Coviretinopathy: COVID-19-induced VEGF-dependent retinopathy

Significance

Effective treatment of COVID-19-associated retinopathy is currently not available in the clinic. Through mechanistic studies in clinically relevant COVID-19 animal models, we have found the essential role of VEGF in mediating the COVID-19-associated retinopathy and have demonstrated VEGF blockade for effective treatment of coviretinopathy. We anticipate that this therapy is most likely successfully translated into clinical practice for effective treatment of COVID-19-related retinal disease.

Abstract

COVID-19 has been associated with high prevalences of retinal diseases in humans. However, cellular and molecular mechanisms that underlie the COVID-19-associated retinopathy remains unknown. Here, we deployed a mouse COVID-19 model to investigate the causative link between SARS-CoV-2 infection and retinopathy development. Our data showed that COVID-19-induced pulmonary hypoxia triggered systemic hypoxia and markedly augmented VEGF expression levels in the retina and plasma. High VEGF levels altered vascular structures and functions in the retina, resulting in neovascularization, vascular disorganization, and increased leakiness. We deployed a terminology of coviretinopathy to accurately describe these COVID-19-induced pathological changes in the retina. Consequently, blocking VEGF by a specific neutralizing antibody (VEGF blockade) completely ablated the COVID-19-associated vascular changes in the retina. Together, these findings provide mechanistic insights into the COVID-19-associated retinopathy and propose a therapeutic paradigm for effective treatment of coviretinopathy.

Retinopathy, often referring to retinal vascular diseases that include diabetic retinopathy (DR), neovascular age-related macular degeneration (neoAMD), retinopathy of prematurity (ROP), and retinal vein occlusion (RVO), is the leading cause of blindness across all age groups (1, 2). Proliferative retinopathy (PR) directly involves vascular damages caused by excessive neovascularization and nonproliferative retinopathy (NPR) is usually linked to dysfunction of blood perfusion and vascular leakiness. Under most pathological circumstances, PR and NPR are intertwined two pathological processes that concomitantly contribute to the onset, development, and progression of retinal diseases (3, 4).

One of the common mechanisms underlying various types of retinopathy is tissue hypoxia, which serves as a potent driver of retinal neovascularization and vascular remodeling (5). Hypoxia potently induces high expression of vascular endothelial growth factor (VEGF) via the hypoxia-inducible factor 1 (HIF-1)-mediated transcription activation (6). VEGF is a key angiogenic factor that displays multifarious vascular functions, including angiogenesis, vascular permeability, vascular remodeling, vascular homeostasis, and endothelial cell survival (7, 8). VEGF levels in various tissues are tightly regulated and excessive or insufficient VEGF production can lead to onset and progression of various diseases (9, 10). The VEGF-executed functions are further regulated by two tyrosine kinase receptors, VEGF receptor 1 (VEGFR1) and VEGFR2, displaying often opposing, but complimentary, vascular functions. While VEGFR2 mediates most of VEGF-stimulated functions, VEGFR1 may act as a decoy receptor to further maintain the homeostatic role of VEGF by preventing excessive signaling (11). Owing to their predominate roles in retinopathy, the VEGF-VEGFR signaling has become one of the most attractive targets for drug development and effective treatment of various retinal diseases. At the time of this writing, anti-VEGF agents are probably the most commonly used and effective drugs in the clinic for treatment of neoAMD, DR, diabetic macular edema (DME), and RVO. They are often prescribed as the first-line regimen in monotherapeutic settings for treatment of retinal diseases (12).

Severe COVID-19 manifests respiratory dyspnea that creates pulmonary hypoxia, which would lead to global hypoxia in various tissues and organs (13–15). In the pulmonary tissue, SARS-CoV-2 infection triggers a robust inflammatory response, angiogenesis, plasma extravasation, and fibrosis (16–18). These pathological changes further exacerbate tissue hypoxia, which augments VEGF expression (19). Since VEGF molecules are synthesized in various spliced isoforms with different molecular weights, the smaller soluble VEGF molecules may function as endocrine hormones by targeting remote tissues and organs (20, 21). Thus, in each tissue local and circulating VEGFs collectively participate in the COVID-19-associated vascular changes and pathology. In the case of retina, it is likely that both locally produced VEGF in the retina and circulating VEGF (cVEGF) molecules contribute to retinopathy development.

In this study, we deployed a mouse COVID-19 model to test the hypothesis of SARS-CoV-2 infection in causing retinopathy. In the clinic, it is known that the most common retinal changes of SARS-CoV-2 infection is the microvascular alterations by manifesting cotton wool spots and retinal microhemorrhages (22). Although these pathological changes have been recognized for some time, the underlying molecular mechanism remains unknown. We showed that in a COVID-19 mouse model both retinal and circulating VEGF levels were markedly elevated and induce vascular leakiness, a key process of NPR, which we designated as coviretinopathy. Blocking VEGF by a specific anti-VEGF neutralizing antibody completely prevented coviretinopathy development. These results provide a therapeutic paradigm for potential treatment of COVID-19-associated retinal disease by blocking VEGF functions.

