Dual regulation of Stat1 and Stat3 by the tumor suppressor protein PML contributes to interferon α-mediated inhibition of angiogenesis

IFNs are effective in inhibiting angiogenesis in preclinical models and in treating several angioproliferative disorders. However, the detailed mechanisms of IFNα-mediated anti-angiogenesis are not completely understood. Stat1/2/3 and PML are IFNα downstream effectors and are pivotal regulators of angiogenesis. Here, we investigated PML's role in the regulation of Stat1/2/3 activity. In Pml knock-out (KO) mice, ablation of Pml largely reduces IFNα angiostatic ability in Matrigel plug assays. This suggested an essential role for PML in IFNα's anti-angiogenic function. We also demonstrated that PML shared a large cohort of regulatory genes with Stat1 and Stat3, indicating an important role of PML in regulating Stat1 and Stat3 activity. Using molecular tools and primary endothelial cells, we demonstrated that PML positively regulates Stat1 and Stat2 isgylation, a ubiquitination-like protein modification. Accordingly, manipulation of the isgylation system by knocking down USP18 altered IFNα-PML axis-mediated inhibition of endothelial cell migration and network formation. Furthermore, PML promotes turnover of nuclear Stat3, and knockdown of PML mitigates the effect of LLL12, a selective Stat3 inhibitor, on IFNα-mediated anti-angiogenic activity. Taken together, we elucidated an unappreciated mechanism in which PML, an IFNα-inducible effector, possess potent angiostatic activity, doing so in part by forming a positive feedforward loop with Stat1/2 and a negative feedback loop with Stat3. The interplay between PML, Stat1/Stat2, and Stat3 contributes to IFNα-mediated inhibition of angiogenesis, and disruption of this network results in aberrant IFNα signaling and altered angiostatic activity.

whereas the function of Isg15-linked Stat1 isgylation is poorly understood. Isg15 is highly induced by type I interferons, including IFN␣ and IFN␤. Protein isgylation, catalyzed by Isg15-activating E1 enzyme (Ube1L), Isg15-conjugating E2 enzyme (UbcH8), and Isg15 E3 ligase (HerC5), is a ubiquitination-like posttranslational modification that covalently conjugates Isg15 peptides to lysine residues of target proteins (20,23). Isg15 conjugation is cleaved by the sole Isg15 deconjugating enzyme Usp18. The exact biological function of protein isgylation remains unclear but is known to play a role in the antiviral response and the control of protein stability (24,25). Although it has been proposed that isgylation positively regulates JAK-Stat signaling and promotes Stat1-associated gene expression (26,27), how Stat1 isgylation is regulated remains elusive. Furthermore, it is unclear whether other Stat family members involved in IFN␣ signaling such as Stat2 and Stat3 are also subjected to Isg15 conjugation.
The PML gene was originally isolated as a fusion partner of the human retinoic acid receptor ␣ (RAR␣) in patients with acute promyelocytic leukemia (28 -30). A significant fraction of PML is localized in discrete nuclear structures alternatively called Kremer bodies, nuclear domain 10, PML oncogenic domains, or PML nuclear bodies (NBs). Ablation of PML results in the loss of these structures and thereby changes PMLrelated biological functions, including transcriptional regulation, protein modification, apoptosis, and cellular senescence (31). PML is highly expressed in endothelium, and its mRNA levels are further elevated upon IFN␣ stimulation (6,32). We recently showed that PML is essential for TNF␣-and IFN␣mediated inhibition of EC migration and angiogenesis (6,33). Furthermore, miR-1246 released from colon cancer cell-derived microvesicles, targets the PML mRNA 3Ј-UTR and downregulates PML expression in HUVECs, thus promoting EC migration and tube formation (34). These results suggest that PML plays an important role in control of angiogenesis.
In this study we observed increased angiogenesis in Pml KO mice when treated with IFN␣. This result suggested a pivotal role of PML in IFN␣-mediated anti-angiogenesis. To further investigate the details, we found that PML has a dual function in this process. Upon IFN␣ stimulation, PML promotes ISGF3 activity for target gene expression and enhances Stat1/Stat2mediated angiostatic effects in part through Stat1/Stat2 isgylation. Additionally, PML negatively regulates nuclear Stat3 stability and thus attenuates Stat3-mediated pro-angiogenic effects. Based on these findings, we propose a model in which PML plays an important role in IFN␣-mediated Stat1/2 and Stat3 activation and disruption of PML blocked IFN␣ angiostatic ability.

