Genetic heterogeneity alleviates behavioral trade-off between exploration and vigilance in Drosophila

Summary

Behavioral trade-offs are a major constraint on animal performance, but group living can reshape such constraints. Using genetically distinct strains of Drosophila melanogaster, we examined how aggregation and heterogeneity shape the exploration-vigilance trade-off under predator-like looming stimuli. A latent negative relationship between exploration and vigilance became apparent when aggregated, suggesting improved behavioral performance compared to solitary condition. Heterogeneity within the group further enhanced performance beyond expected levels, indicating an alleviation of the cost of balancing exploration and vigilance. We then applied a phenome-wide higher level association study (PheHAS) to identify phenotypic features associated with this alleviation across strain pairs. Differences in activity patterns, together with similarity in infection-response traits, were linked to greater trade-off alleviation. These results indicate that aggregation and heterogeneity have distinct effects on group behavior and provide a general framework for linking phenotypic diversity to collective performance.

Graphical abstract

Highlights

• •Aggregation reveals hidden trade-offs between exploration and vigilance in flies
• •Genetic heterogeneity within groups alleviates this behavioral trade-off
• •A phenome-wide analytical framework links phenotypes to trade-off alleviation
• •Diversity can enhance group performance by resolving behavioral dilemmas

Neuroscience; Behavioral neuroscience; Ethology; Behavior genetics

Introduction

Life-history traits are constrained by many trade-offs that arise when one trait cannot increase without a decrease in another.^1,2 This concept is crucial in understanding how animals balance competing demands for multiple resources throughout their life. Prior studies have categorized trade-offs into two types, so-called “allocation trade-off” and “acquisition trade-off.”³ Allocation trade-offs typically arise when animals distribute their limited resources^4,5 and lead to a negative correlation between life-history traits such as reproduction and survival.^6,7 However, many empirical studies have reported positive or negligible correlations between these traits, rather than negative correlations predicted by allocation trade-off theory.^8,9 A long-term monitoring study demonstrated that female reindeer with higher reproductive success also exhibited higher survival rates.¹⁰ These studies highlight the importance of the concept of “acquisition,” meaning that individuals acquiring more resources can allocate more energy to multiple traits. However, some studies have shown that acquisition often comes at a cost.^11,12 Although animals strive to maximize their resource utilization, phenotypic correlations can prevent them from simultaneously maximizing multiple functions. This constraint can generate negative correlations between traits. Such phenomenon, known as the acquisition trade-off, has been an important topic in behavioral ecology.¹³

The foraging-vigilance trade-off is a classic example of an acquisition trade-off. Extensive field and experimental studies have demonstrated its existence and underlying mechanisms.^14,15 Organisms face potential risks of both starvation and predation. Since foraging and avoiding predators cannot be performed simultaneously, a conflict arises in deciding how much to prioritize one behavior over the other. For example, a study using goldfish (Carassius auratus) as preys and little egrets (Egretta garzetta) as predators showed that individuals more motivated to forage were more likely to be preyed upon.¹⁶ However, unlike allocation trade-offs, acquisition trade-offs can sometimes be alleviated through cooperation in group-living animals. For instance, ostriches cannot maintain vigilance while foraging with their heads down, facing an acquisition trade-off between vigilance and foraging in solitary conditions. In contrast, within a group, individuals can focus on foraging while others remain vigilant.¹⁷ In some species, individuals spend more time on foraging without reducing vigilance as group size increases.^18,19 This pattern suggests that living in groups has the potential to resolve the foraging-vigilance trade-off.

Beyond aggregation effects driven by group size, grouping can also exert heterogeneity effects arising from differences among individuals. Accumulating evidence suggests that interindividual heterogeneity in behavior can drive emergent changes in group performance.²⁰ For example, groups of mice foraged more efficiently than individuals due to the presence of key individuals that were particularly good at foraging.²¹ Similarly, fly groups comprising two distinct strains exhibited enhanced exploration, likely due to differences in locomotor activity levels between genotypes.²² Such phenomena are partially explained by the well-established concept of indirect genetic effects, whereby the phenotype of group members influences an individual’s phenotype.^23,24 While these findings highlight the significant role of heterogeneity in group dynamics, previous studies have rarely examined how heterogeneity affects multiple phenotypes and reshapes behavioral trade-off structures. If heterogeneity plays complementary roles in interindividual interactions, it may enable groups to balance foraging and vigilance more effectively at the group level. Disentangling the effects of aggregation and heterogeneity is therefore essential for gaining a clearer understanding of how animal groups navigate behavioral trade-offs in nature.

