Post-transcriptional regulation maintains neurogenic potential of human neural progenitor cells Authors: Natasha M Méndez-Albelo, Yajie Zhang, Soraya O Sandoval, Yu Guo, Yu Gao, Lily A Peters, Carissa L Sirois, Anna B Widicus, Paofue Yang, Drew N Trygstad, Magnus T Eckholm, Zhiyan Xu, Riley V Smith, Rachel M Disher, Christian J Johnston, Xinyu Zhao Published: 2026 ABSTRACT SummaryIncreasing evidence supports important roles of post-transcriptional regulation in brain development; however, the precise mechanism remains unclear. Here, we show that Fragile X autosomal homolog 1 (FXR1), a brain-enriched RNA-binding protein (RBP), is essential for human neurogenesis. FXR1-deficient human cortical neural progenitor cells (NPCs) have reduced proliferation and impaired cell cycle exit during differentiation, leading to impaired neuronal differentiation. Transcriptomic and co-expression network analyses revealed FXR1 as an orchestrator of stage-specific gene programs driving neural differentiation. High-mobility group protein B2 (HMGB2), a master transcriptional regulator in stem cells, is a direct target of FXR1. FXR1 and HuR co-regulate HMGB2 mRNA, and their opposing actions on mRNA stability maintain the levels of HMGB2 in human NPCs. Reducing HMGB2 levels in FXR1-deficient NPCs rescues NPC proliferation and neuronal differentiation. Together, these findings establish post-transcriptional regulation as a key mechanism that maintains neurogenic potency of human NPCs.Keywords: RNA-binding protein, FXR1, HMGB2, human pluripotent stem cells, neural stem cells, neural progenitor cells, neurodevelopment, post-transcriptional regulation, HuR Introduction The development of the human brain requires the orchestrated actions of many molecular pathways critical for the self-renewal and differentiation of radial glia, the proliferation, and cell fate specification of neural progenitor cells (NPCs), and the differentiation, migration, and functional maturation of neurons and glia (Salamon et al., 2025). While transcriptional regulation sets the overall developmental program and has been the focus of neurodevelopmental research (Nowakowski et al., 2025), increasing evidence has put post-transcriptional regulation at the center stage of gene regulation (Hentze et al., 2025; Hoye and Silver, 2021). Post-transcriptional regulations, including RNA splicing, localization, stability, and protein translation, are significantly faster than transcription, thereby enabling rapid control of gene expression to ensure dynamic cell lineage transitions during brain development (Hoye and Silver, 2021). RNA-binding proteins (RBPs) are critical post-transcriptional regulators in many cellular processes, including translational inhibition of cell lineage-specific mRNAs before cell-fate transitions occur (Hoye and Silver, 2021). Consequently, mutations in RBPs or disruptions of RBP-mediated regulations have been implicated in neurodevelopmental disorders, including intellectual disability (Mosti et al., 2025), autism spectrum disorder (ASD) (Jeong et al., 2025), and Fragile X syndrome (FXS) (Richter and Zhao, 2021). However, despite growing evidence for the critical roles of RBPs in neurodevelopment, the precise mechanisms underlying RBP regulation and the contributions of many RBPs in neurodevelopment remain unclear. Fragile X autosomal homolog 1 (FXR1) belongs to a family of brain-enriched RBPs that also includes Fragile X messenger ribonucleoprotein (FMRP) and Fragile X autosomal homolog 2 (FXR2) (Mendez-Albelo et al., 2024). FXR1 is indispensable for development because its loss-of-function is intolerant in humans (Lek et al., 2016) and causes neonatal death in mice (Mientjes et al., 2004). In humans, FXR1 is abundantly expressed in the developing nervous system and persists in cortical neurons throughout postnatal development, suggesting important roles across neurodevelopmental stages (Mendez-Albelo et al., 2024). While a direct causal link between FXR1 and brain disorders remains limited, FXR1 is located within chromosome 3q25-27, a major locus of ASD susceptibility (Auranen et al., 2002), and FXR1 has been shown to interact with ASD-associated proteins (Sakai et al., 2011). Several multi-modal and cross-species studies, including genome-wide association analyses, have identified FXR1 as a high-confidence risk gene for schizophrenia and other psychiatric disorders (Gandal et al., 2018; Li et al., 2018; Liu et al., 2016; Pardinas et al., 2018; Rajarajan et al., 2018; Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014; Takata et al., 2017; Wang et al., 2018). Although most genetic variations identified are in non-coding regions of the FXR1 gene, a single nucleotide polymorphism associated with higher FXR1 mRNA levels in humans is found to be associated with higher mood stability in these individuals (Del’Guidice et al., 2015; Khlghatyan et al., 2018). The function of FXR1 in the brain has only been investigated in a limited number of studies (Mendez-Albelo et al., 2024). In mice, deletion of Fxr1 gene from mature cortical excitatory neurons had little impact (Cook et al., 2014) while deletion of Fxr1 in parvalbumin-expressing interneurons leads to impaired morphology, reduced neuronal excitability, and behavioral deficits associated with schizophrenia (Shen et al., 2021). However, the role of FXR1 in cortical development, particularly human cortical neuronal differentiation, remains unknown. In this study, we demonstrated that FXR1-deficient human NPCs have reduced proliferation and impaired cell cycle exit during neural differentiation, leading to an aberrant increase of proliferating NPCs, even at late stage of neuronal differentiation, and impaired morphological and electrophysiological maturation of neurons. We discovered that FXR1-dependent gene expression programs are disrupted across differentiation stages and revealed high mobility group protein box2 (HMGB2) as the hub of the FXR1-regulated gene network in NPCs. FXR1 deficiency leads to elevated HMGB2 mRNA and protein levels in NPCs and during neural differentiation. We demonstrated that FXR1 directly binds HMGB2 mRNA, which in turn reduces the binding by Human antigen R (HuR), a stabilizer of HMGB2 mRNA. Reducing the levels of HMGB2 in FXR1-deficient NPCs restored NPC proliferation, cell cycle exit during neural differentiation, and neuronal maturation. Together, our data establishes FXR1 as an essential post-transcriptional regulator in maintaining neurogenic potential of human NPCs, which is critical for proper neuronal differentiation. Results FXR1 is required for human NPC proliferation and proper differentiation of dorsal forebrain neurons To investigate the roles of FXR1 in human cortical neurodevelopment, we generated FXR1 knockout (FXR1-KO) induced pluripotent stem cell (iPSC) lines from two healthy control human iPSC lines (Ctrls). CRISPR-Cas9 genome editing using a single-guide RNA (sgRNA) targeting the first common exon (exon 4) shared among all FXR1 isoforms (Mendez-Albelo et al., 2024) produced a 1-bp adenine insertion leading to a frameshift and premature stop codon, as confirmed by Sanger sequencing (Figures 1A and 1B; Table S1). Western blot analysis demonstrated a complete loss of FXR1 protein (Figure 1C). The FXR1-KO (KO1 and KO2) iPSC clones retained typical iPSC morphology, pluripotency marker expression, and a normal karyotype (Figures S1A–S1D), with no off-target mutations at the top six predicted sites (Figure S1E). FXR1 deficiency impairs NPC proliferation and disrupts neuronal differentiation (A) Schematic of the CRISPR-Cas9 targeting strategy to generate FXR1-KO iPSC lines. (B) FXR1-KO Sanger sequencing confirming CRISPR-mediated one base insertion (indicated by arrow) in FXR1-KO iPSCs compared to Ctrl iPSCs. (C) Immunoblot showing complete absence of FXR1 protein in two FXR1-KO iPSC lines (KO-1 and KO-2) compared to parental iPSC lines (Ctrl-1, Ctrl-2). (D) Schematic overview of the experimental timeline and cellular analysis. (E) Representative immunofluorescence images of NPCs stained for proliferation markers Ki67 (red) and EdU and NPC marker SOX2 (white) in Ctrl and FXR1-KO cells. (F) Quantification of the percentage of EdU+ NPCs (EdU+/SOX2+) among total SOX2+ NPCs, normalized to Ctrl. Two-tailed Student’s test with Welch’s correction. (n = 6 independent differentiations derived from 2 independent iPSC lines per group). (G) Representative immunofluorescence images of Ctrl and FXR1-KO cultures at 1-week and 4-weeks of differentiation stages stained for the neuronal markers TUBB3 (red) and MAP2. (H–K) Quantification of TUBB3+ (H and J) and MAP2+ (I and K) cells among total DAPI+ cells at 1-week (H and I) and 4-weeks (J and K) of differentiation. (L) Representative immunofluorescent images of FXR1-KO and Ctrl human excitatory neurons at 4-weeks of differentiation, stained with neuron markers NeuN, MAP2 and SOX2. (M) Quantification of percentage of NeuN+ cells at 4-weeks of differentiation. (N) Representative immunofluorescence images showing Ki67+ (red) and SOX2+ (white) cells at 1-week and 4-weeks of differentiation in Ctrl and FXR1-KO cultures. (O) Proportion of SOX2+ NPCs among FXR1-KO or Ctrl cells across all differentiation stages. (P) Proportion of proliferating NPCs (Ki67+/SOX2+) among FXR1-KO or Ctrl cells across all differentiation stages. Scale bars, 25 μm. (F) Two-tailed Student’s t test with Welch’s correction. n = 6 independent differentiations derived from 2 independent iPSC lines per group. (H–K and M) Two-tailed Student’s t test. Mean ± SEM from n ≥ 4 independent differentiations derived from 2 independent iPSC lines per group. (O and P) Two-way ANOVA, n > 4 independent differentiations derived from 2 independent iPSC lines per group. Each cell line is plotted with a different shape. Data are shown as mean ± SEM. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. We differentiated the FXR1-KO iPSCs (KO1 and KO2) and the two isogenic controls (Ctrl1 and Ctrl2) into NPCs and then dorsal forebrain excitatory neurons using the dual SMAD inhibition method (Li et al., 2020; Shen et al., 2023) (Figure 1D). We first evaluated NPC proliferation using a cell cycle marker Ki67 and pulse labeling with EdU, which is incorporated into the cellular genome during the S phase of the cell cycle. We observed that FXR1-deficient NPCs exhibited a significantly decreased proportion of Ki67+ and EdU+ cells compared to control (Figures 1E, 1F, and S1F). To confirm that the observed phenotypes were a result of FXR1 deficiency, we knocked down FXR1 (FXR1-KD) in three control iPSC lines (Ctrl 1, Ctrl 2, and Ctrl 3) using a published small hairpin RNA (shRNA) against human FXR1 (Shen et al., 2021). FXR1-KD resulted in a significant reduction in FXR1 mRNA levels compared to non-silencing controls (NCs) in both NPCs and 1-week differentiation neurons, without significant effects on FMR1 and FXR2 mRNA levels (Figures S1G–S1L). FXR1-KD NPCs had a significantly reduced proportion of Ki67+ cells compared to control conditions (Figures S2A and S2B), validating the proliferation deficits observed in FXR1-KO NPCs. We next evaluated the neuronal differentiation potential of FXR1-KO NPCs. At 1-week of differentiation, FXR1-KO cells had a 44.3% reduction in TUBB3+ immature neurons (Figures 1G and 1H) and a 20.8% reduction in MAP2+ neurons (Figures 1G and 1I), compared with Ctrl cells. Similar reduction in MAP2+ neurons was also observed in FXR1-KD cells at 1-week of differentiation (Figures S2C and S2D). At 4-weeks of differentiation, FXR1-KO cells showed a significantly reduced proportion of TUBB3+ neurons (36.5% reduction) and even greater reduction in MAP2+ neurons (43.5% reduction), compared to Ctrl cells (Figures 1G, 1J, and 1K). In addition, the proportion of cells expressing NeuN+, a marker of mature neurons, was drastically lower (55.5% reduction) in FXR1-KO compared to Ctrl cells (Figures 1L and 1M). Therefore, the reduced neuronal differentiation of FXR1-KO NPCs was due to a sustained, rather than a delayed, impairment of neuronal differentiation. The substantial reduction of neurons at both 1-week and 4-weeks of differentiation prompted us to investigate whether FXR1-KO NPCs had reduced ability to differentiate. We therefore assessed the presence of SOX2+ NPCs and Ki67+SOX2+ proliferating NPCs during neuronal differentiation. At 1-week of differentiation, although FXR1-KO and Ctrl cultures had similar proportions of undifferentiated SOX2+ NPCs, FXR1-KO cultures had more NPCs that remained in cell cycle (Ki67+SOX2+) compared to Ctrl, indicating impaired cell cycle exit (Figures 1N–1P). Strikingly, at 4-weeks of differentiation, FXR1-KO cultures had significantly higher proportions of cells remaining as SOX2+ NPCs compared to Ctrl culture (31.8% KO vs. 7.8% Ctrl), and nearly half (46.7%) of the NPCs in FXR1-KO cultures were positive for cell cycle marker (Ki67+ SOX2+) compared to 8.3% in Ctrl (Figures 1N–1P). In addition, we detected no cells that were positive for astrocyte markers GFAP and S100β in either FXR1-KO or control cells at both 1- and 4-weeks of differentiation (Figures S2E and S2F). Therefore, FXR1 deficiency did not lead to a developmental shift toward a glial fate. To assess whether the reduced number of postmitotic neurons resulted from increased cell death, we quantified the number of cleaved caspase-3 positive cells. No significant difference was detected in NPCs or neurons at 1-week of differentiation, indicating that reduced neuronal output at the NPC stage is not due to increased apoptosis (Figures S2G–S2I). In contrast, cleaved caspase-3 levels were significantly higher in FXR1-KO neurons at 4-weeks of differentiation (25.9% KO vs. 11.6% Ctrl) (Figures S2G and S2J). Taken together, our data suggest that a loss of FXR1 impairs neurogenic potential of NPCs, leading to failed cell cycle exit during neuronal differentiation and reduced mature neuron survival. FXR1 deficiency impairs morphological and electrophysiological development of human neurons We next investigated whether impaired neuronal differentiation may affect the morphology and function of mature neurons (Figure 2A). FXR1-KO MAP2+ neurons had reduced dendritic ends, nodes, and length compared to Ctrl at both 1-week of differentiation (Figures S3A–S3D) and at 4-weeks of differentiation (Figures 2B–2E). Similar morphological deficits were also observed in 1-week differentiated FXR1-KD neurons (Figures S3E–S3H). FXR1 knockout impairs neuronal morphogenesis and network-level activity (A) Schematics of neuronal differentiation stages used for morphological analyses, calcium imaging, and MEA recordings. (B) Representative neuronal reconstructions for control and FXR1-KO neurons at 4-weeks of differentiation. Scale bars, 10 μm. (C–E) Quantitative analyses of neuronal morphology, including total neurite length (C), number of branching nodes (D), and number of neurite ends (E). (n = 39 individual neurons, from 4 independent batches of differentiation, derived from 2 iPSC lines per group). (F) Representative MEA raster plots and spike traces of Ctrl (top) and FXR1-KO (bottom) recordings (scale bars, 10s). (n = 4 independent batches of differentiation (2 iPSC line per group normalized to isogenic control). (G–H) Mean firing rate (G) and numbers of bursts (H) of FXR1-KO neurons, normalized to Ctrl. (I) Representative long-duration calcium imaging raster plots illustrating neuronal activity in Ctrl and FXR1-KO neurons. (J) Sample field of view showing individual neurons segmented and color-coded by their activity states. (K and L) Calcium imaging analysis and quantification of numbers of GCaMP fluorescent events (K) and numbers of bursts (L) of FXR1-KO and Ctrl neurons; (n = individual ROIs from 4 independent batches of differentiation, derived from 2 iPSCs lines per group). (G–H) Two-tailed Student’s t test with Welch’s correction. (C–E and K–L) Two-tailed Student’s test. Mean ± SEM ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. Data are shown as mean ± SEM. Each cell line is plotted with a different shape. We next measured spontaneous neuronal activity at 5-weeks of differentiation using multielectrode arrays (MEA). FXR1-KO neurons exhibited reduced mean firing rate and reduced number of bursts (Figures 2F–2H), which we also confirmed in FXR1-KD neurons (Figures S3I–S3K). Since FXR1-KO cultures had significantly reduced proportions of neurons at 4-weeks of differentiation, we decided to investigate whether the reduced activity measured via MEA was actually due to the reduced activities of individual neurons. We performed calcium imaging analyses of neurons infected with lentivirus (LV) expressing a genetic calcium sensor driven by the neuronal Synapsin promoter (LV-SYN1-GCaMP7f) (Figures 2I–2L). FXR1-KO neurons exhibited a significant reduction in the number of GCaMP7f fluorescent events detected (Figure 2K) and the number of bursts (Figure 2L) compared to Ctrl neurons. Together, these results indicate that FXR1 is required for both morphological and electrophysiological development of human neurons. FXR1 deficiency disrupts gene expression programs across differentiation stages To investigate the molecular mechanisms underlying FXR1 deficiency-induced neurogenic deficits, we performed transcriptomic analysis of FXR1-KO and isogenic control NPCs, 1-week differentiated neurons, and 4-weeks differentiated neurons (Figure 3A). Principal component analysis (PCA) of the top 5,000 most variable genes across groups showed distinct clustering of control and FXR1-KO cells, indicating robust genotype-dependent transcriptional differences throughout neuronal differentiation (Figures S4A–S4C). In addition, there were largely distinct sets of differentially expressed genes (DEGs) across developmental stages, with minimal overlap among NPCs, 1-week, and 4-weeks differentiated neurons (Figure 3B; Table S2), indicating strong differentiation stage-specific effects of FXR1 loss. FXR1 deficiency disrupts gene expression programs across differentiation stages (A) Schematic of the RNA-seq experiment showing sample collections at NPC stage, 1-week of differentiation, and 4-weeks of differentiation. (B) Venn diagram showing the overlap of differentially expressed genes (DEGs) between FXR1-KO and control (Ctrl) cells in NPCs, 1-week of differentiation, and 4-weeks of differentiation. Transcriptomic analyses using DESeq thresholds of a p value <0.05 and FC > 2. n = 4 independent differentiations derived from 2 independent iPSC lines per group. (C) Heatmap of DEGs across all stages of differentiation, grouped based on differentiation stage-specific and shared expression patterns. (D–F) Volcano plots showing DEGs between FXR1-KO and Ctrl in NPCs (D), 1-week of differentiation (E), and 4-weeks of differentiation (F). (G–I) Gene Ontology (GO) analysis of DEGs in NPCs (G), 1-week of differentiation (H), and 4-weeks of differentiation (I). BP, biological process; CC, cellular component; MF, molecular function. The size of dots represents gene count, and color intensity reflects –log10(P). (J–L) Gene ontology biological process (GOBP) gene set enrichment analysis (GSEA) showing normalized enrichment scores (NES) for pathways enriched in FXR1-KO versus Ctrl cells in NPCs (J), 1-week (K), and 4-week (L) differentiation stages. FXR1, BEX5 (brain expressed X-linked 5), FAM50B (family with sequence similarity 50 member B), DMRTA2 (DMRT like family A2), ENSG00000290654, and LINC02506 (long non-coding RNA) were the only overlapping DEGs among all three differentiation stages (Figure 3B). DMRTA2 has been shown to be important for early forebrain development and cortical progenitor regulation (Fernandez et al., 2015; Maleszewska et al., 2024; Shen et al., 2025). On the other hand, BEX5 and FAM50B have primarily been implicated in cancer prognosis rather than neurodevelopmental processes (Ding et al., 2022; Kazi et al., 2015; Wang et al., 2025). Hierarchical clustering of the DEGs further showed that FXR1-dependent transcriptional changes were segregated by both differentiation stages and genotypes (Figure 3C). Moreover, the number of DEGs increased during differentiation (Figures 3B and 3D–3F), indicating that transcriptomic dysregulation becomes more severe as differentiation progresses. To identify the biological processes dysregulated by FXR1 deficiency, we performed Gene Ontology (GO) analysis (Figure 3; Table S2). In NPCs, the DEGs were associated with early neuronal lineage programs and neuronal functions (Figure 3G), consistent with altered neurogenic transcriptional programs at the NPC stage. At 1-week of differentiation, the DEGs were related to mitotic cell cycle, neuron differentiation, and brain development processes (Figure 3H). In contrast, at 4-weeks of differentiation, the DEGs were related to processes important for neuronal maturation, such as synaptic formation, ion channels, and morphology (Figure 3I). Because GO analysis does not indicate whether these biological processes are coordinately activated or suppressed, we performed gene set enrichment analysis (GSEA) using ranked gene expression data for each differentiation stage. Consistent with the GO analysis, neuronal differentiation (action potential, synaptic membrane, and transmembrane transporter) was negatively enriched in FXR1-KO NPCs (Figures 3J, S4D, and S4E; Table S2) and at 1-week of differentiation (Figures 3K, S4F, and S4G; Table S2), indicating