Epigenetically coordinated GATA2 binding is necessary for endothelium-specific endomucin expression

Abstract

GATA2 is well recognized as a key transcription factor and regulator of cell-type specificity and differentiation. Here, we carried out comparative chromatin immunoprecipitation with comprehensive sequencing (ChIP-seq) to determine genome-wide occupancy of GATA2 in endothelial cells and erythroids, and compared the occupancy to the respective gene expression profile in each cell type. Although GATA2 was commonly expressed in both cell types, different GATA2 bindings and distinct cell-specific gene expressions were observed. By using the ChIP-seq with epigenetic histone modifications and chromatin conformation capture assays; we elucidated the mechanistic regulation of endothelial-specific GATA2-mediated endomucin gene expression, that was regulated by the endothelial-specific chromatin loop with a GATA2-associated distal enhancer and core promoter. Knockdown of endomucin markedly attenuated endothelial cell growth, migration and tube formation. Moreover, abrogation of GATA2 in endothelium demonstrated not only a reduction of endothelial-specific markers, but also induction of mesenchymal transition promoting gene expression. Our findings provide new insights into the correlation of endothelial-expressed GATA2 binding, epigenetic modification, and the determination of endothelial cell specificity.

Introduction

Vascular homoeostasis is a critical component of the embryonic and adult stages. Vascular dysfunction is strongly correlated with various diseases; cancer, atherosclerosis, diabetic retinopathy, and sepsis (Minami and Aird, 2005). The endothelium is not only the inner cell layer of vessels, but also the central player for communication with the outer microenvironment. In many cases, endothelial cell activation responding to extracellular signals leads to the spatial and temporal regulation of gene expression. Thus, investigating the molecular mechanisms behind endothelial-cell-specific gene regulation might uncover the key regulators of vascular homoeostasis and onset of vascular diseases.

Currently, six GATA family members have been identified, GATA1 to 6, as belonging to a class of evolutionally conserved C2H2 zinc-finger transcription factors (Patient and McGhee, 2002). In endothelial cells, GATA2, 3, and 6 are selectively expressed (Minami et al, 2004). However, their expression levels are not equal. GATA3 is predominantly expressed and functions in endothelial cells derived from large vessels, whereas, GATA2 is uniformly expressed in all endothelial cells. Microvascular endothelial cells preferentially express only GATA2 (Song et al, 2009).

GATA2 was initially identified as an activator of endothelin 1 expression in endothelial cells (Wilson et al, 1990). Targeted disruption of the gata2 gene in mice resulted in embryonic lethality between embryonic days 9.5–11.5, due to defects in primitive haematopoiesis and haemogenesis (Tsai et al, 1994). In addition to the developmental stage, GATA2 is also recognized as an important regulator of endothelial selective gene expression, including platelet/endothelial cell adhesion molecule (PECAM)1, endothelial-nitric oxide synthase, von Willebrand factor (vWF), Down syndrome critical region-1, vascular endothelial cell adhesion molecule-1, KDR, and GATA2 itself (Jahroudi and Lynch, 1994; Zhang et al, 1995; Gumina et al, 1997; Kappel et al, 2000; Minami and Aird, 2001; Minami et al, 2001, 2009; Wozniak et al, 2007). Although these findings have suggested that GATA2 has a crucial role in the gene expression profile in vascular endothelial cells, the molecular mechanism by which GATA2 controls many endothelial specifically expressed genes remains largely unknown.

Recently, a hierarchical approach to investigating transcriptional regulation has shown to be the most informative, including consideration of transcription factors, histone modifications, and chromatin conformational changes proceeding gene expression. Modifications associated with trimethylated histone 3 lysine 4 (H3K4me3) in the promoter and monomethylated histone 3 lysine 4 (H3K4me1) in the enhancer, are tightly regulated with lineage specificity (Barski et al, 2007; Kim et al, 2010). More recently, it has been reported that cell-type specific gene activation and silencing can accompany dynamic chromatin conformational changes, resulting in different accessibility of the transcriptional machinery (Fullwood et al, 2009; Visel et al, 2009).

Endothelial cells exhibit a wide range of phenotypic variability depending on the local environment throughout the vascular tree (Aird, 2007). Moreover, recent studies have demonstrated that endothelial to mesenchymal transition (EndMT) can occur in various pathological states in cancer and cardiac fibrosis (Zeisberg et al, 2007a, 2007b). During the EndMT, endothelial cells are characterized by their loss of endothelial markers, such as a PECAM1 and gain of the mesenchymal markers, such as an α-smooth muscle (SM)-actin, resulting in the loss of cell–cell junctions and acquisition of invasive and migratory properties (Potenta et al, 2008).

In this report, we performed chromatin immunoprecipitation with deep sequencing (ChIP-seq) to determine genome-wide occupancy of GATA2 in endothelial cells and compared it with the respective gene expression profiles to understand GATA2 function on a genome-wide scale. Comparison of epigenetic markers and chromatin conformation between endothelial cells and non-endothelial cells, we identified a new GATA2 responsive endothelial-specific marker gene, endomucin. Moreover, abrogation of the GATA2 function in endothelium demonstrated not only reduction of the endothelial-specific marker, but also induction of the EndMT promoting gene expression. These results provide new insights into the cooperation between endothelial-expressed GATA2 binding and epigenetic modification, resulting in the determination of endothelial-cell-specific gene expression.

Results

GATA2 is uniformly expressed in the vascular endothelium

GATA factor has been considered as a key transcription factor regulated with cell-type specificity and differentiation. In endothelial cells, GATA2 has mainly been reported to be involved in the regulation of gene expression. We have previously reported both GATA3 mRNA and protein are predominantly expressed in large vessel endothelium, whereas, GATA2 mRNA is commonly expressed in endothelial cells derived from both microvascular- and larger-endothelium (Song et al, 2009). To identify whether GATA2 protein is indeed expressed in the vascular endothelium in vivo, we generated a monoclonal antibody against the antigen (residues 192–245) of human GATA2 (Figure 1A). The antibody specifically reacts with GATA2, but not GATA3, illustrating its selectivity (Figure 1B). Immunohistochemical staining revealed that GATA2 was specifically expressed in nuclei of dermal microvascular endothelium and artery endothelial cells of the human dermis (Figure 1C). Moreover, GATA2-derived signals were detected in the small vessels of the lung, heart, kidney, and tumour-derived microvascular endothelial cells (data not shown). In cultured endothelial cells, GATA2-positive staining was uniformly found in nuclei (Figure 1D).

