Background

Vertebrate development is a complex, sequential process, which is guided and controlled by a large number of transcription factors acting in a spatio-temporal manner. Many of these transcription factors are pleiotropic in nature, expressing and functioning in multiple tissues simultaneously. Pleiotropy is a feature of some genes, which refers to the ability to affect multiple, distinct and unrelated phenotypes and has important implications in development, disease and evolution [1–4]. Although pleiotropy has been studied over a period of time for genes involved in various biological functions, very few in-depth molecular studies have been conducted looking at a single gene in detail covering its roles in multiple tissues and trying to understand the independent and /or overlapping molecular mechanisms involved in its disparate functions.

The mammalian homeobox gene Bapx1 (Nkx3.2) encodes a transcription factor that belongs to the eponymous Nkx[5] gene family ortholog of the Drosophila bap gene [6, 7]. During mouse embryogenesis Bapx1 is expressed in five distinct tissues - vertebral column, spleen, gut, forelimb and hindlimb, yet appears to have a unique function in only a subset of these. Bapx1 null mice are affected by lethal skeletal dysplasia, with severe malformation or absence of specific bones of the vertebral column and cranial bones of mesodermal origin. The most severely affected parts of the skeleton are the ventral structures, which are in close proximity to the notochord. Bapx1 null mutants also show visceral mesoderm defects leading to asplenia [8–12]. The Bapx1 null embryos exhibit a shortened anterior gut segment and a loss of the pyloric constriction [11]. Although Bapx1 is expressed in the developing limbs there is no overt limb phenotype in the loss-of-function mutants but a Bapx1 gain-of-function mouse model exhibits preaxial polydactyly and hypoplasia of the tibia [13].

Since Bapx1 is expressed in five distinct tissues it was imperative to segregate these tissues as well as the specific cell type within the tissues expressing Bapx1 for a detailed understanding of its molecular mechanism. We achieved this by targeting the Bapx1 locus with enhanced green fluorescent protein (EGFP) driven by an IRES (Bapx1+/tm2.Tlu) into the 3’UTR of Bapx1 allowing for mice of a wildtype genotype expressing both Bapx1 and EGFP proteins or by inserting EGFP immediately after the translational start codon of Bapx1 (Bapx1tm4.Tlu/tm4.Tlu) to create a null genotype. In both cases EGFP was expressed under the control of endogenous Bapx1 transcriptional regulatory elements. This enabled us to isolate specifically only cells expressing Bapx1 from these heterogeneous tissues using fluorescence activated cell sorting (FACS), and to compare the expression profiles of Bapx1 target genes in wildtype versus loss-of-function embryos as a first step to understanding the pleiotropic roles of Bapx1.

It is essential when unraveling the role of a transcription factor in directing organogenesis to take the complementary and supporting roles played by other transcription factors that are active in the same cells into consideration. To understand such complementary roles we focused on the vertebral column, an organ unique to vertebrates providing body structure and support, which undergoes endochondral ossification of its cartilaginous structures to form bones. To decipher the possible molecular mechanism of chondrocyte differentiation and maturation in the vertebral column we co-analyzed the transcription factor Sox9, which has a major role in development and morphogenesis of chondrocytes [14, 15].

Sox9 is a known master regulator in the osteo-chondrogenic lineage but its co-regulatory role with Bapx1 has not been fully explored, though some evidence suggests that it lies in a regulatory cascade upstream of Bapx1[16]. To explore this, we performed similar expression profiling and genome wide binding studies in the vertebral column of E12.5 wildtype and Sox9 null mice, to uncover its role independently, as well as decipher antagonistic and complementary co-regulatory roles between Sox9 and Bapx1, by comparing unique and shared target genes. We uncovered 137 genes, which are regulated by both Bapx1 and Sox9 encompassing very different functions. Finally by combining the genome wide binding and transcriptional profiling data we have generated a gene regulatory network directed by Bapx1 and Sox9 in the mouse vertebral column at E12.5.

Results

Generation of transgenic mouse lines

With the aim of isolating Bapx1 expressing cells by FACS for in vivo comparative gene expression analysis, we targeted the endogenous Bapx1 locus with the enhanced green fluorescent protein (EGFP) reporter gene via homologous recombination in mouse embryonic stem (ES) cells, generating a Bapx1 wildtype (Bapx1tm2.Tlu) and a Bapx1 null (Bapx1tm4.Tlu) transgenic allele, both enabling expression of EGFP under control of the endogenous Bapx1 promoter (Figure 1A adapted from [17]). Bapx1tm2.Tlu was designed with the reporter linked via the internal ribosomal entry site (IRES) at the 3’ UTR of the Bapx1 locus, forming a bicistronic system with two open reading frames [17–20] (Figure 1A). ES cell targeting was verified by genomic Southern blotting (Figure 1B-C adapted from [17]), offspring and embryos were genotyped by PCR (Figure 1D-E). Both Bapx1tm2.Tlu and Bapx1tm4.Tlu showed expression of the EGFP protein in the correct Bapx1-specific domains (Figure 2A) as previously described [6, 8, 11, 12, 17]. The Bapx1tm2.Tlu mice were normal when compared to wildtype Bapx1+/+ littermates [20]. Similar constructs were made for the Sox9 locus to create Sox9tm1.Tlu[20] and Sox9-/-(EGFP) mice that are described in detail in another manuscript.

Figure 1
figure 1

Targeting constructs for Bapx1 . (A) Wildtype Bapx1 allele and targeted Bapx1 alleles producing EGFP via IRES in the wildtype condition (Bapx1tm2.Tlu) and via the endogenous Bapx1 promoter in the loss-of function condition (Bapx1tm4.Tlu) as adapted from [17]. Southern blot with internal and external probes to verify the targeting event of the Bapx1tm2.Tlu(B) and Bapx1tm4.Tlu(C) alleles. Routine PCR genotyping to verify the wildtype and targeted allele (D) and the presence and absence of the neomycin cassette (E).

