75 — Embedding of Cortical Representations by the Superficial Patch System

This paper aims to better understand the superficial patch system, a clustering system found in layers 2 and 3 of many species, including cats and monkeys. Muir et al look to macaque v1 and cat area 17 (homologous primary visual cortex brain regions) to better understand the role of this structure in cortex response.

They hypothesize that the patch system is responsible for encoding a statistical expectation of cortical response to a stimulus: By collecting responses from several microns in every direction, these cells are able to help reduce noise by generating a population-wide smoothing or amplification function.

The first characterization the authors make is determining shape and layout of this system. They acrrue existing literature and generate a generative model to create hexagonal and square grids with random noise added. This hexagonal model looks like what we expect from the biological evidence (statistical agreement on node placement), while the square grid fails this test. From this, Muir et al determine that the superficial patch system node centroids tend to be hexagonally arranged across the cortical surface.

One feature that struck me as very strange is that patch distribution was larger in cats than monkeys: Macaque interdomain centroids are 0.5mm apart, whereas cat centroids are 0.9mm apart, according to the literature review in this paper. This means that more disparate data are encoded and summated in cat than in monkey, assuming visual input machinery is roughly the same size and breadth in these animals (which I appreciate is not necessarily a reliable assumption).

One of the first memories that jumped out at me while reading this paper was a description of pinwheel layout in macaque V1 from an early neuroscience course: The pinwheels were described as “the easiest way to cram that much signal-processing into so small a space.” It makes me wonder if there is something fundamental about the hexagonal tiling layout (rather than square) that makes it particularly well-suited to use in biological neural networks.

Rodents are phylogenetically bracketed by felines and monkeys, but they don’t show signs of this cortical patch system. This means that either ① rodents disposed of the patch system layout, or ② cats’ and monkeys’ ancestors evolved this system in convergence but separately. I find the first solution to be far more convincing of an answer, but it’s troubling and calls into question what signals precisely this system is encoding or communicating.

I wonder if there’s any correlation in occurrence with ocular dominance columns, also found in cats and monkeys but absent in rats.