Dynamics of Information Exchange in Zebrafish: The Role of U-Turns in Visual Communication and Behavior Modulation

C. K. Chan; Hao-Yun Hsu

REVIEW 3 major objections 5 minor 15 references

Dynamics of Information Exchange in Zebrafish: The Role of U-Turns in Visual Communication and Behavior Modulation

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper argues that zebrafish pairs exchange information through U-turn movements, and that increasing the distance between the fish stops the leader's signals and changes the shapes of their trajectories.

desk verdict A plausible and honestly caveated hypothesis that zebrafish U-turns are visual signals, but the evidence is conditional and the simulation is not predictive; worth refereeing, not yet convincing. read the letter →

arxiv 2412.20912 v3 pith:MJ6UQWVK submitted 2024-12-30 physics.bio-ph

C. K. Chan , Hao-Yun Hsu This is my paper · ORCID

classification physics.bio-ph

keywords zebrafishU-turnvisualcommunicationleader-followerrelationtime-laggedmutualinformationcollectivemotionexchangeanimalbehavior

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that a pair of zebrafish separated by a thin transparent barrier maintains its leader-follower relation by using U-turn movements as visual signals. When the barrier separation is small, the two fish's trajectories are nearly synchronized; when it is larger, the same leader-follower relation remains but the trajectories take different forms. The authors interpret side peaks in the detrended time-lagged mutual information as evidence that the leader sends quasi-periodic U-turn signals only when the follower is close, and they show that a simulation built from these signaling rules reproduces the observed trajectory transition. The claim matters because it identifies a concrete, observable motor pattern as the channel of information exchange that shapes collective motion, connecting behavior to known retinal sensitivity to motion reversal.

What carries the argument

The load-bearing machinery is detrended time-lagged mutual information (TLMI) together with the simulated '2UT' maneuver. TLMI measures how much information one fish's position time series carries about the other's at a given time lag; detrending removes the smooth gliding motion so that the U-turn fluctuations become visible. Side peaks beside the central peak of the auto-TLMI indicate quasi-periodic U-turn activity, and the absence of side peaks for the leader at large separation is read as the leader stopping its signals. The 2UT is a pair of consecutive U-turns lasting about one second, which slows a fish down without reversing its overall direction, so it can serve simultaneously as a speed controller for maintaining leader-follower spacing and as a detectable signal. The simulation combines spontaneous leader 2UTs near the boundary, delayed response 2UTs by the follower, and a minimal-separation rule enforced by the follower, and it reproduces the synchronized small-d trajectories, the distinct large-d trajectories, and the multi-peaked cross-TLMI seen in experiments.

What would settle it

Record a zebrafish pair at a fixed small separation while preventing U-turns in one fish, for example by constraining it or by suppressing the relevant motor behavior, and recompute the detrended TLMI; if the side peaks in the other fish's auto-TLMI persist, the peaks cannot be attributed to U-turn signaling. A complementary check is to shuffle or remove the U-turn segments from recorded trajectories and see whether the side peaks disappear.

Watch

Extended reading notes

Core claim

The central claim is that U-turns are functional visual signals in zebrafish pairs, used both to exchange information and to maintain a leader-follower relation (LFR). At small minimal separation d the pair shows a mutual engaging interaction: both fish stay near the barrier, their position trajectories are almost synchronized, and cross time-lagged mutual information shows that one fish consistently leads. At larger d, the same leader-follower relation persists while the trajectory forms change, because the leader stops performing spontaneous double U-turns near the ends of the tank. The evidence comes from detrended time-lagged mutual information, whose side peaks indicate quasi-periodic U-turn signals in the engaging fish, and from a simulation in which the follower responds to the leader's paired U-turn signals (2UTs) with a delay and maintains a minimal separation. The simulation reproduces the main features of both the small-d and large-d trajectories, supporting the interpretation that the trajectory transition is driven by a change in information exchange as d increases. The paper further notes that motion reversal triggers strong firing in the retinas of observing fish, making U-turns a plausible visual signal.

Load-bearing premise

The most fragile premise is that the side peaks in the detrended auto-TLMI are caused by the fish's periodic U-turn signals rather than by detrending or boundary effects; a second load-bearing assumption is that the leader ignores the follower when the separation is large.

