The Complexity of Social Complexity: a Quantitative Multidimensional Approach for Studies of Social Organization

vol. 196, no. 5 the american naturalist november 2020

Synthesis The Complexity of Social Complexity: A Quantitative Multidimensional Approach for Studies of Social Organization

Jacob G. Holland*,† and Guy Bloch*

Department of Ecology, Evolution, and Behavior, Alexander Silberman Institute of Life Sciences, Hebrew University of Jerusalem, Jerusalem, Israel Submitted August 15, 2019; Accepted June 9, 2020; Electronically published September 11, 2020 Dryad data: https://doi.org/10.5061/dryad.x3ffbg7g6. abstract: The rapid increase in “big data” during the postgeno- implies that a system is made from many specialized and mic era makes it crucial to appropriately measure the level of social interacting parts that together contribute to a function complexity in comparative studies. We argue that commonly used (e.g., a complex tissue is composed of more cell types than qualitative classifications lump together species showing a broad a simple tissue, and a complex organism has more tissue range of social complexity and falsely imply that social evolution and cell types than a simple organism; Valentine et al. always progresses along a single linear stepwise trajectory that can 1994; McShea 2000). Studies of insect sociality have typi- be deduced from comparing extant species. To illustrate this point, “ cally used a different approach that is based on a small number we compared widely used social complexity measures in primi- fi tively eusocial” bumble bees with “advanced eusocial” stingless bees, of qualitatively de ned social traits, including reproductive honey bees, and attine ants. We find that a single species can have division of labor within a colony, cooperative brood care, both higher and lower levels of complexity compared with other and an overlap of multiple adult generations (table 1). This taxa, depending on the social trait measured. We propose that mea- approach has been central to defining the “eusocial” (truly suring the complexity of individual social traits switches focus from social) insects (Wilson 1971; Michener 1974; Batra 1966) as semantic discussions and offers several directions for progress. First, well as other classifications and subdivisions. Within an quantitative social traits can be correlated with molecular, develop- evolutionary framework, social complexity is typically used mental, and physiological processes within and across lineages of social animals. This approach is particularly promising for identifying as a proxy for the degree of social group transformation, de- processes that influence or have been affected by social evolution. scribing how groups of formerly independent entities (such Second, key social complexity traits can be combined into multi- as cells or multicellular organisms) become new organism- dimensional lineage-specific quantitative indices, enabling fine-scale like units (Maynard Smith and Szathmáry 1995; Keller comparison across species that are currently bundled within the same 1999; Bonner and Brainerd 2004; Bourke 2011; Szathmary level of social complexity. 2015; West et al. 2015). The richness and diversity of social Keywords: social complexity, social evolution, primitive eusociality, insects have provided fertile grounds for cultivating ideas social insects, bumble bees. on the evolution of complexity and sociality. Social insects have formed the primary examples of the major evolutionary transition from a multicellular organism to a society of Introduction multicellular organisms and for the kin and multilevel se- What Is Social Complexity? lection theories (Hamilton 1964; Maynard Smith and Szath- máry 1995; Keller 1999; Bourke 2011). Although there is no universal definition of “complexity,” in biological systems functional complexity commonly Limitations with the Common Classifications * Corresponding authors; email: [email protected], guy.bloch@mail of Social Complexity in Social Insects .huji.ac.il. † Present address: Institute of Evolutionary Biology, School of Biological The framework in which we assess social complexity is Sciences, University of Edinburgh, Edinburgh, United Kingdom. critical for the interpretation of comparative studies. Well- ORCIDs: Holland, https://orcid.org/0000-0002-4064-0630; Bloch, https:// orcid.org/0000-0003-1624-4926. resolved phylogenies and the rapid increase in sequencing Am. Nat. 2020. Vol. 196, pp. 000–000. q 2020 by The University of Chicago. the genomes, transcriptomes, and epigenomes of an in- 0003-0147/2020/19605-59421$15.00. All rights reserved. creasing number of social and solitary animals has set the DOI: 10.1086/710957 stage for a new wave of studies comparing molecular, Table 1: Common qualitative classifications of social complexity Least complex Most complex Batra 1966/1977 Eusociality: Behavioral Hypersociality: Queen (halicid bees/ division of labor and and worker castes social bees) cooperative brood differ in morphology, care between mono- physiology, and be- morphic parents and havior. Colony repro- offspring. duction by swarming. Wilson 1971 Quasisociality: Cooper- Semisociality:Repro- Eusociality: Reproduc- ative brood care in nest ductive division of la- tive division of labor of same-generation bor and cooperative and cooperative brood individuals. brood care in nest care between parents of same-generation and offspring. individuals. Michener 1974 Parasociality:Multiple Subsociality: Lone Primitive eusociality: High eusociality: Queen (social bees) same-generation foundress mother Queens found alone and worker castes foundresses cohabit a provisions and and morphologically differ in morphology, nest. Possible repro- protects brood until similar to worker physiology, and be- ductive division of emergence of adults caste. Some reproduc- havior. Colony repro- labor between or foundress death. tive and behavioral duction by swarming. foundresses. division of labor. Col- Colony size typically ony size typically in in thousands or tens to hundreds. higher. Crespi and Yanega Quasisociality: Some Semisociality:Tempo- Facultative eusociality: Obligate eusociality: 1995 temporary reproduc- rary reproductive di- The worker caste (but Both worker and tive division of labor, vision of labor with not the reproductive reproductive castes but lifetime repro- bimodal lifetime re- caste) has lost behav- have lost behavioral ductive success is productive success. ioral totipotency. totipotency. unimodal. All indi- All individuals remain viduals remain totipotent. totipotent. Toth et al. 2016 Subsociality: Foundress Facultative sociality: Primitive eusociality: Advanced eusociality: (vespid wasps; mother builds and Some foundresses in a Founding by lone Lone foundress. High also see Jandt inhabits nest alone. population may nest or small groups of queen-worker differ- and Toth 2015) Shows extended with worker females. Queen and entiation. Possible maternal care. daughters. worker castes highly colony reproduction flexible. by swarming. Boomsma and Cooperative breeding: Eusociality:Somecaste Superorganismality:All Gawne 2018 No castes. All indi- distinction. Mating individuals belong to viduals can mate. chance correlated with permanent castes age or body size. lacking totipotency, Queens maintain with no mating for dominance the worker caste. hierarchies. Richards 2019 Solitary/subsociality: Sociality: Nest with Eusociality: Behavioral Hypersociality: Colo- (social bees) Possibly maternal social interactions division of labor and nies with discrete care, but mothers raise between adult cooperative brood castes, with no mating brood alone. conspecifics, with care between mono- in the worker caste. no castes. morphic parents and offspring (after Batra). Note: Shown is a summary of criteria and traits suggested by some key studies for classifying the level of sociality. The term used for each level is highlighted in boldface. The solitary and aggregate levels are omitted. Where studies have primarily focused on certain taxa, this is shown in parentheses in the leftmost column. Quantitative Social Complexity 000 organismal, and ecological data across different levels of However, theoretical considerations and empirical stud- social complexity (e.g., Cardinal and Danforth 2011; Kap- ies suggest that social evolution (or other complex evolu- heim et al. 2015; Rehan and Toth 2015; Toth et al. 2016; Glastad tionary traits) is not constrained in such a way that a pre- et al. 2017; Wittwer et al. 2017; Shell and Rehan 2018; Sum- dictable and stepwise trajectory is consistently followed ner et al. 2018). These comparative studies can drive the across independent lineages (reviewed in Linksvayer and understanding of which traits and mechanisms influence, Johnson 2019). Instead, studies with diverse social insects or are influenced by, the evolution of social complexity point to distinct, lineage-specific trajectories that may have (Fischman et al. 2011). For example, molecular signatures been shaped by different costs or benefits of eusociality (e.g., of caste differentiation, an important component of social Fischman et al. 2011; Sumner 2014; Shell and Rehan 2018). complexity, have already begun to be identified in diverse There is evidence that key traits of “primitive eusociality” taxa (Sumner et al. 2018). However, the array of terms used have been maintained for millions of years and so likely to describe the levels of social complexity, such as “primi- represent successful evolutionary strategies (Linksvayer and tively eusocial,”“advanced eusocial,”“hypersocial,” and Johnson 2019 and references therein). Moreover, the as- “superorganism,” are not always consistently used in the sumption of a linear stepwise progression of traits is not literature or across social lineages (table 1), leading to diffi- consistent with species having different levels of social culties in comparisons across studies and in evolutionary complexity across their social traits in a multidimensional interpretation of the data (Boomsma and Gawne 2018; mosaic fashion (fig. 1)—for example, if species A is more Richards 2019 and references therein). We argue that a re- socially complex than species B in trait X but is less socially assessment of the way social complexity is defined and mea- complex than species B in trait Y. Such a phenomenon, sured is especially timely in the current genomic and post- where different lineages vary in the level of complexity genomic (“-omics”) era in which new molecular data open for different social traits, could result from selection pres- unprecedented opportunities for understanding the evolu- sures that are specific to certain ecological niches, preadap- tion of sociality. We argue that in order to meet this goal, the tions, or life-history traits (e.g., predation pressure, latitudi- field needs to move from inconsistently defined and some- nal clines, habitat selection, nesting biology, or diapause; times unclear qualitative classifications into approaches Rehan et al. 2012; Kocher et al. 2014). In this case, a more that account for multiple quantitative traits as components nuanced understanding of social evolution would be nec- of social complexity. essary, where different components of social complexity The common qualitative classifications of social insects can evolve at different rates across, or even within, distinct (table 1) are built on the presumption of a single axis of so- phylogenetic lineages. We refer to this as the “multidi- cial complexity, which is assumed to be supported by evi- mensional model.” dence for correlations between a number of social traits (Bourke 1999, 2011; Anderson and McShea 2001). For example, multiple social traits have been shown to be associ- A Test Case: Are Bumble Bees Primitively Eusocial? ated with colony size (Thomas and Elgar 2003; Holbrook et al. 2011; Kramer and Schaible 2013; Ferguson-Gow Bumble bees (tribe Bombini) form an especially useful focal et al. 2014; Amador-Vargas et al. 2015) or the degree of caste lineage because there is much research on bumble bee phys- dimorphism (e.g., Fergusson-Gow et al. 2014; Sumner et al. iology and sociobiology as well as that for other apid bees 2018). Thus, while some classifications may consider mul- to which they are often compared. Bumble bees are com- tiple traits, the assumption of strong correlation means that monly considered an intermediate stage of social complex- only a single axis of variation is considered. By using a sin- ity between the orchid bees, which are considered solitary, gle scale, these classifications implicitly assume that the im- “primitively social,” or “facultatively eusocial” (Andrade portant traits underlying social complexity always evolve et al. 2016), and the “advanced eusocial” honey bees and in synchrony with each other in a single narrow trajectory stingless bees. Indeed, Apidae are the most commonly and faithfully reproduce specific stages along a ladderlike used lineage for comparative studies of social complexity evolutionary pathway progressing from solitary to interme- (Kocher et al. 2014; Richards 2019). However, it is not diate sociality (commonly termed “primitive eusociality”) clear how well bumble bees really fitthisassumedposi- to elaborated societies (e.g., defined as “highly eusocial,” tion. Bumble bees are also a good taxon with which to “superorganismal,” or “hypersocial”; see examples above demonstrate inconsistency in the field because their level and in table 1). We herein term this approach the “linear of social complexity has been variously described as “eu- stepwise model.” This common view also asserts, intention- social” (Wilson 1971), “primitively eusocial” (Michener ally or otherwise, that extant solitary species or species with 1974), “highly social” (Kocher et al. 2014), “superorgan- simple societies faithfully represent the ancestors (i.e., lower ismal”(BoomsmaandGawne2018),or“hypersocial”(Rich- rungs on a ladder) of species with more complex societies. ards 2019), among other terms. Below, we use published 000 The American Naturalist nerdprogression Inferred

