Deficiency of the Heterogeneous Nuclear Ribonucleoprotein U locus leads to delayed hindbrain neurogenesis

Enormous progress in genomic technologies has led to the discovery of hundreds of genes associated with various neurodevelopmental disorders (NDDs). These include several genes belonging to the heterogeneous nuclear ribonucleoprotein (hnRNP) family (Gillentine et al., 2021). One of the genes within this family, HNRNPU, which encodes for Heterogeneous Nuclear Ribonucleoprotein U [also known as Scaffold Attachment Factor A (SAF-A)] (Kiledjian and Dreyfuss, 1992), has emerged as a frequently affected gene leading to NDDs such as intellectual disability (ID), autism spectrum disorder (ASD) as well as neurological conditions such as epilepsies (Depienne et al., 2017; Satterstrom et al., 2020; Taylor et al., 2022). The gene was first indicated in NDDs as part of the 1q44 microdeletion syndrome characterized by a severe global developmental delay, ID, seizures, muscular hypotonia, hearth, and congenital malformations such as agenesis of the corpus callosum, heart and skeletal anomalies (Depienne et al., 2017). Later, several smaller deletions and point mutations affecting the HNRNPU locus pinpointed it as the causal gene within the locus for the majority of the brain-related phenotypes (Bramswig et al., 2017; Leduc et al., 2017; Shimojima et al., 2012; Tung et al., 2021; Yates et al., 2017). In addition to the HNRNPU gene, a long non-coding RNA HNRNPU-AS1 maps to the locus. The function of HNRNPU-AS1 is not known, although some reports have indicated its role in different molecular pathways, such as cell proliferation and apoptosis in cancer cells (Niu et al., 2021; Zhang et al., 2022). To date, several reports exist about the pathogenic genetic variants affecting the HNRNPU gene in individuals with HNRNPU-related disorder, and are mostly de novo loss-of-function variants at sequence or copy number level (Brunet et al., 2021; Durkin et al., 2020; Gillentine et al., 2021; Wang et al., 2020). All the reported cases are heterozygous variants, suggesting that homozygous gene-disrupting variants affecting HNRNPU are embryonic lethal in humans, similar to what has been demonstrated in mice (Dugger et al., 2020; Roshon and Ruley, 2005; Ye et al., 2015).

HNRNPU has a key role in three-dimensional (3D) genome organization and regulating RNA processing (Fan et al., 2018; Marenda et al., 2022; Nozawa et al., 2017; Xiong et al., 2020; Ye et al., 2015). For instance, HNRNPU modulates chromatin compaction by promoting chromatin accessibility in a dynamic yet structured fashion, dependent on its oligomerization status (Nozawa et al., 2017). Recent studies have also shown the involvement of HNRNPU in mitosis and cell division by changing its interactions with condensed chromatin and influencing DNA replication and sister chromatid separation (Connolly et al., 2022; Ma et al., 2011; Sharp et al., 2020). Several studies have reported a role for HNRNPU in splicing, promoting both exon inclusion and exclusion (Huelga et al., 2012; Jones et al., 2022; Ye et al., 2015). Mechanistically, it has been shown that HNRNPU can stabilize the pre-mRNA structure, thus inhibiting the splicing of certain exons (Jones et al., 2022). A critical role for Hnrnpu-mediated splicing has also been demonstrated during pre- and postnatal heart development in mice, showing that loss of Hnrnpu leads to increased intron retention events, causing abnormality in heart development and function (Ye et al., 2015).

Recently, few studies shed light on the effects of HNRNPU mutations in cortical development. A mouse model of Hnrnpu haploinsufficiency presented abnormal brain organization and showed at postnatal day 0 altered transcriptome with downregulation of pathways related to neuronal projection and migration and upregulation of genes relevant to cell growth and protein localization in hippocampal and neocortical cells (Dugger et al., 2020). In contrast, a study performed on embryonic mice upon complete conditional Hnrnpu knockout in the cerebral cortex showed upregulation of genes involved in synaptic activity and downregulation of DNA-related ontologies together with changes in alternative splicing regulation (Sapir et al., 2022). Similarly, an isogenic human cortex organoid model with two clonal cell lines carrying different heterozygous mutations in HNRNPU partially resembled what was observed in the embryonic mice but not the postnatal model (Ressler et al., 2023). Therefore, the overall outcome of HNRNPU genetic variants might depend on the stage of development, brain region studied, genetic background, and gene dosage.

So far, the studies have mainly focused on cortex development and forebrain structures. HNRNPU is expressed in different tissues, and postnatally it is highest in cerebellum (Thierry et al., 2012). Accordingly, atrophy of the cerebellum was highlighted in a cohort of patients affected by HNRNPU-related disorder (Durkin et al., 2020). The cerebellum is one of the most studied hindbrain structures, and has emerged as important for typical and atypical development, and abnormalities in cerebellar development have been associated with ASD and ID (Burstein and Geva, 2021; Frosch et al., 2022; Spahiu et al., 2022). Recent observations have highlighted the influence of hindbrain development on the brain cortex and its pivotal role for cognitive functions (Joseph, 2000; Kohlmeier and Polli, 2020) and suggested that analysis of midbrain and hindbrain are needed to expand the knowledge about NDDs (London et al., 2022).

Here, we focused on delineating the molecular and cellular consequences of the HNRNPU locus deficiency in a model of human early hindbrain development using induced pluripotent stem cells (iPSCs) from an individual with HNRNPU-related disorder and a knockdown isogenic cell approach for comparative analyses, and providing evidence of a broad spectrum of affected pathways.

To assess the molecular effects of HNRNPU locus (including HNRNPU and HNRNPU-AS1) haploinsufficiency during human hindbrain differentiation, we generated two different HNRNPU locus deficient cellular models derived from human iPSCs. The iPSCs were induced into neuroepithelial stem (NES) cells with hindbrain profile and further differentiated for 5 (D5) and 28 (D28) days using an undirected protocol as previously described (Becker et al., 2020; Falk et al., 2012) (Fig. 1A-B). This approach generates a cell culture consisting of a mixed population of cells expressing excitatory, inhibitory, and progenitor neural cell markers representing a physiological neuronal environment (Falk et al., 2012).

