In Vivo Vascular Injury Readouts in Mouse Retina to Promote Reproducibility

Noninvasive rodent retinal imaging presents an avenue to study pathology and develop interventions. Previous studies have developed and optimized a mouse model of RVO, limiting variability and allowing for reliable translation of common clinical pathologies in the murine retina^5,7,13. Developments in ophthalmic imaging technology further allow for the use of clinical in vivo imaging techniques such as FA and OCT in experimental animals, granting the ability to compare mouse models with profiles of human disease^6,12,15. However, to maximize the information that can be extracted from these images and the overall translational potential of the model, there is a need for standardized, reproducible, and rigorous quantitative methods for analyzing images. Here we present analysis methods that allow for quantitative representations of damage severity, allowing for more precise and reliable comparisons between mice and across experimental groups. These analyses include leakage quantification in FA images, quantification of mean layer thickness, and areas of DRIL in OCT images.

A critical factor in successful analysis lies in the quality of the acquired images. Poorly resolved OCT images can lead to difficulty tracing individual layers and an inability to distinguish inner retinal disorganization from poor image quality. When imaging, it is important to take care in the positioning of the mouse on the platform, ensuring that the fundus image is in focus, the optic nerve is relatively centered, and the retinal cross-section is horizontal across the image. Consistent lubrication of the eyes while the animal is anesthetized is also important, especially when the same animal is imaged multiple days. Insufficient lubrication may result in corneal cataracts, which will obscure the retina and render it unsuitable for imaging. Various retinal pathologies may occur in RVO imaging, rendering images unsuitable for analysis. These include excessive retinal detachment and excessive hemorrhage, which, along with greatly compromising the quality of imaging, also represent a degree of damage that is too severe to use as a model of RVO. It is additionally possible for all occluded vessels to fully reperfuse shortly after injury, which will not accurately model RVO damage and should be used as an exclusion criterion. However, it is important to note that successful occlusions will naturally resolve by 48-72 h post injury, and the presence of occlusions as an exclusion criterion is best used at or before 24 h post-procedure. Colón Ortiz et al.¹³ detail best practices for limiting variability and calibrating injury in an optimized model for RVO procedure. The identification and judgment of exclusion criteria is also a critical step to image analysis. As this is largely up to the discretion of the evaluator, it is important that evaluators are blinded to treatment groups and practice consistency in the judgment of pathology severity. Some limitations in the application of these methods exist, particularly in the practice of imaging the same mouse at multiple time points. There is a limit to the frequency at which a mouse can be anesthetized for imaging, necessitating the testing and adjustment of time points to determine optimal time course. Our studies employ imaging time points at 4 h, 24 h, 48 h, and 8 days, which we have found capture stages of initial injury, acute inflammatory response, and longer-term injury¹². Additionally, certain mouse strains are more prone to the development of corneal cataracts, which include various diabetic mouse models, which may lead to a large number of exclusions or incomplete time courses^20,21. Studies utilizing such mouse lines may need to tailor experimental group size or imaging time points depending on the sensitivity of the cornea.

Fluorescein angiography imaging has largely been used qualitatively to observe and grade retinal pathologies such as leakage, as well as patterns of altered blood flow RVO⁶. Recently, there have been efforts to develop a quantitative analysis of FA in animal models, such as calculation of vascular area and tortuosity¹⁶ and linear regression analysis of image intensity temporality¹⁷. Segmentation of the major vessels from the fundus background has previously been used, but in a pixel analysis of fill and decay dynamics, testifying to the variability in image intensity in different mice¹⁷. Additionally, the potential for bias was noted in the interpretation of fluorescein pooling¹⁷. The quantitative method discussed here targets the leakage of fluorescein from the major retinal vasculature, indicative of the breakdown of the BRB, which has been demonstrated to play a role in RVO injury^11,12,14. An alternative analysis of leakage quantifies dye leakage on retinal flat mounts²². However, invasive post-mortem analyses are less suitable for studies of the timeline of RVO injury within a single mouse, where the leakage is studied at multiple timepoints. Analyses of fluorescein leakage area at different stages of the retinal disease have previously been used in clinical studies and correlated with other observed disease pathologies²³. This method allows for similar leveraging of FA images to study vessel leakage in vivo, allowing for the study of leakage dynamics within the timeline of RVO injury. As the selection of leakage area relies on evaluator selection of a region, it potentially introduces a greater amount of variability via subjectivity. Further, since the studies of the RVO injury model discussed here investigate leakage throughout the retina, we have instead opted to use a masking technique for calculation. This leakage method reflects a different facet of RVO damage from those revealed by DRIL and OCT layer tracing analysis, and correlation with these measures allows for the creation of a more accurate disease profile.