Figure 1: Models for the evolution of increasing social complexity, represented by hypothetical social insect phylogenies. Different shapes refer to different social traits (e.g., queen-worker dimorphism, across-worker differentiation, or colony size). The common ancestor in each model is assumed to have the least socially complex traits (pale colors), whereas darker colors represent an evolutionary increase in quantitative (or qualitative) complexity for speciﬁc traits (see the main text for details). For simplicity, these models assume that traits cannot become less complex. At the tip of each branch, a present-day species is represented by a letter and its complement set of traits. In the linear stepwise model, the tree can be arranged so that social traits evolve increasing complexity in tandem (e.g., increase in colony size and in worker-queen differentiation), as depicted by the arrow. Under this model, it may be tempting to conclude that species A approximates the ancestral traits of species B, C, and D. Note that lineages that branched early in evolution share many features with the ancestor of the other lineages but themselves are not ancestral or more “primitive” to the other lineages. Moreover, given that nodes in phylogenetic trees can be freely rotated, arrow-like evolutionary inferences are arbitrary. In the multidimensional model, different social traits do not necessarily evolve synchronously along a single trajectory, meaning that species B may be more socially complex than species A in several traits but not in others (e.g., as a result of differing selection pressures). In the multidimensional model, there is no way to rearrange the tree such that it is consistent with a stepwise linear increase in social complexity.

