Dynamic regulation of VEGF-inducible genes by an ERK/ERG/p300 transcriptional network

The growth of new blood vessels is requisite for tissue repair and homeostasis and contributes to the pathogenesis of several diseases, including cancer and diabetic retinopathy. A comprehensive understanding of the signaling pathways and downstream transcriptional networks that control angiogenesis could be leveraged to identify novel therapeutic targets to either promote or inhibit vascular growth. The central mechanism responsible for the majority of vascular growth is angiogenesis. Angiogenesis is a highly coordinated process that requires the interaction of several intracellular and intercellular signaling pathways that ultimately converge on a network of transcriptional pathways to elicit cellular behaviors (Herbert and Stainier, 2011). Vascular endothelial growth factor (VEGF), one of the central drivers of angiogenesis, is required for blood vessel development during embryogenesis (Carmeliet et al., 1996; Ferrara et al., 1996) and contributes to vascular homeostasis, as well as physiological and pathological postnatal vascular growth (Kim et al., 1993; Lee et al., 2007). VEGF activates a number of signal transduction pathways in endothelial cells (ECs) that modulate cytoskeletal dynamics and gene expression (Olsson et al., 2006), resulting in a suite of angiogenic cell behaviors, including directed, polarized cell migration. Although some of the transcriptional networks involved in VEGF signaling have been identified (Herbert and Stainier, 2011), much remains to be discovered regarding the mechanisms by which VEGF coordinates new vessel growth.

ECs receiving a threshold of VEGF stimulation initiate a signal transduction pathway that culminates in the transcription of the Notch ligand delta-like 4 (DLL4) (Lobov et al., 2007), as well as a network of other angiogenic genes (Liu et al., 2008). Phenotypic changes occur in the VEGF receiving cell, endowing it with ‘tip’ cell characteristics, including acquisition of numerous filopodial projections, increased migratory behavior, and elevated VEGF receptor 2 (VEGFR2; also known as KDR) expression (Blanco and Gerhardt, 2013). DLL4 on the surface of a tip cell binds to, and activates, Notch receptors on adjacent stalk cells. Notch activity in stalk cells induces the transcription of Notch-dependent genes, such as those encoding members of the basic helix-loop-helix transcription factor families HEY and HES, and suppresses filopodia formation and cell migration, while also dampening VEGFR2 expression. Importantly, tip and stalk cell phenotypes are dynamic, and in time a stalk cell can become a tip cell, and vice versa (Jakobsson et al., 2010). Coordinating these dynamic cellular behaviors is essential for an effective angiogenic response. The molecular mechanisms responsible for the maintenance and conversion between these phenotypes are only partially understood, and include oscillations in DLL4 induction in the tip cell (Lobov et al., 2007; Ubezio et al., 2016), as well as tight control of Notch signal duration in neighboring stalk cells (Guarani et al., 2011). How VEGF-regulated transcriptional programs control the dynamic expression of DLL4 and other angiogenic genes during sprouting angiogenesis remains poorly understood.

We previously identified a highly conserved enhancer element located within intron 3 of murine Dll4 that directs expression in arteries and angiogenic vessels (Wythe et al., 2013). Activity of this enhancer in arteries is VEGF responsive, and this is at least in part dependent on ETS transcription factors, including ETS-related gene (ERG) (Wythe et al., 2013). The ETS family of transcription factors play crucial roles in multiple stages of vascular development, including angiogenesis (Randi et al., 2009). More than a dozen ETS factors are expressed in ECs, and several of these [e.g. ETV2, TEL (ETV6), ETS1, ETS2, FLI1, ERG] have been implicated in vascular growth (Liu and Patient, 2008; Pham et al., 2007). ETS factors bind to a consensus 5′-GGA(A/T)-3′ sequence in the genome through a highly conserved ∼85 amino acid ETS domain (Sharrocks, 2001). Several of the family members also contain additional functional domains, such as the pointed (PNT) domain, a docking site for the serine/threonine kinase extracellular regulated kinase-2 (ERK2; MAPK1), which phosphorylates ETS1 and ETS2 in response to mitogen-activated protein kinase (MAPK) activation (Seidel and Graves, 2002). Phosphorylation of ETS1 and ETS2 enhances their activity through the recruitment of the transcriptional co-activator proteins p300 (EP300) and Creb-binding protein (CBP; CREBBP) (Foulds et al., 2004). Modulation of ETS factor activity by signal transduction pathways is not unique to ERK2, as other MAPK signaling pathways, such as p38 (MAPK14) and JNK (MAPK8) have been documented (Wasylyk et al., 1998; Yordy and Muise-Helmericks, 2000).

The specificity of MAPK pathways for particular ETS family members has recently been explored in prostate cancer cells in vitro. Interestingly, of the three MAPKs analyzed (p38, JNK and ERK2), only ERK2 phosphorylates ERG (yet multiple MAPK members act on ETS1/2) (Selvaraj et al., 2015). ERK2 primarily phosphorylates three residues on ERG: S96 (amino terminal to the PNT domain), S215 and S276 (both carboxy terminal to the PNT domain). Crucially, mutation of S215 to alanine, an amino acid refractory to phosphorylation (a so-called ‘phospho mutant’), abolishes ERG function in prostate cancer cells (Selvaraj et al., 2015). Although several studies have implicated ERG as a mediator of EC survival, proliferation, motility and vascular integrity (Birdsey et al., 2008, 2012, 2015; Liu and Patient, 2008; Yuan et al., 2011), whether ERG acts as a hub, integrating signals downstream of VEGF to control these diverse EC behaviors is not known.

Here, we explore the signaling and transcriptional pathways activated downstream of VEGF signaling in ECs. We find that the dynamic induction of MAPK/ERK activity controls DLL4 transcription in human ECs and that MAPK/ERK is required for angiogenesis in zebrafish in vivo. Furthermore, we demonstrate that MAPK/ERK activity leads to phosphorylation of ERG, and that ERG is required for the induction of DLL4 and a network of other angiogenic genes in human, mouse and zebrafish ECs. Mechanistically, we show that ERG recruits p300 to enhancer elements to coordinate angiogenic gene expression. These findings provide new insight into the molecular mechanisms of VEGF-mediated angiogenesis, and suggest that MAPK/ERK activation of ERG/p300 might represent a novel therapeutic target for modulating vascular growth.

