Nrf2 in ischemic neurons promotes retinal vascular regeneration through regulation of semaphorin 6A

Significance

Delayed revascularization of ischemic neural tissue is a major impediment to preservation of function in central nervous system (CNS) diseases including stroke and ischemic retinopathies. The key mechanisms governing vascular recovery in ischemic CNS, including regulatory molecules governing transition from tissue injury to repair, are largely unknown. We report here on NF-E2-related factor 2 (Nrf2), a major stress-response transcription factor known for its cell-intrinsic cytoprotective function, in a novel capacity coordinating tissue repair and remodeling, including regulation of cell–cell crosstalk. Nrf2 activity in ischemic neurons reduces their resistance to reparative angiogenesis by suppressing expression of neuronal semaphorin 6A (Sema6A) and its antiangiogenic effects. Pharmacologic activation of Nrf2 or inhibition of Sema6A promote reparative angiogenesis in this ischemic setting, suggesting therapeutic avenues for ischemic retinopathies and other ischemic diseases.

Abstract

Delayed revascularization of ischemic neural tissue is a major impediment to preservation of function in central nervous system (CNS) diseases including stroke and ischemic retinopathies. Therapeutic strategies allowing rapid revascularization are greatly needed to reduce ischemia-induced cellular damage and suppress harmful pathologic neovascularization. However, key mechanisms governing vascular recovery in ischemic CNS, including regulatory molecules governing the transition from tissue injury to tissue repair, are largely unknown. NF-E2-related factor 2 (Nrf2) is a major stress-response transcription factor well known for its cell-intrinsic cytoprotective function. However, its role in cell–cell crosstalk is less appreciated. Here we report that Nrf2 is highly activated in ischemic retina and promotes revascularization by modulating neurons in their paracrine regulation of endothelial cells. Global Nrf2 deficiency strongly suppresses retinal revascularization and increases pathologic neovascularization in a mouse model of ischemic retinopathy. Conditional knockout studies demonstrate a major role for neuronal Nrf2 in vascular regrowth into avascular retina. Deletion of neuronal Nrf2 results in semaphorin 6A (Sema6A) induction in hypoxic/ischemic retinal ganglion cells in a hypoxia-inducible factor-1 alpha (HIF-1α)-dependent fashion. Sema6A expression increases in avascular inner retina and colocalizes with Nrf2 in human fetal eyes. Extracellular Sema6A leads to dose-dependent suppression of the migratory phenotype of endothelial cells through activation of Notch signaling. Lentiviral-mediated delivery of Sema6A small hairpin RNA (shRNA) abrogates the defective retinal revascularization in Nrf2-deficient mice. Importantly, pharmacologic Nrf2 activation promotes reparative angiogenesis and suppresses pathologic neovascularization. Our findings reveal a unique function of Nrf2 in reprogramming ischemic tissue toward neurovascular repair via Sema6A regulation, providing a potential therapeutic strategy for ischemic retinal and CNS diseases.

A critical aspect in the recovery of ischemic neurons is the restoration of vascular supply, including vascular remodeling of the affected area. Strikingly, although the hypoxia that accompanies ischemia is a well-known driver of vascular growth, revascularization of ischemic nervous tissue is often inadequate (1, 2) in conditions such as stroke and ischemic retinopathies. Therapeutic strategies that allow more rapid revascularization would be of tremendous benefit, reducing ischemia-induced cellular damage and also suppressing the harmful aberrant pathologic neovascularization that can occur. Accordingly, there is a great need for additional insights into the key mechanisms governing vascular recovery in the ischemic retina and CNS (3, 4).

Ischemic diseases of the retina are major causes of blindness from conditions including retinopathy of prematurity, diabetic retinopathy, and vein occlusions. A critical element of ischemic retinopathies is the inadequate revascularization of the ischemic tissue; this ischemia leads to pathologic new vessels misdirected toward the overlying vitreous cavity, ultimately resulting in visual loss from vitreous hemorrhage and traction retinal detachment. Indeed, models of ischemic retinopathy have been a powerful system for studying revascularization of neural tissue in ischemia and are particularly convenient for studying neurovascular crosstalk during ischemic injury and repair (3, 5).

A critical theme in neurovascular remodeling in ischemic disease is the shift from an injury to a repair response (4, 6). Toward this end, multiple cellular elements influence neurovascular remodeling in ischemic CNS and retinal disease, including neurons, astrocytes, and vascular cells (3, 4). There is a growing awareness of the pivotal influence of neuronal elements (3). Ischemic neurons possess the ability either to promote or resist revascularization. For example, semaphorin 3A (Sema3A) and Sema3E, secreted by ischemic ganglion cells, serve to suppress vascular regrowth into the ischemic zone in a mouse model of ischemic retinopathy (1, 2). This suggests a therapeutic avenue consisting of treatments that shift neuronal elements toward a repair response, or alternatively, treatments that target neurovascular interactions, including molecular mediators. However, the mechanisms involved in dictating the response of neurons and the neurovascular unit to ischemic conditions remain poorly understood, especially the key regulatory molecules involved in reprogramming ischemic tissue toward a vascular repair response (4).

NF-E2-related factor 2 (Nrf2), a major stress-response transcription factor (7), is well-known to play a critical protective role in many disease settings, offering a mechanism for cell-autonomous cytoprotection (7). The role of Nrf2 in regulating cell–cell crosstalk is less appreciated. Neuroprotective strategies based on Nrf2’s cell-intrinsic properties have focused on neurodegenerative conditions (8, 9), but have also been demonstrated in the context of stroke (10, 11). However, little is known regarding the role of Nrf2 in neurovascular repair and remodeling; namely, its influence on revascularization of ischemic neural tissue. Interestingly, Nrf2 has received some attention in its regulation of angiogenesis, which appears to be strongly context-dependent. Our laboratory has found Nrf2 to promote the angiogenic phenotype in endothelial cells in a cell-autonomous manner during retinal vascular development (12). Knockdown of Nrf2 contributes to the suppression of colon tumor angiogenesis (13), whereas deletion of Nrf2 leads to exacerbation of ischemia-induced angiogenesis in limbs and lung (14, 15). Strikingly, the cellular context of Nrf2’s angiogenic effects, including potential regulation of endothelial cells in paracrine fashion, has received little attention.

In this study, we demonstrate that Nrf2 plays an important role in the revascularization of the neuroretina after ischemia, coordinating neuronal and endothelial elements. This less appreciated facet of Nrf2 as a regulator of cell–cell crosstalk suggests it plays a significant function in ischemic tissue reprogramming. Nrf2 modulated the response of retinal neurons to ischemia, influencing production of the repulsive membrane-associated semaphorin, Sema6A. Treatment with an Nrf2 activator suppressed Sema6A expression and improved revascularization of the ischemic retina. This suggests that Nrf2 may be a participating molecule in the transition of the ischemic neuroretina from an injury to repair response and that Sema6A is an important regulator in neurovascular remodeling.

Results

Nrf2 Is Activated in Neuroretinal Ischemia.

To gain insights into the role of Nrf2 in the regulation of vascular regeneration in ischemic neuronal tissue, we used the mouse model of oxygen-induced retinopathy (OIR), in which mice are exposed to 75% oxygen from postnatal day 7 (P7) to P12, and returned to room air until P17 (16). The hyperoxic phase (P7–P12) results in vaso-obliteration of the central retina. This is followed by a second phase (P12–P17) of vascular regeneration in this central avascular retina (Fig. 1A). Incomplete or delayed vascular recovery of the retina proper, and hence retinal ischemia, results in maladaptive pathologic preretinal neovascularization at P17, reminiscent of human conditions including retinopathy of prematurity and proliferative diabetic retinopathy (5). In this model system of ischemia, we first investigated Nrf2 activation, which is known to be reflected primarily by its nuclear translocation (7). We found evidence for Nrf2 activation soon after onset of ischemia, with increases in nuclear Nrf2 levels as early as 2 h (P12 + 2 h) after the hyperoxic phase (Fig. 1 B and C). The increase in nuclear Nrf2 was sustained through P17 (Fig. 1 B and C). Nrf2 activation was also confirmed by the up-regulation of the Nrf2 target gene, NQO1, from P12 (2 h) to P17 (Fig. 1D), in parallel with the time-course of nuclear Nrf2 increase.

Fig. 1.

Genetic ablation of Nrf2 impedes vascular regrowth and increases pathological neovascularization in OIR. (A) Diagram of OIR. Neonatal mice are put into hyperoxia chamber containing 75% O₂ at P7 and return to room air at P12. The retinal blood vessels are obliterated in hyperoxia and regenerated from P12, when retina becomes hypoxic. The peak of neovascularization tufts (pathological neovascularization) occurs at P17. (B) Immunoblot analysis of Nrf2 in the cytoplasm and nucleus from wild-type retina. β-Actin and lamin B were used as loading controls. (C) Quantification of nuclear Nrf2 shown in B. n = 3. (D) Quantitative RT-PCR analysis of Nrf2 target gene NQO1. n = 5. Nrf2 activated in the ischemic stage from P12 (2 h having after been taken out from hyperoxia) to P17 of OIR. (E) Retinal flat mounts of wild-type (Nrf2^+/+) and Nrf2^−/− mice at P12 and P17 of OIR. Blood vessels were visualized by GS lectin staining. Avascular retina was outlined a by red line, and arrowheads indicate neovascularization tufts. No significant difference in avascular area between Nrf2^+/+ and Nrf2^−/− retinas were observed at P12 (F), whereas dramatic increase in avascular area and area of tufts were exhibited in Nrf2^−/− retinas compared with wild-type at P17 (G). n = 15–21. (Scale bar, top and middle panels, 400 µm; bottom panel, 50 µm.) (H and I) Marked increase in hypoxic area was observed in both vascularized (arrowheads) and avascular (*) Nrf2^−/− retinas at P17. Hypoxic region was detected by hypoxyprobe staining. n = 8. (Scale bar, 200 µm.) (J) Increase in vascular leakage was observed in Nrf2^−/− retinas at P17. Radioactivity of retina was measured 1 h after i.p. injection of [³H] mannitol. n = 8. (K and L) Nrf2^−/− mice exhibited a significant reduction in inner-retinal scotopic b-wave response (L), whereas a-wave (K) reflecting outer retinal function was not affected in response to flash with intensity from −0.9 to 0 log cd/s/m² compared with wild-type. Scotopic electroretinogram was recorded in dark-adapted conditions using flashes of white light ranging in intensity from −6.3 to 0 log cd/s/m² in 0.3-log-unit increments. n = 9–14. Data are presented as mean ± SEM. *P < 0.05; **P < 0.01; NS, not significant.

