A novel multiplex RNAi therapy simultaneously targets Hif1a and Hif2a to defy retinal degeneration in two models of AMD

TL;DR

A new RNA therapy targeting Hif1a and Hif2a in the eye shows promise in preventing retinal degeneration linked to age-related macular degeneration.

Contribution

A dual-acting RNAi therapy using a single virus to target Hif1a and Hif2a in different retinal cells is introduced.

Findings

The therapy preserved photoreceptors and RPE in two AMD models for up to 61 weeks.

Dual-acting virus outperformed single-target viruses in efficacy.

The approach targets a conserved pathway for chronic hypoxia-related diseases.

Abstract

Age-related tissue changes lead to reduced oxygen delivery to photoreceptors and the retinal pigment epithelium (RPE) and contribute to the pathology of age-related macular degeneration (AMD). The implication of hypoxia-inducible factors (HIFs) in this process makes them good candidates as therapeutic targets for AMD. We developed a multiplex dual-acting therapy utilizing the shRNAmir system, delivered by a single adeno-associated virus, that reduces mRNA levels of Hif1a in photoreceptors and Hif2a in the RPE. This RNA interference (RNAi)-based strategy demonstrated a strong therapeutic effect, potently preserving photoreceptors and the RPE in two models of pseudo- and true hypoxia up to 61 weeks post-injection. The efficacy of our dual-acting virus proved superior to single-acting viruses targeting only Hif1a in photoreceptors or Hif2a in the RPE. By targeting a common, conserved disease pathway, this gene-agnostic RNAi therapy shows significant potential to protect tissues from chronic hypoxic insults in complex diseases such as AMD.

A novel multiplexed RNA-based therapy was presented by Samardzija and colleagues, utilizing a single adeno-associated viral vector to silence two hypoxia-inducible transcription factors, Hif1a and Hif2a, in photoreceptors and the retinal pigment epithelium. This provides a potential strategy for preventing retinal degeneration by targeting common disease pathways related to hypoxia.

Figures

Gene expression in the human retina and RPEExpression of *VEGFA*, *ANGPTL4*, *SERPINE1*, *GNAT2*, and *RPE65* was determined by real-time PCR in the retina (A) and RPE (B) of young (Y) and old (O) human donor eyes without diagnosed ocular pathology. Shown are means (line) and individual data points on a log10 scale. Outliers were identified and excluded using the Grubbs method, and significance was determined by unpaired t tests. *n* = 15 (young) and 14 (old).

Evaluation of the pattern and tolerability of shHif expression in WT mice after AAV-mediated delivery(A) Illustration of transvitreal AAV delivery to the subretinal space targeting photoreceptor cells (shown in green due to expression of GFP) and RPE cells (shown in red due to expression of mCherry). (B and C) Fundoscopy, OCT scans, and fundus fluorescence of WT mice injected with shCtrl (B) or shHif (C) at 2 wpi. Orange and red lines in the fundus images indicate positions of the respective OCT scans. Fluorescence indicates transduced areas in the retina (green, GFP) and RPE (red, mCherry). W

Assessment of the *RPE*^*ΔVegfa*^ phenotype(A) *RPE*^*ΔVegfa*^ model schematics: *Vegfa* deletion in the RPE leads to reduced choroidal vessel density, RPE alterations, and photoreceptor degeneration. (B) *Vegfa* (red) *in situ* hybridization and Cre (magenta) immunofluorescence in areas that were or were not transduced with AAV::VMD2-Cre-P2A-EGFP. (C) PODXL immunofluorescence in endothelial cells of choroidal flatmounts (top) and cross-sections (toward bottom) of indicated areas. Contrast-enhanced threshold images allowed pixel intensity scan analysis (bottom) and quantification (right) in *R

Rescue of the *Rod*^*ΔVhl*^ phenotype by the dual-acting shHif RNA therapy(A) Schematic representation of *Rod*^*ΔVhl*^ mice and rescue of RPE (red) and photoreceptors (green) through downregulation of *Hif2a* in the RPE and *Hif1a* in photoreceptors and *Hif2a* in the RPE by shHif. (B) Fundoscopy, OCT scans, and fundus fluorescence assessed at 26 weeks of age (22 wpi) in *Rod*^*ΔVhl*^ mice that were either non-injected (left) or treated with shCtrl (center) or shHif (right). Orange and red lines in the fundus images indicate positions of the respective OCT scans. OCT scans of higher magnifica

Dual-acting multiplex shRNA therapy is more effective than single-acting shRNA vectors in rescuing pathology in *RPE*^*ΔVegfa*^ mice(A) Fundoscopy, OCT scans, and fundus fluorescence of *RPE*^*ΔVegfa*^ mice injected with shCtrl (left) or shHif (right) at 39 wpi. Orange and red lines in the fundus images indicate positions of the respective OCT scans that are shown in lower and higher magnifications. Fundus fluorescence indicates transduced areas in the retina (green, GFP) and RPE (red, mCherry). ∗, optic nerve head. (B) Color-coded ONL thickness maps generated from OCT volume scans of *RPE*^*Δ

Full text

Introduction

Age-related macular degeneration (AMD) is a leading cause of vision loss in the elderly worldwide, primarily affecting the photoreceptors and the retinal pigment epithelium (RPE) in the macular region. These cells degenerate as the disease progresses, leading to reduced visual acuity and eventually central vision loss. Current therapeutic strategies largely focus on targeting secreted pro-angiogenic proteins involved in disease progression. However, emerging RNA-based therapies hold considerable promise as alternatives, offering the potential to provide more effective treatments by targeting intracellular pathways that are otherwise difficult to drug.1

Age-related tissue changes such as reduced choroidal thickness, vascular density and blood flow, ghost vessel formation, thickening of Bruch’s membrane and accumulation of drusen deposits2^,^3^,^4^,^5 lead to reduced tissue oxygenation and increased VEGFA secretion, an established driver of neovascularization in the exudative (wet) form of AMD. Notably, these age-related tissue changes may also be relevant to the non-exudative (dry) form of AMD, as they have been correlated with drusen formation2^,^3^,^4^,^5 and suggested to precede photoreceptor or RPE degeneration.6

Given the critical role of hypoxia-inducible factors (HIFs) in the cellular response to low oxygen levels, chronic activation of the HIF pathway may contribute significantly to the development of AMD and other retinal diseases. This notion is strongly supported by the beneficial effect of 32-134D, a compound that blocks the transcriptional activity of both HIF1 and HIF27 in models of diabetic retinopathy,8 oxygen-induced retinopathy,9 as well as by the genetic inactivation of Hif1a in photoreceptors10^,^11 or Hif2a in the RPE,12 which prevented cell loss and retinal degeneration in experimental models of AMD. Since chronic activation of HIF2, but not HIF1, in the RPE causes RPE atrophy and retinal thinning,12 and constitutively active HIF1, but not HIF2, in photoreceptor cells induces photoreceptor degeneration, a targeted therapy to specifically reduce the expression of the respective HIF isoform in the critical cell type may be an effective approach to manage disease pathology and preserve vision in AMD patients.

The need to target two distinct genes in two different cell types presents a unique challenge in developing an effective therapy. To meet this challenge, we have combined an RNA interference (RNAi) strategy with the adeno-associated virus (AAV) delivery system that allows targeted and long-lasting expression of therapeutic genes in retinal cells. Conventional therapeutic RNA knockdown strategies using small interfering RNAs (siRNAs) are often limited because they interfere with endogenous microRNA (miRNA) biogenesis, lead to an overload of transport proteins, or cause other undesirable side effects.13^,^14^,^15 To circumvent such problems, we have generated a multiplex RNAi approach using the miRNA-based short hairpin RNA mimicking miRNA (shRNAmir) system, which directs the therapeutic RNAs into the miRNA pathway already at the level of DROSHA in the cell nucleus.16^,^17 Sequential processing by DROSHA and DICER ensures the generation of precisely cleaved siRNAs while minimizing interference with the natural miRNA processing machinery.17 This increases efficiency while reducing off-target effects and immune responses.18^,^19 The use of the miR-E instead of the conventional miR-30 backbone for shRNA expression increases the knockdown efficiency of target genes and allows cell-type-specific expression when placed in the 3′ UTR of genes that are controlled by RNA Pol II promoters.20 Using this system, we developed a novel dual-acting RNA-based therapy that simultaneously targets Hif1a in photoreceptors and Hif2a in the RPE, delivered by a single AAV. The strong and long-lasting therapeutic potential of this system was demonstrated in two hypoxia-related animal models of AMD.

