Genome-wide screening reveals essential roles for HOX genes and imprinted genes during caudal neurogenesis of human embryonic stem cells

Summary

Mapping the essential pathways for neuronal differentiation can uncover new therapeutics and models for neurodevelopmental disorders. We thus utilized a genome-wide loss-of-function library in haploid human embryonic stem cells, differentiated into caudal neuronal cells. We show that essential genes for caudal neurogenesis are enriched for secreted and membrane proteins and that a large group of neurological conditions, including neurodegenerative disorders, manifest early neuronal phenotypes. Furthermore, essential transcription factors are enriched with homeobox (HOX) genes demonstrating synergistic regulation and surprising non-redundant functions between HOXA6 and HOXB6 paralogs. Moreover, we establish the essentialome of imprinted genes during neurogenesis, demonstrating that maternally expressed genes are non-essential in pluripotent cells and their differentiated germ layers, yet several are essential for neuronal development. These include Beckwith-Wiedemann syndrome- and Angelman syndrome-related genes, for which we suggest a novel regulatory pathway. Overall, our work identifies essential pathways for caudal neuronal differentiation and stage-specific phenotypes of neurological disorders.

Graphical abstract

Highlights

• •Loss-of-function genomic screen was performed in hESC-derived neuronal cells
• •Screen results revealed early phenotype of multiple neurological disorders
• •Paralogous HOX genes were found to have unique roles in caudal neurogenesis
• •Imprinted genes show essentiality starting in neurogenesis

Benvenisty and colleagues conducted a genome-wide CRISPR-Cas9 knockout screen using haploid human embryonic stem cells to identify neurogenesis essential genes. They discovered the essential role of signaling proteins, identified early phenotypes of neurological disorders, including neurodegenerative disorders, and demonstrated the unique roles of paralogous HOX genes in neurogenesis. Moreover, they established the essentialome of imprinted genes in early human embryogenesis.

Introduction

Human embryonic stem cells (hESCs) are widely used to model embryogenesis and genetic diseases, and as a source of cells in regenerative medicine (Avior et al., 2016; Shahbazi et al., 2019; Trounson and DeWitt, 2016). hESCs are employed extensively because they have unlimited ability for self-renewal without being cancerous, and they have the potential to differentiate into all cell types of the human body. Thus, many in vitro differentiation protocols were established, including those for the differentiation of hESCs into the embryonic germ layers and neurons (Schuldiner et al., 2000; Williams et al., 2012).

In vitro models of neuronal differentiation have been instrumental in studying factors and gene networks that play a role during neurogenesis in humans. The induction of neuronal identity has been previously studied focusing on a handful of factors/genes and their potential to differentiate hESCs or transdifferentiate mature cell types (Chambers et al., 2009; Kawasaki et al., 2000; Pang et al., 2011; Perrier et al., 2004; Schuldiner et al., 2001; Tanabe et al., 2018). Although this approach has been proved to be useful in revealing genes sufficient to induce neurogenesis, it has been largely limited by the narrow scope of factors that are investigated based on the exiting literature. More recently, this approach has been extended to a more high-throughput scale using gain-of-function screens by CRISPR activation testing for the neurogenesis-inducing capacity of transcription factors (TFs) (Black et al., 2020; Ng et al., 2021). However, because of the focus of these studies on TFs, regulators of neurogenesis in humans beyond this gene group are yet to be determined.

Analysis of the transcriptome between undifferentiated hESCs and their neuronal derivatives has also been used extensively as a high-throughput approach to reveal potential regulators of differentiation (Meléndez-Ramírez et al., 2021; Tanabe et al., 2018; Verrier et al., 2018). While these studies are based on an unbiased methodology and have identified several factors and pathways that are regulated during neuronal differentiation, transcriptomics analyses, as well as the gain-of-function screens, are unable to differentiate between essential and redundant genes.

Functional genomics approaches testing gene essentiality in a loss-of-function (LOF) high-throughput manner have improved due to the current developments in haploid genetics. We recently isolated haploid hESCs and differentiated them successfully to the cells of all three embryonic germ layers (Sagi et al., 2016a). Utilizing these cells, we constructed a genome-wide CRISPR-Cas9 LOF library with over 180,000 single guide RNAs (sgRNAs) targeting 18,166 protein-coding genes (Yilmaz et al., 2018). The haploid cells, harboring only one allele for each gene, enable a most robust LOF screening, even though they undergo spontaneous diploidization, occurring in a small percentage (Sagi et al., 2016b). This haploid mutant library allowed us to identify the essential genes for the normal growth and survival of human pluripotent stem cells, showing an enrichment of essential genes in the nucleus as opposed to a small percentage of essential genes coding for cell surface proteins (Yilmaz and Benvenisty, 2019; Yilmaz et al., 2018). Interestingly, this screen also revealed growth phenotypes in hESCs for one-fifth of all the genes responsible for autosomal recessive disorders which have a growth-retardation phenotype, suggesting that a major phenotype of these disorders can be traced back to early embryogenesis.

To study the gene networks essential for germ layer specification events in humans, we utilized the same screening platform and differentiated the mutant library in hESCs into ectoderm, mesoderm, and endoderm (Yilmaz et al., 2020). These studies revealed that the percentage of essential genes coding for the extracellular matrix and plasma membrane proteins was higher compared to those in undifferentiated cells, leading to the identification of the essential signaling pathways for the differentiation of the three germ layers. Moreover, analysis of the essential genes for neuroectodermal cells suggested that these cells can be used to model neurological disorders such as microcephaly and autism (Yilmaz et al., 2020).

Here, we sought to explore the essential genes for later stages of neurogenesis by differentiating the genome-wide LOF library in hESCs into caudal neurogenesis (Schuldiner et al., 2001). Our screen allowed us to identify the essential signaling pathways and TFs for caudal neurogenesis. Among the developmentally relevant TFs, homeobox (HOX) family was represented abundantly within the group of essential TFs. We show that the highest-ranking essential HOX gene, HOXA6, and its essential paralog, HOXB6, regulate the expression of large sets of genes associated with neuronal differentiation. Interestingly, analysis of essential genes for caudal neurogenesis further suggested that these neuronal cultures could serve as a model system to study neurological disorders, including degenerative ones.

In addition, our screen revealed essentiality for a group of imprinted genes during neuronal differentiation. This unique group of genes is only expressed from either the maternal or the paternal allele (Surani et al., 1984), and they are known to be essential for proper embryonic development (Barton et al., 1984). Previous reports suggested that the maternally expressed genes (MEGs) are more essential for neurogenesis than the paternally expressed genes (Sagi et al., 2019). By analyzing the hits of our current screen in neurogenesis, along with data in undifferentiated hESCs and their differentiated germ layers, we could establish the essentialome of imprinted genes across different stages of early human development. Our results suggest that there are no essential MEGs until gastrulation, while there are specific essential MEGs for neurogenesis.

Results

Establishment and analysis of a genome-wide LOF screen to study neuronal differentiation

To investigate the essential gene networks for neuronal differentiation, we first differentiated a CRISPR-Cas9-based genome-wide LOF mutant library of hESCs, previously established by us (Yilmaz et al., 2018), into neuronal cultures (cells differentiated from neural progenitor cells into immature and mature neurons), using retinoic acid (Schuldiner et al., 2001), a protocol which results in mostly caudal cells (Erceg et al., 2008). We previously utilized this mutant library to identify essential genes for hESCs and their differentiation into the three germ layers by using in vitro models of neuroectoderm, definitive endoderm, and early mesoderm differentiation (Yilmaz et al., 2020). In this study, we utilized the same mutant pool to study neural differentiation by thawing and culturing the mutant population of hESCs, in which all sgRNAs were still represented.