Genetic Ablation of Nrf2 Impedes Vascular Regeneration and Increases Pathological Neovascularization in OIR.

We next investigated the role of Nrf2 on reparative angiogenesis in the ischemic retina. Nrf2 has been found to play an important role in angiogenesis in other contexts, both with respect to vascular development (12) and pathologic setting, such as tumor angiogenesis (13, 17). To determine the importance of Nrf2 in ischemic retina, we compared wild-type (Nrf2^+/+) and Nrf2-deficient mice (Nrf2^−/−) with respect to vascular changes in OIR. There was no appreciable difference in avascular area between Nrf2^+/+ and Nrf2^−/− retinas immediately on return to room air at P12 (Fig. 1 E and F). In striking contrast, there was a dramatic increase in avascular area, indicating a marked reduction in reparative angiogenesis, in Nrf2^−/− retinas compared with wild-type at P17. This was accompanied by a considerable increase (about fivefold) in pathologic preretinal neovascularization in Nrf2^−/− retinas compared with wild-type (Fig. 1 E and G). The results indicate an important role of Nrf2 in enhancing revascularization of the ischemic retina between P12 and P17. In addition to revascularization, we also evaluated vascular function. More poorly perfused vessels were observed in the retinal periphery, angiogenic front, and areas associated with neovascular tufts in Nrf2^−/− retinas (Fig. S1 A and B). There was a marked increase in hypoxic area in both the peripheral vascularized region and central avascular zone in Nrf2^−/− retinas (Fig. 1 H and I). Finally, there was an increase in vascular leakage in Nrf2^−/− retinas (Fig. 1J). These results indicate greater impairment of vascular function in Nrf2^−/− retinas at P17 subjected to OIR. Restoration of blood supply via regrowth of functional vessels is crucial for the preservation of neural function in ischemic tissues (18, 19). Given the vascular changes in Nrf2^−/− retinas, we also investigated neuronal function using electroretinography at P40 of OIR. Nrf2^−/− mice exhibited a significant reduction in scotopic b-wave response (Fig. 1L), reflecting an impairment in inner retinal function, whereas the a-wave (Fig. 1K) was not affected in response to a light flash stimulus with intensity from −0.9 to 0 log cd/s/m² compared with wild type. Our results, demonstrating concordance of vascular status and electroretinographic response, are consistent with previous studies that found that neuronal functional changes in OIR primarily involve the inner retina and are caused by dysfunction of retinal blood vessels (2). In addition, we found that both scotopic b-wave and a-wave responses were not affected in Nrf2^−/− retinas without OIR (Fig. S1 C and D), indicating that the impaired visual function in Nrf2^−/− retinas is the result of an impairment in the repair process, rather than a developmental deficit.

Fig. S1.

(A and B) Nrf2^−/− retinal vasculature showed less perfusion compared with wild-type. More poor-perfused vessels were observed in Nrf2^−/− retinas (arrowheads) at P17. n = 7. (Scale bar, 50 µm.) (C and D) No significant difference of scotopic electroretinography was observed in wild-type and Nrf2^−/− mice under room air condition (without OIR) at P40. n = 7. Data are presented as mean ± SEM. **P < 0.01.

Neuronal Nrf2 Plays a Crucial Role in Revascularization in OIR.

In light of Nrf2’s profound effect on reparative angiogenesis, we were interested in determining the cellular context of Nrf2 action. Nrf2^+/+ retina at P12 (2 h after onset of ischemia) was selected for immunofluorescence staining because retinal Nrf2 expression was highest at this point, according to immunoblot analysis (Fig. 1B). We were particularly attentive to Nrf2 expression in the inner retina, the site of vaso-obliteration and reparative angiogenesis in OIR. Nrf2 was ubiquitously expressed in both avascular and vascularized retina. Within the inner retina, there was strong Nrf2 expression in the ganglion cell layer (GCL), with accentuated expression in the avascular retina compared with the peripheral perfused retina (Fig. 2A). Nrf2 was detected in retinal ganglion cells (RGCs), endothelial cells (ECs), and astrocytes. The strongest Nrf2 expression and nuclear localization was observed in RGCs, whereas weaker expression was present in Müller cells (Fig. 2B). The pattern of Nrf2 expression in OIR differs from that in normal adult mouse retina, in which Nrf2 is expressed prominently in Müller glial cells and astrocytes (20).

Fig. 2.

Neuronal Nrf2 plays a crucial role in revascularization in OIR. (A and B) Nrf2 localization in retinal cryosections from P12 (2 h) OIR. Nrf2 is ubiquitously expressed in both avascular and vascularized retina, and strong expression was observed in GCL of avascular retina (arrowheads in A). (B) Nrf2 is expressed in retinal ganglion cells (RGCs), endothelial cells (ECs), and astrocytes, as demonstrated by colocalization with Brn3, GS lectin, and PDGFRα, respectively. Weak expression was shown in Müller cells, as demonstrated by colocalization with vimentin. The strongest Nrf2 expression and nuclear localization was observed in RGCs (arrowhead). GCL, ganglion cell layer; INL, inner nuclear layer; ONL, outer nuclear layer. (Scale bar, 20 µm in A; 10 µm in B.) (C) Retinal flat mounts of control (Nrf2^fl/fl) mice and mice with loss of Nrf2 in nonvascular retinal tissue (Nrf2^fl/fl; Six3-Cre), astrocytes (Nrf2^fl/fl; Gfap-Cre), and endothelial cells (Nrf2^fl/fl; Cdh5-Cre) at P17 of OIR. Blood vessels were visualized by GS lectin staining. Avascular retina was outlined by red line, and arrowheads indicate neovascularization tufts. Nrf2^fl/fl; Six3-Cre retina exhibited a dramatic increase in avascular area (D), but no change in the area of tufts (E). n = 12–15. No appreciable change in either avascular area or area of tufts was observed in Nrf2^fl/fl; Gfap-Cre retina (F and G). n = 10–14. Increased avascular area and area of tufts were observed in Nrf2^fl/fl; Cdh5-Cre retina (H and I). n = 10–16. (Scale bar, 400 µm.) Values represent fold change relative to Nrf2^fl/fl. Data are presented as mean ± SEM. **P < 0.01; NS, not significant.

To dissect the roles of Nrf2 in neurons, astrocytes, and ECs in OIR, we studied Nrf2^flox/flox (Nrf2 ^fl/fl); Six3-Cre, Nrf2^fl/fl; Gfap-Cre and Nrf2^fl/fl; Cdh5-Cre mice, obtained by crossing Nrf2^fl/fl with Six3-Cre, Gfap-Cre and VE-Cadherin (Cdh5)-Cre mice, respectively. These lines accomplish Cre-mediated recombination in retinal neuroglial elements (12, 21), astrocytes (22), and endothelial cells (12, 23), respectively. At P17 of OIR, Nrf2^fl/fl; Six3-Cre retina exhibited a significant increase in avascular area, but no significant change in the area of neovascular tufts (Fig. 2 C–E), and no appreciable change in either avascular area or area of neovascular tufts was observed in Nrf2^fl/fl; Gfap-Cre retina (Fig. 2 C, F, and G). The results suggest an important role of Nrf2 in neurons in regulating revascularization in OIR. Interestingly, there was a modest increase in both avascular area and area of neovascular tufts in Nrf2^fl/fl; Cdh5-Cre retina (Fig. 2 C, H, and I), indicating a role of Nrf2 in ECs in vessel regrowth and preretinal neovascularization in OIR. Taken together, these results suggest that Nrf2 in both neurons and ECs plays a key role in the regulation of vascular regeneration and preretinal neovascularization in OIR. These results also demonstrate a notable difference in the cellular context of Nrf2 action in pathologic ischemia-induced angiogenesis and normal developmental angiogenesis, in which Nrf2 modulates angiogenic sprouting and branching in an EC-autonomous manner, with no discernible contribution of neuronal Nrf2 (12).

Keap1 is known to be a major repressor of Nrf2, binding Nrf2 and preventing its nuclear translocation (7). To confirm the role of endothelial Nrf2 in OIR, we investigated Keap1^fl/fl; Cdh5-Cre mice, in which Nrf2 in ECs is constitutively activated (12). We found that Keap1^fl/fl; Cdh5-Cre retinas exhibited a mild decrease in avascular area and a marked reduction in the area of neovascular tufts (Fig. S2 A–C). Enhanced vascular density and increased vascular sprouts were observed in Keap1^fl/fl; Cdh5-Cre retina (Fig. S2A). Strikingly, significant decrease of hypoxic region was shown in vascularized retina of Keap1^fl/fl; Cdh5-Cre mice (Fig. S2 D and E). These results suggest that Nrf2 in ECs is particularly critical in the suppression of pathologic preretinal neovascularization and preservation of vascular function. It is to be noted that activation of endothelial Nrf2 is not able to largely enhance vessels regrowth into the avascular zone, which implicates a crucial role of neuronal Nrf2 in revascularization.