We first generated an HNRNPU-related disorder patient cell model, hereafter called HNRNPU_del/+. Through genetic screening of a twin cohort focusing on NDDs, we identified a male twin pair carrying a 44 kilobase (kb) heterozygous deletion spanning from COX20 to HNRNPU genes (Stamouli et al., 2018). The twins were diagnosed with ASD, ID, and fever-induced seizures. A detailed phenotypic description of the twin pair is presented in Table S1. As COX20 deficiency is an autosomal recessive disorder, we estimated that the effect of heterozygous deletion of COX20 would not have a major role in the cellular phenotype (Ban et al., 2022). We obtained fibroblasts from skin biopsies of the twins and while the reprogramming of the fibroblasts from twin-1 was not effective, we successfully reprogrammed twin-2 fibroblasts into iPSCs. The iPSCs had a normal karyotype, pluripotent marker expression and showed significant reduction of HNRNPU and HNRNPU-AS1 compared with control iPSCs (Fig. 1C; Fig. S1A, S1B). In addition, we used dual-SMAD inhibition to derive NES cells from the iPSCs as previously described (Falk et al., 2012), followed by staining for neural stem cell markers Nestin and SRY (sex determining region Y)-box 2 (SOX2) to verify their identity (Fig. S1C). To confirm the downregulation of HNRNPU-related RNA and protein product, we measured HNRNPU-AS1 RNA, and HNRNPU RNA and protein in HNRNPU_del/+ cells using cells obtained from a neurotypical male control (CTRL) (Uhlin et al., 2017a) as a reference. HNRNPU_del/+ showed significantly lower RNA expression for both HNRNPU-AS1 and HNRNPU (P<1.0×10^–5, ANOVA and post hoc Tukey test) and an average lower expression of HNRNPU protein spanning from ∼20% to ∼60% downregulation (Fig. 1D-F).

As a complementary approach, we generated an isogenic cell model, hereafter called siHNRNPU, in which we reduced HNRNPU-AS1 RNA, and HNRNPU RNA and protein expression using a pool of small interfering RNA (siRNA) oligos in CTRL cells in parallel with non-target oligo pool (siNTC), similar to approaches successfully used in previous studies of HNRNPU in other cell types (Ma et al., 2011; Nozawa et al., 2017; Zietzer et al., 2020). To achieve consistent knockdown of HNRNPU throughout differentiation, we performed repetitive siRNA treatments every 6 days. Significant downregulation of HNRNPU-AS1 and HNRNPU RNA and a 32% reduction of HNRNPU protein expression were observed at NES stage. Both transcripts were similarly significantly downregulated at D5, but no significant difference was observed in HNRNPU RNA expression after 28 days in differentiation either by RNA sequencing (RNA-seq) or real-time PCR (Fig. S1D). Nonetheless, HNRNPU protein expression was reduced to 29% at D28 compared with siNTC samples (Fig. S1G), therefore, we considered the silencing successful and proceeded with further analyses. The HNRNPU-AS1 was consistently downregulated after differentiation (Fig. S1D).

Our results of the variable RNA and protein expressions after downregulation of HNRNPU by siRNA treatment or mutation are consistent with all the previous studies about HNRNPU mutations in brain tissues and differentiated neuronal populations (Dugger et al., 2020; Ressler et al., 2023; Sapir et al., 2022). Even after knockdown or knockout at gene level, these studies showed that HNRNPU expression levels are similar to the wild-type controls suggesting possible compensatory mechanisms. A possible mechanism might reside in the capacity of proteins belonging to the HNRNP family to directly associate with their own transcripts, thus stabilizing them (Huelga et al., 2012). For instance, single-cell RNA transcriptome comparison of human cortical organoids carrying two different frameshift mutations in HNRNPU revealed no difference in HNRNPU expression between the HNRNPU-mutant and relative control, in any of the identified cell types. At the same time, the downregulation of HNRNPU at the RNA level does not always translate into a similar reduction at the protein level (Dugger et al., 2020; Ressler et al., 2023; Sapir et al., 2022).

Next, we sought to analyze the molecular consequences of HNRNPU haploinsufficiency during human hindbrain development. We extracted total RNA for transcriptomic analyses from NES cells collected at three different time points of differentiation (D0, D5, and D28). To confirm that our model resembles human hindbrain development, we analyzed the expression of several hindbrain and cerebellar markers in our cell line at NES and D28. We first analyzed the expression of HOXA2 and OTX1/OTX2 in neural stem cell/progenitor phases in our cell model as the balance of these markers is fundamental for the specification of early hindbrain development (Lowenstein et al., 2022). At D0 and D5, the cells expressed HOXA2 and not OTX1/OTX2, in line with the developing hindbrain phenotype (Fig. S1H). Accordingly, we show that CTRL cells at D28 express many cerebellar markers (UNC5C, ICMT, CA8, TRPC3, ASTN1, KITLG) (Aldinger et al., 2021; Consalez et al., 2021; Kim et al., 2003; Kim and Ackerman, 2011; Mancarci et al., 2017; Wu et al., 2019) (Fig. S1I). In parallel, we analyzed cerebellar marker expression in our previously published single-cell RNA-seq (scRNAseq) data from a similarly derived cell line at D28 (Becker et al., 2020) and showed that the cerebellar markers are indeed expressed across the cell types (Fig. S1J). Furthermore, to evaluate the specificity of our model, we analyzed the expression of the markers in a previously published dataset from human cortical organoids (Ressler et al., 2023) and observed extremely low or null expression of the markers in all the cell types and samples (Fig. S1K).

When analyzing the expression of the HNRNPU and HNRNPU-AS1 in the CTRL cell line, we observed that HNRNPU-AS1 expression decreases from D5 to D28 (comparison D0-D5: P=0.76; D5-D28: P<0.005, ANOVA and post hoc Tukey test), and HNRNPU expression decreased steadily during differentiation (comparison D0-D5: P=0.019; D5-D28: P =<1.0×10^–5, ANOVA and post hoc Tukey test) (Fig. 1D-E). HNRNPU expression follows a similar decreasing trend during differentiation from iPSCs to neurons from a previously published dataset (Burke et al., 2020) and during development of all the cerebral areas, although in the cerebellum the postnatal expression is the highest compared to the other brain regions (Fig. 1G; Fig. S1L). Since HNRNPU expression was higher in proliferating progenitor cells (Fig. 1D; Ressler et al., 2023; Sapir et al., 2022), we further inspected its expression in our above mentioned scRNA-seq data (Fig. S2A). As expected, HNRNPU expression was highest in the neural progenitor population (Fig. S2B-C). Moreover, the neural progenitor population was further divided into three distinct subclusters, of which one was highly enriched in proliferating markers such as TOP2A, KIFC1, and KIF18B. This cluster also had the highest expression of HNRNPU (Fig. S2D).

