High glucose-induced hyperosmolarity contributes to COX-2 expression and angiogenesis: implications for diabetic retinopathy

Background We tested the hypothesis that glucose-induced hyperosmolarity, occurring in diabetic hyperglycemia, promotes retinal angiogenesis, and that interference with osmolarity signaling ameliorates excessive angiogenesis and retinopathy in vitro and in vivo. Methods and Results We incubated human aortic (HAECs) and dermal microvascular endothelial cells (HMVECs) with glucose or mannitol for 24 h and tested them for protein levels and in vitro angiogenesis. We used the Ins2 Akita mice as a model of type 1 diabetes to test the in vivo relevance of in vitro observations. Compared to incubations with normal (5 mmol/L) glucose concentrations, cells exposed to both high glucose and high mannitol (at 30.5 or 50.5 mmol/L) increased expression of the water channel aquaporin-1 (AQP1) and cyclooxygenase (COX)-2. This was preceded by increased activity of the osmolarity-sensitive transcription factor Tonicity enhancer binding protein (TonEBP), and enhanced endothelial migration and tubulization in Matrigel, reverted by treatment with AQP1 and TonEBP siRNA. Retinas of Ins2 Akita mice showed increased levels of AQP1 and COX-2, as well as angiogenesis, all reverted by AQP1 siRNA intravitreal injections. Conclusions Glucose-related hyperosmolarity seems to be able to promote angiogenesis and retinopathy through activation of TonEBP and possibly increasing expression of AQP1 and COX-2. Osmolarity signaling may be a target for therapy.

Hyperglycemia has been reported to increase the endothelial expression of the proinflammatory proteins intercellular adhesion molecule (ICAM)-1 and vascular cell adhesion molecule (VCAM)-1, and to decrease production of nitric oxide, all through activation of the water channel protein aquaporin (AQP)1 [11]. Hyperosmotic stress inevitably accompanies hyperglycemia, and may therefore contribute to some aspects of hyperglycemiainduced vascular injury.
Against this background, we therefore examined whether hyperosmotic stress, at levels attainable in diabetes, induces COX-2 expression in human micro-and macrovascular endothelial cells. We then analyzed the functional impact of hyperosmotic stress and COX-2 expression on angiogenesis and explored underlying mechanisms, particularly with regard to the involvement of water channels and associated signaling molecules. Finally, we investigated the in vivo implications of such findings for diabetic retinopathy in a mouse model.

Methods
Materials d-glucose, D-mannitol and l-glucose (these two latter devoid of metabolic activities and used as a purely hyperosmolar controls), saccharose, and sodium chloride, were purchased from Sigma (St. Louis, MO, USA). The selective COX-2 inhibitor NS-398 was purchased from Calbiochem (Gibbstown, NJ, USA).

Animal care
Male C57BL/6 mice (body weight: 15 ± 4 g, age: 1 year; n = 21) and male D2.B6-Ins2 Akita diabetic mice, a model of type 1 diabetes developing retinal angiogenesis [23,24] (body weight: 11 ± 2 g, n = 21) were purchased from The Jackson Laboratories (Bar Harbor, ME, USA). All procedures were approved by our Institutional Ethics Committee for Animal Research. Investigations conformed to the Principles of Laboratory Animal Care formulated by the National Society for Medical Research and the Guide for the Care and Use of Laboratory Animals [NIH Publication 86- 23, 1985 revision].
Cell viability after treatments was assessed by cell morphology and cell count at phase-contrast microscopy, Trypan blue exclusion, and total protein content.

Transfection of AQP1 and tonicity enhancer binding protein (TonEBP)/nuclear factor of activated T cells (NFAT)5 siRNA
A pool of three different small interfering RNA (siRNA) oligonucleotides to each target (AQP1 and TonEBP/ NFAT5), as well as scrambled control siRNA were incubated with cells (2 × 10 5 /well) in transfection medium, as previously described [26]. After 24 h, fresh medium was added with or without HG (30.5 or 50.5 mmol/L), or HM (5.5 mmol/L glucose + 25 or 45 mmol/L mannitol), incubated for an additional 24 h, then subjected to Western analysis of COX-2, AQP1, TonEBP/NFAT5 and β-actin.