Results

Age-related tissue changes impair oxygen delivery to the RPE and photoreceptors leading to a chronic molecular response to hypoxia that can be harmful to these delicate cells in the aging eye. RNAi may be an ideal therapeutic strategy to attenuate this response and support long-term cell survival. To investigate the impact of aging on the expression of known hypoxia-regulated genes21^,^22^,^23 associated with AMD,22^,^24^,^25 we determined the expression of VEGFA, angiopoietin-like 4 (ANGPTL4), and serpin family E member 1 (SERPINE1) in the retina and RPE of young (17–55 years old, median: 35 years old) and old (71–94 years old, median: 82 years old) human donor eyes (Table S1). G protein subunit alpha transducin 2 (GNAT2), a gene essential for phototransduction in cones,26 and retinoid isomerohydrolase RPE65 required for the visual cycle in the RPE,27 were included as controls. The observed increased expression of all three hypoxia-regulated genes, but not of the control genes, in the neuronal retina and the RPE of old donor eyes (Figure 1) points to an elevated activity of HIF transcription factors with age. Reducing the expression of HIF transcription factors in the retina and RPE by RNAi may attenuate their activity, reduce the expression of AMD-related HIF target genes, and thus provide significant therapeutic benefits.Figure 1. Gene expression in the human retina and RPEExpression of VEGFA, ANGPTL4, SERPINE1, GNAT2, and RPE65 was determined by real-time PCR in the retina (A) and RPE (B) of young (Y) and old (O) human donor eyes without diagnosed ocular pathology. Shown are means (line) and individual data points on a log10 scale. Outliers were identified and excluded using the Grubbs method, and significance was determined by unpaired t tests. n = 15 (young) and 14 (old).

Efficient targeting of Hif1a and Hif2a using the shRNAmir system

Chronically active HIF1 is the main driver for degeneration of photoreceptors, while HIF2 is the pathologic isoform in RPE cells.10^,^12 To prevent hypoxia-related degeneration (Figure 2A), we developed an RNAi strategy that simultaneously silences HIF1 in photoreceptors and HIF2 in the RPE by targeting the mRNAs for the respective HIF alpha subunits in the specific cell types. To reduce Hif1a expression, we used a published and previously validated shHif1a RNA sequence that targets both the mouse and human sequences,10 and we refer to this sequence as shHif1. To identify an efficient shHif2a sequence, six different shRNAs (shHif2_1–6) were stably expressed in 661W cells using a lentiviral system and tested for their efficiency to reduce expression of Hif2a. All six shRNAs were similarly efficient and reduced Hif2a mRNA levels in hypoxic cells (6 h, 0.2% O_2_) to 2–9% when compared to the short hairpin control (shCtrl) sequence (Figure S1). Since mouse shHif2_6 has only one mismatch to the human HIF2A RNA, all further experiments were performed using this sequence, which we refer to as shHif2.

Hif1a and Hif2a mRNA and protein levels were measured in normoxic and hypoxic (24 h, 0.2% O_2_) 661W cells stably expressing shHif1, shHif2 or shCtrl (Figures 2B and 2C). Hif1a mRNA expression was reduced by about 65% in both normoxic and hypoxic 661W-shHif1 cells when compared to 661W-shCtrl cells, while it remained at control levels in 661W-shHif2 cells (Figure 2B). Hif2a expression was reduced by roughly 80% in normoxic and hypoxic 661W-shHif2 but not 661W-shHif1 cells. Interestingly, an 86% increase in Hif2a levels was observed in hypoxic 661W-shHif1 cells, suggesting a compensatory induction of Hif2a upon Hif1a knockdown in hypoxia, as reported by others.28 The hypoxic state of the cells was confirmed by the significantly increased expression level of solute carrier 2a1 (Slc2a1, alias Glut1; Figure 2B), a gene primarily regulated by HIF1. The diminished but still measurable Slc2a1 expression levels in hypoxic 661W-shHif1 cells indicated a strongly but not fully blocked HIF1 activity, which is expected after robust but not complete Hif1a gene silencing (Figure 2B). Western blots confirmed increased HIF1A and HIF2A protein levels in hypoxic cells and their specific reduction by the corresponding shRNAs (Figure 2C). Thus, the selected shRNAs efficiently downregulated Hif1a and Hif2a at both the mRNA and protein levels.Figure 2In vitro validation of Hif1a and Hif2a shRNAs and design of the dual-acting AAV vector(A) Graphical representation of tissue changes leading to hypoxia and degenerative processes during aging and AMD. (B) Gene expression analysis of 661W cells stably expressing shCtrl, shHif1, or shHif2 after normoxic or hypoxic (24 h, 0.2% O_2_) exposure. Values were normalized to β-actin (Actb) and expressed relatively to normoxic 661W-shCtrl (set to 1). Shown are individual values (n = 3) and means ± SDs. Significance was tested using a one-way ANOVA with Tukey’s multiple comparison test. Asterisks indicate statistical significance; p values of all comparisons are given in Table S2. (C) Western blot analysis for HIF1A, HIF2A, and ACTB in 661W cells stably expressing shCtrl, shHif1, or shHif2 after normoxic and hypoxic exposure. The position of the band for HIF1A is indicated by an arrow. (D) Schematic representation of the DNA constructs packaged into AAV vectors. AAV-shHif, the construct consists of the GRK1 promotor driving photoreceptor-specific expression of shHif1a embedded in the miR-E scaffold in the 3′ UTR of GFP, followed by the VMD2 promotor controlling RPE-specific expression of shHif2a embedded in the miR-E scaffold in the 3′ UTR of mCherry; AAV shCtrl, a scrambled non targeting control shRNA sequence replaced shHif1a and shHif2a; AAV-shHif1, the GRK1 promotor drives photoreceptor-specific expression of shHif1a embedded in the miR-E scaffold in the 3′ UTR of GFP; AAV-shHif2: the VMD2 promotor controls expression of shHif2a embedded in the miR-E scaffold in the 3′ UTR of mCherry in the RPE; BM, Bruch`s membrane; CC, choriocapillaris; ONL, outer nuclear layer; PS, photoreceptor segments; RPE, retinal pigment epithelium. Sequences of shRNAs are provided in Figure S2 and Table S3.

To achieve a simultaneous and effective in vivo downregulation of Hif1a in photoreceptors and Hif2a in the RPE, we placed the corresponding shRNAs (Figure S2) into miR-E scaffolds20 and positioned them in the 3′ UTRs of GFP (shHif1) and mCherry (shHif2). G protein-coupled kinase 1 (GRK1)29 and vitelliform macular dystrophy 2 (VMD2)30 promoters were used to drive the expression of GFP::miR-E-shHif1 in photoreceptors and mCherry::miR-E-shHif2 in the RPE (Figure 2D). Using a single vector to deliver the two shRNAs to their specific cell type allowed for an efficient and precise intervention. A scrambled RNA sequence replaced shHif1 and shHif2 in the shCtrl vector. To compare the potency of this multiplex dual-acting AAV with single-acting AAVs, we generated two additional vectors to deliver either shHif1 to the photoreceptors or shHif2 to the RPE (Figure 2D).