In contrast to CTRL cells the HNRNPU_del/+ cells did not show a change in the expression of HNRNPU-AS1 during differentiation (comparison D0-D5: P=0.92; D5-D28: P =0.84, ANOVA and post hoc Tukey test) (Fig. 1E). Similarly, HNRNPU expression was stable from D0 to D5 (P=0.99, ANOVA and post hoc Tukey test), but followed a significantly delayed decrease at D28 (D5-D28, P=2.6×10^–5, ANOVA and post hoc Tukey test) (Fig. 1D). In the comparison of RNA expression between CTRL and HNRNPU_del/+ at each time point, HNRNPU-AS1 was significantly reduced in HNRNPU_del/+ at each time point (D0 and D5: P<1.0×10^–5; D28: P =0.005, ANOVA and post hoc Tukey test) (Fig. 1E). Instead, at D0 and D5, when the cells express progenitor markers nestin and SOX2 (Fig. S3A), HNRNPU expression was significantly lower in HNRNPU_del/+ cells compared with CTRL cells (P<1.0×10^–5 and 2.3 x10^–5, respectively, ANOVA and post hoc Tukey test), while the expression was similar between the two cell lines at D28 (P=0.58, ANOVA and post hoc Tukey test) (Fig. 1D). HNRNPU protein was predominantly localized in the nucleus at all three analyzed stages of differentiation, and the measured protein levels were extremely variable between different time points in both cell lines and did not mirror mRNA levels (Fig. S3B-C). These results show that physiologically HNRNPU has the highest expression in the neuroepithelial stem cell stage and steadily reduces during the differentiation, demonstrating its important role in early neural differentiation.

Next, we performed differential gene expression analyses using DESeq2 (Love et al., 2014), followed by gene set enrichment analysis (GSEA) of the obtained transcriptomic data. Similar to earlier reported results in other cell types (Nozawa et al., 2017), we found that HNRNPU downregulation had a limited effect on the transcriptional landscape at D0 and D5. The isogenic siHNRNPU cells (replicates n=5) had only 10 differentially expressed genes (DEG) at D0 and 30 DEG at D5 (base mean>20, |log2FoldChange|>0.58, P adjusted<0.05, Wald test and Benjamini-Hochberg procedure) (Table S2A-B). As expected, due to the different genetic backgrounds, when comparing HNRNPU_del/+ cells (replicates n=5) to the CTRL cell line (replicates n=5) at each time point, we identified a higher number of DEG (2091 DEG at D0 of which 1033 upregulated and 1058 downregulated; 2091 DEG at D5, of which 1251 upregulated and 840 downregulated genes) (Table S2C-D). At D28 of the neural differentiation, both HNRNPU-deficient models revealed wider transcriptional rewiring, with 1511 DEG in siHNRNPU and 1608 DEG in HNRNPU_del/+ cells (Table S2E-F). Of these, only 148 DEG genes were shared between the two models at D28, suggesting that downregulation of HNRNPU might affect the expression of upstream transcriptional regulators that increase the transcriptomic landscape variability.

Next, we analyzed gene set and pathway level changes across the two datasets (Table S3). At D0, no specific Gene Ontology (GO) terms or enriched pathways were significantly shared between the two models (Fig. 2A; Fig. S4A, Table S3A-D). At D5, genes affecting the positive regulation of excitatory postsynaptic potential were downregulated in both models. In contrast, among upregulated genes, shared GO terms and enriched pathways included categories referring to the regulation of cell differentiation, growth factor receptors, and constituents of extracellular matrix (Fig. 2A; Fig. S4B, Table S3E-H).

The major transcriptional changes at D28 were clustering in multiple biological processes affected in both HNRNPU_del/+ and siHNRNPU cells. We found 74 significantly enriched GO terms shared between the two models, including several synaptic and transmembrane channel ontologies among the downregulated pathways (Table S3I-L). Interestingly, the shared upregulated pathways from GSEAs included pathways related to cell DNA organization during DNA replication and cell division and more general nucleosome and chromatin remodeling pathways (Fig. 2A and B), confirming the role of HNRNPU in the regulation of chromatin organization and DNA replication even in neural cells. Moreover, we found that two key developmental pathways, epithelium tube and embryonic hindlimb morphogenesis, were significantly upregulated in both model systems. Additional dysregulated developmental pathways were also uniquely affected in HNRNPU_del/+ (Fig. S4C). We also investigated whether the DEG genes at D28 were enriched in genes previously associated with epilepsies, ID, ASD, and general developmental disorders (DD). However, we found no significant enrichment for any gene lists (Hypergeometric test, Table S4). These results suggest that HNRNPU deficiency modulates the transcriptomic variability by altering the expression of genes enriched in neural maturation and chromatin organization and mostly affecting cells to be committed to neuronal differentiation more than cells in the neural progenitor phase.

Since HNRNPU has previously been shown to affect the transcriptome by regulating alternative splicing (Sapir et al., 2022; Xiao et al., 2012; Ye et al., 2015), we analyzed differential exon usage (DEU), which indicates alternative splicing events or differential isoform usage. Similar to the gene level changes, we detected fewer DEU events (ExonBaseMean >10, |Log2FoldChange| >0.58 and P adjusted<0.05) in the earlier timepoints and more at D28 in both model systems (Fig. S4D and Table S5). When comparing siHNRNPU to siNTC samples, we detected 0 and 2 DEU genes at D0 and D5, respectively, whereas a comparison between HNRNPU_del/+ and CTRL yielded 35 and 38 DEU genes at D0 and D5, respectively. At D28, 976 and 1285 DEU genes were detected for siHNRNPU and HNRNPU_del/+ when compared against their respective controls, respectively. Notably, only 5.9% and 9.8% of DEU genes were also differentially expressed in siHNRNPU and HNRNPU_del/+ cells, respectively. Both models revealed more exclusion than inclusion of exons due to HNRNPU deficiency, as 71% of the DEU events were due to downregulation (Fig. S4D). This result is in line with previous findings using other cell types (Huelga et al., 2012; Sapir et al., 2022; Ye et al., 2015). We performed HNRNPU binding motif analysis on differentially used exon (DUE) and flanking intron sequences to evaluate if splicing differences are due to direct binding of HNRNPU. The UGUAUUG binding motif was found on 479 out of 1766 DUEs (padj=1) and 354 out of 1172 DUEs (padj=0.0027) in HNRNPU_del/+ and siHNRNPU, respectively. In comparison, the motif was present in six of 2946 (padj=0.929) and seven of 2013 (padj=1) flanking introns in HNRNPU_del/+ and siHNRNPU, respectively. Our results suggests that the DEU observed in HNRNPU deficiency state can be partially caused by the direct binding of HNRNPU. Over representation analysis (ORA) of the DEU genes shared by both models revealed enriched pathways involved in cell morphogenesis, neuron projection development, and cilium organization (Fig. 2C; Table S6). We also analyzed whether DEU genes were enriched for the genes implicated in the different disorders as earlier and found a strong enrichment for ID gene list [hypergeometric test followed with false discovery rate (FDR) correction: 1.49×10^–14 and 5.26×10^–10 for siHNRNPU and HNRNPU_del/+, respectively], ASD (8.26×10^–8 and 1.96×10^–7), DD (4.68×10^–8 and 2.36×10^–7), and epilepsy (2.28×10^–6 and 0.012) (Fig. 2D; Table S4). Our results are in line with earlier evidence and showcase the role of HNRNPU in regulating exon usage during neural development, of genes previously associated with several NDDs.