Successful differentiation into neuronal cells was assessed by transcriptome analysis, focusing on the expression of stage-specific markers for pluripotent stem cells, neuroectodermal cells (Yilmaz et al., 2020) (9 days of differentiation), and neuronal cells (Schuldiner et al., 2001) (28 days of differentiation; Figure 1A). This analysis showed that the expression of markers for pluripotency and neuroectoderm was downregulated in neuronal cultures, while they expressed more specific neuronal markers. In addition, we also analyzed the expression of a large panel of genes that are classified under the Gene Ontology (GO) term “Neuron Differentiation” in the gene set enrichment analysis (GSEA) database (Mootha et al., 2003; Subramanian et al., 2005) (Figure S1A). Expression of most of these genes was lowest at the pluripotency stage and was progressively upregulated through the neuroectoderm stage, reaching higher levels at the neuronal stage (Figure S1A). In addition, immunostaining with a specific TUJ1 antibody demonstrated that the differentiated cultures were positive for the expression of this marker with intermediate filament and axon-like projections expressing TUJ1 (Figures S1B and S1C). Quantifying immunostaining for different neural markers, namely TUJ1, neurofilament medium polypeptide (NEFM), and a hindbrain-marker HOXA6, in the differentiated neuronal cells revealed expression in more than 85% of cells (Figures S1D and S1E), while immunostaining for OCT4-expressing hESCs revealed the absence of expression of neural markers (Figure S1E). Notably, a major subset of the cells exhibits morphology of neuroblasts without neurites (Singh et al., 2011; Sitnikov et al., 2023) or immature neurons with short neurites.

Figure 1.

Establishment of a genome-wide loss-of-function library to identify essential genes for neuronal differentiation of human embryonic stem cells

(A) Heatmap demonstrating relative transcript levels of the markers of pluripotency, neural stem cells, and neurogenesis in undifferentiated hESCs, neuroectoderm, and neuronal culture (9 and 28 days of differentiation, respectively. n = 3 independent replicates, for each differentiation stage).

(B) Log₂ ratio of the percentage of essential genes in neuroectoderm and neuronal culture to that in hESCs for each cellular compartment.

(C) Volcano plot demonstrating −log₂ (FDR) and CRISPR scores (CS) (the average log₂ ratio of the abundance of mutants in the neuronal cultures to their abundance in the neuroectoderm cultures) of genes that are associated with the plasma membrane and the extracellular matrix. Kolmogorov-Smirnov test was performed to determine the significance of depleted genes (n = 10–40 sgRNAs). 171 essential genes are labeled in pink. Signaling pathways identified by analyzing the predicted protein-protein interactions among the essential genes are highlighted in the schematic illustrations on the right. The thickness of the lines connecting the essential genes in the schematics indicates the level of the confidence for the interaction. Essential genes shown in the schematics within predicted signaling pathways are labeled in red in the volcano plot. See also Figures S1–S3.

Moreover, we performed single-cell RNA sequencing (scRNA-seq) using the 10× Genomics platform of haploid cells and diploid cells and compared their differentiation process. To assess identity and similarity between the cell populations in these samples, we conducted an overlap analysis based on their gene expression profiles (Figure S2A). Notably, a considerable proportion of cells from both samples exhibit overlapping gene expression profiles. To gain further insight into the heterogeneity of cellular populations within our samples, we performed clustering analysis on the single-cell gene expression data (Figure S2B). We identified three major distinct clusters of cells based on their transcriptional profiles. For both haploid and diploid cells, about 50% of the cells are clustered in cluster number 1, about 30%–40% of the cells are clustered in cluster number 2, and about 10% of the cells are clustered in cluster number 3 (Figure S2B). Notably, all the clusters express TUJ1 (TUBB3), a pan-neuronal marker, with major subset of cells expressing MAPT and SNAP25, identifying them as neuronal cells (Figure S2C). In order to classify the neuronal stage of the clusters, we examined the expression of several SRY-related HMG-box (SOX) genes, namely SOX2, SOX4, and SOX11. During neurogenesis, early progenitor cells express SOX2 and late progenitor cells co-express SOX2 and SOX4/11, whereas neuroblasts and immature neurons express SOX4/11 only (Stevanovic et al., 2021). In our neuronal culture, cluster 2 and a subset of clusters 1 and 3 co-express SOX2 and SOX4/11, whereas the rest of the cells in clusters 1 and 3 express SOX4/11 only (Figure S2D). These results indicate that we have heterogeneous population of neural cells, ranging from late progenitor cells to neurons, expressing SNAP25 and MAPT.

To further explore the transcriptional heterogeneity observed within the identified cellular clusters, we examined the expression patterns of various markers of the different regions of the developing human brain and found enrichment of HOX genes. HOX genes have been shown to have a temporal co-linear expression pattern, in which the first genes in the cluster are expressed earlier during development in the anterior hindbrain and the following ones are expressed sequentially at later stages at the posterior hindbrain and spinal cord (Quinonez and Innis, 2014). In our analysis we found that cluster 1 has higher expression of the anterior HOX genes, i.e., HOXB2, and lower expression of the more posterior HOX genes, i.e., HOXB7 (Figures S2E and S2F), whereas clusters 2 and 3 exhibited distinct expression patterns of posterior HOX genes, indicative of a posterior positional identity (Figures S2E and S2F). Notably, none of the clusters exhibit expression of forebrain markers (e.g., SIX3 and FOXG1), and only a small minority expresses midbrain markers (e.g., EN1; Figure S3A). This analysis may suggest that the haploid cells differentiate into more anterior structures, while the diploid cells differentiate into more posterior structures, although both differentiate into hindbrain and anterior spinal cord.

In order to validate our observation of hindbrain differentiation within our culture, we searched for specific markers for each cluster (Figures S3B–S3D), showing that they are expressed in human embryo brain at higher levels in hindbrain than in midbrain in both adult (Uhlén et al., 2015) and fetal brain (Braun et al., 2023) (Figures S3E and S3F), and specifically HOX genes are mostly expressed at 5 and 5.5 post-conception weeks, and much less at 7 post-conception weeks (Figure S3F). We also show that a very small minority of the cells cluster outside the neural linage as endoderm cells (cluster A, ∼3%), neural crest cells (cluster B, ∼1%, Figure S3G), and unidentified low-reads cells (cluster C, ∼1%), whereas a negligible number of cells express mesoderm or astrocytes markers (Figure S3G). To ensure neurogenesis in the HOX-positive cells, we generated pseudobulk expression profile from these cells, enabling the identification of differentially expressed genes compared to hESCs. Subsequently, we conducted GO analysis on the identified differentially expressed genes and found enrichment of neurogenesis-related pathways (Figure S3H).

We next assessed the abundance of the mutants in the differentiated library. To specifically identify the essential genes for caudal neurogenesis, we quantified the changes in abundance of the mutants between the neuronal cultures and the preceding neuroectodermal stage (Yilmaz et al., 2020), using next-generation sequencing. A CRISPR score (CS) was assigned to each gene in the mutant library by taking the average log₂ ratio of the abundance of mutants in the neuronal cultures to their abundance in the neuroectoderm cultures. A statistically significant negative CRISPR score indicates a decrease in abundance of mutants in the neuronal culture, and thus the targeted genes in these mutants are identified as potentially essential for neuronal differentiation (1,910 gene listed in Table S1, with 930 of them having CS < −1). We first analyzed the expression of the essential genes in prenatal brain samples (Lindsay et al., 2016) and found that ∼70% of the essential genes (false discovery rate [FDR]<0.05) are expressed in brain samples corresponding to approximately 30 days post-fertilization.

Next, we analyzed the percentage of essential genes within cellular compartments and compared the percentages between neuronal and pluripotent stem cells using our previously published datasets. This comparison revealed an increase in the percentage of essential genes within outer cellular compartments such as the extracellular matrix and plasma membrane during neuronal differentiation, suggesting a group of essential signaling networks during these differentiation events (Figure 1B). Interestingly, a similar observation was made for neuroectoderm differentiation, highlighting the role of essential signaling networks for distinct stages of differentiation (Figure 1B). To identify the essential signaling pathways and their specific members, we analyzed essential genes associated with extracellular matrix and plasma membrane (Figure 1C). Using the STRING database (Szklarczyk et al., 2019), we performed an analysis for the predicted protein-protein interactions within this specific group of essential genes. This analysis, together with a GO analysis for the interacting groups of genes, revealed several signaling pathways, including axon guidance, neuroactive receptor-ligand interactions, and wingless-related integration site (WNT) pathway, in which we could highlight the essential pathway members based on our screen (Figure 1C).

