Tumour‐secreted miR‐9 promotes endothelial cell migration and angiogenesis by activating the JAK‐STAT pathway

To mimic some of the activation steps in angiogenesis, we established tumour‐endothelial co‐cultures, in which tumour and endothelial cells are separated by a transwell membrane insert (Figure 1A). We paired five human tumour cell lines (H1299, non‐small cell lung cancer (NSCLC); MDA435, melanoma; Panc1, pancreatic cancer; SF‐539, glioblastoma; HM7, colorectal cancer) with microvascular endothelial cells isolated from normal tissues matching each tumour type examined (Figure 1B). After 24 h, we profiled endothelial miRNAs in the presence or absence of tumour cells using Taqman micro fluidic cards (Figure 1A). Hierarchical clustering indicated that different endothelial cells have surprisingly distinct miRNA expression (Figure 1B), while tumour cell stimulation only affected a subset of miRNAs (Supplementary Table 1). Volcano plot of rank products analysis showed a group of most significantly differentially expressed miRNAs among the five endothelial cell types (Figure 1C), the majority of which are highly conserved across species and were confirmed by quantitative RT–PCR (Figure 1D; Supplementary Figure 1A). Importantly, these miRNAs appeared to be selectively activated by tumour cells, since SK23 melanoma cells, but not normal skin melanocytes or fibroblasts, promoted miRNA expression (Figure 1E).

To rule out the possibility that our observations may be restricted to a few tumour cell lines, we screened a large panel of NSCLC lines and found that the vast majority were able to induce miRNA changes in endothelial cells, although to different degrees (Supplementary Figure 1B). Furthermore, human umbilical vein endothelial cells (HUVECs), when co‐cultured with human or mouse tumour cells, also exhibited miRNA changes comparable to those observed in microvascular endothelial cells (Supplementary Figure 1C). Therefore, our observations are consistent with a general endothelial miRNA upregulation induced by tumour cells, which may play a role in tumour angiogenesis.

To follow‐up our observations with tumour‐endothelial co‐cultures, we tested conditioned media from several tumour cell lines and found that they also resulted in enhanced expression of miRNAs in endothelial cells (Supplementary Figure 2A), raising the possibility that soluble factors are responsible for this effect. Recombinant VEGF had no effect on the miRNAs identified in this study. Additionally, wild‐type (C2P) and VEGF‐null (G5) mouse fibrosarcoma cells (Dong et al, 2004) had similar abilities to induce these miRNAs in HUVECs (Figure 2A).

We initially considered the possibility that induction of miRNAs in endothelial cells might be mediated by growth factor(s) other than VEGF. We tested several angiogenic factors (bFGF, TNFα, HGF, PDGF‐A, PDGF‐B, PDGF‐C and angiopoietin‐1) and found that none of these was able to alter the expression of these miRNAs (Supplementary Figure 2B).

We also undertook an initial biochemical characterization of such activity. We chose SK23 tumour cells as a source because they elicited a strong induction of endothelial miRNAs (Figure 1E) and could also be easily maintained in serum‐free medium. The inducing activity was completely retained by Amicon filters with up to 50 kDa MW cutoff (Supplementary Figure 2C), which indicates that this factor(s) has a relatively large molecular mass. We then subjected the tumour cell conditioned media to anion exchange chromatography on a Hi‐Trap Q column, but the miRNA inducing activity did not bind to the column and was almost completely recovered in the flow‐through (Figure 2B). In contrast, a heparin‐sepharose (HS) column bound the activity, which was eluted by 0.5 M NaCl (Figure 2C). These results prompted us to hypothesize that endothelial miRNAs could be stimulated by tumour‐derived soluble factors with heparin‐binding properties.

