To determine if Eed is required for the development of neural crest derivatives, homozygous floxed Eed alleles (Eed^Fl/Fl) (Yu et al., 2009) were used to conditionally delete Eed following neural crest cell induction using Cre recombinase under the control of the Sox10 promoter, which is expressed in the migratory neural crest at embryonic day (E)8.75 and results in complete recombination in post-migratory neural crest cells by E10.5 (Matsuoka et al., 2005; Niethamer et al., 2020). Sox10-Cre; Eed^Fl/Fl mice were not recovered past postnatal day 0, suggesting that loss of Eed following induction of the neural crest is embryonic lethal (Figure 1a). However, Sox10-Cre; Eed^Fl/Fl embryos were recovered at expected genotypic frequencies from E9.5 to E17.5 (Figure 1a), and there were no differences in the penetrance or severity of Sox10-Cre; Eed^Fl/Fl phenotypes whether Cre was maternally or paternally inherited, as has been reported for other phenotypes arising in cells expressing Sox10-Cre (Crispino et al., 2011; Luo et al., 2020). No overt phenotypes were identified at early post-migratory neural crest cell stages E10.5 or E11.5, but craniofacial malformations were seen in Sox10-Cre; Eed^Fl/Fl embryos starting at E12.5 that increased in severity throughout the remainder of embryonic development (Figure 1b). Sox10-Cre; Eed^Fl/Fl phenotypes were broadly consistent with impaired development of craniofacial structures (Ferguson et al., 2018; Schwarz et al., 2014), including frontonasal and mandibular hypoplasia, a prominent telencephalon and exencephaly resulting from underdevelopment of the viscerocranium and frontal calvarium, collapsed nasal cavity bones, and microtia compared to controls (Figure 1c). Sox10-Cre; Eed^Fl/Fl embryos had increased pupillary distance and craniofacial width (Figure 1—figure supplement 1a–d), and whole-mount DAPI imaging of Sox10-Cre; Eed^Fl/Fl embryos revealed severe frontonasal dysplasia with underdeveloped midface and mandible, and an irregular and corrugated facial structure compared to controls (Figure 1d). Although PRC2 is ubiquitously expressed in the developing embryo (O’Carroll et al., 2001), Sox10-Cre; Eed^Fl/Fl embryos did not have cardiac outflow tract defects or other structural heart malformations that can arise from impaired neural crest differentiation (Lindsay et al., 2001; Lindsay et al., 1999; Nakamura et al., 2009; Figure 1—figure supplement 2a–d; Videos 1 and 2). Instead, fetal echocardiography showed subtle structural changes but preserved ejection fraction and fractional shortening in Sox10-Cre; Eed^Fl/Fl embryos compared to controls (Figure 1—figure supplement 2d).

Skeletal stains, micro-computed tomography (microCT), histology, and immunofluorescence were used to define how loss of Eed impacts craniofacial development. Whole-mount Alcian blue/Alizarin Red staining of Sox10-Cre; Eed^Fl/Fl embryos revealed hypoplasia of the viscerocranium, including reduced frontal, temporal, maxillary, and mandibular bones, and complete loss of the tympanic ring and premaxillary and nasal bones compared to controls (Figure 2a,b, Figure 2—figure supplement 1a). MicroCT validated these findings, showing fragmentation and hypoplasia of the frontal, temporal, maxillary, and mandibular bones, loss of frontal calvarial fusion, and absence of tympanic ring and nasal bones (Figure 2c–e and Videos 3 and 4). H&E histology showed that frontal calvarium reduction resulted in anteriorly displaced brain structures in Sox10-Cre; Eed^Fl/Fl embryos compared to controls, with the midbrain and cortex in the same coronal plane as the nasal cavity and developing mandible (Figure 3a, Figure 2—figure supplement 1b). Most of the midfacial structures and mandible were absent in Sox10-Cre; Eed^Fl/Fl embryos, and the tongue and masseter muscles were underdeveloped compared to controls (Figure 3a).

