that upregulated DEGs were particularly involved in synaptic pathways, whereas the downregulated DEGs were mostly involved ECM-related pathways. More importantly, both the up- and downregulated DEGs were primarily connected via co-expression networks with many of these DEGs having putative Arx-binding motifs. These motifs were found in an overwhelming number of ECM-related genes in the downregulated DEGs, suggesting that Arx preferentially binds to ECM genes, making them the likely locus of the hub pathways that led to network abnormalities in Arx CKO mice. Generation of epitope-tagged Arx will enable the production of supporting empirical evidence to validate this conclusion. Nevertheless, it is worth noting that our analyses revealed that the postnatal loss of Arx in PVIs led to the downregulation of proven epilepsy-related genes such as LAMB1, COL18A1, Nid1, and DOCK6 (Pisciotta et al., 2018; Radmanesh et al., 2013; Suzuki et al., 2002; Vasudevan et al., 2010). Thus, these ECM-related DEGs are likely to be causally relevant to the noted network abnormalities in Arx CKO mice.

Central to our RNAseq observations is the preponderance of ECM-associated genes in the DEGs and the identification of several epilepsy causal genes within that group. Consistent with the profound dysregulation of ECM components, we demonstrated a marked reduction in ACAN and a significant loss of Cat-315, suggesting a disruption in the molecular integrity of the PNN. In contrast, the density of PVIs and other GABAergic interneurons was not affected by the loss of Arx in PVIs. Similarly, PVI dendritic complexity was indistinguishable from control and Arx CKO mice. Overall, our results validate the RNAseq data and provide strong evidence that ECM-related biological pathways are central to network abnormalities in Arx CKO mice. Indeed, this conclusion is supported by previous reports of epileptogenesis in mice deficient in ECM molecules (Dityatev, 2010; Vasudevan et al., 2010) and reports of seizures leading to remodeling of the ECM (Dityatev, 2010).

PVIs are connected to pyramidal cells in cortical regions via feedback and feedforward microcircuits (Klausberger and Somogyi, 2008), endowing the intrinsic properties of these interneurons with the power to influence the output of pyramidal cells. Intrinsic membrane active properties were found to be reduced in PVIs of the Arx CKO mice. Notably, the rheobase and spike frequency properties were among the most profoundly affected intrinsic properties. Interestingly, these observations were not correlated with changes in the expression voltage-gated ion channels in our RNAseq dataset. We surmise that they are likely linked to the downregulation of ECM genes in PVIs and subsequent breakdown of the PNNs around these cells. Indeed, several studies have demonstrated direct intrinsic membrane impairments in line with our observations following enzymatic breakdown of the PNNs (Balmer, 2016; Lensjo et al., 2017).

How does the dysregulation of ECM and subsequent breakdown of the PNNs lead to hypo-excitability of PVIs in the absence of dysregulated ion channels? One hypothesis to explain this would be altered dendritic complexity. We were not able to demonstrate a dendritic change, despite the previously reported relationship between dendritic complexity and neuronal excitability (Mason and Larkman, 1990). Another possibility, that our data support, is that dysregulated ECM and changes to the PNNs alters the density of voltage-dependent ion channels at the membrane. Indeed, it has been previously shown that the PNN component tenascin-R interacts with NaV1.2 (Srinivasan et al., 1998), suggesting that clusters of NaV1.2 bound by tenascin-R could have diffused after degradation of the PNNs around PVIs, leading to hypo-excitability. In addition, breakdown of the PNN could lead to the loss of negative charge with decrease in extracellular protein (Morawski et al., 2015) ultimately impacting local electric field sensing by ion channels and subsequent neutralization of membrane surface charge akin to the effect of high extracellular [Ca²⁺ on membrane properties (Frankenhaeuser and Hodgkin, 1957). This neutralizing effect by diffused PNN could have then shifted the voltage-dependence of ion channels toward more depolarized membrane potentials (Frankenhaeuser and Hodgkin, 1957), leading to higher rheobase current and reduced spiking as noted in Arx null PVIs. Although our RNAseq data support an ECM-dependent mechanism in the alterations of PVI membrane properties, we acknowledge that the excitable properties of neurons, which can be influenced by localization or phosphorylation, may not always align to the transcriptomic level of a particular ion channel. Although RNA-Seq transcriptomic profiling provides a comprehensive analysis of actively regulated genes, many studies have found no correlation or seemingly inverse relationships between the expression of various membrane (Adelman et al., 2019; Bomkamp et al., 2019; Földy et al., 2016; Larson et al., 2016; Tripathy et al., 2017) and synaptic (Fazel Darbandi et al., 2018; Harrington et al., 2016; Yook et al., 2019) genes with their electrophysiological properties. Thus, the relationship between gene expression and cellular/behavioral phenotypes is complex, and this complexity must be taken into account in future studies into the ion channel mechanisms mediating hypoexcitability of PVIs in Arx CKO mice.

In addition to impaired intrinsic properties, disruption in the balance of excitation and inhibition likely contributes to network hyper-excitability in Arx CKO mice (He and Cline, 2019). Thus, we recorded both spontaneous EPSC and IPSC currents on PVIs and noted a constellation of changes to the properties of these spontaneous events in presence or absence of TTX that suggests an overall reduction in basal excitatory drives onto PVIs. A reduction in sEPSC frequency and mEPSC amplitude was most prominent among those changes. There are many possible explanations for those prominent changes. For example, fewer excitatory synapses, altered probability of vesicle release, or reduced network activity could all be altered via cell autonomous and non-cell autonomous mechanisms. However, the lack of changes in mini EPSCs frequency and preservation of normal dendritic elaboration suggest that change in synaptic density on PVIs is not a likely contributing factor.