Here, we investigate how group composition shapes the behavioral trade-off between exploration and vigilance using 78 genetically distinct strains of Drosophila melanogaster derived from the Drosophila Genetic Reference Panel (DGRP).²⁵ We set up three social conditions: single individuals, homogeneous groups (composed of a single strain), and heterogeneous groups (composed of two distinct strains). Behavioral recordings were conducted for 5 min under conditions with and without looming stimuli. We quantified exploratory behavior as search comprehensiveness and vigilance behavior as immobility rate. Pareto optimization analysis was applied to quantify and compare the relative performance of flies across the social conditions under the behavioral trade-off (Figure 1). Furthermore, we identified phenotypic diversity potentially resolving the trade-off using a comprehensive database. Our study provides novel insights into how group composition enhances behavioral performance by alleviating the trade-off between exploration and vigilance.

Figure 1.

Experimental scheme and analytical framework

(A) Flies were tested in experimental arenas placed under a display presenting looming stimuli. Behavioral data were collected under three social conditions: single individuals, homogeneous groups of six from the same strain, and heterogeneous groups of three each from two strains. Unstimulated periods (5 min) were used to measure exploratory behavior, and stimulated periods (5 min) were used to measure vigilance.

(B) Calculation of aggregation and heterogeneity effects. Aggregation effects were the deviation of group phenotypic values from the mean of single individuals. In contrast, heterogeneity effects were the deviation of observed heterogeneous groups from their expected values based on homogeneous groups.

(C) Pareto optimization analysis for evaluating the exploration-vigilance trade-off. The diagonal arrow indicates the direction of increasing joint performance in exploration and vigilance. Strategies located in the upper-right region of the exploration-vigilance space (i.e., high exploration and vigilance) were assigned rank 1, with subsequent ranks determined iteratively. This provides a quantitative measure of how group composition influenced trade-off resolution.

Results

Group composition alters mean levels of exploration and vigilance

To assess how group composition affects individual behavior, we compared the mean values of exploration and vigilance traits across three social conditions. For exploratory behavior, homogeneous groups showed significantly higher values than single individuals (W = 247, df = 78, p < 0.001). By contrast, observed values in heterogeneous groups did not differ from their expected values, calculated as the mean of the corresponding homogeneous groups (W = 1,096, df = 65, p = 0.88). These results indicate that aggregation, but not heterogeneity, increases exploratory behavior (Figure 2A).

Figure 2.

Aggregation and heterogeneity effects on exploratory and vigilance behavior

(A) In exploratory behavior, homogeneous groups showed significantly higher values than single individuals, indicating an aggregation effect. No significant difference was observed between expected and observed values in heterogeneous groups.

(B) In vigilance behavior, both aggregation and heterogeneity effects were detected: homogeneous groups differed from single individuals, and observed heterogeneous groups exceeded their expected values. Although overall effects tended to be positive, their magnitude varied considerably across strains and strain pairs (see Figure S1). Lines connect corresponding strains or strain pairs across conditions, with colors indicating fold change: red, >2-fold increase; pink, 1.5- to 2-fold increase; blue, >2-fold decrease; light blue, 1.5- to 2-fold decrease; gray, all other cases.

For vigilance behavior, homogeneous groups showed significantly higher values than single individuals (W = 687, df = 78, p < 0.001). In addition, observed values in heterogeneous groups were significantly higher than their expected values (W = 727, df = 65, p = 0.024). Together, these results suggest that both aggregation and heterogeneity contribute to anti-predator responses (Figure 2B). Although the overall direction of these effects was generally positive, their magnitude varied substantially across strains and strain pairs (Figure S1).