Full size image
Figure 2
figure 2

Expression profiling of E12.5 Bapx1 and Sox9 producing cells from embryos. (A) Endogenous EGFP tagged E12.5 day whole embryo showing EGFP in domains of expression of Bapx1 and the five distinct organs of expression- spleen, gut, vertebral column (VC), forelimb (FL) and hindlimb (HL). (B) Gene expression profiling of cells sorted from five organs in the wildtype embryo as compared to null showing Bapx1 can function as either an activator or repressor in these organs. (C) Venn diagram showing almost non-overlapping regulation in Bapx1 expressing organs. GO annotations of genes regulated by Bapx1 in vertebral column (D), gut (E) and spleen (F) where the x-axis is -log10 (p-value). (G) E12.5 day whole embryo showing EGFP expression under the endogenous Sox9 promoter in expected domains. (H) GO annotations of genes regulated by Sox9 in the vertebral column (VC) where the x-axis is -log10 (p-value).

Full size image

Identification of genes controlled by Bapx1in five different tissues

Since Bapx1 is expressed in multiple tissues in the developing mouse embryo [6, 8, 10, 12, 13], we fine dissected the five principal Bapx1 domains prior to FACS, namely: vertebral column, spleen, gut, forelimb and hindlimb (Figure 2A). At E12.5 the vertebrate hindlimb is slightly delayed in its development compared to the forelimb [17, 21–24]. To capture the distinct molecular events orchestrated by Bapx1 in fore- and hindlimbs, we analyzed them separately. Our microarray analysis revealed that Bapx1 controlled very different sets of genes in all of these five organs, with little overlap between each of the five tissues. While Bapx1 affected a large number of genes (with >2 fold enrichment) in the spleen (n = 980), gut (n = 143) and vertebral column (n = 125), it controlled only a very limited number of genes in the hind- (n = 30) and forelimb (n = 2) (Figure 2B-C), an observation correlating with the of loss-of function phenotypes in spleen, gut and vertebral column and lack thereof in the limbs, although Bapx1 is highly expressed during limb development [8–10], and Bapx1 gain-of-function in the limbs results in polydactyly [13]. Mostly in vitro findings indicate primarily a repressive function for Bapx1[25–33] and in its absence one would expect increased expression of its primary target genes that it is responsible for silencing.

We then focused on the nature of the regulation by Bapx1 in each of the three tissues with a loss-of function phenotype (spleen, gut and vertebral column) where Bapx1 obviously has an important developmental role. In two of these tissues (spleen and gut) Bapx1 was primarily an activator of downstream genes. We found in Bapx1-specific spleen cells 77% of the genes (761/980) and in Bapx1-specific gut cells 88% (126/143) of the genes showed decreased expression in the Bapx1 null embryos. However, in the Bapx1-specific vertebral column cells 91% (114/125) of the genes were increased in expression in the Bapx1 null embryo, indicating that the Bapx1 protein is primarily a repressor of genes during normal vertebral column development (Figure 2B-C & Additional file 1: Table S1) which is consistent with the in vitro findings for Bapx1 mentioned above.

Though it is always possible in multi tissue studies that the varying levels of gene expression changes seen in different organs is a reflection of the different rate of development of each tissue, it is highly unlikely in this case specially in the spleen and limbs. Spleen starts developing as a separate tissue from the surrounding pancreatic mesenchyme at around E12.5 and the role of Bapx1 in this organ is also important from this time point in development [34]. The expression of Bapx1 in the limb is also first detected at E11.5 [17] and hence we are looking at organs at comparable stages of development. The expression of Bapx1 in both gut and vertebral column is first detected at E9.5 [11, 17] and comparable numbers of genes are changing in the Bapx1 null embryo at the stage under investigation. It is also noteworthy that almost 10 times more genes are changing in spleen compared to the gut or vertebral column pointing to the fact that correlation between developmental stage and impact of the loss of transcription factor in an organ can be at times tenuous.

Bapx1represses inappropriate early maturation of chondrocytes

Bapx1 prevents early maturation of chondrocytes in the vertebral column and its role in vertebral column development is its most studied function. The Bapx1 null embryos die at birth exhibiting a truncated axial skeleton owing to the absence of ossification centers in the vertebrae, lack of cartilage condensations and a markedly reduced intervertebral disc formation [8–10]. During chondrogenesis, Bapx1 is considered to primarily suppress the hypertrophic differentiation of chondrocytes [16], but the genes it regulates pertaining to this function are poorly understood. During endochondral ossification, mesenchymal cells of the condensation differentiate into chondrocytes, marked by their ability to secrete Collagen II, IX, XI and Aggrecan, while repressing Collagen type I production. Upon chondrocyte maturation, proliferation of the chondrocytes stops and they progress to hypertrophy, marked by the switch to the production of type X Collagen instead of type II. Terminal hypertrophic chondrocytes then secrete matrix metalloproteinases to degrade the cartilage matrix, facilitating vascular invasion that brings in osteoprogenitors for ossification [35].

