Effect of density and species preferences on collective choices: an experimental study on maggot aggregation behaviours

Evolution of animal species has resulted in various forms of sociality, differing in complexity. In the simple case, social interactions involve only inter-attractions that eventually lead to group formation (i.e. aggregation) (Parrish and Edelstein-Keshet, 1999). These aggregations can reduce the risks of predation, improve foraging and protect individuals against environmental variations. Aggregation can also facilitate information transfer between individuals (Parrish and Edelstein-Keshet, 1999; Krause and Ruxton, 2002) and lead to collective decision making, i.e. the selection of one among several alternatives through social interactions (Conradt and Roper, 2005; Jeanson et al., 2012).

Collective decisions usually result in the maintenance of group cohesion and an increase of decision speed and/or accuracy (i.e. the probability of choosing the best alternative) (Conradt and Roper, 2005; Jeanson et al., 2012). While arthropod species with complex social structures (e.g. eusocial insects) use active recruitment and differential signalling (e.g. trail-laying in ants), collective decisions can also emerge through self-organized processes, involving only local interactions with no need for elaborate perceptive or cognitive abilities (Bonabeau et al., 1997; Sumpter, 2006; Jeanson et al., 2012). For instance, collective decision making for a shelter in cockroaches relies only on the modulation of individual resting time as a function of shelter quality and the number of conspecifics (Deneubourg et al., 2002; Lihoreau et al., 2010).

Collective decisions can thus be strongly influenced by group size and inter-individual variability (Jeanson et al., 2012). An increase in group size generally leads to an increase in decision speed and/or accuracy (Lihoreau et al., 2010, 2016; Canonge et al., 2011; Szopek et al., 2013), but can also trap individuals on sub-optimal alternatives (Dussutour et al., 2007). Inter-individual variability (e.g. personality) can also affect both decision speed and accuracy (Planas-Sitjà et al., 2018) and can reduce the risk of choosing a sub-optimal option (Despland, 2013). This assumption could apply to groups composed of several species, which are frequently observed in arthropods and are known to provide adaptive benefits (Boulay et al., 2019; Goodale et al., 2020). However, the influence of species on collective decisions remains understudied (Leoncini and Rivault, 2005; Boulay et al., 2016; Broly et al., 2016).

The goal of this study was to investigate collective decision-making processes in necrophagous blow fly larvae (Diptera: Calliphoridae). In blow flies, adults lay eggs on vertebrate cadavers, on which newly hatched larvae feed until reaching a critical size, allowing pupation. During the feeding stage, larvae form groups including dozens to thousands of individuals (Rivers et al., 2011). These maggot masses mostly result from attractive and/or retentive social cues (Boulay et al., 2013; Fouche et al., 2018). Blow flies are also attracted by other species, resulting in interspecific aggregates (Woodcock et al., 2002; Joy et al., 2006; Slone and Gruner, 2007; Fouche et al., 2018; Komo et al., 2019). Feeding aggregations can be advantageous for larvae: they generate heat (larval-mass effect; Charabidze et al., 2011) and increase exodigestion (Wilson et al., 2016; Scanvion et al., 2018), which speeds metabolic activity and facilitates food assimilation. These effects reduce mortality and accelerate development, resulting in an Allee effect (Scanvion et al., 2018).

The intense competition between microbes, necrophagous arthropods and vertebrate scavengers for carrion consumption leads to rapid nutrient depletion and the risk of failing to complete development (Campobasso et al., 2001; Devault et al., 2003; Benbow et al., 2015). In this context, the ability to make collective decisions, observed in both conspecific and heterospecific groups of blow fly larvae (Boulay et al., 2016), could allow fast and efficient location of the best food sources on the cadaver and confer a significant competitive advantage. This study aimed to test three hypotheses: (1) blow fly larvae are able to choose the best food source in their environment, (2) their decision speed and/or accuracy increases with increasing larval density, and (3) these factors are modulated by interspecific differences. Two species were studied, Lucilia sericata (Meigen) and Calliphora vicina (Robineau-Desvoidy 1830) (Diptera: Calliphoridae). In a first step, the growth and survival of larval groups of different sizes and species composition were measured on fresh or rotten beef liver. In a second step, groups of L. sericata and C. vicina larvae were tested in binary choice experiments opposing fresh and rotten liver, and collective foraging dynamics were compared between groups of different density and species composition.

