Multispecies collective waving behaviour in fish

Abstract

Collective behaviour is widely accepted to provide a variety of antipredator benefits. Acting collectively requires not only strong coordination among group members, but also the integration of among-individual phenotypic variation. Therefore, groups composed of more than one species offer a unique opportunity to look into the evolution of both mechanistic and functional aspects of collective behaviour. Here, we present data on mixed-species fish shoals that perform collective dives. These repeated dives produce water waves capable of delaying and/or reducing the success of piscivorous bird attacks. The large majority of the fish in these shoals consist of the sulphur molly, Poecilia sulphuraria, but we regularly also found a second species, the widemouth gambusia, Gambusia eurystoma, making these shoals mixed-species aggregations. In a set of laboratory experiments, we found that gambusia were much less inclined to dive after an attack as compared with mollies, which almost always dive, though mollies dived less deep when paired with gambusia that did not dive. By contrast, the behaviour of gambusia was not influenced by the presence of diving mollies. The dampening effect of less responsive gambusia on molly diving behaviour can have strong evolutionary consequences on the overall collective waving behaviour as we expect shoals with a high proportion of unresponsive gambusia to be less effective at producing repeated waves.

This article is part of a discussion meeting issue ‘Collective behaviour through time’.

1. Introduction

Animal collectives in nature often perform highly synchronized and coordinated behaviours [1,2]. Mechanistically, such coordinated collective behaviours have been explained by simple interaction rules that involve attraction and repulsion between adjacent individuals [3–6]. From a functional point of view, collective behaviour has been assumed to provide potential antipredator benefits [7,8], with the occurrence of wave-like phenomena being linked to a reduction in predation pressure for the involved individuals [9–13]. While the evolution of coordination, cooperation and sociality has been studied intensely in eusocial insects, where individuals are genetically highly related [14,15], much less is known about how coordinated and synchronized behaviours can evolve in anonymous groups of non-related animals [13,16,17].

In order to achieve collective behaviours, animals have to time and coordinate their actions with others. This is an intriguing research topic as animal collectives in nature often exhibit substantial phenotypical variation within and between individuals [18,19], which may be expected to hamper such coordination [20]. While these differences are found among individuals of the same populations and species and even among clonal individuals [21], even stronger differences among members of a collective can be assumed when looking at mixed-species aggregations. which are common in birds [22,23], ungulates [24–26] and fishes [27–29]—taxa that also provide examples of some of the largest known animal collectives [30] and are the focus of recent scientific interest [31].

Studying different species that perform collective behaviours together can provide a unique opportunity to look into the evolution of both mechanistic and functional aspects of collective behaviour. This approach may allow us to determine the relative contribution of each species to the collective behaviour in question while constraining certain evolutionary explanations. For example, inclusive fitness benefits are not possible as these can be considered only among individuals that belong to the same species and share (in part) the same alleles, as in eusocial insects or groups of kin in vertebrates. Furthermore, when species share very similar microhabitats, as can be assumed for heterospecific groups, they might also experience similar selection pressures, which consequently leads to convergent evolution of traits directly linked to the performance of the collective.

A particularly suitable system to study heterospecific collective behaviour is the sulphur molly system in Mexico [32]. Sulphur mollies (Poecilia sulphuraria) live in strongly sulfidic water that contains high amounts of H₂S and that is generally toxic to other fish species [33–35]. However, there is another fish species, the widemouth gambusia, Gambusia eurystoma, that has been reported to occur sympatrically with P. sulphuraria [33–35] and is also a member of the family Poeciliidae (see figure 1). For both species, several distinct adaptations to the severe hypoxic and sulfide-toxic conditions have been described [34]. While gambusia densities have been found to be much lower than those of the sympatric sulphur mollies [34], it is unclear to date whether both species form heterospecific shoals. This is especially relevant as both species spend considerable amounts of time at the water surface engaging in aquatic surface respiration (ASR) and, at least for sulphur mollies, extremely dense shoals have been described (greater than 2000 individuals m⁻², see [13]). These gigantic shoals attract many piscivorous bird species, which represent the only known fish predators in this system [13,36–38]. As an adaptation to high rates of bird predation, sulphur mollies perform coordinated collective dives that are repeated for minutes and cause wave-like water ripples, which were causally linked to a reduction in attack frequency and success by bird predators [13]. However, the extent to which the co-occurring gambusia contribute to these collective waves is unknown.

