The transcriptional regulator MEIS2 sets up the ground state for palatal osteogenesis in mice

Haploinsufficiency of Meis homeobox 2 (MEIS2), encoding a transcriptional regulator, is associated with human cleft palate, and Meis2 inactivation leads to abnormal palate development in mice, implicating MEIS2 functions in palate development. However, its functional mechanisms remain unknown. Here we observed widespread MEIS2 expression in the developing palate in mice. Wnt1Cre-mediated Meis2 inactivation in cranial neural crest cells led to a secondary palate cleft. Importantly, about half of the Wnt1Cre;Meis2f/f mice exhibited a submucous cleft, providing a model for studying palatal bone formation and patterning. Consistent with complete absence of palatal bones, the results from integrative analyses of MEIS2 by ChIP sequencing, RNA-Seq, and an assay for transposase-accessible chromatin sequencing identified key osteogenic genes regulated directly by MEIS2, indicating that it plays a fundamental role in palatal osteogenesis. De novo motif analysis uncovered that the MEIS2-bound regions are highly enriched in binding motifs for several key osteogenic transcription factors, particularly short stature homeobox 2 (SHOX2). Comparative ChIP sequencing analyses revealed genome-wide co-occupancy of MEIS2 and SHOX2 in addition to their colocalization in the developing palate and physical interaction, suggesting that SHOX2 and MEIS2 functionally interact. However, although SHOX2 was required for proper palatal bone formation and was a direct downstream target of MEIS2, Shox2 overexpression failed to rescue the palatal bone defects in a Meis2-mutant background. These results, together with the fact that Meis2 expression is associated with high osteogenic potential and required for chromatin accessibility of osteogenic genes, support a vital function of MEIS2 in setting up a ground state for palatal osteogenesis.

