Role of Cdx factors in early mesodermal fate decisions

The murine Cdx genes, Cdx1, Cdx2 and Cdx4, encode homeodomain transcription factors related to caudal in Drosophila. With the exception of early expression of Cdx2 in the trophectoderm of the pre-implantation embryo, Cdx family members are sequentially activated during gastrulation commencing at embryonic day (E) 7.5, forming a nested expression domain in the primitive streak and subsequently in all three germ layers of the tail bud until approximately E13.5 (Beck, et al., 1995; Gamer and Wright, 1993; Meyer and Gruss, 1993; Strumpf et al., 2005). Cdx gene expression is then extinguished in all tissues except the hindgut endoderm and derivative intestinal epithelium, where Cdx1 and Cdx2 are maintained throughout adulthood (Silberg et al., 2000).

Whereas Cdx2^−/− mutants are peri-implantation lethal, owing to a requirement for this transcription factor in the trophectoderm, Cdx1^−/− and Cdx2^+/− mice are viable and exhibit homeotic transformations of the cervical and thoracic vertebrae with concomitant shifts in Hox gene expression (Subramanian et al., 1995; Chawengsaksophak et al., 1997). Cdx4-null mutants are phenotypically normal; however, the loss of Cdx4 exacerbates Cdx1^−/− or Cdx2^+/− phenotypes, indicative of functional overlap between Cdx family members (Faas and Isaacs, 2009; Savory et al., 2011; van den Akker et al., 2002; van Nes et al., 2006; Young et al., 2009). The derivation and analysis of Cdx2 conditional mutants (Savory et al., 2009a; Stringer et al., 2012) and compound mutant derivatives (van den Akker et al., 2002; Savory et al., 2011; Verzi et al., 2011; van Rooijen et al., 2012; Stringer et al., 2012; Grainger et al., 2010; Hryniuk et al., 2012) further substantiated the functional overlap between family members. Analysis of these mutants also revealed a requirement for Cdx genes in diverse processes in development and in the adult intestinal track. These include, for example, roles for Cdx proteins in specification and development of the intestine and its homeostasis in the adult (Beck et al., 1999, 2003; Chawengsaksophak et al., 1997; Gao et al., 2009; Grainger et al., 2010; Stringer et al., 2012; Hryniuk et al., 2012), yolk sac hematopoiesis (Lengerke et al., 2007; Wang et al., 2008; Brooke-Bisschop et al., 2017; Davidson and Zon, 2006; Davidson et al., 2003), neural tube closure, and elaboration of the post-otic embryo (Savory et al., 2009a, 2011; van Rooijen et al., 2012).

Although Cdx function has been shown to be crucial for numerous developmental programs, our understanding of the target genes governing these programs is incomplete. Cdx family members regulate target gene transcription through binding to a cognate response element (CDRE), of the consensus TTTATG, either in the proximity of their target genes (Suh et al., 1994; Subramanian et al., 1995) or at distal enhancer elements (Maier et al., 2005; Taylor et al., 1997; Gaunt, 2017). Although Cdx family members typically positively regulate transcription, there is a growing body of evidence that suggests they can also act as transcriptional repressors. The means by which Cdx factors regulate transcription is also poorly understood, although Cdx function has been shown to impact chromatin architecture (Verzi et al., 2013, 2010; Saxena et al., 2017; Neijts et al., 2017), consistent with their association with Brg1 (Smarca4) and other members of the SWI/SNF chromatin remodeling complex (Yamamichi et al., 2009; Nguyen et al., 2017).

In agreement with their functional overlap (van den Akker et al., 2002; Davidson and Zon, 2006; van Nes et al., 2006; Faas and Isaacs, 2009), Cdx family members appear comparable in their ability to occupy, and regulate, target genes (Charité et al., 1998; Savory et al., 2009a). Targets identified to date include a number of Hox genes, implicated in Cdx-dependent vertebral patterning (Shashikant et al., 1995; Charité et al., 1998; Subramanian et al., 1995; Pownall et al., 1996; Isaacs et al., 1998; van den Akker et al., 2002; Gaunt et al., 2003, 2008), Cyp26A1, which is essential for catabolism of retinoic acid (RA) and crucial for the maintenance of a caudal progenitor population driving axial elongation (Wingert et al., 2007; Savory et al., 2009a; Young et al., 2009; Martin and Kimelman, 2010), genes encoding products involved in intestinal differentiation and homeostasis (Beck et al., 1999, 2003; Chawengsaksophak et al., 1997; Gao et al., 2009; Grainger et al., 2010; Hryniuk et al., 2012), and targets necessary for yolk sac hematopoiesis (Davidson et al., 2003; Davidson and Zon, 2006; Lengerke et al., 2007; Wang et al., 2008; Brooke-Bisschop et al., 2017).

To establish a more comprehensive understanding of Cdx function during mouse development, we conducted chromatin immunoprecipitation coupled with deep sequencing (ChIP-seq) using material from E8.5 embryos. This analysis revealed Cdx2 to be highly enriched at the loci of genes encoding numerous transcription factors, among which were a number of transcription factors involved in hematopoiesis and cardiogenesis. Consistent with this, we found that loss of Cdx function leads to heart anomalies and mis-expression of several cardiogenic genes, including gain of expression of cardiac markers in Cdx mutant yolk sacs concomitant with loss of expression of genes involved in yolk sac hematopoiesis. Finally, Brg1, a catalytic subunit of the SWI/SNF chromatin remodeling complex required for mesodermal differentiation (Alexander et al., 2015) and which interacts functionally with Cdx2 (Yamamichi et al., 2009; Nguyen et al., 2017), colocalized with Cdx2 at a number of cardiogenic loci. Moreover, Brg1 binding and chromatin structure was altered at these genes in Cdx mutants. As Cdx is essential for yolk sac hematopoiesis (Davidson and Zon, 2006; Davidson et al., 2003; Lengerke et al., 2007; Wang et al., 2008; Brooke-Bisschop et al., 2017), these findings suggest that Cdx function regulates chromatin remodeling events necessary for the coordinated expression of transcription factors that direct hematopoietic versus cardiac fates from a common progenitor population.