Figure 1.

GATA2 expression in endothelium. (A) The schematic representation of human GATA2. Zn, zinc-finger domain. The bar indicates the region of the antigen for the monoclonal antibody. (B) Human GATA2 (pcDNA3-GATA2), GATA3 (pcDNA3-GATA3), or mock control (pcDNA3) were transfected in COS7 cells. Whole-cell lysates were prepared and blotted with an anti-GATA2, and anti-GATA3 antibody. Anti-β-actin antibody was used as a loading control. (C) Subcutaneous tissues were sectioned and subjected to immunohistochemical staining. GATA2 was shown with light brown in the nucleus (left). CD34 was with dark brown (right). Indicative of the microvascular and artery endothelium, respectively. Bar: 10 μm. The results are representative of five independent experiments. (D) Immunofluorescent staining of GATA2 (left) and VE-Cadherin (middle) in HMVEC. Merged images with DAPI are shown in right. Bar: 50 μm.

Genome-wide and cell-type specific GATA2 binding was detected in microvascular endothelial cells

Next, we wished to determine how endothelial cell expressed GATA2 regulates the cell-type specific gene expression genome wide. Thus, at first, we performed duplicate ChIP with GATA2 antibody, and then the precipitated genome was sequenced comprehensively (duplicate ChIP-seqs). To test the accuracy of these ChIP assays with GATA2 antibody, we choose the endothelial cell specifically expressed vWF promoter region as a positive control; and non-expressed MyoD1 promoter region as a negative control (Supplementary Figure SI). The GATA2-mediated immunoprecipitated DNA and non-immunoprecipitated whole genome control (input DNA) were used to prepare libraries for deep sequencing and analysed using massively parallel sequencing. The genome-wide GATA2-binding regions were calculated by two independent methods, QuEST (Valouev et al, 2008) and Sole-search (Blahnik et al, 2010) (see Supplementary data in detail). We found a total of 5737 and 5805 regions, which were identified as GATA2-enriched areas from the first and second ChIP-seqs, respectively (Supplementary Table SI). To reconstruct the GATA2-associated binding regions, we clustered the regions into four sections based on the distance from the transcription start site (TSS) in the respective genes. As shown in Figure 2A, up to 57% GATA2-binding regions were positioned around 10 kbp from each transcript. Among them, 9% were in the proximal promoter of the gene, 42% were located in the intron. The remaining 43% were located at the intergene. Collectively, these data suggest that the defined GATA2-binding regions derived from the comprehensive ChIP-seq data were not selectively located within the proximal promoter of each gene, rather, scattered widely in the whole genome. These findings were consistent with other transcription factors, oestrogen receptor, and FOXA1 as performed in tumour cells (Lupien et al, 2008).

Figure 2.

ChIP-seq results in HMVEC and K562. (A) Distribution of GATA2-binding regions in HMVEC. GATA2-binding regions (QuEST score ⩾30) were classified according to the distance from the transcription start sites of known genes. (B) Determination of the sequence recognized by GATA2 in HMVEC and K562. MEME method (Bailey and Elkan, 1994) was used for identification of enriched sequences and displayed with correlation of the size of character and the rate of enrichment. E-value and P-value mean the probability de novo enriched sequences obtained from ChIP-seq are matched to the displayed Weblogo, and known consensus motifs by chance, respectively. (C) Venn diagrams depicting the overlap of genes with GATA2 binding within the up 5 kbp, TSS 1 kbp, 5′UTR, and first intron, in between HMVEC and K562 cells. (D) Comparison of the GATA2 binding and the expression value in HMVEC and K562. The heat map from the expression ratio (HMVEC versus K562) was shown in left. Each gene obtaining the GATA2-binding sites from the ChIP-seq was shown with grey bar in middle. Common indicates the finding the GATA2 association into the gene commonly in HMVEC and K562. HMVEC and K562 design the unique GATA2 association separately in each cell. GSEA analysis showed the correlation between the expression ratio (HMVEC versus K562) and the GATA2-binding gene profiles in right. (E) Each three representative ChIP-seq with GATA2 result derived from HMVEC (left), common (middle), or K562 (right) expressed genes. #1; first ChIP-seq, and #2; second ChIP-seq.

Next, we identified the commonly recognized motif from the whole GATA2-mediated ChIP genome sequences. As shown in Figure 2B, the GATA2-recognized sequence (A/T)GATA(A/G) was indeed determined to be the highest enriched binding element, with E-value 1.3 × e⁻¹⁸⁴⁸. In addition to GATA2, Ets- and AP-1-recognized sequences were also found as the second and third enriched elements.

Recently, GATA2 ChIP-seq results were reported in the erythroid lineage cells, K562, using a similar strategy (Fujiwara et al, 2009). Interestingly, microvascular endothelial cells and K562 both commonly express GATA2, but the GATA2-regulated genes were not similar. Therefore, we compared GATA2 associations among the whole genome between HMVEC and K562 cells. In K562 cells, duplicate ChIP-seq with GATA2 resulted in the GATA2-recognized motif being selected as the highest enriched sequence. However, in contrast to HMVEC, the second and third most enriched motifs were Gfi-1b and TAL1, respectively (Figure 2B). As shown in Figure 2C, in total 896 and 334 genes were found to be GATA2 enriched in HMVEC and K562 cells, respectively. Among them, a total of 118 genes overlapped in both cells (Figure 2C), suggesting that distinct GATA2 associations with recognized elements in a cell-type specific manner could be observed.

