Genetic framework for GATA factor function in vascular biology

Exquisite Cell Type-Specificity of GATA-2 Chromatin Occupancy.

To establish a framework for understanding GATA-2 function in endothelium, we used ChIP-seq to define the GATA-2 chromatin target sites in HUVEC. The ChIP assay was validated by quantitating GATA-2 occupancy at the GATA2 +9.5 kb enhancer (16), which activates a linked LacZ reporter in the vasculature of transgenic mouse embryos (17). GATA-2 occupancy at the +9.5 kb site was ∼15-fold greater than the preimmune control, whereas no significant occupancy was detected at the negative control necdin promoter (Fig. S1). The ChIP-seq analysis was conducted via the standard mode of analysis for the ENCODE Consortium. Immunoprecipitated DNA from two biological replicates was analyzed by ChIP-seq. Because the replicates met established criteria for reproducibility, the datasets were merged, and peaks were called again. Unique sequences were mapped to the genome (HG19). Replicates A and B yielded 9.9 million and 12.6 million unique sequences, respectively. Using a false-discovery rate of 0.0001, replicates A and B yielded 18,233 and 8,830 GATA-2–occupied loci (peaks), respectively. Comparison of replicates revealed major overlap, with 7,619 (86%) of the B peaks present in A. Merging the replicates yielded 15,529 peaks corresponding to 6,081 genes (Dataset S1). The mean peak height was 75 bp (range: 14–229 bp), and the mean width was 417 bp.

The GATA-2–occupied genes encode proteins known to be important for endothelial cell function, including endothelial NOS3, angiopoietins 1, 2, and 3, and intercellular adhesion molecule-2 (Fig. 1). Certain genes encoded transcription factors, including Kruppel-like factor-2 (KLF2) (Fig. 1). KLF2 is induced by shear stress, mediates shear stress signaling (18), determines vascular tone (19), mediates endothelial cell thrombotic function (20), and promotes vessel maturation (21). KLF2 is repressed by the proinflammatory cytokine IL-1β (18) and mediates the anti-inflammatory actions of statins (22). Additional transcription factors included the forkhead factor FOXP1 and multiple v-ets erythroblastosis virus E26 oncogene homolog (ETS)/ETS-variant (ETV) factors (Dataset S2), including ETV6, which is required for hematopoiesis and maintenance of the vascular network (23, 24). GATA-2 occupied myocyte-specific enhancer factor 2C (MEF2C), which is essential for vascular development (25). MEF2C contains an endothelial enhancer, corresponding to the GATA-2 peak at +139 kb, which directs endothelial cell-specific expression of a reporter in E7.5–E9.5 mice (26). The ChIP-seq analysis revealed a rich set of targets, including numerous genes not linked previously to endothelial cell biology.

The majority of GATA-2–occupied peaks (13,256 peaks, 85%) were not occupied in our prior K562 cell analysis (12) (Fig. 2A). In HUVEC and K562 cells (12), the majority of peaks resided distal to the genes in putative enhancer regions (Fig. 2B). The unique chromatin-bound sites in HUVEC vs. K562 cells indicates that GATA-2 occupies a very small subset of genomic GATA motifs in a cell type-specific manner. To explore the relationship between GATA-2 occupancy and the epigenome, we evaluated histone marks in HUVEC at occupied sites. This analysis revealed that 91% of GATA-2–occupied sites overlap with H3K4me1 (Fig. 2C), a diagnostic enhancer mark (27). A smaller subset of occupied sites (35%) overlaps with H3K4me3 (Fig. 2D). As expected for a mark associated with transcription (28), 61% of these overlapping peaks reside in introns and exons. A small cohort of occupied sites overlaps with H3K27me3 (Fig. 2E), H3K36me3 (Fig. 2F), and H4K20me1 (Fig. 2G) (1%, 3%, and 1%, respectively).

Linking GATA-2 with the Inflammatory Regulator AP-1.