Early disease phenotypes for neurological disorders

The essentialome of caudal neurogenesis, which is defined as the entire set of essential genes for this cell fate transition, offers a novel angle to study the phenotypes of neurological conditions. Our dataset allows us to interrogate the involvement of early differentiation phenotypes for this large set of disorders. To this end, we assessed the essentiality of all genes previously suggested to be associated with neurological conditions in the Online Mendelian Inheritance in Man (OMIM) database (Amberger et al., 2015). This analysis revealed essentiality for a group of genes associated with neurological conditions, suggesting that these disorders have early differentiation phenotypes during neurogenesis in the embryo. Nearly 37% of these conditions were associated with movement disorders, while ∼17% were cephalic disorders, with hypotonia, autism, mental disorders, developmental conditions, atrophy, ataxia, and glia-related conditions making up the rest of them (Figure 2A). At least 10% of the causative genes for each movement disorder we analyzed were found to be essential for neuronal differentiation, and this fraction reached nearly 20% for the causative genes of hyperreflexia and spasticity (Figures 2B and S4A). Between 5% and 25% of the causative genes for the major types of cephalic disorders were found to be essential, while up to 35% of the causative genes for less common types, such as colpocephaly, were essential (Figures 2C and S4B). In addition, around 20% of genes associated with ataxia and atrophies, between 15% and 30% of genes related to developmental conditions, and between 2% and 15% of genes causing autism and several mental disorders were also found to be essential for neuronal differentiation (Figures 2D–2F and S4C). Among the disorders that demonstrated a significant phenotype in our neuronal system are LIG4 syndrome and Seckel syndrome 10 associated with microcephaly, HSD10 mitochondrial disease associated with hypotonia, and MRT38 disease associated with autism. These observations demonstrate that a high number of neurological conditions have early phenotypes during neuronal differentiation and suggest novel approaches for disease modeling.

Figure 2.

Analysis of early differentiation phenotypes associated with neurological disorders

(A) Pie chart demonstrating the distribution of percentages of neurological disease groups that are associated with essential genes for neuronal differentiation.

(B–F) Percentage of essential genes for neuronal differentiation within all the genes that are associated with specific neurological conditions under broader categories of movement disorders (B), cephalies (C), ataxia and atrophies (D), autism and mental disorders (E), and developmental conditions (F). Less common types of movement disorders, cephalies, and mental disorders are averaged in one column.

(G) Bar plot demonstrating the percentage of neurodegeneration-related genes with a low score based on their CRISPR scores and the level of their statistical significance (lower than −10, see experimental procedures for further details), in undifferentiated hESCs and in differentiated neuroectoderm cells or neuronal culture as indicated. ND, neurodegeneration.

(H) Dot plots displaying scores of parkinsonism-related genes in undifferentiated hESCs and in differentiated neuroectoderm cells or neuronal culture as indicated. Genes are ordered alphabetically. Genes with a score lower than −10 are labeled with their gene symbol. See also Figure S4.

Given our observation on the early phenotypes of a vast array of neurological conditions, we next investigated neurodegeneration-related genes. Neurodegenerative disorders are age-related conditions causing progressive degeneration and ultimately death of nerve cells. Such conditions can manifest themselves at different stages of life, from birth to old age. We sought to understand whether phenotypes of neurodegeneration-related mutations can be detected in hESC-derived cells and at which developmental stage they would be manifested. To this end, we retrieved lists of neurodegeneration-related genes from the OMIM database and divided the disorders to two categories as the childhood (developmental neurodegeneration) and the adult ones (early- or late-onset neurodegeneration) (Hamosh et al., 2005). We then examined the effects of perturbations in these genes on the growth of undifferentiated hESCs, and their differentiation into neuroectodermal cells and neuronal cells, using the datasets derived from our previous and current LOF screens (Yilmaz et al., 2018, 2020). In order to compare between the screens, we calculated a score as a representation of the size of the effect of the mutated gene as well as its significance. This analysis revealed a growth phenotype at the undifferentiated stage in a few genes related only to the developmental neurodegenerative disorders, but not to adult ones. Notably, as the differentiation progressed, a more extensive phenotype could be detected for mutations in neurodegeneration-related genes, with 9% of them exhibiting phenotypes during the differentiation into neuroectodermal cells and up to 33% of them during differentiation into more mature neuronal cells (Figure 2G). Interestingly, we were also able to detect phenotypes of mutations in genes associated with adult neurodegenerative disorders, and the percentage of these genes with such early phenotypes increased with the level of differentiation (Figure 2G).

We then focused on parkinsonism, a group of brain conditions that cause slowed movements, rigidity, and tremors, but not confined to Parkinson’s disease, as an example of a specific type of neurodegeneration. Out of 66 parkinsonism-related genes, only 2 had a growth phenotype in undifferentiated hESCs, with 6 genes having phenotypes during differentiation into the neuroectodermal cells and 12 of them during differentiation into more mature neuronal cells. One gene, EIF4G1, showed phenotypes at all stages, and 3 others, TAF1, UNTF, and DCTN1, affected neuroectoderm and neuronal cells (Figure 2H).

Essentiality for HOX cluster genes during neuronal differentiation of human pluripotent stem cells

TFs drive the major gene expression alterations required for cell fate changes. Therefore, we sought to identify the essential TFs networks for neuronal differentiation in humans, using our genome-wide screens. For these analyses, we focused on TFs families, such as the forkhead box (FOX) family and the homeobox gene family, which have been previously shown to have widespread roles throughout development (Holland, 2013; Jackson et al., 2010).

The FOX family of TFs has been shown to regulate cellular differentiation, with some members of the family being associated with neuronal development and neurological phenotypes (Devanna et al., 2014; Golson and Kaestner, 2016). We first compared the expression levels of FOX family members in neuronal cultures to their levels at the stages of pluripotency and neuroectoderm. This analysis revealed that a fraction of the genes in these families are mainly expressed in the neuronal cultures, while the expression of the others is higher at the earlier stages (Figure S4D). Within this TFs family, our screen identified RBFOX2, RBFOX3, and FOXRED1 as essential for neuronal differentiation (Figure S4E). Interestingly, RBFOX3 and FOXRED1 are expressed more at earlier stages, suggesting that they may have a priming function for neuronal specification.

The homeobox family of TFs contains more than 200 members, many of which are major regulators of developmental processes such as cellular differentiation and morphogenesis (Mark et al., 1997). Differential expression of several members of this family also plays a role in patterning the nervous system (Kappen et al., 2013). Thus, we profiled the stage-specific expression patterns of essential homeobox genes for pluripotency, neuroectodermal, and later neuronal stages (Figure 3A). For this analysis, we used published transcriptomics and essentialome datasets for hESCs and their neuroectodermal derivatives along with those for neuronal differentiation from this study. Interestingly, one-third of essential homeobox genes with an enriched expression in neuronal cultures as compared to the previous stages were identified as HOX cluster genes despite the suggested redundancy between paralogs of this group.

Figure 3.

Essentiality of homeobox genes for neuronal differentiation

(A) Heatmap demonstrating the relative expression levels of essential homeobox genes with an enriched expression for each developmental stage: hESCs, neuroectoderm and neuronal culture (n = 3 independent replicates, for each differentiation stage). Gene names for these stages are labeled in gray, blue, and red, respectively.

(B) Relative transcript levels of HOX genes normalized to the developmental stage, at which it is expressed the most (hESCs in gray, neuroectoderm in blue, or neuronal culture in red). NA indicates expression values of transcripts per million (TPM) <1.5. Four HOX gene clusters (A–D) are shown with the genomic order of the paralogs they contain. Numbers in the x axis indicate the specific gene in the HOX cluster.

(C) Volcano plot showing the −log₂(FDR) values and CRISPR scores for HOX genes that are expressed in neuronal culture (TPM > 1.5, Kolmogorov-Smirnov test, n = 10–40 sgRNAs). Upper dashed line shows a significance cutoff of FDR <0.05, while the lower dashed line indicates a more permissive cutoff of FDR <0.25. Essential genes are labeled in red based on the more stringent statistical cutoff and in pink based on the more permissive cutoff. See also Figure S4.