Tube formation assays
Tube formation assays were performed as previously described [26]. Before the assay, cells were incubated in low serum (2.5 %) endothelial basal medium (EBM, BD Biosciences) without endothelial growth factors, with concentrations of control glucose (5.5 mmol/L, control), high glucose (30.5 or 50.5 mmol/l), high mannitol (5.5 mmol/L glucose + 25 or 45 mmol/L mannitol) for 24 h, in the presence or absence of the selective cyclooxygenase(COX)-2 inhibitor NS-398 at 1-10 μg/ mL. In the assay testing the effect of siRNA to aquaporin (AQP)1 or TonEBP, cells were pretreated with high glucose or high mannitol for 24 h. At the end of incubations, cells were gently detached with 0.25 % trypsin-EDTA; trypsinized cells were then collected, counted and plated on Matrigel (at a density of 2 × 10 5 cells/50 μL Matrigel) for a further 16 h, in the presence or absence of control glucose, high glucose, high mannitol, with or without NS-398, or siRNA to AQP1 or TonEBP.

Cell migration assay
Cell migration was measured using the CytoSelect ™ cell migration assay (Cell Biolabs, San Diego, CA, USA), according to the manufacturer's instructions. Following a 24 h starvation period in 2 % fetal calf serum (FCS)/ endothelial cell growth medium (EGM), cells were resuspended at 4 × 10 5 cells/mL in EGM and then placed in a separate well of a 96-well feeder tray containing 150 μL of migration media (EGM with control glucose concentration (5.5 mmol/L, control), high glucose (HG 30.5 mmol/L) or high mannitol (HM HM, 5.5 mmol/L glucose + 25 mmol/L mannitol)). One hundred and fifty microliters of assay controls, consisting of EGM supplemented with 10 % FCS (positive control) and basal EGM (negative control), were also tested in triplicate.

Tissue preparation and histological evaluation of vessel density
After the induction of anesthesia, retinas from D2.B6-Ins2 Akita diabetic mice and age/sex-matched C57/ BL6 control mice were removed, embedded in optimalcutting-temperature (OCT) medium, frozen on dry ice, and stored at −70 °C until sectioning. For immunofluorescence staining, 5 μm thick sagittal and frontal cryosections were obtained by using a sliding cryotome. These sections were permeabilized, blocked for 30 min in PBS containing 1 % bovine serum albumin, and incubated for 1 h at 4 °C with purified primary antibody anti-CD31 or Cy3-conjugated anti-α-smooth muscle actin (ASMA). A non-immune IgG (Becton, Dickinson and Co., Franklin Lakes, New Jersey, USA) was used as the isotype control. After washing with PBS, retinal sections stained with anti-CD31 antibody were incubated with phycoerythrin (PE)-conjugated secondary antibody, washed, mounted, and viewed with an immunofluorescence microscope. Sections were counterstained with 4′, 6-diamidino-2-phenylindole (DAPI) to identify nuclei. For immunohistochemistry, retinas were embedded in five separate paraffin blocks. Paraffin-embedded limb sections were deparaffinized, rehydrated, and stained with a primary antibody against CD31. Sections were counterstained with hematoxylin and eosin for structural analysis and to identify nuclei. Arterioles and capillaries were counted in a blinded manner in five randomly selected fields at 10× magnification. Vascular images were captured by using an inverted light microscope (Olympus IX71), and analyzed by using the Image-J software (Media Cybernetics, Rockville, MD, USA). Vessel density was expressed as the number of arterioles or capillaries per mm 2 .

Statistical analysis
Two-group comparisons were performed by the Student's t test for unpaired values. Comparisons of means of ≥3 groups were performed by analysis of variance (ANOVA), and the existence of individual differences, in case of significant F-values at ANOVA, tested by Scheffé's multiple contrasts.

