R1 mESCs were grown on mouse embryonic fibroblasts (MEFs) and maintained in DMEM (Thermo Fisher Scientific), 12.5% heat inactivated fetal calf serum (Fisher Biotec), 10 mM β-mercaptoethanol, 1x non-essential amino acids (Thermo Fisher Scientific), 1X nucleosides (Merck) and 1X leukemia inhibitory factor (LIF). Cells were passaged at 70% confluency, 2–3 days after seeding onto pre-plated MEFs. For chimera generation, mESCs were maintained in 2i/LIF media (Ying et al., 2008) for at least 2 passages before use.

For Lhx1 overexpression in chimeric embryos, two doxycycline inducible A2. loxCre mESC lines were used that either express a FLAG tagged wild-type Lhx1 or tagged truncated Lhx1 coding region lacking the functional LIM domains and homeodomain (Sibbritt et al., 2018).

Extraembryonic endoderm (XEN) cells were generated from blastocyst stage embryos as previously described (Niakan et al., 2013). ARC/s and DsRed.T3 mice (from the Australian Animal Resources Centre) were maintained as homozygous breeding pairs. ARC/s females were crossed with DsRed.T3 males, blastocyst stage embryos were collected and plated onto MEFs in TS cell medium; RPMI 1640 (Gibco), 20% fetal calf serum (Fisher Biotec), 2 mM l-glutamine (Gibco), β-mercaptoethanol, 1 mM sodium pyruvate (Gibco), 1% penicillin-streptomycin plus 24 ng/ml FGF4 (Sigma-Aldrich, cat. no. F8424) and 1 μg/ml heparin (Sigma-Aldrich, cat. no. H3393) for 20 days. The dsRed expressing XEN cells were then expanded on gelatin and maintained without FGF4 and heparin.

Foxd4 edited mESCs were generated as described previously (Sibbritt et al., 2019). Guide RNAs targeting the N-terminal region of Foxd4 were designed using (Benchling [Biology Software], 2021) (gRNA 1: 5′-CAG​TCC​TCT​AAG​TTC​CGA​CC, gRNA 2: 5′-GGA​GCG​ATC​CCT​GCA​GAG​GC) and ligated into pSpCas9(BB)-2A-Puro (PX459) V2.0 (a gift from Feng Zhang (Ran et al., 2013)). To induce editing, 5 × 10⁶ R1 mESCs were electroporated with 2.5 µg of plasmid DNA and plated onto mouse embryonic fibroblasts (MEFs) for 24 h before puromycin selection for 48 h. Individual clones were expanded on MEF coated plates and genotyped for correct edits in the Foxd4 coding region.

The genotyping PCR products were gel purified and sub-cloned into the pGEM-T Easy Vector System (Promega) as per manufacturer’s protocol. At least 10 plasmids from each cell line were Sanger sequenced to identify mutations in each allele.

Assemblies of mESCs and XEN cells (neuruloids) were generated as described previously (Bérenger-Currias et al., 2020) with some modifications. Approximately 2.5 × 10⁶ mESCs and 0.5 × 10⁶ XEN cells were mixed and placed in each well of a 24-well plate on 400 µm Aggrewells (Stem Cell Technologies) with 2 ml of N2B27 media, and then spun at 400 g for 3 minutes. The cells were cultured in Aggrewells for 48 h, and next transferred to low adhesion plates on a shaking platform with a 24-h pulse of 3 µm CHIR99021. CHIR99021 was then removed, and neuruloids were collected after a further 24 h of culture for RNA preparation and whole mount immunofluorescence microscopy.

Neural precursor differentiation of mESCs was initiated using embryoid bodies (EBs) as previously described (Varshney et al., 2017; Fan et al., 2021). After 4 days of differentiation, EBs were collected and plated on laminin (5 μg/ml) coated tissue culture plates in N2B27 media for a further 4 days of culture, then collected for RNA preparation or fixed in 4% paraformaldehyde for immunofluorescence imaging.

Chimeras were generated as previously described (Sibbritt et al., 2019; Fan et al., 2021). Briefly, ARC/s females were crossed with Ds. RedT3 stud males, at E2.5 the uteri and oviducts were flushed to collect 8-cell stage embryo collection. 13–15 mESCs were injected per 8-cell DsRed.T3 embryo, which were incubated overnight. 10 to 12 injected blastocyst-stage embryos were transferred to each E2.5 pseudo-pregnant ARC/s female recipient. E8.0—E11.5 embryos were collected 5–8 days after embryo transfer and imaged immediately on the Zeiss SteREO Lumar. V12 stereomicroscope to determine relative contribution of dsRed host cells versus injected mESCs. Relative intensity of dsRed.T3 fluorescence of each chimeric embryo was measured using ImageJ. The mean fluorescence of the dsRed.T3 channel was collated for each embryo and the background signal was subtracted. The mean fluorescence value was then displayed relative to embryo without ESC contribution at each stage. Animal experimentations were performed in compliance with animal ethics and welfare guidelines stipulated by the Children’s Medical Research Institute/Children’s Hospital at Westmead Animal Ethics Committee under protocol number C346.