Results

SARS-CoV-2-Induced Pulmonary Hypoxia, Inflammation, and Altered VEGF Expression.

To study the impact of COVID-19 on the retinal vasculature, we developed a mouse COVID-19 model using the original “wild-type” SARS-CoV-2 virus by intranasal administration. This COVID-19 model efficiently induced pulmonary infection in transgenic mice (K18-hACE2 C57BL/6J) that express human angiotensin-converting enzyme 2 (hACE2) (23). To validate the cell types that expressed hACE2, we employed an unbiased and integrated single-cell RNA-seq (scRNA-seq) approach in the lung and retinal tissues. Isolated 13,307 retinal and 11,750 lung tissue cells from the wide-type (WT) C57BL/6J mice, 9,855 retinal and 12,444 lung tissue cells from K18-hACE2 C57BL/6J mice were subjected for scRNA-seq analysis. Data from scRNA-seq showed the presence of hACE2 only in the hACE2 C57BL/6J mice but not in WT C57BL/6J mice (SI Appendix, Figs. S1 and S2). Expectedly, alveolar type 1 (AT1) and type 2 (AT2) of lung epithelial cells predominantly expressed hACE2 (SI Appendix, Figs. S1 and S2). Additionally, several cell types in the retinal tissue expressed hACE2, including neuroretinal cells, corn, rod, bipolar cells, and astrocytes. The hACE2 distribution pattern in K18-hACE2 C57BL/6J mice highly relevant to hACE2 expression in human retinal tissues, including the retinal ganglion cell layer, inner plexiform layer, inner nuclear layer, and photoreceptor outer segments express hACE2 (24).

In addition to our previously described pathological changes in lung tissues (21), we focused on pulmonary hypoxia and inflammation after SARS-CoV-2 infection. Consistent with the COVID-19 pulmonary pathology, SARS-CoV-2 induced marked inflammation and alveolar solidification at days 5 and 7 postinfection (Fig. 1 A and B). Lung tissue hypoxia was detected by immunohistochemical staining using carbonic anhydrase 9 (CA9) and pimonidazole as probes. Both CA9 and pimonidazole staining showed severe tissue hypoxia at days 5 and 7 postinfection (Fig. 1 A, C and D). High HIF-1α expression further corroborated the severe lung hypoxia in the SARS-CoV-2-infected lungs relative to noninfected controls (Fig. 1 A and E). Quantitative PCR (qPCR) analysis of Ca9 and Hif1a further support the severe pulmonary hypoxia in our mouse COVID-19 model (Fig. 1 F and G). In concordance with elevated mRNA levels, Western blotting demonstrated marked increases of pimonidazole, CA9, and HIF-1α protein levels in the SARS-CoV-2-infected lung tissues (Fig. 1 H–L).

Fig. 1.

SARS-CoV-2-induces pulmonary hypoxia and upregulates VEGF expression. (A) H&E staining was used to reveal lung tissue structures in noninfected (NI) mice and SARS-CoV-2 (SC)-infected mice at days 5 and 7 postinfection. Immunofluorescence staining of lung tissues for pimonidazole⁺, CA9⁺, and HIF-1α⁺ hypoxic areas, with nuclei counterstained using DAPI. (Scale bar, 25 μm.) (B) Quantification of alveolar diameter in lung tissues of NI or SC-infected mice (n = 8 random fields per group). (C–E) Quantification of pimonidazole-, CA9-, and HIF-1α-positive signals in lung tissues (n = 8 random fields per group). (F and G) qPCR analysis of Ca9 and Hif1a mRNA expression (n = 6 to 7 samples per group) in lung tissues of NI or SC-infected mice at days 5 and 7 postinfection. (H–J) Western blot and quantification of HIF-1α and CA9 expression in lung tissues of NI or SC-infected mice across days 1 to 7 postinfection (n = 3 to 5 samples per group). β-actin was used as a loading control. (K and L) Quantification of tissue hypoxia in lung tissues of mice infected with NI or SC at days 5 and 7 postinfection (n = 4 samples per group). (M) qPCR quantification of Vegf mRNA in lung tissues of mice infected with NI or SC at days 5 and 7 postinfection (n = 11 samples per group). (N) Quantification of circulating VEGF (cVEGF) protein levels in plasma of NI or SC-infected mice at days 5 and 7 postinfection (n = 4 mice per group). (O) Immunofluorescence staining of lung tissues for VEGF⁺ area colocalized with F4/80⁺, CD80⁺, or CD206⁺ macrophage. (Scale bar, 25 μm.) (P–R) Quantification of VEGF positive signals in the macrophages in lung tissues of mice infected with NI or SC (n = 6 random fields per group). Data are presented as mean ± SEM. NI, noninfected; SC, SARS-CoV-2; CA9, carbonic anhydrase 9; HIF, hypoxia-inducible factor; DAPI, 4’,6-diamidino-2-phenylindole; cVEGF, circulating vascular endothelial growth factor.