Targeted expression in photoreceptors and RPE of wild-type mice

The AAVs carrying shHif or shCtrl were injected into the subretinal space of wild-type (WT) mice (Figure 3A) to evaluate transduction and expression profile of delivered constructs. Two weeks after administration, the injection site was visible as a hyperreflective spot in the fundus (Figures 3B and 3C, arrows). Optical coherence tomography (OCT) scans taken at the injection site showed minor layer distortions likely attributable to retinal injury caused by needle penetration. Beyond the injection site, however, tissue integrity was preserved in shCtrl- or shHif-treated areas, as shown by OCT scans through the optic nerve head, indicating that neither vector was toxic to retinal cells nor the RPE (Figures 3B and 3C). Fluorescence funduscopy showed widespread GFP (as surrogate marker for shHif1 expression) and mCherry (indicating shHif2 expression) positive areas surrounding the injection site (Figures 3B and 3C). Retinal cross-sections collected at 3 weeks post-injection (wpi) showed GFP and mCherry signals localized to photoreceptors and the RPE, respectively (Figures 3D and 3E). However, weak mCherry fluorescence was also detected in the neural retina close to the injection site. The signal was much less robust than in the RPE and decreased to undetectable levels with increasing distance from the injection site. GFP fluorescence was observed in the RPE across the transduced photoreceptor area. While this may be attributed to phagocytosis of shed GFP^+^ outer segments by RPE cells, we cannot completely rule out expression of the GRK1-driven gene in the RPE, as it has been observed for AAV8 in macaque before,31 or a functional crosstalk between transcriptional enhancers after concatemer formation.32 To evaluate the transduction perimeter, GFP and mCherry fluorescence was examined on retinal and RPE flatmounts at 3 wpi. The GFP^+^ area covered approximately 40%–50% of the total retina and closely matched the mCherry^+^ area in the corresponding RPE both in size and position (Figures 3F and 3G). Together with the reduced levels of Hif1a in photoreceptors and Hif2a in the RPE in the transduced area (Figures S3A and S3B), this demonstrates successful reduction of the target mRNAs in the intended cell types. Potential inflammation was assessed by immunofluorescence staining for allograft inflammatory factor 1 (IBA1, alias AIF1) on retina and RPE flatmounts, in retinal cross-sections at 3 wpi (Figures 3F, 3G, and S3C). Whereas reactive (ameboid) microglia and/or macrophages were detected in the retina close to the injection site, independent of the injected substance, ramified resting microglia populated the surrounding tissue. RPE flatmounts presented few IBA1^+^ cells only close to the injection site. The presence of microglia or macrophages in this restricted area might have been caused by retinal detachment and scar formation inflicted by the subretinal injections. Additionally, we have used CD68 and CD11b markers to identify blood-derived macrophage and monocyte infiltration on retinal cross-sections (Figure S3D). Apart from very few positive cells close to the injection site, no infiltrating macrophages/monocytes were observed in the treated area. Apart from the perturbations at the injection site, the transduced areas showed normal staining patterns for cone arrestin (ARR3) and β-catenin (CTNB1) (Figure S3E), and showed no signs of degeneration or functional impairment up to 22 wpi (Figures 3H, 3I, and S3F). These results demonstrate widespread and selective transduction of photoreceptors and RPE cells by the AAV delivering the multiplex RNA therapeutics with no apparent adverse effects beyond the injection site.Figure 3. Evaluation of the pattern and tolerability of shHif expression in WT mice after AAV-mediated delivery(A) Illustration of transvitreal AAV delivery to the subretinal space targeting photoreceptor cells (shown in green due to expression of GFP) and RPE cells (shown in red due to expression of mCherry). (B and C) Fundoscopy, OCT scans, and fundus fluorescence of WT mice injected with shCtrl (B) or shHif (C) at 2 wpi. Orange and red lines in the fundus images indicate positions of the respective OCT scans. Fluorescence indicates transduced areas in the retina (green, GFP) and RPE (red, mCherry). White arrows, injection site; ∗, optic nerve head. (D and E) Retinal cross-sections of WT mice injected with shCtrl (D) or shHif (E) at 3 wpi. Green/white: GFP^+^ photoreceptors/RPE. Red/white: mCherry^+^ RPE cells. Shown are merged images (left) and individual fluorescence channels (center and right). (F and G) Retina and their corresponding RPE flatmounts from WT mice injected with shCtrl (F) or shHif (G) at 3 wpi. Shown are native GFP (green) and mCherry (red) fluorescence and stainings for IBA1 (white). Boxed areas in the fluorescence images are shown in higher magnification (center). Arrow, injection site. (H) Representative retinal morphologies of WT mice injected with shCtrl (left) or shHif (right) at 22 wpi. (I) Scotopic and photopic electroretinograms (ERG) of WT mice injected with shCtrl (gray) or shHif (orange) at 22 wpi. Shown are scotopic A- and B-wave and photopic B-wave amplitudes as a function of stimulus intensity. Shaded areas indicate SD; n = 4 per group. Scale bars: as indicated. GCL, ganglion cell layer; INL: inner nuclear layer; SRS, subretinal space.

Mouse models for AMD

Our therapeutic strategy centers on targeting the alpha subunits of HIF1 and HIF2 transcription factors, which are stabilized under reduced oxygen levels. Stabilization is initiated when the activity of HIF-prolyl hydroxylases decreases due to reduced availability of O_2_ as substrate. Reduced hydroxylation of HIF1A and HIF2A prevents their recognition by the VHL protein complex and subsequent degradation by proteasomes, allowing them to accumulate and function as transcription factors. This accumulation (and activation) of HIF1 and HIF2 is mimicked in Rod^ΔVhl^ mice, which lack VHL in rods. These mice represent a pseudohypoxia model in which the hypoxic response is elicited in the presence of normal oxygen levels, leading to photoreceptor degeneration.10^,^11 The stabilization of HIF1A by the chemical prolyl hydroxylase inhibitor rodaxustat resulted in comparable cell death,33 supporting the validity of the model. However, eyes of aging individuals and those afflicted by AMD experience hypoxia affecting both photoreceptors and the RPE. Thus, we established a second model by knocking out Vegfa in RPE cells of adult mice (RPE^ΔVegfa^; Figure 4A). As reported by others, the lack of VEGFA disrupts the homeostasis in the choriocapillaris, leading to vasoconstriction in the choroid. This results in the development of true hypoxic conditions, including HIF activation, consequently causing RPE and photoreceptor degeneration.12^,^34^,^35 To generate a local phenotype, we delivered Cre recombinase to the nasal RPE of 28-day-old Vegfa^fl/fl^ mice36 by the subretinal injection of an AAV4 expressing Cre-P2A-EGFP under control of the VMD2 promoter (Table 1). Inactivation of Vegfa expression in the transduced RPE was verified by in situ hybridization, and transduction was confirmed by staining for Cre in the consecutive section at 3 wpi. Vegfa expression was evident in the non-transduced area but absent in Cre^+^ RPE cells in the transduced area (Figure 4B). This mutually exclusive expression pattern indicated an effective suppression of Vegfa in transduced RPE cells. Choroidal flatmounts stained for podocalyxin (PODXL), an endothelial cell marker, demonstrated a strong reduction in choroidal vessel density at 3 wpi (Figure 4C). To quantify the phenotype, we analyzed the staining pattern of PODXL in sections through control and transduced regions lacking Vegfa expression. Transduced areas showed notable gaps in the PODXL pattern and a significant increase in the gap size (Figure 4C), suggesting a collapse of the choriocapillaris and rearrangement of endothelial cells in the absence of RPE-derived VEGFA.Figure 4. Assessment of the RPE^ΔVegfa^ phenotype(A) RPE^ΔVegfa^ model schematics: Vegfa deletion in the RPE leads to reduced choroidal vessel density, RPE alterations, and photoreceptor degeneration. (B) Vegfa (red) in situ hybridization and Cre (magenta) immunofluorescence in areas that were or were not transduced with AAV::VMD2-Cre-P2A-EGFP. (C) PODXL immunofluorescence in endothelial cells of choroidal flatmounts (top) and cross-sections (toward bottom) of indicated areas. Contrast-enhanced threshold images allowed pixel intensity scan analysis (bottom) and quantification (right) in RPE^ΔVegfa^ mice at 3 wpi. Data from n = 3 eyes, 18–19 measurements/area, analyzed by unpaired t test. (D) Fundoscopy, OCT scans, fundus fluorescence, and angiography of RPE^ΔVegfa^ mice at 3 (top) and 39 (bottom) wpi. Fundus images with orange/red lines indicate OCT scan positions. GFP-labeled images show green fluorescence confirming the presence of Cre at 3 wpi; background fluorescence dominates at 39 wpi. Arrows, injection site; ∗, optic nerve head. (E) Representative color-coded ONL thickness maps generated from OCT volume scans at 39 wpi. Right: boxplots of ONL thickness in the two nasal subfields in mice. Shown are average values of individual eyes (dots) and means of all eyes (line). Values for control injected mice (Vegfa^fl/fl^ + PBS and Vegfa^fl/fl^ + GFP) were combined. n = 4 (Vegfa^fl/fl^); n = 11 (RPE^ΔVegfa^); n = 6 (Vegfa^fl/fl^ + GFP/Vegfa^fl/fl^ + PBS). One-way ANOVA with Tukey’s multiple comparison test. p values as shown. (F) Representative retinal histology of Vegfa^fl/fl^ mice (left) subretinally injected with AAVs expressing Cre (RPE^ΔVegfa^) or GFP (Vegfa^fl/fl^ + GFP) or with PBS (Vegfa^fl/fl^ + PBS) at 39 wpi. (G) ONL thickness and PS length quantified from histological sections at 39 wpi. Shown are spidergrams and average values in the transduced nasal area (barplots, 400–2,000 μm from the optic nerve head [ON]) of indicated groups at 39 wpi. Shown are means ± SDs and individual data points. n = 3 (Vegfa^fl/fl^; Vegfa^fl/fl^ + GFP; Vegfa^fl/fl^ + PBS); n = 2 (RPE^ΔVegfa^). One-way ANOVA with Tukey’s multiple comparison test; p values as indicated. DIC, differential interference contrast.Table 1. Treatment groups and AAVsTreatment groupMouse strainCapsidDNA constructSingle AAV injectionsWT + shCtrlB6/J-Crl1AAV8-TMGRK1-GFP-shCtrl-VMD2-mCherry-shCtrlWT + shHifB6/J-Crl1AAV8-TMGRK1-GFP-shHif1-VMD2-mCherry-shHif2Rod^ΔVhl^+ shCtrl**Vhl^fl/^^fl^;Opsin-CreAAV8-TMGRK1-GFP-shCtrl-VMD2-mCherry-shCtrlRod^ΔVhl^+ shHif**Vhl^fl/^^fl^;Opsin-CreAAV8-TMGRK1-GFP-shHif1-VMD2-mCherry-shHif2Vegfa^fl/^^fl^ + GFP**Vegfa^fl/^^fl^AAV4VMD2-EGFPRPE^ΔVegfa^Vegfa^fl/^^fl^AAV4VMD2-Cre-P2A-EGFP**Double AAV injections (combined with AAV4/2::VMD2-Cre-P2A-GFP)**RPE^ΔVegfa^+ shCtrl**Vegfa^fl/^^fl^AAV8-TMGRK1-GFP-shCtrl-VMD2-mCherry-shCtrlRPE^ΔVegfa^+ shHif**Vegfa^fl/^^fl^AAV8-TMGRK1-GFP-shHif1-VMD2-mCherry-shHif2RPE^ΔVegfa^+ shHif1**Vegfa^fl/^^fl^AAV8-TMGRK1-GFP-shHif1RPE^ΔVegfa^+ shHif2**Vegfa^fl/^^fl^AAV8-TMVMD2-mCherry-shHif2PBS and uninjectedVhl^fl/^^fl^Vhl^fl/^^fl^NANAVegfa^fl/^^fl^Vegfa^fl/^^fl^NANAVegfa^fl/^^fl^ + PBS**Vegfa^fl/^*^fl^*NANA (PBS)All AAVs have type 2 inverted terminal repeats. AAV8-TM, AAV8-triple mutation (Y447F + Y733F + T494V); NA, not applicable.