Editorial extensions

If this is right

If the central claim is right, the transition from synchronized to dissimilar trajectories at larger separation is a direct consequence of reduced information exchange, not a change in the pair's social relationship.
The leader-follower relation is maintained by the follower's U-turn-based waiting behavior, so cohesion is an active, signal-mediated process rather than a passive attraction force.
A single paired U-turn serves both as a speed controller and as a signal, meaning the same motor pattern can carry information without requiring an extra behavioral act.
A simulation based only on sensory detection and decision rules reproduces observed collective trajectories without introducing ad hoc attraction or repulsion forces.
The proposed signaling role of U-turns is consistent with motion reversal evoking strong retinal responses in observing fish, linking this behavioral pattern to known neural sensitivity.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A testable extension the authors do not attempt is to suppress U-turns in one fish and check whether the other fish's detrended TLMI side peaks disappear; this would directly test whether the side peaks are signals rather than detrending artifacts.
The detrended TLMI side-peak signature could be used as a general detector of rhythmic signaling in other species or larger groups, where identifying who is signaling by eye is harder.
If the leader's silence at large d reflects ignoring rather than inability to see, then changing visual contrast or adding visual noise should shift the critical distance at which the trajectory transition occurs.
The dual speed-control and signaling role of the 2UT suggests that other rhythmic or repeated motor patterns in collective behavior may have similarly dual functions, an idea that goes beyond zebrafish.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: this is a hypothesis-generating paper with an honest set of caveats, not an established demonstration of visual U-turn signaling. The distance-controlled pair experiment is a nice idea and the trajectory classification (EE/EL/LL) with d is a clean observation. But the central inference—that side peaks in detrended auto-TLMI are U-turn signals, and that their disappearance for the leader at large d reflects the leader stopping signaling—rests on a conditional assumption the authors themselves flag. The simulation does not rescue it, because the large-d behavior is put in by setting Ps=0.

What's genuinely new: a two-fish setup with a transparent barrier letting them see each other while controlling minimal separation d; TLMI-based leader identification; a 1D simulation with a '2UT' double-U-turn rule that reproduces gross trajectory forms. The authors are appropriately careful in places: they write 'If these side peaks come from signals used by the fish...' and 'we assume that the leader will ignore the follower when d is large.' They also note the model does not capture all features. That honesty counts.

Soft spots, in order. First, the signal interpretation is underdetermined. Side peaks in a single-trajectory autocorrelation can come from boundary turning, the detrending window, or common looming stimuli when both fish are near the windows; none of these require sender-receiver signaling. The cross-TLMI being significant only at small d is also consistent with correlated responses to a shared boundary. Second, the simulation is parameterized to reproduce the phenomena, not tested against new data; no quantitative model-data comparison is given, and the large-d transition is imposed by setting the leader's spontaneous 2UT probability to zero. Third, the sample is small: conclusions are said to hold for at least 4 pairs, but the main figures are from a single pair, and there's no control (e.g., preventing visual contact or manipulating the leader's visibility) that would break the ambiguity.

The observation itself—that pair trajectories change form with d while maintaining leader-follower structure—is plausible and worth reporting. The U-turn signaling hypothesis is a reasonable candidate mechanism, and the retinal response citation gives indirect support. But 'consistent with' is not 'established by.' If I were the editor, I'd send this to review: the experimental setup and the hypothesis are worth airing, and a good referee can push for the missing control and a more predictive model. I'd also note that the authors' conditional language is more careful than the abstract's 'Our finding that UTs as important visual signals' implies; the abstract overreaches a bit relative to the evidence.

Final: this is a paper for people working on collective animal behavior and sensory communication. It deserves a serious referee, but I'd want substantial revision before accepting the signaling claim.