Figure 1.

Widemouth gambusia (Gambusia eurystoma, left) and sulphur molly (Poecilia sulphuraria, right) in their natural habitats. (Online version in colour.)

It is assumed that the diving behaviour in response to predators first evolved as selfish escape in individuals [13]. Diving responses in nearby neighbours presumably provided important information regarding the presence of predators and led to the evolution of collective diving, which translates into surface waves when fish hit the water surface with their caudal fins during their initial escape movement. While these highly visible waves cause piscivorous birds to delay their attacks or to leave the area even before launching an attack [13]; the mechanism remains unknown. It might be that birds visually track the waves that run at the surface (instead of the fish that dive down) and that the repeat waves ensure that there is continuous motion for the bird to be distracted by. Another possibility is that the waves act as a perception awareness signal [13]. The prey sending the signal could benefit because they do not have to invest in costly escape behaviour if the predator leaves. In turn, the predator benefits because it does not waste energy on attacking prey that is harder to catch because it is highly vigilant [39,40]. Little is known about the mechanisms and functions of collective behaviours that occur in predator–prey relationships in the field [10,41,42], and the co-occurrence of sulphur mollies and gambusia provides an opportunity to investigate both aspects in the context of heterospecific shoaling.

We investigated in a field survey whether both species co-occur in the same shoals by taking photographs of fish shoals at the surface and characterizing body size distributions from catches from several years. In a set of subsequent laboratory experiments, we explored the mechanisms underlying potential associations between the two species. First, we compared the oxygen-dependent ASR rates to find out at what oxygen concentration both species typically rise to the surface. Given that both species co-occur in a fast-flowing river, we also tested in a second experiment for differences in the maximal sustained swimming speeds (U_crit), because swimming performance might be another factor that constrains heterospecific shoals. Third, we analysed how conspecific as well as heterospecific groups respond to simulated predator attacks to gain insights into the degree to which both species contribute to collective waves.

2. Methods

(a) . Study site and study system

Our study took place at a sulfidic spring complex located within the area of the Hacienda ‘Los Azufres’ near the city of Teapa, Tabasco, Mexico (‘Banos del Azufre’ site, 17°33′ N, 93°00′ W). Here a freshwater river merges with the outflows of springs fed by sulfidic groundwater aquifers that contain high concentrations of hydrogen sulfide (H₂S, up to 990 µmol l⁻¹; see [34]) generated from volcanic deposits [32,38]. After the inflow of the sulfidic spring water, the river stays sulfidic for at least 2.5 km [34], and within the sulfidic part of the river, a diurnal pattern of oxygen levels associated with temperature changes has been described. The sulfide levels vary both during the day and along the sulfidic part of the river but are more constant close to the sources (above 170 µmol l⁻¹ for the first kilometre at least [34,38]).

(b) . Body size distribution

In order to establish body sizes of both species, we seine-fished in April (2017–2019) in the main river and photographed the captured fish on millimetre paper. Body sizes (standard length (SL) in mm) were compared between species with a linear mixed model (LMM) that included species as fixed factor and year of observation as a random subject factor (Gaussian error distribution and identity link function, restricted maximum likelihood (REML) estimation).

