Nervous System Regionalization Entails Axial Allocation before Neural Differentiation

Abstract

Neural induction in vertebrates generates a CNS that extends the rostral-caudal length of the body. The prevailing view is that neural cells are initially induced with anterior (forebrain) identity; caudalizing signals then convert a proportion to posterior fates (spinal cord). To test this model, we used chromatin accessibility to define how cells adopt region-specific neural fates. Together with genetic and biochemical perturbations, this identified a developmental time window in which genome-wide chromatin-remodeling events preconfigure epiblast cells for neural induction. Contrary to the established model, this revealed that cells commit to a regional identity before acquiring neural identity. This "primary regionalization" allocates cells to anterior or posterior regions of the nervous system, explaining how cranial and spinal neurons are generated at appropriate axial positions. These findings prompt a revision to models of neural induction and support the proposed dual evolutionary origin of the vertebrate CNS.

Keywords: ATAC-seq; CDX; WNT signaling; chromatin; computational genomics; embryonic development; gene regulation; neural induction; spinal cord; stem cells and development.

Graphical abstract

Figure 1

Regulatory Element Usage Distinguishes Cell State during Neural Induction (A) Schematic of mouse ESCs differentiated to NPs with anterior (top), hindbrain (middle), or spinal cord (bottom) identity. Spinal cord progenitors are generated via an NMP state induced by the addition of FGF and WNT signals from day (D) 2 to 3 (light pink shading). (B) D5 immunofluorescence reveals hindbrain progenitors generate a mixture of PHOX2B expressing visceral and OLIG2 expressing somatic MNs. Spinal cord progenitors lack visceral but generate OLIG2 expressing somatic MN progenitors. Scale bars represent 20 μm. (C and D) ATAC-seq accessible regions present in ESCs (D0, gray) compared with D5 anterior (D5A; blue), hindbrain (D5H; yellow), and spinal cord (D5SC; red) progenitors and associated gene expression levels determined by mRNA-seq (Gouti et al., 2014; error bars = SEM). cis interactions indicated below each plot represent known genomic interactions from published data (Table S2). ESCs express Oct4 and show accessibility at Pou5f1/Oct4 enhancers (C, arrow). D5H and D5SC NPs have open regions flanking neural expressed Olig2 (D, arrow). (E) Genome-wide accessibility comparison in D5 spinal cord (D5SC) versus D0 ESCs (false discovery rate [FDR] < 0.01 and |log2(FC)| > 1). (F) The proportion of differential sites present in each condition compared with ESCs. (G) Both neural and AP-specific sites, but not ESC sites, are enriched in H3K27ac marks from NPs (Peterson et al., 2012). bFGF, basic fibroblast growth factor; D, day; ESC, embryonic stem cell; FC, fold change; kbp, kilobase pairs; RA, retinoic acid; SHH, sonic hedgehog; TPM, transcripts per million.

Figure S1

Quality Control for All ATAC-Seq Samples Generated in This Study, Related to STAR Methods and Figure 1 (A) The proportion of mitochondrial fragments recovered across each sample. (B) Representative example showing the distribution of fragment lengths recovered from ATAC-seq, using paired-end sequencing. (C) Average level of Tn5 enrichment (score = maximum(number of insertions)/minimum(number of insertions)) observed across transcription start sites (TSS) for each sample. (D) Summarized Tn5 insertion profile covering ± 2kb around annotated TSS for sample D5H (replicate 2). Red line corresponds to a 50bp running average. (E) The fractions of fragments that map to in vitro consensus peak regions. (F) CTCF footprint present in ESC accessible regions as determined by ATAC-seq.