Whole-mount immunostaining of chimeric embryos was performed as described in Fan et al. (2021), while immunostaining of neuruloids was performed as described in Dekkers et al. (2019). A list of antibodies and concentrations used are outlined in Supplementary Table S2. Embryos and neuruloids were imaged using Zeiss Cell Observer Spinning Disk Confocal Microscope. Three-dimensional images of the samples were produced using optical slices and tiling. Zeiss Zen microscopy analysis software was used to collapse the confocal stacks and stitch together tiles to generate maximum intensity projection (MIP) images.

Immunofluorescence imaging of neural precursor cells on glass cover slips was performed as described in Sibbritt et al. (2018) and imaged using the Zeiss Axio Imager M1 microscope.

RNA was extracted using the RNeasy Mini Kit (Qiagen) for cells and Rneasy Micro Kit (Qiagen) for embryos, according to manufacturer’s protocol. cDNA was synthesised from 1 µg of RNA (or 0.3 µg for E8.0 embryos) using the SuperScript III First-Stand Synthesis System (Invitrogen, Cat. No. 18080-051) as per the manufacturer’s protocol, using random hexamers to prime the single-stranded RNA. Unless otherwise stated, quantitative PCR (qPCR) primers were designed using Primer-BLAST to span exon junctions of the functional mRNA transcript (all qPCR primers are listed in Supplementary Table S1). PowerUp SYBR Green PCR Master Mix (Thermo Fisher Scientific) and 0.4 µM of both forward and reverse primers were made to a total volume of 10uL PCR reaction. Samples were loaded into a 384 well plate (Thermo Fisher Scientific) and run on the QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems). All reactions were performed in technical triplicates, relative gene expression was calculated using the comparative CT method, normalised to the housekeeping genes, Actb or Ubc.

Statistical significance was determined using an unpaired, two-tailed Student’s t-test, assuming unequal variances for single comparisons. p values were obtained relative to wild-type cells/chimeras if not indicated otherwise. Differences were considered significant if the *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

The processed read counts and metadata from Pijuan-Sala et al. (2019) were downloaded from https://github.com/MarioniLab/EmbryoTimecourse 2018. Reads were converted to a Seurat object and quality control of scRNA-seq data was performed with the Seurat package version 4.0.0 (Stuart et al., 2019) in R version 4.0.3. The data consisted of 29,452 genes with 139,331 single cells. The scater/Bioconductor package (McCarthy et al., 2017) was used to create QC metrics for the genes of interest. The dittoSeq package/Bioconductor (Bunis et al., 2020) was used for visualization of reduced dimension plots. Hierarchical cluster analysis was performed on a subset of cells expressing Foxd4 for each stage using hclust (Müllner, 2013) with parameters “complete” method and “Euclidean” distance.

A previous study has shown that conditional ablation of Lhx1 in the epiblast reduced the expression of Foxd4 in the anterior tissues of embryos (Sibbritt et al., 2018). Based on this finding, we hypothesise that the LHX1/FOXA2/OTX2 transcription factor complex drives the expression of Foxd4 in the AME at E7.75 and in the ANE at E8.25. We validated our hypothesis that LHX1 can affect transcription of Foxd4 using an Lhx1-overexpressing embryo model. Doxycycline inducible FLAG-Lhx1 and FLAG-Lhx1-Δ (lacking functional domains) mESC lines were used to generate mouse chimeras (Figure 1A). Chimeras with high mESC contribution were collected at E7.75 following 24h of doxycycline treatment. Expression of wild-type Lhx1 mRNA was 60-fold higher in FLAG-Lhx1 vs FLAG-Lhx1-Δ (Figure 1B). A significant increase in Foxd4 transcripts (Figure 1C) indicates that enhanced LHX1 activity affected the expression of Foxd4.

Using the publicly available eGastrulation spatial transcriptome dataset (Peng et al., 2019), we are able to investigate the location and relative level of Foxd4 expression in the late gastrulation stage mouse embryo. The highest level of Foxd4 expression can be seen in the anterior midline cell population and the neurectoderm (Figure 1C), consistent with previous in situ hybridization data (Tamplin et al., 2008). The spatial expression of Foxd4 in late gastrulation mouse embryos overlaps with the known locations of genes that have been shown to be critical for embryonic head development such as Lhx1, Foxa2, Otx2 and Hesx1.

