Descending networks transform command signals into population motor control

To convert 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 behaviours1—the circuit mechanisms for which remain unclear. Here we show that command-like DNs in Drosophila directly recruit networks of additional DNs to orchestrate behaviours that require the active control of numerous body parts. Specifically, we found that command-like DNs previously thought to drive behaviours alone2–4 in fact co-activate larger populations of DNs. Connectome analyses and experimental manipulations revealed that this functional recruitment can be explained by direct excitatory connections between command-like DNs and networks of interconnected DNs in the brain. Descending population recruitment is necessary for behavioural control: DNs with many downstream descending partners require network co-activation to drive complete behaviours and drive only simple stereotyped movements in their absence. These DN networks reside within behaviour-specific clusters that inhibit one another. These results support a mechanism for command-like descending control in which behaviours are generated through the recruitment of increasingly large DN networks that compose behaviours by combining multiple motor subroutines.


Supplementary Discussion
Alternative mechanisms for DN recruitment In this work we highlight that direct synaptic connections between DNs are sufficient to explain the recruitment of DN populations by command-like DNs.However, there are other mechanisms may contribute in parallel.
The first additional mechanism is via ascending feedback to the brain.Interestingly, the GNG receives a large number of inputs from ascending neurons (ANs) that project from the VNC to the brain [9,10].Among these are a set of ANs which play a role in the decision between locomotion and feeding [11].Connections from ANs may thus allow DNs to integrate information from the VNC to regulate switching between actions.DN-DN recruitment might also potentially arise indirectly via a DN-AN-DN 'zigzag' motif that has previously been observed in low numbers in the brain connectome of Drosophila larvae [12] (i.e., a DN targets an AN in the VNC which then projects back to the brain targeting a different DN).In the adult, ANs encompass 17% of all DN post-synaptic partners in the VNC [13] and 10% of all DN pre-synaptic partners in the brain.Although our experiments in headless animals cannot test the contribution of ANs (decapitation eliminates both DN-DN and putative DN-AN-DN connections), our VNC resection experiments show that ANs are not required for DN recruitment.These connections are far less numerous than direct DN-DN connectivity in the larval brain.Efforts aiming to bridge the existing brain and VNC connectomes [8,14,15] and to generate complete adult nervous system connectomes will further reveal the relative contribution of ANs to DN recruitment.
DN recruitment might also arise even more indirectly via sensory feedback during a change in behavioral state: Active DNs may drive a new behavior, resulting in limb sensory feedback that in turn may be transmitted via ANs to influence other DNs.We would expect to see such sensory-induced DN activation in spontaneous behavior but we instead observe that, in general, fewer DNs are strongly activate during spontaneous behavior compared with during optogenetically elicited behavior.This argues that strong GNG-DN activation is a specific response to optogenetic stimulation of commandlike DNs.In particular, we found that the 10 most active neurons during DNp09 stimulation are not active during spontaneous forward walking.These results, as well as those from our VNC resection experiments blocking ascending feedback to the brain, suggest that DN recruitment likely does not result from sensory feedback arising during optogenetically-induced changes in behavioral state.
We find that DNp09 stimulation elicits additional DN activity beyond what is normally seen during spontaneous walking.This suggests a distinction between DN populations becoming active during spontaneously generated, sensory-induced, and optogenetically activated walking.We note that we previously observed that a large fraction of DNs in the CRG (rather than GNG-DNs recorded in this study) become active during spontaneous and odor-evoked forward walking [16].Thus, spontaneously generated and sensory-induced walking may principally be driven by CRG-DNs.For example, DNp09 are thought to mediate courtship-related forward walking [3].The possibility that DNp09 is active only during specific courtship-related behavioral contexts and inactive during spontaneous walking is supported by recent electrophysiological evidence [17].Thus, it appears that forward walking can be controlled by distinct DNs depending on the context.

The various roles of DNs in motor control
In our work we hypothesized that DNs might each drive 'motor primitives' which, when composed, give rise to complete, coordinated behaviors.Beyond a direct role in motor control other DNs may be modulatory, controlling behavioral vigor or persistence.Other DNs may be inhibitory, potentially playing an important role in action selection.Finally, other DNs may have a role in 'gating' behaviors by increasing the excitability of downstream motor circuits.
For DNs driving specific motor actions, we propose a framework in which DNs coordinate complex behaviors by recruiting additional DNs driving simpler motor primitives.Further evidence for such a model could come from activating command-like DNs while silencing downstream DNs.For example, DNp09 is connected to and requires the actions of a large number of DNs to drive movements of the six legs for goal-directed walking during courtship [3].Both DNp09 (for forward walking) and MDN (for backward walking) synapse upon DNa01 and DNa02, two DNs involved in turning [18,19].Thus, one might silence DNa01 or DNa02 to test the prediction that they control specific turning kinematics during either asymmetric forward or backward walking.Notably, we found that DNa01 and DNa02 also synapse onto other DNs.However, their activation alone (i.e., in headless animals) is not sufficient to drive turning.Thus, DNa01 and DNa02 may also sit atop the hierarchy of DN recruitment, recruiting other DNs which control individual leg degrees of freedom.As an example of a DN controlling similarly few degrees of freedom, we found that stimulating DNg14 drives lowering of the abdomen.In line with it being at lowest level in a DN hierarchy, DNg14 does not have any downstream DNs, but receive inputs from twelve upstream DNs.

Table 1 :
Transgenic fly lines generated in this study.

Table 2 :
SpGAL4 fly lines used in this study.

Table 6
: Exact p-values for statistical tests in tibia-tarsus leg amputation animal experiments (Extended Data Fig.10).