Reduction of forebrain interneurons alters cellular organization and cortical neurogenesis in a mouse model of X-linked intellectual disability

Abstract

Arx is an X-linked transcription factor that is expressed in inhibitory neurons within the ganglionic eminences during embryogenesis. In humans, ARX variants are one of the leading causes of X-linked intellectual disabilities (XLID) and related syndromes. While most often studied in males, recent evidence indicates that >50% of heterozygous female carriers of ARX variants also display mild to severe phenotypes. The mechanisms by which Arx mutations lead to a wide range of disease pathologies remains poorly understood. Here we removed Arx in the medial ganglionic eminence (MGE) to generate conditional knockout male and heterozygous female mice. MGE-derived interneurons were reduced in a gene dosage, subtype and brain region dependent manner in Arx mutant mice. Knockout male mice displayed severe phenotypes with infantile spasms and epilepsy, while Het female mice displayed moderate anxiety and locomotor defect mimicking patient symptoms. Single cell sequencing and spatial transcriptomics revealed striking cell autonomous and non-cell autonomous changes throughout development. Notably, loss of MGE-derived interneurons alters dorsal cortical neurogenesis and cell-cell communication resulting cortical malformations. This study reveals mechanistic insights into genetic and cellular changes that occur when Arx is removed from MGE-derived interneurons that advances our understanding of Arx-related neurodevelopmental disorders.

INTRODUCTION

The X-chromosome contains ~5% of the human genome, yet ~15% of genes (> 100 genes) associated with intellectual disability (ID) are located on the X-chromosome^1,2. This X-linked recessive inheritance of ID-related genes in large part explains why the prevalence of ID is ~50% higher in males³. X-linked intellectual disabilities (XLIDs) are a complex and heterogeneous set of syndromes, with ~50% of these disorders being comorbid with seizures or epilepsy⁴. A better understanding of the genetic mechanisms underlying XLIDs is necessary to understand these disorders and develop new therapeutic treatments.

One of the most frequently mutated genes in XLIDs is the transcription factor Aristaless-related homeobox, Arx. Males with ARX variants display wide ranging symptoms including ID, epilepsy, lissencephaly, and autism-like phenotypes. The most severe syndromes involving brain and genital malformations, (X-linked lissencephaly associate with abnormal genitalia, XLAG, and Proud Syndrome) often arise from point mutations or deletions near the homeodomain while non-malformation phenotypes (Partington Syndrome and West Syndrome) arise from polyalanine expansions and duplications outside of the homeodomain region^5–8. While most clinical and scientific studies on Arx function focus on males, recent evidence indicates that ~60% of female heterozygous carriers of ARX variants display mild to severe XLID-like symptoms⁹. Some of this variability is likely due to X chromosome inactivation (XCI). There is a bias of paternal XCI in the brain, and evidence that local XCI mosaicism of WT vs. mutant alleles can correlate with disease penetrance¹⁰.

The comorbidity of ID and epilepsy implicates abnormal function of GABAergic interneurons in XLIDs and ARX variants. Indeed, Arx is strongly upregulated in postmitotic neurons within the medial, lateral and caudal ganglionic eminences (MGE, LGE & CGE, respectively)^11,12, the source of nearly all GABAergic interneurons and projection neurons in the forebrain. Several studies found that loss of Arx in the GEs disrupts normal migration of interneuron precursors^11,13,14, leading to reduced cortical interneurons^14–16 and increased seizures¹⁷. Notably, postnatal removal of Arx from parvalbumin-expressing (PV+) interneurons altered network excitability and induced spontaneous seizures¹⁸, indicating that Arx plays an important role in interneuron function in adult mice as well. Other studies generated mice mimicking human ARX variants that display variations in symptoms and phenotype severity^19–24. Additionally, perturbation of ARX in organoids and human primary fetal neuronal cultures recapitulate some of these migration defects and cell fate changes^25,26. Despite this progress, the mechanisms by which Arx mutations lead to disease pathology remain poorly understood.

To characterize how loss of Arx induces cell autonomous and non-cell autonomous changes in interneurons throughout development, we generated conditional knockout male and heterozygous female mice where Arx is specifically removed from the MGE. While male KO mice have a near complete loss in MGE-derived interneurons in the cortex and hippocampus, female Het mice show a ~50% reduction. Arx cKO male mice die by 6 weeks, but Arx Het females display increased in seizure susceptibility and mild impairment in locomotor activity and sensorimotor gating, consistent with human female heterozygous carriers of ARX variants⁹. We combined single nuclei Multiome analyses and spatial transcriptomics to characterize striking cell autonomous and non-cell autonomous changes in the embryonic and adult brains of Arx cKO males and het female mice. Loss of Arx in the MGE produces a distinct, transcriptionally defined cohort of cells in male cKO, while females contain both normal and aberrant neuronal populations. Arx-deficient MGE-derived interneurons also exert non-cell-autonomous effects on cortical progenitor cells and Cajal-Retzius cells, resulting in a thinner cortex and abnormal cortical lamination. In the adult, we observe cellular and transcriptional changes associated with epilepsy, especially in the hippocampus. This comprehensive dataset provides a valuable resource for understanding how loss of Arx effects gene expression, cell fate and behavior in relation to human XLIDs.

RESULTS

MGE-derived interneuron numbers decline in a Arx dosage-dependent manner

We generated conditional Nkx2.1-Cre^C/+;Arx^F/Y;Ai9^F/+ knockout males (cKO) and Nkx2.1-Cre^C/+;Arx^F/+;Ai9^F/+ heterozygous females (Het) to remove Arx from the MGE, with Nkx2.1-lineage cells expressing the red fluorescent reporter tdTomato. We confirmed that Arx protein is strongly downregulated in the MGE (Figure S1A). Most cortical MGE-derived interneurons express Arx at E13.5 and P0, but Tom+ cells in the cortex of cKO male mice were Arx-negative (Figure S1A–B). Arx cKO male mice exhibited reduced body and brain weights, with most dying before weaning, whereas Het females had normal body weight, brain weight and viability (Figure S2A). The MGE gives rise to nearly all parvalbumin- and somatostatin-expressing interneurons (PV+ and SST+, respectively) in the cortex, as well as a population of neuronal nitric oxide synthase-expressing (nNos+) neurogliaform and ivy cells in the hippocampus. At P21, we observed a near complete loss of MGE-derived interneurons in the cortex, with significantly decreased density and percentage of PV+ and SST+ interneurons in Arx cKO males (Figure 1A–B). Note that most Tom+ cells in the cortex of Arx cKO males do not express PV or SST, and conversely most PV+ and SST+ cells are Tom-negative. In the Nkx2.1-Cre mouse, Cre is not efficiently expressed in the most dorsal portion of the MGE²⁷, so these PV+/Tom- and SST+/Tom− cells likely originate from this region where Cre is not expressed, and thus Arx expression is presumably normal in these cells. Arx Het females exhibited a ~50% reduction in PV+ and SST+ interneuron numbers, but their relative proportions of these subtypes remained unchanged compared to WT (Figure 1A–B). We observe a ~2:1 ratio of PV+:SST+ cells in WT and Het females, but this ratio is reversed in cKO male mice (Figure 1C). This decreased tendency and altered proportion in MGE-derived cells was present in both the deep and superficial cortical layers (Figure S3).

Figure 1. MGE-derived interneuron loss in a Arx dosage-dependent manner.

A. P21 somatosensory cortex from Nkx2.1-Cre;Arx;Ai9 WT male and female, Het female and cKO male mice stained for tdTomato, PV, SST and DAPI. Scale bar = 50 μm. B. Quantification of cell density (left) and percent of PV+ and SST+ cells (right) in the somatosensory cortex. n = 3 for each condition. C. Average relative percent of PV+ and SST+ cells in the somatosensory cortex. D. P21 hippocampus in WT and cKO male mice stained for PV, SST and nNos. Scale bar = 100 μm. E. Quantification of cell density of PV+, SST+ and nNos+ cells in the hippocampus. n = 5 WT male, 5 WT female, 5 Het female and 3 cKO male mice. F. P21 hippocampus in WT and cKO male mice stained for SST and Calb1. Scale bar = 100 μm. G. Quantification of cell density of Tom+/Calb1+ and Tom+/Calb1+/SST+ cells in the hippocampus. n = 3 WT and cKO brains, 12 sections/brain, each brain labeled with a different shape. All stats are one-way ANOVA followed by Tukey’s multiple comparison tests (B, E), or Welch’s 2-tailed t-test (G: left) and Mann-Whitney 2-tailed test (G: right) based on normal distribution when only WT and cKO mice tested: * = p ≤ .05, ** = p ≤ .01, *** = p ≤ .001, **** = p ≤ .0001.

There was also a strong reduction of MGE-derived interneurons in the hippocampus of Arx cKO males, but less extreme compared to the cortex. The overall number of Tom+ cells was similar between Arx Het females and cKO males, ~50% reduction compared to WT mice (Figures 1D–E & S4A–B). nNos+ cells were nearly absent in the Arx cKO hippocampus whereas the number of SST+ interneurons was similar between Arx Het females and cKO males (Figures 1D–E & S4A–B). The PV+ presynaptic marker synaptotagmin-2 (Syt2)²⁸ was strongly decreased throughout the hippocampus of Arx cKO males (Figure S2B). This distribution of interneuron loss was not uniform across hippocampal regions: Tom+ cell loss was most pronounced in the dentate gyrus (DG) relative to CA1 and CA2/3 regions of Arx cKO males (Figure S4C–E). We observed a significant increase in the number of Tom+/Calb1+ interneurons in the hippocampus of Arx cKO males, with many of these cells co-expressing SST (Figure 1F–G). The MGE gives rise to a population of oligodendrocytes in CA3 that perdure into adulthood^29,30, and the density of these oligodendrocytes is unaffected in Arx cKO males (Figure S2C–D). In sum, forebrain MGE-derived interneurons were reduced in a gene dosage, subtype, and region dependent manner in Arx Het female and male cKO mice.

Increased seizure sensitive and moderate impairment in locomotor and sensorimotor gating of Arx Het mice

Most Arx cKO male mice exhibited spontaneous seizures from P10-P21, characterized by severe forelimb clonus and rearing with hypoactivity. These seizures are likely a leading cause of their premature death, which prevented us from performing any behavior assays on these mice. Over half of human female carriers of ARX variants display seizures, ID and other phenotypes^9,17, and we observed spontaneous seizures in a few Het female mice. To assess seizure susceptibility in Het female mice, we induced seizures with a single injection of Pentylenetetrazole (PTZ). Het female mice exhibited increased seizure scores and seizure severity compared to WT mice, indicating increased seizure susceptibility in Het females (Figure 2A).

Figure 2. Increased seizure sensitivity and impaired locomotor and sensorimotor gating of Arx Het mice.

