The Epigenetic Reader PHF23 Is Required for Embryonic Neurogenesis

Abstract

Epigenetic mechanisms are crucial in the tightly regulated process of neurogenesis from radial glial cells (RGCs) to intermediate progenitor cells (IPCs) to neurons during embryonic brain development. Plant homeodomain (PHD) finger proteins as important epigenetic readers are implicated in development and diseases, yet their roles in embryonic neurogenesis remain largely unexplored. In this study, we found different PHD finger proteins are differentially expressed along the neurogenesis trajectory. Among them, we investigated the function of PHF23 using mouse models, which is highly expressed in RGCs and IPCs, but not in neurons. Our findings demonstrate that PHF23 is essential for proper neurogenesis, and Phf23 knock-out (Phf23-KO) results in cortical developmental defects due to differentiation blockade of RGCs. Mechanistically, PHF23 bind with HDAC2, inhibiting its deacetylation activity on the active histone mark H3K27ac, thereby promoting the expression of neuronal differentiation pathway genes such as Tcf4 and Eya1. Overexpression of Tcf4 rescues the differentiation defects of Phf23-KO NSCs. These results establish PHF23 as a pivotal regulator of neurogenesis, indicating cell type-specific functions of PHD finger proteins.

Significance Statement

Epigenetic regulation of neurogenesis is orchestrated by writers, readers, and erasers of histone codes; however, the roles of a class of epigenetic writers, PHD finger proteins, remain largely unexplored. We have systematically explored their expression patterns using single-cell datasets and revealed the essential role and underlying molecular mechanism of PHF23 in regulating the differentiation of embryonic neural stem cells, using genetic models, RNA-seq and CUT&TAG, and functional assays, shedding light on cell type-specific functions of PHD figure proteins.

Introduction

During mammalian neural development, the vast majority of cells in the brain, including neurons and glial cells, originate from embryonic neural stem cells (NSCs; Gotz and Huttner, 2005; Kriegstein and Alvarez-Buylla, 2009). In mice, neurogenesis begins at approximately embryonic day 9.5–10.5 (E9.5–E10.5), when neuroepithelial cells transform into radial glial cells (RGCs) in the ventricular zone (VZ; Gotz and Huttner, 2005; Kriegstein and Alvarez-Buylla, 2009). RGCs as embryonic NSCs undergo rapid and asymmetric division, generating neurons directly or indirectly through intermediate progenitor cells (IPCs; Noctor et al., 2004; Gotz and Huttner, 2005; Kriegstein and Alvarez-Buylla, 2009; X. Wang et al., 2009b). IPCs migrate out of the VZ along the radial scaffold of RGCs into the subventricular zone (SVZ) and intermediate zone (IZ), giving rise to differentiated neurons units in cortical layers (Noctor et al., 2001; Gotz and Huttner, 2005; Kriegstein and Alvarez-Buylla, 2009). This process is tightly regulated to ensure proper brain development and function (Gotz and Huttner, 2005; Kriegstein and Alvarez-Buylla, 2009).

Epigenetic mechanisms such as histone modifications, DNA methylation, and noncoding RNAs play crucial roles in neurogenesis, controlling spatiotemporal expression of key genes in the proliferation, fate determination, and differentiation of RGCs and their progeny (Yao et al., 2016). Plant homeodomain (PHD) finger proteins are a family of proteins that act as epigenetic readers, which recognize and bind to specific histone modifications to regulate gene expression (Sanchez and Zhou, 2011). Genetic aberrations in the PHD finger regions have been associated with neurological diseases, immunological disorders, and cancer (Baker et al., 2008). It has been shown that PHF21B promotes cell cycle exit during neurogenesis through epigenetic silencing of cell cycle genes in RGCs (Basu et al., 2020), while PHF6 is required for neuronal migration (Zhang et al., 2013). However, whether and how other PHD finger proteins are implicated in embryonic neurogenesis are yet to be determined.

PHF23 is a member of the PHD protein family which specifically recognize trimethylated lysine 4 of histone H3 (H3K4me3; G. G. Wang et al., 2009a). Its fusion with NUP98 through chromosomal translocation promotes human acute myeloid leukemia (G. G. Wang et al., 2009a; Gough et al., 2014). Recently, it has been shown that PHF23 can bind with the SIN3-HDAC1/2 forming a PSH protein complex to repress the deacetylation activity of HDAC1/2 on acetylated lysine 27 of histone H3 (H3K27ac), thereby activating downstream genes of B cell differentiation and suppressing B cell lymphoma (Chen et al., 2021). Nevertheless, the role of PHF23 outside the hematopoietic system remains largely unexplored.

In this study, integrating mouse model, transcriptional/epigenetic profiling, and functional assays, we show that PHF23 is a pivotal regulator of embryonic neurogenesis, which is required for RGC differentiation through epigenetic regulation of neural differentiation pathway genes.

Materials and Methods

Ethics

The mice were bred and maintained in the Experimental Animal Center of the State Key Laboratory of Biotherapy, Sichuan University. All animal studies complied with relevant ethical regulations and were approved by the Animal Care and Use Committee of Sichuan University.

Mice

Phf23^+/− heterozygous (Phf23-HET) mice were previously established by Dr. Chong Chen lab through genome editing (Chen et al., 2021). Time breeding was setup to generate Phf23^−/− (Phf23-KO), Phf23^+/−(Phf23-HET), and Phf23^+/+ (Phf23-WT) embryos, collected at E11.5 or E13.5. For staging of embryos, the day when vaginal plug was identified was considered as E0.5 (Deb et al., 2006).

Tissue processing

Tissue processing follows protocols in a previous study (Wang et al., 2021). Briefly, pregnant mice were anesthetized and the embryos were removed and fixed in 4% paraformaldehyde (PFA) at 4°C overnight. Embryos were dehydrated in a paraffin dehydrator, embedded in paraffin, and sectioned at 5 μm. For genotyping, tail DNA was extracted and amplified by PCR.