Effects of aggregation and heterogeneity on the exploration-vigilance trade-off

To examine how group composition alters the relationship between exploration and vigilance, we compared correlations between these traits across social conditions. Under single-individual conditions, no significant correlation was observed (r = −0.091, p = 0.43). In contrast, a significant negative correlation emerged under homogeneous group conditions (r = −0.48, p < 0.001), revealing an exploration-vigilance trade-off. To evaluate the effects of group composition on the behavioral trade-off, we applied Pareto optimization analysis and obtained Pareto ranks of each strain (Figure 3A). Pareto ranks were significantly smaller under homogeneous group conditions than under single-individual conditions (W = 2,850, df = 78, p < 0.001), indicating better performance conferred by aggregation (Figure 3B). Under heterogeneous group conditions, vigilance and exploration were negatively correlated for both expected (r = −0.49, p < 0.001) and observed values (r = −0.50, p < 0.001) (Figure 3C), confirming the presence of the exploration-vigilance trade-off. Pareto ranks from observed values were significantly smaller than those from expected values (W = 1,435, df = 65, p < 0.001), indicating that heterogeneity within the group substantially enhanced performance under the behavioral trade-off (Figure 3D).

Figure 3.

Effects of aggregation and heterogeneity on the exploration-vigilance trade-off

(A) Relationship between exploration and vigilance under single-individual and homogeneous group conditions. A clear negative correlation emerged in groups, indicating an exploration-vigilance trade-off.

(B) Pareto optimization analysis showed that homogeneous groups achieved smaller Pareto ranks (i.e., better performance) than single individuals.

(C) Relationships between exploration and vigilance in expected and observed heterogeneous groups, both showing significant negative correlations.

(D) Pareto ranks were smaller in observed than expected heterogeneous groups, indicating trade-off alleviation driven by heterogeneity effects.

Phenotypic heterogeneity underlies the alleviation of the behavioral trade-off

To identify phenotypic differences associated with the alleviation of the exploration-vigilance trade-off, we conducted a phenome-wide association study. In this analysis, we used 190 phenotypes of the 73 DGRP strains used in this study, obtained from the DGRPool database.²⁶ Hierarchical clustering grouped the 190 phenotypes into 16 clusters, which were further classified into four higher order categories: physiology, morphology, cuticular component, and life history (Figure S2). Following the concept of a genome-wide higher level association study,²⁷ which examines associations between the genetic characteristics of a group and its group-level traits, we refer to the present analysis as a phenome-wide higher level association study (PheHAS).

We calculated pairwise distances between strains for each phenotype and examined their correlations with trade-off alleviation scores (Figure 4A). To assess whether such associations were consistent within related traits, we analyzed correlation coefficients at the cluster level (Figure 4B). In the physiological category, the activity pattern cluster showed a significant positive correlation (r = 0.18, p < 0.001), indicating that greater differences in activity patterns were associated with stronger alleviation of the exploration-vigilance trade-off. In contrast, the infection response cluster exhibited a significant negative correlation (r = −0.16, p = 0.0049), suggesting that smaller differences in infection responses correlated with better trade-off alleviation. The startle response cluster showed no significant correlation (r = 0.016, p = 0.75). At the phenotype level, waking activity (r = 0.27, p = 0.046, q = 0.88) and sleep bout duration (r = 0.40, p = 0.0030, q = 0.25) displayed detectable positive correlations with trade-off alleviation.

Figure 4.

PheHAS linking phenotypic distances to trade-off alleviation

(A) Conceptual overview of the PheHAS framework. For each strain pair, a trade-off alleviation score was quantified based on the difference between expected and observed Pareto ranks. Phenotypic distances for 190 traits were calculated from the DGRPool database for each strain pair. For each phenotype, the relationship between phenotypic distance and trade-off alleviation score was evaluated across strain pairs, yielding a correlation coefficient.

(B) Correlation coefficients were grouped into 16 phenotypic clusters and four higher order categories (physiology, morphology, cuticular components, and life history). Points represent correlation coefficients for each phenotype. Diamonds and horizontal error bars indicate pooled correlation coefficients and 95% confidence intervals for each phenotypic cluster, respectively. Several clusters (yellow diamonds) showed significant associations with trade-off alleviation, revealing phenotypic bases through which genetic heterogeneity reshapes the exploration-vigilance trade-off at the group level.

