Different bees, different needs: how nest-site requirements have shaped the decision-making processes in homeless honeybees (Apis spp.)

Abstract

During reproductive swarming, a honeybee swarm needs to decide on a new nest site and then move to the chosen site collectively. Most studies of swarming and nest-site selection are based on one species, Apis mellifera. Natural colonies of A. mellifera live in tree cavities. The quality of the cavity is critical to the survival of a swarm. Other honeybee species nest in the open, and have less strict nest-site requirements, such as the open-nesting dwarf honeybee Apis florea. Apis florea builds a nest comprised of a single comb suspended from a twig. For a cavity-nesting species, there is only a limited number of potential nest sites that can be located by a swarm, because suitable sites are scarce. By contrast, for an open-nesting species, there is an abundance of equally suitable twigs. While the decision-making process of cavity-nesting bees is geared towards selecting the best site possible, open-nesting species need to coordinate collective movement towards areas with potential nest sites. Here, we argue that the nest-site selection processes of A. florea and A. mellifera have been shaped by each species' specific nest-site requirements. Both species use the same behavioural algorithm, tuned to allow each species to solve their species-specific problem.

This article is part of the theme issue ‘Collective movement ecology’.

1. Introduction

Social insects in general, and ants and bees in particular, are wonderful model systems to investigate how a collection of individuals makes decisions. Both bees and ants need to make collective decisions about where to forage or nest, and where and when to migrate. Collective decision-making has been studied extensively in both honeybees (genus Apis) and myrmicine ants (genus Temnothorax). Temnothorax ants live in small colonies of fewer than 500 worker ants [1] and conveniently accept nest sites constructed from matchsticks and microscope slides [1,2]. By keeping colonies in the laboratory and offering the ants potential nest sites that differ in attributes such as volume, light levels and entrance size, one can study the processes by which the ants select new nest sites [2–4]. From studies in which each individual ant was marked and followed, we know that they carefully assess the quality of potential nest sites [1,3]. Nest-mates are recruited to sites deemed acceptable via more experienced ants ‘teaching’ naive ants the route from the old nest to the new nest [5]. More favourable sites attract more ants because individuals that have discovered an attractive site will more quickly start to recruit nest-mates [1]. Once a quorum threshold of ants is present at a potential nest site [6], ants simply carry their nest-mates to the chosen site, thus speeding up the colony's move [6]. Although individual ants may make errors when assessing the quality of sites, collectively the colony tends to select the best available site [7].

A swarm of Apis mellifera honeybees is also capable of selecting the best available nest site most of the time [8]. Bee researchers can mimic the bees' natural swarming process by making artificial swarms (the female reproductive unit of a honeybee colony) comprised of a large number of bees full of honey or sugar water and a single queen. The artificial swarm is placed on a vertical board so that researchers can directly observe the decision-making process as it unfolds. Over the next day or so, scout bees leave the swarm and search the environment for suitable cavities in which to form a new home. Bees have strong preferences for cavities that are ideally dry, lofty, clean, unoccupied and cosy, with a defendable entrance and a volume of 40 l [8]. Instead of teaching nest-mates the route to the new-found site by guiding them directly, as do Temnothorax ants, scout bees perform dances on return to the swarm that indicate the distance and direction to the potential nest site [9]. Dances for high-quality sites last longer and those sites therefore attract more nest-mates. As with Temnothorax ants, the bees finalize the decision-making process when a quorum of scout bees is reached at one of the potential nest sites under consideration. Because dances for high-quality sites last longer, better sites are more likely to reach the quorum than sites of lesser quality. At the same time, if no superb site is found, sites of lesser quality will ultimately reach the required number of bees, so that the swarm will move to a new nest site and not succumb to bad weather to which the temporary cluster, the swarm, is vulnerable.

