Cell Type-specific Alternative Splicing Governs Cell Fate in the Developing Cerebral Cortex

RNA Sequencing of Sorted NPCs and Neurons from Developing Mouse Cerebral Cortex Uncovers Extensive Alternative Exon Usage during Cortical NPC Differentiation

Using a Tbr2-EGFP transgene driving EGFP in dorsal cortex (Figure S1A) (Gong et al., 2003), we found that E14.5 VZ NPCs (Sox2+; EGFP−) are well separated from IPs in the subventricular zone (SVZ, Tbr2+; EGFP+) and differentiating neurons in the intermediate zone (IZ) and cortical plate (CP, Sox2-; EGFP+, Figure 1A and Figure S1A). We isolated VZ NPCs (EGFP−) and non-VZ cells (EGFP+) from E14.5 Tbr2-EGFP cerebral cortex (Figure 1B), and confirmed their identities: (1) strand-specific RNA-Seq and quantitative PCR of sorted cells showed that NPC genes Sox2, Pax6, Nes and Hes1 were highly enriched in the (EGFP−) cells while Tbr1, Fezf2 and Neurod2 were enriched in the (EGFP+) cells (Figure 1C, Figure S1B); (2) the fourth exon of REST, expressed in differentiating neurons (Raj et al., 2011), was depleted from sorted (EGFP−) cells (Figure S1C); (3) 93.5% of sorted (EGFP−) cells were Sox2 positive (Figure S1D–E); (4) gene ontology analysis revealed that cell cycle, chromosomal and DNA metabolic genes were enriched in (EGFP−) cells, while neuron differentiation and projection genes were enriched in (EGFP+) cells (Figure S1F). These results indicate that we successfully isolated and analyzed VZ NPCs (EGFP−) and a mixture of IPs and neurons outside the VZ (hereafter referred to as non-VZ or neuron) from developing mouse cerebral cortex.

We compared alternative exon usage between E14.5 VZ NPCs and non-VZ cells using the mixture-of-isoforms (MISO) statistical model, which assigned a ‘percentage spliced in’ (PSI) value to each exon by estimating its abundance compared to adjacent exons (Katz et al., 2010). We found that 622 exons were differentially spliced between mouse NPCs and neurons (|ΔPSI|≥10% and Bayes factor ≥5, the same criteria used hereafter if not specified), with 345 showing higher inclusion in neurons and 277 higher in NPCs. We analyzed VZ and CP samples from two additional RNA-Seq datasets (Ayoub et al., 2011; Fietz et al., 2012), and found 742 AS changes shared by at least two of the three datasets (Figure 1D, Figure S1G). 272 cassette exons or skipped exons (SE) comprised the largest portion of AS events, with 198 (73%) SEs showing higher inclusion in neurons than in NPCs (Figure 1D), 255 (94%) SEs shorter than 300 nt (Figure S1H), and 61 (22%) SEs causing frame shifts (Figure 1E). These results indicate that hundreds of alternative exon usages occur when VZ NPCs differentiate in mice.

Extensive Alternative Exon Usage during NPC Differentiation in the Developing Human Cerebral Cortex

The developing human cortex has an enormously expanded oSVZ housing the largest population of dividing progenitors, the most prominent being outer radial glial (oRG) cells that resemble aRG cells but lack apical processes. We compared laser microdissected oSVZ to VZ, inner SVZ (iSVZ) and CP (Fietz et al., 2012), as well as RNA-seq data from purified aRG and oRG cells (Johnson et al., 2015), and found that most splicing switches occur between neural progenitors in the VZ and neurons in the CP (2582 exons, Figure S1I–S1J), with 31% of human cassette exons (|ΔPSI|≥15%, Bayes factor ≥10) causing a translational frame shift (Figure S1K). When SE, A3SS, A5SS, MXE and RI (Figure 1D) were considered, 38% of mouse differential splicing events were also differentially spliced in human, including 113/272 (42%) mouse SEs (Figure 1F).

