Müller glia cells regulate Notch signaling and retinal angiogenesis via the generation of 19,20-dihydroxydocosapentaenoic acid

sEH protein and activity were detected in retinas from adult and postnatal wild-type mice (Fig. 1 A), and although the highest sEH activity was detected in the liver, activity in the retina was greater than that detected in either the spleen or lung (Fig. 1 B). To study the importance of the sEH in sprouting angiogenesis in vivo, we focused on retinal angiogenesis during the first postnatal week. sEH^−/− mice displayed a significant delay in the radial extension of the vascular plexus from the optic nerve to the periphery at postnatal days 2 (P2) and 5, as well as fewer branch points (Fig. 1 C). The delay in retinal angiogenesis in sEH^−/− retinas at P5 was associated with a significant reduction in endothelial cell proliferation (BrdU incorporation) at the angiogenic front (Fig. 1 D) as well as with fewer tip cells and filopodia (Fig. 1 E).

Because all of these processes are highly dependent on the Notch pathway (Hellström et al., 2007; Tammela et al., 2008), our findings suggested that Notch signaling may be activated in the sEH^−/− animals. We therefore studied the expression of the Notch receptor and ligands as well as the Notch downstream targets Hes1 and Hey1. Retinal Hey1 increased from P2 to P7 in sEH^−/− retina, whereas Hes1 levels were elevated in sEH^−/− retina on P2 and P5 but decreased by P7 (Fig. 1 F). Notch1 was up-regulated in sEH^−/− mice immediately after birth (P0) but RNA levels then decreased (after P2). A transient increase in retinal delta-like ligand (Dll) 1 was detected on P5 but there were no significant changes in Dll4, which is also a target of activated Notch receptors (Phng and Gerhardt, 2009). A more detailed analysis of purified (CD31 beads) sEH^−/− retinal endothelial cells confirmed the significant up-regulation of Dll4, Hes1, and Hey1 in 5-d-old sEH^−/− mice (Fig. 1 G). Although coverage of the retina by the primary vascular layer was almost complete at day 7 in both genotypes, delayed angiogenesis persisted so that the formation of the secondary capillary plexus was also attenuated at P9 in sEH^−/− mice (Fig. 1 H).

Different approaches were taken to ensure that the phenotype observed was related to the lack of sEH activity during retinal development. First, floxed sEH mice (sEH^fl/fl) mice were bred with CreERT2 mice to generate an inducible sEH knockout strain (sEH^iKO). Then sEH^iKO pups were treated (on P1) with tamoxifen for 4 d, a procedure which decreased sEH expression by ∼80% from day 3 onwards (Fig. 2 A). The acute postnatal down-regulation of the sEH also attenuated retinal angiogenesis and increased retinal expression of Dll4, Hes1, and Hey1 (Fig. 2, B–D). The apparent difference in the sEH^−/− and sEH^iKO retinas can be most likely attributed to the fact that the sEH was still expressed at day 2 in the latter group of animals. The mammalian sEH protein is a homodimer, and each monomer consists of an N-terminal domain which displays lipid phosphatase activity and a larger C terminal which possesses classical α/β-hydrolase activity (Cronin et al., 2003; Newman et al., 2003). Therefore, in a second approach to ensure that the defects observed in the sEH^−/− mice were due to the loss of epoxide hydrolase activity, newborn wild-type mice were treated with trans-4-[4-(3-adamantan-1-ylureido)cyclohexyloxy]-benzoic acid (t-AUCB), a specific sEH inhibitor which does not affect the activity of the lipid phosphatase domain (Hwang et al., 2007). Daily inhibitor treatment from P1 onwards significantly decreased retinal vascularization (assessed at P5) compared with vehicle-treated littermates, and this delay was accompanied by a significant increase in the expression of Dll4, Hes1, and Hey1 (Fig. 2, E–G).

In retinal cross sections, the sEH was detected in the ganglion cell, inner plexiform, and outer plexiform layers but did not colocalize with the Isolectin B4 staining of the vasculature. Rather, the sEH colocalized with glutamine synthase (GS; Fig. 3 A), as well as vimentin (Fig. 3 B), aquaporin 4 (AQP-4), and, to a certain extent, glial fibrillary acidic protein (GFAP; Fig. 3 C). In whole mount preparations, the sEH was detected in short AQP-4–positive processes surrounding the vasculature but not in the vessels themselves—i.e., consistent with expression in Müller cell processes (Fig. 3 D). Indeed, Müller cells freshly isolated from adult retinas displayed the typical morphology with multiple end-feet and expressed the sEH protein (Fig. 3 E). Although the morphology of the Müller cells in culture was markedly different from that of freshly isolated cells, they continued to express the sEH, as well as AQP-4 and the Müller cell marker cellular retinaldehyde binding-protein (CRALBP; Bunt-Milam and Saari, 1983; Fig. 3 F).