At 3 wpi, the transduced area was clearly visible as a brighter region in the fundus that correlated with the GFP^+^ area and thus expression of Cre (Figure 4D). Angiography indicated a predominantly regular retinal vasculature, and OCT showed normal retinal thickness, with the expected small perturbations around the injection site at this time (Figure 4D, upper panel). At 39 wpi, however, the same eye presented various abnormalities, including fundus discoloration, RPE hypertrophy, retinal thinning with an almost complete loss of the outer nuclear layer (ONL), and vascular leakage (Figures 4D–4F and S4). The weak GFP signal (Figure 4D) suggested either a loss of transgene expression or the death of transduced cells at this time point. To quantify the effect in vivo, we employed OCT volume scans and measured the ONL thickness in two nasal subfields. Direct comparisons showed a significantly reduced ONL thickness in RPE^ΔVegfa^ but not in non-injected (Vegfa^fl/fl^) or control-injected (Vegfa^fl/fl^ +PBS or +GFP) mice (Figure 4E). Light microscopy and quantification of equatorial nasal-temporal sections supported the in vivo findings and revealed significant thinning of the ONL and reduced length of photoreceptor segment (PS) that was restricted to the transduced nasal retina of RPE^ΔVegfa^ mice (Figures 4F and 4G). The phenotype was not caused by subretinal injections or the activity of the VMD2 promotor per se since the respective controls showed no signs of retinal degeneration (Figures 4F and 4G).

Dual-acting shHif therapy restores tissue integrity in RodΔVhl mice

Having validated the expression pattern of the shHif AAV vector in WT mice (Figure 3), we proceeded to investigate whether this vector can effectively interfere with the hypoxia pathway in our two mouse models. We first used the Rod^ΔVhl^ mouse with its chronically activated molecular response to hypoxia in rod photoreceptors due to the rod-specific deletion of Vhl (Figure 5A). The model is characterized by a HIF1-dependent, late-onset progressive loss of photoreceptors, occasional disturbances in the RPE, and a decline in visual function.10^,^11 As expected, untreated 26-week-old Rod^ΔVhl^ mice showed patches of photoreceptor cell loss in the fundus, as well as RPE and photoreceptor irregularities in OCT images (Figure 5B). Age-matched Rod^ΔVhl^ mice injected with shCtrl vector displayed a comparable phenotype at 22 wpi, with patches of degenerated areas in normal fundus that were prominently visible as dark flecks in fundus fluorescence images capturing vector-derived GFP and mCherry signals. In contrast, fundi and OCT images of shHif-treated mice displayed fewer irregularities and no obvious signs of RPE disturbances (Figure 5B). To quantify the extent of rescue in vivo, we measured ONL thickness in the different groups using volumetric OCT scans. Although not reaching the thickness of control Vhl^fl/fl^ mice (without Opsin-Cre), treatment of Rod^ΔVhl^ mice with shHif resulted in a significantly thicker ONL compared to Rod^ΔVhl^ mice treated with shCtrl (Figure 5C). Light microscopy of retinal morphology at 22 wpi supported these findings and showed that retinal areas that had received the shHif virus had less severe ONL thinning and RPE disturbances than non- or shCtrl-transduced areas (Figure 5D).Figure 5. Rescue of the Rod^ΔVhl^ phenotype by the dual-acting shHif RNA therapy(A) Schematic representation of Rod^ΔVhl^ mice and rescue of RPE (red) and photoreceptors (green) through downregulation of Hif2a in the RPE and Hif1a in photoreceptors and Hif2a in the RPE by shHif. (B) Fundoscopy, OCT scans, and fundus fluorescence assessed at 26 weeks of age (22 wpi) in Rod^ΔVhl^ mice that were either non-injected (left) or treated with shCtrl (center) or shHif (right). Orange and red lines in the fundus images indicate positions of the respective OCT scans. OCT scans of higher magnifications are included. Fundus fluorescence images (last two rows) show the transduced areas in the retina (green, GFP) and RPE (red, mCherry). ∗, optic nerve head. (C) Color-coded ONL thickness maps of Vhl^fl/fl^, Rod^ΔVhl^ + shCtrl and Rod^ΔVhl^ + shHif mice generated from OCT volume scans at 22 wpi. Grids were aligned to the optic nerve head (ON). The ONL thickness was measured in seven subfields and indicated with numbers (artificial units). Right: quantification of ONL thickness in the seven subfields (outlined by the orange line in the grid). Shown are average values of the subfields in individual eyes (dots) and the means of all eyes (line) in box blots (n ≥ 9). One-way ANOVA with Tukey’s multiple comparison test. p values are as shown. (D) Representative retinal morphologies of transduced regions of Vhl^fl/fl^ control and Rod^ΔVhl^ mice injected with shCtrl or shHif. All mice were 26 weeks of age (22 wpi) at the time of analysis. (E and F) Relative gene expression levels in retinas (E) or eyecups (F) of Rod^ΔVhl^ + shCtrl (gray) and Rod^ΔVhl^ + shHif (orange) mice at 22 wpi. Expression was normalized to β-actin (Actb) and shown relative to expression in Rod^ΔVhl^ + shCtrl mice (set to 1). Shown are means ± SDs of n = 11–12. Unpaired t test. p-values as indicated.

Gene expression analysis at 22 wpi showed reduced Hif1a but not Hif2a mRNA levels in the retina (Figure 5E) and reduced Hif2a but not Hif1a mRNA expression in the eyecup (EC) tissue that included the RPE (Figure 5F). This suggests a high degree of cell-type specificity of our dual-acting shHif gene silencing approach. The rather moderate reduction of Hif1a and Hif2a expression levels by 20%–30% was not surprising given the cell content of the samples, which included cells from both transduced and non-transduced areas, as well as cells that were not targeted by our vector (e.g., 2^nd^- and 3^rd^-order neurons). Thus, the observed RNA reduction indicated a robust knockdown efficiency, considering that, on average, only 30% of all rods and RPE cells were targeted. This may also be the reason why levels of Casp1 (marker for degenerative processes) and Pde6b (marker for rod integrity) in the retina and of Rpe65 (marker for RPE integrity) in the EC were only slightly reduced or increased, respectively (Figure 5E). Together with the significant rescue of retinal thickness, our gene expression data collectively support a considerable potential of the shHif approach for alleviating the harmful effects of a chronic hypoxic response in photoreceptors.