We assessed whether the GATA-2 cistrome in HUVEC resembles that of K562 cells, in which the majority of GATA-2–occupied peaks reside at sites containing the GATA motif (12). Of the 15,529 GATA-2 peaks, 12,930 (83%) contained at least one WGATAR motif. Of the 6,976,111 GATA motifs in the human genome, 25,823 (0.37%) resided within the 15,529 peaks. Constrained de novo motif finding with Cosmo (29) identified WGATAA that we defined previously for GATA-1 occupancy in K562 cells (Fig. 3A) (12). A broader GATA consensus, [CG][AT]GATAA[GAC][GAC], resided at 8,176 (43.3%) of the GATA-2–occupied K562 cell peaks and 4,698 (35.4%) of the HUVEC peaks.

De novo motif finding using MEME (30) to pinpoint cis-elements at GATA-2–occupied sites in HUVEC revealed canonical motifs for the inflammatory regulator AP-1 (14) (Fig. 3B) and c-ETS (Fig. 3C), which has important functions in endothelial cells (31). AP-1 motifs (TGA[G/C]TCA) (32) resided in 45.4% of the HUVEC GATA-2 peaks (7,055 of 15,523 peaks). This was highly significant (P < 0.01) relative to 16.3% of GATA-2 peaks in K562 cells (3,451 of 21,167 peaks). Because 69% of GATA-2–occupied peaks overlapped with those occupied by the jun proto-oncogene (c-JUN) component of AP-1 (Fig. 3D and Dataset S3), 29% overlapped with FBJ murine osteosarcoma viral oncogene homolog (c-FOS) (Fig. 3E and Dataset S4), and only 1% (137) overlapped with peaks of c-Myc occupancy (23,525 total c-Myc peaks), this GATA-2–AP-1 link has broad importance. The location with respect to neighboring genes of sites containing both GATA-2 and c-JUN did not differ significantly from the location of the complete cohort of GATA-2 occupancy sites. Examples of AP-1 target genes in which GATA-2 overlapped with c-JUN and c-FOS include IL-8 (33), CXCL2 (34), and CSF2 (35). GATA-2, c-JUN, and c-FOS overlapped at RSPO3, which is not known to be an AP-1 target (Fig. 3F). A high fraction of chromatin sites bound by both GATA-2 and c-Jun were enriched in H3K4me1 or H3K4me3 (76 and 83%, respectively).

We compared the frequency in which the c-ETS motif resided at GATA-2–occupied sites in HUVEC vs. K562 cells. In HUVEC, 5,446 (35.1%) of the 15,523 GATA-2–occupied sites contained c-ETS motifs. In K562 cells, 2,737 (12.9%) of the 21,167 GATA-2–occupied sites contained c-ETS motifs. Based on a binomial test for proportions, the differential frequency of c-ETS motifs at GATA-2–occupied sites in HUVEC vs. K562 cells was highly significant (P < 2.2e⁻¹⁶). Only 2,249 (14.5%) of the 15,523 GATA-2–occupied sites in HUVEC contained both c-ETS and AP-1 motifs. A GATA-2 peak was more likely to contain an AP-1 motif if it lacked a c-ETS motif (P < 0.001). Thus, it is likely that GATA-2 does not function frequently as a ternary complex with AP-1 and c-ETS in HUVEC.

Direct GATA-2 Target Gene Ensemble in Endothelium.

The frequent occurrence of AP-1 motifs at GATA-2–occupied sites and GATA-2 sharing of occupancy sites with a key component of AP-1 suggest that GATA-2 and AP-1 interact functionally. We used RNA interference to establish GATA-2 target genes in HUVEC. GATA2 mRNA was considerably lower in HUVEC than in K562 cells (Fig. 4A) (12), but endogenous GATA-2 in HUVEC was detectable by Western blotting (Fig. 4C). GATA2 siRNA significantly reduced GATA2 mRNA and protein in HUVEC vs. control siRNA-transfected cells (Fig. 4 B and C) without overtly affecting HUVEC morphology. Expression profiling of two biological replicates was conducted, and genes dysregulated upon GATA-2 knockdown were analyzed further. The direct GATA-2 targets were identified by comparing the differential expression with GATA-2 occupancy determined by ChIP-seq. This analysis revealed 116 direct targets differentially expressed by twofold or more upon GATA-2 knockdown (Fig. 4D). Because knockdowns do not eliminate expression, the analysis underestimates the target gene ensemble.