During early axial skeleton development the somite segmentation proceeds from rostral to caudal [35]. Gene ontology (GO) analysis of data from E12.5 embryos using “Database for Annotation, Visualization And Integrated Discovery” (DAVID) [36, 37] revealed enrichment of cartilage condensation/development and neuron development genes. From this, it was evident that Bapx1 controls key genes in cartilage development like Uncx, Chad, Ctgf and Wwp2. We observed down-regulation of these transcripts in the rostral vertebral column of Bapx1-null embryos by in situ hybridization (Figure 3A-J). Thus Bapx1 impacts the level of expression of some chondrogenic genes, but is dispensable for their initiation of transcription. Furthermore, Bapx1 positively regulates Sox5, which regulates extra cellular matrix genes typical for early chondrocyte differentiation like Col2a1 and Acan, in conjunction with Sox9 and Sox6[38, 39]. In this way, Bapx1 indirectly regulates early chondrocyte differentiation.

Figure 3
figure 3

Validation of Bapx1 target genes by RNA in situ hybridization. RNA in situ hybridization for Bapx1 regulated genes in E12.5 wildtype and Bapx1 null embryos validating the expression profiling data for Bapx1 (A and B), Chad (C and D), Ctfg (E and F) , Uncx (G and H), Wwp2 (I and J) in the vertebral column. The arrow marks the loss of expression of the genes in the rostral part of the vertebral column. Similar loss of expression is seen in the spleen of the null embryo for Bapx1 (K and L), Akap12 (M and N) and Gyg (O and P). Abbreviations: Sp, Spleen; St, Stomach; P, Pancreas.

Full size image

At the same time Bapx1 suppresses neural differentiation by inhibiting genes like Alcam, Ntrk3, Corin and Dcx and prevents premature osteogenesis by inhibiting osteoblast marker (Col1a1), cartilage degradation genes with metallopeptidase activity (Ctsk, Adamts4, Adamts2 and Anpep) and prevents blood vessel development by inhibiting Klf5, Vegfc and Nrp1 (FC ≤ 1.5) (Figure 2D and Additional file 1: Table S1). Correspondingly, Ingenuity Pathway Analysis (IPA) analysis showed enrichment of genes in the Wnt and VEGF signaling pathways, as well as the inhibition of matrix metalloproteases (Additional file 2: Table S2). Wnt/B-catenin signaling is known to suppress chondrogenesis, and cartilage degradation and bone matrix formation are associated with chondrocyte maturation [40]. This points out that Bapx1, which largely functions as a repressor in the vertebral column, does not regulate the early differentiation of chondrocytes as much as it represses the inappropriate early maturation of these chondrocytes.

Bapx1controls cell survival in the spleen

Bapx1 has a strong effect on spleen morphogenesis given the Bapx1 null embryos are asplenic at birth [8–10]. The underlying mechanism for this phenotype was unclear as genes controlled by Bapx1 were previously largely unknown. Our transcriptional profiling of Bapx1 specific spleen cells at E12.5 showed that wildtype Bapx1 is an activator of a large number of genes involved in cell division and cell cycle like Bmi1, Cdc14a, Cep55, Ccng2, Lats2, Ccne2, Top1, Stag2, Tfdp1 and Ckap2 (Figure 2E and Additional file 2: Table S2). Moreover, Nkx2.5, a known spleen anlage marker and a regulator of spleen growth through the inhibition of cell cycle inhibitor p15, was one of the targets in Bapx1 null embryos showing loss of expression [41]. Pdx1, another spleen development marker is also known to regulate Nkx2.5. Bapx1 regulates Nkx2.5 as well, yet without affecting Pdx1 itself, indicating the possible presence of a parallel pathway that converges at Nkx2.5 to regulate proliferation of splenic primordial cells [41]. It is further possible that Pdx1 is upstream of Bapx1 and mediates Nkx2.5 via Bapx1. Besides Nkx2.5, Slc40a1 and Bmi1 were activated by wildtype Bapx1 and the loss-of function mouse mutants of these genes exhibit a hypoplastic spleen phenotype. Functional ablation of Slc40a1[42] results in increased splenocyte apoptosis while Bmi1 is a cell cycle regulator, which shows impaired splenocyte proliferation upon knock-out [43], suggesting that these genes also mediate the cell viability function of Bapx1 in the spleen.

IPA analysis predicted the proliferation of cells to be decreased (z-score: -3.922; p: 7.23E-11) while apoptotic genes were enriched (z-score: 1.256; p: 9.23E-06) (Additional file 2: Table S2). In situ hybridization showed the presence of remnant splenic primordial mesenchyme at E12.5 in Bapx1-null embryos, which is progressively lost with development, resulting in asplenia at birth (Figure 3K-P). Thus, the disruption of cell survival via the deregulation of cell cycle and apoptotic genes in the splenic primordial mesenchyme is likely causative for the hypoplasia of the spleen at E12.5 and its complete absence at later developmental stages in the Bapx1 null mutants. IPA further revealed the enrichment of genes of the FGF signaling pathway, a pathway essential for cell survival, cell growth, differentiation, morphogenesis and angiogenesis. Other enriched genes included members of the ERK/MAPK signaling, epithelial adherens junction signaling, VEGF signaling and epithelial-mesenchymal transition pathways (Additional file 2: Table S2).

Validation of novel Bapx1target genes during spleen development

We confirmed two genes with no previously attributed role in spleen development by section RNA in-situ hybridization (SISH). Kinase (PRKA) anchor protein (gravin) 12 (Akap12) previously described in the pancreas [44] was shown here to be expressed and controlled by Bapx1 in the spleen (Figure 3M-N). Similarly Glycogenin (Gyg) previously described in the heart and lungs [45], was found to be expressed and controlled by Bapx1 in the spleen as seen by a significant decrease in expression in the E12.5 day Bapx1 null spleen (Figure 3O-P). The entire repertoire of genes regulated by Bapx1 in the spleen would thus serve as a suitable database of splenic development genes for future investigation.