Transcriptional changes at the pathway level strongly indicated differences in the cell proliferation rate of HNRNPU_del/+ and siHNRNPU cells at D28; therefore, we focused on analyzing the neural progenitor pool at D28, a timepoint in which generally most of the cells are postmitotic and committed for neuronal differentiation. First, we analyzed the cell type proportions using deconvolution of the transcriptomic data similar to previously described (Becker et al., 2020). The deconvolution predicted a higher proportion of neural progenitors in HNRNPU_del/+ compared to CTRL (Fig. 3A; Fig. S5A). To validate the presence of neural progenitors across the differentiation and investigate the difference between the two models, we analyzed SOX2 positive nuclei at the three time points for both HNRNPU_del/+ and CTRL cell lines. As expected, each cell line showed a decreasing number of SOX2-positive cells during the differentiation time course. However, HNRNPU_del/+ cells displayed a significantly higher number of SOX2-positive cells at D5 and D28 compared to CTRL (P<10^–4 and 0.002, respectively, χ² test) (Fig. 3B; Fig. S5B). Also, the siHNRNPU model had an increased number of SOX2-positive cells at D28, but the difference was insignificant (P=0.16, χ² test) compared with siNTC cells (Fig. 3B). To evaluate the characteristics of the progenitor cells at D28, we measured cell proliferation by Bromo-deoxy-uridine (BrdU) incorporation in both conditions. As expected, the cell proliferation rate decreased throughout the differentiation and was significantly higher in both siHNRNPU and HNRNPU_del/+ cells at D28 compared to the relative controls at the same time point (P=0.002 and 0.01, respectively, two-sided t-test). The rate was similar between HNRNPU_del/+ and CTRL cell lines at D0 but diverged already at D5 (P=6×10^–3, two-sided t-test) (Fig. 3C; Fig. S5C). In summary, downregulation of the HNRNPU locus leads to an enrichment of proliferating neural progenitor cells in the mixed population at D28 compared to controls. This suggests either a delay in the maturation trajectory during differentiation or a reduced proportion of cells committing to differentiation.

Since the dysregulated pathways in both HNRNPU-deficient conditions were related to membrane channels, synaptic formation, and extracellular matrix, we hypothesized that HNRNPU deficiency could affect the stage of neuronal maturation. To validate this hypothesis and the transcriptomic results, we first analyzed the proportion of cells with primary cilia during neural differentiation, as primary cilia guide axon tract development (Guo et al., 2019). Additionally, our results for DEU genes showed the enrichment of cilium organization in siHNRNPU and HNRNPUdel/+ cells at D28. Therefore, we analyzed ciliary proteins ARL13 and PCNT expression by immunofluorescence in CTRL and HNRNPU_del/+ cells throughout differentiation and observed a higher number of ciliated cells at D5 compared to D0 in both cell lines and no difference between HNRNPU deficiency and control cells at D0 and D5 (P=0.37 and P=0.64, respectively, χ²-test). However, a significant difference in the proportion of ciliated cells was observed at D28 (P=1.497×10⁻⁶, χ²-test), wherein HNRNPU_del/+ cells had a higher percentage of ciliated cells (Fig. 3D). Since the alteration of cilium organization pathways was similarly significant in both experimental conditions, we considered sufficient to validate the finding by immunofluorescence only in the HNRNPU_del/+ condition.

Additionally, we analyzed the expression of the presynaptic marker Synapsin 1/2 and the postsynaptic marker Homer1. The HNRNPU_del/+ cells had significantly smaller presynaptic particle sizes than the CTRL cells (P=0.009, t-test, two-sided) (Fig. 3E). In contrast, the number of presynaptic particles and the number and size of postsynaptic signals were comparable between HNRNPU_del/+ and CTRL. To monitor the neuronal maturation, we performed patch clamp electrophysiology to study the intrinsic membrane properties and synaptic activity of HNRNPU_del/+ and CTRL cells at D8 (n=4), D28 (n=12 cells for CTRL, and 17 cells for HNRNPU_del/+), and D50 (n=7 cells for CTRL and 10 cells for HNRNPU_del/+) (Fig. 3F). However, despite the clear neuronal morphology (Fig. S5D), the cells were characterized by high membrane resistance in response to voltage-step commands, and no spontaneous excitatory or inhibitory synaptic currents could be measured in any of the recorded cells, therefore at this stage of differentiation neither CTRL nor HNRNPU_del/+ cells can be considered mature neurons.

The results of our transcriptional profiling revealed a role of HNRNPU in chromatin organization, in line with previous reports that demonstrated the function of HNRNPU in chromatin compaction, DNA synthesis, and chromosome folding during mitosis in different cell types and conditions (Connolly et al., 2022; Nozawa et al., 2017; Sharp et al., 2020). Therefore, we sought to investigate whether HNRNPU deficiency affects chromatin organization in our human neural cell model. First, we performed chromatin accessibility analyses by treating CTRL and HNRNPU_del/+ cells with DNaseI; however, no large-scale differences were visible (Fig. S6A). We then performed single-cell immunofluorescent analysis of cell nuclei stained for heterochromatin marker, histone 3 tri-methylation at lysine 9 (H3K9me3). Both HNRNPU-deficient cell models displayed profound changes in nuclear architecture at D28 (Fig. 4A). Specifically, upon HNRNPU deficiency, a subpopulation of cells displayed differential number and total volume of heterochromatic foci (NbF, VFTotal, respectively), intensity and volume of the relative heterochromatic fraction (Intensity RHF, Volume RHF), and nuclear shape characteristics like surface area, volume, and radius of a sphere of equivalent volume (ESR) (Fig. 4A).

Next, we performed Hi-C to investigate the changes in chromatin organization at a higher resolution (Lieberman-Aiden et al., 2009). Since the NucleusJ analyses showed similar alterations of the heterochromatic fraction and nuclear shape in the two experimental conditions, we performed HiC on CTRL and HNRNPU_del/+ cells at D0 and D28 (Fig. S6B-C; Materials and Methods) as a model for HNRNPU haploinsufficiency. The number of chromatin interactions throughout the genome was similar at D0 between the two cell lines. However, at D28, a diverging pattern appeared in the chromatin organization maps, with the largest differences observed at the level of A/B compartments (10⁵–10⁶ base pairs) (Fig. S6D). Furthermore, while the ratio of short and long-range contacts per chromosome was similar at D0, at D28 CTRL cells showed a higher ratio for the short- and long-range contacts than HNRNPU_del/+ cells (Fig. S6E). An analysis of differential chromosome contact frequencies between the HNRNPU_del/+ and CTRL cells revealed little differences at D0 and higher contact differences at D28, as shown by the log2 ratio of the contacts at the chromosome level (Fig. 4B).

