Endorepellin evokes an angiostatic stress signaling cascade in endothelial cells

Endorepellin, the C-terminal fragment of the heparan sulfate proteoglycan perlecan, influences various signaling pathways in endothelial cells by binding to VEGFR2. In this study, we discovered that soluble endorepellin activates the canonical stress signaling pathway consisting of PERK, eIF2α, ATF4, and GADD45α. Specifically, endorepellin evoked transient activation of VEGFR2, which, in turn, phosphorylated PERK at Thr980. Subsequently, PERK phosphorylated eIF2α at Ser51, upregulating its downstream effector proteins ATF4 and GADD45α. RNAi-mediated knockdown of PERK or eIF2α abrogated the endorepellin-mediated up-regulation of GADD45α, the ultimate effector protein of this stress signaling cascade. To functionally validate these findings, we utilized an ex vivo model of angiogenesis. Exposure of the aortic rings embedded in 3D fibrillar collagen to recombinant endorepellin for 2–4 h activated PERK and induced GADD45α vis à vis vehicle-treated counterparts. Similar effects were obtained with the established cellular stress inducer tunicamycin. Notably, chronic exposure of aortic rings to endorepellin for 7–9 days markedly suppressed vessel sprouting, an angiostatic effect that was rescued by blocking PERK kinase activity. Our findings unravel a mechanism by which an extracellular matrix protein evokes stress signaling in endothelial cells, which leads to angiostasis.

Perlecan, one of the largest heparan sulfate proteoglycans of basement membranes and cell surfaces, plays a major role in vasculogenesis and angiogenesis (1). For instance, morpholinobased knockdown of perlecan in zebrafish causes profound disruption of all vessels formed via angiogenesis from the dorsal aorta and other major blood vessels (2).
In mice, the effect of global perlecan knockdown is even more pronounced, as perlecan-null mice (Hspg2 Ϫ/Ϫ ) are embryonic lethal because of intrapericardial hemorrhage and malformation of coronary arteries and cardiac outflow tracts (3). Perlecan expression is finely regulated during cardiovascular development (4,5) and is one of the few extracellular matrix constituents expressed in vascular and avascular matrices.
Perlecan is a modular proteoglycan composed of five domains that are involved in growth factor regulation, cancer, inflammation, lipid metabolism, basement membrane assembly, cell adhesion, and mechanosensing, among many other reported functions (6 -16). The parent protein perlecan is proangiogenic because of the presence of three heparan sulfate chains covalently attached at its N terminus, which act as a reservoir of growth factors. In contrast, the C-terminal module of perlecan, called endorepellin (17,18), is anti-angiogenic and inhibits cancer progression by interfering with the blood supply of the growing tumors (19 -21).
Endorepellin interacts with VEGFR2 via its two proximal laminin-like globular (LG1/2) 3 domains and to the ␣2␤1 integrin via its terminal LG3 module, evoking dual receptor antagonism (22,23). Indeed, soluble endorepellin inhibits angiogenesis by interfering with several VEGFR2-evoked signaling pathways, primarily by evoking transcriptional repression of hypoxia-inducible factor 1 (HIF-1␣) and vascular endothelial growth factor A (VEGFA) and concurrently inhibiting nuclear factor of activated T cells (NFAT1) (24). We recently discovered a link between endorepellin-induced inhibition of angiogenesis and autophagy in endothelial cells (20). In our quest to find novel autophagic targets for endorepellin through NanoString analysis (25), we discovered that the mRNA of growth arrest and DNA damage-inducible (GADD45␣) protein was dynamically regulated by the endorepellin/VEGFR2 axis. This protein has been reported to have a dual role: one in activating autophagy and the other in inhibiting angiogenesis (25). As an angiogenic suppressor, GADD45␣ interacts with mTOR and prevents Stat3 phosphorylation from driving VEGFA expression, inhibiting angiogenesis (26). As an autophagy inducer, GADD45␣ inhibits mTOR and induces the levels of LC3-II, an autophagic marker (27).
GADD45␣ belongs to the canonical PKR-like endoplasmic reticulum kinase (PERK)/eIF2␣/activating transcription factor 4 (ATF4)/GADD45␣ axis, which is usually activated by external stresses such as exposure to UV light, nutrient deprivation, hypoxia, or oxidative stress (28,29). Notably, ATF4 promotes skeletal muscle atrophy via a unique interaction with the transcriptional regulator C/EBP␤ (30). All of these external stimuli are known to trigger the unfolded protein response, which ultimately leads to stress activation in cells (31,32). The stress pathway was originally considered to help cells cope with stress stimuli by promoting survival. However, under prolonged stress, the survival mechanism can switch to cellular elimination via autophagic activation, which can eventually lead to cell death. For instance, in glioblastoma cells, eIF2␣ phosphorylation can up-regulate ATF4 levels, which induce autophagy by mTORC1 inhibition, causing cell death (33). Alternatively, ATF4 can lead to accumulation of transcription factor CCAATenhancer-binding protein homologous protein and apoptotic induction by GADD34 activation in endothelial cells (34). Protein misfolding was originally understood to be essential for activating intracellular stress; however, more recent discoveries have shown that ligand-receptor interactions (i.e. estrogen receptor 1 with 3,3-bis(4-hydroxyphenyl)-7-methyl-1,3-dihydro-2H-indol-2-one) also activate stress and cellular necrosis (35). Together, these reports suggest that the PERK/eIF2␣/ ATF4/GADD45␣ stress axis can be activated in cells without accumulation of misfolded proteins.
Given that endorepellin and GADD45␣ concurrently evoke autophagy and angiostasis, we hypothesized that the biological repertoire of endorepellin could include activation of a stress signaling pathway in vascular endothelial cells. We discovered that endorepellin evoked profound and sustained stress signaling in in vitro and ex vivo models of angiogenesis primarily by interacting with VEGFR2. Our studies highlight the role of endorepellin in inducing the stress pathway. This, in turn, would significantly curtail the generation of new vessels in the tumor matrix with clear implications for impeding cancer progression.