There are 39 HOX genes in humans, and they are spread over four genomic clusters known as HOXA, HOXB, HOXC, and HOXD, with as many as 13 paralog genes presented within each cluster. An expression analysis in our culture system across pluripotency, neuroectodermal, and neuronal stages demonstrated the recapitulation of the in vivo temporal expression pattern throughout differentiation. The first genes of HOX clusters were still expressed at the neuroectodermal stage, while the later ones between the 3^rd and 9^th paralogs were specifically expressed in neuronal cultures (Figure 3B). Interestingly, our screen identified several HOX genes that were highly expressed in neuronal cultures as essential for neuronal differentiation (Figure 3C). Among these, HOXA6 was identified as the most essential one, while it also ranked among the top 50 highest-ranking essential genes genome-wide. A paralog of HOXA6, namely HOXB6, was also identified as essential based on a less stringent statistical threshold. To validate these findings, we performed a second genome-wide LOF screen in differentiated neuronal cells. In this iteration, we sorted the cells based on TUJ1 positivity (Figure S4C). Due to the nature of the intra-cellular fluorescence-activated cell sorting analysis, and the enrichment of neural cells in our population, we collected only the portion that exhibited high expression of TUJ1 (Figures S4F and S4G), which in neural cells could be as low as ∼50% (Turaç et al., 2013), and observed that the most prominent HOX hits identified in the unsorted library also retained significance in the sorted library (Figure S4H).

Regulation of neuronal differentiation by HOXA6 and HOXB6

To understand the roles of HOXA6 and HOXB6 during neuronal differentiation, we generated mutant haploid hESC lines for these two TFs by CRISPR-Cas9 mutagenesis. By differentiating these lines into neuronal cells along with a control hESC line, we were able to analyze genes that are normally upregulated upon differentiation of control cultures, termed as neuronal marker genes, and the differentially expressed ones among those marker genes in ΔHOXA6 and ΔHOXB6 neuronal cultures. The majority of the differentially expressed neuronal marker genes were downregulated in both ΔHOXA6 and ΔHOXB6 neuronal cultures, suggesting that these TFs positively regulate the differentiation, in accordance with the results of our LOF screen (Figures 4A and 4B). Moreover, around 75% of the significantly downregulated genes in the mutant cultures were found to be neuronal marker genes, reinforcing the essential roles of these TFs during neuronal differentiation (Figure S5A). Downregulated neuronal marker genes in differentiated ΔHOXA6 and ΔHOXB6 cultures had a large overlap of 400 genes, while each mutant also had its own smaller unique set of response genes (Figures 4C and 4D). A prediction analysis on the target binding sites of these TFs suggested that about 29% of the downregulated neuronal marker genes in ΔHOXA6 cultures and about 34% in ΔHOXB6 cultures might potentially be direct targets of these TFs. Among the neuronal marker genes downregulated in both mutants, 79 of them were predicted to be direct targets of both TFs, while smaller groups, namely 40 and 56 genes, were predicted to be direct targets of HOXA6 only or HOXB6 only, respectively (Figure S5B). In order to further study this observation, we performed a cleavage under targets and release using nuclease (CUT&RUN) experiment, using antibodies against HOXA6 and HOXB6 in differentiated wild-type neuronal cells. We identified HOXA6 peaks at the promoters of 714 genes and HOXB6 peaks at the promoters of 444 genes. Accordant with our expression and prediction analyses, 51 of these genes were common for both of the TFs. GO analysis on the genes that have binding sites near their promoter for both HOXA6 and HOXB6 showed a significant enrichment of terms related to neuron projection, somatodendritic compartment, and dendritic tree (Figure S5C). Notably, previous research highlighted HOXC6’s interaction with a Foxp1 enhancer in mice, located in an intron, activating its expression (Lacombe et al., 2013). The FOXP1 gene has an important regulatory role within the comprehensive HOX-dependent program governing spinal motor neuron diversity and connectivity (Dasen et al., 2008). We extend these findings by suggesting that the remaining set of HOX6 paralogs also plays a regulatory role in modulating FOXP1 expression, as both HOXA6 and HOXB6 were found to have binding sites within FOXP1 introns. These observations argue that HOXA6 and HOXB6 may be binding on a group of promoters at the same time, suggesting that this group of common targets may regulate a large set of common indirect targets of these TFs. A GO analysis of all downregulated neuronal marker genes in differentiated ΔHOXA6 or ΔHOXB6 cultures showed a significant enrichment of GO terms related to secreted molecules, cell adhesion, signaling molecules, and disease-causing mutations for both HOX mutants (Figure 4E). The strong enrichment of GO terms associated with cell signaling suggested that both TFs may be regulating neuron-related signaling pathways. Thus, we analyzed the differential expression for the downregulated marker genes that are associated with the plasma membrane and extracellular matrix. This analysis revealed that genes in the neuroactive receptor-ligand interaction pathway were among the top 35 significantly downregulated genes in both differentiated ΔHOXA6 and ΔHOXB6 cultures, suggesting that both TFs regulate this neuronal signaling pathway (Figures 4F and 4G).

Figure 4.

Regulation of neuronal differentiation by HOXA6 and HOXB6

(A and B) The number of up- and downregulated neuronal marker genes in differentiated cultures of ΔHOXA6 (A) and ΔHOXB6 (B) as compared to differentiated control cultures. Marker genes were defined as those that were differentially expressed between the neuroectodermal cells and control neuronal cultures (fold change ≥ 2, n = 3 independent replicates, FDR < 0.05).

(C) Heatmap demonstrating the relative expression levels of neuronal marker genes, which are downregulated in mutant cultures, between neuroectoderm, control, and ΔHOXA6 and ΔHOXB6 neuronal cultures.

(D) Venn diagram showing the overlap and the differences between the downregulated neuronal marker genes in ΔHOXA6 and ΔHOXB6 neuronal cultures.

(E) GO analysis of downregulated marker genes in ΔHOXA6 and ΔHOXB6 neuronal cultures. Shown are the highest-ranking common GO terms between the two mutants.

(F and G) Volcano plots showing the expression ratio of plasma membrane- and extracellular matrix-associated genes between the ΔHOXA6 (F) or ΔHOXB6 (G) and the control neuronal culture. Dashed lines indicate a significance threshold of FDR <0.05 (n = 3 independent replicates). Genes that are labeled in gray are those among the top 35 significantly downregulated genes that are associated with the neuroactive receptor-ligand interaction in both mutants.

(H and I) Relative transcript levels of proximal (H) and distal (I) HOXA cluster genes in neuroectodermal cells (light gray), control (dark gray),ΔHOXA6 (blue), and ΔHOXB6 (red) neuronal cultures. Expression values are normalized to the levels of control neuronal cultures for each gene. Statistically significant changes are labeled with an asterisk (n = 3 independent replicates, FDR <0.05, error bars represent standard error of the mean).

(J) Schematics illustrating the roles of HOXA6 and HOXB6 in neuronal differentiation. Both genes have essential roles during differentiation (left). While they directly or indirectly regulate the expression of similar neuronal signaling pathways, they also have their unique targets among neuronal marker genes, coding, and non-coding genes within the HOX clusters (right). See also Figure S5.

Aside from the signaling-related genes, a group of HOX cluster genes were also differentially expressed in the mutant cultures. To understand to what extent HOXA6 and HOXB6 regulate the expression of HOX cluster genes, we analyzed the expression of HOXA cluster in neuroectoderm and in control and mutant neuronal cells. Interestingly, central paralogs, HOXA3, HOXA4, HOXA5, and HOXA7, which are expressed specifically in neuronal cultures, were significantly downregulated in both ΔHOXA6 and ΔHOXB6 cultures (Figure 4H), suggesting that both TFs have roles in regulating this central area of the HOXA cluster. In contrast, the more distal paralogs, HOXA9, HOXA10, HOXA11, and HOXA13, were upregulated in only ΔHOXA6 cultures (Figure 4I), arguing that HOXA6 has a negative regulatory role on the expression of distal region of HOXA cluster.

