Conflict frequently arises in animal societies over reproduction. In many social insects, this conflict is centred on the queen because she is the sole reproductive in a colony of potentially thousands of individuals: workers cannot (or do not) reproduce themselves and only gain fitness indirectly. The identity of the queen is therefore of critical importance as the fitness of all individuals in the colony depends on her. In some species workers may have opportunities to select their queen, and might be expected to do so based on fecundity, relatedness, longevity, and minimising the delay to egg laying 1 . However, therein lies a conflict of interest at multiple levels: whereas colony level selection will favour rapid emplacement of a high quality queen, individual level selection acting on workers can select for behaviour that favours closely related queens (nepotism), while queens are presumably also under selection to fight among themselves 2 3 .

Queen selection occurs in three main contexts in social insect colonies: i) after worker emergence in pleometrotic (multiple queen) assemblages 4 , ii) associated with colony division by dependent colony founding (fission or budding) 3 5 , and iii) during emergency queen replacement 3 6 7 . While queens clearly benefit from the elimination of rivals 8 1 , killing of queens may be also in the interest of workers who seek to maximise colony efficiency and restore monogyny 2 8 . Unrestrained direct conflict between queens may be unfavourable because it entails the risk of both queens being injured or killed 9 , potentially leading to colony failure. This is particularly true during emergency queen replacement because the supply of brood that can be reared into new queens is limited to that laid by the previous queen, so that colonies can ill afford losing new queens. This risk can be reduced by worker intervention in queen-queen contests, whereby workers collectively influence or decide the outcome of contests 2 . Previous studies suggest that in many cases workers decide contests by culling supernumerary queens 10 3 1 , though direct competition between queens is also important 11 12 1 . Queens can therefore maximise their chances of prevailing via direct conflict (fighting) or indirect conflict designed to influence workers (eg: ritualised aggression (dominance displays) or advertisement of fertility). Where workers can influence the outcome of contests, they should act to favour queens that maximise their indirect fitness, that is queens that i) are more related, ii) are likely to survive the longest, and iii) have the highest fecundity 1 13 . Selective elimination of supernumerary queens by workers has been demonstrated based on abdomen size 14 , proximity to brood 15 , loss of mass 16 17 , fertility 18 , recent social environment 17 , size 15 16 and chemical signalling of reproductive condition 19 20 21 22 . The presence of nepotism remains equivocal in social insects despite considerable investment in its study 23 24 25 3 (and references therein). When queens are equivalent or there is no information available to workers with respect to queen quality or relatedness, they should select for rapid resumption of laying activity by favouring first eclosing queens, as occurs in army ants 26 .

Aphaenogaster senilis is a monogynous, monandrous ant that founds colonies by dependent colony foundation (= fission 27 ). Supernumerary queens are produced as part of the colony reproductive process and also during emergency queen rearing 28 5 . In orphaned colonies, workers produce several (up to 5) new queens as replacements which fight together until monogyny is restored 7 . As this species is monandrous and monogynous, workers and replacement queens are full sisters and there is thus no option for nepotism 7 . There should therefore be no conflict of interest between workers during queen selection, and workers should favour new queens based on quality and the rapidity of replacement. Queens, on the other hand, are expected to fight between themselves, as they are more related to their own brood than their sister's brood. Chéron et al 7 demonstrated that contests during emergency queen replacement are resolved in favour of first-eclosed queens, which are physically aggressive toward, and dominant over, subsequently produced queens. They proposed that second and subsequently produced queens are produced as 'insurance' against death of the first-born queen, and that workers select the first-born queens to expedite the re-queening process. However, first-born queens also have a head-start in that they are produced on average seventeen days before subsequent queens, permitting them relatively more time to develop musculature, harden the cuticle, develop ovaries, and produce pheromones, which may give them a competitive advantage in direct and indirect contests with other queens.