Since the Vegf is one of the key HIF-1-targeted genes, high expression of HIF-1 in the SARS-CoV-2-infected lungs would likely to augment VEGF expression. We quantitatively measured VEGF mRNA and protein levels in lung tissues. Expectedly, Vegf mRNA levels were markedly elevated in the SARS-CoV-2-infected lung tissues relative to controls (Fig. 1M). Because small molecular weight VEGFs lack heparin-binding affinity, they often enter the circulation and target vasculatures in remote tissues and organs, we measure the cVEGF protein levels in the plasma of SARS-CoV-2-infected and -noninfected animals. Indeed, exceptionally high cVEGF levels that exceeded more than 25-fold increases above control levels were detected in the plasma at days 5 and 7 after SARS-CoV-2 infection (Fig. 1N). These data demonstrate that SARS-CoV-2 causes severe lung hypoxia, which subsequently stimulates VEGF expression. Immunofluorescence staining showed that VEGF increased in the macrophages of the SARS-CoV-2-infected lung tissues relative to controls (Fig. 1 O–R).

Retinal Hypoxia and High VEGF Expression.

We hypothesize that pulmonary hypoxia would inevitably trigger a systemic hypoxic insult, which affect nearly all tissues and organs owing to the essential role of oxygen exchange in the lung tissue. To investigate hypoxia in the retina, we isolated retinas from SARS-CoV-2-infected and noninfected mice on day 6 or day 7 after inoculation of SARS-CoV-2 virus at the experimental endpoint. Ca9, Hif1α and Vegf mRNA levels was markedly increased in the retina (SI Appendix, Fig. S3 A–C). Consistent with the increased mRNA levels, retinal CA9 and HIF-1α protein levels were accordingly increased (SI Appendix, Fig. S3 D, F, and G). Quantification of pimonidazole levels further validated the existence of severe hypoxia in the retina of the SARS-CoV-2-infected animals (SI Appendix, Fig. S3 E and H). Together, these results show that VEGF protein is upregulated in the retina of the SARS-CoV-2-infected animals.

Retinal Neovascularization.

Given high local retinal and circulating VEGF, we next investigated retinal neovascularization in the retinas of SARS-CoV-2-infected and noninfected mice. We performed CD31 and Ki-67 double immunostaining, which allowed us to detect proliferating endothelial cells in the retina. Consistent with an angiogenic phenotype, the SARS-CoV-2-infected retinal showed increased CD31⁺/Ki-67⁺ double-positive signals (Fig. 2 A–C). These data demonstrate that retinal neovascularization occurs in the SARS-CoV-2-infected mice.

Fig. 2.

SARS-CoV-2 induces retinal neovascularization. (A) Immunofluorescence staining of retinal sections for Ki-67⁺ proliferating cells and CD31⁺ microvessels, with nuclei counterstained using DAPI. (Scale bar, 25 μm.) (B and C) Quantification of Ki-67-positive signals in retinal sections (B) and whole-mount retinas (C) (n = 6 to 8 random fields per group). (D) Schematic diagram depicting the positional segmentation of whole-mount retinas into three analyzed regions: 1) central, 2) paracentral, and 3) peripheral areas. (E) Immunofluorescence staining of paracentral retinal regions for CD31⁺ microvessels across different retinal vascular layers. (Scale bar, 100 μm.) (F–I) Quantification of microvascular density, branching points, intermicrovessel lacunarity, and vessel tortuosity across different retinal vascular layers in the paracentral retinal regions of the retina (n = 8 to 11 random fields per group). (J) Immunofluorescence staining of peripheral retinal regions for CD31⁺ microvessels across different retinal vascular layers. (Scale bar, 100 μm.) (K–N) Quantification of microvascular density, branching points, intermicrovessel lacunarity, and vessel tortuosity across different retinal vascular layers in the peripheral retinal regions of the retina (n = 8 to 14 random fields per group). Data are presented as mean ± SEM. NI, noninfected; SC, SARS-CoV-2; DAPI, 4’,6-diamidino-2-phenylindole.