To target sEH expression in Müller cells, sEH^fl/fl mice were crossed with platelet-derived growth factor receptor α polypeptide (Pdgfra)–cre deleter mice (Roesch et al., 2008). The resulting Müller cell–specific sEH knockout mice (sEH^ΔMC) showed a greatly reduced expression of the sEH in the inner and outer plexiform layers of the retina (Fig. 4 A), decreased retinal epoxide hydrolase activity (Fig. 4 B), and reproduced the delay in angiogenesis (Fig. 4, C–E) as well as the altered tip cell morphology (Fig. 4, F–H) of the sEH^−/− mice. These phenomena were also linked to altered Notch signaling, as Hes1 and Hey1 expression were also increased in retinas from 5-d-old sEH^ΔMC mice (Fig. 4 I).

Several studies have highlighted the role of the astrocyte framework as a template for the primary plexus outgrowth as well as a source of angiogenic growth factors (Stone and Dreher, 1987; Gerhardt et al., 2003). Given that the sEH is expressed in astrocytes (Rawal et al., 2009) and that the sEH was partially colocalized with GFAP, we generated mice with astrocyte-specific deletion of the sEH (sEH^fl/fl mice crossed with GFAP-cre deleter mice). Astrocyte-specific deletion of the sEH (Fig. 5 A) resulted in the development of a less dense upper capillary layer characterized by fewer branching points in central areas and at the leading edge (Fig. 5 B). The marked delay in radial extension observed in global sEH^−/− mice was, however, not observed. In addition, tip cell numbers and filopodia length were comparable in sEH^fl/fl and GFAP-sEH mice (Fig. 5 C), as were the expression of Hes1 and Hey1 (Fig. 5 D). These data indicated that the sEH expressed in Müller cells played a more prominent role in regulating angiogenesis than the sEH in astrocytes.

Müller cells are a rich source of angiogenesis-promoting growth factors, in particular vascular endothelial cell growth factor (VEGF) A (Pierce et al., 1995; Stone et al., 1995). Therefore, we compared the effects of conditioned medium (CM) from wild-type and sEH^−/− Müller cells on endothelial cell proliferation. CM from sEH-expressing Müller cells potentiated endothelial cell proliferation, whereas the CM from sEH^−/− Müller cells exhibited a weaker effect (Fig. 6 A). There were no differences in VEGFR1, VEGFR2, or VEGFR3 expression in retinas from wild-type and sEH^−/− mice over the first postnatal week (unpublished data). VEGF levels were also similar in whole retinas from wild-type and sEH^−/− mice up to day 9 (Fig. 6 B), but the generation of VEGF was reduced in Müller cells and in CM from sEH^−/− mice (Fig. 6, C and D). The expression of the anti-angiogenic pigment epithelium-derived factor (Dawson et al., 1999) was not different in CM from wild-type and sEH^−/− Müller cells. Despite the apparent difference in VEGF secretion, VEGF neutralizing antibodies elicited similar effects on endothelial cells treated with wild-type-CM or sEH^−/−-CM (Fig. 6 E).

However, a soluble factor was responsible for the observed phenotype, as when sEH^−/− Müller cells were co-cultured with, but kept physically separated from, endothelial cells (Transwell chambers), endothelial cell proliferation (assessed by BrdU incorporation) was attenuated. Moreover, the reexpression (adenoviral transduction) of the sEH in Müller cells normalized endothelial cell proliferation (Fig. 6 F). Similarly, the expression of Hes1 and Hey1 increased in endothelial cells cultured with sEH^−/− versus wild-type Müller cells (Fig. 6 G), and a rescue experiment, in which the sEH was reintroduced into sEH^−/− Müller cells, reversed these effects.

To determine which of the sEH substrates or products could account for the delay in retinal angiogenesis, lipid profiles (LC-MS/MS) were generated from the retinas of 5-d-old wild-type and sEH^−/− mice. As expected, levels of 11,12- and 14,15-EET were increased in samples from sEH^−/− mice, whereas the levels of the corresponding diols were slightly decreased (Fig. 7 A). However, the most pronounced changes were in the levels of the DHA derivative 19,20-dihydroxydocosapentaenoic acid (DHDP). In CM from sEH^−/− Müller cells, 11,12-EET, 14,15-EET, and 19,20-epoxydocosapentaenoic acid (EDP) levels were significantly increased in CM and were restored to control by adenoviral transduction of the sEH (Fig. 7 B). The opposite was true for 19,20-DHDP, which was attenuated in CM from sEH^−/− Müller cells and restored by sEH overexpression. There was no significant change in 11,12-dihydroxyeicosatrienoic acid (DHET) levels, 14,15-DHET levels were below the limit of detection, and there were no apparent changes in the levels of the linoleic acid–derived fatty acids, 12,13-epoxyoctadecenoic acid (EpOME), or 12,13-dihydroxyoctadecenoic acid (DiHOME).