A. Maximum seizure scores for WT (n = 9) and Het (n = 11) females following PTZ injection (left), and during the 20-minute observation period (right). B. Representative trajectory plots (left) and time in the center zone (right) of female WT (n = 10) and Het (n = 12) mice during the open field test. C. Number of entries into light zone of WT (n = 12) and Het (n = 12) mice in the light-dark box. D. Beam break counts of fine movements (left) and rearing (right) of WT (n = 10) and Het (n = 13) in home cage test. E. Front base (left) and front left paw stride (right) of WT (n = 5) and Het (n = 8) mice in the free walk test. F. Percent of prepulse inhibition (PPI) in WT (n = 6) and Het (n = 9) mice. G. Discrimination index (left) and total distance traveled (right) in the 2-object novel object recognition test of WT (n = 12) and Het (n = 12) mice. Stats are ordinary two-way ANOVA (A: right, F) followed by Sidak’s multiple comparisons test (B), Mann-Whitney 2-tailed test (A: left) and unpaired 2-tailed t-test (C, D, E, G) based on normal distribution: * = p ≤ .05, ** = p ≤ .01, *** = p ≤ .001.

We performed a series of behavioral tests to assess phenotypes associated with neurodevelopmental disease in WT and Het female mice. To assess anxiety-like behaviors, we performed the open field test and light-dark box test. Het female mice spent less time in the center zone during the open field test, with this difference reaching significance during the final 5-minute interval (Figure 2B). In the light-dark box test, Het female mice exhibited a significant decrease in the number of entries of the light zone (Figure 2C). To assess general locomotor activity, we continuously recorded singly housed mice using the Photobeam Activity System over four days. Het female mice exhibited an increase in fine movements with a decrease in rearing movements compared to WT mice (Figure 2D), mimicking dystonic movements in human patients with ARX variants⁷. Gait and motor coordination was assessed in a free-walk test, but no significant differences were observed between WT and Het female mice in any parameter measured (Figure 2E).

We then assessed sensorimotor gating and memory behaviors in Het female mice. Sensorimotor gating prevents the brain from overwhelmed by environmental stimuli and reflect the ability to suppress a motor response³¹. Prepulse inhibition (PPI) of the startle response assesses sensorimotor gating, and deficits in PPI are associated with schizophrenia and ADHD^31,32. Het female mice exhibited a significantly reduced PPI percentage compared to WT mice, indicating deficits in sensorimotor processing (Figure 2F). To assess memory deficits, we employed the 2-object novel object recognition test. Although no significant difference in discrimination index was observed between groups, Het female mice traveled a greater total distance during the task compared to WT controls (Figure 2G), potentially reflecting hyperactivity and/or heightened anxiety, consistent with results from the open field and light-dark box tests. In summary, Het female mice display behavioral deficits including increased seizure susceptibility, mild anxiety-like behavior, and impairments in locomotor activity and sensorimotor gating, as observed in human female carriers.

Building a spatiotemporal transcriptome atlas in Arx mutant mice

The above findings reveal that Arx cKO male and Het female mice represent two levels of disease phenotypes. Arx cKO mice recapitulate severe clinical features of human ARX variants including postnatal microcephaly, intractable epilepsy, and early lethality⁷. Het female mice, by contrast, display a moderate phenotype consistent with human female carriers of ARX variants, characterized by mild to moderate intellectual disability, stereotyped hand movements, and dystonia, deficits that are associated with dysfunction of the cerebral cortex, hippocampus, and basal ganglia^7,17. To elucidate the genetic mechanisms underlying these two distinct phenotypes, we constructed a transcriptomic atlas spanning developmental stages through adulthood using single nuclei Multiome sequencing (snRNA-seq and snATAC-seq) and spatial transcriptomics (Figure 3A).

Figure 3. Spatiotemporal transcriptome atlas in Arx mutant mice.

A. Summary of ages & genotypes for snMultiome-seq and spatial transcriptomics workflow. B. UMAP plots of snRNA (top) and snATAC (bottom) data of E13.5 and P70 Nkx2.1-lineage cells annotated by age, gender and genotype. C. UMAP plots of snRNA annotated by marker genes for apical progenitors (Nestin), basal progenitors (Ccnd2), post-mitotic immature neurons (Dcx), and mature SST+ (Sst) and PV+ (Pvalb) interneurons. D. E13.5 and E15.5 spatial transcriptomics data clustered by gene expression profiles that correspond to defined anatomical regions. E13.5 and E15.5 datasets were individually analyzed, and colors only matched in individual timepoint across genotypes. E. Postnatal spatial transcriptomics brain slices colored by molecular clusters based on gene expression profiles. F. UMAP plots (downsampled to 5,000 spots per tissue) of postnatal spatial transcriptome slices annotated by clusters (top) and samples (bottom). Cluster colors in UMAP are same as in panel E. G. UMAP plots annotated by marker genes for excitatory neurons (Slc17a7+), inhibitory neurons (Gad1+), astrocytes (Gfap+) and oligodendrocytes (Mbp+).

We generated WT, Het female and cKO male Nkx2.1-Cre;Arx^F;Sun1-sfGFP mice to harvest Nkx2.1+ lineage GFP+ nuclei via flow cytometry. We performed single nuclei Multiome sequencing on MGE from E13.5 WT male, cKO male, WT female and Het female mice. We also harvested Nkx2.1+ lineage nuclei from the cortex of P70 WT and Het females for Multiome sequencing, but the lack of MGE-derived cortical interneurons in adult cKO male mice prevented similar analyses. 18,758 E13.5 nuclei and 16,214 P70 nuclei passed quality control and were used for analysis (Supplementary Table 1). Visualization of RNA-only and ATAC-only data revealed clear segregation between embryonic MGE and adult cortex datasets (Figure 3B). Distinct neurogenic cell populations such as apical progenitors (Nes+), basal progenitors (Ccnd2+), and postmitotic neurons (Dcx+) were visible in the E13.5 dataset, whereas the two principal MGE-derived interneuron subtypes, Pvalb+ and Sst+, were clearly segregated in the P70 dataset (Figure 3C).

To characterize how loss of Arx in MGE-derived cells alters gene expression in all cell types throughout development, we generated spatial transcriptomics datasets spanning embryonic development through adulthood (Figure 3A & Supplementary Table 1). Spatial transcriptomics was performed on all sexes and genotypes at E13.5, E15.5 and postnatal days 16–21. We also performed spatial transcriptomics in male cKO mice at P42, the oldest age we obtained, and female Het mice at P120, following the behavior assays described above. In the E13.5 and E15.5 brain, we identified 27 cell clusters based on gene expression profiles that correspond to defined anatomical regions, such as the dorsal cortical ventricular zone (Pax6+), cortical plate (Neurod6+), future striatum (Ebf1+) and MGE (Lhx6+) (Figure 3D). We observed a striking difference in cluster composition in the postmitotic mantle region of the MGE of male cKO mice compared to WT in both ages (Figure 3D).

For the P16-P120 brain sections, 49 cell clusters were generated based on gene expression profiles with a clear correspondence to anatomical regions and cell types (Figure 3E). We then downsampled each section to 5,000 spots to facilitate UMAP visualization (Figure 3F). Within these datasets, we identified several major cell types, including excitatory neurons (Slc17a7+), inhibitory neurons (Gad1+), astrocytes (Gfap+) and oligodendrocytes (Mbp+) (Figure 3G). Thus, we could identify the expected developmental trajectory and cell types in the single nuclei Multiome and spatial transcriptomics dataset from embryogenesis through adulthood.

Ectopic cell population with distinct genetic signatures in the MGE of Arx mutant mice revealed by spatiotemporal transcriptomics

We subsetted the E13.5 MGE Multiome data and integrated the snRNA and snATAC datasets using weighted nearest neighbor (WNN). Postmitotic cells in male cKO mice were completely segregated from postmitotic cells in male WT mice, with female Het mice having cells in both the endogenous and ectopic clusters (Figures 4A–B & S5A–B). We defined cell clusters based on their gene expression profiles, revealing 4 progenitor cell types, 14 postmitotic cell types, and 1 cluster of oligodendrocyte precursors (Figures 4B & S5A). There were no changes in the proportion of progenitor cell types in male or female Arx mutant mice, likely due to Arx expression being enriched in postmitotic MGE neurons¹¹ (Figure 4C).

Figure 4. Arx dose-dependent alteration of MGE cell fate at E13.5.

A-B. UMAP plots of integrated snRNA and snATAC data via weighted nearest neighbor (WNN) of Nkx2.1-lineage cells from E13.5 MGE, annotated by genotype (A) and inferred cell clusters based on gene expression profiles (B). C. Relative proportions of E13.5 MGE cell clusters of WT, Het and cKO mice. D-E. mRNA expression of Lmo1, Lmo3, Lmx1b, Lhx8, Sst, and Nkx2.1 in E13.5 brains visualized in spatial transcriptomics data (D, left), UMAP plot from snMultiome data (D, right) and FISH (E). Scale bar = 200 μm (D) or 100 μm (E). F. Tracks showing accessible chromatin peaks from snATAC data (pseudobulk) for individual genotypes at the Lmo1, Lmo3, Lmx1b, Lhx8, Sst and Nkx2-1 genomic loci.

In male cKO mice, we observe a near complete loss of Maf+ cells, which are predictive of PV-fated interneurons³³, and a strong reduction of normal SST-fated interneuron cluster (Figure 4B–C); this is consistent with the downregulation of these interneurons in cKO male mice. Nearly all Lhx8+ clusters, which predominantly give rise to MGE-derived striatal interneurons and projection neurons comprising the globus pallidus and other basal ganglion nuclei^34–36, are lost in male cKO mice (Figure 4B–C). These Lhx8+ cells are replaced by cell clusters identified by strong upregulation of several Lim homeodomain transcription factors (Lmo1, Lmo3, Lmo4, Lmx1b) and the transcription factor Ebf3 (Figure 4B–C). We identified two clusters expressing calbindin (Calb1+), one of which is completely lost (Hs3st4+) while the other is expanded (Fstl4+) in male cKO mice (Figure 4B–C). Whether this expanded Calb1+ population in the MGE gives rise to the increased Calb1+ interneuron population in the hippocampus of adult male cKO mice (Figure 1F–G) remains unknown. The MGE of female Het mice contained both the normal and ectopic clusters, with ~50% reduction of endogenous Maf+, Sst+ and Lhx8+ clusters compared to female WT mice, and the proportion of mutant-specific ectopic clusters is ~50% compared to male cKO mice (Figure 4A–C). These results are consistent with the ~50% reduction of PV+ and SST+ interneurons in adult female Het mice.