Flies of L. sericata and C. vicina were collected in the surroundings of Lille (Hauts-de-France, France) and then reared separately in the laboratory. Rearings were composed of tulle cages (50×50×50 cm), stored at room temperature (25±2°C) in daylight, in which water and caster sugar were provided ad libitum. New individuals from the field were added to the cages 4 times a year. To provide proteins required for oocyte maturation, pork heart was placed inside the cage for a week, and then placed daily for only 2 h to trigger egg laying. Eggs were deposited on 10 g of fresh minced beef liver in a plastic box (108×83×64 mm) and stored in an incubator (Pol-Eko-Aparatura, model ST BASIC) at either 25±1°C (for development tests) or 30±1°C (for binary choice tests). Young second instar larvae used for experiments were collected 24±2 h after egg laying. For binary choice tests, C. vicina larvae were used 30±2 h after egg laying, as preliminary observations showed a high mortality for younger larvae (C. vicina being better adapted to cooler temperatures; Donovan et al., 2006).

Development tests were adapted from Scanvion et al. (2018). In short, larvae were placed in plastic boxes (108×83×64 mm) with 50 g of minced beef liver stored in an incubator (Pol-Eko-Aparatura, model ST BASIC) in the dark at 25±1°C. Post-feeding larvae were transferred on a 2 cm layer of sand in new small boxes and kept in the incubator until fly emergence. Flies were left inside the box until their natural death and were then collected, counted, dried and all weighed simultaneously (Kern ALT 310-4 precision balance).

Three larval densities were tested: 100, 250 and 500 individuals per box (respectively labelled d100, d250 and d500). For conspecific groups, all larvae were of the same species, either L. sericata or C. vicina. For heterospecific groups, each density comprised half L. sericata and half C. vicina larvae (i.e. 50 L. sericata+50 C. vicina, 125 L. sericata+125 C. vicina, and 250 L. sericata+250 C. vicina). Liver (i.e. larval food) was initially kept in a freezer and either directly used after thawing (fresh liver) or left at 25±1°C in vacuum packaging for 7 days (rotten liver). This last treatment allowed decomposition in anaerobic conditions, which was previously shown to slow down larval development and growth of C. vicina (Richards et al., 2013). All conditions (18 in total) were repeated 9 times, except d500 conditions, which were repeated 6 times.

The spatial distribution of larvae over time was observed in a circular arena showing two food spots of different quality. The arena, a glass Petri dish (Pyrex^®, 2 cm height, 20 cm diameter), was filled with an agar solution (4%, 1 cm height; Fig. 1). To create food spots, two holes were dug in the agar, with 7.5 cm between them and at an equal distance from the edge of the arena. Each spot had an equal size of 1 cm height and either 2 cm (small spots) or 3.45 cm (big spots) diameter. Both holes were filled with 3.14 ml (small spots) or 9.42 ml (big spots) of a mixture made of 10 g of minced beef liver and 3 ml of saline solution (0.9% of NaCl). One hole was filled with fresh liver (fresh spot) and the other with rotten liver (rotten spot). The agar was then covered with a piece of tulle, which prevented larvae from digging into the food (while allowing them to feed). The whole device was placed in an incubator (Pol-Eko-Aparatura, model ST BASIC) at 30±1°C in darkness. At the beginning of the experiment, larvae were evenly spread on the middle line of the arena at an equal distance from the two spots (Fig. 1). Their spatial distribution was followed by taking one picture every hour for 24 or 48 h (Canon, model EOS 750D; preliminary observations showed the flash did not affect the spatial distribution of larvae).

The conditions tested are summarized in Table 1. For each condition, the number of larvae on each spot was counted from pictures taken hourly. When a single larva was tested, the time spent on each spot throughout the experiment was calculated. It was assumed that when located on one spot at a time t, the larva spent the following hour on this spot. When a group of larvae was tested, analyses focused on the spatial distribution of larvae at the beginning of experiment. Two parameters were calculated to characterize this distribution: (1) latency: the time (t₇₅) before 75% of larvae had made their first spot choice; and (2) mean larval number: the mean number of larvae on the fresh spot observed during the time interval between t₇₅ and the moment when the percentage of larvae on a fresh spot exceeded 25% variation (only minimum intervals of 5 h were considered); this number was computed from the values measured every hour.