To further validate neovascularization, we divided each retina into 3 segments: 1) the central area; 2) paracentral area; and 3) peripheral area (Fig. 2D). In the central area, no overt changes of platelet endothelial cell adhesion molecule 1 (PECAM-1, CD31)-positive microvascular density between the SARS-CoV-2-infected retinas versus controls (SI Appendix, Fig. S4). In contrast, increased microvascular densities were detected in both paracentral (Fig. 2 E–I) and peripheral areas (Fig. 2 J–N) of the retina. Several distinct vascular alterations, including increases of microvascular densities, increases of branching points, morphologically twisted microvessels, and decreases of intermicrovessel distances, corroborated an angiogenic phenotype in the SARS-CoV-2-infected retina (Fig. 2 D–N). These angiogenic phenotypes existed throughout the retina layers, including superficial, intermediate, and deep layers of the retina (Fig. 2 D–N).

To exclude the direct impact of SARS-CoV-2 virus on developing retinopathy and retinal neovascularization, we examined viral components in retinal tissues (SI Appendix, Fig. S5). The expression of SARS-CoV-2 Envelope (E), Nucleocapsid (N), and Spike (S) genes was not detectable in retina by qPCR (SI Appendix, Fig. S5B). The expression of Spike receptor-binding domain (Spike-RBD) was also not detectable in retina by immunoblotting and immunohistochemistry (SI Appendix, Fig. S5 D and E). By contrast, high expression of SARS-CoV-2 virus components were detected in the SARS-CoV-2 virus-infected lung tissues (SI Appendix, Fig. S5 A, C, and E). Thus, it is unlikely that the SARS-CoV-2 virus would have direct effect on retinal vasculatures.

Pericyte Loss and Vascular Instability.

Unlike vasculatures in most other tissues, the retinal microvasculature contains high numbers of neuron-glia antigen 2 (NG2)⁺ pericytes that protect microvessels against leakiness, stabilize vascular structures and architectures, and maintain vascular homeostasis (25, 26). These pericyte-executed functions are particularly important in the retina because the retinal tissue lacks the lymphatic system to collect the extravasated fluids. Our previous work related to cancer-associated retinopathy (CAR) demonstrated that high levels of VEGF ablate pericytes from retinal vessels (27). To study the impact of COVID-19 on retinal pericytes, retinal vasculatures from noninfected and the SARS-CoV-2-infected animals were double immune-stained with CD31 and NG2 (Fig. 3A). Quantitative analysis demonstrated a loss of pericyte numbers in the SARS-CoV-2-infected group versus the control group in the superficial layer (Fig. 3 A and E). Total numbers of vascular pericytes located in the paracentral and peripheral areas are particularly decreased, indicating retinal microvasculatures were primarily affected by SARS-CoV-2 infection (Fig. 3 A and E). In contrast, NG2⁺ pericytes located in relatively large vessels of the central area remained unaffected. In addition to cell number changes, pericyte also underwent a phenotypical alteration by manifesting round-shape and less-coverage (arrow) in retinal vessels (Fig. 3 A and F). These phenotypes are defined as a selective loss of intramural pericytes from retinal capillaries, early signs of pericyte degeneration. Quantification analysis showed an increased number of rounded pericytes in the SARS-CoV-2-infected retinas (Fig. 3F). Unlike NG2⁺ pericytes, alpha-smooth muscle cell actin (α-SMA) in relatively large vessels remain unchanged (SI Appendix, Fig. S6).

Fig. 3.

SARS-CoV-2 triggers retinal microvascular dysfunction. (A) Immunofluorescence staining of retinal whole-mounts for NG2⁺ pericytes and CD31⁺ microvessels. (Scale bar, 50 μm.) (B) Immunofluorescence staining of retinal whole-mounts for collagen IV⁺ vascular basement membranes and CD31⁺ microvessels. (Scale bar, 100 μm.) (C) Immunofluorescence staining of retinal whole-mounts for VE-cadherin⁺ endothelial junctions, CD31⁺ microvessels, and Laminin⁺ vascular basement membranes. (Scale bar, 25 μm.) (D) Immunofluorescence staining of retinal whole-mounts for occludin⁺ endothelial junctions, and CD31⁺ microvessels. (Scale bar, 50 μm.) (E and F) Quantification of NG2-positive pericytes and rounded pericytes in retinal whole-mounts (n = 8 random fields per group). (G) Quantification of collagen IV-positive but CD31-negative regions in medium-sized vessels (n = 8 random fields per group). (H) Quantification of collagen IV-positive but CD31-negative regions in acellular capillaries (n = 8 random fields per group). (I and J) Quantification of retinal whole-mounts for VE-cadherin⁺ and occludin⁺ endothelial junctions within CD31⁺ microvessels (n = 6 random fields per group). Data are presented as mean ± SEM. NI, noninfected; SC, SARS-CoV-2; NG2: neuron-glia antigen 2; No., Number.