Most differentially expressed genes (DEGs) were increased in the MGE of both male cKO and female Het MGE (Figure S6A), consistent with Arx functioning primarily as a transcriptional repressor^37–39. Markers of the ectopic cell clusters (Lmo1, Lmo3, Lmo4, Lmx1b, Ebf3) were some of the strongest upregulated genes in male cKO mice (Figure S6A). Several genes associated with cell differentiation and fate commitment (Etv1, Pbx1, Pbx3)^40–43, alternative splicing (Rbfox1, Nova1), and epilepsy and neurodevelopmental delay (Kcnq3, Glyctk, Magel2)^44–46 were also significantly upregulated in the MGE of male cKO and female Het mice (Figure S6A–B). In contrast, genes associated with GABAergic neuron identity (Lhx8) and PV+ interneuron specification (Maf) were significantly downregulated in Arx mutant mice (Figure S6A). Many of these genes also had corresponding differentially accessible peaks (DAPs) (Figure S6C), indicating a clear correlation between chromatin accessibility and gene expression in these mice.

Several genes involved in interneuron migration were altered in Arx mutant mice. Neuregulin 1 (Nrg1), Ackr3 (Cxcr7) and Cxcr4 are critical regulators of tangential interneuron migration to the cortex^41,47,48. Nrg1 was one of the strongest upregulated genes in Arx mutant MGE while Ackr3 and Cxcr4 were both downregulated in Arx mutants (Figure S6A). Dysregulation of these guidance factors are likely causes for abnormal interneuron migration in the ventral forebrain.

Complementing this Multiome data with spatial transcriptomics provided additional insight into gene expression changes in Arx mutant mice. While Lmo1 is expressed in cycling MGE progenitors in WT mice, Lmo1 is strongly upregulated in postmitotic cells in the MGE mantle in male cKO mice and appears to be downregulated as these cells migrate out of the MGE (Figure 4D). Lmo3 is upregulated in cells slightly deeper in the mantle and is maintained in the Islr2+ and Calb1+/Fstl4+ postmitotic neuron clusters (Figure 4D). Lmx1b+ is specifically upregulated in the deepest portion of the mantle, almost complementary to Lmo1 (Figure 4D). These observations were confirmed with FISH (Figures 4E & S7). We also observed a concurrent increase in chromatin accessibility at the promoters of Lmo1, Lmo3 and Lmx1b in male cKO mice compared to WT, with the largest difference at Lmo3 (Figures 4F). Female Het mice displaying a more moderate increase in accessibility at these loci. Conversely, Lhx8 was strongly reduced in the MGE mantle of male cKO mice, concomitant with reduced accessibility at the Lhx8 promoter, but Lhx8 expression was retained in the subventricular zone (SVZ) (Figure 4D–F).

Despite the reduction of MGE-derived interneurons, we observed a significant increase in Nkx2.1 and Sst expression in the MGE of Arx mutant mice (Figure S6A). Although the normal Sst+ cell cluster was nearly absent in male cKO MGE, many cells in the ectopic clusters express Sst (Figure 4D–E). Notably, the Sst locus had the largest increase in accessibility in male cKO mice in terms of fold change (Figures 4F & S6C). Nkx2.1 is normally restricted to cycling MGE progenitors, but Nkx2.1 expression is maintained in postmitotic cells in both male cKO and female Het mice (Figures 4D–E & S7).

The transcriptional changes were even more extreme in E15.5 Arx mutant mice. We extracted the entire E15.5 subpallium to ensure that all postmitotic migratory cells were included in this analysis. There were more DEGs at E15.5 compared to E13.5 in both cKO male and Het female mice (Figure S6D). Consistent with E13.5 results, markers of the ectopic cell population (Lmo1, Lmo3, Lmo4, Lmx1b, Ebf3) are significantly increased in Arx mutant mice (Figures S6D & S8). Nkx2.1, Sst, Nrg1, and Kcnq3 were also significantly increased while Lhx8, Maf, Ackr3, and Cxcr4 were decreased in E15.5 cKO mice (Figures S6D & S9A–B). Het female mice exhibited fewer DEGs compared to cKO male mice, but still displayed enrichment of Lmo1, Lmo3, Lmo4, Sst and reduction of Lhx8 and Ackr3 at E15.5 (Figure S6D). We confirmed many of these transcriptome changes via FISH (Figure S8 & S9). Notably, we observe a bolus of Nkx2.1+ MGE-derived interneurons in E13.5 and E15.5 cKO male mice stuck in the GE mantle, specifically the MZ migration stream (Figure S10A–B). These Nkx2.1+ MGE-derived cells were still confined to its periventricular zone in the ventral forebrain at P0 (Figure S10C), consistent with previous study¹³. We did observe some differences between E13.5 and E15.5. For example, Fgf17 and Fgf15 are expressed in the Nkx2.1+ septum at E13.5, in agreement with a previous study⁴⁹, but both genes are strongly reduced in E15.5 cKO mice (Figure S9C).

A recent study using primary human neural progenitor cultures demonstrated that ARX promotes interneuron formation by repressing LMO1, and ARX knockdown in these cells produced an ectopic cell cluster containing over 700 DEGs²⁶. 106 of these human DEGs were also dysregulated in our E13.5 MGE dataset. Among these, LMO1, LMO3, LMO4 and EBF3 were top genes upregulated in both datasets (Figure S10D), indicating strong similarities in transcriptome changes when ARX is removed from human and mouse neural progenitors.

In sum, integration of single nuclei Multiome-seq, spatial transcriptomics and FISH in Het female and cKO male mice revealed how reduction of Arx disrupts postmitotic cell fate specification, cell migration and developmental trajectories in a dosage-dependent manner via alteration in gene expression and chromatin accessibility of critical fate determining genes.

Reduced cell activity and enhanced neuroprotective responses in the cortex of Arx mutant mice

To analyze P16-P120 spatial transcriptomics datasets, cortical regions were segmented using BANKSY⁵⁰, which integrates a cell’s transcriptome with its local microenvironment. Cell type composition was determined using Robust Cell Type Decomposition (RCTD)⁵¹ in combination with the Allen Brain Institute reference scRNA-seq dataset⁵². Only singlet-assigned spots were retained for downstream analysis. Cells from P16, P21 and P42 male WT and cKO datasets largely co-clustered regardless of genotype (Figure 5A). Unsupervised clustering identified 23 hierarchically organized cell subtypes: 2 glia subtypes, 3 MGE-derived interneuron subtypes, 3 CGE-derived interneuron subtypes, and 15 excitatory neuron subtypes spanning superficial to deep cortical layers comprising intratelencephalic (IT), pyramidal tract (PT), near-projecting (NP), and corticothalamic (CT) neuron subtypes (Figures 5B–C & S11).

Figure 5. Non-cell autonomous genetic and cellular changes in the cortex of Arx cKO mice.

A-B. UMAP plots of spatial transcripts labeled by age and genotype (A), and cell types based on gene expression (B). C. Spatial localization of major cortex cell types with singlet-assigned spots. D. Relative proportion of major cortical neuron types in WT and cKO mice. E. Increased proportion of L2–3 IT neurons and reduced proportion of L5 IT Ctx neurons in cKO mice. n = 3 for each genotype. F-G. P21 cortical sections through S1 stained with Cux1 to label layers II-IV (F), with reduced thickness across all cortical layers in cKO mice (G). Scale bar = 100 μm. n = 3 brains/genotype, 12 WT slices and 11 cKO slices counted/genotype, each brain labeled with a different shape. H. Scatter chart showing DEGs downregulated (blue) or upregulated (red) in cKO mice in distinct cortical cell types. I. clusterProfiler GO enrichment of top biological processes for DEGs of L2–3 IT cells between WT and cKO mice. J. Normalized expression of Fos, Mt1, Cck and Mapk3 in L2–3 IT cells in WT and cKO mice at P16, P21 and P42. K. Spatial distribution of mRNA in P16 S1 cortex visualized in cell segmented spatial transcriptomics data. Scale bar = 100 μm. L. P21 S1 cortex from WT and cKO mice stained for cFos (top) and quantification of cFos+ cell density (bottom). Scale bar = 100 μm. n = 4 WT and 3 cKO mice. M. P21 layer 2–3 S1 cortex stained for CCK (top) and quantification of relative fluorescent intensity (bottom). Scale bar = 100 μm. n = 3 mice/genotype. Paired 2-tailed t-test (E), unpaired 2-tailed t-test (L) and Welch’s 2-tailed t-test (G, M) was performed: * = p ≤ .05, ** = p ≤ .01, *** = p ≤ .001, **** = p ≤ .0001.

We quantified the proportions of major cortical neuron types across the three postnatal time points. As expected, MGE-derived interneurons were markedly reduced in the cortex of male cKO mice at all 3 ages (Figure 5D). Removing glia from the analyses, we observed a proportional increase in cortical L2–3 IT neurons and a corresponding decrease in L5 IT neurons in male cKO mice compared to WT (Figure 5D–E). Despite this proportional difference in thickness between superficial and deep cortical layers, all cortical layers were significantly reduced in male cKO mice (Figure 5F–G).

To identify cell autonomous and non-cell autonomous DEGs in cortical cell types across postnatal development, we integrated the P16, P21 and P42 datasets and combined the 15 excitatory neuron subtypes into four broad subgroups: L2–3 IT, L4–6 IT, PT/Car3 and NP/CT/L6b (Figure 5B, H). Among these populations, L2–3 IT neurons exhibited the greatest number of DEGs (296 genes upregulated and 223 genes downregulated in cKO), which were enriched for processes including synaptic function, cell growth, neural development, and ERK signaling (Figure 5H–I).

Activity-regulated genes (ARGs) such as Fos, Arc, Egr1 and Nr4a1 were strongly downregulated across most excitatory neuron subgroups in male cKO mice (Figures 5H, J–K & S12). We confirmed a significant decrease in cFos protein levels in cKO males, with the strongest reduction in the most superficial and deep layers (Figure 5L). In contrast, metallothionein family genes Mt1 and Mt2, which confer protection against oxidative stress and immune challenge⁵³, were significantly upregulated in male cKO cortex (Figure 5H, J–K). The Apoe gene encodes Apolipoprotein E (ApoE), which mediates lipid transportation between cells. APOE4 variants are a major risk for Alzheimer’s disease while ApoE2 and ApoE3Ch variants confer resistance from excitotoxicity and protect neurons from ferroptosis⁵⁴. Apoe expression was significantly increased in cKO cortex (Figure 5H, K). CCK+ neurons co-express neurotransmitters (GABA or glutamate), mediate synaptic transmission, and are integral components of cortical circuits^55,56. Cck expression was strongly downregulated in cKO pyramidal cells, particularly in the L2–3 IT population, which we confirmed via immunostaining (Figure 5H–K, M). Islr2, a marker for L6a excitatory neurons that is associated with neurite extension⁵⁷, was strongly upregulated in deep cortical layers in cKO male mice (Figure 5K). Mapk3, a core component of the MAPK/ERK signaling cascade that is associated with neurodevelopmental disorders in patients with 16p11.2 deletions⁵⁸, was broadly increased in cKO cortex (Figure 5H–K).