Haploinsufficiency of Meis homeobox 2 (MEIS2), encoding a transcriptional regulator, is associated with human cleft palate, and Meis2 inactivation leads to abnormal palate development in mice, implicating MEIS2 functions in palate development. However, its functional mechanisms remain unknown. Here we observed widespread MEIS2 expression in the developing palate in mice. Wnt1 Cre -mediated Meis2 inactivation in cranial neural crest cells led to a secondary palate cleft. Importantly, about half of the Wnt1 Cre ;Meis2 f/f mice exhibited a submucous cleft, providing a model for studying palatal bone formation and patterning. Consistent with complete absence of palatal bones, the results from integrative analyses of MEIS2 by ChIP sequencing, RNA-Seq, and an assay for transposase-accessible chromatin sequencing identified key osteogenic genes regulated directly by MEIS2, indicating that it plays a fundamental role in palatal osteogenesis. De novo motif analysis uncovered that the MEIS2bound regions are highly enriched in binding motifs for several key osteogenic transcription factors, particularly short stature homeobox 2 (SHOX2). Comparative ChIP sequencing analyses revealed genome-wide co-occupancy of MEIS2 and SHOX2 in addition to their colocalization in the developing palate and physical interaction, suggesting that SHOX2 and MEIS2 functionally interact. However, although SHOX2 was required for proper palatal bone formation and was a direct downstream target of MEIS2, Shox2 overexpression failed to rescue the palatal bone defects in a Meis2-mutant background. These results, together with the fact that Meis2 expression is associated with high osteogenic potential and required for chromatin accessibility of osteogenic genes, support a vital function of MEIS2 in setting up a ground state for palatal osteogenesis.
Myeloid ectopic viral integration site (MEIS) 4 transcription factors, the members of the three-amino-acid loop (TALE) superclass, are highly conserved in vertebrates (1). Humans and mice possess three MEIS proteins (MEIS1, MEIS2, and MEIS3) that contain an atypical TALE homeodomain for DNA binding. On a genome-wide scale, the prevailing view is that MEIS proteins act as cofactors for homeobox (HOX) proteins, a subset of homeobox proteins, with or without pre-B cell leukemia transcription factor (PBX) to regulate target gene expression and endow the initially similar segments with their distinct features, except in the first branchial arch (HOX proteins are not expressed in the first branchial arch). MEIS proteins improve the binding affinity and specificity of HOX proteins by directly regulating HOX protein activity. On the other hand, MEIS proteins can also bind DNA independently or cooperatively with other transcription factors (2). Interestingly, studies using mouse models uncovered that MEIS proteins occupy a broad spectrum of common regions involved in the development of entire visceral skeleton in the HOX-negative first branchial arch and HOX-positive second branchial arch, suggesting that MEIS proteins provide a common ground state in all arches (3).
Although all MEIS proteins share a highly similar identity and function redundantly, MEIS1-MEIS3 appear to have distinct functions in embryonic development. Meis1-null mice die on embryonic day 14.5 (E14.5) with liver hypoplasia, hemorrhage, and impaired erythropoiesis (4). Meis3-null zebrafish show abnormal hind brain formation (5). Although Meis2-null mice displayed embryonic lethality between E13.5 and E14.5, with hemorrhage and a small liver size, similar to Meis1-null mice, conditional knockout of Meis2 by AP2␣ Cre in developing neural crest cells results in a defective craniofacial skeleton and abnormal palate (6). Significantly, patients with a de novo sequence variant in MEIS2 or a deletion on chromosome 15q14 affecting MEIS2 suffered from recurrent features, including cleft palate. The degrees of cleft palate in such patients varies from mild (submucous cleft palate or bifid uvula) to severe (overt cleft lip and palate). Thus, the MEIS2-related syndrome was attributed to haploinsufficiency of this gene, indicating that MEIS2 plays an indispensable role in normal palate development (7). However, the detailed role of MEIS2 in palatal development remains known.
Development of the mammalian palate, derived from the HOX-negative first branchial arch, is a highly dynamic process that involves outgrowth, elevation, and fusion of the primary and secondary palate. This complex process is governed by the interactions of multiple signaling pathways and transcription factors. The primary palate originates from embryonic frontonasal prominence and develops into a small part of the adult hard palate anterior to the incisive fossa, whereas the secondary palate arises from paired maxillary processes of the first pharyngeal arch, known as palatal shelves, and consists of the dominating portion of the hard palate and the entire posterior soft palate (8,9). Although the initial palatal mesenchymal cells are completely derived from the cranial neural crest (CNC), the anterior two-thirds of the palate undergo osteogenesis to form the bony hard palate, which consists of the palatine process of the maxilla and the horizontal plate of the palatine bone. The rest of the palate is also populated by mesoderm-derived muscle progenitors to form muscular soft palate.
Although a growing number of studies have revealed the importance of numerous genes in palate development (10), most of these targeted mutant mouse models exhibit complete cleft of the secondary palate. Submucous cleft palate, also known as incomplete cleft palate, is a rare type of cleft palate, defined as malformed palatal bone and/or abnormal soft palate (11), allowing mechanistic studies of osteogenesis and patterning of the palatal bone during development. Gene inactivation studies imply a critical role of Bmpr1a and Tbx22 in palatal bone formation (12,13). Our previous studies showed that short stature homeobox 2 (Shox2), encoding a paired-like homeodomain transcription factor, is expressed specifically in the anterior palatal mesenchyme, overlapping with the future palatine process of the maxilla (14,15). Shox2-deficient mice exhibit a rare type of incomplete cleft in the anterior palate and defective palatal bone formation (14). SHOX2 ChIP-Seq on the developing limb revealed substantial co-occupation of HOX-TALE factors around skeletogenic genes (16). Furthermore, unbiased de novo motif discovery identified high enrichment of MEIS/PBX motifs in genome-wide SHOX2-bound sites in the developing palate, suggesting that SHOX2 acts as transcription partner for TALE proteins in the HOX-negative developing palate (16). However, whether SHOX2 and MEIS/PBX have functional interactions and exhibit genome-wide co-occupancy in the HOX-negative palate remains unknown.
In this study, we showed that MEIS2 is widely expressed in the developing palate and that tissue-specific inactivation of Meis2 in CNC-derived palatal mesenchyme, but not the palatal epithelium, causes cleft palate, including submucous cleft palate (cleft soft palate) in mice. We subsequently conducted a comprehensive investigation of the functional mechanisms of MEIS2 in palatogenesis at the cellular, molecular, and genomic levels.

Conditional inactivation of Meis2 in CNC cells leads to cleft of the secondary palate
To unmask the role of MEIS2 in palatogenesis, we started with examination of MEIS2 expression by immunostaining in the developing palate. We found that MEIS2 is strongly expressed in the epithelium and mesenchyme along the anterior-posterior (A-P) axis throughout the entire phase of palate development, from E11.5 until postnatal day 0 (P0) (Fig. 1). Additionally, we observed, at the palatine level (defined as the