To further investigate the correlation between GATA2 binding and resulting expression in HMVEC and K562 cells, we performed comprehensive mRNA expression analysis by triplicate microarrays in both cell types. We selected the gene set probes that exhibited significant expression, as a >100 with average difference in at least either HMVEC or K562 cells (Supplementary Table SII). Subsequently, a total of 16 873 probes were sorted by the ratio of the expression levels in HMVEC versus K562 cells as illustrated with a heat map (Figure 2D, left column). To integrate the ChIP-seq data from HMVEC and K562, a grey bar was shown for each probe corresponding to the positive within the up 5 kbp, TSS 1 kbp, 5′UTR and first intron of the ChIP-seq results (Figure 2D, middle column). Interestingly, roughly one third of genes expressed were categorized as highly HMVEC-expressed genes (means red in Figure 2D), which were also relatively dense in ChIP-seq in HMVEC, but not in K562 cells. In contrast, one third of genes were highly expressed in K562 cells (green in Figure 2D), which were also dense in ChIP-seq in K562 cells, yet were sparse in HMVEC (Figure 2C, middle column). Moreover, based on a ranking list of the expression ratio, HMVEC versus K562, a highly significant (P<10⁻⁶) GSEA score (Subramanian et al, 2005) was observed with GATA2 binding only in HMVEC, whereas a minus score was found with GATA2 binding specifically in K562 cells. Significant correlation was not observed within the common GATA2 bindings in both cells (Figure 2D, right column). Interestingly, the HMVEC highly expressed group included well-known endothelial-cell-specific genes, such as vWF, endothelin 1, PECAM1, while K562 cells highly expressed many important genes for erythroid cell differentiation, such as erythroid krüppel-like factor, GATA1, Gfi-1b, and NF-E2 (Figure 2D). The representative GATA2 binding at the proximal promoter derived from duplicate ChIP-seqs was shown in Figure 2E. Consistent with the expression, HMVEC specifically expressed endothelin 1, vWF, and PECAM1 showed GATA2 binding only in HMVEC (Figure 2E, left). In contrast, erythroid-specific NF-E2, GATA1, and SAM domain-containing protein (SAMSN1) revealed specific GATA2 binding in K562 cells alone (Figure 2E, right). Both stable GATA2 binding in HMVEC and K562 cells were found with commonly expressed genes, ICAM2; TAL1; and Friend of GATA (FOG) 2 (Figure 2E, middle). Taken together, these findings suggest that the presence of the GATA sequence alone in the genome cannot explain the association of GATA2. Rather, distinct cell-type specific regulation in endothelial cells and erythroid cells should exist for regulating GATA2 association.

GATA2 regulates endothelial-cell-specific gene expression and cell specificity

Next, to survey those genes regulated by GATA2 in microvascular endothelial cells, we used siRNA to knockdown GATA2 expression in HMVEC. As shown in Figure 3A, transfection of HMVEC with siRNA to GATA2 (si-GATA2), but not control (si-control), resulted in a 92% reduction in GATA2 mRNA. Moreover, virtually all GATA2 protein expression was blocked by the si-GATA2 treatment (Figure 3A). To comprehensively identify possible GATA2-regulated genes, we employed duplicate microarrays performed in parallel with HMVEC transfected with either si-control or si-GATA2. To identify genes regulated by GATA2, the selection criteria was set at >0.75 log fold upregulated or downregulated genes in the presence of si-GATA2, compared with si-control. A total of 788 (upregulated) and 818 (downregulated) genes were identified and clustered (Figure 3B, whole gene list shown in Supplementary Table SIII). Among them, GATA2 knockdown resulted in the reduction of the endothelial-specific marker, endothelin 1, vWF, KDR, endothelial-specific adhesion molecule, and endothelial differentiation factor-1. In contrast, GATA2 reduction led to the upregulation of the non-endothelial markers, SM-actin, SM22α, and claudin 1. Surprisingly, epithelial mesenchymal transition (EMT)-directed transcription factor, Snail, Slug, and HMGA2, were upregulated (Figure 3B). EndMT-directed TGF-β and the mediator SMAD were also increased (Figure 3B). The representative microarray-based findings were subsequently validated by quantitative real-time PCR (Figure 3B, right graphs). Significant GATA2 binding to the proximal promoter was also detected in TGF-β2 and Slug from duplicate ChIP-seqs (Supplementary Figure SII). To test whether GATA2 reduction leads to the loss of the endothelial specificity, we examined immunofluorescent staining with anti-vWF and SM-actin antibodies as endothelial and mesenchymal markers, respectively. As shown in Figure 3C, si-GATA2-treated cells exhibited a reduction of vWF-positive staining and a concomitant induction of SM-actin-positive staining. To quantify this ratio, we performed FACS analysis. Control siRNA-treated HMVEC resulted in 96.5% of cells staining gated-positive for vWF and 6.5% positive for SM-actin expression. In contrast, si-GATA2-treated HMVEC revealed a decrease in gated-positive cells to 88.6%, with a dramatic increase in cells gated-positive for SM-actin (77.6%) (Figure 3D). Moreover, GATA2 knockdown caused a roughly 50% reduction in the endothelial marker KDR, and up to a 63.5% induction of the mesenchymal marker; claudin 1 (Figure 3D). Collectively, these findings suggested that GATA2 is not only a key regulator of the expression of several endothelial-cell-specific genes, but also necessary for the maintenance of endothelial specificity.

Figure 3.

Essential role for GATA2 in microvascular endothelial cells. (A) (Left) Quantitative real-time PCR analysis of GATA2 mRNA in HMVEC transfected with si-control or si-GATA2. Data are expressed as the mean and the standard deviations of expression levels relative to cyclophilin A obtained in triplicate from five independent experiments, n=3. ^*P<0.001 compared with si-control. (Right) Western blot analysis of HMVEC transfected with si-control or si-GATA2. The membrane was immunoblotted with anti-GATA2 antibody. Anti-β-actin antibody was used as a loading control. The results are representative from three independent experiments. (B) (Left) Heat map of the HMVEC transfected with si-control or si-GATA2 derived from microarrays. Colour intensity: green—lower and red—higher, shown relative to median (black). (Right graphs) Quantitative real-time PCR analysis of the representative genes. Data are expressed as the mean and the standard deviations of expression levels relative to cyclophilin A obtained from three independent experiments, n=3. ^*P<0.05 compared with si-control. (C) Immunofluorescent stainings of HMVEC treated with si-control or si-GATA2, by using the anti-vWF (green) or SM-actin (red) antibodies. Nuclei was stained with DAPI (blue). The merged images show the representative from four independent experiments. Bar: 100 μm. (D) FACS analysis of HMVEC treated with si-control or si-GATA2. The results show the representative from three independent experiments. EC, window from the endothelial fraction.