Fig. S2.

Deletion of the Nrf2 repressor Keap1 in ECs leads to a dramatic reduction in preretinal neovascularization and a modest enhancement in vessel regrowth in OIR. (A) Retinal flat mounts of control (Keap1^fl/fl) and Keap1^fl/fl; Cdh5-Cre mice at P17 of OIR. Blood vessels were visualized by GS lectin staining. Avascular retina was outlined by red line, and arrowheads indicate neovascularization tufts. Keap1^fl/fl; Cdh5-Cre retina exhibited a mild decrease in avascular area (B) and a marked reduction in the area of tufts (C). n = 18. Enhanced vascular density (peripheral in A) and increased vascular sprouts (sprouting front in A) were observed in Keap1^fl/fl; Cdh5-Cre retina. (Scale bar, whole-mount, 400 µm; central, 200 µm; peripheral, 100 µm; sprouting front, 50 µm.) (D and E) Reduction in hypoxic area was observed in vascularized (arrowheads) Keap1^fl/fl; Cdh5-Cre retinas at P17 of OIR. White stars indicate hypoxic area in avascular retina. Hypoxic region was detected by hypoxyprobe staining. n = 8. (Scale bar, 200 µm.) Data are presented as mean ± SEM. *P < 0.05; **P < 0.01.

Deletion of Nrf2 Leads to Increased Sema6A Expression in Ischemic Inner Retina in OIR.

The findings of intense expression and activation of Nrf2 in RGCs and the increased avascular area in Nrf2^fl/fl; Six3-Cre retina strongly suggest a critical role of Nrf2 in RGCs in regulating vascular regeneration. Neuronal guidance molecules, the crucial cues directing axon growth in neuronal development, have also been found to play an important role in modulating vascular development and angiogenesis (24). We therefore explored the idea that the suppression of vascular regrowth in the retina associated with loss of Nrf2 in neurons could be attributed to neuronal guidance molecules. To identify potential candidate mediators, we first analyzed the available Nrf2 Chip-seq databases (25, 26) and found five axon guidance genes (ephrinA5, ephrinB2, Sema4A, Sema6D, Sema3E) that might represent the direct candidate targets of Nrf2. We therefore investigated this potential regulatory relationship, using quantitative RT-PCR, and also included analysis of Sema3A, Sema6A, Slit1, and Netrin1, all of which have been shown to regulate angiogenesis in previous studies (1, 2, 27). The result showed that Sema6A mRNA was significantly increased in Nrf2^−/− retinas at P15 of OIR, whereas no significant changes in expression were observed for the other genes (Fig. S3A). A time-course study of Sema6A expression revealed that Sema6A mRNA levels in Nrf2^−/− retinas from P12 to P15 of OIR were significantly up-regulated, compared with wild-type, from P12 (2 h) to P15 (Fig. 3A and Table S1).

Fig. S3.

(A) Quantitative RT-PCR analysis of axon guidance genes in wild-type and Nrf2^−/− retinas at P15of OIR. n = 5. (B) Quantitative RT-PCR analysis shows the increase in Sema6A, whereas no change in Sema3A or Sema3E was seen in the central GCL of Nrf2^fl/fl; Six3-Cre retinas compared with Nrf2^fl/fl harvested by laser-capture microdissection at P14 of OIR. n = 5. Data are presented as mean ± SEM. *P < 0.05; NS, not significant.

Fig. 3.

Deletion of Nrf2 leads to increased Sema6A expression in ischemic inner retina in OIR. (A) Quantitative RT-PCR analysis shows the increased Sema6A mRNA in Nrf2^−/− retinas compared with wild-type from P12 to P15 in OIR. n = 5. (B) Retinal section demonstrating the area harvested by laser-capture microdissection. (Right) The retinal layers were cut as shown. P, peripheral; C, central. (Scale bar, left, 200 µm; right, 20 µm.) (C) Quantitative RT-PCR analysis of Sema6A mRNA in retinal layers harvested by laser-capture microdissection reveals a marked induction in GCL of Nrf2^fl/fl; Six3-Cre retinas compared with Nrf2^fl/fl at P14 of OIR. n = 5. (D) Retinal sections of immunofluorescence staining show that Sema6A expression is increased in central GCL and inner plexiform layer compared with peripheral of Nrf2^fl/fl mice, and the accentuated induction is exhibited in both central and peripheral retina of Nrf2^fl/fl; Six3-Cre compared with Nrf2^fl/fl at P14 of OIR. The strongest Sema6A expression was observed in GCL (arrowheads). (Scale bar, 50 µm.) (E) Retinal flat mounts of immunofluorescence staining show the Sema6A expression in RGCs (arrows) and the robust expression surrounding angiogenic front and tufts (arrowheads) in Nrf2^fl/fl; Six3-Cre retina at P14 of OIR. (Scale bar, 20 µm.) Data are presented as mean ± SEM. *P < 0.05; **P < 0.01; NS, not significant.

Table S1.

Ct values from quantitative RT-PCR shown in Fig. 3A

Strain	Gene	OIR
		P12-0h	P12-2h	P14	P15
Nrf2+/+	Sema6A	23.34 ± 0.29	22.29 ± 0.20	23.29 ± 0.08	23.23 ± 0.15
	Cyclophilin A	18.76 ± 0.34	17.50 ± 0.84	18.16 ± 0.05	18.02 ± 0.08
Nrf2−/−	Sema6A	22.96 ± 0.11	21.97 ± 0.09	22.86 ± 0.05	23.44 ± 0.24
	Cyclophilin A	17.90 ± 0.04	17.47 ± 0.04	18.14 ± 0.07	18.67 ± 0.20

Data are presented as mean ± SEM.

To further delineate the regulation of Sema6A by neuronal Nrf2 and to localize Sema6A expression, we collected retinal layers from Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre mice by laser-capture microdissection to analyze mRNA levels by quantitative RT-PCR. The results showed that Sema6A mRNA was significantly increased in central GCL of Nrf2^fl/fl retinas compared with peripheral GCL. Accentuation of Sema6A mRNA expression was observed in Nrf2^fl/fl; Six3-Cre retinas compared with Nrf2^fl/fl, in both the central and the peripheral retina. There was also a slight increase in Sema6A mRNA expression, shown in Nrf2^fl/fl; Six3-Cre retinas compared with Nrf2^fl/fl in the inner nuclear layer (INL), whereas no difference was observed in the outer nuclear layer (ONL) (Fig. 3C and Table S2). Sema3A and Sema3E expression have been shown to increase in GCL in OIR retinas and to inhibit vascular regeneration and/or promote pathologic preretinal neovascularization (1, 2). However, we observed no appreciable change in the mRNA level of either Sema3A or Sema3E in the central GCL of Nrf2^fl/fl; Six3-Cre retinas (Fig. S3B). With respect to Sema6A, we performed further localization studies using immunofluorescence staining, which confirmed the higher expression of Sema6A in the GCL compared with other retinal layers (Fig. 3D), consistent with the mRNA expression pattern in the laser-captured microdissected retinal layers. Costaining of Sema6A and GS lectin in retinal flat mounts demonstrated the Sema6A expression localized to RGCs, with robust expression surrounding the angiogenic front and neovascular tufts (Fig. 3E). These results suggest that Sema6A was induced in the hypoxic (ischemic) GCL and that deletion of Nrf2 in neurons exacerbates Sema6A expression in the GCL (RGCs) in OIR.

Table S2.

Ct values from quantitative RT-PCR shown in Fig. 3C

Strain	Gene	Central			Peripheral
		ONL	INL	GCL	GCL
Nrf2fl/fl	Sema6A	33.96 ± 0.24	32.83 ± 0.73	33.26 ± 0.35	33.14 ± 0.41
	Cyclophilin A	28.09 ± 0.11	26.57 ± 0.40	27.98 ± 0.20	27.34 ± 0.35
Nrf2fl/fl;Six3-Cre	Sema6A	34.22 ± 0.39	31.67 ± 0.45	31.69 ± 0.38	32.65 ± 0.73
	Cyclophilin A	28.73 ± 0.18	26.28 ± 0.26	27.63 ± 0.19	27.76 ± 0.54

Data are presented as mean ± SEM.

Hypoxia Accentuates Sema6A Expression in Nrf2-Deficient RGCs via Hypoxia-Inducible Factor-1 Alpha.

Our in vivo studies indicate that Sema6A expression is regulated in RGCs by Nrf2 in the ischemic phase of OIR. To further confirm this regulation and to gain insights into the underlying mechanisms, we isolated RGCs from Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre retinas and treated them with hypoxia. No Nrf2 mRNA was detected in Nrf2^fl/fl; Six3-Cre RGCs, as expected (Fig. S4A). Exposure to hypoxia for 4 h induced Sema6A mRNA expression in both Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre RGCs. With prolonged exposure to hypoxia (8–24 h), Sema6A mRNA in Nrf2^fl/fl; Six3-Cre RGCs was significantly higher than Nrf2^fl/fl (Fig. 4A and Table S3). The results indicate that Sema6A is induced by hypoxia in RGCs and that Nrf2 deletion accentuates Sema6A expression in hypoxic RGCs. Interestingly, hypoxia led to an induction of VEGF mRNA in both Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre RGCs, and accentuated levels were observed in Nrf2^fl/fl; Six3-Cre RGCs after 12 h of exposure to hypoxia (Fig. 4B and Table S3). An increase in VEGF expression was also observed in Nrf2^fl/fl; Six3-Cre retina compared with Nrf2^fl/fl at P14 of OIR (Fig. 4C). The results indicate that accentuation of Sema6A expression by Nrf2 deficiency is accompanied by an increase in VEGF in RGC. The pattern of Sema6A induction by hypoxia is different from what has been reported for Sema3A, which is induced after a longer period of exposure to hypoxia (36 h) when VEGF induction has subsided (2).