To further probe the contribution of impaired synaptic drive to network hyper-excitability in Arx CKO mice, we measured I-O relationship of evoked EPSCs and IPSCs on PVIs and found a decreased excitability of PVIs primarily driven by a reduction in eEPSCs. The locus of this reduction in synaptic excitability appeared to be the postsynaptic side as evidenced by changes in conduction and ratio of AMPA and NMDA receptors. The reduction in postsynaptic GluR current at PVI synapses is inconsistent with the upregulation of DEGs associated with excitatory transmission. We surmise that this may be part of an ongoing compensatory response to restore synaptic excitability at PVI synapses following the loss of Arx. The reduction in excitatory transmission in the face of increased expression of glutamate receptor signaling genes can be explained by reduction in lateral diffusion of AMPA and NMDA receptors from extracellular to synaptic compartments (Groc et al., 2006). Previous studies have shown that the ECM/PNN components do indeed operate as passive diffusion barrier that control the lateral diffusion exchanges of glutamate receptors between synaptic and extrasynaptic compartments (Shi and Ethell, 2006). Interpretation of the RI of the AMPA and NMDA receptors herein may be complicated by their reversal potentials, which were considerably different from 0 mV, an observation previously noted in another study (Suyama et al., 2017). This error in voltage could be due to series resistance, activation of K+ channels at positive potential, liquid junction potential, and poor space clamp among others. We limited the source of this error by employing a cesium-based internal solution to block K⁺ leak currents and compensated for liquid junction potential as well as series resistance, albeit up to 80%. We suppose that this error in voltage might then be due to poor space clamp, a serious limitation in slice recordings that is currently difficult to overcome. To mitigate its impact on our analysis, we excluded cells whose reversal potentials were substantially far from 0 mV and made sure the reversal potentials in the remaining cells from both genotypes were comparable. Nonetheless, these findings strengthen our conclusion that the reduction in PVI synaptic excitability was primarily associated with impaired excitatory drives with a postsynaptic locus.

Taken together, our findings demonstrate that continued expression of Arx has an important role in the functional maintenance of PVIs and that alterations in Arx expression in these cells specifically during early postnatal brain development contributes to the pathologic mechanisms of ARX-related EOEEs. Although the precise molecular mechanisms mediating this postnatal role remain to be elucidated, our results establish a role for Arx transcriptional activity in the functional properties of PVIs and in the proper maintenance of ECM integrity around these cells. Loss of this postnatal transcriptomic regulation of PVIs resulted in a profound reduction in intrinsic and synaptic excitability, leading to behavioral and network abnormalities. Our results suggest that restoration of cellular properties and network functioning in adult mice may be a viable therapeutic approach toward ARX-related EOEEs. Notably, functional restoration of circuit activity in mature mice by Cre-mediated re-expression of certain genes (Cisse et al., 2017; Guy et al., 2007; Rotaru et al., 2018; Silva-Santos et al., 2015; Ure et al., 2016) or by cell transplantation (Casalia et al., 2017; Cunningham et al., 2014; Howard et al., 2014) has shown great promise in a number of NDDs including Rett syndrome, Dravet syndrome, and Angelman syndrome. Similar approaches could be applied to the Arx-linked EEOE mouse models in the near future to test whether reintroduction of Arx or downstream effectors, including epigenetic and epitranscriptomic regulators, could provide more precise and enduring restorative therapeutics in Arx-linked EOEEs.

Limitations of the study

We aimed to determine the role of an early acting developmental TF Arx in neural circuits and its contribution to the progression of EOEEs by using cre-lox technology. Although our new PVI CKO model allowed dissection of the functional consequence of Arx ablation in PVIs, this model does not mimic the normal disease process, which involves both developmental and ongoing dysfunction in Arx signaling in all interneurons. Another limitation is that although we focused our investigations on the hippocampus, a region vital for learning and seizure generation, Arx expression was also ablated in PVIs from other brain regions in our Arx CKO model. Therefore, we cannot rule out the possibility of Arx-fulfilling distinct functions at other brain regions nor the possibility of the noted effects in CA1 hippocampus being mediated by connective or interactive defects between those regions. Despite these limitations, cell-type-restricted gene knockout models represent necessary models to understand the biological underpinnings of neurological disorders including the EOEEs. This is particularly relevant to Arx-linked EOEEs, given the pleiotropic phenotypes, strong genotype-phenotype relationship, and the numerous symptoms associated with ARX disorders are probably related to cell-specific alterations in the function of Arx. This study has now defined important roles for postnatal Arx in PVIs in the specific control of neural circuit function in the hippocampus. Further studies are needed, however, to establish the specific roles of this TF in PVIs from other brain regions as well as in other interneuron subtypes throughout the brain. Insights from those studies along with those from this study should provide a more comprehensive view of the full range of control exerted by postnatal Arx transcriptional activity in the complex regulation of neural circuits and facilitate the development of targeted circuit-based therapies in EOEEs.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to Eric D. Marsh (marshe@email.chop.edu).

Materials availability

No new unique materials or reagents were generated in this study.

Data and code availability

RNA sequencing data are available in GEO depository: GSE157689

https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE157689.

Datasets used for generating figures are available in Mendeley Data depository: DOI: 10.17632

https://doi.org/10.17632/r94jkwmd5z.1.

MATLAB codes are available from the Lead contact on reasonable request.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.

Published: January 22, 2021