To understand the means by which Cdx family members regulate developmental programs, Cdx2 occupancy in the E8.5 embryo was assessed by chromatin immunoprecipitation followed by deep sequencing (ChIP-seq). Of the 13,991 total peaks returned from this analysis, Cdx2 binding was found to be distributed primarily in introns (32% of peaks) and proximal to transcriptional start sites (20% of peaks; Fig. 1A,B). Genomic regions associated with Cdx2 occupancy were also highly enriched for canonical CDREs (Fig. 1C; Dearolf et al., 1989; Suh et al., 1994; Subramanian et al., 1995).

Putative Cdx target genes were selected for further analyses using a minimum peak score of 800 as a cut-off, based on our recovery of previously characterized Cdx2 targets with peak scores ranging from 800 (e.g. Dll1; Grainger et al., 2012) to 3100 (e.g. Scl; Savory et al., 2011; Brooke-Bisschop et al., 2017). Gene ontology (GO) term enrichment analysis was conducted using the 441 peaks annotated to a reference gene that met this threshold. This analysis revealed that Cdx2-occupied loci were significantly enriched for genes involved in transcription, positive and negative regulation of gene expression, and cell fate commitment (Fig. 1D), suggesting that Cdx2 coordinates the expression of gene regulatory networks essential to embryonic development. Consistent with this, 226 of the 441 peaks were associated with genes encoding factors related to regulation of transcription, and six corresponded to genes involved in signaling pathways (Wnt3, Wnt7b, Wnt11, Cyp26A1, Bmp2 and Dll1). This analysis likely excluded potential candidate targets, owing to the relatively high stringency used for cutoff, as well as additional targets that may not be accessible at the developmental stage employed for this analysis. It is, however, interesting to note that a number of annotated peaks that fall below the cutoff are also enriched for genes encoding transcription factors and signaling molecules (Table S1).

Among transcription-related targets, Cdx2 was enriched at a number of Hox genes (e.g. Hoxa1, Hoxa2, Hoxb1, Hoxb2, Hoxb13, Hoxc4 and Hoxd1), consistent with the known roles for Cdx family members in vertebral patterning through regulation of Hox gene expression (Subramanian et al., 1995; Charité et al., 1998; van den Akker et al., 2002; Shashikant et al., 1995; Pownall et al., 1996; Isaacs et al., 1998; Gaunt et al., 2003, 2008). Certain transcriptional regulators of tissue patterning and specification were also enriched for Cdx2 occupancy (e.g. Twist1, Irx2, Gsc, Pitx1, Pax1, Pax3, Pax9, Alx3, Sox1, Sox2, Foxa2 and Foxd3).

As previously described (Wang et al., 2008; Brooke-Bisschop et al., 2017), ChIP-seq identified Scl (Tal1), which encodes a master regulator of hematopoiesis, as a Cdx2 target (Fig. 2A; Shivdasani et al., 1995; van Handel et al., 2012; Org et al., 2015). In addition, other regulators of yolk-sac hematopoiesis, including Lyl1, Lmo2 and Meis1, showed significant Cdx2 binding (Fig. 2A). Of particular note was the finding of a number of genes encoding cardiogenic transcription factors that were highly enriched for Cdx2 occupancy. Such genes encompassed some of the earliest known markers of cardiac progenitors, including Gata4, Nkx2-5, Tbx5 and Isl1, as well as genes required for cardiac looping, such as Hand1 and Hand2 (Fig. 2B; reviewed by Srivastava, 2006).

We validated Cdx2 binding by ChIP-PCR within the genomic loci of select cardiac transcription factors, using primers flanking canonical CDREs within Cdx2 ChIP-seq peaks and chromatin isolated from E8.5 embryos (Fig. 2C). This analysis confirmed Cdx2 occupancy at a number of cardiac gene loci, including Tbx5, Gata4, Hand1 and Hand2 (Fig. 2D). Cdx2 binding to the promoter region of Cyp26A1, a previously characterized Cdx2 target (Savory et al., 2009a), was included as a positive control. Exon 4 of Wnt3a is not occupied by Cdx2 (Nguyen et al., 2017) and no increase in Cdx2 binding relative to IgG was seen at this region. Additionally, Mef2c, a cardiac marker specific to the second heart field (Verzi et al., 2005), was found to have a peak score of only 56.26 in the ChIP-seq dataset, and was not detected by ChIP-PCR. Thus, this locus served as a cardiac-specific negative control for Cdx2 binding in these experiments.