Immunofluorescence demonstrated disorganized ALPL and Osteocalcin, markers of osteoblasts and mineralizing bone (Liu et al., 2018), in craniofacial tissues from Sox10-Cre; Eed^Fl/Fl embryos compared to controls (Figure 3b, Figure 2—figure supplement 1c). Immunofluorescence for Sox9, a marker of chondrocytes that are derived from the neural crest (Mori-Akiyama et al., 2003; Yan et al., 2002), was also disorganized in Sox10-Cre; Eed^Fl/Fl embryos compared to controls (Figure 3b). There was a small increase in immunofluorescence labeling index for cleaved Caspase 3 (Figure 2—figure supplement 1d and e), suggesting that apoptosis may play a subtle role in craniofacial phenotypes from Sox10-Cre; Eed^Fl/Fl embryos compared to controls. In contrast to Wnt1-Cre Ezh2^Fl/Fl embryos (Ferguson et al., 2018; Schwarz et al., 2014), the craniofacial region of Sox10-Cre; Eed^Fl/Fl embryos had marked decreased immunofluorescence labeling index for Ki67 (Figure 3c and d), a marker of cell proliferation (Gerdes et al., 1984), and decreased immunofluorescence staining for Vimentin (Vim), a marker of mesenchymal cells (Mendez et al., 2010; Figure 3c). Immunofluorescence for Runx2, a transcription factor that is required for osteoblast differentiation and proliferation (Kawane et al., 2018; Komori, 2009), was also decreased in craniofacial tissues from Sox10-Cre; Eed^Fl/Fl embryos compared to controls (Figure 3e, Figure 2—figure supplement 1f). In support of these findings, (1) BrdU labeling index was decreased in the craniofacial region of Sox10-Cre; Eed^Fl/Fl embryos compared to controls (Figure 3—figure supplement 1a and b), (2) immunofluorescence for Eed, Runx2, Ki67, and ALPL was decreased in primary Sox10-Cre; Eed^Fl/Fl craniofacial cell cultures compared to controls (Figure 3f, Figure 3—figure supplement 1c), and (3) there was no change in immunofluorescence for Sox10 in either primary Sox10-Cre; Eed^Fl/Fl craniofacial cell cultures or Sox10-Cre; Eed^Fl/Fl embryos compared to controls (Figure 3—figure supplement 1c and e). These results suggest that craniofacial mesenchymal differentiation, proliferation, and osteogenesis are impaired in Sox10-Cre; Eed^Fl/Fl embryos compared to controls (Honoré et al., 2003; Kim et al., 2003).

To define cell types and gene expression programs underlying craniofacial phenotypes in Sox10-Cre; Eed^Fl/Fl embryos, single-cell RNA sequencing was performed on litter-matched E12.5 Sox10-Cre; Eed^Fl/WT and Sox10-Cre; Eed^Fl/Fl heads (n=3 biological replicates per genotype). Uniform manifold approximation and projection (UMAP) analysis of 63,730 single-cell transcriptomes revealed 23 cell clusters (C0-C22) that were defined using automated cell-type classification (Ianevski et al., 2022), cell signature genes, cell cycle analysis, and differentially expressed cluster marker genes (Figure 4a, Figure 4—figure supplements 1–3, and Supplementary file 1 and 2). Differentiating osteoblasts marked by Runx2 (C0) and proliferating mesenchymal cells marked by Ki67 (C7) were enriched in Sox10-Cre; Eed^Fl/WT samples (Figure 4b–d, Figure 4—figure supplement 3). Mesenchymal stem cells marked by Col6a3 (C4) or Dcn (C5) were enriched in Sox10-Cre; Eed^Fl/Fl samples (Figure 4b-d, Figure 4—figure supplement 3), suggesting that loss of Eed prevents craniofacial mesenchymal stem cell differentiation (Jang et al., 2016; Lamandé et al., 2006). There were also subtle differences in the number of interneurons (C17, 1.4% vs 1.9% of cells, p=0.01), Schwann cells (C19, 1.0 vs 1.3% of cells, p=0.03), pericytes (C20, 0.7% vs 0.9%, p=0.05), and spinal neurons (C21, 0.5% vs 0.9% of cells, p=0.003) in Sox10-Cre; Eed^Fl/WT versus Sox10-Cre; Eed^Fl/Fl samples, but each of these comprised a minority of recovered cell types (Student’s t-tests) (Supplementary file 1). There were no differences between genotypes in the number of chondrocytes (C10, 3.1% vs 2.7%, p=0.09), fibroblasts (C11, 2.6% vs 2.7%, p=0.24), endothelia (C16, 1.9% vs 1.6%, p=0.22), hematopoietic cells, or other recovered cell types (Student’s t-tests) (Supplementary file 1).