Hematoxylin & eosin staining

Paraffin sections were dried on a roaster at 37°C for 20 min and successively placed in xylene for dewaxing, ethanol for hydration, deionized water for rehydration, hematoxylin (Sigma) for nuclear staining, 0.5% ethanol hydrochloride for clearing, 0.1% sodium bicarbonate for bluing, eosin (Sigma) for cytoplasmic staining, ethanol for dehydration, and xylene for clearing. Stained slides were coverslipped and imaged under an Olympus BX51 microscope.

Immunofluorescence

Immunofluorescence staining on tissues follows protocols in a previous study (Wang et al., 2021). Paraffin sections were dried on a roaster at 37°C for 20 min and successively placed in xylene for dewaxing, ethanol for hydration, and deionized water for rehydration. Sections were permeabilized in 0.3% Triton X-100 for 20 min, blocked with 2% goat serum for 1 h at room temperature, and incubated with primary antibody overnight at 4°C. On the following day, sections were rinsed and incubated with secondary antibody for 1 h at room temperature and incubated with DAPI (1 μg/ml) for 5 min. Slides were coverslipped and imaged under an Olympus BX51 microscope.

For immunofluorescence staining on differentiated NSCs, cells grown on slides were fixed in PFA for 15 min, permeabilized in 0.2% Triton X-100 and 0.1% Tween-20 for 10 min, blocked with 10% goat serum for 1 h at room temperature, and incubated with primary antibody overnight at 4°C, followed by secondary antibody staining for 1 h at room temperature, and DAPI (1 μg/ml) staining for 5 min. Slides were imaged under an Olympus BX51 microscope.

The primary antibodies used in this study include Sox2 (1:2,000, rabbit, Abcam, ab92494), Tbr2 (1:500, rabbit, Abcam, ab23345), Ctip2 (1:500, rat, Abcam, ab18465), Ki67 (1:500, mouse, BD biosciences, 550609), BrdU (1:500, rat, Abcam, ab6326), Cleaved Caspase-3 (1:500, rabbit, Cell Signaling Technology, 9664S), Map2 (1:5,000, rabbit, Proteintech, 17490-1-AP), GFAP (1:5,000, chicken, Abcam, ab4674), and Ng2 (1:1,000, rabbit, Millipore, AB5320).

The secondary antibodies include goat anti-rabbit IgG Alexa Fluor 488 (1:1,000, Invitrogen, A-11008), goat anti-mouse IgG Alexa Fluor 555 (1:1,000, Invitrogen, A-21422), goat anti-rat IgG Alexa Fluor 555 (1:1,000, Invitrogen, A-21434), and goat anti-chicken IgG Alexa Fluor 647 (1:1,000, Invitrogen, A-21449).

Brdu labeling and detection

BrdU labeling and detection follows a previous study (Hu et al., 2017). BrdU (50 mg/kg) was injected intraperitoneally into pregnant mice at E12.5 to label S-phase cells. Embryos were collected at E13.5, fixed in 4% PFA, and sectioned. For BrdU staining, 1 M HCl was used for DNA hydrolysis and neutralized by 0.1 M sodium borate. The rest of the steps are similar to standard immunofluorescence staining

Lentiviral packaging and viral infection

Lentivirus packaging follows the protocol of a previous study (Wang et al., 2021). Briefly, lentivirus was produced in 293T cells using the calcium phosphate method and concentrated using Lenti-X Concentrator (Takara). During the passage of NSCs, the concentrated lentivirus was added to the culture medium for 12 h. The NSCs were cultured in fresh media for 48 h and then screened with puromycin (Invitrogen) to establish stable cell lines.

Neurosphere assay

For the primary neurosphere assay, cortical tissues from E13.5 VZ/SVZ were isolated and dissociated. Cells were placed in 96-well plates with 2,000 cells per well and cultured in the DMEM/F12 medium containing 1× B-27 supplement, Invitrogen; 1× N-2 supplement, Invitrogen; 20 ng/ml Human bFGF recombinant protein, Invitrogen; and 20 ng/ml Mouse EGF Recombinant Protein. For the secondary neurosphere assay, primary neurospheres were dissociated enzymatically into single cells. The NSCs were placed in 96-well plates with 2,000 cells per well and cultured in the same condition as the primary neurosphere assay. The number and the diameter of neurospheres are larger than 50 μm in multiple wells.

NSC differentiation assay

NSC differentiation assay follows the protocol of a previous study (Chojnacki and Weiss, 2008). Briefly, the neurospheres were dissociated enzymatically into single cells and adhesively grown on poly-d-lysine-coated (Invitrogen) and laminin-coated (Invitrogen) glass slides. The differentiation medium is the DMEM/F12 (Invitrogen) containing B27 (Invitrogen) and N2 (Invitrogen) without bFGF or EGF. The medium was half-changed every 2 d and immunofluorescence staining was performed 10 d later. The cell culture slides were imaged under an Olympus BX51 microscope.

Western blot and Co-IP

Western blot analysis follows the protocol of a previous study (Wang et al., 2021). The NSCs were lysed with RIPA buffer (Sigma) with 1 mM PMSF (Sigma) and protein concentrations were determined by BCA method. The proteins were separated by SDS-PAGE, transferred to PVDF membranes (Millipore), sealed with 5% skim milk for 1 h at room temperature, incubated with primary antibody overnight at 4°C, and incubated with secondary antibody for 1 h at room temperature. The signals were detected by ECL or ECL Plus (GE HealthCare).

For Co-IP experiments, NSCs were lysed with RIPA buffer (Beyotime) with 1 mM PMSF and were rotationally incubated with primary antibody overnight at 4°C. The primary antibody magnetic beads were added to rotationally incubate for 1 h at room temperature, and the protein complex bound to the primary antibody was separated by a magnetic rack.