Dll4 is dynamically expressed in tip cells during sprouting angiogenesis (Hellström et al., 2007; Jakobsson et al., 2010; Suchting et al., 2007; Ubezio et al., 2016). We first delineated the kinetics of VEGF-dependent DLL4 transcription in vitro. We assayed DLL4 unspliced pre-mRNA (as a surrogate of transcription) and mature mRNA transcript levels in VEGF-stimulated serum- and growth factor-starved human microvascular ECs (MVECs) or human umbilical vein ECs (HUVECs). In both cell types, DLL4 transcription responded dynamically to VEGF stimulation, peaking at 15-30 min (15-30′) post addition of VEGF, and returning to baseline levels by 2 h (Fig. S1A; Fig. 1A). The expression of spliced DLL4 mRNA was also transient and dynamic, with expression peaking at 1 h and returning to near baseline levels by 2 h (Fig. S1A; Fig. 1A).

As VEGF engagement of its principal angiogenic receptor, VEGFR2, can activate multiple downstream signaling pathways, we employed a panel of pharmacological cell signaling inhibitors to define the pathway(s) responsible for the rapid and transient induction of DLL4 transcription. Inhibition of the MAPK/ERK signaling pathway [using inhibitors of either MEK or protein kinase C (PKC)] abrogated DLL4 induction in response to VEGF (Fig. S1B; Fig. 1B). Immunofluorescent staining of VEGF-stimulated HUVECs revealed the presence of phosphorylated ERK (pERK) in both the nucleus and cytoplasm 15-30′ after treatment, with levels returning to baseline after 1 h (Fig. 1C). Measurement of pERK by western blotting mirrored the rapid and dynamic MAPK/ERK activation observed in immunofluorescence experiments, as pERK levels returned to near baseline levels by 1 h after stimulation (Fig. 1D). The kinetics of MAPK/ERK activation therefore parallels that of DLL4 transcription in response to VEGF treatment.

Several VEGF/MAPK/ERK-responsive genes have been characterized, including the immediate early gene early growth response 3 (EGR3) (Liu et al., 2008) and the ERK phosphatase dual specificity phosphatase 5 (DUSP5) (Bellou et al., 2009; Kucharska et al., 2009). The transcriptional induction of EGR3 and DUSP5 (as measured by qRT-PCR analysis of unspliced pre-mRNA) largely mirrored that of MAPK/ERK activation and DLL4 transcription (Fig. 1E). As expected, the induction of DLL4, EGR3 and DUSP5 mRNA was completely inhibited in the presence of the highly selective small molecule MEK inhibitor U0126 (Fig. 1F). In addition, the induction of these genes by VEGF stimulation was attenuated in HUVECs in which ERK1 (MAPK3) and ERK2 (MAPK1) were knocked down by siRNA (Fig. S1C). To determine whether MAPK/ERK activity in the absence of VEGF signaling was sufficient to induce expression of these genes, we treated serum-starved HUVECs with a PKC activator (and therefore an activator of MEK/ERK signaling), phorbol-ester myristate acetate (PMA) (Franklin et al., 1994; Schultz et al., 1997). PMA treatment markedly elevated DLL4, EGR3 and DUSP5 mRNA levels, and this response was blocked by pre-treatment with U0126, demonstrating that the MEK/ERK pathway is necessary and sufficient to activate transcription of a subset of angiogenic genes (Fig. 1G,H).

We further assessed the physiological relevance of MAPK/ERK signaling during sprouting angiogenesis in vivo. In agreement with recent reports (Costa et al., 2016; Shin et al., 2016), pERK was enriched in angiogenic sprouts (i.e. intersomitic vessels) in developing zebrafish embryos, indicative of active MAPK/ERK signaling (Fig. 2A; Fig. S2A). Importantly, inhibition of MAPK/ERK signaling using the MEK inhibitor SL327 completely abrogated the pERK signal throughout the embryo, including in the sprouting vessels (Fig. 2A; Fig. S2A). Inhibition of MEK signaling had a functional effect on angiogenesis, as sprout length (Fig. 2B) and the number of ECs per sprout (Fig. S2B) were decreased, as demonstrated previously (Shin et al., 2016). At this dose, SL327 did not cause developmental delay or necrosis (Fig. S2C). Inhibition of MAPK/ERK signaling diminished the expression of dll4 mRNA, as determined by qRT-PCR (Fig. 2C). Furthermore, time-lapse microscopy using a Notch biosensor revealed that attenuation of MEK activity reduced Notch signaling within the developing vasculature in vivo, and these results were confirmed by static confocal microscopy of a conventional Notch reporter (Fig. 2D,E; Fig. S3; Movies 1 and 2).

To determine whether ERG is required for the dynamic induction of DLL4 downstream of VEGF, we knocked down ERG using siRNA in HUVECs. ERG knockdown reduced the basal levels of DLL4 and completely abrogated the induction of DLL4 in response to VEGF stimulation (Fig. 3A,B). Furthermore, activation of MAPK/ERK with PMA stimulation failed to elevate DLL4 transcription in ERG knockdown cells, confirming that ERG functions downstream of VEGF and MAPK/ERK (Fig. 3C).

To explore further the relationship between ETS factors and MAPK activity, we tested whether MAPK/ERK signaling modulates ETS factor activity by creating a luciferase reporter construct under the control of a concatemer (eight tandem copies) of the ETS-DNA binding site within the intron 3 enhancer of murine Dll4 [identified by Wythe et al. (2013)]. ETS reporter activity in bovine aortic ECs (BAECs) was attenuated by both MEK and PKC inhibition (Fig. 3D). This suggests that ETS factor transactivation is controlled by MAPK/ERK signaling.

Selvaraj et al. recently demonstrated that ERK2 preferentially bound and phosphorylated ERG at serines 96 (S96), 215 (S215) and 276 (S276), and that S215 phosphorylation was required for ERG activity in prostate cancer cells (Selvaraj et al., 2015). We found that S215 was dynamically phosphorylated in ECs in response to VEGF stimulation, with peak phosphorylation occurring at 15-30′, which coincides with increased MAPK/ERK activity following VEGF treatment (Fig. 3E). Pretreatment with a MEK inhibitor abolished S215 phosphorylation (Fig. 3F). To determine the functional importance of ERK-mediated phosphorylation of ERG, we eliminated endogenous ERG using an siRNA directed to the 3′ UTR of ERG and then reintroduced wild-type or phospho-mutant ERG. Expression of wild-type ERG restored DLL4 transcription, whereas expression of ERG containing a mutation of one phosphorylation site (S215A) had less activity, and ERG containing mutations in all three ERK-phosphorylated residues (S96A, S215A, S276A) failed to rescue DLL4 transcription (Fig. 3G). This suggests that ERK phosphorylation is functionally important in dictating ERG activity.

