Multimodal analyses reveal genes driving electrophysiological maturation of neurons in the primate prefrontal cortex

SUMMARY

The prefrontal cortex (PFC) is critical for myriad high-cognitive functions and is associated with several neuropsychiatric disorders. Here, using Patch-seq and single-nucleus multiomic analyses, we identified genes and regulatory networks governing the maturation of distinct neuronal populations in the PFC of rhesus macaque. We discovered that specific electrophysiological properties exhibited distinct maturational kinetics and identified key genes underlying these properties. We unveiled that RAPGEF4 is important for the maturation of resting membrane potential and inward sodium current in both macaque and human. We demonstrated that knockdown of CHD8, a high-confidence autism risk gene, in human and macaque organotypic slices led to impaired maturation, via downregulation of key genes, including RAPGEF4. Restoring the expression of RAPGEF4 rescued the proper electrophysiological maturation of CHD8-deficient neurons. Our study revealed regulators of neuronal maturation during a critical period of PFC development in primates and implicated such regulators in molecular processes underlying autism.

eTOC

Gao et al. report the identification of the genes, including RAPGEF4, and regulatory networks underlying neuronal maturation in the primate prefrontal cortex. Using an orthogonal multimodal approach, they reveal that knocking down CHD8 hinders proper maturation of excitatory neurons, which they were able to rescue by overexpressing RAPGEF4.

Graphical Abstract

INTRODUCTION

The dorsolateral prefrontal cortex (dlPFC), a neocortical area highly derived in primates^1–3, is required for higher-order brain functions, including executive functions such as working memory, planning, and decision-making^3–5. These functions are significantly impaired in numerous brain disorders, including autism spectrum disorders (ASD), schizophrenia, obsessive-compulsive disorder, and neurodegenerative disorders^6–8. Much of our knowledge on PFC development is largely extrapolated from rodent studies. Although rodents possess a PFC that is homologous to a small portion of the primate PFC, they lack the granular dlPFC¹. Several comparative studies have identified gene expression signatures that are unique to the PFC of humans and non-human primates (NHP)^3,9–12. Therefore, it is essential to understand the mechanisms regulating dlPFC development in primates.

The cells that compose the primate dlPFC, especially excitatory and inhibitory neurons, undergo extensive and dynamic maturation throughout midfetal and late-fetal development, during which critical neurodevelopmental events, such as circuit assembly and electrophysiological maturation of neurons occur¹³. Midfetal and late-fetal development are a convergent period of expression for many ASD genes^14,15, yet the function of these ASD genes during this period of development remain unclear. Furthermore, although we have a robust knowledge of the molecular and cellular mechanisms that govern cell-fate specification during cortical development^16–22, and of the transcriptomic and electrophysiological properties of mature neurons in the adult neocortex^23–27, our understanding of the gene networks that regulate the early stages of neuronal maturation, especially electrophysiological maturation, remains elusive. Therefore, a multimodal investigation of primate dlPFC neurons that combines electrophysiological and functional genomic analyses during critical periods for neuronal maturation is essential to understanding the mechanisms driving neuronal maturation.

Here, to uncover the molecular mechanisms that regulate the maturation of dlPFC neurons, we first generated single-nucleus multiomic (snMultiomic) data and performed integrated analyses of the dlPFC from rhesus macaques (Macaca mulatta), ranging from midfetal to late-fetal periods. We then performed Patch-seq on acute dlPFC slices from 16 macaques, including the same specimens profiled for snMultiome. This approach allowed us to integrate our electrophysiological analyses with gene expression profiles to identify the genes that are important for the maturation of specific electrophysiological features. Furthermore, we have evaluated the function of select genes – RAPGEF4 and CHD8 – and demonstrated that a loss of function of these genes impaired the morphological and electrophysiological maturation of cortical neurons in organotypic slices of macaque and human, as well as in cultured human primary neurons. Conversely, overexpression of these genes led to morphologically and electrophysiologically more mature human primary cortical neurons. Finally, Patch-seq analysis of brain slices with CHD8 knockdown revealed RAPGEF4 as a mediator for CHD8 regulation of electrophysiological maturation of human cortical neurons.

RESULTS

Cell type-specific transcriptomic changes underlying the maturation of rhesus macaque dlPFC cells during midfetal and late-fetal development

We performed snMultiomic analyses on 9 rhesus macaque dlPFC samples across eight prenatal ages: postconceptional day (PCD) 85, PCD95, PCD100, PCD105, PCD110, PCD125, PCD145, and two PCD155 (Figures 1A–1B; Table S1). After implementing stringent quality control measures, we retained 76,855 nuclei for further analyses (Figures S1; Tables S1–S2). After removing batch effects, we used unsupervised clustering to group nuclei into transcriptomically defined groups (Figures S1B–S1C). Using this strategy, together with cell type-specific marker genes^16,23,28, we defined 9 major cell classes that included glutamatergic excitatory neurons (ExN), GABAergic inhibitory neurons (InN), glial progenitor cells (GPC), astrocytes, oligodendrocyte precursor cells (OPC), oligodendrocytes, microglia, vascular leptomeningeal cells (VLMC), and endothelial cells (Figure S2A). We re-clustered the ExN and InN nuclei to identify their subclasses based on the layer (L) and projection identity (intratelencephalic [IT], extratelencephalic [ET], near-projecting [NP], corticothalamic [CT], and L6B) for ExN, and the developmental origin (medial ganglionic eminence [MGE], caudal ganglionic eminence [CGE], or dorsal lateral ganglionic eminence [dLGE]) for InN (Figures 1C and S2B–S2C). Interestingly, one small population of IT neurons exhibited molecular features of both ExN L2-3 IT and ExN L3-5 IT, indicating that this small population of neurons is yet to be fully specified; we classified this population as ExN L2-3 IT/L3-5 IT. At the lowest hierarchical level of cell type classification, we transcriptomically defined 85 cell subtypes (Figure S2E; Table S3). By integrating our dataset with two published snRNA-seq datasets that analyzed early fetal²⁸ and adult²³ macaque dlPFC, we observed that our data bridges the embryonic/early fetal cell types with the adult cell types, facilitating the tracing of the developmental origins of distinct adult cell subtypes (Figures 1E–1G).

Figure 1. Single-nucleus transcriptomic atlas of rhesus macaque dlPFC cells across midfetal and late-fetal development.

(A) The frontal cortex of Rhesus macaque (Macaca mulatta) brains, aged from postconceptional day (PCD) 85 to 155, were dissected and sectioned at 300 μm. Acute dorsolateral prefrontal cortices were analyzed using Patch-seq and single-nucleus multiome (snRNA-seq and snATAC-seq). Tissue from the same specimens, as well as human midfetal tissue, was also used to culture organotypic slices for functional analyses of select genes (RAPGEF4 and CHD8). Validation of key findings was also performed by multiplexed immunostaining and single-molecule fluorescent in situ hybridization on dlPFC sections from the contralateral hemisphere of the same specimens (Table S1). Finally, we have performed an integrative analysis of the generated multimodal dataset.

(B-C) UMAP visualization of all the cell classes and subclasses. Cells are colored according to age in postconceptional days (PCD), from PCD85-PCD155 (B) cell subclass (C).

(D) Bar plot showing the cell composition of the dlPFC throughout midfetal and late-fetal development. A breakdown of the cell compositions is provided in Table S1.

(E-G) Integrated UMAP visualization depicting cells obtained from both the current study and published datasets^22,27. Cells are color-coded based on their respective dataset (current study in medium blue) (E), age of donor specimen (embryonic/early fetal in blue, current study in orange, adult in dark green) (F), and cell type (G). ExN, excitatory neuron; InN, inhibitory neurons; Astro, astrocytes; Oligo, oligodendrocytes; IPC, intermediate precursor cells; enIPC, excitatory neuron IPC; inIPC inhibitory neuron IPC; oIPC, oligodendrocyte IPC; OPC, oligodendrocyte precursor cells; GPC, glial precursor cells; Endo, endothelial cells; RG, radial glia cells; oRG, outer radial glia cells; tRG, truncated radial glia cells; vRG, ventricular radial glia cells; NESC, neuroepithelial stem cells; VLMC, vascular leptomeningeal cells; AntVen domain cell, antero-ventral domain cells; NA, unknown cell type.

See also Figures S1–2, and Table S1–3.

As expected, we observed an increasing number of glial cells, including astrocytes, OPCs, and oligodendrocytes throughout late-fetal development (Figure 1D; Table S1), a period characterized by minimal cortical neurogenesis and robust gliogenesis²⁹. Within the excitatory neurons, we observed that IT neurons were more abundant than any other subclass, as previously described in the adult dlPFC²³. The selective expansion of IT neurons in the primate lineage, especially of upper-layer (L2-3) neurons²³ and their functional relevance in the cortico-cortical circuits that govern some of the unique cognitive and behavioral abilities of primates³⁰, led us to focus our analysis on the maturational profiles of these IT neuronal populations: ExN L2-3 IT and ExN L3-5 IT (Figures 2A–2D). We performed differential gene expression analysis along the developmental trajectory to identify genes that display dynamic, highly variable changes during neuronal maturation. Among the top 100 genes (Table S4), we observed genes that exhibit cell type-specific correlation with maturation (Figures 2C–2D), including MEIS2, a transcriptional regulator that is induced by retinoic acid and is important for the arealization of the prefrontal cortex¹⁰, in L2-3 IT, and FOXP2, a gene encoding a transcription factor that has recently been shown to have primate-specific expression in L3-5 IT excitatory neurons²³. Disruptions in these genes have been linked to neuropsychiatric disorders^31,32.

Figure 2. Maturational trajectories of L2-3 and L3-5 intratelencephalic-projecting neurons.

(A-B) UMAP visualization of the maturation trajectory of ExN L2-3 IT (A), and ExN L3-5 IT (B) cells; the top panel is colored according to postconceptional days, whereas the bottom panel is colored according to inferred pseudotime.

(C-D) Gene expression variation with pseudotime of genes specific to ExN L2-3 IT (C), ExN L3-5 IT (D) (from the top 100 highly variable genes along the developmental trajectory).

See also Figures S1–2, and Table S1–4.

Together, our analyses uncovered the transcriptional landscape of primate midfetal and late-fetal PFC development, highlighting differences between distinct excitatory neuronal populations throughout maturation.

Chromatin accessibility signatures and gene regulatory networks of macaque midfetal and late-fetal dlPFC cells

The simultaneous profiling of gene expression and chromatin accessibility in the same cell allows linking putative regulatory elements to genes, thus providing insights into cell-type gene regulatory mechanisms in the developing primate brain. After filtering out low-quality nuclei and implementing other stringent quality control steps, snATAC-seq data from 64,941 nuclei were retained for downstream analyses (Figure S1D). We observed that the clustering of cells based on accessible chromatin was similar to the clustering utilizing gene expression data (Figures S1E–S1H), showing the congruence between the two datasets (Figure 3A). However, the snATAC-seq data revealed more pronounced differences in the maturation states of cells within each cell subclass, especially for IT excitatory neurons (Figure S1E). To gain a better understanding of the association of regulatory elements and their target genes during neuronal maturation, we estimated peak-to-gene links, which identify co-activity between genes and their nearby accessible chromatin regions, for the excitatory and inhibitory neurons (Figure S3A), thus providing a list of potential regulatory elements for each gene (Table S5). We found that peak-gene combinations were hierarchically divided into three distinct categories: a cluster specific to excitatory neurons during early midfetal development, a cluster specific to excitatory neurons during late-fetal development, and a cluster specific to inhibitory neurons during late-fetal development. Functional enrichment analysis of each group revealed that the clusters specific to late-fetal development were enriched for terms related to calcium signaling and dendritic development in excitatory neurons, and synaptic processes in inhibitory neurons, whereas the early-midfetal development cluster was enriched for transcriptional regulation terms (Figures S3C–S3E; Table S5).

Figure 3. Single-nucleus multiome gene expression and chromatin accessibility of rhesus macaque dlPFC cells across mid- to late-fetal development.

(A) Joint (snRNA-seq and snATAC-seq) UMAP visualization of cell subclasses.

(B) Differential gene regulatory networks inferred for ExN L2-3 IT and ExN L3-5 IT cell subclasses. Hexagonal nodes indicate transcription factors, and the elliptical nodes indicate the target genes. The edge colors represent the cell subclass specificity of transcription factor-target gene relationships (black, conserved in both cell subclasses; red, ExN L2-3 IT specific; blue, ExN L3-5 IT specific).

(C) UMAP visualization of MEIS2 expression and motif activity scores for MA0774.1 (MEIS2 transcription factor motif) in L2-3 IT and L3-5 IT cells (top panels); MEIS2 transcription factor footprints for ExN L2-3 IT and ExN L3-5 IT cells across four developmental periods (bottom panel).

(D) Immunofluorescent detection of MEIS2 in the macaque dlPFC at PCD155, in L2-3 IT and L3-5 IT cells shows that a significantly higher number of cells express MEIS2 in L2-3 IT than in L3-5 IT cells (L3-5 was identified using in situ hybridization against RORB, as shown in Figure S3E). Scale bar, 20 μm. (**p < 0.01, t-test).

(E) Peak-to-gene link plot for NFIA together with the gene expression of MEIS2 and NFIA in ExN L2-3 IT and ExN L3-5 IT across four developmental periods. The links indicate the co-activity links between NFIA gene expression and chromatin accessibility peaks. The highlighted regions within the chromatin accessibility peaks show the MEIS2 transcription factor motif (MA0774.1) binding sites.

(F) Dual-Luciferase reporter assay of NFIA genomic sequences containing MEIS2 motifs. OE, overexpression. *p < 0.05, **p < 0.01, ****p < 0.0001 (one-way analysis of variance with Tukey multiple comparisons). Each dot represents one replicate.

See also Figures S1 and S3 , and Tables S1 and S5–6.

Next, since gene expression is regulated by gene regulatory networks (GRNs)³³ and expression of genes associated with ASD converge in excitatory neurons during these developmental stages in ExN L2-3 IT and ExN L3-5 IT cell subclasses^14,15, we inferred cell-type gene regulatory networks (GRNs) for these two cell subclasses using transcriptomic and epigenomic data together with transcription factor (TF) motif binding sites (STAR Methods). These cell-type GRNs link regulatory elements with TF binding sites to target genes. We particularly focused on analyzing the subnetworks regulating the genes identified in our developmental maturation trajectory analysis (Table S4) and found nine major TFs regulating the driver genes of the two cell subclasses. Among these nine TFs, we identified MEIS2 and FOXP2 as TFs that are specific for ExN L2-3 IT and ExN L3-5 IT, respectively, and MEF2C as an important TF for both ExN subclasses (Figure 3B).

Since upper-layer (L2-3) neurons exhibit selective expansion in the primate lineage²³ and form the cortico-cortical circuits that are important for the unique cognitive and behavioral abilities of primates³⁰, we investigated putative regulatory elements that underlie the transcriptomic differences between ExN L2-3 IT and ExN L3-5 IT. We calculated differential accessibility and overrepresented motifs between these neuronal populations throughout midfetal and late-fetal development. Using this strategy, we identified the MEIS2 motif MA0774.1 as one of the overrepresented motifs in the ExN L2-3 IT subclass (Table S6). Furthermore, we found that ExN L2-3 IT has both higher MEIS2 gene expression and a higher MEIS2 motif activity score when compared to ExN L3-5 IT (Figure 3C). Because MEIS2 is a transcription factor enriched in upper-layer neurons in primates^10,23 and is implicated in neuropsychiatric disorders³², we calculated MEIS2 transcription factor footprints across the whole genome in the two cell subclasses and found that, throughout the analyzed developmental periods, ExN L2-3 IT had a higher footprint enrichment than ExN L3-5 IT, with the highest enrichment in ExN L2-3 IT during midfetal development (PCD105-110) and the lowest footprint enrichment in ExN L3-5 IT during late fetal development (PCD155), indicating the selective involvement of MEIS2 in the regulation of target genes in ExN L2-3 IT (Figure 3C). Indeed, quantitative immunofluorescence showed that there were significantly fewer MEIS2-expressing cells in L3-5, identified by the expression of RORB (Figure S3B), compared to L2-3 of dlPFC at PCD155 (Figure 3D).

