Summary: Diabetic retinopathy (DR) is a complex disease displaying diverse vascular-associated complications, including upregulation of vascular endothelial growth factor (VEGF)-A165, enhanced vascular permeability, and retinal angiogenesis. Current animal models do not fully replicate the spectrum of DR pathologies. In this study, we aim to evaluate the efficacy of adeno-associated virus (AAV)-mediated long-term expression of human VEGF to establish angiogenic DR-related phenotypes in Brown Norway rats, as well as validate a novel artificial intelligence (AI) framework for autonomous quantification of retinal angiogenesis in a newly established DR model.
Abstract
INTRODUCTION: Diabetic retinopathy (DR) is a complex disease displaying diverse vascular-associated complications, including upregulation of vascular endothelial growth factor (VEGF)-A165, enhanced vascular permeability, and retinal angiogenesis. Current animal models do not fully replicate the spectrum of DR pathologies. In this study, we aim to evaluate the efficacy of adeno-associated virus (AAV)-mediated long-term expression of human VEGF to establish angiogenic DR-related phenotypes in Brown Norway rats, as well as validate a novel artificial intelligence (AI) framework for autonomous quantification of retinal angiogenesis in a newly established DR model.
MATERIALS AND METHODS: Brown Norway rats received single unilateral intravitreal injection of AAV-hVEGF165.V5 (5×1010 vg/eye) in the right eye on Day 0. The progression of retinal pathology was monitored weekly via fluorescein angiography (FA). Six weeks post-AAV administration eyes were enucleated, and retinal flat-mounts were stained with Isolectin B4 and panoramic flat-mount images were captured using fluorescent microscope. For the automated quantification of retinal angiogenesis, we employed a combination of deep learning with traditional computer vision algorithms. The artificial intelligence (AI) component for blood vessel recognition in retinal flat-mounts was based on transfer learning approach when a pre-trained U-Net architecture neural network [1] was fine-tuned with Recovery-FA19 [2] fluorescein angiography dataset derived from human subjects. The neural network-generated vascular masks were processed to quantify vascular area, total vessel length, and number of branch points.
RESULTS:
Figure 1.AAV-hVEGF Injection Led to Vascular Pathologies in Rat Eyes. (A) IVT injection of AAV-hVEGF led to vascular tortuosity, focally dilated or constricted vessels (mild DR phenotype, arrowheads), and microaneurysms (moderate DR phenotype, circles) 2 weeks afterinjection as observed by FA images. (B) Increasedvascular leakage was observed in 52% of AAV-hVEGF injected rats 2 weeks post-AAV administration (severe DR phenotype, arrows).
Figure 2. AAV-hVEGF Injection Led to Retinal Angiogenesis. (A) Representative images of Isolectin B4-labeled retinal flat-mount, grayscale image of retinal flat-mount, blood vessel segmentation mask and retinal flat-mount mask, used for calculations of vascular area. (B) Representative images of segmentation mask skeleton used for total vessel length calculations, and the same skeleton with marked branching points (C), used for branching point calculations. Scale bar = 500 μm for the Isolectin B4-labeled retinal flat-mount.
Figure 3. A significant increase was found in (A) Vascular area, (B) Total vessel length, and (C) Branching points 6 weeks post-AAV-hVEGF administration, as compared to contralateral non-injected eyes, *** P < 0.001. Branching points and total vessel length are presented per 100×100 px square. Statistical analysis was done by unpaired t-test. Data are presented as mean ± SEM, n = 4 -14 eyes per group.
CONCLUSIONS:
1. AAV-mediated expression of human VEGF in rat retinas demonstrates an easy-to-use model that recapitulates several aspects of DR pathology.
2. Our data validates the effectiveness of our novel AI algorithm in quantifying retinal vasculature within the newly established DR model.
Authors: Inesa Lelytė | Symantas Ragauskas | Marius Dragašius | Zubair Ahmed | Giedrius Kalesnykas| Nerija Kvietkauskienė
References
- [1] Ronneberger, O & Fischer, P & Brox, T, 2015. U-Net: Convolutional Networks for Biomedical Image Segmentation. LNCS. 9351. 234-241. 10.1007/978-3-319-24574-4_28
[2] Li Ding, Mohammad H. Bawany, Ajay E. Kuriyan, Rajeev S. Ramchandran, Charles C. Wykoff, Gaurav Sharma, June 3, 2019, “RECOVERY-FA19: Ultra-Widefield Fluorescein Angiography Vessel Detection Dataset”, IEEE Dataport, doi: https://dx.doi.org/10.21227/m9yw-xs04.