To test further the functional importance of ERG phosphorylation, transplantation experiments were performed in zebrafish. Wild-type or mutant (S96A, S215A, S276A) ERG mRNA was injected into kdrl:nls-EGFP donor embryos, followed by transplantation of these cells into kdrl:mCherry recipient hosts at sphere stage. The location of the donor cells within the trunk vasculature was scored at 28-30 hours post-fertilization (hpf) to determine whether expression of mutant ERG affects the ability of these cells to contribute to angiogenesis. There appeared to be no overt phenotypic consequence following mosaic overexpression of wild-type or mutant ERG. However, the percentage of ERG mutant-expressing cells contributing to intersegmental vessels (ISVs) (but not other vascular structures) was significantly reduced compared with wild-type ERG-expressing cells (Fig. 3H).

p300 is recruited to VEGF-dependent enhancers and is required for regulating the expression of many angiogenic genes (Zhang et al., 2013). As the earliest time-point previously examined was 1 h post-VEGF treatment, we sought to define the dynamic nature of p300 recruitment to the DLL4 enhancer by ChIP assays in VEGF-stimulated HUVECs. ERG recruitment was modestly enhanced by VEGF treatment (Fig. 4A), but there is high basal ERG occupancy at this enhancer (Wythe et al., 2013). Strikingly, p300 was transiently recruited to the intronic DLL4 enhancer 15-30′ after VEGF stimulation, mirroring the robust and transient increase in MEK/ERK activity and ERG phosphorylation (Fig. 4A). Of note, p300 recruitment did not coincide with increased acetylation of K27 of histone H3 (H3K27ac), although acetylation is already high at this region in ECs (Wythe et al., 2013). Importantly, we found that p300 recruitment to the intronic DLL4 enhancer required ERG (Fig. 4B). In addition, co-immunoprecipitation in ECs demonstrated that ERG and p300 physically interacted following VEGF stimulation (Fig. 4C), and that this interaction was lost in cells expressing a phospho-mutant ERG (S96A, S215A, S276A) protein (Fig. 4D). To determine the functional importance of p300 in DLL4 induction, we utilized a small molecule inhibitor of p300 and CBP histone acetyltransferase activity, c646 (Bowers et al., 2010). Inhibition of p300/CBP activity in HUVECs in vitro did not affect basal levels of DLL4, but completely blocked VEGF induction of DLL4 mRNA (Fig. 4E). Furthermore, inhibition of p300/CBP in zebrafish suppressed elongation of intersomitic vessels (Fig. 4F), but did not result in other gross developmental defects (Fig. S2C).

To determine the extent of the genetic network regulated by ERG, we transfected HUVECs with control or ERG siRNAs and performed microarray analysis of gene expression in serum-starved and VEGF-stimulated cells. We focused on transcripts induced at early stages of VEGF stimulation (i.e. within 1 h) to identify genes directly regulated by VEGF/ERK/ERG. Knockdown of ERG resulted in the downregulation of 202 genes, including CLDN5, RASIP1 and ARHGAP28 in serum-starved cells (Fig. S4A), consistent with previous studies (Birdsey et al., 2012; Yuan et al., 2012), and the upregulation of 68 genes. Gene ontology (GO) analysis revealed that the most frequent functional categories altered following loss of ERG were: response to wounding, inflammation, cell migration, cell motility and cell adhesion (Fig. S4B). VEGF treatment increased the expression of 160 genes and downregulated only four genes (Fig. S5A). Analysis of the genes modulated by VEGF revealed GO terms associated with transcriptional regulation and gene expression, cell proliferation and vascular development (Fig. S5B). Of the VEGF-induced genes, 30 (representing ∼19% of all VEGF-inducible genes) were attenuated in ERG knockdown cells, and 14 genes (∼9% of all VEGF-inducible genes) were further elevated in ERG knockdown cells (Fig. 5A,B). GO analysis revealed that these genes (e.g. NRARP, HLX, DUSP5, EGR3 and PIK3R1; Bellou et al., 2009; Herbert et al., 2012; Liu et al., 2008; Nicoli et al., 2012; Phng et al., 2009) are implicated in angiogenesis, blood vessel morphogenesis and development, homotypic cell-cell adhesion, cell proliferation and differentiation (Fig. 5C).

Examining the kinetics of a subset of the ERG-dependent, VEGF-induced genes revealed that their transcription is increased transiently by VEGF stimulation, with peak transcription occurring between 15′ and 1 h (Fig. S6). To query the requirement of MAPK/ERK signaling and p300 activity in this response, we measured the induction of these genes in the presence of MEK or p300/CBP inhibitors. We found that 12 of 16 (75%) were MEK dependent, and all required p300/CBP activity (Fig. 5D,E).

To probe the relevance of this pathway in vivo, we assessed the expression of several of the identified genes by in situ hybridization in zebrafish embryos treated with inhibitors of VEGF, MEK or p300/CBP. Importantly, dll4, hlx1 and dusp5 were regulated by this pathway within ISVs (Fig. 6A,B). In addition, flt4, which is regulated by MAPK/ERK signaling during sprouting angiogenesis (Shin et al., 2016), was also dependent on p300/CBP and MEK (Fig. 6A,B). To induce ectopic activation of the MAPK/ERK pathway in a VEGF-independent manner, kdrl:GFP zebrafish embryos were exposed to PMA for 2 h (until 24 hpf). PMA treatment induced the phosphorylation of ERK1/2 in a MEK-dependent manner (Fig. 6C) and led to the induction of dll4, hlx1, dusp5 and egr3 expression in the endothelium, as determined by qRT-PCR from fluorescence-activated cell sorting (FACS)-isolated ECs (Fig. 6D). Importantly, pre-treatment of the embryos with c646 inhibited the PMA-induced induction of dll4, hlx1 and egr3, whereas dusp5 was refractory to c646 inhibition (Fig. 6E).