Given the roles of long non-coding RNAs (lncRNAs) in differentiation (Pauli et al., 2011), we also assessed whether there were neuronal-specific lncRNA expression signatures and if these were regulated by these two essential HOX cluster genes. A comparison between the neuroectoderm and control neuronal cells revealed that there was indeed a group of lncRNAs that had an enriched expression in neuronal cells. 11 of such neuronal marker lncRNAs were downregulated in both ΔHOXA6 and ΔHOXB6 cultures, suggesting that these essential TFs also regulate the expression of non-coding marker genes for neuronal cells (Figures S5D and S5E). One of these common neuronal marker lncRNAs, HOXB-AS3, is located within the HOXB cluster. In addition to this common target, another neuronal marker lncRNA, HOXA-AS3, was downregulated only in ΔHOXA6 cultures (Figure S5F).

In parallel to the knockout experiments conducted in haploid cells, we extended our investigation to include knockouts in diploid cells using CRISPR-Cas9 mutagenesis. (Figure S5G). Expression analysis of the diploid mutants showed an impact of the mutation on neural differentiation in diploid cells similar to our observations in haploid cells (Figure S5H). Furthermore, cell staining for the neural markers TUJ1 and EGR2 in control and ΔHOXB6 neuronal cells demonstrated a decrease in cells co-expressing both markers, accompanied by an increase in cells solely positive for TUJ1 or negative for both markers in the ΔHOXB6 cell population (Figure S5I). It is important to note that the ΔHOXB6 cells did not lose their neuronal morphology, but only their EGR2 expression (Figure S5J). The consistency of the results across different ploidy states and different cell lines suggests a conserved role of the targeted genes in neural differentiation pathways.

Taken together, we propose a mechanism, in which HOXA6 and HOXB6 regulate neuronal differentiation by regulating the expression of neuronal signaling pathways, neuronal marker genes, and coding and non-coding genes of HOX clusters (Figure 4J). While working together for some of these gene expression changes, they also have their own unique targets, which they can regulate directly or indirectly, and are therefore functionally not redundant as previously believed.

The essentialome of MEGs during neurogenesis

Next, we sought to identify the essentialome of imprinted genes. Imprinted genes are genes that are differentially expressed depending on whether they are inherited from the mother or the father. These genes are thought to be important in a variety of developmental and disease-related processes, and in recent years much attention has been paid to their importance. There is growing evidence suggesting that imprinted genes are involved in a variety of disorders, including cancer, obesity, and diabetes (Jelinic and Shaw, 2007; Mitchell and Pollin, 2010; Weinstein et al., 2010). Imprinted genes are also thought to be important in normal embryogenesis (Surani et al., 1984). The identification of essential imprinted genes is important for providing insights into the etiology of diseases, and it can help to identify new targets for therapeutic intervention.

To establish the essentialome of imprinted genes during early embryogenesis, we have utilized data from genome-wide LOF screens in hESCs (Yilmaz et al., 2018) and their derived germ layers (ectoderm, mesoderm, and endoderm) (Yilmaz et al., 2020), together with the dataset from the current screen in neurogenesis. These screens rely on parthenogenetic cells which only contain maternally originated DNA, thus only expressing MEGs. For each MEG in the haploid hESCs, we determined the CRISPR score and p value (Figure 5A). Our analysis demonstrated that the MEGs are not essential for pluripotency, and there are no essential expressed imprinted genes during differentiation of hESCs into the three embryonic germ layers, judged by the absence of genes with a negative CRISPR score and a p value smaller than 0.05 in both cases (Figure 5A). One gene, KCNK9, was identified with a significant negative CRISPR score in neuroectodermal cells; however its expression levels in hESCs and neuroectoderm were rather low, suggesting that it may be expressed in higher levels in a transient stage during neuroectodermal differentiation. In contrast, during differentiation into neuronal cells, we found seven essential imprinted genes: UBE3A, PHLDA2, SLC22A18, SLC22A2, ANO1, MAGI2, and ZNF597.

Figure 5.

The essentialome of maternally expressed genes during neurogenesis

(A) Volcano plot demonstrating −log₂(p value) and CRISPR scores of maternally expressed genes in loss-of-function screens of early embryonic stages using hESCs, endodermal, mesodermal, neuroectodermal, and neuronal cells. Kolmogorov-Smirnov test was performed to determine the significance of depleted genes.

(B) Principal-component analysis plot of the transcriptome of three embryonic stages during neuronal differentiation (hESCs, neuroectoderm, and neuronal culture) and mutant neuronal cultures of UBE3A, PHLDA2, or SLC22A18 genes, and of both PHLDA2 and SLC22A18 genes (ΔPHLDA2/SLC22A18).

(C) Bar plots showing the expression levels of markers of pluripotency (top), neurogenesis (middle), and five neuronal stage-specific HOX genes (bottom) in control hESC, neuroectodermal, and neuronal cells, and in mutant neuronal cultures of UBE3A, PHLDA2, or SLC22A18 genes, and of both PHLDA2 and SLC22A18 genes (ΔPHLDA2/SLC22A18). Statistically significant changes compared to control neuronal culture are indicated by asterisk (FDR < 0.05, n = 3 independent replicates for all samples except ΔPHLDA2/SLC22A18 where n = 2 independent replicates, error bars represent standard error of the mean).

(D) −log(FDR) values of GO terms enriched within the downregulated genes in mutant compared to control neuronal cultures. Shown are the highest-ranking GO terms in ΔUBE3A. See also Figure S6.

Of the imprinted genes essential for neuronal differentiation, the MEG UBE3A had the lowest CRISPR score (Figure 5A). Loss of expression of UBE3A causes Angelman syndrome (AS), which affects development of the brain, leading to mental impairment, seizures, and behavioral disorders (Kishino et al., 1997). Other essential imprinted genes include PHLDA2 and SLC22A18, a set of genes that reside in the Beckwith-Wiedemann syndrome (BWS) region on chromosome 11, and their expression is controlled by the differentially methylated region (DMR) in the KCNQ1 gene (Fitzpatrick et al., 2002), which controls the uniparental expression of several genes in this region. BWS could arise from mutations in several genes in the genome; most of them are expressed from the maternal allele of the BWS region (Choufani et al., 2010; Cooper et al., 2005; Weksberg et al., 2005). Although its phenotype is mostly non-neuronal, specific cases of BWS that originated from loss of methylation of the KCNQ1-DMR involve brain abnormalities (Gardiner et al., 2012). Moreover, SLC22A18 is associated with promoting the growth of head circumference of newborns (Lambertini et al., 2012). Our expression analysis across pluripotency, neuroectoderm, and neuronal stages showed that UBE3A, PHLDA2, and SLC22A18 are expressed during neuronal differentiation of hESCs (Figure S6A). Thus, these genes were chosen for further research.

To analyze the differentiation phenotypes of these imprinted genes, we generated LOF mutations in hESCs for UBE3A, PHLDA2, or SLC22A18 (ΔUBE3A, ΔPHLDA2, and ΔSLC22A18, respectively) using CRISPR-Cas9 mutagenesis (Figure S6B). Because the maternal expression of PHLDA2 and SLC22A18 is controlled by the same DMR, in cases of BWS that results from the disruption of the DMR on the maternal allele, both of these genes should have lower expression which might lead to a more severe phenotype. To test this, we also generated a double-knockout mutant hESCs line for PHLDA2 and SLC22A18 (ΔPHLDA2/SLC22A18, Figure S6C). Next, we differentiated the mutant cell lines into neuronal cells and analyzed their transcriptome during neurogenesis together with the transcriptome of hESCs and neuroectoderm from a previous study (Yilmaz et al., 2020). The expression profiles of the mutant neuronal cells are clustered away from those of the control neuronal cultures (Figure 5B), with ΔUBE3A cluster being the most segregated and the ΔPHLDA2/SLC22A18 double-knockout cluster positioning very close to the clusters of the single knockouts of PHLDA2 or SLC22A18. Specific markers of pluripotency in the mutant lines remained highly expressed even after the differentiation, most prominently in ΔUBE3A cells, suggesting incomplete exit from pluripotency (Figures 5C [upper panel] and S6D). Accordingly, the mutant neuronal cultures presented low expression of several markers of neurogenesis, such as PAX6, FOXA2, and SHH, whereas other markers of neurogenesis such as WNT4 and WNT5A were similarly expressed in the control and the mutants (Figures 5C [middle panel] and S6E). The expression pattern of HOX genes further validated the aberrant neurogenesis of the mutant lines, as several genes in the HOXB cluster between HOXB3 and HOXB7 showed lower expression in the mutants compared to the control neuronal cells (Figure 5C, lower panel). In order to assess the overall differentiation potential of the mutants, we performed GO analysis on the downregulated genes in mutant lines as compared to the control neuronal cells. In the majority of mutants, the most significantly enriched terms were found to be neurogenesis-related terms (Figure 5D). ΔSLC22A18 neuronal cells did not show significant enrichment of these terms, suggesting a weaker phenotype for these during neurogenesis as compared to neuronal cultures mutated for UBE3A and PHLDA2. In light of the role of UBE3A in the manifestation of AS, we further confirmed its involvement in neurogenesis by employing CRISPR-Cas9 mutagenesis in diploid cells (Figure S6F). Similar to the effect of HOX genes, analyzing the expression in diploid mutant cells reveals a comparable impact of UBE3A knockout on diploid cells as observed in haploid cells (Figure S5G).

