Wiring similarity of fly wing steering motor neurons in four connectomes matched flight co-activation of basalar and first-axillary muscles (Spearman 0.65) but not of the other six muscles (-0.12)

Question: sci:sha256:66ad0d92d95cfafcb44299b13e579f1d10dd8d43a336f4c4216914befc0abf1f found that shared input to wing steering motor neurons (MNs) correlated only 0.08-0.11 with muscle co-activation, in short movies triggered on fast muscle changes (about 6 min of flight). Does this weak match remain with all continuous flight data and a fourth connectome, and which muscles cause it? Approach: we calculated co-activation (Pearson r of GCaMP7f signals in flight periods) from the continuous traces of Melis et al. 2024: 16.9 h of tethered flight from 74 flies, about 170 times more flight time. We calculated MN input similarity (cosine of presynaptic synapse counts and nine other measures) in seven hemisegments of four animals: Male CNS v1.0 L/R, MANC v1.2.1 L/R, BANC v888 L/R and FANC L. The main test used 11 muscles (55 pairs). Results: both matrices were reproducible (co-activation split-half Spearman 0.98; wiring 0.96 between hemisegments, FANC 0.91-0.97). The overall match was weak, which reproduces sci:sha256:66ad0d92d95cfafcb44299b13e579f1d10dd8d43a336f4c4216914befc0abf1f: Spearman 0.12 (fly-bootstrap 95% CI 0.09-0.15; FANC 0.13; Mantel p = 0.37). But the weak value hid two groups. For b1, b2, b3, i1 and i2, a split set beforehand from the haltere clusters of Ludlow et al. 2026, wiring predicted co-activation: Spearman 0.65 (0.65-0.71 in each hemisegment). For iii1, iii3 and hg1-hg4 it was -0.12. Post hoc, the 21 pairs without a fourth-axillary (hg) MN gave 0.54 (p = 0.009) and the 34 pairs with one gave -0.12. The strongest flight synergy (i1, i2, hg1, hg2; mean r 0.71) received only 3.7% of its input from neurons that contact all four MNs; AN07B024 and IN11B002 did so in all six full hemisegments. The wiring module hg1-hg3-hg4 (46% common input) was not co-active (mean r -0.12). Over 330 four-MN groups, common-partner count did not track co-activation (-0.04). The best of 564 signed linear whole-nerve-cord models, with noise into one of eight source sets, gave 0.31 (ascending-neuron noise; max-statistic p = 0.46). Overlap of muscle outlines in the unmixing model did not explain co-activation (Spearman 0.01). Conclusion: premotor modules match flight synergies for basalar and first-axillary muscles; the fourth-axillary MNs cause most of the mismatch. Limits: 11 muscles with known MNs; one imaged side and driver line; no electrical synapses; predicted transmitter signs; the post hoc split is exploratory.

Drosophila · connectome · wing steering muscles · motor neurons · premotor modules · muscle synergy · MANC · BANC · Male CNS · FANC · calcium imaging · structure-function · flight control · ventral nerve cord · replication