High glucose induces COX-2 expression in human endothelial cells through a hyperosmolar mechanism
Both short-term and longer-term exposure to high concentrations of glucose or mannitol, increased the expression of COX-2 protein, as shown by Western analysis ( Fig. 1a-g). Similar results were obtained by incubating HAECs and HMVECs with l-glucose and NaCl at equimolar concentrations (30.5 mmol/L), as a further osmotic controls ( Fig. 1h-i).

AQP1 mediates high glucose-induced COX-2 expression in endothelial cells
Since AQP1 regulates responses to hyperosmolarity in endothelial cells [28], we tested whether AQP1 is involved in hyperosmolarity-induced COX-2 expression. We used targeted disruption of AQP1 gene expression by siRNA technology, with 3 different siRNA species, to specifically block AQP1. AQP1 siRNA did not affect cell morphology or cell proliferation (Fig. 2a), but triggered a substantial (80 %) reduction in AQP1 protein expression (Fig. 2b). By contrast, serving as a sequence-unrelated control, scrambled siRNA did not exert any detectable effect. Proteins from AQP1 siRNA-transfected cells were then probed for COX-2 abundance. COX-2 expression in high glucose-and high mannitol-treated cells was substantially suppressed after transfection with siRNA against AQP1 (Fig. 2c-d). Taken together, these results showed that hyperosmolarity caused by exposure to high glucose or mannitol induces COX-2 expression through the activation of AQP1.

High glucose promotes angiogenesis through hyperosmolarity-induced expression of COX-2 and its transcription factor TonEBP/NFAT5
COX-2 is involved in human angiogenesis, although with mechanisms still incompletely understood [29]. To explore the involvement of hyperosmolarity in COX-2-induced angiogenesis, we seeded cells on Matrigel after incubation with high glucose or high mannitol in the presence or absence of siRNA against AQP1 or the COX-2 inhibitor NS-398. In this functional assay of angiogenesis, HAECs were treated with endothelial growth factor (EGF) as positive control for tubulization. EGF did induce tubule formation, and high glucose or high mannitol produced the highest number of network formations. Treatment with siRNA against AQP1 strongly reduced the number of these structures (Fig. 3A, C), while NS-398 reduced them to a lesser extent (Fig. 3B). TonEBP, also called nuclear factor of activated T cells (NFAT)-5, is a master transcription factor of genes responsive to hypertonic stress [30][31][32], TonEBP/NFAT5 siRNA did not affect cell morphology or cell proliferation (Fig. 3D), but triggered a substantial (80 %) reduction in TonEBP/NFAT5 protein expression (Fig. 3D). By contrast, serving as a sequence-unrelated control, scrambled siRNA did not exert any detectable effect. TonEBP/ NFAT5 silencing drastically decreased the network number and area formed by HAECs seeded in Matrigel, compared with those obtained from glucose and mannitol-treated cells (Fig. 3E). Cell migration induced by high glucose or high mannitol in a modified Boyden chamber system was also quantified with a fluorimetric assay. Migration of TonEBP/NFAT5-deficient endothelial cells towards high glucose and high mannitol was significantly inhibited (Fig. 3F), indicating a role of TonEBP/NFAT5 in mediating hyperosmolarity-induced endothelial migration and angiogenesis. TonEBP/NFAT5 expression and binding activity was increased in cells exposed to high glucose and mannitol ( Fig. 4a-d), and siRNA against TonEBP/NFAT5 24 h prevented the hypertonic induction of TonEBP/NFAT5 (Fig. 4a) and COX-2 proteins (Fig. 5a), indicating that TonEBP/NFAT5 expression and activity are required for the hyperosmolarity induction of COX-2. Taken together, these results showed that hyperosmolarity contributes to the stimulatory effect of COX-2 on endothelial angiogenesis mostly through the activation of upstream and downstream mediators of hyperosmolarity, such as AQP1 and TonEBP/NFAT5, respectively.
The promoter region of the COX-2 gene also contains consensus sequences for NF-κB [33][34][35][36], and high glucose induces the expression of some genes also by activating NF-κB [33,37]. In order to analyze the contribution of NF-κB to the hyperosmotic stress-induced expression of COX-2, we performed EMSA using NF-κB binding probe and nuclear extracts from osmotically stressed HAECs for 1-24 h. We found low, but detectable amounts of NF-κB DNA binding activities in unstimulated cells. After treatment with high glucose (1 and 3 h) and TNF-α (up to 24 h), but not with high mannitol or sodium chloride, we observed a time-dependent increase in NF-κB DNA-binding activity, suggesting that this transcription factor can cooperate to the expression of the COX-2 gene by high glucose, but not in response to high glucose-related hyperosmotic stress (Fig. 5b-c).