Collective decision-making requires that the individuals involved use information obtained from their environment to decide what to do next. While individuals have incomplete information about their environment, collectively the group possesses much more complete information [10]. The behavioural algorithm, or the set of behavioural rules used by individuals [11], ultimately determines the collective decision, and the algorithm can be tuned depending on the circumstances. For example, Temnothorax ants make faster decisions when their nest is destroyed than when they are selecting a nest site while their old nest is intact [12]. By initiating a search for an alternative nest site earlier, accepting a new nest site more quickly and lowering the quorum threshold required, the ants make faster but less accurate decisions when the need to find a new nest is urgent. Hence, they use a modified version of the same behavioural algorithm [11].

If within the same species the behavioural algorithm can be tuned by exogenous conditions, we might expect that similar behavioural algorithms are also tuneable by natural selection. That is, it is likely that an ancestral algorithm is tweaked by natural selection, so that it is appropriate for the ecological conditions experienced by different species that are descended from the same ancestral species. Within the genus Apis, there are currently 11 recognized species [13] that diverged from a common ancestor 6–10 million years ago [14]. These 11 species can be broadly divided into three clades — the dwarf bees, the giant bees and the cavity-nesting bees. Both the dwarf and giant bees nest in the open, building a single comb under (giant bees) or around (dwarf bees) a branch or twig, respectively [15]. Thus far, the nest-site selection process has been described in detail for just two species: A. mellifera, a cavity nester, and A. florea, a dwarf species.

Of the extant Apis species, the behaviour of A. florea is probably most similar to that of the common ancestor to Apis [16]. Here, we will argue that the nest-site selection process as seen in A. mellifera is the result of a fine-tuning of the ancestral nest-site selection process of the open-nesting honeybees. When the ancestor of the cavity-nesting clade moved into cavities, most likely as a means to better defend their honey stores [15], it required more precise mechanisms to select nest sites. Therefore, we regard the bees’ decision-making process as a tuneable algorithm, selected to allow the bees to make the best decision given their specific nest-site requirements.

2. Nest-site selection

When honeybees move nest, the swarm not only needs to decide on a new nest site, but most of the bees need to be guided to the chosen location by the scouts. The two processes, nest-site selection and swarm guidance, are tightly linked because the bees that are involved in nest-site selection need to guide the swarm in flight to the chosen nest site. Because both processes have been best described for A. mellifera (reviewed in [17]), we will use this species to outline how honeybees decide on where to nest and how they move to the chosen site.

Natural colonies of A. mellifera live in tree cavities. The quality of the cavity is often critical to the survival of the colony: it should be defendable, and large enough to allow the colony to grow to full size. While a swarm needs to move to a permanent home as quickly as possible (an unprotected swarm is vulnerable to weather and cannot produce young or store honey without a comb), it cannot generally afford to settle for an inferior site. The scouts must thus search widely to ensure that they do not settle for a poor cavity when a better one is available.

When a swarm leaves the colony, it normally clusters a few tens of metres from it [18]. Scout bees then begin to search the environment for suitable nest sites while the majority of bees are not involved in the decision-making process; only approximately 5% of all the bees in the swarm participate [19]. When a scout has found a suitable nesting site, it will return to the swarm and report the location of the site found to other bees by performing communication dances on the surface of the swarm cluster [9]. These dances recruit others to visit their discovery. Both the duration of nest-site dances and the number of circuits per dance that a scout or recruit (hereafter ‘bee’) performs upon return to the swarm are positively correlated with the bee's perception of the quality of the site danced for [20]. (A communication dance comprises many dance circuits, each providing information about the direction and distance to the advertised site.) Bees do not continue to dance for ‘their’ site. After each return to the swarm (as bees keep inspecting the site they are dancing for), the bee will reduce the number of dance circuits until she stops dancing altogether. Because bees that have rated their site as being of high quality start with more dance circuits per dance than those that have visited a poor quality site [21], sites of high quality are advertised for longer than those of low quality. The outcome of this process is an increase in the number of bees visiting and dancing for sites of high quality, and a decreasing number of bees dancing for sites of lesser quality [21]. Eventually, dances tend to converge to one site only, a process that may take several days in A. mellifera [22]. Once a site has attracted a sufficient number of bees, a quorum [23], bees that have sensed the quorum will now return to the swarm and signal the end of the decision-making process by producing an auditory signal known as piping [24]. This signal informs the quiescent bees in the cluster that they should prepare themselves for flight [25]. In temperate zones, the piping signal is essential to ensure that the formerly inactive bees warm up their flight muscles to the 35°C required to sustain flight [25,26].