Differential expression analysis of single-cell transcriptomes from differentiating osteoblasts (C0), which were enriched in Sox10-Cre; Eed^Fl/WT samples (Figure 4c), compared to mesenchymal stem cells marked by Col6a3 (C4), which were enriched in Sox10-Cre; Eed^Fl/Fl samples (Figure 4c), showed Sox5 and Sox6 were reduced in mesenchymal stem cells (Figure 4e, Figure 4—figure supplement 4, and Supplementary file 3). These data are consistent with the known role of Sox transcription factors in craniofacial specification and development (Smits et al., 2001). Iroquois homeobox (Irx) transcription factors 3 and 5, which contribute to craniofacial osteogenesis and mineralization (Bonnard et al., 2012; Cain et al., 2016; Tan et al., 2020), were also suppressed in mesenchymal stem cells compared to differentiating osteoblasts (Figure 4e, Figure 4—figure supplement 4a, and Supplementary file 3).

Differential expression analysis of single-cell transcriptomes from differentiating osteoblasts compared to mesenchymal stem cells marked by Dcn (C5), which were also enriched in Sox10-Cre; Eed^Fl/Fl samples (Figure 4c), showed Mecom, Trps1, and Ptch1 were suppressed, and Sparc and Tnmd were enriched in mesenchymal stem cells (Figure 4f, Figure 4—figure supplement 4b, and Supplementary file 4). Loss of Mecom causes craniofacial malformations in mice and zebrafish (Shull et al., 2020), Ptch1 is a key component of the Hedgehog pathway that is crucial for craniofacial osteogenesis (Jeong et al., 2004; Metzis et al., 2013), and Trps1 regulates secondary palate and vibrissa development (Cho et al., 2019; Fantauzzo and Christiano, 2012). Consistently, Sox10-Cre; Eed^Fl/Fl samples had decreased vibrissa compared to controls (Figure 1d, Figure 2—figure supplement 1b). Sparc drives morphogenesis of the pharyngeal arches and inner ear (Rotllant et al., 2008), and Tnmd regulates mesenchymal differentiation (Shukunami et al., 2006). Pax3 and Pax7, key regulators of neural crest migration and development of skeletal structures (Maczkowiak et al., 2010), were also suppressed in mesenchymal stem cells compared to differentiating osteoblasts (Figure 4f, Figure 4—figure supplement 4b, and Supplementary file 4).

Differential expression analysis of single-cell transcriptomes from the two clusters of mesenchymal stem cells that were marked by Col6a3 (C4) versus Dcn (C5), both of which were enriched in Sox10-Cre; Eed^Fl/Fl samples (Figure 4c), showed Twist2, Irx3, and Irx5, which regulate mesenchymal differentiation to osteoblasts (Lee et al., 2000; Tamamura et al., 2017; Tan et al., 2020), distinguished mesenchymal stem cell populations (Figure 4g, Figure 4—figure supplement 4c, and Supplementary file 5).