One of the genes potentially regulated by MEIS2 in IT excitatory neurons is the Nuclear factor I/A (NFIA), which has been shown to be regulated by MEIS2 in retinal progenitor cells³⁴ and is linked to callosal hypoplasia, likely reflecting malformations of upper-layer excitatory neurons³⁵. By analyzing the distribution of accessible chromatin peaks in the vicinity of NFIA across the two neuronal subclasses and developmental periods, we identified five active MEIS2 binding sites within 200kb upstream of the NFIA transcription start site (Figure 3E). We also found three peak-to-gene links (peaks that show higher co-activity with NFIA gene expression), with one of the links containing two predicted MEIS2 binding motifs (chr1-163553896-163555161) approximately 181kb upstream of the NFIA transcription starting site, suggesting that it may be a potential regulatory element for NFIA. To validate this finding, we conducted a dual luciferase assay to assess the activity of the genomic sequence containing the two MEIS2 motifs. Overexpression of MEIS2 resulted in a significant increase in transcription activity driven by the genomic sequence. Furthermore, when we introduced mutations into two potential MEIS2 binding motifs, one of these mutations significantly reduced the activity of the genomic sequence (Figure 3F). Taken together, our analyses have identified key regulators, such as MEIS2, and gene regulatory networks that likely underlie the maturation of distinct IT neuronal populations during midfetal and late-fetal primate brain development.

Patch-seq of midfetal and late-fetal macaque dlPFC cells identified transcriptomic signatures correlated with electrophysiological maturation

The midfetal and late-fetal periods are critical stages in cortical development, marked by notable alterations in neuronal morphology, electrophysiology, and gene expression¹³. Although all the analyzed neurons displayed a relatively immature morphology, PCD155 neurons exhibited significantly greater complexity, with greater total dendritic length, increased branching points, and more endings when compared to PCD100 neurons (Figures S4B–S4F). To gain a comprehensive understanding of the molecular mechanisms that drive neuronal maturation, we employed Patch-seq, which allows for sequential electrophysiological and transcriptomic analyses of the same cell³⁶, in 16 rhesus macaques ranging from PCD85 to PCD155 (Figures 4A and S4; Table S1). We used our snRNA-seq data (Figure 2) as the reference dataset to infer cell types by mapping the 256 cells profiled by Patch-seq onto the reference UMAP (Figures 4B and S4I; Table S7). Approximately 72% of all cells were identified as neurons based on their transcriptomic signature, which included 111 IT neurons (Table S7). Our evaluation of the electrophysiological features of three major IT subtypes (L2-3, L3-5, and L6) across different ages revealed that these IT neurons exhibited similar electrophysiological features (Figures S4J–S4N), showing no significant differences among them during this developmental window. Therefore, we combined all IT neuron types for downstream analysis.

Figure 4. Identification of midfetal and late-fetal neuronal maturation-related genes using Patch-seq.

(A) Schematic representation of the Patch-seq experimental design for analyzing midfetal and late-fetal rhesus macaque dlPFC cells.

(B) Integration of the 256 cells profiled by Patch-seq onto the snRNA-seq reference UMAP.

(C) Pseudotime trajectory based on the gene expression profiles of patched IT neurons from 16 individual rhesus macaque fetal brains at 9 distinct fetal ages, with representative electrophysiological sample traces (voltage responses to depolarizing and hyperpolarizing current injections, and inward sodium currents).

(D) In situ hybridization detection and quantification (right panel) of CAMK2A, MICAL2, and RAPGEF4 in PCD100 and PCD155 dlPFC. Scale bar, 20 μm. (****p < 0.0001, t-test).

(E-L) Plots illustrating the developmental changes in electrophysiological properties, including membrane capacitance (C_m, E), input resistance (R_in, F), resting membrane potential (RMP, G), and inward sodium current (I_Na, H) of IT neurons across six different age groups. Bottom panels show the expression levels of C_m-correlated genes (I), R_in-correlated genes (J), RMP-correlated genes (K), and I_Na-correlated genes (L) in IT neurons across the same age groups. (**p < 0.01, ***p < 0.001, ****p < 0.0001, one-way ANOVA, Tukey’s multiple comparisons test).

See also Figures S4–5, and Tables S1 and S7–9.

We employed a pseudotime trajectory analysis to infer a maturation trajectory by integrating differentially expressed genes across age groups in all identified IT neurons (STAR Methods). Our analysis revealed a trajectory characterized by the progressive maturation of IT neurons from earlier to later developmental periods (Figure 4C). We observed that the proportion of IT neurons exhibiting action potentials (APs) across six age groups (PCD85-90, PCD95-100, PCD105-110, PCD125, PCD145, and PCD155) gradually increased from 0% in the PCD85-90 group to 95% in PCD155, providing strong evidence for electrophysiological maturation (Figure S4O). We then assessed neuronal maturation through the lens of the following features that offer unique perspectives on maturation: membrane capacitance (C_m), input resistance (R_in), resting membrane potential (RMP), and inward sodium current (I_Na). Our analysis revealed that, during early midfetal development (from PCD85 to PCD100), none of the electrophysiological features showed significant change (Figure 4). We observed a significant increase in C_m from PCD100 to PCD105, after which the values remained stable (Figure 4). This increase in C_m potentially reflects the enlargement of the cell body and more complex cellular structures during the maturation of neurons. In contrast, there was a gradual decline in the averaged value of R_in after PCD125 (Figure 4F). This effect may be attributed to an increase in the number of ion channels in the neuronal membrane. Moreover, the morphological changes that occur in dendrites (Figures S4B–S4F) and spines during maturation may contribute to alterations in signal processing and integration, further enhancing neuronal function. We also observed a conspicuous pattern of increasingly negative RMP from PCD100 to PCD155 (Figure 4G), which reflects the developmental maturation in the ion channel composition of these neurons. During neuronal development, the current density of I_Na amplifies, thereby increasing the capacity of neurons to generate action potentials. Our results indicate a significant increase in the magnitude of the I_Na from PCD105 to PCD155 (Figure 4H). This suggests that IT neurons in the dlPFC acquire the ability to initiate and transmit APs after PCD105. In summary, our analyses showed a progression in maturation across four distinct electrophysiological properties. However, it is noteworthy that the temporal dynamics of the four electrophysiological properties were not identical. While RMP and I_Na exhibited a progressive increase from PCD95-100 to PCD155, C_m and R_in displayed drastic changes only at PCD105-110. Our study has uncovered distinct profiles in the temporal dynamics of electrophysiological maturation, where each property exhibits a different rate of maturation.

Similar to IT ExNs, InNs exhibited progressively more negative RMP and I_Na from PCD105-110 to PCD155, whereas R_in did not show any significant differences and C_m became higher throughout development (Figures S5D–S5H). The proportion of InNs with APs increased from 12.5% in PCD85-90 to 100% in PCD125, PCD145, and PCD155, indicating a faster electrophysiological maturation than ExNs (Figure S5I).

We hypothesized that differential expression of distinct genes drives the divergent trends of the four electrophysiological features across maturation. By implementing a correlation analysis between the electrophysiological data and the gene expression levels across all IT neurons, we identified 39, 129, 224, and 227 genes that are associated with the temporal dynamics of electrophysiological features of C_m, R_in, RMP, and I_Na, respectively (Figure S5A; Table S9). Importantly, genes that are not expected to exhibit correlation with electrophysiological features, including the housekeeping genes GAPDH and TBP, showed no significant correlation with any of the electrophysiological features (Figure S5B). Several genes, including CAMK2A and MICAL2, exhibited high correlation with all four investigated electrophysiological features (Figures 4I–4L). Other genes, such as RAPGEF4, exhibited significant correlation with one or two of the electrophysiological features. To validate our Patch-seq data, we performed in situ hybridization on dlPFC sections of PCD100 and PCD155 macaques for select genes that demonstrated significant correlation with all (CAMK2A, MICAL2) or some (RAPGEF4, NRGN) of the electrophysiological features. Consistent with the Patch-seq results and the multiome expression data, the expression levels of all tested genes, including the voltage-dependent calcium channel CACNA1C and the voltage-gated sodium channel SCN2A were significantly higher in PCD155 compared to PCD100, whereas the expression of the choline transporter SLC44A5 was lower in PCD155 compared to PCD100 (Figures 4D, S3F–S3H, and S4Q).

Therefore, our Patch-seq analysis of dlPFC neurons across midfetal and late-fetal development revealed evidence of neuronal maturation across three distinct aspects: morphological, transcriptomic, and electrophysiological properties, and identified genes associated with maturation of specific electrophysiological features.

RAPGEF4 is crucial for the morphological and electrophysiological maturation of excitatory neurons

Our Patch-seq analysis demonstrated a marked upregulation of RAPGEF4 expression in IT neurons during cortical development (Figures 4K–4L), which was also corroborated by our snRNA-seq data (Figure 5A). RAPGEF4 (Rap Guanine Nucleotide Exchange Factor 4), also known as EPAC2 (exchange protein directly activated by cyclic AMP 2), is a regulator of small GTPase signaling and a protein kinase A (PKA)-independent cAMP target^37–39. We found that RAPGEF4 expression exhibited a high degree of correlation with two electrophysiological features, with the highest correlation observed with I_Na and the second highest correlation with RMP (Figure S6A). To confirm that RAPGEF4 is required for neuronal maturation, we infected ex vivo macaque brain slices with lentivirus (LV) expressing mScarlet and shRNA against RAPGEF4 (LV-mScarlet-U6-shRAPGEF4) together with LV expressing mNeonGreen and control shRNA (LV-mNeon-U6-shNC) (Figures 5B and S6B–S6C). The neurons with RAPGEF4 knockdown (shRAPGEF4, mScarlet⁺) exhibited reduced dendritic complexity, total dendritic length, branching points, and number of endings compared to controls (mNeonGreen⁺mScarlet⁻) (Figures 5C–5H). In addition, LV-shRAPGEF4-infected neurons exhibited smaller I_Na (Figures 5I and 5M) and less negative RMP (Figures 5I and 5L), without significant changes in C_m and R_in, compared to control neurons (Figures 5J–5K; Table S10). These results indicate that a deficiency of RAPGEF4 in neurons leads to delayed maturation of I_Na and RMP, which is consistent with our finding that RAPGEF4 had a strong correlation with I_Na and RMP (Figures 4K–4L and S6A).

Figure 5. Knockdown of RAPGEF4 impairs morphological and electrophysiological maturation of cortical neurons in macaque organotypic dlPFC slices.

(A) Violin plots showing the expression of RAPGEF4 in IT neurons from PCD85 to PCD155.

(B) Schematic of the experimental setup for investigating the morphological and electrophysiological alterations after RAPGEF4 knockdown in ex vivo dlPFC slices.

(C) Representative image depicting RAPGEF4 knockdown (both mScarlet⁺[red] and mScarlet⁺/mNeon⁺[yellow]) and control (mNeon⁺[green only]) neurons in macaque dlPFC slices. Scale bar, 100 μm.

(D) Sample traces depicting the morphology of both RAPGEF4 knockdown (bottom panel) and control (top panel) neurons.

(E) Sholl analysis of dendritic complexity in RAPGEF4 knockdown neurons compared to control neurons [F(1,150) = 14.531, ***p < 0.001, multivariate analysis of variance].

(F-H) Knockdown of RAPGEF4 decreases total dendritic length (F), number of nodes (G), and number of endings (H) (Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; ***p < 0.001, **p < 0.01).

(I) Representative traces of inward sodium currents and voltage responses to depolarizing and hyperpolarizing current injections for macaque RAPGEF4 knockdown (right panel) and control (left panel) neurons.

(J-M) Knockdown of RAPGEF4 in macaque organotypic slices leads to reduced I_Na (M) and less negative RMP (L), without significantly change in C_m (J) and R_in (K) (R_in and RMP: two-tailed unpaired t-test; C_m and I_Na: Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; ***p < 0.001, *p < 0.05). Each dot represents one patched cell. All cells were collected from four individual fetal brains, with each fetal brain indicated by a different color: black, dark gray, medium gray, and light gray.

(N) Representative traces of inward sodium currents and voltage responses to depolarizing and hyperpolarizing current injections for human RAPGEF4 knockdown (right panel) and control (left panel) neurons.

(O-R) Knockdown of RAPGEF4 in human organotypic slices leads to reduced I_Na (R) and less negative RMP (Q), without significantly change in C_m (O) and R_in (P) (R_in: two-tailed unpaired t-test; C_m, RMP and I_Na: Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; *p < 0.05). Each dot represents one patched cell. All cells were collected from four individual fetal brains, with each fetal brain indicated by a different color: black, dark gray, medium gray, and light gray.

See also Figure S6, and Tables S1 and S10.

RAPGEF4 is enriched in neurons compared to other cell types³⁹. The levels of RAPGEF4 increase in the human cortex throughout development, compared to other brain regions or organs^9,40. Based on our observations in macaque, we hypothesized that RAPGEF4 expression is also associated with the maturation of electrophysiological properties in human. We infected human midfetal cortical brain slices with LV-shRAPGEF4 and performed patch-clamp recording. As in macaque, we observed reduced I_Na and RMP, but not C_m and R_in, in LV-shRAPGEF4-infected compared to LV-shNC-infected neurons (Figures 5N–5R; Table S10). To confirm that the observed effect was not caused by off-target effects of the shRNA, we replicated the experiment and observed the same results using a different shRNA targeting RAPGEF4 in human neurons (Figures S6D–S6H). To ensure that our results were not dependent of our methodological approach, we knocked down RAPGEF4 in primary neurons isolated from midfetal human cortex and observed similar morphological (Figures S6I–S6M) and electrophysiological (Figures S6N–S6R; Table S10) deficits in these neurons. To corroborate that the observed deficits of shRAPGEF4 neurons are specific to a loss of function of RAPGEF4, we expressed ‘shRAPGEF4-resistant’ RAPGEF4 in human primary cortical neurons that also had RAPGEF4 knocked down. We observed that exogenous RAPGEF4 rescued both morphological (Figures S6S–S6W) and electrophysiological (Figures S6X–S6AB; Table S10) deficits of human cortical neurons, confirming the specificity of the RAPGEF4 shRNA.

Since RAPGEF4 levels increase from mid- to late-fetal development (Figure 5A), we overexpressed RAPGEF4 in human early midfetal brain slices, when RAPGEF4 levels are low, to determine whether the overexpression would lead to accelerated maturation (Figure 6A). In contrast to RAPGEF4 knockdown, RAPGEF4 overexpressing neurons exhibited increased neuronal complexity and longer dendrites (Figures 6B–6D), as well as increased RMP and I_na, without significant effects on C_m and R_in (Figures 6E–6I; Table S10). Similarly, treatment of human early midfetal organotypic slices with a RAPGEF4 activator, 8-CPT³⁹, promoted neuronal maturation morphologically (Figures 6J–6N), and increased RMP and I_Na (Figures 6O–6S; Table S10).

Figure 6. RAPGEF4 overexpression or activation promotes maturation of human cortical neurons.

(A) A schematic diagram depicting the strategy for expressing exogenous RAPGEF4 in ex vivo human fetal cortical slices.

(B) Rrepresentative traces of neurons showing the morphology of both RAPGEF4 overexpression (RAPGEF4 OE) and control neurons in human cortical slices.

(C) Sholl analysis of dendritic complexity in RAPGEF4 OE neurons compared to control neurons [F(1,94) = 9.881, **p < 0.01, multivariate analysis of variance].

(D) RAPGEF4 OE increases total dendritic length (Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; **p < 0.01).

(E) Representative traces of inward sodium currents and voltage responses to depolarizing and hyperpolarizing current injections in RAPGEF4 OE (right panel) and control (left panel) neurons. (F-I) RAPGEF4 OE leads to higer I_Na (I) and less negative RMP (H), without significantly change in C_m (F) and R_in (G) (I_Na: two-tailed unpaired t-test; C_m, R_in and RMP: Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; **p < 0.01, *p < 0.05). Each dot represents one patched cell.

(J) Sample traces showing the morphology of neurons in human cortical slices treated with agonist for RAPGEF4, 8-CPT or vehicle (control).

(K) Sholl analysis of dendritic complexity in neurons treated with 8-CPT compared to those treated with vehicle [F(1,121) = 9.135, **p < 0.01, multivariate analysis of variance].

(L-N) Neurons treated with 8-CPT have longer total dendritic length (L), more nodes (M), and more dendritic endings (N) (Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; ****p < 0.0001).

(O) Representative traces of inward sodium currents and voltage responses to depolarizing and hyperpolarizing current injections for human neurons treated with 8-CPT (right panel) and control (left panel) neurons.