Cdx2^F/FActin::Cre-ER^T mutants (Savory et al., 2009a) were crossed with germline Cdx1-nulls (Subramanian et al., 1995), generating Cdx1^−/−Cdx2^F/FActin::Cre-ER^T compound mutants as previously described (Savory et al., 2011). Post-implantation Cdx1-Cdx2 compound conditional null mutants (hereafter referred to as DKO) lose expression of Cdx4, becoming functionally Cdx-null and thus circumventing both the peri-implantation lethality of Cdx2-null mutation and functional overlap between family members (Savory et al., 2011, 2009a; van den Akker et al., 2002; van Nes et al., 2006; Young et al., 2009). DKO mutants in which Cdx2 deletion was induced at E5.5 (i.e. prior to the normal onset of embryonic Cdx expression) developed beating heart tubes at E9.5 but exhibited a distended pericardial sac and aberrant heart tube morphology, and typically died at E10.5 (Savory et al., 2011; Fig. 3A). Additionally, in situ hybridization (ISH) revealed that the ventricular marker Mlc2v (Myl2) (Franco et al., 1999) was reduced in intensity and spatially restricted in the mutant heart tubes relative to controls (Fig. 3B).

The heart defects observed in DKO mutants were variable, potentially as a result of differences in the embryonic stage at which Cdx2 deletion was induced. To investigate this, tamoxifen was administered at E4.5 to assess whether loss of Cdx function at an earlier stage might exacerbate the cardiac phenotype. This earlier deletion, however, resulted in fully penetrant embryonic lethality, likely because of the requirement for Cdx2 during implantation (Chawengsaksophak et al., 1997). Although tamoxifen treatment at E6.5 also resulted in cardiac defects (Fig. S1), these were less severe than those elicited by Cdx2 deletion at E5.5, suggesting a narrow developmental window during which loss of Cdx function impacts heart development.

To characterize further the effect of Cdx loss of function on cardiogenesis, the expression of Nkx2-5 and Tbx5, early markers of cardiac progenitors (Bruneau et al., 1999; Komuro and Izumo, 1993; Lints et al., 1993) and highly enriched for Cdx2 occupancy (Fig. 2B), were assessed by ISH. At E8.5, Nkx2-5 transcripts were distributed in a triangle-like expression domain in DKO embryos rather than the cardiac crescent labeling typically observed in controls (Fig. 3C). At this stage, Tbx5 transcripts were also localized to the cardiac crescent in control embryos, whereas DKO mutants exhibited broader lateral distribution of Tbx5-positive cells that tapered toward the midline (Fig. 3D). At E9.5, Tbx5 expression in DKO embryos was even less spatially restricted relative to controls (Fig. 3E), consistent with prior observations in cdx1a/4-deficient zebrafish embryos (Lengerke et al., 2011).

Prior work has shown that Cdx function is required for proper yolk sac vascularization and hematopoiesis, and this may occur, in part, through Cdx-dependent regulation of Scl expression (Brooke-Bisschop et al., 2017; Wang et al., 2008). In this regard, both cardiac and yolk sac hematopoietic progenitors are derived from a common pool of multi-potential Mesp1⁺ mesodermal precursors (Chan et al., 2013). Mesp1⁺ mesoderm fated to give rise to early blood lineages must therefore initiate the expression of a hematopoietic gene expression program while simultaneously repressing the expression of genes involved in cardiogenesis (Chan et al., 2013). To determine whether the loss of hematopoietic potential in DKO mutant yolk sacs is accompanied by ectopic cardiac differentiation, indicative of a fate switch in the absence of Cdx, cardiac gene expression was assessed by qRT-PCR in E8.5 and E9.5 DKO yolk sacs and compared with littermate controls. Ectopic expression of Tbx5 was observed in mutant yolk sacs at both stages (Fig. 4A), consistent with a role for Cdx in repressing cardiac differentiation in this lineage. In contrast, variable changes in expression were observed for yolk sac Nkx2-5 and cTnT (Tnnt2) (Fig. 4B,C), suggesting that such Cdx-dependent repression may be specific to Tbx5.

In addition to mesoderm-intrinsic processes, signals from both the endoderm and the ectoderm impact heart development (Zaffran and Frasch, 2002). As the conditional mouse model employed here uses a CMV-Actin::Cre-ER^T transgene to induce global Cdx2 gene deletion, the loss of Cdx function on cardiogenesis may occur through non-mesodermal pathways. To investigate this, we utilized a Mesp1-Cre knock-in allele (Mesp1^+/Cre; Saga et al., 1999) to effect Cdx2 deletion. Mesp1 expression initiates at E6.5, and descendant cells contribute to cardiac progenitors of both the first and second heart fields, hematopoietic and endothelial progenitors in the yolk sac, and myogenic mesoderm (Devine et al., 2014; Chan et al., 2013; Saga et al., 1999).

At E11.5, Cdx1^−/−Cdx2^F/FMesp1^+/Cre mutants appeared smaller than littermate controls, lacked red blood cells, exhibited a malformed heart with pericardial edema (Fig. 5A), and were never recovered beyond E12.5. E8.0 Cdx1^−/−Cdx2^F/FMesp1^+/Cre mutants and littermate controls were assessed for Nkx2-5 expression, which marks cardiac progenitors of both the first and second heart fields (Stanley et al., 2002). In contrast to Cdx DKO mutants, Nkx2-5 expression in these embryos was indistinguishable from controls (Fig. 5B), suggesting that the cardiac defects observed in Cdx1^−/−Cdx2^F/FMesp1^+/Cre mutants either manifest after E8.0, or occur through mechanisms that do not impact Nkx2-5.