coordinated dysregulation of early neurogenic programs. Interestingly, there was a positive enrichment for terms related to cell cycle (mitotic cell cycle checkpoint and mitotic sister chromatid segregation) in 1-week FXR1-KO neurons (Figures 3K, S4F, and S4G; Table S2), suggesting impaired cell cycle exit. At 4-weeks of differentiation, pathways associated with neuronal differentiation and synaptic function were negatively enriched, indicating downregulation of maturation-associated gene networks in FXR1-KO cells (Figures 3L, S4H, and S4I; Table S2). Together, these findings demonstrate that FXR1 deficiency impairs the neurogenic potential of NPCs, leading to progressive disruption of transcriptional networks governing neuronal differentiation, maturation, and synaptic development. Co-expression network analysis identifies coordinated transcriptional programs disrupted by FXR1 deficiency To better understand coordinated gene expression programs underlying FXR1-dependent neuronal differentiation, we performed weighted gene co-expression network analysis (WGCNA) using DEGs identified across all three differentiation stages and identified four major gene co-expression modules (M1-M4) with distinct temporal expression during neuronal differentiation (Figures 4A and S5A–S5C; Table S3). Heatmap visualization of module gene expression showed that these co-expression modules segregate strongly by differentiation stages in both control and FXR1-KO samples, while also revealing genotype-dependent alterations within specific modules (Figure 4B). FXR1 deficiency disrupts the temporal organization of gene co-expression modules during neuronal differentiation (A) Summary of WGCNA-derived gene modules (M1-M4) showing average log2 fold change between FXR1-KO and Ctrl samples in NPCs, 1-week of differentiation, and 4-weeks of differentiation. The size of dots reflects the number of genes within each module. (n = 4 independent differentiations derived from 2 iPSC lines per group). (B) Heatmap showing expression patterns of genes within modules M1-M4 across differentiation stages in Ctrl and FXR1-KO samples. (C) Gene Ontology (GO) enrichment analysis of the M1 module. (D) Gene set variation analysis (GSVA)-based correlations between module activity and known developmental traits, including stem cell maintenance, proliferation, neuronal differentiation, maturation, and action potential–related programs. (E) Correlation heatmap showing module eigengene relationships with differentiation stages and comparisons between FXR1-KO and controls. (F) Global gene co-expression network, with nodes colored by module assignment (M1-M4 shown), illustrating organization and connectivity among modules. (G) Network visualization of the top 25 hub genes within the M1 module, identified based on intramodular connectivity. (H) Network-trait association plot showing the relationships between the top 3 hub genes in each module (M1-M4) and key processes of neuronal development, including stem cell proliferation, neuronal differentiation, and neuronal maturation, illustrating coordinated regulation of early neurodevelopmental programs. M1 genes exhibited high expression in NPCs and decreased progressively during neuronal differentiation (Figures 4A and 4B). GO analysis of genes within M1 showed enrichment in processes related to mitotic progression, chromatin organization, and early neurodevelopmental programs (Figure 4C; Table S3). In contrast, M2 and M3 genes, enriched in neuronal development and function (Figures S5D and S5E; Table S3), had higher levels of expression in neurons compared to NPCs (Figures 4A and 4B), and were induced during early and late neuronal differentiation, respectively. M2 genes were progressively induced during neuronal differentiation in Ctrl cells but failed to increase in FXR1-KO. M4 genes, enriched in functions of mature neurons such as calcium signaling and action potentials (Figure S5F; Table S3), had the highest expression in 4-weeks Ctrl neurons, but failed to upregulate in FXR1-KO neurons (Figures 4A and 4B). Together, these results indicate that FXR1 deficiency disrupts coordinated stage-specific transcriptional programs during neuronal differentiation. To quantify the relationship between each co-expression module and neuronal differentiation, and determine how these associations were altered by FXR1 deficiency, we performed a gene set variation analysis (GSVA)-based module trait analysis (Figure 4D) and a temporal and genotype-based module-trait analysis (Figure 4E). Module-trait analysis confirmed that the co-expression modules identified by WGCNA were strongly associated with distinct stages of neurodevelopment (Figure 4D). We next determined how these developmental associations were altered in FXR1-KO compared to Ctrl. The M1 module exhibited a less negative correlation with FXR1-KO compared to Ctrl at 1-week differentiation (Figure 4E), indicating that genes within this module are insufficiently repressed during FXR1-KO neuronal differentiation. In contrast, the M2, M3, and M4 modules were positively correlated with neuronal differentiation and maturation traits, with M2 exhibiting the most negative correlation with NPCs and M4 exhibiting the most positive correlation with 4-weeks of differentiation (Figure 4D). However, M2, M3, and M4 modules were differentially affected in FXR1-KO (Figure 4E). The positive correlations between 1-week of differentiation and M2, and between 4-weeks of differentiation and M2 and M4, are either reduced or absent in FXR1-KO neurons compared to Ctrl. Among these modules, the M4 module exhibited the most dramatic reduction in 4-weeks of differentiation, suggesting impaired neuronal maturation. On the other hand, the M3 module displays a significantly higher correlation with 4-weeks FXR1-KO neurons than Ctrl neurons, suggesting abnormal induction of gene expression. In fact, many genes upregulated in FXR1-KO 4-weeks neurons are associated with RNA metabolism, stress response, and cell-cycle regulation, consistent with the abnormal presence of proliferating NPCs in FXR1-KO 4-weeks neurons (Figures 1N–1P). To identify the gene module most likely to mediate FXR1-dependent regulation of neuronal differentiation, we performed gene network analysis. Because M1 exhibited the highest network connectivity, we hypothesized that changes within the M1 might have the greatest impact on neuronal differentiation (Figure 4F). We then performed gene expression network analysis and identified highly interconnected hub genes in each module (Figures 4G and S5G–S5I). Many hub genes in M1 are implicated in cell cycle regulation and developmental control. HMGB2 (high mobility group box 2), CDT1 (chromatin licensing and DNA replication factor 1), and NOTCH3 (notch receptor 3) were the three highest-ranked hub genes in M1 module, with HMGB2 showing the highest intramodular connectivity and strongest contribution to the module (Figure 4G). We then performed association analysis of the top three hub genes from each module with neurodevelopmental processes, including NPC proliferation, neuronal differentiation, and neuronal maturation (Figure 4H). NOTCH3 was not differentially expressed in FXR1-KO NPCs but was upregulated at 1-week of differentiation (Table S2). Because Notch signaling is known to be important for NPC proliferation and neuronal commitment (McLaren and Butts, 2025), we treated FXR1-KO NPCs with the Notch signaling inhibitor, γ-secretase inhibitor Compound E (CE). However, CE treatment at 10× higher than what we normally use for neuronal differentiation (2 vs. 0.2 μM) failed to rescue neuronal differentiation deficits of FXR1-KO NPCs (Figures S6A and S6B). These findings suggest that NOTCH3 upregulation is more likely a downstream consequence of impaired neuronal differentiation rather than a primary driver of the NPC phenotype. CDT1 and HMGB2 showed not only the strongest positive associations with NPC proliferation but also the strongest inverse correlation with neuronal differentiation and maturation (Figure 4H). CDT1 is implicated in DNA replication in proliferating cells (Pozo and Cook, 2016; Spella et al., 2007), but CDT1 mRNA levels did not show significant changes in FXR1-KO NPCs (Table S2). HMGB2 is a chromatin-binding transcription factor that has been shown to play a key role in NPC proliferation and neurogenic-gliogenic fate transitions in mouse cortical development (Abraham et al., 2013a, 2013b; Lee et al., 2010). In addition, HMGB2 is a target of FXR1 in human K562 cancer cells (Li et al., 2014). We found that both HMGB2 mRNA and HMGB2 protein levels were upregulated in both FXR1-KO NPCs and at 1-week of differentiation compared to Ctrl (Figures 5A–5D). We therefore focused on HMGB2 as a potential mediator of FXR1-mediated regulation of the neurogenic potential of human NPCs. FXR1 binds and regulates stability of HMGB2 mRNA (A) HMGB2 mRNA levels measured by quantitative reverse-transcription PCR (RT-qPCR) in Ctrl and FXR1-KO cells in NPCs, 1-week, and 4-weeks of differentiation samples, normalized to GAPDH mRNA levels. (B–D) Representative immunoblots (B) and quantification of HMGB2 protein levels in in FXR1-KO and control in NPCs (C) and 1-week differentiated neurons (D) normalized to GAPDH protein levels. (E and F) FXR1 RIP-qPCR analysis showing enrichment of HMGB2 mRNA in Ctrl compared to FXR1-KO in NPCs (E) and 1-week of differentiation (F). Two-tailed Student’s t test with Welch’s correction. (G) Schematic (top) of human iPSC with 3XFLAG knocked into the endogenous FXR1 gene (in Ctrl3 iPSC line) to express C-terminal FLAG-tagged FXR1. Immunoblot (bottom) showing the FXR1-FLAG expression in FXR1-FLAG iPSCs compared to parent Ctrl3 iPSCs. (H) RIP-qPCR analysis showing enrichment of HMGB2 mRNA in immunoprecipitation using an FLAG antibody normalized to that using control IgG in NPCs. Two-tailed Student’s test. (I) HMGB2 mRNA stability analysis, showing elevated HMGB2 mRNA stability in FXR1-KO compared to Ctrl NPCs. The half-lives (T1/2) of HMGB2 mRNA were calculated. Two-way ANOVA with Bonferroni correction. (J) Proposed model for opposite effects of FXR1 and HuR on HMGB2 mRNA stability and impact of FXR1-KO. (K) RIP-qPCR showing elevated binding of HMGB2 mRNA by HuR in FXR1-KO NPCs compared to controls. Two-way ANOVA. (L) Schematic illustration for lentivirus-mediated overexpression of FXR1 (LV-FXR1-OE) and HuR (LV-HuR-OE) in NPCs, as well as control (LV-NC), followed by RNA-IP (M) FXR1 mRNA levels in NPCs infected with LV-FXR1 (FXR1-OE) and control LV-NC (NC), measured by RT-qPCR. (N and O) Representative immunoblots (N) and quantification of FXR1 protein levels in FXR1OE and NC in NPCs. (P) ELAVL1 mRNA levels measured by RT-qPCR in HuROE compared to NC. (Q and R) Representative immunoblots (Q) and quantification of HuR protein levels in HuROE and NC in NPCs (R). (S) RIP-qPCR analysis showing enrichment of HMGB2 mRNA in immunoprecipitation using a FXR1 antibody normalized to NC in NPCs. (T) RIP-qPCR analysis showing enrichment of HMGB2 mRNA in immunoprecipitation using a HuR antibody normalized to NC in NPCs. (A–F, I, and K) Data are presented as mean ± SEM from n ≥ 4 independent differentiations derived from 2 iPSC lines per group. (H) n = 5 independent differentiations derived from 1 iPSC line. (M–T) N = 3 independent iPSC lines (1 independent differentiation batch per iPSC line). Each cell line is plotted with a different shape. Data are shown as mean ± SEM. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001 FXR1 directly binds and destabilizes HMGB2 mRNA in NPCs and during differentiation We hypothesized that FXR1 regulates HMGB2 mRNA post-transcriptionally to preserve neurogenic potential of human NPCs. We first examined whether HMGB2 mRNA is a direct target of FXR1 in human NPCs. RNA immunoprecipitation (RIP) using an anti-FXR1 antibody followed by quantitative PCR (RIP-qPCR) showed enrichment of HMGB2 mRNA bound by FXR1 in Ctrl but not in FXR1-KO NPCs (Figures 5E and 5F). To further validate FXR1 binding to HMGB2 mRNA, we generated an FXR1-tagged human iPSC line by knocking in a 3XFLAG tag to the endogenous FXR1 gene (Figures 5G, 5H, and S6C–S6E). RIP-qPCR with an anti-FLAG antibody revealed significant enrichment of HMGB2 mRNA compared to IgG RIP controls in NPCs (Figure 5H). FXR1 is known to regulate mRNA through multiple post-transcriptional mechanisms, including RNA stability (Khan et al., 2024; Mendez-Albelo et al., 2024). Since HMGB2 mRNA levels are higher in FXR1-KO NPCs (Figure 5A), we assessed HMGB2 mRNA stability upon addition of actinomycin D to block transcription. In control NPCs, HMGB2 transcript abundance declined progressively after transcriptional inhibition, with an estimated half-life of 6.68 h. In contrast, the half-life of HMGB2 mRNA was significantly longer (10.44 h) in FXR1-KO NPCs, indicating that the loss of FXR1 leads to increased HMGB2 mRNA stability (Figure 5I). It has been shown that HuR (ELAVL1), a well-established stabilizer of target transcripts, binds HMGB2 mRNA in cancer cells (Brennan and Steitz, 2001; Li et al., 2014). Since both FXR1 and HuR are expressed in human NPCs and 1-week neurons (Table S2), we hypothesized that FXR1 antagonizes HuR-mediated regulation of HMGB2 mRNA in human NPCs (Figure 5J). Indeed, there was increased HuR binding of HMGB2 mRNA in FXR1-KO NPCs (Figure 5K). We next tested whether overexpression (OE) of one RBP alters the binding of the other to HMGB2 mRNA (Figure 5L). NPCs derived from three independent control iPSCs (Ctrl 1, Ctrl 2, and Ctrl 3) were transduced with LV expressing HuR (HuR-OE: ENST00000407627.7) or FXR1 (FXR1-OE: ENST00000445140.6) (Figures 5L–5Q). HuR-OE significantly reduced FXR1 binding to HMGB2 mRNA compared with NC (Figure 5S), whereas FXR1-OE reduced HuR binding to HMGB2 mRNA (Figure 5T), supporting the antagonistic effect between these two RBPs in regulating HMGB2 mRNA. These results demonstrate that FXR1 directly binds and promotes the degradation of HMGB2 mRNA, therefore controlling the abundance of HMGB2 transcripts in NPCs and during neuronal differentiation. HMGB2 knockdown in FXR1-KO NPCs rescues neuronal differentiation deficits Based on our results, we hypothesized that HMGB2 is a key post-transcriptional effector of FXR1 in regulating NPC proliferation and neurogenic potential during human cortical neuronal differentiation. To test this hypothesis, we knocked down HMGB2 in FXR1-KO NPCs and assessed its impact on NPC proliferation, neuronal differentiation, and electrophysiological maturation (Figures 6A and S6F–S6J). HMGB2 knockdown rescued proliferation (EdU+ SOX2+) of FXR1-KO NPCs to the Ctrl level (Figures 6B and 6C). In addition, HMGB2 knockdown led to a significant reduction of proliferating (Ki67+SOX2+) NPCs in FXR1-KO cultures at 4-weeks of differentiation (Figures 6D–6F), although not at 1-week of differentiation (Figures S6K–S6M). Moreover, HMGB2 knockdown in FXR1-KO NPCs led to an increased proportion of TUBB3+ and MAP2+ immature neurons at both 1-week (Figures S6N–S6P) and 4-weeks (Figures 6G–6I) of differentiation. We next examined whether HMGB2 knockdown rescues the neuronal maturation deficits in FXR1-KO. HMGB2 knockdown in FXR1-KO NPCs rescued neurite length and complexity of FXR1-KO neurons at both 1-week (Figures S6Q–S6T) and 4-weeks of differentiation (Figures 6J–6M). In addition, HMGB2 knockdown increased the number of NeuN+ mature neurons in 4-weeks of differentiation (Figures 6N and 6O). Furthermore, HMGB2 knockdown increased both the firing rate and burst number above Ctrl levels at 5-weeks of differentiation (Figures 6P and 6Q). Taken together, these findings demonstrate that elevated HMGB2 levels contribute to reduced neurogenic potential, which can be rescued by reducing HMGB2 levels in FXR1-KO NPCs. HMGB2 knockdown in FXR1-KO NPCs rescues NPC proliferation and neuronal differentiation (A) Schematic of the experimental design for lentiviral shRNA knockdown of HMGB2 (LV-shHMGB2-mScarlet) in NPCs. (B) Representative immunofluorescent images of NPCs transduced with control (NC) Ctrl (Ctrl-NC) cells and NC (KO-NC) and shHMGB2 (KO-shHMGB2) in FXR1-KO cells, stained for SOX2 (white) and EdU. LV infected cells expressed mScarlet (red fluorescence protein reporter for LV-shRNA transduction). (C) Quantification of proliferating NPCs (EdU+/SOX2+) among total LV-shRNA infected cells, presented as the percentage of EdU+SOX2+mScarlet+ cells among total mScarlet+ cells. (D) Representative immunofluorescence images of neurons at 4-weeks of differentiation, stained for SOX2 (white) Ki67, and mScarlet (red). (E and F) Quantification of (SOX2+) NPCs (E) and proliferating (Ki67+/SOX2+) NPCs (F) among total LV-shRNA infected (mScarlet+) cells at 4-weeks of differentiation. (G–I) Representative immunofluorescence images (G) and quantification of TUBB3+ (H) and MAP2+ (I) neurons at 4-weeks of differentiation. (J–M) Representative immunofluorescent images (J) and morphological analysis (K–M) of MAP2+ and mScarlet+ neurons. Scale bars, 10 μm. (N and O) Representative immunofluorescent images (N) and quantification (O) of neurons at 4-weeks of differentiation stained with the mature neuronal marker NeuN (magenta). (P and Q) Mean firing rate (P) and number of bursts (Q) of LV-shRNA infected (mScarlet+) cells at 5-weeks of differentiation. Scale bars (except for J), 25 μm. (C, E, F, H, I, and O–Q) Two-way ANOVA. Data are shown as mean ± SEM from n ≥ 3 independent differentiations derived from 2 iPSC lines per group. (K–M) Two-way ANOVA. Error bars are shown as mean ± SEM from n = 22 fields of view from 4 independent differentiations derived from 2 independent iPSC lines per group. Each cell line is plotted with a different shape. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. Discussion Neurodevelopment requires precise cell fate transitions from proliferating NPCs to differentiated cells. Here, we identified FXR1 as a critical post-transcriptional regulator of human neuronal differentiation. By controlling the transcript abundance of functionally important genes, such as HMGB2, FXR1 maintains the neurogenic potential of human NPCs. Our study also reveals that FXR1 is a part of the RBP post-transcriptional regulatory network in human NPCs. In this context, FXR1 and HuR exhibit antagonistic regulation of the same transcripts, and their binding leads to distinct outcomes. Our results unveil a critical role for post-transcriptional regulation in human neurodevelopment. Precise control of stem cell self-renewal and differentiation is critical for brain development, and disruption of this balance causes developmental arrest and disorders (MacPherson et al., 2021). Therefore, transcriptional mechanisms regulating lineage differentiation have been a major focus of neurodevelopmental research, and recent technological advancements have provided substantial insights into the roles of transcription factors in this process. For example, SOX2, paired box 6 (PAX6), Notch, and HMGB2 are highly expressed in NPCs and have been shown to be critical for maintaining NPC proliferation, while NGN2 (neurogenin 2), NEUROD1 (neuronal differentiation 1), and ASCL1 (achaete-scute family basic helix-loop-helix [bHLH] transcription factor 1) induce neuronal differentiation (Nowakowski et al., 2025). However, an increasing number of studies, including this one, demonstrate that NPCs express neuronal mRNAs prior to differentiation, even though these mRNAs are not translated into proteins until after differentiation occurs (Telley et al., 2019; Yoon et al., 2017; Zahr et al., 2018). Hoye and Silver (2021) have proposed that in developing cortex, NPCs are “primed” with mRNAs for differentiated lineages, but these differentiation-associated mRNAs are post-transcriptionally repressed in NPCs, which allows for fast initiation of protein expression upon induction of differentiation, because post-transcriptional regulation is at least 10× faster than transcriptional regulation (Hoye and Silver, 2021). This is an intriguing model, similar to the bivalent chromatin state found in human pluripotent stem cells (hPSCs) (Bernstein et al., 2006). However, only limited studies have provided experimental evidence for this model. One example is DDX3X (DEAD-box helicase 3 X-linked), an RBP whose functional loss leads to DDX3X syndrome with severe cortical malformation and ASD. It has been shown that DDX3X binds many transcripts, including neurogenic transcripts in NPCs, and regulates the cell cycle of NPCs during mouse cortical neurodevelopment (Hoye et al., 2022). However, whether similar post-transcriptional regulation of NPCs occurs during human neurodevelopment remains unclear. Our data show that FXR1 is one of the post-transcriptional regulators in human NPCs and that it preserves the neurogenic potential of human NPCs by controlling transcript abundance of key genes such as HMGB2. Using WGCNA, we found that several gene co-expression modules exhibit dynamic changes across differentiation stages. FXR1 deficiency shifts the balance between genes involved in cell cycle and genes associated with differentiated neurons. The M1 module, which contains many cell-cycle and chromatin-related genes, was most active in NPCs and dysregulated when FXR1 was absent. These results suggest that FXR1 controls the abundance of transcripts associated with NPC stage, such as HMGB2, allowing NPCs to properly exit the cell-cycle upon