In sum, we identified both broad and layer specific non-cell autonomous DEGs in cortical excitatory neurons in Arx cKO male mice, likely arising from the significant loss of MGE-derived interneurons in these mice. DEGs span a broad range of cellular activity including neuropeptide signaling, synaptic function, and neuroprotective responses across multiple cell types, highlighting the extent of non-cell-autonomous transcriptional consequences of interneuron loss.

Hippocampal sclerosis and dysregulation CA3 gene network induced by BDNF in Arx cKO hippocampus

Temporal lobe epilepsy (TLE) is the most common form of focal epilepsy, and mesial temporal lobe epilepsy with hippocampal sclerosis (MTLE-HS) is a well characterized disorder, most commonly occurring in the hippocampus⁵⁹. Since Arx mutant mice display seizures, we extracted the hippocampus from our spatial transcriptomics dataset to identify DEGs in distinct hippocampal cell populations in male cKO mice. Cells could be cleanly divided into principal cells from distinct hippocampal subdomains (CA1, CA2, CA3, DG), MGE-derived (PV and SST) and CGE-derived interneurons (Lamp5 and Sncg), and astrocytes (Figures 6A–C & S13A–B). Hippocampal oligodendrocytes were excluded from the analysis because they could not be cleanly separated from the adjacent corpus callosum in these sections. Consistent with our immunostaining results (Figure 1D–E), PV+ and SST+ interneurons were markedly reduced in the cKO hippocampus (Figure 6B). Notably, CA1 and CA3 cells were largely segregated based on genotype and age (Figure 6A–B), which is in stark contrast to the comingling of cortical cells from WT and cKO mice in a salt-and-pepper fashion (Figure 5A). This indicates greater transcriptional differences in hippocampal excitatory cells compared to the cortex.

Figure 6. Indicators of epilepsy and genetic dysregulation in the hippocampus of Arx cKO hippocampus.

A-B. UMAP plots of hippocampus spatial transcriptomics data labeled by age and genotype (A), and cell types (B). C. Spatial localization of major hippocampal cell types with singlet-assigned spots. D. Scatter chart showing DEGs downregulated (blue) or upregulated (red) in cKO mice in distinct hippocampus cell types E. clusterProfiler GO enrichment of top biological processes for DEGs of CA1 (left) and CA3 (right) between WT and cKO mice. F-L. Spatial distribution of mRNA in P16 hippocampus of WT and cKO mice visualized in cell segmented spatial transcriptomics data for astrocytes (F, with GFAP immunostaining), activated astrocytes (G), Bdnf-TrkB signaling (H, with BDNF immunostaining), Vgf (I), synaptic plasticity (J, with CCK immunostaining), immediate early genes (K, with cFos immunostaining), and calcium regulation (L, with Calb2 immunostaining). Synpr mRNA (J) was visualized in P42 hippocampus. M. Volcano plot depicting hippocampus DEGs between P16 and P42 cKO mice. Two-sided Wilcoxon rank sum test with Bonferroni correction (D, M) and one-sided Fisher’s exact with BH correction (E) were used. Scale bar = 100 μm.

Integrating the P16, P21 and P42 spatial transcriptomics datasets together revealed that CA1 and CA3 had the largest number of DEGs, with most DEGs being upregulated in cKO mice (Figure 6D). Many DEGs in both CA1 and CA3 were associated with synaptic structure/function, synaptic plasticity and ion transport (Figure 6E). DEGs related to hormone and peptide secretion were specifically found in CA3 whereas DEGs involved in neurogenesis and dendrite development were prominent in CA1 (Figure 6E), highlighting hippocampal subregion specificity of DEG families. Very few DEGs were identified in MGE-derived interneurons, likely due to very few SST+ and PV+ cells in these mice. Notably, Calb1 was significantly increased in Sst+ interneurons (Figure 6D), consistent with our immunostaining results (Figure 1F). Vgf, which is required for activity-dependent scaling in PV+ interneurons⁶⁰, is increased in both PV+ and SST+ interneurons (Figure 6D).

Many hippocampal DEGs are associated with epilepsy and seizures. TLE with hippocampal sclerosis has the hallmark of increased astrocyte activity⁵⁹. There was an apparent increase in astrocytes in the P16 cKO hippocampus (Figure 6B–C), which was confirmed with increased GFAP mRNA and protein (Figure 6F), consistent of severe gliosis. Other gliosis-related genes such as Apod⁶¹ and Id3⁶² were significantly upregulated throughout the hippocampus of male cKO mice (Figure 6G). Apoe mediates lipid transfer from hyperactive neurons to reactive astrocytes in TLE⁶³, and Apoe was upregulated in both astrocytes and principal neurons in cKO mice (Figure 6D,G). BDNF is strongly upregulated in TLE where it can paradoxically increase neuronal excitability through interaction with Trk receptors while also producing antiepileptic effects via upregulation of NPY that can suppress seizures^64,65. In male cKO mice, Npy is upregulated in MGE-derived interneurons, Bdnf mRNA and protein are strongly upregulated in the CA3 of male cKO mice, and its receptor Ntrak2 is also upregulated throughout the hippocampus (Figures 6D,H & S14A). BDNF stimulates Vgf expression to modulate hippocampal synaptic transmission⁶⁶, and Vgf is also upregulated in CA3 pyramidal cells (Figures 6D,I & S14A).

We observed dysregulation of numerous genes involved in synaptic transmission and plasticity. Neuronal pentraxin 2 (Nptx2) and Sortilin-related VPS10 domain-containing receptor 3 (Sorcs3) are postsynaptic proteins normally restricted to CA1 where they play critical roles in synaptic plasticity^67,68. Both Nptx2 and Sorcs3 are significantly upregulated in CA3 of male cKO mice (Figures 6D,J & S14A). Synaptoporin (Synpr) is a synaptic vesicle protein expressed in DG mossy cells and CA3 cells where it modulates homeostatic plasticity^69,70, but Synpr expression was lost in the CA3 of cKO mice (Figure 6D,J). Cck is normally expressed throughout CA1-CA3 pyramidal cells where it regulates excitatory cell output⁷¹, but it is significantly downregulated CA3 of male cKO mice (Figures 6D,J & S14A). The dysregulation of genes associated with synaptic plasticity in the hippocampus likely reflect a hyperexcitable state in cKO mice, confirmed by increased cFos mRNA & protein in CA1 and CA3 (Figures 6D,K & S14A).

Hippocampal calcium plays a critical role in the generation and establishment of epilepsy. Calbindin (Calb1) and Calretinin (Calb2), two seizure sensitive calcium-binding proteins, are often dysregulated in epilepsy^72,73. Calb2 mRNA and calretinin protein are upregulated in the DG and aberrantly expressed in CA3 in cKO hippocampus, while Calb1 mRNA and calbindin protein are downregulated in the DG (Figures 6D,L & 1F). Neuronatin (Nnat) can modify dendritic calcium levels in hippocampal neurons⁷⁴, and Nnat was increased in the cKO DG and CA3 of hippocampus (Figure 6D,L & S14A). Testican-3 (Spock3), a calcium-binding proteoglycan associated in the extracellular matrix⁷⁵, was significantly upregulated in cKO CA3 (Figures 6D,L & S14A).

In addition to common DEGs throughout all 3 ages, we also identified age-specific DEGs between P16 and P42. Genes associated with seizure susceptibility (Ptprn)⁷⁶, spine and axon growth (Camk4, Olfm1)^77,78 and mitochondrial organization (mt-Co2, mt-Co3)⁷⁹ were significantly changed in the hippocampus between P16 and P42 cKO male mice (Figure 6M & S14B), suggesting dynamic changes in gene expression likely due to seizure progression and altered brain function in Arx cKO male mice over time. These data demonstrate that loss of MGE-derived hippocampal interneurons induce both broad and region-specific transcriptome changes in hippocampal principal neurons related to seizures and hippocampal sclerosis, and altered neuronal and synaptic function.

Dysregulation of energy metabolism and initiation of transcription and translation in Arx Het cortex and hippocampus

Since Arx Het female mice & humans also display phenotypes, we analyzed cell autonomous and non-cell autonomous transcriptional changes in juvenile and adult Arx Het females. Single nuclei Multiome sequencing was performed on GFP+ Nkx2.1-lineage cortical interneurons from one WT and two Het P70 female mice (Figure 7A). MGE-derived interneurons were cleanly segregated into SST+ and PV+ interneurons, which could be further divided into 8 PV+ and 10 SST+ subtypes^30,80 (Figure 7B). There were no significant differences in the cell clusters or proportional distribution of PV+ and SST+ interneuron subtypes between WT and Het female mice (Figure 7B). A small number of DEGs were identified in both PV+ and SST+ interneurons (Figure 7C), likely because surviving interneurons express WT Arx with the mutant Arx allele undergoing X-inactivation. Stmn3, which is involved in microtubule formation and dynamics⁸¹, was upregulated in both PV+ and SST+ cells in Het mice (Figure 7C). Dysregulation of Stmn3 protein was also reported in a different Arx mutant mouse line⁸², consistent with our findings.

Figure 7. Downregulation of genes associated with energy metabolism, transcription and translation in Arx Het cortex and hippocampus.

A-B. UMAP plots of integrated snRNA and snATAC data via weighted nearest neighbor (WNN) of Nkx2.1-lineage cells from P70 female WT and Het cortex, annotated by genotype (A), and PV+ and SST+ interneuron subtypes with relative proportions (B). C. Volcano plot depicting DEGs in P70 female Het PV+ (left) and SST+ (right) interneurons. D-E. UMAP plots of spatial transcripts labeled by age and genotype (D), and major cortical cell types (E). F. Spatial localization of major cortical cell types with singlet-assigned spots. G. Relative proportion of major cortical neuron types of female WT and Het mice. H. Scatter chart showing DEGs downregulated (blue) or upregulated (red) in distinct cortical cell types in P21 Het mice. I. clusterProfiler GO enrichment of top biological processes for DEGs of L2–3 IT cells between WT and Het mice. J. Spatial distribution of mRNA in P21 cortex of WT and Het mice visualized in cell segmented spatial transcriptomics data for mt-Nd2, Sub1 and Eif4a2 (top), with Eif4a2 immunostaining and quantification (bottom). n = 3 for each genotype. K. UMAP plots of P21 and P120 hippocampal spatial transcriptomics labeled by age and genotype (left) and cell types (right). L. Scatter chart showing DEGs downregulated (blue) or upregulated (red) in Het mice in distinct hippocampus cell types in P21 mice. M. clusterProfiler GO enrichment of top biological processes for DEGs of P21 CA1. N. Spatial distribution of mRNA in P21 hippocampus of WT and Het mice visualized in cell segmented spatial transcriptomics data. Two-sided Wilcoxon rank sum test with Bonferroni correction (C, H, L), one-sided Fisher’s exact with BH correction (I, M) and Welch’s 2-tailed t-test (J) were used. Scale bar = 100 μm.