To compare AS between pure NPC and neuronal populations, we further analyzed RNA-Seq data of 224 single cells from three human fetal cortices (Camp et al., 2015). Using the unbiased pathway and gene set over dispersion analysis approach (PAGODA) (Fan et al., 2016), we were able to identify distinct transcriptional subpopulations corresponding to RGs, IPs, immature neurons, and neurons (Fig 1G). We then pooled single RGs and single neurons in silico and extended our MISO analysis to these putatively pure RG and mature neuron populations, which gave rise to 298 significant cassette exon changes. Consistent with bulk samples, more SEs show higher inclusion in neurons (166) than in RGs (132). For 26 SEs that were also identified in human bulk samples, ΔPSIs are significantly correlated between bulk and pooled single cell analyses (r^2 = 0.44, p-value = 1.7e-4, Fig 1H–1I). These results indicate that extensive alternative exon usages occur when VZ NPCs differentiate in mouse and human cerebral cortex and many of them may critically regulate gene function by shifting translational frames, though we focus here mainly on those shared by mouse and human to allow functional analysis of their mechanism.

Cytoskeleton Genes Are Preferentially Regulated by Alternative Splicing during Neurogenesis

Gene ontology analysis for differentially spliced exons revealed that cytoskeleton genes are overrepresented in mouse and human (Figure 2A, Figure S2A–S2B). KEGG pathway analyses of alternative exons also revealed enriched functions in regulating tight junctions and actin cytoskeleton (Figure S2C and data not shown). Differential mouse SEs between E14.5 NPCs and neurons were validated by RT-PCR, and many of these events were switch-like (|ΔPSI| > 50%, Figures 2B–2F, and S2D–S2E, Table S1). Notably, alternatively spliced cytoskeleton genes encoded functionally interconnected protein networks (Figure 2G–2I) involved in NPC proliferation, neuronal migration and neuronal differentiation. Over a dozen genes alternatively spliced during the NPC-to-neuron transition--including Add1, Ank2, Flna, Kif2a, Macf1, Scrib, and Syne2--have previously been shown to be required for normal cerebral cortical development in mice (Table S2), and certain alternative exons harbor mutations that associate with human brain disorders such as microcephaly and autism (Figure 2J–2M and below), suggesting cell type-specific mechanisms of human diseases.

Alternatively Spliced Exons Alter Protein Domains and Subcellular Localization

Exons that are differentially spliced between NPCs and neurons are remarkable in the extent to which they critically involve essential protein domains. Among 61 splicing changes in 49 genes regulating cytoskeleton (Table S3), 20 (33%) of them caused insertion or deletion of one or more entire protein domains (Figure 3A–3B, Figure S3A–S3C). Multiple genes showed AS of membrane-targeting domains (Figure 3B), which typically regulate subcellular localization. Another 13 (21%) alternative exons--including a microexon--inserted extra amino acids (extraAA) into well-defined protein domains (Figure 3A, Figure S3A and S3C). Thus, more than half (54%) of differentially spliced regions directly regulate protein domains.

Alternative Splicing Switches Ninein from Centrosome in NPCs to Non-centrosomal Loci in Neurons

We have shown a large Nin alternative exon 18 (> 2000 nt) that is almost exclusively included in mouse and human NPCs but skipped in neurons (Figure 2J–2K, ΔPSI = −83.5%). In contrast, we also identified a conserved 61 nt exon (exon 29) of Nin that is specifically included in neurons (CP) but skipped in NPCs (Figure 3C–3D, ΔPSI = 74.0%), introducing a frame shift and truncating the C-terminus of Nin (Figure 3E–3F). Nin associates with the mother centriole to anchor microtubules, with its C-terminal segment targeting efficiently to centrosomes (Delgehyr et al., 2005). Although Nin translocates from centrosome to noncentrosomal loci during brain progenitor cell differentiation, the underlying mechanism remains unknown (Baird et al., 2004; Ohama and Hayashi, 2009).

We examined the effects of exon 18 and exon 29 on Nin subcellular localization and found that the C-terminal centrosome-localization signal of Nin (Nin-NPC-CT) is lost with the inclusion of alternative exon 29, causing the neuronal Nin isoform (Nin-Neuron) to be diffusely localized in cytoplasm when highly expressed (Figure 3F–3G), associating with microtubules at non-centrosomal loci (Figure S3D–S3F). Targeted co-immunoprecipitation (co-IP) screen of centrosomal proteins allowed us to uncover that the Nin-NPC-CT specifically pulled down CEP250 (Figure 3H), and CEP250 knockdown disrupted the centrosomal localization of Nin (Figure 3I, Figure S3G). We further found that CEP170 preferentially interacts with Nin-NPC through exon 18, but not the Nin-Neuron isoform (Figure 3H, Figure S3I). Given that CEP170 relies on Nin for its centrosomal localization (Graser et al., 2007), our data suggest that the Nin-CEP170 interaction is mediated by Nin exon18 and is dynamically regulated by alternative splicing. These results indicate a two-fold alternative splicing mechanism: (1) inclusion of neuronal exon 29 dissociates Nin from CEP250 and centrosome, causing Nin to adopt a non-centrosomal localization in neurons; and (2) skipping of Nin exon 18 further dissociates CEP170 from Nin in neurons.