Endorepellin induces GADD45␣ by evoking a stress signaling axis
To dissect the mechanism involved in endorepellin-evoked anti-angiogenic activity, we investigated GADD45␣, a gene that is highly modulated in a NanoString analysis of endothelial cells exposed to recombinant endorepellin (25). The rationale is based on the fact that GADD45␣ increases in response to various stressors (28,36) and suppresses tumor angiogenesis by blocking the mTOR pathway (26) while concurrently inducing LC3 (37). Moreover, the GADD45␣-mTOR interaction prevents Stat3 phosphorylation from driving VEGFA expression, inhibiting angiogenesis (26). We performed time course experiments using the same concentrations of endorepellin (200 nM) under nutrient-rich conditions as in the NanoString studies (25). We discovered that endorepellin evoked a rapid and sustained increase in GADD45␣ (Fig. 1, A and B). As GADD45␣ is known to be induced by ATF4 (38), we tested ATF4 levels in parallel experiments. We found a significant induction of ATF4 that peaked at 3 h and declined thereafter (Fig. 1, A and B).
ATF4 is known to be regulated by phosphorylation of eIF2␣, the ␣ subunit of eukaryotic initiation factor 2 (39). Phosphoryl-ation of eIF2␣ at Ser 51 results in global protein synthesis suppression and plays a key role in response to various stresses, such as viral infection, buildup of misfolded proteins, and starvation. Paradoxically, phosphorylation of eIF2␣ at Ser 51 induces transcription and translation of ATF4 (40). There are only four eIF2␣ mammalian kinases (PERK, heme-regulated inhibitor, PKR, and GCN2) (29,39,41), all of which phosphorylate eIF2␣ at Ser 51 , leading to similar downstream effects. In initial screenings, we found that human umbilical vein endothelial cells (HUVECs) predominantly express PERK, PKR, and GCN2 (Fig. S1A). Thus, we tested PERK and found time-dependent phosphorylation of PERK at Thr 980 within its catalytic site (42), evoked by endorepellin (Fig. 1, C and D). Moreover, we found sustained activation of eIF2␣ at Ser 51 ( Fig. 1, C and D). We note that activation of PERK occurred rapidly, peaking at 1 h and preceding that of eIF2␣ (Fig. 1, C and D). We also saw temporal phosphorylation of PERK by LG1/2 domains of endorepellin, but not LG3, mirroring the effect of endorepellin (Fig. S1, C-F). As an internal positive control, we utilized tunicamycin, an antibiotic that inhibits N-linked glycosylation and induces the unfolded protein response (43) (Fig. S1B).
To address the mechanism of stress signaling induction by exogenous endorepellin, we blocked the VEGFR2 kinase with the small-molecule inhibitor SU5416 (44,45) and found complete suppression of endorepellin-evoked PERK (Fig. 1E). Thus, it appears that VEGFR2 is directly involved in endorepellin biological activity, in agreement with our previous studies (20,24,45).
Next we performed imaging and cell fractionation to ascertain whether GADD45␣ and its activator ATF4 would translocate into the nuclei of stressed endothelial cells, as shown before in other cells (46,47). We discovered that endorepellin evoked marked nuclear translocation of GADD45␣ at 4 h, similar to that induced by tunicamycin (Fig. 1, F and G). Cell fractionation studies followed by endorepellin treatment for 4 h showed that the trimeric form of GADD45␣ (68 -72 kDa doublet, Fig. 1H) that was mainly observed in the nucleus under basal conditions, as seen in other studies (46), was markedly increased by endorepellin (Fig. 1H). In contrast, the monomeric form of GADD45␣ (ϳ22 kDa) did not change appreciably (Fig.  1H). In support of these findings, ATF4 levels and nuclear translocation were concurrently increased in HUVECs at 4 h ( Fig. 1, I and J). These findings were further validated in two other endothelial cell models, telomerase human aortic endothelial cells (TeloHAECs) and porcine aortic endothelial cells expressing VEGFR2 (PAER2s). Notably, we found even faster induction of PERK phosphorylation, with maximal activation at 30 min in TeloHAECs (Fig. S2, A-C) compared with HUVECs, where p-PERK peaked at 1 h (Fig. 1C). In PAER2s, the kinetics of PERK phosphorylation at Thr 980 were similar to that of HUVECs (Fig. S2, E-G). These biochemical data were corroborated by immunofluorescence studies, which showed up-regulation of p-PERK and GADD45␣ in TeloHAECs and PAER2s together with nuclear translocation of GADD45␣ (Fig. S2, C and G).
Collectively, our results provide the first evidence that endorepellin activates the PERK/eIF2␣/ATF4/GADD45␣ stress signaling pathway. This could be a dynamic nexus for endorepellin bioactivity connecting stress signaling and angiostasis.