Discussion
Clinical consequences of type-1 diabetes result from the combination of a number of different factors, ranging from acidosis to metabolic changes, to biophysical derangements, such as hyperosmotic stress. Previous research has mostly focused on metabolic disorders induced by hyperglycemia and the underlying signaling pathways. Several lines of evidence presented here however indicate that a substantial contribution to the induction of vascular disease is given by the hyperosmolar stress related to hyperglycemia. We show here that concentrations of glucose attainable in diabetes induce COX-2 in human endothelial cells through an AQP1dependent hyperosmolar mechanism and, through this mechanism, promote angiogenesis. Specifically, in diabetic Ins2Akita mice, retinal COX-2 and AQP1 are upregulated concomitant with increased angiogenesis, suggesting the relevance of our observations in an in vivo setting where hyperglycemia is the main driver of microvascular disease. Finally, hyperosmolarity, through an AQP1-dependent mechanism appears to be causally involved in angiogenesis in vivo, and interruption of the AQP1 axis mitigates in vivo increased angiogenesis in response to high glucose. The effect of siRNA to AQP1 on COX-2 expression in endothelial cells exposed to high glucose and high mannitol. Serum-starved HAECs were transfected with Cy5-coniugated-siRNA to AQP1 or non-coniugated-siRNA to a scrambled sequence for 24 h and then treated with stimuli for 24 h. At the end of treatments, transfection efficiency was checked by observing the plates under fluorescence microscopy (a), while AQP1 gene knockdown efficiency was detected at the protein level by Western analysis (b). Protein expression for COX-2 was detected by Western analysis, with β-actin serving as a loading control (c). LPS (1 μg/mL) and tumor necrosis factor (TNF)-α (10 ng/mL) were here used as positive controls for the expression of COX-2. Representative blots (c) are shown. The results of scanning densitometry (n = 3 independent experiments) are expressed as arbitrary units in panel d, where columns and bars represent the mean ± SD (**, P < 0.01 mannitol-or glucose-treated vs control HAECs; §, P < 0.05 vs without AQP1-siRNA) This report expands our previous findings that hyperosmotic stress induces endothelial ICAM-1 and VCAM-1 expression [11] by showing the involvement of COX-2-dependent pathological phenomena. The current findings provide evidence that hyperosmolarity is a biophysical mechanism through which excessive angiogenesis may occur in diabetes. Additionally, our findings detail part of the molecular signaling through which hyperosmolarity promotes such effects.
The involvement of AQP1 in hyperosmolarity-induced endothelial tubulization and angiogenesis [38][39][40][41] is of biological importance. AQP1 is a water channel protein widely expressed in vascular endothelial cells, which, when activated, increases cell membrane water permeability [42][43][44]. AQP1 protein is strongly expressed in human and animal models of highly proliferating microvessels [43] and the chick embryo chorioallantoic membrane [44]. In human retinal epithelial cells, osmotic stress decreases AQP4 expression [45]. Reduction in AQP1 expression by targeted siRNA inhibits hyperosmolarity-dependent angiogenesis. In general, angiogenesis requires endothelial cell proliferation, adhesion and migration. We found a significant decrease in network number and areas in AQP1-silenced cultures, indicating a lost capacity to organize tubuli in networks. This is in line with previous reports showing that AQP1 plays a key role in endothelial cell migration, an early fundamental step in angiogenesis [10,46]. Data from the present study also provide a mechanistic explanation on how AQP1 is linked to endothelial cell TonEBP/NFAT5, a rel/NF-κB family member, has been identified as a transcription factor that in response to hyperosmolarity activates the transcription of osmosensing genes, including aldose reductase and protein kinase C delta [47], TNF-α and monocyte chemoattractant protein (MCP)-1 [30,[34][35][36]. In renal medullary epithelial cells, hypertonic stress increased COX-2 promoter activity in a luciferase reporter assay [37,48].  