and new data to compare social traits of bumble bees with t p 1:36, n p 7, P p :23) and comparable to those of other social insect taxa (see the appendix for a description many stingless bees (Wilcoxon rank sum test, W p 55, of our methods). n p 27, P p :46), the “advanced eusocial” close relatives of bumble bees (Cardinal and Danforth 2011; Bossert et al. 2019). Notably, bumble bee queen-worker size dimorphism was significantly larger than that of attine ant Queen-Worker Variation species taken together (Wilcoxon rank sum test, W p Morphological differentiation between queens and work- 151, n p 36, P p :008); the degree of bumble bee dimor- ers has been often used as a key criterion for the highest phism was actually numerically similar to that of the most level of social complexity (table 1). In bumble bees, queens extreme attines, Atta and Acromyrmex (the leaf-cutting are typically much larger than workers, and in many spe- ants), which have been used to exemplify an extreme degree cies the size ranges of workers and queens do not overlap of social complexity (e.g., Holldobler and Wilson 1990). (Cumber 1949; Michener 1974; Goulson 2010; Prŷs-Jones Bumble bee queens and workers apparently differ in a and Corbet 2011). We quantified the size ratio of bumble few additional morphological traits (such as color patterns bee queens to workers and compared these ratios with and possibly ommatidia number and the density of anten- those of the honey bee Apis mellifera, stingless bees, and nal sensilla; Goulson 2010; Chole et al. 2019) but do not attine ants (fig. 2; see the appendix for methodological de- show the strong allometric variation typical to queens tails). These ratios were higher than those in the honey of some “highly” social species (Wilson 1971; Michener bee (although not statistically significant, probably as a re- 1974). Queens of many ants, honey bees, and stingless bees sult of the small sample size; t-test assuming equal variance, show more substantial morphological variation, reflecting Quantitative Social Complexity 000

3.0

2.5

2.0

1.5 Queen-worker size ratio

1.0

Figure 2: Queen-worker body size dimorphism in selected social insect genera. Bumble bees (red) are compared with 11 attine ants (black), 10 stingless bees (blue), and Apis mellifera (green). Each point represents queen-worker size ratio (mean queen size/mean worker size) for a single species of the genus indicated on the X-axis. Bombus data are from Cumber (1949; raw data) and from I. Medici de Mattos and G. Bloch (unpublished data). Bombus terrestris is highlighted with a black outline. Stingless bee and attine ant data are from Toth et al. (2004) and Ferguson-Gow et al. (2014), respectively. A phylogeny is shown to highlight relationships. Additional details on our methods are provided in the appendix.

selection on the same genome to produce two distinct de- ony development, even though relatively few worker off- velopmental trajectories. Allometric or other morpholog- spring seem to make it to adulthood in queenright colonies ical variability may be used for characterizing the degree (Owen and Plowright 1982; Bloch et al. 1996; Bloch and of caste differentiation (Wilson 1971; Michener 1974), but Hefetz 1999; Brown et al. 2003; Takahashi et al. 2008). How- it has not been precisely quantified and currently cannot ever, worker egg laying may be similarly (and even more) be used in our analysis. substantial in many species of the “highly social” stingless Division of labor in reproduction is a hallmark of insect bees. In some species workers lay numerous “trophic” eggs sociality, and various measures of reproductive skew have for the queen diet (Michener 1974), and in other species been used as proxies for social complexity (e.g., the propor- workers lay many male eggs (Toth et al. 2004). Thus, the tion of reproductive females, the proportion of queen- highly eusocial stingless bees and the “primitively eusocial” derived males, and worker ability to mate; e.g., Sherman bumble bees cannot be clearly separated on the basis of et al. 1995; Rubenstein et al. 2016). In bumble bees, queens the degree of worker egg laying or the proportion of repro- and workers have a similar number of ovarioles, but the ductive workers. queenhaslargerovarieswithahighermaximalnumberof Morphological differences between adult queens and mature oocytes, and she is significantly more fertile. Only workers (including in overall body size) are determined queens attract and copulate with males and are physiologi- by developmental processes during the embryo (egg), larva, cally capable of diapause (Röseler and Röseler 1986; Am- or pupa stages. In particular, early differentiation is typi- salem et al. 2015). Nevertheless, in sharp contrast to honey cally associated with more distinct queen-worker phenotypes bees, a large proportion of bumble bee workers activate (Wheeler 1986) and has been suggested as a component of their ovaries and attempt to reproduce at later stages of col- social complexity (Bourke 2011). In Bombus terrestris,the 000 The American Naturalist known critical period for caste determination occurs early Variation in body size by itself is not necessarily linked in development, a few days after larval hatching (Cnaani to social complexity because it may not relate to functional et al. 2000; Bortolotti et al. 2001). variation in behavior. However, the profound worker The paragraphs above indicate that bumble bees show body size variation in bumble bees has been repeatedly substantial queen-worker body size polymorphism, few shown to relate to the propensity of workers to perform worker-produced adults, caste-specific physiology, and different tasks (reviewed in Michener 1974; Alford 1975; early caste determination, which are comparable to those Goulson 2010; Chole et al. 2019). Specifically, many stud- found in some “highly eusocial” species. On the other hand, ies have shown that small workers disproportionately per- they do not show caste-specific structures or significant al- form in-nest tasks, such as brood care, whereas larger lometric differences, and compared with honey bees (but workers disproportionately perform foraging activities. not many stingless bees) reproductive bias is relatively small This behavioral variability is associated with a range in terms of the number of eggs laid by workers and the pro- of size-related morphological and physiological features portion of reproductive females. (summarized in Chole et al. 2019) that apparently make larger workers more efficient foragers than smaller workers (Goulson et al. 2002; Spaethe and Weidenmuller 2002). There is also evidence consistent with a reduction Between-Worker Morphological Variation in the performance of free-foraging B. terrestris colonies Highly eusocial societies typically show clear division when the largest and smallest workers are replaced with of labor between workers specializing in different tasks, middle-sized workers, at least under some conditions and in some highly social ants and termites this is asso- (Holland et al. 2020). This association with task perfor- ciated with morphological worker caste differentiation mance links body size variation to social complexity be- (Wilson 1971). At intermediate levels of social complexity cause division of labor is a key organizational principle (e.g., “primitive eusociality”), division of labor between of insect societies. Other social insect taxa also exhibit workers is assumed to be less structured and morpholog- an association between worker size variation and task spe- ical variation is expected to be absent or weak (table 1). It cialization, which can be functionally significant even is then perplexing to note that bumble bees exhibit pro- without allometric distinctions (Trible and Kronauer 2017). found variation in worker size, with up to eightfold vari- Thus, the high degree of between-worker body size varia- ability in mass in B. terrestris (Goulson 2010), tenfold var- tion coupled with its associated task specialization is not iability in B. lucorum (Cumber 1949) and B. impatiens consistent with most definitions of primitive eusociality. (Mares et al. 2005), and probably even more in “pocket- Instead, it bears some similarity to the morphological making” species such as B. pascorum and B. hortorum worker castes found in highly eusocial societies of some (Alford 1975; Pouvreau 1989; Prŷs-Jones and Corbet 2011). ants and termites, despite having a monomorphic rather This variation is mediated by juvenile hormone and social than polymorphic body size distribution, as in some ants. inhibition, among other factors (reviewed in Chole et al. It should be appreciated, however, that division of labor 2019). To investigate differences in this component of so- can also be efficient without morphological variation, as cial complexity, we compared variation in worker body seen in many other taxa. size between bumble bees and “highly eusocial” honey bees, stingless bees, and attine ants (see the appendix for details). Following previous studies, we used the coefficient Colony Size of variation as a simple way to compare all genera (including those with morphological worker castes, such as Atta Colony size (i.e., the total number of individuals, or only and Acromyrmex). We found that bumble bee colonies workers, in a colony) is broadly correlated with various have a significantly larger variation in worker body size indices of social complexity and has been commonly used compared with their close relatives, the stingless bees as a proxy for the degree of social complexity (Wilson (Wilcoxon rank sum test, W p 280, n p 47, P ! :0001; 1971; Bourke 1999; Bonner and Brainerd 2004). Social bum- fig. 3). Bumble bee size variation was also larger than that ble bees meet the upper range of Michener’s description in A. mellifera (although not quite statistically significant, stating that primitively eusocial bees typically have !20 work- probably as a result of the small sample size; t-test assum- ers and no more than a few hundred (Michener 1974). ing equal variance, W p 2:39, n p 8, P p :054) and These numbers contrast with “advanced eusocial” colonies overall was comparable to that in attine ants (Wilcoxon of honey bees and many stingless bees that can be two rank sum test, W p 125, n p 43, P p :98). In fact, bum- orders of magnitude larger (Michener 1974, 2007); ant and ble bees show higher variation than all attine ant genera termite societies can be even larger still (Wilson 1971). Nota- studied except the leafcutter ants. bly, however, other stingless bee or ant species may have Quantitative Social Complexity 000