Fig. S4.

(A) Quantitative RT-PCR analysis of Nrf2 in isolated primary RGCs from Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre mice. (B) Quantitative RT-PCR analysis of Nrf2 and NQO1 in RGC-5 treated with control siRNA or Nrf2 siRNA for 36 h. (C) Quantitative RT-PCR analysis of Sema6A in RGC-5 treated with control siRNA or Nrf2 siRNAin 1% O₂. Data are presented as mean ± SEM. *P < 0.05; **P < 0.01.

Fig. 4.

Hypoxia accentuates expression of Sema6A in Nrf2-deficient RGCs via HIF-1α. (A and B) Quantitative RT-PCR analysis of Sema6A (A) and VEGF (B) in primary RGCs isolated from Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre mice. Exposure to hypoxia for 4 h induces an increase in Sema6A mRNA in both Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre RGCs, and as hypoxic exposure is prolonged (8 h to 24 h), Sema6A mRNA in Nrf2^fl/fl; Six3-Cre RGCs is significantly higher than Nrf2^fl/fl. Hypoxia leads to an induction of VEGF mRNA in both Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre RGCs and the accentuated levels were observed in Nrf2^fl/fl; Six3-Cre RGCs after 12 h exposure. (C) Nrf2^fl/fl; Six3-Cre retina exhibits a significant increase in VEGF protein level compared with Nrf2^fl/fl at P14 of OIR. n = 5. (D) Immunohistochemical staining of retinas from human fetus (24–28 wk gestation) shows that Sema6A (blue) expression is increased in NFL (“[”) and GCL (outlined by dashed line) in nonvascularized retina [no CD34 (blue) and strong HIF-1α (blue) staining (arrows indicate nuclear localization)] compared with central vascularized retina [CD34 (arrowheads) and weak HIF-1α staining)]. Nuclei were stained in red. (E) Immunohistochemical staining of retinas from human fetus (24–28 wk gestation) reveals the presence of Nrf2 (blue) in NFL and GCL and the increased Nrf2 expression in nonvascularized retina, similar to the expression of Sema6A (blue). Arrowheads indicate nuclear localization of Nrf2. Nuclei were stained in red. (F and G) The increase of Sema6A mRNA (F) or protein (G) in Nrf2-deficient RGC-5 exposed to hypoxia for 10 h is abrogated by Digoxin. Digoxin (100 nM) was added 2 h before exposure to hypoxia. (Scale bar, 25 µm.) Data are presented as mean ± SEM. *P < 0.05; **P < 0.01.

Table S3.

Ct values from quantitative RT-PCR shown in Fig. 4 A and B

Strain	Gene	Time of exposure to 1% O2 (h)
		0	4	8	12	24
Nrf2fl/fl	Sema6A		23.61 ± 0.26	22.54 ± 0.16	24.66 ± 0.21	26.13 ± 0.27
	Cyclophilin A	24.09 ± 0.35	21.93 ± 0.39	21.19 ± 0.30	21.92 ± 0.21	22.67 ± 0.26
Nrf2fl/fl;Six3-Cre	Sema6A	24.28 ± 0.46	22.51 ± 0.38	22.51 ± 0.53	24.51 ± 0.37	24.37 ± 0.31
	Cyclophilin A	21.92 ± 0.39	20.51 ± 0.22	20.91 ± 0.33	22.55 ± 0.42	22.12 ± 0.64
Nrf2fl/fl	VEGF	30.80 ± 0.62	26.15 ± 0.09	25.46 ± 0.46	25.89 ± 0.30	26.80 ± 0.19
	Cyclophilin A	23.75 ± 0.58	21.05 ± 0.46	20.78 ± 0.23	21.38 ± 0.26	22.27 ± 0.20
Nrf2fl/fl;Six3-Cre	VEGF	28.34 ± 0.35	25.26 ± 0.40	24.50 ± 0.53	25.77 ± 0.26	25.36 ± 0.42
	Cyclophilin A	21.34 ± 0.33	20.21 ± 0.26	20.40 ± 0.49	21.87 ± 0.60	21.92 ± 0.80

Data are presented as mean ± SEM.

We performed complementary expression studies of Sema6A in human retina specimens. Immunohistochemical staining of human fetal eyes showed that Sema6A was present in the central, vascularized retina and localized to the nerve fiber layer (NFL), GCL, inner plexiform layer, and outer plexiform layer. Strikingly, Sema6A expression was increased in both the NFL and GCL in nonvascularized peripheral retina compared with expression in the central vascularized retina; this was accompanied by hypoxia-inducible factor-1 alpha (HIF-1α) induction in the peripheral retina (Fig. 4D). The localization of Sema6A in nonvascular fetal retina was similar to that observed in the retina from embryonic mouse, in which Sema6A mRNA was found predominantly in inner neuroblastic layers (28). The expression pattern of Sema6A in NFL and GCL was consistent with that observed for Nrf2, with strong expression of Nrf2 in peripheral (avascular) retina (Fig. 4E).

Given the critical role HIF-1α is known to play in regulating gene expression in hypoxia/ischemia, we investigated whether the induction of Sema6A in Nrf2-deficient RGC might involve modulation of HIF-1α. For this purpose, we used RGC-5 cells, a retinal neuronal cell line that has been a useful cell culture system for investigating neurovascular interactions in retinal ischemia (2, 27, 29, 30). We found that knockdown of Nrf2 (Fig. S4B) led to an induction of both HIF-1α (Fig. 4G) and Sema6A (Fig. S4C and Fig. 4 F and G) in RGC-5 exposed to hypoxia. Suppression of HIF-1α abrogated the induction of Sema6A that resulted from Nrf2 knockdown under hypoxic conditions (Fig. 4 F and G). These results suggest that the regulation of Sema6A by Nrf2 in hypoxic stress-injured neurons is HIF-1α-dependent.

Extracellular (Immobilized) Sema6A Induces Endothelial Cell Contraction and Suppresses Migration via Activation of Notch Signaling.

The in vivo studies indicate a strong role for neuronal Nrf2, and particularly Nrf2 in RGCs, in the regulation of reparative angiogenesis into the avascular retina in OIR. The associated modulation of expression of the transmembrane molecule Sema6A suggests this guidance gene may be an effect or molecule that mediates the suppression of revascularization in Nrf2-deficient retinas. To explore this concept, we investigated the effects of extracellular Sema6A on EC motility. To mimic the extracellular presentation of the transmembrane Sema6A by RGCs to the neighboring endothelial cells, human retinal endothelial cells (HRECs) were cultured on a surface coated with immobilized recombinant Sema6A (Fig. 5A). Sema6A induced a dose-dependent cellular contraction and reduction in EC surface area (Fig. 5 B and C), consistent with a modulation of ECs away from the migratory phenotype. Similar to its effects on cell contraction, Sema6A suppressed both EC migration in a wound scratch assay (Fig. 5 I and J) and EC tube formation (Fig. 5 D and E) in a dose-dependent fashion. The secreted semaphorin, Sema3E, has been found to suppress Notch signaling in developmental angiogenesis (31). In addition to its critical regulatory role of EC sprouting, Notch signaling in ECs is also known to regulate EC migration (32). In this study, we found that Sema6A increased EC levels of cleaved Notch1, Notch target genes, and the Notch 1 ligand Dll4 in a dose-dependent fashion (Fig. 5 F–H). Sema6A at a higher dose also up-regulated levels of Notch1. The pattern of dose-dependent induction of cleaved Notch1 and Notch targets are consistent with the dose-dependent induction of Dll4, suggesting that Notch1 activation in Sema6A-treated EC is more likely mediated by Dll4. In addition, treatment of ECs with DAPT abrogated the antimigratory effects of Sema6A on ECs, even at the highest dose of Sema6A (Fig. 5 I and J), indicating that Sema6A suppresses migration via activation of Notch signaling.

Fig. 5.

Extracellular (immobilized) Sema6A leads to dose-dependent endothelial cell contraction and suppression of migration through Notch signaling. (A) Schematic diagram of treatment with immobilized Sema6A in a coating plate. (B and C) The F-actin network of HRECs stained with rhodamine-phalloidin. HRECs cultured in a Sema6A-coated plate exhibited a dose-dependent cellular contraction. (Scale bar, 20 µm.) (D and E) Sema6A leads to a dose-dependent decrease of HREC tube formation. (Scale bar, 100 μm.) (F and G) Immunoblot analysis shows that HRECs treated with 0.2 mg/L or 1 mg/L Sema6A for 8 h exhibits an increase in Notch1 intracellular domain (cleaved Notch1) and its ligand Dll4. GAPDH was detected as a loading control. (H) Quantitative RT-PCR analysis shows that the Notch target genes Hey1 and Hey2 were increased in HREC treated with 0.2 mg/L or 1 mg/L Sema6A for 8 h. (I and J) Migration of HRECs was assessed using a scratch wound assay. HRECs treated with Sema6A exhibited a dose-dependent reduction of wound closure. Inhibition of Notch signaling with DAPT (5 µM) rescues inhibition of migration in HRECs treated with Sema6A. (Scale bar, 100 µm.) Data are presented as mean ± SEM. *P < 0.05; **P < 0.01.