Furthermore, a compartment interaction analysis revealed similar interaction strengths between the samples at D0 (compartmentalization strength: 1.30 and 1.28 for CTRL and HNRNPU_del/+ cells, respectively). In contrast, we observed stronger compartmentalization in HNRNPU_del/+ at D28 (1.31 versus 1.59 for CTRL and HNRNPU_del/+ cells, respectively), with higher interactions between active compartments (A compartment) and fewer trans-interactions compared to the CTRL cells (Fig. 4C). These results align with the Hi-C experiments previously conducted on Hnrnpu-deficient mouse hepatocytes, showing decreased A-B interactions and increased A-A and B-B interactions upon Hnrnpu downregulation (Fan et al., 2018).

Next, we analyzed the compartment composition of each sample to identify genomic regions that switch compartments between CTRL and HNRNPU_del/+ (Materials and Methods). At D0, 3% of the compartments switched from inactive (B) to active (A) compartment and 5.6% from A to B comparing CTRL and HNRNPU_del/+, while at D28, 4.2% switched from B to A and 8.6% from A to B (Fig. 4D). To investigate whether these compartment switches affect disorder-related genes, we assessed the enrichment for the genes mapping in genomic regions that switch compartments. Genes mapped in regions switching from A to B compartment were enriched in all the gene lists (ASD FDR: 0.0043; ID: 0.0018; epilepsy: 0.0035, DD: 0.0136; hypergeometric test), whereas genes mapping in regions switching from B to A compartment were enriched only in the ASD gene list (FDR=0.0136) (Fig. 4E and Table S4). In contrast, we found no significantly enriched GO terms.

Lastly, we investigated whether the compartment changes were related to the transcriptional changes described above. Therefore, we mapped the DEGs at D28 (|log2FC|> 0.58, base mean>20) with gene mapping in the switching compartments. The concordant genes were defined as genes mapping to a region with a compartment switching from A to B and downregulated in the transcriptome analysis, or genes mapping in the B to A compartment switch and upregulated. We identified 144 concordant genes at D0 and 241 concordant genes at D28 (Table S7). Only concordant genes mapping in the compartment switching from A to B at D28 were significantly enriched for ASD (FDR: 0.014) and epilepsy (FDR: 0.007) gene lists (Fig. 4E; Table S4). By further analyzing the concordant genes to identify the enriched pathways, we showed that no ontologies were enriched in the concordant B to A genes, while the most significant enriched GO term in the concordant A to B genes was ‘GABA-gated chloride ion channel activity’ (FDR<0.05) (Fig. S6F), driven by genes such as CACNB2, GABRA2, GABRA4, GABRB1, GABRG1, and SCN1A. Interestingly, the GABR family genes downregulated in HNRNPU_del/+ at D28 belong to a cluster on chromosome 4 with an approximately seven megabase (Mb) region that maps in the A compartment in CTRL and B compartment in HNRNPU_del/+ at D28 (Fig. 4F).

Overall, the Hi-C analysis showed that, similar to the transcriptome analyses, the compartment organization is more affected in HNRNPU_del/+ at D28 than at D0, and HNRNPU deficiency led to an enrichment of B compartments at both timepoints. Interestingly, genes mapping in the enriched B compartments are associated with ASD and epilepsy, suggesting that the chromatin remodeling dependent on HNRNPU expression might ultimately be the driver of the observed phenotypes.

Heterozygous genetic variants in the HNRNPU locus lead to various disorders with predominant brain phenotypes. Recently, few studies focused on the effects of mutations in HNRNPU gene in mouse and human cortical organoids. Here, we model microdeletions of the HNRNPU locus responsible of HNRNPU-related disorder and provide novel evidence of the molecular and cellular consequences of HNRNPU deficiency in differentiating human neuronal cells with hindbrain phenotype, adding knowledge on the effect HNRNPU deficiency on the brain region where HNRNPU expression is the highest.

We demonstrate that adequate levels of transcripts from the HNRNPU locus are needed in the early developmental transition from neural progenitors to developing neurons for proper neurogenesis. Our results, consistent with the earlier reports, show that HNRNPU expression is highest at early neural stem cell and progenitor stages and in a subpopulation of neural progenitors at the later stage of neuronal differentiation (Connolly et al., 2022; Sapir et al., 2022). Despite this high expression, results indicate that HNRNPU deficiency does not affect neural cells at this early stage.

In contrast, as the neural progenitor cells commit to differentiation, the reduced HNRNPU levels led to a higher number of dividing neural progenitors compared to the control stage, in a phase where the proportion of progenitor cells should decrease. This delayed transition from progenitors to differentiating neurons could explain the lack of differences in HNRNPU expression in HNRNPU_del/+ from D0 and D5, and the almost equal expression levels with CTRL cells at D28, despite the heterozygous deletion of one allele. A recent study showed opposite effects for the complete loss of Hnrnpu, as it led to decreased proliferation followed by cell death of neural progenitors and postmitotic neurons in mice (Sapir et al., 2022). As heterozygous mutations in HNRNPU are not lethal, and the severity of the phenotypes in HNRNPU-related disorders is variable (Balasubramanian, 2022), embryonic cells can likely adapt to low levels of HNRNPU and still proliferate and differentiate. Therefore, it is reasonable that iPSCs and NES cells that retain 30–70% of the physiologic HNRNPU levels do not show a major phenotype, as we have shown here. Instead, we propose that cells with HNRNPU haploinsufficiency are inadequate to drive efficient cell fate transition of mitotic cells int differentiating neurons and other neural cells through multiple regulatory pathways, leading to stochastic rewiring of the hindbrain development. Indeed, altered regulation of proliferation has been demonstrated to cause defects in the progenitors’ fate and, ultimately, neuronal development trajectories (Lalli et al., 2020; Pilaz et al., 2016). In several cellular models of ASD, unbalanced neural progenitors population due to both hyper- and hypo-proliferation of the progenitors have been documented (Connacher et al., 2022; Marchetto et al., 2017; Mowat et al., 2003; Zucco et al., 2018). Accordingly, we hypothesize that the observed downregulation of the synaptic and neuronal maturation markers is due to abnormal enrichment of progenitors at D28 stage and a consequence of the delayed maturation process. This observation is in contrast with what was observed in a recently published study on human cortical organoids, where HNRNPU mutations are shown to associate with downregulation of ontologies referring to nucleic acid binding and upregulation of neurogenic pathways (Ressler et al., 2023). The dysregulated genes are partially resembling transcriptomic alterations in embryonic mice carrying a heterozygous mutation in HNRNPU but are discordant with the perinatal mice. Thus, the stage of cell maturation and development in which the analyses are performed seem to be critical for studying effects of HNRNPU. Moreover, in this study, we are modeling the effect of the microdeletion of the whole HNRNPU locus on a hindbrain cell model, in contrast with single HNRNPU mutations on brain cortex systems, likely contributing to the discrepancy of the observed effects of HNRNPU mutations.

