projects / fly fast weights / where step speed comes from

The speed of each step: it traces back to cells nobody has recorded

To add up its steps in the dark, a fly must sense how fast it is turning and walking from its own body. The neurons that deliver these two speeds to the brain’s navigation centre are known, but not where their speed signals come from. We traced their inputs back through the wiring diagram toward neurons that carry signals up from the body. Both speeds trace back to cell types that nobody has recorded, PS196_b for turning and FB3A for walking.

What is known

Each step has a direction and a length. The direction comes from the fly’s compass (finding 2). The length depends on how fast the fly is moving. The central complex, the brain region that navigates, needs two speeds: how fast the fly is turning, to keep the compass up to date, and how fast it is walking, to know how far each step goes.

In the light, the fly could judge both speeds from the motion of the visual scene. In the dark, it must get them from its own body: from sense organs in the legs, or from copies of the commands it sends to them. Signals from the legs reach the brain through ascending neurons. These run from the ventral nerve cord, the fly’s spinal cord, where the leg circuits sit, up into the brain.

The neurons that deliver each speed are known:

• Turning. GLNO neurons carry the fly’s turning speed to the PEN neurons that shift the compass. Silencing GLNO impairs the compass in the dark¹.
• Walking. PFNd neurons combine the compass heading with walking speed and pass each step on to hΔB, the step neurons of finding 3. Their speed signal comes from two inhibitory cell types, LNO2 and SpsP².

Hulse et al.¹ found that GLNO uses information from both vision and movement, but could not tell whether the movement signal comes from the legs or from motor commands. For walking speed, what feeds LNO2, SpsP and PFNd from the body had not been traced.

So our goal was to trace the inputs of these speed neurons back toward the body, and to find which cell types sit on the routes from ascending neurons to them.

What we found in the wiring

We listed every input to GLNO, LNO2, SpsP and PFNd across the whole brain, without choosing cell types in advance. We also listed every route from ascending neurons into the central complex, either direct or through one other cell type. Fig. 1 summarises the routes we found, and Table 1 gives their synapse counts.

For turning, GLNO’s largest input from outside the central complex is PS196_b, a cell type in a neighbouring region called the posterior slope. It supplies 19% of GLNO’s input (1,801 synapses). Hulse et al.¹ had already named PS196 as GLNO’s largest input, and proposed that GLNO inherits its turning signal from it. We add two things:

• Where PS196_b’s signal comes from. Its largest input from outside its own brain region is an ascending neuron, AN07B037_a (7%). None of its twelve largest inputs is a descending neuron, the kind that carries motor commands from the brain to the body. So the wiring points to a signal from the body rather than a copy of a motor command.
• What else PS196_b feeds. PS196_b also supplies 8% of the input to ExR2 (899 synapses). These dopamine neurons are active in proportion to turning speed and help the compass learn the visual scene³. So one cell type is placed to send turning speed both to the neurons that shift the compass and to the neurons that tell it when to learn.

For walking, PFNd’s largest input is LNO2, one of its two known speed inputs (20%). Its third-largest, after PFNd itself, is FB3A: four cells spanning the fan-shaped body that supply 12% of PFNd’s input (5,765 synapses). This is more than double the input from SpsP, PFNd’s other known speed source (5%). FB3A receives an ascending neuron directly, AN06B009 (3%), and gets more direct input from ascending neurons than any other input to PFNd. It also receives visual motion, through LAL143 (6%), and signals from Johnston’s organ, the antenna’s sense organ for sound, wind and gravity, through PS326 (4%). FB3A therefore pools several senses that could each report how fast the fly is walking.

FB3A is also a loop. PFNd supplies 34% of FB3A’s input, and FB3A sends PFNd 12% of its input back. So FB3A may adjust how strongly PFNd responds to speed, rather than supply the speed itself.

AN07B037_a also feeds PS196_a, a sister type of PS196_b, which supplies 3% of the input to LNO2, the walking-speed neuron. So a single ascending neuron stands upstream of both speed signals.

Table 1 | Synapse counts on the routes from the body to the speed neurons. A share is the fraction of all synapses onto the target type.

The wiring shows only who contacts whom, not what signals they carry. Ascending neurons can carry copies of motor activity as well as sensory signals, and Hulse et al.¹ saw hints of a motor contribution in GLNO. Three experiments would test the idea. Recording PS196_b as the fly walks in the dark should show it tracking turning speed before the compass shifts. Recording FB3A with and without vision, and with the antennae held still, should show which senses carry its speed signal. Silencing FB3A should weaken PFNd’s response to walking speed and impair the fly’s judgement of distance.