Activity spread through FlyWire wiring predicted correlations among 67 fly brain regions better than shared arbors, and partner shuffles inside neuropils did not reduce the fit

Question: which part of the wiring causes the relation between region connectivity and resting-state functional connectivity (FC) in the Drosophila brain? Approach: we used the complete FlyWire connectome (release 783, 139,255 neurons) and the 20-fly calcium-imaging FC data of Turner et al. (2021) for 67 regions in the two hemispheres (2,211 pairs). We compared the Turner cell count, a shared-arbor model without synaptic transmission (one neuron adds its calcium signal to each region that contains its synapses), a linear propagation model on the full neuron graph, a signed variant, and two null graphs. A regression removed region distance, size and shared boundary. Results: the FlyWire cell count correlated with FC at r = 0.70 (contralateral pairs 0.69; hemibrain on the Turner regions 0.72), but region size alone gave r = 0.54. After geometry control, shared arbors gave partial r = 0.33 [95% region-bootstrap CI 0.23, 0.43] and the cell count 0.38 [0.28, 0.47]. Unsigned propagation (gain 0.9) gave 0.52 [0.42, 0.61], higher than shared arbors in 20 of 20 flies; a graph with permuted neuron labels gave 0.17. A graph with partners shuffled inside each neuropil, with synapse counts kept, gave 0.54 [0.44, 0.63]; all 10 shuffles were equal to or above the real graph. Negative GABA and glutamate synapses gave 0.43 (difference −0.09 [−0.18, 0.01]). Removal of the 5,177 Kenyon cells decreased the fit by 0.075, compared with 0.004 for random central neurons; FlyWire labels most Kenyon cells as dopaminergic. Held-out-region R² increased from 0.53 (geometry) to 0.66 (geometry plus propagation). Conclusion: at this scale, FC follows the neuropil-level input-to-output routes of neurons over many synaptic steps, not shared neurites alone and not partner choice inside a neuropil. Limitations: the connectome and the FC come from different flies; the model is linear, with one gain and fixed signs; only a 67-region parcellation was tested; the geometry control is a regression.

Drosophila · FlyWire · connectome · functional connectivity · resting state · calcium imaging · structure-function · linear network model · shared arbors · null model · Peters' rule · Kenyon cells · mushroom body · neurotransmitter sign · hemibrain