Dual-acting shHif therapy restores tissue integrity in RPEΔVegfa mice

We assessed the potential of the shHif approach to mitigate the degenerative effects of true chronic hypoxia in the RPE^ΔVegfa^ model. Since this model also includes the pathogenic activation of HIF2A in the RPE,12 we compared the efficacy of the dual-acting shHif AAV with single-acting shHif1 (targeting photoreceptors) and shHif2a (targeting the RPE) AAVs. Subretinal co-injections of AAV::VMD2-Cre-P2A-EGFP to inactivate Vegfa in the RPE, along with either shCtrl, shHif, shHif1, or shHif2 were performed at 4 weeks of age followed by analysis at 39 wpi. Fundoscopy and OCT of RPE^ΔVegfa^ mice injected with shCtrl showed discolored patches and affected tissue layers, indicating RPE disturbances and degeneration at 39 wpi (Figure 6A), a phenotype similar to untreated RPE^ΔVegfa^ mice (Figure 4D). In contrast, shHif-treated animals displayed only minor disturbances around the injection site, but no obvious retinal degeneration (Figure 6A). The rescue effect by the treatment with the various viruses was quantified in vivo using OCT volume scans. Treatment with the dual-acting shHif virus preserved ONL thickness in RPE^ΔVegfa^ mice significantly better than treatment with the single-acting or control AAVs. (Figure 6B). The superior rescue effect by shHif was further verified by light microscopy of retinal sections (Figure 6C) and measurements of ONL thickness and PS length (Figure 6D) in a subset of mice. Eyes treated with shCtrl displayed a drastic thinning of the ONL, loss of PS, and variable thickness of the RPE in transduced areas. In contrast, mice treated with shHif had preserved PS with few disturbances, a thicker ONL, and only sporadic regions with irregular RPE morphology. shHif1 and shHif2 viruses rescued the phenotype at intermediate levels (Figures 6C and 6D). Areas that were not transduced by the Cre-expressing virus and thus with regular Vegfa expression in the RPE showed normal retinal morphologies for all injected eyes. The superior rescue effect by the dual-acting shHif virus underscores the importance of simultaneously targeting Hif1 in photoreceptors and Hif2 in the RPE to efficiently mitigate the consequences of chronic hypoxia. The dual-targeting approach was not only superior to targeting individual HIF transcription factors but also long-lasting and protected tissue integrity at least up to 61 wpi (Figure 7). While RPE^ΔVegfa^ mice treated with shCtrl exhibited significant loss of GFP and mCherry fluorescence at 61 wpi when compared to 4 wpi, shHif-treated mice still showed strong fluorescence at this time point, suggesting preservation of photoreceptors and RPE (Figures 7A and 7B). This was confirmed by light microscopy of retinal sections and quantification of ONL thickness and PS length, which showed strong preservation of retinal morphology in shHif- but not shCtrl-treated mice (Figures 7C and 7D). The eyes injected with the GFP control virus provided a baseline for non-degenerated retinas, allowing us to assess the overall efficacy of the dual-acting shHif virus. Our results demonstrate the efficacy of our multiplex shHif therapy in preserving retinal structures for more than 1 year despite the continued presence of the toxic hypoxic stimulus. The strong protection in two different models of hypoxia-induced retinal degeneration indicates the potential of the approach for a future translation into a clinical setting.Figure 6. Dual-acting multiplex shRNA therapy is more effective than single-acting shRNA vectors in rescuing pathology in RPE^ΔVegfa^ mice(A) Fundoscopy, OCT scans, and fundus fluorescence of RPE^ΔVegfa^ mice injected with shCtrl (left) or shHif (right) at 39 wpi. Orange and red lines in the fundus images indicate positions of the respective OCT scans that are shown in lower and higher magnifications. Fundus fluorescence indicates transduced areas in the retina (green, GFP) and RPE (red, mCherry). ∗, optic nerve head. (B) Color-coded ONL thickness maps generated from OCT volume scans of RPE^ΔVegfa^ + shCtrl, RPE^ΔVegfa^ + shHif, RPE^ΔVegfa^ + shHif1, and RPE^ΔVegfa^ + shHif2 mice at 39 wpi. Grids were aligned to the optic nerve head (ON). The ONL thickness was measured in the two nasal subfields (outlined by the orange line in the grid) that were most affected in RPE^ΔVegfa^ mice and indicated with numbers (artificial units). Right: quantification of ONL thickness in the nasal two subfields. Shown are average values in individual eyes (dots) and means of all eyes (line) in box blots. n = 12 (shCtrl) n = 12 (shHif); n = 9 (shHif1); n = 8 (shHif2). One-way ANOVA with Dunnett’s multiple comparison test comparing all groups to shHif. p values as shown. (C) Representative retinal morphologies of transduced and non-transduced areas in RPE^ΔVegfa^ mice treated with AAVs as indicated. Analysis was at 39 wpi. (D) Quantification of ONL thickness and PS length of the four treatment groups. Shown are spidergrams and average values of the transduced nasal area (400–2,000 μm from the optic nerve head [ON]) of a subset of mice that were already analyzed by volume scans (B) at 39 wpi. Gray, shCtrl; orange, shHif; green, shHif1; blue, shHif2. Shown are means ± SDs and individual data points. One-way ANOVA with Dunnett’s multiple comparison test comparing all groups to shHif. p values as shown. shHif: n = 4 (ONL) and n = 3 (PS); shHif1: n = 9 (ONL) and n = 4 (PS); shHif2: n = 8 (ONL) and n = 6 (PS); shCtrl: n = 2 (ONL and PS).Figure 7. Long-lasting rescue of the RPE^ΔVegfa^ phenotype by the dual-acting shRNA therapy(A and B) Fundoscopy and fundus fluorescence of RPE^ΔVegfa^ mice injected with shCtrl (top) or shHif (bottom) analyzed at 4 wpi (A) and 61 wpi (B). Fluorescence indicates transduced areas in the retina (green, GFP) and RPE (red, mCherry). ∗, optic nerve head. (C) Representative retinal morphologies of transduced and non-transduced areas of RPE^ΔVegfa^ mice injected with shCtrl or shHif, analyzed at 61 wpi. (D) Quantification of ONL thickness and PS length at 61 wpi, presented as spidergrams (left) and boxplots (right) presenting the averaged values of the transduced nasal area (400–2,000 μm from the optic nerve head [ON]). Treatment groups: Vegfa^flox/flox^ mice injected with GFP control virus (green), RPE^ΔVegfa^ + shHif mice (orange), and RPE^ΔVegfa^ + shCtrl mice (gray) at 61 wpi. ONL thickness and PS length in Vegfa^flox/flox^ + GFP mice are not shown in spidergrams to reduce complexity of the figure images. Shown are means ± SDs and individual data points. One-way ANOVA with Tukey’s multiple comparison tests. p values as shown. shHif, n = 11 (ONL) and n = 9 (PS); shCtrl, n = 10 (ONL) and n = 6 (PS); GFP, n = 3 (ONL and PS).

Discussion

Age is the major non-genetic risk factor for the development of AMD. The aging ocular tissue undergoes several changes that result in reduced oxygen levels in the outer retina and increased expression of HIF target genes.2^,^3^,^10 Some HIF target genes, such as VEGFA and ANGPTL4, are angiogenic factors that ultimately contribute to the development of neovascular AMD.22^,^25 Others, such as SERPINE1, may facilitate disease development by remodeling the extracellular matrix24 or through their connection to senescence and aging.37 Expression of all three hypoxia-related genes was increased in old donor eyes. In addition, a hypoxic environment triggers an intracellular response that adjusts major metabolic pathways, including glycolysis, oxidative phosphorylation, lipid metabolism, and others to the new environmental condition. Such a metabolic shift likely contributes to cell death in the delicate photoreceptor neurons and the RPE, highlighting the complex interplay between aging, hypoxia, and cellular responses involved in AMD development.10^,^12^,^38^,^39^,^40^,^41 Thus, targeting the hypoxia pathway could serve as a mutation-independent therapeutic approach that may be beneficial for a wide range of patients affected by wet or dry AMD. Here, we established an efficient RNAi strategy that simultaneously targets HIF1 in the photoreceptors and HIF2 in the RPE. This multiplex approach prevented retinal degeneration associated with the chronic hypoxic response in both mouse models of pseudo- and true hypoxia. Another key finding was the demonstration that this dual-acting therapy proved to be more effective than single-target treatments. This finding highlights the potential and superiority of multi-target therapies for complex multifactorial diseases such as AMD.

The use of shRNAs driven by RNA Pol II promoters to downregulate target genes has several advantages. The system allows targeting those cell types that require treatment without affecting other cells. In contrast to the transient effects of short-lived synthetic siRNAs, it also ensures long-lasting effects through continuous expression of the therapeutic RNAs for at least 61 week as illustrated by the sustained presence of the surrogate GFP and mCherry fluorescent proteins (Figure 7). It also minimizes the possibility that a bolus injection of siRNAs overwhelms the endogenous RNAi pathway, leading to toxic side effects.19^,^42 Furthermore, this system is based on a single AAV vector and does not rely on double transduction of individual cells, trans-splicing, or other mechanisms that may reduce efficiency. The therapeutic approach is not only well tolerated but also potent and has protected tissue integrity in two different models of hypoxia-related retinal degeneration that share key aspects of AMD.

While long-term, global inhibition of HIFs in healthy cells could have problematic effects, including impaired defense against oxidative stress, metabolic dysregulation, altered immune adaptation, and others, their impact depends on the tissue and cellular context.43 Studies in mice have shown that knocking down HIFs from RPE12^,^35^,^44 or photoreceptors45^,^46^,^47 has no impact on normal retinal physiology and shows no increased sensitivity to acute hypoxic insults. Collectively, these findings suggest that a therapy aiming at dampening chronic HIF-driven pathologic processes in photoreceptors and RPE, should not cause harm for healthy cells. Furthermore, our RNAi therapy allows for residual HIF activity, which is likely to preserve critical cellular functions and responses.

