Comparing cell-type specific response across retinal diseases and degeneration

Abstract

Purpose : Photoreceptor cell death, culminating in vision loss arises from a variety of diseases such as inherited retinal diseases, diabetic retinopathy, and age-related macular degeneration. While each disease originates from variable alterations in physiological and molecular pathways, commonalities in cell-type specific responses across retinal degeneration is not well understood. The influence of environment on disease progression may impact studies, making it difficult to compare across the field. We hypothesize that using machine learning on large cohorts undergoing in-depth phenotypic and molecular analysis, we can identify commonalities in cell-type response across diseases.

Methods : In this study, we compare 14 mouse models encompassing 7 diseases to investigate cell-type specific response in connection to physical, functional, and molecular changes throughout early, mid, and late retinal degeneration. At each time point per line, we carried out in vivo analyses including ERG, OCT, angiography, visual acuity, and intraocular pressure. Similarly, in a second cohort we focused on pathology, performing blood work and 12 IHC stains along with H&E (>3000 slides). In a third cohort we performed flat mount immunostaining, and in the 4^th, bulk RNA-seq (>152 samples) and scRNA-seq (>60 samples) on the retina. The resulting assays, images, and analyses were then reduced to features for machine learning.

Results : While developing this dataset for interrogating commonalities across retinal degeneration diseases, we have built an online resource encompassing data collected from over 1000 mice. Pilot machine learning on this dataset validated known features associated with aging from our C57BL6/J line with 12 time points ranging from p14 to 30 months old. Furthermore, the scRNA-seq curated datasets from this study alone suggest that rod-specific responses have shared gene programs across the modeled diseases, correlating with kinetics of rod-degeneration.

Conclusions : While this data continues to be generated and analyzed with two additional mouse lines to collect soon, these data suggests that machine learning across murine mouse models’ molecular, physical, and functional phenotypes is viable for interrogating commonalities across disease. With further investigation, this dataset and overall resource will provide many avenues to investigate retinal degeneration disease etiology and potential therapeutic avenues to pursue.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.