Admitted 2026-11-09 17:16 UTC4 citing publications

References

Shared input in three fly maps had correlations of 0.08 to 0.11 with calcium coactivity of 11 wing muscles in selected female flight bouts sci:sha256:66ad0d92d95cfafcb44299b13e579f1d10dd8d43a336f4c4216914befc0abf1f
[Legacy relation: reproduces] We repeat its test (input cosine vs co-activation, 55 pairs) with 16.9 h of continuous, non-selected flight traces instead of triggered movies, and add FANC; we get Spearman 0.12 (FANC 0.13) vs its 0.08-0.11.
supportsAt gain 0.8, predicted transmitter signs gave higher wing coactivity scores than all 199 shuffles within cell class and side in two linear fly models sci:sha256:a4b621870894c9f34b527bc6f402efc45b69ea239380ca6ca4d4bae42dd181d2
Our 564 signed whole-nerve-cord model variants with source-specific noise also gave weak fits that did not exceed the muscle-label null (best 0.31, max-statistic p = 0.46).
At fixed target scales and gain 0.8, source variances bounded by one million fitted 11 wing muscle correlations in two signed fly connectome models sci:sha256:0908f9ae0d2a901ff79b480478d5ad37fe06bc32e31f78804e6766226bbd7681
Shows that nearly free source variances can fit the co-activation matrix; our common-partner analysis names candidate high-variance drivers.
A fixed ridge model gave correlations near -0.15 between input and motion similarity for 11 wing muscles in three fly maps sci:sha256:d6bc392d0e040b6f8c874b9dc2fe5a3b674c999250d7ee58eae903e51ff6c033
Related test of input similarity against muscle-kinematics profiles in the same data.
A prior from shared motor input did not reduce mean error in a fixed linear model of wing motion from calcium on 31 fly recording days sci:sha256:055115fec0febfe821fd12f3a33c05e0511eaca2cec1b9b86a024e8c62fe3d97
Related use of shared input as a prior for wing-motion decoding.
usesdoi:10.1038/s41586-024-07293-4
Source of the steering-muscle GCaMP7f flight data; our co-activation matrix reproduces its correlation slopes (Extended Data Table 2).
usesdoi:10.22002/aypcy-ck464
CaltechDATA deposit main_muscle_and_wing_data.h5 (CC0); continuous unmixed muscle traces of 74 sessions.
challengesdoi:10.1038/s41586-024-07600-z
We use its FANC wing preMN-to-MN table. Its premotor steering modules, proposed to be recruited together, did not predict flight co-activation outside the basalar/first-axillary muscles (within-module r 0.26 vs between 0.17, p = 0.15).
supportsdoi:10.7554/eLife.112125
Our b1, b2, b3, i1, i2 subset (wiring vs co-activation Spearman 0.65 in six hemisegments) agrees with its BANC finding of two haltere-input clusters (b1/b2 vs i1/i2/b3).
usesdoi:10.1016/j.cub.2016.12.018
Muscle outline model of the unmixing method, used for the imaging-geometry control; muscle nomenclature.
usesdoi:10.1101/2025.10.09.680999
Male CNS v1.0 connectome (left and right T2 wing MNs and whole-VNC network).
usesdoi:10.7554/eLife.97769
MANC v1.2.1 connectome.
usesdoi:10.7554/eLife.97766
MANC cell-type and hemilineage annotations.
usesdoi:10.7554/eLife.96084
MANC wing MN and premotor annotations.
usesdoi:10.7910/DVN/7WTH1N
BANC v888 Dataverse deposit (meta and edge list v3).
usesdoi:10.1101/2025.07.31.667571
BANC connectome paper; its annotation maps MANC MNwm35 to the iii4 MN.
usesdoi:10.1038/s41586-024-07389-x
FANC reconstruction and wing MN identification.
doi:10.1016/j.cell.2020.12.013
FANC dataset.
doi:10.7554/eLife.106548
Split-GAL4 lines and MANC matching of wing MNs; no co-activation test.

Cited by

4 publications

Public catalog entries on Sciscape.

usesSensory input fraction and median calcium skew had a rank correlation of -0.52 across 11 fly wing muscles
We use its unchanged record feature table, synapse edge subsets, source labels, motor matches, and record quality table. We test sensory input fraction against calcium skew. That question differs from its co-activation test.
Shared input gave no clear phase separation beyond muscle position in 20 motor output pools from one larval fly wiring release
This related test compared common premotor input with adult wing muscle activity. It motivates explicit functional scope. We reuse no data or code from it.
With normalized count weights and assumed 25% threshold gaps, a twofold source rate range allowed reversed order in 185/206 pairs in one fly leg map
This related study tested shared wiring against measured wing muscle activity. We test a separate leg drive bound and reuse no files from that study.
Shared input gave no clear phase separation beyond muscle position in 20 motor output pools from one larval fly wiring release
This related test compared common premotor input with adult wing muscle activity. It motivates explicit functional scope. We reuse no data or code from it.