The involvement of TonEBP/NFAT5 in glucose-induced endothelial COX-2 expression. a Western analysis of the effect of siRNA to TonEBP/ NFAT5 on COX-2 expression in endothelial cells exposed to high glucose and high mannitol. Serum-starved human aortic endothelial cells (HAECs) were transfected with fluorescein isothiocyanate (FITC)-coniugated-siRNA to TonEBP/NFAT5 or siRNA to scrambled sequence for 24 h, and then treated with stimuli for 24 h with high glucose, or high mannitol, or equimolar concentration of sodium chloride. At the end of treatments transfection efficiency was checked by observing the plates under fluorescence microscopy. Protein expressions for cyclooxygenase (COX)-2 was detected by Western analysis, with β-actin serving as loading control. Data represent for each condition the mean ± S.D. from three separate experiments. **, P < 0.01 mannitol-or glucose-treated vs control HAECs; §, P < 0.05 vs without TonEBP/NFAT-siRNA. b The effect of hyperosmotic stress on NF-κB activation in human aortic endothelial cells. Electrophoretic mobility gel shift assay (EMSA), showing the effect of hypertonic stress on NF-κB activation. EMSA was performed by mixing the 32 P-oligonucleotide encoding for the NF-κB binding probe with nuclear extracts from osmotically stressed HAECs for 1-3 h. The electrophoretic run with nuclear protein extracts from unstimulated HAECs is shown in lane 1 and lane 7. The supershift analysis with a polyclonal rabbit antibody against p65, shows upward shift of binding complexes. Results are representative of three separate experiments. c Scanning densitometry of the shift bands (highlighted within the rectangle) of EMSA gels, expressed as arbitrary units of optical density. Data represent for each condition the mean ± S.D. from 3 separate experiments. *, P < 0.01 vs control HAECs Also, hypertonic saline had been reported to induce COX-2 in rat macrophages [49] and HUVECs [50], but in experimental conditions that do not reproduce the diabetic milieu. The participation of NFAT transcription factors in the activation of COX-2 expression had also been already demonstrated in several cell types [31,32,[51][52][53][54]. Indeed, previous research had shown that the angiogenic signaling involved in the induction of COX-2 by vascular endothelial growth factor (VEGF) in human endothelial cells requires the activation of NFAT proteins [51,55]. Furthermore, TonEBP/NFAT5 is required for the hypertonic induction of COX-2 in renal epithelial cells [31]. An upregulation of the NFAT-MMP-2 pathway had been also demonstrated in the development of metastatic osteosarcoma, where the secretion of active MMP-2 is under the transcriptional regulation of NFAT [32]. Here however we provide the first evidence that the induction of COX-2 in endothelial cells is largely hyperosmolarityrelated, and that TonEBP/NFAT5 is here involved, since (a) TonEBP/NFAT total abundance was increased in cells exposed to high glucose or high mannitol; (b) EMSA and Western analysis demonstrated a cytoplasm-to-nucleus redistribution of TonEBP/NFAT in endothelial cells; and (c) TonEBP/NFAT gene disruption prevented hyperosmolarity-related induction of COX-2. Of note, our data show that at least another signal transduction pathways . c The extent of angiogenesis was determined by measuring total vessels (arterioles and capillaries) density in light microscopy sections. N = 2 eyes per animal, with each experimental group consisting of n = 7 mice. The bar graph represents for each value the mean ± S.D. from 3 separate experiments. **, P < 0.01 vs C57BL/6 control mice. d, e: Western analysis of COX-2 and AQP1 expression in retinas from D2.B6-Ins2 Akita mouse and from sex-and age-matched (1 year-old) C57BL/6 non-diabetic control mice, with β-actin serving as a loading control. N = 2 eyes/group, with each group consisting of n = 7 mice. The bar graph represents for each value the mean ± S.D. from three separate experiments. **, P < 0.01 vs C57BL/6 control mice  Fig. 8 The involvement of AQP1 in COX-2 expression in the Ins2Akita and C57BL/6 mice. a