One of the defining features of advanced eusocial groups is reproductive division of labor, where one or a few individuals specialize on reproduction while others perform strictly nonreproductive tasks such as brood care, defense and foraging. Recent theoretical work suggests that the rudiments of division of labor may originate spontaneously during initial group formation as an emergent property, rather than requiring a secondary adaptation. Empirical studies on nonreproductive tasks support the emergence hypothesis, but it is unclear whether this mechanism also extends to reproduction. To test whether reproductive division of labor can be produced as an emergent property, we assessed the extent and mechanisms of both nonreproductive and reproductive division of labor in forced associations of normally solitary queens of the harvester ant Pogonomyrmex barbatus. We find that division of labor in both types of tasks can be induced in groups of individuals with no evolutionary history of social cooperation. Specialization in excavation behavior was more pronounced than reproduction, which tended to be incomplete although significantly skewed. In addition to reproductive division of labor, enhanced productivity in forced pairs relative to solitary queens suggests that both queens contributed cooperatively to brood care despite unequal maternity. Thus, two of the three defining features of eusociality may have originated through self-organizing mechanisms concurrently with the evolution of grouping, exposing these social strategies to selection early on in the evolution of social life.
The evolution of reproductive altruism has been an enduring puzzle that has long fascinated evolutionary biologists. Although reducing personal reproduction appears maladaptive by definition, reproductive division of labor is a fundamental feature of the eusocial insects and likely critical to their striking ecological success (Wilson, 1971). This paradox has made deciphering the context in which such behavior would be favored a popular, but contentious, field of inquiry. Indeed, the study of the evolution of eusociality has been plagued by disagreements: over how to describe the attributes of social groups (Crespi & Yanega, 1995; Sherman et al., 1995); the level at which selection acts (Wilson, 1975; Foster, Wenseleers & Ratnieks, 2006); and the proper focus of fitness calculations (Taylor, Wild & Gardner, 2007). These arguments continue in the literature to this day (Nowak, Tarnita & Wilson, 2010; Abbot et al., 2011; Boomsma et al., 2011; Bourke, 2011).
One challenge in evaluating alternative selective hypotheses for the evolution of eusociality is that eusociality is itself composed of multiple social traits, whose evolutionary history must be properly reconstructed in order to understand the conditions in which they evolved. Traditional models of social evolution posit that initial social groups formed due either to a loss of dispersal (the subsocial route) or active group formation (the parasocial route). These groups presumably functioned as aggregations, with all individuals producing and caring for their own offspring (Oster & Wilson, 1978). Asymmetries in these behaviors evolved as secondary kin-selected adaptations (Hamilton, 1964), which were then amplified as concentration of reproduction into one or a few individuals released constraints on morphological adaptation to more and more specialized subsets of tasks (West-Eberhard, 1989; Gadagkar, 1997). Thus, to arrive at eusociality from a solitary ancestor required multiple successive evolutionary changes: first, a shift from solitary breeding to either loss of dispersal or group joining; second, a shift from independent to cooperative brood care; and third, a shift from equal reproductive efforts to cessation of reproduction by some group members.
Although this stepwise view follows a logical progression of social structures, the number of evolutionary transitions involved suggests that eusociality should be relatively difficult to evolve, as a multistep process would require adaptive benefits at every stage along the route, not just an adaptive end point. An alternative possibility that avoids this problem is that the evolution of eusociality may have occurred in a single evolutionary step: a shift in dispersal tendency or timing is accompanied by other eusocial traits as an automatic side effect (Michener, 1985; Linksvayer & Wade, 2005; Nowak et al., 2010). In the last two decades, there has been increasing appreciation that many of the attributes of social groups, including movement patterns, spatial organization, nest architecture, decision making and division of labor, can emerge from self-organizing processes as individuals interact with and respond to a dynamic social and physical environment (Bonabeau et al., 1997; Fewell & Page Jr, 1999; Parrish, Viscido & Grünbaum, 2002; Theraulaz et al., 2003; Couzin, 2008). Importantly, emergence mechanisms require only spatial proximity among individuals, leading to novel behaviors and patterns without underlying genetic changes in behavioral strategy as individuals interact with one another and their shared environment. If the defining features of eusociality are similarly self-organizing in nature, this would provide a mechanism for their appearance in a single step at the origin of group formation.