MEIS2 regulates palatal osteogenesis
middle level), two unique MEIS2-negative domains that locate at either end of the nasal passage just dorsal to the palate and eventually become part of maxillary bone (Fig. 1D').
To determine the tissue-specific requirement of MEIS2 in palatal development, we generated Meis2 conditional mutants by compounding the floxed Meis2 conditional allele (6) with K14 Cre or Wnt1 Cre allele to delete Meis2 from either palatal epithelium or mesenchyme. Although epithelium-specific inactivation of Meis2 did not produce any overt phenotype (data not shown), Meis2 deletion in CNC-derived palatal mesenchyme by Wnt1 Cre (Fig. S1) indeed led to cleft of the secondary palate, similar to that seen in AP2␣ Cre ;Meis2 f/f mice (6). Close gross examination of mutant mice identified two types of cleft palate defects: complete cleft palate (CCP) (47.6%, 81 of 170) and submucous cleft, referred to here as incomplete cleft palate (ICP) (52.4%, 89 of 170), manifested as disorganized palatal rugae and cleft soft palate (Fig. 2, A and A', and Fig. S2, A and A'). Whole-mount skeletal staining of newborn pups showed that all Meis2 mutants with ICP or CCP displayed complete absence of palatal bones, including the palatine process of the maxilla and palatine bone, as well as loss of lamina obturans and pterygoid (Fig. 2, B and B', and Fig. S2, B and B'). Histological examination of mutant embryos compared with controls revealed that all mutants had smaller palatal shelves in the posterior portion, where the soft palate would form ( Fig. 2 and Fig.  S2). A BrdU labeling assay confirmed significantly reduced cell proliferation rates in the posterior palatal mesenchyme but not in the anterior palate, in line with a down-regulated expression level of Cyclin D1 and A2 ( Fig. S3 and data not shown). Moreover, the comparable levels of caspase 3 in control and mutant palates at E13.5 and E16.5 indicated that Meis2 deletion does not affect cell apoptosis (data not shown). However, although in some mutants, the anterior portion of the palatal shelf also appeared to be deformed and failed to elevate ( Fig. S2 and data

MEIS2 regulates palatal osteogenesis
not shown), some mutant embryos exhibited relatively normal anterior palatal shelves that were able to elevate and meet at the midline (Fig. 2, E', G', and I'). Histological examination of P0 mutants with ICP further confirmed failed bone formation in the hard palate domain (Fig. 2, I and I'). Additionally, the tongue of mutant mice appeared to be smaller and exhibited disorganized muscle fibers compared with controls ( Fig. S2, E-F').

Deletion of Meis2 down-regulates the expression of osteogenic genes
The complete absence of the palatal bones in Wnt1 Cre ; Meis2 f/f mice indicates an essential role of MEIS2 in osteogenesis during palatogenesis. Because most genetically modified animal models present a complete cleft palate phenotype, Wnt1 Cre ;Meis2 f/f mice with ICP are an excellent model for studying MEIS2's function in palatal osteogenesis in situ. We chose mutant embryos whose anterior palatal shelves had elevated at E14.5 and fused at the following stages for detailed analyses. Because palatal bone formation undergoes intramembranous ossification, which requires RUNX2 and SP7, we conducted immunostaining for these two initial bone markers. At E14.5-E16.5, we failed to detect RUNX2 and SP7 expression in the entire future hard palate domain of the mutants (Fig. 3, A-F'), indicating failure of initial osteogenesis in the developing palate. Interestingly, the parts of the maxillary bone at the either end of the nasal passage within the palatine domain formed nevertheless, consistent with the lack of MEIS2 expression in these two sites (Figs. 1D' and 3, B', D', and F').
To profile MEIS2-regulated gene expression involved in palatal osteogenesis, we performed RNA-Seq in the anterior palatal mesenchyme of Wnt1 Cre ;Meis2 f/f and control mice at stages before osteogenic differentiation (E12.5) and after initiation of

MEIS2 regulates palatal osteogenesis
osteogenesis (E15.5) with biological triplicates. At E12.5, 1613 genes were responsive to Meis2 inactivation, with p Ͻ 0.05 and a false discovery rate (FDR) Ͻ 0.05. Among them, 782 genes were expressed at significantly lower levels, and 831 genes were up-regulated in Meis2 mutants compared with controls (Table  S1). Gene ontology (GO) analysis showed that most of the significant terms associated with those down-regulated genes are osteogenesis-related, such as bone morphogenesis, bone development, ossification, and skeletal system morphogenesis (Fig.  3G). To validate the RNA-Seq results, we assayed the expression of some selected genes (Runx2, Shox2, Pbx1, Bmp2, Col9a1, Alx4, Mecom, Igf1, Aldh1a2, and Gsc) by RT-PCR and confirmed significant down-regulation of these genes in the mutant palatal mesenchyme compared with controls (Fig. 3H). Concordantly, immunostaining and in situ hybridization assays further verified down-regulation of SHOX2, PBX1, and Bmp2 expression in the mutant palatal shelves (Fig. 3, I-K'). Furthermore, RNA-Seq results of E15.5 palate revealed a total number of 1240 differentially expressed genes (DEGs), including 471 down-regulated and 769 up-regulated ones with p Ͻ 0.05 and FDR Ͻ 0.05. Similarly, these DEGs were also highly related to ossification, biomineral tissue development, bone mineralization, regulation of biomineral tissue development, and bone development (Fig. S4A). In addition to down-regulated expression of Runx2, Shox2, Pbx1, and Bmp2, RT-qPCR also confirmed dramatically reduced expression of osteogenic differentiation markers, including Sp7, Dmp1, Col1a2, Spp1, and Osr1 (Fig. S4, B-E').