Figure 2

Differential Enhancer Usage Reveals NP Axial Identity (A) Self-organizing map (SOM) of regulatory regions showing differential accessibility relative to D0. Each plot represents the chromatin accessibility Z score for each genomic region in the cluster for each condition (see key in A’). Many sites are common (“neural sites”) to all NPs (black cluster; n = 5,584). These differ from epiblast regulatory regions that are accessible at early stages of the differentiation (Epi; green; n = 1,714). Region-specific sites are identified in anterior (blue; n = 1,863), hindbrain (orange; n = 2,509), and spinal cord (red; n = 1,538) progenitors. A distinct set of regulatory regions identifies D3 NMPs (pink; n = 454 regions). A/H represents shared anterior and hindbrain accessible sites (purple; n = 1,276); H/SC, shared hindbrain and spinal cord sites (lime; n = 1,840); and NMP/SC, shared NMP and spinal cord sites (brown; n = 421). Grey are unclassified sites. (B–D) Examples of ATAC-seq accessible regions that define anterior (B), hindbrain (C), or spinal cord (D) D5 progenitors, identified using the SOM (A). Gene expression from mRNA-seq (error bars = SEM) is shown to the right. Anterior progenitors display region-specific open sites at Shh (B), and hindbrain progenitors demonstrate a Phox2b site (C) and a Hoxc8 site opens in spinal cord (D). (E–G) In vivo ATAC-seq correlates with in vitro. NPs obtained from brain (E; blue shading) and spinal cord (E; red shading) of E9.5 Sox2eGFP embryos. The fold change in accessibility at anterior (blue; n = 1,863) and spinal cord (red; n = 1,538) sites identified in vitro in spinal cord NPs relative to anterior NPs in vivo correlates with AP identity (F). Common neural sites in vitro (black) are similar in both populations in vivo. Anterior sites identified in vitro show preferential accessibility in vivo in anterior NPs (G), and spinal cord in vitro sites show more accessibility in vivo in spinal cord NPs (p values; Wilcoxon rank-sum test). (H–J) ChIP-seq enrichment analysis in anterior (H), hindbrain (I), and spinal cord sites (J). SOX2 ChIP-seq in D5 hindbrain (D5H) and spinal cord (D5SC) cells reveals that the binding site preference is condition specific (I and J). CDX2^∗ denotes CDX2 ChIP-seq performed in the presence of FGF signaling (Mazzoni et al., 2013). FPM, fragments per million; neural EB, embryoid bodied-derived NPs; NMP, neuromesodermal progenitors; NP, neural progenitors; NT, neural tube; pMN, motor neuron progenitors. See also Figure S2.

Figure S2

Tissue Specificity and Genomic Location of Regulatory Regions that Define Neural and Region-Specific Identity, Related to Figure 2 (A–D) Comparison of ATAC-seq identified regions with previously published DNase hypersensitivity sites present across a range of in vivo tissues and time points from the ENCODE regulatory element database (ENCODE Project Consortium, 2012, Sloan et al., 2016). Genomic regions correspond to neural (A), anterior (B), hindbrain (C) and spinal cord (D) specific sites from Figure 2A. Each set of genomic regions demonstrates an enrichment in embryonic and neural samples in vivo. (E) Comparison of ATAC-seq identified regions with the Vista enhancer database (Visel et al., 2007) shows that accessible regions correspond to enhancers that have neural tissue specificity in vivo. (F–I) Classification of neural (F), anterior (G), hindbrain (H) and spinal cord (I) sites according to genomic position. Neural sites are enriched at promoter regions (F), in contrast to the region-specific sites, which predominantly occupy distal intergenic and intronic regions (G-I). p values determined using one-sided Binomial test and multiple testing corrected using Benjamini–Hochberg procedure. (J and K) In vivo-derived neural progenitors display accessibility at known enhancers depending on their axial identity. Genome browser view (mm10 assembly) showing ATAC-seq from anterior (blue track) and spinal cord (red track) neural progenitors obtained from E9.5 mouse embryos at the Shh (J) and Olig2 (K) locus. Arrows indicate known enhancers that direct Shh expression in the midbrain (Epstein et al., 1999; J) and Olig2 in the spinal cord (Oosterveen et al., 2012 and Peterson et al., 2012; K). Gene expression levels determined by mRNA-seq (Gouti et al., 2014) are shown as bar plots from in vitro D5 anterior (blue) and spinal cord (red) conditions (error bars = SEM). Chromatin interactions recovered from indicated tissues are presented below for comparison. Peak regions are represented with black bars. A = anterior neural progenitor; DR = dorsal root; NP = neural progenitor; NSC = neural stem cell; SC = spinal cord progenitor; TPM = transcripts per million.

Figure S3

Motif Analysis of Region-Specific Sites that Define Anterior, Hindbrain, and Spinal Cord, Related to Figure 2 (A–C) Motif enrichment analysis performed using Homer on anterior (A), hindbrain (B) and spinal cord (C) specific sites shows distinct and common neural factors enriched at each AP level. For each factor (indicated on the left), the corresponding expression level determined by mRNA-seq (Gouti et al., 2014) in the same condition at D5 of the in vitro differentiation is shown (central column; error bars = SEM). The top 6 predicted motif logos are presented on the right. TPM = transcripts per million.

