top of page

Neural correlates of free behaviour

Predicting mouse behaviour from neural activity

Using a water-deprived self-initiated lever pull task we sought to determine how well we could predict behaviours such as lever pulls or random body movements from preceding neural activity.  We found that:

  • water-seeking behaviors as well as spontaneous movements can be decoded using linear and nonlinear methods by up to several seconds prior to movement.

  • in several mice, decoding improves with learning (e.g. over periods of several weeks) suggesting voluntary movements are not randomly generated but increasingly neuro-stereotyped over learning (consistent with some studies in humans).

  • spontaneous behavior preparation is widely distributed across the dorsal cortex with somatosensory and association cortex (e.g. retrosplenial cortex) being most activated during self initiated movements.

These findings show that preparation of high-value behaviours could be prepared in the mammalian cortex up to 10 seconds or more prior to movement and point the way to additional rodent studies as well as future studies in humans.


Clustering and modeling natural behaviour in gerbil families

Using an enriched home cage paradigm, we recorded the social activity of gerbil family groups over multiple weeks continuously. We found that:

  • animal identity can be corrected using convolutional neural network (CNN) classifiers trained on animal body markings.

  • group behaviour follows circadian rhythms

  • social interactions between pairs of animals are heterogenous and some behaviours (e.g. chasing or following) can be predicted above chance in the first 1 sec following a social interaction.

These findings show that spontaneous behaviour in rodents can be tracked and modeled to predict future behaviour.


Neural structure of spontaneous neural activity

Relating spontaneous single neuron activity to meso-scale (cortex-wide) neural networks

Understanding the basis of brain function requires knowledge of cortical operations over wide-spatial scales, but also within the context of single neurons. We carried out in vivo, wide-field calcium imaging and single neuron electrophysiology recordings to characterize single-neuron to cortex-wide functional relationships.

We found that:

  • cortical neurons are coactivated with brain-wide activity patterns suggesting they carry out computations that are similar to nearby neurons in well established single-synapse anatomical (consensus map) networks

  • subcortical neurons were coactivated during specific cycles of wide-scale cortical inhibition/excitation and were more dissimilar to neighbouring neurons.

These findings show that the relationship between the micro and meso-scale can be  established using single neuron spiking and mesoscale cortical maps.


Relating spontaneous single-neuron activity to multi-neuron activity in multiple species

Neurons are spontaneously active during awake, sleep and anesthetized states. Numerous studies have identified default-mode-networks at the root of quiescent states and during non-movement. In this study we sought to identify whether there is temporal or spiking order structure in spontaneous neural activity recorded in mice and other animals. We found that:

  • large-scale (local-field-potential) functional activity can be clustered into different groups by the cortical dynamics of each cluster

  • cortical and subcortical neurons exhibited a firing order that was broadly preserved over time.

These findings show that spontaneous neural activity contains a firing order structure which can be detected from large scale neural activity events and may represent a type of neural code.

bottom of page