(P-S) Neurons in human organotypic slices treated with 8-CPT have higer I_Na (S) and less negative RMP (R), without significantly change in C_m (P) and R_in (Q) (C_m and RMP: two-tailed unpaired t-test; R_in and I_Na: Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; ***p < 0.001, *p < 0.05). Each dot represents one patched cell.

See also Figure S6, and Tables S1 and S10.

Together, these findings highlight the important role of RAPGEF4 in neuronal maturation during midfetal and late-fetal development in primate brains. This role of RAPGEF4 is consistent with the functional enrichment for “regulation of GTPase activity” in genes driving neuronal maturation (Figure S5C). These data validate our approach to identify genes that drive the maturation of specific electrophysiological features and reveal the novel role of RAPGEF4 in morphological and electrophysiological maturation of neurons in both macaque and human cortical development.

CHD8 knockdown leads to impaired maturation of human cortical excitatory neurons

Recent advances in genomics have revealed important roles of chromatin regulators during midfetal PFC development and implicated them in neurodevelopmental disorders^41,42. CHD8 (Chromodomain Helicase DNA-binding protein 8) is a chromatin remodeling protein that plays a crucial role in regulating gene expression during brain development and has been linked to ASD^42–44. To investigate the role of CHD8 during human midfetal cortical development, a critical developmental period for ASD^14,15, we knocked down CHD8 expression in ex vivo human midfetal organotypic brain slices using LV expressing shCHD8 and mScarlet together with LV expressing shNC and mNeonGreen (Figures 7A and S7A–S7B). LV-shCHD8-infected neurons displayed significantly shorter dendrites compared to control neurons (Figures 7B–7D and S7C–S7E), suggesting that a deficiency in CHD8 disrupts morphological maturation of cortical neurons during midfetal development, which is consistent with previous findings in mice⁴⁴.

Figure 7. Knockdown of CHD8 impairs maturation of cortical excitatory neurons and down-regulates RAPGEF4 in human organotypic dlPFC slices.

(A) Schematic diagram illustrating the experimental setup used to study the morphological and electrophysiological changes after knockdown of CHD8 in ex vivo human fetal cortical slices.

(B) Representative image showing CHD8 knockdown (mScarlet⁺[red], mScarlet⁺ and mNeon⁺[yellow]) and control (mNeon⁺[green]) neurons in human cortical slices. Scale bar, 100 μm.

(C) Sample traces showing the morphology of both CHD8 knockdown and control neurons.

(D) CHD8 knockdown reduces total dendritic length (Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; **p < 0.01).

(E) Representative traces (voltage response to depolarizing [50 pA] and hyperpolarizing [−50 pA] current injections) for control neurons and CHD8 knockdown.

(F-I) CHD8 knockdown increases R_in (G) and reduces RMP (H) but does not significantly alter C_m (F) and I_Na (I) (Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; *p < 0.05, **p < 0.01).

(J) Integration of the 156 cells profiled by Patch-seq onto the snRNA-seq reference UMAP.

(K) Volcano plot depicting (−log₁₀(P_adj)) vs log₂(fold-change)) differentially expressed genes between LV-shNC and LV-shCHD8 -infected excitatory neurons. Genes with expression profiles that significantly correlate with the RMP (green), R_in (purple), or both (red) electrophysiology features are labeled.

(L-O) Violin plots showing the expression levels of NRGN (L), CACNA1I (M), RAPGEF4 (N), and RGS6 (O), highlighting the effect of CHD8 knockdown on these neuronal maturation genes (Figures 4I–4L).

(P) Schematic of lentiviral vectors used for knocking down CHD8 and overexpression of RAPGEF4 in human cortical ex vivo organotypic slices.

(Q) Representative traces (voltage response to depolarizing [50 pA] and hyperpolarizing [−50 pA] current injections) for neurons with CHD8 knockdown and neurons with CHD8 knockdown and RAPGEF4 overexpression.

(R) Representative traces of inward sodium currents to depolarizing current injections for neurons with CHD8 knockdown and neurons with CHD8 knockdown and RAPGEF4 overexpression.

(S-V) RAPGEF4 rescues RMP (U) and I_Na (V) but does not significantly alter C_m (S) and R_in (T) of neurons with CHD8 knockdown (C_m and I_Na: two-tailed unpaired t-test; R_in and RMP: Mann-Whitney test; Normality was assessed using the Shapiro-Wilk test; *p < 0.05, **p < 0.01).

See also Figure S7, and Tables S1 and S11–12.

We then performed Patch-seq on 156 cells from slices that contained both CHD8-knockdown and control cells (Figure 7J). To determine the cellular identity of recorded cells, we compared their transcriptome with the reference snRNA-seq data and assigned each queried cell a specific cell type (Figures 7J and S7F; Table S11). Out of the 156 cells examined, there were 55 excitatory neurons infected with LV-shNC and 42 infected with LV-shCHD8 (Table S11). LV-shCHD8-infected excitatory neurons had less negative RMP and a significant increase in R_in (Figures 7E, 7G and 7H) but not in C_m and I_Na (Figures 7E, 7F and 7I), compared to the controls. To rule out the possibility that the observed effect is due to shRNA off-target effects, we repeated the experiment with an alternative shRNA in human cultured neurons and observed similar deficits in both R_in and RMP (Figures S7G–S7K). We then knocked down CHD8 in primary neurons isolated from human midfetal cortex and observed similar morphological (Figures S7L–S7P) and electrophysiological (Figures S7Q–S7U; Table S11) deficits as observed in human cortical organotypic slices. To further assess the role of CHD8 in human neuronal maturation, we performed a gain of function assay (STAR Methods). Due to the large size of CHD8, we opted to use a dCas9-activator human pluripotent stem cell line⁴⁵ and designed LV-sgRNAs targeting the promoter of CHD8 (Figures S7V–S7W). While CHD8 upregulation did not have a significant effect on neuronal morphology (Figures S7X–S7AC), it led to increased C_m, decreased R_in, and increased RMP (Figures S7AD–S7AH; Table S11), indicating that neurons with upregulation of CHD8 were electrophysiologically more mature.

Therefore, CHD8 knockdown and overexpression produce a discernible impact on the electrophysiological properties of midfetal human neurons, which is consistent with its role during early development^43,44.

RAPGEF4 rescues electrophysiological deficits of CHD8-deficient human cortical neurons

To determine which genes underlie the observed differences in R_in and RMP maturation caused by CHD8 knockdown, we identified differentially expressed genes between shCHD8 and shNC and intersected this list with the genes that were associated with the maturation of these two electrophysiological features (Table S9). We found 30 genes associated with R_in, 32 genes associated with RMP, and 20 genes associated with both (Figure 7K). These included genes that are associated with the overall maturation of neurons, such as NRGN, CACNA1I, and RGS6, a GTPase regulator important for activity-dependent morphological and electrical maturation⁴⁵, and RAPGEF4, which we found to significantly impact the maturation of RMP (Figures 7K–7O and 5).

To determine whether reduced RAPGEF4 levels might contribute to the observed reduction of RMP in CHD8 deficient neurons, we expressed RAPGEF4 in human cortical neurons that had CHD8 knockdown (shCHD8) (Figure 7P). Exogenously expressed RAPGEF4 restored normal maturation of both RMP and I_Na, with no significant effect on R_in or C_m, in CHD8-deficient neurons (Figures 7Q–7V; Table S12). Therefore, CHD8 deficiency in primate cortical neurons leads to reduced expression of genes important for electrophysiological maturation and we identified RAPGEF4 as one of the mediators for CHD8 regulation of maturation of cortical neurons during human midfetal development.

DISCUSSION

In this study, we employed multimodal approaches including snMultiome and Patch-seq to interrogate the mechanisms driving neuronal maturation in the primate dlPFC during midfetal and late-fetal development. We identified dynamic gene expression changes across cell types in their maturation trajectories as well as critical developmental periods for the maturation of specific electrophysiological properties in excitatory and inhibitory neurons.

We confirmed that, based on their molecular profiles, excitatory neurons gain their laminar and projection identities earlier than interneuronal final specification. This corroborates previous findings that pointed to the relevance of extrinsic factors to the final specification of interneurons during perinatal development^20,46. We focused our analyses on the IT excitatory neurons due to their relevance both to primate brain evolution and their association with neuropsychiatric disorders. Our integrative analysis of gene expression and chromatin accessibility data unveiled gene regulatory networks associated with neuron maturation of both excitatory and inhibitory neurons and highlighted MEIS2 as a key transcription factor for L2-3 IT neuronal development and maturation.

By integrating electrophysiology, morphology, and functional genomics analyses of primate dlPFC development, we discovered that different electrophysiological properties mature at different rates. While RMP change gradually and linearly during maturation, C_m and R_in exhibit a jump at the end of neurogenesis during midfetal developmental, and I_Na exhibited significant maturation only at the late-fetal period, which coincides with a significant increase in the number of neurons able to fire action potentials. The distinct rate of maturation among electrophysiological properties reveals the intricate and finely tuned regulation of neuronal maturation during midfetal and late-fetal development of the primate PFC.

Our Patch-seq approach allowed us to investigate the genes associated with the maturation of specific electrophysiological features. We have uncovered genes, including kinases, small GTP signaling molecules, neurotransmitter receptors, ion channels, and transporters that are associated with the maturation of specific electrophysiological features that likely underlie the differences we observed in their maturation kinetics. One of these genes is RAPGEF4, which we have identified as being significantly correlated with the maturation of resting membrane potentials and inward sodium current. RAPGEF4, a guanine exchange factor in small GTPase pathway, is a direct target of cAMP, has been shown to regulate dendritic spine remodeling in mice, and its deficiency in mice leads to learning and memory deficits^39,47. However, the role of RAPGEF4 in primate brain development or human disease is unknown. Dysregulation of the cAMP pathway has been implicated in a wide range of neurological disorders and a potential treatment to elevate cAMP levels is currently in clinical trial for fragile X syndrome^48–50. The small GTPase pathway has been shown to have critical roles during maturation of human neurons⁵¹. Our study demonstrates that RAPGEF4 is one of the small GTPases regulating the maturation of certain intrinsic electrophysiological properties in cortical neurons during early primate brain development.

Finally, to demonstrate the potential of our experimental platform in exploring the developmental origin of human diseases, we showed that knocking down of a high-confidence ASD gene – CHD8 – impaired morphological and electrophysiological maturation of excitatory neurons in human organotypic slices. This impairment was driven by the downregulation of the expression of several key genes that we identified to be relevant for the electrophysiological maturation of neurons, including RAPGEF4 and RGS6, two regulators of small GTPase signaling critical for neuronal maturation⁵⁰. By restoring the expression of RAPGEF4 in CHD8-deficient neurons we were able to rescue the deficits in electrophysiological maturation of these neurons. The chromatin regulator CHD8 is expressed at high levels during early-fetal and midfetal development and the levels of CHD8 do not exhibit significant upregulation during midfetal to late-fetal neuronal maturation. How CHD8 and other early expressing chromatin regulators converging at midfetal development contribute to ASD remains unclear. Our finding sheds light on how a chromatin remodeler may regulate the physiological maturation of neurons, thus impairing their normal functioning.

Together, our study has revealed key genes and gene networks driving neuronal maturation during a critical period of prefrontal cortical development in primates and linked the function of a novel regulator of the maturation of specific electrophysiological features with a high-confidence ASD risk gene. Our results demonstrate the power of our multimodal approach and its application for future studies aimed at understanding how genes regulate neuronal maturation and its implication in brain disorders. Finally, we have summarized our results including cell clusters, genes and networks in an open-access, searchable web app, which will allow researchers to investigate the role of different genes on neuronal maturation.

Limitations of the study

Although our study has identified several genes associated with the maturation of intrinsic neuronal properties in the developing primate prefrontal cortex, there are limitations that should be addressed. All experiments were performed in ex vivo organotypic slices and primary neuronal cultures, which do not fully capture the diversity of long range connectivity, as well as the microenvironment, of the in vivo brain. Thus, future work involving in vivo validation of our findings will be important for corroborating the role of the identified genes in neuronal maturation. Our work was focused on the prefrontal cortex due to its role in highly derived cognitive functions in primates. It will be valuable to analyze other cortical areas to establish whether these mechanisms that underlie neuronal maturation are shared across different areas or are specific to the prefrontal cortex. Although we found that RAPGEF4 is important for the electrophysiological and morphological maturation of cortical excitatory neurons, the molecular mechanisms driving this maturation remain elusive and should be further explored in future studies. Finally, our work was centered on the maturation of intrinsic electrical properties of neurons during prenatal development, which have been thus far understudied. Future work on other aspects of maturation, including connectivity and synaptic plasticity, among others, will establish a more complete understanding of neuronal maturation.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Andre M. M. Sousa (andre.sousa@wisc.edu).

Materials availability

All unique reagents generated in this study are available from the lead contact.

Data and code availability

The genomic data has been deposited at Gene Expression Omnibus (GEO; RRID:SCR_005012), under the persistent identifier GSE235493. The data can be interactively visualized at https://daifengwanglab.shinyapps.io/DevMacaquePFC/. The code and data used for generating figures can be accessed at Zenodo (RRID:SCR_004129) https://doi.org/10.5281/zenodo.15243470. Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

STAR Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Postmortem macaque and human brain specimens

Macaque brain specimens were collected postmortem from 19 specimens, from postconceptional day (PCD) 85 to 155. All experiments using non-human primates were carried out in accordance with a protocol approved by the University of Wisconsin’s Institutional Animal Care and Use Committee and NIH guidelines. Specimens from fetal (17 to 21 postconceptional weeks; Table S1) human brain were obtained from the Birth Defects Research Laboratory at the University of Washington, with ethics board approval and maternal written consent. This study was performed in accordance with ethical and legal guidelines of the University of Wisconsin-Madison Institutional Review Board. Tissue was handled in accordance with ethical guidelines and regulations for the research use of human brain tissue set forth by the NIH (https://oir.nih.gov/sourcebook/ethical-conduct/special-research-considerations/policies-procedures-use-human-fetal-tissue-hft-research-purposes-intramural/policies#acquisition) and the WMA Declaration of Helsinki (https://www.wma.net/policies-post/wma-declaration-of-helsinki-ethical-principles-for-medical-research-involving-human-subjects/). All clinical histories, tissue specimens, and histological sections were evaluated to assess for signs of disease, injury, and gross anatomical and histological alterations. No obvious signs of neuropathological alterations were observed in any of the specimens analyzed in this study. A detailed list of all specimens used in this study is available in Table S1.

Primary neurons isolated from human cortical tissue

Isolation of primary neurons from human fetal cortical tissue was performed using a published method⁵¹, with modifications. Brifely, cortical tissue was dissected and arachnoid mater and large vasculature removed from the cortical surface. Samples were transferred to a 15mL tube in Hibernate^™-E Medium (Gibco, A1247601) and centrifuged for 2 minutes at 300g. The supernatant was then aspirated and tissue resuspended in neuron plating medium (Neurobasal A medium [Gibco, 10888-022], 1X B-27 Supplement minus vitamin A [Gibco, 12587-010], 1X Glutamax [Gibco, 35050061], 1X Antibiotic-Antimycotic [Gibco, 15240-062]) by pipetting. Resuspended neurons were then filtered through a 70um cell strainer and plated in a 24-well plate on poly-d-lysine hydrobromide (0.1mg/mL final conc., Sigma-Aldrich, P6407) and laminin (5ug/mL final conc., Sigma-Aldrich, L2020)-coated glass coverslips. Two days after plating, medium was changed to primary neuron maturation medium (Brainphys Neuronal Medium [StemCell Technologies, 05790], 1X B-27 Supplement minus vitamin A [Gibco, 12587-010], 1X N2, 1X Antibiotic-Antimycotic [Gibco, 15240-062], 1X Glutamax [Gibco, 35050061], 20ng/mL BDNF [Peprotech, 450-02], GDNF [Peprotech, 450-10], 1μM dibutyryl cyclic-AMP [Sigma, D0260], 200μM l-ascorbic acid [Sigma, A8960]) for neuronal maturation. Half medium changes of primary neuron maturation medium was performed every 2-3 days.

