Fig 1.
(a) Partial connectome from [12] containing 198 sensory neurons, interneurons, and motor neurons (most motor neurons that form neuromuscular junctions with muscle cells are excluded).
The 15 neurons that will be selected as core neurons are shown in purple. Additionally, the partial connectome contains 137 neurons that are directly connected to the core neurons either presynaptically or through gap junctions. (b) Core neurons selected for the premotor network model and the connections among them. We categorize most core neurons as either forward or reversal based on the connectome and previous experimental work.
Fig 2.
(a) Relative gap junction weights between core neurons and between signal neurons and core neurons (select signal neurons are shown).
Weights are taken directly from [12]. (b) Synaptic signed weights between core neurons and between signal neurons and core neurons derived from regression (select signal neurons are shown). Signal neurons are loosely sorted as forward-promoting or reversal-promoting. (c) Cartoon of principal structure between signal neurons, forward core neurons, and reversal core neurons.
Fig 3.
(a) Core neuron activity in the dataset 2023-01-23-15 [14] versus the simulation (101 neurons labeled). (b) The dominant cluster of neurons—forward or reversal—is a proxy for behavior (see Appendix). (c) Pairwise correlations between neurons. (d) Distribution of neural activity. (e) Dwell times in the forward and reversal states. Histograms are generated from the forward/reversal sequences in (b) using bin size 0.1 and excluding dwell times less than 0.05 min.
Fig 4.
Forward and reversal cluster dynamics using connected neurons as signals.
(a) Forward neuron activity from dataset 2023-01-23-15. (b) Simulated forward neuron activity using both gap junctions and synapses. (c) Simulated forward neuron activity using only gap junctions. (d) Simulated forward neuron activity using only synapses. (e) Reversal neuron activity from dataset 2023-01-23-15. (f) Simulated reversal neuron activity using both gap junctions and synapses. (g) Simulated reversal neuron activity using only gap junctions. (h) Simulated reversal neuron activity using only synapses.
Fig 5.
Simulations using signal neurons from dataset 2023-01-05-01 [14].
Promoter and suppressor neurons from dataset 2023-01-05-01 [14] are ranked by the strength of their promotion or suppression (left column). Top promotors/suppressors are then set to zero, and resulting activity strips are shown (middle column). Fractions of time spent in forward and reversal motion are summarized (right column). (a) Signal neurons that promote the forward neurons. Behavioral time series in data (row 1), with the top promoters set to zero (rows 2,3,4). Analogous results are presented in (b), (c) and (d) for signal neurons that suppress reversal, promote reversal, and suppress forward neurons respectively.
Fig 6.
(a) Core neurons simulation with no ablations, using dataset 2023-01-23-15 for the signal neurons.
(b) Core neurons simulation with RIB neurons ablated. (c) Simulation with AIB neurons ablated. (d) Simulation with AVD, AVA, and AVE neurons ablated. In all simulations, forward and reversal intervals of less than 2 seconds dividing a pause state were set to a pause state. Pause intervals of less than 5 seconds dividing a forward or reversal interval where set to the enclosing locomotive state (F or R).
Fig 7.
(a) Signal propagation map reproduced from Ref [39] data.
(b) Post-stimulus average displacement computed from the simulation displacement trajectories.
Fig 8.
(a) Event-triggered averages of neural activity aligned to the heat stimulus for some neurons with (D) excitatory or (E) inhibitory responses to the stimulus [14].
(b) Event-triggered averages of behavior of 32 animals in response to the heat stimulus [14]. (a,b: Reprinted from Cell, 186(19), Atanas, A. A. & Kim, J., et al., Brain-wide representations of behavior spanning multiple timescales and states in C. elegans, 4134-4151, Copyright (2023), with permission from Elsevier.) (c) Positive and negative perturbations applied to heat responsive neurons and resulting time series of heat responsive neurons with the perturbation applied. (d) Fraction reversing before and after perturbations are applied, averaged over 44 perturbation events.
Fig 9.
(a) Behavioral state time series from the model simulation using dataset 2023-01-09-22 [14], and three locomotion paths simulated from this behavioral time series.
The simulation procedure is described in the Methods. (b) Analogous results using dataset 2023-01-23-15 as input for the simulation.
Fig 10.
Polarities obtained from regression compared to polarities determined from previous work [16,34,38,42].
Synaptic weights are grouped and averaged over each neuron class (e.g. AVBL and AVBR and combined to form AVB). Ref [34] determines polarities for synapses that are not in the Ref [12] dataset which is used to establish the location of synapses for the model in this study.
Fig 11.
(a) Voltage clamp experiments from Ref [32].
Reprinted from Neuron, 20(4), Goodman M, Hall D, Avery L, and Lockery S, Active currents regulate sensitivity and dynamic range in C. elegans neurons, 763–772, Copyright (1998), with permission from Elsevier. Voltage clamp experiments from Ref [33]. (b) Intrinsic dynamics approximated from voltage clamp data in terms of voltage, . (c) Calcium imaging time series and distribution over premotor neurons. The pdf of the time series of the premotor neurons is used to set fixed points for the intrinsic dynamics, xfp = −0.8,0.1,1. (d) Negative of approximate intrinsic dynamics,
, in terms of the calcium imaging variable, x.
Fig 12.
(a) Average hybrid error as a function of .
The average error is a combination of the regression error and the derivative reconstruction error. Derivative and reconstruction errors are the L2 norm of the difference between the LHS and RHS of Eqs 2 and 4 respectively. (b) Example regression LHS and derivative reconstruction for AVAL and AVBL for .
Fig 13.
Dataset compared to simulations with different timescales and magnification factors.
Decreasing the timescale parameter makes the system respond quicker to stimuli. Because the system is receiving partial signals we also magnify
by 1.4x. Increasing the amount of stimulus to the premotor neurons increases their average activity levels.
Fig 14.
Time series of core neurons in whole-brain imaging data compared to the simulation using signal neuron input.
Behavior is measured as the dominant cluster (forward or reversal). , magnification on synaptic input is 1.4.
Fig 15.
Simulations using signal neurons from dataset 2023-01-23-15 [14].
Promoter and suppressor neurons from dataset 2023-01-23-15 are ranked by the strength of their promotion or suppression. (a) Signal neurons that promote the forward neurons. Behavioral time series with the top promoters set to zero. (b) Signal neurons that suppress the reversal neurons. Behavior with the top suppressors set to zero. (c) Signal neurons that promote reversal neurons and resulting behavior with the top reversal promoters set to zero. (d) Signal neurons that suppress the forward neurons and resulting behavior with the top forward suppressors set to zero.
Fig 16.
(a) Forward and reversal neuron average activity levels along with velocity.
(b) Difference of cluster averages and velocity. (c) Behavioral state classification of the difference variable (z(t)) and true velocity. (d) Correlations between the true velocity and measures taken from the neural activity—F(t), R(t), and F(t)–R(t).