Endogenous Nin-Neuron is significantly elevated in postmitotic neurons during cortical neurogenesis (Figures S4A–S4B). Ectopic expression of Nin-Neuron, but not Nin-NPC, led to significant neural progenitor depletion from the VZ (Figure 4A–4B) and defective neuronal migration into the CP (Figure S4C). We performed pair-cell assays, and found Nin-Neuron expression causes significantly more NPCs to become neurons compared to control or the Nin-NPC isoform (Figure 4C–4D). Expression of Nin-Neuron disperses endogenous Nin and Dctn1--a key dynactin subunit mediating cytoplasmic dynein-cargo binding--from the centrosome (Figure 4E–4F), suggesting a dominant negative mechanism in regulating cell fate. We thus created a NIN knockout cell line (Chr14:51,259,547delT (hg19), p.Arg107Thrfs*16, Figure S4D) and found that loss of centrosomal NIN disrupted normal mitotic cleavage planes (Figure 4G–4H). These results indicate that endogenous switching of Nin-NPC to the Nin-Neuron isoform is sufficient to convert cortical NPCs to neurons, suggesting that neurogenesis is normally controlled at least in part by alternative splicing of Nin.

Aberrant Splicing of A Cryptic Poison Exon Causes A Unique Brain Malformation

Splicing analysis identified a highly conserved but not previously annotated 57 nt exon in Flna that is skipped in E14.5 NPCs (VZ) and most adult tissues, but included in mRNAs from E14.5 neurons (CP) as well as adult cortex and cerebellum (Figure 5A). Surprisingly, an in-frame stop codon was embedded in this neuron-specific Flna exon (Figure 5B and Figure S5A), inclusion of which causes early truncation of Flna protein (482 of 2647 amino acids, Figure 5C), and/or nonsense-mediated mRNA decay (NMD, Figure 5D). We thus name the Flna SE “exonN” (N stands for Null in Neuron), and found that the human homologous exonN is skipped in VZ while included in CP (Figure 5E). The skipping of exonN in cortical NPCs creates a full-length protein, whereas splicing into this exon in neurons would give a truncated protein or unstable mRNA. Remarkably, genomic sequences of exonN including the embedded stop codon are conserved across multiple placental mammalian species, though it is not observed outside mammals (Figure 5F).

The presence of poison exons encoding translational stop codons reported here and by other groups (Yan et al., 2015) suggests a potential mechanism of human disease that to our knowledge has not been previously described: mutations that disrupt normal suppression of a poison exon would inactivate protein expression in a cell that normally skips the exon. We tested this model by studying FLNA, since heterozygous null mutations in females typically cause periventricular heterotopia (PVNH), where neurons form nodules along the VZ, reflecting failure of normal migration (Fox et al., 1998). We hypothesized that, since abnormal inclusion of exonN will create a FLNA null allele (Figure 5B–5F), mutations that promote exonN inclusion in NPCs may lead to PVNH.

Sequencing FLNA exonN and its flanking introns in 221 individuals with genetically unexplained PVNH revealed a rare variant FLNA-as (ChrX: 153,594,210 C>T, c.1429+182 G>A) in a large family in which multiple individuals show PVNH that is notably milder than that caused by FLNA null alleles (Figure 5G–5I). Unlike typical females with heterozygous FLNA null mutations, who have abnormally located neurons lining the entire lateral ventricles bilaterally, the PVNH in FLNA-as patients is striking, but limited to smaller segments of the VZ, often asymmetrically (Figure 5G). The FLNA-as mutation lies in the upstream intron of exonN (Figure 5J), and leads to abnormal exonN inclusion in patient blood cells, where exonN is normally skipped (Figure 5K). Furthermore, FLNA null mutations are usually male lethal, yet affected males in this family do not show early lethality. Males and females with null FLNA mutations also show variably penetrant congenital heart disease and severe vascular catastrophes (Fox et al., 1998), but the FLNA-as pedigree shows no clear evidence of cardiovascular disease, suggesting that the FLNA-as mutation may cause a central nervous system (CNS)-specific and cell type-specific phenotype.

