New tools for recording and interpreting brain-wide activity in C. elegans

Recent studies have shown that behavioral information is richly distributed across the brain, but how individual neurons across the brain encode behavior is largely unknown. However, it is challenging to record/interpret high quality single neuron traces across the brain. Here, we describe new hardware/software and analytical tools to address this. We engineered a microscope to record brain-wide activity and behavior of freely-moving C. elegans, wrote new image analysis software, and developed a probabilistic encoder model to delineate how each neuron encodes specific behavioral features.
Extracting high quality neural traces from images is difficult due to the large amount of deformations as animals freely move. In our automatic extraction software, we construct registration graphs based on posture similarity, connecting timepoints that are similar enough to perform volumetric registration. Then using a custom clustering method, we link the ROIs over time to construct a neural time series for each neuron. The behavioral features are automatically extracted using custom neural networks and other computer vision methods. Overall, the new system gives us ~10x SNR improvement compared to the previous best system.
We next constructed a single-neuron encoder model based on our observations which demonstrates that C. elegans neurons can: encode behavior at multiple timescales (e.g. current vs recent); encode multiple behaviors (i.e. “mixed selectivity”); and display rectified encodings based on locomotion state. To be able to interpret the fitted parameters, we used probabilistic inference (a resample move sequential Monte Carlo inference algorithm validated with simulation based calibration), which gives a posterior distribution for each parameter instead of a single estimate. Overall, this allows us to determine the precise tuning of how each neuron encodes different behavioral features.
These new tools enable high-SNR brain-wide recordings from moving animals and interpretable modeling of how neurons across the brain encode behavior.