Can retinal ganglion cell dipoles seed iso-orientation domains in the visual cortex?

TL;DR

This study investigates whether retinal ganglion cell dipoles can initiate the formation of orientation preference maps in the visual cortex, finding that their spatial structure is unlikely to be the seed for such cortical patterns.

Contribution

The paper introduces a novel point process to analyze spatial correlations in RGC mosaics and demonstrates that these mosaics lack the necessary spatial structure to seed cortical orientation maps.

Findings

RGC dipole angles are spatially uncorrelated in cat and primate mosaics.

A new point process effectively models RGC mosaic spatial statistics.

Weak angular correlations in RGC mosaics are very unlikely given current data.

Abstract

It has been argued that the emergence of roughly periodic orientation preference maps (OPMs) in the primary visual cortex (V1) of carnivores and primates can be explained by a so-called statistical connectivity model. This model assumes that input to V1 neurons is dominated by feed-forward projections originating from a small set of retinal ganglion cells (RGCs). The typical spacing between adjacent cortical orientation columns preferring the same orientation then arises via Moir\'{e}-Interference between hexagonal ON/OFF RGC mosaics. While this Moir\'{e}-Interference critically depends on long-range hexagonal order within the RGC mosaics, a recent statistical analysis of RGC receptive field positions found no evidence for such long-range positional order. Hexagonal order may be only one of several ways to obtain spatially repetitive OPMs in the statistical connectivity model. Here, we investigate a more general requirement on the spatial structure of RGC mosaics that can seed the emergence of spatially repetitive cortical OPMs, namely that angular correlations between so-called RGC dipoles exhibit a spatial structure similar to that of OPM autocorrelation functions. Both in cat beta cell mosaics as well as primate parasol receptive field mosaics we find that RGC dipole angles are spatially uncorrelated. To help assess the level of these correlations, we introduce a novel point process that generates mosaics with realistic nearest neighbor statistics and a tunable degree of spatial correlations of dipole angles. Using this process, we show that given the size of available data sets, the presence of even weak angular correlations in the data is very unlikely. We conclude that the layout of ON/OFF ganglion cell mosaics lacks the spatial structure necessary to seed iso-orientation domains in the primary visual cortex.