Endorepellin induces an anti-angiogenic stress pathway PERK and eIF2␣ are required for endorepellin-dependent stress axis induction
To confirm the biological relevance of PERK and eIF2␣ in endorepellin-mediated activation of stress signaling, we transiently depleted these two proteins via RNAi in HUVECs. We observed that PERK depletion (Ͼ80% knockdown) led to significant abrogation of eIF2␣ phosphorylation at Ser 51 induced by endorepellin or tunicamycin (Fig. 2, A and B). Notably, even basal eIF2␣ phosphorylation at Ser 51 was suppressed by PERK depletion (Fig. 2, A and B), further indicating a primary role of PERK as the key kinase regulating eIF2␣ phosphorylation in vascular endothelial cells.
Next we found that, when eIF2␣ was depleted (Ͼ90% knockdown), there was marked suppression of GADD45␣ under , ranging from 0 -6 h) and probed for ATF4 and GADD45␣. Data were obtained from endorepellin-treated HUVECs from three independent experiments (n ϭ 3) and normalized to GAPDH. B, quantification of ATF4 and GADD45␣ from A. C, representative immunoblots of time course experiments of HUVECs treated with endorepellin (200 nM, 0 -6 h) and probed for p-PERK Thr980 , total PERK, p-eIF2␣ Ser51 , and total eIF2␣. Data were acquired from three independent experiments (n ϭ 3) and normalized to their respective total protein levels. D, quantification of p-PERK Thr980 and p-eIF2␣ Ser51 from C. One-way ANOVA was performed on all data. E, immunoblots of time course experiments of HUVECs pretreated for 30 min with SU5416 (a VEGR2 kinase inhibitor, 30 M), followed by addition of endorepellin ranging from 0 -24 h. The last lane shows the positive control, i.e. endorepellin alone for 2 h. ␣-Tubulin served as the loading control. F, immunofluorescence images of HUVECs showing cytoplasmic or nuclear distribution of GADD45␣ (red) seen with respect to DAPI (blue) after treatment with vehicle (PBS), endorepellin (200 nM), or tunicamycin (10 g/ml) for 4 h. Nuclear localization of GADD45␣ is seen as a magenta hue under endorepellin or tunicamycin treatment conditions because of the merging of red and blue tones in the nucleus. G, quantification of cells with nuclear localization of GADD45␣ from F. An average of 100 cells per treatment (vehicle, endorepellin, or tunicamycin) were analyzed from five independent experiments. H, representative immunoblots of cell fractionation experiments of HUVECs treated for 3 h with or without endorepellin (200 nM). The membranes were probed for lamin A/C to label nuclear fractions (Nu), GAPDH to label cytoplasmic fractions (Cy), and GADD45␣. Only the nuclear fraction was enriched in the trimeric form of GADD45␣, and these levels were increased by exposure to endorepellin. I, immunofluorescence images of HUVECs treated with vehicle, endorepellin, or tunicamycin for 4 h and probed for filamentous actin (red) and ATF4 (green). The nuclei are outlined by dotted lines. J, quantification of cells with nuclear localization of ATF4 from I. An average of 500 cells per treatment (vehicle, endorepellin, or tunicamycin) were analyzed from three independent experiments (n ϭ 3). All statistical analyses were calculated via one-way ANOVA (***, p Ͻ 0.001).

Endorepellin induces an anti-angiogenic stress pathway
basal conditions as well as endorepellin or tunicamycin treatments ( Fig. 2, C and D). Collectively, our results underscore the requirement of PERK as the major activator of the endorepellin-evoked stress axis in endothelial cells and provide novel mechanistic insights into sequential activation of the PERK/ eIF2␣/GADD45␣ axis by endorepellin.