MEIS2 directly binds to osteogenic genes in the palate
To define the direct targets of MEIS2 in the palatal mesenchyme, we generated a Meis2 HA knock-in allele that allows effective immunoprecipitation of HA-tagged MEIS2 (Fig. S5A). The expression pattern of HA-tagged MEIS2 was confirmed by immunostaining (Fig. S5B), and heterozygous and homozygous Meis2 HA mice appeared to be normal and fertile, indicating that the HA tag does not disrupt MEIS2's function (data not shown). To match the RNA-Seq results, we conducted MEIS2 ChIP-Seq using an HA antibody on the palatal mesenchyme from E12.5 Meis2 HA mice. The majority of the MEIS2-bound peaks have a great distance to its nearest transcription start sites (TSSs; Fig.  4A) and were assigned to 1891 coding genes. GO analysis revealed association of these genes to ossification (including Shox2, Runx2, Pbx1, and Bmp2), osteoblast development, and embryonic organ morphogenesis ( Fig. 4B and Table S2).
To further confirm the ChIP-Seq results and to define the chromatin landscape in palatal mesenchymal cells with and without MEIS2, we performed assay for transposase-accessible chromatin sequencing (ATAC-Seq) on palatal cells from E12.5 Meis2 mutants and controls. Quality control (QC) analysis indicated that all ATAC-Seq libraries possessed the expected fragment size distribution, with a large proportion of fragments less than 100 bp, representing nucleosome-free regions (Fig.  S5C). In controls, 85.83% of reads were mapped to mm10 reference genomes, and about 140,000 regions of accessible chromatin were identified. On the other hand, 88.85% of reads were

MEIS2 regulates palatal osteogenesis
mapped, and about 120,000 regions were identified in the Meis2 mutant group. Comparison of the distribution of genomic annotation between controls and mutants showed that most ATAC-Seq peaks were mapped to regions within several hundred base pairs of TSSs but exhibited a prominent difference in the percentage of the promoters, introns, and intergenic regions in the mutants compared with controls. These results demonstrate that the chromatin accessibility of the promoters and enhancers is affected by deletion of Meis2 (Fig. S5, D and E).
To establish the correlation between chromatin accessibility and transcriptional regulation, we annotated the MEIS2-related open chromatin regions to the nearest genes, with p Ͻ 0.05 and FDR Ͻ 0.05 (Table S3). In the absence of Meis2, 1067 chromatin regions annotated to 761 genes lost their accessibility, whereas 1721 chromatin regions annotated to 1211 genes gained the accessibility. We subsequently conducted integrative analyses of the ATAC-Seq, RNA-Seq, and MEIS2 ChIP-Seq datasets. As expected, hundreds of genes linked to MEIS2-related open chromatin regions were occupied by MEIS2 and showed a transcriptional change (Fig. 4C). The key osteogenic genes bound by MEIS2, including Runx2, Shox2, Pbx1, Bmp2, Osr1, and Twist1, were down-regulated at the transcriptional level with p Ͻ 0.01 and FDR Ͻ 0.01 (Table S1) and lost their chromatin accessibility, including the coding regions and MEIS2-bound distal regulatory regions (Figs. 4D and Fig. 5D). These results indicate that MEIS2 binds directly to the osteogenic genes loci, arranges chromatin accessibility, and eventually regulates transcriptional profiles during palatal osteogenesis.