Critically evaluating these alternative trajectories of social evolution is not straightforward, as the initial characteristics of extant social species, whether emergent or not, are likely to have long since been superseded by secondary adaptations to social life. One approach to recovering what incipient groups may have been like is to assemble artificial social groups in species that are normally solitary, but tolerant enough of conspecifics to persist in groups without fatal aggression and group dissolution. Because such individuals have no evolutionary history of social cooperation, their behaviors under experimental group formation should be a function of their intrinsic behavioral repertoires and any emergent properties resulting from interactions with the shared physical environment and/or other group members. As predicted by the emergent property hypothesis, artificially assembled groups of insects that are normally solitary during the life stage being investigated show pronounced division of labor in nonreproductive tasks such as nest construction and defense, suggesting that these can emerge from self-organizing processes (Fewell & Page Jr, 1999; Helms Cahan & Fewell, 2004; Jeanson, Kukuk & Fewell, 2005; Jeanson & Fewell, 2008; Holbrook et al., 2009). However, whether self-organization can also cause the emergence of division of labor in reproduction has scarcely been investigated, despite its centrality to the origin and elaboration of eusociality (Sakagami & Maeta, 1987). This has led authors to question whether emergent property scenarios have any applicability to the evolution of eusociality (Bourke, 2011; Duarte et al., 2011; Herre & Wcislo, 2011).
In this study, we experimentally test whether self-organizing mechanisms can spontaneously generate reproductive division of labor by creating forced associations of colony-founding queens of the harvester ant species Pogonomyrmex barbatus. Although ants show highly derived eusocial structure during most of the life cycle, queens of many species found colonies alone, excavating the nest and rearing the first cohort of workers in complete social isolation. Because queens are strictly solitary during this period, they should be selected for a behavioral repertoire similar to that of a hypothetical solitary ground-nesting ant ancestor. Thus, any changes in behavior induced by a social context should be solely a proximate response to novel conditions, rather than an evolved strategy (e.g. Fewell & Page Jr, 1999; Helms Cahan & Fewell, 2004; Jeanson & Fewell, 2008). To investigate emergence of reproductive division of labor, we quantified the allocation of reproduction between two queens when forced to cofound a nest, and compared the extent of division of labor in reproduction to that of nest excavation, a nonreproductive behavior previously shown to emerge spontaneously in forced associations (Fewell & Page Jr, 1999). To investigate how division of labor in these two tasks might be generated mechanistically, we tested whether task specialists could be predicted by their relative size, aggressive behavior or performance of other tasks.
Study species and collection
The harvester ant P. barbatus is a large-bodied granivore whose range extends from central Mexico to southeast Arizona in the west and south-central Texas to the east (Johnson, 2000). All North American species in this genus are exclusively solitary founding with the exception of a single group-founding population of the desert specialist P. californicus, distantly related to P. barbatus (Parker & Rissing, 2002; Helms Cahan & Fewell, 2004). Colonies reproduce after monsoon rains in mid-summer by releasing large numbers of winged males and virgin queens that congregate in dense mating swarms. Following mating, queens disperse aerially from the mating swarm site, remove their wings and begin to excavate an incipient nest in the soil. When the nest is complete, the queen seals the entrance from the inside and begins egg-laying and brood care. Queens do not forage during this period and feed the brood from excess eggs and other secretions derived from the queen's muscle and fat reserves. Depending on the temperature, the first workers emerge in 6–8 weeks, open the nest and assume all nonreproductive tasks.
Experiments were conducted over 2 years: 90 newly mated queens were collected on the ground following a mating flight in Hidalgo Co., New Mexico, on 27 July 2011, and 108 were collected from a similar mating flight in Santa Cruz Co., Arizona on 7 July 2012. These populations contain two distinct genetic lineages, referred to as J1 and J2, and produce workers solely from interlineage crosses (genetic caste determination, Julian et al., 2002; Volny & Gordon, 2002a). The two lineages can be distinguished genetically (Helms Cahan & Keller, 2003; Schwander, Helms Cahan & Keller, 2006) but not visually or behaviorally, and queens were paired without regard to their lineage identity. Although previous studies have found productivity differences between lineages due to differences in fecundity and/or sperm stores (Anderson et al., 2006, 2011; Helms Cahan et al., 2010), comparison of the reproductive output of J1 and J2 queens kept alone in this study revealed no differences in intrinsic productivity between lineages (t26 = 1.62, P = 0.12) and there was no overall bias toward a particular queen lineage in J1/J2 pairings that would suggest that lineage identity systematically affected reproductive role (J1 > J2 in five cases, J2 > J1 in five cases).