Figure 3

Axial Identity Is Established in Cells prior to Neural Identity (A–C) The average accessibility (Z score) of region-specific sites over time in anterior (labeled “A”), hindbrain (labeled “H”), or spinal cord (labeled “SC”) conditions. AP-specific sites become accessible between D3 and D4. Spinal cord progenitors do not transiently open sites corresponding to anterior (A) or hindbrain (B) identity before opening spinal cord-specific sites (C). (D) Neural sites become accessible in all regions at the same time. Error bars = SD. (E) Schematic of the differentiation (H+ condition). (F–N) qRT-PCR of genes at D3 and D5 following the differentiation of cells to hindbrain (D5H), spinal cord (D5SC), or “hindbrain+” (D5H+) identity. The WNT target Notum (F) is observed following WNT signaling treatment at D3 (D3NMP) and D5 (D5H+). Induction of posterior spinal cord Hox genes Hoxb9 and Hoxc8 is dependent on timing: induction in D3NMP follows D2 to D3 treatment with WNT signals, but not at D5 in D5H+ cells following D4 to D5 treatment with the same signals (G and H). Induction of T/Bra and Cdx2 is dependent on timing, responding to early (D2 to D3), but not late (D4 to D5), treatment with WNT signals (I and J). Late treatment of WNT in the D5H+ condition prevents expression of ventral neural genes Phox2b and Olig2 (K and L, compare D5H to D5H+) while dorsal Pax7 (M) and intermediate Dbx1 (N) neural tube genes are induced. Error bars represent the standard deviation. (O) SOX1 immunofluorescence on D3 versus D4 cells cultured in hindbrain (D3A and D4H) or spinal cord (D3NMP and D4SC) conditions. Scale bars represent 20 μm. (P) Sox1 expression, detected by mRNA-seq (Gouti et al., 2014), at the indicated times and conditions. Error bars = SEM.

Figure S4

Expression Dynamics of Cdx TFs during the Spinal Cord Differentiation, Related to Figure 4 (A) Scatterplot showing the fold change of D5H+ or D5H cells relative to the D5SC condition. Note that the majority of hindbrain sites (yellow) remain accessible in D5H+ treated cells, in contrast to spinal cord (red) sites. (B) The distribution of genomic regions, as defined in the self-organizing map (SOM; see Figure 2A), in D5H+ cells that show changes in accessibility when compared to D5H cells (FDR < 0.01 & |log2(FC)| > 1). Note that D5H+ cells do not gain sites associated with D3NMP, which are treated with WNT between D2-3 (see columns labeled “NMP”), but gain additional sites not classified in the SOM (see columns labeled “D5H+ specific). (C) Motif enrichment using iCis Target (Imrichová et al., 2015) on genomic regions that are differentially accessible between D5H and D5H+ reveals an enrichment of TCF/Lef factors following WNT treatment in D5H+ cells, in contrast to the D5H condition which harbours sites enriched in SOX factors. (D) Confocal microscopy of hindbrain and spinal cord cells from D2-4 shows the induction of CDX2 at D3 in FGF/WNT in the spinal cord condition (D3NMP). SOX1 expression, marking neural progenitors, is not detected until D4 in both conditions, following RA and SHH treatment. Right column shows CDX2/SOX2 composites. (E) Flow cytometry in WT (blue) versus Cdx triple mutant ESCs (grey) at the indicated time points and conditions indicates the percentage of SOX2 positive cells that express CDX2. (F) Expression profile determined by mRNA-seq (Gouti et al., 2014) for Cdx1, Cdx2 and Cdx4 from D0 to D5 of the spinal cord differentiation. Error bars = SEM. NMP=neuromesodermal progenitor; SC=spinal cord; TPM=transcripts per million.