To determine if there were differences in the gene expression programs of differentiating osteoblasts (C0), mesenchymal stem cells (C4, C5), or proliferating mesenchymal cells (C7) in Sox10-Cre; Eed^Fl/WT versus Sox10-Cre; Eed^Fl/Fl samples, differential expression analysis was performed on all single-cell transcriptomes from these clusters between genotypes (Supplementary file 6). Consistent with craniofacial findings after loss of Ezh2 (Ferguson et al., 2018; Kim et al., 2018; Schwarz et al., 2014) and vertebral body findings after loss of Eed (Kim et al., 2006), Hox transcription factors were broadly de-repressed in mesenchymal cell populations from Sox10-Cre; Eed^Fl/Fl samples compared to controls (Figure 4h). PRC2 targets Sp7 in bone marrow stroma cells (Liu et al., 2013) and regulates Wnt signaling in dental stem cells (Jing et al., 2016), and differential expression analysis also showed that Sp7 was enriched in mesenchymal single-cell transcriptomes from Sox10-Cre; Eed^Fl/WT samples and that the Wnt and stem cell regulator gene Gata4 was enriched in mesenchymal single-cell transcriptomes from Sox10-Cre; Eed^Fl/Fl samples (Figure 4h). Mki67, Pax3, and Pax7 were enriched in mesenchymal single-cell transcriptomes from Sox10-Cre; Eed^Fl/WT samples, and the cell cycle inhibitors Cdkn2a and Cdkn2b were enriched in mesenchymal single-cell transcriptomes from Sox10-Cre; Eed^Fl/Fl samples (Figure 4h). In sum, these data show that Eed specifies craniofacial osteoblast differentiation and mesenchymal cell proliferation by regulating diverse transcription factor programs required for the specification of post-migratory neural crest cells, and that loss of Eed inhibits mesenchymal stem cell differentiation and mesenchymal cell proliferation. These data contrast with the function of Eed in chondrocytes, where genetic inactivation of Eed at a later stage in mesenchymal development accelerates hypertrophic differentiation, leading to hypoxia and cell death (Mirzamohammadi et al., 2016).

To determine if genome-wide methylation changes underlie phenotypes and gene expression changes after Eed deletion in post-migratory neural crest cells, H3K27me3 CUT&Tag chromatin profiling was performed on nuclei harvested from E12.5 or E16.5 Sox10-Cre; Eed^Fl/WT or Sox10-Cre; Eed^Fl/Fl craniofacial tissues (n=3 biological replicates per genotype per timepoint). At E12.5, conditional Eed deletion was associated with global decreases in H3K27me3 deposition and also decreased H3K27me3 deposition at differentially expressed mesenchymal cell cluster genes from single-cell RNA sequencing (C0, C4, C5, C7) (Figure 5a, Supplementary file 6). Decreased H3K27me3 deposition was observed with conditional Eed deletion at E16.5, albeit to a lesser extent (Figure 5b), likely from relative loss of Sox10+ cells that failed to differentiate or proliferate during development without Eed and compensatory colonizing of the craniofacial region by Sox10- Eed+ cells.

The gene-specific impact of Eed-dependent H3K27me3 deposition was assessed by ranking genes from CUT&Tag chromatin profiling based on relative H3K27me3 signal for identification of methylation-rich regions (MRRs). There were 115 MRRs in Sox10-Cre; Eed^Fl/WT and 111 MRRs in Sox10-Cre; Eed^Fl/Fl craniofacial tissues at E12.5 (72 overlapping, 31.9%) and 147 MRRs in Sox10-Cre; Eed^Fl/WT and 123 MRRs in Sox10-Cre; Eed^Fl/Fl craniofacial tissues at E16.5 (105 overlapping, 38.9%) (Figure 5c and d, Supplementary files 7–10). Ontology analyses of MRRs from Sox10-Cre; Eed^Fl/WT or Sox10-Cre; Eed^Fl/Fl craniofacial tissues showed widespread changes in the H3K27me3 deposition at genes involved in transcriptional regulation and organ development and differentiation (Figure 5—figure supplement 1).

Transcription factor loci involved in differentiation and development of craniofacial progenitor cell populations were aberrantly methylated in Sox10-Cre; Eed^Fl/Fl craniofacial tissues compared to controls, including decreased H3K27me3 signal at Alx1, Alx3, Barx1, Pax7, and Tfap2a loci (Figure 5c and d, Supplementary file 7–10). Mis-regulation of Alx1 and Alx3 expression results in craniofacial dysplasia (Pini et al., 2020), Barx1 controls skeletal precursors by promoting cartilage fates (Nichols et al., 2013), Pax7 expression is critical for craniofacial progenitor cell fate (Murdoch et al., 2012), and Tfap2a influences patterning of the craniofacial skeleton through homeobox gene expression (Van Otterloo et al., 2022; Van Otterloo et al., 2018). Loci that increased in H3K27me3 signal in Sox10-Cre; Eed^Fl/Fl craniofacial tissues compared to controls resided near genes such as Twist1 and Bmp4, which are required for stabilizing neural crest cell signatures to guide migrating potential (Fan et al., 2021; Goodnough et al., 2016) and are expressed in the frontal primordium at E12.5 by Msx1 and Msx2 to control craniofacial neural crest cell differentiation (Han et al., 2007), respectively.