PML is required for IFN␣-mediated angiostatic activity in vivo
We previously demonstrated that PML is essential for the ability of TNF␣ and IFN␣ to inhibit EC migration and capillary tube formation (6,33). To further validate our in vitro data, we carried out in vivo Matrigel plug assays. Pml KO mice exhibited increased angiogenesis in the plugs compared with the wildtype mice (Fig. 1). Consistent with our previous in vitro data, IFN␣ inhibited angiogenesis in the wild-type animals. However, significant increases in angiogenesis were observed in IFN␣-treated Pml KO mice, suggesting that the ability of IFN␣ to inhibit angiogenesis is largely dependent on Pml. PML is also an IFN␣-inducible gene. This observation indicates that PML is required for IFN␣ to inhibit angiogenesis in vitro and in vivo.

Depletion of PML promotes microvessel outgrowth and changes expression of IFN␣and Stat1-inducible genes
Because PML is highly expressed in human endothelium and our data also suggest PML as a key cytokine-inducible angiogenic inhibitor, we focused our study on endothelial system. To further dissect the role of PML in IFN␣ signaling, we examined PML and IFN␣ target genes. Earlier microarray data showed that knockdown (KD) of PML in HUVECs resulted in alterations of a set of target genes (35). Among these altered genes, IFN␣-inducible genes (36) were significantly decreased in PML KD HUVECs (35). More than 100 genes that were induced Ͼ2-fold by IFN␣ in HUVECs were down-regulated Ͼ2-fold in PML KD cells. To validate this, we carried out quantitative realtime (qRT)-PCR and confirmed several IFN␣ and PML common target genes ( Fig. 2A). These data suggest that loss of PML may compromise the ability of IFN␣ to induce its target genes. Stat1 is the major IFN␣-induced transcription factor that is responsible for activation of numerous target genes. Cheon and Stark (37) reported that overexpression of a tyrosine phosphorylation-defective Stat1 mutant (unphosphorylated Stat1 (U-Stat1)) resulted in expression of several known interferon target genes, indicating that unphosphorylated Stat1 is transcriptionally active. Interestingly, these genes are down-regulated in PML knockdown cells as shown in our array (Fig. 2B). Stat1 and PML common target genes include IFIT1, MX1, OAS1, OAS2, OAS3, IFI27, IFI44, and STAT1. By qRT-PCR, we validated that these Stat1-and unphosphorylated Stat1-inducible genes are down-regulated in PML KD ECs (Fig. 2C). These data indicate that Stat1 induces PML expression in response to IFN␣ stimulation (6), and in turn, PML positively regulates Stat1 transcription activity. Left, two representative Matrigel plugs from each group are shown. Right, quantification of angiogenesis. Approximately 200-mg plugs isolated from mice were homogenized in Drabkin's solution, and optical density was measured at 540 nm and normalized to the weight of the plugs. Unpaired two-tail t test (*, p Ͻ 0.05; ***, p Ͻ 0.001). The numbers (N) of mice are indicated for each group.

Stat1-PML-Stat3 axis for IFN␣-mediated angiogenesis inhibition
To test whether the ability of Pml to regulate Stat1 target genes is conserved in mouse ECs, we isolated aortic rings from WT and Pml KO mice and carried out ex vivo aortic ring explant assays. We found that aortic rings isolated from Pml KO mice exhibited increased outgrowth of vessels compared with the rings from the wild-type animal (Fig. 2, D and E). Cells from aortic ring outgrowth were further isolated by FACS and confirmed as CD31-positive endothelial cells (38) (Fig. 2F). Isolated aortic ECs were grown and harvested, and total RNA was prepared. RT-qPCR demonstrated that the expression of Stat1 target genes such as Ifit3, Oas2, Mx1, and Stat1 were significantly reduced in Pml Ϫ/Ϫ ECs (Fig. 2G).