Endorepellin activates the stress pathway in ex vivo angiogenic assays
To directly test whether the anti-angiogenic activity of endorepellin was associated with induction of stress signaling, we performed ex vivo aortic ring assays in 3D type I fibrillar collagen (48). Fully sprouted explanted aortic rings of C57BL/6J mice were subjected to acute treatment with endorepellin or tunicamycin using the same in vitro concentrations. For p-PERK induction, we used 2-h treatment, the time point at which we observed maximal PERK phosphorylation (cf. Fig.  1C). For confocal imaging, we utilized isolectin IB4, which specifically binds to galactosyl residues on endothelial cells (49,50), and antibodies against p-PERK at Thr 980 , considered a marker for its activation status (51)(52)(53). We found a marked increase in p-PERK evoked by endorepellin or tunicamycin (Fig. 3A). These effects could be clearly seen using the line-scanning tool, as described under "Experimental procedures" (Fig. 3B). Quantification of four independent experiments using three randomly selected aortic rings per mouse showed significant induction of p-PERK, measured as fluorescence intensity of the sprouted area normalized on the area occupied by the vascular sprouts (p Ͻ 0.001, Fig. 3C and Fig. S3B). We confirmed the endothelial nature of the sprouts by staining with a second endothelial cell marker, CD31 (red), and counterstaining with P-PERK (green) (Fig. S3D). We obtained similar results as those shown for IB4 and p-PERK staining (Fig. 3A).
To validate our confocal imaging studies, we next performed biochemical analysis of rings treated with endorepellin or tunicamycin for 2 h. We first pooled two aortic rings per sample and removed the rings after solubilization with radioimmune precipitation assay buffer. Thus, the protein extracts were composed exclusively of sprouted endothelial cells growing in 3D collagen. In agreement with the confocal imaging studies, we found marked induction of p-PERK at Thr 980 evoked by either treatment (Fig. 3, D and E).
Next we probed for GADD45␣, the ultimate effector protein of the stress axis. Interestingly, we observed that GADD45␣ was not appreciably altered at 2 h (not shown) but markedly

Endorepellin induces an anti-angiogenic stress pathway
increased at 4 h of treatment with endorepellin or tunicamycin (Fig. 4A). Quantification of four independent experiments using three randomly selected aortic rings per mouse showed significant induction of GADD45␣ (p Ͻ 0.001, Fig. 4B), as determined by the fluorescence intensity of the sprouted area, measured as described under "Experimental procedures" (Fig.  S3B). In agreement with the confocal imaging, we found by immunoblotting that endorepellin or tunicamycin markedly induced GADD45␣ (Fig. 4, C and D). Collectively, these ex vivo functional assays corroborate the in vitro results and show progressive activation of the stress signaling pathway, where PERK is first phosphorylated at Thr 980 , followed by an increase in GADD45␣ levels.

The endorepellin-activated stress axis inhibits angiogenesis
To further investigate the specificity of endorepellin-evoked PERK activation, we utilized a small-molecule inhibitor, GSK2656157, a cell-permeable, ATP-competitive inhibitor of PERK (PERKi) (54). This small molecule is exquisitely specific for PERK with more than 100-fold selectivity for PERK over other eIF2␣ kinases (55) and an IC 50 of ϳ1 nM in various cells, as shown by inhibition of PERK autophosphorylation, eIF2␣ phosphorylation, and ATF4 suppression (55). Oral administration of this PERKi results in tumor suppression and angiostasis (55), consistent with reduced tumor growth in Perk Ϫ/Ϫ mice (56). We found that the PERKi efficiently blocked the activity of endorepellin in a dose-dependent manner (Fig. 5A) with an A, representative confocal images of aortic rings following treatment with endorepellin (200 nM) or tunicamycin (10 g/ml) for 2 h. Endothelial cells in sprouts were visualized with isolectin IB4 (red), a marker for endothelial cells, and counterstained with an antibody against P-PERK Thr980 (green). Insets show magnified sprouts stained for p-PERK Thr980 . B, line intensity profiles of PERK (green) expression in aortic ring sprouts corresponding to the yellow line traced along the enlarged image of sprouts in the inset in A. C, quantification of the fluorescence intensity of PERK in the sprouted area from A; fluorescence intensity data are from a total of 12 vehicle (Veh)-, nine endorepellin (Endo)-, and 10 tunicamycin (Tunic)-treated aortic rings from four independent experiments (n ϭ 4). D, representative immunoblots of three pooled, sprouted blood vessels probed for p-PERK Thr980 and total PERK. E, quantification of p-PERK Thr980 levels over total PERK from D. Three aortic rings were pooled per condition (vehicle, endorepellin, or tunicamycin) and repeated four times; therefore, four independent experiments were performed (n ϭ 4). The data are derived only from newly formed sprouts, as the aortic rings were removed before solubilization. All statistical analyses were calculated via one-way ANOVA (***, p Ͻ 0.001).

Endorepellin induces an anti-angiogenic stress pathway
IC 50 of ϳ10 nM (Fig. 5B). Notably, we observed a parallel reduction in phosphorylation of downstream eIF2␣ at Ser 51 ( Fig. 5A) with an IC 50 of ϳ1 nM (Fig. 5B). We also confirmed that GSK2656157 was 1000-fold more potent in inhibiting PERK phosphorylation in HUVECs with an IC 50 of ϳ1 nM (Fig. 5A) compared with another PERK inhibitor, AMG44, which was discovered recently and has an IC 50 of ϳ1000 nM in HUVECs (Fig. S3C).
Next we performed functional aortic ring assays in 3D type I collagen using the PERKi with or without endorepellin. In these experiments, we added the PERKi or endorepellin after 3 days, when we detected the first sprouts. We added fresh PERKi (10 nM) or endorepellin (200 nM) every other day and continued the treatment for 9 days in total. At the end of the chronic treatment, we observed a drastic abrogation of ring sprouting in endorepellin-treated rings (Fig. 5C) vis à vis vehicle (p Ͻ 0.001, Fig. 5D). However, the PERKi completely blocked the angiostatic activity of endorepellin (Fig. 5, C and D). We acquired phase-contrast images of the rings and calculated the radial distance of the sprouts by subtracting the background and highlighting sprouts using the threshold function in ImageJ. We then drew circles around the rings, encompassing the edges of the sprouts, and measured the radii (Fig. S3A). Next we labeled the rings with IB4 and DAPI, and, using confocal laser scanning microscopy, we obtained comparable qualitative (Fig.  5E) and quantitative (Fig. 5F) findings. Collectively, our results suggest that endorepellin-dependent inhibition of angiogenesis requires activation of the PERK-stress axis.