19,20-DHDP is of particular interest, given that it was the most abundant lipid detected in the retinal lipid profile and that its precursor DPA has previously been reported to modulate the activity of the γ-secretase (Delbosc et al., 2008; Grimm et al., 2011). When activated, the γ-secretase cleaves the Notch intracellular domain (NICD), which then either undergoes rapid proteasome degradation or translocates to the nucleus and binds to transcription factors (Boulton et al., 2008). We therefore compared the ability of 19,20-DHDP and the corresponding epoxide and sEH substrate 19,20-EDP to influence Dll4-stimulated Notch activity in murine endothelial cells. Although 19,20-EDP had no effect on the Dll4-induced increase in NICD cleavage, 19,20-DHDP decreased NICD levels by 54 ± 12% versus 89 ± 7% inhibition in the presence of the γ-secretase inhibitor N-[N-(3,5-Difluorophenacetyl)-l-alanyl]-S-phenylglycine t-butyl ester (DAPT; Fig. 7 C). Neither 19,20-EDP nor 19,20-DHDP affected VEFGR phosphorylation or downstream signaling in vascular endothelial cells (unpublished data). Treatment of cells with other sEH substrates, e.g., 11,12-EET, 14,15-EET, or 12,13-EpOME, or products, e.g., 11,12-DHET, 14,15-DHET, or 12,13-DiHOME, that were altered in the sEH^−/− retinas had no effect on the Dll4-induced cleavage of NICD (Fig. 7, D–F). When the nuclear translocation of the NICD was assessed by immunofluorescence (Fig. 8 A), 19,20-DHDP, but not 19,20-EDP, attenuated the Dll4-stimulated translocation. The fact that 19,20-DHDP also attenuated NICD levels in the presence of the proteasome inhibitor MG 132 (Fig. 8 B) suggested that the generation of the NICD, rather than its degradation, was targeted by the lipid.

The binding of a Notch ligand to a Notch receptor initiates two steps of cleavage to generate the NICD, the first by ADAM (a disintegrin and metalloprotease) proteases and the second by the γ-secretase (Kopan and Ilagan, 2009). The effect of 19,20-DHDP on the γ-secretase was evaluated using cells expressing a Notch 1 mutant (NotchΔE) that lacks the extracellular domain of the protein and possesses only the γ-secretase cleavage site (He et al., 2010). We found that 19,20-DHDP, but not 19,20-EDP, reduced the γ-secretase–dependent generation of NICD from NotchΔE (Fig. 8 C). The small effect observed at higher concentrations of 19,20-EDP can most likely be attributed to the endogenous expression of the sEH in the cells used (Barbosa-Sicard et al., 2009) and its conversion to 19,20-DHDP.

The γ-secretase complex is made up of several proteins, including presenilin 1, nicastrin, anterior pharynx defective 1, and presenilin enhancer 2, and the active complex is localized to lipid rafts (Vetrivel et al., 2004; Hur et al., 2008; Vetrivel and Thinakaran, 2010). We therefore determined whether 19,20-DHDP could inhibit the γ-secretase by altering the localization of constituent proteins and concentrated on presenilin 1. After sucrose density gradient fractionation, presenilin 1 and the raft-associated protein flotillin-1/reggie-2 were detected in detergent-insoluble lipid raft fractions (fractions 5 and 6) isolated from solvent–treated cells. 19,20-DHDP caused the displacement of presenilin 1 from lipid rafts to non-raft fractions (fractions 10 to 12) without affecting the distribution of flotillin-1 (Fig. 8 D). The latter effect that was accompanied by a decrease in cholesterol levels in the raft fractions (fractions 5 and 6) and an increase in the non-raft fractions (Fig. 8 E). Neither 19,20-EDP (in the presence of the sEH inhibitor t-AUCB) nor DAPT affected the colocalization of presenilin 1 with flotillin-1 or the distribution of cholesterol between the different fractions.

Finally, we determined whether or not 19,20-DHDP could rescue the retinal phenotype in sEH^−/− mice. Intravitreal injection of 19,20-DHDP was performed in 5-d-old wild-type and sEH^−/− pups and its effects compared with that of DAPT-treated mice. As before, the sEH^−/− retinas demonstrated a less dense vessel network as well as fewer tip cells and filopodia than their wild-type littermates, but this phenotype was largely reversed after the administration of 19,20-DHDP (Fig. 9). Similar effects were observed in retinas from animals receiving DAPT, whereas 19,20-EDP was without effect. In retinas from wild-type mice, 19,20-DHDP (but not 19,20-EDP) also induced a more complex vessel network, and increased tip cell and filopodia numbers (Fig. 9).

The E3 ubiquitin ligase Fbxw7 is involved in the ubiquitination and degradation of the NICD (Tan et al., 2008; Welcker et and Clurman, 2008), and endothelial cell-specific and inducible Fbxw7 mutant (Fbxw7^iECKO) mice demonstrate retinal blood vessel growth impairment that is attributable to Notch activation (Izumi et al., 2012). To confirm the in vitro data suggesting that 19,20-DHDP inhibits γ-secretase activity, the lipid was given intravitrealy to Fbxw7^iECKO mice. Also in these animals, 19,20-DHDP increased sprouting and filopodia numbers (Fig. 10), confirming the inhibition of NICD generation by the lipid. Similar effects were observed in animals that received DAPT, whereas 19,20-EDP was without effect.