In E9.5 embryos, expression of Mlc2v was comparable in the prospective ventricles in Cdx1^−/−Cdx2^F/FMesp1^+/Cre mutants and littermate controls (Fig. 5C,D). However, the positioning of the heart was affected in the mutants, which also exhibited a narrowed outflow track (Fig. 5D, arrowhead) and underdevelopment of the right ventricle. These observations suggest that Mesp1^+/Cre-mediated deletion of Cdx2 results in a milder cardiac phenotype than that elicited by CMV-Actin::CreER^T. Consistent with this, although the tail buds of E9.5 Cdx1^−/−Cdx2^F/FMesp1^+/Cre embryos were foreshortened relative to controls, they did not exhibit the severe axial truncation consistently observed in DKO mutants (Fig. 5C; Savory et al., 2011). These differences may be due to Cdx deletion in axial progenitors in the tail bud of Cdx1^−/−Cdx2^F/F CMV-Actin::CreER^T mutants, whereas Mesp1^+/Cre is not anticipated to impact this niche (Saga et al., 1999; Chan et al., 2013).

Cre-negative Cdx1^−/−Cdx2^F/F embryos did not display obvious cardiac defects, nor did Cre⁺ littermates heterozygous for Cdx2 (Cdx1^−/−Cdx2^+/FMesp1^+/Cre; data not shown), consistent with prior findings (Subramanian et al., 1995; van den Akker et al., 2002). These defects were also not induced by expression of Cre alone in either model (Mesp1-Cre or CMV-Actin::CreER^T), with or without tamoxifen treatment (data not shown).

In addition to aberrant cardiac morphogenesis, Cdx1^−/−Cdx2^F/FMesp1^+/Cre mutants exhibited defective yolk sac vascularization similar to that observed in DKO mutants (Fig. S2A; Brooke-Bisschop et al., 2017). Deficiency in yolk sac hematopoiesis can lead to pericardial edema (Shivdasani et al., 1995; Moser et al., 2004). It is therefore possible that the cardiac defects in Cdx mutants are secondary to the failure of yolk sac hematopoiesis. To assess this, a Tie2^+/Cre transgenic line, in which expression of the Cre recombinase is largely restricted to endothelial and hematopoietic lineages (Tang et al., 2010), was used to delete Cdx2 in a Cdx1-null background as previously described (Brooke-Bisschop et al., 2017). In addition to being smaller than littermate controls and deficient in yolk sac hematopoiesis (Fig. 5E,F; Brooke-Bisschop et al., 2017), Cdx1^−/−Cdx2^F/FTie2^+/Cre embryos displayed aberrant cardiac morphology at E10.5, with distended pericardial sacs and a narrow, misshapen heart tube (Fig. 5F). It may be that the loss of Cdx function impacts cardiac development secondary to yolk sac hematopoiesis; however, mis-expression of Tbx5 and Nkx2-5 in DKO mutant embryos was observed at E8.5, before circulatory defects are anticipated to influence cardiac morphogenesis. Furthermore, Cdx2 binding occurs at the genomic loci of a number of cardiac targets, also at E8.5 (Fig. 1B-D). Although it is impossible to distinguish between the two, these data suggest that the aberrant cardiac morphogenesis observed in Cdx mutants likely results from a combination of defects that are both primary and secondary to Cdx deletion.

Cdx2 expression has been reported in the chorion, ectoplacental cone, primitive streak and allantois as early as E7.5 (Beck et al., 1995); however, the contribution made by Cdx2⁺ cells and their descendants to the cardiac lineage is unknown. Moreover, this pattern of Cdx2 expression is inconsistent with a role in cardiac mesoderm specification from Mesp1⁺ progenitors, which commences at E6.5 (Saga et al., 1996; 2000; Bondue et al., 2008). To address these discrepancies, we analyzed single cell RNA-seq data generated by Scialdone and colleagues (Scialdone et al., 2016), which revealed that Cdx2 transcripts are indeed present in the murine embryo as early as E6.5 (Fig. 6A-D). Cdx1 transcripts were also identified at this stage, whereas Cdx4 transcripts were rare (Figs S3 and S4). Cdx2⁺ cells were found within the epiblast at E6.5 in nascent mesoderm, posterior mesoderm, allantois and blood progenitors, among others (Fig. 6C). Whole-mount immunohistochemistry also revealed Cdx2 in the E6.5 embryo, including the anterior region of the primitive streak (Fig. 6E), corresponding with the region associated with cardiac mesoderm specification and egression. In addition, Cdx2 transcripts were detected in E6.5 embryos by RT-PCR in both embryonic and extra-embryonic material (Fig. 6F), and Cdx1, and to a lesser extent Cdx4, were also expressed (Fig. S5). Consistent with previous reports (Beck et al., 1995; Mcdole and Zheng, 2012), Cdx2 levels were more robust at E7.5 (Fig. 6E). Taken together, these data demonstrate that Cdx2 is present in the appropriate region and at the appropriate developmental stage to be able to impact cardiac progenitors directly.

Cdx2-positive cells frequently co-expressed Mesp1 (Fig. 6G), but rarely expressed Tbx5 or Mef2c, markers of the first and second heart fields, respectively (Fig. 6I,J; Bruneau et al., 1999; Verzi et al., 2005). Although there was a weak correlation between Cdx2 expression and that of additional heart markers (Fig. S6), these genes, as well as Cdx, are also expressed in other mesodermal lineages. Correlations between Cdx1 or Cdx4 expression and the expression of Mesp1, Tbx5 and Mef2c were similar to those observed between Cdx2 and these transcripts (Scialdone et al., 2016; data not shown), in agreement with functional overlap among Cdx family members. Conversely, a subset of Cdx2⁺ cells co-expressed Scl (Fig. 6H) and other hematopoietic markers (Fig. S6). Taken together, these data are consistent with a role for Cdx2 in repressing cardiogenic, and promoting hematopoietic, differentiation in Mesp1⁺ progenitors.