We then investigated non-cell autonomous gene expression changes in P21 and P120 WT and female Het mice via spatial transcriptomics. Similar to cKO male cortices, cells from female P21 and P120 WT and Het mice co-clustered regardless of genotype (Figure 7D). We identified the same 23 cortical cell types, with no obvious changes in the proportion of any subtypes except for the expected moderate reduction of MGE-derived interneurons (Figures 7E–G & S15A). No significant difference in cortical thickness was found between WT and Het mice (Figure S15B). There were many more DEGs at P21 compared to P120, with the majority of DEGs downregulated in Het pyramidal cells at P21 whereas most DEGs were upregulated in Het pyramidal cells and astrocytes at P120 (Figures 7H & S15C). These DEGs were enriched for processes related to synapse structure or activity, calcium ion transport, neuron projection extension and dendritic spine morphogenesis (Figure 7I). Notably, mitochondrial genes (mt-Nd2, mt-Nd3) were broadly downregulated across cortical cell types at P21 (Figures 7H,J & S15A), indicating impaired energy metabolism in Het female mice. Sub1, a component of the transcriptional regulatory complex, was similarly decreased throughout the cortex of Het female mice (Figures 7H,J & S15A). Eif4a2, an ATP dependent RNA helicase whose dysfunction is associated with neurodevelopmental disorders featuring ID and seizures⁸³, was markedly reduced in Het cortex, with Eif4a2 protein most strongly decreased in superficial cortical layers (Figures 7H,J & S15A). Eif4a2 was also dysregulated in mouse model of polyalanine repeats at the Arx locus where protein translation is suppressed⁸².

As in the cortex, WT and Het hippocampal cell types showed substantial overlap between genotypes, although there was some age-based clustering in CA1 and DG principal cells (Figure 7K). CA1 and DG regions exhibited the greatest number of DEGs at both P21 and P120, with nearly all DEGs downregulated in Het females (Figures 7L & S15D). In P21 CA1, DEGs were enriched for protein transport and localization, ATP metabolism, and hormone secretion, whereas in the DG, DEGs were associated with transcription and translation, RNA splicing and lamellipodium organization (Figures 7M & S15E). Similar to the cortex, Eif4a2, Eif1 and Sub1 were downregulated in the hippocampus of Het female mice (Figures 7L,N & S15A). We did not observe significant changes in epilepsy-associated genes (BDNF, Vgf, Cck, etc.) as we did in the hippocampus of male cKO mice.

Overall, as expected there were many fewer DEGs throughout the cortex and hippocampus of Arx Het female mice compared to cKO males. Many DEGs were associated with fundamental transcriptional and translational processes, as well as mitochondrial organization and ATP synthesis, across both cortex and hippocampus.

Loss of MGE-derived interneurons alters dorsal cortical neurogenesis, cellular organization and induces band heterotopias

Dorsal cortical progenitors progress through temporal specification guided by both intrinsic genetic programs and extrinsic cues^84,85. Interneurons migrating through the cortex can influence pyramidal cell neurogenesis, radial migration and cortical lamination^86–88. The decreased cortical thickness combined with proportional changes in the number of L2–3 IT and L5 IT pyramidal neurons in cKO male mice (Figure 5D–G) suggest that loss of migrating MGE-derived interneurons could influence on dorsal progenitor neurogenesis and cell fate.

We defined nine cell types from the spatial transcriptomics data in E13.5 and E15.5 Arx cKO and WT male mice, including progenitors (radial glial cells (RGCs), intermediate progenitor cells (IPCs)), postmitotic excitatory neurons (immature deep layer cortical projection neurons (Immature DL CPNs), deep layer cortical projection neurons (DL CPNs), upper layer cortical projection neurons (UL CPNs) and migratory excitatory neurons (Migratory ExNs)) and two streams of migrating interneurons (INs_SVZ stream and INs_MZ stream) (Figures 8A & Fig. S16A–B). The vast majority of DEGs were upregulated in cortical excitatory and inhibitory cells of E13.5 cKO male mice but were downregulated at E15.5 (Figures 8B & S16C). As expected, many markers of postmitotic interneurons (Sst, Maf, Lhx6, Nxph1, etc.) were significantly reduced in the cortex of E13.5 and/or E15.5 cKO male mice, while Nkx2.1 and Ebf3 were increased in cortical interneurons in cKO mice (Figure 8B–C & S16C).

Figure 8. Loss of MGE-derived interneurons alters dorsal cortical neurogenesis and cell-cell signaling leading to cortical heterotopias.

A. Spatial localization of E13.5 and E15.5 WT and cKO cell types from cell segmented spatial transcriptomics data. B. Scatter chart showing DEGs downregulated (blue) or upregulated (red) in cKO mice in distinct E15.5 cell types. C. Spatial distribution of mRNA in E15.5 cortex of WT and cKO mice visualized in cell segmented spatial transcriptomics data. D-E: Immunostaining (D) and quantification (E) of Eomes+ IPCs in E15.5 lateral cortex of WT and cKO mice. n = 3 brains/genotype, n = 9 slices/genotype. F. E15.5 cKO brain section stained for Pax6, Eomes and Reln and DAPI highlighting cortical layering abnormality. G. Spatial distribution of Reln and Lhx6 mRNA in E15.5 cortex of WT and Het female mice visualized in cell segmented spatial transcriptomics data (top), and immunostained for Reln and tdTomato (bottom). H. Spatial localization of cortical cell types and area used for CellChat analysis from E15.5 WT and cKO male mice, color coding same as A. I. CellChat-derived heatmap depicting differential interaction strength between cell types from inferred cell-cell communications in E15.5 WT and cKO cortex. J-M. Chord diagram of EphA (J), Sema4 (K), collagen (L), and Adgrl (M) signaling networks between different cell types from inferred communication at the individual cell level in WT and cKO cortices, with spatial visualization of signaling score for EphA (J) and Sema4 (K). Two-sided Wilcoxon rank sum test with Bonferroni correction (B) and ordinary two-way ANOVA followed by Sidak’s multiple comparisons test (D) were used: ** = p ≤ .01, **** = p ≤ .0001. Scale bar = 100 μm.

At E13.5, several postmitotic excitatory neurons markers (Neurod2, Neurod6, Bcl11b, Tbr1) were significantly increased in cKO cortex (Figure S16C), suggesting premature cell cycle exit. Consistent with this, genes associated with cycling progenitor cells and neurogenesis (Pax6, Eomes, Mki67, Sox1) were significantly downregulated in the E15.5 cortex of cKO male mice (Figures 8B & S16C), and we confirmed a significant decrease of Eomes+ cells in the E15.5 cKO cortex (Figure 8D–E). Thus, loss of migratory MGE-derived interneurons in the dorsal cortex alters neurogenesis of excitatory projection neurons.

Patients with ARX variants can display lissencephaly and other cortical malformations^5,89,90. We observed abnormal cortical structures resembling band heterotopias in both cKO male (Figure 8F) and het female mice (Figure 8G), and instances of oddly shaped pointed dorsal cortices in other brains (Figure S16D). This abnormal cortical formation appears to compress and disrupt the normal organization of Pax6+ and Eomes+ dorsal cortical progenitors (Figure 8F). Our spatial transcriptomics data revealed an ectopic layer of Reln+ Cajal-Retzius (CR) cells within this band heterotopia that is surrounded by Tom+ MGE-derived interneurons (Figure 8G). CR cells are required for normal inside-outside cortical patterning, and Reln is also expressed by interneurons where it plays a role in layer-specific distribution of cortical neurons⁸⁷. Notably, Reln is upregulated in the IN_MZ stream in cKO mice (Figure 8B). These results demonstrate that loss of MGE-derived interneurons is sufficient to disrupt normal CR cell organization, likely leading to severe perturbation of dorsal cortical neurogenesis.

The non-cell-autonomous effects of Arx deletion in MGE-derived interneurons is likely due to dysregulated cell-cell communications. We inferred and analyzed cortical cell–cell communication networks from E15 spatial transcriptomic data using CellChat⁹¹. We performed CellChat analysis on ~5000 dorsal cortical cells in E15.5 WT and cKO mice (Fig. 8H). The total number of inferred cell-cell communication networks were increased in cKO cortex, but the interaction strength was decreased (Figures 8I & S16E–F). There was a strong increase in both the number and strength of interactions between progenitor cells (RGCs and IPCs) with nearly all other cell types (Figures 8I & S16F). Conversely, interneurons in the MZ and CR cells had a strong reduction in interaction numbers and strength, with the strongest observed decrease in CR-CR communication (Figures 8I & S16F). This is consistent with our immunostaining results showing significant changes in these progenitor and CR cells.

The most altered signaling network in cKO mice involved two guidance factor families (EphA receptors, Sema4 ligands) and two families involved in cell adhesion (collagen and members of the adhesion GPCR L subfamily, ADGRL) (Figure 8J–M). EphA receptors and their ephrin ligands play critical role in neural proliferation, cell migration, synaptogenesis, and topographical patterning of the cortex⁹². EphA-based interactions between UL CPNs, CR and INs_SVZ stream were lost in cKO mice (Figure 8J). Class 4 semaphorins mediate receptor clustering in neuronal postsynaptic membranes, and deficiencies in Sema4 signaling is associated with defects in cell proliferation, exencephaly and neonatal lethality⁹³. Sema4 interactions were completely lost between most cell types, but there is a predicted new sema4-based signaling network between UL CPNs and INs_SVZ cells in cKO mice (Figure 8K).

Collagen signaling is involved in guiding neuronal outgrowth, maturation and synaptogenesis⁹⁴, and ADGRL signaling controls cell–cell and cell–matrix interactions during neural development⁹⁵. Mutations in the adhesion GPCR gene Adgrg1 (GPR56) and its ligand collagen type III alpha-1 (Col3a1) results in a malformed cerebral cortex that resembles cobblestone lissencephaly^96,97, similar to what we see in Arx mutant mice. The collagen signaling network predominantly involves RGCs, IPCs and CR cells in WT mice, but collagen signaling is strongly downregulated in CR cells and upregulated in postmitotic cortical neurons in cKO mice (Figure 8L). This is consistent with numerous collagen genes (Col3a1, Col4a1, Col4a2, Col1a2) being downregulated specifically in CR cells in cKO mice (Figure 8B). ADGRL signaling was lost in all cell types except for CPNs in cKO mice (Figure 8M). Thus, reduction of MGE-derived interneurons in Arx mutant mice disrupts numerous signaling families causing altered cell-cell communication and cellular disorganization of CR cells and dorsal cortical progenitors, leading to profound defects in cortical neurogenesis.