The results of the present study show that the genetic deletion and inhibition of the sEH activates Notch signaling and attenuates physiological sprouting angiogenesis in the murine retina. Mechanistically, these effects could be attributed to the lack of the sEH product 19,20-DHDP, a fatty acid diol derived from DHA which inhibits the γ-secretase by inducing the redistribution of presenilin 1 from lipid rafts. Moreover, 19,20-DHDP, but not the parent epoxide, was able to rescue the defective angiogenesis in sEH^−/− mice as well as in animals lacking the Fbxw7 ubiquitin ligase which demonstrate strong basal activity of the Notch signaling cascade.

Müller glia cells were identified as the major source of the sEH in the murine retina. These cells develop and maintain a close contact with both superficial vessels and deeper capillaries via their multiple end-feet (Newman and Reichenbach, 1996) and have been implicated in angiogenesis by virtue of their ability to produce angiogenic substances in response to hypoxia (Pierce et al., 1995; Stone et al., 1995; Robbins et al., 1997). However, as the Müller cell–specific deletion of VEGF-A inhibited neovascularization in a mouse model of oxygen-induced retinopathy without affecting physiological vascularization or retinal morphology (Bai et al., 2009), it was assumed that Müller cells play a greater role in proliferative retinopathy than physiological angiogenesis. In our study, the deletion of the sEH in Müller cells clearly affected endothelial cell proliferation as well as Notch signaling and tip cell filopodia formation in mice, indicating that Müller cells make a more important contribution to retinal angiogenesis than generally appreciated. Interestingly, several recent studies also support a role for VEGF-independent signaling by Müller cells in developmental angiogenesis. For example, Müller cells secrete Norrin, a retinal signaling molecule which binds to Frizzled-4 to activate canonical Wnt/β-catenin signaling, in the absence of which the development of the superficial vessels was attenuated and deeper intraretinal capillaries were not formed (Xu et al., 2004; Ye et al., 2009; Ohlmann and Tamm, 2012). Also, the deletion of hypoxia inducible factor-1α in neuroretinal cells (includes Müller cells) results in impaired vascular development characterized by decreased tip cell filopodia and reduced vessel branching (Nakamura-Ishizu et al., 2012), a phenotype very similar to that of sEH^−/− mice.

The Müller cell–specific sEH knockout (sEH^ΔMC) mice were generated using Pdgfra-cre mice (Roesch et al., 2008), which may also target astrocytes. Also, because astrocytes express the sEH (Rawal et al., 2009) and retinal angiogenesis is generally thought to be largely determined by astrocyte-derived VEGF (Stone et al., 1995), we also focused on this interaction. However, although vessel branching was impaired in retinas from astrocyte-specific sEH knockout (GFAP-sEH) mice, the tip cell phenotype and Notch signaling (Hes1 and Hey1 expression) were unaltered, suggesting that another sEH-expressing cell type underlies the effects observed. Our findings do not completely rule out a role for astrocyte sEH in retinal angiogenesis but rather indicate that another source of the sEH plays a more dominant role.

To date, the CYP–sEH axis has been linked with the promotion of angiogenesis, as the epoxides of arachidonic acid (particularly 11,12- and 14,15-EET) promote tube formation in vitro (Munzenmaier and Harder, 2000; Michaelis et al., 2003) and can promote the vascularization of matrix material and tumors in vivo (Jiang et al., 2007). For a while, sEH inhibitors looked like highly interesting compounds for the treatment of cardiovascular disorders (Imig and Hammock, 2009), but interest was dampened by speculation that the associated increase in EET levels could induce or promote cancer progression. Certainly, tumor growth and metastasis can be increased by sEH inhibition in transgenic mice with high vascular EET levels, i.e., animals that overexpressed either the human CYP2C8 or human CYP2J2 specifically in Tie-2–expressing cells or that were treated with high concentrations of 14,15-EET (Panigrahy et al., 2012). The latter models were somewhat artificial and focused on the products of arachidonic acid metabolism, largely ignoring the biological actions of other lipids that feed into the same CYP–sEH axis. It was partly to assess the role of the sEH in angiogenesis under more physiological conditions that we determined the effects of the global and induced deletion of the sEH, as well as its pharmacological inhibition in the postnatal murine retina. We found that sEH deletion and inhibition resulted in markedly decreased capillary branching and endothelial cell proliferation as well as reduced tip cell numbers and filopodia.

If the epoxides of arachidonic acid possess angiogenic properties, why was angiogenesis attenuated in retinas from sEH^−/− mice which demonstrate higher than normal circulating and tissue levels of EETs? One logical explanation is that other lipids generated by the sequential activity of CYP epoxygenases and the sEH exert effects that dominate over, or antagonize those of, the EETs. The retinal phenotype observed could have been due to the accumulation of an anti-angiogenic epoxide or the loss of a pro-angiogenic diol, even though lipid diols are generally assumed to be either inactive or markedly less active than the parent epoxides (Imig and Hammock, 2009). To identify such lipids, we used an LC-MS/MS–based lipid profiling approach to screen for the lipid epoxide or diol most affected by the deletion of the sEH in the retina and found that 11,12-EET, 14,15-EET, and 12,13-EpOME were elevated, whereas 19,20-DHDP was attenuated in retinas from sEH^−/− mice. These findings could be confirmed in isolated Müller cells. Interestingly, DHA levels are known to be higher than those of arachidonic acid in the retina (Arterburn et al., 2006), and we detected higher levels of the DHA-derived lipids 19,20-EDP and 19,20–DHDP than the epoxides and diols of arachidonic acid in retinas from 5-d-old mice.