In contrast to our data for the chronic models presented here, the recent work of Babapoor-Farrokhran and colleagues suggests a beneficial function of HIF1 for the survival of WT photoreceptors.48 However, this conclusion was drawn after the induction of acute oxidative stress by NaIO_3_, which is a short-term stress model without the sustained reduction in oxygen supply that is characteristic of the chronic conditions that develop during the aging process as a key driver in AMD pathophysiology. While valuable for investigating specific aspects of oxidative stress-induced cell death, acute models like NaIO_3_ toxicity may therefore less accurately model the effects of a long-lasting chronic hypoxic response or environment. In addition, the chemicals used in their study can be toxic to photoreceptors per se (digoxin49) or exhibit side effects apart from HIF activation (dimethyloxalylglycine50^,^51). Nevertheless, data from their acute model is interesting and may best be correlated to the protective effects of hypoxic preconditioning against acute light-induced retinal degeneration.52 Since this indicates that acute and chronic hypoxic responses have distinct consequences for retinal cells, we believe that our two genetic models—particularly the RPE^ΔVegfa^ model, which faithfully provides a true and chronic hypoxic environment due to choroidal atrophy—represent the situation in the aged eye more accurately than drug-induced acute models. Therefore, while acknowledging the value of acute models for specific mechanistic questions, findings from such models should be interpreted in light of the distinct chronic hypoxic nature of AMD, which our models aim to recapitulate.

While the Rod^ΔVhl^ model has chronically active HIFs in normoxic rods causing HIF1-dependent photoreceptor degeneration,10 the RPE-specific deletion of Vegfa in the RPE^ΔVegfa^ model affects the choroidal vasculature, leading to true tissue hypoxia and leading to HIF2-dependent RPE damage and to retinal degeneration.12 Since the true tissue hypoxia in RPE^ΔVegfa^ mice affects several cell types and elicits broad effects, including changes in metabolism,12 RPE^ΔVegfa^ mice may better represent the complex, multifactorial environment in the aging eye. To knock down Vegfa in adult RPE cells, we delivered a VMD2-controlled Cre transgene to Vegfa^fl/fl^ mice. This approach is more specific, efficient, and reproducible compared to methods suffering from variable and patchy Cre expression in RPE of transgenic mice35^,^53 or the lack of cell-type specificity after delivery of CjCas9 for insertion or deletion generation.34

The therapeutic vector was designed to express shHif1 in photoreceptors and shHif2 in the RPE, with GFP and mCherry as surrogate markers. Despite that the expression of GFP::shHif1 was controlled by the GRK1 promotor, GFP fluorescence was also visible in the RPE (Figures 3D and 3E). This was likely due to phagocytosis of photoreceptor material by the RPE or to some expression of the gene in the RPE following delivery by AAV8, which has a known tropism for the RPE, as observed in macaque studies.31 Alternatively, a transcriptional crosstalk between the enhancers and promoters of the two gene cassettes may be at the basis of this observation. The phenomenon of intermolecular activation of gene expression delivered by AAVs was been very recently investigated in detail by Coughlin and colleagues in cells transduced with two AAVs.32 They concluded that concatemer formation of the viral genomes generates sufficient proximity between genes to allow functional crosstalk between a transcriptional enhancer from one AAV genome and a gene from the other, leading to its expression. A similar crosstalk may have occurred in our dual approach, in which we delivered two genes with their set of enhancers to the same host cell. This opens the possibility that some molecules targeting Hif1a mRNA reached the RPE. Although phagocytosed material would likely be confined to lysosomes, we cannot exclude some downregulation of Hif1a transcripts in the RPE by such “spills.” However, since the knockout of Hif1a in the mouse RPE does not cause phenotypic changes,35 this possibility does not significantly influence the validity, specificity, and efficacy of our approach.

Although our dual-active RNA therapy rescued the phenotype in both mouse models, our results suggest a higher efficiency in RPE^ΔVegfa^ mice, where degeneration is based on true hypoxia due to pathologic changes in the choroidal vasculature (Figure 4C). The RNA-induced silencing complex that is formed after shRNA expression reduces stability or translation of mRNA transcripts but does not affect preformed proteins. Thus, the efficacy of shRNA-mediated downregulation of protein activity depends on the expression levels of the target RNA. Considering reports suggesting that hypoxia can reduce levels and half-lives of Hif1a and Hif2a transcripts,54^,^55 it appears possible that RNAi is more effective in the model of true hypoxia due to lower levels of target mRNAs. Clearly, unraveling the mechanisms that contribute to the differences in treatment efficacies in the two models needs further research and is important to further improve the RNAi approach. A limitation of our study lies in structural differences between the mouse and the human retina. Unlike humans, mice do not possess a macula, which prevents our models from presenting all aspects of AMD. However, since the molecular response to hypoxia with the activation of HIF transcription factors is quite universal, we nonetheless believe that our RNA therapy targeting this conserved mechanism is translatable in the future to the human situation. Future studies involving larger animals, including non-human primates, are necessary to validate our approach and evaluate its translational potential.

Since reduced oxygen delivery to the outer retina through the choroidal system affects both RPE and photoreceptors, it is important that a therapeutic intervention targets the degeneration-causing factors in both cell types simultaneously and specifically. We approached this requirement through the dual-active RNA therapy system using a single AAV. Delivery of therapeutic genes by AAV is safe and has been evaluated in several clinical trials.56^,^57 The first US Food and Drug Administration- and European Medicines Agency-approved gene therapy (Luxturna) for Leber congenital amaurosis type 2 supports not only the safety but also the efficacy of the delivery of genetic material to target cells by subretinal AAV injections. The approach presented here further expands the potential of AAVs as a therapeutic system by using a single AAV vector with two specific RNA Pol II promotors to drive expression of shRNAs targeting two transcripts in two distinct cell types. We believe that the system demonstrates potential for a widespread applicability to other multifactorial diseases that require alteration of gene expression across various cell types or tissues. Since RNAi-based therapeutics, including siRNA and shRNA, have already been used in several clinical trials (e.g., to target cancer or genetic disorders),58^,^59 the system developed here may be translated for use in patients, particularly for the treatment of retinal disorders associated with a chronic hypoxic response such as AMD.

Materials and methods

Animals

Animal experimentation was approved by the Cantonal Veterinary Office of Zürich, Switzerland (license nos. ZH019/2019 and ZH105/2022) and adhered to the Association for Research in Vision and Ophthalmology (ARVO) statement for the use of Animals in Ophthalmic and Vision Research. WT (B6/J-Crl1) and transgenic mice were maintained as breeding colonies at the Laboratory Animal Services Center (LASC) of the University of Zürich with a 14/10-h light-dark cycle with an average light intensity of 60–150 lux at cage level. Mice had access to food and water ad libitum. Vhl^fl/fl^ mice60 were mated to Vhl^fl/fl^ mice heterozygous for OpsinCre (LMOPC161) to generate Vhl^fl/fl^;OpsinCre (Rod^ΔVhl^) mice and Vhl^fl/fl^ control littermates as described.11 Vegfa^fl/fl^ animals were kindly provided by Christian Stockmann (Institute of Anatomy, University of Zürich).36 Individual eyes of mice of both sexes were randomly assigned to treatment or control.

Human tissue samples

Human donor eyes without cornea and lens in ice-cold PBS were received from the eye bank of the University Hospital Zürich. The use of the tissue for research was approved by the ethics committee of Zürich, Switzerland (Business Administration System for Ethics Committees [BASEC] nos. PB_2017-00550 and 2020-01856) and adhered to the tenets of the Declaration of Helsinki. The summarized donor information is given in Table S1. The vitreous was carefully removed and four incisions were used to flatmount the tissue. Retinal and RPE samples were collected from the donor eyes with no clinical diagnosis of retinal disease. RNA was isolated from the samples using the RNeasy kit (Qiagen, Hilden, Germany). cDNA synthesis and real-time PCR were performed using human-specific primer pairs (Table S4), following the same methods as described for the mouse samples (see below).

