The fly feeding circuit

How does disconnecting Roundup, a feeding-circuit neuron, change motor output?

Roundup outgoing connections

Simulation · Gain 1.0× · Seed 0 · Roundup outputs cut

Roundup’s neighbourhood

Select a neuron to understand its role, evidence and connections.

Graph layout · not anatomical positions or recorded activity

Loading source connections…

Actual source-connectome connections around RoundupA weighted, directed subnetwork from the source model. Select neurons or connections to inspect them. Positions are a graph layout, not anatomical coordinates.
Model-assigned signs:Positive weightNegative weightCut output

Select a neuron or arrow. Drag to pan; use + / − to zoom. Widths use a compressed scale; inspect for exact weights. Dashed outputs are zeroed.

I proposed and directed this focused reproduction and sensitivity study of Shiu et al. (2024). The original authors tested neuron silencing and weight robustness. Controls select saved runs; they do not run a live simulation.

Methods & sources

Explore the input-timing study · A separate experiment from this gain comparison and the 3D body replay.

MN9 and MN6

MN means motor neuron. MN9 innervates muscle 9, involved in rostrum (main mouthpart) protraction. MN6 innervates muscle 6, involved in labellar extension. They control different parts of the feeding apparatus. This experiment measures MN9 only; it does not demonstrate movement or feeding.

Motor anatomy: McKellar et al. (2020), Figures 7 and 9.

Simulation comparison

Thirty one-second saved simulations: three recurrent gains, five matched input seeds and two conditions. Gain scales all signed internal connection weights, including inhibition and sensory-neuron outgoing synapses. Cutting sets only Roundup’s outgoing weights to zero; its incoming connections remain. External input is unchanged. Accepted sensory input also matches across all six gain/condition cells for each seed in this archive.

The graph uses actual directed connections and signed source weights from a bounded Roundup neighbourhood. Layout positions are computed for readability, not measured anatomy. The selected subnetwork omits most of the brain and does not establish that the direct Roundup-to-MN9 connection explains the full effect. Edge widths use a capped square-root scale, adjusted by gain, for readability. The inspector gives exact gain-scaled weights and whether the cut zeros them. The export retains 55 neurons and 230 of their 1,070 induced directed connections; the initial strongest-links view shows a further subset. Full-source degrees and selected-subgraph counts are distinguished in the inspector. This is wiring, not animated neural propagation.

All fifteen saved pairs have fewer MN9 spikes after the cut. Five seeds are not five animals. One neuron, output and input regime do not establish biological effects, general robustness, novelty or full paper replication. The shared-candidate design and an execution correction were post-failure amendments.

Model: Shiu et al. (2024), pinned source, MIT. Replay provenance · Network selection and attribution · Aggregate results · Notices.