To assess further the role of ERG in angiogenesis in vivo, we generated a novel Erg knockout/lacZ knock-in mouse line. Deletion of Erg resulted in embryonic lethality by embryonic day (E) 11.5-E12.5, similar to previous reports (Birdsey et al., 2015; Vijayaraj et al., 2012) (Fig. S7A-F). Following loss of ERG protein (Fig. 7A-B′), we observed major defects in vascular integrity and angiogenesis during embryogenesis, within both the cranial and the trunk vasculature (Fig. 7C-F). Conditional deletion of Erg (Erg^iECKO) using an EC-specific CreERT2 driver [Cdh5(PAC)-CreERT2] (Wang et al., 2010) led to defects in physiological angiogenesis, as determined by examination of angiogenesis within the postnatal retina (Fig. 7G-J; Fig. S7G-K). These data, combined with the embryonic lethality, hemorrhage and reduced angiogenesis all demonstrate a requirement for ERG in physiological angiogenesis.

To determine whether the candidate genes identified by our in vitro screen are downstream of Erg in vivo, we isolated ECs from wild-type or Erg mutant mouse embryos and performed qRT-PCR for several of the ERG- or VEGF/ERG-dependent genes (Fig. 7K,L). We found that several ERG-dependent, VEGF-independent genes identified in our screen (e.g. Rasip1, Sox18) or by others [e.g. Cdh5 (Birdsey et al., 2008; Gory et al., 1998) and Cldn5 (Yuan et al., 2012)] were downregulated in Erg loss-of-function embryos (Fig. 7K). Similarly, we observed a significant reduction in a typical VEGF-induced, ERG-dependent transcript, Dll4, in agreement with previous results (Wythe et al., 2013) (Fig. 7L). Additional candidates in this category, which showed robust sensitivity to MAPK and P300 activity in vitro, were substantially downregulated in vivo (e.g. Fjx1, Pik3r1, Sdf2l1, Nrarp). Collectively, these findings demonstrate that a VEGF/MAPK/ERG/p300 cascade is a crucial regulator of angiogenesis in vitro and in vivo.

We next sought to identify the enhancers/promoters that ERG might directly act upon to regulate this gene network. We previously found that conserved orthologous transcription factor binding can reveal functional enhancers (Ballester et al., 2014). To identify evolutionarily conserved, epigenetically modified enhancers for further functional analyses, we performed ERG ChIP-seq experiments in both human (HUVECs) and bovine (BAECs) ECs cultured in complete media (i.e. containing VEGF) (Fig. 8). We identified 31,175 ERG ChIP-seq peaks in HUVECs and 34,773 peaks in BAECs, and found that 8337 of the human peaks were conserved in cow (Tables S3 and S4). We also performed H3K27ac ChIP-seq and found that 94% of conserved ERG peaks overlapped H3K27ac-enriched regions, supporting their association with active enhancers. We found that the DLL4 locus contains multiple conserved ERG-bound enhancers, including regions ∼12 kb upstream of the transcriptional start site (TSS) and within intron 3 (Fig. 8A), both of which were previously shown to have arterial-specific activity in vivo (Sacilotto et al., 2013; Wythe et al., 2013). We also identified an ERG-bound enhancer, conserved in cows and humans, ∼3.0 kb upstream of the gene H2.0-like homeobox (HLX) (Fig. 8B; Fig. S8). HLX, a homeobox transcription factor, expression of which is induced by VEGF in vitro (Schweighofer et al., 2009), has been implicated in controlling angiogenic sprouting of human cells in vitro (Prahst et al., 2014; Testori et al., 2011), and in ISV formation in zebrafish (Herbert et al., 2012), as well as yolk sac vascular remodeling in the mouse (Prahst et al., 2014). Further analysis of our ChIP-seq data revealed that the majority of ERG/VEGF-regulated genes (25 of 44) had an ERG ChIP-seq peak within 10 kb of the TSS, and ERG binding was conserved in cow for 16 of these genes (Fig. 8C). This is suggestive of direct regulation of these genes by ERG. Furthermore, ERG binding was significantly enriched near ERG- and ERG/VEGF-regulated genes (Fig. 8D).

HLX transcription is transiently induced in response to VEGF stimulation, similar to DLL4 (Fig. 9A). We found that p300 was dynamically recruited to this evolutionarily conserved non-coding region (Fig. 9B) in an ERG-dependent manner (Fig. 9C). We cloned this conserved H3K27ac- and ERG-enriched −3 kb 5′ putative regulatory region (HLX-3a, 1565 bp fragment) upstream of a minimal promoter (SV40) driving a luciferase reporter, and found that it was VEGF responsive, and that the basal and VEGF-induced activity of this enhancer required ETS DNA-binding sequences (Fig. 9D). Furthermore, inhibition of MEK activity abrogated the VEGF responsiveness of this regulatory region (Fig. 9E). Although the full HLX-3a regulatory region failed to drive endothelial expression in vivo (data not shown; n=75), refinement of the element to the region bound by ERG (which was highly conserved across vertebrates; Fig. S8) and the 3′ acetylated region (HLX-3b, 435 bp fragment) drove robust EGFP reporter activity in the vasculature of the embryonic zebrafish (Fig. 9F). EGFP reporter expression was preferentially observed in the ECs of the ISVs (which form by angiogenesis) compared with the axial vessels (which form by vasculogenesis), and reporter activity was ETS element dependent (Fig. 9F). To test further the functional importance of this enhancer, we utilized clustered regularly interspaced short palindromic repeats/Cas9 (CRISPR/Cas9) genome editing to delete a portion (1201 bp; see Fig. 8B for schematic) of the H3K27ac-enriched, ERG-bound region upstream of HLX in TeloHAECs, an immortalized human aortic EC line. Several clonal lines (ΔHLX15, ΔHLX17 and ΔHLX21) heterozygous for deletion of this region were generated and confirmed by PCR and DNA sequencing (data not shown). Comparison was made with a clonal line generated following transfection of scrambled control gRNAs (Scr3). Although the basal expression of HLX appeared to be unaffected in the deletion lines, the VEGF-dependent induction of HLX appeared to be diminished (Fig. 9G). In contrast, DLL4 induction was unaffected. Furthermore, knockdown of ERG appeared to attenuate the induction of HLX to a greater extent in the control line compared with the deletion lines, implying that ERG acts through the deleted enhancer region. Collectively, these findings demonstrate the requirement of a highly conserved ERG-bound regulatory element in the VEGF responsiveness of the angiogenic gene HLX.