Cooperation of MEIS2 and SHOX2 during palatal osteogenesis
To identify potential transcription factors that also bind to the MEIS2-bound regions, we performed de novo motif analysis on the binding peaks of MEIS2 in the palate (Fig. 5A). In addition to factors involved in neural crest development, such as FOXC2 and HOXA10, the MEIS2-bound regions were enriched for binding motifs of transcription factors involved in bone formation, including SHOX2, RUNX2, PBX1, and ASCL2.
We have reported previously that TALE-related motifs are enriched in SHOX2-bound peaks in the palate (16), consistent with enrichment of SHOX2-bound motifs in MEIS2-bound regions. We have demonstrated that most of the SHOX2bound regions are likely distal regulatory elements, as demonstrated by the enhancer activity of several selected SHOX2bound DNA fragments in transient transgenic reporter assays (15,16). Together, these results suggest that SHOX2 and MEIS2 function as partners to regulate gene expression and palatal osteogenesis. To test this hypothesis, we performed double immunostaining and found extensive but not complete colocalization of MEIS2 and SHOX2 in the palatal mesenchyme (Fig. S6, A-B'). A coimmunoprecipitation assay of culture cells and a Duolink proximity ligation assay on palatal mesenchymal cells were performed as described in the supporting experimental procedure, and confirmed the physical interaction between MEIS2 and SHOX2 (Fig. S6, C-D'). We subsequently intersected our MEIS2 ChIP-Seq datasets with our published SHOX2 ChIP-Seq results on the E12.5 anterior palate (16). Indeed, the majority of MEIS2-bound regions were also occupied by SHOX2, indicating genome-wide co-occu-pancy of SHOX2 and MEIS2 (Fig. 5B). Furthermore, functional annotation of MEIS2 ChIP-Seq and SHOX2 ChIP-Seq datasets identified almost the same GO terms, such as embryonic organ development, ossification, osteoblast differentiation, and mesenchyme development (Fig. 5C). Considering the vital role of MEIS2 and SHOX2 in palatal osteogenesis, we focused on genes from the ossification term that were cobound by MEIS2 and SHOX2 for further analysis of their chromatin landscape revealed by ATAC-Seq. As shown in Fig. 5D, the top osteogenic genes Shox2, Runx2, Pbx1, and Bmp2 were found to lose accessibility in the Meis2 mutant palate. Integration of our ChIP-Seq and ATAC-Seq data with the published H3K27ac ChIP-Seq data on the embryonic facial region (17) showed high enrichment of H3K27ac in these regions, indicating active enhancer characteristics of these SHOX2 and MEIS2 cobound regions.

MEIS2 provides the ground state for palatal osteogenesis
Given the dramatic down-regulation of SHOX2 and inaccessibility of the regulatory and coding regions of Shox2 in the Meis2 mutant palate and the important role of SHOX2 in palatal bone formation (15,18), we hypothesized that, as a downstream target of MEIS2, SHOX2 may mediate MEIS2's function in regulating palatal osteogenesis. To test this hypothesis, we overexpressed Shox2 in CNC cells in the Meis2 mutant background by compounding the R26R Shox2 knock-in allele (19) with Wnt1 Cre and Meis2 f/f alleles. All Wnt1 Cre ;Meis2 f/f ; R26R Shox2 mice (n ϭ 14) died soon after birth with ICP (5 of 14, 35.7%) or CCP (9 of 14, 64.3%). The ICP mice likewise manifested a fused anterior palate and cleft soft palate (Fig. S7, A and  A'). Similar to Wnt1 Cre ;Meis2 f/f mice, whole-mount skeletal staining of Wnt1 Cre ;Meis2 f/f ;R26R Shox2 ICP mice revealed complete loss of the palatal bones (Fig. S7, B and B'). Immunostaining confirmed lack of RUNX2 expression in the mutant hard palate compared with controls at E16.5 (Fig. S7, C-D'). Thus, Shox2 overexpression failed to rescue the defect of palatal bones in the absence of Meis2.
We have reported previously that palatal mesenchymal cells expressing Shox2 are divided into two subpopulations: one differentiating to osteoblasts and one becoming fibroblasts (15). The observation that SHOX2 and MEIS2 expression overlaps largely but not completely in the palatal mesenchyme (Fig. S6) prompted us to ask whether Shox2-positive palatal mesenchymal cells with or without Meis2 expression possess distinct osteogenic potential. To address this question, we carried out single-cell RNA-Seq on total Shox2-expressing palatal mesenchymal cells. We took advantage of existing Shox2 Cre knock-in allele and R26R mTmG reporter mice to isolate Shox2 ϩ cells by cell sorting. A total number of 3979 GFP-positive cells from E13.5 Shox2 Cre/ϩ ;R26R mTmG palatal mesenchyme were sequenced, and 2953 cells were selected for the subsequent analysis based on QC metrics. After normalization and dimensionality determination, the cells were automatically clustered into five clusters (C0 -C4) using Seurat v3. Interestingly, using Fea-turePlot, cells in cluster 2 were found to express a remarkably lower level of Meis2 compared with other clusters (Fig. 6A). Using FindMarkers, we identified 455 DEGs between cluster 2 and other clusters with p Ͻ 0.05 and FDR Ͻ 0.05, including 195 genes at lower levels and 260 genes at higher levels in cluster 2

