A novel system for the classification of diseased retinal ganglion cells.

ABSTRACT

Retinal ganglion cell (RGC) dendritic atrophy is an early feature of many forms of retinal degeneration, providing a challenge to RGC classification. The characterization of these changes is complicated by the possibility that selective labeling of any particular class can confound the estimation of dendritic remodeling. To address this issue we have developed a novel, robust, and quantitative RGC classification based on proximal dendritic features which are resistant to early degeneration. RGCs were labeled through the ballistic delivery of DiO and DiI coated tungsten particles to whole retinal explants of 20 adult Brown Norway rats. RGCs were grouped according to the Sun classification system. A comprehensive set of primary and secondary dendrite features were quantified and a new classification model derived using principal component (PCA) and discriminant analyses, to estimate the likelihood that a cell belonged to any given class. One-hundred and thirty one imaged RGCs were analyzed; according to the Sun classification, 24% (n = 31) were RGCA, 29% (n = 38) RGCB, 32% (n = 42) RGCC, and 15% (n = 20) RGCD. PCA gave a 3 component solution, separating RGCs based on descriptors of soma size and primary dendrite thickness, proximal dendritic field size and dendritic tree asymmetry. The new variables correctly classified 73.3% (n = 74) of RGCs from a training sample and 63.3% (n = 19) from a hold out sample indicating an effective model. Soma and proximal dendritic tree morphological features provide a useful surrogate measurement for the classification of RGCs in disease. While a definitive classification is not possible in every case, the technique provides a useful safeguard against sample bias where the normal criteria for cell classification may not be reliable.