Which of the lipids affected by the deletion of the sEH could contribute to the development of the retinal phenotype in sEH^−/− mice? A significant delay in the radial extension of the vascular plexus, coupled with reduced endothelial cell proliferation as well as with fewer tip cells and filopodia, are indicative of Notch signaling activation (Hellström et al., 2007; Tammela et al., 2008). Therefore, we assessed the ability of all four candidate lipids to influence Dll4-stimulated Notch activity and the generation of the NICD. However, the sEH product 19,20,-DHDP and not the sEH substrates 19,20-EDP, 11,12-EET, 14,15-EET, or 12,13-EpOME attenuated NICD levels. Two distinct cleavage steps are required to generate the NICD; the first is dependent on ADAM proteases and the second on the activity of the γ-secretase (Kopan and Ilagan, 2009). To determine which of these proteases could be targeted by 19,20-DHDP, we assessed its ability to influence the Dll4-induced generation of the NICD from a so-called NotchΔE that possesses only the γ-secretase cleavage site (He et al., 2010). The finding that 19,20-DHDP also affected NICD generation in this model indicated that its molecular target was the γ-secretase. To confirm this in vivo, we assessed the consequences of the intravitreal application of 19,20-EPD or 19,20-DHDP to 5-d-old mice with hyper-activated endothelial cell Notch signaling, i.e., animals lacking the E3 ubiquitin ligase involved in the ubiquitination and degradation of the NICD (Izumi et al., 2012). In the latter Fbxw7^iECKO animals, which demonstrate markedly impaired retinal blood vessel growth impairment that is attributable to Notch activation (Izumi et al., 2012), 19,20-DHDP increased vessel branching as well as tip cell and filopodia numbers. The γ-secretase inhibitor DAPT elicited similar effects, whereas 19,20-EDP was ineffective.

Fatty acid epoxides such as the EETs are thought to act via a specific membrane receptor or the activation of peroxisome proliferator-activated receptors (Spector and Norris, 2007; Pfister et al., 2010). How can fatty acid diols exert biological effects that could lead to the inhibition of the γ-secretase? A cell surface receptor may not be essential, as polyunsaturated fatty acids can incorporate into membrane phospholipids and potentially influence cell signaling and biological responses by modulating the lipid microenvironment therein (Kitson et al., 2012). Certainly, the fish oils EPA and DHA are readily incorporated into phospholipids and the resulting polyunsaturated phospholipids are able to infiltrate lipid rafts as well as form non-raft domains (Williams et al., 2012). DHA, from which 19,20-DHDP is derived, is particularly interesting in this respect as it tends to incorporate more readily into lipid rafts than EPA (Williams et al., 2012). Changes in the lipid composition of the plasma membrane can have marked consequences of the formation and stability of the γ-secretase complex (Vetrivel et al., 2004; Vetrivel and Thinakaran, 2010). We found that 19,20-DHDP, but not the corresponding epoxide, altered cholesterol distribution in cellular membranes and displaced presenilin 1 from lipid rafts, where the processing of substrate proteins preferentially occurs, to non-raft fractions, thus disrupting the γ-secretase complex. Little is known about the incorporation of DHA-derived epoxides or diols into phospholipids, but this is highly likely given that the epoxides of arachidonic acid (the EETs) can be esterified to phosphatidylcholine, phosphatidylethanolamine, and phosphatidylinositols (Capdevila et al., 2000). This is not the first time a link between DHA and the γ-secretase has been reported, and although DHA increased γ-secretase activity in vascular smooth muscle cells (Delbosc et al., 2008), it was found to redistribute cholesterol out of lipid rafts (Wassall and Stillwell, 2008, 2009), thereby decreasing γ-secretase activity in a neuroblastoma cell line (Grimm et al., 2011). Our results agree with the latter study. Moreover, given that fatty acids such as DHA are rapidly metabolized by CYP and sEH enzymes and that the DHA epoxide 19,20-EDP had no effect on the γ-secretase, it is possible that the DHA diol 19,20-DHDP and the activity of the sEH may underlie the previously reported changes in lipid raft composition and biological responses rather than DHA on its own.