We expressed wild-type (WT) and mutated FLNA mini-genes in Neuro2a cells (Figure 5J and 5L–5M) and in E14.5 mouse cortical NPCs (Figure 5N) and found that the WT mini-gene produced transcripts that skipped exonN (PSI<5%), while the mini-gene with the FLNA-as mutation produced 40%–50% of FLNA transcripts that abnormally included exonN or the last 8-nt of exonN that creates a frame shift (Figure 5L, 5N, Figure S5B). FLNA-as mutation led to a truncated FLNA mini-protein and reduced levels of mini-proteins (Figure 5M). These results confirm that the c.1429+182 G>A mutation promoted abnormal exonN inclusion in NPCs, creating a cell type- and tissue-specific FLNA partial loss-of-function and an atypical PVNH syndrome.

Ptbp1 and Rbfox1/2/3 Antagonistically Regulate Neuronal Exon Inclusion in the Developing Cerebral Cortex

We used DREME (Bailey, 2011) to identify discriminative cis-regulatory sequence motifs enriched in the flanking introns of cassette exons that are preferentially included in neurons. The only two significant motifs identified correspond to binding motifs of Ptbp1/2 (CU(C/U)UCUU, found within 200 nts in the upstream intron, p-value=3.6e-9), and Rbfox1/2/3 (GCAU(G/A), within 200 nts in the downstream intron, p-value =1.3e-8, Figure 6A). Notably, 61% of alternatively spliced neuronal exons had Ptbp1/2 and/or Rbfox1/2/3 regulatory motifs (Figure S6A). Analyses of Rbfox HITS-CLIP and Ptbp1 iCLIP datasets from neuronal samples (Linares et al., 2015; Weyn-Vanhentenryck et al., 2014) further support that many cortically regulated exons exhibit direct binding by these proteins (Figure S6B, Table S4 and below).

It has been shown that Ptbp1 binding upstream suppresses exon inclusion (Keppetipola et al., 2012), whereas Rbfox binding downstream enhances exon inclusion (Kuroyanagi, 2009; Wang et al., 2008). We found that Ptbp1 mRNAs and proteins were highly expressed in NPCs (VZ), with Rbfox1/2/3 specifically expressed in neurons (CP, Figures 6B and S6C), consistent with the positional enrichment around neuron-specific exons. Knock down of Ptbp1, but not Ptbp2, de-repressed inclusion of predicted neuronal exons including Flna exonN (Figure 6C–6D, Figure S6D–S6E), and resulted in decreased Flna protein expression (Figure 6E). Moreover, the protein levels of Ptbp1 and Flna were high in E12.5 mouse cortex and dropped over similar time-courses during cortical development (Figure S6F). Analysis of Ptbp1 iCLIP-Seq datasets (Linares et al., 2015; Masuda et al., 2012) revealed that Ptbp1 bound to the alternatively spliced Flna exonN in both NPCs and C2C12 cells (Figure 6F). The mouse homologous nucleotide (chrX:71,486,253) of the PVNH-associated FLNA c.1429+182 G>A mutation lies within a Ptbp1 iCLIP cluster (Figure 6F), suggesting that the human mutation disrupts splicing regulation via PTBP1.

Expression of Rbfox1/2/3 proteins significantly promoted inclusion of predicted neuronal exons (Figure 6G, Figure S6G–S6H), especially Nin exon 29 (Figure 6H), which led to decreased endogenous Ninein at the centrosome (Figure 6I). Re-analysis of Rbfox-1/2/3 HITS-CLIP datasets from P15 mouse brains (Weyn-Vanhentenryck et al., 2014) revealed that Rbfox1 and Rbfox3 bound to a conserved GCAUG motif, and Rbfox2 CLIP tags mapped to a different but adjacent GCAUG motif (Figure 6J and S6I). These results indicate that Rbfox proteins bind to the downstream intron of Nin exon 29 and promote exon inclusion. In summary, Ptbp1 in NPCs suppresses splicing of a family of neuronal exons including Flna exonN, while Rbfox1/2/3 in neurons promotes inclusion of over 120 neuronal exons including Nin exon 29. Thus, Ptbp1 and Rbfox1/2/3 proteins antagonistically regulate neuronal exon inclusion during the NPC-to-neuron transition in vivo.