In the morphological category, clusters related to wing (r = 0.096, p < 0.001) and leg structure (r = 0.17, p = 0.0034) showed significant positive correlations, indicating that greater differences in these structures were associated with the alleviation of the exploration-vigilance trade-off. In contrast, the leg coverage ratio cluster exhibited a significant negative correlation (r = −0.25, p < 0.001), suggesting that more similar coverage ratios linked to better trade-off alleviation. No significant correlation was detected for the body structure cluster (r = 0.078, p = 0.17). At the phenotype level, the femur coverage ratio in the second thoracic segment (T2) showed a detectable negative correlation (r = −0.39, p = 0.032, q = 0.86).

In the cuticular component category, the C28–C31 cluster showed a weak positive correlation (r = 0.10, p = 0.068), indicating that greater variation in the proportion of these components was associated with the alleviation of the exploration-vigilance trade-off. Other cuticular clusters (C21–C24, C22–C29, C23–C25, C25–C29, and NI) showed no significant correlations. At the phenotype level, the proportions of C28 and Me-C27 components exhibited positive correlations (r = 0.26, p = 0.048, q = 0.88), whereas the proportion of C24 showed a negative correlation (r = −0.30, p = 0.022, q = 0.69).

In the life history category, no significant correlations were observed at the cluster level, including fecundity (r = 0.088, p = 0.23), lifespan (r = 0.042, p = 0.27), and other life-history traits (r = −0.032, p = 0.094). At the phenotype level, fecundity at 5 weeks (r = 0.39, p = 0.0039, q = 0.25) and its corrected value (r = 0.39, p = 0.0038, q = 0.25) were positively correlated with alleviation score, whereas desiccation survival (r = −0.34, p = 0.0080, q = 0.36) and abdominal pigmentation score in T6 (r = −0.34, p = 0.0094, q = 0.36) were negatively correlated.

Discussion

Aggregation and heterogeneity jointly shape group behavior. Aggregation effects are well established through density-dependent mechanisms such as the dilution effect²⁸ and confusion effect,²⁹ whereby group size alters risk and decision-making. In contrast, heterogeneity effects arise from qualitative differences among individuals, such as indirect genetic effects,^23,24 in which partner traits influence behavioral outcomes. However, most prior work has faced two limitations: conflating aggregation with heterogeneity and examining individual behaviors in isolation. These limitations obscure trade-off structures and behavioral syndromes. Using D. melanogaster, we disentangled these effects and examined their phenotypic bases.

Homogeneous groups showed greater spatial coverage than single individual conditions, indicating a clear aggregation effect on exploratory behavior. Enhanced coverage of exploration likely increases encounters with food, mates, or oviposition sites. Such improvements associated with group living may stem from social facilitation, whereby the presence of conspecifics enhances the intensity of certain behaviors.^30,31 In Drosophila, social facilitation has also been reported in contexts such as grooming³² and long-term memory,^33,34 consistent with our observations. A parallel aggregation effect was observed in vigilance: homogeneous groups showed higher immobility rates than individuals, indicating enhanced anti-predator responses through social interaction.²⁷ Immobility is a widespread defensive strategy observed across mammals,³⁵ fish,³⁶ and insects,³⁷ and it often scales with group size. The many-eyes hypothesis³⁸ proposes that larger groups detect predators earlier by distributing vigilance, while the Trafalgar effect³⁹ highlights the role of social cues in the early detection of threats not directly perceived by all individuals. Together, these mechanisms provide a plausible explanation for the enhanced vigilance observed in our study. Beyond aggregation effects, heterogeneity also shaped vigilance. Heterogeneous groups showed higher immobility rates than expected from homogeneous groups, indicating a heterogeneity effect, consistent with our prior findings.²⁷ In vertebrates, sentinel systems emerge when certain individuals monitor while others forage, thereby boosting group vigilance.^40,41 Although such structured systems are unexplored in flies, variation in sensitivity thresholds could similarly allow some individuals to benefit from the vigilance of more reactive others.