3. Swarm guidance

Once airborne, the swarm needs to be guided towards its chosen home. How exactly can a group of animals be guided by only a small number of knowledgeable individuals? Two theoretical studies have addressed the issue of information transfer from informed to uninformed group members. Janson et al. [27] modelled a situation in which the informed individuals make their presence known by moving at a higher speed than the average group member and into the direction of travel. Guidance is achieved by uninformed individuals aligning their direction of movement to that of fast-moving neighbours. Because the informed individuals initially move faster, they have a larger influence on the direction of movement of the uninformed individuals, thereby steering the group. A second model [28] shows that a group can be guided by a few informed individuals without these individuals providing explicit guidance signals and even without any individual in the group ‘knowing’ which individuals possess information about travel direction. Only the informed members of the group have a preferred direction, and it is their tendency to go in this direction that steers the group. The main difference between the two models lies in the presence or absence of cues or signals from the informed individuals to the uninformed majority. In Janson et al.'s [27] model, leaders clearly make their presence known, whereas Couzin et al. [28] suggest that leadership can arise simply as a function of information difference between informed and uninformed individuals, without the individuals being able to tell which ones have more information.

Experimental work on honeybee swarms has elucidated that within a flying swarm, two types of individuals can be identified: those that fly fast and are more directed, and those that seem to mill about and fly at a lower speed [29,30]. The fast and more directed bees are mainly found in the top part of the swarm, presumably because there they are more visible when projected against the blue sky or white cloud cover to the bees milling below them. When all scouts are guiding the swarm towards the same nest site, the swarm flies more or less directly to its goal [31]. If, however, the swarm's flight is disturbed by the presence of a large number of bees that fly fast through the swarm into a direction other than the nest site, the swarm is unable to fly straight to the goal or even misses the goal completely [31]. Thus, honeybee swarms are guided by informed individuals that make known to the rest of the swarm they hold directional information by streaking through the swarm, and only if the scouts agree on the direction of travel will the swarm be able to reach its goal.

4. Linking nest-site selection to swarm guidance

We know that the piping signal, performed by bees that have experienced the quorum [32], is not related to directional consensus in dances on the swarm cluster. How, then, does the swarm achieve directional consensus prior to lift-off? Recent work has shown that scout bees produce an auditory stop signal directed at bees dancing for a site other than the site the stop-signalling bee was dancing for [33]. Bees that receive the stop signal tend to cease dancing. Once worker piping has commenced, the stop signal is greatly upregulated, resulting in a reduction of flight and dance activity [33]. This reduction in flight activity ensures that scout bees remain on the swarm surface during the final phase of the decision-making process, so that they can act as guides once the swarm is airborne. Once piping bees have warned the swarm bees that lift-off is imminent, the swarm only needs an activation signal to coordinate the departure of the swarm. This signal, the buzz-run, is produced by scout bees that have experienced the quorum at the chosen nest site [34]. Buzz-running bees run through the swarm cluster physically dislodging the swarm and forcing it into the air [34] (figure 1).

Figure 1.

Flow diagram of the different decision points that scout bees go through during the nest-site selection process and the range of signals used by A. mellifera. Shown are an A. florea swarm awaiting the end of the decision-making, and an A. mellifera swarm in flight on its way to the chosen site. See the text for more details. (Online version in colour.)