DISCUSSION

ARX variants are a leading cause of XLID, a complex set of disorders with a high comorbidity of ID and epilepsy that predominantly affects males, but female carriers also display a variety of phenotypes. In this study, we demonstrate how removal of Arx in the MGE leads to a near complete loss of forebrain MGE-derived interneurons and early lethality in cKO male mice, and reduced interneurons with increased seizure susceptibility and altered behavior in Het female mice. We constructed a transcriptomic atlas spanning embryogenesis through adulthood using single nuclei sequencing and spatial transcriptomics. These data revealed significant transcriptomic changes in the MGE leading to an Arx mutant-specific cell cluster that is consistent with a recent study in human forebrain progenitor primary cultures²⁶. Most notably, the loss/reduction of MGE-derived interneurons in the embryonic cortex had a profound non-cell autonomous effect on gene expression, neurogenesis, cellular organization and cell-cell communication in dorsal cortical progenitors, ultimately leading to cortical abnormalities mimicking band heterotopias in both cKO male and Het female mice. These findings highlight that loss of Arx in the MGE is sufficient to induce both cell autonomous and non-cell autonomous transcriptomic changes that may explain the variation of symptoms and severity in patients ARX variants and XLID phenotypes.

Reduction of cortical MGE-derived interneurons was consistent across subtypes and closely followed Arx dosage, with a near complete loss of PV+ and SST+ cells in cKO male mice and ~50% reduction in Het females, although there was some variability in subtype loss between different hippocampal regions. cKO male mice displayed infantile spasms and seizures, similar to removing Arx in all ganglionic eminences¹⁷, which likely led to most cKO males dying before weaning. Most Het female mice did not exhibit spontaneous seizure, and while Het females did show a significant increase in seizure susceptibility overall, ~50% of Het mice had similar response to WT females. Het female mice also displayed significant but variable deficits in locomotor activity, anxiety behaviors and sensorimotor gating. These results are consistent with human studies where female carriers have a large range of mild to severe phenotypes⁹, even in monozygotic female twins with ARX variants⁸⁹, implying mosaic XCI could in part lead to different phenotype severity. This study did not correlate XCI with behavior or cellular phenotypes, but Het female mice have a normal paternal X-chromosome and a mutant maternal X-chromosome based on our mating strategy.

Reduced MGE-derived interneurons are consistent with our embryonic transcriptome data where we found a complete loss of normal Sst+ and PV-fated Maf+ cell clusters in cKO male mice and an ~50% reduction in Het female mice. These clusters were replaced with ectopic postmitotic cell clusters in cKO mice enriched for Lim homeodomain transcription factors (Lmo1, Lmo3, Lmo4) and dysregulation of genes associated cell migration (Cxcr4, Cxcr7, Nrg1); this population likely represents the bolus of cells whose dorsal migration was halted in the ventral telencephalon. Sst was also enriched in these ectopic clusters, though its role in these cells is unclear. Lhx6 expression was maintained in these ectopic postmitotic cell clusters whereas Lhx8, which promotes forebrain cholinergic neuron fates³⁴, was strongly downregulated. Lhx6 directly regulates Arx in the MGE, and Arx expression can partially rescue interneuron maturation in Lhx6^−/− mice⁹⁸, indicating Lhx6 acts upstream of Arx. In Lhx6^−/−;Lhx8^−/− double KO mice, Lmo3 and Nkx2.1 are both downregulated³⁵, which is the opposite effect in Arx mutants. Shh expression in the MGE mantle is reduced in these double KO mice leading to abnormal progenitor differentiation³⁵; Shh was also downregulated in Arx cKO mice. These findings highlight the intricate relationship between Arx, Nkx2.1, Lhx6, Lhx8 and Shh in regulating fate & maturation of MGE-derived neurons.

Loss of MGE-derived interneurons in Arx mutant mice led to striking non-cell autonomous cellular and transcriptional changes in the cortex and hippocampus. Neuroprotective genes were upregulated whereas immediate early genes (Arc, cFos, Egr1) were downregulated in pyramidal cells across all cortical layers in cKO male mice despite their increased seizure activity. This agrees with studies showing chronic network hyperactivity leads to long-term reduced expression of activity-dependent genes^99–102. Many DEGs were associated dendrite development, cell growth and synaptogenesis, which is consistent with sustained hyperactivity leading to more immature neuronal states that may be a shared phenotype by many neuropsychiatric disorders^103,104.

The hippocampus of cKO male mice displayed gliosis and other phenotypes indicative of TLE. Bdnf-Trk signaling is often upregulated in TLE⁶⁴, especially in CA3 where BDNF may act as a central hub for network processing⁶⁵. Bdnf mRNA and protein are both strongly upregulated in the CA3 of cKO mice. In general, the most notable changes in gene expression levels were observed in CA3 compared to other regions. Different cohorts of DEGs were identified between CA1 and CA3, with many CA3 DEGs associated with peptide and hormonal function that were not found in CA1 DEGs. Curiously, cfos protein was strongly upregulated in CA1 and CA3 pyramidal cells in the hippocampus, in contrast to the clear reduction of cfos in the cortex. Thus, loss of forebrain MGE-derived interneurons produces distinct genetic and cellular effects between the cortex and hippocampus, and between different regions of the hippocampus.

As expected, juvenile and adult Het female mice displayed fewer transcriptomic changes in interneurons and excitatory cells compared to cKO males. There was a strong bias for DEGs to be downregulated in cortical and hippocampal excitatory cells in Het female mice, while the distribution was more even or biased toward upregulation in cKO males. This could be due to decreased expression of transcription-associated (Sub1) and translation-associated (Eif4a2, Eif) in both cortical and hippocampal excitatory cells. In contrast to our data, Eif4a2 was significantly upregulated in the cortex (but not hippocampus) in male Arx^(GCG)7/Y mice, a model of the elongated polyalanine ARX variant⁸². This highlights the need to explore transcriptional changes in different cell types and different Arx disease models to fully understand the broad range of phenotypes observed in both males and females.

Lissencephaly is a spectrum of severe brain malformations where normal brain folding fails to occur properly, with ARX variants being one of the leading genetic causes of this disease^90,105. A related phenotype called band heterotopias (or double cortex) occurs when there is an ectopic band of neurons within or below the normal cortex, most commonly arising in females with DCX or LIS1 variants¹⁰⁶. We observe cortical malformations that resemble neocortical heterotopia and subcortical band heterotopia in both cKO male and Het female brains. In both cases we find an ectopic band of Reln+ CR cells, and in a E15.5 Het female brain, these CR cells are encircled by Tom+ MGE-derived interneurons. CellChat analysis predicts a strong downregulation of interaction strength for CR-CR and CR-IN_MZ communications, of which the collagen signaling network may play a prominent role. Thus, decreasing Arx expression in the MGE, and the corresponding loss of MGE-derived interneurons migrating into the cortex, is sufficient to induce cellular rearrangements, abnormal cell-cell signaling, reduced IPCs and ultimately cortical malformations. These findings demonstrate the critical role that migratory MGE-derived interneurons play in regulating dorsal cortical neurogenesis and highlight potential disease mechanisms in ARX-related neurodevelopmental disorders.

METHODS

Animals

All experimental procedures were conducted in accordance with the National Institutes of Health guidelines and were approved by the NICHD Animal Care and Use Committee (protocol #23–047 & 26–047). The following mouse lines were used in this study: Arx-floxed (gift from Drs. Eric Marsh & Jeffrey Golden)³⁸; Nkx2.1-Cre (Jax# 008661)²⁷, Ai9 (Jax# 007909)¹⁰⁷, Sun1-sfGFP (Jax# 030952)¹⁰⁸. Mice were housed under standard conditions (12h light and 12h dark). The morning on which the vaginal plug was observed was denoted E0.5.

Brain harvesting

Embryonic brain dissections: Pregnant dams were anesthetized with an i.p. injection of Euthasol (270 mg/kg, 50 μL injection per 30 g mouse), E13.5 embryos were removed and placed in ice-cold carbogenated artificial cerebral spinal fluid (ACSF, in mM: 87 NaCl, 26 NaHCO₃, 2.5 KCl, 1.25 NaH₂PO₄, 0.5 CaCl₂, 7 MgCl₂, 10 glucose, 75 sucrose, saturated with 95% O₂, 5% CO₂, pH 7.4). MGE was dissected from individual brains in ice-cold ACSF, transferred to individual 0.5 mL Eppendorf tubes, then flash frozen on dry ice and stored at −80°C. For analyzing fixed tissue, E13.5 and E15.5 brains were removed and drop-fixed in 4% paraformaldehyde (PFA) in PBS overnight, washed in PBS, incubated in 30% sucrose in PBS at 4°C overnight, embedded in OCT and stored at −80°C.

Postnatal and adult brain dissections: P0-P120 mice were anesthetized with Euthasol. For harvesting brain regions, the somatosensory and motor cortex of P70 mice were dissected, transferred to 1.5 mL Eppendorf tubes, then flash frozen on dry ice and stored at −80°C. For analyzing fixed brains, P0-P120 mice were perfused with PBS followed by 4% PFA, brains were removed and post-fixed in 4% PFA overnight, washed in PBS, incubated in 30% sucrose in PBS at 4°C overnight, then embedded in OCT and stored at −80°C.

Tissue staining and microscopy

E13.5 embryonic brains were cryosectioned at 14μm, mounted on slides, dried and stored at −20°C or −80°C. Slides were incubated in block solution (10% Normal Donkey Serum in PBS + 0.3% Triton X-100) for 1 hour RT, then incubated overnight with primary antibodies in block solution at 4°C and secondary antibodies for 2 hours RT. After washing, brain sections were mounted.

P0-P120 brains were cryosectioned at 30 μm, sections transferred to 96-well plates containing antifreeze solution (30% ethylene glycol, 30% glycerol, 40% PBS) and stored at −20°C. Prior to staining, brain sections were washed in PBS (3 × 10 minutes). Brain sections were incubated in block for 1 hour RT, then incubated for 24–48 hour in primary antibodies in block solution at 4°C. Then sections were washed in PBS (3 × 10 minutes) and incubated with secondary antibodies and DAPI for 2 hours at RT or overnight at 4°C. Brain sections were washed and mounted.

The following primary antibodies were used in this study: rat-anti SST (1:300, Millipore MAB354), goat-anti PV (1:1000, Swant PVG213), rabbit-anti nNos (1:500, Millipore MAB5380), rat-anti EOMES (1:500, Invitrogen 14–4875-82), rabbit-anti CUX1 (1:500, Proteintech 11733–1-AP), rabbit-anti c-Fos (1:500, Synaptic Systems 226008), rabbit-anti Pax6 (1:500, BioLegend 901301), mouse-anti Reelin (1:500, Sigma-Aldrich MAB5364), rabbit-anti BDNF (1:500, Invitrogen PA5–85730), rabbit-anti proCCK (1:1000, Frontier Institute MSFR105040), rabbit-anti Calb1 (1:200, Invitrogen 711443), mouse-anti Calb2 (Calretinin) (1:1000, Millipore MAB1568), mouse-anti GFAP (1:500, Sigma-Aldrich G3893), rabbit-anti Olig2 (1:500, Sigma-Aldrich AB9610), rabbit-anti eIF4A2 (1:500, Invitrogen PA5–79195), mouse-anti SYT2 (1:1000, Abcam ab154035), chick-anti VGAT (1:500, Synaptic Systems 131006), sheep-anti ARX (1:100, R&D systems AF7068). Species-specific fluorescent secondary antibodies (1:500) were used conjugated to AlexaFluor 488, 594, 647 and 790.