Group composition also reshaped behavioral trade-offs. No correlation between exploration and vigilance was detected in the single-individual condition, whereas a strong negative correlation emerged in homogeneous groups. We interpret the absence of a negative correlation at the individual-level in our experiment as being driven by a small subset of individuals exhibiting extreme phenotypes, particularly those with markedly low performance. Previous studies have explained similar failures to detect expected trade-offs in terms of variation in “individual quality.” Garland illustrated this phenomenon using speed and stamina, showing that behavioral trade-off may remain hidden among average performers but become evident among top athletes.² This analogy suggests that the visibility of trade-offs can depend on the overall performance level—higher performance can make previously hidden trade-offs detectable. In our study, aggregation raised both exploration and vigilance, likely driven by social facilitation and the many-eyes effect, thereby revealing the underlying behavioral trade-off. Indeed, we found that excluding a limited number of individuals identified as having extremely low values revealed a negative correlation between exploration and vigilance at the individual level (r = −0.32, p = 0.010) (Figure S3). While we acknowledge that data filtering requires careful interpretation, this result suggests that a latent behavioral trade-off may indeed exist in a singly housed condition. Pareto optimization analysis confirmed better performance under group conditions compared to individual conditions, also supporting our interpretation rather than a property unique to group contexts. Importantly, aggregation alone did not resolve the trade-off. In heterogeneous groups, observed performance exceeded expected performance despite the persistence of the trade-off, indicating that heterogeneity can alleviate the cost of balancing exploration and vigilance. While trade-off alleviation has been discussed in ecology⁴² and evolution,⁴³ it has not been explicitly demonstrated in animal behavior. Our findings therefore provide insight into how genetic heterogeneity can help resolve fundamental behavioral dilemmas.

Using our novel framework, PheHAS, we identified within-group phenotypic diversity associated with trade-off alleviation. In the physiological category, greater heterogeneity in activity patterns predicted stronger trade-off alleviation. The response threshold model^44,45 explains how inter-individual variation in task-specific sensitivity can drive the division of labor. Extensions incorporating copying behavior^46,47 and observational conditioning^48,49 further suggest that such diversity in sensitivity enables flexible responses to environmental cues. From this perspective, relatively insensitive individuals—often described as bold—may promote exploration in their groupmates, whereas more sensitive individuals—often described as shy—may induce immobility in others, allowing the group as a whole to balance exploration and vigilance. Consistent with this idea, previous studies have shown that social cues help flies (D. melanogaster) both initiate freezing in response to predator-like stimuli and terminate freezing at appropriate timing, providing a plausible explanation for the trade-off alleviation observed here.^50,51 Conversely, lower heterogeneity in infection responses correlated with a greater trade-off alleviation. In social insects, cooperation and synchronized collective behavior enhance resistance to parasites and pathogens, a phenomenon known as social immunity.^52,53 Reduced heterogeneity in infection responses may therefore reflect stronger social cohesion, potentially reinforcing social facilitation.^30,33 This amplification may be indirectly related to trade-off resolution, although this interpretation warrants caution given the small sample size of this cluster.

Beyond physiological traits, several other traits were also associated with the trade-off alleviation. Within the morphological category, heterogeneity in wing structure, leg structure, and leg coverage ratio was positively associated with trade-off alleviation and may influence locomotion (flying⁵⁴ and walking⁵⁵) as well as social behaviors (aggression⁵⁶ and grooming⁵⁷). Within the cuticular component category, only the structurally complex C28–C31 cluster—which are known to be involved in hormonal regulation and kin recognition^58,59—showed a weak positive association with trade-off alleviation. These results suggest that the effects of heterogeneity may depend on the ability of individuals to recognize genetically related conspecifics. Within the life history category, fecundity and desiccation survival were associated with trade-off alleviation, indicating that both short-term behavioral or physiological traits and long-term life-history traits can shape group performance. Nevertheless, the specific interactions underlying these associations remain hypothetical.

One important implication of this study is that the phenotypic bases underlying heterogeneity effects can be examined by focusing on group-level performance and phenotypic distances, even when individual-level mechanisms are not fully understood. At the same time, our results point to several promising directions for future work. First, experimental systems that allow individual genotypes to be tracked within heterogeneous groups (e.g., using color marker tags) will be essential for distinguishing genotype-specific enhancement from global shifts at the group level. Second, leveraging existing phenotypic data from resources (e.g., DGRPool) to predict the behavioral performance in group with unexplored strain combinations, and then empirically validating these predictions, would be valuable for evaluating the robustness and effectiveness of the PheHAS framework. Finally, experimental designs that further increase the complexity of heterogeneous group composition would provide deeper insights into collective behavior. For example, groups composed of three or more genotypes, or serial dilution designs that systematically vary genotype frequencies within groups, would allow more detailed tests of how genetic heterogeneity reshapes behavioral trade-offs.