Discussion
Classical pathways evoked by protracted activation of the unfolded protein response and ER stress are known to interfere with angiogenesis (57). Specifically, the PERK/eIF2␣/ATF4/ GADD45␣ pathway, a major ER stress signaling axis, can generate angiomodulatory and angiostatic cues (57). Although this axis is usually activated under stressful stimuli, such as osmotic/ oxidative stress, irradiation, or nutrient deprivation, it may also be activated in the absence of external stresses (58). In this study, we present endorepellin-mediated activation of this stress axis through ligand-receptor interaction in the absence C, representative immunoblots of three pooled, sprouted blood vessels probed for GADD45␣. GAPDH was used as a loading control. D, quantification of GADD45␣ levels normalized to GAPDH from C. Data are from four independent biological experiments (n ϭ 4), with each condition comprised of three pooled sprouts. All statistical analyses were calculated via one-way ANOVA (***, p Ͻ 0.001).

Endorepellin induces an anti-angiogenic stress pathway
of external stress. We provide a working model summarizing our current findings regarding the mechanism of action for endorepellin-evoked endothelial cell stress (Fig. 6). According to this model, endorepellin engages the ectodomain of VEGFR2 via LG1/2 globular domains, acting as a partial agonist and evoking phosphorylation of the receptor tyrosine kinase Tyr 1175 , a key residue from which the canonical phospholipase C-␥ cascade originates (59). Phosphorylation of VEGFR2 at Tyr 1175 is necessary for stress activation because downstream signaling is completely abrogated by inhibiting VEGFR2 kinase activity with SU5416 (44,45). The signal emanating from VEGFR2 causes activation of the stress-sensing kinase PERK at Thr 980 . PERK activation subsequently leads to eIF2␣ phosphorylation at Ser 51 , enhanced translation of ATF4, and up-regulation of GADD45␣ (Fig. 6). ATF4 and GADD45␣ translocate into the nucleus, where, presumably, a number of anti-angio-genic and pro-autophagic genes are induced. Ultimately, endorepellin-mediated activation of the stress axis causes inhibition of angiogenesis.
The path to this study was paved by identification of GADD45␣ as one the targets of endorepellin in endothelial cells, as discovered by NanoString analysis (25). As GADD45␣ is a part of the stress axis and is modulated by endorepellin in endothelial cells, we hypothesized that GADD45␣ may affect angiogenesis. We found that GADD45␣ up-regulation inhibits angiogenesis. These data are in congruence with a previous study that showed that GADD45␣ inhibits tumor angiogenesis in mice injected with transformed mouse embryonic fibroblasts (26). Interestingly, we saw up-regulation of GADD45␣ and upstream stress axis proteins upon endorepellin treatment but a down-regulation of GADD45␣ mRNA evoked by endorepellin. A plausible cause of this observation may be mRNA degra- , plotted as mean Ϯ S.E. One-way ANOVA was performed on the data. C, representative phase-contrast images of aortic rings treated with endorepellin or endorepellin and the PERKi for 7 days in total after sprouting. D, quantification of the radial distance of endorepellin-or endorepellin and PERKi-treated sprouts from C. Phase-contrast data are from 12 vehicle-, 10 endorepellin-, and 10 endorepellin and PERKi-treated rings from four independent experiments (n ϭ 4). E, representative confocal images of aortic rings treated with endorepellin or endorepellin and PERKi-treated rings. Rings were labeled with IB4 (red) and DAPI (blue). F, quantification of fluorescence intensity of the sprouted area from E. Confocal data are from eight vehicle-, eight endorepellin-, and nine endorepellin and PERKi-treated rings from four independent biological experiments (n ϭ 4). All statistical analyses were calculated via one-way ANOVA (***, p Ͻ 0.001).