For development tests, the percentage of emerged adults and their mean individual dry mass were analysed by mean comparisons between conditions. The Kruskal–Wallis test and Nemenyi post hoc test were used to compare conditions of different densities, and the Mann–Whitney test was used to compare conspecific and heterospecific groups and fresh and rotten food. Comparisons of mean dry mass were performed only when a minimum of 4 values were measured in the condition. The significance level was set at α=0.05/n, with n being the number of statistical comparisons performed for one given condition.

For binary choice tests, when a single larva was tested, the percentage of time spent on a spot was compared between spots using the Wilcoxon test. The percentage of time spent on the fresh spot was then used to analyse results of group tests: it was considered as the probability of observing one larva on a fresh spot at time t under the assumption that larvae did not interact with each other (henceforth named the ‘no group effect hypothesis’).

When a group of larvae was tested, the latency of larvae first choices was compared between conditions using the Kruskal–Wallis test and the Nemenyi post hoc test. Then, binomial tests were used to analyse the group effect on larval preferences, using the probability assessed from the individual tests. When the mean larval number on the fresh spot was significantly different from that expected under the no group effect hypothesis, the label ‘aggregation on the rotten spot’ (lower number) or ‘aggregation on the fresh spot’ (higher number) was attributed to the replicate. Then, binomial tests were used to test (1) the total number of aggregations, (2) the number of aggregations on the rotten spot and (3) the number of aggregations on the fresh spot, with the respective probabilities of success expected under the no group effect hypothesis; that is, (1) 0.05, (2) and (3) 0.025. Finally, binomial tests were used again to test the number of aggregations on the fresh spot in relation to the total number of aggregations (probability of success of 0.5). The significance level was set at α=0.05 for all these analyses. Development and binary choice test analyses were performed with R software (version 3.6.2). Datasets are available in Tables S1 and S2.

On either fresh or rotten liver and in either the conspecific or the heterospecific group, the percentage of emerged adults of both L. sericata and C. vicina was not significantly different between densities (d100, d250 and d500; Table 2). Furthermore, no significant difference in mean adult mass was observed between densities in L. sericata. A significant difference was observed in the C. vicina conspecific group on fresh liver: the adult mass was significantly lower at d500 than at d250 and d100 (Table 2). Accordingly, percentage and mass of emerged adults at different densities were pooled for subsequent analysis, but comparisons of mass were detailed at each density for C. vicina in the conspecific group on fresh liver (Table 3).

The percentage of emerged adults was significantly lower on rotten than on fresh liver in either the conspecific or the heterospecific group and for both species (Fig. 2, Table 3). The mean adult mass of L. sericata was also significantly lower on rotten liver in the conspecific group but was not different in the heterospecific group (Fig. 3, Table 3). For C. vicina, adult mass was significantly lower on rotten liver in both groups (Fig. 3, Table 3).

On either fresh or rotten liver and for both species, the percentage of emerged adults was not significantly different between the conspecific and heterospecific groups (Fig. 2, Table 3). The mean adult dry mass of L. sericata was also not significantly different between these groups on fresh liver (Fig. 3, Table 3). However, on rotten liver, it was significantly higher in the heterospecific than in the conspecific group (Fig. 3, Table 3). No significant difference in mean adult dry mass between groups was observed for C. vicina, regardless of food type (Fig. 3, Table 3).

The estimated probability of observing one larva on the fresh spot at a time t was 0.73 for L. sericata (95% confidence interval, 95% CI: [0.51, 0.91]) and 0.67 for C. vicina (95% CI: [0.35, 0.93]). Lucilia sericata larvae spent significantly more time on the fresh spot than on the rotten spot (Wilcoxon test, mean percentage of time spent on the fresh spot, m_fresh=73%; mean percentage of time spent on the rotten spot, m_rotten=26%, V=168, P=0.021), in contrast to C. vicina (Wilcoxon test, m_fresh=67%, m_rotten=30%, V=43, P=0.14). Both spots were occupied by the larva in only 1 out of 20 tests for L. sericata and 3 out of 10 tests for C. vicina.