We provide mechanistic insights that both RNA processing and chromatin regulation in early brain development play a critical role in the observed brain phenotypes in HNRNPU-related disorders. It has been earlier demonstrated that the correct pool of mRNA isoforms from the alternative splicing process is important in the transition from progenitor cells to neurons in the developing cerebral cortex (Zhang et al., 2016), and RNA splicing is one of the key enriched pathways from molecular and genetic studies of ASD (Gandal et al., 2018; Satterstrom et al., 2020). Therefore, our and others’ results pinpoint that further delineation of the RNA splicing program during the early steps of brain development is essential for understanding the origins of NDDs.

Furthermore, the importance of 3D genome organization for cell fate decisions during neural development is starting to emerge, showing that dynamic changes at multiple levels of chromatin organization are needed for these processes (Bonev et al., 2017; Hu et al., 2021). This is in line with our results showing major reorganizations in a later stage of neural differentiation. Indeed, multiple NDD cell models have shown that the dysfunction of the chromatin organization leads to changed neuronal maturation (Calzari et al., 2020; Markenscoff-Papadimitriou et al., 2021; Parisian et al., 2020; Su et al., 2021).

We additionally provide evidence of cellular processes, such as cilia organization and initial synaptogenesis, that are affected by the molecular changes in the HNRNPU deficiency state. For instance, we demonstrate an increase in ciliated cells. Recently, HNRNPU was indicated to localize occasionally to cilium in mice brain cells (Sapir et al., 2022). As HNRNPU-related disorders share many phenotypic features with ciliopathies (Balasubramanian, 2022; Focșa et al., 2021), the connection between cilia organization and HNRNPU warrants more studies.

In conclusion, we present the first suggestive evidence that HNRNPU mutations in human hindbrain neural progenitors result in inadequate transition to commitment to neurogenesis. This, in turn, leads to large-scale effects on chromatin organization and transcriptional landscape at later stages of neural development and presumably to diverging trajectories of neurons and other neural cells. Follow-up studies of direct targets, different developmental stages, and brain regions using both two-dimensional (2D) and organoid models will be needed to assess better the impact of HNRNPU haploinsufficiency on neurogenesis and its role in the pathogenesis of HNRNPU-related disorders.

We acknowledge that this study has limitations, including that we only compare the isogenic model constructed with siRNA with one patient cell line, providing the limited possibility to analyze the genetic background effects, recently shown to be highly important to study (Paulsen et al., 2020 preprint). Moreover, while mimicking the heterozygous deletion of the whole locus, we cannot distinguish if the observed effects are mainly driven by HNRNPU or if they are the results of a combined role of HNRNPU and HNRNPU-AS1. Furthermore, we focused on a very early model of neural development using NES cells and undirected differentiation, which cannot represent the complexity of the human hindbrain. The analyses described here are from a pool of cells not sorted for the cell type or analyzed as single cells, thus, we only give a general overview of the transcriptional and chromatin organization landscape related to HNRNPU haploinsufficiency. Further studies using single-cell techniques, different neuronal differentiation models and hindbrain human organoids should be employed to present the cell-specific effects of HNRNPU mutations in later stages of brain development.

We have earlier performed a screening for copy number variation in a twin sample from The Roots of Autism and ADHD study in Sweden (RATSS) (Bölte et al., 2014; Stamouli et al., 2018) in which we identified a male twin pair carrying the deletion on chr1:244997953-245042312 (hg19) encompassing HNRNPU. The twin pair has undergone an extensive phenotypic characterization, including evaluation for ASD, ID, and other NDDs, cognitive testing, and medical examination (Table S1). Written informed consent was obtained from individuals and their parents prior to the study. The study was approved by the regional and national ethical boards in Sweden and has been conducted in accordance with the Declaration of Helsinki for medical research involving human subjects, including research on identifiable human material and data.

Human iPSC cells were derived from fibroblasts of one of the twins from the RATSS study, a male carrying a 44Kb mutation of the HNRNPU locus spanning through COX20, AS1-HNRNPU, and HNRNPU, using a previously described protocol (Uhlin et al., 2017b). Human iPSC cells were cultured in mTeSR Plus (Stem Cell Technology) on 5 µg/ml BioLaminin 521LN (Biolamina) precoated vessels. Pluripotency and normal karyotype were confirmed, as shown in Fig. S1. Dual-SMAD inhibition was applied to derive neuroepithelial stem (NES) cells from human iPS cells as described previously (Chambers et al., 2009; Falk et al., 2012). Patient-derived and previously established NES cells from a male neurotypical donor (Uhlin et al., 2017a) were seeded on 20 µg/ml poly-L-ornithine (Sigma Aldrich), and 1 µg/ml laminin (Sigma Aldrich) precoated plastic surfaces in DMEM/F12+Glutamax medium (Gibco) supplemented with 0.05X B-27 (Gibco), 1× N-2 (Gibco), 10 ng/ml bFGF (Life Technologies), 10 ng/ml EGF (PeproTech) and 10 U/ml penicillin/streptomycin (Gibco), replacing media every day. For immunofluorescence, the glass coverslips were precoated with increased concentrations of poly-L-ornithine to 100 µg/ml and laminin to 2 µg/ml. Cells were maintained in a 5% CO₂ atmosphere at 37°C. Upon starting the differentiation, cells were changed to medium with an increased 0.5× B-27 concentration without bFGF and EGF. Two-thirds of media supplemented with 0.4 ug/ml laminin was changed every other day until differentiation 15; whereafter media was changed every third day. NES cells were harvested two days after culturing and neural cells were harvested after 5, 28 or 50 days of differentiation.

Cells were grown and harvested as earlier described and resuspended in cold RSB buffer: 10 mM Tris-HCl (pH7,4), 10 mM NaCl, and 3 mM MgCl_2. Cells were then lysed in cold lysis buffer (RSB buffer and 0,2% Triton X-100) and centrifuged to collect nuclei in the pellet. The nuclei were then incubated with different amounts of DNase I enzyme (0, 1, 3, 5, 0, 15 units) at 37°C for 10 min. The digestion was blocked by adding 50 mM of EDTA followed by incubation at 55°C for 1 min. The results of the digestion were visualized on an 0.8% agarose gel.

Cells were transfected with 0.5 µM Accell SMARTpool siRNA targeting HNRNPU mRNA (Dharmacon E-013501-00-0010) or 0.5 µM Accell Nontargeting siRNA (Dharmacon D-001950-01-20) according to manufacturer's protocol. The siRNA pool consisted of the following oligos: oligo1: UCUUGAUACUUAUAAUUGU, oligo2: CUCGUAUGCUAAGAAUGGA, oligo3: GUUUCAGGUUUUGAUGCUA, oligo4: CUAGUGUGCUUGUAGUAGU. NES cells were transfected one day after seeding, and samples were collected 24 h after treatment. Cells under differentiation were transfected once in the NES phase, once when changing media to differentiation media, and thereafter every sixth day until D28 sample harvesting.