GATA2 regulates endothelial-cell-specific endomucin gene expression

We decided to further focus our analysis on endomucin gene expression, since it showed the greatest reduction under si-GATA2 treatment. Human endomucin gene structure has been studied by two independent groups (Kinoshita et al, 2001; Samulowitz et al, 2002). ChIP-seq data suggested that GATA2 was recruited to the proximal promoter in the endomucin gene. The peak extended from 500 bp upstream to 200 bp downstream of the TSS (Figure 4A). This segment includes three consensus GATA-binding motifs (Figure 4C). To confirm whether these regions were in an active chromatin state, we performed ChIP and quantitative PCR (ChIP–qPCR) analysis using antibodies against acetylated histone 3 (H3Ac), acetylated histone 4 (H4Ac), trimethylated lysine 4 of histone 3 (H3K4me3), and monomethylated lysine 4 of histone 3 (H3K4me1), as well as GATA2. In addition, we employed ChIP–qPCR analysis with antibodies against serine 5 phosphorylated active polymerase II (polII), and the transcriptional cofactor, p300. As a negative control, we selected the promoter region from zinc-finger protein (znf) 649, which is known to be completely silent in endothelial cells. As shown in Figure 4B, the endomucin promoter indicated a profound enrichment with active PolII, H4Ac, H3Ac, H3K4me3, H3K4me1, p300, and GATA2 itself. In contrast, control IgG failed to enrich the endomucin promoter in HMVEC.

Figure 4.

GATA2-mediated endomucin gene expression. (A) Signals of duplicate ChIP-seqs data. Displayed enrich scores were calculated by QuEST, and visualized in the IGB software (Affymetrix). Significant GATA2-binding signals are shown in grey. – denotes Refseq genes located on the minus strand. (B) Formalin-fixed chromatin was immunoprecipitated with antibodies as shown. Immunoprecipitated DNA was quantified by the qPCR using the specific primers for endomucin and the znf 649, as a negative control. Fold enrichment were compared with non-immunoprecipitated chromatin, %INPUT. Data are expressed as mean and standard deviations obtained from three independent experiments, n=3. ^*P<0.001 compared with negative control, znf 649. NS, non-significant. (C) Schematic representation of the endomucin (−1968/+108)-luc and three consensus GATA motifs. (D) Endomucin (−1968/+108)-luc and TK-luc were transiently transfected into HUVEC, HMVEC, K562, or fibroblasts, and assayed for the luciferase activities. Each cell was cotransfected with pRL-CMV to normalize the transfection efficiency. The results show the mean and the standard deviations from the ratio of the endomucin-luc versus TK-luc, with three independent experiments, n=3. ^*P<0.001 compared with the activity from fibroblasts. (E) HMVEC were transiently transfected with either wild-type or a GATA-point mutated-luc. The results show the mean and the standard deviations of luciferase light units relative to the wild-type endomucin-luc, obtained in triplicate, n=4. ^*P<0.02 and ^**P<0.001, compared with the activity from the wild type. NS, non-significant. (F) COS7 cells were transiently cotransfected with the combination both endomucin-luc or GATA-point mutant, and expression plasmids for GATA2 (pcDNA3-GATA2) or mock control (pcDNA3). The results show the mean and the standard deviations of luciferase light units relative to the pcDNA3 alone, obtained in triplicate from three independent experiments, n=3. ^*P<0.01 compared with the activity from the pcDNA3 alone. NS, non-significant.

To determine whether GATA2 is required for endomucin gene expression, we employed luciferase reporter analysis. We isolated a fragment from the human endomucin promoter from −1968 to +108 relative to the TSS, and subcloned it into the luciferase vector pGL3 (Figure 4C). The plasmid, endomucin-luc, was transiently transfected into primary cultured human umbilical vein endothelial cells (HUVEC), HMVEC, human skin fibroblast, and K562 cells. As shown in Figure 4D, the endomucin promoter was greatly upregulated in HUVEC and HMVEC, but not in K562 nor fibroblasts. To determine whether the three GATA elements at +85 (motif 1), −40 (motif 2), and −330 (motif 3) contribute equally to endomucin gene expression in HMVEC, we generated point mutations of the GATA motif to TTTA in each region, and systematically measured luciferase activities. As shown in Figure 4E, compared with the wild-type promoter, mutation of motif 3 failed to reduce promoter activity. However, mutation of motif 1 and 2 resulted in a 37 and 48% reduction of promoter activity, respectively. Mutation of all three sites in combination further reduced (72%) promoter activity (Figure 4E), with mutation in motif 1 and 2 resulted in a similar reduction (∼70%) of promoter activity (data not shown). These results suggest that the two proximal GATA motifs are necessary for endomucin promoter activity in endothelial cells.

Subsequently, to test whether GATA2 transactivates the endomucin gene expression, COS7 cells were cotransfected with either an endomucin-luc or a GATA mutant-luc, along with a GATA2 expression vector or a mock control vector, pcDNA3. As shown in Figure 4F, GATA2 overexpression resulted in 6.9-fold increase in endomucin promoter activity, while GATA2 overexpression failed to transactivate the GATA-mutated endomucin promoter. In large vessel endothelium, GATA2, 3, and 6 were expressed (Song et al, 2009). Thus, we next verified whether endomucin expression was commonly regulated by the GATA family, or selectively regulated by GATA2 alone. In cotransfection assays with GATA2, 3, or 6, comparable GATA-mediated transactivation activities were detected with core-GATA element-containing thymidine kinase (TK)-luc (Supplementary Figure SIIIA). Importantly, under these conditions, GATA2 potentiated the highest transactivation activity in endomucin-luc (Supplementary Figure SIIIB). Compared with GATA2, GATA3 and 6 indicated weaker, but statistically significant upregulation of endomucin promoter activity (Supplementary Figure SIIIB). Moreover, knockdown experiments in HUVEC revealed that endomucin was reduced ∼90% via si-GATA2 treatment. Knockdown of GATA3 or 6 also showed more than half of the reduction. Combined siRNAs against GATA2, 3, and 6 resulted in a far more profound reduction of endomucin expression (Supplementary Figure SIIIC). Collectively, these findings suggest that among the GATA family, GATA2 is the most prominent factor for endomucin gene expression. The endomucin promoter specifically functions in endothelial cells, with the +85 and −40 GATA elements important to confer promoter activity.