When 40 L. sericata larvae were simultaneously tested in an arena with small food spots, 74% of larvae were on the fresh spot and 23% were on the rotten spot during the first 18 h. Subsequently, the distributions balanced until the end of the experiment, with 40–60% of larvae located on each spot (Fig. 4A). In 3 out of 10 replicates, regular shifts of larvae distribution were observed during the last 24 h of the experiment. Larvae stayed all on one spot from 15 to 60 min, moved collectively to the other spot for the same duration, then came back and started again, with several successive switches (Fig. 4B,C; Movie 1).

The latency of larvae first choices was not significantly different between high- and low-density conditions (Kruskal–Wallis test, H=2.42, P=0.49; Fig. 5). The number of replicates showing an aggregation is summarized in Fig. 6 (see also Table 4). There were significantly more aggregations than expected under the no group effect hypothesis in all conditions at high density, but only in one condition at low density (40 L. sericata on big spots; Table 4). Moreover, there were significantly more aggregations on the fresh spot than on the rotten spot at high density (binomial test, n_total=15, n=12, P=0.035) but not at low density (binomial test, n_total=9, n=6, P=0.51). Finally, although the number of aggregations on the rotten spot was still higher than expected at high density, it was not significantly higher in the largest group (120 larvae; Fig. 6, Table 4).

The latency of larvae first choices was significantly higher in the heterospecific group than in the L. sericata group [Kruskal–Wallis test, H=10.46, P=0.005; Nemenyi post hoc test, 40 L. sericata versus 40 C. vicina, P=0.054, 40 L. sericata versus (20 L. sericata+20 C. vicina), P=0.006, 40 C. vicina versus (20 L. sericata+20 C. vicina), P=0.71; Fig. 5]. In all conditions, there were significantly more aggregations than expected under the no group effect hypothesis (Table 4). However, the number of aggregations on the fresh spot was significantly higher than expected in the conspecific groups, but not in the heterospecific group (Fig. 6, Table 4). Finally, no significant difference was observed when comparing the number of aggregations on the fresh spot with that on the rotten spot (Table 4).

This study was designed to assess the costs and benefits of food decomposition, larval density and group species composition on larval development and to investigate the effects of these three factors on larval collective behaviour. The results of the development tests showed (1) a negative effect of decomposition on both survival and growth of larvae, (2) no benefit of increasing larval density, and (3) a benefit on L. sericata larval growth of developing in a heterospecific group on decomposed food.

Liver decomposed under anaerobic conditions is a sub-optimal food source for both blow fly species (as previously evidenced in C. vicina; Richards et al., 2013). The adverse effect of decomposition could be due to two main factors: (1) an alteration of nutritional value with putrefaction (Carter et al., 2007; Metcalf et al., 2016) and (2) a modification of food microbial community, causing nutrient depletion, the proliferation of pathogenic bacteria and/or the decrease in the relative abundance of beneficial bacteria (Ahmad et al., 2006; Andersen et al., 2010; Lam et al., 2009a,b; Metcalf et al., 2016). Despite the single observed benefit of the heterospecific group, both increasing density and species number did not allow larvae to reach a fitness on rotten liver as high as that on fresh liver. Therefore, the following observed effects of density and species number in binary choice tests cannot be explained solely by changes in food quality induced by density or species. Indeed, according to the ideal free distribution (Giraldeau and Caraco, 2000), larvae would always benefit most from aggregating on fresh than decomposed food (assuming that the food quantity is high enough to prevent competition).

The results of binary choice tests with single larvae are consistent with a fresh food-associated benefit (first hypothesis): L. sericata larvae spent significantly more time (73%) on the fresh than on the rotten spot. Calliphoravicina larvae also tended to spend more time (67%) on the fresh spot, although not significantly. These results indicate that larvae were able to assess the quality of the food and preferred fresh to rotten meat. This preference is consistent with the natural behaviour of blow fly larvae on carrion: flies can lay eggs a few minutes after death of the animal (e.g. Payne, 1965), and larvae feed on fresh food and leave it at the post-feeding stage when the tissue gets decomposed. However, larvae can sometimes be laid on several day old carcasses and feed only on decomposed food (e.g. Matuszewski et al., 2010). Repulsion from rotten food is a common phenomenon among animals: it relates to the spoiling theory, which supposes that microbes release repellent compounds allowing them to compete with animals (Janzen, 1977; Ruxton et al., 2014). For L. sericata, the fact that almost no larvae (5%) visited both spots suggests they first oriented themselves preferentially towards the fresh odour source. Then, the food exerted a strong retention effect, preventing larvae on the rotten spot from switching to the fresh one. For C. vicina, a few more larvae visited both spots (30%), suggesting species-specific preference or increased tolerance.