However, in conflict with this hypothesis, we did not detect any significant changes in endothelial pri‐miRNA or pre‐miRNA in response to conditioned medium or HS fractions. Therefore, to determine the source of upregulated miRNAs, we knocked down Drosha, which is required for miRNA biogenesis. Surprisingly, Drosha knockdown in tumour cells impaired miRNA induction (Supplementary Figure 2D). Accordingly, when we individually inhibited miR‐9 or miR‐183 in SK23 melanoma cells, tumour cells were less capable of inducing endothelial miR‐9 or miR‐183, respectively (Supplementary Figure 2D). The level of inhibition induced by siDrosha or antagomirs on endothelial miRNAs was ∼50%. This incomplete inhibition may be due to the residual miRNAs in knockdown conditions. These results suggest that the observed increases in endothelial miRNAs largely originate within tumour cells in the co‐culture system. Indeed, we were able to detect miRNAs in tumour conditioned media by RT–PCR (Figure 2D). Considering that our data suggest these miRNAs are in relatively large molecular weight complexes, we hypothesized that tumour‐secreted MVs might be the carriers (Balaj et al, 2011; Lee et al, 2011; Chen et al, 2012), transporting miRNAs into endothelial cells. In agreement with this hypothesis, isolated MVs from either SK23 or 4T1 cells induced miRNAs in endothelial cells (Figure 2E). To further verify that the miRNAs were indeed assembled within the MVs, we treated purified MVs with RNase and found that miRNAs were substantially protected from RNase digestion. Pre‐incubation with 0.5% SDS, but not with proteinase K, significantly reduced miRNA levels after RNase treatment (Supplementary Figure 2E). To determine whether MVs are truly taken up by recipient endothelial cells, we labelled isolated MVs with PKH67 green fluorescence and visualized them while internalized into endosome‐like structures by HUVECs (Figure 2F, left panel). Similar results were observed when we incubated SK23 cells with DiI and co‐cultured the labelled cells with HUVECs (Figure 2F, middle panel). Furthermore, we verified that purified MVs from DiI‐labelled tumour cells are also transferred into HUVECs (Figure 2F, right panel). Following miR‐9 knockdown in SK23 cells, we observed a concurrent decrease in miR‐9 levels in purified MVs or HUVECs co‐cultured with MVs (Supplementary Figure 2F). Taken together, these data demonstrate that miRNAs are secreted by tumour cells and are delivered into endothelial cells via a specific population of MVs that bind heparin.

To determine whether MVs carrying miRNA contribute to angiogenesis, we assessed endothelial cell migration in real time using the xCELLigence system. Consistent with previous reports (Skog et al, 2008; Grange et al, 2011), both SK‐23 tumour conditioned media and isolated MVs led to enhanced migration of HUVECs (Figure 3A). Following knockdown of Drosha in tumour cells, tumour‐derived MVs showed reduced ability to promote endothelial cell migration, indicating that miRNAs in MVs play an important role in this process (Figure 3B).

To identify functionally important miRNAs, we overexpressed each miRNA found to be upregulated in endothelial cells and determined its ability to affect endothelial cell motility. We found that the vast majority of miRNAs had no obvious effect (data not shown). However, miR‐9 induced striking morphological changes in HUVECs, including elongated cellular shape, fewer stress fibres and increased lamellipodia at the cell edge (Figure 3C). These observations suggested a potential role for miR‐9 in modulating endothelial cell motility. To explore these events in greater details, we performed gain‐ and loss‐of‐function studies in cell migration assays. We found that overexpression of miR‐9 in HUVECs results in enhanced cell motility (Figure 3D), while miR‐9 knockdown significantly inhibits MV‐induced cell migration (Figure 3E). Additionally, inhibition of miR‐9 in SK23 cells partially suppressed migration of co‐cultured endothelial cells in transwell assays (Supplementary Figure 3A). We also evaluated the role of miR‐9 in an angiogenic sprouting assay. Expression of miR‐9 markedly increased sprouting angiogenesis of HUVECs, suggesting an increase in the number of tip cells and/or branching activities of endothelial cells (Figure 3F). We tested the hypothesis that miR‐9 may indirectly affect endothelial cells by regulating tumour cell proliferation. However, we found no changes in tumour cell growth in the presence of anti‐miR‐9 or siDrosha (Supplementary Figure 3B). Also, the composition of the MVs did not appear to be markedly affected by miR‐9 inhibition, as assessed by antibody arrays to evaluate a panel of angiogenesis‐related proteins (Supplementary Figure 3C). Interestingly, angiogenic factors such as IL8 and VEGF were readily detectable in purified SK23 MVs, which may account for the incomplete inhibition of endothelial cell migration by anti‐miR‐9. These findings indicate that tumour cells activate endothelial cell migration at least in part through miR‐9.