The direct targets encode regulators of cell migration, including semaphorin 3A (36), and podocalyxin-like (37). GATA-2–activated direct targets encode proinflammatory mediators, including interleukins, the chemokine ligand CXCL2, and the cytokine CSF2 (GM-CSF). GATA-2 directly repressed KLF2, the key mediator of shear stress signaling in endothelium that opposes inflammation. GATA-2 activated RSPO3, which controls placental vascular plexus development and regulates angiogenesis during embryogenesis (38). Secreted RSPO3 activates wingless-type/β-catenin signaling, which exerts a proinflammatory function (38). GATA-2 induced expression of activating transcription factor 3 (ATF3), a stress-regulated transcription factor that protects endothelial cells from TNF-α–induced cell death (39) and negatively regulates allergic airway inflammation (40).

A subset of GATA-2–activated and –repressed targets was subjected to additional validation by real-time RT-PCR (Fig. 4 E and F) and quantitative ChIP analysis using the human β-globin (HBB) promoter as a negative control (Fig. 4G). ChIP-seq peak height correlated (R² = 0.68) with the quantitative ChIP signal (Fig. 4H).

Our results indicate that GATA-2, c-JUN, and c-FOS commonly occupy the same chromatin sites containing GATA and AP-1 motifs. Taken together with the fact that AP-1 is a key regulator of inflammatory genes such as IL-8 and CSF2 that are GATA-2 targets (Fig. 4 D and E), this finding suggests the existence of a GATA factor–AP-1 regulatory network to control inflammatory genes. Ontological analysis of the direct GATA-2 target-gene ensemble revealed an enrichment of inflammatory genes, the majority of which are GATA-2–activated (Fig. 5A).

GATA-2 Requirement for Assembly of Functional AP-1 Complexes on Chromatin.

In our loss-of-function analysis, HUVEC were cultured in medium containing FBS, EGF, and FGF2, which induce signaling that can activate AP-1 (13). Our results suggested that GATA-2 and AP-1 confer maximal AP-1 target gene transcription. We tested whether activity of presumptive AP-1 targets requires AP-1 in HUVEC. Transfection with dominant-negative c-FOS (A-FOS) (41, 42) significantly reduced IL-8 (P < 0.01), CSF2 (P < 0.01), and IL-6 (P < 0.01) expression (Fig. 5C). We tested genes that emerged from our direct GATA-2 target screen that were not established AP-1 targets. Expression of RSPO3 (P < 0.01), CXCL1, and RUNX1 was reduced by A-FOS. c-JUN, and c-FOS knockdowns confirmed AP-1 regulation of select GATA-2 targets; CSF2 and CXCL1 expression were not affected significantly (Fig. 5 B and C). Thus, GATA-2 and AP-1 ensure maximal activity of a cohort of the target genes.

We considered potential mechanisms underlying the GATA-2–AP-1 cooperation (Fig. 6A). At target genes (Fig. 6AI), reduced GATA-2 occupancy might decrease AP-1 occupancy (Fig. 6AII), decrease AP-1 phosphorylation (Fig. 6AIII), or induce AP-1–independent molecular alterations (Fig. 6AIV). To distinguish among these possibilities, occupancy of c-FOS, c-JUN, and c-JUN phosphorylated at serine 73 was analyzed in control siRNA- and GATA2 siRNA-transfected cells. The analysis was conducted under conditions in which GATA2 expression was greatly reduced (Fig. 6B). Knocking down GATA-2 significantly reduced phosphorylated c-JUN occupancy (P < 0.05), but not total c-JUN and c-FOS occupancy, at the IL-8 promoter and distal sites at RSPO3 and ATF3 genes (Fig. 6D). Loss of chromatin-bound phosphorylated c-JUN was not associated with a change in total c-JUN (Fig. 6C).

To assess the specificity of the GATA-2–dependent phosphorylated c-JUN occupancy, occupancy was analyzed at the GATA-2–insensitive IL-1B and IL-17D genes (Fig. 6D). GATA-2 knockdown did not affect phosphorylated c-JUN occupancy at these genes. Occupancy of all factors was low at the HBB promoter and an unoccupied site 1 kb upstream of the IL-8 promoter (Fig. 6D). The context-dependent GATA-2 requirement for phosphorylated c-JUN occupancy may explain why certain AP-1 target genes that are occupied by GATA-2 and AP-1 (Datasets S3 and S4) are insensitive to reducing GATA-2. In principle, GATA-2 may promote c-JUN phosphorylation indirectly or may promote phosphorylated c-JUN chromatin occupancy, and it has been reported that GATA-2 and c-JUN can interact (43). Coimmunoprecipitation analysis with HUVEC whole-cell lysates did not provide evidence for a stable complex containing GATA-2 and c-JUN.