1.5

1.0

0.5

0.0 Worker size variation (log scale) Worker

-0.5 Atta Apis Scaura Plebeia Trigona Bombus Friesella Melipona Tetragona Trigonisca Partamona Geotrigona Paratrigona Lestrimelitta Mycetarotes Acromyrmex Leurotrigona Tetragonisca Mycetosoritis Apterostigma Mycocepurus Nannotrigona Scaptotrigona Frieseomelitta Sericomyrmex Cyphomyrmex Mycetagroicus Myrmicocrypta Trachymyrmex Kalathomyrmex

fi Figure 3: Log10 coef cient of variation of worker body size in selected social insect genera. Bumble bees (red) are compared with 12 attine ants (black), 16 stingless bees (blue), and the honey bee Apis mellifera (green). Each point represents the coefficient of variation for a single species of the genus indicated on the X-axis (using the sample-size-weighted mean of study populations). Workers in each case included both in-nest and foraging workers. The Bombus terrestris data represent more than 6,000 workers from three studies (Cumber 1949; Goulson et al. 2002; Holland et al. 2020) and are highlighted with a black outline. Additional Bombus data (five more species) are from Cumber (1949) and Couvillon et al. (2010). Attine ant data and methods are from Ferguson-Gow et al. (2014). Stingless bee data are from Waddington et al. (1986) and Grüter et al. (2017). A phylogeny is shown to highlight relationships. Additional details on our methods are provided in the appendix. smaller colonies than bumble bees, indicating the colony worker morphological variation and specialization, larger size does not always separate “advanced” from “primitively” colony size than some stingless bee and ant species, lower eusocial species (e.g., Wille 1983; Holldobler and Wilson proportion of egg-laying workers than some stingless 1990). bees; fig. 4A). No matter how trait variables are rescaled or classifications repositioned, these taxa will not neatly fitintoaclassification system that assumes that all social traits evolve in synchrony and in which one taxon is more Multidimensional Patterns of Social socially complex than another in all traits. Complexity in Other Social Insects The literature suggests that a multidimensional mosaic Our analyses show that social bumble bees, compared combination of social traits is not unique to comparative with some honey bees, stingless bees, and attine ants, in studies within the Apidae. Examples are implied in An- some traits are less socially complex (generally smaller derson and McShea’s(2001)classification of social com- colony size, less extreme reproductive division of labor plexity among ants, in which they categorized some species between queens and workers) but in other traits are com- as having an overall higher level of social complexity than parable or even more socially complex (similar or larger others on the basis of such factors as greater polymorphism, queen-worker body size ratio, similar or larger between- complex division of labor, and large colony size. One such 000 The American Naturalist

Colony size

Colony longevity

Reproductive skew

Queen-worker dimorphism

Worker variation

0.00 0.25 0.50 0.75 1.00 Rescaled value B. C.

1.00

0.75

0.50

0.25

Combined social complexity index 0.00 Atta Trachymyrmex Apis Melipona Bombus Euglossa sexdens septentrionalis mellifera fasciata terrestris viridissima