Figure 4

WNT Establishes Spinal Cord Identity via CDX-Dependent Chromatin Remodeling (A) Proportion of NMP, NMP/SC, or epiblast genomic sites from the SOM (Figure 2A) that overlap with accessible regions in vivo (Neijts et al., 2016; this study). 71% of all NMP sites identified in vitro are found in the posterior E7.5 embryo (E7.5-P; Neijts et al., 2016). This contrasts with NPs from the spinal cord (E9.5-SC) or anterior nervous system (E9.5-A; this study), which show little overlap with these sites. (B) The average accessibility profile of NMP/SC and NMP-specific sites in wild-type versus T/Bra^−/− mutant cells. These sites remain accessible in T/Bra^−/− mutant cells at D3 of the spinal cord differentiation. (C) T/Bra^−/− mutant cells differentiated to D5 under spinal cord conditions retain accessibility at spinal cord genomic sites. (D) Heatmap showing NMP (top panels) and NMP/SC (bottom panels) site accessibility in D3NMP conditions from WT, T/Bra^−/−, and Cdx1/2/4 (Cdx^−/−) mutant cells. These sites are maintained in the absence of T/Bra but are reduced in the absence of the three Cdx TFs. (E–H) qRT-PCR of Hox genes at D5 indicates AP identity of hindbrain (D5H) and spinal cord (D5SC) cells in wild-type compared with T/Bra^−/− and Cdx^−/− mutant cells differentiated under spinal cord conditions. Bra^−/− mutant cells retain expression of spinal cord Hox genes Hoxb9 (E) and Hoxc6 (F) in contrast to Cdx mutant cells, which express Hoxb4 (G) and Hoxc4 (H). Error bars represent the standard deviation. (I) ChIP-seq enrichment analysis reveals that CDX2 is highly enriched at NMP-specific sites (p values; one-sided Fisher’s exact test; multiple testing corrected using Benjamini-Hochberg procedure). (J) Tn5 insertion frequency, across all SOM regions containing at least one CDX2 motif, at nucleotide resolution in D3NMP cells reveals the presence of a footprint centered on the CDX2 motif. (K) CDX2 (cyan) and Sox2 (red) immunofluorescence at D3. Scale bars represent 20 μm. (L) Histogram of CDX2-positive cells at D3 comparing WT (blue) with Cdx^−/− mutant cells (gray), determined by flow cytometry. Numbers indicate percentage of SOX2-positive cells that are CDX2 positive. (M) Removal of the three Cdx transcription factors Cdx1/2/4 results in the continued induction of OLIG2 (gold) and ectopic production of cranial MN progenitors, marked by PHOX2B (magenta). Scale bars represent 20 μm. (N) The average profile of spinal cord sites (left plot) shows that, relative to D5 spinal cord (D5SC, red), accessibility at these sites is reduced in Cdx^−/− mutant cells differentiated under the same conditions (D5SC Cdx^−/−, green), to the same extent as D5 hindbrain cells (yellow). Under spinal cord conditions, Cdx mutant cells show increased accessibility at hindbrain sites (right plot).

Figure 5

CDX2 Can Replace WNT and Prolong Spinal Cord Competency (A) Schematic of the differentiation using iCDX2 ESCs (Niwa et al., 2005) to induce CDX2 between D2 and D3 (SCind). (B) Immunofluorescence of NPs at D5 PHOX2B (cyan) and SOX2 (red). Cranial MNs in hindbrain, but not spinal cord or SCind. (C) qRT-PCR analysis at D5 shows that the induction of CDX2 between D2 and D3 maintains Olig2 expression and upregulates Hoxb9 and Hoxc6. Error bars represent the standard deviation. (D) Chromatin accessibility, measured by ATAC-seq, at hindbrain (yellow) and spinal cord (red) sites at D5. Spinal cord sites are more open in WT spinal cord cells than WT hindbrain cells (left plot). The induction of CDX2 between D2 and D3 (D5SCind) increases accessibility at spinal cord sites versus D5H WT cells (middle plot) and similar levels of accessibility in hindbrain and spinal cord sites when compared to D5SC WT cells (right plot). (E) Schematic of the differentiation using iCDX2 ESCs to induce CDX2 between D4 and D5 under hindbrain conditions (Hrep) versus hindbrain+ (H+) conditions, in which WNT signaling is activated between D4 and D5. (F and G) qRT-PCR data indicate that, by D5, the induction of CDX2 is sufficient to repress Phox2b but maintain Olig2 (F), in contrast to H⁺ cells, which repress Olig2 expression (Figure 3). The induction of CDX2 between D4 and D5 upregulates posterior Hox genes Hoxb9, Hoxb8, and Hoxc6 (G). Error bars represent the standard deviation. (H) Accessibility at hindbrain (yellow) and spinal cord (red) sites reveals that D5Hrep cells display increased accessibility at spinal cord sites and loss of hindbrain sites compared to D5H cells. (I and J) WNT reporter embryos cultured for 14 hr in control versus WNT signaling conditions (Table S3) from E7.0 (I) or E7.5 (J). Images show embryos cultured in media containing bFGF (control) versus FGF and CHIR99021 (FGF/WNT) conditions. Embryos are oriented with anterior at the top of the image. Dashed white lines demarcate the anterior limit. (I) Ventral view of E7.0 cultured embryos. Ectopic induction of WNT activity (n = 10/13) and CDX2 (green, white bracket) is observed in the presence of WNT signaling (n = 10/13), but not control conditions (n = 0/10). SOX2 marks the epiblast (red). Asterisk demarcates the position of the node. Scale bars represent 250 μm. (J) Ectopic induction of WNT signaling in E7.5 cultured embryos (n = 13/16) versus control conditions (n = 0/6). No CDX2 (green) expansion was detected in the anterior neural plate (n = 0/31) marked by SOX1 (red) versus control conditions (n = 0/30). Top panels in (J) show ventral views; bottom panels in (J) show dorsal views. Scale bars represent 250 μm. (K) Chick whole-mount in situ hybridization for Cdx2 following 10-hr ex vivo embryo culture from HH stage 3+ (top panels) and stage 7 (bottom panels). The addition of WNT signals promotes ectopic (white arrows) anterior Cdx2 expression at early stages (15/15 in the FGF/WNT versus n = 0/13 control at stage 3+). By contrast, no expansion is observed in response to WNT signaling in stage 7 embryos that already contain a neural plate (n = 0/9 in FGF/WNT and n = 0/12 in control). White arrowheads mark the anterior limit of Cdx2 expression. Scale bars represent 500 μm. cc, cardiac crescent; mb, midbrain; nt, neural tube; p, posterior.