Loss of pericytes in retinal microvessels could possibly affect vascular stability, which causes vascular dysfunctions (28). To study vascular stability, we performed immunohistochemistry using collagen IV and CD31 double staining. Intriguingly, a number of medium-sized vessels only showed collagen IV positivity and lacked CD31 positive signals (Fig. 3 B and G, triangle). These data demonstrate vascular regression occurred in response to pericyte loss. Under other physiological and pathological conditions such as vascular remodeling and tumors, it is known that pericyte loss may result in vascular instability and regression. In contrast to vascular regression, retinal capillaries show a sprouting-phenotype by increasing vascular branching points and decreasing intercapillary distance (Fig. 3 B and H, arrow). These results show opposing effects of COVID-19 on retinal medium-sized vessels and capillaries, most likely due to pericyte loss-committed vascular instability and capillary sprouting.

Endothelial tight-junctions are one of the key structures that determine vascular permeability, which are adhered by VE-cadherins and other adhesion molecules (29). In the SARS-CoV-2-infected retinas, we have found significant decreases of inter-endothelial cell junction molecules, including VE-cadherin and occludin in capillaries (Fig. 3 C, D, I, and J). These results suggest potential leakiness of retinal microvessels in response to COVID-19.

Vascular Leakage.

Pericyte loss, vascular regression, and neovascularization often concomitantly occur with increased vascular permeability and edema (30). To study vascular leakiness, we used 70-kDa dextran (red) to monitor vascular leakage at various time points after inoculation of SARS-CoV-2 viruses. At day 5 postinfection, leakiness of 70 kDa-LRD (arrow) became overt and persisted until the end of the experiments (Fig. 4 A–C). Quantification of both total leaky areas and numbers of leaky acnes demonstrated marked increases of vascular leakage in the retina of SARS-CoV-2-infected animals (Fig. 4 B and C). Vascular perfusion was determined using 2,000 kDa-LRD and CD31 double immunostaining. In concordance with vascular regression, a number of retinal vessels lacked blood perfusion along SARS-CoV-2 infection (Fig. 4 A and D). In-depth analysis of vascular permeability demonstrated that leakiness of 70 kDa-LRD (arrow) occurred throughout the retina, including paracentral and peripheral areas (Fig. 4 E and F). The leaky phenotype exists in all three retinal vascular layers, including the superficial, intermediate, and deep plexuses (arrows) (Fig. 4 G–I). These results demonstrate that COVID-19 has functional impacts on vascular permeability and blood perfusion in the retina.

Fig. 4.

SARS-CoV-2 infection causes retinal vascular leakage. (A) Micrographs showing leakage of 70-kDa lysine-fixable rhodamine dextran (LRD, red), and perfusion of 2,000-kDa LRD (red) in association with CD31⁺ microvessels (green) in the retinas of noninfected (NI) and SARS-CoV-2 (SC)-infected mice. (Scale bar, 50 μm.) (B–D) Quantification of leakage area, leakage numbers, and vascular perfusion in retinas of NI or SC-infected mice (n = 8 random fields per group). (E) Micrographs of leakiness of 70-kDa LRD (red) in central, paracentral, and peripheral retinal regions of NI, day-5 or day-7 post-SC-infected mice. (Scale bar, 100 μm.) (F) Quantification of leakiness in central, paracentral, and peripheral retinal regions of NI, day-5 or day-7 post-SC-infected mice (n = 8 random fields per group). (G) Schematic presentation of the positions analyzed on whole-mount retinas. (H) Micrographs of leakiness of 70-kDa LRD (red) in superficial, intermediate, and deep retinal layers of SC-infected mice. (Scale bar, 50 μm.) (I) Quantification of leakiness in different retinal layers (n = 8 random fields per group). Data are presented as mean ± SEM. NI, noninfected; SC, SARS-CoV-2; LRD, Lysine-fixable rhodamine dextran; No., Number. Panel G was created using Figdraw (www.figdraw.com).

Exposure to Hypoxia Alone in Retinal Vascular Changes.

In addition to the SARS-CoV-2 infection model, we independently assessed the role of hypoxia alone without SARS-CoV-2 infection in regulation of VEGF expression and retinal vascular changes. Adult mice were successively exposed to low oxygen from 21% O2 to 7% O2 by reducing 1% O2 per day. The mice were maintained in 7% O2 for 2 wk. After exposing adult mice to this severe hypoxic environment for 14 d, we analyzed the retinal vascular changes. We did not see retinal neovascularization under hypoxia alone (SI Appendix, Fig. S7). These results are consistent with published findings (31). One of the explanations of lacking retinal neovascularization by hypoxia alone is the limited upregulation of VEGF expression. We detected the VEGF levels in hypoxia-exposed mice in the plasma and lung tissues. Marginal increases of plasma VEGF levels by less than 1.5-fold are detected under hypoxia alone (SI Appendix, Fig. S7C). By contrast, more than a 25-fold increase in plasma VEGF levels was detected in SARS-CoV-2-infected mice compared with controls (Fig. 1N).

