Networks of descending neurons transform command-like signals into population-based behavioral control

Abstract

To transform intentions into actions, movement instructions must pass from the brain to downstream motor circuits through descending neurons (DNs). These include small sets of command-like neurons that are sufficient to drive behaviors—the circuit mechanisms for which remain unclear. Here, we show that command-like DNs inDrosophiladirectly recruit networks of additional DNs to orchestrate flexible behaviors. Specifically, we found that optogenetic activation of command-like DNs previously thought to drive behaviors alone in fact co-activate larger populations of DNs. Connectome analysis revealed that this functional recruitment can be explained by direct excitatory connections between command-like DNs and networks of interconnected DNs in the brain. The size of downstream DN networks is predictive of whether descending population recruitment is necessary to generate a complete behavior: DNs with many downstream descending partners require network recruitment to drive flexible behaviors, while neurons with fewer partners can alone drive stereotyped behaviors and simple movements. Finally, DN networks reside within behavior-specific clusters that inhibit one another. These results support a mechanism for command-like descending control whereby a continuum of stereotyped to flexible behaviors are generated through the recruitment of increasingly large DN networks which likely construct a complete behavior by combining multiple motor subroutines.

Fine-grained descending control of steering in walkingDrosophila

“…For example, they both receive input from many cells that arborize in motor-associated brain regions. Both DNs also receive substantial input from other DNs; this is not surprising, as many DNs communicate via specific networks of dendro-dendritic connections in the brain 63 . Finally, both DNa02 and DNg13 receive input from visual neurons and ascending neurons.…”

“…Bottom: rotational and forward velocity over time for these examples. (E) Pivots and swerves produce different changes in stride length (two-way ANOVA with legs and pivot/swerve as factors: pivot/swerve p=6.11×10 -31 ; leg identity p=4.96×10 -66 ; interaction: p=0.738, n = 389 (64) and 155 ( 53) bouts (flies), respectively). Post-hoc Tukey-Kramer tests show a significant difference between pivots and swerves for every leg (see Table S1).…”

“…The return and power stroke positions are plotted as positions in the anterior-posterior axis, and step direction is plotted as the difference from the mean when the fly was not turning (two-way ANOVA with legs and pivot/swerve as factors: return stroke: pivot/swerve p=3.68×10 -15 , interaction between pivot/swerve and leg identity: p=0.0198; power stroke: pivot/swerve p=1.75×10 -27 , interaction between pivot/swerve and leg identity: p=0.794; post-hoc Tukey-Kramer tests show a significant difference between pivots and swerves (Table S1); step direction: parametric Watson-Williams multi-sample test for equal means, comparing pivot/swerve for each leg, out-front: p = 2.22×10 -16 , out-mid: p=2.22×10 -16 ; out-hind: p=0.445; in-front: p=4.13×10 -12 ; in-mid: p=9.94×10 -5 ; in-hind: p = 0.140). n = 389 (64) and 155 (53) bouts (flies) for pivots and swerves, respectively. In ascending order of rotational velocity: out-front [0.0734, 6.67×10 -12 , 0, 1.47×10 -8 ]; out-mid [0.0409, 7.15×10 -18 , 0, 0]; out-hind [0.807, 1.68×10 -4 , 0.00437, 7.18×10 -10 ]; in-front [0.00167, 0, 0, 0]; in-mid [1.00, 0, 0, 0]; in-hind [0.627, 0, 0, 0]…”

Inhibitory control of locomotor statistics in walkingDrosophila

“…We next sought to develop a physiologically-inspired conceptual model of locomotor control that could account for the statistics we observed experimentally— in particular the shifts in statistics that occur between baseline walking and search behavior, and the changes in odor-evoked run length with baseline ground speed state. As our starting point, we considered that (1) multiple DNs contribute to both forward and angular velocity (Rayshubskiy et al 2020, Yang et al 2023, Braun et al 2023), (2) different units make different contributions to forward versus angular velocity (Rayshubskiy et al 2020, Yang et al 2023, Bresovec et al 2024, Aymanns et al, 2022), (3) bilateral activity correlates with forward velocity while activity differences between hemispheres correlate with angular velocity, both in some single neurons (Bidaye et al 2020, Yang et al 2023), and in population imaging (Bresovec et al 2024, Aymanns et al 2022), and (4) distinct sets of DNs promote stopping (Lee and Doe 2021, Sapkal et al 2023). Based on these considerations, we developed a simple model of locomotor control (Fig.…”

“…Therefore, each unit in our model receives an independent Gaussian noise input, μ i . Experiments in headless flies also suggest that DNs interact directly or indirectly within the brain to generate walking and turning behavior (Braun et al 2023). Thus, our units can interact with themselves and each other through an interaction matrix M (Fig 3C).…”