UBE3A regulates lineage determination during neurogenesis

UBE3A encodes for the E6-associated protein (E6AP), which was discovered as an E3 ubiquitin ligase (Huibregtse et al., 1993; Scheffner et al., 1993), and is associated with several processes in the cell such as transcription, replication, and cell cycle by creating complexes with other proteins (Martínez-Noël et al., 2018). A TF expression analysis revealed that the ΔUBE3A neuronal cultures failed to upregulate the expression of ∼100 out of ∼500 TFs that were significantly upregulated in the control neuronal cultures as compared to hESCs (Figure 6A). Using the online predicting associated transcription factors from annotated affinities (PASTAA) tool (Roider et al., 2009), we found 30 TFs that were significantly (p value < 0.05) predicted to regulate the expression of the ∼100 downregulated TFs. Among the highest-ranking predicted TFs, three of them, HOXA5, Churchill domain containing 1 (CHURC1), and WT1, are related to neurogenesis and were downregulated in the ΔUBE3A neuronal cultures, suggesting that they contribute to the differentiation phenotype seen in ΔUBE3A cells (Figures 6B and 6C). Whereas WT1 protein is required for the differentiation of several cell types and is not neurogenesis specific (Ambu et al., 2015; Armstrong et al., 1993; Sharma et al., 1992) and HOXA5’s role in neurogenesis is to govern the formation of the anterior-posterior pattern (Carpenter, 2002), CHURC1 has a more general role of regulating cell fate during neurulation (Londin et al., 2007; Sheng et al., 2003). Regulation of the neuronal cell fate by CHURC1 is mediated by its inhibitory effects on the activity of pro-mesendodermal ACTIVIN/NODAL pathway (Figure 6D). During mesendodermal differentiation, the ACTIVIN/NODAL pathway is active, and it activates the expression of mesendodermal genes such as NODAL, LEFTY1, and LEFTY2 (Fei et al., 2010) (Figure 6D). In contrast, this pathway is inhibited in ectoderm differentiation protocols that have been established (Chambers et al., 2009). In vivo, in cells that are destined to become neurons, CHURC1 transcription is induced by fibroblast growth factor (FGF)4/FGF8 and, in turn, upregulates the expression of ZEB2 (Sheng et al., 2003), while inhibiting the continuous expression of FGF4/FGF8 in the tissue as a negative feedback mechanism. ZEB2 interacts with the ACTIVIN/NODAL pathway and inhibits the expression of NODAL, LEFTY1, and LEFTY2 (Chng et al., 2010). In addition, ZEB2 itself binds the promoter of mesoderm-inducing TBXT and inhibits its expression (Postigo et al., 2003) (Figure 6D). Our transcriptome analysis revealed that the expression of FGF4/FGF8 is more than 10-fold higher in ΔUBE3A neuronal cultures as compared to control neuronal cultures, whereas CHURC1 expression remains low (Figures 6C and 6E). The lower expression of CHURC1 is expected to lead to lower expression of ZEB2, which was indeed about 2-fold higher in control neuronal cultures. In addition, we show high expression of the mesendodermal genes NODAL, LEFTY1, LEFTY2, and TBXT, likely to be a consequence of the lower expression of their inhibitor, ZEB2 (Figure 6E). ZEB2 was also suggested to be a main regulator of epithelial-mesenchymal transition (Vandewalle et al., 2005). In accordance with this, GSEA revealed enrichment of this pathway among the downregulated genes in ΔUBE3A neuronal cultures (Figure S6G). Clinical studies commonly attempt to reactivate the suppressed UBE3A in the brain of patients (Margolis et al., 2015), and, from our findings, we may suggest a new strategy of treating AS patients, by directly addressing the molecular pathway of neurogenesis.

Figure 6.

UBE3A regulates lineage determination during neurogenesis

(A) Heatmap demonstrating the relative expression levels of neuronal marker genes, which are downregulated in the mutant culture, between hESCs, control, and ΔUBE3A neuronal cultures (n = 3 independent replicates.

(B) Heatmap showing the predicted genome-wide nuclear factor binding ranks of top three transcription factors that are predicted to regulate the neurogenesis-related nuclear factors that are downregulated in ΔUBE3A neuronal cultures. The statistical significance for the predicted regulation of the top three transcription factors is indicated underneath their gene name.

(C) Bar plot representing the expression of the top three transcription factors suggested to be regulated by UBE3A. Shown are expression levels in undifferentiated hESCs, control, and ΔUBE3A neuronal cultures. Statistically significant changes compared to control neuronal culture are indicated by asterisk (FDR < 0.05, n = 3 independent replicates, error bars represent standard error of the mean).

(D) Putative schematics showing the role of CHURC1 in lineage determination.

(E) Bar plot demonstrating expression levels of the genes in CHURC1 pathway in control and ΔUBE3A neuronal cultures. Statistically significant changes are indicated by asterisk (FDR < 0.05, n = 3 independent replicates, error bars represent standard error of the mean). See also Figure S6.

Discussion

We aimed at identifying and analyzing essential genes for the differentiation into caudal neurogenesis in humans. In the past, we defined the essentialome of undifferentiated hESCs and their three embryonic germ layer progenies, and in the current study we focused, for the first time, on neurogenesis. This study paves the way to define the essential genes for differentiation of other somatic cells. Although genome-wide CRISPR screens have been performed on early human neurons (Liu et al., 2018; Tian et al., 2019, 2021), these studies used dCas9 to test gain-of-function or knockdown phenotypes of targeted genes and thus did not achieve complete LOF mutants. Here we present the first genome-wide LOF CRISPR-Cas9 screen on caudal neuronal cultures derived from hESCs. In our analysis, we observed an increased essentiality of extracellular matrix- and membrane-associated genes during neuronal differentiation, compared to the essentialome of hESCs. We also identified essential TFs from multiple TF families, such as the HOX and FOX.

The use of haploid hESCs for high-throughput genetic screens provides unique advantages owing to the presence of a single allele for each gene in their genome. We previously estimated around 50% higher efficiency in generating LOF mutations in haploid cells upon site-directed mutagenesis (Yilmaz et al., 2018). This prediction was supported by our analysis of a dataset generated by a similar genome-wide screening methodology in a near-haploid cancer cell line. We showed that the only diploid chromosome in this cancer cell line performed more poorly in the essentiality scores of its genes as compared to all other haploid chromosomes of the same cell line. In addition, our previous pioneering genome-wide screen in haploid hESCs was followed by other genome-wide screens performed in diploid hESCs. When our dataset was re-analyzed by Mair et al. in comparison to their essentialome data from diploid hESCs, they showed that nearly 50% higher number of essential genes were detected in our haploid screen (Mair et al., 2019). Moreover, haploid hESCs exhibit comparable behavior to diploid hESCs in terms of transcriptome and differentiation capacity, differing only in subtle aspects, primarily related to X inactivation (Sagi et al., 2016a). Moreover, haploid hESCs undergo spontaneous diploidization event (Sagi et al., 2016a), meaning that some percentages of the cells are even more similar to diploid cells, with homozygous mutations (see the experimental procedures section). Even though genetic screens of such magnitude are not able to identify all essential genes and for some genes might show only “relative” essentiality, due to the inherent noise of such complex experiments, these previous observations demonstrate the distinctive utility and advantage of haploid hESCs and a higher level of detection of essential genes when using these unique cellular models. The differentiation of our previously established haploid library into neuronal cultures revealed essential roles of HOX genes for proper embryonic neuronal development (of more posterior neuronal cells). The majority of HOX genes are expressed in the central nervous system where they have important roles in neuronal specification and target connectivity in mice (Philippidou and Dasen, 2013), yet they do not necessarily actively aid in all types of neuronal differentiation. Because HOX genes have been duplicated throughout mammalian evolution and exist in humans in four different clusters, it was suggested that the entire effect of mutations in a single gene may be masked by functional redundancy of other paralogs (Quinonez and Innis, 2014). However, in our analysis, several HOX genes, mainly HOX4-6, exhibit essentiality upon differentiation into caudal neurogenesis. By analyzing single knockout hESCs of HOXA6 and HOXB6, we showed here that these paralogs regulate the expression of several neuronal markers. In this context, we propose a synergistic effect involving these paralogs, augmenting the recognized synergistic interactions observed in various other HOX paralogous pairs, such as HOXA9 and HOXB9 (Chen and Capecchi, 1997), HOXA9 and HOXD9 (Fromental-Ramain et al., 1996), and HOXA3 and HOXD3 (Condie and Capecchi, 1994).