(c) . Species composition of shoals

To investigate the composition of fish shoals in their natural habitats, we took photographs of shoals at the surface with a Canon 400D SLR camera and a 200 mm zoom objective, which allowed proper species identification of individuals. We analysed N = 154 photographs taken in April 2018. Photographs used for analysis were taken during the course of the day (8.00–17.00) and over a period of 14 days. We photographed shoals whenever we saw at least 10 adult fish (body size above 1.5 cm) within a frame of roughly 50 × 50 cm at the water surface. Recording locations were chosen along the river where the shore line allowed photographs to be taken almost vertically down. When pictures were taken at the same location, we left at least 5 min between recordings to ensure sufficient time for fission–fusion of shoals and for individuals to change positions within shoals. We used a custom computer vision processing pipeline to mark species ID as well as individual position and orientation within each picture. This was done by first drawing a line manually from the midpoint of the mouth tip to the end of the caudal fin for each fish of each species. Second, these manual annotations were extracted and processed by the software. To ensure correct species identification, three researchers evaluated each picture independently. Individuals that could not be categorized as either P. sulphuraria or G. eurystoma by all three researchers were scored as ‘unidentified’ (317 (2.35%) out of 13 488 total fish visible). Based on the manually annotated positions and orientations, the software calculated the nearest neighbour distance (NND) and the relative swimming polarization (tail to head centred unit vector) for each individual in a respective picture. We further calculated the relative frequency of sulphur mollies in a picture as the proportion of molly number to total number of fish that could be identified (N_molly/(N_molly + N_gambusia)), average NND, average polarization (mean unit vector) and average number of neighbours within two body lengths (average degree). We computed the proportion of mollies in all 154 pictures (93.3%) and compared it with the proportion that is obtained if we only use those pictures where all fish could be identified (93.2%). Based on the similarity of proportions we decided that our analysis is robust against the small number of unidentified fish and used all 154 pictures for all reported results. We used Spearman's correlation to quantify relations among the listed variables. To compare the proportions of mollies per picture we tested for equality of proportions using the function prop.test in R. For this test, we made the conservative assumption that 93.3% of the unidentified individuals in each picture were mollies, because this was the observed overall percentage of mollies.

(d) . Oxygen-dependent aquatic surface respiration

To understand how much time each species spends at the surface and how this relates to oxygen concentration in the water, we investigated the oxygen-dependent ASR. The procedure followed the experimental protocol previously published in Lukas et al. [38]. All experiments were performed in a field laboratory close to the field site at Hacienda ‘Los Azufres’. We collected individuals of both species from the El Azufre river and left them in insulated coolers for at least 1 h prior to testing. During this phase, fish were held in aerated water (5.5 mg l⁻¹ dissolved oxygen (DO)) in order to minimize effects of differences in prior oxygen conditions among tested fish. In the test tank, water was taken from the El Azufre river and aerated to the desired DO level under constant mixing before testing with commercially available battery-driven air pumps. We exposed fish to DO concentrations ranging from near-anoxic to normoxic conditions (0.6–5.3 mg l⁻¹ DO), which is representative of the conditions in their river habitat. DO and temperature (average 27.8°C) were monitored with a multiprobe (OxyGuard Polaris 2) directly before and after each trial to calculate a mean treatment value (average increase of DO during the trials: +0.15 mg l⁻¹ DO; 0.0–0.3 min. to max. increase). Water was exchanged after each trial and testing was done under natural light conditions indoors through a window but without direct sunlight. To ensure that fish of both species were exposed to the same water conditions, the test arena (30 × 20 × 30 cm, water depth of 20 cm) was separated into two equally sized compartments by an opaque permeable plastic separator. To initiate a trial, for each species a group of four fish was netted haphazardly and introduced into one of the two compartments. A trial lasted 15 min and each fish's position in the water column was recorded with a camera (GoPro Hero6). Each group was used once, and thus only exposed to one oxygen concentration. As part of the acclimation protocol, we did not analyse the first 5 min of each trial to ensure fish had recovered from handling and resumed normal swimming behaviour. Owing to technical difficulties with the GoPro battery duration, four trials only lasted 10 min. We analysed the videos using EthoVision 12 (Noldus Information Technology). The front view of the water column allowed monitoring of fish's depth position, but the view of individual fish was sometimes occluded as they moved slowly along the surface, and thus did not allow trajectories with individual identity. As a prerequisite for ASR, fish need to have direct surface contact and we thus defined the region less than 1 cm below the water surface as the zone of interest, which was tracked for the presence/absence of fish and corrected manually. We note that this approach may slightly overestimate ASR duration (e.g. owing to brief transitions during up- and downward diving); nonetheless, we deem this negligible considering the overall observation period of 10 min. We quantified the cumulative time spent at the surface by all four fish and calculated an average percentage surface time per group. For each species, we obtained data for 19 independent groups (N = 152 individual fish in total). Data were analysed with a dose–response model (variable slope) separately for both species with the formula: Y = Bottom + (100 − Bottom)/(1 + 10^{((logEC₅₀ − X) × HillSlope)}) in GraphPad Prism 9.3. The log(EC₅₀) value then depicts the half-maximal DO concentration (ASR₅₀).

