Deep learning-driven characterization of single cell tuning in primate visual area V4 unveils topological organization

Deciphering the brain’s structure-function relationship is key to understanding the neuronal mechanisms underlying perception and cognition. The cortical column, a vertical organization of neurons with similar functions, is a classic example of primate neocortex structure-function organization. While columns have been identified in primary sensory areas using parametric stimuli, their prevalence across higher-level cortex is debated. A key hurdle in identifying columns is the difficulty of characterizing complex nonlinear neuronal tuning, especially with high-dimensional sensory inputs. Here, we asked whether area V4, a mid-level area of the macaque visual system, is organized into columns. We combined large-scale linear probe recordings with deep learning methods to systematically characterize the tuning of >1,200 V4 neurons usingin silicosynthesis of most exciting images (MEIs), followed byin vivoverification. We found that the MEIs of single V4 neurons exhibited complex features like textures, shapes, or even high-level attributes such as eye-like structures. Neurons recorded on the same silicon probe, inserted orthogonal to the cortical surface, were selective to similar spatial features, as expected from a columnar organization. We quantified this finding using human psychophysics and by measuring MEI similarity in a non-linear embedding space, learned with a contrastive loss. Moreover, the selectivity of the neuronal population was clustered, suggesting that V4 neurons form distinct functional groups of shared feature selectivity, reminiscent of cell types. These functional groups closely mirrored the feature maps of units in artificial vision systems, hinting at shared encoding principles between biological and artificial vision. Our findings provide evidence that columns and functional cell types may constitute universal organizing principles of the primate neocortex, simplifying the cortex’s complexity into simpler circuit motifs which perform canonical computations.