Whole-body 3D kinematics of freely behaving Drosophila

Key results

1. ShownOn held-out running frames, triangulated foreleg-tip predictions reproject with a mean error of 18.2 µm across seven camera views, about 2-3% of the length of a fly femur.

   Front tarsal tips were hand-labelled in all 7 views in 51 frames: reprojection error 1.48 ± 0.69 pixels (18.2 ± 8.5 µm, mean ± s.d.), against a femur of ~0.72 mm. Calibration landmarks: 0.48 ± 0.26 pixels (5.9 ± 3.2 µm) for N=272 landmarks, computed on the calibration training data. Cameras resolve ~12.3 µm/pixel at 800 fps. · 2.1 The rainbow rig tracks whole-body 3D…; Camera calibration

2. SuggestedScutellum height and forward speed (used as a stand-in for the centre of mass) peak in phase twice per step cycle at every running speed, with no aerial phase. This fits grounded running: one locomotor program adjusted continuously with speed, with no switching between discrete gaits.

   2213 running bouts (~654,000 frame-sets) from 53 flies. Peaks fall at 0.5-0.6 rad (mid L1 swing) and -2.8 to -2.9 rad (mid L1 stance), each during one tripod's swing. Mean phase and spread change little across speed tertiles. Oscillation amplitude rises with step-cycle mean speed (linear-fit slopes 0.159 for height, 0.567 for speed) while the phase relationship stays stable. · 2.2 3D joint kinematics of freely running flies; 2.3 Fly kinematics are consistent with grounded…; Amplitude vs. speed; Figure legends

3. ShownCoupling between legs tightens with speed. Tripod groups share phase offsets in fast running, while slow running shows weaker, more variable coupling, most clearly in the hind legs. The increase in joint range of motion with speed falls on different joints in the middle legs than in the front and hind legs.

   Median mean resultant length R of each leg's phase offset to L1 rises from the slow to the fast tertile for every leg (R1 0.70 to 0.93, L2 0.72 to 0.95, R2 0.73 to 0.95, L3 0.69 to 0.93, R3 0.63 to 0.93; 33-35 paired flies per leg). Two-sample Kuiper tests: V = 0.30-0.57, all p < 0.01 after Bonferroni correction. · 2.2 3D joint kinematics of freely running flies; Inter-leg coordination across speeds; Joint-speed correlation analysis; Figure legends

4. ShownIn courting pairs tracked from video alone, the unextended wing moves in phase with the extended, singing wing during sine song, and the phase does not break when the singing wing switches. Male thorax pitch follows the angle from the male's body to the female's centre of mass.

   11 male-female pairs filmed at 800 Hz. During sine song, phase differences between the wings cluster near 0 rad, and amplitude passes from one wing to the other at switches with no phase discontinuity. Pulses detected from wing-tip kinematics split into Pslow and Pfast under a two-component Gaussian mixture. Pitch versus angle to the female's centre of mass: median offset 23.2° across bouts. · 2.4 Multi-animal tracking of flies during…; 3.2 Courtship kinematics in three dimensions; Courtship data; Figure legends

5. SuggestedWith little extra annotation, the pipeline also tracks grooming, optogenetically driven running in headless flies, and running over stepped terrain. Running induced through DNg100 uses less leg flexion than wild-type running but stays inside the wild-type kinematic space.

   About 600 extra annotated frames were enough for grooming. Midleg stance centroids shift 0.41 mm (L2) and 0.47 mm (R2), p = 0.002 (Hotelling T², Holm-Bonferroni; 14 grooming vs 19 standing bouts, one male fly). DNg100>CsChrimson vs speed-matched wild type: principal-component (PC) centroids 0.17 PC units apart, n=50 cycles each. On steps, flies searched with the forelegs and swung legs again after failed landings. · 2.5 3D tracking of other behaviors and…; Figure legends