Endorepellin induces an anti-angiogenic stress pathway
dation by microRNA, which may act on surplus GADD45␣ being formed. Notably, a GADD45␣-repressing microRNA, miR-362-5p, promotes malignancy of chronic myelocytic leukemia cells (60). Another possibility is that there is simultaneous stabilization of GADD45␣, which usually requires an accessory protein, such as S7 ribosomal protein, to prevent its ubiquitination and degradation (61). In future studies, it would be interesting to unravel how the endorepellin-VEGFR2 interaction leads to stabilization of GADD45␣ in endothelial cells; this could help us to better understand vascular stress.
A critical observation in this study was sustained nuclear translocation of GADD45␣ evoked by soluble endorepellin. Previous studies have shown that nuclear GADD45␣ interacts with proliferating cell nuclear antigen, protein 21 (p21 Waf1 ), and cell division control 2 (Cdc2), playing a role in DNA repair as a checkpoint protein to prevent a severely damaged cell from proliferating (28). More recently, a novel role of GADD45␣ has been reported in causing DNA demethylation (62). Specifically, GADD45␣ recruits the demethylase TET1 at the promoter of the powerful transcription factor TCF21 (62). This bioactivity provides an explanation for the mechanism of epigenetic regulation by GADD45␣. We hypothesize that GADD45␣ may play a direct role in demethylating anti-angiogenic and pro-autophagic gene promoters. Especially important are genes that are known to be controlled by differential methylation, includ-ing paternally expressed gene 3 (Peg3) which is differentially methylated in paternal and maternal alleles (63). Notably, we discovered that Peg3 is a master regulator of autophagy (64) and is markedly induced at the mRNA and protein levels in endothelial cells exposed to endorepellin (45). Moreover, we found that overexpression of Peg3 concurrently evokes transcriptional up-regulation of pro-autophagic Beclin 1 and anti-angiogenic thrombospondin 1 (65).
To unravel the mechanism of action of endorepellin-induced activation of the canonical PERK/eIF2␣/ATF4/GADD45␣ stress axis, we performed inhibitory studies of PERK, the most upstream molecule of the stress axis. We used two independent approaches to inhibit PERK action: we knocked down total PERK using RNAi, and we blocked PERK phosphorylation with a specific-kinase small-molecule inhibitor. Notably, genetic and pharmacological inhibitors of PERK markedly suppressed the downstream stress signaling pathway. These results were further validated by knockdown of eIF2␣, which also efficiently blocked the downstream effectors of this stress axis. Use of PERK inhibitors required careful selection of the most efficient inhibitory molecule from the cohort of available drugs. Of the two most recently reported PERK inhibitors, GSK2656157 and AMG44, we selected GSK2656157 (55,66). Although GSK2656157 is the most extensively studied PERK inhibitor and tumor repressor (55), with an IC 50 of ϳ10 nM (55) and kinase inhibition, AMG44 has a higher specificity for PERK. Unfortunately, the IC 50 for AMG44 is ϳ84 nM, which is eight times that of GSK2656157, making GSK2656157 the more effective inhibitor (66). To confirm these findings, using both inhibitors in dose-response studies in HUVECs revealed a greater difference in their efficiencies, with GSK2656157 having a 1000-fold higher capability of inhibiting PERK phosphorylation (IC 50 , ϳ1 nM) compared with AMG44 (IC 50 , ϳ1000 nM). Because the advantages of GSK2656157 outweighed that of AMG44, we used GSK2656157 to block PERK phosphorylation and study its impact on angiogenesis. Overall, our results support a high degree of specificity for GSK2656157 because we had the same effects by suppressing PERK with a pool of three to five PERK-specific siRNAs.
The mode of activation of PERK determines its ultimate mechanism of action. For instance, VEGFA-based PERK activation leads to induction of angiogenesis (58), whereas endorepellin-based PERK induction causes angiostasis. This opposing effect could in part be attributed to allosteric inhibition of VEGFA by endorepellin (20). This response could also be due to differential binding of VEGFA and endorepellin to VEGFR2. Indeed, VEGFA binds VEGFR2 at position 2-3 of the immunoglobulin-like sub-domains' on the extracellular domain, whereas endorepellin binds at position 3-4 of the immunonoglobulin-like sub-domains (20,22). This raises the strong possibility that they may evoke distinct VEGFR2 phosphorylation signatures, leading to activation of diverse downstream signaling pathways. Stress axes are mainly comprised of three pathways: PERK, IRE1, and ATF6. Notably, VEGF simultaneously activates two of these stress pathways, PERK (eIF2␣) and ATF6 (58), in contrast to endorepellin, which activates the former. However, we cannot exclude the possibility that endorepellin may also be involved in the ATF6 pathway.