MEIS2 regulates palatal osteogenesis
cells. GO analysis revealed that the genes with lower expression levels are associated with roof of mouth development, regulation of ossification, and ossification (Fig. 6B). Indeed, an expression heatmap manifested significantly low expression levels of osteogenic genes in cluster 2 cells (Fig. 6C). By comparison with the RNA-Seq data from the E12.5 Wnt1 Cre ;Meis2 f/f palate, we found that, among the 22 ossification-associated genes with lower expression levels in Meis2-negative cells, 13 were also down-regulated in the mutant palate (Fig. 6D). The results suggest that the osteogenic potential of Shox2-positive cells is Meis2-dependent during palatal osteogenesis. Meis2 appears to be a key determinant in directing the osteogenic fate of Shox2positive palatal cells. Together with the fact that the key osteogenic genes bound by MEIS2 lost their chromatin accessibility in the absence of Meis2, these results pinpoint a fundamental role of MEIS2 in palatal bone formation by setting up a ground state for palatal osteogenesis.

Discussion
In humans, substantial lines of evidence support that MEIS2 is a candidate gene for cleft palate (7,(20)(21)(22), but the mechanisms underlying MEIS2-mediated defects are completely

MEIS2 regulates palatal osteogenesis
unknown. In this study, we conducted a comprehensive investigation of the role of MEIS2 in palatal development. Tissuespecific inactivation of Meis2 in CNC cells led to complete cleft palate or submucous cleft with complete loss of palatal bones. Molecular and genomic analyses demonstrate direct regulation of key osteogenic genes by MEIS2. In addition, we also showed, that in the HOX-negative first branchial arch-derived palate, MEIS2 may act together with SHOX2 to regulate the expression of osteogenic genes by binding to distal regulatory elements. The fact that Meis2 is highly associated with osteogenic potential and required for chromatin accessibility of osteogenic genes supports a vital function of MEIS2 in providing a ground state for palatal osteogenesis.
Similar to the relatively stabilized but varying features of cleft palate defects in humans with MEIS2 haploinsufficiency, Meis2 mutant mice exhibited a typical CCP or ICP, indicating a conserved role of MEIS2 in regulating palatogenesis. Importantly, Meis2 mutant mice with ICP lack all palatal bones, in contrast to several previously reported animal models with submucous cleft that presented partial defects of palatal bones (12,13,23). Although a critical role of MEIS2 in embryonic skeleton development has been reported previously (3,6), the functional mechanisms of MEIS2 in osteogenesis remain elusive. In this study, we showed that Meis2 mutant palatal shelves failed to express osteogenic markers, indicating that palatal mesenchymal cells lacking Meis2 lose their osteogenic fate. These results raise the intriguing possibility that MEIS2 functions in cell fate determination during palatal development. Palatal bones, including the palatine process of the maxilla and the palatine bone, are formed through intramembranous ossification, a process that requires expression of RUNX2 (24). Many other genes have also been implicated in skeleton formation in the craniofacial region, including the palate (18,(25)(26)(27)(28). All of these osteogenic genes, including Runx2, Shox2, Pbx1, Bmp2, Twist1, and Aldh1a2, were dramatically down-regulated in the palatal mesenchyme lacking Meis2, indicating that MEIS2 functions upstream of a spectrum of osteogenic genes. Intersection of RNA-Seq with MEIS2 ChIP-Seq and ATAC-Seq datasets on the palatal shelves confirmed direct targets of these osteogenic genes by MEIS2. These results demonstrate that MEIS2 functions to establish osteogenic cell fate in palatal mesenchymal cells by directly activating the expression of osteogenic genes. Given the fact that the distal regulatory regions of osteogenic genes that were bound by MEIS2 became inaccessible along with their coding regions, MEIS2 may also acts as an epigenetic modifying factor required for chromatin poising or opening. This is in agreement with previous reports showing that TALE proteins, including PBX and MEIS, have an early role in poising target gene loci for gene activation by HOX proteins (29,30). It appears that this unique role of TALE proteins as transcription factors is conserved in HOX-negative domains.
In the MEIS2-bound regions, the highly enriched SHOX2, PBX1, and RUNX2 binding motifs implicate that these proteins