Altogether, we show that aggregation and heterogeneity exert distinct, measurable effects on group behavior and identify their phenotypic bases. To our knowledge, this is the first study to disentangle aggregation from heterogeneity effects and to show that genetic heterogeneity can alleviate a fundamental behavioral trade-off between exploration and vigilance. PheHAS provides a generalizable framework for identifying traits associated with trade-off alleviation. These findings suggest that acquisition trade-offs, rather than being immutable constraints, can sometimes be alleviated—a perspective that may help explain how evolution overcomes behavioral dilemmas. Genetic diversity may represent one such evolutionary strategy. Elucidating the functions and mechanisms of heterogeneity effects will offer deeper insights into the dynamics of animal group behavior.

Limitations of the study

Although exploratory behavior was conceptually and computationally defined to be distinct from locomotor activity in this study, the influence of baseline activity cannot be fully excluded. Because an established method for effectively disentangling the influence of locomotor activity is not currently available, this should be considered a limitation of the present study.

In addition, a more rigorous treatment of trade-offs would require separating within-individual and between-individual covariance components.⁶⁰ Because our experimental design was not intended for such partitioning, this should also be considered an important limitation of the present study.

Resource availability

Lead contact

Further information and requests for resources used in this study should be directed to and will be fulfilled by the lead contact, Yuma Takahashi (takahashi.yum@gmail.com).

Materials availability

All DGRP lines used in this experiment are available from the Bloomington Drosophila Stock Center (BDSC) (https://bdsc.indiana.edu/).

Data and code availability

• •The behavioral dataset for group conditions generated in this study has been deposited at Figshare and is publicly available at https://doi.org/10.6084/m9.figshare.30293386. The behavioral dataset for the single-individual condition was obtained from Sato et al. (https://doi.org/10.5061/dryad.ttdz08m60). Phenotypic data for DGRP strains were obtained from the DGRPool database (https://dgrpool.epfl.ch/).
• •The analysis code used in this study has been deposited at Figshare and is publicly available at https://doi.org/10.6084/m9.figshare.30293386.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We thank Masashi Murakami for his valuable comments on the interpretation and discussion of the results. We are also grateful to the mechanical unit at our university for their kind assistance in constructing the experimental arenas. This work was supported by the Japan Society for the Promotion of Science (Grants-in-Aid for Scientific Research 22H05646 and 23H03840 to Y.T.).

Author contributions

Conceptualization, T.O., D.X.S., and Y.T.; data curation, T.O. and D.X.S.; formal analysis, T.O.; funding acquisition, Y.T.; investigation, T.O.; methodology, T.O. and D.X.S.; project administration, Y.T.; supervision, D.X.S. and Y.T.; validation, D.X.S. and Y.T.; visualization, T.O.; writing – original draft, T.O.; writing – review and editing, T.O., D.X.S., and Y.T.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used ChatGPT 5.2 in order to assist with grammar editing. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
DGRPool phenotypic database	DGRPool	https://dgrpool.epfl.ch/
Behavioral dataset for group conditions	This study	https://doi.org/10.6084/m9.figshare.30293386
Behavioral dataset for single-individual condition	Sato et al.50	https://doi.org/10.5061/dryad.ttdz08m60
Experimental models: Organisms/strains
DGRP lines	Bloomington Drosophila Stock Center (BDSC)	Table S1; https://bdsc.indiana.edu/stocks/wt/dgrp.html
Software and algorithms
R	https://www.r-project.org	Version 4.3.1

Experimental model and study participant details

Fly strains and rearing conditions

We used 78 inbred strains of Drosophila melanogaster, all derived from the Drosophila Genetic Reference Panel (DGRP), which was established from a natural population in the United States²⁵ (Table S1). Genotype information for these inbred lines is available through the DGRP resource. Flies were maintained under standardized rearing conditions (12 L:12D cycle, 25°C, ∼50% humidity) on standard cornmeal food (Bloomington Formulation, Nutri-Fly, #66–113, Genesee Scientific), which also served as the substrate for egg laying and larval development. As Drosophila melanogaster is an invertebrate model, no institutional animal ethics approval was required for this study.