Human neurons differentiated from pluripotent stem cells

The idCas9A-H9 human pluripotent stem cell line⁴⁵ was differentiated using the NGN2-induced differentiation method, as described⁴⁵. Briefly, hPSCs were plated onto MEFs for 5 days, and neural differentiation was then induced by switching the hESC medium to neural induction medium (NIM) [DMEM/F12 (Thermo Fisher Scientific, 11330032): Neurobasal (Thermo Fisher Scientific, 21103049) 1:1, supplemented with 1X N2 (Waisman Center Stem Cell Core), 1X L-Glutamine (Thermo Fisher Scientific, 25030081), 1X Anti-Anti (Thermo Fisher Scientific, 15240062), 10 μM SB432542 (Biogems, 3014193), 100 nM LDN193189 (Selleck, S2618), and 2 μM XAV-939 (Tocris, 3748)]. Cells were cultured in NIM for 9 days with a daily medium change. Cells were then dissociated with TrypLE (Thermo Fisher Scientific, 12605010) and replated 1:1 on Cultrex-coated (R&D Systems, 3433-005-01) plates in neural progenitor cell (NPC) medium [Neurobasal medium (Thermo Fisher Scientific, 21103049), 1X L-Glutamine (Thermo Fisher Scientific, 25030081), 1X N2 (Waisman Center Stem Cell Core), 0.5X B27 without vitamin A (Thermo Fisher Scientific, 12587020), 1X Anti-Anti (Thermo Fisher Scientific, 15240062)) supplemented with 10 μM ROCK inhibitor (Y-27632 dihydrochloride, Tocris, 1254)]. Cells were patterned for 7 days with daily change of NPC medium. The NPCs were re-plated for the differentiation of neurons. For neuronal differentiation, NPCs were dissociated with Accutase (Thermo Fisher Scientific, A111050) and re-plated on Cultrex-coated (R&D Systems, 3433-005-01) plates. For neuronal differentiation, NPCs were plated at a density of 300,000 cells/cm² in NPCs medium [DMEM/F12 (Thermo Fisher Scientific, 11330032), 1X N2 (Waisman Center Stem Cell Core), 1X B27 without vitamin A (Thermo Fisher Scientific, 12587020), 1X Anti-Anti (Thermo Fisher Scientific, 15240062), 20 ng/ml FGF2 (Waisman Manufacturing), 1 μg/ml mouse laminin (Thermo Fisher Scientific, 23017015) supplemented with rock inhibitor (Y-27632 dihydrochloride, Tocris,1254)]. To transduce NPCs with LV, the media was aspirated and replaced with NPCs medium containing the lentiviruses LV-TetO-NGN2-Neo and LV-rtTA the following day. Viral media was replaced with fresh NPCs medium containing 1 μg/ml doxycycline (Thermo Fisher Scientific, D3447) after 24 h. On the following day, media was replaced with fresh NPCs medium containing 400 μg/ml G418 (Sigma-Aldrich, 10131027) and 1 μg/ml doxycycline (Thermo Fisher Scientific, D3447) for 3 days. After G418 selection, cells were dissociated with Accutase (Thermo Fisher Scientific, A111050) and re-plated on Poly-D-Lysine (Sigma-Aldrich, P7886)/Cultrex (R&D Systems, 3433-005-01)-coated plates in Neurons medium [DMEM/F12 (Thermo Fisher Scientific, 11330032), 1X N2 (Waisman Center Stem Cell Core), 1X B27 without vitamin A (Thermo Fisher Scientific, 12587020), 1X Anti-Anti (Thermo Fisher Scientific, 15240062), 20 ng/ml BDNF (Peprotech, 450-02), 20 ng/ml GDNF (Peprotech, 450-10), 500 ng/ml cAMP (Sigma-Aldrich, D0260), 200 μM ascorbic acid (Sigma-Aldrich, A8960 ), 1 μg/ml mouse Laminin (Thermo Fisher Scientific, 23017015) supplemented with 1 μg/ml doxycycline (Thermo Fisher Scientific, D3447), 10 μM ROCK inhibitor (Y-27632 dihydrochloride, Tocris, 1254) and 0.1 μM Compound E (Sigma-Aldrich, 565790)]. Neurons were fed every 3-4 days in Neurons medium supplemented with 1 μg/ml doxycycline (Thermo Fisher Scientific, D3447).

METHODS DETAILS

Brain dissection, processing, and ex vivo organotypic culture

Tissue processing and organotypic slice culture were performed as described in our previous publications^9,52,53. The prefrontal cortex was dissected in chilled hibernate E solution (Thermo Fisher Scientific Inc, A1247601) with 1X B27 supplement (Thermo Fisher Scientific Inc, 17504044), 1X Glutamax (Thermo Fisher Scientific Inc, 35050061) and 1X Penicillin-Streptomycin (Thermo Fisher Scientific Inc, 15140122). After being rinsed with fresh hibernate E solution, the prefrontal cortex was then embedded in 3.2% low melting point agarose (Promega Corporation, V2111) in PBS. Acute coronal brain slices (300 μm thickness) were prepared with a VT1200S (Leica Biosystems) vibratome. The dorsolateral prefrontal cortex was dissected from acute slices for snMultimome and was immediately frozen in isopentane (J.T. Baker)/dry ice at −40 °C and stored at −80 °C. The slices for patch clamp recording were transferred to oxygenated artificial cerebral spinal fluid (aCSF) composed of (in mM): 124 NaCl, 2.5 KCl, 2.5 CaCl₂, 1.2 MgCl₂, 1.25 NaH₂PO₄, 26 NaHCO₃, and 15 glucose (pH 7.35) for incubation before being used for recording.

For ex vivo organotypic culture, slices were rinsed with hibernate solution and slice culture medium (BrainPhys^™ Neuronal Medium (Stem Cell Technologies, 05790) supplemented with N2 (1:100), BDNF (25ng/ml), B27 (1:50), and 1x Penicillin-Streptomycin (Thermo Fisher Scientific Inc, 15140122) and then transferred to a 6-well plate for culture and viral transduction. The slices were placed on 0.4 μm pore-sized PET membrane cell culture inserts (Falcon, 353090) and the wells were filled with 1 ml of culture medium. Culture plates were placed in a humidified 5% CO₂ incubator at 37 °C. The medium was changed every other day, with half of it being replaced each time. For RAPGEF4 or CHD8 knockdown experiments, cultured slices were infected with shRNA-expressing lentivirus by directly applying a concentrated lentivirus solution over the surface of the slice one day after the slices were plated. For maturation morphological analysis, cultured slices were infected with mScarlet expressing lentivirus immediately after the slices were plated.

For validation experiments, dorsolateral prefrontal cortex from PCD100 and PCD155 rhesus macaque fixed specimens were dissected and embedded in OCT. Sections were cut at 35 μm thickness on a Leica CM1950 Cryostat and mounted on TOMO^® adhesion slides (Matsunami Glass USA, TOM-11/90).

Treatment of ex vivo organotypic cultured brain slices with RAPGEF4 agonist

For agonist treatment, human fetal cortical brain slices were treated with 8-CPT (8-CPT-2Me-cAMP sodium salt, Tocris Bioscience, 1645) at a final concentration of 50 μM. In treatments shown in Figure 6, fetal cortical brain slices (~11-13 DIV) were treated with 8-CPT for 11 days before electrophysiological recording or fixation for morphological analysis. The experimental slices and corresponding control slices were processed on the same day.

Nucleus Isolation for single-nucleus multiome (snATAC-seq and snRNA-seq)

Isolation of nuclei was performed according to our previous publication²³ with minor modifications. Frozen dorsolateral prefrontal cortex sections were pulverized into a powder in liquid nitrogen on dry ice, using a mortar and pestle (Fisherbrand, FB961A, FB961K) before performing nuclei isolation. All buffers were prepared fresh and maintained on ice. 500 μL of chilled nuclear lysis buffer (10 mM Tris-HCl (pH 7.4; Sigma, T2194), 10 mM NaCl (Sigma, 59222C), 3 mM MgCl₂ (Sigma, M1028), 0.1% Tween-20 (Bio-Rad, 1662404), 0.1% Nonidet P40 Substitute (Sigma, 74385), 0.01% Digitonin (Thermo Fisher Scientific, BN2006), 1% BSA (Miltenyi Biotec, 130-091-376), 1 mM DTT (Sigma, 646563), 1 U/μl RNase inhibitor (Sigma, 3335402001)) was added to a tube containing approximately 10 mg of tissue. Once the tissue was resuspended in nuclear lysis buffer, the suspension was carefully transferred to a sterile, chilled 1 ml Dounce tissue grinder (DWK Life Sciences, 357538). The homogenization process involved 30 gentle strokes with a loose pestle, followed by 30 more gentle strokes with a tight pestle. Following a 5-min incubation on ice, the homogenate was gently mixed by pipetting it 15 times. Subsequently, the sample was further incubated on ice for 10 min. The homogenate was mixed with 500 μl of chilled nuclei wash buffer (10 mM Tris-HCl (pH 7.4; Sigma, T2194), 10 mM NaCl (Sigma, 59222C), 3 mM MgCl2 (Sigma, M1028), 1% BSA (Miltenyi Biotec, 130-091-376), 0.1% Tween-20 (Bio-Rad, 1662404), 1 mM DTT (Sigma, 646563), 1 U/μl RNase inhibitor (Sigma, 33335402001)). The homogenate mixture was passed through a pre-wetted 70 μm tube top cell strainer (Corning, 352350) using 250 μl of nuclear wash buffer. Subsequently, the filtered homogenate was further passed through a pre-wetted 40 μm tube top cell strainer (Corning, 352340) using 250 μl of wash buffer. The filtered homogenate was transferred to a 15 ml tube and subjected to centrifugation with a swing-out rotor (Eppendorf, 5943000343) at 500 rcf for 5 min at 4 °C. After centrifugation, the supernatant was removed, and the pellet was resuspended in 1 ml of chilled nuclear wash buffer. After gently pipetting the sample 5 times, it was centrifuged at 500 rcf for 5 min at 4 °C. Following centrifugation, the supernatant was discarded, and the resulting pellet was resuspended in 1 ml of chilled nuclear wash buffer. A volume of 10 μl from the sample was loaded onto a hemocytometer to perform a nuclei count. The obtained counting result was then used in a subsequent calculation to determine the appropriate final volume for resuspension targeting a final concentration of 5 million nuclei ml⁻¹. The solution was then subjected to centrifugation at 500 rcf for 5 min at 4 °C. Once the centrifugation was complete, the supernatant was removed without disturbing the pellet. The pellet was then resuspended in the volume calculated beforehand using diluted nuclei buffer (1X Nuclei Buffer (10x Genomics, 2000153), 1 mM DTT (Sigma, 646563), 1 U/μl RNase inhibitor (Sigma, 3335402001)). After the final resuspension, a volume of 10 μl from the sample was loaded onto a hemocytometer and counted to determine the final concentration.

Library construction for single nucleus multiome (snATAC-seq and snRNA-seq)

Only samples that met the minimum concentration requirement of 3.23 million nuclei per ml were selected for the generation of snATAC-seq and snRNA-seq libraries. The libraries were constructed in accordance with the guidelines provided in the “Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Kits User Guide” (10x Genomics, CG000338). To accomplish this, the Chromium Next GEM Single Cell Multiome ATAC Kit A (10x Genomics, PN-1000280), Chromium Next GEM Single Cell Multiome Reagent Kit A (10x Genomics, PN-1000282), and Chromium Next GEM Chip J Single Cell Kit (10x Genomics, PN-1000234) were employed. The GEM generation process was carried out utilizing a Chromium Controller (10x Genomics). For this study, the goal was to achieve a target recovery of 10,000 nuclei per capture. The snATAC-seq libraries were generated following the manufacturer’s protocol, utilizing the Single Index Kit N Set A (10x Genomics, PN-1000212) for sample-indexed libraries. The snRNA-seq libraries were generated using the Library Construction Kit (10x Genomics, PN-1000190) and Dual Index Kit TT Set A (10x Genomics, PN-1000215), following the manufacturer’s recommended protocol. Sequencing was performed by following 10xGenomics recommendations, on the NovaSeq 6000 platform (Illumina) for a targeted depth of 30,000 reads per nucleus.

10x multiome data alignment

The 10x Multiome dataset (ATAC + Gene Expression) was processed using the Cell Ranger ARC pipeline. To analyze the macaque data, we used the Cell Ranger ARC mkref to create a rheMac10 reference and performed a quality check on the output of the Cell Ranger ARC pipeline for downstream analyses.

Preprocessing of snRNA-seq and snATAC-seq multiome data

Seurat (v4.2.1)^54,55 and Signac (v1.8.0)⁵⁶ R packages were used to preprocess and analyze the single nucleus multiome data after alignment. The ten multiome samples across eight ages (PCD85, PCD95, PCD100, PCD105, PCD110, PCD125 (2 samples), PCD145, and PCD155 (2 samples)) consist of 106,728 single nuclei and 35,432 genes (Tables S1 and S2). The “mmulatta_gene_ensembl” dataset of the biomaRt (v2.50.3)⁵⁷ R package was used to convert Ensembl gene ids to gene names. Among 35,432 genes, 21,553 genes were protein-coding and 16,639 genes had unique gene names. Further, the genes not expressed in at least three cells were removed. Additionally, to mitigate the influence of sample sex, we excluded Y chromosome genes and Y chromosome peaks from our calculations, resulting in 15,659 genes.

snRNA-seq data processing

First, the SoupX (v1.6.2) package⁵⁸ was utilized to remove the mRNA contamination due to cell-free ambient RNA that could distort cell type identification and downstream analysis. The SoupX package uses empty droplets to model the ambient RNA expression profiles and return corrected count data. The low-quality nuclei (i.e., a nucleus that has a UMI count of less than 800 and greater than 25,000, ribosomal read greater than 20%, and mitochondrial content greater than 5%) were removed from the downstream analysis (Figure S1A). Second, the DoubletFinder (v2.0.3) R package⁵⁹ was used to predict doublets in each sample separately. After filtering low-quality nuclei, the dataset retained 76,855 nuclei (PCD85: 10,338; PCD95: 8,570; PCD100: 5,317; PCD105: 7,277; PCD110: 10,482; PCD125: 17,633, PCD145: 8,610 PCD155: 8,628). Then, soupX-corrected counts were normalized using Seurat with a scale of 10,000⁶⁰. The top 2,000 highly variable genes were obtained with the default variance stabilizing process in Seurat. Fifty principal components were estimated using the RunPCA() function in Seurat and the dimensionality of the data was identified using an elbow plot. The top 30 principal components were utilized for nonlinear dimensional reduction using Uniform Manifold Approximation and Projection (UMAP) method⁶¹. The RunUMAP() function in Seurat with the option of using umap-learn python package was utilized to estimate UMAP embeddings.

Batch effect corrections of snRNA-seq data

The batch correction across our samples was performed using the Canonical Correlation Analysis (CCA) based approach implemented in Seurat⁵⁵, Harmony (v0.1.1)⁶², and FastMNN⁶³ wrapper functions implemented within SeuratWrappers (v0.3.1) R package. The top 2000 highly variable genes in each sample were identified using the FindVariableFeatures() function in Seurat with default parameters, and the union of these genes was used for the batch correction pipeline (a total of 4,347 genes). In the CCA-based method, Seurat uses highly variable genes to identify integration anchors across samples and provides a corrected gene expression. The Harmony-based UMAP embeddings were obtained by following the same procedure explained earlier but using Harmony components instead of principal components. The FastMNN method used the top 2,000 highly variable genes and provided corrected gene expressions and ‘mnn’ components. The corresponding UMAP embeddings were obtained using the ‘mnn’ components. All three approaches overall exhibited very similar results/characteristics in terms of how they grouped similar cell classes together (Figure S1C). However, the CCA-based approach provides extreme mixing among batches and offers little to no distinction between ages. In contrast, the FastMNN approach provides an apparent variation among ages in the UMAP, which could lead to cell clusters that group cells from a specific age. Since the ages considered here (i.e., PCD 85-155) have the same set of cell subclasses, it is expected to have a moderate amount of mixing among batches. Thus, Harmony based batch correction was utilized for cell clustering, cell type annotations and UMAP visualizations.

Cell clustering and cell type annotations

To infer cell types, transcriptomically similar cells were clustered together using the Louvain clustering method⁵⁵ implemented in Seurat. The shared nearest neighbors in low dimensional space for each cell were identified using the FindNeighbors() function in Seurat with the top 30 Harmony components. The cluster resolution values of 2 and 2.5 (with 42 and 49 clusters, respectively) were considered and the cluster resolution of 2.5 was taken as the optimum resolution. Nuclei were clustered into nine major classes based on the expression of principal marker genes: excitatory neurons (ExN – SATB2), inhibitory neurons (InN – GAD1, GAD2), glial precursor cells (GPCs – EGFR, IGFBP2, MKI67), oligodendrocyte precursor cells (OPCs – OLIG2, PDGFRA), oligodendrocytes (Oligo – MBP), astrocytes (Astro – AQP4), microglia (MG – PTPRC, C1QC, APBB1IP), endothelial cells (Endo – FN1, CLDN5), and vascular leptomeningeal cells (VLMCs – CEMIP, COL1A1)²³. The marker gene expression profiles across the major cell classes clearly delineated cell types throughout all ages in this study (Figure S2A).

