Understanding feedforward and feedback signaling between neuronal populations

Summary Most perceptual, cognitive, and motor functions rely on neuronal activity distributed across multiple networks, often located in different brain areas. In many systems, including the visual system, signaling between areas is bidirectional: lower areas communicate with higher ones via feedforward connections, and higher areas signal to lower areas via feedback. Feedforward pathways are thought to underlie the increasingly sophisticated receptive fields as one ascends the visual hierarchy. The role of feedback signaling in visual processing, in contrast, is poorly understood. Feedback has been proposed to underlie a diverse set of interrelated functions including providing contextual information, predictions, learning signals, and attentional and expectation signals. Testing these proposals has proven experimentally difficult: it requires assessing not only what signals are sent from higher to lower cortex but also how feedback signals interact with ongoing population activity in the target area to influence the feedforward signals relayed back to higher areas. In this project we aim to understand how inter-areal feedforward and feedback signaling work together to underlie visual function. We will do so by determining the signals conveyed by neuronal population spiking responses—which underlie cortical representation—in the feedforward and feedback direction. We will use high yield multi-area neuronal recordings; a new conceptual framework of how inter-areal signaling is implemented; and new analytical tools that will allow us to disentangle the influence of feedforward, recurrent, and feedback signaling, even when these are concurrently active. Our working hypothesis is feedforward-feedback loops implement a form of predictive coding, a concept that to date has been tested primarily using single neuron responses rather than the hierarchical flow of population signals. In Aim 1, we will test this hypothesis by analyzing simultaneously recorded neuronal population responses evoked in macaque V1/V2 and V1/V4, by a broad but targeted set of visual stimuli. In Aim 2, we will develop a hierarchical spiking network model of predictive coding. The model will allow us to relate existing theoretical constructs to the responses measured in our experiments and to understand how the pattern of inter-areal signaling observed in data contributes to (or constrains) predictive coding computation. In Aim 3, we will test how active predictions, made by animals performing a perceptual decision-making task, are relayed between cortical areas and shape visual cortical representations. Our ambitious goals will be accomplished by pooling the complementary expertise of three PIs, building on an established and successful collaboration. Successful completion of this project will shift the study of inter- network signaling from single neuron to population-based interactions and will test a central concept in neuroscience—hierarchical predictive coding. We expect the understanding we gain, and the analytic and conceptual tools we develop, will be broadly applicable. Because inter-areal signaling is dysregulated in several disorders, our findings may also lay the groundwork for developing treatments in future work.