An arXiv preprint reports that feedback from body bending is associated with the way zebrafish match their swimming to nearby fish and form organized schools. The proposed mechanism involves proprioceptors—spinal sensors that register the fish’s own movement—and was examined in live-fish recordings and behavior, fluid simulations, and a physical fish-inspired robot.
A measurable pattern in the school
Control fish formed tighter, more aligned groups than shuffled comparisons. Their polygonal area was smaller than the shuffled null model in eight batches (p = 0.0117), while group polarization was higher in nine batches (p = 0.002).
Swimming timing also changed with the distance between fish, a relationship the paper calls vortex phase matching. The fit was R² = 0.571 in static water (p = 0.0138; nine batches) and R² = 0.548 in a swim tunnel (p = 0.00718; 10 pairs). These results describe an association between spacing and timing; they do not by themselves show that the spinal circuit caused the schooling pattern.
Signals from a fish’s own body
The experiments used wild-type and genetically modified zebrafish of both sexes at juvenile and adult stages. They combined electrophysiology, calcium imaging, optogenetics, and behavioral analysis with a neuromechanical model and a physical robot. The study also created a piezo2b mutant line with CRISPR/Cas9 and used Cre-dependent DTA expression to eliminate centrally located, piezo2b-expressing proprioceptors.
During spinal bending, calcium activity in intraspinal proprioceptor cells was higher (p < 0.0001; five cells from two fish), while activity in contralateral V2a cells was lower (p = 0.0005; six cells from three fish). The finding came from cell-level measurements in a small number of fish, so it is evidence about this circuit, not a population-wide estimate.
When feedback from the fish’s own bending was blocked, locomotor bursts were slower, V2a excitation lasted longer, and inhibitory currents were smaller than during free-tail swimming. The comparison was nonrandomized and was made in ex vivo or head-fixed preparations, which limits how directly it represents a freely swimming fish.
In separate pacing tests, swimming frequency tracked imposed tail-bending frequencies of 2, 4, 6, 8, and 10 Hz, with R² = 0.9964 across 10 fish. In an ex vivo test with eight fish, dynamic schooling signals were associated with time-locked V2a inhibition when the preparations were silent and with synchronized V2a firing and muscle-activity bursts during swimming; both before-versus-pacing and pacing-versus-post comparisons had p < 0.0001.
The same rule appeared in engineered tests
In a fluid simulation, a follower changed its swimming timing with the distance from a leader when proprioception was switched on. The phase-distance fit was R² = 0.558 across five networks; with proprioception off, followers decoupled and the fit fell to R² = 0.074 across five networks (ON versus OFF, p < 0.0001).
The physical robot showed a similar pattern relative to a flapper-generated wake rather than a live fish: R² = 0.711 with feedback on and R² = 0.035 without it, across six trials in each condition (ON versus OFF, p < 0.0001).
With vortices present, power use was about 17% lower in the proprioception-on simulation (50 networks) and about 7% lower in the robot (six trials). No significant off-condition benefit was reported (p = 0.06 in simulation; p = 0.1061 in the robot), and these power readings were not direct metabolic measurements in live fish.
The genetic evidence was suggestive, not simple
The genetic test pointed in the same direction but also exposed a complication. Wild-type fish showed a phase-distance association in the swim tunnel (R² = 0.548, p = 0.00718; 10 pairs), whereas piezo2 mutants did not show a clear association (R² = 0.141, p = 0.281; six pairs); a control-mutant permutation test gave p = 0.001.
Piezo2 mutants also showed larger dispersion and less cohesive, less polarized schooling. Fish whose centrally located proprioceptors were targeted for elimination likewise showed loss of cohesive schooling and polarization. Because the Piezo2 mutation was also associated with locomotor effects, it was not a fully specific test of social coordination.
The findings remain specific
The biological evidence is confined to zebrafish. The simulations and robot add engineered tests, but their outcomes depend on model assumptions and settings such as body stiffness, and their power measures cannot be treated as direct evidence of metabolic savings in live fish.
The work combines different assay types rather than reporting one overall sample size, and some recordings came from multiple cells taken from the same fish. The study did not fully isolate spinal feedback from descending brain inputs or other sensory systems, leaving open how the proposed circuit interacts with vision and lateral-line sensing.
It is an arXiv preprint; no journal, DOI, PMID, or PMCID is listed. Further work would need to test whether the mechanism generalizes beyond zebrafish and whether the modeled energy benefit appears in live fish under natural schooling conditions.
Paper data and sources
Original title: A spinal circuit for collective coordination
Authors: Laurence Picton, David Madrid, Alessandro Pazzaglia et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text