neuronal differentiation. Without FXR1, elevated HMGB2 levels inhibit cell cycle exit, therefore a subset of NPCs remains in the cell cycle throughout neuronal differentiation. These results are consistent with our previous observation in FXR1-deficient adult mouse NPCs (Patzlaff et al., 2017), suggesting a common mechanism of FXR1 in NPCs and neurodevelopment. FXR1-deficient NPCs exhibit not only reduced neuronal differentiation but also impaired morphological and electrophysiological maturation, indicating that proper NPC differentiation is essential for neuronal maturation. In fact, it has been shown that deficiencies in NPC genes, such as SOX2 (Zhang and Cui, 2014) and PAX6 (Hu et al., 2025; Xu et al., 2021), lead to impaired neuronal differentiation and functional maturation (Hewitt et al., 2023; Marchetto et al., 2017). It is possible that impaired neuronal maturation is a combined result of NPC differentiation and the neuron-specific functions of FXR1, since silencing FXR1 in postmitotic mouse hippocampal neurons decreased neurite length (Shimizu and Hohjoh, 2023) and deletion of FXR1 in postmitotic parvalbumin interneurons reduced neurite complexity and excitability (Shen et al., 2021). Future studies of neuron-specific functions of FXR1 will provide further insight into FXR1 functions in human brain development. Our network identified HMGB2 as the central hub gene in the most prominently affected M1 gene module. HMGB2 is a chromatin-associated architectural factor that has been shown to transcriptionally regulate NPC proliferation, differentiation, and neurogenic output (Bronstein et al., 2017; Kimura et al., 2018; Lee et al., 2010; Maddhesiya et al., 2025). We demonstrated that FXR1 directly binds HMGB2 mRNA in both NPCs and 1-week-differentiated neurons and limits HMGB2 transcript abundance as cells exit the cell cycle and commit to neuronal differentiation. Many NPC- and cell cycle-associated transcripts harbor adenylate-uridylate (AU)-rich elements (AREs) and are regulated by competing RBPs that either stabilize (e.g., HuR) or destabilize (e.g., FXR1) the transcripts (Herman et al., 2018). Although HMGB2 is not a classical ARE-containing transcript, the presence of conserved and accessible AU/U-rich motifs provides a plausible substrate for antagonistic regulation by RBPs with opposing effects on mRNA stability, including stabilizing factors such as HuR and decay-promoting factors such as FXR1 (Fallmann et al., 2016; Herman et al., 2018). Our data identify FXR1 and HuR as antagonistic post-transcriptional regulators of HMGB2 mRNA abundance in human NPCs: FXR1 destabilizes HMGB2 mRNA, while HuR stabilizes it. It is unlikely that FXR1 and HuR directly compete for the binding sites on HMGB2 mRNA, because published binding studies have shown that FXR1 primarily binds within the coding sequence, whereas HuR-binding motifs are predominantly located within the 3′ UTR (Ascano et al., 2012; Lebedeva et al., 2011; Tran et al., 2019). Our data support the notion that complex interactions among RBPs are important for regulating dynamic cellular processes (Schieweck et al., 2021). Our data showed that reducing HMGB2 levels in FXR1-deficient NPCs is sufficient to restore NPC proliferation, neuronal differentiation, and morphological and electrophysiological maturation, establishing HMGB2 as a key functional effector downstream of FXR1. We observed that HMGB2 knockdown did not simply restore neuronal activity in FXR1-KO 5-weeks differentiated neurons but increased the activity to above the control levels, suggesting that HMGB2 dosage is critical for maintaining normal neuronal network excitability. Extensive studies have shown that FXR1 regulates proliferation in non-neuronal systems, mostly cancer cells (Majumder et al., 2016, 2020; Qian et al., 2015). Our data demonstrate important roles of FXR1 in human neurodevelopment and support a model in which FXR1 promotes neurodevelopmental progression by actively repressing NPC-enriched transcripts through post-transcriptional mechanisms. Future studies are needed to determine how FXR1 selectively engages transcripts such as HMGB2 and whether FXR1 regulation of NPCs involves storing transcripts through liquid-liquid phase separation, as in spermatogenesis (Kang et al., 2022). Limitations of the study We did not investigate the effects of FXR1 reduction on human inhibitory interneurons. Given the impact of FXR1 deletion in parvalbumin interneurons in mouse models, we anticipate a significant function of FXR1 in interneuron differentiation. In addition, we have identified HMGB2 as a target of FXR1; however, it is possible that disruption of FXR1-dependent post-transcriptional regulatory networks, rather than HMGB2 alone, contributes to the NPC deficits. Furthermore, FXR1 may control distinct RNA targets within other gene modules (e.g., M2, M3, and M4), which contribute to the deficits we observed in mature neurons, such as reduced neuronal activity. Furthermore, our results suggest that FXR1 reduces the abundance of HMGB2 by destabilizing its mRNA. However, it remains to be investigated how FXR1 selects its targets and regulates the stability of mRNA targets. Since FXR1 and HuR may not bind the same sites of HMGB2 mRNA, how the binding of one RBP inhibits the binding of the other RBP is unclear. Additionally, although human iPSC-derived models offer a robust platform to investigate human-specific neurodevelopmental mechanisms, in vivo studies are necessary to elucidate how FXR1-dependent post-transcriptional programs interact with cell-type-specific and circuit-level processes during cortical development. Resource availability Lead contact Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Xinyu Zhao (xinyu.zhao@wisc.edu). Materials availability Requests for materials can be made to the lead contact. Data and code availability RNA sequencing (RNA-seq) data have been deposited to Gene Expression Omnibus (GSE318782). This paper does not report original codes. Acknowledgments We thank Dr. Sara Knaack, Ethan Dayley, Ruthika Ajit and Jensen Weik, and Dr. Nithin Raveendran for their technical assistance. We thank Dr. K. Knobel and C. Sorefat the Waisman IDD Models Core (grant from the National Institute of Child Health and Human Development P50HD105353); we especially thank Dr. A. Bhattacharyya, J. Svaren, and A.M.M. Sousa for their input. This work was supported by grants from the National Institutes of Health (R01MH118827, R01MH136152, R01NS138268 to X.Z.; P50HD105353 to Waisman Center; a diversity supplement to R01MH118827 and R36MH136790 for S.O.S.; R21NS142497 to Y. Gao.), DOD IIRA grant W81XWH-22-1-0621 (to X.Z.), Simons Foundation SFARI pilot grant, Eagles Autism Foundation pilot and translational grants, Jenni and Kyle Professor for Novel Neurodevelopmental Diseases, Kellet Mid-Career Award, and Vilas Distinguished Achievement Professorship (to X.Z.), Science and Medicine Graduate Research scholarship (to S.O.S. and N.M.M.-A.), T32 GM141013 molecular pharmacology training grant and Stem Cell and Regenerative Medicine Center (SCRMC) (to N.M.M.-A.), Morris Aprison Scholarship for undergraduate research (to L.A.P. and P.Y.), Simons Foundation Shenoy Undergraduate Research Fellowship in Neuroscience (SURFiN, to R.M.D.), postdoctoral fellowships from FRAXA, the SCRMC, and the Autism Science Foundation (to C.L.S.), and Warren Alpert Distinguished Scholarship (to Y. Guo). Author contributions X.Z. conceived the concept, designed experiments, wrote the manuscript, and secured funding; N.M.M.-A. designed experiments, collected and analyzed data, and wrote the manuscript; Y.Z. performed the transcriptomic analyses; S.O.S performed calcium imaging analysis and collected data; Y. Guo isolated mouse astrocytes; Y. Gao designed shRNAs and OE plasmids; C.L.S. designed FXR1-KO iPSCs lines; L.A.P., R.V.S., R.M.D., P.Y., A.B.W., D.N.T., Z.X., C.J.J., and M.T.E. collected data. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesAnti-FLAGSigma-Aldrich (MilliporeSigma)Cat# F1804; RRID: AB_262044Anti-FXR1ProteintechCat# 13194-1-AP; RRID: AB_2110702Anti-MAP2Sigma-Aldrich (MilliporeSigma)Cat# M1406; RRID: AB_477171Anti-MAP2AbcamCat# ab5392; RRID: AB_2138153Anti-NeuNCell Signaling Technology (CST)Cat# 24307; RRID: AB_2651140Anti-TRA1-81 (PODXL)Millipore (Merck/MilliporeSigma)Cat# MAB4381; RRID: AB_177638Anti-Oct4 (POU5F1)Santa Cruz BiotechnologyCat# SC-5279; RRID: AB_628051Anti-PAX6BioLegendCat# 901301;RRID:AB_2565003Anti-SOX2R&D Systems (AF = goat polyclonal series)Cat# AF2018; RRID: AB_355110Anti-SOX2Aves LabCat# SOX2-0020; RRID: AB_3068326Anti-B-Tubulin (TUBB3)BioLegend (formerly Covance/BioLegend)Cat# MMS-435P; RRID: AB_2313773Anti-Ki67 (SP6)InvitrogenCat# MA5-14520RRID:AB_10979488Anti-Ki67eBioscienceCat# 14-5698RRID:AB_2865120Anti-HMGB2ProteintechCat# 14597-1-AP; RRID: AB_2117919Anti-HuRProteintechCat# 11910-1-AP; RRID: AB_11182183Anti-mCherryThermofisherCat# M11217; RRID: AB_2536611Anti-GFAPThermofisherCat# 13-0300; RRID: AB_2532994Anti-S100BProteintechCat# 66616-1-Ig; RRID: AB_2881976Anti-GFPInvitrogenCat # A10262; RRID:AB_2534023IRDye® 800CW Anti-Rabbit IgG Goat Secondary AntibodyLICORbioCat# 926-32211; RRID: AB_621843IRDye® 800CW Goat Anti-Mouse IgG Polyclonal Secondary AntibodyLICORbioCat# 926-32210; RRID: AB_621842IRDye® 680LT Goat Anti-Rabbit IgG Polyclonal Secondary AntibodyLICORbioCat# 926-68021; RRID: AB_10706309IRDye® 680LT Goat Anti-Mouse IgG Polyclonal Secondary AntibodyLICORbioCat# 926-68020; RRID: AB_10706161Alexa Fluor Goat anti-Mouse 488ThermofisherCat# A11029; RRID: AB_2534088Alexa Fluor Goat anti-Mouse 568ThermofisherCat# A11031; RRID: AB_144696Alexa Fluor Goat anti-Mouse 647ThermofisherCat# A21236; RRID: AB_2535805Alexa Fluor Goat anti-Mouse 546ThermofisherCat# A11003; RRID: AB_2534071Alexa Fluor Goat anti-Rabbit 488ThermofisherCat# A11008; RRID: AB_143165Alexa Fluor Goat anti-Rabbit 568ThermofisherCat# A11036; RRID: AB_10563566Alexa Fluor Goat anti-Rabbit 647ThermofisherCat# A21245; RRID: AB_2535813Alexa Fluor Goat anti-Chicken 488ThermofisherCat# A11039; RRID: AB_2534096Alexa Fluor Goat anti-Chicken 568ThermofisherCat# A11041RRID:AB_2534098Alexa Fluor Goat anti-Chicken 647ThermofisherCat# A21449; RRID: AB_1500594Alexa Fluor Goat anti-Rat 568ThermofisherCat# A11077; RRID: AB_2534121Alexa Fluor Goat anti-Rat 546ThermofisherCat# A11081RRID:AB_2534125Chemicals, peptides, and recombinant proteinsDMEM/F12Thermo Fisher ScientificCat# 11330032NeurobasalThermo Fisher ScientificCat# 21103049KnockOut Serum Replacement (KOSR)Thermo Fisher ScientificCat# 10828028Antibiotic-AntimycoticThermo Fisher ScientificCat# 15240062Dispase IIThermo Fisher ScientificCat# 17105041Collagenase, Type IVThermo Fisher ScientificCat# 17104019StemPro™ AccutaseThermo Fisher ScientificCat# A1110501TrypLE Express EnzymeThermo Fisher ScientificCat# 12605010B-27 (without vitamin A)Thermo Fisher ScientificCat# 12587020GlutamaxThermo Fisher ScientificCat# 35050061L-glutamineThermo Fisher ScientificCat# 25030081Laminin Mouse Protein, NaturalThermo Fisher ScientificCat# 23017015MEM NEAAThermo Fisher ScientificCat# 11140050SB431542BiogemsCat# 3014193CAS: 301836-41-9LDN-193189Selleck ChemicalsCat# S2618; CAS: 1062368-24-4XAV 939TocrisCat# 3748; CAS: 284028-89-3Y-27632TocrisCat# 1254; CAS: 129830-38-2GlucoseDot ScientificDSG32040Recombinant Human/Murine/Rat BDNFPeproTechCat# 450-02Recombinant Human GDNFPeproTechCat# 450-10DMEM, High GlucoseSigma-AldrichCat# D5796L-ascorbic acidSigma-AldrichCat# A8960; CAS: 1713265-25-8Cyclic AMP (cAMP)Sigma-AldrichCat# D0260; CAS: 241-059-4γ-Secretase Inhibitor XXI, Compound ESigma-AldrichCat# 565790; CAS: 209986-17-42-mercaptoethanolSigma-AldrichCat# M6250; CAS: 60-24-2Dimethyl sulfoxide (DMSO)Sigma-AldrichCat# D2650; CAS: 67-68-5Poly-L-ornithineSigma-AldrichCat# P4638; CAS: 27378-49-0Critical commercial assaysVF 488 Click-iT EdU Universal Cell Proliferation Detection Kit 100TMedChem ExpressCat# HY-K1087Deposited dataRNA-seqGene Expression OmnibusGSE318782Experimental models: Cell linesGM1 (GM00498-4)Coriell InstituteRRID: CVCL_7290WC031i-5907-6 (WISCi008-C)WiCellRRID: CVCL_VF57WC032i-6007-1 (WISCi009-A)WiCellRRID: CVCL_VF586007FXR1KO_C14This paper–GM1FXR1KO_C13This paper–WC031i-5907-6-FLAG-C6This paper–Experimental models: Organisms/strainsLV-shNC-mScarletThis paper–LV-shHMGB2-mScarletThis paper–LV-Syn1-GCaMP7fThis paper–LV-shFXR1-mCherryShen et al. (2021)–LV-shNC-mCherryShen et al. (2021)–LV-HuR-mScarletThis paper–LV-FXR1-mScarletThis paper–OligonucleotidesPlease see Table S4––Recombinant DNApLV-shNC-mScarletThis paper–pLV-shHMGB2-mScarletThis paper–pLV-Syn1-GCaMP7fThis paper–pLV-shFXR1-mCherryShen et al. (2021)–pLV-shNC-mCherryShen et al. (2021)–pLV-HuR-mScarletThis paper–pLV-FXR1-mScarletThis paper–Software and algorithmsPrism (v10)GraphPadhttps://www.graphpad.com; RRID:SCR_000306ImageJNational Institute of Healthhttps://imagej.nih.gov/ij/docs/guide/user-guide.pdf; RRID:SCR_002285ImageStudio (v5.2)https://www.licor.com/bio/products/software/image_studio/;RRID: SCR_015795R (v4.5.0)https://www.r-project.org/RRID:SCR_001905Trimmomatic v0.39https://www.usadellab.org/cms/index.php?page=trimmomaticRRID: SCR_011848STAR v2.7.11bhttp://code.google.com/p/rna-star/RRID: SCR_004463featureCounts v2.2.1https://github.com/byee4/featureCountsRRID: SCR_012919DESeq2 v1.46.0https://bioconductor.org/packages/release/bioc/html/DESeq2.htmlRRID: SCR_015687AxIS Navigator (v3.7.1)Axion Biosystemshttps://www.axionbiosystems.com/products/mea/mea-softwareRRID: SCR_016308AxIS Metric Plotting Tool (v2.5.0)Axion Biosystemshttps://www.axionbiosystems.com/products/mea/mea-softwareMetaXpress (v6.7.2.290)Molecular Deviceshttps://www.moleculardevices.com/products/cellular-imaging-systems/acquisition-and-analysis-software/metaxpress; RRID:SCR_016654StereoInvestigatorMBF Biosciences, Inchttps://www.mbfbioscience.com/products/stereo-investigator; RRID:SCR_002526NeurolucidaMBF Biosciences, Inchttps://www.mbfbioscience.com/products/neurolucida; RRID:SCR_001775Python (v3.8)Python Software Foundationhttps://www.python.org/downloads/release/python-380/; RRID:SCR_008394CaImANPnevmatikakisLabhttps://github.com/flatironinstitute/CaImAn; RRID:SCR_021152LASXLeicahttps://www.leica-microsystems.com/products/microscope-software/p/leica-las-x-ls/; RRID:SCR_013673 Experimental model and study participant details Generation of FXR1KO and FXR1-FLAG gene edited hiPSC lines Previously published human iPSC lines used as controls in this study include Ctrl1 (GM00498-4, GM1, skin fibroblast, 3-year-old, male, Coriell Institute), Ctrl2 (WC032i-6007-1, skin fibroblast, neonatal, male, WiCell) and Ctrl3 (WC031i-5907-6, skin fibroblast, neonatal, male, WiCell). FXR1-KO iPSC lines (KO1 and KO2) were generated from Ctrl1 and Ctrl2, respectively (Guo et al., 2023; Shen et al., 2023; Yin et al., 2019). FXR1-FLAG iPSC line was generated using Ctrl3. Detailed information in Table S1. Generation of the two FXR1-KO hiPSC lines (KO-1, KO-2), was carried out using published methods (Li et al., 2020). Briefly, control hiPSCs (Ctrl1 and Ctrl2), were dissociated to single cells with Accutase (Thermo Fisher Scientific) and washed with PBS. 5×106 cells were electroporated using the 4D-NucleofectorTM System (Lonza) and the P3 Primary Cell 4D Nucleofector X Kit (Lonza, V4XP-3024), parameters CG104. Using 2.5ug of plasmid of FXR1_sgRNA2 was used (Integrated DNA Technologies). The electroporated cells were plated at low density onto mouse embryonic fibroblast (MEFs) in hESC Media (DMEM/F12 (Thermo Fisher Scientific, 11330032), KnockOut™ Serum Replacement (KOSR) (Thermo Fisher Scientific, 10828028), L-glutamine (Thermo Fisher Scientific, 25030081) and 2x FGF2 (Waisman Biomanufacturing, 8 ng/mL) supplemented with 10μM ROCK inhibitor (Y-27632 dihydrochloride, Tocris, # 1254)) About two weeks later, individual colonies were picked for expansion and screening for indels by PCR using primers targeting the sequence (AGAGCCATGTGGGTGGTGGTTGGCTAAAGTTCGGATGATGAAAGG) and subjected to Sanger sequencing. The indels were analyzed using Tracking of Indels by DEcompostion (Shen et al.). The absence of FXR1 protein in FXR1-KO iPSC lines was confirmed by western blot, using an antibody against FXR1 (Proteintech). FXR1-FLAG hiPSCs were generated using published methods (Li et al., 2020). Briefly, hiPSCs (Ctrl3) were cultured in TeSR-PLUS media (StemCell Technologies) on Matrigel coated plates until around 80% confluency. The sgRNA (TCTAAGCTACAGCGTACTCC) targeting the C-terminal region of FXR1 was designed using CRISPOR design too (Haeussler et al., 2016) and WU-CRISPR (Hiranniramol et al., 2020; Wong et al., 2015) and was ordered from Synthego. Donor plasmid (pDonor-hFXR1-3xFLAG "DYKDDDDK") was constructed by VectorBuilder 5’ and 3’ homology arms (Position 1-1500 and 1660-3159) and 3x FLAG tag (Position 1591-1656). sgRNA was pooled with 4ug Cas9 Nuclease protein (TrueCut Cas9 Protein V2, Thermo Fisher Scientific) and 5uL of Neon Buffer R (Invitrogen) to promote Cas9-RNP complex formation. After 15mins of incubation, 3.0ug of donor plasmid was added to the complex. hPSCs were disassociated using 0.5mM EDTA: Accutase (1:1) for 3-4 minutes and washed with PBS. 4x106 PSCs were resuspended in Buffer R and mixed with the Cas9-RNP complex. Electroporation was done using NEON electroporation format using 1200V, 30 msec, 1x pulse settings. Cells were pooled following four rounds of electroporation and plated in TeSR-Plus media at low density. Selection was done using Puromycin for 3 days and clones were expanded then screened using PCR. The top 5 predicted off target sites for each sgRNA predicted by the CRISPOR algorithms were tested by PCR and gel purification (Zymoclean Gel DNA Recovery Kit), then Sanger sequenced by Quintara Biosciences. For the iPSC lines generated in this study, pluripotency was assessed by immunofluorescent staining for stem cell markers OCT4 (POU5F1), SOX2, and TRA-1-81 (PODXL), and a neuroepithelial marker, PAX6 (Li et al., 2020). Karyotype and mycoplasma analysis were performed using WiCell services. Human iPSCs culture and neural differentiation Human iPSCs culture and differentiation were carried out using published protocols (Guo et al., 2023; Shen et al., 2023; Sirois et al., 2024). Briefly, to start differentiation, hPSCs were thawed onto MEF feeder plates in hESC media supplemented with 10μM ROCK inhibitor (Y-27632 dihydrochloride, Tocris, # 1254). Cells were passaged every seven days onto new MEF plates using collagenase Type IV (Thermo Fisher Scientific #17104019, 1mg/ml in DMEM/F12). Human iPSCs were fed every day with hESC media. Neural differentiation of human PSCs was carried out using our previously published dual SMAD inhibition-based protocol (Guo et al., 2023; Li et al., 2020; Shen et al., 2023). Briefly, hPSCs neural differentiation was induced 5 days after passaging of hPSCs onto MEFs by switching hESC medium to neural induction medium (NIM), consisting of DMEM/F12: Neurobasal ( Thermo Fisher Scientific, 21103049) 1:1, 1x N2 (Waisman Center), 1x L-glutamine, 1x Antibiotic-Antimycotic (Thermo Fisher Scientific,15240062), 10 μM SB432542 (Biogems, #3014193), 100nM LDN193189 (Selleck Chemicals, #S7507), and 5 μM XAV-939 (Selleck, #3748). Cells were cultured in NIM for 9 days with a daily medium change and small molecules added fresh, daily. Cells were then dissociated with TrypLE Express (Thermo Fisher Scientific, #12605010) and re-plated 1:1 onto Cultrex (R&D Systems) -coated plates in neural stem cell medium (Neurobasal medium, 1x L-glutamine, 1x N2, 0.5x B27 without vitamin A (Thermo Fisher Scientific), 1x Anti-Anti), supplemented with 10 μM ROCK inhibitor. Cells were maintained for 6 days with daily change of NSC medium, during which they transition from neuroepithelia to neural progenitors (NPCs) and reorganized into neural rosettes. At this stage NPCs were dissociated with Accutase (Thermo Fisher Scientific), frozen in NPC freeze medium (90% FBS, 10% DMSO), and stored in liquid N2. For experiments, NPCs were thawed onto cultrex-coated plates in NPC media containing 10μM ROCK inhibitor. For experiments using NPCs, NPC cells were maintained in Neuron progenitor (NPCs) medium [DMEM/F12 (Thermo Fisher Scientific, 11330032), 1× N2 (Waisman Center Stem Cell Core), 1× B27 without vitamin A (Thermo Fisher Scientific, 12587020), 1× Anti-Anti (Thermo Fisher Scientific, 15240062), 20 ng/ml FGF2 (Waisman Manufacturing), 1 μg/ml mouse laminin (Thermo Fisher Scientific, 23017015), with daily media change. NPCs were either maintained in NPC medium for 3 days, or infected with lentivirus expressing shRNA and mCherry or mScarlet in NPC medium, then plated for experiments after 72hrs onto either Cultrex-coated plastic plates or PEI/Cultrex coated coverslips or plates (as indicated below). For experiments using neurons, NPCs were plated at low density in NPC medium containing 10uM ROCK inhibitor and 0.2uM γ-Secretase Inhibitor XXI, Compound E (Calbiochem). The next day, the media was switched to neuronal differentiation medium (NDM; Neurobasal medium, 1X GlutMAX (Thermo Fisher Scientific), 1x Anti-anti, 1x N2, 0.5x B27 without vitamin A, 0.3% glucose, 20 ng/ml BDNF (Peprotech), 20 ng/ml GDNF (Peprotech), 500 ng/ml cAMP (Millipore-Sigma), 200uM ascorbic acid (Millipore-Sigma), supplemented with 0.2uM Compound E. Post-mitotic neurons were maintained in NDM, with half media changes performed every 3 days. Method details DNA constructs The lentiviral vectors