To determine if locus-specific H3K27me3 loss translated to increased gene expression, as predicted from functional deletion of PRC2, single-cell RNA sequencing and CUT&Tag data from E12.5 embryos were integrated. Differentially enriched genes in mesenchymal cell clusters (C0, C4, C5, C7) from single-cell RNA sequencing of Sox10-Cre; Eed^Fl/Fl samples compared to controls included the developmental transcription factors Vax2 (Ventral Anterior Homeobox 2) and Foxa1, Foxa2, and Foxd3 (Supplementary file 6), each of which had gene body or distal element loses in H3K27me3 signal from CUT&Tag chromatin profiling (Figure 5e). Forkhead box (FOX) transcription factors are required for craniofacial bone and cartilage development, and FOX mis-regulation expands cartilage domains and inhibits bone fate decisions (Mundell and Labosky, 2011; Xu et al., 2018). These data suggest that Vax2, Foxa1, Foxa2, or Foxd3 may contribute to craniofacial phenotypes observed in Sox10-Cre; Eed^Fl/Fl embryos downstream of H3K27me3 deposition in response to functional deletion of PRC2. More broadly, these data suggest that loss of H3K27me3 signal at developmental gene loci may correlate with gene expression changes that result in aberrant cell fate decisions of craniofacial mesenchymal stem cells, osteoblasts, and proliferating mesenchymal cells from the neural crest.

Mice were maintained in the University of California San Francisco (UCSF) pathogen-free animal facility in accordance with the guidelines established by the Institutional Animal Care and Use Committee (IACUC) and Laboratory Animal Resource Center (LARC) protocol AN191840. Mice were maintained in a 70°F, 50% humidity temperature-controlled barrier facilities under a 12–12 hr light cycle with access to food and water ad libitum. Sox10-Cre and Eed^Fl mice were obtained from the Jackson Laboratory (B6;CBA-Tg(Sox10-cre)1Wdr/J, 025807 and B6;129S1-Eed^tm1Sho/J, 022727, respectively). The presence of the floxed Eed allele was determined through standard PCR genotyping using the following primers: 5’ GGGACGTGCTGACATTTTCT 3’ (forward) and 5’ CTTGGGTGGTTTGGCTAAGA 3’ (reverse). To generate embryos at specific time points, Sox10-Cretg^tg+ Eed^Fl/WT mice were bred overnight with Eed^Fl/Fl mice. Females were checked for copulation plugs in the morning, and the presence of a vaginal plug was designated as E0.5.

Ultrasound studies were acquired with a Fujifilm Vevo 2100 Imaging system, and instrument specifically designed for lab animal imaging studies. Pregnant females were anesthetized using an isoflurane/oxygen mixture with an isoflurane concentration of 3% during imaging. The pregnancy date (at least p17) was checked to observe the correct development of fetal hearts. Isoflurane was increased until 5% and the abdomen of pregnant females was open surgically to expose the gravid uterus. The number of fetuses was counted, and craniofacial architecture was evaluated for phenotypic confirmation. Once each fetus was defined as mutant or wild type, echocardiography was performed to evaluate left ventricle anatomy and physiology, including left ventricle size and contractility. Three different measurements were obtained in B and M modes, and functional parameters, including heart rate, fraction shortening, ejection fraction, and left ventricle mass, were obtained.

Embryos were stained as previously described (Sandell et al., 2018). In brief, embryos were harvested, and heads were fixed overnight in 4% PFA. Embryos were dehydrated through a methanol series up to 100% before being treated with Dent’s bleach (4:1:1, MeOH, DMSO, 30% H₂O₂) for 2 hr at 25°C. Embryos were serially rehydrated in methanol up to 25%, washed with PBST, and stained with DAPI solution (1:40,000, Thermo Scientific, cat# 62248) overnight at 4°C. Samples were mounted in low melt agarose for imaging.