Bapx1controls enzyme inhibitors and viability in the gut cells

Bapx1 is the mammalian ortholog of the Drosophila bagpipe, a gene that is essential for directing the development of the fly visceral mesoderm into gut musculature [46, 47]. In contrast, Bapx1 does not seem as critical during morphogenesis of the mammalian gut, given most of the previous knock-out studies failed to report a significant gut phenotype, except for a shortening of the antrum-pylorus segment and the loss of the pyloric constriction [8–10]. Bapx1 expressing cells are reportedly present in the gut mesenchyme with a rostral boundary in the hind stomach, near the junction of the gastric corpus and antrum [11]. By analyzing genes under the control of Bapx1 we defined its potential role in the gastrointestinal tract. Our GO annotation of genes clearly indicated that Bapx1 primarily controls genes involved in endopeptidase and other enzyme inhibitor activities (Figure 2E). These include KngI, Ambp, Serpina1B, SerpinaA6, Spink3 and Itih3 among others. In analogy to the spleen our IPA analysis revealed that Bapx1 regulates a significant proportion of proliferation (z-score: -3.054; p: 2.22E-05) and apoptotic genes (p: 3.01E-03) in the gut, thus explaining the shortening of the antrum-pylorus segment in the null embryo (Additional file 2: Table S2). Our genome wide transcriptional profiling revealed genes previously unknown to be controlled by Bapx1 in the gut and the control of a physiological activity (enzyme inhibitory action) that has not been attributed to it previously. Furthermore, signaling pathways such as ERK/MAPK signaling, VEGF signaling and the epithelial-mesenchymal transition pathway were enriched in the gut similar to our observations in the spleen (Additional file 2: Table S2).

Bapx1-Sox9act synergistically in the developing vertebral column

Previous in vitro analyses showed Bapx1 and Sox9 connect in an auto-regulatory loop and Sox9 repressing Runx2 via Bapx1, thereby preventing the inappropriate hypertrophic differentiation of chondrocytes [16]. Hence we investigated the regulatory targets of Sox9 in the developing E12.5 vertebral column in vivo. Similar to our work regarding Bapx1, we enriched for Sox9-specific cells from the vertebral column of Sox9+/tm1.1Tlu and Sox9-/-(EGFP) E12.5 embryos using FACS (Figure 2G). Our microarray data on the Sox9-specific cells revealed that the Sox9 protein, unlike Bapx1, is neither an overt activator nor a repressor of transcription in the vertebrate column as 42.2% (1514/3593) genes showed decreased expression (FC1.5; p ≤ 0.05) compared to 57.7% (2073/3593) showing increased expression (FC1.5; p ≤ 0.05) in the Sox9 null embryo relative to the Sox9 wildtype embryo (Additional file 1: Table S1). IPA of these Sox9 controlled genes revealed that most had known roles in chondrocyte differentiation (p: 3.43E-06), proliferation (p: 3.82E-28) and apoptosis (p: 3.92E-14) (Figure 2H and Additional file 2: Table S2) consistent with the known role of Sox9 as a master regulator of osteo-chondrogenesis. Though many of the downstream genes are known to be involved in chondrogenesis through in vitro studies, our data demonstrates that Sox9 regulates these genes in vivo in the developing mouse embryo at E12.5, a time when the developing skeleton is transitioning from mesenchymal condensations to a chondrogenic framework. Similar to the pathways regulated by Bapx1 in the vertebral column, IPA analysis for Sox9 regulated genes showed enrichment of the Wnt/B-Catenin and VEGF signaling pathways, and the inhibition of matrix metalloproteases (Additional file 2: Table S2).

With Bapx1 and Sox9 functioning in the same tissue and regulating similar pathways, the question arose whether these two transcription factors act on the same set of genes in a co-regulatory manner. Comparing Sox9 controlled gene expression with Bapx1 target genes allowed us to filter for chondrogenic genes, as Bapx1 is naturally absent in the neural tube. We focused on three sets of genes: (1) common genes that are activated by Sox9 but repressed by Bapx1, (2) genes repressed by both and (3) genes activated by both. For analysis of gene co-regulated by Sox9 and Bapx1 we have used a cut off of FC1.5 to make a larger gene set to avoid overlooking any gene which might be strongly regulated by one and weakly by other. We found 59 genes to be repressed by Bapx1 and activated by Sox9 all primarily involved in neurogenesis and neuronal differentiation like Gbx2, Dbx, Nrn1 and Ctnna2 (Figure 4A and Additional file 3: Table S3).

Figure 4
figure 4

Co-regulation by Bapx1 and Sox9 . (A) Venn diagram showing the 59 common genes activated by Sox9 and repressed by Bapx1 in the vertebral column of E12.5 wildtype embryos and the GO annotation of those common genes, which are primarily neuronal markers. (B) Venn diagram showing the 72 common genes repressed by both Sox9 and Bapx1 in the vertebral column of E12.5 wildtype embryos and the GO annotation of those common genes.

Full size image

Analyzing genes repressed by both transcription factors in the vertebral column, we found 72 genes, which included the osteogenic factor Col1a1, cartilage degrading metalloproteases (Mmp11, Adamts2, Adamts4) and most highly enriched, genes controlling blood vessel development (Vegfc, Emcn, Nrp1), an essential process during chondrocyte maturation [48, 49] (Figure 4B and Additional file 3: Table S3 ). Though the role of Sox9 as a repressor of cartilage vascularization has been shown recently [50], our data suggests that Bapx1 plays an equally critical role in this process. Thus by repressing these genes at an early time point in development it prevents premature ossification of the developing vertebral column. Moreover, Bapx1 and Sox9 function as activators on only six common target genes, including Sox5, Wwp2, Chad and Ctgf, which are all known to be involved in cartilage development [14, 50–52] (Additional file 3: Table S3).