Identification of the GATA-mediated distal enhancer regulating endomucin expression

From the above GATA2-mediated ChIP-seq findings, defined GATA2-binding regions were not concentrated in the proximal promoter regions, as many were found in the introns and intergenic regions. It has previously been reported that β-globin gene expression is regulated through a distal enhancer region bound by GATA factor (Kim et al, 2007). Thus, we wished to evaluate whether similar GATA2-mediated distal enhancers also existed in the genome of endothelial cells, and whether they regulated endomucin gene expression in an endothelial-cell-specific manner. To that end, we first evaluated ChIP-seq information with H3K4me1 and H3K4me3 bindings in HMVEC. Regions that showed specifically higher H3K4me1 signals compared with H3K4me3 were considered enhancer regions (Robertson et al, 2008). Thus, we searched the H3K4me1 signals surrounding the endomucin locus. As shown in Figure 5A, H3K4me3 signals (means the active core promoter definition) were detected only at the TSS. More importantly, H3K4me1-specific signals were detected at several points in the locus. In all, −219, −171, −139, −132, and +138 kbp regions (Figure 5A, bars) were chosen and validated with ChIP–qPCR analysis with H3K4me1 (Figure 5B). Compared with the negative control region at −84 kbp, H3K4me1 showed statistically significant binding to the positive region. Among them, −139 kbp region showed the highest and −171 kbp showed the lowest binding (Figure 5B). Subsequently, we searched the GATA2 binding within the locus. From the duplicate ChIP-seqs, GATA2 associations were commonly detected at −139 kbp and at a weaker level at −132 kbp (Figure 5A). ChIP–qPCR validation resulted in significant bindings of GATA2 to −132 and −139 kbp regions (Figure 5B). To verify which region from the −132 or −139 kbp is epigenetically designed as an enhancer, we performed ChIP–qPCR analysis with antibodies against active PolII, H3Ac, H4Ac, and p300. As shown in Figure 5B, compared with the negative control region at −84 kbp, only the −139 kbp region showed strong merged binding by p300, H3Ac, H4Ac, and active PolII. Moreover, the comparative genome analysis among mammalian species resulted in sharp homologous signals at −139 kbp (Figure 5A, bottom). Collectively, these findings suggest that the −139 kbp region is the main GATA2-binding enhancer within the endomucin locus.

Figure 5.

Identification of the GATA2-binding enhancer for the endomucin locus. (A) In all, 390 kbp UCSC hg18 genome browser view around the endomucin gene. The gene is transcribed from right to left. In HMVEC, GATA2-binding regions (blue) and H3K4me1 (red), H3K4me3 (pink) enriched regions were shown. Each black bar below the frame indicates the ChIP–qPCR amplification region, and the number show the distance (kbp) from TSS. At lower panel, magnified image around the −139 kbp region in which GATA2 signals and H3K4me1-positive signals are coexisted. Green bar shows the conservation rate (%) between mammalian species. (B) ChIP analysis with specific antibodies. Immunoprecipitated DNA were amplified with specific primers described in (A) and Supplementary Table SIV. Fold enrichment were calculated with the ratio from the znf 649 signals as a negative control. Data are expressed as the mean and the standard deviations obtained from three independent experiments, n=3. (C) HMVEC were transfected with TK-luc, GATA enhancer-TK-luc, or GATA enhancer (GATA mut)-TK-luc. HMVEC were cotransfected with pRL-CMV to normalize for the transfection efficiency. The results show the mean and the standard deviations of luciferase light units obtained in triplicate from three independent experiments, n=3. ^*P<0.001 compared with TK-luc, ^**P<0.01 compared with GATA enhancer-TK-luc. (D) HMVEC were transfected with endomucin (−1968/+108) promoter containing either −139 kbp core enhancer or GATA-mutated enhancer, n=3. ^*P<0.01 compared with endomucin (−1968/+108)-luc. NS, non-significant.

To further investigate whether the −139 kbp region enhanced promoter activity, we employed TK promoter-based luciferase assays. The core area (822 bp) at the −139 kbp region contains the consensus GATA element (Supplementary Figure SIV). Thus, we cloned and ligated the core 822 bp in upstream of the TK promoter. As shown in Figure 5C, addition of the GATA enhancer resulted in more than a six-fold increase in TK promoter activity over the TK basal promoter alone. In contrast, a GATA-mutated enhancer led to a >80% reduction of enhancer activity. Subsequently, to test whether the core area at the −139 kbp region works as an enhancer connected with the intact proximal endomucin promoter, we generated constructs and performed reporter assays. The endomucin promoter (2.2 kbp) contained high luciferase activity. Ligated with the core enhancer resulted in a further 2.2-fold increase in promoter activity (Figure 5D). While a GATA-mutated enhancer failed to transactivate the endomucin promoter. Moreover, an all GATA motifs mutated construct revealed a profound reduction in the basal endomucin promoter activity (Figure 5D). Taken together, these findings suggest that the −139 kbp region is in an active chromatin state and the core-GATA-binding element works as an enhancer in endothelial cells.

GATA2 regulates the chromatin structure of the endomucin locus in endothelial cells

GATA factor binds and activates not only a proximal promoter but also a distal enhancer region. It was also reported that GATA factor binding regulates the chromatin conformation (Jing et al, 2008). Therefore, to evaluate whether GATA binding at the proximal promoter and distal enhancer of the endomucin locus also induce chromatin remodelling, we performed chromatin conformation capture (3C) assays (Dekker et al, 2002). Endomucin was specifically expressed in endothelial cells but not K562 cells (Figure 6A). To compare the chromatin structure between HMVEC and K562 cells by 3C assay, we picked nine different points within the endomucin locus (Figure 6B). As shown in Figure 6C, 3C-mediated PCR amplification was detected not only in the TSS region, but also at the GATA associated −132 and −139 kbp regions (see Figure 5A and B) of the endomucin locus in HMVEC. In contrast, the chromatin derived from K562 cells exhibited 3C products only detected in the TSS region (Figure 6C).

Figure 6.

