How do neurons in neocortex work in concert to encode the natural world, construct a representation of its behaviorally relevant content, and coordinate behavior based on those representations? This is a complex problem that we are addressing in multiple ways, with an interdisciplinary and comparative approach.
Mapping circuit structure from dynamics
Individual synaptic connections are weak, and patterns of spiking are complex and variable, making the mapping between structure and functional dynamics far from straightforward. However, we have found that it is possible to identify a small set of synapses that are actively involved in the recruitment of post-synaptic neurons – i.e., those synaptic inputs that occur at just the right time to drive the post-synaptic neuron to threshold. These ‘recruitment’ synapses (see Chambers and MacLean (2016), for our full network taxonomy) are a very small subset of the synaptic connections that a neuron receives but are key to the propagation of spiking. Specifically, we found that recruitment synapses are directly linked to particular higher-order motifs in the functional wiring diagrams. To make this observation we accounted for the non-linear integrative properties of neurons, using a combination of spiking neuronal network models and experimental measurements. In this way, we have gained insight into how spiking is propagated through a network by the coordination of synaptic input and firing by multiple neurons in specific triplet motifs. Like water finding the steepest channels as it flows downhill, spiking activity follows the path of least resistance and propagates through triplet motifs of connectivity and suggests this is a canonical motif to propagate information through the cortical microcircuit.
A circuit representation of visual input
A powerful aspect of summarizing microcircuit activity as a functional network is that the reliability of neuronal activity can be captured and summarized as the weight of a connection in that network. Thus, the network summarizes both the spatiotemporal pattern of activity as well as the variability of the patterned activity. By imaging microcircuit responses to visual input during locomotion, we have found that trial-averaged tuning properties of neurons explain only a small fraction of the single-trial activity of neurons. By summarizing the dynamics as a functional network, we are able to use the neighbors of a neuron, including tuned and untuned neurons, to predict both individual neuron activity on a single-trial basis and also recapitulate average tuning properties of tuned neurons. Perception and behavior take place in real time, after all, so it is necessary that any encoding model of stimulus representations in cortex encompass single trial responses. Moreover we identified a specific triplet motif that improved predictions of single trial responses. Only through study of the functional network were we able to find this emergent signature of informative correlations. Thus, predicting neuronal activity at any moment requires both knowledge of external variable(s) (the visual stimulus in this case) and knowledge of the functional interrelationships between neurons. Having developed a better encoding model that successfully predicts spiking in neurons on a trial-by-trial basis (rather than just averaged activity), our goal is to build single trial decoders, again employing a network science framework to temporal graphs.
A circuit representation of motor output
We also study motor cortex (M1) in order to link network analysis of dynamics to output in the form of motor behaviors. Our goal is to reveal the design principles of the M1 code by quantitatively characterizing patterned activity at the level of the cortical microcircuit during both ethologically relevant behaviors and motor skill learning. To record M1 microcircuit activity in unrestrained animals, we use an ultra-light head-mountable intracortical microscope. This allows us to image the activity of up to two hundred M1 pyramidal neurons across all cortical layers during a learned reach-to-grasp task as well as innate, naturalistic behaviors including grooming, locomotion and climbing. To inform and constrain theories that specify the encoding of movement in single-trial M1 microcircuit activity, we are generating: (a) a detailed statistical description of single-trial microcircuit dynamics in forelimb M1 during innate behaviors, (b) measures of similarity and dissimilarity between functional networks associated with each of the specified behaviors, and (c) a quantitative evaluation of the effects of motor learning (acquisition of a reach and grasp motor skill) on M1 microcircuit spatiotemporal activity patterns, incorporating laminar location of the recorded neurons and corresponding forelimb kinematics.
We also design spiking neuronal network models capable of reproducing realistic activity to test our circuit-centric hypotheses and generate new ones. Simulation is a valuable complement to experiments in living tissue because it reveals the consequence of design assumptions and provides omniscient access to simulated synaptic connectivity patterns. By moving back and forth rapidly between experiment and theory, our group is working to achieve a better understanding of the fundamental principles that govern neocortical computation.