Figure 4: Possible strategies for using quantitative multidimensional approaches for social complexity. We used real data in five social traits from six species of bees and ants. A, Separately comparing social traits across all six species. Colony size p number of workers at peak colony size; colony longevity p estimated lifespan of colony; reproductive skew p the proportion of males produced by the queen/ foundress in queenright colonies; queen-worker dimorphism p mean queen size/mean worker size; worker size variation p coefficient of variation in worker size. Log10 values are used for worker size variation and colony size. All values are rescaled so that the minimum and maximum across species are 0 and 1, respectively. Each symbol represents a different species, as shown at the bottom of B and C. The social complexity rank of the species differs for each measured social trait. B, C, Example of combining the traits in A into a single social complexity index, with traits weighted equally and rescaled separately for ants (B) and for bees (C). Illustrations are by Guy Bloch. Additional details on our methods are provided in the appendix. example is the genus Cataglyphis, which they classify as timated at 5,000) than many species they classify as more having a lower level of social complexity because of its au- socially complex. The inverse is true for another species, tonomous individuals and a lack of complex recruitment Acromyrmex landoli, which they classify as having a higher behavior in foraging but which has a larger colony size (es- level of social complexity because of fungus farming and Quantitative Social Complexity 000 worker polymorphism but which has a colony size of only lecular features are compared for species differing in their about 1,000 workers. Consistent with this evidence, some level of social complexity. As illustrated above, the com- studies with ants have failed to find any relationship be- monly used linear stepwise qualitative classification suffers tween colony size and worker reproduction (Hammond from several important limitations. First, there is inconsis- and Keller 2004; Helantera and Sundstrom 2007), despite tency in the way different studies label qualitatively defined both being considered important measures of social com- levels of social complexity for the same taxon (table 1). Sec- plexity. In vespine wasps, worker-queen differentiation is ond, much of this classification is very crude and limits the restricted, since queens establish nests alone and workers resolution of the analyses. For example, small Halicitinae, can lay eggs (Evans and West-Eberhard 1973), but colony Euglossini, and Xylocopinae societies composed of few in- sizes can be very large, including the production of nearly dividuals are frequently placed at the same level of social 10,000 larvae in Vespula colonies (Donovan et al. 1992). complexity as bumble bee societies numbering several hun- By contrast, the paper wasp Ropalidia ignobilis forms small dred individual workers and showing substantial caste and colonies of 100 or fewer adults and yet has female castes between-worker variability. Similarly, ant societies showing differing in size, wing length, coloration, and allometric orders of magnitude variability in colony size and consider- relationships among body parts (Wenzel 1992). Lower ter- able variation in the degree of queen-worker and worker- mites live in complex societies but still show little, if any, worker differentiation are all classified as having the same brood care, which is one of the defining characters of euso- level of sociality (e.g., “advanced eusocial”). Third, social com- ciality. This lack of brood care probably relates to their plexity is commonly treated as a single trajectory passing wood-dwelling lifestyle (Korb et al. 2012). through similar stages that bundles multiple social traits The multidimensional model of social evolution is also together, but in reality these traits are not always tightly congruent with findings of substantial variation in mo- correlated (fig. 4A). There is currently no convincing the- lecular mechanisms regulating social traits in different oretical or empirical support for the notion that social com- taxa, even within the same lineages (Sumner 2014; Shell plexity evolves along the same narrow restrictive stepwise and Rehan 2018). For example, there is evidence that var- trajectory in different social lineages (Linksvayer and John- ious bee species differ in whether queen-biased genes or son 2019). Fourth, low social complexity in one or more worker-biased genes undergo higher levels of selection traits should be seen as a successful evolutionary strategy (Harpur et al. 2017) and in the microRNAs associated (adaptation) rather than a transitional or “primitive” stage with caste determination (Collins et al. 2017), while some representing a low rung in the ladder of social evolution. caste-related genes in wasps also appear to be taxon spe- Fifth, the current approach may advocate focus on species cific (Ferreira et al. 2013). In other studies, a mixture of that neatly fit the common classifications, potentially ob- shared and lineage-specific pathways and genes are asso- scuring important natural variation among species. ciated with social regulation across taxa (e.g., Woodard For some studies, such as those specifically focusing on et al. 2011; Berens et al. 2015; Kapheim et al. 2015; Warner a “point of no return” to eusociality (e.g., Holldobler and et al. 2019), suggesting that it may be fruitful to test cor- Wilson 1990; Boomsma 2009), on the initial transition from relations of these patterns with different components of solitary to social life, or on species differing greatly in several social complexity rather than with the common qualitative social complexity components, using broad qualitative clas- stages along the social ladder. To test whether these cor- sifications may be useful. Nonetheless, we argue that for relations hold over broad taxa, such comparisons should many studies, particularly those using molecular and phys- also be phylogenetically corrected to control for pseudo- iological variables, it will be more fruitful to measure social replication of clades. complexity on a trait-by-trait basis. The logic is that molecular, developmental, and physiological processes affect specific organismal traits, such as behavior, communication A New Approach for Comparative Studies systems (e.g., pheromonal or odorant receptor repertoire), of Social Complexity body size, morphology, or reproductive potential. Many of these traits underlie key components of social complex- A Need to Move from Stepwise Qualitative Definitions ity, and it is important to know whether they are regulated to Quantitative Multidimensional Approaches by similar or different processes across different lineages of The “-omic” era presents an unprecedented opportunity social insects and how this regulation is modified through to study social evolution in molecular terms as a wealth evolution. For example, the developmental processes that of new studies using large amounts of data become avail- determine body size underlie the variability between queens able (e.g., transcriptomics, genomics, proteomics, and and workers and among workers. The number of enzymes epigenomics). Many of these studies rely on the compar- that are involved in pheromone biosynthesis and the num- ative approach in which behavioral, physiological, or mo- ber of odorant receptors may determine communication 000 The American Naturalist complexity. Likewise, the molecular processes that control colony size, termed the “size-complexity hypothesis” (Bon- ovary size and oogenesis influence fertility and the degree ner and Brainerd 2004; Bourke 2011). There is less agree- of reproductive skew. Therefore, molecular processes such ment concerning other metrics. For example, some re- as the expression of genes in certain pathways or the degree searchers may argue that size variation among workers is of epigenetic methylation are likely to show clearer associ- not a good proxy of social complexity if it is not associations, within and across taxa, with such quantitative com- ated with discrete (polymorphic) size distribution or that ponents of social complexity than with the common broadly morphological/allometric differences between queens and defined qualitative states that may stem from diverse com- workers should receive more weight than mere differences binationsof social traits. Indeed, studies that have attempted in body size or physiology. Likewise, variation in behavioral to link molecular processes to the commonly used levels of repertoire (as seen in honey bees) may be seen as a better social complexity revealed weak associations at best (e.g., proxy for social complexity than morphological variability Woodard et al. 2011; Kapheim et al. 2015; Glastad et al. (as seen in some ants, termites, and bumble bees). Deciding 2017). between these conflicting opinions requires good understanding of the social biology, natural history, and life history of the studied species. The value of different traits prob- Can Components of Social Complexity ably differs between lineages of social insects. Thus, further Be Reliably Quantified? development of our proposed approaches requires addi- Previous studies have already compiled and compared tional discussion that is based on a deep understanding commonly used quantitative traits that are components of the natural history and life history of the different social of social complexity (Michener 1974; Anderson and lineages. There is also an obvious need for more empirical McShea 2001; Gardner et al. 2007; Aviles and Harwood and theoretical studies that are beyond the scope of the 2012). These include reproductive skew, variation in body current article. size, colony size, behavioral repertoires, behavioral specialization (i.e., division of labor), the number of genes How to Use Quantitative Multidimensional Approaches? differentially expressed between queen- and worker- destined larvae, the richness of communication signals As an illustrative example of a quantitative