Inhibition of Sema6A Abrogates the Suppression of Reparative Angiogenesis and the Induction of Preretinal Neovascularization in Nrf2-Deficient Mice in OIR.

To implicate up-regulation of neuronal Sema6A as a critical mechanism for the negative regulation of revascularization by Nrf2 deficiency in OIR, we used a strategy for Sema6A knockdown in wild-type and Nrf2-deficient eyes by intravitreal injection of lentivirus containing small hairpin RNA (shRNA) targeting Sema6A (Lv.shSema6A).We first evaluated the ability of the lentivirus system to transduce the retina and drive transgene expression, using a lentivirus–GFP vector. Intravitreal injection of lentivirus at P7 effectively infected the retina, especially the inner retina (GCL and INL, predominantly the cells surrounding the blood vessels) (Fig. S5 A and B), as demonstrated by GFP-positive cells observed at P12, and led to a ∼50% reduction of Sema6A mRNA in Nrf2^+/+and Nrf2^−/− mice at P12 or P14 of OIR (Fig. S5C). Nrf2^+/+retinas treated with Lv.shSema6A exhibited a slight decrease in avascular area at P12, and a significant ∼50% reduction in avascular retinal area at P17 of OIR. In Nrf2^−/− retina, treatment with Lv.shSema6A did not affect avascular area at P12, but exhibited a significant ∼50% reduction in avascular retinal area at P17 of OIR, which is equal to the degree of retinal avascularity in Nrf2^+/+retina injected with Lv.shControl. Pathologic preretinal neovascular tufts area was significantly decreased by Lv.shSema6A treatment in both Nrf2^+/+and Nrf2^−/− retina at P17 of OIR (Fig. 6 A–C). A similar effect on avascular area and neovascular tufts area at P17 of OIR was observed in Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre retina treated with Lv.shSema6A (Fig. 6 D–F), which further confirms the role of Sema6A in the regulation of reparative angiogenesis by Nrf2 in neurons. Taken together, the data demonstrate that neuronal Nrf2 promotes vascular regeneration through inhibition of Sema6A in OIR.

Fig. S5.

(A) GFP-positive cells were observed in retinal sections at P12 of OIR from mice intravitreally injected with a lentiviral vector containing GFP (Lv.GFP) at P7, and the strong GFP staining was located in GCL and INL. (Scale bar, 25 µm.) (B) GFP was highly expressed in the cells surrounding the blood vessels (lectin-positive) in retinas intravitreally injected with Lv.GFP. (Scale bar, 10 µm.) (C) Quantitative RT-PCR analysis shows that Sema6A mRNA in retina with intravitreal injection of a lentiviral vector expressing Sema6A shRNA (Lv.shSema6A) at P7 is about twofold decreased compared with the retina injected with Lv.shControl (expressing scrambled control shRNA) in both Nrf2^+/+and Nrf2^−/− mice at P12 or P14 of OIR. n = 6. Data are presented as mean ± SEM. *P < 0.05; **P < 0.01.

Fig. 6.

Inhibition of Sema6A abrogates the restrained vascular regeneration and increased preretinal neovascularization in Nrf2-deficient mice in OIR. (A) Retinal flat mounts show that intravitreal injection of Lv.shSema6A at P7 leads to a dramatic decreases in avascular area (B) and tufts (C) in both Nrf2^+/+ and Nrf2^−/− retinas at P17 of OIR compared with Lv.shControl injection. n = 8. (D) Retinal flat mounts show that intravitreal injection with Lv.shSema6A at P7 resulted in a significant decrease in avascular area (E) and tufts (F) in both Nrf2^fl/fl and Nrf2^fl/fl; Six3-Cre mice at P17 of OIR. n = 8. In A and D, blood vessels were visualized by GS lectin staining. Avascular retina was outlined by red line, and arrowheads indicate neovascularization tufts. (Scale bar, 400 µm.) Data are presented as mean ± SEM. *P < 0.05; **P < 0.01; NS, not significant.

Pharmacologic Activation of Nrf2 by CDDO-Im Improved Retinal Revascularization and Suppressed Pathologic Neovascularization.

The important role of Nrf2 in OIR suggests that pharmacologic enhancement of Nrf2 could be a therapeutic strategy for this condition. Synthetic triterpenoids have been developed as potent inducers of Nrf2 that act by promoting its nuclear translocation (33, 34). We have previously demonstrated therapeutic effects of the triterpenoids CDDO-Me and CDDO-Im in retinal ischemia-reperfusion, including a protective effect against retinal neurodegeneration (20, 35). We therefore sought to determine whether retinal revascularization and pathological neovascularization could be improved by the administration of CDDO-Im. Intravitreal injection of CDDO-Im twice at P12 and P14 of OIR led to up-regulation of Nrf2 target genes (Fig. 7 A and B) and suppression of Sema6A (Fig. 7C) in wild-type retinas, but not in Nrf2^−/− retinas (Fig. S6 A and B). Strikingly, a significant decrease in avascular area and a robust reduction of preretinal neovascularization were observed in CDDO-Im-injected wild-type retinas at P17 of OIR (Fig. 7 D–F), whereas injection of CDDO-Im did not rescue the defective vascular phenotype in Nrf2^−/− retinas (Fig. S6 C–E). Taken together, these results indicate that Nrf2 activation enhances vascular regeneration and suppresses preretinal neovascularization in OIR. Therefore, Nrf2 could be a therapeutic target in diseases related to ischemia-induced angiogenesis in the retina and CNS.

Fig. 7.

Pharmacologic activation of Nrf2 by CDDO-Im improves retinal revascularization and reduces pathologic neovascularization. (A) Schematic of CDDO-Im administration in OIR. Wild-type mice were intravitreally injected with 1 µL 24 nM CDDO-Im twice at P12 and P14 of OIR. (B and C) Quantitative RT-PCR analysis shows that the Nrf2 target genes (B) and Sema6A (C) expression were increased 6 h after CDDO-Im injection at P14. n = 4. (D) Retinal flat mounts of wild-type mice at P17 of OIR show that injection with CDDO-Im results in a significant decrease in avascular area (E) and preretinal neovascularization (tufts) (F). Avascular retina was outlined by red line, and arrowheads indicate neovascularization tufts. n = 9. (Scale bar, 400 µm.) Data are presented as mean ± SEM. *P < 0.05; **P < 0.01.

Fig. S6.

Pharmacologic activation of Nrf2 by CDDO-Im has no effect on retinal revascularization in OIR in Nrf2^−/− mice. Nrf2^−/− mice were intravitreally injected with 1 µL 24 nM CDDO-Im twice at P12 and P14 of OIR. (A and B) Quantitative RT-PCR analysis shows that no change of the Nrf2 target genes (A) or Sema6A (B) expression was observed in retinas after CDDO-Im injection at P14 (6 h later). n = 4. (C) Retinal flat mounts of Nrf2^−/− mice at P17 of OIR show that injection with CDDO-Im does not lead to a decrease in avascular area (D) and preretinal neovascularization (tufts) (E). Avascular retina was outlined by red line, and arrowheads indicate neovascularization tufts. n = 9. (Scale bar, 400 µm.) Data are presented as mean ± SEM. NS, not significant.

Discussion

An important challenge in ischemic diseases of the CNS and retina is the need to accelerate revascularization. Although a highly desirable goal, no clinically approved therapies exist to achieve this end. It is likely that multiple cellular elements participate in dictating this revascularization process involving cell–cell communication. It is speculated that transition from an injury to repair state might play a role in promoting revascularization, but the signals that mediate this transition continue to be explored. In this study, we introduce several concepts relating to molecular regulators of reparative angiogenesis. First, we demonstrate that Nrf2 is activated in the ischemic retina and significantly influences the extent of revascularization. Endothelial Nrf2 promotes vascularization by enhancing vascular sprouting and branching, and neuronal Nrf2 plays a crucial role in vascular regrowth toward the avascular zone. Nrf2 activity in ischemic neurons, beneath the regenerating vasculature, suppresses expression of a cell-surface semaphorin, Sema6A, through a HIF-1α-dependent mechanism. This Sema6A appears to be part of a neuronal stress response, and its antiangiogenic effect on EC motility renders the neuroretina resistant to reparative angiogenesis. We postulate that an important aspect of Nrf2’s reprogramming of the ischemic neuroretina in favor of angiogenesis includes the suppression of this antiangiogenic effect. Importantly, this neuronal Nrf2 activity, as well as the endothelial Nrf2 activity, are subject to pharmacologic modulation that can further tilt the scales in favor of revascularization by regulating the balance between pro- and antiangiogenic molecules including Sema6A (Fig. 8 A and B).

Fig. 8.

Schematic of principal findings depicting a crucial role of Nrf2 in promoting vascular regeneration in ischemic retinopathy. (A) Nrf2 is activated in EC and RGC during vascular regeneration of ischemic retinas. Loss of Nrf2 in EC leads to a mild decrease in revascularization and an increase in preretinal neovascularization. Nrf2 deficiency in RGC results in an induction of Sema6A, which causes a marked reduction of vessel regrowth. The vascular regeneration is improved by pharmacologic activation of Nrf2, as evidenced by enhanced revascularization and inhibitory preretinal neovascularization with a decrease of Sema6A. (B) Upon ischemia, endothelial Nrf2 promotes sprouting, branching, and vascularization, which plays an important role in the restraint of preretinal neovascularization. Nrf2 activation in RGC promotes vessel regrowth into the avascular zone by suppression of hypoxia-induced Sema6A through a HIF-1α-dependent mechanism.

