Unexpected functional diversity among mouse retinal ganglion cell types

Abstract

Purpose: Retinal circuits compute visual features in parallel and send this information to the brain through dedicated channels represented by the retinal ganglion cell (RGC) types. Anatomical studies have suggested up to 22 morphological RGC types, indicating that as many functional channels may form the output of the retina. Here we show that this estimate may need to be at least doubled to reflect the true functional diversity.

Methods: To reliably record from every cell in the ganglion cell layer we used bulk-electroporation (Briggman & Euler, 2011) and two-photon Ca²⁺ imaging. A standardized stimulus set, including temporal full-field stimulation, local motion, and dense noise for receptive field mapping, was presented to the retina. Some recordings were obtained from transgenic mice (PV, Pcp2), in which distinct RGC subsets are fluorescently labelled. Also, electrical single-cell recordings were performed to relate RGC spiking to somatic Ca²⁺ signals and to retrieve RGC morphologies.<br /> We implemented a probabilistic clustering framework for separating our sample of almost 10,000 cells (42 retinas) into functional clusters solely based on features extracted from their light responses using sparse PCA and mixture of Gaussians clustering. Then, the 70+ functional clusters were post-processed into “RGC groups” by a small number of cluster-split and -merge operations using meta data, such as immunolabels and morphological features.

Results: We found that RGCs can be divided into at least 30 functional types. Many of these matched known types, e.g. the “alpha” RGCs or the local-edge-detector (“W3”). In addition, we identified new RGC types, including an OFF DS RGC that does not co-stratify with starburst amacrine cells.<br /> To test if our RGC groups indeed correspond to single RGC types, we measured how well the dendritic fields of each group covered the retinal surface. Most RGC groups had a coverage factor (CF) of ~1, suggesting that none of them has been spuriously split. Some groups had a CF >>1, suggesting that they consist of multiple subtypes (e.g. the ON-OFF DS RGC had a CF ~4, presumably corresponding to the known subtypes pointing in the 4 cardinal directions). The summed CFs from all groups (57) indicate that there may be more than 50 types of RGCs in the mouse.

Conclusions: Taken together, our data indicates that information channels from the eye to the brain may be much more diverse than previously thought.