Brg1, the catalytic subunit of the SWI/SNF chromatin remodeling complex, is required for mesoderm induction and cardiac progenitor specification in differentiating embryonic stem cells (ESCs; Alexander et al., 2015), as well as heart development (Stankunas et al., 2008; Hang et al., 2010). Cdx2 physically associates with Brg1, and other members of the SWI/SNF complex, and Cdx2-dependent expression of at least some targets is associated with chromatin remodeling events (Nguyen et al., 2017; Verzi et al., 2010, 2013).

To determine whether a similar association between Cdx2 and Brg1 was observed at cardiogenic loci, the Cdx2 ChIP-seq data were intersected for Brg1 chromatin occupation derived from ChIP-seq analysis of ESCs at the mesoderm stage of directed differentiation to cardiomyocytes (Alexander et al., 2015). This exercise revealed a number of cardiac loci occupied by both Cdx2 and Brg1, including Tbx5, Nkx2-5, Gata4, Hand1 and Hand2, whereas the Isl1 locus was occupied by Cdx2 only (Fig. 7A).

To determine whether Brg1 occupies Cdx2-bound cardiac genes in a Cdx-dependent manner, ChIP-PCR was performed for Brg1, comparing wild-type with DKO embryos at E9.5 (Fig. 7B). Brg1 binding was impaired within Cdx2-occupied regions of Tbx5, Gata4 and Nkx2-5 genes as well as the previously characterized Cdx targets Cyp26A1 and Dll1 (Savory et al., 2009a; Grainger et al., 2012). These findings are consistent with a model whereby Cdx2 recruits Brg1 to select cardiac targets to regulate gene expression via SWI/SNF chromatin remodeling.

To determine whether impaired Brg1 binding to cardiac targets results in altered chromatin organization, formaldehyde-assisted isolation of regulatory elements (FAIRE) coupled with PCR was used to compare chromatin accessibility between wild type and DKO mutants. Given the ectopic cardiac gene expression observed in yolk sacs, and that hematopoietic yolk sac mesoderm shares a common Mesp1⁺ progenitor with the cardiac lineage, yolk sacs from wild-type and DKO embryos at E9.5 were used for these experiments. Changes in chromatin accessibility at Cdx2-occupied CDREs were observed at Tbx5, Gata4 and Nkx2-5 loci, whereas accessibility at the β-actin locus remained unaffected in DKO yolk sacs, and thus served as a negative control (Fig. 7C). Although these changes were variable and did not reach statistical significance, likely owing to the cellular heterogeneity of the yolk sac, they are consistent with Cdx-dependent chromatin organization and in agreement with prior observations in other systems (Verzi et al., 2010, 2013; Nguyen et al., 2017; Saxena et al., 2017).

ChIP-seq analysis of Cdx2 occupancy in E8.5 embryos revealed a marked enrichment at loci encoding transcription factors, suggesting that Cdx coordinates gene regulatory networks essential to numerous developmental processes, a finding consistent with the complex phenotype of Cdx-null mutants (Beck, et al., 1995; Gamer and Wright, 1993; Meyer and Gruss, 1993; van den Akker et al., 2002; Strumpf et al., 2005; Savory et al., 2009a, 2011; van Rooijen et al., 2012). A subset of these transcription factors are involved in cardiogenesis and hematopoiesis, both of which are known roles for Cdx (Lengerke et al., 2007; Wang et al., 2008; Brooke-Bisschop et al., 2017). Cardiac defects observed in Cdx DKO embryos, together with altered gene expression patterns, suggest that Cdx function is required for normal heart development, potentially governing fate decisions between hematopoietic and cardiac lineages from a common progenitor pool. Finally, Brg1 was recruited to a subset of cardiogenic loci in a Cdx-dependent manner. Together with the finding of altered chromatin architecture at some of these genes, these observations lead us to propose a model in which Cdx factors, transiently expressed in mesodermal progenitors within the primitive streak, occupy the loci of fate-determining transcription factors with Brg1 and the SWI/SNF complex, leading to establishment of the chromatin architecture required for the fidelity of subsequent gene expression programs at later developmental stages (Fig. 8).

ChIP-seq analysis revealed that many of the robust Cdx2 peaks localized to transcription factor loci. Consistent with this, GO terms highly over-represented in this dataset included ‘transcription’, ‘positive and negative regulation of gene expression’ and ‘cell fate determination’, suggesting that Cdx2 impacts the expression of a number of transcriptional networks implicated in development. Indeed, transcription factors represented 226 of the 441 annotated peaks exhibiting binding above the threshold of 800. This association is further reflected in some of the phenotypes exhibited by Cdx mutants. For example, Cdx1^−/− and Cdx2^+/− mice and compound mutant derivatives thereof typically display vertebral homeotic transformations associated with altered expression of Hox target genes (Subramanian et al., 1995; Chawengsaksophak et al., 1997). Cdx DKO and triple-knockout mutants exhibit precocious axial truncation, which may be ascribed, at least in part, to illicit RA signaling in the tailbud caused by loss of Cyp26A1 expression (Savory et al., 2009a; Young et al., 2009; Martin and Kimelman, 2010).