MEIS2 regulates palatal osteogenesis
act as cofactors for MEIS2 to activate transcription of downstream genes. Indeed, TALE proteins have been shown to act as a functional unit with or without additional transcription factors in all species tested to date (2,31). Interestingly, these transcription factors were also prominently down-regulated in the absence of Meis2, suggesting a complicated regulatory network between MEIS2 and these osteogenic factors in palate development. Among these factors, SHOX2 appears to be a critical cofactor for MEIS2 in palatal bone formation and patterning. This is because Shox2 is expressed in the hard palate, and its inactivation leads to anterior cleft of the secondary palate with defective palatal bone (15,18), and MEIS/PBX binding motifs are highly enriched in the SHOX2-bound regions unraveled by SHOX2 ChIP-Seq on the palate (16). As expected, integration of MEIS2 ChIP-Seq and SHOX2 ChIP-Seq datasets on the palate revealed extensive genome-wide co-occupancy of these two factors, suggesting that, in the HOX-negative palate, SHOX2 and TALE proteins act together to regulate gene expression. In the developing palate, these two factors appear to primarily regulate osteogenesis, as both of them bind to distal regulatory elements of osteogenic genes. Although SHOX2 is a downstream target of MEIS2, they appear to function in parallel, as Shox2 overexpression failed to rescue the bone defects in the palate lacking Meis2.
We have reported previously that, although the Shox2 expression domain is restricted within the anterior hard palate, only a subpopulation of Shox2-expressing palatal mesenchymal cells differentiates into osteoblasts, with a subpopulation of Shox2-expressing cells eventually becoming fibroblasts (15). Interestingly, although MEIS2 is widely expressed in developing palatal shelves, a closer look identified partial overlap of MEIS2 and SHOX2 in the palatal mesenchyme. Despite the lack of direct evidence showing that Shox2-positive cells lacking Meis2 have a fibroblast fate, unbiased single-cell RNA-Seq analysis revealed a high osteogenic potential of palatal cells expressing Shox2 and Meis2 compared with Shox2-positive cells without Meis2 expression. These results indicate a positive role of Meis2 in the osteogenic trajectory. Taken together with the fact that MEIS2 is required for proper chromatin poising/opening of osteogenic genes, our results support a role of MEIS2 in setting up the ground state for palatal bone formation and patterning.

Mouse strains
Generation and genotyping of Wnt1 Cre , Meis2 f/f , R26R mTmG , Shox2 Cre , R26R Shox2 and Shox2 HA alleles have been described previously (6,19,(32)(33)(34)(35). The Meis2 HA allele was generated by using the Alt-R CRISPR-Cas9 crRNA kit (Integrated DNA Technologies), following the manufacturer's protocol. Briefly, crRNA was designed to target the 5ЈATG of Meis2 (RRID: SCR_015723). The synthetic crRNAs were duplexed with trans-activating crRNA (tracrRNA) to form single guide RNA. The single guide RNA and Alt-R S.p. Cas9 nuclease were mixed to generate a ribonucleoprotein complex, which was then immediately electroporated into the C57BL/6J zygotes together with the donor DNA, a single-strand oligo DNA nucleotide containing a 2ϫ HA sequence. Viable electroporated embryos at the two-cell stage were transferred to pseudopregnant females to create founders. Two positive male founders were obtained and backcrossed with C57BL/6J females to generate F1 mice. The primers for genotyping of Meis2 HA alleles were 5Ј-CGTTTTCTTGACTGGGCTTTCC-3Ј (forward) and 5Ј-CGCCATCATCATCAAGCAACC-3Ј (reverse), which amplified a 348-bp product from the HA tag allele and a 291-bp product from the WT allele. All experiments involving animals in this study were approved by the Institutional Animal Care and Use Committee of Tulane University.

RNA isolation, sequencing, data analysis, and quantitative RT-PCR (RT-qPCR) analyses
For the bulk RNA-Seq assay, the anterior palatal mesenchyme from E12.5 and E15.5 Wnt1 Cre ;Meis2 f/f and WT embryos was collected after removal of the palatal epithelium using Dispase II (Sigma). Three biological replicates from each group and genotype were collected for analyses. Total RNA was isolated using the RNeasy Micro Plus Kit (Qiagen). Reverse transcription and amplification of RNA were done and followed by library preparation with the TruSeq RNA

MEIS2 regulates palatal osteogenesis
Sample Preparation Kit v2 (Illumina). Subsequently, singleend high-output flow cell sequencing was performed on an Illumina NexSeq 500. Sequencing reads were mapped to a reference mouse genome (GRCm38/mm10) with HISAT2 (default parameters) (38). Gene expression was counted using featureCounts (39), and the DEGs were further processed with DESeq2 (40). Genes with a log 2 -fold change (FC) of Ϫ1 or less were termed "down-regulated genes," and those with a log 2 -fold change of ϩ1 or more were defined as "upregulated genes". GO analysis was performed using the clus-terProfiler R package (41).
For RT-qPCR, RNA from embryonic palatal mesenchyme was reverse-transcribed into complementary DNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific). Quantitative PCR was performed with iTaq Universal SYBR Green Supermix (Bio-Rad) using a 7500 Fast Real-Time PCR System (Applied Biosystems). The primer sequences are listed in Table S4. Cycle threshold (Ct) values were normalized to Gapdh, and the fold changes were calculated according to the 2 Ϫ⌬Ct method. Student's t test was applied to determine the significance of statistical differences between two groups, and results were presented as mean Ϯ S.D. p Ͻ 0.05 was perceived as statistical significance.