Embryos were stained as previously described (Rigueur and Lyons, 2013). In brief, embryos were dissected in cold PBS and scalded in hot water to facilitate the removal of eyes, skin, and internal organs. Embryos were fixed in 95% ethanol overnight, then subsequently incubated in acetone overnight. Embryos were stained with Alcian blue solution (Newcomer Supply, cat# 1300B) for 8 hr, washed in 70% ethanol, and destained in 95% ethanol. Embryos were cleared in 95% ethanol and 1% KOH and stained in Alizarin Red S solution (SCBT, cat# 130-22-3) for 4 hr. Embryos were stored in glycerol for imaging.

microCT studies were acquired with a Perkin Elmer’s Quantum GX2 microCT Imaging system, an instrument specifically designed for lab animal imaging studies. The samples were imaged in Eppendorf tubes containing preservative media, which were placed on the scanning bed and fixed with tape to minimize movement. Acquisition parameters were set as follows: FOV 10 mm, acquisition time 14 min, current voltage 70 kV, amperage 114 µA. Each study was composed of a 512×512×512 voxels matrix with a spatial resolution of 0.018 mm³. Study reconstruction was based on Feldkamp’s method using instrument dedicated software. For image analysis, tridimensional renders were obtained using 3D SLICER software. The threshold range applied for bone segmentation was 300–1400 Hounsfield units. For the segmentation and visualization of individual frontal bones, cranial base, and mandible, Avizo Lite software (v9.1.1) was used. The threshold range applied was 570–1570 Hounsfield units.

Embryos were fixed in 4% PFA overnight at 4°C and subjected to a sucrose gradient from 15% to 30% before embedding in OCT (Tissue-Tek, cat# 4583) and storage at –80°C. Embryos were sectioned at 25 µm on a Leica CM1850 cryostat. Slides were immediately subjected to H&E staining as previously described (Cardiff et al., 2014) or stored at –80°C for immunofluorescence.

Immunofluorescence of cryosections was performed on glass slides. Immunofluorescence for primary cells was performed on glass coverslips. Primary cells were fixed in 4% PFA for 10 min at room temperature. Antibody incubations were performed in blocking solution (0.1% Triton X-100, 1% BSA, 10% donkey serum) for 2 hr at room temperature or overnight at 4°C. Slides or coverslips were mounted in ProLong Diamond Antifade Mountant (Thermo Scientific, cat# P36965). Primary antibodies used in this study were anti-Sox9 (1:500, Proteintech, cat# 67439-1-Ig), anti-Sox9 (1:500, Abcam, cat# ab185230), anti-Runx2 (1:500, Cell Signaling, cat# D1L7F), anti-Ki67 (1:2000, Abcam, cat# ab15580), anti-Vimentin (1:1000, Abcam, cat# ab8069), anti-BrdU (1:1000, Abcam, cat# ab6326), anti-Sox10 (1:500, Cell Signaling, cat# D5V9L), anti-ALPL (1:1000, Invitrogen, cat# PA5-47419), anti-Eed (1:1000, Cell Signaling, cat# E4L6E), and anti-Cleaved Caspase-3 (Asp175) (1:500, Cell Signaling, cat#9661). Secondary antibodies used in this study were Donkey anti-Rabbit IgG (H+L) Alexa Fluor Plus 555 (1:1000, Thermo Scientific, cat# A32794), Donkey anti-Rabbit IgG (H+L) Alexa Fluor 488 (1:1000, Thermo Scientific, cat# A21206), and Donkey anti-Rabbit IgG (H+L) Alexa Fluor Plus 568 (1:1000, Thermo Scientific, cat# A10042). DAPI solution (1:5000, Thermo Scientific, cat# 62248) was added to secondary antibody solutions. Slides or coverslips were mounted in ProLong Diamond Antifade Mountant (Thermo Scientific, cat# P36965) and cured overnight before imaging.

Timed mating dams were subjected to intraperitoneal injection of 100 mg/kg sterile BrdU in PBS (Abcam, cat# ab142567). After 4 hr, mice were euthanized, and embryos were harvested in cold PBS. Embryos were fixed in 4% PFA overnight at 4°C and cryosectioned. Sections were incubated with 1 M HCl for 2 hr then 0.1 M sodium borate buffer for 15 min before being subjected to standard immunofluorescence.