Dynamic control of gene expression and the resultant cellular outputs in tip cells and adjacent stalk cells are central to the growth of nascent angiogenic sprouts (Blanco and Gerhardt, 2013; Lobov et al., 2007). The pathways that regulate the temporal VEGF-dependent expression of DLL4 in tip cells are of particular importance considering the central role of this ligand, and its receptor, Notch, in directing tip and stalk cell behaviors (Hellström et al., 2007; Jakobsson et al., 2010; Suchting et al., 2007; Ubezio et al., 2016). Here, we identify one potential mechanism for the transient VEGF-dependent induction of DLL4 transcription. VEGF stimulation initiates a rapid and transient burst of MAPK/ERK activity, with similar kinetics to VEGF induction of DLL4 transcription. Downstream target proteins modulated by ERK kinase activity are also dynamically modified, as illustrated by the transient phosphorylation of ERG at serine 215. Additionally, we find that the co-activator p300 is recruited to angiogenic enhancers in an ERG-dependent manner, with kinetics mirroring ERG phosphorylation. This VEGF/ERK/ERG/p300 transcriptional pathway also dynamically regulates a network of genes shown to positively (e.g. HLX, FJX1, EGR3, APOLD1, ADRB2, EPHA2, FUT1, MAP2K3, NDRG1) and negatively (e.g. DUSP5, NRARP) regulate angiogenesis (Fig. 9H) (Al-Greene et al., 2013; Bellou et al., 2009; Herbert et al., 2012; Iaccarino et al., 2005; Liu et al., 2008; Mirza et al., 2013; Moehler et al., 2008; Phng et al., 2009; Pin et al., 2012; Prahst et al., 2014; Toffoli et al., 2009; Zhou et al., 2011).

The ETS family of transcription factors has previously been implicated as signal-dependent effectors (Wasylyk et al., 1998), but how ETS factors act downstream of VEGF has not been explored in detail. Interestingly, we find that DLL4 induction by VEGF signaling requires MAPK/ERK signaling, as well as ERG expression. Previous studies in cancer cells revealed that phosphorylation of ERG at S96, S215 and S276 is mediated by ERK2 (Selvaraj et al., 2015). We find that VEGF signaling leads to dynamic ERK-dependent phosphorylation of ERG at S215, and that S96, S215 and S276 are required for maximal ERG activity. The ability of ERG to drive expression of VEGF target genes appears to be p300 dependent, as VEGF initiates a physical interaction between ERG and p300, and ERG is required for p300 recruitment to DLL4 and HLX enhancer elements. Furthermore, p300/CBP inhibition abolishes VEGF/ERG-dependent gene expression. Mutation of ERK-phosphorylated residues in ERG prevents its interaction with p300, suggesting a role for ERG phosphorylation in recruitment of p300 to target genes. Furthermore, the termination of p300 recruitment temporally coincides with loss of ERG phosphorylation, implying a functional role for these phosphorylation events. It will be of interest to determine whether all DNA-localized ERG, or only those molecules involved in VEGF signaling output, become phosphorylated in response to VEGF signaling. It will also be of interest to determine how diverse activators of MAPK/ERK signaling, which have distinct effects on angiogenesis (e.g. ANG1/TIE2 (ANGPT1/TEK) signaling), might differentially activate ERG. Answering these questions will be vital for the development of targeted therapeutics to suppress angiogenesis. Because ERG regulates vascular integrity (presumably in cells lacking active ERK), it is possible that ERG functions to maintain vascular stability in an ERK-independent manner, suggesting the possibility of selectively blocking angiogenesis through targeting ERG phosphorylation, while maintaining vascular stability. Furthermore, ERG is known to function as an oncogenic fusion protein (e.g. TMPRSS2-ERG) in prostate cancer (Adamo and Ladomery, 2016). Of note, the amino terminus of ERG (included in many of these fusion proteins) appears to contain the same serine residues phosphorylated by ERK2. It will be of interest to determine how upstream signaling pathways (e.g. activated RAS/MAPK/ERK) influence ERG transcriptional activity in cancer. Perhaps targeting ERG phosphorylation could be of interest to quell ERG oncogenic activity.

Previous studies identified a role for another ETS factor, TEL (ETV6), in the repression, rather than the activation, of DLL4 (Roukens et al., 2010). In this case, TEL bound to the DLL4 promoter under basal conditions to recruit a co-repressor protein, CTBP. Addition of VEGF led to the rapid disassembly of this repressive complex. The kinetics of this repressive TEL/CTBP complex disassembly are comparable to the assembly of the activating ERG/p300 complex that we report here, suggesting that TEL and ERG dynamically control co-activator/co-repressor recruitment. Recently, VEGF has also been shown to stimulate dynamic exchange of co-repressors for co-activators bound to MEF2 transcription factors (Sacilotto et al., 2016), suggesting that several families of transcription factors may coordinate VEGF-dependent sprouting angiogenesis. Indeed, ETS proteins interact with multiple transcription factor families (Carrère et al., 1998; De Val et al., 2008). Of note, we have identified a number of transcription factor binding motifs that are enriched under ERG ChIP-seq peaks in the vicinity of ERG- and ERG/VEGF-dependent genes that might functionally interact with ERG to control gene expression (Table S3). Although a subset of VEGF inducible genes are regulated by ERG, it is equally important to note that many VEGF-dependent genes are ERG independent. This could be attributable to redundancy of ETS factors, but could also imply that additional transcriptional pathways responsible for angiogenic gene regulation remain to be uncovered.

In summary, our study has identified a VEGF/MAPK/ERK/ERG/p300 network that is required for the induction of a subset of VEGF-inducible genes in ECs, allowing us to propose a model for how transient activation of an angiogenic program might be regulated to orchestrate sprouting (Fig. 9H).

Zebrafish protocols were approved by the Animal Care Committee at the University Health Network, the University of Texas MD Anderson Cancer Center and Baylor College of Medicine. The following transgenic lines were utilized: Tg(kdrl:mCherry)^ci5 (Proulx et al., 2010), Tg(kdrl:GFP)^s843 (Jin et al., 2005), Tg(fli1a:nls-EGFP)^y7 (Roman et al., 2002), Tg(TP1bglob:EGFP)^um14 (Parsons et al., 2009) and Tg(TP1bglob:VenusPEST)^S940 (Ninov et al., 2012).