Western analysis of COX-2 and AQP1 expression in retinas from from D2.B6-Ins2 Akita and sex-and age-matched (1 year-old) C57BL/6 non-diabetic control mice, after the intravitreal injection of saline (sham), or siRNA against scrambled sequence (siRNA-scr) or siRNA against AQP1 (siRNA-AQP1), with β-actin serving as a loading control. b n = 2 eyes per animal, with each experimental group consisting of n = 7 mice. The bar graph represents the means ± S.D. from three separate experiments. **, P < 0.01 vs C57BL/6 control mice °°, P < 0.01 vs sham and siRNA-scr. Glyco AQP1 represents the glycosylated form of AQP1 able to insert itself in the plasma membrane activated by high glucose, but with lesser relevance to COX-2 expression and angiogenesis, such as the NF-κB pathway, is-conversely-not substantially induced by high mannitol, indicating a larger dependence of NF-κB activation on metabolic effects of glucose. A number of studies have shown that hyperosmolarity (as tested by high mannitol) does not influence angiogenesis and/or COX-2 expression in endothelial cells [55,56]. Culture conditions used in these studies may not reflect those used in our study. Our data show more robust effects when using glucose concentrations as high as 50.5 mM, or with lower concentrations but for prolonged times, which may more closely mimic diabetic pathophysiology. Concentrations of glucose outside the range tested here might exert different effects, therefore not completely ruling out the additional involvement of mechanisms different from hyperosmolarity. Nevertheless, our data clearly indicate that hyperosmolarity is an important part of the mechanisms through which chronic exposure to high glucose may promote angiogenesis and cardiovascular complications of diabetes.
We recognize some limitations of our study. Main ones are that (a) there is limited translability of murine data to humans; (b) molecular mechanisms by which TonEBP/ NFAT operates on the COX-2 promoter are still incompletely understood; (c) we still need to close the circle of the in vivo causality demonstration involving the transcription factor TonEBP/NFAT5 as an additional osmosignaling effector by showing that the in vivo interruption of the TonEBP signaling mitigates the in vivo increased angiogenesis in response to high glucose. Such experiments are complicated by the lack of available transgenic mouse models of TonEBP knockouts. Such limitations might however be overcome in the future through the prolonged administration of drugs targeting TonEBP activation.
Nonetheless, data provided here lend support to the hypothesis that hyperosmolarity is an important mechanism of vascular disease in diabetes and a potential target for future therapies, with several potential practical implications. In principle, promotion of inflammation and angiogenesis by COX-2 through a hyperosmolarityrelated mechanism may have a role not only in microvascular disease (proliferative retinopathy), as shown here, but also in macrovascular disease (e.g., plaque rupture), both increased in diabetes. The unraveling of the hyperosmolarity-related signaling, as sketched in Fig. 9, pointing both to the sensing molecule AQP1 and the effector molecule COX-2, allows identifying potential pharmacological targets to reduce hyperosmolar stressrelated excessive angiogenesis. In addition, the in vivo expressions of AQP1 and COX-2 might serve as new markers for diabetic complications. The in vivo implications of these observations are therefore worth further assessment. Fig. 9 A scheme of the putative signaling pathway through which hyperosmotic stress contributes to some of the vascular events in diabetes, based on findings from the present study. In the plasma membrane, hyperglycemia activates "osmosensing" structures, such as aquaporin (AQP)-1, which can sense osmolarity changes; and an "osmosignaling" pathway, involving the transcription factor tonicity enhancer binding protein (Ton/EBP), which can transmit the signal towards effector regulatory sites in the nuclei, able to further promote the expression of proinflammatory genes such as adhesion molecules and cycloxygenase (COX)-2