Queens were isolated with moist paper towels in individual plastic shipping tubes and shipped overnight to the University of Vermont. Upon arrival, queens were individually weighed to the nearest 0.01 mg with a Mettler Toledo microbalance (AX 205 Microbalance, Mettler-Toledo, Columbus, OH, USA) and painted with one of three different colors of Testors paint pens on the thorax. Pairs of queens differing in paint color and an equal number of single ‘control’ queens were placed into 600-mL bottles 2/3 filled with damp soil in which the queens could excavate a nest and rear brood in a seminatural soil-filled tunnel. Thirty sets of bottles were set up in 2011, and 36 in 2012.
Division of labor could emerge as a result of multiple types of self-organization mechanisms (Duarte et al., 2011), including agonistic social interactions (Jeanson et al., 2005). To determine whether agonistic interactions drive division of labor between queens, we quantified the extent and symmetry of aggressive behavior when queens were first introduced. All pairs of queens in both nest types were observed in groups of six nests for the first 15 min following their release into the nest. All instances of aggressive behaviors (Table 1) performed by each queen during this period were recorded.
Table 1. Set of aggressive behaviors scored during behavioral interactions of Pogonomyrmex barbatus queens. Behaviors are listed in order of increasing escalation intensity
Rapid beating of antennae on the body of second queen
Opening mandibles to full width while facing second queen
Rapid raising and lowering of gaster when in contact with second queen
Following retreating queen
Standing over or on the second queen with forelegs
Biting the second queen on the body or legs
Carrying the second queen in mandibles
The contribution of each queen to excavation behavior was quantified by intensive observations of groups of 20 nests for 15-min intervals in which all instances of excavation behavior by each queen were noted. A subset of five nests in a set was scanned by a single observer for 3–5 s before moving to the next subset, resulting in approximately two scans per minute per nest over the entire 15-min interval. All observations were conducted over a period of 2 days, after which excavation behavior had ceased and the majority of nests were sealed with soil. In 2011, nests were observed for a total of 10 observation periods; this was increased to 15 in 2012 to better capture high-intensity excavation bouts in the first few hours following queen introduction.
Colonies were collected in week eight, when the brood in the majority of colonies contained darkening pupae and/or workers. All surviving queens, larvae, pupae and workers were counted and preserved in 95% ethanol. Any pairs in which one or both queens had died prior to collection were excluded from reproduction comparisons. To determine queen lineage identity and reproductive apportionment in paired nests, DNA was extracted from a leg or the head of each queen from both the paired and control nests, and the whole body for all brood from paired nests using a standard Chelex-100 rapid extraction protocol (Helms Cahan et al., 2006). To determine queen lineage identity, the Cox1 mitochondrial gene was amplified as described in Schwander et al. (2006) and cut with the restriction enzymes MfeI or PvuII, which both have been shown to have a diagnostic restriction site in the J1 but not the J2 sequence (Schwander et al., 2006; Anderson et al., 2011). To avoid accidental misidentification of lineage due to enzyme inactivity, all digestion assays included a known J1 sample as a positive control. Because mtDNA is inherited matrilinearly, offspring from nests with queens that differed in lineage were typed in the same manner to identify matriline. The offspring of pairs of queens from the same lineage were distinguished using six highly variable microsatellite loci: Pb2, Pb4 and Pb9 (Volny & Gordon, 2002b), and Po3, Po7 and Po8 (Wiernasz, Perroni & Cole, 2004). Both queens and all offspring were genotyped for each pair. Offspring maternity was assigned using a strict maternity exclusion criterion (i.e. no alleles in common between the offspring and one queen at a microsatellite locus). Maternity exclusion is facilitated in these populations because worker offspring are exclusively of J1/J2 hybrid ancestry (Julian et al., 2002; Volny & Gordon, 2002a); because the J1 and J2 lineages show diagnostic or strong allele frequency differences at most microsatellite loci (Volny & Gordon, 2002b; Table 2), the paternal allele was invariably from the alternate lineage and thus was rarely shared with either putative mother. This combination of loci allowed one parent to be excluded as the mother at one or more loci for all offspring in all but 2 out of 20 surviving pairs, which were excluded from parentage analysis.