Hypoxia exposure alone was unable to trigger a robust inflammatory response as seen in the COVID-19 animals (Fig. 1 and SI Appendix, Fig. S7). We show that pulmonary F4/80⁺ macrophages were important cellular sources of VEGF production (Fig. 1 O–R). Huge differences of VEGF levels existed between SARS-CoV-2-infected adult mice and hypoxia-exposed adult mice. We believe that the exceptionally high VEGF levels in SARS-CoV-2-infected mice contribute to retinal neovascularization, whereas the hypoxia exposure alone is unable to induce retinal neovascularization in adult mice.

Improvement of Pulmonary Hypoxia and Coviretinopathy by VEGF Blockade.

To investigate the role of VEGF in coviretinopathy, we deployed an anti-mouse neutralizing antibody (VEGF blockade) that is known to block VEGF functions (32). VEGF blockade was intraperitoneally administrated to each mouse on the day of SARS-CoV-2 virus inoculation. To investigate tissue distribution of VEGF blockade and pharmacokinetic half-life, we performed liquid chromatography–mass spectrometry (LC–MS/MS) analysis in plasma, lungs, and retina following a single-dose i.p. injection. As shown in SI Appendix, Table S1 and Fig. S8, the half-life of this antibody was 32.20 ± 8.37 h in plasma; 48.59 ± 7.16 h in lung tissues; and 120.00 ± 50.48 h in retina (SI Appendix, Fig. S8). It is very interesting that this anti-VEGF neutralizing antibody had very long half-life in the retina.

Systemic treatment of VEGF blockade markedly alleviated hypoxia in lung tissues of SARS-CoV-2-infected animals (Fig. 5 A–O), corroborating our clinical findings of anti-VEGF for effective treatment of severe COVID-19 (21). The anti-VEGF-treated lung tissues showed decreased positive signals of CA9, HIF-1α, and pimonidazole, indicating improved oxygen perfusion in these SARS-CoV-2-infected animals (Fig. 5 A–O). Along hypoxic improvement, pulmonary VEGF expression levels were also significantly decreased (Fig. 5 J and P–S).

Fig. 5.

Blocking VEGF prevents SARS-CoV-2-induced pulmonary hypoxia. (A) Schematic illustration of the COVID-19 animal model and anti-VEGF treatment regimen. (B) Hematoxylin and eosin (H&E) staining of lung tissues from nonimmune IgG (NIIgG)- and anti-VEGF-treated noninfected (NI) or SARS-CoV-2 (SC)-infected mice. (Scale bar, 25 μm.) (C) Quantification of alveolar diameter in lung tissues (n = 8 random fields per group). (D) Immunofluorescence staining of lung tissues for pimonidazole⁺, CA9⁺, and HIF-1α⁺ hypoxic areas. Sections were counterstained with 4’,6-diamidino-2-phenylindole (DAPI). (Scale bar, 25 μm.) (E–G) Quantification of pimonidazole-, CA9-, and HIF-1α- positive signals in lung tissues (n = 8 random fields per group). (H–J) qPCR analysis of Ca9, Hif1a, and Vegf mRNA expression in lung tissues of NIIgG- or anti-VEGF-treated NI or SC-infected mice (n = 6 to 14 samples per group). (K–M) Western blot and quantification of CA9 and HIF-1α expression in lung tissues of NIIgG- or anti-VEGF-treated NI or SC-infected mice (n = 3 samples per group). β-actin was used as a loading control. (N and O) Quantification of tissue hypoxia in lung tissues of NIIgG- or anti-VEGF-treated NI or SC-infected mice (n = 3 samples per group). (P) Immunofluorescence staining of lung tissues for VEGF⁺ area colocalized with macrophage markers F4/80, CD80, or CD206. (Scale bar, 25 μm.) (Q–S) Quantification of VEGF⁺ macrophages in lung tissues of NIIgG- or anti-VEGF-treated NI or SC-infected mice (n = 6 random fields per group). Data are presented as mean ± SEM. NI, noninfected; SC, SARS-CoV-2; CA9, carbonic anhydrase 9; HIF, hypoxia-inducible factor; DAPI, 4’,6-diamidino-2-phenylindole; VEGF, vascular endothelial growth factor.

In the retina, anti-VEGF treatment completely blocked the COVID-19-caused neovascularization, vascular permeability, and vascular regression (Fig. 6 A–O). In line of vascular normalization, retinal microvessels showed improved blood flow (Fig. 6 A–O). These anti-VEGF-treated retinal vasculatures in SARS-CoV-2-infected animals were morphologically and functionally indistinguishable from those of noninfected animals. We should emphasize that anti-VEGF treatment alone in the noninfected animals had no impact on healthy retinal vasculatures (Fig. 6 A–O). These results provide compelling evidence to demonstrate that coviretinopthy is completely dependent on VEGF and anti-VEGF treatment may provide a therapeutic paradigm for effective treatment of COVID-19-associated retinopathy (Fig. 7).