To avoid sexual behavior that could interfere with responses to conspecifics and predator-like stimuli, all experiments used single-sex groups. Flies were collected 0–2 days after eclosion and reared for an additional 2 days under the same conditions. During this period, 5 males and 15 females were housed together in a single vial to ensure mating and to randomize re-mating experience. This was based on previous reports indicating that most flies are mated in natural environments⁶¹ and that mating history can significantly influence behavioral patterns.^62,63 Adult females aged 2–4 days were used for the behavioral assay. Because only females were assayed, the influence of sex on the results was not evaluated.

Method details

Behavioral assay and tracking

Behavioral assays were conducted primarily during zeitgeber time 6–12, corresponding to the daytime portion of the flies’ light-dark cycle (lights on at ZT 0, lights off at ZT 12). The incubator conditions were kept consistent with rearing (12 L:12D, 25°C, ∼50% humidity). Prior to the experiment, females were briefly anesthetized with CO₂ for less than 1 min to minimize after-effects on behavior.^64,65 They were then introduced into the experimental arenas (30 mm diameter, 2 mm depth), which were shallow enough to prevent flight.

The experimental setup followed previous methods,⁵⁰ as illustrated in Figure 1A. After an initial 30-min recovery period from anesthesia, 12 arenas arranged on a single acrylic plate were placed onto an LED board (147 mm × 115 mm, 7500 lx white light; TLB-MP, Asone Co., Japan) and given an additional 10 min for acclimation. Subsequently, fly behavior was recorded for 10 min.

During the last 5 min of recording, flies were exposed to looming stimuli consisting of 20 trials of a black circle expanding over 500 ms on a white background, presented every 15 s, following previous protocols.^50,51,66 The stimuli were displayed at 60 Hz on a 13.3-inch EVICIV monitor, tilted at a 45° angle toward the arenas.

Locomotor behavior of flies was recorded at 640 × 480 pixel resolution and ∼50 frames per second using a USB3 camera (DMK33UX290, The Imaging Source, Germany) controlled by a custom Python script. The recorded videos were cropped to isolate each arena, and fly trajectories were extracted using Flytracker.⁶⁷

Indices of exploration and vigilance

Tracked coordinates were standardized based on the position of the arena, such that the center of the arena was set to (0,0) and the diameter to 30. From the standardized coordinates, the distance traveled from the previous time point and the speed (distance divided by frame interval) were calculated at each time point and summarized into 0.5-s bins.

We then quantified both exploration and vigilance. The exploration index was measured as the evenness of spatial coverage during an unstimulated 5-min period, as previously described.²² To calculate this, the arena was divided into five concentric regions of equal area, and Simpson’s diversity index⁶⁸ was computed based on the time spent in each region. Importantly, the exploration index quantified here is conceptually and computationally distinct from general locomotor activity. While locomotor activity reflects how much an individual moves, spatial evenness captures how flies utilize the limited space.

The vigilance index was measured as the rate of immobility in response to predator-like looming stimuli during a 5-min stimulation period. Immobility was defined as a state in which an individual’s movement speed fell below a fixed threshold of 4 mm/s within 0.5 s after stimulus onset. The immobility rate was calculated as the proportion of trials meeting this criterion. In our previous study, freezing-related responses were analyzed with a primary focus on changes in locomotor speed following looming stimulus presentation.²⁷ In contrast, this study defined immobility rate from a survival-strategic perspective, focusing on the balance between exploration and vigilance. Therefore, the two studies differ fundamentally in how vigilant behavior is conceptualized and quantified.

Quantification and statistical analysis

Calculation of the aggregation effect and heterogeneity effect

To examine how fly behavior changes with aggregation and heterogeneity among individuals, we compared their behavior across three social conditions (Figure 1B). This analysis quantified aggregation and heterogeneity effects, defined as synergistic effects of interindividual interaction on group performance. We collected behavioral data from 78 inbred strains of D. melanogaster, including both single individuals and homogeneous groups (6 individuals of the same strain per group). In addition, we analyzed 65 heterogeneous groups, each composed of two distinct strains, with 3 individuals from each strain. The 65 heterogeneous combinations were drawn from a subset of DGRP lines previously shown to span a wide range of inter-strain phenotypic distances.²²

The aggregation effect for a given strain was calculated as the discrepancy between phenotypic values of single individuals and homogeneous groups, as follows:

Aggregation effect = ${\mathit{log}}_{10}\left(\frac{P_A^g}{P_A^s}\right)$, where $P_A^s$ represents the mean phenotypic value of single individuals, and $P_A^g$ represents that of the homogeneous group.