Table 2. Microsatellite genotypes and maternity apportionment for queens in same-lineage pairs. Gray shading indicates loci at which paired queens shared no alleles in common
We took a uniform statistical approach to quantify the relative contribution of each queen to aggression, excavation and reproduction. For each behavior, we used a simple measure of task sharing that ranged from 0 (all actions by a single queen) to 1 (equal task performance): the number of times that the lower frequency (LF) queen performed the task divided by the number of times the higher frequency queen (HF) performed the task. For aggression, a decreasing value in this index measures social dominance of one queen over the other, as indicated by the extent to which aggressive behaviors are one sided. For excavation and reproduction, a decrease in the index represents more pronounced division of labor.
Although queens perform many individual acts of excavation during nest construction, relatively few worker offspring are produced in the first cohort of workers (range = 0–17). This makes it difficult to achieve sufficient statistical power to test whether queens contribute equally to offspring production at the level of individual colonies. Instead, for both tasks, we focus here on whether the entire distribution of values matches the distribution expected if any bias in task performance were produced solely by intrinsic variation among queens in propensity to perform the behavior in question (excavation or reproduction). A custom Python script (provided as Supporting Information Text S1)was used to generate a predicted distribution of LF/HF values by calculating this value for 1000 simulated queen pairs, each generated from the observed task performance of two single-queen controls sampled randomly with replacement. To test whether the observed and expected distributions differed significantly, we used a second script (Supporting Information Text S2) to calculate the distribution of median LF/HF values expected from intrinsic variation alone given the sample size available by calculating median LF/HF values for sets of simulated values of the same sample size as the observed number of paired-queen colonies. This calculation was repeated 1000 times and the bottom 5% provided the cutoff for statistical significance.
To determine the relationships among relative size differences, social dominance, excavation performance and reproduction, paired queens were ordered at random and the proportional difference between paired queens in each variable was calculated as the natural log of the ratio of the first over the second queen. For all variables except size, 1 was added to the individual queen values to prevent division by zero. This created an index symmetrical around 1 (equal values), with roughly half of pairs above 1 and half below for any given variable. We then tested for relationships among these variables using multiple linear regressions. Given the sequential nature of dominance, excavation and reproductive behaviors, we analyzed each of these dependent variables against only those independent variables which preceded its expression (i.e. only size differences were considered to affect dominance, while both size and dominance were included when analyzing excavation).
Although P. barbatus queens can be maintained in pairs, queens were generally aggressive toward one another when forced to associate. Over both years, pairs of queens displayed at least one aggressive behavior in 78.5% of cases, ranging from 1 to 18 discrete aggressive actions displayed in a 15-min observation period. Of these, in 47% (24 of 51 pairs) all aggression was displayed by one of the two queens. In the remaining pairs, aggressive behaviors were initiated to some extent by both queens. Aggressive behaviors did not appear to produce any visible injuries that could have physically impaired excavation behaviors. Aggressive behaviors were not more likely to be initiated by the larger queen in the pair (t53 = 1.05, P = 0.30, Fig. 1a). Despite the high level of initial aggression, continued aggression after excavation had commenced was relatively rare, observed between queens more than 2 hours after introduction in only 8 of the 65 total nests (12.3%).
A queen performed at least one instance of excavation behavior in 61 of 63 single nests (97%) and in all 65 paired nests. Total excavations observed in single and paired nests did not differ in 2011 (singles: 15.33 ± 1.66 excavations, doubles: 14.14 ± 1.18; Student's t-test, t53 = 0.71, P = 0.48) or in 2012 (singles: 33.97 ± 3.66, doubles: 40.18 ± 3.85; t53 = −1.17, P = 0.25; Fig. 2). Although per capita excavation by queens in pairs was approximately half that of singles, however, the two queens did not reduce their task performance equally (Fig. 2). Consistent with previous work on this species (Fewell & Page Jr, 1999), the distribution of task sharing in excavation of observed pairs was significantly lower than that predicted from the extent of intrinsic variation in excavation behavior displayed by queens when founding colonies alone (2011: predicted median = 0.55, observed median = 0.19, P < 0.01; 2012: predicted = 0.44, observed = 0.27, P < 0.05; Supporting Information Fig. S1), with the largest excess in the most extreme level of excavation skew, where one queen performed little or no excavation while the behavior of the excavation specialist was either slightly but significantly lower (2011: t44 = 2.25, P < 0.05) or not significantly different (2012: t62 = 0.61, P = 0.54) from that of solitary queens (Fig. 2). Relative performance of excavation behavior was significantly predicted by patterns of queen–queen aggression during the first 15 min of pair formation (t50 = 2.02, P < 0.05; Fig. 1c), but not by relative size differences (t50 = −0.24, P = 0.81; Fig. 1b).