Fig. 6.

VEGF blockade improves coviretinopathy. (A) Immunofluorescence staining of retinal sections for Ki-67⁺ proliferating cells and CD31⁺ microvessels from nonimmune IgG (NIIgG)- and anti-VEGF-treated noninfected (NI) or SARS-CoV-2 (SC)-infected mice. Sections were counterstained with DAPI (blue). (Scale bar, 25 μm.) (B and C) Quantification of Ki-67-positive signals in retinal sections (B) and whole-mount retinas (C) (n = 6 to 8 random fields per group). (D) Immunofluorescence staining of retinal whole-mounts for NG2⁺ pericytes and collagen IV⁺ vascular basement membranes within CD31⁺ microvessels (n = 8 random fields per group). (Scale bar, 100 μm.) (E and F) Quantification of NG2-positive pericytes and rounded pericytes in retinal whole-mounts (n = 8 random fields per group). (G) Quantification of collagen IV-positive but CD31-negative regions in medium-sized vessels (n = 8 random fields per group). (H) Quantification of collagen IV-positive but CD31-negative regions in acellular capillaries (n = 8 random fields per group). (I–K) Immunofluorescence staining and quantification of retinal whole-mounts for VE-cadherin⁺ and occludin⁺ endothelial junctions within CD31⁺ microvessels, as well as Laminin⁺ vascular basement membranes (n = 6 to 8 random fields per group). (Scale bar, 50 μm.) (L) Micrographs of leakiness of 70-kDa lysine-fixable rhodamine dextran (LRD, red), and perfusion of 2,000-kDa LRD (red) in association with CD31⁺ microvessels (green) in retinal whole-mounts of NIIgG- or anti-VEGF-treated NI or SC-infected mice. (Scale bar, 100 μm.) (M–O) Quantification of leakage area, leakage numbers and vascular perfusion in retinal whole-mounts of NIIgG- or anti-VEGF-treated NI or SC-infected mice (n = 8 fields per group). Data are presented as mean ± Â SEM. NI, noninfected; SC, SARS-CoV-2; NG2: neuron-glia antigen 2; No., Number; LRD, Lysine-fixable rhodamine dextran.

Fig. 7.

Schematic illustration of a therapeutic strategy for coviretinopathy. SARS-CoV-2 infection-induced pulmonary hypoxia leads to both systemic and retinal hypoxia, triggering VEGF upregulation and resulting in neovascularization, vascular disorganization, and leakage—collectively termed coviretinopathy. VEGF blockade effectively reverses these pathological changes, offering a potential therapeutic approach for COVID-19-associated retinopathy.

K18-hACE2 C57BL/6J Female Mice Exhibit Similar Coviretinopathy.

Differential expression of ACE2 has been reported in male and female COVID-19 patients, which is associated with estrogen levels (33). To study if female mice would exhibit similar pathological phenotypes in retinal and pulmonary tissues, we performed the retinopathy analysis and anti-VEGF responses in SARS-CoV-2-infected and noninfected K18-hACE2 C57BL/6J female mice. Our data show that SARS-CoV-2 infection induced the same retinopathy and pulmonary phenotypes in female mice as seen in male mice (SI Appendix, Figs. S9 and S10 and Figs. 1–6). Similar to male animals, coviretinopathy in female mice was markedly improved by anti-VEGF therapy (SI Appendix, Fig. S10). Our results show that both male and female mice develop similar coviretinopathy that is responsive to anti-VEGF therapy.

Discussion

COVID-19-associated ophthalmological complications in human patients frequently manifest retinovascular abnormalities, including bilateral diffuse peripapillary cotton wool spots, microvascular hemorrhages, and Purtscher-like retinopathy (22). These clinical observations suggest that pulmonary infection by SARS-CoV-2 likely triggers a systemic disease by targeting microvasculatures in other tissues. In this study, we provide compelling evidence to demonstrate that COVID-19-augmented pulmonary hypoxia is the primary driver for causing coviretinopathy. The consequence of pulmonary hypoxia would inexorably result in global tissue hypoxia in various organs, including eyes. Thus, both retinal local and systemic hypoxia contribute to development of coviretinopathy.