Order of eclosion and any advantages relating to acquired asymmetries are correlated so that it remains unclear which factor is determinant in queen selection. We investigated this by manipulating the ability of first queens to dominate subsequent queens, by ablating a mandible of the first-born queen in experimentally orphaned colonies. In Apis mellifera the ablation of a mandible results in queen avoidance of other queens: mutilation of all queens yields stable polygynous colonies despite this species being strictly monogynous 29 . We infer that mutilation of the first-born queen in A. senilis should increase the proportion of second-born queens inheriting the colony if queen choice is based on phenotypic characteristics such as fighting ability, whereas there will be no effect of the treatment if queen choice is based on order of emergence. Mutilation also allows testing whether A. senilis queens show evidence of self-assessment, as has been argued for Apis mellifera 29 . Specifically, we expect that, i) if fighting is the primary means of queen selection, then ablated queens will have a lower success rate in treatment groups; ii) if fighting is ritualised (a dominance display) and queens are instead selected by workers, ablation may have no effect on queen success; iii) These outcomes may be influenced by the behaviour of ablated queens, which we expect to be less aggressive and more evasive if self-assessment is occurring.

Twenty-nine colonies of A. senilis were collected from Aznalcázar, near Seville, Spain, between the 8th and 12th of November, 2008. One additional colony used was collected in February 2007 and had undergone laboratory hibernation. Entire colonies were placed in open plastic boxes of 16 × 26 cm, the walls of which were treated with fluon to prevent ant escape, and kept in a constant temperature room (~28°C) under a 12/12 hr day/night light cycle. Colonies were housed in nests made of two plastic Petri dishes of 10 × 5.7 cm each (diameter × height) placed on top of one another. The top Petri dish provided the nesting space while the bottom Petri dish contained water, creating a humid environment in the top box via a thin metallic mesh. After one week, queens were removed from all colonies and workers and brood were divided evenly between two identical nest boxes, creating a total of 60 orphaned groups. One half of each colony was assigned to the control group and the other to the treatment group, and thus control and treatment groups were of identical size. Groups were provided with water and sugar cubes ad libitum, and fed freeze-dried crickets and an artificial diet three times a week. Colony size in the initially collected colonies ranged from 550 to 1698 (mean ± SE: 1031 ± 57), with experimental/treatment groups half this size.

In the absence of a queen, A. senilis workers rear existing brood into replacement queens, and groups were monitored over the following 18 weeks for developing queen pupae. Queens were weighed and paint marked two days after emergence to permit them time to develop a hardened cuticle and be accepted by workers. In experimental groups, the left mandible of the first emerging queen was ablated at this time using small scissors. Queens were given several minutes rest before being returned to their group. Workers seemed to detect the injury to ablated queens, but behaviour soon returned to normal following reintroduction in both groups. Any sexual (queen) pupae reared after the first two queens had emerged were removed to simplify the selection process.

Groups were monitored daily for eclosion and survival of queens for the duration of the experiment (18 weeks). In addition, behavioural observations were made for the duration of the association. The scan method 30 was employed as in a previous study of this species 7 , with repeated observations made daily. Classification of behaviour followed table 1, and included a behavioural and location component for each observation. Aggression observed included antennation, 'stand-over' and biting (figure 1). The latter was almost always associated with curling of the gaster of the aggressor toward the victim. Although clipped queens had trouble actually biting the victim, they harassed subordinates in the same manner as intact queens. They occasionally managed to grasp narrow appendages such as antennae but in general could not bite effectively. Stand-overs were not necessarily coupled with antennation or biting, and involved the aggressor standing stationary over the victim with legs extended. While antennation was not overtly aggressive, it was usually associated with other dominance behaviours and, as all forms of interaction were usually unilateral (see below), it is included as a measure of dominance here.

Queen selection is a critical event in the life of a colony. It represents not only a potential conflict of interest between queens, workers, and the colony as a whole, but also a possibility for colony failure. Workers are known to be responsible for selection of queens in several species of ants, and cull queens based on a range of criteria 1 . It is in the interests of queens, however, to improve their chances of survival directly by fighting among themselves or indirectly through improvement of characters on which workers may select queens, such as reproductive readiness. A previous study 7 of queen replacement in Aphaenogaster senilis indicated that first-born queens were behaviourally dominant over subsequently produced queens and usually survived the queen selection process. Chéron et al 7 proposed that additional queens were produced as insurance against death or shortcomings of the first queen, and that queen selection was based on order of eclosion, presumably to minimise the delay to reproductive activity. They further suggested that production of insurance queens was delayed to permit first queens every advantage in selection, minimising chances of colony failure from both queens being injured during conflict.

