Volitional modulation of optically recorded calcium signals during neuroprosthetic learning

Abstract

Brain-machine interfaces are not only promising for neurological applications, but also powerful for investigating neuronal ensemble dynamics during learning. We trained mice to operantly control an auditory cursor using spike-related calcium signals recorded with two-photon imaging in motor and somatosensory cortex. Mice rapidly learned to modulate activity in layer 2/3 neurons, evident both across and within sessions. Learning was accompanied by modifications of firing correlations in spatially localized networks at fine scales.

Figure 1. Mice learn to intentionally modulate calcium dynamics

a. Example imaging field (left) and recordings from cells in E1 and E2 (top right). Red stars indicate hits. Bottom right, mean ensemble fluorescence around hits. b. Performance over 8 days of training. Mean performance is shown in black, individual animals in gray. Error bars denote s.e.m. Shaded region denotes chance performance. c. Performance rapidly dropped compared to normal task levels (T) during the contingency degradation (CD). Performance returned to previous levels during reinstatement (R). d. E1 ΔF/F increases during the task and decreases during CD. Likewise, target hits (red) increase in frequency over training, and decrease during CD. e. At the beginning of day CR2, the animal initially performs as if the previous day's transform algorithm were still in use, but quickly learns the new transform.

Figure 2. Local network reorganization accompanies neuroprosthetic learning

a. Mean fluorescence increases in E1 cells over the course of a session. Inset: Mean target-locked fluorescence. b. E1 cells with low baseline activity increase their activity more during the task than cells with high baseline activity. E2 cells suppress their activity evenly. Note logarithmic scale. c. Activity in E1, E2 and indirect cells time-locked to large events in E1 cells. d. Activity in E1, E2 and indirect cells time-locked to large events in E2 cells. e. Correlations increase between output cells (cyan) during the session, with no similar increase in correlations between indirect cells (black). f. Indirect cells near output cells have more task-related activity than those far from output cells. Circles are individual cells, bars indicate s.e.m. g. Early in a session (solid lines), target-related modulations in indirect cells decrease with distance from E1 cells (blue) and increase with distance from E2 cells (green). Later in the session (dashed lines) there are no significant modulations in indirect cells, regardless of distance from output cells.