lenti-EF-ELAVL1-IRES-GFP and lenti-EF-FXR1-IRES-GFP, designed to express human ELAVL1 and FXR1 in NPCs, were generated by separately inserting the ELAVL1-IRES-GFP and FXR1-IRES-GFP expression cassettes into the lentiviral backbone. The lenti-EF-GFP-IRES-GFP vector was used as a control. Lentiviral vector construction and production of lentivirus Lentiviral vector expressing shRNA against FXR1 and control shRNA was previously published (Shen et al., 2021). Lentiviral vector expressing shRNA against HMGB2 and control shRNA was cloned into a modified pCSCPW vector as described (Gao et al., 2025). Lentivirus production was performed as described previously (Guo et al., 2023; Li et al., 2020) with some modifications. Briefly, lentiviral DNA was co-transfected with packaging plasmids pMDL, REV and pCMV-VSVG into HEK293T cells using the PEI method. The viral transfer vector DNA and packaging plasmid DNA were transfected into 5x15 cm dishes of cultured HEK293T cells using PEI. The medium containing lentivirus was collected at 48- and 60-, 96-hours post-transfection, pooled, filtered through a 0.2-μm filter, and concentrated using an ultracentrifuge at 19-22k rpm for 2 hours at 4°C using a SW27 rotor (Beckman). The virus was washed once and then resuspended in 100 μl PBS. We routinely obtained 1x108 infectious viral particles /ml for lentivirus. Immunocytochemistry Immunocytochemical analysis of human cells was carried out as described (Guo et al., 2023; Shen et al., 2023). Briefly, NPCs and neurons were cultured on PEI/ Cultrex-coated 96-well 1.5p glass bottom plates (Cellvis, P96-1.5P) or glass coverslips and fixed with 4% PFA for 10 minutes at room temperature and then washed twice with PBS. Cells were blocked with PBS-T (PBS containing 5% goat serum and 0.2% Triton X-100) for 1h at room temperature. Primary antibodies were diluted in PBS-T 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 (1:500) in PBS-T for 60 minutes at room temperature. After 3 washes with PBS, neurons were counterstained with the nuclear counterstain, DAPI (4’, 6-diamidine-2’-phenylindole dihydrochloride, 1:2000, Roche Applied Science, Indianapolis, IN) and stored in PBS at 4°C (plates), or coverslips were mounted using DABCO-PVA mounting medium. Cell proliferation assays NPCs were plated at a density of 60,000 cells per well of a 96-well 1.5p glass bottom plate (Cellvis, P96-1.5P) coated with PEI and Cultrex (R&D Systems, 3433-005-01). After 24hrs, EdU pulsing was performed using the VF 488 Click-iT Universal Cell Proliferation Detection Kit. In brief, 10uM of EdU was added to the NPCs for ∼2hrs. After EdU was removed, the cells were washed 2x with PBS. The NPCs were fixed using 4%PFA for 10mins and then washed twice with washing solution (PBS: 3%BSA) then washed once with penetrant solution (PBS: 0.5% Triton X-100). The Click-it solution (Reaction Buffer: CuSO4: VF 488 Azide: Additive Solution) was added for 30mins and then washed with washing solution. After 10mins, immunocytochemistry was performed for Ki67 and SOX2 (following the protocol discussed above). Quantification analysis Quantitative analyses of cell proliferation and differentiation were carried out using unbiased stereology with StereoInvestigator software (StereoInvestigator, MBF Biosciences, Inc) as previously described (Gao et al., 2020; Shen et al., 2019). Briefly, imaging was performed using an Apotome microscope at 20x. At least 3 biological replicates (independent iPSC lines) were used for these experiments. For quantification of proliferation, differentiation and cell death using high-content imaging, a nano high-content imaging system (Molecular Devices) equipped with a 20x objective was used. Briefly, for each well, a tile consisting of 21 individual fields was acquired. Nuclei segmentation was performed using DAPI images, applying intensity thresholding with size and shape constraints. Marker-specific signals were quantified from their respective fluorescence channels following background subtraction and uniform thresholding to generate marker-positive masks. Marker readouts were normalized to DAPI-positive nuclei to account for variations in cell density. Images that did not meet focus or intensity criteria were excluded using consistent thresholds across all experimental conditions. Morphological analysis Dendritic morphological analysis was carried out using an Olympus BX51 microscope with 20X lens, a motorized stage and digital camera as described (Guo et al., 2015, 2023; Shen et al., 2019). The soma and dendrites of neurons were traced and analyzed using Neurolucida software (MicroBrightField). Approximately 40 neurons were analyzed for each group in each experiment. Neurons from at least 3 independent differentiations replicates were analyzed for each experimental condition. For morphological analysis using high content imaging, NPCs were plated at a density of 20,000 cells per well of a 96-well 1.5p glass bottom plate (Cellvis, P96-1.5P) coated with PEI and Cultrex (R&D Systems, 3433-005-01). The cells were replated in NSC media supplemented with 10μM ROCK inhibitor (Y-27632 dihydrochloride, Tocris, 1254) and 0.2uM Compound E (Sigma-Aldrich, 565790). The next day a full media change to NDM was made (Neurobasal medium, 1x GlutaMAX (Thermo Fisher Scientific), 1x Anti-anti, 1x N2, 0.5x B27 without vitamin A, 0.3% glucose, 20 ng/ml BDNF (Peprotech), 20 ng/ml GDNF (Peprotech), 500 ng/ml cAMP (Millipore-Sigma), 200uM ascorbic acid (Millipore-Sigma), supplemented with 0.2uM Compound E). Neurons were infected with LV-shNC or LV-shHMGB2 3 days after replating. Neurons were fixed 1 week after replating and stained for mCherry (1:1000) and MAP2 (1:500) markers as described above. Imaging was done using the Nano (Molecular Devices) high content imaging system with a 20x objective. For each well a tile of 25 individual fields was imaged. The neurite outgrowth parameters were analyzed using the Neurite Outgrowth module in MetaXpress (Molecular Devices) set to detect neurons positive for mCherry and the pipeline adjusted for brightness and size in each imaging session for batch neurite analysis. Experimental and control conditions were always imaged on the same day. Transcriptomics analysis Transcriptomics analysis was carried out as described (Gao et al., 2017; Shen et al., 2023). Briefly, NPCs and neurons (n = 4 independent differentiation batches derived from 2 iPSC lines per group) were collected using Trizol. RNA was isolated using Direct-zol RNA Microprep Plus (ZYMON RESEARCH #R2062) kit following the manufacturer’s instructions. RNA quality was evaluated using Agilent tape station. The sequencing library construction, library quality assessment, and sequencing were performed by Novogene Bioinformatics Institute (Sacramento, CA, USA). Bioinformatic analysis Raw RNA-seq FASTQ files were first subjected to quality control to assess sequencing quality metrics (Dawadi et al., 2025). Adapter contamination and low-quality bases were removed using Trimmomatic v0.39. Following trimming, all cleaned reads were independently aligned to the human reference genome (hg38) using STAR v2.7.11b. Gene-level read counts were subsequently generated using featureCounts v2.2.1. The resulting raw count matrices were used for downstream transcriptomic analyses. Differential gene expression (DEG) analysis was performed using DESeq2 v1.46.0. Pairwise comparisons were conducted between FXR1 knockout (FXR1-KO) and control samples at each differentiation stage, including NPCs, 1-week differentiated neurons (1-wk-diff), and 4-week differentiated neurons (4-wk-diff). Genes with a nominal p value < 0.05 and an absolute log2 fold change > 1 were considered significantly differentially expressed (Collins et al., 2023). DEG sets identified at each stage were retained for downstream enrichment and network analyses. Gene Set Enrichment Analysis (GSEA) was performed separately at each differentiation stage. Genes were ranked according to DESeq2-derived differential expression statistics (Subramanian et al., 2005). Enriched gene sets were considered significant based on an adjusted p value < 0.05. Gene ontology (GO) enrichment analysis was performed on DEG sets using Metascape (Zhou et al., 2019) with human gene annotations. Biological process (BP), molecular function (MF), and cellular component (CC) categories were analyzed. Enriched GO terms were selected based on a nominal p value < 0.05. GO enrichment results were visualized using custom plotting scripts implemented in R. Weighted gene co-expression network analysis (WGCNA) was performed using the union of DEGs identified across all three differentiation stages (Li et al., 2023). Transcript abundance in TPM was calculated from featureCounts outputs and used for network construction. Genes with low expression were filtered by requiring expression levels >1 TPM in at least 50% of samples. Expression values were log2-transformed [log2(TPM + 1)] prior to network analysis. Sample clustering was performed to identify potential outliers. A soft-thresholding power was selected based on the scale-free topology criterion, and an unsigned adjacency matrix was constructed. Gene co-expression modules were identified using hierarchical clustering with dynamic tree cutting, followed by module merging based on eigengene similarity. Modules were assigned unique identifiers (M1, M2, M3, etc.) according to module size. Genes assigned to the grey module were excluded from downstream correlation and visualization analyses. Module eigengenes, defined as the first principal component of each module, were calculated to summarize module-level expression patterns. Eigengene dynamics were examined across differentiation stages and genotypes. Heatmaps of gene expression within each module were generated to visualize stage- and genotype-dependent expression patterns. To identify key regulatory genes within co-expression modules, intramodular connectivity (kWithin) and module eigengene-based connectivity (Sjöstedt et al., 2020) were calculated using the WGCNA framework. Highly interconnected hub genes were defined based on high intramodular connectivity and strong correlation with the corresponding module eigengene. Connectivity metrics for all genes were compiled for downstream interpretation. Gene Set Variation Analysis (GSVA) (Zhang et al., 2023) was performed using curated gene sets representing neuronal stem cell population maintenance, stem cell proliferation, neuron fate commitment, forebrain neuron differentiation, neuron development, neuron maturation, neuron remodeling, and neuronal action potential. Gene sets were obtained in GMT format and filtered to retain sets containing between 10 and 5,000 genes. GSVA scores were calculated using a Gaussian kernel. Module–trait correlation analyses were conducted by correlating module eigengenes with GSVA-derived pathway activity scores and categorical traits representing differentiation stage and genotype using Pearson correlation. Corresponding p values were calculated using Student’s t distribution. Module–trait relationships were visualized using heatmaps. Western Blot and quantification Western blot analysis was performed as described (Shen et al., 2023). Briefly, cells were lysed in RIPA buffer (50 mM Tris, pH 8.0, 150 mM NaCl, 1% NP-40, 0.1% SDS) supplemented with protease inhibitors (Roche Applied Science). After centrifugation for 10 min at 4°C, supernatants were quantified by the Bradford Assay (Protein Assay Dye Reagent Concentrate, Bio-Rad). 