Cdx2 was found to be resident at a number of genes encoding regulators of yolk sac hematopoiesis, consistent with previous work underscoring a requirement for Cdx in hematopoiesis in diverse vertebrate species (Brooke-Bisschop et al., 2017; Lengerke et al., 2007; Davidson and Zon, 2006; Davidson et al., 2003; Wang et al., 2008). In the present study, Cdx2 binding was also enriched at the loci of numerous cardiogenic transcription factors (e.g. Nkx2-5, Tbx5, Gata4, Isl1, Hand1 and Hand2).

Prior work has suggested a role for Cdx in early mesoderm fate decisions (Mendjan et al., 2014), including suppression of cardiogenesis (Lengerke et al., 2011; Rao et al., 2016). Although the basis underlying these observations is poorly understood, occupancy of cardiogenic and hematopoietic transcription factors by Cdx2, coupled with ectopic cardiogenic gene expression in DKO mutants, is consistent with a role for Cdx2 in suppressing cardiogenesis and promoting hematopoietic gene expression programs within a common Mesp1⁺ mesodermal precursor, an event believed to occur during early gastrulation at E6.5 (Saga et al., 1999; Chan et al., 2013). However, the onset of Cdx gene expression in the embryo proper is reported to occur at around E7.5 (Beck et al., 1995; Gamer and Wright, 1993; Meyer and Gruss, 1993; Strumpf et al., 2005), and is therefore inconsistent with a role in this specification event. Recent single cell RNA-seq performed and published by Scialdone et al. (2016) suggests that these early Mesp1⁺ progenitors do, in fact, express Cdx family members as early as E6.5. This was further supported by whole-mount immunohistochemistry, which revealed Cdx2 protein in the primitive streak at E6.5, thus placing Cdx in the appropriate developmental context to influence early Mesp1⁺ mesoderm specification events. Furthermore, an onset for Cdx function at the early primitive streak stage supports previously described roles for Cdx family members in early post-otic embryonic development (van Rooijen et al., 2012; Chawengsaksophak et al., 2004, 1997; Savory et al., 2009a; Amin et al., 2016; Subramanian et al., 1995) and craniorachichisis (Savory et al., 2011).

In contrast to its role at hematopoietic loci, Cdx2 occupancy at cardiac loci is likely inhibitory, as previous studies using directed differentiation of mouse and human ESCs in vitro demonstrated increased differentiation to cardiac fates in the absence of Cdx function (Lengerke et al., 2011; Rao et al., 2016) and decreased cardiac gene expression in models of Cdx gain of function (Lengerke et al., 2011). Consistent with this, single cell RNA-seq revealed that all three Cdx family members are expressed in Mesp1⁺ progenitor cells in the mouse, but are not co-expressed with Tbx5 or Mef2c, suggesting that Cdx expression is incompatible with the acquisition of a cardiac fate. A role for Cdx in suppressing cardiogenesis is further evidenced by ectopic Nkx2-5 and Tbx5 expression in the cardiac crescent of DKO mutants, and ectopic Tbx5 expression in DKO yolk sacs.

Our observation that Tbx5 expression is misregulated in DKO mutants is consistent with previous work in zebrafish that revealed expanded tbx5a expression in cdx1a/4 compound mutants (Lengerke et al., 2011). Moreover, ectopic Tbx5 expression in DKO mutant yolk sacs indicates that Cdx-deficient yolk sacs maintain sufficient plasticity to be redirected to a cardiac fate. Notably, a similar role has been described for Scl (Van Handel et al., 2012; Org et al., 2015), which functions downstream of Cdx (Brooke-Bisschop et al., 2017; Wang et al., 2008; Davidson and Zon, 2006; Davidson et al., 2003). Although Cdx may impact mesodermal cell fate decisions in part via Scl, our finding of direct binding of Cdx2 to numerous cardiogenic loci, together with the mis-expression of some of these genes in DKO mutants, suggests that Cdx also plays a direct role in repressing cardiogenic gene expression.

Although Tbx5 expression was consistently upregulated in DKO mutant yolk sacs, random alterations in Nkx2-5 expression suggests that Cdx factors may not repress all cardiac targets similarly within a given context. In this regard, in the zebrafish, embryonic nkx2-5 expression remains unaffected in Cdx mutants until they are treated with RA antagonists, suggesting that retinoid-mediated compensatory mechanisms are necessary to effect nkx2-5 expression in the absence of Cdx (Lengerke et al., 2011). RA signaling not only regulates Cdx1 expression (Houle et al., 2000, 2003; Prinos et al., 2001) but is also involved in heart development, and altered retinoid signaling can lead to aberrant cardiac mesoderm specification and subsequent defects in cardiac morphogenesis (Zile, 1998; Dickman and Smith, 1996). This suggests that murine Cdx factors are necessary, but not sufficient, for the regulation of certain Cdx targets, such as Nkx2-5, and that RA, or other signaling pathways, impact expression of such genes in Cdx mutants.

Defects in yolk sac hematopoiesis, such as those observed in Cdx mutants (Savory et al., 2011; Brooke-Bisschop et al., 2017; Lengerke et al., 2008; Wang et al., 2008; Davidson and Zon, 2006; Davidson et al., 2003), can also affect cardiac development (Shivdasani et al., 1995; Moser et al., 2004). Consistent with this, Cdx1^−/−Cdx2^F/FTie2-Cre mutants exhibited aberrant cardiogenesis, suggesting that the absence of blood contributes to cardiac defects in these mutants. Nevertheless, cardiac defects in DKO mutants are preceded by altered expression of Nkx2-5 and Tbx5 as early as E8.5, likely before yolk sac hematopoiesis is sufficiently compromised to impact heart tube morphogenesis. Furthermore, Cdx2 binding to a number of cardiac loci supports a direct role for Cdx during heart field specification and argues against aberrant cardiogenesis in DKO mutants being entirely secondary to circulatory deficiencies.