Our results are similar with respect to timing of queen replacement and number of queens produced, and concur with Chéron et al 7 in that first-born queens were generally dominant and survived in the majority of cases. The importance of order of eclosion is reinforced by the lack of any effect of mandibular ablation on contest outcome or queen behaviour. However, in contrast to their results, our data show that while dominant individuals were more likely to succeed overall, dominance does not guarantee success, and behaviourally dominant queens occasionally lose contests. In addition, we also demonstrate that Q2s can be behaviourally dominant over Q1s. Finally, the finding that mandibular ablation has no effect is in marked contrast to the impact of this treatment in honeybees, where ablated queens actively avoid conflict 29 .

While dominance is important for survival in queens of A. senilis, the proximate mechanism triggering queen execution remains unclear. Indeed, disentangling worker and queen roles in queen elimination can be problematic: workers attack already injured queens in Solenopsis invicta 15 16 32 , and queen elimination by workers was not significantly distinct from queen-queen aggression in Messor (Veromessor)pergandei 33 . The long duration of cohabitation after fighting bouts in A. senilis argues against queens using chemical means to 'mark' subordinates for execution or immobilisation (eg: 34 35 ) and no behaviour that could be construed as marking was observed. Aggression can have direct affects (eg: injury) or indirect affects (eg: ovarian suppression). Mandibular ablation in A. senilis clearly affected the capacity to fight as queens predominantly use biting as a means of aggression, and ablated queens had difficulties grasping opponents. Several factors however, point to a lack of any direct affect of fighting in contest resolution. Firstly, the duration of queen cohabitation was prolonged (mean 16 days) and did not differ between control and treatment groups. We witnessed multiple bouts of aggression in some groups, some of which were extended (>35 minutes), yet at no time did this clearly result in an injury to either queen or lead to queen culling. In contrast, queen contests in honeybees are resolved fatally and relatively quickly (4s-15 mins 36 ). Secondly, whereas mandibular ablation had a profound effect on the behaviour of honeybee queens, which avoided contests when ablated 29 , the behavioural profile of A. senilis queens did not differ between treatments. This suggests mutilated Q1s did not detect any reduced chance of victory despite having ample opportunity to assess their condition during bouts. In addition, Q2s did not behave differently when aggressed by a mutilated versus an un-mutilated Q1, suggesting that they did not assess the mutilated Q1s as weakened. Finally, aggression was always unilateral: subordinates never fought back or attempted to escape (whereas they did flee from mass worker aggression) suggesting they did not consider themselves under direct threat from queen aggression. These data suggest that aggression in A. senilis is ritualised and serves primarily as a dominance display rather than a direct means of resolving contests. This may explain the different response to ablation in honeybees, where fighting between queens is the primary means of contest resolution and bouts are usually fatal 3 .