Rbfox Proteins and Ptbp1 Antagonistically Regulate Neurogenesis in Cerebral Cortical Development

Conditional knockout (cKO) of Ptbp1 in mouse cerebral cortex causes NPCs to detach from the VZ and differentiate prematurely, leading to lethal hydrocephalus, but the pathogenic mechanism and direct downstream targets remain unknown (Shibasaki et al., 2013). We found that Ptbp1 cKO strikingly phenocopied human PVNH caused by FLNA mutations (Figure 7A), and that protein levels of both Flna and its paralog Flnb were significantly decreased in Ptbp1 knockout cells (Figure 7B). Flna exonN was skipped in controls but abnormally included in Ptbp1 cKO (Figure 7C). We identified an unannotated 98 nt exon of Flnb in the homologous location to Flna exonN, and this Flnb exon was also aberrantly included in Ptbp1 cKO (Figure 7C), and is predicted to cause a translational reading frame shift. These results indicate that Ptbp1 represses inclusion of Flna and Flnb poison exons in NPCs, and Ptbp1 loss-of-function leads to decreased Flna/Flnb levels and PVNH.

FLNA mutations cause PVNH in humans (Fox et al., 1998) and Flna knockout mice show microcephaly, defective adherens junctions and reduced cortical progenitors (Feng et al., 2006; Lian et al., 2012). To test whether Flna functions downstream of Ptbp1, we knocked down Ptbp1 in E13.5 mouse cortical progenitors and found this led to fewer NPCs in the VZ due to abnormal cell cycle exit (Figures 7D–7E, and S7A–S7C). Introduction of exogenous FLNA by in utero electroporation (IUE) partially rescued the premature NPC depletion phenotype (Figure 7D–7E), supporting the model that Ptbp1 functions upstream of Flna in cortical neurogenesis.

Ptbp1 represses neuronal fate in tissue culture cells (Xue et al., 2013), and we found that Ptbp1 expression in the developing mouse cortex was highly correlated with Sox2 (Figure S7D). Sox2 is a master transcriptional regulator of NPC identity and indeed, Sox2 bound to the highly active Ptbp1 promoter region in NPCs (main peak = 398nt, fold enrichment= 7.6, p-value =1.0e-16, Figure S7E), suggesting that Ptbp1 is an important target of Sox2 in suppressing neuronal differentiation.

Single knockout of Rbfox1 or Rbfox2 causes minimal structural malformation in mouse cerebral cortex possibly due to gene redundancy (Gehman et al., 2012; Gehman et al., 2011). However, of particular interest is the consequence of Rbfox1/2/3 gain-of-function in NPCs, because Rbfox proteins promote neuronal exon inclusion and have long been used as pan-neuronal markers. Ectopic expression of Rbfox proteins in NPCs not only significantly decreased the number of VZ NPCs but also dramatically reduced the number of neurons in the CP (Figure 7F–7G). Furthermore, combining Rbfox3 expression with Ptbp1 knock down led to a more severe depletion of VZ NPCs (Figure 7F–7G). These data indicate that Rbfox1/2/3 and Ptbp1 antagonistically regulate neurogenesis in the developing cerebral cortex (Figure 7H).

Mouse Maintenance

Mouse protocols were reviewed and approved by Institutional Animal Care and Use Committee (IACUC) at Boston Children’s Hospital (BCH) and all colonies were maintained following animal research guidelines at BCH. The Tbr2-EGFP transgenic mouse line, for which an EGFP-PolyA cassette was inserted in front of the translational start codon in a FVB/NTac mouse stain, was obtained from GENSAT (Gong et al., 2003) and crossed to the FVB inbreed strain for at least two generations. Heterozygous transgenic mice did not display any obvious developmental defects. Dorsal cerebral cortices were dissected from embryonic day 14.5 (E14.5) heterozygous Tbr2-EGFP transgenic embryos and pooled for splicing analyses without distinguishing genders.

Human Subjects

All human protocols were reviewed and approved by the institutional review board of BCH. Informed consent was obtained from all subjects involved in this study or from parents of those who were younger than 18 years old. PVNH cases screened in this study were recruited from diverse ethic groups and a Caucasian pedigree carrying the FLNA-as mutation is described. Mutations were detected with DNA extracted from blood samples. Splicing analyses of affected individuals were performed with RNA samples extracted from immortalized lymphoblasts.