Focusing on miR‐9, we conducted a series of in‐vivo experiments to determine whether this miRNA affects tumour angiogenesis. Initially, we extracted miRNAs from plasma of naïve or tumour‐bearing mice. Subcutaneous implantation of HM7 human colorectal tumours resulted in elevated miR‐9 levels in the plasma, suggesting that tumour cells also secrete miRNAs in vivo (Figure 4A). Next, we intratumorally injected miR‐9 antagomirs, which specifically decreased miR‐9 but not miR‐126 in HM7 tumours (Figure 4B). We observed a statistically significant delay in tumour growth (Figure 4C), in spite of the fact that the knockdown of miR‐9 was only ∼50%. Tumour angiogenesis, as assessed by CD31 immunostaining, was significantly decreased in the presence of miR‐9 antagomirs (Figure 4D and E). To specifically quantify functional tumour vasculature, we performed FITC‐lectin perfusion before collecting tumour tissues. Both the numbers of perfused vessels and FITC‐labelled vessel areas were significantly reduced in anti‐VEGF or anti‐miR‐9 treated tumours, suggesting that the function of tumour vessels was compromised (Supplementary Figure 4A). We found that inhibition of miR‐9 significantly reduces tumour cell proliferation without perturbing apoptosis, as assessed by Ki67 and cleaved caspase3 staining, respectively (Supplementary Figure 4B). A likely explanation for the decreased tumour cell proliferation is inhibition of tumour angiogenesis. Indeed targeting miR‐9 did not affect HM7 growth in cell culture (Supplementary Figure 3B). The ability of miR‐9 antagomirs to inhibit tumour growth was confirmed in the LLC, a murine lung carcinoma model that, in agreement with previous studies (Shojaei et al, 2007), was only moderately responsive to anti‐VEGF treatment (Figure 4F). Interestingly, the combination of miR‐9 antagomirs with anti‐VEGF showed a clear trend towards reduced tumour progression compared to single agent treatment, although it did not achieve statistical significance (P=0.06). We conclude that miR‐9 plays a role in promoting tumour angiogenesis in vivo.

To investigate the molecular mechanisms by which miR‐9 enhances endothelial cell migration, we undertook an unbiased approach by overexpressing miR‐9 in HUVECs and examining the major signalling pathways involved in endothelial cell function. Among 46 components tested, cAMP response element binding (CREB) and signal transducer and activator of transcription 1 (STAT1) exhibited the most marked increases in phosphorylation, while c‐Jun showed noticeable decreases (Figure 5A). Using a different antibody, we confirmed that phospho‐STAT1 is augmented following miR‐9 transfection in human or mouse endothelial cells. Phospho‐STAT3 was also dramatically elevated (Figure 5B). In agreement with these findings, MVs or tumour conditioned media increased STAT1 and STAT3 phosphorylation in HUVECs (Figure 5B), a finding that was further verified by ELISA (Figure 5C). Other STAT members were either unchanged or undetectable. Upstream of STAT1 and STAT3, JAK2 (Janus Kinase 2) was phosphorylated in basal conditions in endothelial cells. We also found that JAK1 phosphorylation is upregulated upon miR‐9 overexpression (Figure 5B). Suppressor of cytokine signalling 5 (SOCS5), a negative regulator of JAK‐STAT pathway, was substantially decreased in response to miR‐9, MVs or conditioned media (Figure 5B).

In addition, we predicted the gene targets of miR‐9 in silico (Xiao et al, 2009) using four independent algorithms: miRanda (Betel et al, 2008), miRtarget2 (Wang and El Naqa, 2008), PicTar (Krek et al, 2005) and TargetScan (Lewis et al, 2005). Each program yielded a large number of genes. However, the 93 top candidates were common to all four methods (Figure 5D). Interestingly, SOCS5 was identified as one of the top candidates, having two putative miR‐9 binding sites in its 3′ UTR. Similar to miR‐9, the two predicted binding sites, separated by ∼80 bases, are highly conserved (one shown in Figure 5E). Ectopic expression of miR‐9 suppressed the activity of a luciferase reporter in frame with wild‐type SOCS5 3′ UTR by ∼40%, but not SOCS5 3′ UTR with mutated miR‐9 binding sites, suggesting that SOCS5 gene is a direct target of miR‐9 (Figure 5F). Taken together, we conclude that miR‐9 targets SOCS5 in endothelial cells and activates the JAK‐STAT signalling pathway.

In addition to miR‐9, multiple miRNAs identified in our screening were able to activate STATs, at least to various extents (Supplementary Figure 5A). Therefore, JAK‐STAT appears to be a major signalling pathway regulated by miRNAs in endothelial cells. These findings prompted us to determine whether interfering pharmacologically with JAK‐STAT signalling could impair miR‐9 induced cell migration and tumour angiogenesis. First, we showed that JAK proteins play an essential role in mediating miR‐9 induced STATs activation by knocking down JAK1 or JAK2. When we combined both siRNAs, STAT1 and STAT3 phosphorylations triggered by miR‐9 were completely abolished (Figure 6A), indicating an indispensible role of JAK kinases in mediating miR‐9 induced STATs activation.