Referee Report

3 major / 5 minor

Summary. This manuscript reports experiments on pairs of zebrafish separated by transparent windows at varying minimal distances d. The authors find that at small d the two fish's x-trajectories are nearly synchronized (EE type), while at larger d the trajectories take different forms (EL type) while a leader-follower relation persists. Time-lagged mutual information (TLMI) between the two positions shows a positive-lag peak, indicating that Fish 2 leads Fish 1, and the peak height decreases with d. After detrending with a 1-s smoothing window, auto-TLMI exhibits side peaks, which the authors interpret as quasi-periodic U-turn signals emitted by the engaging fish. They propose a simulation model with spontaneous double U-turns (2UTs) by the leader, follower responses with a delay, and a minimum-following distance, and they show that the simulation can reproduce the qualitative forms of the small-d and large-d trajectories. The paper concludes that the EE-to-EL transition reflects a change in information exchange: the leader stops sending U-turn signals when d is large.

Significance. The central hypothesis, that U-turns serve as visual signals maintaining leader-follower coordination and that their range is limited by inter-individual distance, is interesting and testable. The experimental design is a strength: physically separating the fish with windows while allowing visual contact and systematically varying d is a clean way to isolate visual information exchange, and the use of TLMI is appropriate for detecting directional coupling. The paper is also transparent about its assumptions, explicitly labeling the signal interpretation as conditional and acknowledging limitations of the simple model. If the signal interpretation can be validated with direct U-turn event analysis and quantitative model comparison, the work would provide a concrete mechanistic link between sensory detection, decision rules, and emergent collective motion in zebrafish. However, because the current support for the signal identification is indirect, the significance is conditional.

major comments (3)

[Figure 4 and following paragraph] The central inference that side peaks in the detrended auto-TLMI are U-turn signals is not established. The text itself states, 'If these side peaks come from signals used by the fish to send information to the other fish, our result indicates...', and the subsequent conclusion that information is shared only for small d depends entirely on this conditional identification. Quasi-periodic side peaks in a single-fish auto-TLMI can also arise from (i) the 1-s detrending window imposing a preferred fluctuation timescale, (ii) periodic turning at the ends of the channel, or (iii) common visual looming when both fish are near the windows; none of these alternatives requires sender-receiver communication. The random-shuffle noise estimate reported in Figure 3 establishes a baseline for uncorrelated data but does not identify the mechanism producing the correlations. The authors should mark U-turn events directly from the trajectories (e.g., local extrema or curvature thresholds) and show that the auto-TLMI side peaks are specifically associated with those events, or provide a surrogate control that preserves boundary-turning statistics but destroys any coupling.
[Simulation model and Figure 5] The simulation cannot independently support the claim that the large-d transition is caused by a change in information exchange, because the key large-d behavior is inserted by hand. In the upper inset of Figure 5 the leader's spontaneous 2UT probability is set to Ps = 0, following the text's assumption that 'the leader will ignore the follower when d is large.' The absence of leader U-turns at large d is therefore an input to the model, not a derived consequence, and the simulation's reproduction of the EL-type trajectory is a restatement of that assumption. Moreover, the large-d simulation also changes the speeds (v1 = 0.009, v2 = 0.005) relative to the small-d case (v1 = v2 = 0.003), so two parameters are varied together and the separate effects of Ps = 0 and the speed difference are not identified. A mechanistic explanation would need d to enter the model explicitly with Ps (or the leader's decision) as an output.
[Figure 5 and 'Details of the parameters' in the SI] There is no quantitative model-data comparison. The statement that Figure 5 reproduces the essential features of the observed trajectories is based on visual similarity of sample trajectories and a single cross-TLMI inset. The simulated cross-TLMI shape follows directly from the implemented 2UT mechanism and the chosen delay tau_d; it is not an independent prediction. With at least seven free parameters (Ps, ds, tau_2U, tau_d, d_m, v1, v2, and the detrending window), one needs a systematic parameter search, error bars on experimental summary statistics, and a goodness-of-fit measure to show that the proposed rules are constrained by the data rather than merely consistent with one hand-picked realization.

minor comments (5)