Isolation of Primary Cells from the Developing Mouse Cerebral Cortex

Dissected brain tissues are dissociated with the Papain Dissociation System (Worthington). Single cell suspension (Neural Basal medium supplemented with 2% B27, 1% N2, 1% Penicillin Streptomycin (PenStrep), 10ng/ml EGF, 10ng/ml FGF2, 0.5% FBS and 0.25% HEPES) was cooled on ice and immediately sorted on the FACSAria II (BD). For immunostaining, sorted cells were plated in Lab-Tek chamber slides (Thermo) and cultured for 4 hours before fixation.

Tissue Culture Cells

Cell lines including Neuro2a (a neuroblastoma cell line with a NPC-like state), Hela and Ptbp1 cKO MEF cells were used for transfection (Lipofectamine 2000, Life Tech) or lentiviral transduction. Lentiviral vectors were packaged and infected following the The RNAi Consortium (TRC) protocols. Briefly, the envelop plasmid (pMD2.G, Addgene, 12259), packaging plasmid (psPAX2, Addgene, 12260), and pLKO.1 based vectors were cotransfected into HEK 293T/17 (ATCC, CRL-11268), and viral particles were harvested 42 hours post-trasfection. Ptbp1 cKO MEF cells were transduced with lentiviral particles carrying control plasmid or CMV-Cre for protein and RNA analyses. To generate NINEIN knockout cell line, pX459 (Addgene 48139) carrying guide sequences were transfected into Hela cells and puromycin (5µg/ml) selected colonies were screened for mutant alleles.

RNA Sequencing analysis

Total RNA was extracted using RNeazy Mini Kit (Qiagen) and treated with Turbo DNase (Ambion). RINs measured by Bioanalyzer (Agilent) for all samples were above 9.0. 500–2000ng total RNA was treated with Ribo-Zero Gold (Epicentre), and mRNA was enriched using Oligo(dT)25 Dynabeads (Life Tech). RNA-Seq libraries (2 replicates each for VZ and non-VZ cells, n or number of embryos= 7 and 11 for VZ, n=7 and 8 for non-VZ replicates) were prepared using ligation-based directional RNA-Seq kit (Bioo Scientific), and sequenced on Illumina HiSeq2000. 36 to 42 million single reads (50bp) were obtained for each replicate, trimmed, aligned (mm9, Tophat-2.0.7, allowing 2 mismatches) and analyzed using the Tuxedo protocol (Kim et al., 2013; Trapnell et al., 2012). Gene expression levels are presented as FPKM (Fragments Per Kilobase of exon per Million reads). Gene ontology analyses were performed using the DAVID online tools (Huang da et al., 2009).

Alternative Splicing Analysis

Aligned bam files (mm9) were analyzed using MISO (version 0.4.6). PSI values of each non-VZ replicate were compared to each VZ replicate, only alternative splicing events that were consistent across all comparisons were considered for downstream analysis. Additional RNA-Seq datasets of laser microdissected cortical tissues (Ayoub et al., 2011; Fietz et al., 2012) were analyzed using the same pipeline. Sashimi plots of RNA-Seq reads were generated in Integrative Genomics Viewer (IGV) (Robinson et al., 2011; Thorvaldsdottir et al., 2013). To study alternative splicing changes during human brain development, published RNA-Seq datasets (Fietz et al., 2012) on laser microdissected fetal human cortical tissues across GW13–16 were aligned to human genome (hg19) and analyzed using the MISO pipeline. Alternatively spliced human exons (|ΔPSI|≥10%, and Bayes factor ≥5) were lifted over to the mouse genome (mm9) and compared to splicing events identified in E14.5 mouse cortex (Figure 1D). RNA-Seq datasets from the ENCODE project were analyzed for tissue distribution of alternative exons.

Motif Analysis

We performed DREME (Bailey, 2011) motif discovery in each of the four regions around differentially spliced cassette exons in mouse: the first 200 nts (i.e. 5’ end) of the upstream intron, the last 200 nts (i.e. 3’ end) of the upstream intron, the first 200 nts of the downstream intron, and the last 200 nts of the downstream intron. For each region, we used exons with higher PSI in neurons as foreground, and exons with higher PSI in NPCs as background, and verse versa. DREME (version 4.9.0) were then run with the option ‘-norc –e 0.05’ to identify significant motifs in each region.

