Population temporal structure supplements the rate code during sensorimotor transformations

Abstract

Sensory-to-motor transformations are critically mediated by neurons in premotor brain networks whose evolving activities represent sensory, cognitive, and movement-related information. However, this multiplexing poses a conundrum: how does a decoder know precisely when to initiate a movement if such neurons are also active at times other than when a movement occurs? Extant models of movement generation that rely on firing rate, including rise-to-threshold, inhibitory gating, and dynamical switches at the population level, leave certain explanatory gaps unfilled. Here, we propose a novel hypothesis: movement is triggered not only by an increase in firing rate, but critically by a reliable temporal pattern in the population response. We show that in brain regions involved in orienting eye movements - the superior colliculus and the frontal eye fields - the temporal dynamics between neurons during visually-driven activity is different from that during pre-movement activation. Specifically, using a measure that captures the fidelity of the population code, we show that the temporal structure fluctuates in the visual response but remains stable during the pre-movement response, thus distinguishing incoming sensory input from motor output. This is an important attribute because SC and FEF “visuomovement” neurons project directly to the brainstem which houses the controller for gaze shifts, and any increase in the incoming drive is poised to trigger a gaze shift. We also demonstrate that a simple firing rate-based network with synaptic facilitation can distinguish between stable and fluctuating population codes, suggesting a putative mechanism for interpreting temporal structure. These findings offer an alternative perspective on the relationship between spatial attention and motor preparation by situating the correlates of movement initiation in the temporal features of activity in shared neural substrates. They also suggest a need to look beyond the instantaneous rate and consider the effects of short-term population history on neuronal communication and behaviour.SummarySensorimotor transformations are mediated by premotor brain networks where individual neurons represent sensory, cognitive, and movement-related information. Such multiplexing poses a conundrum – how does a decoder know precisely when to initiate a movement if its inputs are active at times when a movement is not desired (e.g., in response to sensory stimulation)? Here, we propose a novel hypothesis: movement is triggered not only by an increase in firing rate, but critically by a reliable temporal pattern in the population response. Laminar recordings in the macaque superior colliculus (SC), a midbrain hub of orienting control, and pseudo-population analyses in SC and cortical frontal eye fields (FEF) corroborated this hypothesis. We also used spatiotemporally patterned microstimulation to causally verify the importance of temporal structure. A spiking neuron model with dendritic integration was able to decode this temporal information. These findings offer an alternative perspective on movement generation and highlight the importance of short-term population history in neuronal communication and behaviour.

Instantaneous movement-unrelated midbrain activity modifies ongoing eye movements

“…The only difference in this case was that the overall behavioral impact on the movement amplitudes ( D ) was smaller than with the near visual bursts (consistent with Figure 3—figure supplement 1 ). Note that this could reflect a lower likelihood of proper temporal alignment of far peripheral visual spikes with the population motor bursts for microsaccades, according to the novel hypothesis of Jagadisan and Gandhi, 2019 . Indeed, in Figure 6 and Figure 6—figure supplement 2 , we found that with proper temporal alignment, the impacts of individual movement-unrelated spiking on microsaccade amplitudes were quantitatively similar for both near and far neurons.…”

“…This is achieved through multiplexing of functions, either at the level of individual brain areas or even at the level of individual neurons themselves. For example, in different motor modalities like arm ( Alexander and Crutcher, 1990 ; Shen and Alexander, 1997 ; Breveglieri et al, 2016 ) or eye ( Goldberg and Wurtz, 1972b ; Goldberg and Wurtz, 1972a ; Wurtz and Goldberg, 1972 ; Mohler and Wurtz, 1976 ; Bruce and Goldberg, 1985 ; Jagadisan and Gandhi, 2019 ) movements, a large fraction of the neurons contributing to the motor command are also intrinsically sensory in nature, hence being described as sensory-motor neurons. In this study, we aimed to investigate the implications of such sensory and motor multiplexing using vision and the oculomotor system as our model of choice.…”

“…Any other spiking, by the same or other neurons, before or after the eye movement is irrelevant. This causes a significant readout problem, since downstream neurons do not necessarily have the privilege of knowing which spikes should now count for a given eye movement implementation and which not ( Jagadisan and Gandhi, 2019 ).…”

Time-course of population activity along the dorsoventral extent of the superior colliculus during delayed saccade tasks

“…One way to reconcile these seemingly discrepant results, particularly with respect to the latter set of studies, is that the transformation may occur in dorsal, visually-dominant layers and that the population activity doesn’t transition to the motoric layers until after the animal receives permission to produce a movement. This notion suggests that the gaze-centered signal, although thought to be a movement signal, is in fact not interpreted as a movement command perhaps because the population activity exhibits state space dynamics that are not optimal for evoking a movement 45-47 .…”

“…It is also in line with the spike modulation times observed in burst generator neurons that participate in saccade generation 69,70 . It is generally stated the impact of reaching a threshold (equivalently, entering burst mode), either at individual or population activity, is to inhibit the brainstem omnipause neurons 67,71,72 , although other frameworks that depend on state space dynamical systems may facilitate direct communication between SC and pontine burst neurons 45-47 .…”