Embryos were dechorionated, then treated from 18-20 hpf until 26-28 hpf (unless noted otherwise) with the following inhibitors: SU5416 (VEGFR2 inhibitor, 5 μM, LC Laboratories), SL327 (MEK inhibitor, 30 μM, Sigma) or c646 (p300 inhibitor, 3 μM, Sigma), with all inhibitors prepared as 1000× stocks in DMSO, and embryos treated in E3 supplemented with PTU to prevent pigmentation. PMA (Bioshop) was used at a concentration of 1 μM. DMSO (0.1%) was used as a vehicle control. Of note, repeated freeze thaw of c646 stocks diminished efficacy and higher doses of c646 produced serious developmental delay and growth defects (data not shown).

See supplementary Materials and Methods for details regarding confocal imaging.

Treated Tg(kdrl:GFP)^s843 embryos were processed following the protocol of Inoue and Wittbrodt (2011), with the modifications suggested in Le Guen et al. (2014) (see supplementary Materials and Methods for details).

Tg(TP1bglob:VenusPEST); Tg(kdrl:mCherry) animals at 18-20 somites were mounted in 1% low-melt agarose on a four-compartment glass-bottom cell culture dish (Cellview, #627975), treated with PTU and tricaine in E3, along with the same concentration indicated above for either DMSO (vehicle control) or SL327 (MEK inhibitor) (see supplementary Materials and Methods for details).

Embryos were washed with PBS (without calcium/magnesium) and 1 ml of pre-warmed 0.25% Trypsin was added to the embryos. Embryos were incubated at 28°C and gently pipetted up and down every 5 min until digestion was complete. After digestion, 100 μl of fetal bovine serum (FBS; 100%) was added to stop digestion. The cells were spun at 1100 rpm (300 g) for 5 min at 4°C and the supernatant was removed and the cell pellet was re-suspended in 500 μl of FACS solution (450 μl PBS+50 μl 10% bovine serum albumin). Sytox Red (0.5 μl) was added and the samples were incubated at room temperature for 15 min and then passed through a cell sieve. FACS was performed by the UHN Flow Cytometry Facility using a low differential pressure (20 psi). Cells were sorted directly into RLT buffer (Qiagen) for RNA extraction using the RNeasy Micro Kit (Qiagen).

Experiments were performed as described previously (Wythe et al., 2011). Riboprobes, with the exception of flt4, were amplified by PCR with primers containing SP6 and T7 overhangs and sense and digoxigenin-labeled antisense probes were synthesized from the PCR template. The hlx1 template was provided by Dr Saulius Sumanas (Cincinnati Children's Medical Center, OH, USA), dll4 in pGEM-T was from Dr Jiandong Liu (University of North Carolina, NC, USA), dusp5 was from GE Dharmacon (Clone ID: 4199935), and flt4 was from Dr Jeffrey Essner (Iowa State University, IA, USA) (digested with EcoRI, transcribed with T7). See supplementary Materials and Methods for primers used.

pTol2 enhancer injections were performed as previously published (Wythe et al., 2013). Briefly, an injection mixture consisting of 100 ng of DNA, 125 ng of Tol2 transposase mRNA (Kawakami et al., 2004), 1 µl 0.8% Phenol Red/0.1 M KCl, pH 7.0, and ddH₂O in 10 µl total volume was combined and 1 nl injected directly into the cell of one-cell stage Tg(kdrl:mCherry)^ci5 zebrafish embryos. Embryos were then maintained at 28.5°C and scored at 42 hpf for enhancer activity within and outside of the vasculature (mCherry⁺).

Sense-strand-capped mRNA was transcribed using the mMessage mMachine kit (Ambion) from either a pCS2-3xFlag-ERG or pCS2-3xFlag-ERG-3xS→A template (see ‘Cloning’ in supplementary Materials and Methods for details). mRNA (5 µl of 125 ng/µl) was mixed with 1 µl of 2 mg/ml Alexa 647 10,000 MW, Anionic, Fixable Dextran (Life Technologies, D22914) as a lineage tracer, then 1 nl of this mixture was injected into one-cell stage Tg(fli1a:nls-EGFP)^y7 donor embryos. Dividers were pulled 1-2 h later for Tg(kdrl:mCherry)^ci5 host embryos. Animals were dechorionated on agarose dishes using pronase in E2 media supplemented with penicillin and streptomycin, as suggested by Westerfield (2007). When donor embryos reached approximately sphere stage to 30% epiboly (about 4 hpf), they were transferred to agarose wells (Adaptive Science Tools, PT-1) in E2 plus antibiotics, and 20-40 cells from the lateral margin were transferred from donor to host embryos (at a similar location). Embryos were reared at 28.5°C in 1.5% agarose dishes in E2 supplemented with antibiotics until their fixation in 4% paraformaldehyde (PFA) at 28-30 hpf. See supplementary Materials and Methods for details on imaging. The percentage contribution of donor-derived cells contributing to a location in the host trunk vasculature was quantified as the number of donor-derived cells within the structure divided by the total number of donor-derived cells within the entire vasculature in the region of interest.

Generation of Erg knockout mice was carried out by the KOMP consortium (project ID: 48771; Erg^{tm1a(KOMP)Wtsi}). Cryopreserved sperm were received and in vitro fertilization was performed at the Genetically Engineered Mouse (GEM) core at Baylor College of Medicine. In this Erg^tm1a allele, insertion of a splice acceptor-IRES-lacZ-stop, human beta actin promoter driving Neomycin between exon 5 and 6 acts as a null mutation (i.e. a gene trap) (Fig. S7A). For these studies, a global Erg null allele (Erg^lacZΔNEO/+, Erg^KO/+ or Erg^tm1b) was generated by Cre-mediated removal of the hBact::Neo cassette by crossing Erg^tm1a/+ male mice to females harboring a Tg(ACTB::Cre) (MGI#2176050) (Lewandoski et al., 1997) driver (see Fig. S7A). To generate a conditional allele (Erg^flox/+ or Erg^tm1c), an Erg^tm1a/+ male was crossed to a Tg(ATCB::FlpE) female (MGI#2448985) (Rodríguez et al., 2000) for Flp-mediated removal of the FRT-flanked promoterless lacZ gene trap. The resulting animals, with two loxP sites flanking exon 6, were incrossed to generate Erg^flox/flox animals for postnatal studies (see below). For further details regarding PCR genotyping, see Table S1.