Endothelial-cell-specific epigenetical regulation in the endomucin locus. (A) RT–PCR was carried out with specific primers for GATA2, endomucin, or cyclophilin A by using the total RNA from HMVEC and K562 cells. The data show the representative from three independent experiments. (B) Long range interactions between the endomucin promoter and enhancer measured with 3C-qPCR. The organization of the endomucin locus is displayed. A grey bar represents each BstYI fragment. The grey bar indicates the bait region of the endomucin promoter. The numbers depict the distance (kbp) from the TSS. (C) 3C assay in HMVEC (left) and K562 (right). The x axis in each graph represents the location of the primers relative to the TSS. The y axis represents the cross-link frequency relative to it from the BAC DNA encompassing the endomucin region. Data are shown as the mean and the standard error from data in triplicate from two independent 3C experiments. (D) ChIP analysis with GATA2 antibody. Immunoprecipitated DNA were quantified with primer pairs described in Supplementary Table SIV. Fold enrichment was calculated with the INPUT. Data are expressed as the mean and the standard deviations obtained from three independent experiments, n=3. (E) HMVEC were transfected with si-control or si-GATA2. After 48 h, 3C assays were employed. Data are shown as the mean and the standard error from data in triplicate from two independent 3C experiments. (F) ChIP analysis with GATA2 antibody in K562. Primer pairs in TAL1 and MyoD1 promoters were used as a positive and a negative control, respectively. Data are expressed as the mean and the standard deviations obtained from triplicate samples, n=3. (G) ChIP analysis with H3K9me3 antibody in HMVEC and K562. Primer pairs in znf 180, 283, and β-actin were used as a positive and a negative control, respectively. Data are expressed as the mean and the standard deviations obtained from triplicate samples, n=3.

To verify whether GATA2 association directly mediates chromatin conformation at the endomucin locus, we treated cells with siRNA against GATA2 and again performed 3C assays. As shown in Figure 6D, si-GATA2 treatment led to more than a 90% reduction in GATA2 association with GATA sites in the endomucin proximal promoter (TSS) and distal enhancer (−139 kbp). Importantly, GATA2 knockdown reduced chromatin-loop formation at −139 kbp and the TSS of the endomucin locus (Figure 6E).

Next, we investigated why the endomucin gene was silent in K562 cells. GATA2 was highly expressed in both K562 and HMVEC cells (Figure 6A). However, GATA2-mediated ChIP-seq in K562 cells revealed non-GATA2 binding to the endomucin locus of the TSS and −139 kbp. The ratio was comparable to the non-expressed MyoD1 gene promoter (Figure 6F). In contrast, significant GATA2 binding was found in the well-expressed TAL1 gene promoter in K562 cells (Figure 6F). Epigenetically, silent gene loci are often determined by the binding of histon 3 lysine 9 trimethylation (H3K9me3) (Kouzarides, 2007). As shown in Figure 6G, high H3K9me3 enrichments were detected in the endomucin locus in K562 cells but not in HMVEC. Moreover, significant H3K9me3 enrichments were found in non-expressive genes, znf 180 and 283, in both K562 and HMVEC, while the ubiquitously expressed β-actin gene indicated neglectable H3K9me3 signals in both cells. Compared with HMVEC, the entire endomucin locus showed no active histone marks in K562 cells (Raney et al, 2011). Taken together, these findings suggest that GATA2 expressed in endothelial cells binds to the H3K9me3 free regulatory element on several lineage specific genes, including endomucin, and mediates endothelial-cell-specific chromatin conformation and determines endothelial cell specificity.

GATA2-regulated endomucin is essential for cell growth, migration, and tube formation in microvascular endothelial cells

Having established a critical role for GATA2 in endomucin gene expression in microvascular endothelial cells, we next wished to evaluate the functional relevance of endomucin in endothelial cells. To that end, we used siRNA to knockdown expression of endomucin. Transfection of HMVEC with two independent siRNAs against endomucin (si-endomucin oligo 1 and 2) resulted in >90% inhibition of the mRNA expression (Supplementary Figure SV). Of these, we first analysed the cell growth level. Consistent with previous findings (Mammoto et al, 2009), GATA2 knockdown impaired growth factor-mediated cell growth. Importantly, endomucin knockdown also attenuated the growth rate to 5.6 and 25.7% by siRNA oligo 1 and 2, respectively (Figure 7A). There was no evidence of increased apoptosis in siRNA-treated HMVEC (data not shown). Next, we performed cell migration assays using a modified Boyden chamber. Fluorescently labelled HMVEC transfected with control siRNA or siRNA against GATA2 or endomucin were plated in the fibronectin-coated upper chamber. The addition of endothelial culture media with growth factors resulted in significant increases in migration of control siRNA-treated cells (Figure 7B). In contrast, GATA2 knockdown abolished, and endomucin knockdown impaired by 80.5 and 88.9% (oligo 1 and 2, respectively), the growth factor-mediated cell migration (Figure 7B). Coordinate function from endothelial cell growth and migration are important for angiogenesis activities. Thus, we finally performed tube formation assays of HMVEC. HMVEC were fluorescently labelled and incubated in a collagen gel. As shown in Figure 7C, control siRNA-treated HMVEC promoted growth and migration towards each cell, leading to the formation of capillary-like cords or tubes. Quantitation revealed a 68.4 and 81.6% decrease in tube length in control versus endomucin siRNA (oligo 1 and 2, respectively)-treated cells (Figure 7D, red). Similarly, total vessel tree number was greatly attenuated by siRNAs against GATA2 or endomucin (Figure 7D, blue). Taken together, these findings suggest that endomucin is critical for cell growth, migration, and angiogenesis activities. Cell-specific expression regulating transcription factor; GATA2, and the downstream key target; endomucin, are indispensable for endothelial cell maintenance and physiological function.

Figure 7.

Knockdown of endomucin impair endothelial cell growth, migration, and matrix tube formation. (A) HMVEC were treated with siRNA against control, GATA2, or endomucin (oligo 1 and 2). After 6 h, HMVEC were re-seeded the concentration at 2 × 10⁵. After 1 and 2 days, each cell number was counted. Data are shown with the mean and the standard deviation from three independent experiments, n=4. ^*P<0.001 compared with si-control. (B) Migrated cells were quantified by cell image analyser in four independent experiments, n=3. ^*P<0.001 compared with si-control. (C) Tube formation assays of si-control, si-GATA2, or si-endomucin-treated HMVEC plated on collagen gel. Capillary-like morphology was observed under the fluorescent microscopy. Bar: 100 μm. (D) Quantification of the total tube length (red) and total vessel tree number (blue) was calculated using the cell image analyser from three arbitral optical images per two independent experiments. Data are expressed as mean and standard deviation from six independent results, n=3. ^* and ^#P<0.01 compared si-control in tube length and vessel tree number, respectively.