GNE‐372 is a potent JAK2 inhibitor (manuscript in preparation) that specifically targets JAK family kinases including JAK2 and JAK1 (Figure 6B). As expected, this compound effectively inhibited STAT1 and STAT3 phosphorylation in miR‐9 transfected endothelial cells (Figure 6C). We used ELISA to quantify its efficacy in cell‐based assays. GNE‐372 inhibited pSTAT1 with an IC50 of 0.11 μM and pSTAT3 with an IC50 of 0.08 μM (Figure 6D). Addition of the inhibitor to HUVECs overexpressing miR‐9 significantly reduced cell migration (Figure 6E). Also, GNE‐372 inhibited endothelial cell proliferation and survival (Supplementary Figure 5B and C). To assess the impact of JAK inhibition on tumour growth in vivo, we treated mice bearing LLC tumours with GNE‐372. Such treatment significantly inhibited tumour growth relative to vehicle‐treated controls, and was comparable to the anti‐VEGF cohorts (Figure 6F).

Primary microvascular endothelial cells from human brain, colon, and pancreas and HUVECs were purchased from Lonza (Walkersville, MD). Human dermal and lung microvascular endothelial cells were from Cell Systems (Kirkland, WA). Endothelial cells were cultured in EGM‐2 or EGM‐2MV medium (Lonza). Tumour cell lines were obtained from the Genentech cell line repository (gCell) and were cultured in DMEM (high glucose) supplemented with 10% fetal bovine serum (Invitrogen). The HM7 cell line is a variant of LS 174T human colorectal carcinoma cell line, as previously described (Bresalier et al, 1991; Warren et al, 1995). Tumour‐endothelial co‐culture experiments were performed in six‐well plates with 8‐micron inserts (BD Biosciences). miRNA mimics (Applied Biosystems) and siRNA sequences (Dharmacon) were transfected with Lipofectamine RNAiMAX reagent (Invitrogen).

Total RNA was prepared with mirVana miRNA isolation kit (Ambion) according to manufacturer's protocol. miRNA in 500 ng total RNA was subjected to reverse transcription using the Megaplex RT primer pools and the Taqman microRNA reverse transcription kit (Applied Biosystems). The resultant cDNA was pre‐amplified using Megaplex PreAmp primers and TaqMan PreAmp master mix (Applied Biosystems), and then loaded into the Taqman array microRNA cards. The real‐time PCR was run on the Applied Biosystems 7900HT machine. Relative expression levels of each miRNA were normalized to U6 snRNA. Expression of individual miRNAs was quantified by TaqMan miRNA assays (Applied Biosystems) following manufacturer's instructions. U6 snRNA was used as the endogenous control for all experiments. At least three biological replicates were included for each condition.

Real‐time monitoring of endothelial cell migration was performed using the xCELLigence system with the CIM‐Plate 16 (Roche). The upper chamber was coated with 20 μg/ml Fibronectin (BD Biosciences) and seeded with 50 000 HUVECs. When cells migrated through the membrane into the bottom chamber in response to attractants, they contacted and adhered to the electronic sensors, resulting in an increase in impedance. The cell‐index values reflecting impedance changes were automatically and continuously recorded every 15 min.

Twenty hours after transfection, HUVECs were mixed with Cytodex microcarrier beads (Sigma‐Aldrich) in a ratio of 10⁶ cells per 2500 beads and incubated for 4 h at 37°C. Coated beads were then cultured in a six‐well dish overnight. The following day, ∼200 coated beads were dissolved in a solution of 2 mg/ml fibrinogen (Sigma‐Aldrich) in EGM‐2, and added to 0.625 U/ml of thrombin (Sigma‐Aldrich) in one well of a 24‐well plate. Skin fibroblasts (D551) were then plated on top of the clot and incubated with D551 conditioned medium and EGM‐2 (3:1). HUVEC sprouts were visualized by immunostaining with Alexa Fluor 488 phalloidin (1:100) and Hoecsht 33258 (1:1000) (Invitrogen) overnight at 4°C. Image Xpress Micro was used for capturing images and HUVEC sprouting was analysed in MetaXpress software.

Tumour cells were cultured in serum‐free DMEM. Conditioned medium was collected after 48 h and concentrated 10‐fold. MVs were isolated using two independent methods. For differential centrifugation, conditioned medium was filtered through a 0.22‐μm membrane and centrifuged for 30 min at 16 500 g. Supernatants were filtered again and MVs were pelleted by ultracentrifugation for 2 h at 110 000 g. In the other approach, ExoQuick precipitation reagent (System Biosciences) was mixed with concentrated conditioned medium and MVs were precipitated by centrifugation.