Cloning and generation of AAVs

shRNAmir DNA fragments containing shHif1a, shHif2a, or shCtrl sequences were designed using the miR-E backbone following the recommendation by Fellmann et al.20 (Figures S2A–S2C) and produced by Bio-Synthesis (Bio-Synthesis, Lewisville, TX). The shHif1a and shHif2a RNAmiR fragments were cloned into the 3′ UTRs of GFP and mCherry, respectively. To achieve cell specificity, the GRK1 promoter was used for the expression of GFP (and thus shHif1a) in photoreceptors, and the VMD2 (alias bestrophin 1) promoter was used for the expression of mCherry (and thus shHif2a) in the RPE. (This dual-acting viral construct will be called shHif from this point on [Figure 2D].) The shCtrl construct had a scrambled shRNA sequence cloned into the miR-E backbone in the 3′ UTRs of both fluorescent genes (Figure 2D). The complete plasmid sequences of plasmids “shHif” and “shCtrl” are shown in Table S3. Plasmids for the single-acting AAVs contained either the GRK1-GFP-shHif1a or the VMD2-mCherry-shHif2a cassette (Figure 2D). shHif, shHif1, shHif2, and shCtrl viruses were packed into an AAV8 with the three mutations Y447F, Y733F, and T494V (AAV8-TM).62 AAV4::VMD2-Cre-P2A-GFP63 was used to knock out Vegfa in the RPE of Vegfa^fl/fl^ mice, and AAV4::VMD2-GFP served as the control virus. Specifics of all AAVs are provided in Table 1. All AAVs had type 2 inverted terminal repeats and were produced by the Viral Vector Facility (VVF) of the University of Zürich.

Subretinal injections

Injections were performed between post-natal days (PNDs) 27 and 30 as described by Gasparini et al.64 Briefly, pupils were dilated with 1% cyclogyl (Alcon Pharmaceuticals, Fribourg, Switzerland) and one drop of 5% neosynephrin (Ursapharm Schweiz GmbH, Hünenberg, Switzerland), and corneas were kept moist with viscotears (Bausch + Lomb Swiss AG, Zug, Switzerland). Mice were anesthetized with a subcutaneous injection of ketamine (85 mg/kg, Parke-Davis, Berlin, Germany)/xylazine (10 mg/kg, Bayer AG, Leverkusen, Germany) and placed on a heating pad set to 37°C during the procedure. The head was held by a stereotactic adapter (Hugo Sachs Elektronik–Harvard Apparatus GmbH, March-Hugstetten, Germany) and the temporal sclera punctured with a 30G needle just beneath the ora serrata. A 5-μL Hamilton syringe (Hamilton Bonaduz AG, Bonaduz, Switzerland), placed in a micromanipulator (H. Saur Laborbedarf, Reutligen, Germany), was inserted gently through the pre-punctured site for transvitreal subretinal injections of 1 μL AAV solution approximately 15°–25° nasally of the optic nerve. Three standard doses of AAV (1 × 10^8^, 5 × 10^8^, and 1 × 10^9^ vector genomes [vg]/μL) were tested in a preliminary experiment, resulting in a similar transduction efficiency of 5 × 10^8^ and 1 × 10^9^ vg/μL and a reduced transduction efficiency at a concentration of 1 × 10^8^ vg/μL. Thus, all AAVs were applied at a concentration of 5 × 10^8^ vg/μL, except for AAV4::VMD-Cre-P2A-EGFP (1 × 10^9^ vg/μL) when used to generate RPE^ΔVegfa^ mice for characterization (Figure 4). To visualize and control injections, the injection solution was supplemented with 10% fluorescein (0.1 mg/mL, Akorn, Decatur, IL). After injections, anesthesia was reversed with atipamezole (2 mg/kg, Graeub, Bern, Switzerland), and mice were placed on a heating pad until fully awake.

Fundoscopy, OCT, angiography, and volume scans

Anesthesia and pupil dilation were performed as described above. One drop of 2% Methocel (OmniVision AG, Neuhausen, Switzerland) was applied to keep the eyes moist. Fundus images and OCT scans were acquired using the Micron IV system (Phoenix Research Labs, Pleasanton, CA) as described.65 For fluorescein angiography, mice were injected intraperitoneally with 30 μL 2% fluorescein (Akorn Inc.), and retinal vasculature was imaged with the Micron IV system. For in vivo quantification of retinal thickness, volume scans were performed using a Spectralis OCT-2 device (Heidelberg Engineering, Heidelberg, Germany) with a 25-diopter lens. Animals were placed on a custom-built holder and kept warm with an integrated water heating system. An atraumatic focal contact lens was placed on the cornea moistened with a drop of viscotears (Bausch + Lomb) to improve imaging quality. The optic scan area was aligned to the optic nerve head, and 25 consecutive high-resolution B-scans, each consisting of 1,024 A-scans, were acquired. Automatic real-time tracking was set to 50. The built-in semi-automated segmentation tool (Spectralis software version 6.7.13.0, Eye Explorer software version 1.9.14.0) was utilized, and manual adjustments were done by blinded, experienced users (L.J.A.E., C.I., and M.S.) to measure ONL thickness in vivo. To delineate fundus areas, an Early Treatment Diabetic Retinopathy Study (ETDRS) circular grid with 9 subfields (1, 2.22, 3.45) was aligned to the optic nerve head (ON). ONL thickness was quantified in two (RPE^ΔVegfa^ mice) or seven subfields (Rod^ΔVhl^ mice). These subregions were marked with orange outlines in the respective images. For quantification, the data were first averaged per subfield using ETDRS grids (left), and then the average across all subfields was calculated and plotted for each eye (graphs, right).

Scotopic and photopic electroretinogram

Mice were dark adapted overnight. Pupils were dilated and the animals anesthetized as described above. Electroretinograms (ERGs) were recorded simultaneously from both eyes using the Diagnosys (Cambridge, UK) Celeris rodent ERG device. Stimulus intensities ranging from 8 × 10^−6^ to 3 cd∗s/m^2^ were used for scotopic measurements. After a 5-min exposure to rod-suppressive background light (30 cd/m^2^), photopic recordings were acquired using single-flash intensities ranging from 1 to 200 cd∗s/m^2^ with the background light on. For each stimulus intensity, 5 traces were averaged under scotopic and 10 traces under photopic conditions.

RNA isolation and real-time PCR

Mice were euthanized with CO_2_ inhalation followed by decapitation. Retinas were isolated through a slit in the cornea and snap-frozen in liquid nitrogen. In some instances, the remaining EC containing the RPE was cleaned from excessive tissue including the cornea and snap-frozen. RNA from tissues or cells was isolated using an RNA isolation kit (NucleoSpin RNA, Macherey-Nagel GmbH, Düren, Germany) according to the manufacturer’s instructions, including a DNase digestion step on the column. First-strand cDNA synthesis was performed by Moloney Murine Leukemia Virus reverse transcriptase (Promega, Dübendorf, Switzerland) using 1 μg total RNA and oligo-dT primers. Gene expression was analyzed via semi-quantitative real-time PCR (QuantStudio 3, Thermo Fisher Scientific, Waltham, MA) using 10 ng cDNA template and PowerUp SYBR Green Master Mix (Thermo Fisher Scientific). Primers pairs (Table S4) were designed to span large intronic regions and avoid known single-nucleotide polymorphisms in the mouse sequence. Normalization was performed with β-actin (Actb) as housekeeping gene, and relative expression was calculated using the comparative threshold cycle method (2^−ΔΔCT^).66

Morphology and immunofluorescence

Morphology and immunofluorescence analyses were performed as described.67 For morphology, dorsally marked eyes were enucleated from non-perfused (Rod^ΔVhl^) or perfused (RPE^ΔVegfa^; 4% paraformaldehyde [PFA]) mice and subsequently fixed in 2.5% glutaraldehyde at 4°C overnight. Eyes were postfixed in 1% osmium tetroxide and embedded in Epon812. Nasal-temporal sections (0.5 μm) were cut through the optic nerve head, counterstained with toluidine blue, and analyzed by light microscopy (AxioImager Z2, Carl Zeiss AG, Feldbach, Switzerland). ONL thickness and PS length were measured every 200 μm (from the optic nerve head) along reconstructed panorama images using Adobe Photoshop CS6 ruler tool (Adobe Systems, San Jose, CA).

Eyes for immunofluorescence analysis were labeled dorsally and fixed in 4% PFA for 1 h. After removal of the lens and post-fixation for an additional 1 h, eyes were cryoprotected in 30% sucrose (Sigma-Aldrich, St. Louis, MO) overnight, embedded, and frozen in freezing medium (O.C.T., Leica Biosystems, Nussloch, Germany). Eyes were stored at −80°C until sectioned into 12-μm-thin nasal-temporal sections. Immunolabeling was performed overnight at 4°C with primary antibodies (Table S5) in blocking solution containing 3% normal goat serum (Sigma-Aldrich) and 0.3% Triton X-100 (Sigma-Aldrich) in phosphate buffer. The next day, sections were washed and incubated with the appropriate secondary antibodies in blocking solution for 1 h. Counterstaining was done with 4′,6-diamidino-2-phenylindole before sections were mounted with MOWIOL (Sigma-Aldrich) antifade medium and imaged using a fluorescence microscope (AxioImager Z2, Carl Zeiss AG). Pixel intensity measurements were performed using ImageJ (version 1.52a) by analyzing 250-μm-long regions within transduced and non-transduced areas. The automatic threshold function was used to identify signal and background pixels. Pixels with an intensity of zero were counted as background/gaps. The cumulative length of these zero-intensity regions was calculated and presented as gap size per 250 μm.