(e) . Maximum sustained swimming speed (U_crit)

Both gambusia and mollies live in environments with strong variation in flow rates, which means that swimming speeds could be a strong determinant of whether they occur together in their natural habitat [34]. We measured maximal locomotor capacity as the critical sustained swimming speed (U_crit) according to published protocols for poeciliid fishes [43,44]) of individuals of both species caught the same day in the same locations. U_crit was measured in a Blazka-style swimming flume consisting of a cylindrical clear Perspex flume (150 mm length and 38 mm diameter). The flume was fitted tightly over the intake end of a submersible pump (12 V DC, iL500, Rule, Hertfordshire, UK). A bundle of hollow straws at the inlet end of the flume helped maintain laminar flow. Two of those flumes and pumps were submerged in a Sterilite plastic tank (38 × 62 cm) that contained continuously aerated water (DO 5.3 mg l⁻¹ ± 0.2 s.d., T 28.4°C ± 1.5 s.d.). We controlled water flow speed by changing the voltage input into the pump with a variable DC power source (NP9615, Manson Engineering Industrial, Hong Kong, SAR China). The water flow in each flume was measured in real-time by a flow metre (DigiFlow 6710M, Savant Electronics, Taichung, Taiwan) connected to the outlet of each pump. We tested two fish (one of each species) at the same time and started with an initial flow rate of 0.08 m s^–1 for 10 min followed by an increase in flow speed by 0.04 m s^–1 every 5 min until the fish could no longer hold their position in the water column or a total duration of 60 min was reached. When fish fell back onto the grid, the flow was stopped for 5–10 s before restarting and increasing the speed to the last setting again. We terminated the trial when fish stopped swimming for the second time at the same speed setting. After a trial, fish were measured for body size (P. sulphuraria = 27.1 mm ± 2.2 s.d. (N = 15), G. eurystoma = 26.7 mm ± 1.8 s.d. (N = 15); paired t-test: t₁₄ = −0.54, p = 0.59) and released them back into their capture site. We report U_crit as body length per second (BL s^–1) and compared species' performances using a linear model (LM) with 'species' as fixed factor and 'BL' as a covariate. We initially included an interaction term between both independent variables but removed it as it was not significant (F_1,26 = 1.11, p = 0.301) and did not improve model fit.

(f) . Antipredator dive responses in con- and heterospecific shoals

To understand to what degree mollies and gambusia contribute to collective waves, we compared the anti-predator dives in response to a simulated bird attack in con- and heterospecific shoals, i.e. shoals of four mollies or four gambusia in the conspecific treatment and shoals of two mollies and two gambusia in the heterospecific treatment. A typical diving sequence of the sulphur molly consists of an initial fast-start triggered by a simulated bird attack, a small period of hovering during which the diving velocity decreased significantly, and subsequent resurfacing, or, in some cases, a second dive (see details in [37]). As sulphur mollies perform these dives when performing ASR, e.g. when at the surface [37], we tested our groups at DO levels low enough that both sulphur mollies and gambusia were found at the surface performing ASR. Under these conditions, diving can be induced through a visual stimulus that simulates an attacking bird silhouette (see below). For sulphur mollies, similar experiments revealed almost 100% dive response rates when tested in small groups [37]. Water for the experiment was taken from the source pool and DO and temperature were monitored with a multiprobe (OxyGuard Polaris 2). Water parameters did not differ between trials (average water temperature: 29.9°C; LMM F_1,22 = 0.005, p = 0.99; average DO: 0.5 mg l⁻¹; F_1,22 = 0.36, p = 0.71). Water was exchanged after each trial and testing was done under natural light conditions in a field laboratory close to the site of fish capture.

This experiment consisted of two consecutive parts, in which single-species fish groups were first presented with a set of simulated aerial attacks and in a second part this stimulation was repeated with the fish being in a mixed-species group. We tested fish in mixed-sex groups of four adults. We selected size-matched fish to minimize size differences between species and sexes (LMM: species: F_1,54 = 0.40, p = 0.5; sex: F_1,57 = 1.57, p = 0.2; overall average body size = 20 mm) as well as between test groups (between-group variance estimate not significant).

