Neural Substrates for Visual Decision Making in the Visual Cortex

National Eye Institute

Investigators

Linked publications & trials

Abstract

During the past funding period we made progress in three projects: 1) What is the role of neuromodulator serotonin on visual processing in the early visual cortex? Serotonergic signaling in the brain mainly originates from neurons in the dorsal raphe nucleus, which have widespread projections throughout the brain. The serotonergic system is therefore ideally suited to modulate information processing in an orchestrated way in multiple brain regions, including in sensory areas depending on the behavioral context. Recent work in rodents (Seo et al. 2019, Science) found that while serotonin promoted increased motor activity under intense threat, it reduced motor activity in a neutral or moderately appetitive behavioral context, thought to promote quiet vigilance. However, the corresponding effects on sensory processing were not examined. The primary visual cortex receives substantial serotonergic projections, serotonin receptors are highly expressed throughout layers in the primary visual cortex, and particularly pronounced in the input layers. But in spite of this long held anatomical knowledge the physiological role of these projections on visual processing has been limited. This project builds on our recent work in the primary visual cortex, in which we identified that in non-aversive or moderately appetitive behavioral context, serotonin lowers the responses of visual neurons by lowering the gain of the responses (Seillier*, Lorenz* et al. 2017, J Neurosci). This modulation is consistent with promoting quiet vigilance by reducing the saliency of visual stimuli and suppressing an orienting response to those stimuli. In the current follow-up project we hypothesized that we might identify signatures of increased vigilance at the network level. We therefore examined how serotonin affects processing at the local network level in the primary visual cortex in a non-aversive behavioral context. We found that serotonin decreased the coherence between the spikes or an individual unit and the local network (quantified by the local field potential, LFP), decreased the power of the LFP and that the spiking activity was less predictable based on the LFP, suggesting decreased functional connectivity. These findings mirror previously reported signatures in the primary visual cortex of directing spatial attention to the neurons processing the visual information, consistent with a processing state for increased vigilance. These results are currently prepared for publication. 2) Exploring the flexibility of task-selective feedback in mid-level visual processing: Anatomical feedback connections are a ubiquitous property of the cerebral cortex but its role for computation and behavior is not understood. One of the most substantial differences between biological and state-of-the art artificial vision circuits concerns feedback: artificial systems typically do not incorporate feedback, whereas feedback is evident between every stage in the visual-processing hierarchy. This discrepancy underscores the large gap in our knowledge about the role of feedback in visual processing. A prediction common to theoretical accounts of feedback is that it targets the relevant neuronal subpopulations. For simple settings, this is supported by decades of empirical findings: If a localized region of space is relevant to a behavioral task, processing at that region can be selectively enhanced by feedback (spatial attention). If a given visual feature is relevant, processing of that feature can be selectively modulated throughout the visual field (feature selective attention). But the real world is more complex than such simple settings. Oftentimes behaviors depend on some combination of mechanisms, and whatever mechanisms enable behavior need to be flexibly and dynamically deployed, and in light of the enormous number of ethologically possible tasks and contexts, such selectivity could become anatomically costly. In this project we trained animals to perform a visual discrimination task that combines aspects of both feature and visual space, and we leverage a powerful behavioral analysis tool that enables us to be certain about what aspect of the task is used by the animal in making its decision. While the animals performed the task, we used multi-contact neurophysiological array recordings targeted at mid-level visual areas (V2, V3, V3a). We found that while the animals behavior was highly spatially selective, the task-relevant feedback was not. It suggests a common mechanism across tasks, independent of the spatial selectivity those tasks demand. Such common mechanism may reflect biological contraints to limit the anatomical cost of the feedback, while facilitating generalization across tasks. This work is currently being submitted for publication. 3) With the ability to collect increasingly high-dimensional behavioral and neural data, we need ways to characterize and quantify relevant behavioral features in an unbiased way, in order to relate these to neural signals. To address this challenge (addressed in Macke and Nienborg, 2019, Curr Opinion Neurobiology), summary statistics or hand-picked features are often used. But by doing so it is possible that information about the underlying structure of the data is missed. Dimensionality reduction and unsupervised learning methods have become increasingly popular as a means to comprehensively analyze behavior, such as for video motion tracking or acoustic vocalizations. We used a holistic, multi-view dimensionality-reduction approach based on t-SNE (t-Distributed Stochastic Neighbor Embedding) on a comprehensive dataset of eye measurements during fixation over extended training periods during which animals learned a complex visual task. This unbiased approach revealed characteristic eye-movements (directed microsaccades) predictive of behavioral performance. We are currently preparing these results for publication. 4) Organisms process sensory information in the context of their own moving bodies, an idea referred to as embodiment. This idea is important for developmental neuroscience, and increasingly plays a role in robotics and systems neuroscience. The mechanisms that support such embodiment are unknown, but a manifestation could be the observation in mice of brain-wide neuromodulation, including in the primary visual cortex, driven by task-irrelevant spontaneous body movements. In this project we tested this hypothesis in macaque monkeys, a primate model for human vision, by simultaneously recording visual cortex activity and facial and body movements. Activity in the visual cortex (V1, V2, V3/V3A) was associated with the animals own movements, but this modulation was largely explained by the impact of the movements on the retinal image. These results suggest that embodiment in primate vision may be realized by input provided by the eyes themselves.