These studies highlight the fact that consequences of sEH inhibition in vivo cannot be solely attributed to the accumulation of the epoxides or arachidonic acid. In fact, more and more evidence is being accumulated to demonstrate that other lipid products (epoxides and diols) of the CYP–sEH axis can affect different steps in vessel development and maturation. For example, the sEH products 12,13-DiHOME (generated from the linoleic acid epoxide 12,13-EpOME), and to a lesser extent 11,12-DHET (generated from 11,12-EET), attenuate angiogenesis and vascular repair via the activation of the canonical Wnt pathway (Frömel et al., 2012). Moreover, the combination of 19,20-EDP with an sEH inhibitor was reported to inhibit primary tumor growth and metastasis by preventing the VEGF-induced phosphorylation of VEGFR2 (Zhang et al., 2013). However, why the latter lipid failed to affect retinal angiogenesis is unclear at the moment but fits with our finding that the phenotype observed in the sEH^−/− mice was independent of VEGF signaling. Our study also highlights the role of the neuroretina and particularly the sEH in Müller cells in the regulation of retinal angiogenesis. It will be interesting to determine whether or not this mechanism can be exploited and whether the pathological retinal angiogenesis associated with Müller cell activation (Coorey et al., 2012), such as that observed in oxygen-induced retinopathy in preterm infants, can be linked to altered lipid signaling and modulated by sEH inhibition.

The sEH inhibitor t-AUCB was provided by B. Hammock (University of California, Davis, Davis, CA). The proteasomal inhibitor MG-132 was obtained from EMD Millipore. Mouse and human Dll4 protein were obtained from R&D Systems. DAPT, tamoxifen, and all other chemicals were purchased from either Sigma-Aldrich or Merck and Roth.

C57BL/6 mice (6–8 wk old) were purchased from Charles River. sEH^−/− mice (Sinal et al., 2000) were provided by F. Gonzalez (National Institutes of Health, Bethesda, MD) and cross-bred for ∼20 generations onto the C57BL/6 background in the animal house facility at Frankfurt University. Inducible and endothelial cell–specific Fbxw7^iECKO mutant mice, generated as described previously (Izumi et al., 2012), were treated with tamoxifen (1 mg/ml i.p., 2% ethanol in sunflower oil, 50 µl) once a day from days 1–4.

Floxed sEH (sEH^fl/fl) mice (Taconic) were generated using C57BL/6 embryonic stem cells for gene targeting. Positive selection cassettes flanked by FRT/F3 sites were deleted by crossing with a ubiquitously expressing FLP1 recombinase strain (Taconic). These animals were cross-bred with animals expressing Cre under the control of the GFAP promoter (FVB-TgGFAP-cre25Mes [Zhuo et al., 2001] backcrossed in the C57BL/6 background) to generate mice lacking sEH in astrocytes. Inducible sEH knockout mice (iCreER-sEH) were generated by breeding sEH^fl/fl mice with the RosaCre-ER-T2 strain (Indra et al., 1999; cross-bred to C57BL/6) and Müller cell–specific sEH-deficient mice (sEH^ΔMC) were generated by crossing the sEH^fl/fl mice with C57BL/6-TgPdgfra-cre1Clc/J mice (Roesch et al., 2008; The Jackson Laboratory).

Mice were housed in conditions that conform to the Guide for the Care and Use of Laboratory Animals published by the U.S. National Institutes of Health (publication no. 85–23). All experiments were approved by the governmental authorities (Regierungspräsidium Darmstadt: F28/06, F28/23, and F28/38; Münster: 84-02.04.2011.A257). Age-, gender-, and strain-matched animals (usually littermates) were used throughout.

To inhibit epoxide hydrolase activity, C57BL/6 pups received t-AUCB (2 mg/kg i.p., twice per day) from days 1–4. To postnatally delete the sEH, iCreER-sEH transgenic mice were treated with tamoxifen (1 mg/ml i.p., 2% ethanol in sunflower oil, 50 µl) once a day from days 1–4 as previously described (Benedito et al., 2009). On day 5, retinas were isolated for Western blotting and immunohistochemistry.

Mice were anesthetized with 2% isoflurane. A drop of 0.5% proparacaine was administered as a topical local anesthesia. A fine glass micropipette connected to a 10 µl Hamilton glass syringe was inserted through the incision in the cornea and slid between the iris and the lens into the posterior chamber of the eye. Each eye received 0.5 µl 19,20-EDP, 100 µmol/liter 19,20-DHDP, or 200 µmol/liter DAPT or 0.5 µl of solvent (1% DMSO in PBS) as control. Injections were administered slowly over approximately one minute and monitored with a stereo microscope.

Proliferating endothelial cells in P5 retinas were labeled by administering BrdU (50 mg/kg i.p.; BD) 4 h before sacrifice. After Isolectin B4 staining, retinas were post-fixed for 30 min in 4% paraformaldehyde (PFA), washed three times with PBS, incubated for 1 h in 1 M HCl, washed five times in PBS, blocked, and incubated overnight with anti-BrdU antibody (1:50; BD). Secondary detection was performed with Alexa Fluor–coupled secondary antibodies.