PML promotes Stat1 and Stat2 isgylation and affects the ISGF3 complex
The transcriptional activity of Stat1 is regulated by IFN␣induced phosphorylation, nuclear translocation, acetylation, sumoylation, and Isg15 conjugation (isgylation). Although isgylation has been proposed to positively regulate Stat1 transcriptional activity, it is not clear whether this modification is present in ECs or how this modification is regulated (27,39,40). To address these issues, we determined the effect of IFN␣ and Usp18, the sole Isg15-deconjugating enzyme, on isgylation of nuclear Stat1 (nStat1). Due to the transient and reversible nature of this modification, the detectable levels of isgylated proteins are relatively low. To overcome this issue, HUVECs were transiently transfected with a control siRNA (siCtrl) or two USP18 siRNAs (siUSP18) followed by IFN␣ treatment for 16 h and nuclear and cytoplasmic fractionation for Western blotting. RT-qPCR was performed to confirm that efficient knockdown of USP18 was achieved (Fig. 3A). IFN␣ induced accumulation of nStat1 and a distinct slower migrating species (ϳ100 kDa) (Fig. 3B, lanes 1 versus 4). This Stat1 species was further induced in USP18 KD cells (Fig. 3B, lanes 5 and 6, L.E.).
Similar observations were made for unmodified nuclear Stat2 and a slower migrating species of nStat2 (lanes 7-12, arrowhead). These results suggest that these slower-migrating proteins are likely to be Isg15-conjugated nStat1 (Isg15-nStat1) and nStat2 (Isg15-nStat2) and that knockdown of USP18 significantly increased this modification. Similar to most post-translational modifications including acetylation, sumoylation, and methylation, only a small fraction of the Stat1 and Stat2 was isgylated.
To further verify whether these slower-migrating species are Isg15-conjugated Stat1 and Stat2, we performed immunoprecipitation followed by Western blotting. We first carried out knockdown experiments with a control (siCtrl), ISG15 (siISG15), or USP18 (siUSP18) siRNA followed by a 16-h IFN␣ treatment. Knockdown of ISG15 decreased protein isgylation, and knockdown of USP18 increased the isgylation signal. Cells were harvested, and subcellular fractions were prepared. Nuclear fractions were immunoprecipitated with a control antibody, anti-Isg15, anti-Stat1, or anti-Stat2 antibody. Immunopellets were Western-blotted with anti-Stat1, anti-Stat2, or anti-Isg15 antibody. Our data indicate that endogenous Stat1 is isgylated in IFN␣-treated HUVECs (Fig. 4, A and B) as well as Stat2 (Fig. 4, C and D). The observation of Stat1 isgylation is consistent with a previous report (21). Stat2 has not been previously reported to be Isg15-conjugated. Based on the mobility of the bands, we also detected unmodified Stat1 and Stat2 bands in the anti-Isg15 immunoprecipitates (Fig. 4, B and D, lanes 7-9, marked with asterisks). Together, these observations suggest that Stat1 and Stat2 are Isg15-conjugated.
We further carried out time-course experiments to determine whether isgylation of Stat1 and Stat2 occurs early or late after of IFN␣ stimulation. To investigate this, HUVECs were transiently transfected with a control or two independent  (37). The log-fold change was scaled in a green-dark-red color scheme. Each row represents a gene designated by the official gene symbol. Two independent PML siRNAs (siP1 and siP2) was compared with a non-targeting control siRNA (siC). C, RT-qPCR of U-Stat1 target genes in PML KD ECs. HUVECs were transiently transfected with a non-targeting siRNA or a mixture of siP-1 and siP-2. Total RNA was isolated 48 h after transfection followed by qRT-PCR. Relative mRNA levels are shown. D, representative pictures showing microvessel outgrowth in explanted aortic rings isolated from WT and Pml Ϫ/Ϫ mice at days 3, 5, and 6. Note that more microvessels sprouting from aortic rings prepared from Pml Ϫ/Ϫ mice than those isolated from the WT animals. E, quantification of microvessel outgrowth. F, cells from aortic ring outgrowth at the end of the experiments (12 days after explanting) were isolated and analyzed for CD31 expression by FACS analysis. G, aortic ECs were isolated from WT and Pml Ϫ/Ϫ mice as described in F. The expression of putative Stat1 target genes was quantified by RT-qPCR. HUVECs were transiently transfected with a control (siCtrl) or two independent siRNAs (siU-1 and siU-2) against USP18. 72 h after transfection cells were treated with vehicle or IFN␣ (10 3 units/ml) for 16 h. An aliquot of cells was used to isolate total RNA followed by RT-qPCR (A), and the rest of cells were used for subcellular fractionation (B). L.E., long exposure; S.E., short exposure.