Approximately 80% confluent SK23 cells were incubated in serum‐free medium for 3 days. Conditioned media were collected and concentrated 10‐fold as starting material for chromatography. The heparin‐sepharose column (Hi‐Trap, HS, 5 ml) was pre‐equilibrated with 20 mM Tris, pH 7.4. The Hi‐Trap Q sepharose column, 1 ml (GE Healthcare) was pre‐equilibrated with 20 mM Tris, pH 8.0. Samples were injected into the column and bound material was eluted with a linear gradient of NaCl. Aliquots of fractions were tested on HUVECs for miR‐9 expression.

Cells were lysed in RIPA buffer (Tris pH 7.4 50 mM, NaCl 150 mM, NP‐40 1%, SDS 0.1%, EDTA 2 μM) containing proteinase inhibitors (Roche) and phosphatase inhibitors (Sigma). The cell lysates (20 ug protein) were subjected to SDS–PAGE and western blot. Antibodies against the following proteins were used: pTyr‐701 STAT1, STAT1, pTyr‐705 STAT3, STAT3, pTyr‐1022/1023 JAK1, JAK1, pTyr‐1007/1008 JAK2, JAK2, PARP, Caspase‐3, Actin (Cell Signaling Technology); SOCS5 (Santa Cruz Biotechnology). Human phospho‐kinase array and human angiogenesis antibody array were performed following manufacturer's instructions (R&D Systems). For phosphor‐kinase array, ∼300 μg HUVEC lysates were analysed 48 h after transfection of control oligo or miR‐9. For human angiogenesis array, MVs were purified from 5 × 10⁶ SK23 cells transfected with scrambled sequence or anti‐miR‐9, and subjected to the antibody array.

For miR‐9 target validation, 3′ UTR segment of human SOCS5 was cloned into a mammalian expression vector with dual luciferase reporter system (GeneCopoeia). COS‐7 cells were transfected in six‐well plates using Lipofectamine 2000 (Invitrogen). Transfections were performed using 1 μg dual luciferase reporter plasmids and a final concentration of 10 nM synthetic miRNA mimics (Applied Biosystems). Forty‐eight hours after transfection, dual luciferase assays were performed using Luc‐Pair miR luciferase assay kit (GeneCopoeia) according to manufacturer's instructions. Firefly luciferase activity was first normalized to Renilla luciferase expression control. For each reporter construct, the normalized value for miR‐9 transfection was then normalized to the value obtained from the same reporter construct co‐transfected with control miRNA. Mean values, standard deviations and Student's t‐test were calculated from seven independent transfections for each condition. The QuikChange II XL site‐directed mutagenesis kit (Agilent) was used to mutate miR‐9 binding sites at SOCS5 3′ UTR.

Tumour cells (1 × 10⁶) were subcutaneously implanted in the dorsal flank of BALB/c Nude mice. Anti‐VEGF mAb B‐20.4.1 (Liang et al, 2006) or anti‐Ragweed control was IP injected at the dose of 10 mg/kg twice weekly. Antagomirs (Exiqon) were mixed with in vivo‐jetPEI transfection reagent (Polyplus) and injected intratumorally at 2 μg per mouse twice a week. JAK2 inhibitors GNE‐372 (100 mg/kg) or vehicle control (0.5% methylcellulose/0.2% Tween‐80) were administered twice daily by oral gavage. Tumour volumes (10 animals per group) were measured with digital caliper and calculated as length × width² × 0.52. To quantify perfused tumour vessels, fluorescein‐labelled lectin (Vector Lab) was injected intravenously (2 mg/kg) 10 min before collecting tumours. All animal protocols were approved by the Genentech animal care and use committee.

Tumours from six mice per treatment group were collected and embedded in OCT blocks. Tissues were cryo‐sectioned to 16 μm thickness on Leica CM3050S, and stained with CD31 antibody (BD Biosciences). Images were acquired with Zeiss AxioImager Z1 fluorescence microscope controlled by TissueFAXS software. Image files were loaded into the TissueStudio analysis package (v1.5, Definiens). Necrotic tissues and staining artifacts such as skin tissues and folds were automatically identified and excluded based on nuclei staining. Blood vessel density was calculated as the ratio of CD31‐positive pixels to the total viable tumour area. Statistical analysis was performed in JMP software.

Rank products analysis was performed on the miRNA profiling results to identify most significantly changed miRNAs in response to tumour stimulation, and a permutation approach was used to compute the empirical P‐values (Breitling et al, 2004). The results were graphed in volcano plot, and miRNAs with P<0.001 were highlighted. In all experiments, comparisons between two groups were based on two‐sided Student's t‐test and one‐way analysis of variance (ANOVA) was used to test for differences among more groups. P‐values of <0.05 were considered statistically significant.