Discussion

Recent technological advancements now allow the defining of transcription factor binding regions and unveiling of the histone code on a genome-wide scale. In this report, we identified GATA2 binding in human primary cultured endothelial cells, and then comprehensively mapped and compared these signals with epigenetic histone marks. Duplicate ChIP-seqs with GATA2 demonstrated that only a small number of the GATA2-binding elements were located within the proximal promoter, the majority of which were consistent with a previous promoter study of GATA2 in endothelial cells. Most GATA2 associations were defined in the intergenic or distal regions, far from the core promoter of the gene (see Figure 2A). This GATA2-binding pattern is not unique, but consistent with other reports of ChIP-seq analysis on the whole genome (Bieda et al, 2006; Chong et al, 2010; Schmidt et al, 2010). Such a finding suggests that not only transcription factor binding to the core promoter, but also coordinated transcriptional regulations involved in distal enhancer, silencer, and insulator, might occur in cells. For example, the gata2 gene itself, GATA2 was reported to bind to the 9.5-kbp downstream of the mouse gata2, which enhances endothelial-cell-specific GATA2 gene expression (Khandekar et al, 2007). In another group, GATA2 has been shown to bind to the 77-kbp upstream of the mouse gata2 (Grass et al, 2006). In our ChIP-seq data with HMVEC, GATA2 associations were observed at the +9.8 and −80.6 kbp of the human gata2 gene (Supplementary Figure SVI), which exactly corresponds to the +9.5 and −77 kbp of the mouse gata2 gene, suggesting that sparse GATA2 binding at the genome-wide level could be functional. Endothelial-cell-specific human gata2 expression might be regulated with GATA2 binding itself.

Our microarray analysis with siRNA against GATA2 revealed that GATA2 is indispensable for many endothelial-cell-specific genes consistent with previous reports for vWF, KDR, and endothelin 1 (Wilson et al, 1990; Minami et al, 2001; Liu et al, 2009). Unexpectedly, GATA2 knockdown also indicated a role in the induction of cell conversion markers; Snail, Slug, HMGA2, and TGF-β2. EMT and EndMT were observed in pathological stages of cancer metastasis and cardiac fibrosis (Zeisberg et al, 2007a, 2007b). Moreover, we observed that GATA2 and 3 expression significantly declined in endothelium under inflammatory and tumour-invaded conditions (Song et al, 2009 and data not shown). While interesting, our cell staining studies showed that abrogation of GATA2 alone resulted in only a limited conversion. When lacking GATA2, the mesenchymal markers; SM-actin and claudin 1 were greatly induced, in turn, endothelial marker; KDR was markedly reduced. However, vWF was weakly reduced. A possible reason for only a partial reduction of vWF may be due to protein stability; another is based on the compensation from other endothelial-cell-specific transcription factors. Indeed, vWF is stored in the Weibel-Palade body (Ewenstein et al, 1987), which might affect the protein stability. Alternatively, we have recently shown the vWF gene is specifically transactivated by not only GATA but also Ets family member, ERG in endothelium in vivo (Liu et al, 2009, 2011). Correlating these findings, the combination of transcription factors, FOXC2 and Ets family; Etv2, was shown to have pivotal roles in endothelial specificity and differentiation in zebrafish (De Val et al, 2008). Thus, loss of a combination of GATA2 and another transcription factor might result in complete EndMT even in fully differentiated endothelial cells. Collectively, these findings strongly support that GATA2 is a key transcription factor for the maintenance of endothelial cells.

In investigating GATA2-mediated regulation, we chose to focus on the endomucin gene, as it exhibited the largest reduction in expression upon GATA2 knockdown. Endomucin has been reported to be an endothelial-cell-specific mucin involved in cell–cell or cell–extracellular matrix interactions (Ueno et al, 2001). Angiogenic stimulus increases endomucin gene expression in HUVEC (Liu et al, 2001). Moreover, here we show that knockdown of the endomucin by specific siRNA treatment resulted in the suppression of growth, migration, and matrix tube formation in HMVEC (see Figure 7). Recently, it has been shown that reduction in GATA2 levels leads to the abrogation of tube formation (Mammoto et al, 2009). Therefore, it appears that the GATA2-endomucin axis has an important role in vascular stability and angiogenesis. Further study with GATA2-ChIP-seq under detailed various angiogenic and inflammatory stimulus, or the generation of endomucin null mutations could serve to uncover the detailed mechanisms.

Duplicate ChIP-seqs with GATA2 in HMVEC indeed enriched for the sequence (A/T)GATA(A/G), a well-recognized consensus binding sequence. Such an element is never randomly enriched, since the associated E-value is 1.3 × e⁻¹⁸⁴⁸ (see Figure 2B). In our data, GATA2 was commonly expressed within HMVEC and K562 cells, but different GATA2-binding patterns indicated a cell-type specific manner. Moreover, GATA2 occupancy was positioned with H3K4me1, p300, and active PolII-positive enhancer region in the endomucin locus in endothelial cells. Therefore, it is interesting to consider the role of GATA2 in regulating endothelial cell specificity. Indeed, it is not well understood how GATA factor binds the appropriate sites with cell-type specificity. One possibility is that GATA factor interacts with other transcription factors, which select the binding region and define the tissue-specific expression. As shown in Figure 2B, the second and third enriched sequences via ChIP-seqs with GATA2 were different between endothelial cells and erythroids. K562 cells coenriched with Gfi-1b and TAL1, as reported with essential factors for erythroid differentiation (Saleque et al, 2002; Hall et al, 2003). In contrast, HMVEC coenriched with Ets family and AP-1. Our newly finding −139 kbp endomucin enhancer also involved AP-1 and Ets binding region (Supplementary Figure SIV). Functional interaction between GATA and AP-1 was reported in endothelial-specific endothelin 1 expression (Kawana et al, 1995). The importance of Ets family in endothelium specificity was discussed above. Moreover, we recently reported that GATA interacts with Ets factor, ELF1, which cooperatively induces endothelial-specific Tie2 gene expression (Song et al, 2009).

