Brain Computer Interface#
This dataset contains in vivo 2P calcium imaging recordings from layer 2/3 neurons in mouse primary motor cortex (M1) during an optical brain–computer interface (BCI) learning task. In this paradigm, mice control the position of a motorized reward port with the activity of a single “conditioned neuron” (CN) in layer 2/3 of M1. Specifically, at the start of the trial, the reward port is far away from the mouse and out of reach. The activity of CN controls the speed of the reward port toward the mouse. If the mouse can move the reward port within reach in 10 seconds from the trial start, a drop of water is earned as a reward (hit). If the mouse fails to bring the reward port close within 10 seconds, the reward port returns to the starting position (miss), and a new trial starts. Increasing the activity of the CN results in shorter and more rewarded trials. Mice typically learn to increase the activity of the CN within ~30 trials (~5 minutes), leading to higher reward rates. Activity changes following learning are remarkably sparse, with only a small fraction of neurons changing their activity as much as the CN.
In addition to neural and behavioral measurements, targeted 2P single-cell photostimulation was performed before and after the task to assess functional connectivity and its changes with learning.
Background#
Learning a new task or skill relies on synaptic plasticity to rewire neural circuits. The circuits modified by plasticity form novel sensory-to-motor associations making the learned behaviors feel almost automatic. The goal of the BCI dataset is to identify the “learning rules” that govern this plasticity. Learning rules are algorithms that determine which synapses change, when they change and by how much. Decades of work in brain slices and invertebrate preparations have identified rules governing plasticity in these simpler systems, but the relevance of these rules to in vivo plasticity in mammalian cortex remain unknown.
A detailed understanding of the rules governing plasticity during learning in mammalian circuits generally, and in cortical circuits in particular, is currently lacking. While it is well-established that cortical plasticity is critical for learning, understanding how the basic building blocks identified in brain slices operate within the dynamic, recurrent, and highly interconnected networks of the intact brain remains a major challenge. Addressing this gap requires approaches that can both precisely define the neural activity patterns linked to a behavioral outcome and measure the resulting changes in connectivity within the same local circuit during learning.
To address this challenge we developed two complimentary optical approaches in mouse primary motor cortex (M1). First, we use optical connection-mapping techniques that combine cellular-resolution 2P optogenetics with calcium imaging to measure the causal connectivity between each neuron in a recorded population, allowing us to track changes in connectivity over time. Second, we use optical brain–computer interface (BCI) learning tasks that explicitly define the relationship between the activity of imaged MC motor cortical neurons and behavioral outcomes, enabling precise control over which activity patterns are rewarded. Together, these approaches allow us to study the learning rules that govern plasticity in motor cortex during learning.
Technique#
We used resonant-scanning 2P calcium imaging to record, and 2P photostimulation to perturb, the activity of populations of neurons in layer 2/3 of primary motor cortex (M1). These experiments rely on neurons that co-express the calcium indicator GCaMP and the light-activated ion channel ChRmine. We used a variety of expression strategies in this dataset, including Cre-driver transgenic mouse lines and viral gene transfer, to target excitatory or inhibitory populations.
Because GCaMP and ChRmine are optimally excited by different wavelengths (920 nm for GCaMP; ~1080 nm for ChRmine), we used a dual-path microscope with separate light paths for imaging and photostimulation. Photostimulation targeted single neurons, one at a time, to measure their influence on the recorded population. In this dataset we imaged a single 800x400 microns large plane containing ~500 neurons, with GCaMP expression present in both excitatory and inhibitory neurons.
Questions to explore#
How much does the conditioned neuron increase its activity during the BCI task?
Do other neurons change their activity during BCI learning?
Using the photostimulation data:
Does connection strength depend on pairwise distance?
Does connection strength depend on pairwise correlation?