Experiments testing groups of L. sericata showed that larvae interacted with each other and that these interactions influenced their food preferences. The fact that aggregations occurred either on the fresh spot or on the rotten spot demonstrates that this effect of congeners can lead to both optimal and sub-optimal choices. However, there were more aggregations on the fresh spot than on the rotten spot at high density. Moreover, in groups of 120 larvae, the number of aggregations on the rotten spot was not significantly different from that expected if larvae did not interact with each other. These two results validate the second hypothesis of the study: choice accuracy increases with increasing larval density, with effects of both the absolute larval number and the ratio between larval number and food quantity. The absence of a significant difference in latency shows, however, that the speed of larval choices was not affected by either group size or density. This result suggests that the increase in choice accuracy results not from a density-dependent increase in larval movement (as such an increase would lead to finding the spots more quickly) but rather from larval inter-attractions.

Social influences occurring in collective decisions are often mediated by short-range cues, such as trail pheromones and contact cues (Jeanson et al., 2012). In cockroaches, the tactile perception of compounds left on the ground induces an increase in resting time, which can trigger a collective choice (Deneubourg et al., 2002; Lihoreau et al., 2010). The same mechanism is likely to occur in blow fly collective decisions, as they are known to leave retentive chemical cues (Boulay et al., 2013; Fouche et al., 2018). This mechanism could also explain the alternation in spot selection observed between 38 and 48 h with 40 L. sericata (Movie 1). A spatial saturation of the most-crowded spot could induce a shift of some larvae to the less-crowded spot. This shift could also be due to a group effect, such as waste accumulation (e.g. ammoniac), to a decrease in food quantity, or a decrease in food quality through a change in microbiome, induced or not by larvae (Johnson et al., 2013; Metcalf et al., 2016). Then, the formation of a trail would raise the probability to leave one spot for the other (Boulay et al., 2013; Fouche et al., 2018). The decrease in larvae number on the initial aggregate increased the probability of the remaining larvae leaving, leading to the collective movement. Further studies using modelling are necessary to investigate these hypotheses.

In the present study, social effects led to both optimal and sub-optimal larval choices. Optimal collective decisions are usually achieved by the amplification of individual preferences for the best resource (Canonge et al., 2011; Jeanson et al., 2012). However, the amplification mechanism can also induce trapping on sub-optimal locations (Dussutour et al., 2007). Such apparently non-adaptive behaviours can be caused by a lack of flexibility in the response to congeners. In forest tent caterpillars, although individuals increased their exploration level on the unbalanced food, this increase was too weak to counterbalance the trail-following behaviour and induce a switch in group decision (Dussutour et al., 2007). In the present study, the strong food retention effect also suggests a lack of behavioural plasticity: larvae trapped on the rotten spot were not able to leave even when other larvae were on the fresh spot. Nevertheless, an increase in density offers a way to limit sub-optimal choices: when density increases, the influence of social cues is enhanced, and the initial preference for the best resource has more chance of being amplified by the group (Canonge et al., 2011; Szopek et al., 2013).

Finally, the results showed that larval preferences were affected by the group's species composition, confirming the third hypothesis of this study. Although conspecific groups of both species made both optimal and sub-optimal choices, when these species were mixed together, they were unable to make optimal collective choices. Moreover, L. sericata larvae took significantly more time to make their first decision in the heterospecific than in the conspecific group. Therefore, the presence of another species led to a cost in choice accuracy through a reversal of larval preferences towards rotten liver, in addition to a cost in choice speed for L. sericata. Previous studies reported a change in individual preferences when environmental conditions were modified, such as by adding predation-related cues (e.g. spiny lobsters: Eggleston and Lipcius, 1992; snails: Gerald and Spezzano, 2005). This change was shown to be adaptive, leading to a decrease in predation risk. Preferences for food can also switch depending on the microbial community, as shown with L. sericata larvae (Rhinesmith-Carranza et al., 2018). In the present study, development tests did not show equal or greater benefit of developing on rotten than on fresh liver when species were together. However, such a benefit could arise in a natural environment, as discussed below.