All images were captured on an Olympus VS200 scanner (VS200 ASW). Post-processing was performed with Adobe Photoshop and/or ImageJ2 (v2.9.0).

HiPlex12 RNAscope

Brains were cryosectioned at 14 μm. RNAscope HiPlex12 reagents kit (Advanced Cell Diagnostics, #324409) were used according to the manufacturer’s instructions. All RNAscope HiPlex probes and reagents from Advanced Cell Diagnostics. The following probe sets were used in this study: Lmo1 (511211-T1), Lmo3 (497631-T2), Lmo4 (485491-T3), Lhx6 (422791-T4), Lhx8 (515101-T5), Ebf3 (576871-T6), Lmx1b (412931-T7), Nkx2-1 (434721-T8), Sst (404631-T9), Maf (412951-T10), Shh (314361-T11) and Nrg1 (418181-T12).

Cell counting and analysis

All cell counts on Nkx2.1-Cre;Arx;Ai9 mice were performed manually using Photoshop and were blind to genotypes. Cortex: Cells were counted from 3–4 non-consecutive brain sections through the somatosensory cortex for each P21 brain. DAPI staining was used to divide the cortex into superficial (I-III) and deep (IV-VI) layers. Tom+, PV+ and SST+ cells were counted, and the density was calculated by cells/mm². Any PV+ and SST+ cells that were Tom− were excluded from the counts. Hippocampus: Cells were counted from 4–6 non-consecutive sections spanning the anterior-to-middle P21 dorsal hippocampus. Tom+, PV+, SST+ and nNos+ cells counted, and the density was calculated by cell numbers/mm². Small Tom+ cell bodies in CA2/3 that are primarily Olig2+ oligodendrocytes were excluded from these interneuron counts and counted as separate group. The number of Tom+/Calb1+ or Tom+/Calb1+/SST+ were counted from 4 sections through whole hippocampus, and 3 male WT and cKO brain were used. The thickness of the primary somatosensory cortex (S1) from P21 male mice (bregma: from 0.97 to 0.49 mm) and female mice (bregma: from −1.55 to −1.91 mm) were measured. Cux1 immunostaining was used to label layers II-IV and 3 mice per genotype, 3–4 sections measured per brain. cFos+ cells were counted from 2–4 sections through P21 the somatosensory cortex of male WT and cKO mice. The relative fluorescent intensity of CCK in layer 2–3 of somatosensory cortex was measured in three P21 male WT and cKO mice with 4 sections per mice. The E15.5 lateral cortex of male WT and cKO was divided into 12 bin and the quantification of EOMES+ cells were calculated from 3 sections for each brain.

Pentylenetetrazole (PTZ) injection and recording

PTZ (Sigma-Aldrich, P6500) was prepared on the day of use with in sterile 0.9% (w/v) NaCl. PTZ injections and recording were performed on 4-month-old female mice. All mice receive an IP injection of PTZ at 35 mg/kg and were recorded immediately via video tracking system ANY-maze (Stoelting Company) for 60 min. Epileptic behaviors were scored using a modified Racine scoring system¹⁰⁹: stage 0 = normal behavior, stage 1 = immobility and rigidity; stage 2 = head bobbing, facial, forelimb, or hindlimb myoclonus; stage 3 = continuous whole-body myoclonus, myoclonic jerks or tail held up stiffly; stage 4 = tonic seizure, continuous rearing and falling, stage 5 = clonic-tonic seizure, stage 6 = death. Total seizure scores were calculated at every 5-min block in the first 20 min. N (WT) = 9 and N (Het) = 11.

Behavioral tests

All behavioral tests were performed on WT, Het and Hom littermates according to approved NICHD and/or NIMH standard operating procedures (SOPs) and cleaning protocols. Animal behavioral tests were carried out during the daylight cycle in NICHD animal behavioral room or at the NIMH Rodent Behavior Core. All mice were habituated to the testing environment for 30–60 minutes before behavior tests. All behavior assays were performed blind to genotype, and 1–4 independent cohorts were used for each behavioral test.

Home-cage test

Individually housed juvenile mice were observed in their home cages placed in the Photobeam Activity System-Home Cage (PAS-HC, SD Instruments) to assess activity (ambulation, fine movements and rearing) that can be quantified by beam breaks. Mice were kept in the cage for 4 days with appropriate food and water. During this testing, mice were housed in a ‘quiet’ room to minimize outside interactions. Mice were returned to group housing after experiment. The total ambulatory movements, fine movements, rearing and general locomotor activity were analyzed and processed using the PAS-HC data collection transferred to Excel. N (WT) = 10 and N (Het) = 13.

Open-field test

Juvenile mice were originally placed in the center of open field apparatus (51 cm × 51 cm × 30 cm Plexiglas box) and recorded for 20 minutes. The time spent in the center zone of open field areas were recorded and analyzed with the video tracking system ANY-maze (Stoelting Company). N (WT) = 10 and N (Het) = 12.

Light-dark box test

The apparatus (40 cm × 40 cm × 30 cm) consists of dark chamber (40 cm × 15 cm × 15 cm) and light chamber compartment. There is a door between two chambers which allow mice to freely move between chambers. Bright white lights illuminated the box from above as mice were placed in the box for 10 minutes. For analysis, the session was divided into two 5-min bins (0–5 min and 6–10 min), and data from the second bin (6–10 min) were used for statistical comparisons between WT and Het to reduce the variability caused by acute novelty stress. The test was recorded by ANY-maze video trace camera to determine the amount of time each mouse spent in the light chamber. N (WT) = 12 and N (Het) = 12.

Prepulse inhibition (PPI) of the acoustic startle reflex

Startle response and pre-pulse inhibition were performed via SR-LAB-Startle Response System (San Diego Instruments). Adult mice are placed in a startle apparatus with sound-attenuated chamber. Whole-body startle movements were detected by a piezoelectric accelerometer mounted beneath the cage, converted to electrical signals, and digitized and stored by a computer. A loudspeaker mounted to the side of the cage produced white noise acoustic stimuli. A continuous 64 dB background noise was present throughout the test. The test began with a 5-minute acclimation to the apparatus, followed by 12 pulse-alone trials. To measure PPI, 12 blocks containing 6 trials were presented: 1 pulse-alone trial, 4 prepulse-pulse trials, and 1 no-stimulus trial. Pulse-alone trials consisted of a 40 ms startle pulse at 120 dB; prepulse-pulse trials consisted of a 20 ms prepulse at 66, 68, 72, or 76 dB followed 100 ms later by the pulse; and no-stimulus trials consisted of background noise only. Trials will be presented in pseudorandom order, with an inter-trial interval of 10–20 s (mean of 15 s). Startle response will be determined as the peak amplitude within 100 ms of the startle pulse onset. The entire test lasted 30 minutes. N (WT) = 6 and N (Het) = 9.

2 Object Novel object recognition test

Novel object recognition test was conducted in an open field arena (40 cm × 40 cm white Plexiglas box). The protocol used a 3-day paradigm that includes habituation, training, and testing phase. During habituation (day 1), adult mice were placed into the center of an open field arena and allowed to explore for 10 minutes. For training (day 2), two identical objects were placed on either side of the center of the arena, and the mice explored the arena for 10 minutes. For testing (day 3), a familiar object and a novel object (dissimilar object) were placed in the same position as in the training day and mice explored the arena for 10 minutes. ANY-maze was used to record each trial. The amount of time each mouse spent with their nose oriented toward the object within 2.5 cm of the object edge was considered ‘investigating time’. The discrimination index (DI) was calculated as: DI = (time investigating novel − time investigating familiar)/(time investigating novel + time investigating familiar). N (WT) = 12 and N (Het) = 12.

Free-walk test

FreeWalkScan 2.0 (CleverSys Inc.) was used to assess mice gait. Adult mice move freely in a 40 cm × 40 cm × 30 cm (length × width × height) chamber. A high-speed camera below a clear bottom plate captures mouse movement for 5 minutes with red light in dark room. Videos are analyzed using FreewalkScanTM 2.0 software for various characteristic of gait, including front base (distance between sequential footprints of front limbs), stride length (distance between two sequential footprints of same paw). N (WT) = 5 and N (Het) = 8.

Multiome sequencing and analysis

Pregnant dams and P70 mice were terminally anesthetized with an i.p. injection of Euthasol (270 mg/kg, 50 μl injection per 30 g mouse). Tissue from male and female mice were processed separately for E13.5 and P70 samples. Frozen E13.5 and P70 tissue from Nkx2.1-Cre;Arx;sun1-GFP mice were mechanically lysed, washed, filtered and sorted to generate single nuclei suspensions for the 10x Genomics Multiome assay as previously described^30,110. We combined tissue from 2 E13.5 embryo for each Multiome reaction. Nuclei were reconstituted in 1x Nuclei buffer at a density of 3,000 nuclei/μl. 5 μl of this solution was then used for the 10x Genomics Single Cell Multiome Assay per manufacturer’s protocol. Our single cell Multiome workflow ‘multiome-wf’ can be found at github (https://github.com/NICHD-BSPC/multiome-wf) with documentation located here (https://nichd-bspc.github.io/multiome-wf/).

Sequencing analysis: Gene expression and open chromatin state were profiled using 10X Chromium Single Cell Multiome ATAC + Gene Expression kit (10X Genomics, 1000285). Joint libraries were simultaneously created by following standard protocol. Sequencing was conducted with paired-end (50 × 50 bp) using an Illumina HiSeq 2500 or NovaSeq 6000. The raw sequencing data were processed using Cell Ranger ARC (v2.0) pipeline. The cellranger-arc mkfastq command was used to generate the demultiplexed FASTQ files from BCL files. The sequencing reads were aligned to custom built mouse (GRCm38/mm10) reference genome using cellranger-arc count. The cellranger-arc count command was used to generate gene-by- barcode (snRNA-seq) and peak-by-barcode (snATAC-seq) matrices. The cellranger-arc aggr command, without depth normalization (--normalize = none) was used to aggregate sample datasets into a single feature-barcode matrix file.