Retinas for whole-mount immunohistochemistry were fixed in 4% PFA for 2 h at room temperature, or overnight at 4°C. After fixation, retinas were blocked and permeabilized in 1% BSA and 0.5% Triton X-100 overnight at 4°C. Then retinas were washed three times in Pblec buffer (0.5% Triton X-100, 1 mM CaCl₂, 1 mM MgCl₂, and 1 mM MnCl₂ in PBS, pH 6.8) and incubated overnight in Pblec containing Fitc-labeled Isolectin B4 (1:100; Sigma-Aldrich). The following primary antibodies were diluted in 1% BSA and 0.5% Triton X-100 and incubated overnight: sEH (1:250; provided by M. Arand, Institute of Pharmacology and Toxicology, Uinversity of Zurich, Zurich, Switzerland), GFAP (1:500; DAKO; or 1:1,000; EMD Millipore), NG-2 (1:1,000; EMD Millipore), AQP-4 (1:500; Santa Cruz Biotechnology, Inc.), GS (1:1,000; EMD Millipore), and vimentin (1:500; Nococastra). For secondary detection, Alexa Fluor–coupled secondary antibodies (1:200) were used. Cell nuclei were visualized with DAPI (1:200; Molecular Probes). After antibody staining, retinas were post-fixed in 4% PFA for 15 min before flat-mounting in mounting medium (Dako). 10-µm cryosections were cut after retinas were embedded within Tissue-Tek OCT Compound (Sakura). All quantification was done with high-resolution images taken using a meta laser scanning confocal microscope (LSM-510; Carl Zeiss) as previously described (Kovacevic et al., 2012).

Samples were extracted twice into ethyl acetate, evaporated under nitrogen, and resuspended in methanol/water (vol. 1:2). The eicosanoid profiles generated were determined with a mass spectrometer (API4000; AB Sciex) operating in multiple reaction monitoring (MRM) mode as previously described (Michaelis et al., 2005; Frömel et al., 2012). Chromatography was performed on a Gemini C18 column (150 mm length, 2 mm inner diameter; particle size, 5 µm; Phenomenex). All fatty acid epoxides, diols, and deuterated analogues were obtained from Cayman Europe.

Müller cells were isolated from 8–12-d-old mice, as previously described (Hicks and Courtois, 1990), and cultured in DMEM/F12 containing 10% FCS (Biochrom), 100 U/ml penicillin, and 100 µg/ml streptomycin. After 6 d, the cultures were rinsed free of nonadherent tissue, and culture medium was replenished every 3–4 d. The purity of the Müller cell cultures was characterized after each passage by testing for the expression of vimentin, AQP-4, GS, and CRALBP (Bunt-Milam and Saari, 1983). Müller cell CM was generated by culturing confluent wild-type or sEH^−/− Müller cells for 48 h.

The human sEH (Ephx2) cDNA sequence, with a C-terminal Myc/His tag, was excised via NotI–PmeI from pcDNA3.1myc/His as previously described (Barbosa-Sicard et al., 2009) and cloned into the NotI–EoRV site of the cloning plasmid pAdTrackCMV. The recombination into the adenoviral transfer plasmid and the generation of the viruses was performed as previously described (Luo et al., 2007). For the sEH rescue experiments, sEH^−/− Müller cells were infected with sEH adenovirus for 6 h and then washed four times with Hank’s solution. Experiments were performed after an additional 48 h and sEH expression was confirmed by immunoblotting.

1 × 10⁶ endothelial cells isolated as previously described (Fleming et al., 2005) were seeded on 6-well plates coated with fibronectin. After 4 h, the nonadherent cells were removed and the medium was replaced with 1.1 ml DMEM/F12 containing 10% FCS. Then Transwell inserts (Millipore) were added to the wells and seeded with 6 × 10⁵ Müller cells. After 8 h, the insert was removed and RNA was extracted from the endothelial cells.

5 × 10⁴ endothelial cells were seeded in 24-well plates and stimulated with M-CM from wild-type or sEH^−/− Müller cells (mixed with endothelial cell growth medium at 1:1 ratio), or solvent (DMEM/F12). Mouse VEGF at a final concentration of 50 ng/ml was used as a positive control. Viable cells were counted (CASY; Roche) after 24, 48, and 72 h of stimulation. For the BrdU incorporation, endothelial cells were seeded in 96-well plates (1 × 10⁴ cells/well) and stimulated with the same condition as described above for 48 h. BrdU was added to cells 4 h before the end of stimulation and then BrdU incorporation was assessed using BrdU ELISA kit (Roche) according to the manufacturer’s instructions.

Retinas were isolated from P5 mice. After digestion with 0.1% trypsin (Worthington Biochemical Corporation) and 70 U/ml collagenase type I (Biochrom) in DMEM/F12 medium at 37°C for 30 min, retinal endothelial cells were isolated with anti–mouse CD 31 antibody based on magnetic bead separation similar to that described above. CD-31–positive cells were used for gene expression analysis.