multidimen- (e.g., pheromones, antennation, trophallaxis, “dances” sional approach, we have used data available from the etc.), the fraction of the life cycle that individuals remain literature to compare the relative levels of five commonly in their social group, and the efficacy of nest homeostasis used quantitative components of social complexity in six (e.g., Sherman et al. 1995; Gorelick and Bertram 2007; species of bees and ants (fig. 4A; see the appendix for de- Avilés and Harwood 2012; Sumner et al. 2018). However, tails). This approach reveals the uncorrelated nature of to realize the full potential of the multidimensional ap- social traits, showing that the order of relative complex- proach it is important to develop quantitative and stan- ity between species is different for each trait, which em- dardized measures for additional components of social phasizes the multidimensional pattern and highlights the complexity, such as the degree of morphological differ- need to take multiple quantitative traits into account. entiation other than body size (including allometry and To this end, we suggest two approaches for using quan- other morphological differences among workers and be- titative multidimensional data on social complexity to un- tween queens and workers). derstand the causes and consequences of social evolution. The first approach, which explicitly recognizes the multidimensional pattern of social complexity, is to separately What Are the Best Quantitative Measures focus on individual components of social complexity. In of Social Complexity? the ideal case, this involves comparing species that vary in The selection of the most appropriate indices of social one focal social trait but not in others (e.g., where colony complexity—and in some cases also the best ways to size varies but queen-worker differentiation is comparable), quantify them—is not always easy and may differ between which allows the importance of the focal trait to be assessed lineages of social insects. There is good agreement and empirically. A clear separation of social complexity compo- much literature supporting the importance of several com- nents facilitates the recognition of alternative preadapta- monly used metrics. For example, the degree of reproduc- tions or modifications that may be important for social evo- tive skew is considered in many studies to be particularly lution. For example, the annual colony life histories of some important for determining social complexity overall (Sher- bees and wasps restrict increases in colony size and may also man et al. 1995; Crespi and Yanega 1995; Hughes et al. restrict queen-worker differentiation (because the queen 2008; Boomsma 2009; Boomsma and Gawne 2018). Similar needs to care for the brood and forage effectively during col- arguments have been made for the causal importance of ony foundation), but they do not appear to constrain the Quantitative Social Complexity 000 complexity of other social traits, such as between-worker plexity (e.g., under some classifications, all of these species differentiation (as seen in bumble bees). In addition, it will except Euglossa viridissima would be considered hyper- also help in the recognition of different selection pressures social or superorganismal; table 1). Another advantage in different lineages. For example, most bumble bee species is for lineages, such as spiders, in which there are no are principally adapted to cold climates, whereas stingless accepted classifications for the overall level of social com- bees and honey bees are mostly tropical. This may account plexity (Avilés and Harwood 2012). However, studies us- for some of the differences they show in social traits, such as ing this approach should acknowledge the intrinsic lim- queen-worker variation, which may be related to diapause itations of assuming a linear stepwise model that we discuss in bumble bees but not in honey bees or stingless bees. above as well as the need to carefully consider the weighting The same logic has been used to assess the effect of abiotic of different social traits according to the biology and natu- factors (variation in temperature and precipitation) on col- ral history of the studied lineages. Our example in figure 4 ony size, queen-worker dimorphism, and worker size vari- demonstrates some of these difficulties—for example, by ation in attine ants (Fergusson-Gow et al. 2014). This ap- showing bumble bees as more similar to stingless bees and proach, for example, sets the stage for asking questions honey bees than is commonly accepted. We propose that such as whether similar processes (e.g., molecular) are asso- this approach is especially powerful for comparing species ciated with colony size in meliponine stingless bees and within the same lineage (e.g., Attini, Halictinae, Vespidae) attine ants or whether the degree of caste differentiation is by allowing for a finer resolution that can be later used to regulated by similar molecular pathways in bees and wasps. identify molecular or other processes that have been impor- This approach will be particularly powerful with molecular tant inthe evolution of social complexity.For this reason, we data, such as the level of DNA methylation or differential separately compared our basic “social complexity” metric gene expression. Rather than correlating the molecular for ants and bees as a simple form of phylogenetic control. variables with qualitatively defined stages of social complex- Using this metric to compare bees, in which all five traits ity (e.g., Kapheim et al. 2015; Glastad et al. 2017; Sumner were equally weighted, Apis mellifera, Bombus terrestris, 2018), the same data could be separately correlated to each and Melipona fasciata all showed very similar levels of com- relevant component of social complexity (e.g., reproductive plexity, with E. viridissima having a relative complexity of skew, queen-worker size variation, colony size). zero. A second approach is to combine multiple components of social complexity into a single quantitative measure of Conclusions and Future Perspectives social complexity and use this multidimensional index to identify new features and processes correlated with social The remarkable increase in the number of species for which complexity. This would be similar to the approach of pre- “-omics” data are available, coupled with improved phylog- vious comparative studies (see the introduction) but using enies and steadily growing knowledge on the social biology a quantitative and less ambiguous measure of social com- of species showing diverse forms of social lifestyles, calls for plexity. An important advantage of this approach over new methods and strategies for studies of the evolution of qualitatively defined levels of sociality is that it allows finer social complexity. We suggest that multidimensional arrays analyses and distinctions between species that are cur- of quantitative metrics more precisely describe the diverse rently bundled within the same level of sociality. A first forms of social complexity than the broad and inconsistent move in this direction was suggested by Avilés and Har- qualitative classifications that are currently used. Moving wood (2012), who combined multiple quantitative indices the field into a quantitative multidimensional approach will of social complexity into a single value to compare social enable finer-scale analyses within lineages and a better com- spiders. We used a similar approach to demonstrate an- parison of processes influencing or influenced by different other example using several species of bees and ants (see components of social complexity across lineages, recogniz- the appendix for details on our methods). As in Avilés ing that social evolution is not always strictly linear and and Harwood (2012), we assigned the same weight to each stepwise. The proposed approach is particularly appropri- of the social complexity components in figure 4A to gen- ate for linking molecular or physiological processes, pre- erate a social complexity metric (fig. 4B,4C). Such overall adaptations, selection pressures, and regulatory processes social complexity metrics could use a complement of dif- to specific phenotypic social traits, such as colony size, ferent social traits that are tailored to the study in question reproductive skew, or the degree of caste differentiation. but would be most reliable if they explain as much varia- The various quantitative measures of complexity can po- tion as possible. Our example demonstrates an important tentially be further combined into unified multidimensional advantage of this approach, which is the ability to separate indices or models for overall social complexity. Although the level of social complexity for species that are com- we believe that some previous and current comparative monly considered to have the same level of social com- studies can already significantly benefit from relating 000 The American Naturalist molecular (or other) data to individual indices of social In Cumber’s publication, wing lengths of females (n p complexity (fig. 4A), we advocate that additional theoreti- 944) present in nine nests of six species were provided cal and empirical work is needed in order to realize the full without distinguishing between workers and queens. potential of quantitative multidimensional approaches. However, all colonies were stated to contain queens and Some of the most pressing challenges are to identify the the size distribution was clearly bimodal in each colony, most important components of social complexity and find which we assumed to represent worker and queen modes. the best ways to quantify them. Developing social complex- In the pollen-storing species (Bombus lapidarius, Bombus ity indices, while potentially useful, is challenging because lucorum, Bombus terrestris, Bombus pratorum), there was it requires deciding not only which traits to include but also clear separation of worker and queen distributions. For how much weight to assign to each social trait. These de- the pocket-making species (Bombus agrorum, Bombus cisions will require deep knowledge on the biology, behav- hortorum), where