Stat1-PML-Stat3 axis for IFN␣-mediated angiogenesis inhibition
USP18 siRNAs, treated with IFN␣, harvested at different time points, and analyzed by Western blotting. In the siCtrl-transfected ECs, we did not observe any difference in total protein levels of Stat1 or Stat2 for the first hour, but a significant increase in the levels of nStat1 and nStat2 was observed after 16 h of IFN␣ treatment (Fig. 5, A and B, lanes 4 versus 8 and lanes 12 versus 16). Similar to that observed in Fig. 3B, a small fraction of a major slower-migrating Stat1 and Stat2 species was detected after a 16-h of IFN␣ treatment in control ECs (Fig.  5, A and B, lanes 1-4). In USP18 KD ECs, these slower migrating Stat1 and Stat2 species were increased (Fig. 5, A and B, lanes 4 versus 8 and lanes 12 versus 16). However, we did not observe slower-migrating species of Stat3 in control or USP18 KD HUVECs in response to IFN␣ stimulation, implying that Stat3 is not isgylated (Fig. 5C).
Previous studies that knockdown of PML decreased IFN␣ and Stat1 target gene expression suggest that knocking down PML might have an effect on posttranslational modification, nuclear translocation, or abundance of ISGF3 (Stat1, Stat2, and IRF9). To test this, HUVECs were transfected with a control siRNA or a PML siRNA, treated with IFN␣, harvested, and nuclear and cytoplasmic fractions were prepared. Western blotting was performed with anti-Stat1, -Stat2, -Tyr-701-Stat1, and anti-IRF9 antibodies. In the control cells, in response to IFN␣ treatment, we observed the following. 1) An additional nStat1 species appeared transiently at 0.5 h of IFN␣ (Fig. 6A, lane 2). This species is likely to be Tyr(P)-701-nStat1 (see Fig.  6B). 2) A significant increase appeared in nuclear Stat2, but not Stat1, at 0.5 h of IFN␣ treatment (Fig. 6A, lanes 1 versus 2). 3) A significant increase appeared in nStat1 and Stat2 at 16 h of IFN␣ treatment (Fig. 6A, lane 5). 4) A significant increase appeared in nuclear PML, Isg15, and IRF9 at 16 h of treatment (Fig. 6A, lane 5). 5) Isg15-conjugated nStat1 and Stat2 appeared at 16 h of IFN␣ treatment (arrows) (Fig. 6A).
Furthermore, knockdown of PML resulted in a marked decrease in Isg15-conjugated nStat1 and nStat2 at 16 h of IFN␣ treatment but had little or no effect on unmodified nStat1 and nStat2 (Fig. 6A, lanes 5 versus 10). Additionally, knockdown of PML significantly increased Isg15-conjugated and unmodified cytoplasmic Stat2. However, knockdown of PML did not decrease Tyr(P)-701-Stat1 (Fig. 6B, lanes 2 versus 7). These observations suggest that PML promotes Stat1 and Stat2 activity at a late stage of IFN␣ signaling by increasing Isg15 conjugation of nStat1 and Stat2. We further examined whether Pml regulates Isg15 conjugation of nStat1 and Stat2 in rodent ECs and found that loss of Pml resulted in decreases in Isg15-conjugated nStat1 and nStat2 (Fig. 6C, lanes 4 and 8) and blocked IFN␣-induced nIRF9 (lanes 2 and 6) in mouse ECs.

Stat1-PML-Stat3 axis for IFN␣-mediated angiogenesis inhibition The effect of USP18 knockdown on IFN␣-mediated Stat1 isgylation, EC migration, and capillary tube formation
Our results indicated that loss of PML decreased nuclear isgylated Stat1 and Stat2 and nuclear IRF9 and reduced ISGF3 target gene expression. PML is angiostatic, as supported by both in vitro (6) and in vivo data ( Figs. 1 and 2D). Our observations that knockdown of USP18 increased isgylation of nStat1 and nStat2 (Figs. 3B and 5A, lanes 4 versus 8 and 12 versus 16) and that knockdown of PML decreased isgylation of nStat1 and nStat2 (Fig. 6A, lane 5 versus 10) suggested that USP18 is proangiogenic. To test this idea, we knocked down USP18, PML, or both in HUVECs and examined their effects on IFN␣-mediated inhibition of EC migration and network formation. We found that knockdown of USP18 increased Stat1 isgylation and decreased EC migration and capillary network formation (Fig.  7A, lanes 1-3, and Fig. 7B), whereas knockdown of PML did the opposite (Fig. 7, B and C). Furthermore, knockdown of PML compromised the USP18 KD effects on Stat1 isgylation and EC migration and capillary tube formation (Fig. 7A, lanes 4 -6, B,  and C). Collectively, our data suggest that PML functions as an IFN␣ downstream effector to promote Stat1/2 isgylation and by this means enhances Stat1/2 transactivation of their gene expression and IFN␣-mediated angiostatic activity.