[References] Reference [10] appears mis-cited: the idtracker.ai tracking software should cite Romero-Ferrero et al., Nature Methods 16, 179 (2019), not Van der Walt et al., PeerJ 2, e453 (2014), which is the scikit-image paper.
[Experimental methods] The manuscript reports results from one series of experiments with the same fish pair to avoid variability, but also claims the results are common to at least four pairs; please provide the per-pair data or supplementary figures so that this generality claim can be checked.
[Figure 4 caption] The text says 'A shorter ts will give smaller cross-TLMI peaks' but gives no sensitivity analysis; please show how the side-peak structure depends on ts over a range, since the choice ts = 1 s could otherwise be seen as selecting the phenomenon.
[Figure 5 caption] In the caption, 'Ps, τ2U = 40' should presumably read 'Ps = 0.15, τ2U = 40'.
[Text after Figure 4] The sentence 'Once this happens, the two fish will engage in mutual UTs responses seen in Figure 1' contains a typo ('UTs responses'), and the causal claim that the leader performs U-turns 'because the leader will then start to see the follower' is not directly evidenced and should be marked as a hypothesis.

Circularity Check

3 steps flagged · score 7.0 of 10

The simulation's 2UT signal mechanism and the large-d Ps=0 shutoff are built in from the features the paper claims to explain, so the central communication claim is partly circular.

fitted input called prediction [Simulation section, large-d inset of Figure 5 and model rules paragraph]
"Here, d is important because we assume that the leader will ignore the follower when d is large. ... Simulated trajectories for the case of large d are shown in the upper inset of Figure 5. Here, the leader is not engaging with the follower and it will not perform any spontaneous 2UTs. In this case, Ps = 0 and therefore the leader will be moving with a linear trajectory."

The decisive large-d feature, namely the absence of leader U-turn signals, is not an emergent consequence of the model but is inserted by setting Ps = 0, under the stated assumption that the leader ignores the follower at large d. The simulation is then cited as reproducing the observed absence of side peaks in Fish 2 and as support that the trajectory transition is caused by a change in information exchange. That support is circular because the outcome is a direct restatement of the parameter choice.
self definitional [Simulation model, 2UT implementation paragraph and Figure 5 lower-inset discussion]
"One important mechanism in our model is the implementation of a 2UTs; two consecutive U-turns with a total duration of τ2U. These 2UTs are inspired by the quasi-periodic nature of the TLM shown in Figure 4 ... The location of the main (tallest) peak is determined by the delay in the response of the follower to the UTs of the leader."

The model is constructed around the 2UT rule, whose quasi-periodic output is 'inspired by' the very auto-TLMI side peaks it is then used to explain. The simulated cross-TLMI's multiple peaks and their delay are direct outputs of the implemented 2UTs and the chosen τd, not independent confirmations. Thus the simulation's match to Figure 4 reduces to the construction of the input rules and cannot independently establish that the side peaks are U-turn signals.

1 more flagged steps

self definitional [Figure 4 discussion paragraph and 'picture emerges' paragraph]
"If these side peaks come from signals used by the fish to send information to the other fish, our result indicates that Fish 1 is sending signal for all d while Fish 2 is doing the same only for d = 0 .4 and 1 .4 cm. ... Since UTs are not needed just for following, most likely the UTs of the follower shown in the lower inset of Figure 2 are also used as signals to acknowledge its detection of the leader."

The signal interpretation is introduced as a conditional ('If these side peaks come from signals...') and is then used to conclude that U-turns are signals. No independent control or model comparison rules out detrending artifacts or boundary-related turning correlations as the source of the side peaks. The conclusion therefore re-states the assumption that defines the side peaks as signals rather than deriving it from the data.

full rationale

The paper does not rely on a load-bearing self-citation chain; the TLMI methodology and retinal-firing citations are external or standard and are not the circular core. The circularity is in the model and interpretation chain. The 2UT rule is inferred from the auto-TLMI side peaks and then the simulation is said to reproduce those peaks; the large-d absence of leader U-turns is enforced by setting Ps = 0 under the explicit assumption that the leader ignores the follower. Hence the simulation cannot independently support the central claim that the trajectory transition is caused by a change in information exchange. The paper itself flags the fragility with 'If these side peaks come from signals...' but then treats the signal interpretation as the conclusion. This is partial, not total, circularity: the empirical TLMI measurements and leader-follower identification are independent observations, and the retinal-firing fact provides plausibility, but the load-bearing 'prediction' reduces to construction choices.