The presence of a specific signal aimed at reaching consensus in the dances prior to swarm departure, the stop signal, is intriguing. It suggests that directional consensus prior to lift-off is important for swarm guidance in A. mellifera. In his classic paper on the swarming behaviour of A. mellifera, Lindauer [9] described one swarm in which the bees did not seem to agree on the preferred nest site before the swarm departed. While airborne, two sets of scout bees tried to coerce the swarm to fly into their chosen direction. Because the preferred nest sites were in opposite directions and because each was supported by roughly the same number of bees, the swarm was unable to fly to either site and reclustered. Makinson & Beekman [35] investigated more precisely if A. mellifera swarms indeed require consensus in the sites danced for prior to lift-off, by forcing swarms to the air before the bees had been through the complete process described above. Swarms were forced to become airborne when the total number of bees dancing was similar to the number of bees dancing at the time of lift-off in non-manipulated swarms, but prior to the onset of the piping signal and while bees were dancing for multiple sites. None of the experimental swarms were able to fly towards one of the nest sites the scouts had been dancing for. Clearly, consensus is required for successful guidance of swarms in flight.

Which individuals guide the swarm in flight? There are two possibilities. A swarm could either be guided by all bees actively involved in the decision-making process (all bees dancing prior to the swarm taking to the air), or only by those that have experienced the quorum. If all bees involved guide the swarm once airborne, Makinson & Beekman's [35] experimental swarms would have flown into the general direction of all dances present prior to lift-off. If, on the other hand, only bees that have experienced the quorum act as swarm guides, then none of the swarms in their experiments would have been able to fly. Makinson & Beekman [35] concluded that only bees that have directly experienced the quorum at the nest site guide the swarm, thereby ensuring that different groups of bees (those that have visited different potential nest sites) do not guide the flying swarm in different directions. We then need to assume that in the particular swarm Lindauer studied (see above) two nest sites reached a quorum simultaneously. While such scenario is unusual, it is certainly not impossible, as Seeley & Visscher [24] discovered. They too observed one swarm in which bees were advertising two alternative nest sites. As with Lindauer's, this swarm was unable to reach either of the advertised sites. The need to restrict swarm guidance to those bees that have visited the site at which the quorum is reached explains why the stop signal is upregulated once a nest site has attracted sufficient bees to end the decision-making process.

5. Different bees, different needs

What if you do not care that much about where to live because of limitless equally good options? This is the situation typically faced by the red dwarf bee, A. florea. Apis florea is endemic to Southeast Asia, India and eastern regions of the Middle East [15], and is now spreading as an invasive species into the Middle East [36]. It builds a small nest comprised of a single comb [37] constructed around a twig of a shrub or a tree [38]. Notwithstanding the fact that A. florea colonies prefer well-shaded nest sites away from predatory ants, their natural environment is replete with suitable twigs. Therefore, it seems that there is no strong need for A. florea to go through an elaborate decision-making process like that seen in A. mellifera.

In contrast to A. mellifera, where at the time of lift-off dances approach or reach consensus about the chosen nest site [9,39], artificial swarms of A. florea show large variation in directional information prior to lift-off; bees continue to dance for multiple sites [40,41]. This is also true for natural swarms that depart from the mother colony. (Natural swarms of A. florea do not form an interim cluster [42].) Instead, it appears that the swarm lifts off once a vectorial consensus has been reached [40]. The level of vectorial consensus is determined by both the number of bees dancing on a swarm and the level of agreement regarding the average direction indicated by these dances (figure 2). Unlike A. mellifera, A. florea scouts do not seem to use a quorum at the nest site to finalize the decision-making process [40]. Evidence for a lack of quorum is circumstantial, as the bees cannot be enticed to select specific nest sites making it impossible to study the behaviour of scouts as they inspect a site. From observing individually marked bees as artificial swarms go through their decision-making process, Makinson et al. [40] concluded that the majority of bees do not leave the swarm once they start dancing. Re-evaluation of the site danced for is therefore unlikely and scouts are most likely unable to determine the number of other bees present at the same site. In addition, both dancing and non-dancing bees produce a piping signal throughout the decision-making process, suggesting that this signal is not linked to a quorum at a chosen nest site in A. florea [40].

Figure 2.