The primary antibodies used include PHF23 (1:2,000, rabbit, Proteintech, 25862-1-AP), FLAG (1:2,000, rabbit, Cell Signaling Technology, 14793S), HDAC2 (1:1,000, rabbit, Proteintech, 12922-3-AP), GAPDH (1:5,000, rabbit, Huabio, ET1601-4), Histone H3 (1:2,000, rabbit, Abcam, ab201456). The second antibody was anti-rabbit IgG (1:5,000, SeraCare, 5220–0336 or 1:1,000, Lablead, RN0100).

Construction of plasmids

To prepare overexpression lentiviral plasmids (EF1a-mCherry) targeting Tcf4 and Eya1, deoxy-oligonucleotides containing targeting sequences of the genes were amplified from the cDNA of neurospheres. The amplified products were inserted between AgeI and BsrGI restriction sites in a lentiviral plasmid pLentiCRISPRV2 (Addgene), replacing the mCherry sequences by homologous recombination. All plasmids were confirmed by Sanger sequencing (Sangon Biotech).

RT-PCR analysis

The total RNA of neurospheres were isolated with RNA Purification Kit (EZBioscience). RNA for each sample was reverse transcribed into cDNA by ExonScript RT SuperMix with dsDNase (exongen) and prepared in iTaq Universal SYBR Green Supermix (Bio-Rad). Reactions were run on Bio-Rad CFX96 Touch Real-Time PCR Detection System. Primers used for qPCR are listed below:

Tcf4-cDNA-F: 5′ GCTCTAGCAGCGGAATGAA 3′

Tcf4-cDNA-R: 5′ AGCTAGAGGACTGAGGGATATG 3′

Eya1-cDNA-F: 5′ CAGTTTCTCAGGTTCAGCTCTC 3′

Eya1-cDNA-R: 5′ TGTGGGTATGATTTGGAAGGATAA 3′

FLAG-cDNA-F: 5′ ATTGCCAGAAATGCAAGGAAC 3′

FLAG-cDNA-R: 5′ CCGTCGTGGTCTTTGTAGTC 3′

beta-Actin-cDNA-F: 5′ GGCTGTATTCCCCTCCATCG 3′

beta-Actin-cDNA-R: 5′ CCAGTTGGTAACAATGCCATGT 3′

Bulk RNA-seq analysis

Bulk RNA-seq analysis follows the protocol of a previous study (Qiao et al., 2018). Primary mouse neurospheres were lysed with TRIzol (Thermo Fisher Scientific) and snap-frozen in liquid nitrogen after repeated blowing with pipettes until no clumps of cells could be seen. Sequencing libraries were generated using NEBNext Ultra RNA Library Prep Kit for Illumina (NEB, catalog #E7530L), purified with AMPure XP system. Library quality was assessed by Agilent 5400 (Agilent). Qualified libraries were pooled and sequenced on Illumina platforms by Novogene. Raw FASTQ files were processed with fastp (v0.23.4; Chen et al., 2018) to remove adaptor sequences, and the trimmed files were aligned to the mouse reference genome mm10 using STAR (v2.7.1a; Dobin et al., 2013). Data normalization and differential expression analysis were performed with DESeq2 (v1.38.3; Love et al., 2014). Genes with Benjamini–Hochberg-adjusted p value < 0.05 and an absolute log2(fold change) >0.5 were counted as differentially expressed genes. Gene ontology (GO) enrichment analysis was performed using the R package clusterProfiler (v4.6.2; Yu et al., 2012).

CUT&Tag-seq analysis

CUT&Tag-seq analysis follows the protocol of a previous study (Jin et al., 2021). CUT&Tag libraries were generated using NovoNGS CUT&Tag3.0 High Sensitivity Kit. The neurospheres were dissociated enzymatically into single cells, then treated with ConA beads, incubated with H3K27ac antibody (1:50, rabbit, Abcam, ab4729) overnight at 4°C, incubated with secondary antibody for 1 h at room temperature, and incubated with pAG-Tn5 for 1 h. Fragment DNA at 37°C for 1 h and stop it with 10% SDS at 55°C for 10 min. Extract DNA fragments and amplify libraries. The libraries were quality checked by Agilent 5400 and then subjected to paired-end sequencing on the Illumina NovaSeq 6000. CUT&Tag data processing and analysis were performed according to a previously described protocol (Kaya-Okur et al., 2019; Na et al., 2022). Reads were aligned to the mouse mm10 reference genome using Bowtie2 (v1.2.2; Langmead and Salzberg, 2012) with the parameters: –local, –very-sensitive, –no-mixed, –no-discordant, –phred33, -I 10, and -X 700. PCR duplicates were removed using the MarkDuplicates from GATK (v4.1.3) using the option -REMOVE_DUPLICATES = true. H3K27ac peaks were called using SEACR (v1.3; Meers et al., 2019), and peak annotation was performed with R package ChIPseeker (v1.32.1; Yu et al., 2015). Differential peaks analysis was performed by DESeq2 (v1.38.3; Love et al., 2014). GO enrichment analysis for genes associated with differential H3K27ac peaks was performed using clusterProfiler (v4.6.2; Yu et al., 2012). Normalized bigwig files were generated with bamCoverage from deepTools (v3.5.3; Ramirez et al., 2016) with bins per million (BPM) normalization. Heatmaps and metaplots were generated with computeMatrix and plotHeatmap from deepTools (v3.5.3), and sashimi plots showing sequencing coverage over gene regions were created using trackplot (v0.3.6; Zhang et al., 2023). Venn diagrams were generated using R package ggvenn (v0.1.10).

Mouse embryonic scRNA-seq reanalysis

The single-cell expression matrix for the E14.5 mouse embryonic brain was obtained from previously published data (GSE152281; Dennis et al., 2024). Low-quality cells were strictly filtered following the approach described by Borrett et al., and Seurat (v5.0.1; Hao et al., 2024) was used for downstream analyses, including data normalization (NormalizeData), feature scaling (ScaleData), identification of variable genes (FindVariableFeatures), and dimensional reduction analysis (RunPCA and RunUMAP). Unsupervised clustering was performed using the original Louvain algorithm (FindClusters), and cell types were identified based on the signature genes provided in the original study.