We present two methods for the evaluation of OCT images. Acute inflammation and subsequent degeneration of the retinal layers is a hallmark of RVO injury^6,12. The OCT layer tracing methodology detailed here allows for the precise study of individual layers and reveals more subtle effects and differences in dynamics in different regions of the retina. This analysis technique builds upon other commonly used protocols for the quantification of retinal layer thickness in OCT imaging. This method addresses the variation across protocols in the area used to estimate layer thickness, as well as the number of measurements taken across the image¹¹. As thinning is not uniform within each retinal layer, methods using fewer point measurements are unlikely to give a complete picture of injury effects. Meta-analysis of multiple measurement strategies for retinal layer thickness reported that protocols averaging across larger areas of the OCT image showed a higher correlation with disease severity, as well as greater repeatability¹¹. By averaging across the whole image, this method captures a more accurate representation of the retinal thinning present in long-term RVO injury. Studies also differ in terms of the location where OCT images are taken-many studies center imaging on the optic nerve. By contrast, the presented method centers relative to the occlusions. A recent development in the analysis of human OCT imaging is the usage of machine learning algorithms to classify and quantify features²⁴. Such analyses could be a promising future direction for the analysis of animal retinal imaging.

Additionally, we present a translation of DRIL, a clinical measure of capillary ischemia, into a rodent model. In humans, DRIL has been found to be a predictor of visual acuity loss and retinal thickness differences and has demonstrated high diagnostic sensitivity and specificity^18,19. Quantifying the DRIL in mice by measuring the proportion of the retina that is disorganized has shown correlation with the fraction of occluded veins, ERG b wave amplitude at 7 days post-RVO, and retinal thinning at 8 days post-RVO¹². An alternative to DRIL measurement is the usage of HYPOX-4 to measure retinal hypoxia and ischemic damage. HYPOX-4 joins pimonidazole anime hydrochloride, a hypoxia marker, with a fluorescent probe to detect retinal hypoxia²⁵. Most protocols using HYPOX-4 are invasive and require retinal flat mount analysis, which may be less suitable for the building of injury timelines, though an in vivo imaging protocol using a HYPOX-4 probe has recently been piloted²⁵. DRIL analysis is also useful as a quick readout of retinal damage, as single measurements in each OCT image are more time-efficient than analyses such as retinal layer tracing. However, it should be noted that these measures are not interchangeable and reveal different retinal pathologies. Rather, they should be used in concert, where DRIL can be used as an initial readout for effect size or intervention efficacy, and layer tracing can be subsequently employed for a thorough analysis of more subtle effects in the retinal layers.

These methods are orthogonal in nature, which allows for the creation of a disease profile for each experimental subject. As the pathologies reported by each of these methods are distinct, they are not guaranteed to scale proportionally, and obtaining a more holistic picture of pathology will allow for a more rigorous investigation of the varying manifestation configurations of RVO damage. The ability to maximize the amount of information that can be extracted from the imaging of each experimental animal will reduce the number of animals necessary to draw significant conclusions, enhancing the efficiency of the experimental process. Applying these methods on recently refined RVO protocols allow for greater reproducibility and study of the translation of clinical phenotypes into animal models. Beyond the study of RVO models, the usage of these methods has applications to other models of retinal diseases that employ FA and OCT imaging. Examples of such mouse models include those for age-related macular edema (AMD)²⁶, diabetic macular edema (DME)²³, choroidal neovascularization (CNV)²⁷, experimental autoimmune uveoretinitis (EAU)²⁸, and retinopathy of prematurity (ROP)¹⁵. These methods can further be generalized to studies using FA and OCT imaging in studying models of these diseases in other species. These quantifications are also sensitive to more subtle changes in disease mechanism, making them useful in the evaluation of treatment efficacy, such as in Figure 3D and Figure 4C. Utility also extends to the use of imaging in toxicity testing in tolerability studies of drug compounds. The standardization and reproducibility of these analysis protocols can serve to improve the translational validity of animal models and expand our understanding of the pathogenesis and pathophysiology of the retinovascular disease.