15-20 μg of total proteins was resolved by SDS-PAGE, transferred to nitrocellulose membrane, blocked with 5% BSA, and probed with primary antibodies. Secondary antibodies conjugated with near infrared fluorescent dyes (IRDye 800CW or IRDye 680LT, LI-COR) were used at a dilution of 1:5,000 for visualizing protein bands with an Odyssey Imager (LI-COR). Quantification of band intensity was performed using Image Studio software (LI-COR). Four independent differentiation batches from two iPSC lines per group were used for analysis. Real-time quantitative PCR (qPCR) Cells were collected in TRIzol reagent (Thermo Fisher Scientific), then RNA was extracted and stored at -80°C for all RNA expression experiments. Real-time PCR assays were performed using standard methods as previously described (Gao et al., 2020; Shen et al., 2019). The first-strand cDNA was generated by reverse transcription with both oligo dT and random primers using PrimeScript™ RT Reagent Kit (Takara, #RR037A). To quantify the mRNA levels using real-time PCR, aliquots of first-strand cDNA were amplified with gene-specific primers and universal SYBR Green PCR super mix (Bio-Rad, #172-5124) using a Step-1 Real-Time PCR System (Applied Biosystems). The PCR reactions contained 20–40 ng of cDNA and 300nM of forward and reverse primers in a final reaction volume of 20μl. Oligonucleotide primers used are provided (Table S4). RNA immunoprecipitation followed by quantitative PCR (RNA-IP-qPCR) RIP-qPCR analysis was performed as described (Shen et al., 2023). Briefly, human NPCs and 1-week differentiated neurons were homogenized in 2 ml of homogenization buffer (25 mM Tris (pH 7.0), 25 mM Tris (pH 8.0), 100 mM KCl, 12 mM MgCl2, 10% IGEPAL, 1 mM DTT, protease inhibitors, 1× RNase inhibitor, 200 units/mL, 100μg/mL cycloheximide, 1 mg/ mL heparin). Nuclei and debris were pelleted at 14,000 g for 15 min. An aliquot of input was saved for RNA extraction (50μl). A polyclonal antibody against FXR1 or HuR (4ug) was incubated with supernatant at 4°C for 4 hours before adding Dynabeads (Life Technologies). The antibody–protein conjugation was rotated at 4°C overnight. After the third wash with the high-salt buffer (25 mM Tris (pH 7.0), 25 mM Tris (pH 8.0), 300 mM KCl, 12 mM MgCl2, 1 mM DTT, 100μg/mL cycloheximide), the immunoprecipitations were resuspended into Trizol (Invitrogen) for RNA isolation. RNA from immunoprecipitation and input was used, and all qPCR reactions were carried out in duplicate for each sample on each amplicon. mRNA stability assay RNA stability analysis was performed as described previously (Shen et al., 2023). Briefly, NPCs were treated with 10μg/ml of actinomycin D (Sigma-Aldrich) to inhibit gene transcription as described (Shen et al., 2019, Shen et al., 2023) and NPCs were collected at various time intervals for RNA isolation and qPCR analysis. HMGB2 mRNA levels were normalized to GAPDH. RNA decay kinetics and half-life were analyzed using a published method (Shen et al., 2019). Briefly, we used exponential function M =M0e-ƛt (M: amount of mRNA at t time, M0: amount of mRNA t t=). ƛ= ln(2)/T1/2 (T1/2 is the half-life of the mRNA). Isolation of mouse astrocytes We performed all procedures involving live mice in accordance with the NIH Guide for the Care and Use of Laboratory Animals (IACUC) and the protocols approved by the University of Wisconsin-Madison Animal Care and Use Committee. Mouse astrocytes were isolated from the telencephalon of newborn wild-type C57BL/6J mice using a published protocol (Barkho et al., 2006). Briefly, P0-P3 mouse forebrains were dissected and digested with trypsin for 5 min. Cells were then dissociated by harsh trituration to avoid neuronal growth and plated onto T25 flasks in DMEM supplemented with 10% FBS. The isolated cells were plated in T25 flasks and cultured in a 37°C, 5% CO2 incubator until cells reached confluence. The flasks were then shaken at 100 rotations/min for 2 days at 37°C to shake off proliferating cells and neurons. The media was changed. The cells were passaged a total of three times before the astrocytes were used for coculture experiments with human neurons. Microelectrode array (MEA) analysis MEA analysis was carried out as described (Guo et al., 2023; Shen et al., 2023). Briefly, NPCs were seeded onto PEI/Cultrex 96-well Cytoview MEA plates (Axion Biosystems) at a density of 50,000 cells per well using the procedures described above. One week after plating, 10,000 primary mouse astrocytes resuspended in NDM + 10% FBS were added per well. FBS was removed 48 hours later by replacing medium with standard NDM. MEA recordings occurred starting one week after the addition of astrocytes and were performed weekly, on the same day and always at a minimum of 24 hours after medium changes. Half NDM changes were performed every 3 days. Extracellular recordings were performed in a Maestro MEA system preset at 37°C and 5% CO2 equipped with AxIS Navigator software (Axion Biosystems) using a bandwidth with a filter for 200 Hz to 3 kHz cutoff frequencies. Each plate was acclimatized for 5 min in the Maestro Pro Instrument then recorded for 15 min for each measurement. Spike detection was performed using an adaptive threshold set to 6 times the standard deviation of the estimated noise on each electrode. MEA analysis was performed using the Axion Biosystems Neural Metrics Tool and Metric Plotting Tool. An electrode was considered active at a threshold of 2 spikes/min. Sample size for each treatment was indicated in figure legends. Single-electrode and network bursts were determined using the default parameters in AxIS Navigator. Mean firing rate was estimated with a 10s detection window. Calcium imaging and analysis NPCs were plated in 96-well 1.5p glass bottom plates (Cellvis, P96-1.5P) coated with PEI and Cultrex (R&D Systems, #3433-005-01) in NSC medium supplemented with 0.2uM Compound E, then 24hrs post plating, medium was changed to NDM supplemented with 0.2uM Compound E. One week after replating into 96-well plates, 104 mouse astrocytes per well were added onto cultures in NDM supplemented with 10% FBS. After 48hrs of co-culture, 50% of the medium in each well was exchanged to normal NDM, followed by twice a week media change. Two weeks prior to recording (3 weeks of differentiation) LV-Syn1-GCaMP7f was added to neurons and 24hrs before recording, media was changed to BrainPhys imaging optimized medium (STEMCELL Technologies, 05796) 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). GCaMP7f fluorescence was recorded on a Leica Stellaris 8 inverted epi-fluorescence microscope with the 488nm filter under environmental control (37°C, 5% CO2) using Leica Application Suite X (LAS X) at the Waisman Center CMN imaging core at UW-Madison. Neuronal cultures were imaged for 5mins at a frame rate of 1.34hz (∼400 frames per time lapse) using a 20x objective. Data analysis was performed using a custom pipeline in Python (v3.8) built with the CaImAn library (Giovannucci et al., 2019). Before uploading the files to CaImAn, the videos were exported as TIFF files from LASX software and bleach corrected using an exponential fit and reduced to 512 x 512 using FIJI (ImageJ) software. Segmentation of neuronal regions of interest (ROIs) was performed using Constrained Non-negative Matrix Factorization (Pnevmatikakis et al., 2016) (CNMF). The model was initialized with a target of K = 200 components and an expected neuron size (gSig = 4,4). To ensure only neuronal somas were detected, we applied a filtration step. ROIs were filtered based on signal quality (min SNR >5.0, min PNR >4) and evaluated by a Convolutional Neural Network (CNN) classifier (CaImAn default). Additional filtering of ROIs was done by application of a morphological filter. Deconvolved neural activity (S) was estimated from the ΔF/F traces using the OASIS algorithm (Friedrich et al., 2017). We calculated mean ΔF/F by taking the time average of the ΔF/F trace for each neuron. A burst was defined as a sequence of at least three events within two imaging frames, and the number of bursts was calculated per neuron. Quantification and statistical analysis Power analysis was used to pre-determine sample sizes, and our sample sizes are similar to those reported in previous publications (see citations within each procedure). Data distribution was assumed to be normal, but this was not formally tested. Statistical analysis was performed using ANOVA and Student's t test, unless specified, with the aid of GraphPad Prism 9. Unpaired or paired two-tailed Student’s t test were used for two-group comparisons, or two-tailed Welch’s t-test were used for normalized data to account for unequal variances. Two-way ANOVA with Bonferroni post hoc test was used for analyzing multiple groups. Outliers were identified using GraphPad Prism 9 software with the ROUT method (Q = 1%). Probabilities of p < 0.05 were considered significant. For sample size, N indicates the number of biological replicates (e.g. number of cell line), while n indicates sample size including experimental replicates (e.g. batches of differentiation) used for statistical analyses (see Table S1). Schematics drawings were created using Power Point and Biorender. Published: September 17, 2026 Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.stemcr.2026.103082. Supplemental information Document S1. Figures S1–S6 and Tables S1 and S4 Table S2. Stage-specific DEGs identified by bulk population RNA-seq Table S3. WGCNA and modules Document S2. Article plus supplemental information References Associated Data Supplementary Materials Document S1. Figures S1–S6 and Tables S1 and S4 Table S2. Stage-specific DEGs identified by bulk population RNA-seq Table S3. WGCNA and modules Document S2. Article plus supplemental information Data Availability Statement RNA sequencing (RNA-seq) data have been deposited to Gene Expression Omnibus (GSE318782). This paper does not report original codes.