Mouse embryonic snATAC-seq reanalysis

The BAM files and filtered peak-by-barcode matrix from the single nucleus assay for transposase accessible chromatin followed by sequencing (snATAC-seq) of the E14.5 mouse embryonic cortex were obtained from a previously published dataset (GSE167047; Rhodes et al., 2022). The snap.pp.make_fragment_file function in SnapATAC2 (v2.7.0; Zhang et al., 2024) was used with default parameters to convert the BAM files into fragment files. Subsequent analyses followed the workflow previously described by Rhodes et al. (2022). For downstream processing, the Signac (Stuart et al., 2021; v1.12.0) was used for normalization of the peak-by-barcode matrices and detection of highly variable peaks. To estimate transcriptional activity, ATAC-seq counts within each gene body and a 2 kb upstream region were quantified using the GeneActivity function in Signac (v1.12.0). FindTransferAnchors function in Seurat (v5.0.1) was used to integrate the snATAC-seq data with scRNA-seq data from the E14 cortex published by Borrett et al. (Dennis et al., 2024). The TransferData function in Seurat (v5.0.1) was then used to the coembedding of the scRNA-seq and snATAC-seq datasets, transferring cell type predictions from scRNA-seq to the snATAC-seq nuclei. Finally, RunPCA and RunUMAP were applied to visualize the coembedding of the scRNA-seq and snATAC-seq datasets.

Data availability

The bulk RNA-seq and CUT&Tag data generated in this study are available in the NCBI Gene Expression Omnibus (GEO) under accession codes GSE277737 (RNA-seq: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE277737) and GSE277736 (CUT&Tag-seq: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE277736). Reviewer access tokens (cfchqysczlkhfon for GSE277736 and khqlgogczzchtmx for GSE277737) were provided for the reviewers and editors to validate our data and analysis results. Previously published mouse embryonic scRNA-seq and snATAC-seq data that were reanalyzed here are downloaded from the GEO under accession numbers GSE152281 (scRNA-seq) and GSE167047 (snATAC-seq).

Statistical analysis

Data are presented as mean ± SEM, with sample size (n) indicated in the figure legends. All n values were ≥3 per group. The statistical significance between two groups was calculated by the unpaired t test. For multiple groups, one-way ANOVA with Tukey’s multiple-comparisons test was used. p < 0.05 was considered statistically significant. All statistical analyses were performed with GraphPad Prism (8.0).

Results

PHD finger proteins are differentially expressed along the neurogenesis trajectory

We first systematically explored the expression of all 75 PHD finger proteins (Basu et al., 2020) along the embryonic neurogenesis trajectory. We reanalyzed a public single cell RNA-seq (scRNA-seq) dataset of mouse embryonic brain at E14.5 (Borrett et al., 2020; Fig. 1A). Excluding genes that are not expressed or lowly expressed, the remaining 64 genes can be categorized into six groups based on their cell type-specific expression patterns (Fig. 1B). A total of 9, 9, and 11 genes exhibit specific expression in RGCs, IPCs, and excitatory neurons, respectively (Fig. 1B). Meanwhile, 35 genes are highly expressed in more than one cell type, including 9 in RGCs/IPCs, 22 in IPCs/neurons, and 4 in RGCs/neurons (Fig. 1B). These patterns indicate that individual PHD figure proteins may have cell type-specific functions during neurogenesis, the majority of which remains unexplored except for PHF21B and PHF6. Consistent with the previous report that PHF21B induces differentiation by promoting cell cycle exit (Basu et al., 2020), it is highly expressed in IPCs and neurons, but not in RGCs (Fig. 1B). In contrast, PHF23 belongs to the RGC-IPC group, indicating that its mechanism of action during neurogenesis may be different from PHF21B (Fig. 1B,C).

Figure 1.

Differential expression of genes encoding PHD finger proteins along the neurogenesis trajectory in the mouse brains at E14.5. A, Reanalysis of Borrett et al. E14.5 scRNA-seq dataset (Borrett et al., 2020) presented by t distributed stochastic neighbor embedding (t-SNE) plot. Cells are colored according to the cell type. B, Dot plot showing expression of PHD finger proteins (rows) in RGCs, IPCs, and excitatory neurons (columns). Dot size represents the percentage of cells expressing each PHD gene, and the red intensity indicates scaled average gene expression. C, Violin plot with box plot overlay showing normalized Phf23 expression in RGCs, IPCs, and excitatory neurons. p values were calculated using the Wilcoxon test. RGCs, radial glial cells; IPCs, intermediate progenitor cells; GE TAPs, ganglionic eminence transit amplifying cells.

Loss of Phf23 disrupts neurodevelopment

We have previously generated Phf23^+/− mice through CRISPR/Cas9-mediated genome editing (Chen et al., 2021), introducing a 14 bp insertion into exon 5 and a premature stop codon. Phf23^−/− mice (Phf23-KO) are embryonically lethal at ∼E15.5. At E13.5, however, we were able to harvest Phf23-KO embryos slightly below the Mendelian ratio (∼21%; Fig. 2A). To ascertain Phf23-KO in the neural lineage, we performed Western blot analysis on Phf23-WT NSCs cultured from E13.5 or P0.5 brains and compared with Phf23-KO NSCs from E13.5 brains. PHF23 proteins are robustly expressed in WT NSCs at both timepoints, but not in Phf23-KO NSCs at E13.5, confirming the KO status (Fig. 2B). To investigate the impact of Phf23 loss on embryonic neurogenesis, we performed H&E histological analysis on Phf23-KO, Phf23-HET, and Phf23-WT brains at E13.5 (Fig. 2C). The area of cortical plate (CP) is markedly reduced in Phf23-KO embryos (Fig. 2C,D). We further performed binned cell density analysis at the medial telencephalon, by quantifying cells in eight segments (75 μm wide × 25 μm tall) along the radial column from the ventricular wall to the pial surface. In Phf23-KO embryos, cell density was significantly higher near the VZ but sharply declined in distal regions closer to the pial surface (Fig. 2E). Phf23-HET mice exhibit a similar trend of a decrease in cortical size and an increase of cellularity near the ventricular zone (Fig. 2C–E) but do not reach statistical significance. Thus, loss of Phf23 results in apparent neurodevelopmental defects.