To start a trial, fish were introduced into the test tank (30 × 20 × 30 cm, water depth of 20 cm) and were acclimated to the arena for 10 min, a period that pilot tests identified as sufficient for the fish to resume their natural swimming behaviour (i.e. fish swam close to the surface, performing ASR; [38]). After the acclimation to the test tank, we started to present the single-species groups with a so-called looming stimulus [45] consisting of a black dot quickly increasing in size presented on a Tablet (ACEPAD A121, 10.1 zoll, 3G Tablet) placed centrally on top of the test aquarium. We confronted the fish with 11 consecutive identical stimulations all separated by 1 min (counting when all fish reached the surface again after the prior stimulation). After this first part, we caught all fish from the test tank and placed them back in mixed-species groups and repeated the loom stimulation for another 11 times. After the experiments, all fish were released directly into their capture site.

Experiments were recorded using a Canon XF200 camcorder (full HD resolution at 50 frames s⁻¹) facing the front of the test tank. Diving trajectories in the XY-plane were tracked with EthoVision 12 (Noldus Information Technology, Wageningen, The Netherlands). Individual IDs were kept constant throughout a dive (i.e. tracking started 1–2 s before stimulus application and lasted until all fish had resurfaced), but could not be tracked throughout a full experiment (i.e. between stimulations) as fish were visually occluded while swimming at the surface. The resulting trajectories were manually checked and corrected for minor tracking errors and processed using a custom Python script to perform smoothing by moving average (i.e. window width of three frames) and to extract variables of interest. We acknowledge the loss of positional information on the Z-axis owing to two-dimensional tracking. While this constitutes a systematic error across treatments, we minimized it by (i) confining fish to the front wall by means of a bended white plastic sheet inserted into the arena, which promoted movement in the XY-plane (see [37]), and (ii) choosing response parameters that were robust against this type of error. From the processed data, we assessed the diving probability as the proportion of fish that responded with diving down after a stimulus had been presented. For all diving fish, we extracted additional diving depth until reaching the hovering point, i.e.. the depth at which they stopped after the stimulation.

In order to compare parameters among con- and heterospecific shoals, we fitted (generalized) LMMs with the fixed factors ‘species’ (two levels: P. sulphuraria or G. eurystoma) as well as ‘social context’ (two levels, conspecific versus heterospecific), and 'trial number' as a covariate (continuous from 1 to 11). We included 'group ID' as a random factor to account for repeated testing of groups. As our experiments were further performed in an arena with two identical compartments (see above), group ID was nested within the arena. For analysing the diving probability, we used a generalized model with binomial error and logit link. We initially included the interaction term ‘species by social context’ but removed it as it did not improve the model fit based on corrected Akaike information (AICc) scores (ΔAIC = 409). We used a model with Gaussian error and identity link for the diving depth. We included the interaction term ‘treatment by trial’ as it improved the model fit.

As the probability of diving after a stimulation was close to 100% for sulphur mollies, regardless of the social environment (see Results) and the probability of the gambusia diving was independent of treatment, always below 50%, we further asked whether the sulphur mollies might adjust their diving depth as a response to the gambusia diving or not in the mixed-species treatment. We thus compared the diving depth of sulphur mollies when their gambusia partners were diving with when the gambusia were not diving. For this subsequent analysis, we used only the subset of P. sulphuraria in the mixed-species treatment and another general linear mixed model (GLMM) with similar random effects structure to that described above. We included the fixed factor ‘gambusia diving’ (two levels; yes or no) and ‘trial’ as a covariate in our model. The interaction term of our fixed factor with trial was not included in the final model (based on AIC scores; ΔAIC = 3.5) (figure 1).

All data and SPSS (IBM Software) analysis code are provided as electronic supplementary material.

3. Results

Sulphur mollies were the more common species compared with gambusia based on surface shoal photographs (N_molly = 12 282 individuals, 93.3%; N_gambusia = 889, 6.7%; N = 154 shoal pictures analysed). This numerical dominance of sulphur mollies in surface shoals is further supported by the fact that mollies were absent in only seven of the analysed pictures (4.5%), while 77 photographs did not include a single gambusia (50%, figure 2a). The proportion of sulphur mollies per picture differed significantly (test for equality of proportions, χ² = 7757.1, d.f. = 153, p < 0.001, figure 2b), which means the fish did not mix randomly, but assortatively by species. The trend for the proportion of gambusia to decrease with shoal size (figure 2b) is a by-product of the fact that there are far fewer gambusia than mollies in the population, in combination with the assortative behaviour. In photographs with a numerical dominance of mollies, fish were significantly more polarized and more dense (lower NND and higher neighbour count within 2 BL radius, figure 2d–f) than in shoals containing more gambusia.