A significant facet of Nrf2 biology in the context of neuroischemic disease is its activation by, and modulation of, hypoxia. We found that Nrf2 is activated early on during the ischemic phase of oxygen-induced retinopathy, via nuclear translocation, which is known to be the primary mode of activation of this transcription factor (7). This activation is sustained during the ischemic phase of OIR. The inner retina, specifically the central zone, is most greatly affected by retinal vaso-obliteration in the OIR model. We were therefore struck by the strong immunolocalization of Nrf2 in this ischemic region (Fig. 2A), including the ganglion cell layer and endothelial cells. This is consistent with an important functional role for Nrf2 in the setting of ischemic neuronal disease.

In cell culture studies, we and others have found that promotion of Nrf2 promotes the angiogenic phenotype in endothelial cells (12, 36). We previously reported that endothelial Nrf2 activation has a proangiogenic effect in the setting of developmental angiogenesis in the retina. Endothelial Nrf2 promoted tip cell formation and vascular branching in a cell-autonomous fashion, and the effect was mediated by the regulation of Dll4/Notch signaling (12). Consistent with this, we found in the present study that enhancement of Nrf2 (in the EC-specific Keap1 knockout) increased angiogenic sprouting and vascular density in OIR and slightly increased vascular regrowth into the central ischemic zone. As a result of increased vascularization, the retina was rendered less ischemic, and preretinal neovascularization was suppressed. Conversely, loss of Nrf2 in ECs impaired angiogenic capability, resulting in compromise of revascularization, more severe retinal ischemia, and increased pathologic preretinal neovascularization. Previous studies have found that deletion of Nrf2 led to exacerbation of angiogenesis in peripheral ischemia in limbs and lung (14, 15). This aggravation of ischemia might well be attributable, at least in part, to Nrf2 deficiency in ECs and its cell-autonomous effect on EC function.

Apart from endothelial cells, Nrf2’s effect on revascularization largely depended on its activity in retinal neuronal cells, particularly the ganglion cells (RGCs). RGCs are a primary neuronal cell type affected in OIR, consistent with the profound effect of this stress in promoting vaso-obliteration of the superficial retinal vasculature. The robust Nrf2 activation in neurons in the GCL may reflect the marked stress exerted by OIR. RGCs are thought to be key regulators of revascularization in OIR (3). An important theme that has emerged is that ischemic RGCs might have a maladaptive effect on neurovascular remodeling by secreting semaphorins, specifically Sema3A and Sema3E, guidance genes that serve to inhibit reparative angiogenesis, and thereby prevent vascular ingrowth into the ischemic zone of RGCs (1, 2). Our studies implicate Sema6A as an additional molecule that mediates the vasorepulsive effect of ischemic RGCs. Interestingly, and in contrast to Sema3A and 3E, Sema6A is known to be a membrane-associated guidance molecule. This suggests a direct contact-mediated regulation of endothelial cells by RGCs, consistent with the physical apposition of these cells with the superficial retinal vasculature (3, 37).

In our study of Nrf2, we find that the enhancement of revascularization by Nrf2 is mediated by the suppression of the membrane-associated guidance molecule Sema6A in neurons. Interestingly, the literature indicates that Sema6A can have both pro- and antiangiogenic effects, depending in part on the context of its action. Two studies investigating Sema6A in the context of endothelial cells demonstrated contrasting effects of this molecule on angiogenesis. One study demonstrated a proangiogenic effect of Sema6A in the setting of retinal vascular development and found that endothelial cell expression of Sema6A promoted EC survival, VEGFR2 expression, and responsiveness to VEGF (38). A second study of endothelial cells demonstrated that Sema6A can exert an antiangiogenic effect. siRNA-mediated knockdown of Sema6A increased EC sprout length, whereas recombinant Sema6A significantly decreased EC sprout length (39). One possible explanation for these paradoxical effects could be whether the Sema6A is operating in an autocrine (cell-autonomous) versus paracrine fashion. Importantly, the second study demonstrated that exogenously administered recombinant Sema6A was antiangiogenic and that endothelial cells were repelled from neighboring ECs that expressed Sema6A (39). Interestingly, Sema6A has been demonstrated to have opposing effects on responding neurons, depending on whether Sema6A is acting in a cis versus trans manner (40). Consistent with this notion, administration of soluble extracellular domain of Sema6A inhibited both growth factor- and tumor-induced angiogenesis (41). Our study demonstrates that Sema6A expressed by ischemic neurons exerts an antiangiogenic effect on adjacent ECs. Exogenously presented Sema6A led to dose-dependent cellular contraction and suppression of the migratory phenotype of ECs through activation of Notch signaling. In the specific setting of oxygen-induced retinopathy, exogenously presented Sema6A forms a barrier to prevent ECs from growing into the avascular (ischemic/hypoxic) zone.

Hypoxia has been found to induce up-regulation of several semaphorins, including Sema3A (2), 3E (42), 4D (43), and 7A (44), in multiple cell types, and it also induces Sema6A in the CNS. In developing mouse retina, Sema6A expression is increased at P0 (28) (a stage at which it is avascular, and therefore hypoxic), helping drive the formation of the retinal vasculature. In human fetal retina, we similarly observed higher Sema6A expression in the inner retina (GCL and NFL), and the expression was decreased in central retina, which is vascularized and less hypoxic. In mouse retinas subjected to OIR, increased Sema6A expression was observed in the central GCL, where retina is poorly vascularized and severely hypoxic. In vitro studies have demonstrated that hypoxia induces Sema4D in cancer cells (43) and Sema7A in endothelial cells (44) in a HIF-1α-dependent manner. Likewise, in our study, we found that Nrf2 exerted its effects on hypoxic induction of Sema6A by regulating HIF-1α in neuronal cells. The regulation of HIF-1α by Nrf2 has been investigated in tumor angiogenesis. Suppression of Nrf2 inhibited HIF-1α and resulted in a reduction of VEGF and blood vessel formation in human glioma cells (17) and colon cancer cells (13). Reduced mitochondrial O₂ consumption accounts for the degradation of HIF-1α in Nrf2-deficient cancer cells. In contrast to these observations, we found in our model system that loss of Nrf2 in hypoxic neuronal cells led to an increase in HIF-1α. The associated increase in VEGF, a well-known target of HIF-1α in hypoxia, also indicates the activation of HIF-1α in these Nrf2-deficient cells. Given that reactive oxygen species is an important regulator of HIF-1α (45–48), exacerbated oxidative stress caused by Nrf2 inactivation may account for the increase of HIF-1α in Nrf2-deficient cells under hypoxia.

The finding that Nrf2 and Sema6A may participate as regulators of the neuroretinal response to ischemia suggests a therapeutic strategy directed at shifting the neuroretina toward a repair response, specifically one promoting vascularization. This might involve enhancing Nrf2 activation to influence the overall neurovascular response, or suppressing Sema6A and its critical antiangiogenic effect. In our experimental system, we found that both strategies had a beneficial effect. It will be of great interest to evaluate the role of Nrf2 and Sema6A in other models, including the brain, to determine their broader involvement in ischemic disorders.

Taken together, our findings suggest an important role for Nrf2 in neurovascular regeneration in ischemic conditions. We find that Nrf2 is a sensor of hypoxic/ischemic stress, one whose activation can modulate the balance of factors that ultimately determine the extent of revascularization. Vascular regeneration is initially suppressed as part of the injury response, both at the level of the endothelial cell and the milieu that governs the angiogenic drive. The level of Nrf2 activity shifts the neuroischemic environment toward a repair response that favors revascularization. At this time, there are no approved therapies for promoting vascular regeneration in the ischemic CNS. Our study provides proof of concept for the modulation of Nrf2 or Sema6A, so as to tilt the balance toward neuronal recovery, representing a therapeutic strategy for ischemic retinopathies and ischemic CNS disorders such as stroke.

Materials and Methods

Animals.

The Nrf2^+/+, Nrf2^−/− (49), Nrf2^fl/fl (12, 50), and Keap1^fl/fl (12, 50) mice have been published previously. C57BL6/J, VE-Cadherin-Cre (#006137), and Gfap-Cre mice (#012886) were obtained from Jackson laboratory, and Six3-Cre mice (21) were generously provided by Yasuhide Furuta, M.D. Anderson Cancer Center. All animal procedures performed were approved by the Johns Hopkins University Institutional Animal Care and Use Committee. Animal care was conducted in accordance with institutional guidelines.

OIR.

The OIR model was described previously (16, 27, 51, 52). Mice pups were exposed to 75% (vol/vol) oxygen for 5 d from P7 to P12. At P12 or P17, retinas were harvested for analyzing avascular area and/or neovascularization (tufts) area. Staining and quantification were performed as previously described (52, 53). Briefly, retinal vasculatures were visualized by incubation overnight with Alexa-conjugated isolectin GS-IB₄ from Griffonia simplicifolia (GS lectin) (Life Technologies). Avascular area and tufts area were quantified by comparing the number of pixels in the vaso-obliteration or tufts area with the total number of pixels in the retina in a masked fashion. Fold change was obtained by comparison with control mice in the same background, subjected to OIR at the same time. Mice with body weight lower than 6 g (P17) were excluded from analysis (52).

Electroretinography.