Figure 2.

Phf23 loss leads to neurodevelopmental defects. A, Pie chart depicting the genotype ratios of embryos (n = 77) collected at E13.5 derived from Phf23-HET male and female parental mice. B, Western blot for PHF23 and GAPDH of Phf23-WT and Phf23-KO mNSCs at E13.5. Phf23-WT mNSCs at P0.5 were used as a control. C, Left, Representative H&E staining of Phf23-WT, Phf23-HET and Phf23-KO dorsal forebrains at E13.5. Scale bar: 200 μm. Right, High magnification images from left showing radial columns. Scale bar: 20 μm. D, Quantification of the thickness of Phf23-WT, Phf23-HET, and Phf23-KO CP at E13.5. Biological replicates n = 5 for Phf23-WT group, n = 3 for Phf23-HET group, and n = 4 for Phf23-KO group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. E, Quantification of the binned cell counts along 75-µm-wide radial column of Phf23-WT, Phf23-HET, and Phf23-KO medial telencephalon at E13.5. Biological replicates n = 4 for Phf23-WT group, n = 3 for Phf23-HET group, and n = 4 for Phf23-KO group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. LV, lateral ventricle; VZ, ventricular zone; SVZ, subventricular zone; IZ, intermediate zone; CP, cortical plate.

Loss of Phf23 blocks differentiation of RGCs

To determine which cell type(s) is most affected by Phf23 loss, we immunostained brain sections with Sox2/Ctip2 and Tbr2 (Fig. 3A,B). The number of Sox2⁺ RGCs/NPCs in the Phf23-KO VZ/SVZ is increased compared with Phf23-WT, while the number of Tbr2⁺ IPCs in the SVZ/IZ and Ctip2⁺ neurons in the CP are markedly reduced to less than half of the Phf23-WT group (Fig. 3C–E). Notably, the thickness of VZ/SVZ is apparently increased in Phf23-KO cortex, in contrast to decreased thickness of IZ or CP (Fig. 3A,B), consistent with cell density quantification (Fig. 2E). Phf23-HET mice exhibit a similar trend of an increase in Sox2⁺ cells and a decrease of Tbr2⁺ IPCs and Ctip2⁺ neurons (Fig. 3A–E) but do not reach statistical significance. These observations suggest that loss of Phf23 may block the differentiation of RGCs to IPCs and neurons. This is consistent with cell type-specific expression pattern of Phf23 in RGCs and IPCs (Fig. 1B,C).

Figure 3.

Phf23 loss disrupts cortical neurogenesis. A, Representative immunofluorescence staining of Sox2 (green) and Ctip2 (red) on Phf23-WT, Phf23-HET, and Phf23-KO cortex at E13.5. Scale bar: 20 μm. B, Representative immunofluorescence staining of Tbr2 (green) on Phf23-WT, Phf23-HET, and Phf23-KO cortex at E13.5. Scale bar: 20 μm. C–E, Quantification of Sox2⁺ cell numbers in the VZ/SVZ (C), Ctip2⁺ cell numbers in the CP (D) of A, and Tbr2⁺ cell numbers in the IZ (E) of B. Biological replicates n = 4 for Phf23-WT group, n = 3 for Phf23-HET group, and n = 4 for Phf23-KO group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. VZ, ventricular zone; SVZ, subventricular zone; IZ, intermediate zone; CP, cortical plate.

In addition to E13.5, we analyzed Phf23-KO, Phf23-HET, and Phf23-WT brains at E11.5, a stage preceding the generation of Ctip2⁺ deep-layer neurons. At E11.5, Phf23-KO already exhibited increased Sox2⁺ RGC/NPCs (Fig. 4A,B) and reduced Tbr2⁺ IPCs compared with WT (Fig. 4C,D). Phf23-HET mice exhibit intermediate phenotypes. These data strengthen the notion that Phf23 loss disrupts differentiation of RGCs to IPCs.

Figure 4.

Phf23 loss disrupts the initial phase of cortical neurogenesis. A, Representative immunofluorescence staining of Sox2 (green) on Phf23-WT, Phf23-HET, and Phf23-KO cortex at E11.5. Scale bar: 20 μm. B, Quantification of Sox2⁺ cell numbers in the VZ/SVZ of A. Biological replicates n = 4 for each group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. C, Representative immunofluorescence staining of Tbr2 (green) on Phf23-WT, Phf23-HET, and Phf23-KO cortex at E11.5. Scale bar: 20 μm. D, Quantification of Tbr2⁺ cell numbers in the IZ of C. Biological replicates n = 4 for each group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. VZ, ventricular zone; SVZ, subventricular zone; IZ, intermediate zone.

To rule out the possibility that enlarged RGC pools in Phf23-KO brains at E11.5 and E13.5 is due to increased proliferation and/or decreased apoptosis, we performed immunostaining with the proliferation marker Ki67, BrdU, and apoptosis marker Cleaved Caspase-3 (CC3), using Tbr2 staining to demarcate the boundary between VZ/SVZ and IZ (Fig. 5A–F). There is no significant difference in the percentage of Ki67⁺ cells in the VZ/SVZ or IZ at E11.5 and E13.5 (Fig. 5B,D). BrdU labeling also confirms that there is no apparent difference in proliferation (Fig. 5E). As for apoptosis, there are few CC3⁺ cells in either Phf23-KO and Phf23-WT brains (Fig. 5F). These observations indicate that proliferation or apoptosis in RGCs/IPCs unlikely contribute to the neurogenesis defects observed in Phf23-KO brains.