Figure 2.

(a) Species composition of shoals. Each column represents a single picture (N = 154). (b) Mixing of sulphur mollies and widemouth gambusia in relation to the observed number of fish per picture. Shown are observed proportions of mollies (blue) as well as the intervals between the 0.025 and the 0.975 quantiles (orange) expected for random mixing. (c) Body size distributions of Poecilia sulphuraria and Gambusia eurystoma caught at the same river locations. (d) Polarization, (e) nearest neighbour distance (NND) and (f) neighbour count within 2 BL in relation to the proportion of mollies in the respective photo. (Online version in colour.)

Sulphur mollies were on average 3.4 mm larger than widemouth gambusia that were caught at the same time and location (P. sulphuraria: N = 1789; average size (SL) = 22.5 mm; min.–max.= 10.4–49.2 mm; G. eurystoma: N = 1372; average size = 19.1 mm; min.–max. = 10.4–31.2 mm; LMM: species: F_1,1199 = 375.5, p < 0.001; figure 2c).

(a) . Oxygen-dependent aquatic surface respiration

When exposed to DO levels below 1.3 mg l⁻¹, P. sulphuraria switched to full ASR behaviour (R² = 0.69, median surface time of the groups below DO 1.3 mg l⁻¹ = 89.91%; median above DO 1.3 mg l⁻¹ = 1.1%). By contrast, no change in behaviour was detectable for G. eurystoma, which kept swimming at the surface during the full range of DO levels (R² = 0.13, overall median surface time = 97.28%; figure 3a).

Figure 3.

(a) Oxygen-dependent ASR tendencies of Poecilia sulphuraria and Gambusia eurystoma. Time spent at the water surface performing ASR increased with decreasing oxygen concentrations for mollies while gambusia spent most of their time at the surface throughout the tested oxygen range. Each data point equals the cumulative time spent at the surface for one group of four adult fish (n = 19 groups per species). The half-maximal concentration at which fish spend 50% of their time at the surface (ASR₅₀) was estimated at 1.3 mg l⁻¹ DO for mollies. (b) Maximum sustainable swimming speed (U_crit) for mollies and gambusia in relation to body size. (Online version in colour.)

(b) . Maximum sustainable swimming speed (U_crit)

Sulphur mollies showed a significantly higher maximum sustainable swimming speed (U_{crit,Poecilia} = 11.1 BL s⁻¹) than gambusia (U_{crit,Gambusia} = 6.8 BL s⁻¹; LM; species: F_1,27 = 26.9, p < 0.001; figure 3b). For both species, maximum sustainable swimming speeds were dependent on body size, with smaller individuals maintaining higher U_crit values (body size: F_1,27 = 5.4, p = 0.027; figure 3b).

(c) . Anti-predator dive responses in con- and heterospecific shoals

We found a significant difference in the probability of gambusia and mollies diving after a simulated predator attack (GLMM; factor ‘species’: F_1,1404 = 310, p < 0.001), which was independent of the social context (factor ‘social context’: F_1,1404 = 2.22, p = 0.14; figure 4a). On average, gambusia dived down with a 50% probability in both con- and heterospecific contexts, while mollies were diving down with above 97% probability in both contexts after a simulated attack. Interestingly, for those gambusia that dived down, no difference in the diving depth was found regardless of whether they were in con- or heterospecific contexts, while diving mollies significantly reduced their diving depth when together with gambusia (GLMM; interaction term ‘species by social context’: F_1,943 = 11.2, p < 0.001; factor ‘species’: F_1,943 = 147, p < 0.001; factor ‘social context’: F_1,943 = 18.6, p < 0.001; figure 4b). The repeated testing (trial) had no significant effect on either diving probability or depth (diving probability: F_1,1404 = 3.44, p = 0.064; diving depth: F_1,943 = 0.07, p = 0.79).

Figure 4.