Both paired queens survived to brood collection in 6 of 28 colonies in 2011 and in 14 of 35 colonies in 2012. In contrast to excavation, total productivity in colonies with two surviving queens was significantly higher than productivity in single nests (main effect of queen number, F1,60 = 23.51, P < 0.001), with the two nest types not differing significantly in average per capita productivity (Student's t-test, t57 = 1.38, P = 0.17; Fig. 3). As with excavation, however, the allocation of offspring in paired nests was significantly more skewed toward a single queen than that predicted by the distribution of productivity values from single queens (predicted median = 0.60, observed median = 0.40, P < 0.01; Supporting Information Fig. S2). This was caused by both a significant increase in reproduction by HF queens (t38 = 2.05, P < 0.05) and by a reduction in reproduction by the LF queen (t46 = 5.39, P < 0.001) relative to single queens (Fig. 3). Reproductive role was not significantly associated with relative size (t13 = 0.52, P = 0.61; Fig. 4a), relative social dominance (t13 = 0.39, P = 0.71; Fig. 4b) or excavation role (t13 = 0.49, P = 0.63; Fig. 4c).
Social life involves a complex interplay between individual behavior and patterns expressed at the level of the group as a whole, with the potential for complex group-level patterns and collective behavior from relatively simple individual decision rules (Camazine et al., 2001). Critically, higher level patterns emerge whenever individuals form interactive groups, which may or may not be adaptive but in many cases mimic the properties of socially adapted taxa (Parrish et al., 2002). Our experimental results support the notion that self-organization can produce reproductive division of labor, as predicted by an emergent property model. Forced associations of normally solitary ant queens showed incomplete but significant variation in reproduction that was more biased toward one queen than that predicted by intrinsic individual variation in productivity. Moreover, paired queens were nearly twice as productive as single queens; given that individual queens are limited in their maximal contribution to offspring biomass by their own fat and muscle reserves, this suggests that both queens contribute to brood care despite their unequal genetic contributions to the offspring. Thus, even in the absence of adaptation to social colony founding, the ‘default’ character of these queen groups includes a rudimentary form of two of the three essential features of eusociality: reproductive division of labor and cooperative brood care.
Self-organization can produce division of labor via a number of different mechanisms, which vary in the how individuals interact with their environment and one another. Intrinsic variation in stimulus response thresholds, for example, can result in specialization if the task stimulus induces the lower threshold individual to initiate the behavior sooner than the higher threshold individual, resulting in a feedback loop as task performance by that individual further reduces the task stimulus encountered by the other (Page Jr & Mitchell, 1991; Page Jr & Robinson, 1991). Previous work on excavation specialization in P. barbatus queen associations was consistent with this mechanism (Fewell & Page Jr, 1999): which queen would become the excavation specialist could be predicted by their excavation propensities when alone, and the primary change in behavior when groups were formed was the cessation of excavation by the lower frequency queen. Similarly, we found that the primary change in excavation behavior when pairs were formed was task reduction by one of the two queens; in c. 40% of cases, one queen performed little to no excavation (Supporting Information Fig. S1), an exceedingly rare rate of task performance in solitary queens.