Assumptions & free parameters 8 free parameters · 4 assumptions · 1 invented entities

The central mechanism rests on eight fitted or hand-chosen parameters, four domain assumptions, and one model construct. The free parameters are mostly tuned so that the simulation reproduces the experimental trajectory forms and TLMI structure. The only external anchor is the cited retinal-firing result, which suggests U-turns are salient visually but does not establish that they are signals in this pair context.

free parameters (8)

Ps (leader spontaneous 2UT probability) = 0.15
Chosen to reproduce boundary fluctuations in small-d simulated trajectories; no estimation procedure is described.
ds (slow-down zone size) = 0.1
Chosen as the boundary region where leader 2UTs occur.
tau_2U (duration of double U-turn) = 40 simulation steps (about 1.3 s)
Set to match the quasi-periodic TLMI side peaks with characteristic time near 1 s.
tau_d (follower response delay) = 13 simulation steps (about 0.43 s)
Set to produce the location of the main cross-TLMI peak in the simulated data.
d_m (minimum separation maintained by follower) = 0.05
Enforces the leader-follower relation and repulsion in the model.
v1, v2 (swimming speeds) = small d: 0.003 each; large d: v1=0.009, v2=0.005
Adjusted to match trajectory slopes and the spacing between fish in the two regimes.
ts (smoothing window for detrending) = 1 s
Detrending window chosen to expose U-turn fluctuations; shorter windows give smaller cross-TLMI peaks.
TLMI discretization resolution = 2 bits
Binning choice affects absolute information values and peak structure.

assumptions (4)

domain assumption Leader-follower relation is intrinsic to the pair, with the same fish always leading.
Inferred from cross-TLMI peak positions in one representative series; causality is not manipulated or tested across many pairs.
ad hoc to paper The leader ignores the follower when d is large.
Stated in the model section: 'we assume that the leader will ignore the follower when d is large.' This assumption sets Ps=0 and drives the large-d transition.
ad hoc to paper Side peaks in the detrended auto-TLMI represent U-turn signals.
The interpretation is conditional in the text: 'If these side peaks come from signals...' and is not independently verified.
domain assumption Each fish moves at constant speed between instantaneous U-turns, and a U-turn is implemented as a direction reversal.
Simulation model approximates real fish turning kinematics with a binary direction indicator.

invented entities (1)

2UTs (double U-turn signal unit)
purpose: Serves as both a speed-control maneuver and a visual signal between fish; the leader's 2UTs trigger delayed 2UTs from the follower.
The 2UT construct is inferred from quasi-periodic TLMI side peaks and then embedded in the simulation; no direct behavioral or neural evidence shows that fish produce and detect 2UTs as a discrete unit.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Dynamics of Information Exchange in Zebrafish: The Role of U-Turns in Visual Communication and Behavior Modulation." pith.science (2026). https://pith.science/paper/MJ6UQWVK

@misc{pith2026241220912,
  author       = {Pith},
  title        = {Pith review of: Dynamics of Information Exchange in Zebrafish: The Role of U-Turns in Visual Communication and Behavior Modulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MJ6UQWVK}},
  note         = {Machine review of arXiv:2412.20912}
}

abstract

Motions of visually coupled zebrafish pairs are studied to understand the effects of information exchange on their behavior as a function of their minimal separation ($d$). We find that when $d$ is small, the pair can display a leader-follower relation (LFR) with trajectories of almost synchronized form. However, with larger $d$, although the same LFR is still maintained, the originally similar trajectories turn into different forms. Detailed analysis of their motion trajectories suggests that the pair might be using U-turns (UTs) to exchange information and to maintain a LFR at the same time. A simulation model based on UTs with inferred and proposed rules is able to reproduce prominent features of observed trajectories; indicating that the transition of trajectories can be understood as the result of a change in information exchange between the fish as $d$ increases. Our finding that UTs as important visual signals is consistent with the fact that UTs can induce a large amount of firings in retinas of observing fish.

Figures

Figures reproduced from arXiv: 2412.20912 by the authors.