Surprisingly, we did not detect Bapx1 as one of the downstream targets of Sox9 or vice versa at E12.5 in the vertebral column (Additional file 1: Table S1 and Additional file 3: Table S3). However, we found Sox9 to be downstream of Bapx1 in the spleen and gut tissue. Besides, Bapx1 was clearly upregulated in the Bapx1-null embryos in the vertebral column (Figure 3A, B), indicating potential auto-regulation.

Genome wide binding of Bapx1 and Sox9 proteins to DNA target sites in the vertebral column

The pool of differentially expressed genes from the expression profiling consisted of target genes that were regulated directly by Bapx1 and Sox9, and indirect target genes whose expression was altered owing to the loss of these two transcription factors affecting their upstream regulators (secondary effectors). To find genes directly controlled by Bapx1 and Sox9 proteins during vertebral chondrogenesis, we performed chromatin immunoprecipitation sequencing (ChIP-Seq) on vertebral columns dissected from E12.5 embryos. For Sox9, we performed ChIP-Seq with an anti-Sox9 antibody. However for Bapx1, owing to the absence of a suitable ChIP grade quality Bapx1 antibody, a mouse line with a S-Peptide and Calmodulin binding peptide (CBP) epitope tagged to Bapx1 at its N-terminus was generated (Figure 5A and Additional file 4: Figure S1) similar to other work we have described [53–56]. S-peptide is a 15 amino acid tag and CBP is a 26 amino acid tag, routinely used for protein purification purposes [57–59] and owing to their small size suitable for endogenous tagging of transcription factors in the mouse [53, 54]. Their absence from the wildtype mouse proteome allows for highly specific ChIP. The homozygous Bapx1 protein tagged mice we generated were phenotypically normal compared to their wildtype littermates, assuring normal biological activity of the tagged protein and thus allowed us to use the S-peptide antibody to pull down tagged Bapx1.

Figure 5
figure 5

In vivo ChIP-Seq for Bapx1 and Sox9 in the vertebral column of E12.5 embryos. (A) Targeting of the endogenous Bapx1 locus with a CBP- S peptide tag to generate N-terminal tagged BAPX1 protein. (B) The Bapx1 binding site motif enriched in the top 1000 binding sites from ChIP-Seq. (C) Whole genome binding site distribution for Bapx1 in the vertebral column of E12.5 wildtype embryos. The number on top of each peak represents the total number of peaks found at that distance from the TSS. Note that most peaks are found distally from the TSS. (D) GO annotation of the nearest genes to the peak shows enrichment of genes expressed in chondrocytes. (E) Similar binding site distribution for Sox9 and (F) GO annotation for the nearest genes showing enrichment of genes expressed in chondrocytes, x-axis is -log10 (p-value).

Full size image

Using the model-based analysis for ChIP-Seq (MACS) algorithm [60] we detected 2815 Bapx1 peaks with 19% (540) of the binding sites located on the transcription start sites (TSS) of genes (Figure 5C). Interestingly a large proportion of the binding sites were distal from the TSS, with 37% (1052) located at >25 kb distally in the intergenic region and 26% (722) located >25 kb in the intragenic regions (Figure 5C). Likewise for Sox9, we identified 3722 peaks. Similar to Bapx1, most of the binding sites were located distal to the TSS, with only ~ 10% (359) of the Sox9 binding sites at the TSS of various genes and 66% (2470) of the binding sites more than 25 kb distal to the TSS (Figure 5E). This shows that in midgestation embryos, critical transcription factors like Bapx1 and Sox9 tend to bind primarily to distal regulatory elements in order to direct downstream gene expression in target cells and tissues rather than activating transcription at the TSS. To ascertain which are potential Bapx1 direct targets in the vertebral column we looked at genes that are differentially expressed (fold change = 1.5) in the microarray dataset for the vertebral column and discovered that of all the genes that have a binding site within 100 kb of their TSS 84% are up-regulated in the Bapx1-null. This adds confidence to the data that the Bapx1 protein primarily acts as repressor in the vertebral column (Additional file 5: Table S4).

We identified the previously reported Bapx1 binding motif T(G/A)AGTG [28], enriched in about 40% of the Bapx1 binding sites in the genome (Figure 5B) specifically near genes controlling cartilage and bone morphogenesis (Figure 5D). Notably, this motif was reported to be involved in chondrogenesis in vitro, in cell culture and chicken embryo explant assays [28, 29, 61]. Our data is the first to demonstrate the in vivo binding of Bapx1 to genomic DNA by this motif in mouse embryo chondrogenic cells. We further validated the Bapx1 binding motif by performing electrophoretic mobility shift assays (EMSA) with four randomly selected genomic regions containing this motif (Additional file 6: Table S5). This allowed us to test the binding specificity of the motif in its native context. All four of the randomly chosen genomic regions containing the motif were able to bind Bapx1 in vitro (Additional file 7: Figure S2A). To further confirm the specificity of the motif, we mutated all the nucleotides of the core sequence of one of the regions by transversion and noted a complete abrogation of binding (Additional file 7: Figure S2B). We ascertained the specificity of this binding of the probe to Bapx1 using a supershift assay. We generated NIH3T3 cells overexpressing V5-tagged Bapx1 and used an anti-V5 antibody on nuclear extract for specificity and an anti-Flag antibody as a non-specific control. Our results showed a supershift with the anti-V5 antibody, whereas the addition of anti-Flag antibody had no effect on the shift (Additional file 7: Figure S2B). We further analyzed if some of the Bapx1 bound regions could act as enhancers in a transactivation assay using a luciferase reporter. We cloned seven randomly selected genomic regions (including the four tested in EMSA) (Additional file 6: Table S5) in a minimal promoter driven luciferase vector (pGL4.23) and co-transfected it with the V5-tagged Bapx1 vector into NIH3T3 cell lines. Five out of the seven regions showed activity over the basal level thus acting as an enhancer. To further test if the core-binding motif for Bapx1 was necessary and sufficient for this activity we used mutated oligonucleotides for the core sequence in two of the regions. The mutated oligonucleotides failed to drive expression over the basal level, thus proving that the core Bapx1 motif is the essential component of these enhancers (Additional file 7: Figure S2C). Similarly, most of the genes nearest to the Sox9 binding regions were involved in chondrogenesis, validating the specificity of the data (Figure 5F). We also identified the DNA binding motif for Sox9 (A/T)(A/T)CAA(A/T)G that had been previously identified mostly by in vitro studies [62–64]. Our data shows that Sox9 binds near genes involved in chondrogenesis in the developing mouse embryo by this same motif.