Referring to a previous study,²² the heterogeneity effect for a given mixed-strain group was calculated as follows:

Heterogeneity effect = ${\mathit{log}}_{10}\left\{\frac{P_{MIX}^g}{\left(P_A^g+P_B^g\right)/2}\right\}$, where $P_A^g$ and $P_B^g$ represent the mean phenotypic values of the homogeneous groups of the two strains comprising the heterogeneous group, and $P_{MIX}^g$ represents the phenotypic value of the heterogeneous group.

To calculate the mean phenotypic values for each social condition (i.e., $P_A^s$, $P_A^g$, $P_B^g$, and $P_{MIX}^g$), we used a linear mixed model with behavioral indices as the response variable, strain identity as a fixed effect, and arena position, plate identity, and time of day as random effects. Experiments for each strain or strain pair were conducted 17–20 times for single individuals and 3–6 times for groups, with the arena position within the plate randomized for each repetition.

Evaluating exploration and vigilance trade-off

We first defined “behavioral strategy” of a given strain and/or strain pair based on its coordinate in a two-dimensional space defined by exploration and vigilance. We then applied Pareto optimization analysis, a method widely used in engineering and economics,⁶⁹ to evaluate relative performance. Specifically, as illustrated in Figure 1C, strain pairs that had no other strategies located in the upper-right region—representing simultaneously higher exploration and vigilance—were assigned to rank 1. These rank 1 strategies were then removed, and the same procedure was repeated to assign rank 2, rank 3, and so on. Thus, smaller Pareto rank values correspond to better performance. This ranking approach allowed us to identify strategies that resolved the trade-off between exploration and vigilance. Using these ranks, we evaluated how group composition influenced the structure of this behavioral trade-off.

Uncovering the mechanism of trade-off resolution with phenome-wide association analysis

To elucidate how group composition influences the exploration–vigilance trade-off, we conducted a comprehensive analysis of the phenotypic basis underlying the trade-off alleviation, using a framework we termed phenome-wide higher-level association study (PheHAS). Specifically, we screened 190 phenotypes obtained from the DGRPool database²⁶ to identify those for which the phenotypic differences between strains were strongly associated with the degree of trade-off alleviation observed across strain pairs. Measures of variability (e.g., SE, SD, or CV) were excluded from analysis.

The trade-off alleviation score for a given strain pair was calculated as follows

Alleviation score = $R_{Expected}^g-R_{Observed}^g$, where $R_{Expected}^g$ represents the rank expected from the average phenotypic values of the homogeneous groups constituting the focal pair (calculated using Pareto optimization as described above), and $R_{Observed}^g$ denotes the actual rank of the corresponding heterogeneous group under the same trade-off.

For each strain pair, the phenotypic distance was defined as the absolute difference in a given phenotype between the two strains, across all 190 phenotypes. We computed a distance matrix based on pairwise correlations among the phenotypes and applied hierarchical clustering (Ward’s method), classifying them into 16 clusters. These clusters were further grouped into four higher-order categories based on biological relevance.

For each phenotype, we calculated Pearson’s correlation coefficient between the alleviation score and phenotypic distance across 65 strain pairs, following the method described in our previous study.²² We then applied Fisher’s z-transformation and used a random-effects model to integrate the correlation coefficients within each cluster, yielding pooled estimates and confidence intervals in a meta-analytic framework.

Statistical analysis

All statistical analyses were conducted using R version 4.3.1. Wilcoxon’s signed rank tests were performed using the ‘ggpubr’ package to evaluate differences in mean values between conditions. Pearson’s correlation analysis was performed to evaluate the association between two variables. The ‘ggeffects’ package was used to predict the mean phenotypic value of each strain or strain pair by considering random effects. We used the ‘rPref’ package to derive relative ranks along the behavioral trade-off dimension, applying Pareto optimization to two indicators: exploration and vigilance.⁷⁰