Unlike other tissues and organs, the retinal vasculature mainly consists of microvessels that are highly sensitive to blood oxygen levels (34). To a certain extent, retinovascular changes may serve as an open window to dictate blood oxygen levels. For example, exposure of newborns to hyperoxia followed by normoxia can sufficiently induce retinal neovascularization and retinopathy (34). Similarly, systemic exposure of adult zebrafish to hypoxic water results in robust retinal neovascularization (35, 36). Since VEGF is a key targeted gene of hypoxia, it is not surprising that VEGF participates in retinal neovascularization under various pathological conditions. In fact, the hypoxia-VEGF axis plays a central role in the onset, development, and progression of various retinal disorders (37). Extraordinary benefits of anti-VEGF drugs in the clinic for treatment of DR, DME, neoAMD, and RVO proves the central role of the VEGF-induced vascular abnormalities in retinal disease (4, 12). Unlike these clinically approved indications, coviretinopathy involves retinal and systemic VEGF molecules in development of retinal complications. We show that circulating VEGFs in the SARS-CoV-2-infected animals are elevated to an exceptionally high levels, most likely originated from pulmonary and other hypoxic tissues. Circulating VEGFs probably represent a small fraction of soluble VEGF molecules because the large isoforms of VEGF, owing to their high affinity to heparan sulfate proteoglycans in the extracellular matrix, preferentially exert their biological functions in the local tissues where they are synthesized (38–40). Along this line of thinking, the ratio between small and large molecular weight VEGF isoforms also altered in the retina. At the time of this writing, it is uncertain if any functional differences exist between the small and large VEGF isoforms in causing coviretinopathy. Perhaps, increased proportions of smaller isoforms are responsible for the clinically manifested COVID-19-associated diffuse peripapillary cotton wool spots, microvascular hemorrhages, and Purtscher-like retinopathy (22). This interesting issue warrants future investigation. Although the clinically approved indications of treatment of ophthalmological disorders with neovascularization by anti-VEGF drugs, these drugs should be considered for effective treatment of other nonneovascularization-related eye diseases by blocking VEGF-mediated nonangiogenic effects. For example, VEGF displays potent permeability effect that contributes to the onset, development of and progression of various ocular disorders (41). Anti-VEGF therapy warrants clinical evaluation and perhaps authority approval for treatment of coviretinopathy.

Although in the present study we focus our efforts on gaining mechanistic insights into the causative link between COVID-19 and retinopathy, our findings have broad implications in other infectious and noninfectious diseases. In principle, various pulmonary disorders, including virus infections, bacterial infections, lung tissue damage, inhalation of toxic chemicals, smoking, and malignant diseases can all cause retinopathy through the same hypoxia-VEGF mechanism as described in this study. For example, one of our previous studies demonstrated lung cancer–induced cancer-associated retinopathy (CAR) is also driven by VEGF-induced vascular changes (27). In further support of this notion, a large number of various infections have been associated with the onset of retinopathy, including cytomegalovirus (CMV), measles, influenza, Epstein–Barr virus, and Rift Valley fever virus, typically occurs subsequent to an acute viral systemic illness (42). In experimental animal models, Bronchopulmonary dysplasia (BPD) has been causatively linked to ROP development owing to hypoxia (34, 43). In this regard, our data suggest that targeting VEGF provides a therapeutic paradigm for effective treatment of pulmonary disorder-associated retinopathy. To the best of our knowledge, this important concept remains unprecedented and needs further exploration.

In our study, a humanized rabbit anti-VEGF neutralizing antibody was used to block VEGF functions. This antibody has recently been approved by the China-FDA for treatment of human ovarian cancer (44). We show that anti-VEGF drugs effectively prevent coviretinopthy development in experimental mice. As anti-VEGF drugs are broadly used in various ophthalmological disorders and cancers, expansion of clinical indications of these drugs for treating coviretinopthy would be feasible. In particular, this therapeutic concept should be considered for effective treatment of retinovascular diseases caused by epidemic and pandemic respiratory viruses.

Materials and Methods

K18-hACE2 transgenic mice were used to establish SARS-CoV-2 infection models and were treated with a VEGF-neutralizing antibody or isotype control. Antibody pharmacokinetics in plasma and tissues were quantified using a validated LC–MS/MS assay (SI Appendix, Table S1). Lung and retinal tissues were collected for histology, immunohistochemistry, whole-mount staining, vascular perfusion assays, and molecular analyses using established methods. Single-cell RNA sequencing of lung and retinal tissues was performed using the 10x Genomics platform followed by standard bioinformatic analyses. Detailed procedures, reagents, and data analysis methods are provided in the SI Appendix.

Ethics Statement

All animal experiments were conducted in accordance with approved ethical permits (10712–2020 and 16975–2024) at the Astrid Fagraeus Laboratory, Karolinska Institutet, following Swedish Board of Agriculture regulations. The protocols were approved by the local ethics committee (Stockholms Norra Djurförsöksetiska Nämnd). Animals were maintained under standard conditions and handled according to ARRIVE guidelines. Further details are provided in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

RNA sequencing data have been deposited in NCBI GEO (GSE315089) (45). All study data are included in the article and/or SI Appendix.