Direct co-regulation by Bapx1 and Sox9 in the vertebral column

Next we established how many of the genes in these groups have binding sites for both transcription factors within 200 kb of their TSS. Of the group of genes activated by Sox9 and repressed by Bapx1, 86% (51/59) had both the Bapx1 and Sox9 binding sites within 200 kb of their TSS (Additional file 8: Table S6). These genes represent potential primary targets of Bapx1 and Sox9, which are under opposite regulatory pressure in the vertebral column to maintain a distinct neuronal and osteo-chondrogenic identity in the developing embryo, they include genes like Gata3, Ctnna2, Nrn1 and Dcx, whose knockouts have a reported neuronal phenotype [65–69] but have never been associated with a regulation by either Sox9 or Bapx1. Our dataset thus connects Bapx1 and Sox9 downstream genes with a dual regulatory mechanism. Similarly, 78% (56/72) of the genes repressed by both Sox9 and Bapx1 had binding sites for both Sox9 and Bapx1 within 200 kb of their TSS (Additional file 3: Table S3 and Additional file 8: Table S6). These genes are also potential primary targets of both Sox9 and Bapx1, but repressed by both. This group includes the osteogenic, metalloproteinase and angiogenesis genes Col1a1, Mmp11, Adamts4, Ctsk, Emcn and Nrp1. We looked for binding sites that were overlapping for both the transcription factors but failed to find any at this stage of development. This was not entirely unexpected since both are large transcription factors with distinct motifs thus it is slightly unlikely they could occupy a single enhancer without causing steric hindrance at the same time in development.Our results show that combining genome wide data for binding and transcriptional profiling of two critical transcription factors (Bapx1 and Sox9) in a single organ can lead to a detailed understanding of the diverse levels of regulation involved in its organogenesis. Some neuronal genes were found under a competing control mechanism, receiving both positive and negative inputs: primarily neuronal genes and genes controlling processes essential for chondrocyte maturation were repressed in chondrocytes and those promoting chondrogenesis were activated by both Bapx1 and Sox9 (Figure 6A). Finally by assembling information from all the genes regulated by Bapx1 and Sox9 either directly or indirectly a large-scale integrative gene regulatory network could be delineated (Figure 6B).

Figure 6
figure 6

A gene regulatory network in vertebral column driven by Bapx1 and Sox9 . (A) Three layer mechanism of genomic control in the vertebral column by Bapx1 and Sox9. Bapx1 suppresses neuronal genes whereas Sox9 activates them; both activate chondrogenic genes whereas both suppress genes involved in osteogenesis, cartilage degradation and blood vessel development. (B) A large gene regulatory network in the vertebral column of genes that are either activated (green lines) or repressed (red lines) by Sox9 and Bapx1. Each blue dot represents a gene and the common regulated genes are in the center on the network connected to both Sox9 and Bapx1.

Full size image

Discussion

It is becoming increasingly clear that an in-depth understanding of organogenesis in vertebrates will require comprehensive systems biology approaches to tease out all the players in various pathways and integrate them into a mechanistic network. Gene expression profiling coupled with genome wide binding studies for particular organs have over the years shed some light on these complex processes [70, 71]. However, transcriptome analysis of an entire tissue or organ only gives a broad and often confusing picture of the precise transcriptional changes taking place during development owing to a large heterogeneity of different cell types that comprise individual tissues that make up a functional organ. By conducting expression profiling only on cells that express a particular set of transcription factors we generated a highly focused dataset elucidating pleiotropic roles played by the transcription factors in organogenesis.

We have analyzed five different organs expressing the Bapx1 gene to explain the disparate roles it plays in each of them. Although the pleiotropic role of Bapx1 was known and many phenotypes defined, our approach is the first large-scale effort looking at all the major tissues expressing Bapx1. This allowed us to make connections between many new genes previously unknown to be controlled by Bapx1. The large scale dataset generated enabled us to place these genes within pathways they control, essentially explaining the lack of a limb phenotype, the absence of an overt morphological phenotype in the gut and the impact on cell cycle regulation in the spleen leading to asplenia in the Bapx1 loss-of-function mice. Importantly, Bapx1 largely regulates different genes and pathways in the different tissues to execute its pleiotropic roles. Interestingly, spleen and gut tissue, which are closely associated during gastrointestinal development [72] share more similar pathways with each other than with vertebral column tissue. Indeed, among the five tissues, spleen and gut share the most number of common genes. Intriguingly, while Bapx1 represses VEGF signaling in the vertebral column and gut, it is largely activated in the spleen. The spleen is a repository of red blood cells and its development involves vascularization that allows the infiltration of lymphoid progenitors as early as E12.5-E13.5 [73]. Hence, positive regulation of angiogenesis in the spleen is not too surprising. At the same time, consistent with its role in preventing premature endochondral ossification, Bapx1 inhibits VEGF signaling in the vertebral column.