To infer excitatory and inhibitory subclasses, the FindSubCluster() function in Seurat was used to recluster excitatory neurons (32 subclusters) and inhibitory neurons (25 subclusters). Excitatory neuron subclasses were defined by expression of canonical cortical layer and projection markers: layer 2-3 intratelencephalic (L2-3 IT – CUX2, PCDH8), layer 3-5 intratelencephalic (L3-5 IT – RORB), layer 2-3/3-5 intratelencephalic (L2-3 IT / L3-5 IT – combinatorial expression of CUX2 and RORB), layer 5 extratelencephalic (L5 ET – POU3F1), layer 6 intratelencephalic (L6 IT – OPRK1), layer 6 corticothalamic (L6 CT – SYT6), layer 6 near-projecting/corticothalamic (L6 NP/CT – combinatorial expression of HTR2C and SYT6), and layer 6B (L6B – NR4A2) (Figure S2B)²³. Inhibitory neuron subclasses were defined by the expression of markers corresponding to their developmental origins within the medial ganglionic eminence (MGE – SOX6, LHX6, MEF2C), caudal ganglionic eminence (CGE – NR2F2, PROX1), and dorsal lateral ganglionic eminence (dLGE – TSHZ1, FOXP2, ETV1) (Figure S2C)¹⁶.

Subtype marker genes were defined by comparing the gene expression of a given subtype with the residual subtypes of its respective subclass and class. The FindMarkers() function in Seurat was used to perform the Wilcoxon Rank Sum test, and genes with expression in more than 25% of cells within the respective subtype, and Bonferroni corrected p-value less than 0.01 were identified as potential subtype marker genes. A subtype-restricted marker gene was selected from this list following a comparison of gene expression levels and expression fold change among all subtypes of the given subclass and class. The final annotation of all 85 subtypes (32 ExN subtypes, 25 InN subtypes and 28 non-neuronal subtypes) derived from this study included the specific class (i.e., ExN), subclass (when applicable, i.e., ExN L2-3 IT), and subtype marker genes (ExN L2-3 IT CUX2 SNTG2) derived from the aforementioned methods (Figure S2E; Table S3).

Evaluation of the clustering stability and cell type separability

The accuracy of the cell type annotation is highly dependent on the capability of the clustering method to identify transcriptomically similar cells. Since the clustering method primarily uses the shared neighbor graphs obtained at reduced dimensions, the cluster stability or the accuracy of the cell type annotations can be determined by investigating their cell neighbors. Thus, a neighbor voting-based approach was adopted²³. First, a selected percentage of total cells were randomly sampled without replacement. Secondly, the gene expression values were scaled, and principal components were estimated using RunPCA() function in Seurat. Next, the 20 nearest neighbors for each cell were identified in the reduced dimensional space. Finally, new cell-type predictions were made using the neighbor voting approach. Here the most abundant subclass label among the neighbors (i.e., 60% or more neighbors) was assigned as the new cell class prediction. The area under the curve (AUC) ROC scores were calculated for each subtype using the original and predicted (based on neighbor voting) cell classes (Figure S2D). In the current study, two subsamples were considered (i.e., samples with 10% and 25% of the original dataset) (Figure S2D).

Cell developmental trajectory and pseudotime inference (snRNA-seq)

Cell developmental trajectories and pseudotimes for snRNA-seq were calculated using Monocle3 (v1.3.1)⁶⁴ (Figures 2A–2B). Developmental trajectories of ExN L2-3IT and ExN L3-5IT excitatory subclasses were compared. The Monocle3 function, order_cells() was used to infer pseudotime values by projecting each cell onto a principal graph with a given starting point (i.e., a root cell/group). The selection of the root cell could be done in two ways. One would be to manually select a starting trajectory node based on the prior knowledge and expertise of the user, and the second and more acceptable method is to choose a node that neighbors most of the early timepoint cell groups. We used the latter method, and the cells derived from the earliest timepoint analyzed in this study, PCD85, were considered the root cell group for pseudotime calculations. Moran’s I test implemented in the graph_test() function in Monocle3 was utilized to obtain differentially expressed genes along the developmental trajectory (Figures 2C–2D; Table S4).

Integration of snRNA-seq data with published datasets

Comparisons were made with our data and two published snRNA-seq datasets of embryonic/early fetal (PCD37-110)²⁸ and adult²³ rhesus macaque dorsolateral prefrontal cortex to investigate the developmental transitions from early-fetal to adult. Due to the prominent batch effects among these datasets, the Harmony batch correction method was utilized to remove batch effects and create an integrated UMAP across all three datasets (Figures 1E–1G). A unified cell type annotation across all three datasets was created by annotating the major class (i.e. ExN, InN, Astro, etc…) of all identified clusters.

snATAC-seq data processing

The Signac (v1.8.0)⁵⁶ R package was used to analyze snATAC-seq data. Ensembl 105 EnsDb for the Macaca mulatta NCBI database was used for annotations and reference genomic ranges. First, genomic ranges for each batch were obtained, and the reduce() function in Signac was used to convert them to a common genomic range, allowing proper comparisons across samples. Similar to snRNA-seq, low-quality nuclei were identified and filtered based on per-cell quality control metrics such as ATAC peak count, the strength of the nucleosome banding pattern, and transcriptional start site (TSS) enrichment scores. Nuclei with ATAC count greater than 100,000 and less than 1000, nucleosome signal greater than 4 and TSS enrichment lower than 2 were removed, resulting in 64,941 nuclei across 6 ages (PCD85: 9,660; PCD95: 8,042; PCD105: 6,328; PCD110: 9,480; PCD125: 17,364; PCD145: 6,558; PCD155: 7,509) (Figure S1D). It should be noted that the PCD100 sample was excluded from our multiome calculations as less than 700 nuclei passed our quality control metrics.

Batch effect corrections of snATAC-seq data

Similar to snRNA-seq processing steps, the potential batch effects in the snATAC-seq data samples were investigated. The data were normalized using the term frequency inverse-document frequency (TF-IDF) method⁵⁵, and the latent semantic indexing (LSI) components were calculated. The top 2-30 LSI components were used to obtain UMAP embeddings. The resultant UMAP showed significant levels of batch effects across samples. Seurat and Harmony were used to remove batch effects (Figures S1G–S1H). In the Seurat-based approach, the LSI components were used with the “reciprocal LSI” approach instead of CCA to find integration anchors among samples. Similarly, the LSI components were utilized for the RunHarmony() function in the Harmony approach. The cell subclass UMAP visualizations for the reciprocal LSI and Harmony methods show similar characteristics except for some minor mixing of InN dLGE and ExN L5 ET cell types in the reciprocal LSI approach. Thus, the Harmony based batch correction method was utilized for visualization purposes.

Joint UMAP visualization

A joint UMAP embedding (Figures 3A and S1F), characterized by both transcriptome and epigenome was obtained using the weighted nearest neighbor method implemented in Seurat. Using the FindMultimodalNeighbors() function in Seurat, we estimated a joint neighbor graph that used the top 50 harmony components calculated in snRNA-seq and snATAC-seq. UMAP embeddings were calculated using the weighted nearest neighbor graph.

Peak calling

The accuracy of the chromatin accessibility analysis relies heavily on precise peak calling, particularly because peak calling in the entire cell population may inadvertently eliminate peaks that are specific to smaller cell populations⁵⁶. Therefore, CallPeaks() function in Signac was used to identify distinct peaks using the MACS2 method⁶⁵ after grouping across cell subclasses to preserve peaks related to all subpopulations.

Inferring cell-type peak-to-gene links

Here, we estimated peak-to-gene links (i.e., peaks that are correlated with gene expression of a given gene) across inhibitory and excitatory neuronal cell types across developmental stages to examine characteristics of regulatory elements at different developmental stages. Peak-to-gene links within the genomic distance of 500 kb from the corresponding gene’s TSS were obtained using the LinkPeaks() function in Signac. The expected coefficient values for the peak were then used to compute a z-score and p-value. Only the high-confidence peak-to-gene links were retained using the thresholds of p-value < 0.05 and the absolute value of z-score > 0.05. Overall, 57,698 peaks-gene links, involving 35,016 peaks and 9,948 genes, were identified across cell types and developmental stages.

Considering the sparsity of these data and the complexity of handling a large number of cells, pseudo-cells were utilized instead of individual cells. The nuclei were clustered using the Louvain clustering method in Seurat per gestational age and cell type using the top 50 Harmony features. Then, each inferred cluster was considered as a pseudo-cell (total of 271 pseudo-cells) and used their average gene expression and chromatin accessibility for visualization.

In order to further narrow down the number of peak-to-gene links, we calculated differentially expressed genes and peaks between five aging groups (Early-midfetal PCD85-95, Midfetal PCD105-110, and three separate late-fetal groups, PCD125, PCD145, and PCD155), as well as between excitatory and inhibitory neuronal cells. It was done using the FindMarkers() function in Seurat. The top 200 differential expressed peaks for each aging group and top 100 differentially expressed genes were isolated (resulting in 128 peak-gene links) for visualization and downstream analysis (Figure S4A; Table S5). A heatmap containing the gene expression and chromatin accessibility of pseudo-cells (columns) for each unique peak-to-gene link (rows) was depicted in Figure S4A. Since a peak may be linked to several genes and vice versa, multiple rows may correspond to a single gene or peak in the heatmap. High variability in clustering the associated peak-gene links, corresponding to excitatory (ExN) and inhibitory (InN) cell types, as well as within the five distinct age groups, were observed. Distinct gene expression and chromatin accessibility patterns across cell types and developmental stages were identified and utilized for further analysis (Figures 4B–4D; Table S5).

Transcription factor binding motif enrichment analysis

Regulatory elements specific to cell types and brain development were analyzed by calculating differential accessibility and overrepresented transcription factor (TF) binding motifs, specific to the differential peaks. Comparisons were made using ExN L2-3 IT and ExN L3-5 IT excitatory subclasses. Due to the unavailability of macaque motifs, human motifs from the JASPAR2020 (v0.99.10) database⁶⁶ were used for our analysis together with the macaque genome (BSgenome.Mmulatta.UCSC.rheMac10 (v1.4.2). The differential accessibility peaks between ExN L2-3 IT and ExN L3-5 IT excitatory subclasses were obtained using the FindMarkers() function in Seurat, with test.use = ‘LR’ and latent.vars = ‘nCount_ATAC’. Peaks with adjusted p-value (Bonferroni-corrected) less than 0.001 and log₂(fold change) greater than 0.5 were identified as differentially accessible peaks and utilized for motif enrichment. A hypergeometric test was then used to test the probability of observing a particular motif at a given frequency while comparing it with a set of background peaks matched for GC content. The background peaks were identified using AccessiblePeaks() and MatchRegionStats() functions in Signac. The AccessiblePeaks() function filters out the differentially accessible peaks identified using FindMarkers() function by keeping only the peaks open in specified cell types (i.e., ExN L2-3 IT and ExN L3-5 IT subclasses in our example). Then, the MatchRegionStats() function creates a GC content-matched set of peaks. Once the background peaks were identified, they were used to identify enriched motifs using the FindMotifs() function in Signac (Table S6). RunChromVAR() function in Signac (wrapper for chromVAR (v1.16.0)⁶⁷) was utilized to estimate motif activity score for each cell. It provides a complementary approach to identifying overrepresented motifs across cell populations (Figure 3D).

Transcription factor footprint analysis

MA0774.1 (MEIS2 motif) is one of the examples of overexpressed motifs in ExN L2-3 IT cells compared to ExN L3-5 IT (Figure 3C; Table S6) and was chosen for the downstream analysis. The locations of active MEIS2 binding sites were explored using the Footprint() function in Signac, which inquires about the overlaps between binding sites and open chromatin regions in each cell. Then the motif binding site footprints were visualized using the PlotFootprint() function in Signac. In the current study, the footprint visualizations were done by grouping nuclei according to the cell subclass (i.e., ExN L2-3 IT, ExN L3-5 IT) and age ranges (i.e., PCD85-95, PCD105-110, PCD 125, and PCD145-155) to investigate the binding site enrichment across cell subclasses and maturation (Figure 3C).

Linking potential transcription factors and peak-to-gene links to target genes at the cell type level

As shown in Figure 3E, gene NFIA was chosen as an example target gene for MEIS2 binding as it was previously reported in the development of retinal progenitor cells³⁵. First, the availability of MEIS2 binding sites upstream of the NFIA transcription start site (TSS) was investigated. Five potential binding sites within 200 kb upstream of NFIA TSS were identified (the highlighted regions in Figure 3E). Then gene-to-peak links for gene NFIA were estimated to identify peaks that are highly correlated with the NFIA expression. This was done by using the LinkPeaks() function in Signac⁶⁸, which considered the peaks within 200 kb from TSS. Multiple links were identified (Figure 3E), and two of the highly correlated links overlapped with MEIS2 binding sites. Thus, these links were identified for further analysis.

Inference of cell-type gene regulatory networks in development using multiomic data

Further extending the analysis on regulatory elements, we inferred cell-type gene regulatory networks (GRNs) using transcriptomic and epigenomic multiomic data to link TFs with binding sites on regulatory elements to target genes. Our computational pipeline for inferring such cell-type GRNs has the following steps:

Step 1. Identifying TFBS motifs related to a specific transcription factor (we considered JASPAR 2020 – human TFBS motifs).

Step 2. Identifying TFBSs in the upstream open chromatin regions (200 kb from TSS) of all possible target genes. Here, we considered 200 kb distance from TSS to include both proximal and distal regulatory elements. It should be noted that the transcription directions of the genes were considered in identifying the upstream binding region.

Step 3. Matching the directions of gene transcription and binding motifs. This step filters out the TFBSs that have an opposite direction to the transcription direction.

Step 4. Identifying peak-to-gene links – see the inferring cell-type peak-to-gene links section for further details (i.e., peaks that are correlated with gene expression of a given gene).

Step 5. Overlap the peaks identified in Step 3 and Step 4. Suppose there is an overlap between these two sets of peaks (i.e., potential open chromatin regions that have the potential of binding a specific TF and peaks correlated with a particular target gene, TG). In that case, we establish a regulatory link between the TF and TG.

We inferred cell-type GRNs for ExN L2-3 IT and ExN L3-5 IT cells following the steps mentioned above. We limited our analysis to the highly variable genes (576 genes) along the maturation trajectory identified in snRNA-seq cell type trajectories (Figure 2; Table S4). Nine transcription factors (PBX1, MEF2C, FOXP2, NFIA, NFIB, GLIS3, CUX2, MEIS2 (two motifs), POU6F2) among these highly variable genes were utilized considering their availability of binding site motifs in the JASPAR 2020 database. The cell type specificity of these TFs was investigated using their log₂(fold change) across ExN L2-3 IT and ExN L3-5 IT cells (i.e., ExN L2-3 IT specific if log₂(fold change) change > 1, ExN L3-5 IT specific if log₂(fold change) < 1, otherwise conserved in both cell types). We found that CUX2, MEIS2 were specific to ExN L2-3 IT and FOXP2, POU6F2 were specific to ExN L3-5 IT cells. Figure 3B depicts the differential GRN across two cell types. It should be noted that only the autism-related genes (https://gene.sfari.org/database/human-gene/) were depicted in the figure. The cell type specificity of regulatory elements was visualized based on the edge color (red – ExN L2-3 IT specific, blue – ExN L3-5 IT specific, and black – conserved across cell types).