Figure 6

Proposed Model of Nervous System Development (A) Pluripotent epiblast cells in the early embryo are first allocated into anterior (blue) or posterior (red) populations before acquiring neural identity. Posteriorized cells form spinal cord; anterior epiblast cells generate the anterior nervous system. (B) Comparisons between cnidarian and bilaterian animals provide support for the dual evolutionary origin of the vertebrate CNS (Arendt et al., 2016). Cnidarians display two distinct nerve centers: apical (blue) and blastoporal (red). Blastoporal centers show expression of putative CDX orthologs (Arendt et al., 2016, Ryan et al., 2007). In bilaterians, these separate nerve centers are proposed to have expanded and merged.

Figure S5

CDX2 Occupancy in Open Chromatin Sites and Associated Gene Ontology Enrichment, Related to Figure 4 (A) The proportion of accessible regions bound by CDX2, as indicated by CDX2 ChIP-seq analysis from neuromesodermal progenitors (NMP, light blue bars, Amin et al., 2016) and motor neuron progenitors (pMNs, dark blue bars, Mazzoni et al., 2013) derived in vitro, compared with the accessible regions recovered from the self-organizing map (SOM) in this study (refer to Figure 2A). The overlap demonstrates that CDX2 binds to NMP, NMP/SC (NMP and spinal cord shared) and spinal cord (SC) sites identified by ATAC-seq. Furthermore, in pMN conditions, CDX2 binds accessible regions that are shared between the hindbrain and spinal cord (A’, boxed region outlined in green). CDX2 also targets hindbrain accessible sites (A’’). (A’) The Phox2b genomic region represents a shared hindbrain/spinal cord accessible site that is bound by CDX2 in pMN conditions. (A’’) A hindbrain-accessible site is bound by CDX2 at the Mafb locus in pMN conditions. (B) Region heatmap showing CDX2 binding at open chromatin sites recovered from the SOM (pMN; Mazzoni et al., 2013). (C–F) Gene ontology enrichment analysis for CDX2-bound regions shown in (B). In hindbrain accessible regions (D), CDX2 binding is associated with neural genes in contrast to either the NMP and spinal cord (NMP/SC) shared (E) or SC-specific sites (F), which target genes involved in anterior-posterior patterning. (G) Comparison of log2 fold gene expression changes in D5 spinal cord (D5SC) versus D5 hindbrain (D5H), determined by mRNA-seq (Gouti et al., 2014), and wild-type (WT) versus Cdx2-induced motor neuron progenitors (iCdx2-pMNs) determined by microarray (Mazzoni et al., 2013). CDX2 induction positively correlates with Hoxb9 and other 5′ Hox genes while it negatively correlates with Aldh1a2 in the spinal cord, in agreement with previous studies (Gouti et al., 2017). CDX negatively correlates with hindbrain genes including Phox2b. Color filling indicates –log10(adj. pvalue) from the D5SC versus D5H comparison using DESeq2 (Love et al., 2014). (H) Motif enrichment analysis of CDX2-bound regions shown in (B). Heatmap colors represent the normalized enrichment score computed using iCis Target (Imrichová et al., 2015). CDX2 binds to hindbrain accessible regions that are enriched with SOX factor motifs, in contrast to HOX motifs found in spinal cord sites.