Electroretinographic recording was performed as previously described (2), with minor modification, and scotopic electroretinographs were obtained. Scotopic intensity responses were generated with flashes of white light ranging in intensity from −6.3 log to 0.3 log cd/s/m² in 0.3-log increments, using a photostimulator with an interstimulus interval of 10 s, flash duration of 20 ms, and an average of 2–5 flashes. Signals were detected and a-wave and b-wave were measured using an UTAS visual electrodiagnostic system (LKC Technologies).

Cryosection and Whole-Mount Immunofluorescence Staining.

Eyeballs were enucleated and directly embedded in optimal cutting temperature medium (Sakura Finetek) at −80 °C. Section staining was performed as previously described (53). Eyeballs were fixed for 30 min in 4% (wt/vol) paraformaldehyde (PFA) at room temperature. Whole-mount immunofluorescence staining was performed as previously described (53).

Laser-Capture Microdissection.

Laser-capture microdissection was performed as previously described (2, 12). Sections (8 µm) mounted on RNase-free polyethylene naphthalate glass slides (Leica) were microdissected with the LMD 6000 system (Leica Microsystems) and collected directly into RLT lysis buffer in the RNeasy Micro Kit (QIAGEN).

Primary RGCs, HRECs, and RGC-5 Cells.

RGCs were purified as previously described (54). HRECs (Cell Systems) were maintained in endothelial cell growth medium-2-MV (Lonza), as previously described (53), and used before passage 9. RGC-5 cells were maintained in DMEM with 10% (vol/vol) FBS and terminally differentiated by exposing them to 333 nM staurosporine (Sigma) for 12 h, as described previously (55).

Knockdown of Sema6A With Lentivirus in Vivo.

The infectious lentivirus were produced by transfecting HEK293T cells with a vector containing the shRNA targeting Sema6A or scramble shRNA (Origene) and packaging plasmids (pV-SVG, pMDL, and pREV) (Open Biosystems), using Lipofectimine 2000 according to the manufacturer’s instructions. P7 mice received an intravitreal injection of 1 µL lentivirus (100–200 ng/µL) before exposure to 75% O₂.

Immunohistochemistry.

The paraffin-embedding human fetal eyes (24–28 wk gestation) were obtained from the Wilmer Eye Institute Ophthalmic Pathology Laboratory with approval from Institutional Review Board. All eyes were fixed within 15 h postmortem. Immunohistochemistry was performed using the VECTASTAIN ABC-AP kit (Vector Labs), as previously described (20).

Statistical Analysis.

Results were expressed as mean ± SEM. Data were first analyzed by KS normality test. Statistical differences of normal distributed data were assessed by Student’s t test or one-way ANOVA. Mann–Whitney test or Kruskal–Wallis test was used to analyze statistical differences of nonnormally distributed data. P values less than 0.05 were considered statistically significant.

SI Materials and Methods

Animals.

The Nrf2^+/+, Nrf2^−/− (49), Nrf2^fl/fl (12, 50), and Keap1^fl/fl (12, 50) mice have been published previously. C57BL6/J, VE-Cadherin-Cre (#006137), and Gfap-Cre (#012886) mice were obtained from Jackson laboratory, and Six3-Cre mice (21) were generously provided by Yasuhide Furuta, M.D. Anderson Cancer Center. The animals were maintained on an AIN-76A diet and water ad libitum and housed at a temperature range of 20–23 °C under 12:12-h light–dark cycles. All animal procedures performed were approved by the Johns Hopkins University Institutional Animal Care and Use Committee. Animal care was conducted in accordance with institutional guidelines.

OIR.

The OIR model was described previously (16, 27, 51, 52). Mice pups were exposed to 75% (vol/vol) oxygen for 5 d from P7 to P12. At P12 or P17, retinas were harvested for analyzing avascular area and/or neovascularization (tufts) area. Staining and quantification were performed as previously described (52, 53). Briefly, retinal vasculatures were visualized by incubation overnight with Alexa-conjugated isolectin GS-IB₄ from Griffonia simplicifolia (GS lectin) (Life Technologies). Avascular area and tufts area were quantified by comparing the number of pixels in the vaso-obliteration or tufts area with the total number of pixels in the retina in a masked fashion. Fold change was obtained by comparison with control mice in the same background, subjected to OIR at the same time. Mice with body weight lower than 6 g (P17) were excluded from analysis (52).

Labeling of Perfused Blood Vessels.

Labeling of perfused retinal blood vessels was performed as previously described (56). Briefly, P17 mice subjected to OIR were anesthetized and injected with 50 μL 0.33 mg/mL Alexa-Fluor 595-conjugated GS-lectin (Life Technologies) in PBS with 1 mM CaCl₂ into the left cardiac ventricle. Eyeballs were removed and fixed in 4% (wt/vol) PFA 5 min after injection. Retinas were carefully dissected and incubated with Alexa-Fluor 488-conjugated GS-lectin (Life Technologies) diluted with phosphate buffered saline Tween-20 (PBST) plus 0.1 mM CaCl₂ and 10% (vol/vol) normal goat serum (NGS) for 1 h at room temperature. Fluorescently stained retinas were washed, mounted in Fluoromount-G (Electron Microscopy Sciences), and analyzed by confocal laser scanning microscopy using a Laser Scanning Microscope 710 (Zeiss). Six random fields (20× objective) of vasculature (three from central area of retina, close to the angiogenic front, and three from peripheral area) of each retina were taken for analysis; 488-conjugated GS-lectin-positive/595-conjugated GS-lectin-negative vessels were counted as nonperfused blood vessels.

Hypoxyprobe.

The hypoxic retinal area was assessed by Hypoxyprobe, as previously described (56). P17 mice subjected to OIR were injected with Hypoxyprobe (pimonidazoleHCl; Hypoxyprobe) intraperitoneally at 2.5 mg per pup 1 h before they were killed. Retinas were isolated after fixation of eyeballs in 1% PFA for 1 h, washed 3 times with PBST, blocked with 10% (vol/vol) NGS for 1 h, and incubated with rabbit antipimonidazole antibodies (PAb2627; HPI) diluted with PBST plus CaCl₂ overnight at 4 °C. The retinas were then stained with secondary antibody and GS lectin.

Vascular Permeability Assay.

Retinal vascular permeability was assessed as previously described (53, 57). Briefly, P17 mice subjected to OIR were injected intraperitoneally with 1 µCi/g body weight of [³H]-mannitol 1 h before sacrifice. Retinas were dissected, and vitreous and blood were carefully removed. Retinas were then incubated overnight at 50 °C with 1 mL NCSII solubilizing solution (GE Healthcare). After decolorization with 20% (wt/vol) benzoyl peroxide, 5 mL scintillation fluid cytoscint ES (ThermoFisher Scientific) and 30 μL glacial acetic acid were added and the vials were kept in darkness for several hours at 4 °C. Radioactivity was counted with a scintillation counter (LS 6500 Liquid Scintillation Counter; Beckman-Coulter). Results were normalized to the dry weight of each retina.

Electroretinography.

Electroretinographic recording was performed as previously described (2), with minor modification, and scotopic electroretinographs were obtained. Briefly, P40 mice were anesthetized after 12 h dark adaptation. Drops of 1% cyclopentolate hydrochloride (Alcon) were used to dilate the pupils. A platinum loop electrode was placed on the cornea with 2.5% (wt/vol) hypromellose. The reference and ground electrodes were placed s.c. under the scalp and tail, respectively. Scotopic intensity responses were generated with flashes of white light ranging in intensity from −6.3 log to 0.3 log cd/s/m² in 0.3-log increments, using a photostimulator with an interstimulus interval of 10 s, a flash duration of 20 ms, and an average of 2–5 flashes. Signals were detected and a-wave and b-wave were measured using a UTAS visual electrodiagnostic system (LKC Technologies).

Western Blot.

Retina samples were isolated and snap-frozen in liquid nitrogen, and nuclear and cytosolic extracts were prepared using NE-PER Nuclear and Cytoplasmic Extraction Kit (Thermo Scientific) according to the manufacturer’s instructions. Cell lysates were prepared with RIPA (Sigma) according to the manufacturer’s instructions. Western blot was performed as previously described (53). Protein extracts were separated by SDS/PAGE, and the proteins were transferred to Hybond ECL nitrocellulose membrane (GE Healthcare). The membranes were then blocked and probed with monoclonal rabbit anti-Nrf2 (Abcam), monoclonal rabbit anti-β-actin (Cell Signaling Technology), polyclonal goat anti-Lamin B (Santa Cruz Biotechnology), polyclonal rabbit anti-HIF-1α (Abcam), polyclonal goat anti-Sema6A (R&D systems), monoclonal rabbit anti-cleaved Notch1 (Cell Signaling Technology), polyclonal rabbit anti-Dll4 (Cell Signaling Technology), and monoclonal mouse anti-GAPDH (Abcam). HRP-conjugated secondary antibodies were detected using SuperSignal West Pico or Femto chemiluminescent substrates (Thermo Scientific). Band intensity was quantitated using the Image J.

Quantitative RT-PCR.

Total RNA from retina was isolated using the RNeasy Mini Kit (Qiagen), then treated with DNase I (Qiagen), as previously described (12, 35). Single-stranded cDNA was synthesized using oligo (dT)12–18 primer (Invitrogen) and MMLV Reverse Transcriptase (Invitrogen). For laser-capture microdissected vessels, total RNA was purified using the RNeasy Micro Kit (Qiagen), following the manufacturer’s instructions. cDNA was synthesized using the iScript cDNA Synthesis Kit (Bio-Rad). Real-time PCR was performed using SYBR Premix Ex Taq (Takara) with Stratagene Mx3005P qPCR system (Angilent Technologies). Cyclophilin A was used for normalization. Each cDNA sample was run in duplicate.