queen-worker size distinction is less ior, natural history, and life history of the studied species obvious, the size measurement with the minimum amount and will likely depend on the specific lineages and ques- of records between the worker and queen modes (a clear tions in focus. The value of a quantitative multidimensional trough in all three colonies) was assumed to consist of approach will increase as more data are collected for more workers and queens equally, with all values smaller and species along gradients of social complexity. This will al- larger than this assumed to be workers and queens, re- low for better resolution in comparative studies and for spectively. Additional size data using thorax width were better understanding of the interactions between different taken from all emerging workers (n p 434) and gynes components of complexity. We hope that our contribution (n p 26) in five B. terrestris colonies kept enclosed in will encourage additional researchers to contend with these standard laboratory conditions from incipient colonies challenges. to queen death (I. Medici de Mattos and G. Bloch, unpublished data). A B. terrestris value was calculated from both Acknowledgments studies using a mean weighted by sample size. Queen worker We thank Jenny Jandt and Igor Medici de Mattos for size ratio (queen mean size/worker mean size) was calcu- sharing unpublished data used for the analyses. We also lated for comparison with stingless bee data from Toth et al. (2004) and attine ant data from Fergusson-Gow et al. thank Daniel Bolnick, Timothy Linksvayer, and two anon- ymous reviewers for detailed and helpful comments on (2014). Additionally, a value for Apis mellifera was calcu- the manuscript. Financial support was provided by grants lated using thorax size and taking a mean of values from European and African strains in DeGrandi-Hoffman et al. from the US-Israel Binational Agricultural Research and Development Fund (BARD; project IS-4418-11 and IS- (2004). To statistically compare bumble bees to stingless 5077-18) and the US-Israel Binational Science Founda- bees and (separately) to attine ants, we used species-level Wilcoxon rank sum tests (given the nonnormality of the tion (BSF) to G.B. as well as a Lady Davis Postdoctoral Fellowship to J.G.H. Finally, we thank Eamonn Mallon’s data). For comparing bumble bees and A. mellifera (single group for hosting J.G.H. during much of the development value only), a Wilcoxon rank sum test does not provide sufficient power to allow such comparisons; thus, we assumed of this article. a normal distribution and equal variances in a t-test. Statement of Authorship To quantify bumble bee among-worker size differentiation, we used the following: B. terrestris data obtained from The ideas in this piece were conceived and developed by (i) wing marginal cell length of all workers (n p 1,832) J.G.H. and G.B., data gathering and analyses were per- produced by nine freely foraging colonies over the majority formed by J.G.H., and the manuscript was written by of colony growth (Holland et al. 2020), (ii) thorax width of J.G.H. and G.B. all workers (n p 4,492) found in 28 freely foraging colonies around the peak of colony growth (Goulson et al. 2002), and Data and Code Availability (iii) wing length of workers (n p 139) found in wild colo- Data in support of this publication have been deposited nies (Cumber 1949). Bombus impatiens data were obtained in the Dryad Digital Repository (https://doi.org/10.5061 from thorax width of all workers (n p 1,133) produced /dryad.x3ffbg7g6; Holland and Bloch 2020). by 12 enclosed colonies over the later stages of colony growth (Couvillon et al. 2010; J. Jandt, personal communi- APPENDIX cation). Other Bombus spp. data were obtained from wing lengths of workers (n p 570) found in seven wild colonies Methods of five species (Cumber 1949). For the data from Cumber’s To quantify bumble bee queen-worker size differentia- (1949) study, workers were determined as described above. tion, we used data from Cumber (1949) and new data. The coefficient of variation, 100 # (worker head width Quantitative Social Complexity 000 standard deviation/worker head width mean), was calcu- worker-laid eggs in Dijkstra et al. 2005; E. viridissima p lated for comparison with attine ant data from the meta- the mean proportion of eggs laid by dominant mother, study of Fergusson Gow et al. (2014). In addition, stingless Cocom Pech 2008; M. fasciata p Toth et al. 2004); (4) col- bee data from Waddington et al. (1986) and Grüter et al. ony size p log10 number of workers at peak colony size (2017) were also used to calculate coefficients of variation (A. mellifera p approximated at 10,000 workers; A. sex- (there was no overlap in the species covered by each study). dens, T. septentrionalis p Fergusson-Gow et al. 2014; B. A value for A. mellifera was obtained from Roulston and terrestris p approximated at 200 workers; E. viridissima p Cane (2000). Although size in bumble bees was measured Cocom Pech 2008; M. fasciata p Toth et al. 2004); and as thorax width or wing marginal cell length, we assume this (5) colony longevity p best estimate of lifespan in colo- to be comparable to head width (which was used for ants nies surviving foundation (A. mellifera p mean of feral col- and stingless bees) because both of these measures are onies, Seeley 1978; A. sexdens p Keller 1998; B. terrestris p strongly correlated with head width across the full size approximated at 0.5 years; E. viridissima p longevity of range of B. terrestris workers (Pearson correlation, thorax adults, Skov and Wiley 2005; M. fasciata p Roubik 1983; width-head width, r p 0:92, R2 p 0:85; wing cell length- T. septentrionalis p approximated at 2 years, on the basis head width, r p 0:95, R2 p 0:90; n p 69; J. G. Holland, of comments in Beshers and Traniello 1996 and Weber unpublished data). To statistically compare bumble bee 1966). For plotting, traits were rescaled across species so species to stingless bees and (separately) to attine ants, that the minimum was set to 0 and the maximum was we used species-level Wilcoxon rank sum tests (given set to 1. This rescaling was performed both for all species the nonnormality of the data). As with the queen-worker together (presented in fig. 4A) and for ants and bees sep- comparison, for comparing bumble bees and A. mellifera arately (fig. 4B,4C). To produce the combined indices (single value only), a Wilcoxon rank sum test does not (fig. 4B,4C), unweighted means of the six scaled trait provide sufficient power to allow such comparisons; thus, values were then calculated. This is not an exclusive list we assumed a normal distribution and equal variances in of traits that could be quantified as components of social a t-test. complexity, and several of these traits could be quantified We produced a genus-level phylogenetic tree for graph- by alternative or complementary methods. These compar- ical comparison using relationships from the following isons are obviously limited by the amount of comparable studies: within Attini, Fergusson-Gow et al. (2014); within data available. Meliponini, Rasmussen and Cameron (2009); among Meli- ponini, Bombini, and Apini, Cardinal and Danforth (2011); Literature Cited and between Apidae and Formicidae, numerous studies Alford, D. V. 1975. Bumblebees. Davis-Poynter, London. (e.g., Johnson et al. 2013). The trees were produced, without Amador-Vargas, S., W. Gronenberg, W. T. Wcislo, and U. Mueller. differences in branch length, using TreeGraph2. 2015. Specialization and group size: brain and behavioural cor- As an example of comparing social complexity be- relates of colony size in ants lacking morphological castes. Pro- tween species, we chose six species (A. mellifera, Atta sex- ceedings of the Royal Society B 282:20142502. https://doi.org dens, B. terrestris, Euglossa viridissima, Melipona fasciata, /10.1098/rspb.2014.2502. Amsalem, E., D. A. Galbraith, J. Cnaani, P. E. A. Teal, and C. M. and Trachymyrmex septentrionalis) for which quantitative fi fi Grozinger. 2015. Conservation and modi cation of genetic and data were available in ve social traits. The traits were physiological toolkits underpinning diapause in bumble bee fi p quanti ed as follows: (1) worker size variation log10 co- queens. Molecular Ecology 24:5596–5615. https://doi.org/10.1111 efficient of variation in worker head width (A. mellifera p /mec.13410. Roulston and Cane 2000; A. sexdens, T. septentrionalis p Anderson, C., and D. W. McShea. 2001. Individual versus social Fergusson-Gow et al. 2014; B. terrestris p see above; complexity, with particular reference to ant colonies. Biological – M. fasciata p Waddington et al. 1986; E. viridissima p Reviews 76:211 237. https://doi.org/10.1017/s1464793101005656. Andrade, A. C. R., E. A. Miranda, M. A. Del Lama, and F. S. variation among females, Eltz et al. 2011); (2) queen-worker p Nascimento. 2016. Reproductive concessions between related body size dimorphism mean queen size/mean worker and unrelated members promote eusociality in bees. Scientific size (A. mellifera p DeGrandi-Hoffman et al. 2004, see Reports 6:26635. https://doi.org/10.1038/srep26635. above; A. sexdens, T. septentrionalis p Fergusson-Gow Avilés, L., and G. Harwood. 2012. A quantitative index of sociality et al. 2014; B. terrestris p see above; E. viridissima p equal and its application to group-living spiders and other social organ- size assumed since all females can mate and disperse; isms. Ethology 118:1219–1229. https://doi.org/10.1111/eth.12028. M. fasciata p Toth et al. 2004); (3) reproductive skew p Batra, S. 1966. Nests and social behavior of halictine bees of India (Hymenoptera: Halicidae). Indian Journal of Entomology 28:375– the proportion of adult males that are queen produced 393. rather than worker produced in queenright colonies (A. ———. 1977. Bees of India (Apoidea), their behavior, manage- mellifera, B. terrestris, Trachymyrmex spp. p Wenseleers ment, and a key to the genera. Oriental Insects 11:289–324. https:// and Ratneiks 2006; A. sexdens p 1 implied given no viable doi.org/10.1080/00305316.1977.10433811. 000 The American Naturalist