The formation of gene regulatory networks in vertebrate development continues to be a long arching goal of developmental biology. Yet, to establish such detailed maps, many times input from more than one transcription factor is required. Hence, we co-analyzed in similar fashion cells specific for Sox9, another key transcription factor during chondrogenesis and vertebral column development. Combining the genes regulated by Bapx1 and Sox9 in the vertebral column we have generated a network of genes activated and/or repressed by these transcription factors during the fine tuned process of organogenesis. We have identified 137 genes co-regulated by these two transcription factors providing insight into some of the mechanisms employed during chondrocyte maturation; there is a group of neuronal genes, which are acted upon antagonistically by Bapx1 and Sox9, chondrogenic differentiation genes which are activated by both Sox9 and Bapx1, and a group of genes which are repressed by both Sox9 and Bapx1 that are involved in processes essential for hypertrophic chondrocyte differentiation. Furthermore, Bapx1 and Sox9 contribute to several similar pathways during vertebral column development, especially the repression of osteogenic differentiation and cartilage degradation. We also studied the genome wide binding of Bapx1 and Sox9 in vertebral column tissue to identify potential primary target genes. Our ChIP-Seq data revealed that although Bapx1 and Sox9 bound to the TSS of some genes, the majority of the genes regulated by these two transcription factors had their binding sites distal to the TSS, highlighting the fact that during the midgestation development in mouse embryos many critical transcription factors bind to distal cis-regulatory elements.

Conclusions

In all we have managed to create a comprehensive list of genes controlled by a single transcription factor namely Bapx1 from five different tissues, explaining its diverse roles and the various genetic pathways Bapx1 contributes to. This is the first study analyzing all major Bapx1 dependent organs in molecular detail and providing explanations for the phenotypes observed in the loss-of-function animals. Many of the here identified target genes were previously unknown to be either directly or indirectly regulated by Bapx1 owing to a lack of feasible strategies for determining downstream effectors on a genome wide scale. This genomic resource and the underlying mechanisms uncovered will form the basis for a deeper understanding of the complex process of organogenesis in the developing mammalian embryo.

Methods

Ethical statement

All animal procedures were performed according to the Singapore A*STAR Biopolis Biological Resource Center (BRC) Institutional Animal Care and Use Committee (IACUC) guidelines which are set by the National Advisory Committee for Laboratory Animal Research (NACLAR). The IACUC protocols employed were reviewed and approved by the aforementioned committee before any animal procedures were undertaken for this study described here (IACUC Protocol No: 110689 and 110648). The mouse strains used in this study were housed, maintained and provided by the A*STAR Biopolis Biological Resource Center. The lines described here will be made available to the research community upon acceptance of the manuscript.

Gene targeting and mouse generation

The over 160 kb murine BAC clone RP24-148P5 containing genomic C57BL/6 J DNA flanking the Bapx1 gene was obtained from the BACPAC Resources Centre at Children’s Hospital Oakland Research Institute (CHORI) and modified using a Quick and Easy BAC modification kit (Gene Bridges) according to the manufacturer’s instruction. The Bapx1tm2.Tlu allele was generated as described in [17] to generate wildtype mice expressing EGFP in the Bapx1 domains. The Bapx1tm4.Tlu allele was generated by inserting the EGFP-FRT-PGK-gb2-Neo-FRT cassette immediately after the translational start codon of Bapx1, deleting 90 bp of the endogenous sequence to create the Bapx1 null mice expressing EGFP. The gene targeting events were confirmed by Southern blotting and for each allele, at least two independent ES cell clones with normal karyotype were subsequently microinjected into 2-8-cell stage embryos isolated from C57BL/6 mice to generate chimeric mice as previously described [74]. The chimeric mice were bred to wildtype mice to generate stable lines. The FRT-flanking neomycin cassette in the targeted Bapx1 allele was removed by breeding to the FLPe–deleter mice 129S4/SvJaeSor-Gt(ROSA)26Sor tm1(FLP1)Dym /J (Stock # 3946) from Jackson Laboratories [75, 76]. Routine PCR genotyping of Bapx1 lines was performed as essentially described in [17].

Sorting Bapx1-EGFP cells from mouse embryos

The mouse embryos were harvested in ice-cold Leibovitz L-15 medium (Gibco) at E12.5. Embryos expressing EGFP in the Bapx1-expression domains were identified under a fluorescent dissection microscope (Leica). The embryos were dissected to separate out the vertebral column, spleen, gut, hindlimb and forelimb, which were then separately dissociated into single cells with a solution comprising of 100U/ml Collagenase I & II, 50U/ml DNAse and 0.05% Trypsin (Invitrogen). The cells were filtered serially through a 100uM and 40uM cell strainer, and centrifuged at 2000 rpm for 5 min. The cell pellet was resuspended in 5% FBS, 4 mM EDTA in Leibovitz L-15 medium for cell sorting using FACSAria (BD Biosciences).