Endorepellin induces an anti-angiogenic stress pathway
A novel and unique aspect of our study is the combination of biochemical and confocal imaging assays performed on sprouted capillary-like vessels emanating from aortic rings grown in a 3D collagen network. This ex vivo assay is a useful model to study angiogenesis insofar as the sprouts emerging from the aorta are purely endothelial in nature, with only a minute layer of pericytes supporting the sprouting vessels (67). This allowed us to efficiently dissect the angiogenic vessels and the effects of soluble endorepellin, tunicamycin, and PERK inhibitor. We successfully demonstrated endorepellin-based activation of the stress axis in the rings and concurrent abrogation of sprouting by performing imaging and biochemical analysis, providing robust evidence of protracted stress inhibiting angiogenesis in the mouse system.
One of the limitations of the aortic ring assay is the lack of blood flow through the newly formed vessels, which is essential for hemodynamic and mechanochemical functions (67). However, as this study focuses only on endorepellin's effect on formation of neovessels and activation of stress proteins, the absence of blood flow does not have a negative impact on our results. Another challenge concerning the aortic ring assay is reliable quantification of neovessel formation. Different techniques have been described previously (68 -70). These methods mainly include counting the sprouts or evaluating the branching patterns. However, counting sprouts can be challenging, especially when the sprouts are numerous and densely packed. A hurdle of quantifying branching pattern is inconsistency in network formation, which may lead to erroneous calculations. To circumvent these issues, we modified the quantification process by using fluorescently labeled rings (marked with IB4 or CD31) to trace the area of vascularization or sprouting and compared the angiogenic footprint between different conditions. As our approach excludes the ring area and exclusively focuses on the newly formed vessel sprouts and branches, it is very sensitive in obtaining detailed information. The margin of error is also primarily reduced because of digital processing and quantification of fluorescence and not manual evaluation of phase-contrast images.
The discovery of activation of the stress pathway by endorepellin is important because the effector molecule GADD45␣ is the convergence point for various pathways (28). For instance, GADD45␣ is known to inhibit mTOR and may have a role in evoking autophagy in endothelial cells, leading to angiostasis (26). GADD45␣ could be the connector protein between the two processes, induction of autophagy and suppression of angiogenesis, which we have proposed in our previous studies (20,45). It would be interesting to evaluate the role of GADD45␣ in enabling cross-talk between these two pathways. This study reinforces previous findings that endorepellin induces broad activation of many anti-angiogenic pathways. We showed previously that endorepellin inhibits the PI3K/PDK1/Akt and PKC/c-Jun N-terminal kinase/AP1 axes, which subsequently inhibit HIF1␣and AP1-based VEGFA promoter activation, inhibiting angiogenesis (24). Additionally, we recently discovered that endorepellin induces autophagic degradation of hyaluronan synthase 2 (HAS2), the major producer of pro-angiogenic and pro-inflammatory hyaluronan. This, in turn, evokes dynamic interaction of HAS2 and ATG9A and downstream angiogenic inhibition in vascular endothelia (71). Thus, endorepellin could evoke multiple pathways in endothelial cells emanating from its interaction with VEGFR2, which would synergistically block angiogenesis.
Endorepellin-based shotgun activation of these pathways ensures failsafe inhibition of angiogenesis, making it a viable candidate for combinatorial treatment. Overactivation of stress in the vasculature using endorepellin could be used to enhance susceptibility of endothelial cells to antiangiogenic drugs. Notably, hyperactivation of the stress axis in myeloma cells evoked by proteasome inhibitors enhances susceptibility to chemotherapy (72). Even though endorepellin does not target cancer cells directly, the concept of stress hyperactivation making cells susceptible to annihilation by drugs could be applied in the antiangiogenic paradigm.
Collectively, our findings posit endorepellin as an activator of cellular stress that simultaneously inhibits angiogenesis and evokes autophagy. Given this information, endorepellin may prove to be a protein-based therapeutic agent for overcoming drug resistance against angiogenic inhibitors and, thus, improve cancer therapy. Furthermore, we believe that subsequent investigation will add more extracellular matrix proteins to the list of stress activators and bring to the surface many more key players associated with the stress axis. Thus, we attribute an additional meaningful function to endorepellin's already extensive repertoire and further underscore the importance of the extracellular matrix in improving cancer therapy.

Nuclear fractionation and immunoblotting
HUVECs (ϳ10 6 cells) treated with vehicle or endorepellin for 4 h were harvested and spun down at 500 ϫ g for 5 min. Pelleted cells were washed in PBS, and fractionation was performed using NE-PER nuclear and cytoplasmic extraction reagent (78833, Thermo Scientific). The nuclear pellets were washed twice with PBS to remove any cytoplasmic contaminants before extraction. To perform Western blotting, following treatment, endothelial cells were lysed using radioimmune precipitation assay buffer (50 mM Tris-HCl, 50 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 0.5% sodium deoxycholate, 1 mM sodium orthovanadate, 0.5% SDS, 1 g/ml leupeptin, 1 g/ml aprotinin, 1 mM PMSF, 100 M L-1-tosylamido-2-phenylethyl chloromethyl ketone, and 1 EDTA-free protease inhibitor tablet) for 20 min while rocking on ice. Proteins were separated on an SDS-PAGE gel, transferred to a nitrocellulose membrane (Bio-Rad), incubated with the appropriate antibodies, and visualized using SuperSignal West Pico chemiluminescence substrate (Thermo Scientific) and an Image Quant LAS-4000 (GE Healthcare).