PML negatively regulates nStat3 stability and attenuates Stat3 function in EC network formation
It has been proposed that Stat1 and Stat3 antagonize each other's activity (15,16). Consistent with the notion, several Stat1-inducible genes described above (Fig. 2C) were repressed when Stat3 was overexpressed in A549 lung cancer cells (41). Using an siRNA KD approach and microarray hybridization, a recent study has examined gene expression patterns in control and Stat3 KD HUVECs (42) (data set no. GSM688389). We analyzed Stat3 target genes by perl and R programs and identified Stat3 target genes in HUVECs. Comparison of these Stat3 target genes with the PML target genes identified in our array study (35) suggested significant overlap. We found 426 genes that showed a Ͼ2-fold decrease in expression in PML KD ECs, 143 of which were induced Ͼ2-fold when STAT3 was knocked down (supplemental Table 1). Furthermore, among 167 genes up-regulated Ͼ2-fold in PML KD ECs, 36 of which were found to be down-regulated Ͼ2-fold in STAT3 KD ECs (supplemental Table 2). These observations indicate that PML and Stat3 share a large set of common target genes and that PML and Stat3 mutually antagonize a subset of each other's target genes.
To dissect the mechanism by which PML affects Stat3 expression in response to IFN␣ stimulation, we first examined whether knockdown of PML in HUVECs alters Tyr(P)-705-nStat3, nuclear translocation, or abundance of nStat3. In the control cells nStat3 and Tyr(P)-705-nStat3 significantly increased 0.5 h after IFN␣ treatment, markedly declined after 1 h of treatment, and Tyr(P)-705-nStat3 completely disappeared after 2 h of treatment (Fig. 8, A and C). These observations are consistent with previous reports indicating that Stat3 translocates to the nucleus in response to 0.5 h of IFN␣ treatment (14,19). We further discovered that knockdown of PML 1) had little or no effect on nuclear translocation of Stat3 (Fig.  8A, lanes 2 and 7), 2) increased nStat3 accumulation after 1 h of IFN␣ treatment ϳ2-fold (Fig. 8, A, lanes 8 -10 and B), and 3) prolonged Tyr(P)-705-nStat3 (Fig. 8C, lanes 2-4 versus 7-9). These observations are significant because nStat3 antagonizes nStat1 activity, and a 2-fold increase in nStat3 accumulation will have a profound effect on IFN␣ signaling. In summary, these observations provide an important new finding that nStat3 abundance is altered by PML.
IFN␣ induces rapid nuclear translocation of Stat3. However, it remains unclear how nStat3 sharply declines afterward. Because the decrease in nStat3 occurs rapidly (0.5-1 h), we hypothesized that nStat3 is subjected to degradation after 0.5 h of IFN␣ treatment. To explore this possibility, HUVECs were treated with or without the proteasome inhibitor MG132 after 0.5 h of IFN␣ treatment. We found that MG132 increased nStat3 accumulation in IFN␣-treated ECs, indicating that nStat3 is subjected to proteasome-mediated degradation after 0.5 h of IFN␣ stimulation (Fig. 8D). However, knockdown of PML blocked MG132-mediated increase in nStat3 accumulation (Fig. 8E).
Based on the transcriptome analysis from PML and STAT3 KD HUVECs as well as biochemical data, we hypothesized that the increased Stat3 abundance observed in PML KD ECs contributes to elevated EC migration and angiogenesis. To test this

Stat1-PML-Stat3 axis for IFN␣-mediated angiogenesis inhibition
hypothesis, we carried out wound healing and in vitro capillary tube formation and determined whether LLL12, a selective Stat3 small molecule inhibitor, has effects on EC migration and tube formation in PML KD ECs. Our data indicated that LLL12 inhibited increased EC migration and capillary tube formation in PML KD HUVECs upon IFN␣ stimulation (Fig. 8, F and G).