Cryosection Immunofluorescence Staining.

Eyeballs were enucleated and directly embedded in optimal cutting temperature medium (Sakura Finetek) at −80 °C. Section staining was performed as previously described (53). Briefly, sections (8 µm) were cut and fixed for 10 min with 4% (wt/vol) PFA. After blocking with 5% (vol/vol) normal serum, sections were incubated overnight with primary antibody diluted with PBS plus 0.1% Triton X-100 at 4 °C. The following antibodies were used: monoclonal rabbit anti-Nrf2 (Abcam), polyclonal goat anti-Brn3 (Santa Cruz Biotechnology), monoclonal rat anti-PDGFRα (BD Biosciences), monoclonal rabbit anti-Vimentin (Sigma), polyclonal goat anti-Sema6A (R&D Systems), and monoclonal rabbit anti-Tuj1 (Covance). Secondary detection was performed with appropriate Alexa-conjugated secondary antibodies (Invitrogen). Stained samples were covered with Vectashield mounting medium containing 4.6-diamidino-2-phenylindole (Vector Laboratories) and visualized using a Laser Scanning Microscope 710 (Zeiss).

Whole-Mount Immunofluorescence Staining.

Eyeballs were fixed for 30 min in 4% (wt/vol) PFA at room temperature. Whole-mount immunofluorescence staining was performed as previously described (53). After fixation, retinas were dissected and blocked for 1 h in 10% (vol/vol) normal serum in PBS with 0.3% Triton X-100. Retinas were then incubated overnight with diluted polyclonal goat anti-Sema6A (R&D Systems) at 4 °C. Alexa-conjugated secondary antibodies (Invitrogen) and GS lectin (Invitrogen) were added and incubated overnight at 4 °C. Fluorescently stained retinas were washed, mounted and analyzed by confocal laser scanning microscopy, using a Laser Scanning Microscope 710 (Zeiss).

Laser-Capture Microdissection.

Laser-capture microdissection was performed as previously described (2, 12). Eyeballs were enucleated and directly embedded in optimal cutting temperature medium at −80 °C. Sections (8 µm) mounted on RNase-free polyethylene naphthalate glass slides (Leica) were fixed in 50% (vol/vol) and 75% (vol/vol) ethanol and washed with DEPC-treated water. Sections were then incubated for 20 seconds with Histogenes Staining Solution (Life Technologies). After dehydration with ethanol, layers were microdissected with the LMD 6000 system (Leica Microsystems) and collected directly into RLT lysis buffer in the RNeasy Micro Kit (QIAGEN). One-third of retinal layers from central or peripheral were collected separately.

VEGF ELISA.

VEGF protein concentration was assessed using Mouse VEGF DuoSet ELISA kit (R&D Systems), as previously described (12, 58). Briefly, retina was ultrasonically homogenized in 0.1% Triton X-100 in PBS containing a mixture of protease inhibitors (Invitrogen). The samples were cleared by centrifugation and supernatant and were subjected to measurement according to the manufacturer’s instructions. Each sample was performed in duplicate. VEGF concentration was calculated based on a standard curve and normalized by protein concentration.

Isolation of Primary RGCs.

RGCs were purified as previously described (54). Briefly, P0–P2 mice were killed, and retinas were dissociated with papain. The suspension of retinal cells was immunopanned on plates coated with anti-Thy1.2 antibody (Serotec; MCA028) and goat anti-mouse IgM (Jackson Immunoresearch) at room temperature. RGCs were released from the plate by a cell lifter, counted, and seeded at a density of 10,000 cells per well in 96-well plates in the media containing Neurobasal (Life Technologies), B27, N2 supplement, l-glutamine, and penicillin/streptomycin.

HRECs and RGC-5 Cells.

HRECs (Cell Systems) were maintained in endothelial cell growth medium 2-MV (Lonza), as previously described (12, 53), and used before passage 9. RGC-5 cells, a retinal neuronal cell line, were maintained in DMEM with 10% (vol/vol) FBS and terminally differentiated by exposing them to 333 nM staurosporine (Sigma) for 12 h, as described previously (55). For silencing experiments, HRECs or RGC-5 were transfected with 25 nM negative control small-interfering RNA (siRNA, Cat. 4390846; Ambion) or human Nrf2 siRNA (Cat. 4392420; Ambion), using Lipofectamine 2000 transfection reagent (Invitrogen) according to the manufacturer’s instructions. For hypoxia experiment, RGC5 cells cultured in fresh DMEM were put into chamber with 1% O₂.

F-Actin Network Staining.

HRECs were seeded on the coverslips coated with human recombinant Sema6A-Fc (R&D Systems) or BSA. The coating coverslips were prepared using attachment factor (Life Technologies), as previously described (12, 59). Eight hours later, HRECs were fixed for 10 min with 4% (wt/vol) PFA and permeabilized with 0.2% Triton X-100. The actin network was visualized by staining with rhodamine-phalloidin (Life Technologies) for 30 min, and the images were obtained by confocal laser scanning microscopy, using a Laser Scanning Microscope 710 (Zeiss). Fifty cells in each well were randomly selected for analysis.

Tube Formation Assay.

Tube formation assay was performed as previously described (12, 60, 61). HRECs were cultured in the middle of two layers of polymerized bovine type I collagen (Advanced Biomatrix) in a 24-well plate (7 × 10⁴ per well). Images were taken after 24 h, using Axiovert 200 M microscope (Zeiss) with 10× objective. Ten random fields in each well were selected for analysis. Total tube length per field was quantified using Image J software in a masked fashion.

Wound Healing Migration Assay.

HRECs were seeded on a 12-well plate coated with human recombinant Sema6A-Fc (R&D Systems) or BSA (12, 59, 62). The wound healing scratch assay was performed as previously described (63). When reaching 80–90% confluence, cell layer was scraped with a plastic pipette tip and washed three times with endothelial cell basal medium 2. The remaining cells were cultured in endothelial cell growth medium 2-MV for 10 h. Fourteen injured fields in each well were photographed, and wound closure area was quantified using Photoshop.

Knockdown of Sema6A with Lentivirus in Vivo.

The infectious lentivirus were produced by transfecting HEK293T cells with a vector containing the shRNA targeting Sema6A or scramble shRNA (Origene) and packaging plasmids [pV-SVG (for expression of VSV-G envelope glycoprotein), pMDL (for expression of gag-pol gene), and pREV (for expression of viral rev gene)] (Open Biosystems), using Lipofectimine 2000 according to the manufacturer’s instructions. The virus supernatant (Lv.shSema6A or Lv.shControl) was collected after 48 h and concentrated by ultracentrifugation at 50,000 × g for 2 h (about 250-fold). The titers of lentivirus were assessed with Lenti-X p24 Rapid Titer Kit (Clontech). P7 mice were received an intravitreal injection of 1 µL lentivirus before exposure to 75% (vol/vol) O₂. For each mouse, one eye was injected with Lv.shSema6A, and the contralateral eye was injected with Lv.shControl. The concentration of injected Lv.shSema6A and Lv.shControl was 100–200 ng/µL. The lentiviral vectors contain a gene encoding GFP driven by CMV promoter, which allows for conformation of infection.

Immunohistochemistry.

The paraffin-embedding human fetal eyes (24–28 wk gestation) were obtained from the Wilmer Eye Institute Ophthalmic Pathology Laboratory with approval from Institutional Review Board. All eyes were fixed within 15 h postmortem. Immunohistochemistry was performed using the VECTASTAIN ABC-AP kit (Vector Labs), as previously described (20). Briefly, sections were first subjected to deparaffinization and antigen unmasking. Sections were then blocked with normal serum and incubated overnight with diluted primary antibodies at 4 °C. The biotinylated secondary antibodies, ABC reagent, and Vector blue AP substrate (Vector Labs) were sequentially applied to the sections. Sections were counter stained with nuclear fast red (Vector Labs). The following primary antibodies were used: polyclonal rabbit anti-Sema6A (Abcam), monoclonal mouse anti-CD34 (Biolegend), polyclonal rabbit anti-HIF-1α (Abcam), and monoclonal rabbit anti-Nrf2 (Abcam). The normal IgG was used as a negative control.

CDDO-Im Treatment.

Wild-type C57BL6/J mice or Nrf2^−/− mice subjected to OIR were intravitreally injected twice with 1 µL 24 nM CDDO-Im [kindly provided by Michael B. Sporn, Dartmouth School of Medicine, dissolved in 10% (vol/vol) DMSO, 10% (vol/vol) cremophor-EL, 80% (vol/vol) PBS] or vehicle at P12 and P14. For each mouse, one eye was injected with CDDO-Im, and the contralateral eye was injected with vehicle. Results were normalized to vehicle-injected eye.

Statistical Analysis.

Results were expressed as mean ± SEM. Normal distribution of data were analyzed by KS normality test. Statistical differences of normally distributed data were determined using Student’s t test (two groups’ comparison) or one-way ANOVA, followed by Newman-Keuls test (three or more groups’ comparison). Mann–Whitney test (two groups’ comparison) or Kruskal–Wallis test, followed by Dunn’s test (three or more groups’ comparison), were used to analyze statistical differences of nonnormally distributed data. Paired t test was used for results of intravitreal injection experiments, and all others were analyzed by unpaired t test. Unpaired t test with Welch’s correction was used to analyze data with unequal variances. P values less than 0.05 were considered statistically significant.