Patch-seq - Patch clamp recording

All surfaces and equipment were thoroughly cleaned and treated with RNaseZap RNase Decontamination Wipes (Thermo Fisher Scientific Inc., AM9786) before experiments to ensure sample integrity. All solutions were prepared using DEPC-treated water. The patch clamp recording was performed as previously described⁶⁹. Briefly, slices were transferred to a recording chamber (Warner Inc.) perfused with 95%O₂ / 5%CO₂ saturated aCSF (124 mM NaCl, 2.5 mM KCl, 2.5 mM CaCl₂, 1.2 mM MgCl₂, 1.25 mM NaH₂PO₄, 26 mM NaHCO₃, and 15 mM glucose. pH 7.35). Slices were visualized with an Olympus BX51WI microscope and infrared differential interference contrast (IR-DIC) optics. Patch pipettes were pulled from thick-wall filamented borosilicate glass (Sutter Instruments, BF150-86-10) using a Sutter Micropipette Puller (Sutter Instruments P-1000) to a tip resistance of 5-8 MΩ, and filled with the internal solution, which contains : 140 mM K-gluconate, 7.5 mM KCl, 10 mM HEPES-K, 0.5 mM ethylene glycol-bis(b-aminoethyl ether)-N,N,N’,N’-tetraacetic acid (EGTA)-K, 4 mM Mg-ATP, 5 mM Li-GTP, and 0.25 U/μl RNase inhibitor (Takara, 2313A) (pH 7.2 titrated with KOH). Whole-cell patch clamp recordings were made using a Multipatch 700B preamplifier, digitized with a Digidata 1440A, and acquired with pCLAMP 10 (Molecular Devices, USA). To be included in the analysis, a cell needed to have a >1 GΩ seal before whole-cell mode and an initial access resistance <20 MΩ and <15% of the input resistance (R_in). In the case of acute brain slices, the targeted cells were in the dorsolateral prefrontal cortex (dlPFC). Conversely, for recordings from cultured slices that had been infected with lentivirus, only cells displaying either mNeon or mScarlet fluorescence were selected and subjected to recording.

After the formation of a stable whole-cell patch clamp, the resting membrane potential (RMP) of the cell was recorded. Each cell was recorded using a standardized stimulus paradigm, including voltage and current clamps, in order to extract features that could be compared across cells. Subthreshold features, including the input resistance (R_in), membrane time constant (τ), and membrane capacitance (C_m), were obtained for all cells. The R_in and C_M are calculated from voltage clamp recording data because the hyperpolarized response of immature neurons in current clamp mode is unstable. To calculate Rin, a series of small voltage steps without eliciting voltage-gated ion currents were applied, and the stable current responses were measured. The ΔV to ΔI was fitted linearly to calculate the cell’s R_in. We also use step response analysis to calculate the C_m. The decay phase of the transient current induced by a small voltage step is fitted with an exponential function to obtain the decay time constant (not the neuron’s time constant) and calculate the charge mediated by this process, from which the cell membrane capacitance is calculated⁷⁰. Most of the cells recorded from the early stage of development (earlier than PCD110) did not exhibit typical action potentials. For cells showing action potentials, the amplitude, width (at half-height), and dV/dt were calculated. Additionally, the amplitude of depolarization-induced sodium current (I_Na) was also measured. A series of depolarizing voltage steps were applied, and the amplitude of the inward current was measured. The I_Na was expressed as the current density, which is the ratio of the peak current amplitude to the cell membrane capacitance. Whole-cell and pipette capacitance were corrected for the I_Na recording. Most of the cells recorded from the early stage of development (earlier than PCD110) did not exhibit typical action potentials. For cells showing action potentials, the amplitude, width (at half-height), and dV/dt were calculated. Additionally, the amplitude of depolarization-induced sodium current (I_Na) was also measured. For the analysis of RAPGEF4 knockdown cells, the cells must exhibit non-zero I_Na to ensure that the analyzed cells are indeed neurons, while for CHD8 knockdown cells, we analyze all ExNs with reliable annotations.

Statistical analysis was conducted using GraphPad Prism software. Ordinary one-way ANOVA was utilized to analyze the data across four age groups, followed by multiple comparisons. For the statistical analysis comparing two groups in electrophysiological and morphological data, normality was assessed using the Shapiro-Wilk test. If the data were not normally distributed, comparisons were made using the Mann-Whitney test. For normally distributed data, comparisons were made using a two-tailed unpaired t-test..

Patch-seq – SMART-seq

After completing electrophysiological recordings, the recording pipette was carefully repositioned to the center of the soma, followed by the application of slight negative pressure to extract cytosol and/or nucleus. Subsequently, the pipette was gradually withdrawn along the x and z axes. Once the pipette was outside the tissue slice, individual PCR tubes containing 6 μl of patch-seq lysis buffer (5% Polyethylene Glycol 8000, 0.1% Triton X-100, 0.5 U/μl RNAse Inhibitor, 0.5 μM OligodT30VN, and 0.5 mM dNTPs) were used to collect the expelled nucleus and a minimal amount of cytosol present at the pipette tip. The tube was immediately frozen on dry ice to preserve the cell content. The collected samples were subsequently preserved for future use by storing them at a temperature of −80 °C. The Smart-seq3 protocol was employed in combination with the patch-clamp technique for Patch-seq analysis⁷¹. The lysates were incubated at 72 °C for 10 min to release the cellular content into the buffer. Reverse transcription was then performed with a master mix containing 100 mM Tris-HCl, 120 mM NaCl, 10 mM MgCl2, 4 mM GTP, 32 mM DTT, 2 U/μl RNAse Inhibitor, 8 μM TSO, 8 U/μl Maxima H-minus RT enzyme and adjusted to pH 8.0. The samples were subjected to reverse transcription under the following program: initial reverse transcription at 42 °C for 90 min, followed by 10 cycles of further reverse transcription at 50 °C for 2 min and 42 °C for 2 min. An enzyme deactivation step was performed at the end of the program with an 85 °C incubation for 5 min. The resulting cDNA molecules were amplified through PCR using a final reaction concentration of 1X Kapa HiFi HotStart buffer, 0.3 mM dNTPs, 0.5 mM MgCl2, 0.5 μM Forward primer, 0.1 μM Reverse primer, and 0.02 U/μl polymerase. The cDNA library amplification was conducted in a thermocycler, and the purified cDNA libraries were evaluated for size distribution using the Agilent 2100 Bioanalyzer and the Agilent High Sensitivity DNA Kit (Agilent Technologies Cat, 5067-4626) and quantified using the Qubit dsDNA HS kit (Thermo Fisher Scientific, Q32851). The final library synthesis was carried out only for cDNA libraries meeting the following two criteria: (1) a distinct peak at approximately 2000 bp indicating clear enrichment and (2) a proportion of DNA fragments greater than 300 bp comprising at least 35% of all DNA fragments in the libraries. The Nextera XT DNA Library Preparation Kit (Illumina, FC-131-1096) was employed for tagmentation, and Phusion High-Fidelity DNA Polymerase (Thermo Scientific, F530L) was used for PCR amplification with the Nextera Index primers (IDT^® for Illumina^® DNA/RNA UD Indexes, illumina, 20027213, 20027214, 20042666, and 20042667). The amplified DNA libraries were purified using AMPure XP beads (Beckman Coulter, A63881). The assessment of the size distribution of DNA libraries was carried out utilizing the Agilent TapeStation 4200, which employed the High Sensitivity D5000 ScreenTape (5067-5592) and High Sensitivity D5000 Reagents (5067-5593). Additionally, the quantification of the libraries was performed using the PicoGreen^® dsDNA reagent kit (Molecular Probes catalog, P-11496). Subsequently, the DNA libraries were subjected to sequencing using the Novaseq 6000 platform, utilizing paired-end reads of 150 base pairs in length.

Patch-seq data preprocessing

The patch-seq single cell RNA-seq reads from human were aligned to reference genome GRCh38 and the reads from rhesus macaque were aligned to reference genome rheMac10. Trim Galore was used for the adaptor trimming. We used STAR (v2.7.2b)⁷² in two pass Mode for the alignment and removed PCR duplicates. STAR quantMode was used for gene counts on uniquely aligned reads, and the intergenic regions were excluded.

The same preprocessing steps as in snRNA-seq were followed to preprocess human and macaque patch-seq’s single-cell RNA-seq data (Smart-seq3). However, minor modifications were made in the preprocessing pipeline of the macaque patch-seq cells. In the 10x snRNA-seq datasets, the genes that did not have a unique gene name were removed from the analysis. The ‘mmulatta_gene_ensembl’ dataset in the biomaRt (v2.50.3)⁵⁷ package comprises 16,480 unique protein-coding gene names. Due to the differences in sequencing depths between 10x multiome data (~25,000) and Smart-seq3 in patch-seq (8 million), removing gene features that did not have a unique gene name became more erroneous in patch-seq data than in multiome data. The percentage of UMI counts from genes that did not have a gene name in 10x snRNA-seq has a mean value of 2.82% (s.d 0.96%), while in Smart-seq3 has a mean value of 7.26% (s.d 2.78%). Thus, all protein-coding genes were kept for downstream analysis, regardless of whether it has a gene name. The patch-seq data comprised cells from postconceptional days (PCD) 85, 90, 95, 100, 105, 110, 125, 145, and 155 and were sequenced in four batches.

Patch-seq cell type annotation

Patch-seq cell types were inferred using the label transfer method implemented in Seurat and with our 10x snRNA-seq data as the reference. The method identifies anchor cells between the query and reference datasets using the FindTransferAnchors() function based on canonical correlation analysis. Then the MapQuery() function in Seurat was used to transfer cell types and create a projection UMAP. It creates a score matrix (query cells × the number of cell types) that each score would provide how likely a particular cell belongs to a particular cell type. These scores are based on the identities of the neighboring anchors to a query cell. Even though this method has proven to provide accurate label transfers, a discrepancy between the assigned cell type and their location of the projected UMAP was observed. This was due to the smaller size of the query dataset (i.e., not enough anchors to make an accurate prediction). Thus, a minor modification was made in the label transfer step. Instead of assigning cell types based on neighboring anchors, the new method projects the query dataset onto the reference UMAP using projected UMAP embedding. Then for each query cell, the closest neighbors from the reference cells were identified, and the abundant cell type of the neighbors was assigned to the query cell.

Since the transcriptome of cells in cultured slices is often significantly different from that of cells in acute slices, special care needs to be taken when annotating cells in cultured slices. To remove the influence of transcriptomic differences due to culture, differentially expressed genes in cells from cultured slices were identified and removed from the cell type annotation. A list of removed genes from monkey and human cultured datasets was provided in (Table S8).

Trajectory inference of Patch-seq data

The developmental trajectory for the acute, excitatory neuron IT patch-seq cells (i.e., ExN L2-3 IT, ExN L3-5 IT, ExN L2-3 IT/L3-5 IT, and ExN L6 IT, a total of 111 cells) was obtained using the Monocle 2 (v 2.26.0)⁷³ with default parameters. Monocle 2 was chosen for trajectory inference considering the limited number of cells. First, the top 2000 highly variable genes were identified and scaled using FindVariableFeatures() and ScaleData() functions in Seurat. Then, the scaled gene expression of the highly variable genes was imported into Monocle 2 and variable genes across developmental trajectory were identified using the differentialGeneTest() function. Here, we used individual postconceptional days as separated age groups to obtain the highly variable genes. Differentially expressed genes that were expressed in at least 10% of the cells (i.e., 11 in our analysis) and a q-value cutoff of 0.01 were identified and used for trajectory inference (Table S8). Dimensionality reduction was performed using the reduceDimension() function with the DDRTree method and arguments of norm_method = ‘none’ and pseudo_expr = 0). The resulting developmental trajectory is visualized in Figure 4C.

Correlation analysis between gene expression and electrophysiological properties

Gene expression data and paired electrophysiological feature data were extracted from the same set of cells (111 IT neurons from acute slice Patch-seq data). Subsequently, correlation tests were performed to analyze the relationship between the gene expression levels and the corresponding electrophysiological features. In brief, the gene expression data for each gene within the entire set of cells was extracted as LogNormalize values. Additionally, the electrophysiological feature data obtained from patch clamp recordings, including membrane capacitance (C_m), input resistance (R_in), resting membrane potential (RMP), and sodium current (I_Na), were extracted as paired electrophysiological feature data from the same set of cells. Due to the zero-inflated and thus non-normally distributed gene expression data, we performed a non-parametric method, Kendall correlation analysis, to assess the associations between gene expression levels and electrophysiological features. We applied the Bonferroni correction method to control the familywise error rate. Genes were then ranked based on their absolute correlation coefficients. We selected genes with adjusted p-values below the significance threshold (p < 0.05) as those significantly correlated with specific electrophysiological features. In total, 39 genes were identified as correlated with Cm, 129 with Rin, 224 with RMP, and 227 with INa, respectively (Table S9).

GO enrichment analysis

We conducted Gene Ontology analyses with Metascape (version v3.5.20230501, available at https://metascape.org/)⁷⁴ using “Homo sapiens” as the selected species and “all genes” as the background gene set. In the snRNA-seq dataset, we conducted separate differential gene expression analyses for the ExN L2-3 IT and ExN L3-5 IT cell subclasses along their maturation trajectories. We identified the top 250 differentially expressed genes along these trajectories for subsequent GO analyses. Enriched GO terms for these top genes were determined and a comprehensive compilation of the differentially expressed genes, background genes, and the results of the GO term enrichment analysis for both ExN L2-3 IT and ExN L3-5 IT cells in Table S4. For GO analyses with electrophysiological feature correlated genes, we utilized genes that exhibited correlations with each specific electrophysiological feature as our input genes. For the GO analysis, we chose four databases: “GO Molecular Functions”, “GO Biological Processes”, “GO Cellular Components” and “DisGeNET”. We subsequently focused on the top 50 GO terms with the highest “−LogP” values for additional analysis.

In Figure S3A, which presents the heatmap for peak-to-gene linkages, we have discerned three distinct gene expression-chromatin accessibility patterns associated with ExN PCD85-95, ExN PCD155, and InN PCD155. We subsequently utilized the corresponding gene sets within each of these patterns to conduct the GO enrichment analysis. It is worth noting that we employed all the genes referenced in this study as our background genes, keeping all other parameters at their default settings. Detailed information concerning these gene sets, as well as the pathways and functions they represent, is available in Table S5. Furthermore, visual representations of the pathways and functions are available in Figures S3C–S3E.

Gene manipulation in primary neurons isolated from human fetal cortical tissue

On day 5 post-isolation (5 DIV), the primary neurons were infected with lentivirus. On 25-29 DIV, the virus-infected cells were used for patch-clamp recording. Both the experimental group and correspondent control group were recorded on the same day.

Transcriptional activation of CHD8 in human neurons differentiated from pluripotent stem cells

Four days after replating, the differentiated neurons were infected with the guide RNA lentivirus (LV-CHD8 gRNA) together with control lentivirus (LV-mScarlet). For morphological analysis, neurons were plated at a density of approximately 316,000 cells/cm² in Poly-D-Lysine (Sigma-Aldrich, P7886)/Cultrex-coated glass coverslips and collected after 2 weeks. For electrophysiology analysis, neurons were plated at a density of approximately 474,000 cells/cm² and mouse astrocytes were added (~21,000 cells/cm²) after 1 week of differentiation in Neurons medium supplemented with 10% FBS (R&D Systems, S12450). Fresh media change was performed after 2 days of astrocyte plating. The virus-infected cells were used for electrophysiology recordings 14-17 days after replating.

For morphological analysis, neurons cultured on Poly-D-Lysine (Sigma-Aldrich, P7886)/Cultrex (R&D Sytems, 3433-005-01) coated glass coverslips were fixed with 4% PFA for 10 min at room temperature and then washed with PBS. Neurons were blocked with PBS⁺⁺ (PBS containing 5% goat serum and 0.2% Triton X-100) for 1 h at room temperature. Primary antibodies were diluted in PBS⁺⁺ and incubated overnight at 4 °C. Samples were washed three times with PBS and incubated with Alexa Fluor-conjugated secondary antibodies (Thermo Fisher Scientific) diluted in PBS⁺⁺ for 60 min at room temperature. After three washes with PBS, neurons were counterstained with a nuclear counter stain, DAPI (4’, 6-diamidine-2’-phenylindole dihydrochloride, 1:2000, Roche Applied Science, Indianapolis, IN). Samples were then washed twice prior to being coverslipped in DAVCO-PVA.