(a) Proportion of fish diving after a simulated bird attack in con- and heterospecific shoals of Poecilia sulphuraria and Gambusia eurystoma. (b) Dive depth of both species. (c) Dive depth of only mollies in the mixed-species treatment when at least one gambusia was diving compared with none. In (b) and (c), only those individuals that performed a dive were included; bars denote means and error whiskers standard deviations. Asterisks show significance of p < 0.01 from post hoc comparisons; n.s., not significant.

As the less deep dives of mollies in the heterospecific context could be due to companion gambusia not diving, we compared the depth of molly dives when at least one gambusia in their test group was diving with cases when no gambusia joined the dives. We found that mollies dived less deep when gambusia did not join their dives (GLMM; factor ‘gambusia diving’: F_1,329 = 26.0, p < 0.001; trial: F_1,329 = 0.001, p = 0.98; figure 4c).

4. Discussion

Widemouth gambusia and sulphur mollies caught at the same locations in the wild had a large body size overlap and were frequently found together in surface shoals, with mollies often making up a much larger proportion of the shoal members. Both species responded with rapid dives to the looming stimulus simulating a bird attack, but in the mollies this response was much more frequent and associated with deeper dives. Interestingly, gambusia were not influenced by mollies in their dive behaviour, whereas mollies dived less when in the presence of non-diving gambusia, which suggests that mollies are highly responsive to public information even from heterospecifics.

Bird attacks are very frequent in the sulfur system investigated here. Several bird species hunt fish in this system, with peak bird activities during noon and afternoon, when O₂ levels force fish to spend their time at the surface performing ASR [13,36–38]. As previously described for the sulphur molly, fish respond to attacking birds by performing repeated collective dives that delay predator attacks, drive predators off altogether and in some bird species reduce the capture success of subsequent attacks [13]. The occurrence of mixed-species shoals of sulphur mollies and widemouth gambusia at the surface of the sulfidic El Azufre river suggests that both species contribute to these repeat waves when attacked by avian predators. This fact has been previously unnoted likely due to the numerical abundance of mollies compared with gambusia (see [34] for a similar observation), as well as the very similar body size, body shape and coloration of both species when observed from the outside the water.

Gambusia seem to be much more bound to the water surface than sulphur mollies as they do not leave the surface even when oxygen levels are high and no ASR behaviour is needed [38]. This surface preference of gambusia is accentuated by the fact that gambusia respond much less with diving than sulphur mollies when presented with a simulated bird attack. The responsiveness in gambusia was not affected by the presence of diving mollies, which in turn dived less deep when confronted with the predator stimulus in a heterospecific context, especially when gambusia remained at the surface. First of all we can speculate that this is indicative of a species-specific discrepancy in the use of social information, with sulphur mollies being more susceptible to the actions of their neighbours than gambusia. Differential responses to avian predation have also been reported in three-spined sticklebacks, Gasterosteus aculeatus [46], where individuals that were parasitized by Schistocephalus showed a greatly reduced or no response to simulated heron attacks. Unparasitized sticklebacks dived less when in the presence of parasitized fish, but parasitized individuals were not influenced by unparasitized ones [46] resulting in potential consequences for the escape behaviour of fish shoals [47]. In Schistocephalus-infected sticklebacks and other fishes carrying parasites with vertical transmission, a reduced or absent response to predators has been interpreted as a way in which the parasite increases the transmission probability to its final host (which is often a bird predator [48]). Nevertheless, a reduced responsiveness to social information in gambusia cannot fully explain why this species per se dived less in response to an approaching predator, especially given that gambusia spend so much time at the surface. We thus postulate two non-mutually exclusive explanations: First, we conducted dive experiments at low oxygen concentrations typical of the conditions in the field when frequent waving behaviour is seen in fish shoals. It thus remains possible that gambusia are more oxygen-limited than mollies and therefore require higher oxygen levels for more frequent dives. This could be connected to the fact that sulphur mollies have been reported to have better adaptations to the toxic sulfidic environment than gambusia [34,35]. Second, gambusia might be less preferred by predators, because of their smaller body size or by being harder to detect owing to their more cryptic coloration. As a result, they might face a lower risk of predation, which eventually led to the observed decrease in predator-avoidance reactions [49,50]. For example, in the closely related guppy (Poecilia reticulata), risk-sensitive differences in anti-predator behaviours are known [51], and similar observations have also been made in sulphur-adapted mollies, including the sulphur molly, when confronted with piscivorous and non-piscivorous fishes [52]. In addition to these hypotheses, a detailed habitat analysis at the Banos del Azufre showed that gambusia were positively associated with flow patterns and fine sediment, while sulphur mollies did not show specific associations with particular environmental factors [34]. However, this does not preclude the possibility of micro-habitat differences between the species, such as distance to the river bank, use of cover or other micro-structures, that may have consequences for predation risk and/or the ASR and diving behaviour. The strong assortedness by species in shoals could be a consequence of such habitat preferences (unless the fish actively prefer shoals with a higher proportion of conspecifics). Thus, we propose future experimentation that aims at elucidating the observed species differences in the reaction to bird predation in general as well as the use of social information.