To investigate the expression of these transcription factors at higher resolution, we used previously published single-cell RNA-seq data of wild-type mouse embryos (Pijuan-Sala et al., 2019). At E7.75 Foxd4 is highly expressed in the notochord cell lineage (Supplementary Figure 1A), which gives rise to the midline mesendoderm tissues (Yamanaka et al., 2007). At this stage Lhx1 and Foxa2 share similar expression profiles (Supplementary Figure 1A) and our analysis identified co-expression of Foxd4 with either Lhx1 or Foxa2 expressing cells in the notochord and definitive endoderm populations (Figure 1D). At E8.25 Foxd4 is highly expressed in rostral neurectoderm and forebrain/midbrain/hindbrain cell populations. Similarly, the head organizer genes Otx2 and Hesx1 are expressed in these cell populations (Supplementary Figure 1A). Our analysis highlighted several groups of cells that share co-expression of Foxd4, Otx2 and Hesx1 at E8.25 (Figure 1D). Foxd4 expression in the whole embryo is increasing at E7.5 and peaks at E8.0 (Figure 1E).

To elucidate if these transcription factors bind to the regulatory region of the Foxd4 locus in mouse cells we retrieved the binding data from publicly available ChIP-seq dataset. LHX1 ChIP-seq data in differentiated P19 carcinoma cells shows a low confidence peak ∼1k bp upstream of the Foxd4 transcriptional start site (TSS) (Figure 1F) (Costello et al., 2015). ChIP-seq data of OTX2 in epiblast like-cells (Buecker et al., 2014) and FOXA2 in mesendoderm cells (Cernilogar et al., 2019) show high confidence peaks in the same locus on chromosome 19: 24,902,170–24,902,900 (mm10 genome). These data suggest the binding of a LHX1/OTX2/FOXA2 transcription factor complex upstream of the Foxd4 TSS.

In Foxd4/5, the Foxd4 paralog in Xenopus, the AB domain has been shown to be a transcriptional activator of neural transcription factors (Neilson et al., 2012). We targeted this AB domain region in mESCs with CRISPR-Cas9 mediated genome editing (Figure 2A). Following screening, we chose clones with a bi-allelic frameshift mutation in the N-terminal region of Foxd4, at the beginning of the AB domain (Foxd4 ^Δ7/Δ8 , Foxd4 ^Δ2/Δ2 (Figure 2A, Supplementary Figure 2A). We also used a different gRNA targeting the region between the AB and forkhead domain of Foxd4, to exclude off-target effects of CRISPR-Cas9 genome editing. Using the second gRNA we obtained a clone with a 2 bp deletion and a 1 bp insertion in respective alleles (Foxd4 ^Δ2/Δ1 , Supplementary Figure 2D). Despite trying numerous antibodies from different manufacturers, we were unable to get a reliable signal to assay the expression of the predicted truncated FOXD4 protein in our knockout mESC line (data not shown).

To study the downstream genetic targets of FOXD4 during development, we used an in vitro model of the anterior epiblast (neuruloid) generated through the co-culture of mESCs and extraembryonic endoderm (XEN) (Figure 2B) (Bérenger-Currias et al., 2020). The XEN cells express genes that are highly expressed in the extraembryonic endoderm including Foxa2, Sox17 and Gata4, but do not express pluripotency markers, Oct4 and Sox2 (Supplementary Figure 3A, B). The dsRed expressing XEN cells colonized the exterior portion of the neuruloid (Figure 2C, Supplementary Figure 3C), where they may act in a similar way to the anterior visceral endoderm population in the embryo. In contrast to embryoid bodies differentiated for the same period, we showed significantly higher expression of anterior markers Otx2, Lhx1, Hesx1 as well as Foxd4 (Supplementary Figure 3D). Wild-type neuruloids expressed early neurectoderm marker SOX1 in distinct regions of the neuruloids (Figure 2D, Supplementary Figure 3E).

Compared with neuruloids generated using wild-type mESCs, Foxd4 ^Δ7/Δ8 neuruloids had significantly reduced expression of Foxd4 transcripts (Figure 2E). Lhx1, Hesx1 and Foxa2 transcripts were also significantly reduced (Figure 2F). This result indicates that FOXD4 is crucial for the appropriate specification of the precursor tissues to the embryonic head and notochord. Comparable to results seen in Xenopus (Neilson et al., 2012), knock-out of Foxd4 caused the reduction in expression of neural ectodermal genes Zic1 and Zic2 (Figure 2F). Expression of the neural progenitor gene Sox2 was not changed in Foxd4 ^Δ7/Δ8 neuruloids, whereas the transcripts of early neural crest cell (NCC) marker Sox9 was significantly reduced, indicating a role for Foxd4 in the establishment of the NCC population. Our analysis of scRNA-seq data from Pijuan-Sala et al. (2019) showed Foxd4 is co-expressed with Zic2 and Sox9 but not Zic1 at E7.75 and E8.25 in the ANE tissues (Supplementary Figure 1B).