We also tested whether c-JUN promotes GATA-2 chromatin occupancy, but knocking down c-JUN did not affect GATA-2 chromatin occupancy (Fig. S2).

To test whether GATA-2 facilitation of AP-1–mediated transcription has biological implications, we asked whether GATA-2 regulates secretion of the critical inflammatory mediator IL-8. GATA-2–knockdown HUVECs were treated with the proinflammatory factor TNF-α, which induced IL-8 secretion several fold as measured by ELISA (Fig. 6E). Knocking down GATA-2 significantly reduced IL-8 secretion (P = 0.0015), consistent with the mRNA expression data.

Linking GATA-2, AP-1, and Inflammatory Genes.

The results described herein, representing the genome-wide analysis of a GATA factor in endothelium, provide a rigorous genetic framework for understanding GATA-2 function in vascular biology. GATA-2 controls expression of genes involved in establishing endothelial cell phenotypes and inflammation. Analysis of 116 GATA-2–occupied genes differentially regulated upon knocking down GATA-2 using Ingenuity Pathways Analysis (http://www.ingenuity.com/products/pathways_analysis.html) revealed a complex network (Fig. S3) in which factors were knitted together based on established protein–protein interactions and abilities to regulate expression of other factors. The network highlights a GATA-2 function in regulating genes mediating inflammation (Fig. 6F). GATA-2 targets also included genes encoding anti-inflammatory factors, including TNFAIP3, which is induced by proinflammatory factors and suppresses proinflammatory signaling (44, 45) and ATF3 (40), as described above. Certain AP-1 targets are regulated by the proinflammatory factor NF-κB (46). Scanning the GATA-2 peaks with the NF-κB position/weight matrix from the JASPAR database using the FIMO tool (47) revealed that only 9.1% of the peaks match significantly (P ≤ 1e⁻⁴) to the matrix, a finding that is inconsistent with a frequent GATA-2–NF-κB association at HUVEC chromatin sites.

Because certain GATA-2 target genes with established roles in inflammation were co-occupied by GATA-2, c-JUN, and c-FOS, GATA-2 knockdown reduced phosphorylated c-JUN occupancy and IL-8 secretion, and AP-1 is a pivotal mediator of inflammation, it is attractive to propose that GATA-2–AP-1 cooperation is a common mode of orchestrating inflammatory processes. Vascular endothelial dysfunction involving inflammation underlies diseases including atherosclerosis, rheumatoid arthritis, and type II diabetes (15).

Given the biological importance of genes constituting the GATA-2 genetic network in HUVEC, why does the Gata2-knockout mouse appear to have morphologically normal vasculature, despite dying at E10.5 because of severe anemia (3)? Since multiple GATA factors are expressed in endothelium (48), a rigorous loss-of-function approach to ascertain GATA-2 function in the adult vasculature might require multigenic disruptions to reveal critical GATA factor-dependent pathways; alternatively, an important vascular activity of GATA-2 might not be apparent given the early embryonic lethality.

The direct regulation of inflammatory genes by GATA-2 has considerable biological and pathophysiological implications. It will be important to implement additional functional analyses and to determine whether endothelial cell subtypes differ in this mechanism. During the review of our manuscript, a genome-wide analysis of GATA-2 occupancy in human microvascular endothelial cells (HMVEC) was reported (49). Comparison of the HUVEC and HMVEC datasets will be informative.

Given the oncogenic activity of AP-1 subunits (50, 51), the role for GATA-3 in breast cancer (52), and links between GATA-2 dysregulation and leukemia (53), it will be important to determine whether GATA factors cooperate with AP-1 in diverse biological contexts and to what extent disruption of this mechanism underlies human pathologies. Because strong evidence implicates inflammation in a plethora of disease processes, including cancer, studies to evaluate GATA-2 function in endothelium in diverse disease models will be particularly instructive.

Cell Culture.