Total RNA from retinas or cultured primary cells was extracted using an RNeasy kit (QIAGEN), and equal amounts (1 µg) of total RNA was reverse transcribed (Superscript III; Invitrogen). Gene expression levels were detected using SYBR Green (Absolute QPCR SYBR Green Mix; Thermo Fisher Scientific). The TaqMan probe used for the detection of the Dll4 RNA was from Applied Biosystems. The relative expression levels of the different genes studied was calculated using the 2^−ΔΔCt method with the 18S RNA as a reference. The primer sequences used were as follows: 18S forward, 5′-CTTTGGTCGCTCGCTCCTC-3′, reverse, 5′-CTGACCGGGTTGGTTTTGAT-3′; Dll1 forward, 5′-CATCCGATACCCAGGTTGTC-3′, reverse, 5′-ACGGCTTATGGTGAGTACAG-3′; Jagged1 forward, 5′-TCTCTGACCCCTGCCATAAC-3′, reverse, 5′-TTGAATCCATTCACCAGATCC-3′; Notch1 forward, 5′-CCTCAGATGGTGCTCTGATG-3′, reverse, 5′-CTCAGGTCAGGGAGAACTAC-3′, Hes1, forward, 5′-GAGGCGAAGGGCAAGAATAAA-3′, reverse, 5′-GTGGACAGGAAGCGGGTCA-3′; Hey1 forward, 5′-TGAGCTGAGAAGGCTGGTAC-3′, reverse, 5′-ACCCCAAACTCCGATAGTC-3′; and VEGF forward, 5′-GCACTGGACCCTGGCTTTACTGCTGTA-3′, and reverse, 5′-GAACTTGATCACTTCATGGGACTTCTGCTC-3′.

Endothelial cells were seeded onto 6-well plates coated with or without 500 ng/ml recombinant mouse Dll4 protein (R&D Systems) in 0.1% BSA/PBS for 1 h at room temperature. Experiments were performed in the presence of solvent, 10 µmol/liter 19,20-EDP/DHDP, 10 µmol/liter 11,12-EET/DHET, 10 µmol/liter 14,15-EET/DHET, 3 µmol/liter 12,13-EpOME/DiHOME, 20 µmol/liter DAPT, or 10 µmol/liter MG 132. After an additional 4 h, cells were washed three times with PBS and lysed in a Triton X-100–containing buffer. Intracellular domain of Notch (NICD) levels were assessed by Western blotting with an S3 (γ-secretase) cleavage-specific antibody Val1744 (Cell Signaling Technology). For the nuclear translocation of NICD, cells were fixed with 4% PFA for 15 min at room temperature, followed by NICD and DAPI immunostaining. Co-localization of NICD with DAPI was quantified with ImageJ software (National Institutes of Health).

HEK 293 cells were seeded in 10 cm culture dishes and transfected with 6 µg NotchΔE construct (provided by P. Greengard, Rockefeller University, New York, NY). Cells were then incubated with solvent, 19,20-EDP, 19,20-DHDP (0.1–10 µmol/liter), or DAPT (20 µmol/liter) for 24 h. The γ-secretase–dependent generation of NICD was detected as described above.

HEK293 cells were incubated with solvent (0.1% DMSO), 10 µmol/liter 19,20-EDP, 10 µmol/liter 19,20-DHDP, or 20 µmol/liter DAPT in serum-free medium containing 10 µmol/liter of the sEH inhibitor t-AUCB for 2 h at 37°C. Cells were then washed twice with ice-cold PBS, harvested by scraping, recovered by centrifugation (300 g for 4 min), and resuspended in 2 ml MBS buffer (containing 1% CHAPSO in 25 mM MES, pH 6.5, 150 µmol/liter NaCl supplemented with a protease inhibitor mix, and 1 µmol/liter phenylmethylsulfonyl fluoride), and homogenized by using a glass homogenizer. Lipid rafts were isolated as previously described (Vetrivel et al., 2004). In brief, after 30 min at 4°C, equal amounts of protein were adjusted to a final sucrose concentration of 45% (final volume, 4 ml) and transferred to 12 ml ultracentrifuge tubes. A discontinuous sucrose gradient was then formed by sequentially overlaying 4 ml of 35% and 4 ml of 5% sucrose. Samples were subjected to ultracentrifugation (35,000 rpm, 4°C for 20 h) using a rotor (SW 41; Beckman). After centrifugation, twelve 1 ml fractions were collected using a capillary tube connected to a peristaltic pump, and equal volumes of each faction were analyzed by SDS-PAGE using antibodies against flotillin 1 (BD) and presenilin 1 (Santa Cruz Biotechnology, Inc.). Cholesterol levels in the different fractions were determined using the Amplex Red cholesterol assay (Invitrogen) according to the manufacturer’s instructions.

Data are expressed as mean ± SEM, and statistical evaluation was performed using Student’s t test for unpaired data or one-way ANOVA, followed by a Bonferroni’s t test where appropriate. Values of P < 0.05 were considered statistically significant.

The authors are indebted to Isabel Winter and Katharina Engel-Herbig for expert technical assistance.

This work was supported by the Deutsche Forschungsgemeinschaft (SFB-TR 23/A6, GRK 880 and Exzellenzcluster 147 “Cardio-Pulmonary Systems”). J. Hu was supported by a research scholarship from China Scholarship Council (CSC).

The authors declare no competing financial interests.