Embryonic Multiome-seq data analysis: For snRNA-seq data analysis, the aggregated feature-barcode matrix was used as input to Seurat (v5.2.1)¹¹¹ in R (v4.4.2, https://cran.r-project.org). Low-quality cells were firstly removed based on the following QC metrics: unique molecular identifiers (UMIs) and mitochondrial counts. Upper outliers using standard deviation ‘sd’ and remove cells automatically, while lower outliers kept cells with UMI counts more than 100 and mitochondrial counts more than sd. The standard Seurat workflow with default parameters was performed unless otherwise noted: NormalizeData (LogNormalize), FindVariableFeatures (nfeatures = 3000), ScaleData, RunPCA, FindNeighbors, FindClusters and RunUMAP (Dims = 1:30) steps. For snATAC-seq data analysis, the aggregated peak-by-barcode matrix was used as input to Signac (v1.14.0)¹¹² in R. Low-quality cells were removed based on the following QC metrics: number of chromatin accessibility peaks and transcription start site enrichment score. Upper outliers using standard deviation ‘sd’ and remove cells automatically, while lower outliers kept cells with more than 1000 chromatin accessibility peaks and the transcription start site enrichment score greater than 2. After filtering out the low-quality cells, we proceeded with normalization, identified highly variable peaks and reduced dimensions using the function of RunTFIDF, FindTopFeatures (min.cutoff = “q0”) and RunSVD in Signac with default parameters. The regular non-linear dimension reduction and clustering were performed by RunUMAP (dims = 2:30), FindNeighbors (dims = 2:30) and FindClusters in Seurat. The gene activity scores were calculated using the GeneActivity function and added to Seurat object. Multimodal analysis: The shared cells across the two modalities (snRNA-seq and snATAC-seq datasets) were merged, and the RNA and ATAC dimensionality reduction was recomputed using default parameters. Multimodal data integration was performed by finding the Weighted Nearest Neighbor (WNN)¹¹³ using Seurat::FindMultiModalNeighbors. The dimensionality reduction was separately set to 1:30 (snRNA assay) and 2:30 (snATAC assay). The joint UMAP and clustering was performed using WNN graph with default parameters. Cell types of E13.5 MGE were identified with FindAllMarkers function within clusters and integrated with mature MGE cell type biomarkers. Notably, cells didn’t belong to MGE lineage were also removed for further analysis.

Adult Multiome-seq data analysis: For snRNA-seq data analysis, the aggregated feature-barcode matrix was used as input to Seurat (v5.2.1)¹¹¹ in R (v4.4.2, https://cran.r-project.org). Cells with percent of ribosomal counts less than 2, percent of mitochondrial counts less than 1, RNA feature counts between 1000 and 7500, and UMIs counts less than 40000 were kept. The Seurat workflow was performed as previous described except with FindVariableFeatures (nfeatures = 2000). For snATAC-seq data analysis, the aggregated peak-by-barcode matrix was used as input to Signac (v1.14.0)¹¹² in R. Cells with more than 1000 chromatin accessibility peaks, the transcription start site enrichment score greater than 2 and nucleosome signal less than 4 were kept. Multimodal analysis was performed as pervious described except with FindClusters (resolution = 1.2). Joint clusters with few cell numbers were removed.

Annotation analysis and merge datasets: All cell types were defined using the RNA expression only. Cell type annotation was based on clusters marker using FindAllMarkers function and public articles. Filtered RNA assay of Embryonic (E13.5) and adult (P70) datasets were merged and run default Seurat workflow, while filtered ATAC assay of both datasets were also merged and run previous described steps.

Pseudotime analysis: Single cell trajectory analysis of E13.5 snMultiome-seq datasets was performed using Monocle 3 (v1.3.7)¹¹⁴ (https://cole-trapnell-lab.github.io/monocle3/docs/trajectories/). Seurat objects were converted to a CellDataSet object using as.cell_data_set function and RNA counts were extracted. The reduce dimensionality of CellDataSet object was set to WNN.UMAP of Seurat object. Learning trajectory graph using learn_graph function. The root of trajectory was set to Nes+ clusters and order cells using order_cells function with default parameters. Plotting trajectory colored by pseudotime with plot_cells function.

Visualization: Uniform Manifold Approximation and Projection (UMAP) coordinates and WNN clustering, computed by Seurat on multimodal integrated datasets, were visualized using DimPlot function. The expression of genes of interest was visualized using FeaturePlot, VlnPlot or DoHeatmap function. Packages ggplot2 (v3.5.1) and patchwork (v1.3.0) were used to visualize during Multiome datasets analysis process.

Visium HD Spatial Gene Expression

Sample preparation and spatial transcriptomic sequencing: All brain tissues used in Visium HD were cryosectioned at 12 μm, which were taken from Fixed Frozen Tissue (FFT) blocks following the Visium HD FFT Preparation Handbook (CG000764) except one batch (one P120 female WT and one Het) that was taken from formalin fixed & paraffin embedded (FFPE) block following the Visium HD FFPE Preparation Handbook (CG000684). Hematoxylin (Sigma-Aldrich, GHS216) and Alcoholic Eosin (Sigma-Aldrich, HT110132) (H&E) staining and imaging were performed using preparation Handbook, then processed and sequenced following the Visium HD Spatial Gene Expression Reagent Kits User Guide (CG000685) with Visium HD Slides (1000670), Visium Mouse Transcriptome Probes v2 (1000667), Visium HD Reagents (1000668) and Visium HD Cassettes (1000669).

Space Ranger pipeline: Loupe Browser (v9.0.0) was used to align the CytAssist and H&E images, and Space Ranger (v 4.0) was used for analysis. Sequences were aligned with mm10 reference genome to gene expression counts using Space Ranger default setting. Space Ranger outputs Visium HD data binned to 2 μm, 8 μm and 16 μm resolution and cell segmented output. Unless otherwise specified, downstream analyses for all samples were performed on the 8 μm resolution datasets, which were recommended by 10x Genomics due to enriched UMI reads and closed single-cell scale compared to other resolution. Downstream analyses for E13.5 brain samples and E15.5 cortex were performed using cell segmented datasets to allow transcripts to be assigned at the cell level. Key metrics of 8 μm resolution (Number of binned squares under tissue, mean reads, mean UMIs and total genes detected) and cell segmentation metrics (number of cells, mean reads per cell and median genes per cell) are listed (Supplementary Table 1).

Spatial transcriptomic data processing: 8 μm resolution Visium HD spatial datasets were analyzed with Seurat (v5.4.0) following standard Visium HD workflow (https://satijalab.org/seurat/articles/visiumhd_analysis_vignette). Brain dataset was loaded with Load10X_Spatial function with bin.size = 8. Unless otherwise specified, LogNormalize, FindVariableFeatures and ScaleData were performed for “Spatial.008um” assay with default parameters. The sketch-based analysis approach was adopted with cells = 50,000 and perform clustering workflow, then sketch assay was projected to “Spatial.008um” assay. To identifying and segmenting spatial domain, Building Aggregates with a Neighborhood Kernel and Spatial Yardstick (BANKSY) was performed with default parameters. After dimensional reduction and clustering, we subset out cortex and hippocampus based on BANKSY clusters for further downstream analysis. To accurately annotate spatial data with scRNA-seq reference, Robust Cell Type Decomposition method was introduced. Here, cortex or hippocampus were separately integrated with ALLEN scRNA-seq dataset using RCTD with default workflows. Output from RCTD contained the weights for cell types in the reference in each spot and the class (‘singlet’, ‘doublet_certain’, ‘doublet_uncertain’ and ‘reject’) of each bin. Only the ‘singlet’ (one cell type) spots and a minute UMI threshold of 100 for a spot were used for further analysis. Upper limitation of UMI depended on datasets. When needed, cell segmentations Visium HD spatial datasets were analyzed with Seurat (v5.4.0) following standard workflow (https://satijalab.org/seurat/articles/visiumhd_analysis_cell_segmentations). Brain datasets were loaded with Load10X_Spatial with bin.size = segmented_outputs. Spatial transcripts of Visium HD datasets were visualized by using Seurat and Loupe Browser.

Differential analysis

Multiome-seq datasets: Differentially expressed genes between genotypes were computed using the Wilcoxon rank sum test implemented in the Seurat FindMarkers function with Bonferroni correction (significance: log₂FC > ± 0.2, adjusted P-value < 10⁻⁶) for the E13.5 MGE and P70 cortex datasets. Differentially expressed peaks between different genotypes were computed using the Wilcoxon rank sum test implemented in the Seurat FindMarkers function with Bonferroni correction (significance: log₂FC > ± 0.2, adjusted P-value < 10⁻⁶). The EnhancedVolcano function from the EnhancedVolcano package (v1.24.0, https://github.com/kevinblighe/EnhancedVolcano) was used to visualize the DEGs between different conditions per population and parameters.

Spatial transcriptomic datasets: Unless otherwise specified, differentially expressed genes between genotypes were computed using the Wilcoxon rank sum test implemented in the Seurat FindMarkers function with Bonferroni correction (significance: log₂FC > ± 0.2, adjusted P-value < 0.05). When compared E15.5 cortex DEGs, parameters were set to adjusted P-value < 10⁻⁵⁰ and log₂FC > ± 0.1 for significant determination. When compared E15.5 subpallium DEGs, parameters were set to adjusted P-value < 10⁻⁶ and log₂FC > ± 0.2 for significant determination.

Cell-cell communication analysis

Cell-cell communication was analyzed in E15.5 male cortex using CellChat⁹¹ version 3 (Spatial CellChat) following tutorial (SpatialCellChat analysis of spatial transcriptomics data). E15.5 segment Visium HD datasets of cortex from WT and cKO were input and only part of dorsal cortex (around 5, 000 cells) was kept and used to create individual spatial CellChat object. All CellChatDB database except for “Non-protein Signaling” was used for cell-cell communication analysis. The communication probability was computed with 250 μm interaction range and 10 μm contact range. Cell-cell communication at signaling pathway level and network centrality scores were computed with default setting. Spatial CellChat objects from WT and cKO were integrated and comparison analysis was performed following tutorial (Comparison analysis of multiple datasets using CellChat).

Enrichment analysis

Gene ontology pathway analysis was performed on each group’s (e.g., genotype and sex) DEGs with adjusted p values <0.05 using the enrichGO function from the clusterProfiler package (v4.14.0)¹¹⁵. The parameters were set as following: OrgDb = org.Mm.eg.db (v3.20.0), keyType = “SYMBOL”, ont = “BP”, universe = “All detected genes in specific population”. ClusterProfiler::simplify function was also used to reduce the redundancy among enriched terms.

Data & Code availability

snMultiome sequencing and spatial transcriptomics data generated in this study will be deposited in the Gene Expression Omnibus (GEO) and made publicly available upon publication of the manuscript. The code for snMultiome and spatial transcriptomics analysis will be publicly available on Github upon publication of the manuscript. Additional source data, statistics, etc. used in this study will be available at Synapse.org upon publication of this manuscript. For any inquiries about data accessibility and analyses, please email tim.petros@nih.gov.