Another possibility is that GATA factor interacts with an epigenetically modified enzyme, which regulates chromatin conformation and drives tissue specificity. For example, GATA2 interacts with histone deacetylase (HDAC)3 and HDAC5 in haematopoietic stem cells (Ozawa et al, 2001). GATA1 interacts with EZH2 and Suz12 in erythroids that regulates the erythroid cell differentiation (Yu et al, 2009). However, it is not well understood whether pre-setting epigenetically modified enzymes allow the association of tissue-specific transcription factors, or the transcription factors positively regulate the epigenetic circumstance. Further analysis would be needed to define the possibility for endothelial cell differentiation.

Chromatin conformation undergoes dynamic changes in various differentiated cells. In erythroid cells, 3C assays were performed to show the communication between the promoter and locus control enhancer region in β-globin gene. GATA1 and FOG1 interaction regulates chromatin-loop formation (Kim et al, 2007). Here, we first show the endothelial-cell-specific GATA2-mediated chromatin conformation. Our 3C assays revealed a chromatin loop in the endomucin locus. GATA2 was indispensable for such loop formation. Similar to the β-globin locus in erythroids, a region over 100 kbp apart from the TSS region communicated with the core promoter, mediated by GATA2, p300, and active PolII. Such interactions might be associated with the active histone code in an endothelial-cell-specific manner, since H3K4me1 and H3K4me3 bindings were also detected in distal enhancer and core promoter, respectively, in HMVEC, while K562 cells did not show H3K4me1-positive or H3K4me3-positive, but rather H3K9me3-positive heterochromatin formation. In addition, our reporter assays revealed that the −139 kbp distal regulatory region also functioned as an enhancer for the heterologous (TK) promoter and the intact proximal endomucin promoter. Especially in endomucin, both of the GATA motifs in the distal enhancer and proximal promoter were essential for the expression in endothelial cells. Collectively, the −139 kbp distal regulatory region could function for not only the endothelial-specific chromatin-loop conformation, but also downstream endomucin gene expression.

In summary, our findings reveal a mechanism by which GATA2 binds and regulates the tissue-specific endomucin gene expression (Figure 8). Moreover, loss of GATA2 initiates cell-type conversion in several populations occurring even in epigenetically defined endothelial cells. Cooperation with epigenetically regulatory enzymes and tissue-specific transcription factors would define the endothelial cell commitment and differentiation.

Figure 8.

Schematic model for the endothelial-cell-specific endomucin gene expression.

Materials and methods

Additional information is available in the Supplementary data.

Cell culture

HUVEC and HMVEC were purchased from LONZA. All endothelial cells were cultured in EGM-2MV complete medium (LONZA). Human skin fibroblast (LONZA) and COS7 cells (ATCC CRL-1651) were cultured in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum (FBS). K562 cells (ATCC CCL-243) were grown in RPMI 1640 medium plus 10% FBS.

Immunohistochemistry

HMVEC or frozen tissue sections obtained from human dermis were fixed and incubated with an anti-VE-Cadherin antibody (R&D systems), anti-vWF antibody (Abcam), anti-SM-actin (Sigma), anti-CD34 (NOVOCASTRA), or an anti-GATA2 antibody. Sections were washed and incubated with secondary antibody labelled with Alexa-Fluor 488 or Alexa-Fluor 594 (Invitrogen) for 1 h. The slides were then washed and mounted in ProLong Gold antifade reagent with DAPI (Invitrogen), and examined by fluorescence microscopy.

DNA microarray

HMVEC were transfected with either si-control or si-GATA2 for 48 h. RNA was harvested and purified with Trizol (Invitrogen). Preparation of cRNA and hybridization of probe arrays were performed according to the manufacturer's instructions (Affymetrix). Annotation of the probe numbers and targeted sequences are shown on the Affymetrix web page.

ChIP-seq analysis

All protocols for Illumina/Solexa sequence preparation, sequencing, and quality control are provided by Illumina. A brief summary of the technique and minor protocol modifications are described in Supplementary data.

Chromosome conformation capture (3C) assay

HMVEC or K562 cells were cross-linked with 1% formaldehyde for 10 min. The reaction was stopped by the addition of 125 mM glycine for 5 min. Nuclei were re-suspended with restriction enzyme buffer, treated with 7.5 μl of 20% SDS at 37°C for 1 h, and then incubated with 50 μl of 20% Triton X-100 for 1 h. Samples were treated with 400 U of BstYI at 37°C for 16 h. After enzyme digestion, 40 μl of 20% SDS was added and incubated at 65°C for 10 min. The samples were diluted with ligation buffer and treated with T4 DNA ligase (2000 U) at 16°C for 16 h. Samples were finally reverse cross-linked and purified with spin columns (Qiagen). All primers were shown in Supplementary Table SIV. Three BAC clones (RP11-1041D18, RP11-760F14, RP11-891J2) were used to validate 3C assay.

FACS analysis

si-control-or si-GATA2-treated HMVEC were harvested with 0.05% trypsin plus 0.6 mM EDTA. Following cells were diluted with PBS plus 1% FBS, and fixed with ethanol. After washing by PBS twice, samples were re-suspended with PBS plus 0.2% Triton X-100, and incubated with antibodies for α-SM-actin (Sigma), vWF (Abcam), claudin 1 (Abnova), or KDR (Cell Signaling). PE-labelled F(ab′)₂ fragments of donkey anti-mouse IgG (for α-SM-actin or claudin 1) or goat anti-rabbit IgG labelled with Alexa-Fluor 488 (for vWF or KDR) were used as secondary antibodies. Finally, samples were re-suspended ∼6 × 10⁵ cells/ml with PBS plus 1% FBS, and immediately analysed by flow cytometry (Becton-Dickinson).

Data access

The data indicated in this publication are accessible through National Center for Biotechnology Information; Gene Expression Omnibus (GSE 28304) for microarrays, and Sequence Read Archive (SRA 030934.1) for ChIP-seqs.