Whole-mount skeletal and H&E stains were imaged using a Zeiss Stemi 305 Stereo Zoom microscope running Zeiss Blue v2.0. Whole-mount embryo DAPI stains were imaged using an upright Zeiss Axio Imager Z2 running AxioVision v4.0. Immunofluorescence images were obtained using a Zeiss LSM800 confocal laser scanning microscope running Zen Blue v2.0. Quantification of embryo measurements and immunofluorescence staining intensities was performed in ImageJ using standard thresholding and measurement of signal integrated density. Signal to nuclei normalization was performed by dividing signal intensity by DAPI signal per image. Quantification was performed using n>6 images per embryo. Proliferation index was calculated by dividing the number of Ki67+ or BrdU+ nuclei by the total number of DAPI-stained nuclei per image.

Timed mating embryos were dissected on ice-cold PBS and heads were removed. Brains and eyes were removed from each head, and the remaining craniofacial structures were minced with a sterile blade on a glass surface. Tissue was enzymatically dissociated in 3 mg/ml Collagenase Type 7 (Worthington, cat# CLS-6) in HBSS for 2–3 hr at 37°C with frequent trituration using an Eppendorf ThermoMixer. Dissociated cells were filtered through a 70 µM MACS smart strainer (Miltenyi Biotec, cat# 130-110-916) and either plated on coverslips or cell culture dishes. Primary craniofacial cells were grown in DMEM with 10% FBS. For RNA extraction and immunofluorescence, cells were harvested 1 day following initial dissociation.

Matched littermate timed embryos were dissected on ice-cold PBS. Embryo heads were removed and minced using sterile razor blades on a glass surface. Tissue was enzymatically dissociated in 3 mg/ml Collagenase Type 7 (Worthington, cat# CLS-6) in HBSS for 2–3 hr at 37°C with frequent trituration using an Eppendorf ThermoMixer. The quality of the dissociation was frequently monitored by looking at the cell suspension on an automated cell counter. Cells were spun down at 350 rcf, resuspended in MACS BSA stock solution (Miltenyi Biotec, cat# 130-091-376), and serially filtered through 70 µm and 40 µm MACS smart strainers (Miltenyi Biotec, cat# 130-110-916). Dissociation quality checks, cell viability, and counting were performed using an Invitrogen Countess 3 automated cell counter. 10,000 cells were loaded per single-cell RNA sequencing sample. Single-cell RNA sequencing libraries were generated using the Chromium Single Cell 3’ Library & Gel Bead Kit v3.1 on a 10x Genomics Chromium controller using the manufacturer’s recommended default protocol and settings.

Library demultiplexing, read alignment to the mouse reference genome mm10, and unique molecular identifier (UMI) quantification was performed in Cell Ranger v7.2.0. Cells with greater than 200 unique genes were retained for analysis. Data were normalized and variance-stabilized by SCTransform in Seurat v5.0 (Hao et al., 2024). UMAP and cluster analysis were performed using the Seurat function RunUMAP with parameters of mindist = 0.7, res = 0.4, dims = 1:30. Cluster markers were identified using the Seurat function FindAllMarkers with parameters min.pct=0.25, thresh.use=0.25. Differential gene expression analysis was performed using DElegate v1.1.0 (Hafemeister and Halbritter, 2023) using the DESeq2 method, with biological replicate animals (n=3 per genotype) as the replicate_column parameter. FeaturePlots were generated using SCPubr v1.1.2 (Blanco-Carmona, 2022), and scale bars represent log2 UMIs corrected by SCTransform.