To analyze the function of FOXD4 during mouse development, wild-type or Foxd4-LOF mESCs were injected into 8-cell host embryos ubiquitously expressing dsRed (Figure 3A). Host embryos were injected with either wild-type or Foxd4 ^Δ7/Δ8 mESCs (15 embryos each), chimeric embryos were collected at E8.0 and the relative contribution of mESCs was quantified using fluorescence microscopy. Three chimeras of each genotype with high (>60%) contribution were kept for RNA assay (Supplementary Figure 4A). E8.0 chimeras that showed high contribution of mESCs in fluorescence imaging had significantly lower dsRed expression compared to un-injected embryos (Figure 3B). Chimeras with high contribution of Foxd4 ^Δ7/Δ8 mESCs showed significantly reduced Foxd4 expression compared to wild-type mESC injected chimeras (Figure 3B).

Foxd4-LOF chimeric embryos had visible neural tube closure defects and truncated forebrain tissue at E9.5, E10.5 and E11.5, whilst none of the wild-type mESC derived chimeras that were collected displayed an abnormal head phenotype (Figure 3C). At E9.5, 3/13 Foxd4 ^Δ7/Δ8 chimeras displayed anterior defects, compared to 0/9 for wild-type chimeras (Supplementary Figures 5A,B and 6A,B). No anterior defects were evident in E10.5 wild-type chimeras (0/12) (Supplementary Figure 7A,B). 14/18 Foxd4 ^Δ7/Δ8 , 2/2 Foxd4 ^Δ2/Δ2 , and 2/4 Foxd4 ^Δ2/Δ1 chimeras collected at E10.5 displayed anterior developmental defects (Supplementary Figures 2C,F and 8A,B). Finally, 0/6 wild-type E11.5 chimeras and 2/8 E11.5 Foxd4 ^Δ7/Δ8 chimeras showed anterior defects (Supplementary Figure 9A,B).

Foxd4 ^Δ7/Δ8 chimeras collected at E9.5 with high contribution had comparable expression of neuroectoderm marker SOX1, though displayed severe truncation of the head tissue compared to wild-type control (Figure 3D, Supplementary Figure 4B). At E10.5 a range of head defect phenotypes were evident in Foxd4 ^Δ7/Δ8 , Foxd4 ^Δ2/Δ2 and Foxd4 ^Δ2/Δ1 chimeras (Figure 3E, Supplementary Figure 2C,F). NEFM (neurofilament) staining shows exencephaly in the midbrain and hindbrain, though there were no defects in the caudal neural tube in any Foxd4-LOF chimeras (Figure 3F). Craniofacial defects were also common among Foxd4-LOF chimeras including truncated facial tissue and abnormal forebrain patterning (Figures 3E,G, Supplementary Figure 4D).

In Foxd4 ^Δ7/Δ8 chimeras collected at E11.5, exencephaly was evident in the rostral neural tube (Figures 3H,I, Supplementary Figures 4E,F). The protein OTX2 that is normally expressed in the midbrain and eyes of E11.5 wild-type embryo, was not detected in Foxd4 ^Δ7/Δ8 chimeras (Figure 3H). All the defects seen were in anterior head and neural tube tissues, indicating the specific role of FOXD4 in the anterior neural and midline tissue in late gastrulation/early organogenesis.

We adapted a protocol from Varshney et al. (2017) for the differentiation of mESCs to neural precursor cells (NPC) (Figure 4A), revealed by high levels of Foxd4 expression in wild-type NPCs at Day 8 of culture (Figure 4B). NCC markers Twist1 and Sox9 were also highly expressed in the NPCs compared to undifferentiated mESCs (Figure 4B). Both wild-type and Foxd4 ^Δ7/Δ8 mESCs expressed a high level of neural ectoderm marker SOX1 and neuron specific Class III β-tubulin (TUBB3) (Figure 4C). Foxd4 ^Δ7/Δ8 day 8 NPCs had significantly reduced Foxd4 mRNA expression compared to the wild-type (Figure 4D). mRNA expression of neurectoderm markers Pax4 and Nestin (Nes) were not significantly different in Foxd4 ^Δ7/Δ8 NPCs (Figure 4D). In contrast, the head organizer genes Otx2, Lhx1 and Foxa2 were significantly reduced (Figures 4E,F). The loss of Otx2 expression is consistent with the reduction in OTX2 expression seen in the in vivo model (Figure 3H). NCC markers Twist1 and Sox9 were also downregulated in Foxd4 ^Δ7/Δ8 NPCs (Figure 4F) compared to wild-type, though our scRNA-seq analysis did not show significant co-expression of Foxd4 and Twist1 at E7.75 or E8.25 (Figure 1B).