Discussion
Upon IFN␣ stimulation, both Stat1 and Stat3 are phosphorylated and translocated into the nucleus binding to specific DNA regions to control gene expression. However, the activated Stat3 in the nucleus is transient and only found during the early phase of IFN stimulation (19,43). The simultaneously activated opposing activities of Stat1 (angiostatic) and Stat3 (angiogenic) ensure forced amplification (feedforward) or timely attenuation (feedback) of the biological effects of a given dose of IFN␣. PML has been known as an interferon-stimulated gene with the potential to block angiogenesis (6,44). Thus, a link between IFN␣, PML, and anti-angiogenesis is intriguing. Our results demonstrate that PML has a dual activity in positively regulating Stat1 activity and negatively regulating Stat3 activity through isgylation and protein stability control, respectively. Thus, as one of the ISGs, PML has a unique role in controlling Stat1 and Stat3 activity, thereby contributing to IFN␣-mediated inhibition of angiogenesis. Our findings further support a concept that under certain circumstances, such as PML deficiency, IFN␣ may reduce or even lose its anti-angiogenic activity due to alteration of the balance between Stat1 and Stat3, [Stat1]/[Stat3] (Fig. 1). Several anti-cancer or antiinflammation drugs are direct or indirect PML inducers or suppressors and require PML for their therapeutic activity (45,46). As such, our data raise the possibility for anti-angiogenic ther- Figure 6. PML promoted Stat1 and Stat2 isgylation and affected ISGF3 complexes. A, HUVECs were transiently transfected with a control (siCtrl) or a PML siRNA (siPML). 72 h later cells were treated with IFN␣, harvested at 0, 0.5, 1, 2, and 16 h, and subcellular fractions were prepared. Nuclear (lanes 1-10) and cytoplasmic (lanes [11][12][13][14][15][16][17][18][19][20] fractions were subjected to Western blotting with the indicated antibodies. L.E., long exposure; S.E., short exposure. The arrows indicate Isg15-nStat1 and Isg15-nStat2. The asterisk marks a unknown Stat1 species, likely to be Tyr(P)-701-Stat1. Lamin B and ␣-tubulin were used as markers for nuclear and cytoplasmic fractions, respectively. B, the effect of PML KD on the abundance of Tyr(P)-701-Stat1. Nuclear fractions from A were subjected to Western blotting with anti-Tyr(P)-701-Stat1 antibody. C, aortic ECs from Pml Ϫ/Ϫ show decreased Isg15-nStat1, Isg15-nStat2, and nIRF9. Aortic ECs isolated from aortic rings of WT and Pml Ϫ/Ϫ mice were treated with IFN␣ (10 3 units/ml) and harvested at the indicated times, and subcellular fractions were prepared. The nuclear fractions were subjected to Western blotting with the indicated antibodies. apies in which patients are simultaneously treated with IFN␣and PML-altering agents.
There are several reports demonstrating that Stat1 transcriptionally activates PML expression (6,47,48). Our study is the first reporting that PML directly regulates Stat1 activity, indicating that there is a PML-Stat1 axis that forms a positive feedforward loop to mediate IFN␣-induced angiostatic activity (6). A large cohort of ISGs is regulated by both PML and the unphosphorylated active form of Stat1, supporting a role for PML and Stat1 in the same IFN regulatory axis (Fig. 2, B and C). It is known that Stat1 transcriptionally activates its own promoter (49). Thus, the decrease in STAT1 mRNA and Stat1 target gene expression is a consequence of reduced Stat1 activity in PML KD cells. Among these PML target genes, Stat1 has been demonstrated as a negative regulator of angiogenesis and Mx1, a GTPase known to inhibit cell migration and senescence (50,51). Furthermore, we also observed decreased expression of several senescence secretory cytokines, such as IL-1, CXCL1, and CXCL11, in PML KD HUVECs. Taken together, our findings support the concept that PML mediates IFN-induced EC
Mechanistically, PML mediates IFN␣ activity by promoting isgylation of Stat1 and Stat2 (Figs. 3-6). This is evident by the experiments with knockdown of ISG15 or USP18 and further validated in primary ECs isolated from WT and Pml KO mice (Fig. 6). Although Stat1 isgylation has been reported previously (21), we are the first to demonstrate that Stat2 is isgylated. It was proposed that the isgylation potentiates ISGF3 function and augments IFN␣ signaling to promote cellular senescence and anti-viral activity (39,40). Notably, our data demonstrated that protein isgylation promotes the inhibitory effect of the IFN-PML axis on EC migration and network formation (Fig. 7). However, the detailed mechanism by which isgylation enhances Stat1/Stat2 activity remains unknown, whereas it has been proposed that isgylation may regulate protein synthesis or protein stability (24,25,54). Mapping the isgylated residues in Stat1 and Stat2 and validating the defect of Stat1/2 isgylation incompetent mutants will help address this issue in the future.
Stat3 is a well known oncogene and angiogenesis promoter (43,55,56). PML is known to directly interact with Stat3 and inhibit Stat3 activity by sequestering Stat3 in PML NBs (57,58) or by interfering Stat3-HDAC3 interaction (59). Interestingly, we found that PML negatively regulates nuclear Stat3 abundance in response to IFN␣ treatment, partially by increasing proteasome-mediated Stat3 turnover (Fig. 8, A--E). Although several studies have suggested partial co-localization between components of proteasomes and PML NBs (60, 61) and a potential role of PML NBs in promoting protein degradation (62), the detailed mechanism by which PML promotes Stat3 turnover awaits future investigation. Furthermore, our data indicate that PML and Stat3 share a set of common target genes Procedures were the same as that described in Fig. 6, except that Stat3 and Tyr(P)-705-Stat3 antibodies were used for Western blotting. B, quantification of nStat3 protein abundance from A. The relative abundance of Stat3 at each time point was normalized to that at 0 h of IFN␣ treatment (lanes 1 and 6). n ϭ 5. C, samples from A were subjected to Western blotting with anti-Tyr(P)-705-Stat3 antibody. D, MG132 increased nStat3 after 0.5 h of IFN␣ treatment. HUVECs were treated with 0.5 h of IFN␣ (10 3 units/ml) followed by 0.5 h of treatment with vehicle or MG132 and harvested, nuclear fractions were prepared and Western-blotted with the indicated antibodies. E, knockdown of PML abolished MG132-mediated accumulation of nStat3 in response to IFN␣ stimulation. F, control and PML KD HUVECs were pretreated with LLL12 (100 nM) for 3 h and trypsinized. Equal numbers of cells were subjected to woundhealing assays and capillary tube formation assays in the presence of IFN␣ (G). Statistical analysis was performed by counting numbers of branch points per field at 6 and 21 h of the assay. n ϭ 6 per group (*, p Ͻ 0.05; ***, p Ͻ 0.001; unpaired two-tailed t test).
Stat1 and Stat3 can form either homodimers or heterodimers, and the type and stoichiometry of Stat1-Stat3 complex formation may affect its DNA binding and alter the IFNinduced gene expression pattern (67). Several studies suggest that Stat3 is a negative regulator of Stat1 and that depletion of Stat3 increases IFN-mediated anti-viral activity (43,68). Interestingly, a recent study demonstrated that knock-out of Stat1 in murine macrophage sustained type I IFN-induced Stat3 phosphorylation (69), an observation similar to what we observed in PML KD HUVECs (Fig. 8, A--E). As such, the prolonged pStat3 in Stat1 knock-out cells is presumably the outcome of the disturbed Stat1-PML-Stat3 regulatory network.
Our study uncovers PML as a key IFN␣-signaling regulator that amplifies a positive feedforward loop with ISGF3 complex through isgylation system and counteracts IFN signaling suppressor Stat3 by controlling Stat3 abundance. Through this intricate regulation, PML augments the IFN␣-Stat1 signaling in ECs and thus inhibits angiogenesis. Once PML function is lost, the overall IFN␣ activity tips toward enhancing Stat3 activity and, thus, enhances angiogenesis. Although our current study mainly focused IFN␣-PML-Stats axis function on EC regulation, the outcome of reduced in vivo angiogenesis inhibition in Pml knock-out mouse may partially result from other microenvironmental factors and awaits further investigation. In conclusion, our observations suggested that elevation of PML protein levels in endothelium would promote IFN␣ anti-angiogenic activity and therapeutic effects.