Dominance is important in reproductive success in many species, and can be correlated with productivity or fitness 2 . In Leptothorax sp. A dominance was correlated with ovarian status and attractiveness to workers, and lead to preferential feeding and eventual expulsion or killing of subordinates 37 38 . We did not see preferential feeding, but differential treatment of Q1 and Q2 may be indicated by lower observed allogrooming of Q2s. Aggression probably acts to suppress the development of characters upon which worker selection acts, such as reproductive ability or pheromone production 39 40 . The observation of one apparently healthy queen being attacked and killed by workers supports the idea of worker mediated queen culling. Yet if workers are controlling selection and presumably would favour rapid resolution of contests 2 8 , this raises the question of why some contests are so prolonged. One explanation is that asymmetries between queens take time to develop and workers are waiting for an asymmetry threshold to be reached. Differential ovarian development and chemical signalling of this condition is one possibility (eg: 41 ). Development of ovaries and signalling of this condition takes around 6-8 weeks in the ponerine ant Dinoponera quadriceps 42 , which is considerably longer than our mean coexistence time of 16 days. Worker awareness of ovarian development has been demonstrated in Aphaenogaster cockerelli and workers kill "cheating" (laying) workers 43 44 . Another possibility is that workers await one queen to be mated before culling the other queens. As mating did not occur in the laboratory, this could result in longer co-inhabitation of queens than under natural conditions. Q1s have a distinct head start in development of any such characters, and Q2s can only prevail presumably if Q1 is of relatively poor fitness in general. These data support the "supernumerary queens as life insurance" hypothesis of Chéron et al 7 and indicate that, rather than fighting ability, the process of queen selection is based on advantages relating to acquired asymmetries, which are in turn at least partially associated with order of eclosion. Aggression based dominance may aid first-born queens in the contest to reach the acquired asymmetry threshold that triggers worker intervention, but aggression is probably not directly involved in contest resolution.

Aggression is, however, important in the establishment of dominance, and a problem that remains is explaining the lack of aggression from subordinate queens. Although some subordinate queens were able to prevail, dominant queens succeeded in 73% of cases for which dominance could be ascertained. Queens produced during emergency queen production in A. senilis have few options other than inheriting the colony, as dispersal is only possible through colony fission which can only occur in large colonies and under certain conditions 5 . Thus, not fighting may effectively be a form of suicide for Q2s. Reduced aggression in animal contests is often associated with the use of conventions to decide contests, which act to remove or minimise the risks of overt fighting 45 . Prior residency is one convention that is common in a range of animal taxa 46 , though the mechanisms underlying the effect have been the subject of much debate 47 . The prior residency effect operates when challengers for a resource back down in favour of residents, as in the 'bourgeois' strategy of game theory 9 48 . For example, in the parasitoid wasp Trissolcus basalis, the first arriving female at a patch of host eggs is more likely to win contests than intruders, who back down before escalated conflict 49 . Similarly, first-born queens in A. senilis are effectively prior residents, and in most cases this translated to behavioural dominance and eventual victory. Queens are thus basically following a 'bourgeois' convention. However, game theory models also predict that when the value of the contested resources represents the majority of an individual's possible lifetime fitness, contestants should fight regardless of ability, kinship or asymmetries 50 51 52 . Hence, while asymmetries may discourage lesser individuals from overt fighting, fighting should occur regardless when reproductive alternatives are scarce or non-existant. For example, queens of Apis mellifera with one mandible ablated avoid contests, but will fight if challenged 29 . Our data suggest that subordinate queens are almost never aggressive, despite the high value of the contested resource and the very low success of subordinates. One aspect of the biology of A. senilis that may be important is the high relatedness between queens, which are full sisters (r = 0.75; 7 ). In this case a loser can still gain considerable indirect fitness even if it is killed, and this indirect fitness may be greater than the benefits of fighting for inheritance if fighting entails risk to both queens. By obeying a bourgeois strategy and ceding to first-born queens, second-born queens can minimise the chance of both queens being injured while also facilitating rapid commencement of reproductive activity in the colony.

Queen selection is an important process in the lifecycle of a colony as it influences the fitness of all colony members. First-born Aphaenogaster senilis queens were behaviourally dominant over second-born queens, and inherited the colony in most cases. Ablating one mandible of the first-born queen did not affect contest outcomes, suggesting aggression is probably a form of ritualised dominance. Queens are likely selected by workers based on reproductive quality, with first-born queens gaining a significant time-related advantage over subsequently-born queens. Queens may follow a bourgeois strategy in contests, with second-born queens effectively committing suicide by ceding to first-born queens. This minimises the chance of colony failure and reproductive delay through injury or death of both queens, and is possible because queens are full sisters and thus highly related.

The authors declare that they have no competing interests.

Both authors conceived of the study, designed and executed the experiment, and contributed to the final manuscript. ALC performed analyses and drafted the manuscript.