Vectorial consensus in dancing bees. To determine the level of vectorial consensus in the dances performed by bees, one determines the average direction danced for by an individual bee to construct a unit vector of each bee's dance direction. The consensus vector is then determined by adding all the unit vectors of bees dancing head to tail. When most bees are dancing for different locations, as in (a), the consensus vector is shorter than when more bees dance roughly for the same direction, as seen in (b). Photos show two A. florea swarms in which all individuals were uniquely marked to collect the dance direction data on individual bees. See Makinson et al. [40] for a detailed description on quantifying on-swarm agreement using consensus vectors. (Online version in colour.)

There is also no evidence that A. florea scouts judge the quality of the sites they are dancing for. Makinson et al. [40] found no evidence for ‘dance decay’ in A. florea: bees did not reduce the number of dance circuits after leaving and returning to a swarm as they do in A. mellifera where the quality of the site is encoded in the total length of time a site is advertised [21]. The absence of dance decay combined with the observation that most bees do not repeatedly leave the swarm led Makinson et al. [40] to conclude that A. florea scouts do not differentiate between sites of low and high quality. Most likely, scouts indicate a direction in which they had previously encountered food sources (see below). It remains possible that individual bees adjust the number of dance circuits depending on the perceived quality of the site without reducing this number as time goes on. Ultimately, though, for the swarm to be able to select the best site, the number of dance circuits produced for a site must be quality-dependent, and the bees need to cease dancing even when dancing for the best site possible [43]. This does not seem to be the case for A. florea.

6. Swarm guidance in Apis florea

Given that the selection of a nest site to move to and guidance of the flying swarm are linked processes, what is the effect of the lack of directional consensus at the time of lift-off on the ability of A. florea swarms to move cohesively? Schaerf et al. [44] used agent-based modelling to examine whether the decision-making process as described by Makinson et al. [40] allows a swarm to build sufficient vectorial consensus for successful swarm guidance. Their work showed that, provided potential nest sites are abundant, the rather laissez faire decision-making process of A. florea is sufficient for the swarm to be guided, by streaker bees, towards the average direction advertised by the bees dancing at the end of the decision-making process. The area containing potential nest sites needs to be large because only then can sufficient bees locate nest sites in the same general direction. In other words, when nest sites are restricted to a small area only, assuming the lack of a quorum threshold, the swarm cannot reach the level of vectorial consensus required for successful swarm guidance. When potential nesting sites were widely separated in space, resulting in bees dancing for areas that were wide apart, simulated A. florea swarms performed poorly when using vectorial consensus [44]. This is because the model assumes that all bees that are dancing at the time of lift-off will participate in swarm guidance once airborne, resulting in the swarm flying in the average direction indicated by the dances [45]. The assumption that all bees dancing prior to lift-off participate in swarm guidance seems reasonable, given that A. florea does not use a quorum at the chosen nest site nor seems to have a mechanism to silence dances for alternative sites.

By following swarms as they moved towards new nesting sites, Makinson et al. [40] concluded that A. florea decides on the exact location to build a nest once the swarm has reached the stand of trees it was dancing for. Thus, it appears that the purpose of dancing on an A. florea swarm is to increase the level of vectorial consensus to allow coordinated flight into an area where nest sites are abundant. Frequently, the swarms simply appear to follow the availability of floral resources in their environment. By studying natural swarming events, Makinson et al. [42] concluded that swarms of A. florea tend to move into areas with abundant floral resources, as swarms often departed into the direction the bees were found to be foraging both before and after the departure of the swarm. Presumably once resources become scarce, the bees will move again.