Immunofluorescence and confocal laser microscopy
HUVECs (ϳ5 ϫ 10 4 cells) were grown on gelatin-coated four-chambered slides (Thermo Fisher Scientific) and given the stipulated treatments, and then the cells were briefly washed with PBS and fixed with 4% (w/v) paraformaldehyde for 20 min on ice (75). Cells were blocked in 1% BSA for 1 h, washed three times in PBS, incubated with primary antibodies for 1 h, washed three times in PBS (76), and then incubated with secondary antibodies for 1 h, as described before (77,78). Stained cells were mounted using a hard-set mounting medium containing DAPI (Vector Laboratories, Burlingame, CA) and sealed with a coverslip. Immunofluorescence images were captured using ϫ63 objective on a Leica DM5500B microscope installed with Leica Application Suite v1.8 software (Leica Microsystems, Frankfurt, Germany). For higher-resolution images, confocal microscopy was used, and images were captured using a ϫ63, 1.3 oil immersion objective of a Zeiss LSM-780 confocal laser-scanning microscope equipped with the Zen Imaging software. Images were acquired as single optical sections of 1 m and collected with the pinhole set to 1 Airy unit for all channels. Post-acquisition, images were analyzed/processed using ImageJ and Photoshop CS6 (Adobe Systems).

siRNA-mediated knockdown
Individual transient knockdown of PERK and eIF2␣ was achieved in HUVECs using siRNA specific for PERK (sc-36213) or eIF2␣ (sc-35272). Scrambled siRNA (sc-37007) was used as the negative control. HUVECs (ϳ2 ϫ 10 5 cells/well) were seeded in 6-well plates to achieve ϳ80% confluence after attachment and transfected with 100 pM siScr, siPERK, or sieIF2␣ and 10 l of Lipofectamine RNAiMax (Life Technologies) diluted in 150 l of serum-free Opti-MEM medium (Gibco) overnight. The following day, medium containing siRNA was aspirated, and cells were incubated in 250 l/well Lifeline Cell Technology medium containing 4% FBS and 1% PenStrep for 1 day. On day 3, 4% medium was replaced with 2% FBS-containing medium. On day 4, we started respective treatments with vehicle (PBS), endorepellin (200 nM), or tunicamycin (10 g/ml). siRNA-mediated knockdown of PERK or eIF2␣ was verified by immunoblotting. Cell lysates were further analyzed by immunoblotting for proteins downstream of PERK and eIF2␣.

Aortic ring assay
For all animal-based experiments, instructions were followed according to the Guide for Care and Use of Laboratory Animals and the Institutional Animal Care and Use Committee of Thomas Jefferson University. Thoracic aortae from 6-to 7-week-old WT C57/BL6 mice (The Jackson Laboratory) were surgically isolated, cleaned off the fibro-adipose tissue, and sectioned serially into 0.5-to 1-mm rings. Rings were sandwiched between two 3D collagen type I gel layers (1 mg/ml) in 48-well plates and incubated in EGM-2 medium (Lonza) at 37°C as described elsewhere (48,79). Following initial sprouting, i.e. 3 days after embedding, the rings were given acute or chronic treatment. For immediate measurement of stress induction, rings were treated acutely (2-4 h) with vehicle (PBS), endorepellin (200 nM), or tunicamycin (10 g/ml). They were subsequently processed for immunofluorescence or Western blotting. For observing long-term effects of the reagents on sprouting, the rings were grown over 10 days with alternative-day treatment of vehicle (PBS), endorepellin (200 nm), tunicamycin (10 g/ml), or PERKi. The rings were then processed for immunofluorescence or Western blotting.

Processing the aortic ring data
Sprouting of the rings was measured as radial distance using low-magnification (ϫ5) phase-contrast images (79). We subtracted the background and highlighted sprouts using the threshold function. Then we manually encircled the edges of the sprouts and measured the radii (Fig. S3A).
During the immunofluorescence analyses, the rings were labeled with IB4, CD31, p-PERK, and GADD45␣. Z-stack images of the rings were captured with a Zeiss LSM780 NLO Endorepellin induces an anti-angiogenic stress pathway confocal/multiphoton microscope, and the images were analyzed using ImageJ software. Two methods of analyses were used: fluorescence of the traced-out sprouts (79) and intensity profiles of the sprouts, both using ImageJ. Briefly, fluorescence intensity was uniformly normalized across all conditions using the threshold function, and regions of interest outlining the sprouts were selected using the polygon drawing tool. We then measured the integrated intensity of the traced areas (Fig. S3B) and normalized on the frame areas. The normalized fluorescence intensity of each protein was plotted as the mean fluorescence intensity. IB4 or CD31 expression was also used to quantify the extent of sprouting in the rings. We also performed line-scanning analyses of protein levels in the sprouts using the profile plot tool in ImageJ. A line was drawn along the sprout, and fluorescence intensity was measured along the length.

Quantification and statistical analysis
The immunoblots were analyzed by scanning densitometry using ImageJ. Statistical significance was calculated by twotailed, unpaired Student's t test. One-way ANOVA was performed on data with multiple groups. All experiments were conducted at least three times, and the mean differences were considered significant at p Ͻ 0.05.

Data availability
All data are contained in this manuscript.
Author contributions-A. K., C. G. C., and R. V. I. conceptualization; A. K., C. G. C., and R. V. I. data curation; A. K. and C. G. C. software; A. K., C. G. C., and R. V. I. formal analysis; A. K., C. G. C., and R. V. I. validation; A. K., C. G. C., and R. V. I. investigation; A. K., C. G. C., and R. V. I. visualization; A. K. and C. G. C. methodology; A. K. and R. V. I. writing-original draft; C. G. C. and R. V. I. writingreview and editing; R. V. I. supervision; R. V. I. funding acquisition; R. V. I. project administration.