Given that collective behaviours are assumed to largely depend on activity synchronization [53] and coordination, we might expect shoals with a higher proportion of unresponsive gambusia to be less effective at producing repeat waves capable of delaying predator attacks or driving predators off altogether. The predicted relationship between shoal species composition and waving performance (in terms of wave number but also speed and size) and the consequences for antipredator effects provide exciting directional hypotheses for future fieldwork in the area of collective behaviour and its evolution. For example, it is unclear whether sulphur mollies might discriminate against gambusia shoal members, possibly by aggregating in stream areas with higher flow rates in which gambusia might not be able to swim for long given their lower average maximal sustainable swimming speeds (see for example studies on guppies where females prefer deeper areas with stronger flow rates so that the harassing smaller males cannot follow [54]). Consistent with this assumption is the fact that shoals with higher proportions of sulphur mollies exhibit stronger polarization of individuals, which indicates that fish are swimming in stronger currents. Future studies should further aim at elucidating how robust the waving behaviour of the heterospecific fish shoals will be when the proportion of less diving gambusia individuals increases.

Further, it is an intriguing question as to whether gambusia gain antipredator protection from sulphur molly waves even when they do not participate themselves. It is conceivable that the waves produced by mollies redirect the attention away from the gambusia to the moving wavefront. This is likely because the wavefront is much more conspicuous than individual fish. Another possibility is that the white wavefront creates such a strong contrast with the surrounding water that it reduces the ability of bird predators to detect fish that remain at the surface in the short intervals between waves [55]. Eye-tracking in birds [56,57] would provide an exciting possibility to find out where their attention focuses and thereby could give insights into the selective processes at play.

Ethics

Experiments reported in this study were carried out in accordance with the recommendations of ‘Guidelines for the treatment of animals in behavioral research and teaching’ [58] and under the authorization of the Mexican government (H. Ayuntamiento Constitucional Tacotalpa, Direction Fomento Economico y Turismo; DGOPA.09004.041111.3088, PRMN/DGOPA-003/2014, PRMN/DGOPA-009/2015 and PRMN/DGOPA-012/2017, PRMN/DGOPA-2018 issued by SAGARPACONAPESCADGOPA).

Data accessibility

The datasets supporting this article have been uploaded as part of the electronic supplementary material [59].

Authors' contributions

J.L.: conceptualization, data curation, formal analysis, investigation, methodology, project administration, supervision, validation, writing—review and editing; J.K.: conceptualization, funding acquisition, investigation, project administration, resources, supervision, validation, writing—original draft, writing—review and editing; A.S.T.: data curation, investigation, methodology, writing—review and editing; J.M.P.: investigation, methodology, software, writing—review and editing; P.R.: funding acquisition, investigation, methodology, project administration, resources, software, writing—review and editing; H.S.: conceptualization, funding acquisition, writing—review and editing; L.A.-R.: investigation, project administration, resources, writing—review and editing; S.K.: data curation, formal analysis, software, validation, writing—review and editing; C.S.: investigation, writing—review and editing; D.B.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, validation, visualization, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed herein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

We acknowledge funding by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy—EXC 2002/1 ‘Science of Intelligence’—project number 390523135 (J.K., P.R. and H.S.), DFG—‘Sachbeihilfe’—BI 1828/3-1 (D.B.), DFG—Emmy Noether Programme—RO 4766/2-1 (P.R.), the Alexander von Humboldt Foundation (C.S.) and the Elsa-Neumann Scholarship (J.L.).