Figure 5.

Phf23 loss blocks NSC differentiation without affecting proliferation or apoptosis. A, Representative immunofluorescence staining of Ki67 (red) and Tbr2 (green) on Phf23-WT, Phf23-HET, and Phf23-KO cortex at E11.5. Scale bar: 20 μm. B, The percentage of Ki67⁺ cells among total DAPI cells in the VZ/SVZ and IZ of A. Biological replicates n = 4 for each group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. C, Representative immunofluorescence staining of Ki67 (red) and Tbr2 (green) on Phf23-WT, Phf23-HET, and Phf23-KO cortex at E13.5. Scale bar: 20 μm. D, The percentage of Ki67⁺ cells among total DAPI cells in the VZ/SVZ and IZ of C. Biological replicates n = 4 for Phf23-WT group, n = 3 for Phf23-HET group, and n = 4 for Phf23-KO group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. E, Representative immunofluorescence staining of BrdU (red) on Phf23-WT and Phf23-KO cortex at E13.5. Scale bar: 20 μm. F, Representative immunofluorescence staining of Cleaved Caspase-3 (green) on Phf23-WT and Phf23-KO cortex at E13.5. Scale bar: 20 μm. G, Representative images of cortical primary neurospheres and second neurospheres on Phf23-WT, Phf23-HET, and Phf23-KO mNSCs. Scale bar: 100 μm. H, Quantification of the primary neurospheres number and diameter of G. Biological replicates n = 4 for each group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. I, Quantification of the second neurospheres number and diameter of (G). Biological replicates n = 4 for each group. p values, One-way ANOVA test. Error bars indicate mean ± SEM. VZ, ventricular zone; SVZ, subventricular zone; IZ, intermediate zone; CP, cortical plate.

We further performed in vitro neurosphere assay to determine whether Phf23-KO NSCs exhibit self-renewal defects. We cultured RGCs from Phf23-KO, Phf23-HET, and Phf23-WT brains at E13.5 under standard nonadhesive NSC culture conditions (Chojnacki and Weiss, 2008; Fig. 5G) and quantified the number and size of neurospheres (NSs) in primary and secondary culture. The number of primary NSs from Phf23-KO NSCs is reduced compared with Phf23-HET and Phf23-WT, but the sphere sizes are relatively normal (Fig. 5H). In contrast, both the number and the size of secondary NSs from Phf23-KO NSCs are reduced compared with Phf23-WT (Fig. 5I). These results demonstrate that loss of Phf23 also impairs the self-renewal of NSCs in vitro.

Loss of Phf23 in NSCs leads to epigenetic downregulation of neurogenesis pathways

To investigate the molecular mechanism underlying differentiations defects of Phf23-KO NSCs, we first harvested Phf23-KO and Phf23-WT NSCs for bulk RNA-seq. A total of 1,214 genes were downregulated while 1,095 genes were upregulated in Phf23-KO NSCs compared with the control (Fig. 6A, Extended Data Table 1). GO analysis shows that genes downregulated in Phf23-KO NSCs enrich in neurogenesis and neuronal differentiation pathways, as well as histone modification/acetylation pathways (Fig. 6B, Extended Data Table 2). Key factors regulating neuronal differentiation such as Dlx1/2, Eya1, and TCF4 are among the significantly downregulated genes (Fig. 6B). Interestingly, genes in the eye development pathway are also downregulated (Fig. 6B). Meanwhile, genes upregulated in Phf23-KO NSCs enrich for metabolism pathways (Fig. 6B). Thus, PHF23 plays an important role in activating the neuronal differentiation program in NSCs.

Figure 6.

PHF23 complexes with HDAC2 to sustain H3K27ac levels in NSCs. A, Volcano plot of differentially expressed genes (adjusted p value < 0.05 and absolute log₂ fold change >0.5) in Phf23-KO versus WT NSCs, with red dots indicating upregulated genes and blue dots indicating downregulated genes. B, Gene ontology (GO) enrichment analysis of differentially expressed genes (adjusted p value < 0.05 and absolute log₂ fold change >0.5) in Phf23-KO versus WT NSCs. Downregulated genes associated with the GO term “regulation of neuron differentiation” are shown. C, Representative neurospheres viral transfected with Phf23-FLAG-IRES-mCherry plasmids. Scale bar: 50 μm. D, Western blot of FLAG and HDAC2 of input and Phf23-FLAG Co-IP group, with IgG group and mNSCs without Phf23-FLAG as blank controls. E, Heatmaps and profiles showing normalized H3K27ac signals at all gene TSS regions in WT and Phf23-KO NSCs, measured by CUT&Tag. n = 3 for each group. Kb, kilobases. F, Profiles showing average H3K27ac signals at all gene TSS regions in WT and Phf23-KO NSCs. G, GO enrichment analysis of genes with differentially regulated H3K27ac (p value < 0.05 and absolute log₂ fold change >0.5) at promoters in Phf23-KO versus WT NSCs. Downregulated genes associated with the GO term “regulation of neuron differentiation” are shown.

