Abstract: The analysis of neural population activity in several brain cortices has consistently uncovered low-dimensional subspaces that capture a significant fraction of neural variability. These “neural manifolds” are spanned by specific patterns of correlated neural activity, whose activation are often called “latent dynamics”. I will discuss a model of brain function in which these latent dynamics, rather than the independent modulation of single neurons, drive behaviour. Animals readily execute learnt behaviours in a consistent manner. How does the brain achieve this stable control? We recorded from neural populations in premotor, primary motor, and somatosensory cortices for up to two years as monkeys performed the same reaching task. Remarkably, despite the unavoidable changes in recorded neurons, the population latent dynamics remain stable. Such stability allows reliable decoding of behavioural features for the entire timespan, while fixed “decoders” based on the recorded neural activity degrade substantially. Local field potentials (LFPs) arise from synaptic input currents to neural populations, and are thus thought to provide a glimpse on neural population function. Using simultaneous LFP and neural population recordings during the experiments described above, we found that the relationship between latent dynamics and LFPs is largely frequency-dependent and varies significantly across areas. However, for any given area, this relationship is stable across different aspects of behaviour, such as movement planning and movement execution. Interestingly, these results could not be predicted based on the relationship between the LFPs and single neuron activity. A neural population view of how the brain works allows revealing stable patterns of neural activity that mediate behaviour. Given that neural manifolds are found throughout the brain, from prefrontal to visual cortex and even hippocampus, similar principles may apply to the study of non-sensorimotor processes as well.
Dr. Devika Narain