Background
Figure 1. In everyday life situations (e.g. finding a cab) our brain relies on networks composed of subcortical and cortical structures for processing sensory cues. Cortex and subcortex interact through feedforward (black) and feedback (red) projections. Cortical feedback helps focus on relevant information (e.g the color yellow in this task).
Our brain perceives the world dynamically, zooming in into relevant stimuli and fading out irrelevant ones, generating skewed moment-to-moment snapshots of reality. For example, if you are in New York City in need of a cab, you will be susceptible to yellow objects, the colour of cabs in NYC (Fig. 1). This “adaptive sensing” of the world is at the essence of the flexibility that has allowed mammals to flourish in varying environmental conditions. Adaptive sensing depends on the interaction between incoming sensory input and the feedback that can modulate it, in other words, it depends on a processing loop. Anatomists have known of feedback projections for decades. These projections often arise in the cortex and innervate numerous subcortical nuclei at various levels of the sensory processing, creating cortico-subcortical loops. Importantly, feedback projections can exceed in number their feedforward counterparts. And yet, we continue to view sensory processing as a feedforward transformation of information. Feedforward networks, however, fail to capture the high proficiency of mammalian brains to flexibly interpret a complex world on a moment-to-moment basis, according to current needs and previous experience. We must venture away from the streetlight to incorporate the role of feedback projections into our understanding of adaptive sensing. While theories on the function of feedback projections have been put forward, their investigation has been limited mostly because existing tools were not sufficiently refined to selectively target these projections. Thanks to recent technical advances, this is now possible. Indeed, we can now use genetic tools to unprecedentedly target and even manipulate both the cortical neurons that send feedback, as well as the subcortical neurons that receive it. Combined with recent advancements in high density electrophysiology and brain-wide imaging, these techniques allow us to sample and manipulate the activity of cortical and subcortical neurons even in alert behaving animals during adaptive sensing. The momentum generated by these technical developments is paralleled at the theoretical level, with the incorporation of feedback projections into current artificial network models.
The core of this SPP is to provide a deeper understanding of the role of cortico-subcortical loops in adaptive sensing, across modalities and in behaving animals.
The reasons for the emerging interest in cortico-subcortical loops are manifold. i) It is becoming increasingly clear that the brain is not an input-output machine. To understand the brain we need to include feedback projections. ii) There is an urge to rethink the role of subcortical structures that had been classically regarded as “relay” stations. (iii) The question is emerging simultaneously across all sensory modalities. Since cortical feedback is a common feature across modalities, a comparison will help derive general principles of feedback on sensory processing. (iv) Recent technological advances allow now to target individual neurons by their projection, to sample en masse simultaneous single cell activity for many days/weeks, and to perturb brain activity with cell-type specificity in cortico-subcortical networks, enabling us to characterize specific functional loops in a manner that was not possible before. And (v), the automated, quantitative analysis of flexible behavior, indicative of adaptive sensing, is now possible thanks to the implementation of deep learning algorithms in behavioural measurements.