Gene expression microarrays and analyses

Total RNA was extracted form 4 biological replicates for both wild type and null embryos for each of the tissues. Each biological replicate was isolated from each population of sorted cells using TRIzol (Invitrogen) followed by the RNeasy Micro Kit (Qiagen). Sample integrity was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies). 25 ng total RNA from each sample was labelled using TargetAmp™-Nano Labelling Kit for Illumina Expression BeadChip (Epicentre Biotechnologies) and hybridized on MouseWG-6 v2.0 Expression BeadChip microarrays (Illumina) according to Illumina guidelines. GenomeStudio software (Illumina) was used to prepare background-subtracted data. The background-subtracted data was then imported into Partek Genomics Suite (Agilent) and the data normalized. The probe intensities from various biological replicates were normalized using scale normalization using median absolute deviation as a spread measure. This was further filtered on percentile (lower 20 and upper 100) by expression of probe sets without averaging over replicates. These consistent probe sets, which passed this filter, were further used to calculate gene expression changes between different conditions. ANOVA with nominal alpha value set to 0.05 was then used to determine the probe sets significantly different between the different genotypes compared. Benjamini and Hochberg Multiple testing correction was applied to reduce the false positive rate, and probe sets with expression fold change >2.0 or >1.5 between the different genotypes were selected for further validation and analyses. All data has been deposited in Gene Expression Omnibus (GEO) under accession number GSE35877. Ingenuity Pathway Analysis (Ingenuity® Systems, http://www.ingenuity.com) and DAVID (http://david.abcc.ncifcrf.gov/) gene ontology analyses were performed using the respective web-based tools.

ChIP-assay and peak calling

For Bapx1 ChIP vertebral columns from ~100 E12.5 (both S-peptide tagged and wildtype each) embryos were dissected and 2 mg of chromatin was used for ChIP as previously described [77] with anti S-Peptide antibody (Bethyl laboratories, A190-134A). For Sox9 ChIP anti-Sox9 antibody (R&D Systems, AF3075) was used for immunoprecipitation. 10-15 ng of purified ChIP DNA from each sample was used to synthesize the sequencing library as instructed by the ChIP-Seq DNA sample Prep Kit (Illumina). The libraries were then subjected to Solexa sequencing according to Illumina’s instruction. Sequence reads produced by Illumina Genome Analyzer II/IIx that passed the signal purity filtering were mapped to the mouse genome mm9, using the Illumina Genome Analyzer Pipeline. All uniquely mapped reads that are with two or fewer mismatches were retained. Genomic binding sites in the ChIP-Seq datasets were identified using the peak calling algorithm MACS (version 1.4.0 beta) with default settings (band width = 300, model fold = 10, 30, p-value cutoff = 1.00e-05, range for calculating regional lambda = 1000 and 10000 bps) [62]. The corresponding control libraries were used for all the peak callings. For Annotation of the ChIP-Seq peaks, first the TSS of the nearest Refseq gene nearest to peaks were found, then the distance from nearest TSS was found and reported. If the peaks resided in intragenic region of nearest gene it was called “intragenic”, otherwise “intergenic”. If the peaks resided within 5000 bp of TSS they were called as “promoter” peaks. All data has been deposited in Gene Expression Omnibus (GEO) under accession number GSE35877 and also provided in BED format (Additional file 9: Table S7).

Motif analysis

Peaks called by MACS were ranked according to the total tags count as defined in the MACS output file. The top 200 peaks were used for motif analysis and the repeat masked genome sequence +/- 50 bp from the summit of these 200 peaks was downloaded from the UCSC genome browser (http://genome.ucsc.edu/). After masking repeats to N, we performed de novo motif finding using MEME ver. 4.3.0 with the sequences. MAST was used to scan for the occurrences of the primary de novo motif obtained using all the sequences +/-50 bp of the ChIP-Seq peak summit. The cut off for motif match in MAST used was the default p = 1.0e-4.

Gene enrichment test for regions bound by Bapx1 and Sox9

We submitted the analyzed ChIP-Seq data as bed format to Genomic Regions Enrichment of Annotations Tool (GREAT) [78] version 1.8 with the following parameters- Species assembly-mm9 and Association rule-Single nearest gene: 100,000 bp maximum extension, curated regulatory domains included.

RNA in-situhybridization

In situ hybridizations were performed using DIG labeled RNA probes as previously described [17]. For analysis of the vertebral column the E12.5 embryos were sectioned through the sagittal plane and for analysis of the spleen through the dorsofrontal plane.

Luciferase reporter assay and EMSA

Luciferase assay was performed to determine the transcriptional activation property of some of the binding sites bound by Bapx1. Mouse Bapx1 full-length cDNA (40131175, Openbiosystems) was cloned upstream of the V5-tag in the pcDNA6/V5-HisABC vector (Invitrogen). 400 ng of firefly luciferase vector (pGL4.23, Promega Corporation) containing the binding sites, 500 ng each of Bapx1 expression vector, and 2 ng of Renilla luciferase vector (transfection control) was transiently transfected into mouse fibroblast cell line (NIH/3 T3) (5 – 6 × 104 cells/well) using 6ul of FuGENE HD transfection reagent (Roche Diagnostic, USA), in 100 μl of OPTI-MEM I medium (Invitrogen, USA). The cells were grown for 48 hours and the luminescence was measured using Dual Luciferase Reporter Assay System on a Glomax multidetection system Luminometer as per manufacturer’s instructions (Promega Corporation).

Electrophoretic mobility shift assays were performed as previously described [79]. Briefly V5 tagged Bapx1 vector was transfected in NIH/3 T3 cells using FuGENE HD transfection reagent. Cells were harvested 48 hours post transfection and nuclear proteins were extracted using the NE-PER nuclear and cytoplasmic extraction kit (Thermo Scientific, USA). EMSAs were carried out using 10 nM DNA probes modified with 5’ Cy5 labels (Sigma Proligo) incubated with 10ug nuclear extracts. For supershift assays the probe and the nuclear extract were incubated with 1ug of V5 antibody (ab15828, Abcam) or Flag (F7425, Sigma) was used as a non-specific antibody.

Gene regulatory network

The gene regulatory network was constructed using Cytoscape [80], by considering all genes that are showing >1.5 fold change in the microarray (activated as well as repressed).