E12.5 or E16.5 timed mating embryos were dissected on ice-cold PBS. Craniofacial tissues were microdissected under a Leica Stereoscope, minced, and weighed such that 30 mg of craniofacial tissue was harvested per embryo. Tissue was then processed, and libraries were generated according to the manufacturer’s instructions using an end-to-end CUT&Tag-IT Assay Kit (Active Motif, cat# 53170). In brief, nuclei were mechanically isolated in a Dounce homogenizer with lysis buffer containing protease inhibitors and filtered through a 40 µm strainer. Nuclei were pelleted, washed, and counted such that each reaction was normalized to 500,000 nuclei per embryo. To enable quantitative normalization across samples, 20,000 Drosophila melanogaster spike-in nuclei (Active Motif, cat# 53173) were added prior to bead conjugation. Nuclei were then bound to pre-equilibrated concanavalin A-coated magnetic beads and incubated with a primary antibody against H3K27me3 (Active Motif, cat# 39055) along with a Drosophila-specific anti-H2Av spike-in antibody (Active Motif, cat# 61686) overnight at 4°C in antibody binding buffer containing 0.05% digitonin, 2 mM EDTA, and 0.1% BSA. Following washes with digitonin-containing buffer, samples were incubated with species-appropriate secondary antibodies for 1  hr at room temperature and subsequently exposed to a pre-loaded pA-Tn5 transposome complex (1:20 dilution) in Dig-300 Buffer for 1  hr to enable tethering. Tagmentation was initiated by the addition of 10 mM MgCl₂ at 37 °C for 1 hr, after which the reaction was terminated with EDTA, 0.1% SDS, and proteinase K digestion at 55°C for 1 hr. DNA was purified using magnetic bead-based cleanup and amplified with i7 and i5 indexed primers using standard PCR amplification. Libraries were cleaned with SPRI beads and quality-assessed on an Agilent Tape Station. Sequencing was performed on an Illumina NovaSeq X with paired-end reads targeting 25 million reads per sample.

Raw H3K27me3 reads were processed using the nf-core Cut&Run pipeline (v1.0.0) and aligned to the mm10 genome. Signal tracks were generated as bigWig files. Replicate bigWigs were merged, averaged, and CPM-normalized using deepTools bamCompare (v3.5.0) with options --normalizeUsing CPM, --scaleFactorsMethod None, and --binSize 50. Peaks were called on merged bigWigs using SEACR (v1.3) in stringent mode with an FDR threshold of 0.001. Signal metaplots were generated using deepTools computeMatrix and plotProfile with --referencePoint center and flanking windows of ±10 kb. Metaplots were generated for all protein-coding genes and for selected differentially expressed genes (DEGs). DEG regions were defined by identifying DEGs from differential expression analyses and retrieving their transcription start site (TSS) coordinates using biomaRt (Ensembl release 106). TSS regions were expanded by ±10 kb, and BED files were generated by excluding genes located on non-autosomal chromosomes (MT, X, Y). The enhancer ranking and super-enhancer identification strategy was adapted from previously described methods (Lovén et al., 2013; Whyte et al., 2013). Peaks overlapping TSS were removed using BEDTools intersect (v2.31.1) with the -v option. MRRs were identified using ROSE (Lin et al., 2020) with default settings, and putative MRRs ranked by area under the curve (AUC) from bigWig signal. Inflection points in AUC-ranked plots were used to define MRRs. MRRs were annotated to the nearest RefSeq gene using the TxDb.Mmusculus.UCSC.mm-10.knownGene (v3.10.0) and org.Mm.eg.db (v3.18.0) packages in R. Hockey-stick plots were generated by plotting H3K27me3 signal versus rank.

All experiments were performed with independent biological replicates and repeated, and statistics were derived from biological replicates. Biological replicates are indicated in each figure panel or figure legend. No statistical methods were used to predetermine sample sizes, but sample sizes in this study are similar or larger to those reported in previous publications. Data distribution was assumed to be normal, but this was not formally tested. Investigators were blinded to conditions during data collection and analysis. Bioinformatic analyses were performed blind to clinical features, outcomes, and molecular characteristics. The samples used in this study were nonrandomized with no intervention, and all samples were interrogated equally. Thus, controlling for covariates among samples was not relevant. No data points were excluded from the analyses.

No custom software, tools, or packages were used. The open-source software, tools, and packages used for data analysis in this study are referenced in the methods where applicable and include Cell Ranger v7.2.0, Seurat v5.0, DElegate v1.1.0, SCPubr v1.1.2, nf-core Cut&Run (v1.0.0), deepTools bamCompare (v3.5.0), SEACR (v1.3), BEDTools intersect (v2.31.1), TxDb.Mmusculus.UCSC.mm-10.knownGene (v3.10.0), and org.Mm.eg.db (v3.18.0).