Chip-Seq Analysis

Chip-Seq reads for Sox2 (Lodato et al., 2013) and H3K27ac (Creyghton et al., 2010) on NPCs cultured in vitro were trimmed, and aligned to the mouse genome (mm9) with Bowtie 2.2.4 (Langmead and Salzberg, 2012) allowing zero or one mismatch. Mapped reads were visualized in IGV. Peaks were called using MACS 2.1.0 (Zhang et al., 2008) and statistical results were presented.

Molecular Cloning

pCAGIG (Addgene, 11159) vector and its derivatives were used for ectopic expression of genes in the neuronal system. First, the multiple cloning sites (MCS) between EcoR I and Not I was replaced with EcoR I-ATG(start codon)-HA(1x)-Flag(1x)-Asc I-Not I. Next, the EGFP expression cassette between Msc I and Fse I was replaced with either an mCherry or Puromycin resistance cassette. Coding or genomic sequences of specific genes (e.g., genomic sequences of WT human FLNA between exon 8 and exon 12 and the same sequence carrying the c.1429+182 G>A mutation) were PCR amplified and inserted between Asc I and Not I sites. To knockdown Ptbp1 and Ptbp2 in Neuro2a cells, oligos were annealed and cloned into pLKO.1 TRC cloning vector (Addgene, 10878) between EcoR I and Age I sites. To knockdown Ptbp1 expression in mouse brains, effective hairpins tested in Neuro2a cells (#3) and a control hairpin were end-modified and cloned into the pLL3.7 vector (Addgene, 11795) between Hpa I and Xho I sites. For reverse transcription and polymerase chain reaction (RT-PCR), cDNAs from independent samples were amplified with Phusion (NEB) for 26–30 cycles and analyzed on agarose gels or Novex TBE gels (Life Tech). qRT-PCR (quantitative RT-PCR) were carried out using SYBR Green (ABI). Oligo sequences are listed in Table S5, and plasmids are listed in Table S6.

Protein Analysis

Protein domains were analyzed using Pfam (Finn et al., 2014). 3-D protein structures were predicted on the SWISS-MODEL and Phyre2 servers (Kelley and Sternberg, 2009). Protein lysates were resolved on SDS-PAGE gels and Western Blots were carried out using the Licor Odyssey system. For immunofluorescence staining (IF), embryonic mouse brains were dissected out, fixed in 4% paraformaldehyde, and coronal sections were obtained using a Vibratome or Cryostat (Leica Biosystems). Tissue sections or fixed tissue culture cells were rinsed twice in 1x PBS with 0.2% Triton X-100 (PBST) and incubated at room temperature in a blocking solution (PBST and 4% normal donkey serum), followed by incubation with primary antibodies at 4°C overnight. Samples were then washed 3 times with PBST and incubated with fluorescence conjugated secondary Alexa antibodies (Life Technologies) at room temperature for 2 hours. Slides were mounted with Fluoromount G (Southern Biotech) and imaged on Imager M2 (Carl Zeiss). Primary antibodies are listed in Key Resource Table.

In utero Electroporation (IUE)

IUE was performed following standard protocols (Saito, 2006). In brief, a timed pregnant CD-1 mouse (E13.5) was anesthetized with isoflurane, laid on a 37°C warm plate, the uterine horns exposed and bathed in warm 1x HBSS (1% Pen/Strep), and ~1 µl of plasmid DNA (1–2µg/ µl) mixed with Fast Green was microinjected into the lateral ventricles of embryos using a glass micropipette (Drummond Scientific). Five 50 ms pulses of 30–50 mV at 950-ms intervals were delivered across the uterus with a pair of electrode paddles placed on each side of the embryo’s head using a square-pulse electroporator (BTX, ECM 830). Following electroporation, the uterus was repositioned back into the abdominal cavity and the wound was surgically sutured. The whole process was performed in a sterile environment. After surgery, the animals were closely monitored until they recovered and resumed normal activity. Pair-cell assays were performed following published protocols (Shen et al., 2002). Briefly, in utero electroporated NPCs expressing ectopic Nin isoforms were dissociated, FACS sorted, plated at clonal density and their progeny pair classified as progenitors (Sox2+; Tuj1−), neurons (Sox2−; Tuj1+) or a mixture.