Timed matings between heterozygous mutant animals (Erg^KO/+) mice were conducted, and noon of the day a vaginal plug was detected was considered day 0.5. Embryos were collected at E10.5, and the yolk sac was used for PCR genotyping of the embryos. Embryos were fixed overnight in 4% PFA then subjected to whole-mount IHC using an endomucin antibody (eBioscience, V.7C7) (1:200); biotinylated goat anti-rat secondary antibody (Vector, BA-9401) (1:250), ABC Elite Kit (Vector Labs, PK-6100) and Alexa 488 Tyramide (Thermo Fisher Scientific, T20948) and imaged by light-sheet microscopy, as previously published (Wang et al., 2016). For frozen sections, embryos were fixed in fresh 4% PFA at 4°C for 20 min, washed in PBS, cryoprotected in 30% sucrose overnight at 4°C, equilibrated in optimal cutting temperature compound (Sakura, #25608) in peel-away molds (Electron Microscopy Sciences, #701081), solidified on dry ice, then stored overnight at −80°C. For additional detail, see supplementary Materials and Methods.

Whole-mount embryos were processed as detailed previously (Wythe et al., 2013). See supplementary Materials and Methods for further details.

Embryos were collected at E10.5 (their yolk sac was removed for PCR genotyping) and dissociated to single cells by a 20-min incubation at 37°C in collagenase type I (1 mg/ml) (ThermoFisher Scientific, #17100017). See supplementary Materials and Methods for details. Next, 25,000 live ECs per embryo were collected into 350 µl RLT Buffer (Qiagen) for downstream processing. Total RNA was isolated from sorted cells using the RNAeasy Mini Kit (Qiagen, #74104). RNA was eluted in 30 µl RNAse-free H₂O, and 20 µl of purified RNA was reverse transcribed to cDNA with 5 µl of SuperScript VILO Master Mix (Thermo Fisher Scientific, #11755050). The resulting cDNA was diluted 1:10 with RNAse-free dH₂O, and 4 µl was used for each qPCR reaction (15 µl total reaction volume). qPCR reactions were performed in technical triplicate using iTaq Universal SYBR Green Supermix (Bio-Rad, #1725124) and run on a ViiA 7 Real-Time PCR System (Applied Biosystems). Relative abundance of mRNA transcripts was calculated by normalization to Gapdh using the ΔΔCt method (Livak and Schmittgen, 2001). All embryonic RNA samples used for analysis had a deviation no greater than two Cts (range 16-18) for Gapdh. For a list of all primers used, see Table S2.

Erg^flox/flox (aka Erg ^tm1c/tm1c) females were crossed to Erg^KO/+ (aka Erg ^tm1a/+); Cdh5(PAC)CreERT2 (Wang et al., 2010) males and 30 µl of tamoxifen was administered by subcutaneous injection at a concentration of 10 mg/ml (≈30 µg total per mouse) to their progeny at postnatal days (P) 1 and 3. See supplementary Materials and Methods for details regarding quantification of vascular branching and radial expansion in the retina.

The following cells were utilized for experiments: human umbilical vein endothelial cells (HUVECs, ScienCell), telomerase-immortalized aortic endothelial cells (TeloHAECs, ATCC), dermal microvascular endothelial cells (MVECs, Life Technologies) and bovine aortic endothelial cells (BAECs, Lonza). Cells were cultured according to manufacturer's recommendations. ECs were serum-starved in basal EC media (ScienCell) containing 0.1% FBS and no growth factors for at least 6 h (typically overnight) prior to stimulation with VEGF or PMA. Pathway inhibitors were added 1 h prior to stimulation with VEGF-165 (50 ng/ml, recombinant human protein, R&D Systems or Thermo Fisher Scientific), except for c646, which was added 20 min prior to stimulation. The following inhibitors were used: U0126 (MEK inhibitor, 20 μM, InvivoGen), GF109203X (PKC inhibitor, 5 μM, Tocris Bioscience), LY294002 (PI3K inhibitor, 10 μM, Cell Signaling), SB203580 (p38 inhibitor, 10 μM, Tocris Bioscience), DAPT (γ-secretase inhibitor, 38.5 μM, Sigma) and c646 (p300/CBP inhibitor, 5 μM, Sigma). All drugs were dissolved in DMSO, and comparison was made with vehicle (i.e. DMSO, 0.1%) treated controls. PMA was from BioShop and was used at a concentration of 100 nM.

HUVECs were transfected at 30-50% confluency with 40 nM siRNA (Silencer Select, Thermo) targeting the coding region of ERG (assay ID: s4813), ERK1 (MAPK3) (assay ID: S230180) or ERK2 (MAPK1) (assay ID: S11138) using RNAiMax (Invitrogen), and cellular assays were performed 48-72 h later. Western blotting and qRT-PCR were used to assess ERG, ERK1 and ERK2 knockdown. Comparison was made with cells transfected with 40 nM Silencer Select negative control #1. For ERG rescue experiments, an independent siRNA recognizing the 3′ untranslated region of ERG (40 nM, Silencer Select, custom synthesis) was utilized. After 48-72 h, cells were electroporated using the P5 Primary Cell 4D Nucleofector kit and a Lonza 4D Nucleofector; ∼0.5×10⁶ cells were electroporated with 2.5 μg of pCS2 control or pCS2-Flag-ERG expression constructs (wild type and phospho-mutants; see ‘Cloning’ in supplementary Materials and Methods), and 0.2 μg of pmaxGFP (to assess electroporation efficiency). After 18 h, cells were serum-starved for 6 h prior to VEGF stimulation (1 h) and harvested for RNA/protein analyses.

For details regarding cloning of ETS concatamer reporter constructs, HLX enhancer reporter constructs and wild-type and mutant ERG expression constructs, please see the supplementary Materials and Methods.