ChIP, library preparation, sequencing, and data analysis
The anterior palate mesenchyme from E12.5 Meis2 HA mouse embryos was fixed in 1% formaldehyde for 10 min and quenched with glycine for 5 min at room temperature. After washing with ice-cold PBS containing protease inhibitor mixture, tissues were dispersed into single-cell suspension on ice. The following ChIP assays were performed with Pierce TM Magnetic ChIP Kit (Thermo Scientific) according to the manufacturer's instructions. Briefly, cells were pelleted in PBS and then resuspended in membrane extraction buffer to isolate nuclei. Purified nuclei were subsequently equilibrated in micrococcal nuclease digestion buffer. Micrococcal nuclease was then added to the nucleus suspension to fragmentize chromatin into one to five nucleosomes (150 -900 bp). Immunoprecipitation was performed with a rabbit antibody against mouse HA overnight at 4°C, followed by 2-h incubation with ChIP-grade protein A/G magnetic beads. After washing, the beads were resuspended with IP elution buffer containing 5 M NaCl and 20 mg/ml proteinase K, and then all of the immunoprecipitate and 10% total input in IP elution buffer were incubated at 65°C for 1.5 h. DNA was purified using a DNA Clean-Up Column, and sequencing libraries were prepared subsequently using the ChIP-Seq DNA Library Prep Kit for Illumina (Cell Signaling Technology). Paired end 2 ϫ 150-bp reads were gained using Illumina HiSeq 2000 and aligned to the mouse genome mm10 with Bowtie2. After filtration, the bigwig file of ChIP was obtained using bamCompare through normalization to input. Peaks were called with MACS2 callpeak controlled to input. The ChIPseeker R package was used to annotate and visualize peaks, followed by GO analysis with clusterProfiler and de novo motif analysis with DREME.

FACS, ATAC-Seq library preparation, sequencing, and data analysis
The GFP-positive anterior palatal shelves were isolated from E12.5 Wnt1 Cre ;Meis2 f/ϩ ;R26R mTmG and Wnt1 Cre ;Meis2 f/f ; R26R mTmG embryos. Samples were subjected to digestion with a mixture of 0.02% collagenase I, II, and IV in Hank's Balanced Salt Solution for 30 min, followed by Accutase solution (Sigma) treatment for 5 min at 37°C. After stopping the digestion with culture medium containing 10% FBS, cells were washed with PBS containing 3% FBS and then filtered through a 70-m filter. Suspended cells were sorted by FACS on a BD FACSAria Fusion cell sorter (BD Biosciences). Around 50,000 single GFPpositive cells from each group were obtained and promptly subjected to generation of an ATAC-Seq library, as described previously (42). Libraries were pooled equimolarly and sequenced using a paired-end 75-bp read mid-output kit on an Illumina NexSeq 500 platform. Three biological replicates were implemented for each group. After removal of the adaptor sequences, truncated reads were mapped to mm10 reference genomes with bowtie 2 (43). The bamCoverage was used to generate a bigwig file, which is visible on the Integrative Genomics Viewer (44). Peaks from each group were identified on a pool of triplicates using MACS2 callpeak (45). DiffBind was used to identify sites that exhibited differential accessibility between control and Meis2 mutant groups (46). ChIPseeker R package was used to annotate ATAC-Seq peaks to identify genes regulated by MEIS2. By comparison with control group, peaks with a -fold change of Ϫ1 or less were considered as "lost accessibility," and those with a-fold change of ϩ1 or more were regarded as "gained accessibility." The clusterProfile R package was used to perform the GO analysis.

Single cell RNA-Seq and data analysis
GFP-positive cells from the E13.5 Shox2 Cre/ϩ ;R26R mTmG anterior palatal shelves were collected after cell suspension and purified by FACS. Immediately, about 5000 live cells were loaded onto the 10ϫ Genomics Chromium system, and the library was prepared using the Single-Cell 3Ј Library and Gel Bead Kit v3, following the manufacturer's protocol. Sequencing was performed on an Illumina NextSeq 500, and Illumina basecall files (*.bcl) were gained for each sample. After conversion to FASTQ, the files were aligned to mouse genome mm10 with CellRanger 3.1.0. The Seurat 3.0 R package was used for QC and visualization, followed by GO analysis with clusterProfiler.

Data availability
All of the RNA-Seq, ChIP-Seq, ATAC-Seq, and single-cell RNA-Seq data were deposited into the GEO with SuperSeries accession number GSE143914. The rest of data are contained within the manuscript.