June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Large scale excitatory and inhibitory synapse detection using deep learning in an experimental mouse model of glaucoma
Author Affiliations & Notes
  • Luca Della Santina
    Vision Sciences, University of Houston College of Optometry, Houston, Texas, United States
    Ophthalmology, University of California San Francisco, San Francisco, California, United States
  • Yvonne Ou
    Ophthalmology, University of California San Francisco, San Francisco, California, United States
  • Footnotes
    Commercial Relationships   Luca Della Santina None; Yvonne Ou None
  • Footnotes
    Support  NIH grant EY028148, NIH grant EY002162, NIH grant EY007551, NVIDIA GPU grant, Glaucoma Research Foundation Shaffer grant, All May See Foundation grant, Research to Prevent Blindness unrestricted grant to UCSF Ophthalmology
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 487. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Luca Della Santina, Yvonne Ou; Large scale excitatory and inhibitory synapse detection using deep learning in an experimental mouse model of glaucoma. Invest. Ophthalmol. Vis. Sci. 2023;64(8):487.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : Glaucoma causes progressive dysfunction of the retina, ultimately leading to blindness. Animal models have shown that synapses between bipolar and ganglion cells are being dismantled early after IOP elevation. To understand whether degeneration differentially impacts excitatory (E) and inhibitory (I) circuits, we labeled and quantified E/I synapses in alpha retinal ganglion cells (aRGCs) as well as in large volumes of the inner plexiform layer (IPL) using automatic quantification via deep learning algorithms.

Methods : Transient IOP elevation was induced by photocoagulation of the episcleral and limbal vessels in adult CD1 mice. Retinal whole mounts were immunolabeled for the excitatory and inhibitory synaptic proteins PSD95, RibeyeA, Gephyrin and VGAT. Individual RGCs and their postsynaptic excitatory and inhibitory sites were labeled by biolistic transfection of PSD-95 and Gephyrin tagged with fluorescent proteins. The automatic pipeline for detection of synaptic proteins was developed in MATLAB within our open-source program ObjectFinder, deep learning models were trained using Nvidia Quadro RTX 8000 GPU using trained human annotations on biolistically labeled RGCs as ground truth. Statistical comparisons were performed using Wilcoxon-Mann-Whitney test.

Results : Automatic detection of both excitatory and inhibitory proteins was achieved with an accuracy rate greater than 90% by training DeepLab v3 plus classifier models with a minimum of 10,000 synapses manually validated by an expert annotator. Trained models were able to generalize across RGC types as well as across labeling methods (transfection versus immunolabeling). In individual aRGCs and across the IPL, loss of of excitatory synapses is observed as early as 7 days after IOP elevation (p<0.05), while loss of inhibitory synapses occurs at later time points (p<0.05 at 14 days for IPL, p<0.05 at 30 days for aOFF-Sustained RGCs).

Conclusions : Excitatory and inhibitory synapses are unevenly disassembled across the inner retina, with loss of excitatory synapses occurring prior to inhibitory synapses. Within the IPL, specific alpha RGC types are are differentially affected in their E/I balance. These findings underscore the capability of the adult retina to rearrange its connectivity in the face of degeneration, and identify potential targets to test early functional impairments in glaucoma.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×