**Figure 1.** FIG. 1. Time course [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 3.** FIG. 3. Effects of minimal separation distance on the cross [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: FIG. 4. Effects of detrending on the form of cross-TLMI be [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]

**Figure 5.** Figure 5: Similar to that from the experiment (Figure 4), [PITH_FULL_IMAGE:figures/full_fig_p003_5.png]

**Figure 5.** Figure 5: FIG. 5. Simulated trajectories of the follower (Fish 1) and [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages

[1]

Vicsek and A

T. Vicsek and A. Zafeiris, Physics reports 517, 71 (2012)

work page 2012
[2]

Mora et al., Nature physics 12, 1153 (2016)

T. Mora et al., Nature physics 12, 1153 (2016)

work page 2016
[3]

Katz et al., Proceedings of the National Academy of Sciences 108, 18720 (2011)

Y. Katz et al., Proceedings of the National Academy of Sciences 108, 18720 (2011)

work page 2011
[4]

C. W. Reynolds, in Proceedings of the 14th annual con- ference on Computer graphics and interactive techniques (PUBLISHER, ADDRESS, 1987), pp. 25–34

work page 1987
[5]

I. D. Couzin et al., Advances in the Study of Behavior 32, 10 (2003)

work page 2003
[6]

Ito and N

S. Ito and N. Uchida, PNAS nexus 3, 264 (2024)

work page 2024
[7]

Lecheval et al., Proceedings of the Royal Society B: Biological Sciences 285, 20180251 (2018)

V. Lecheval et al., Proceedings of the Royal Society B: Biological Sciences 285, 20180251 (2018)

work page 2018
[8]

Lecheval et al., BioRxiv 138628 (2017)

V. Lecheval et al., BioRxiv 138628 (2017)

work page 2017

Show all 15 references

[9]

Schwartz et al., Neuron 55, 958 (2007)

G. Schwartz et al., Neuron 55, 958 (2007)

2007
[10]

Van der Walt et al., PeerJ 2, e453 (2014)

S. Van der Walt et al., PeerJ 2, e453 (2014)

2014
[11]

Chou et al., Physical Review E 103, L020401 (2021)

P.-Y. Chou et al., Physical Review E 103, L020401 (2021)

2021
[12]

Key insights emerge when re-examining the phe- nomenological cohesion and repulsion rules

could have taken place but difficult to detect. Key insights emerge when re-examining the phe- nomenological cohesion and repulsion rules. The param- eter dm in our model does not only enforce the repulsion rule but also provide long enough distance between the animals such th...
[13]

Chen et al., Entropy 26, (2023)

W.-J. Chen et al., Entropy 26, (2023)

2023
[14]

J.-H. Yu, J. L. Napoli, and M. Lovett-Barron, Current Opinion in Neurobiology 86, 102866 (2024)

2024
[15]

Portugues and F

R. Portugues and F. Engert, Current opinion in neurobi- ology 19, 644 (2009)

2009

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

Vicsek and A

T. Vicsek and A. Zafeiris, Physics reports 517, 71 (2012)

work page 2012

[2] [2]

Mora et al., Nature physics 12, 1153 (2016)

T. Mora et al., Nature physics 12, 1153 (2016)

work page 2016

[3] [3]

Katz et al., Proceedings of the National Academy of Sciences 108, 18720 (2011)

Y. Katz et al., Proceedings of the National Academy of Sciences 108, 18720 (2011)

work page 2011

[4] [4]

C. W. Reynolds, in Proceedings of the 14th annual con- ference on Computer graphics and interactive techniques (PUBLISHER, ADDRESS, 1987), pp. 25–34

work page 1987

[5] [5]

I. D. Couzin et al., Advances in the Study of Behavior 32, 10 (2003)

work page 2003

[6] [6]

Ito and N

S. Ito and N. Uchida, PNAS nexus 3, 264 (2024)

work page 2024

[7] [7]

Lecheval et al., Proceedings of the Royal Society B: Biological Sciences 285, 20180251 (2018)

V. Lecheval et al., Proceedings of the Royal Society B: Biological Sciences 285, 20180251 (2018)

work page 2018

[8] [8]

Lecheval et al., BioRxiv 138628 (2017)

V. Lecheval et al., BioRxiv 138628 (2017)

work page 2017