Although similar morphological defects were observed in Cdx mutant heart tubes produced by CMV-Actin::CreER^T and Mesp1-Cre mediated recombination, perturbations in the expression of cardiac transcription factors were only observed in CMV-Actin::CreER^T mutants, arguing against Cdx function in Mesp1⁺ progenitors as causal to these heart defects. It may be that cell Cdx-dependent non-autonomous cues instructive to cardiac morphogenesis originating from the endoderm and/or ectoderm are disrupted only in the CMV-Actin::CreER^T mutant model. For instance, BMP2 expressed in the endoderm (Schultheiss et al., 1997; Somi et al., 2004) is required for the formation of a heart in zebrafish (Frasch, 1995) and Xenopus (Shi et al., 2000), as is its homolog dpp in Drosophila (Kishimoto et al., 1997). Although cardiac mesoderm is induced in BMP2-deficient mice, subsequent heart morphogenesis is perturbed (Zhang and Bradley, 1996). In this regard, we identified Bmp2 as a potential Cdx2 target gene by ChIP-seq in E8.5 embryos, the stage at which the heart tube emerges and commences looping. Thus, Cdx regulation of BMP signaling from non-mesodermal germ layers may be preserved in Mesp1-Cre driven Cdx mutants, yet lost in CMV-Actin::CreER^T mutants, possibly contributing to the stronger phenotype in the latter. Further work will be required to examine this potential relationship.

Cdx proteins act as intermediaries that relay caudalizing signals to regulate expression of transcription factors, such as the Hox genes (Shashikant et al., 1995; Charité et al., 1998; Subramanian et al., 1995; Pownall et al., 1996; Isaacs et al., 1998; van den Akker et al., 2002; Gaunt et al., 2003, 2008). Consistent with this, previous work has demonstrated Wnt- and RA-dependent regulation of Cdx family members (Houle et al., 2000, 2003; Prinos et al., 2001; Ikeya and Takada, 2001; Pilon et al., 2006). Directed differentiation of ESCs to cardiomyocytes in vitro mimics signaling gradients in the primitive streak, and requires an initial phase of Wnt signaling for mesoderm induction (Lindsley et al., 2006; Gadue et al., 2006; Haegel et al., 1995), followed by Wnt inhibition to permit differentiation to cardiac cell fates (Lickert et al., 2002; Ueno et al., 2007; Burridge et al., 2012; Yang et al., 2008; Murry and Keller, 2008; Marvin et al., 2001; Schneider and Mercola, 2001; Tzahor and Lassar, 2001); continued Wnt signaling promotes mesodermal progenitor commitment to alternative lineages, including blood (Woll et al., 2008; Wang et al., 2008; Rao et al., 2016). Such models have placed Cdx function downstream of Wnt to repress cardiogenesis, thereby promoting differentiation to such alternative mesodermal lineages (Lengerke et al., 2011; Mendjan et al., 2014; Rao et al., 2016). Thus, cardiac inhibition by Cdx factors as shown here, coupled with activation of hematopoietic regulators, is consistent with their role downstream of Wnt during mesodermal cell fate decisions.

Cdx2 has been shown to influence chromatin accessibility in the intestinal epithelium (Verzi et al., 2010, 2013), and interacts with Brg1 at some target loci in other models (Nguyen et al., 2017; Yamamichi et al., 2009). Furthermore, Cdx2 appears to be required for Brg1 recruitment to cardiac genes. Chromatin organization within these regions was also disrupted in DKO yolk sacs, suggesting that the interaction between Cdx2 and Brg1 at cardiac genes impacts chromatin accessibility. It is possible that Cdx2 recruits Brg1 and SWI/SNF chromatin remodeling activity to cardiogenic loci to establish non-permissive chromatin, thereby epigenetically silencing gene expression in non-cardiac lineages at later developmental stages. However, not all Cdx2-bound loci were co-occupied by Brg1. For example, at the Tbx5 gene locus, Brg1 binding was robust at the Cdx2 ChIP-seq peak most distal to the transcriptional start site (TSS), whereas the more proximal peak was occupied by Cdx2 alone (Fig. 7A). Cdx2 and Brg1 were also present at the Hand1 locus, although Brg1-bound regions flanked Cdx2 peaks, suggesting that these factors may both impact Hand1 expression, but do not co-occupy genomic intervals at this particular gene. Conversely, Isl1 was occupied by Cdx2 alone, suggesting that regulation by Cdx at some cardiac targets might occur independent of SWI/SNF complex recruitment. Although we cannot exclude Brg1 occupation of such genes in lineages not surveyed here, and/or at other times, these observations suggest that Cdx2-dependent transcription occurs through Brg1-dependent and -independent mechanisms.

In summary, the findings presented here suggest a model in which Cdx factors regulate gene expression through epigenetic mechanisms following recruitment of the SWI/SNF complex. This putative mechanism allows for Cdx⁺ cells and their progeny to initiate the appropriate gene regulatory networks at developmental stages when cell fate decisions are made, after Cdx expression is extinguished.