Except for HOXD4, which was not found to be essential in our screen, there are no human phenotypes known to be associated with the HOX4-6 genes. However, mouse knockout models of these genes demonstrated nervous system phenotypes (Quinonez and Innis, 2014), supporting our observations and establishing hESC-derived neuronal cultures as effective models to study the function of the HOX family members during human neurogenesis.

In this study, we were able to further expand the repertoire of neurological disorders that show a phenotype in early embryonic development, including early- and late-onset neurodegenerative disorders. Due to their progressive and age-related nature, modeling of neurodegenerative disorders in culture poses a major challenge. As of today, most models are based on in vivo systems or complex in vitro systems, such as two-dimensional cultures, or three-dimensional organoids, comprising induced pluripotent stem cells that are differentiated into neuronal and glial cells (Bhargava et al., 2022). Establishing systems that can examine the stage of differentiation of hESCs, at which we can expect a phenotype of the mutations in genes related to neurodegenerative disorders, can advance the field significantly. In turn, establishing such models for neurodegeneration can expedite drug development.

Our data show that in several rather rare cases, the effect of neurodegeneration-causing mutations is evident already in undifferentiated hESCs. At this early pluripotency stage, these are mostly mutations in genes associated with developmental neurodegenerative conditions, but, with the progression of the differentiation process, more gene mutations begin to affect cell growth, including mutations in genes that cause early/late-onset conditions. In neurogenesis, the phenotype of a considerable number of developmental neurodegenerative conditions and a few of early/late-onset mutations is already apparent. This observation suggests that for several neurodegenerative conditions a rather easily accessible disease model can be established by differentiating hESCs into embryonic neuronal cells. As an example, we focused on parkinsonism-related genes and looked at the effect of gene perturbation on the growth of hESCs and their differentiation into neuroectoderm and neuronal cells. Our analysis within this group of genes showed a gradual increase in the number of mutated genes with an essentiality phenotype along the differentiation process, suggesting that differentiated hESCs, even with differentiation process encompassing a range of neural lineages, can serve as a model to study the etiology of even late-onset neurodegenerative conditions and be utilized for drug screening.

hESC-derived neuronal cultures also allowed us to explore the role of different imprinted genes during neuronal differentiation. It is well established that the entire set of imprinted genes is essential for normal embryonic development (Surani et al., 1984) and that the MEGs are more essential for neurogenesis than paternally expressed genes (Sagi et al., 2019). We could demonstrate that the MEGs are not essential for undifferentiated hESCs, modeling the post-implantation stage embryos, and for the differentiation into the three embryonic germ layers, modeling gastrulation stage embryos. However, they start to show essential features during neurulation.

UBE3A was found to have an essential role in perinatal brain development, and mouse models have demonstrated that loss of UBE3A actually affects many areas of the brain, leading to increased neuronal excitability and a loss of synaptic spines, along with changes in a number of distinct behaviors (Rotaru et al., 2020), without defining the exact stage of its essentiality. Using multiple CRISPR-Cas9 screens across different stages of early human development, we were able to show that UBE3A is essential during early neurogenesis, and our observations suggest a possible mechanism in which UBE3A positively regulates the activity of the FGF pathway to induce CHURC1 expression and drives the cells into the neuronal fate instead of the mesendodermal one.

The most recognized symptom of loss of UBE3A in AS is paroxysms of laughter, leading to the alternative naming of the disorder as the “happy puppet” syndrome (Angelman, 1965; Bower and Jeavons, 1967). There are multiple Angelman-like syndromes that present a similar phenotype to that of AS without the molecular defect in UBE3A expression. One of those Angelman-like syndromes, Mowat-Wilson syndrome, was shown to share several symptoms with AS, including the happy behavior phenotype, and is sometimes clinically mislabeled as AS (Luk, 2016; Mowat et al., 2003). The molecular basis of Mowat-Wilson syndrome is heterozygotic mutations in ZEB2 that leads to haploinsufficiency. Besides Mowat-Wilson-associated ZEB2, genes linked to several other Angelman-like syndromes showed lower expression in our ΔUBE3A neuronal cells, suggesting a molecular link between different disorders sharing similar phenotypes (Figure S6H).

In contrast to AS, which has a strong neuronal phenotype, BWS phenotype involves the central nervous system only in rare cases. It was reported that BWS patients with loss of methylation of the KCNQ1-DMR or uniparental disomy could also be diagnosed with autism spectrum disorder, but the molecular connection of these two disorders was not studied (Kent et al., 2008). We suggest that reduced expression of PHLDA2 and SLC22A18, which are both regulated by the KCNQ1-DMR, may cause the improper neurogenesis phenotype.

In previous reports, we suggested that by using hESC-based LOF screens we can build a hierarchy of essentiality (Yilmaz and Benvenisty, 2019). In this study, we found several HOX genes to be essential, including HOXB6. In addition, we showed that UBE3A, PHLDA2, and SLC22A18 regulate the expression of HOXB3-6 genes, suggesting another example of hierarchy of essentiality.

Our system reflects the essentialome of caudal neurogenesis. The differentiation protocol ended with neural/neuronal cells of mostly hindbrain cells of the beginning of the second month of human embryonic life. Our data show that our differentiation, although heterogeneous, represent almost exclusively these type of neural/neuronal cells, as other lineages show extremely low percentages. Previous analysis demonstrated the essentialome of neuroectoderm at the second week of the embryo, and thus we can now learn more of the later processes of human embryonic brain development. Although our assay was not robust enough to conclusively state the essential genes for each neural/neuronal cell at this stage, we can learn about caudal neurogenesis and the involvement of HOX and imprinting genes during this stage of development, in addition to the ability to model neuronal disorders during this period of embryo development.

In conclusion, we present an effective tool for studying the regulatory network of neurogenesis in humans by using an hESC-based model. The use of such LOF screens in different embryonic stages and somatic cells will lead to mapping of the essential gene networks for human development.

Experimental procedures

Cell lines and maintenance

The following cell lines were used in this work: haploid female pES10 hESCs, h-pES10-based mutant library previously established by us (Yilmaz et al., 2018), diploid pES10 cells, CSES2 cells, and female 293T cells, obtained from R. Weinberg (Whitehead Institute). All cells were Mycoplasma free. hESCs were used under the Israeli guidelines concerning hESC research, and FAIR (Findable, Accessible, Interoperable, and Reusable) and CARE (Collective benefit, Authority to control, Responsibility, Ethics) principles were followed. pES10 cells were verified for euploidy by G-banding karyotype at the beginning of the experiments. All cell lines were thawed and cultured as previously described (Yilmaz et al., 2018). pES10 were enriched for haploid cells as previously described (Yilmaz et al., 2018). For further information see supplemental experimental procedures.