In addition to a response threshold mechanism, we also found evidence that interindividual social interactions may play a role in mediating excavation role. As expected for queens that typically repel conspecifics from their nest site, forcing queens into a restricted shared nesting space led to aggressive displays in the majority of nests. Importantly, aggressive behaviors were often asymmetrically performed, and the ‘winner’ of these agonistic interactions was more likely to become the excavation specialist. It is likely that agonistic interactions reinforce existing propensity differences, leading to more extreme task specialization. Aggressive interactions tended to produce spatial segregation of queens within the arena, as losers of encounters tended to avoid the winner, either remaining immobile on the soil surface or attempting to enter the incipient nest. It is notable that the extreme skew in excavation performance seen in this species exceeded that observed in another Pogonomyrmex species where agonistic interactions were rare (Jeanson & Fewell, 2008).
As with excavation behavior, we found that reproductive division of labor also emerged when normally solitary queens were placed in a novel social context. Although few colonies were completely monopolized by a single queen, the relative contribution of the lower frequency reproducer was significantly lower than that expected solely from intrinsic variation in productivity, with a median output of only 40% of that of the higher frequency queen. Thus, despite its fundamental importance for fitness, in a mechanistic sense, reproduction is not exceptional and appears to be responsive to the same types of social modulators as other behaviors. Unlike excavation, reproductive role was unrelated to social dominance status, which may explain why relatively few pairs displayed complete reproductive specialization in the manner seen with excavation (Fig. 4). Although aggression was common while queens were initiating excavation behavior, it rarely extended more than a few hours into nest excavation and thus was resolved by the time egg-laying commenced days later. Instead, we hypothesize that the emergence of reproductive division of labor resulted primarily from a signal-response mechanism: initially, small individual variation in the onset of egg-laying became amplified as the queen who initiated the egg pile was further stimulated by physical contact with the existing eggs. Queen pairs tended to maintain a single brood pile at the end of a narrow tunnel at the bottom of the nesting bottle (E. Gardner-Morse, unpubl. data), potentially permitting a single queen to monopolize contact with the brood by blocking the tunnel with her body.
Reproduction was also unrelated to excavation role, indicating that the emergence of reproductive division of labor was inherent to this task rather than resulting from a trade-off in investment among potential tasks (Jeanson et al., 2007). This differs from the congener P. californicus, in which the two tasks are negatively related (Jeanson & Fewell, 2008); a key difference may be that P. californicus nests in loose, sandy soil and does not seal the nest entrance (Johnson, 2004; Enzmann & Nonacs, 2010), extending the duration of this task well into the egg-laying period and creating an excavation-reproduction tradeoff in individual time budgets. This tradeoff is further exacerbated in P. californicus by the fact that queens actively forage for resources, limiting the time available for other tasks and physically separating the forager from the nest and brood (Johnson, 2002; Dolezal et al., 2009).
One striking difference between the two tasks is the effect of queen pairing on total task performance. Unlike excavation, in which the total number of excavation trips did not differ between single-queen and paired-queen nests, paired nests produced double the number of worker offspring as single-queen nests. Reproductive efforts were altered in both directions relative to solitary queens, with HF queens significantly increasing productivity while LF queens were less productive than solitary queens (Fig. 3). This is surprising, as initial egg number is presumably homeostatically regulated in order to preserve energy for brood rearing, and thus queen egg-laying should be responsive to the total quantity of eggs regardless of queen number. One possible explanation is that the increase in HF productivity stems not from increased investment into egg-laying, but increased brood survival. Because queens do not forage during the founding period, they are inherently constrained in their initial productivity by their own energetic reserves, which are used to provide resources to the developing worker cohort. Founding queens typically rear only a small proportion of the initial batch of eggs laid, with the remainder consumed by developing larvae (Baroni Urbani, 1991; Wheeler, 1994; Liu et al., 2001). The fact that single and paired nests did not differ in per capita productivity suggests that the LF queen fully invests in brood rearing despite significantly lower maternity. If LF queens provide generalized brood care, a larger proportion of eggs may hatch and develop successfully for both queens and result in an effective increase in productivity for the HF queen.