HUVEC were maintained in Medium 200 (Cascade Biologics) supplemented with Low Serum Growth Supplement (Cascade Biologics) and 1% antibiotic/antimycotic (Invitrogen). Cells were passaged at ∼70–80% confluence with 0.05% trypsin, and cells from passages four and six were used.

Antibodies.

Rabbit anti–GATA-2 polyclonal antibody was described previously (54). Anti–c-FOS (SC-253) and anti–c-JUN (SC-45) were from Santa Cruz Biotech. Anti–p-c-JUN (Ser73) was from Cell Signaling Technology (#9164). Mouse monoclonal α-tubulin (#CP06) was from Calbiochem.

Transfection and RNA Interference.

HUVEC were seeded 3–5 d before siRNA or plasmid transfection and grown to ∼70% confluence. After trypsinization, ∼10⁶ cells per experiment were resuspended in HUVEC Nucleofector Solution (Lonza), and 0.45–0.60 nmol of siRNA, 5 μg A-FOS expression vector, or empty vector was transfected using the Nucleofector system (Lonza). siGENOME SMARTpool GATA2 siRNA, c-JUN siRNA, c-FOS siRNA, and control siRNA (SMARTpool #1) were from Dharmacon (Thermo Scientific). CMV-500-A-Fos and CMV-500 empty vector were provided by Charles Vinson (National Cancer Institute, Bethesda, MD). Cells were analyzed 24 or 40 h after siRNA or A-Fos transfection, respectively.

Generation of ChIP-Seq Data.

HUVEC histone ChIP-seq data were generated at the Broad Institute and by the B. Bernstein group (Massachusetts General Hospital, Boston, MA). HUVEC c-JUN ChIP-seq data were generated by D. Raha and Mike Snyder (Stanford University, Stanford, CA). HUVEC c-Myc ChIP-seq data were generated by Vishy Iyer (University of Texas at Austin, TX). HUVEC GATA-2, HUVEC c-FOS, and K562 GATA-2 ChIP-seq data were generated by the P. Farnham group. These data were collected as part of the ENCODE Consortium (http://www.genome.gov/10005107) and are available on the University of California, Santa Cruz Genome Browser.

Quantitative ChIP Analysis.

ChIP was conducted as described (55) using 1 × 10⁸ HUVEC cells per condition for ChIP-seq, or 1 × 10⁷ HUVEC cells per condition for quantitative ChIP.

ChIP-Seq Analysis.

HUVEC were crosslinked for 10 min with 1% formaldehyde, snap frozen, and stored at −80 °C. ChIP was conducted as described (http://www.genomecenter.ucdavis.edu/farnham/pdf/FarnhamLabChIP%20Protocol.pdf). Chromatin from 10⁸ cells was diluted with three volumes of immunoprecipitation (IP) dilution buffer and incubated at 4 °C overnight with 60 μL of rabbit anti–GATA-2 antibody. StaphA cells were blocked with BSA. StaphA and Staph-Seq cells (SIGMA) were used. Blocked StaphA/Staph-Seq cells (100 μL) were added to the antibody/chromatin mixture and rotated for 15 min at room temperature. StaphA/Staph-Seq cells were washed twice with dialysis buffer and four times with polyclonal IP wash buffer. After reversal of crosslinks and RNase treatment, DNA was purified and analyzed as described with the Illumina GA2 platform (56). Short-sequence reads were aligned to the genome, and peaks were called using Sole-search (false-discovery rate, 0.0001; α value 0.001) with sequenced HUVEC input DNA as background (57). ChIP-seq data produced by the ENCODE Project Consortium were downloaded from the University of California, Santa Cruz browser and analyzed with Sole-search. Peak overlap was analyzed with Sole-search, allowing a distance of 200 bp between peaks.

Expression Profiling.

Control siRNA- and GATA2 siRNA-treated samples collected 24 h posttransfection were analyzed by hybridization to human 244k expression microarrays (Agilent Technologies). For each sample, 1.2 μg RNA was labeled with Cy-5 (control siRNA) or Cy-3 (GATA2 siRNA), and labeled cRNAs were combined and hybridized. Data were collected with an Agilent scanner. Two biological replicates for control siRNA/GATA-2 siRNA pairs were each hybridized in duplicate, yielding similar results, and average signals between the four replicates were used to establish fold-change cutoffs.

PCR Primers.

Primer sequences are indicated in Dataset S5.