Three-neuron circuit may set the fly’s walking rhythm, connectome simulations find

In 636 of 1,024 pruning runs on the male fly nerve-cord connectome, the rhythm survived down to the same two excitatory and one inhibitory interneuron, yet those three cells have not been switched off in a living fly.

Peer reviewed · Computational model of four connectomes · Two predictions tested by optogenetics in 7 and 10 headless flies

Simulations built on the fruit fly’s wiring diagram have picked out a three-neuron loop that may set the rhythm of walking. Sarah Pugliese, John Tuthill and Bingni Brunton of the University of Washington and colleagues report it in Nature. A central pattern generator, a circuit that turns a steady command into a repeating output, has been assumed to drive walking since the 1910s, but its cells had never been named in any animal. The catch: the circuit comes from a model, and its three cells are untested in living flies.

Evidence at a glance

Is this a big deal?

For neuroscience, a solid step rather than a settled answer. A walking pattern generator with named cells and synapse counts is new, and the method found something a wiring map alone could not. But the circuit is a prediction until someone silences E1 and E2 in a fly and the stepping stops.

What scientists found

A fly’s six legs are run by the ventral nerve cord, the insect counterpart of the spinal cord. Physiologists have held since Thomas Graham Brown’s cat experiments of 1914 that such cords contain central pattern generators (CPGs), because cats, crayfish and locusts kept producing motor rhythms with their sensory nerves cut. Which cells did it stayed unknown.

In the simulations, the core is small. With the walking command neuron DNg100 switched on in a model of the male nerve cord (the MANC dataset), the smallest network that still made leg motor neurons oscillate was two excitatory interneurons, E1 and E2, and one inhibitory, I1.

Across four connectomes from male and female flies, E1 and E2 appeared in 3,521 of 4,096 minimal circuits, with either I1 or a second inhibitory cell, I2. Silencing E1 or E2 in the full simulated network killed the rhythm in every dataset; silencing I1 or I2 alone did not.

How they found it

The raw material was four connectomes, electron-microscope maps of every neuron and synapse in a fly’s nerve cord or whole nervous system, published between 2024 and 2026. A map shows which cells connect, not what they do, so the team wrote an equation for each cell’s activity, with synapse counts as connection strengths and a sign from each neuron’s predicted neurotransmitter. Four properties per neuron, such as gain and threshold, are in no connectome; the team drew them at random from measured ranges and reran each simulation 128 to 1,024 times, so any result that survived came from the wiring. No simulated cell could burst on its own.

They switched on each of 933 excitatory descending neurons, the cells carrying commands from brain to cord, and scored how rhythmic the leg motor neurons became. Only 29 scored above 0.5 on a 0-to-1 scale; DNg100, known from Salil Bidaye’s lab as a walking command neuron, was strongest. Then came pruning: silence one interneuron at random, keep it off if the rhythm survived, restore it if not, repeat. In the male nerve cord, 62% of 1,024 runs stopped at the same three cells.

The mechanism works like an old electromechanical buzzer: a steady current pulls an arm that breaks its own contact, releases, and repeats. DNg100 drives E1, E1 excites E2, E2 recruits I1, and I1 switches both excitatory cells off; because I1 gets no command input, its inhibition fades and E1 restarts. The analogy’s limit: the circuit needs two excitatory cells, since fly neurons almost never synapse onto themselves. The linearised loop runs at about 14 Hz, inside the 7 to 15 Hz at which flies step.

Two predictions were tested in headless flies, which walk when DNg100 is activated by light. Brighter light raised walking speed and stepping frequency in 7 flies. Activating DNb08, a flagged neuron type with no known function, made 10 flies wave their legs rhythmically, like insects searching with legs off the ground.

Why it matters to you

Nothing changes for your body this year. The payoff is a way of working: predicting what a circuit does from its wiring, then checking in an animal. Engineers building legged machines already borrow pattern-generator ideas, as our explainer on embodied AI describes.

What the evidence does not show

It does not show that E1, E2 and I1 generate walking in a living fly. No genetic line targets these cells, so nobody has silenced or recorded them; the authors list that as the chief test to come.

The model also walks badly. Some muscles that should fire during stepping, including the main tibia flexors, stayed silent or arrhythmic. Driving both DNg100 neurons in the full 23,532-neuron male connectome made all six legs oscillate, but with no consistent left-right timing and no tripod gait. DNb08 failed in one of the four datasets. And the connectomes omit ion channels, receptors, neuromodulators and gap junctions.