Differential gene expression analysis of excitatory neurons with CHD8 knockdown

EBseq v1.14⁷⁵ was employed to determine differentially expressed genes between 55 LV-shNC and 42 LV-shCHD8-infected ExNs using Patch-seq data. Median normalization was applied to genes with counts greater than or equal to 200 across all 97 ExNs, reducing the number of genes from 19,021 to 14,753. Differential expression analysis was specifically conducted between ExNs subjected to two different groups: shNC and shCHD8. In this analysis, a modified form of median normalization, known as median-by-ratio normalization, was implemented on the raw aggregated counts. This adjustment aimed to account for a prevailing trend of at least one zero count per gene within this dataset, which was deeply sequenced single-cell analysis. The EBSeq package’s MedianNorm() function was utilized with the “alternative=TRUE” option to apply this adjusted median normalization. This modified median normalization approach yielded results that were more interpretable. Specifically, it led to greater divergence between up- and down-regulated genes when visualized in a −log₁₀(P) vs. log₂(fold change) volcano plot, in comparison to the standard median normalization typically used for bulk sequencing data. In this analysis 20 full iterations of the EBSeq algorithm were applied, instead of the default 5 iterations. DE genes were selected based on significance levels of P < 0.05 (or PPDE ⩾ 0.95, following the recommendation of EBSeq authors) and a log fold change (LFC) magnitude of at least log₂(1.5), selecting 5,279 DE genes.

As depicted in Figure 7K, the DE genes include several identified as correlated with both RMP and R_in electrophysiological features. Within our DE gene set, we identified 32 genes exhibiting RMP-correlation and 30 genes displaying R_in-correlation. Additionally, 20 genes correlated with both RMP and R_in electrophysiological features. Notably, one of these genes is RAPGEF4, which is correlated with three electrophysiological features: RMP, C_m and I_Na. Figures 7L–7O display violin plots of Seurat-normalized data (log-transformed with scale=10,000), providing a visual representation of the differential expression patterns for four specific genes of interest between shNC and shCHD8 ExNs.

In situ hybridization (ISH) and immunohistochemistry (IHC)

For multiplex fluorescent ISH, the RNAscope Multiplex Fluorescent Reagent Kit v2 Assay was used (ACD, Bio-Techne, 323110). The manufacturer’s pretreatment protocol for fixed frozen tissue was carried out with modifications (ACD, Bio-Techne, document no. 323100-USM). Sections were rinsed in PBS, fixed for an additional 48 h in 10% neutral buffered formalin, dehydrated in progressive solutions of 50%, 70%, and 100% EtOH, and air-dried for 15 min at room temperature. Antigen unmasking was performed with the Retriever 2100 (Electron Microscopy Services) using buffer AG (EMS, 62706-10). Sections were rinsed in PBS and incubated in 3% hydrogen peroxide for 20 min at room temperature. Primary antibody was diluted in RNAscope Co-detection Antibody Diluent (ACD, Bio-Techne, 323180) and incubated for 48 h at 4 °C. Sections were washed in PBST (1X PBS + 0.3% Triton X-100) and post-fixed in chilled 10% neutral buffered formalin for 30 min. Sections were then washed in PBS and incubated in RNAscope Protease Plus (ACD, Bio-Techne, 32330) for 10 min at 40 °C. RNAscope multiplex fluorescent v2 detection protocol was performed with modifications (ACD, Bio-Techne, document no. 323100-USM). All washes were performed with PBST (1X PBS + 0.3% Triton X-100). Tyramide signal amplification was performed with TSA Plus Fluorescein (Akoya Biosciences, NEL741001KT) or TSA Plus Cy3 (Akoya Biosciences, NEL744001KT) diluted 1:200 in RNAscope Multiplex TSA Buffer. Following the RNAscope detection protocol, sections were incubated with secondary antibodies (Jackson ImmunoResearch Labs) diluted 1:250 in RNAscope Co-detection Antibody Diluent for 1 hour at room temperature. Sections were washed, treated with Autofluorescence Eliminator Reagent (Millipore, 2160) according to manufacturer instructions, and cover-slipped with Vectashield Plus Antifade Mounting Medium (Vector Laboratories, H-1000).

DNA constructs

The lentiviral shRNA constructs were generated by cloning shRNA cassettes into a lentivector, as previously described⁷⁶. The shRNAs were designed in “Stem-Loop” structures, where the shRNA sequence formed the “stem” and a seven-nucleotide sequence “TCAAGAG” served as the loop sequence. The U6 promoter, along with the complete “Stem-Loop” structure, was cloned as the U6-shRNA cassette. For the creation of LV-mNeon and LV-mScarlet constructs, the CMV-GFP cassette of the lentivector was replaced with the EF1α promoter driven mNeon and EF1α-mScarlet cassettes, respectively. Subsequently, the U6-shNC cassette was inserted into LV-mNeon to generate LV-shNC-mNeon, and the U6-shRAPGEF4 cassette or the U6-shCHD8 cassette were inserted into LV-mScarlet to generate the specific shRNA constructs. All shRNAs were designed to target the conserved sequences between rhesus macaque and human. The shRNA sequences for RAPGEF4, CHD8 and negative control (shNC) are listed as follows:

shRAPGEF4-1: 5′-GCCCACTGTAAGGAGTATAAA-3′

shRAPGEF4-2: 5′-ATACGCCTTGAGGCAACTTTA-3′

shRAPGEF4-3: 5′-CGGAGTTATGTACGGCAATTA-3′

shRAPGEF4-4: 5′-GAGTTATGTACGGCAATTAAA-3′

shRAPGEF4-5:5′-CAAGCGTGTTCAGCTATTAAA-3′

shRAPGEF4-6:5′-CACATTTGGAAGGCATAATTT-3′

shCHD8-1:5′-GCAGATTGGATCCGGAAATAT-3′

shCHD8-2:5′-CCGTGAAGCTTGCCATATTAT-3′

shCHD8-3: 5′-TAGACATCCTAGAGGATTATT-3′

shNC: 5′-GGAATCTCATTCGATGCATAC-3′⁷⁶

The sequences of the shRNA constructs were confirmed by sequencing. To validate shRNA knockdown efficiency, HEK293T cells were transfected with individual shRNA and shNC. After 48 h, cells were harvested for RNA isolation using TRIzol reagent, followed by reverse transcription and quantitative real tme PCR (qPCR). The relative expression of RAPGEF4 or CHD8 was assessed by comparing shRNA-transfected cell samples and the shNC control. Statistical significance was evaluated using an ordinary one-way analysis of variance (ANOVA) with multiple comparisons, employing shNC as the control, and analyzed using Graphpad Prism software. The most efficient knockdown shRNA was selected for further experiments (Figures S6B–S6C and S7A–S7B).

An adjacent ATAC-seq peak region proximal to NFIA was identified to harbor two potential MEIS2 binding sites. To investigate whether MEIS2 regulates the activity of this genomic region through these putative binding motifs, we opted to conduct a dual luciferase assay. This entailed the creation of a MEIS2 overexpression construct and four distinct luciferase vectors, each driven by a different variant of the DNA sequence from this region, chr1:163,554,484-163,555,125: A luciferase vector incorporating the unaltered DNA sequence, facilitating firefly luciferase expression; A luciferase vector harboring the same DNA sequence with the first potential MEIS2 binding motif mutated, driving firefly luciferase expression; A luciferase vector with the same DNA sequence, wherein the second potential MEIS2 binding motif was mutated, driving firefly luciferase expression; A luciferase vector bearing the same DNA sequence with mutations in both potential MEIS2 binding motifs, facilitating firefly luciferase expression.

The MEIS2 overexpression construct was created by insertion of the EF1α -MEIS2 cassette into a lentivector. The genomic region, chr1:163,554,484-163,555,125, encompassing the aforementioned potential MEIS2 motifs and associated with the adjacent NFIA ATAC-seq peak, was cloned into a pGL4.23[luc2 minP] firefly luciferase vector (Promega, E8411) using XhoI and HindIII restriction sites. To generate the first MEIS2 binding motif mutation construct, we substituted the core sequence “TGACAG” within the cloned peak region with “AGTCTG”. Similarly, for the second MEIS2 binding motif mutation construct, we replaced the core sequence “CTGTCA” with “CAGACT”. The fully mutated MEIS2 binding motif construct, involving mutations in both the first and second MEIS2 binding motifs, was generated by incorporating both of the aforementioned mutations

To assess the specificity of shRAPGEF4 in neurons (Figure S6), we created a LV vector, Lenti-shRAPGEF4 (Lenti-U6-shRAPGEF4-CAG-DIO-GFP) expressing U6 promoter-shRAPGEF4 (or Lenti-shNC[(Lenti-U6-shNC-CAG-DIO-GFP)]) and CAG promtoer-driving CRE-dependent (DIO) GFP. To express exogenous RAPGEF4 in either control or RAPGEF4 knockdown human neurons, we created a LV-RAPGEF4-IRES-CRE vector that expresses EF1α promoter driving the RAPGEF4 with a silent mutation at the shRAPGEF4 targeting sites, preventing shRAPGEF4 from targeting the exogenously expressed RAPGEF4. Double infection of neurons with LV-U6-shRAPGEF4-CAG-DIO-GFP and LV-RAPGEF4-IRES-CRE leads to exogenous RAPGEF4 expression specifically in neurons with RAPGEF4 knockdown. Similarly, to determine whether exogenous RAPGEF4 could rescue CHD8 deficiencies, we creaed a LV-shCHD8 lentiviral vector expressing both shCHD8 and CRE-dependent GFP expression (Lenti-shCHD8 [Lenti-U6-shCHD8-CAG-DIO-GFP]).

To transcriptionally activate CHD8 expression in neurons, we designed and screened several LV guide RNAs (LV-sgCHD8-mScarlet) and selected the most efficient one for CHD8 transcriptional activation. The EF1α-driven mScarlet lentivirus vector (LV-mScarlet), lacking the U6-guide RNA cassette, was used as a control. These LV were used to infect a dCas9-Activator human pluripotent stem cell line (dCas9-A-H9)⁴⁵ derived NPCs to activate the endogous CHD8 gene expression in human neurons. The guide RNA sequences for CHD8 were:

sgCHD8-1: 5′-ACGTCTTCAGAGGAAGATGG-3′

sgCHD8-2: 5′-GTAGCCATCTTGCTCCATGA-3′

sgCHD8-3: 5′-TGTATTTGTATTCCTACAAC-3′

sgCHD8-4: 5′-CCTGCAGCTTGCTTGAAAAA-3′

sgCHD8-5: 5′-CTATCCGGTTTCTAGGGCCG-3′

sgCHD8-6: 5′-TAGGTTGAGAGCGCACGGAG-3′

Droplet digital PCR

Droplet digital PCR (ddPCR) was utilized to quantify CHD8 expression in dCas9/activator-sgRNA infected neurons. ddPCR was performed following our published protocol, with modifications⁷⁷. Briefly, 175 ng of total RNA was used for cDNA synthesis with both Random 6 mers primer and Oligo dT primer (Takara Bio USA, RR037A). CHD8 primers and probe were designed using NCBI/Primer-BLAST (http://www.ncbi.nlm.nih.gov/tools/primer-blast/) and PrimerQuest tool (IDT). ddPCR was carried out using the Bio-Rad QX200 system. After each PCR reaction mixture consisting of ddPCR master mix and custom primers/probe set was partitioned into ~17,000 droplets, parallel PCR amplification was carried out. Endpoint PCR signals were quantified and Poisson statistics was applied to yield target copy number quantification of the sample. Two-color PCR reaction was utilized for normalization of gene expression by the housekeeping gene TBP.

The primers and probes used in ddPCR were:

CHD8: Forward: CGAGAAGATGAAGAAAGCAGCA

Probe (5’FAM/ZEN/3’IBFQ): AGAGACGCTCAAACCGCCAAGT

Reverse: GACCAGTTACATCCACCTCTT

TBP: Forward: CCACTCACAGACTCTCACAAC

Probe (5’HEX/ZEN/3’IBFQ): CCATCACTCCTGCCACGCCA

Reverse: GTACAATCCCAGAACTCTCCG

CHD8 expression was compared between groups using a two-tailed unpaired t-test (Graphpad Prism).

Dual luciferase assay

To assess the impact of MEIS2 binding motif mutations on the activity of the NFIA adjacent ATAC-seq peak region, a co-transfection experiment was performed using 100 ng of the promoter-Fluc, along with 100 ng of the pGL4.73[hRluc/SV40] Vector (Promega, E6911) and 100 ng of MEIS2 overexpression or control vectors (pCDNA3). To investigate the effect of elevated MEIS2 expression on the activity of the cloned genomic peak region, we conducted co-transfections with varying quantities of the MEIS2 overexpression plasmid, specifically: (1) 0 ng of pMEIS2 (accompanied by 100 ng of pCDNA3); (2) 50 ng of MEIS2 (along with 50 ng of pCDNA3); and (3) 100 ng of pMEIS2. The functional impact of the mutation on the activity of the peak region was assessed through co-transfection with 100 ng of the pMEIS2 plasmid. Transfection was carried out in HEK293 cells using PEI (4 μL of 1 μg/μl PEI per 1 μg of plasmid DNA). Luciferase activities were assessed at 48 hours post-transfection utilizing dual luciferase assays (Promega, E1960), and promoter activities were determined by normalizing firefly luciferase readings to Renilla luciferase readings. Statistical significance was assessed through a one-way analysis of variance (ANOVA) with multiple comparisons utilizing Tukey’s test. The “MEIS2 overexpression plasmid: 100 ng of MEIS2” served as the control.

RNA isolation, cDNA synthesis and qPCR

RNA was isolated from TRIzol samples using the TRIzol Reagent following the manufacturer’s instructions. Reverse transcription and Real-time PCR was performed using standard methods as described⁷⁶. The first-strand complementary DNA (cDNA) was generated by reverse transcription with both Random 6 mers primer and Oligo dT primer (Takara Bio USA, RR037A). To quantify the mRNA expression using real-time PCR, aliquots of first-stranded cDNA were amplified with gene-specific primers and iTaq^™ Universal SYBR^® Green Supermix (Bio-Rad, 1725125) using a StepOne Real-Time PCR System (Applied Biosystems). The PCR reactions contained cDNA, SYBR^® Green Supermix (Bio-Rad), and 0.3 μM of forward and reverse primers in a final reaction volume of 20 μl. The mRNA expression of different samples was calculated using 2^−ΔΔCT method⁷⁸. The primers used for qPCR were:

RAPGEF4 forward primer: 5′-ATTAATGGACGCCTGTTTGC-3′

RAPGEF4 reverse primer: 5′-CATGCACGCAGTTGAAGAGT-3′

CHD8 forward primer: 5′-CAGCCCAGTTCACCAAACTT-3′

CHD8 reverse primer: 5′-CTCCAGGTCCACATCAAGGT-3′

GAPDH forward primer: 5′-GTCAGTGGTGGACCTGACCT-3′

GAPDH reverse primer: 5′-TGCTGTAGCCAAATTCGTTG -3

Production of lentivirus

Lentivirus productions were performed using our published protocol^76,79 with modifications. Briefly, lentiviral vector DNA was co-transfected with packaging plasmids pMDL, REV and pCMV-Vsvg into HEK293T cells using the PEI method. Briefly, for one 15 cm petri-dish of cultured HEK293T cells, 12.2 μg of lentiviral DNA, 8.1 μg pMDL, 3.1 μg REV and 4.1 μg pCMV-Vsvg were mixed followed by adding 110 μl of 1 μg/μl PEI. The mixture was then added to HEK293 cells. The medium was then replaced with fresh medium at 5-hour post-transfection. The medium containing lentivirus was collected twice at 36 hours and 60 hours post-transfection, pooled, filtered through a 0.2 μm filter, and concentrated using an ultracentrifuge at 22,000 rpm for 2 hours at 4 °C using a SW32Ti rotor (Beckman). The virus was washed and then resuspended in DPBS (20 μl per petri-dish). We routinely obtained 1x10⁹ infectious viral particles per ml.

Lentivirus infection of targeted neurons

Morphological analysis of primate neurons with gene knockdown were performed as previously described⁴⁵. Briefly, cultured slices were infected with shRNA lentivirus by applying a concentrated lentivirus solution over the surface of the slice 20-24 hours after plating. The infected slices were fixed between 17 and 22 days post-infection. For morphological analysis of PCD100 and PCD155 neurons, the brain slices were infected with lentivirus expressing mScarlet (LV-EF1α-mScarlet) virus immediately after the slices were plated and fixed at 48 hours post-infection. Approximately, 4 x 10⁵ viral particles were used for infection of 1 cm² of the cultured slice. The cultured brain slices were initially fixed using 10% neutral buffered formalin (Tissue-Tek^®, 5990) at 4 °C in a dark environment for 48 h. Subsequently, the sections were sequentially transferred to 10%, 20%, and 30% sucrose solutions in PBS (phosphate-buffered saline), also in the absence of light. Once the whole slice had sunk to the bottom of the 30% sucrose solution, the slice was subjected to treatment with Autofluorescence Eliminator Reagent (Millipore, 2160). The procedure involved immersing the slice in PBS for 5 min at room temperature, transferring it onto a slide, and then washing it with 70% ethanol for 5 min. Following this, the slice was immersed in Autofluorescence Eliminator Reagent for 5 min, washed again with 70% ethanol for 3 min by gently dripping 70% ethanol onto the slice, dried, and finally mounted with DABCO for further analysis. The entire Autofluorescence Elimination process was carried out in dark.

