Figure 8. Integrated neuromechanical computational models test postulated roles of inhibition in proprioceptive (feedback) driven pattern generation, as compared with CPG (purely feedforward) models. A, Repeating units of the neuromuscular circuit were used to model C. elegans locomotion control. Single-unit schematic motor circuit of the ventral nerve cord used in our proprioceptive model of locomotion. During forward locomotion, VB and VD (DB and DD) motoneurons innervate ventral (dorsal) muscles and VD motoneurons innervate VB motoneurons. During backward locomotion, A-type motoneurons take the place of B-type motoneurons. Postulated mechanosensation (dash lines) encode a proprioceptive signal by integrating the extent of bending over its receptive field. Suprathreshold proprioceptive currents trigger the activation of VB and DB motoneurons during forward locomotion (or VA and DA motoneurons during backward locomotion). The reciprocal synaptic connections between VD and DD motoneurons (blue thin lines) are not considered in the model. Adapted from Cohen and Denham (2019; their Fig. 1B). B, Schematic traces of muscle activation in the proprioceptive control model (left) and feedforward control model (right), under different model perturbations. Positive activation denotes dorsal muscle activation. By construction, dorsal and ventral activation are in exact antiphase. Three hypotheses for the role of inhibition (cross-inhibition of the opposing muscles; disinhibition of the innervated muscles; and inhibitory reset of VB by VD) were tested in proprioceptive models of motor control; postulated roles of muscle cross-inhibition and disinhibition were tested in feedforward models (parameters and parameter sweeps in Extended Data Figs. 8-1, 8-2, respectively). When muscle inhibition is disrupted, the amplitude or waveform of muscle input is modified from the model wild type (black). Shown are schematics of perturbations corresponding to hypothesis 1 (reduced amplitude, red), hypothesis 2 (slower/smoother waveform, yellow), hypotheses 1 and 2 combined (orange), and hypothesis 3 (removal of neural inhibition, brown, proprioceptive model only). C, Relative frequency changes because of model perturbations. Under models of proprioceptive control, the undulation frequency is reduced by ∼10–35% relative to the wild type under a single perturbation to muscle inhibition, or up to ∼40% under the combined perturbation, suggesting that sufficient muscle activation amplitude and rapid activation onset are both required to sustain rapid undulation, during both crawling and swimming. Under proprioceptively driven locomotion, VB inhibition by VD serves to reset the neural rhythm during rapid undulation but has negligible effect in slow crawling-like motion. Bar plots with color scheme as in B depict the undulation frequency of model mutants normalized by the respective frequency of the model wild type. D, Under a model of proprioceptive control, inhibitory reset is necessary for rapid swimming in a liquid environment (bottom), but not for slow crawling on agar-like environments (top). In an agar-like environment, the elimination of VD-to-VB inhibition (right) has no effect. In a water-like environment, elimination of this neural inhibition eliminates the rhythmic pattern altogether. Red and blue shaded areas represent dorsoventral curvature; vertical axis is the length of animal from the head (H) on the top to the tail (T); color bar is ventral (blue) to dorsal (red) curvature -10–10 mm-1; along time (horizontal axis, common scale bar is 1 s).