7. It is a numbers game

Notwithstanding the above, is it possible that A. florea is capable of making decisions as precisely as A. mellifera, but our inability to observe the behaviour of A. florea scouts as they evaluate potential nest sites obstructs our ability to discern their real decision-making capacities? We do not think so. The beauty of modelling studies is that you can manipulate the system in ways that are impossible in natural systems. Schaerf et al. [44] changed the level at which bees could use the dance to gain information about the location of potential nest sites and studied the effect this has on the decision-making ability of in silico swarms. If more bees used dance information to inform themselves about the location of potential nest sites, the simulated swarms were more likely to make a decision on where to move to within the set time limit of the simulation even when nest sites were scarce. Thus, the number of individuals involved in making an informed decision affects the quality of the decision-making process. To further investigate the effect of the number of individuals on the quality of the decision made, Schaerf et al. [46] studied large and small swarms of A. mellifera both empirically and in a simulation model. Both sets of data clearly indicate that small A. mellifera swarms are less able to select the best site available when there is an abundance of nest sites to choose from. Small swarms are unable to settle on the best site because they do not have sufficient scouts to sample the environment thoroughly and to build a quorum. To select the best nest site possible, the number of individuals involved in the decision-making process needs to be large. From this it follows that even if A. florea employed the same decision-making process as A. mellifera, due to the much smaller number of individuals involved, A. florea swarms would be unable to select the best nest site available, unless they took a very long time to do so. We therefore think it highly unlikely that A. florea's decision-making process is more complex than we have been able to decipher, simply because it would not improve the quality of the decision unless associated with a significant increase in colony size.

An effect of the number of involved individuals on the quality of collective decision-making is not unique to nest-site selection. Beekman et al. [47] compared foraging behaviour of small and large A. mellifera colonies and found that small colonies were less able to focus their foraging efforts on the best available patches. Instead, small colonies foraged at many more patches relative to their size than large colonies, indicating their inability to use the dance language to the greatest advantage. This effect of colony size on effective use of the dance language was later confirmed experimentally by Donaldson-Matasci et al. [48]. Here too, larger colonies benefitted more from dance communication that allowed them to dominate high-value resources through rapid recruitment. Similarly, by comparing ant species that differ in the complexity of their recruitment mechanism (ranging from individual foraging to relying solely on mass recruitment via pheromones), Beckers et al. [49] showed that large colonies use more complex means to recruit nest-mates. In species that use pheromone trails to recruit nest-mates to food sources found (the most complex form of recruitment in Beckers et al.'s [49] analysis), experimentally reducing colony size reduces the ability of colonies to effectively communicate the existence of profitable food sources [50]. Similarly, small ant colonies are less capable of connecting nests via the shortest possible route using pheromone trails compared with larger colonies [51].

If A. florea swarms would perform better if they were larger, why has natural selection not led to an increase in colony size, and by extension swarm size? We argue that the decision-making process of A. florea does not need to be more precise. Its low key and perhaps messy decision-making process is sufficient to allow the bees to move to areas suitable to nest in without putting the swarm at risk of breaking up during flight. This line of reasoning also allows us to predict what the nest-site selection process will be of species that nest in the open, but have larger colonies. The giant honeybee Apis dorsata is such a species. Apis dorsata builds a single comb that can be up to 2 m wide on the underside of rocky outcrops or branches of smooth-barked trees [15]. Like A. florea, potential nest sites are relatively abundant but because the bees prefer to nest in large aggregations [52] combined with the fact that selected nesting sites need to be able to carry the weight of the colony, choices are most likely more restricted. The combination of large colonies and slightly more specific nest-site requirements leads to the prediction that A. dorsata's decision-making process is intermediate between that of A. florea and A. mellifera. This prediction has been shown to be correct [53]. The large colony size of A. dorsata allows the bees to be more discerning when selecting a place to live, even when potential nest sites are common.