How this fits what we already knew

Named CPG cells exist for other rhythms, such as the crab stomach, the leech heartbeat and the breathing centre in mammals, and many rely on neurons that burst on their own. The proposed fly circuit does not, and it lacks the mutual inhibition between two half-centres that textbook walking models assume. Each DNg100 neuron contacts more than 1,400 cord cells, so a feedforward trace of the map would have found E1 but missed E2 and I1, the authors say.

The two-excitatory, one-inhibitory motif occurs 21,544 times in the front-leg network, against 307 expected by chance; E1-E2-I1 stands out for its link strength and the highest intrinsic frequency of any copy.

Independent perspective

SciQuest found no independent comment on the Nature version as of 31 October 2026, 12:48 UTC, and has not yet interviewed anyone. When the work was a preprint, Rune Berg of the University of Copenhagen told The Transmitter it breaks with the half-centre model: “There’s no reciprocal inhibition in [the putative fly CPG] network, which is a bit surprising.” Graziana Gatto of University Hospital Cologne called knowing the neurons “a big step forward”.

A separate preprint from Bidaye’s group at the Max Planck Florida Institute for Neuroscience traced a five-neuron motif from behaviour, with one cell overlapping the loop, The Transmitter reported. It is unreviewed, and Bidaye is thanked in the paper for sharing fly stocks, so it is not independent confirmation.

How other headlines framed it

Only The Transmitter has reported the work, in May 2026 as a preprint, under the line that a long-sought walking circuit had been found, with the Bidaye preprint as a second study. The UW Neurobiology and Biophysics page said researchers had located the pattern generator. The paper’s own abstract is more careful: a “putative CPG circuit”. SciQuest keeps that word.

What happens next

The decisive experiment is to build driver lines for E1 and E2 and silence them during DNg100-driven walking; recordings would test whether the cells lack bursting properties. The authors also plan to hunt for the neurons that couple left and right legs and to connect the simulation to a physical model of the fly body. Their code is public on GitHub.

In one sentence

Simulations of four fly connectomes keep converging on a three-neuron loop as the walking rhythm generator, and two of its predictions held in headless flies, but the loop itself awaits a direct test.

Read next

• What is embodied AI? Why machines that move need their bodies, and the control ideas they borrow from biology.
• Octopus and squid DNA forms flexible loops linked to brain development: another nervous system read from its molecular blueprint.
• The fastest robot in the world: what legged machines can do today.

Sources and disclosure

1. Pugliese SM, Chou GM, Abe ETT, Turcu D, Lancaster JK, Tuthill JC, Brunton BW. “Connectome simulations identify a central pattern generator for fly walking.” Nature, published online 28 October 2026. doi:10.1038/s41586-026-11195-y. Abstract, figure and extended-data captions, funding and competing-interest statements read on nature.com; main text paywalled.
2. Pugliese SM et al. “Connectome simulations identify a central pattern generator circuit for fly walking.” bioRxiv preprint, version 2, 30 April 2026. doi:10.1101/2025.09.12.675944; full text read via PMC13142387. Source of all numbers not in the Nature abstract.
3. Code and subnetworks: github.com/smpuglie/Pugliese_2026; simulation outputs on Zenodo, doi:10.5281/zenodo.22260923.
4. Mesa N. “Long-sought walking circuit found in fruit flies.” The Transmitter, 13 May 2026 (source of the Berg and Gatto quotes).
5. Sapkal N, Kumar DS, Sunke S, Mancini N, Pitchford J, Murakami K, Bidaye SS. “Central versus peripheral neural control of a coordinated walking pattern in Drosophila.” bioRxiv preprint, version 2, 5 May 2026. doi:10.64898/2026.04.29.721658 (abstract read).
6. Crossref metadata for doi 10.1038/s41586-026-11195-y (online date check).

Disclosure: This article was drafted with AI assistance and checked against the sources above. The Nature main text was not accessible; figures are taken from the authors’ open preprint and may differ in minor ways from the final version. The authors declare no competing interests; funding came from the US National Institutes of Health, the Air Force Office of Scientific Research, the Swartz, Shanahan Family and New York Stem Cell foundations, a Pew award and a university chair. Calculations by SciQuest: 636 ÷ 1,024 = 62.1%; 3,521 ÷ 4,096 = 86.0%; 29 ÷ 933 = 3.1%; 3,817,772 synapses rounded to 3.8 million. The graphic is a SciQuest original drawn from the paper’s reported values. Spotted an error? See our corrections policy.