Figure 1. Neuronal activity is consistently different in different individuals.

(A) Calcium signals (${\displaystyle \mathrm{\Delta }F/F}$) recorded in one animal for ∼15 min by Kato et al. (2015). Each row represents a single neuron. The top 15 rows (above the red line) correspond to neurons unambiguously identified in all animals (shared neurons). (B) Trace of the AVA neuron colored by behavioral state as defined by Kato et al. (2015). (C) Neuronal activity of representative neurons plotted as a function of behavioral phase (Materials and methods) in a single behavior grouped by animal. Colored solid lines show mean activity for each animal. Shaded regions show 95% confidence intervals. Mean and confidence intervals are computed across multiple cycles of the same behavior in each animal. The dashed black line shows the mean across all cycles of the behavior in all animals. The cross-animal mean for ALA and RIML (Dorsal turn) is a good approximation of activity in each animal individually. In contrast, the cross-animal mean of RIML (Backward locomotion) does not represent any individual. Thus, activity of RIML is consistently different among individuals during backward locomotion. (D) Probabilities that neuronal activity from different individuals was drawn from the same distribution (Materials and methods) computed for each neuron in each locomotor behavior. Activity of most neurons differs consistently among individuals in at least one locomotor behavior. (E) Attempts to decode the onset of backwards locomotion are successful within each animal individually using either all neurons or the 15 shared neurons. This confirms that neuronal activation is stereotyped in each animal. Yet decoding fails across animals. Probability of decoding onset of backwards locomotion in one animal on the basis of neuronal activity averaged across other four animals is indistinguishable from chance. Thus, averaging neuronal activity across individuals disrupts behaviorally relevant information. Box plots show distribution of decoding rates bootstrapped across animals and bi-partitions of the data into training and validation datasets (Materials and methods).

Figure 1—figure supplement 1. Projection of calcium imaging data onto Temporal PCs.

Time derivative of neuronal activity in each animal is projected onto it’s corresponding first two principal components similar to Kato et al. (2015). The only preprocessing applied to the data was a Gaussian smoothing with sigma of one time step. Colors correspond to the fictive behaviors assigned by Kato et al. (2015). Note that the projection of each individual reveals clear loops in the dynamics of the nervous system. Kato et al. refer to this projection as Temporal PCA. Note that activity from each animal is projected onto its own coordinate system. Also see Figure 1—video 1.

Figure 1—video 1. Projection of calcium imaging data onto shared TPCAs.

Download video file ^{(2.5MB, mp4)}

DOI: 10.7554/eLife.46814.005

The shared set of 15 neurons recorded in each animal is used to construct shared Temporal PCs using the same method as in Kato et al. (2015). The video shows the data from all animals projected onto the first 3 components of the shared Temporal PCA. To produce these PCs the data from Figure 1—figure supplement 1 (without z-scoring) was concatenated together and subjected to PCA. This results in 15000 observations, which, accounting for a mean autocorrelation in the data of 150 time steps yields approximately 100 independent observations. Colors correspond to the fictive behaviors assigned by Kato et al. (2015). The structure apparent in the individual animal Temporal PCA projections has been lost (Figure 1—figure supplement 1).