Cells on glass coverslips were fixed in 4% formaldehyde for 20 min at room temperature and washed with 1× TBS. Blocking was performed with 5% Donkey Serum and 0.1% Triton X-100 in 1× TBS for 1 h. Primary antibodies diluted in blocking buffer (Nestin 1:1000; SOX2 1:1000; Synapsin 1/2 1:500; Homer1 1:250; MAP2 1:500; HNRNPU 1:500; H3K9me3 1:500, ARL13B 1:10,000, PCNT 1:250) were incubated at +4°C overnight. Careful washing was done with 1× TBS, and secondary antibodies diluted 1:1000 in blocking buffer were incubated at room temperature for 1 h, covered from light. Coverslips were washed and mounted with Diamond Antifade Mountant (ThermoFisher Scientific). Images were acquired with LSM 700 Zeiss Confocal Microscope using 20× or 63× magnification and 0.5 µm z-stack through the sample.

At least three replicates per sample were stained and analyzed for the immunocytochemistry analyses. The images stained with synapsin 1/2 were analyzed with ImageJ plugin Synapse Counter, and the statistical Student's t-test analyses were performed in R. The statistical analyses for SOX2 positive nuclei were performed in R by χ²-test. For primary cilia analysis, a cell was considered ciliated if positive for both cilia markers ARL13B and PCNT; the statistical testing was performed in R by χ²-test. For the images of the nuclei stained with the H3K9me3 antibody, the single nuclei were cropped using an in-house built macro for ImageJ. A minimum of thirty nuclei per replicate was analyzed with NucleusJ, a plugin of ImageJ, using the default settings (Poulet et al., 2015).

Cells were collected in extraction buffer (50 mM Tris-HCl, 100 mM NaCl, 5 mM EDTA, 1 mM EGTA) supplemented with a 1× protease inhibitor cocktail (ThermoFisher Scientific) using a plastic cell scraper. The samples were sonicated with six short pulses at 36% amplitude (Vibra-Cell VCX-600, Sonics). Protein concentrations were measured with Qubit Protein Assay Kit (ThermoFisher Scientific). The total protein sample (150 ng/µl) was loaded on the capillary Western blot system Simple-Western-JESS (Bio Techne), multiplexed for total protein and chemiluminescence detection of HNRNPU (1:10 dilution). Data were analyzed using Compass for S.W. software (5.0.1), and the HNRNPU peak area was normalized against the total protein area. Three to five biological replicates were analyzed for each time point, and the significance at each time point was evaluated by Student's t-test.

Cells were lysed in TRIzol reagent (Invitrogen), and RNA isolated with ReliaPrep RNA Cell Miniprep kit (Promega Z6012). We extracted RNA samples for three to five biological replicates per cell line and time point. Samples were delivered to NGI Sweden for library preparation and sequencing. The samples were subjected to library preparation with Illumina Truseq Stranded total RNA RiboZero GOLD kit, except for siNTC-D28 and siHNRNPU-D28 libraries prepared with Illumina TruSeq Stranded mRNA kit due to low RNA yields. All libraries were sequenced on the NovaSeq6000 platform with a 2×151 setup using NovaSeqXp workflow in S4 mode flowcell. We obtained, on average, 35 and 65 million reads per sample for early time points and D28, respectively, with a minimum 82,7% alignment rate.

Differential gene expression analysis was performed using DESeq2 (v1.24.0) (Love et al., 2014) in R (v4.1.2). The significance thresholds used were adjusted P value<0.05 (Benjamini-Hochberg adjustment), base mean >20, and absolute log2 fold change >0.58. Pathway analysis was done according to previously described protocols (Reimand et al., 2019). In short, the ranked gene expression list was used in gene set enrichment analysis (GSEA) (Version 4.3.0) to analyze the enrichment in the gene ontology, molecular function, and biological process gene lists (v. 7.4), and enriched categories were visualized in Cytoscape (v3.8.2) with Enrichment Map (v3.1.0) and AutoAnnotate (Merico et al., 2010; Reimand et al., 2019; Shannon et al., 2003).

Differential exon usage analysis was performed using the DEXSeq package (v1.40.0) (Anders et al., 2012). A flattened annotation file was created using provided python script, excluding the aggregate exon bins, and exon counts were counted using provided python script. The analysis was performed with the formula ∼ sample+exon+condition: exon. A gene was called to have evidence for DEU if at least one exon bin was differentially used between conditions. The difference was considered significant with FDR (Benjamini-Hochberg adjustment) adjusted P<0.05, exon base mean >10, and an absolute log2 fold change>0.58. Over-representation analysis (ORA) was performed using the online tool WebGestalt (Liao et al., 2019). In addition, hypergeometric tests were used to test for enrichment between differentially expressed genes and DEU genes and specific NDD-related gene lists: ASD gene list (SFARI genes selected for score 1,2 and syndromic, release 07-20-2022), ID gene list (green and amber genes from https://panelapp.genomicsengland.co.uk/ v3.1632), epilepsy gene list (green and amber genes from https://panelapp.genomicsengland.co.uk/ v2.547) and general development [compiled gene list (Becker et al., 2020)].

HNRNPU binding motif analysis was performed using AME function (McLeay and Bailey, 2010) of MEME-Suite v5.5.3 with default settings and background datasets of all expressed exons with exon base mean >10 (151878 and 150565 exons) and their flanking introns (156090 and 159915) in HNRNPUdel and siHNRNPU. The position weight matrix for UGUAUUG motif was downloaded from RBPmap. The exon bins obtained from DEXSeq were collapsed with annotated exons using Intersect function of Bedtools v2.31.1. In instances of overlapping exons, the longest exon was selected for the analysis. The intron regions were extracted from the annotation file using intronicParts function of GenomicRanges v1.50.4.

The RNA was reverse-transcribed using iScript cDNA Synthesis Kit (BioRad) and cDNA quantified with SsoAdvanced Universal SYBR Green Supermix (BioRad) following the manufacturer's protocols on a CFX96 thermal cycler (BioRad). The primer used are: HNRNPU-AS1 (AGGAAGCTGTACACTGGAGG, CAATGTCTTCACCAATAACAAAGC); HNRNPU (AGTTTAACAGAGGTGGTGGCC, GCCCCTCCTATTATATCCGCC); GAPDH (AAGGTGAAGGTCGGAGTCAAC, GGGGTCATTGATGGCAACAATA). CFX Manager software was used to record amplification curves and to determine Ct values. RT-qPCR reactions were performed in technical triplicates. We calculated the ΔCt to the GAPDH housekeeping gene and ΔΔCt to control cell lines. We used three biological replicates of cells seeded at different passages. Statistical significance between cell lines was determined with ANOVA and post hoc Tukey HSD in R (v. 4.1.2).