Cell lines, reagents, and antibodies
HUVECs were purchased from Lonza and cultured in endothelial cell basal medium (EBM-2; Lonza) with EGM-2 Single-Quot growth supplements (Lonza). Cells that were Ͻ6 passages were used in this study. LLL12 was purchased from EMD Millipore. MG132 was purchased from Sigma, recombinant human IFN␣ was purchased from R&D system, and recombinant mouse IFN␣ was purchased from PBL assay science. siRNAs were purchased from Thermo Scientific and are listed in the supplemental Table 4. Antibodies used in this study are listed in supplemental Table 3 and are tested in supplemental Figure 1. The transfection reagent DharmaFECT1 (T-2001) was purchased from Thermo Scientific.

siRNA knockdown and subcellular fractionation
A non-targeting siRNA or independent siRNAs against USP18, PML, or ISG15 were transfected into HUVECs using DF1 transfection reagent (Thermo Scientific) according to previous protocol (70). Cells were harvested at 72 h after siRNA transfection before preparation of total RNA or nuclear and cytoplasmic fractionation. Before harvest cells were treated with MG132 (50 g/ml) or IFN␣ (10 3 units/ml) for the indicated times. For nuclear and cytoplasmic fractionation, HUVECs were washed with 1ϫ PBS one time and spun down using a desktop centrifuge at 4°C, 5000 rpm for 5 min. The cell pellets were resuspended in 5ϫ the pellet volume in cytosolic fractionation buffer (10 mM Tris, pH 7.4, 10 mM NaCl, 3 mM MgCl 2 , 0.5%Nonidet P-40, and 5% glycerol) and kept on ice for 10 min. The nuclei were collected by centrifugation at 4000 rpm for 6 min, and the supernatants (1st cytosolic fraction) were transferred to a new tube. The pellets were resuspended in cytosolic fractionation buffer and spun down again. The supernatant was collected and combined with the 1st extract as the cytosolic fraction. The resulting nuclear pellets were resuspended in nuclear fraction buffer (10 mM HEPES, pH 7.5, 400 mM NaCl, 5 mM EDTA, 0.5% Nonidet P-40, 1 mM DTT, and 5% glycerol) on ice for 15 min and vortexed every 5 min until completely homogenized. The prepared lysates were mixed with an equal volume of 2ϫ SDS buffer (100 mM Tris-Cl, pH 6.8, 4% SDS, 0.2% bromphenol blue, 20% glycerol, and 2% ␤-mercaptoethanol) and boiled for 10 min. For Western blots, antibodies against Lamin B or ␣-tubulin served as the internal loading and recovery controls for nuclear and cytoplasmic fractions, respectively.

