Drifting population dynamics with transient resets characterize sensorimotor transformation in the monkey superior colliculus

Abstract

To produce goal-directed eye movements known as saccades, we must channel sensory input from our environment through a process known as sensorimotor transformation. The behavioral output of this phenomenon (an accurate eye movement) is straightforward, but the coordinated activity of neurons underlying it is not well understood. We searched for a neural correlate of sensorimotor transformation in the activity patterns of simultaneously recorded neurons in the superior colliculus (SC) of rhesus monkeys performing a standard delayed saccade task. Neurons in its intermediate layers produce a burst of spikes both following the appearance of a visual (sensory) stimulus and preceding an eye movement command, but many also exhibit a sustained activity level during the intervening time ("delay period"). Each session's population activity was summarized in a low-dimensional framework and assessed on a scale of visual- to motor-like throughout the delay period using a novel measure we call the Visuomotor Proximity Index (VMPI). On average, population activity slowly evolved from a more visual- to a more motor-like pattern throughout the delay period, but microsaccade perturbations transiently deviated it to a visual-like pattern. A correlation was also found between the VMPI and single trial saccadic reaction time, even hundreds of milliseconds before the cue to initiate a movement. Therefore, we conclude that SC population activity contains a neural signature of the sensorimotor transformation process, systematically drifting toward a motor-like representation and intermittently reverting to a visual-like representation following a microsaccade.

Distinct context- and content-dependent population codes in superior colliculus during sensation and action

“…Since subspaces are formed by a weighted linear combination of activity across all recorded neurons, it logically follows that the population-level representations are distinct across epochs. This result is consistent with the notion that the trajectory formed by the neural population traverses through different regions of the state space as the trial transitions through stages of sensation, cognition, and action ( 18 , 25 , 46 , 49 ) and is confined to lower-dimensional “manifolds” ( 50 ). The present study therefore builds another case for the importance of characterizing neural activity through a measure that accounts for the relationship between neurons rather than by averaging the signal across all neurons, as is the standard in single-unit studies.…”

“…The top 4 latent factors accounted for at least 95% of the variance across neurons in the firing rate profiles for 18 out of 20 datasets ( 25 ). Unless noted otherwise, we chose to analyze only the top 3 latent dimensions because the third and higher dimensions do not show much modulation during either epoch.…”

“…Such efforts have yielded deeper insights into neural systems, although this technique has scarcely been applied to brain structures involved in producing visually guided eye movements; see refs. 22 , 25 , and 46 for notable exceptions.…”