The results of this study suggest that the component elements of eusociality are surprisingly easy to produce through self-organizing mechanisms, and raise the possibility that initial social groups might have already possessed the rudiments of a ‘derived’ social structure without requiring intervening secondary adaptations. This is consistent with earlier suggestions that nonreproductive helping originated as an automatic consequence of failing to disperse, arising spontaneously when nondispersing adult offspring are exposed to the provisioning stimulus of begging siblings (West-Eberhard, 1987; Jamieson, 1989). A recent paper suggested that evidence for an emergent origin of eusocial traits would argue against the importance of kin selection in the evolution of eusociality (Nowak et al., 2010); however, these two types of explanation operate at different levels of analysis and are not actually alternatives. It is important to distinguish between the evolutionary origin of a trait, which may well be emergent, and its evolutionary fate, which depending on its fitness returns, may be eliminated, reinforced or enhanced through underlying genetic changes (Ligon & Stacey, 1991). The fact that division of labor emerges from solitary behavioral strategies placed in a social context does not mean that such patterns are inevitable; if division of labor were not advantageous, selection could lead to changes in the underlying individual properties that form the building blocks of its expression, such as decreasing individual response thresholds to social cues, changes in the connections between stimulus inputs and behavioral outputs to minimize performance of high-cost tasks, or adaptive modification of behavioral responses in response to group size or composition.
Although solitary colony founding is the rule in the genus Pogonomyrmex, group colony founding has evolved repeatedly in the social Hymenoptera (Bernasconi & Strassmann, 1999). Thus, comparing the behavioral features of naturally evolved associations to the emergent patterns observed in P. barbatus associations may provide some insights into the fitness consequences of division of labor under different ecological and social contexts. Most notably, reproductive skew in natural ant queen associations tends to be lower than that observed here, ranging from intermediate (∼50% of pairs in Lasius niger, Aron, Steinhauer & Fournier, 2009) to low or absent (Messor pergandei, Rissing & Pollock, 1986; Pachycondyla cf. ‘inversa’, Kolmer & Heinze, 2000; Crematogaster morphospecies 2, Feldhaar, Fiala & Gadau, 2005). This would imply an evolutionary reduction in division of labor, if incipient groups displayed reproductive specialization as an emergent property. Importantly, ant associations are unrelated (Hagen, Smith & Rissing, 1988; Helms Cahan & Helms, 2012), so there may often be a direct opposition between a queen's individual fitness interests and reduction of individual reproduction, particularly if queens compete for reproductive dominance and can increase their likelihood of survival via enhanced fecundity (Balas, 2005; Holman, Dreier & D'Ettorre, 2010).
On the other hand, division of labor that is expressed in a context in which it is advantageous, either through direct or indirect fitness returns, may be maintained or evolutionarily enhanced. In species in which reproductive turnover is likely, individuals that initially reduce their reproductive output may later inherit a well-established nest, providing a direct fitness benefit to assuming the LF role (e.g. allodapine bees, Schwarz et al., 2011). Strong reproductive division of labor also occurs in many wasp foundress associations, which are typically composed at least partially of full-sisters and thus nonreproducers have more potential to reap indirect benefits (Strassmann, 1981; Uddin & Tsuchida, 2012). In such circumstances, traits that may have been initially context dependent can be converted into much more canalized phenotypic plasticity in response to social and environmental cues, and ultimately result in discrete, specialized polyphenisms. This is clearly the case for division of labor among workers, in which self-organization mechanisms are an important mediator of colony-level patterns and are enhanced by within-colony genetic variability, nest spatial complexity, and size- and age-related changes in behavioral propensities (Bonabeau, Theraulaz & Deneubourg, 1996; Huang & Robinson, 1996; Julian & Fewell, 2004).
Like physiological and molecular models of social evolution, group-level behavioral patterns are likely to begin with a ‘behavioral toolbox’ shaped by the requirements and regulatory architecture of a solitary life history, which can then be modified and co-opted for the novel selective pressures imposed by social life. As this and other studies make clear, solitary behavioral strategies can produce surprising and novel phenotypes simply by being moved into a social context, which may play an important role in determining the raw material on which selection acts at the origin of social life. Evolutionary self-organization models of division of labor have not generally considered what individuals bring to the table behaviorally when they enter into groups from a solitary lifestyle, but they would likely benefit from explicitly considering how starting conditions affect subsequent evolutionary outcomes, for nonreproductive and reproductive behaviors alike (Duarte et al., 2011).
We would like to thank M. Herrmann, J. Grauer, Y. Hernáiz-Hernández and D. Bartolanzo for their dedicated efforts collecting founding queens in the field. A. Nguyen provided valuable assistance with behavioral observations. This work was funded by NSF grant DEB-0919052 to S. Helms Cahan, and a University of Vermont Undergraduate Research grant to E. Gardner-Morse.