8. A tuneable behavioural algorithm

To be able to argue that the bees' nest-site selection process is a tuneable algorithm that each species adapts to its specific needs, we need to know exactly what kind of problem an A. florea- and A. mellifera-type algorithm can solve. Diwold et al. [45], using an individual-based model, directly linked a swarm's decision-making process with the guidance of the swarm. Thus, the decision-making component of the process determines the level of vectorial consensus present at the time of lift-off. Once in flight, swarms are guided by bees that were involved in dancing at the time of lift-off. Diwold et al. [45] then constructed two versions of a behavioural algorithm, one based on the nest-site selection behaviour as observed in A. florea and one based on A. mellifera. The main difference in the decision-making component between the two versions of the algorithm is the accuracy with which directional information is conveyed in the dances and the mechanism by which a decision is made (table 1). Because dances by individual A. florea scouts vary widely within a dance [40], recruits are scattered over a larger area in the A. florea version compared with the A. mellifera version of the algorithm. The decision to prepare the swarm for lift-off is made by vectorial consensus (A. florea version of the algorithm) or quorum (A. mellifera version of the algorithm).

Table 1.

Two versions of the honeybee's tuneable behavioural algorithm as used by Diwold et al. [45]. ‘Dance precision’ relates to the accuracy of the bee's dance, while ‘recruitment precision’ refers to the probability that a dance follower will locate the advertised site. In the A. mellifera version of the algorithm, the decision is made once a site is visited by a threshold number of bees (quorum), while the A. florea version requires a certain level of vectorial consensus in the dances performed on the swarm.

	A. mellifera-like	A. florea-like
‘dance precision’	high	low
‘recruitment precision’	high	low
‘decision measure’	quorum	vectorial consensus

By offering the in silico swarms two regions (the centres of which were separated by 180°) each divided into 60 smaller subregions representing distinct patches of foliage or cavities that differed in ‘quality’, Diwold et al. [45] could determine how well each version of the algorithm performed. One of the regions contained more subregions of higher quality. Each swarm therefore had to first decide which region to select, and then which subregion within that region. At the end of the simulation, Diwold et al. [45] determined where each swarm ended up and how long it took to make a decision. Almost all A. florea-like swarms (95.7%) made a decision within the set time limit, whereas only 28.6% of the A. mellifera-like swarms had finalized their decision at the same time. However, the speed of the A. florea swarms' decision-making process came at a cost: A. mellifera swarms were much better at selecting the best site possible. Even though more A. florea swarms selected the best region compared with A. mellifera swarms, only 48.7% of the A. florea swarms selected the best-quality site within the region, whereas 99.7% of the A. mellifera swarms that made a decision did so.

By combining the two processes bee swarms need to go through in a single decision-making algorithm, Diwold et al. [45] nicely showed how the same basic behavioural mechanism can be used to address species-specific problems. A honeybee species such as A. florea uses a basic form of the algorithm to select a region with food and, by extension, many potential nest sites. When the quality of the chosen site is more important, as is the case in A. mellifera (and most likely all cavity-nesting bees), the algorithm is more refined allowing a more precise decision-making process.

9. How ecological conditions have shaped the bees’ nest-site selection process

Until about 10 years ago, almost everything known about nest-site selection in honeybees was based on studies on A. mellifera. The temptation then exists to assume that other, related species behave in the same way. By examining other species of Apis that have fundamentally different nest-site requirements, we are now able to speculate on the evolutionary origin of the nest-site selection process in honeybees.

The common ancestor to Apis lived in the open and most likely had a nest structure similar to that of the dwarf bees [16]. When communicating the location of resources, the dwarf bees simply ‘point’ towards the resource while dancing on the horizontal surface of the comb, mostly on top of other bees [15]. The giant honeybees, while also open-nesting, are obliged to dance on the vertical curtain of bees surrounding the single comb. Because the comb is built under a branch or cliff overhang, the giant bees cannot simply point and have evolved to use gravity and the current position of the sun as points of reference [54]. For example, to indicate the direction of a resource that is directly in the direction of the sun's current azimuth, a giant bee directs her dances directly up the curtain. A resource 45° west of the current sun's azimuth is indicated by a dance at 45° from the vertical, and so on. Because cavity-nesting species live in the dark, they too rely on gravity as their reference point and dance on the actual comb when advertising forage sites [55]. Because cavity-nesters form a temporary cluster when they swarm, nest-site dances take place on top of other bees and dancers then use the direction of the sun as their reference [9,56]. Despite the different contexts (forage versus nest sites), substrates (comb or other bees), dance orientation (horizontal versus vertical) and point of reference (sun or gravity), the dances of A. mellifera, A. florea and A. dorsata are remarkably similar [57].

