Neural computation in a thalamocortical circuit
Sensory information enters cortex through the thalamus. Layer 4 in the cortex is the main thalamus-recipient layer. There is a great expansion (10-100 x) in the number of neurons from thalamus to cortex. However, the nature of the thalamocortical transformation is not understood. We are studying the thalamocortical circuit in mice during active tactile sensation. We think that active sensation is critical for revealing the computations performed by the thalamocortical circuit.
Mice move their whiskers to detect, localize and identify objects by touch. All sensory information comes in through sensors at the base of the whiskers. By tracking whiskers during behavior we can learn everything there is to know about the movements underlying somatosensation and related sensory input. Tactile information mediated by individual whiskers is spatially segregated all the way into the somatosensory cortex, where information from each whisker is processed in one barrel in Layer 4. Using well-established tricks of the trade we know where to look for the relevant activity to coding information relevant to a single whisker.
We know a great deal about the cortical circuits underlying whisker-based tactile sensation in the mouse. Information from each whisker is processed by a barreloid in the VPM thalamus. L4 stellate cells within one barrel are connected with each other in a recurrent manner. GABAergic Interneurons mediate feedforward and feedback inhibition within L4. L4 neurons, stellate cells and interneurons, receive long-range excitation only from VPM (to a first approximation). L4 is the simplest cortical circuit, the ‘hydrogen atom’ of the cortex.
We are working on the following questions:
What is the computation performed by the thalamocortical circuit? Spike trains recorded from defined neurons in thalamus and cortex suggest that this circuit removes signals related to self-movement during active sensation.
How is the computation implemented at the level of synaptic circuits? Whole-cell recordings made in behaving animals suggest that feed-forward inhibition in layer 4 rejects self-movement signals but transmits touch-related information.
We are testing our understanding of the thalamocortical circuit using detailed computational modeling.
Mapping and perturbing complete representations during behavior
We have been using volumetric calcium imaging in mice expressing GCaMP6s to sample activity in the majority of L2/3 barrel cortex neurons that code for different haptic variables (12,000 neurons per animal) during whisker-based object localization. An encoding model (in collaboration with Jeremy Freeman) relates activity to behavior. Approximately 17 % of neurons each encode touch and whisker movements. This representation is independent of learning.
We are using multi-photon single cell perturbations guided by volumetric two-photon calcium imaging to manipulate specific representations. Ablating a handful (<50) neurons encoding whisker touch reduces touch-related activity in the remaining, unablated members of the touch subnetwork. This argues for strong and specific recurrent excitation in this cortical subnetwork. Our experiments are poised to test the operational principles of cortical circuits during behavior.