Given PHF23 can promote downstream gene expression by forming complex with and inhibit the deacetylation activity of HDAC1/2 on H3K27ac in pre-B cells (Chen et al., 2021), we tested whether this is a conserved mechanism in NSCs. We overexpressed Phf23-Flag-IRES-mCherry in cultured embryonic NSCs (Fig. 6C) and performed Co-IP experiments to show that PHF23 can indeed form complex with HDAC2 (Fig. 6D). To test whether loss of Phf23 loss could promote deacetylation of H3K27ac, we performed CUT&Tag analysis to investigate the level of H3K27ac across the epigenome in Phf23-KO and Phf23-WT NSCs. Consistently, we observed marked reduction of the overall H3K27ac levels in Phf23-KO NSCs (Fig. 6E,F). We further called differential H3K27ac peaks between Phf23-KO and Phf23-WT NSCs and annotated them to genomic regions, identifying 933 genes with significantly reduced H3K27ac levels (Extended Data Table 3). Among them, 488 genes exhibit reduced H3K27ac at promoter regions (Extended Data Table 3). GO analysis reveals that these genes enrich for neurogenesis and neuronal differentiation pathways (Fig. 6G, Extended Data Table 4), consistent with the RNA-seq data. These data demonstrate that loss of Phf23 in NSCs epigenetically downregulates the expression of neurogenesis pathway genes.

Tcf4 and Eya1 are downstream targets of PHF23

To identify candidate genes under the direct epigenetic control of PHF23-HDAC2, we intersected genes downregulated in RNA-seq and with reduced H3K27ac at promoter regions in CUT&Tag (Fig. 7A, Extended Data Table 5). We identified 82 overlapping genes, which as expected are enriched in neuronal differentiation pathways (Fig. 7B, Extended Data Table 6). Among them, Tcf4 and Eya1 are two top hits associated with neuronal differentiation (Fig. 7A,C,D). We further confirm the results by comparing the H3K27ac peaks between Phf23-KO and Phf23-WT NSCs and indeed found marked reduction in promoter regions of these two genes (Fig. 7E,F), indicating that Eya1 and Tcf4 are downstream of PHF23-HDAC2 epigenetic regulation.

Figure 7.

PHF23 epigenetically promotes the expression of Tcf4 and Eya1 in NSCs. A, Venn diagram showing overlap between genes with decreased RNA expression (adjusted p value < 0.05 and log₂ fold change <−0.5) and those with reduced H3K27ac (p value < 0.05 and log₂ fold change <−0.5) at promoters in Phf23-KO versus WT NSCs. p value calculated by hypergeometric test. B, GO enrichment analysis of the 82 overlapping genes in A. C, Normalized expression of Tcf4 (left) and Eya1 (right) in WT and Phf23-KO NSCs from bulk RNA-seq. D, Dot plot showing log₂ fold change and significance of RNA expression and H3K27ac levels for the 82 overlapping genes in A. Dot size indicates log₂ fold change of RNA expression, and red intensity indicates statistical significance. E, F, Genome browser tracks showing H3K27ac binding density at the Tcf4 (E) and Eya1 (F) promoters in WT and Phf23-KO NSCs. G, Coembedding uniform manifold approximation and projection (UMAP) visualization of Borrett et al. scRNA-seq dataset and Rhodes et al. snATAC-seq dataset. Left, Colors indicating cell types. Right, Colors indicating sequencing methods. H, Dot plot showing chromatin accessibility of Phf23, Tcf4, and Eya1 (rows) in RGCs, IPCs, and excitatory neurons (columns). Dot size represents the percentage of cells with chromatin accessibility for each gene, and the red intensity indicates scaled average chromatin accessibility.

In addition, we examined the chromatin accessibility status of these two genes in RGCs, IPCs, and excitatory neurons by reanalyzing Rhodes et al. single nucleus ATAC-seq (snATAC-seq) dataset (Rhodes et al., 2022; Fig. 7G). The chromatin accessibility of Phf23 and Tcf4 are closely correlated, which are highest in RGCs (Fig. 7H). The chromatin accessibility of Eya1, while present in RGCs, appears highest in IPCs and neurons (Fig. 7H).

Overexpression of Tcf4 but not Eya1 rescues differentiation defects of Phf23-KO NSCs

TCF4 is a member of the basic helix-loop-helix (bHLH) family of transcription factors and is essential for the differentiation and migration of neural progenitor cells (Schoof et al., 2020), while EYA1 (Eyes Absent Homolog 1) is a transcriptional coactivator for SIX1 essential for proliferation, survival, and neuronal differentiation during inner ear and sensory neuron development (Xu et al., 1999; Zou et al., 2004). To test whether these genes are responsible for the differentiation defects of Phf23-KO NSCs, we overexpressed Tcf4 and Eya1 in Phf23-KO and Phf23-WT NSCs and compared their differentiation potential with empty vector-overexpressing Phf23-KO and Phf23-WT NSCs (Fig. 8A,B). Tcf4 overexpression in Phf23-KO NSCs completely rescues the total number of DAPI⁺ cells, as well as the percentage of differentiated MAP2⁺ postmitotic neurons (Fig. 8C–E). In contrast, overexpression of Eya1 only rescues the total number of DAPI⁺ cells but does not rescue the percentage of MAP2⁺ neurons (Fig. 8C–E). Meanwhile, the increased percentage of SOX2⁺ progenitor cells derived from Phf23-KO NSCs can be rescued by both Tcf4 and Eya1 overexpression (Fig. 8C,E). As previously reported, the majority of NSCs differentiated upon growth factor withdraw become GFAP⁺ astrocytes, and Phf23-KO NSCs do not appear to have a defect in this lineage (Fig. 8C,F). However, Phf23-KO NSCs do exhibit oligodendrocyte differentiation defects, which can be partially rescued by Tcf4 overexpression (Fig. 8C,F). These data indicate that TCF4 and EYA1 are both downstream effectors of PHF23, and TCF4 is mainly responsible for proper neuronal differentiation.

Figure 8.