Differentiation into neuronal cultures

For neuronal differentiation, hESCs were grown on mouse embryonic fibroblasts combined with an hESC medium until 50%–60% confluency. Cells were washed with PBS and gently dissociated using 0.25% trypsin solution A (Biological Industries) for approximately 5 min at 37°C. Dissociated colonies were collected and transferred to non-adherent 10 cm dishes and suspended in confluency. The following day, the hESCs medium was replaced with embryoid body (EB) medium, which contains same ingredients as the standard hESC medium without basic fibroblast growth factor (bFGF), for 4 days. After 4 days the EB medium was supplemented with 0.1 μM retinoic acid (Sigma). Cells were grown in suspension for 17 more days before being plated on 10 cm Matrigel-coated dishes for up to 10 more days before they were collected for downstream analysis. In this study, we utilized the mutant population of hESCs, in which all sgRNAs were represented. About 60% of the cells were still haploid when the differentiation started (Figure S6I). In our mutant library, the diploidized cells are homozygote for the mutation. For further information see supplemental experimental procedures.

Generation of knockout cell lines

sgRNA sequences for neuronal validation are summarized in Table S2. HOX genes- and UBE3A-targeting sgRNAs were cloned into the lentiCRISPR v2 lentiviral vector (a gift from Feng Zhang, Addgene cat. no. 52961). PHLDA2- and SLC22A18-targeting sgRNA were cloned into lentiCRISPRv2-neo and lentiCRISPRv2-hygro, accordingly (gifts from Brett Stringer, Addgene cat. no. 98292 and 98291). An independent control cell line was generated: lentiCRISPRv2 vector without any sgRNA. Lentiviruses production, cell transduction, and antibiotic selection were performed as previously described (Yilmaz et al., 2020).

Surviving transduced cells of the HOX sgRNAs were expanded and used for the differentiation assays. Surviving transduced cells of the imprinted genes sgRNAs were further sub-cloned to isolate isogenic mutant cells. For further information see supplemental experimental procedures.

DNA and RNA extraction and sequencing

Genomic DNA and total RNA were extracted and sequenced as previously described (Yilmaz et al., 2018). For the imprinted genes' knockouts, genomic DNA of knockout cell lines was amplified using region-specific primers summarized in Table S2 and sequenced using the 3730xl DNA analyzer with ABI’s Data collection and Sequence Analysis software. For further information see supplemental experimental procedures.

Data analysis

Reads containing sgRNA sequences were obtained by aligning sgRNA sequences to DNA sequencing reads using Bowtie 2 aligner (Langmead and Salzberg, 2012), as previously mentioned (Yilmaz et al., 2018). CRISPR score was calculated as previously described (Yilmaz et al., 2020), with NPCs as the “undifferentiated control” and the 4 neuronal replicates as the “differentiated samples.” Transcriptome analysis was performed as previously described (Yilmaz et al., 2020), generating TPM (transcripts per million) and CPM (counts per million) values. Values of biological repeats were averaged. A gene was considered expressed if its TPM level was higher than 1.5. GO analysis was performed using the GSEA tool (Mootha et al., 2003; Subramanian et al., 2005). Enriched GO terms were considered significant if their FDR score was lower than 0.05. For further information see supplemental experimental procedures.

Analysis of cellular compartments

Data on cellular localization were retrieved from the Subcellular Localization Database website http://compartments.jensenlab.org/About) from a previous study (Binder et al., 2014; Yilmaz et al., 2018). For this analysis, we utilized a previously published dataset screening essential genes in haploid hESCs (Yilmaz et al., 2018). All data were generated using the same analysis pipeline to make datasets more comparable and to minimize the technical bias. We analyzed genes that are essential in each screen (neuroectoderm and neuronal essentiality screen) and are associated with a single cellular compartment.

Analysis of developmental neurological disorders

To compare between the results obtained in the different screens, the CRISPR scores in each screen were re-calculated and compared to the same initial time point—1 day after infection with the CRISPR-CAS9 library (Yilmaz et al., 2018). We set a score by ranking all genes in each screen using the formula CSx(−log(p value)), which takes into account the log fold change, which indicates the size of the effect, as well as the significance level of each gene. A score lower than −10 was considered as low rank in our analyses.

Lists of neurodegeneration-related genes, divided by age of onset, were obtained from the OMIM database (https://www.omim.org/) (Amberger et al., 2015), using the following keywords: “‘neurodegeneration’ AND ‘developmental’,” “‘neurodegeneration’ AND ‘early-onset’,” and “‘neurodegeneration’ AND ‘late-onset’.” In case a certain gene was included in several age-of-onset categories, we considered only the earliest one.

Genes related to all other neurological conditions were also retrieved from the OMIM database. In cases where a gene was causing more than one neurological condition, it was independently analyzed for every relevant disease category.

Analysis of downstream targets of HOXA6 and HOXB6

Genomic sequences of promoter regions for neuronal marker genes downregulated in ΔHOXA6 and ΔHOXB6 neuronal cells were retrieved from the UCSC Genome Browser (http://genome.ucsc.edu) (Kent et al., 2002). A promoter was defined as a region spanning 1,000 bp upstream and 100 bp downstream of a transcription start site for the gene of interest. The binding motifs of HOXA6 and HOXB6 were retrieved from the JASPAR database (Castro-Mondragon et al., 2022). To search for the predicted binding sites for these TFs in the promoters of the downregulated genes, an online prediction tool, Find Individual Motif Occurrences (FIMO), was used together with the promoter sequences and the binding motif matrices (Grant et al., 2011). Motif occurrences with lower than the suggested p value threshold of 1e⁻⁴ were reported.

Analysis of upstream regulatory TF

Data regarding the list of TFs were retrieved from Ravasi et al., 2010. To predict the upstream regulatory TFs, neuronal-related TFs that are downregulated in ΔUBE3A neuronal cells were analyzed by the online predicting PASTAA tool (Roider et al., 2009). Predicted upstream regulatory TFs were chosen, if they were upregulated in neuronal cells, they were downregulated in ΔUBE3A neural cells, and their p value was lower than 0.05.

scRNA-seq, immunostaining, flow cytometry, and CUT&RUN

See supplemental experimental procedures.

Ethical statement

All experiments were performed according to the ethical guidelines of the Hebrew University.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Nissim Benvenisty (nissimb@mail.huji.ac.il).

Materials availability

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

Data and code availability

The CRISPR-Cas9 library sequencing data presented in this article are available in ArrayExpress (accession number: E-MTAB-11944). The RNA sequencing (RNA-seq) data of the library and individual mutants are available in E-MTAB-11943, E-MTAB-12504, and E-MTAB-13611. scRNA-seq data are available in E-MTAB-13620. CUT&RUN data are available in E-MTAB-13627. Previously published sequencing data that were re-analyzed here are available under accession codes GSE111309, GSE143369, GSE143373, and E-MTAB-4840. All other data that support the findings of this study are available within the article and its supplementary material.

Acknowledgments

We thank Sara Isaac, Elyad Lezmi, and Mordecai Peretz for their assistance with data analyses and all members of the “Azrieli Center for Stem Cells and Genetic Research” for their input and critical reading of the manuscript. We thank Abed Nasereddin and Idit Shiff from The Genomics Applications Lab of The Core Research Facility of The Hebrew University of Jerusalem for their assistance with the 10× Genomics platform. This work was partially supported by The Azrieli Foundation (N.B.), the Rosetrees Trust (N.B.), the US-Israel Binational Science Foundation (2021278) (N.B.), the Israel Science Foundation (2054/22) (N.B.), the Israel Ministry of Science (0004272) (E.M.), the ISF-Israel Precision Medicine Partnership (IPMP) Program (3605/21) (E.M. and N.B.), and the HEAL project, funded by the European Union, EU Horizon (101056712) (N.B.). Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or EU Horizon. Neither the European Union nor EU Horizon can be held responsible for them. E.M. is the Arthur Gutterman Chair in Stem Cell Research. N.B. is the Herbert Cohn Chair in Cancer Research.

Author contributions

S.K., A.B.-T., A.Y., and N.B. designed the experiments and interpreted the data. S.K., A.B.-T., and A.Y. performed the experiments with the assistance of P.S.L.L., T.G.-L., O.Y., E.M., O.R., and D.E. S.K., R.V.-B., A.Y., and N.B. wrote the manuscript with input from all authors. S.K., A.B.-T., R.V.-B., R.S., and A.Y. analyzed the data. A.Y. and N.B. supervised the study.

Declaration of interests

The authors declare no competing interests.

Published: October 31, 2024