Cdx1^−/−Cdx2^F/F, CMV-Actin::CreER^T, Tie2^+/Cre, and Mesp1^+/Cre mouse lines have been previously described (Savory et al., 2009a; Santagati et al., 2005; Saga et al., 1999, 2000; Kisanuki et al., 2001; Tang et al., 2010). Cdx1^−/−Cdx2^F/F females were paired with Cdx1^−/−Cdx2^F/F CMV-Actin::CreER^T males for timed matings, with noon of the day of vaginal plug discovery considered E0.5, and Cdx2 deletion induced with 2 mg tamoxifen in corn oil by oral gavage at E5.5. Mesp1^+/Cre or Tie2^+/Cre animals were crossed into the Cdx1^−/−Cdx2^F/F line, and Cdx1^−/−Cdx2^+/FMesp1^+/Cre or Cdx1^−/−Cdx2^+/FTie2^+/Cre males were paired with Cdx1^−/−Cdx2^F/F females for timed matings, as above. Animals were maintained according to the guidelines established by the Canadian Council on Animal Care and Animal Care and Veterinary Services at the University of Ottawa.

Rabbit polyclonal anti-Cdx2 antibodies have been previously described (Savory et al., 2009b). Brg1 antibodies were purchased from Abcam (ab110641). Control rabbit IgG antibodies were purchased from Sigma (i8140).

Embryos from timed matings were dissected, crosslinked with 1% formaldehyde, and ChIP performed as previously described (Pilon et al., 2006) with the following modifications: chromatin was sonicated for 1-min intervals six times using a Branson Sonifier 450 at output 3 with 30% duty cycle, and 10 µg of polyclonal rabbit-anti-Cdx2 (Savory et al., 2009b) and pre-immune rabbit-anti-IgG (Sigma) was coupled to 200 µl of magnetic protein G Dynabeads (Invitrogen). Samples were purified using the QIAquick PCR Purification Kit (Qiagen) and library preparation and sequencing performed by the Genome Quebec and McGill Innovation Centre using the mm9 reference genome. ChIP experiments were performed as above using 5 µg of antibody against Cdx2 (Savory et al., 2009b), or IgG control (Sigma), coupled to 50 µl of protein A/G Sepharose beads (Santa Cruz Biotechnology). DNA purification was followed by PCR using primers directed against sequences flanking putative Cdx response elements (CDREs), identified by TRANSFAC, within genomic intervals associated with enriched Cdx2 occupancy as determined by ChIP-seq. Primer sequences are listed in Table S2.

ChIP-seq tracks for Brg1 binding in ESCs at the mesoderm stage of differentiation to cardiomycotes in culture (Alexander et al., 2015) were intersected with Cdx2-ChIP seq data using the UCSC genome browser (https://genome.ucsc.edu/).

Cdx2 ChIP-seq peaks with peak scores greater than 800 was assembled and used for GO term enrichment analysis. Gene IDs were uploaded to the Gene Ontology Consortium enrichment analysis tool (www.geneontology.org/page/go-enrichment-analysis), and only those GO terms with a fold enrichment of 5 or greater were considered. GO terms were sorted by P-value and plotted as –log₁₀(P-value).

Embryos were dissected at E8.5 or E9.5 in chilled PBS, pooled according to genotype, stage-matched based on somite number (or head morphology in cases of axial truncation) and processed in parallel as described (Houle et al., 2000; Wilkinson et al., 1990). Probes were generated from previously described plasmids encoding Nkx2-5 (Komuro and Izumo, 1993), Tbx5 (Bruneau et al., 1999) or Mlc2v (Molkentin et al., 1997).

Whole-mount immunohistochemistry was performed using antibodies against Cdx family members on E6.0-E7.5 embryos as previously described (Qiu et al., 1997; Savory et al., 2009b).

Embryos or yolk sacs were collected at E8.5-9.5 and RNA extracted using Trizol (Gibco BRL) according to the manufacturer's instructions. cDNA was synthesized by standard methods and amplified by semi-quantitative RT-PCR with GoTaq (Promega) using a DNA Engine Tetrad 2 (Bio-Rad), or by RT-qPCR using SYBR green (Promega) and the CFX 96 (Bio-Rad) thermocycler. Either β-actin or Gapdh was used as an input control, and RT-qPCR results were analyzed using the 2^−ΔΔCt method. Oligonucleotides (listed in Tables S3 and S4) were validated before use and dissociation curves were considered for each amplicon to ensure specificity.

Single cell RNA-seq was originally performed using E6.5-E7.75 whole embryos by Scialdone et al. (2016). Single cell read counts were downloaded from http://gastrulation.stemcells.cam.ac.uk and plotted against expression levels of Cdx family members.

Experiments were performed using E9.5 yolk sacs isolated from wild-type and DKO mutant embryos. Chromatin was sheared for seven 1-min intervals on ice using a Branson Sonifier 450 with a 30% duty cycle on power output 3, and DNA accessibility was assessed according to previously published protocols (Simon et al., 2012). Primers are listed in Table S5.

We thank Filippo Rijli for the Actin::Cre-ER^T mouse line, Benoit Bruneau for Nkx2-5 and Tbx5 in situ hybridization probes, Ilona Skerjanc for the Mlc2v in situ hybridization probe, David Cook for further analyses of the single cell RNAseq data, Sanzida Jahan for her expertise on FAIRE-PCR, and the University of Ottawa Animal Care and Veterinary Service for excellent mouse husbandry.