Overexpressing of Tcf4 but not Eya1 rescues differentiation defects in Phf23-KO NSCs. A, Left, The expression of Tcf4 in Phf23-WT NSCs transfected with empty vector (Phf23-WT) or Tcf4 overexpression vector (Phf23-WT-Tcf4 OE) was quantified by qPCR, normalized to Phf23-WT level. Right, The expression of Tcf4 in Phf23-KO NSCs transfected with empty vector (Phf23-KO) or Tcf4 overexpression vector (Phf23-KO-Tcf4 OE) was quantified by qPCR, normalized to Phf23-KO level. Technical replicates n = 4 for each group. p values, unpaired t test. Error bars indicate mean ± SEM. B, Left, The expression of Eya1 in Phf23-WT NSCs transfected with empty vector (Phf23-WT) or Eya1 overexpression vector (Phf23-WT-Eya1 OE) was quantified by qPCR, normalized to Phf23-WT level. Right, The expression of Eya1 in Phf23-KO NSCs transfected with empty vector (Phf23-KO) or Eya1 overexpression vector (Phf23-KO-Eya1 OE) was quantified by qPCR, normalized to Phf23-KO level. Technical replicates n = 4 for each group. p values, unpaired t test. Error bars indicate mean ± SEM. C, Representative immunofluorescence staining of MAP2 (green), SOX2 (green), GFAP (red), and NG2 (green) on Phf23-WT, WT-Tcf4 OE, WT-Eya1 OE, Phf23-KO, KO-Tcf4 OE, and KO-Eya1 OE mNSCs differentiated for 10 d. Scale bar: 20 μm. D–F. Quantification of total DAPI cell number (D), percentage of MAP2⁺ cells, SOX2⁺ cells (E), GFAP⁺ cells, and NG2⁺ cells (F) among total DAPI cells. Biological replicates n = 3 for each group. p values, one-way ANOVA test. Error bars indicate mean ± SEM. G, The mechanism of action of PHF23 on embryonic neurogenesis. Schematic diagram showing that PHF23 forms complex with HDAC2 and inhibit its deacetylation function, leading to increased H3K27 acetylation levels of neuronal differentiation pathway genes to promote the differentiation of RGCs to IPCs and neurons. In the absence of Phf23, the H3K27 acetylation levels of these genes are reduced resulting in RGC differentiation blockade.

Discussion

PHD finger proteins as epigenetic readers play important roles in development and disease, yet their roles in neurodevelopment and cortical neurogenesis remain largely unexplored. In this study, we demonstrate that PHD finger proteins are differentially expressed along the neurogenesis trajectory and PHF23 is a major regulator governing the differentiation from RGCs to IPCs and neurons.

A number of PHD finger proteins are associated with development and diseases, underscoring their functional significance. For example, PHF2 controls the genomic stability and cell cycle progression of neural progenitor cells (Pappa et al., 2019) and is implicated in metabolic disorders (Jeong et al., 2023). PHF5a regulates transcriptional elongation for maintaining pluripotency of embryonic cells and controls differentiation of adult myoblasts (Strikoudis et al., 2016). Mutations in the PHF6 gene are associated with Börjeson–Forssman–Lehmann syndrome, a rare X-linked disease causing neurological disorder (Gecz et al., 2006; Zhang et al., 2013). The majority of known PHD finger proteins are expressed in RGCs, IPCs, and/or neurons during embryonic neurogenesis, yet their expression patterns differ dramatically and clearly exhibits cell type specificity. While they commonly recognize the H3K4me3 histone mark through the PHD domain, the protein complexes they recruit along with other regulatory factors may determine how they regulate downstream genes in a cell type and spatiotemporally specific manner. The high expression of Phf23 in RGCs and IPCs correlates well with its function in regulating RGC differentiation, while the high expression of Phf21b in IPCs and postmitotic neurons correlates with its role in promoting cell cycle exit.

The quantification of progenitor cell numbers in Phf23-KO and Phf23-HET embryonic brains by binning regions and normalizing to total cell counts show consistent trends toward increased progenitor numbers near the VZ/SVZ regions and reduced differentiation. If more samples are analyzed, it could be statistically more significant to show an effect of Phf23 heterozygosity. However, neurosphere analysis indicated that these increases are not due to increases in self-renewing progenitors but more likely reflecting an inability to fully differentiate.

The observation that loss of Phf23 leads to differentiation blockade has been previously described by us in the hematopoietic system (Chen et al., 2021). This study broadens our understanding on the role of PHF23 in the central nervous system. Notably, while many downstream targets (such as Eya1) are conserved across these two systems, there are clearly cell type-specific differentiation programs (such as Tcf4) epigenetically regulated by Phf23. Thus, it is likely that additional players are involved in different systems that work synergistically with Phf23 to promote differentiation. Recently, PHF23 has been show to regulate the proliferation of non-small cell lung cancer cells in part by the regulation of the ERK pathway (Cheng et al., 2023). While we did not observe proliferation defects in Phf23-KO brains, reduced self-renewal of Phf23-KO NSCs in neurosphere assays may be associated with similar mechanisms.

Through combined RNA-seq and CUT&TAG analysis, we identified Tcf4 as a major downstream target of Phf23 and validated its role through rescue experiments. Consistently, loss of function TCF4 mutation in Pitt–Hopkins syndrome–derived brain organoids impaired the capacity of neural progenitor cells to differentiate into neurons (Papes et al., 2022). Admittedly, as an epigenetic regulator, PHF23, may function through additional downstream genes and pathways. Our Cut-Tag analysis of enriched promoters showed a number of genes potentially implicated in NSC self-renewal versus differentiation, such as Hey2, Bmp2, and Bdnf, which may also be contributing to the Phf23-mutant phenotypes. HEY2 regulates maintenance of NSCs (Sakamoto et al., 2003), BMP2 inhibits proliferation of RGCs and induce differentiation (Yao et al., 2014), and BDNF is well known for its function in inducing neuronal differentiation (Leschik et al., 2013). Functionally validating the rescue effects of expressing such genes in Phf23-KO NSCs could provide more detailed mechanism of action of PHF23 during embryonic neurogenesis.

In summary, our study highlights an epigenetic mechanism regulating neurogenic pathway genes such as Tcf4 during embryonic neurogenesis through PHF23-HDAC2 (Fig. 8G), providing insights into cell type-specific functions of PHD finger proteins.