Understanding the neuronal substrates of internal models of the world

Growing evidence suggests that the brain predicts its inputs and builds models of the world such as to best match its predictions to reality. While the idea of mental simulation dates back as early as Plato, and is echoed by modern theories of cortical function, little is known about how the brain builds internal models, and predicts upcoming inputs. During behavior, movements are good predictors of sensory inputs: where we look determines what we see. We leverage olfactory closed-loop behaviors in mice and large scale functional imaging and electrode recordings to identify neuronal circuits that mediate egocentric sensorimotor predictions and errors of sensory inputs given specific motor actions. We aim to understand:
Where and how are sensorimotor predictions represented?
How are predictions compared to sensory inputs to trigger prediction errors?
How are internal models updated, given persistent errors?
 
 

odorant_features_pic_take4

3D_optical_monitoring and manipulation_pic

What features of odorants and their neuronal representations are relevant for olfactory perception?

The relationship between perceived odor quality, the underlying spatial-temporal patterns of activity in the brain and the physical-chemical space of odors remains elusive. The realization that color perception is based on three types of cone photoreceptors enabled the invention of cameras and displays that faithfully reproduce any natural stimulus by mixing a basis set of just three lights. In the case of smell, we lack any comparable conceptual understanding. To a large degree, we still do not understand what properties of odorants lead to particular percepts, and how these properties are represented in the neuronal activity. We aim to identify the basis set of olfactory perception. 
To determine what features in glomerular activity patterns are captured by the brain, we combine odor stimulation with patterned illumination methods and recordings from downstream circuits. Employing artificial optogenetic stimuli informs on detection thresholds and resolution limits of the system. Combining odors with precise manipulations, we test our predictions during odor guided behaviors.

Structured long-range connectivity between the olfactory bulb and cortex: parallel olfactory processing streams

 
A large body of work across several decades has uncovered no apparent structure in the mammalian olfactory bulb projections to the olfactory cortex and the intra-piriform connectivity. Because piriform circuits were assumed to be devoid of any innate structure, models of olfactory learning have assumed unstructured random connectivity and were proposed to rely entirely on plasticity to construct meaningful representations.

We recently investigated the structure in the olfactory cortex circuit using novel high-throughput DNA barcode sequencing-based single-cell projection mapping (BARseq and MAPseq). We found that individual mitral cells in the olfactory bulb project in a graded manner to different locations along the anterior-posterior axis of the piriform cortex, and furthermore these piriform cortex loci project to extra-piriform brain regions that the same mitral cells send collaterals to, completing triadic circuit motifs. This triadic organization is replicated at different positions within the piriform cortex along its anterior-posterior axis for functionally distinct brain regions, such as the anterior olfactory nucleus (AON), cortical amygdala (CoA) and lateral entorhinal cortex (lENT) via specific input-output projection gradients.

Our results support the view that, akin to other sensory modalities, olfactory information leaving the olfactory bulb is segregated into parallel streams that support different computations potentially related to perception, valence and action. In ongoing projects, we test the hypothesis that the olfactory cortex architecture is structured, and thus need not rely on algorithms that assume random connectivity, the relationship to odorant receptors identity and long-range inter-area connectivity.

 

Parallel long-range feedback loops for olfactory processing

Cross-talk between feedforward and feedback signals across different brain areas enables computations ranging from extracting fine sensory features in complex environments to generating predictions on incurring stimuli, and serving as substrates for planning and execution of motor actions. Despite evidence for massive top-down projections, the specificity and logic of interplay between feedforward and feedback neural signals between early sensory processing areas and the cortex remains poorly understood. This is due both to a conceptual bias favoring the prevalence of feedforward hierarchical processing, as well as to technical limitations in assessing the effects of cortical feedback. We investigate the function of such long-range circuits in the mammalian olfactory system.

Recently, we identified two parallel long-range feedback functional loops involving the tufted and mitral cells and their major cortical targets: the anterior olfactory nucleus versus piriform cortex. Cortical feedback regulates specifically the activity of its dominant input cell type and implements different computations. For example, mitral cell representations can be re-shaped by piriform cortex feedback, potentially based on prior experiences, context, and reward contingency. Tufted cells outperform mitral cells in representing odor identity, intensity and timing. Our results suggest that the tufted cell ↔ anterior olfactory nucleus pathway rather than the canonical mitral cell ↔ piriform cortex pathway is the major player in representing sensory olfactory information pertaining to stimulus identity and intensity. In ongoing experiments, we are investigating whether indeed these two olfactory feedforward-feedback loops perform different functions during behavior, enabling robust olfactory perception and behavioral flexibility.