For luciferase assays, BAECs (80% confluent) were transfected with 0.5 µg of luciferase construct (ETS reporter, HLX enhancer, see ‘Cloning’ in supplementary Materials and Methods) and 0.1 µg of pRenilla construct using Lipofectamine 2000 (Invitrogen) (2 µl per 1 µg of plasmid) in 12-well dishes. Cells were treated with cellular signaling pathway inhibitors for 18 h (as above). In some experiments, VEGF (50 ng/ml) was added to OptiMEM medium for 18 h. After 24 h, dual luciferase (Renilla and firefly) was measured using a GloMax20/20 Luminometer (Promega).

Western blotting was performed as before (Fish et al., 2011) using the following antibodies: anti-pERK1/2 (Thr202/Tyr204, rabbit polyclonal, Cell Signaling, #9101; 1:1000), anti-ERK2 (mouse monoclonal, Santa Cruz, D-2; 1:500), anti-ERK1/2 (rabbit monoclonal, Cell Signaling, clone 137F5; 1:500), anti-DLL4 (rabbit polyclonal, Cell Signaling, #2589; 1:1000), anti-GAPDH (mouse monoclonal, Santa Cruz, #0411; 1:500), anti-ERG (rabbit polyclonal, Santa Cruz, C-20 or mouse monoclonal antibody, BioCare Medical, 9FY; 1:1000), anti-pERG^S215 [rabbit polyclonal, a kind gift from Peter Hollenhorst, Indiana University, IN, USA (Selvaraj et al., 2015); 1:500], anti-p300 (rabbit polyclonal, Santa Cruz, C-20; 1:200). All antibodies have been previously validated. See supplementary Materials and Methods for details regarding pERG western blots and co-immunoprecipitation experiments.

HUVECs were plated on Permanox eight-well chamber slides. Following stimulation with VEGF, cells were fixed with 4% PFA followed by permeabilization with 0.25% Triton X-100. Staining with anti-pERK (rabbit polyclonal, Cell Signaling, #9101, 1:500) was performed overnight at 4°C, followed by addition of a secondary antibody (anti-rabbit Alexa Fluor647, Cell Signaling #4414). Slides were mounted using Vectashield mounting medium with DAPI (Vector Labs H-1200) and imaged using an Olympus FV1000 confocal microscope.

RNA was isolated from cells and zebrafish using Trizol and reverse transcription was performed using a high-capacity cDNA reverse transcription kit (Applied Biosystems). qRT-PCR was performed using a Roche Lightcycler 480 with LC 480 SYBR Green I Master Mix (Roche). Data were normalized to TATA-box binding protein (TBP) or Gapdh using the ΔΔCt method. For further details regarding primer sequences, see Table S2.

HUVECs were transfected with control or ERG siRNA and after 48 h the cells were serum-starved overnight and cells were then left unstimulated or were treated with 50 ng/ml VEGF for 1 h. RNA was isolated from four independent experiments using Trizol and analyzed on Agilent microarray, performed at the Princess Margaret Genomics Centre. See supplementary Materials and Methods for details regarding microarray processing and analysis.

Differentially expressed genes were submitted to the Database for Annotation, Visualization and Integrated Discovery (DAVID) bioinformatics resource (https://david-d.ncifcrf.gov/home.jsp) to be classified into gene ontology (GO) annotation groups (Ashburner et al., 2000; Huang et al., 2009). Fisher's exact test was applied to identify significant GO categories. Select representative GO categories are included in figures.

ChIP was performed as before (Wythe et al., 2013), using the Imprint ChIP kit (Sigma) or the Magna ChIP A/G kit (Millipore). HUVECs were serum-starved overnight prior to stimulation with 50 ng/ml VEGF for 15′, 30′ or 2 h. Following fixation and shearing of chromatin, immunoprecipitation was performed overnight at 4°C using 1 μg (or 1 μl) of antibodies to ERG (rabbit polyclonal, Santa Cruz, C-20), p300 (rabbit polyclonal, Santa Cruz, C-20) or H3K27ac (rabbit polyclonal, Abcam, ab4729). Mouse IgG (Sigma) was used as a non-specific background control. qPCR was performed using primers that amplified the DLL4 intron 3 enhancer or an enhancer upstream of HLX (see Table S2 for primer sequences). IP DNA was calculated by subtracting the IgG value from the specific antibody value and dividing by a diluted input sample. In some experiments, control or ERG siRNAs were transfected into HUVECs 48 h prior to stimulation with VEGF (15′), followed by p300 ChIP.

Primary HUVECs and BAEC cells were grown in supplier-recommended EC Growth Media (ScienCell) and cultured at 37°C in a 5%-CO₂ humidified incubator. Approximately 20 million cells were used for the ERG and ∼3 million cells for the H3K27ac ChIPs. ChIP experiments were conducted as previously described (Ballester et al., 2014). Antibodies used for ChIP were mouse anti-H3K27ac (Millipore, 05-1334 monoclonal) and rabbit anti-ERG 1/2/3 (Santa Cruz Biotechnology, sc353 polyclonal). Two replicates were performed. See supplementary Materials and Methods for details regarding processing and analysis of ChIP-seq experiments.

The HLX enhancer region to be targeted for deletion was defined by H3K27ac ChIP-seq enhancer marks (∼1200 bp). The MIT CRISPR/Cas9 design tool (http://crispr.mit.edu/) was used to generate gRNAs targeting the 5′ and 3′ boundaries of the HLX enhancer. Two scrambled sequence gRNAs were used as controls. For further details, see the supplementary Materials and Methods.

Unless otherwise stated, all experiments were performed a minimum of three times and data represent the mean±s.e.m. Statistical analyses were performed using a Student's t-test (for two groups) or ANOVA (for more than two groups), followed by the Newman–Keuls post-hoc test. P<0.05 was considered statistically significant. In all figures, *P<0.05, **P<0.01 and ***P<0.001.

The authors thank the Optical Imaging and Vital Microscopy (OIVM) Core at BCM for support with imaging and data processing; Dr Neil I. Bower (The University of Queensland, Dr Benjamin M. Hogan laboratory) for advice and suggestions regarding pERK staining; Dr Peter Hollenhorst for providing an antibody to phosphorylated ERG and for advice on ERG immunoprecipitation; Dr Jiandong Liu and Ms Nicole Fleming (UNC Chapel Hill) for suggestions on transplantation experiments; and Mrs Karen Berman de Ruiz for excellent mouse husbandry care and colony management.