Cells were cultured as NES for D0 collection or differentiated for 28 days and harvested by briefly rinsing the cells in accutase and then incubating them with TrypLE express (Gibco) and neural isolation enzyme (ThermoFisher Scientific). Samples were then processed using the Arima-HiC kit (PMID: 29779944) according to the Arima Genomics User Guide for Mammalian Cell Lines (catalog number A510008). Briefly, we crosslinked harvested cells with 2% formaldehyde and then used approximately 1 million fixed cells as input for each replicate sample. Subsequently, we used 1.5 μg of Hi-C template for biotin pull-down and library preparation according to the Arima Genomics User Guide for Library Preparation using KAPA Hyper Prep Kit. Specifically, we used eight PCR cycles for library amplification. We then sequenced the Arima-HiC libraries for each time point on one flowcell on the Illumina NovaSeq S Prime system, obtaining an average of 900 M sequencing reads for D0 samples, and 1000 M sequencing reads per sample for D28 samples. Sequencing was carried out at the National Genomics Infrastructure at the Science for Life Laboratory (SciLifeLab) in Stockholm, Sweden.

We processed the raw sequencing reads using the HiCUP pipeline (v0.7.4) (PMID: 26835000) with default parameters. Briefly, the pipeline employs Bowtie2 (v2.4.1) (PMID: 22388286) to align the reads to the human reference genome (GRCh37/hg19) and filter out experimental artifacts (i.e. circularised, re-ligated, and duplicate reads). We generated the ‘digest file’ using the ‘hicup_digester’ command with the Arima option (--Arima). We used the HiCUP output files (BAM format), which contain only valid, non-redundant read pairs, as input for pairtools (v0.3.0). First, we converted the BAM files into .pairsam format using the pairtools parse and sort modules. In addition to individual replicates, we generated pooled samples for each developmental stage by merging replicates using the pairtools merge module. We marked read duplicates using the pairtools dedup module, with option --mark-dups, and filtered the results by selecting only specific pair types (i.e. ‘U.U.’, ‘U.R.’ and ‘R.U.’) via the pairtools select module, which produced a .pairs format output. Finally, we added the fragment information using the ‘fragment_4dnpairs.pl’ convenience script provided alongside the Juicer pipeline and converted the .pairs files into .hic format using the Pre module from Juicer-Tools (v1.22.01). Unless explicitly required by the software/package, the contact matrices were normalized to the smallest library and corrected with ICE using HiCexplorer (v3.7) (Ramírez et al., 2017). HiCexplorer package was used for mapping genomic contacts and contact enrichment between the samples.

The compartments were called at 1 Mb resolution, and the first four PCAs were extracted by HiCexplorer using the hicPCA tool and the gene track from hg19 to assure the correct orientation of the eigenvector. The PCA for each chromosome and each sample were chosen by considering the highest correlating PCA between the samples and the best mapping with the gene density eigenvector. The compartments were defined as A and B according to the eigenvector orientation, with A being the gene-rich and positive eigenvector and B being the gene-poor regions and negative eigenvector. FanC toolkit (v0.9.1) (Kruse et al., 2020) was used to calculate compartment strength and cis-trans- compartment interactions. The R package annotatr was used to annotate the genomic regions switching compartments in the different conditions.

The deconvolution of bulk RNA-seq data to previously published scRNA-seq sample (Becker et al., 2020) was done using the Bseq-SC package (Baron et al., 2016). Bseq-SC uses cell-type-specific marker genes from single-cell RNA transcriptomes to predict cell-type proportions underlying bulk RNA transcriptomes. Deconvolution was done with 30 marker genes for each cell population, except for cluster ‘Undefined maturing neurons’, which had only 25 defining genes. Statistical differences in the estimated cell proportions were calculated by the χ² statistical test in R (v4.1.2).

Data from Human Brain Transcriptome (https://hbatlas.org) were obtained to visualize HNRNPU expression in different human brain regions during development. Additional stem cell time-course data was accessed via LIBD Stem Cell Browser (http://stemcell.libd.org/scb). RPKM values for HNRNPU were extracted for evaluating HNRNPU expression during neuronal differentiation from iPSCs (Burke et al., 2020). Seurat object for human cortical organoid data was downloaded from GEO with accession number GSE219317 (Ressler et al., 2023).

To analyze differences in cellular proliferation, the BrdU assay (ab126556, Abcam) was performed for three replicates at D0, D5, and D28 time points. Twenty-four hours prior to the read-out, 1X-BrdU reagent was added to the cell culture vessels for incorporation by incubation at 37°C and 5% CO₂. The culture media was aspirated for the read-out, and cells were fixed with the supplied fixing solution. This was followed by exposure to the anti-BrdU antibody (primary antibody), incubation at room temperature for 1 h, and washing with the supplied plate wash buffer. The cells were incubated with the HRP-tagged secondary antibody at room temperature for 30 min, followed by TMB exposure and recording absorbance at 450 nm. Afterwards, the cells were incubated in TBS, 0.1% Triton-X 100, and Hoechst 1:1000 for 15 min, followed by fluorescence measurement, which was used to calibrate the BrdU quantification for the number of nuclei in each sample. The quantification of proliferation rate followed by Student's t-test statistical comparison was done in R (v4.1.2).

Whole-cell patch-clamp recordings were performed at 5, 8, 28, and 50 days of differentiation. Immediately prior to recordings, the cells were washed with 1× PBS, and then Krebs-Ringer's solution composed of (in mM): NaCl 119, KCl 2.5, NaH2PO4 1, CaCl2 2H2O 2.5, MgCl2 6H2O 1.3, HEPES 20, D-Glucose 11 (pH 7.4)+Laminin 1:1000 was added in the dish. The recordings were performed in Krebs-Ringer's solution. Recording pipettes were fabricated with a Narishige pc-100 puller and had resistances of 3-5 MOhm when filled with the internal solution composed of (in mM): 120 K-gluconate, 0.1 EGTA, 4 MgATP, 0.3 Na2GTP, 10 HEPES, 20 KCl and 5 Na2-phosphocreatine (pH 7.4). Current and voltage responses were measured at room temperature using a MultiClamp 700B amplifier (Molecular Devices) and digitized with Axon™ Digidata^® 1550B analog-to-digital converter connected to a personal computer running pClamp 11.0.3 (Molecular Devices). Membrane capacitance and resistance were derived from the pClamp 11.0.3 (Molecular Devices) membrane-test function. Data analysis was performed by Clampfit 10.7 (Molecular Devices).

The authors are grateful to the donors of the cell lines for their participation in this research. The authors acknowledge support from the National Genomics Infrastructure in Stockholm funded by Science for Life Laboratory, the Knut and Alice Wallenberg Foundation and the Swedish Research Council. The computations and data handling were enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) and the Swedish National Infrastructure for Computing (SNIC) at Uppsala Multidisciplinary Center for Advanced Computational Science, partially funded by the Swedish Research Council through grant agreements number 2022-06725 and number 2018-05973. We thank A. Smialowska at NBIS (National Bioinformatics Infrastructure Sweden) for technical support with HiC data analysis, which was made possible through application support provided by SNIC. The authors acknowledge support from the iPS Core facility at Karolinska Institutet for assistance with the generation of iPSC and NES cells.