Where the species do differ is in the need to find a very specific location to nest in and the length of time they stay at a particular site. Cavities large enough to house a decent-sized honeybee colony are relatively rare, particularly those that fulfil other requirements such as entrance orientation and size, and the absence of other tenants. Once occupied, colonies can stay in the same cavity for many years [58,59]. While giant honeybee colonies are often philopatric, returning to the same location seasonally [60,61], upon their return they build a new comb [62]. Nest-site fidelity is probably due to a relative lack of suitable nesting sites, as the combs the bees build are massive and require significant trees (or buildings or rock faces) able to bear their weight. The dwarf bees migrate often and are prone to absconding when disturbed [63]. While an A. mellifera colony will normally produce one or two swarms per reproductive cycle, A. florea produces many more [42], and often the original colony perishes at the end of swarming. Interestingly, A. florea is the only species of which bees return to old combs to collect the wax [64], another indication that swarms tend not to move far from the original colony.

When nesting locations are abundant and easy to locate, there is no need to invest in an elaborate decision-making process, especially when it is likely to be ineffective due to small colony size (see above). We therefore see a simplified process, particularly in A. florea, while the process is more complicated in A. dorsata compared with A. florea because the specifics of A. dorsata's nest-site requirements are more restrictive [65]. Once some Apis species moved into cavities, they required a more sophisticated nest-site selection process—one that allows assessment of the quality of the site. They also needed a means for precise guidance of the swarm in flight towards that site. Because a vectorial consensus is insufficient to achieve these goals [35], cavity-nesting bees need to reach a near-unanimous decision prior to the swarm taking to the air. And because consensus is so important to A. mellifera, this species has evolved a ‘stop-signal’, aimed at silencing bees dancing for the non-chosen site [33]. This signal results in bees dancing for a single site only, thus ensuring successful guidance towards the chosen site. The need for unanimity in dances might also explain why cavity-nesting bees form a temporary cluster, despite the obvious dangers of hanging exposed, sometimes for days, while the swarm comes to consensus. While in the absence of a temporary cluster dances occur for both nest sites and forage [42], dances on the temporary cluster are exclusively for nest sites.

Even within species, the behavioural algorithm appears to be flexible. Both A. dorsata and the African subspecies Apis mellifera scutellata migrate seasonally in response to changes in floral resources. Often such migrations are over distances in excess of 20 km [66,67]. Clearly, in such instances, it is impossible for scout bees to physically visit the travel destination. Both species use a modified version of the communication dance specific to long-distance migration, called the migration dance. Migration dances communicate a general route of travel but not a specific location and their main purpose appears to be to ensure that the swarm can travel cohesively [67,68], much like the swarms of A. florea [40]. Hence, in both A. m. scutellata and A. dorsata, the behavioural algorithm is tuned to the type of colony movement and precision required in selecting a destination.

10. Final words

Because A. mellifera is so amenable to experimental manipulation, it is not surprising that until recently, it was the most studied species of honeybee. Studies have revealed an intricate decision-making process that allows the bees to select the best available nest site most of the time. But it is also costly, as making accurate decisions takes time. Thus, if the quality of the decision is less important, the process by which the decision is reached can be less elaborate. Over the past few years, we and our colleagues have studied the different components of the mechanisms by which A. florea, and to a lesser extent A. dorsata, selects and moves to a new nest site. We have argued that the decision-making process of A. florea is perfectly adequate for the task at hand. When some honeybee species gave up their open-nesting habits and moved into cavities, the quality of the potential nesting location became important and the potential sites to choose from decreased. This led to the need to refine the bees' nest-site selection process.

Data accessibility

This article has no additional data.

Competing interests

We declare we have no competing interests.

Funding

Our work is supported by the Australian Research Council.