Total RNA extraction, RT-PCR, and qRT-PCR
HUVECs and mice aortic endothelial cells were harvested, and total RNA was prepared using PrepEase RNA Spin kits (U. S. Biochemical Corp./Affymetrix). Single-strand cDNA pools were generated using transcriptor universal cDNA master (Roche Applied Science) according to the manufacturer's instructions. The cDNAs of the target genes were quantified by qPCR using an iCycler (Bio-Rad) platform with 2ϫiQ SYBR Green Supermix (Bio-Rad) and appropriate primer sets as listed in supplemental Table 3. The PCR program was set for 40 cycles with three steps of 95°C for 30 s, 55°C for 30 s, and 72°C for 30 s. Melting curves were obtained after each cycle to ensure the homogeneity of the PCR products. The relative abundance of target genes was normalized to an internal control (18S rRNA) and are depicted as the mean Ϯ S.D. from three independent experiments.

Examination of IFN␣-induced Stat1/Stat2 isgylation
The siControl, siUsp18, and siISG15 HUVECs were treated with INF␣ (10 3 units/ml) for 16 h. After collection, the nuclear fractions were prepared with a mixture of protease inhibitors (Roche Applied Science). The lysates were precleaned with protein A beads and immunoprecipitated with anti-Stat1, Stat2, or Isg15 antibodies. The immunoprecipitates were pulled down by protein A beads and subjected to Western blotting with the indicated antibodies.

Capillary tube formation assay
HUVECs were transiently transfected with the indicated siRNAs. Cells were treated with trypsin, and 10 4 cells were plated on Matrigel (Millipore) in 96-well plates. After seeding on the gel for 2 h, cells were treated with or without INF␣ (10 3 units/ml) in the culture medium for 12 h before image capture. Six fields per experimental group were randomly picked, and the branch points in each field were counted for statistical analysis. All results are shown as the mean Ϯ S.D.

Stat1-PML-Stat3 axis for IFN␣-mediated angiogenesis inhibition Wound healing assay
HUVECs were transiently transfected with siRNAs. The next day the cells were plated on 6-well dishes. Cells were grown to confluence and treated with IFN␣ (10 3 units/ml) for 4 h and followed by scratching with a sterile pipette tip to generate wounds with continued IFN␣ treatment. At 0 and 12 h after scratching, images of 6 randomly chosen fields were captured through a microscope equipped with a camera, and the width of the same wounds was measured by ImageJ. A portion of the treated cells was used for Western blotting. The cell migration rate was quantified by measuring the distance of the wound closure between 0 and 12 h. The results are shown as the mean Ϯ S.D.

Isolation of mouse aortic endothelial cells
Both Pml ϩ/ϩ (WT) and Pml Ϫ/Ϫ (KO) mice were maintained in the 129 S1/SvImJ background with normal rodent chow and sterile water in the Health Science Animal Facility at Case Western Reserve University. All procedures were approved by the Case Western Reserve University Institutional Animal Care and Use Committees. The genotype of the mouse was confirmed by PCR analysis from tail biopsies. The mouse aortas were isolated, and periadventitial fat was removed. The sliced aortic rings were embedded into Matrigel (BD Biosciences) in EGM2 medium (Lonza) with 10% FBS (Sigma). After 3 days the gel-embedded aortic rings were collected and treated with trypsin. Endothelial cells were resuspended in EGM2 medium with 5% FBS and cultured on the fibronectincoated culture dishes. The endothelial cells used in the study were from passages 2-3.

In vivo Matrigel plug assay
In vivo Matrigel plug assays were performed according to our published protocol (38). To implant plugs, wild-type and Pml KO mice at 8 weeks of age were subcutaneously injected with 1 ml/mouse of Matrigel on an abdominal site adjacent to, but not overlapping abdominal blood vessels. The base Matrigel contained a mixture of Matrigel (BD Biosciences) with 60 ng/ml VEGF-A and 60 units/ml heparin. The vehicle control was 0.1% BSA in 1ϫPBS. For cytokine treatment, mouse TNF␣ or IFN␣ was added to a final concentration of 50 ng/ml. Two weeks after injection, the mice were euthanized, and the Matrigel plugs were isolated. For quantification purposes, ϳ200 mg plugs were homogenized in Drabkin's solution (Sigma, D5941), and optical density was measured at 540 nm and then normalized to the weight of the plugs.

Statistical analysis
Data are presented as the mean Ϯ S.D. of three independent experiments. Two compared groups were analyzed by two-tailed Student's t test. Statistical significance is presented as: n.s., not significance; *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001.