In situ hybridization

In situ hybridization was performed with the RNAscope 2.5 HD RED Detection Kit (catalog no. 322350, BioTechne/ACD, Minneapolis, MN). Cryosections were prepared as described above, baked at 60°C for 30 min, and post-fixed in 4% PFA for 15 min at 4°C. Slides were dehydrated for 5 min each in 50%, 70%, and 100% ethanol and dried for 10 min at 60°C. Antigen retrieval was performed using hydrogen peroxide for 10 min, washed, and followed by overnight incubation with target retrieval buffer (BioTechne/ACD). Protease Plus treatment was performed for 15 min at 40°C, followed by 2 × 30-s washings with H_2_O and incubation with probes (Mm-Hif1a, catalog no. 313821; Mm-Epas1, catalog no. 314371; Mm-Vegfa-O3, catalog no. 521471; BioTechne/ACD) for 2 h at 40°C. Hybridized tissue sections were washed 2 × 5 min with wash buffer and incubated at 40°C with each amplifier (1–6) for 15–30 min, with two 5-min washes conducted between each amplification step. The positive control probe Mm-Polr2a (catalog no. 312471, BioTechne/ACD) and the negative control probe diluent (catalog no. 300041, BioTechne/ACD) were used to validate experimental conditions. Slides were stained in 50% hematoxylin for 20 s, destained in H_2_O, incubated in 0.02% ammonia water for 10 s, and washed again. Finally, slides were dried for 30 min at 60°C and mounted with Ecomount (catalog no. 320409, BioTechne/ACD).

Retina and RPE flatmounts

After eyes were enucleated and incubated in 4% PFA for 10 min, corneas were opened and eyes incubated for another 20 min in PFA. The eyes were cut along the ora serrata to remove the cornea and placed on a nylon membrane (Boehringer Ingelheim, Ingelheim am Rhein, Germany). We made 5–6 radial incisions to flatten the EC and retina, and the lens was removed. The retina and EC were then separated and held flat between two membranes during an additional 1 h of fixation in 4% PFA. Flattened retinas or ECs were removed from the membrane, washed with PBS, and blocked with 3% normal goat serum (Sigma-Aldrich) and 0.3% Triton X-100 (Sigma-Aldrich) in PBS for 1–2 h. Incubation with primary antibodies (Table S5) occurred overnight at 4°C. Flatmounts were washed for 3 × 10 min with PBS and incubated with the appropriate secondary antibodies for 1 h at room temperature. Subsequently, retinas and ECs were washed for 3 × 10 min with PBS, mounted, and imaged using a fluorescence microscope AxioImager Z2 (Carl Zeiss AG) or Leica DMI6000B (Leica Microsystems, Wetzlar, Germany).

Choroidal flatmounts

After an initial 30-min fixation in 4% PFA, eyes for choroidal flatmounts were washed in PBS and incubated at 55°C in pre-warmed 10% H_2_O_2_ for 2–2.5 h to bleach RPE and choroidal pigments. The cornea was removed by cutting along the ora serrata, and the eyes were placed on a nylon membrane. Lens and retina were removed, and four radial incisions were made to flatten the choroidal EC. The flattened tissue, secured between two nylon membranes, underwent an additional fixation step in 4% PFA for 1 h. Staining for PODXL proceeded as above, with the modifications that incubation with the primary antibody lasted for 2 days, and 3% horse serum (Sigma-Aldrich) was used for blocking.

Lentiviral-mediated gene silencing in 661W cells and hypoxic exposure

The 661W68 and HEK293T (American Type Culture Collection CRL-3216) cells were cultured in DMEM, 10% heat-inactivated fetal bovine serum (FBS) (Gibco, Thermo Fisher Scientific), and 1% penicillin-streptomycin (Gibco, Thermo Fisher Scientific) at 37°C and 5% CO_2_. Lentiviral-pseudotyped particles were used to generate stable cell lines expressing shRNA against Hif1a (TRCN0000232222) or Hif2a (TRCN0000082303, TRCN0000082307, TRCN0000436203, TRCN0000416474, TRCN0000428636, and TRCN0000082306) or a non-targeted scrambled shRNA (shCtrl, SHC002) (Sigma-Aldrich). To prepare lentiviral particles, HEK293T cells were seeded in 75 cm^2^ culture flasks and co-transfected with shRNAs in combination with the ViraPower lentiviral expression vector system and Lipofectamine 3000 (Invitrogen, Thermo Fisher Scientific) in serum-free medium. Medium was replaced the next day with medium containing FBS. The culture supernatants containing the lentivirus particles were collected 48–72 h post-transfection, cleared by centrifugation at 200 × g for 5 min, and filtered through a 0.45-μm pore size filter (Merck & Cie, Schaffhausen, Switzerland). Filtrates were applied in a 1:1 mixture with medium containing 6 μg/mL polybrene (Sigma-Aldrich) to infect 661W cells. Infected cells were selected with 2 μg/mL puromycin for 2 weeks. Stable cell lines expressing shHif1a, shHif2a, and shCtrl were seeded at a density of 30,000 cells/mL in a 6-well plate and incubated overnight. Prior to hypoxic exposure, medium was exchanged, and cells were exposed to normoxia (21% O_2_) or hypoxia (0.2% O_2_ for 24 h). Subsequently, cells were harvested and prepared for western blotting (see below) or real-time PCR (see above).

Western blotting

Cells were washed with PBS and lysed with 50 μL lysis buffer (1 M Tris, 150 mM NaCl, 1 mM Na_3_VO_4_, 1% Triton X-100 containing protease inhibitor [1 tablet per 10 mL solution; cOmplete ULTRA Tablets-Mini, F. Hoffmann-La Roche AG, Basel, Switzerland]). Cells were scraped off the plate, homogenized by sonication for 10 min, and centrifuged (3 min at 1,000 RCF) at room temperature. The protein concentration was determined in the supernatant with a bicinchoninic acid protein assay kit (Pierce, Thermo Scientific) according to the manufacturer’s instructions. DNA present in the homogenate was digested at 37°C for 1 h with 125 U benzonase (Sigma-Aldrich) before 70 μg proteins were used for standard western blotting. Primary antibodies (Table S5) were diluted in 5% non-fat blocking milk (BioRad, Cressier, Switzerland) in TBS-T and incubated overnight at 4°C with mild agitation. Horseradish peroxidase-conjugated secondary antibodies were used for signal detection by the Western lightning chemiluminescence reagent (PerkinElmer, Waltham, MA) and exposure to X-ray films.

Statistical analysis

Statistical analysis was performed using GraphPad Prism (GraphPad Software, San Diego, CA). Significance was tested by an unpaired t test when two independent groups were compared. When more than two groups were compared, a one-way ANOVA followed by a Tukey’s (when all groups were compared to each other) or Dunnett’s (when all groups were compared to a specific group) multiple comparison test were applied. Data were considered to be statistically different, with p ≤ 0.05. p values are given directly in figures or in Table S2. All tests and number of animals/samples are indicated in the figure legends. A full record of the animals used and exclusion criteria for each experiment can be found in Table S6.

Data and code availability

All data generated or analyzed during this study are included in this article and its supplemental information files.

Acknowledgments

The VVF of the Neuroscience Center Zurich produced all AAVs, and the LASC of the University of Zurich was responsible for animal care. The analysis of human ocular tissue was approved by the Cantonal Ethic Commission (license numbers BASEC: PB_2017-00550 and 2020-01856) and adhered to the tenets of the Declaration of Helsinki. Animal experimentation was approved by the Cantonal Veterinary Office of Zürich, Switzerland (license nos. ZH019/2019 and ZH105/2022) and adhered to the ARVO statement for the use of Animals in Ophthalmic and Vision Research. This work was supported by Swiss National Science Foundation grants 31003A_173008 and 310030_200798, and a research grant of the University of Zürich to L.J.A.E. (FK-20-028).

Author contributions

L.J.A.E., M.S., and C.G. conceived the study. L.J.A.E. designed and performed the experiments and the analysis. I.M. and F.B. collected human ocular tissue. M.S., C.I., D.K., F.P., E.K., A.F., D.S.F., L.M., G.M.W., I.T., L.P.G., S.L.B., and S.E.B. contributed to data collection and performed the experiments. All authors contributed to project design, organization of the experiments, and interpretation of results. L.J.A.E., M.S., and C.G wrote the manuscript. All authors read and approved the final manuscript.

Declaration of interests

The authors declare no competing interests.