On field carcasses, larval masses can largely exceed the highest density studied here and often contain more than two species (Rivers et al., 2011). Larvae also feed on various tissue types, which results in different impacts on larval fitness (Kaneshrajah and Turner, 2004; Clark et al., 2006). When feeding on hard tissues such as muscles, L. sericata larvae develop faster and survive better at higher density (i.e. Allee effect; Scanvion et al., 2018). Furthermore, benefits of interspecific aggregation were observed for both L. sericata and C. vicina larvae when reared together on such food (Komo et al., 2019). Finally, external selective pressures such as predation (Goodbrod and Goff, 1990) and parasitism (Frederickx et al., 2013) can be limited by group living (Krause and Ruxton, 2002; Rivers et al., 2012). It is therefore possible that the group behaviours observed in the present study could be adaptive under natural conditions.

For example, being trapped on a sub-optimal food source also leads to the maintenance of group cohesion and its associated benefits (e.g. larval-mass effect; Rivers et al., 2011). Furthermore, the adult mass increase observed for L. sericata grown in the heterospecific groups demonstrates that species association can provide benefits on decomposed food. Such benefits could result from the pooling of species-specific antibacterial compounds (Barnes et al., 2010) that would select a microbiome more favourable for development. Under this scenario, aggregating together on rotten food would make new food sources available for participant species, reducing local competition and creating a new developmental niche (Odling-Smee et al., 2013; Schwab et al., 2017; Goodale et al., 2020). In a similar way, collective decisions of Drosophila larvae towards a food spot infected by a pathogenic fungus were explained by the ability of larvae to collectively inhibit fungal growth and improve their development, enabling the colonization of new food spots (Rohlfs, 2005; Rohlfs et al., 2005). In blow flies, interspecific aggregation on rotten food would also be more advantageous when larval competition on fresh spots is high, inducing increased mortality and decreased adult mass and fecundity (Kamal, 1958; Saunders and Bee, 1995; Smith and Wall, 1997). Such higher competition on fresh than on rotten spots could occur at the end of the bloat stage, when the skin cracks and larvae get access to the uncolonized internal organs decomposed by anaerobic bacteria (Hyde et al., 2013; Metcalf et al., 2013; Pechal et al., 2014). Further experiments on the fitness consequences of interspecific aggregations are needed to confirm or not these hypotheses.

Finally, the aggregation behaviour of larvae may be affected by other factors not studied here. For instance, the presence of predators and parasitoids could increase the attraction between larvae such that the group becomes more cohesive and provides greater protection (Rivers et al., 2012). The local temperature has also a critical influence on larval aggregation: the two species studied here differ widely in their optimum temperatures (C. vicina having lower optima than L. sericata; Grassberger and Reiter, 2001; Donovan et al., 2006) and C. vicina aggregate preferentially at 22°C when placed in a thermic gradient (against 33°C for L. sericata; Aubernon et al., 2016). Bacterial growth is slower at cooler temperatures, which could explain the slightly lower tendency of C. vicina to choose the fresh food. It also suggests that C. vicina is usually exposed to a lower competition level with microorganisms and would have evolved less efficient antibacterial secretions compared with L. sericata, as observed in Barnes et al. (2010). However, this would lead to a benefit for C. vicina of being associated with a more competitive species such as L. sericata, while the opposite was observed in this study.

To conclude, this study has demonstrated that (1) blow fly larvae are able to collectively choose the best resource in their environment, (2) their choice accuracy is weak at low larval density but increases with increasing density, and (3) larval preferences are modulated by the presence of other species, leading to a choice reversal in heterospecific groups. These results provide new evidence of the ubiquity of collective decision properties in non-eusocial insects. They show that blow fly larval behaviour results from the perception and integration of environmental cues (food decomposition level), social cues (density) and species-specific cues. Their implications for the ecology of interspecific aggregations stress the importance of further studies of collective decision making in heterospecific groups.

The authors thank the anonymous reviewers for their helpful comments.