QUANTIFICATION AND STATISTICAL ANALYSIS

To quantify differences in MEIS2 expression levels between ExN L2-3 IT and ExN L3-5 IT neurons, PCD155 macaque dlPFC sections were immunostained using an anti-MEIS2 antibody (1:50, Santa Cruz Biotechnology, sc-81986) and hybridized with an RNAscope probe targeting RORB transcript following the methods as described above. ExN L3-5 IT neurons were identified by expression of RORB and resided in a clear laminar position in the developing cortex (Figure S3B). ExN L2-3 neurons were identified above the laminar plane of RORB expression and lacked significant expression of RORB (Figure S3B). Five images of ExN L2-3 and five images of ExN L3-5 were acquired from each section (total n = 3 sections) of PCD155 macaque dlPFC using a Nikon A1 confocal microscope. High resolution (1024 x 1024 pixels, zoom 1.00, pixel size 0.21 μm) z-stack images were acquired with 0.75 μm step size and high and low z-position limits set above and below the section focal plane, respectively, using an oil immersion 60X objective (Plan Apo, numerical aperture 1.4, working distance 130 μm). For each image, identical laser power and detector gain settings were used. The images were blinded, z-projected in maximum intensity, and randomized prior to quantification. MEIS2 intensity was thresholded using a fixed threshold for each image (minimum pixel intensity 250, maximum pixel intensity 600), and the total number of MEIS2⁺ cells and DAPI⁺ cells manually counted. The percentage of MEIS2⁺/DAPI⁺ in ExN L2-3 and ExN L3-5 was calculated for each section and significance tested using a two-tailed unpaired t-test (Graphpad Prism).

To quantify the levels of CAMK2A, MICAL2, RAPGEF4, NRGN mRNA, the PCD100 and PCD155 macaque dlPFC sections were immunostained for SATB2 (a marker of developing excitatory neurons) (1:50, Abcam ab51502) and hybridized with an RNAscope probe of the respective target as described above. Images were acquired from 3 sections of PCD100 and 3 sections of PCD155 macaque dlPFC using a Nikon A1 confocal microscope. High resolution (1024 x 1024 pixels, zoom 1.37, pixel size 0.15 μm) z-stack images were acquired with 0.75 μm step size and high and low z-position limits set above and below the section focal plane, respectively, using an oil immersion 60X objective (Plan Apo, numerical aperture 1.4, working distance 130 μm). For each respective RNAscope probe, identical laser power and detector gain settings were used. Images were subsequently blinded and randomized for quantification. 125 - 200 SATB2-immunopositive cells were identified from each section and the number of target RNA puncta per cell were hand-counted using raw, non-projected image data by a blinded counter. The number of RNA puncta per cell was compared between PCD100 and PCD155 dlPFC and significance tested using a two-tailed unpaired t-test (Graphpad Prism).

To quantify the levels of SLC44A5, CACNA1C, and SCN2A, three sections from both PCD100 and PCD155 macaque dlPFC were immunostained and imaged as described above. For each respective antibody, identical laser power and detector gain setting were used. Five independent locations within the cortical plate were imaged from each section. For SLC44A5 and CACNA1C quantifications, fluorescent signal was enriched within neuronal cell bodies. Therefore, Z-stack images were projected using “sum of all slices” (ImageJ) and 10 ROIs of neuronal cell bodies were drawn per image. The integrated density (sum of all pixel intensities) of each cellular ROI along with one background ROI was recorded and the cell total corrected fluorescence (CTCF) was calculated after background subtraction (CTCF = integrated density of signal – [area of signal ROI * mean intensity of background ROI]). For SCN2A quantification, the fluorescent signal was principally enriched in the neuropil. Therefore, 10 ROIs of neuropil directly adjacent to the neuronal soma were drawn per image and CTCF calculated as above. The CTCF was compared between PCD100 and PCD155 macaque dlPFC and significance tested using a two-tailed unpaired t-test with Welch’s correction (Graphpad Prism).

Morphological analyses of lentivirus infected neurons were carried out following a published protocol⁷⁶. Briefly, for dendritic morphology analysis, red-or green-fluorescent neurons were imaged using a Nikon A1 confocal microscope with 20X objective. The images were then analyzed using Neurolucida software (Micro-BrightField, Burlington, Vermont, https://www.mbfbioscience.com/). A minimum of 25 neurons per condition were analyzed. For cultured slices, the neurons selected for tracing must be intact, with no missing branches. Randomly chosen neurons were zoomed in and traced through z-stack images to capture their 3-dimensional structure. Live tracing was used to trace cultured primary neurons. Virus-infected neurons with triangular or oval-shaped somas were randomly selected for tracing under a fluorescent microscope using Neurolucida with 3D module plug-in. Statistical analysis was conducted using Graphpad Prism software, using a two-tailed and unpaired t-test to compare two conditions. To compare the complexity of neurons using Sholl analysis, multivariate analysis of variance (MANOVA) was performed using SPSS statistical software.

Other statistical analysis details are provided in each metods details’ section.

Supplementary Material

2

Table S1. Specimens used for sn-multiome, patch-seq, and validation, related to Figures 1–7.

3

Table S2. Quality control measures of sn-multiomic data, related to Figures 1 and 2.

4

Table S3. Annotation of putative transcriptomically defined subtypes, related to Figures 1 and 2.

5

Table S4. snRNA-seq maturation trajectory, differentially expressed genes and functional enrichment (ExN L2-3 IT and ExN L3-5 IT), related to Figure 2.

6

Table S5. Gene-peak linkages and GO analysis, related to Figure 3.

7

Table S6. Motif Enrichment on differential peaks in L2-3IT and L3-5IT cells, related to Figure 3.

8

Table S7. Patch-seq data quality control measures, Patch-seq gene expression data, and maturation trajectory genes, related to Figure 4.

9

Table S8. Patch-seq data analysis of acute and cultured slices, related to Figure 4.

10

Table S9. Genes correlated with electrophysiological properties and GO analysis, related to Figure 4.

11

Table S10. Electrophysiological data of macaque and human cortical neurons with RAPGEF4 manipulation, related to Figures 5 and 6.

12

Table S11. Patch-seq or patch-clamp recording of human neurons with CHD8 knockdown and transcriptional activation, related to Figure 7.

13

Table S12. Patch-clamp analysis of human cortical neurons with both CHD8 knockdown and RAPGEF4 overexpression, related to Figure 7.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse anti-SATB2 (1:50)	Abcam	ab51502; RRID:AB_882455
Mouse anti-CASPR2 (1:10)	Developmental Studies Hybridoma Bank	K67/25; RRID:AB_2877287
Mouse anti-MEIS2 (1:50)	Santa Cruz Biotechnology	sc-81986; RRID:AB_2143037
Rabbit anti-GLIS3 (1:100)	Novus Biologicals (Bio-techne)	NBP2-33787; RRID:AB_3285315
Sheep anti-NRGN (1:50)	R&D Systems (Bio-techne)	AF7947; RRID:AB_3644532
Rabbit anti-SLC44A5 (1:100)	Invitrogen	PA5-61522; RRID:AB_2647519
Mouse anti-CACNA1C(1:50)	Abcam	ab84814; RRID:AB_1860052
Rabbit anti-SCN2A(1:5000)	Alomone Labs	ASC-002; RRID:AB_2040005
RNAscope ISH probes
Mmu-CAMK2A-C1	ACD, Bio-Techne	461731
Mmu-NRGN-C2	ACD, Bio-Techne	1231141-C2
Mmu-MICAL2-C2	ACD, Bio-Techne	1231151-C2
Mmu-VXN-C2	ACD, Bio-Techne	1231161-C2
Mmu-RAPGEF4-C2	ACD, Bio-Techne	1231131-C2
Mmu-RORB-C3	ACD, Bio-Techne	876301-C3
Critical commercial assays
Chromium Next GEM Single Cell Multiome ATAC Kit A	10x Genomics	Cat# PN-1000280
Chromium Next GEM Single Cell Multiome Reagent Kit A	10x Genomics	Cat# PN-1000282
Chromium Next GEM Chip J Single Cell Kit	10x Genomics	Cat# PN-1000234
Single Index Kit N Set A	10x Genomics	Cat# PN-1000212
Library Construction Kit	10x Genomics	Cat# PN-1000190
Dual Index Kit TT Set A	10x Genomics	Cat# PN-1000215
Agilent High Sensitivity DNA Kit	Agilent Technologies	Cat# 5067-4626
Nextera XT DNA Library Preparation Kit	Illumina	Cat# FC-131-1096
Maxima H Minus Reverse Transcriptase	ThermoFisher Scientific	Cat# EP0751
KAPA HiFi Hotstart PCR kit	Roche	Cat# KK2502
Phusion High-Fidelity DNA Polymerase	ThermoFisher Scientific	Cat# F530L
Qubit dsDNA HS kit	ThermoFisher Scientific	Cat# Q32851
AMPure XP beads	Beckman Coulter	Cat# A63881
High Sensitivity D5000 ScreenTape	Agilent Technologies	Cat# 5067-5593
PicoGreen® dsDNA reagent kit	Molecular Probes	Cat# P-11496
iTaq™ Universal SYBR® Green Supermix	Bio-Rad	Cat# 1725125
Dual-Luciferase® Reporter Assay System	Promega	Cat# E1960
PrimeScript RT Reagent Kit	Takara Bio	Cat# RR037A
RNAscope Multiplex Fluorescent Reagent Kit v2 Assay	ACD, Bio-Techne	Cat# 323110
RNAscope Co-detection Antibody Diluent	ACD, Bio-Techne	Cat# 323180
TSA Plus Fluorescein	Akoya Biosciences	Cat# NEL741001KT
TSA Plus Cy3	Akoya Biosciences	Cat# NEL744001KT
Recombinant DNA
pLV-mNeon-U6-shNC (LV-EF1α-mNeon-U6-shNC)	This study	N/A
pLV-mScarlet-U6-shCHD8-1 (LV-EF1α-mScarlet-U6-shCHD8-1)	This study	N/A
pLV-mScarlet-U6-shCHD8-2 (LV-EF1α-mScarlet-U6-shCHD8-2)	This study	N/A
pLV-mScarlet-U6-shCHD8-3 (LV-EF1α-mScarlet-U6-shCHD8-3)	This study	N/A
pLV-mScarlet-U6- shRAPGEF4-1 (LV-EF1α-mScarlet-U6-shRAPGEF4-1)	This study	N/A
pLV-mScarlet-U6 -shRAPGEF4-2 (LV-EF1α-mScarlet-U6- shRAPGEF4-2)	This study	N/A
pLV-mScarlet-U6-shRAPGEF4-3 (LV-EF1α-mScarlet-U6- shRAPGEF4-3)	This study	N/A
pLV-mScarlet-U6-shRAPGEF4-4 (LV-EF1α-mScarlet-U6- shRAPGEF4-4)	This study	N/A
pLV-mScarlet-U6 -shRAPGEF4-5 (LV-EF1α-mScarlet-U6- shRAPGEF4-5)	This study	N/A
pLV-mScarlet-U6 -shRAPGEF4-6 (LV-EF1α-mScarlet-U6- shRAPGEF4-6)	This study	N/A
pMEIS2	This study	N/A
pWildTypeBindingSite1	This study	N/A
pMutationBindingSite1	This study	N/A
pMutationBindingSite2	This study	N/A
pMutationBindingSite1And2	This study	N/A
pLV-CRE (LV-EF1α-CRE)	This study	N/A
pLV-RAPGEF4-IRES-CRE(LV-EF1α-RAPGEF4-IRES-CRE)	This study	N/A
pLV-shCHD8(LV-U6-shCHD8-CAG-DIO-GFP)	This study	N/A
pLV-shNC (LV-U6-shNC-CAG-DIO-GFP)	This study	N/A
pLV-shRAPGEF4 (LV-U6-shRAPGEF4-CAG-DIO-GFP)	This study	N/A
pLV-sgCHD8-mScarlet (LV-U6-sgCHD8-EF1α-mScarlet)	This study	N/A
pLV-mScarlet (LV-EF1α-mScarlet)	This study	N/A
Experimental models: Cell lines
dCas9A-H9 human pluripotent stem cell line	45
HEK293T	ATCC	Cat# CRL-3216; RRID:CVCL_0063
Software and algorithms
StepOne Software v2.3	http://downloads.thermofisher.com/Instrument_Software/qPCR/Step-1/SOP23_Release%20Notes_4482516.pdf	RRID:SCR_014281
Neurolucida version 2023.2.3	http://www.mbfbioscience.com/neurolucida	RRID:SCR_001775
Neurolucida Explorer 2022.2.1	https://www.mbfbioscience.com/neurolucida-explorer	RRID:SCR_017348
ImageJ	https://imagej.net/	RRID:SCR_003070
Graphpad Prism 9.0.0	https://www.graphpad.com/	RRID:SCR_000306
Cell Ranger Arc 2.0.0	https://www.10xgenomics.com/support/software/cell-ranger-arc/latest	RRID:SCR_023897
Seurat 4.3.0.1	https://satijalab.org/seurat/	RRID:SCR_016341
Signac 1.9.0	https://stuartlab.org/signac/	RRID:SCR_021158
chromvar 1.20.2	https://github.com/GreenleafLab/chromVAR	RRID:SCR_026570
Gviz 1.42.1	https://bioconductor.org/packages/release/bioc/html/Gviz.html	RRID:SCR_024239
TFBSTools 1.36.0 https://bioconductor.org/packages/release/bioc/html/TFBSTools.htmlJASPAR20200.99.10	https://jaspar2020.genereg.net/	RRID:SCR_024260
SeuratWrappers 0.3.1	https://github.com/satijalab/seurat-wrappers	RRID:SCR_022555
Monocle3 1.3.1	https://cole-trapnell-lab.github.io/monocle3/	RRID:SCR_018685
SingleCellExperiment 1.20.1	https://www.bioconductor.org/packages/release/bioc/html/SingleCellExperiment.html	RRID:SCR_026794
Harmony 0.1.1	https://github.com/immunogenomics/harmony	RRID:SCR_022206
biomaRt 2.54.1	https://bioconductor.org/packages/release/bioc/html/biomaRt.html	RRID:SCR_019214
Monocle 2.26.0	https://cole-trapnell-lab.github.io/monocle-release/docs/	RRID:SCR_016339
DDRTree 0.1.5	https://CRAN.R-project.org/package=DDRTree	RRID:SCR_026795
GenomeInfoDb 1.34.9	https://bioconductor.org/packages/GenomeInfoDb	RRID:SCR_024235
IRanges 2.32.0	https://bioconductor.org/packages/IRanges/	RRID:SCR_006420
BiocGenerics 0.45.2	https://bioconductor.org/packages/BiocGenerics/	RRID:SCR_024226
BSgenome.Mmulatta.UCSC.rheMac10 1.4.2	https://bioconductor.org/packages/BSgenome.Mmulatta.UCSC.rheMac10/	RRID:SCR_026796
complexHeatmap 2.15.2	https://jokergoo.github.io/ComplexHeatmap-reference/book/index.html	RRID:SCR_017270
GenomicRanges 1.50.2	https://bioconductor.org/packages/GenomicRanges/	RRID:SCR_000025
ensembldb 2.22.0	https://github.com/jorainer/ensembldb	RRID:SCR_019103
EBseq v1.14	https://www.biostat.wisc.edu/~kendzior/EBSEQ/	RRID:SCR_003526
Metascape v3.5.20230501	https://metascape.org/	RRID:SCR_016620
MACS2	https://pypi.org/project/MACS2/#description	RRID:SCR_013291
DoubletFinder v 2.0.4	https://github.com/chris-mcginnis-ucsf/DoubletFinder	RRID:SCR_018771

Highlights.

• Identification of genes and regulatory networks driving primate neuronal maturation

• Discrete electrophysiological properties have distinct maturational profiles

• RAPGEF4 promotes the maturation of resting membrane potential and sodium current

• Knockdown of CHD8 impairs the maturation of cortical excitatory neurons