Berens, A. J., J. H. Hunt, and A. L. Toth. 2015. Comparative tran- Friese (Hymenoptera, Apidae, Euglossini). Insectes Sociaux scriptomics of convergent evolution: different genes but con- 55:428–433. https://doi.org/10.1007/s00040-008-1023-4. served pathways underlie caste phenotypes across lineages of Collins, D. H., I. Mohorianu, M. Beckers, V. Moulton, T. Dalmay, eusocial insects. Molecular Biology and Evolution 32:690–703. and A. F. G. Bourke. 2017. MicroRNAs associated with caste de- https://doi.org/10.1093/molbev/msu330. termination and differentiation in a primitively eusocial insect. Beshers, S. N., and J. F. A. Traniello. 1996. Polyethism and the Scientific Reports 7:45674. https://doi.org/10.1038/srep45674. adaptiveness of worker size variation in the attine ant Trachy- Couvillon, M. J., J. M. Jandt, N. Duong, and A. Dornhaus. 2010. myrmex septentrionalis. Journal of Insect Behavior 9:61–83. Ontogeny of worker body size distribution in bumble bee (Bom- https://doi.org/10.1007/bf02213724. bus impatiens) colonies. Ecological Entomology 35:424–435. Bloch, G., B. W. Borst, Z. Y. Huang, G. E. Robinson, and A. Hefetz. https://doi.org/10.1111/j.1365-2311.2010.01198.x. 1996. Effects of social conditions on juvenile hormone mediated Crespi, B. J., and D. Yanega. 1995. The definition of eusociality. reproductive development in Bombus terrestris workers. Physio- Behavioral Ecology 6:109–115. https://doi.org/10.1093/beheco logical Entomology 21:257–267. https://doi.org/10.1111/j.1365-3032 /6.1.109. .1996.tb00863.x. Cumber, R. A. 1949. The biology of humble-bees, with special ref- Bloch, G., and A. Hefetz. 1999. Regulation of reproduction by erence to the production of the worker caste. Transactions of dominant workers in bumblebee (Bombus terrestris) queenright the Royal Entomological Society of London 100:1–45. https:// colonies. Behavioral Ecology and Sociobiology 45:125–135. https:// doi.org/10.1111/j.1365-2311.1949.tb01420.x. doi.org/10.1007/s002650050546. DeGrandi-Hoffman, G., M. Chambers, J. E. Hooper, and S. S. Bonner, J. T., and E. Brainerd. 2004. Perspective: the size-complexity Schneider. 2004. Description of an intermorph between a worker rule. Evolution 58:1883–1890. https://doi.org/10.1554/04-146. and queen in African honey bees Apis mellifera scutellata (Hyme- Boomsma, J. J. 2009. Lifetime monogamy and the evolution of noptera: Apidae). Annals of the Entomological Society of Amer- eusociality. Philosophical Transactions of the Royal Society B ica 97:1299–1305. https://doi.org/10.1603/0013-8746(2004)097 364:3191–3207. https://doi.org/10.1098/rstb.2009.0101. [1299:doaiba]2.0.co;2. Boomsma, J. J., and R. Gawne. 2018. Superorganismality and caste Dijkstra, M. B., D. R. Nash, and J. J. Boomsma. 2005. Self-restraint differentiation as points of no return: how the major evolution- and sterility in workers of Acromyrmex and Atta leafcutter ants. ary transitions were lost in translation. Biological Reviews 93:28– Insectes Sociaux 52:67–76. https://doi.org/10.1007/s00040-004 54. https://doi.org/10.1111/brv.12330. -0775-8. Bortolotti, L., M. J. Duchateau, and G. Sbrenna. 2001. Effect of ju- Donovan, B., A. Howie, and N. Schroeder. 1992. Comparative venile hormone on caste determination and colony processes in characteristics of nests of Vespula germanica (F.) and Vespula the bumblebee Bombus terrestris. Entomologia Experimentalis vulgaris (L.) (Hymenoptera: Vespinae) from Christchurch City, et Applicata 101:143–158. https://doi.org/10.1046/j.1570-7458.2001 New Zealand. New Zealand Journal of Zoology 19:61–71. https:// .00899.x doi.org/10.1080/03014223.1992.10423252. Bossert, S., E. A. Murray, E. A. B. Almeida, S. G. Brady, B. B. Eltz, T., F. Fritzsch, J. R. Pech, Y. Zimmermann, S. R. Ramirez, Blaimer, and B. N. Danforth. 2019. Combining transcriptomes J. J. G. Quezada-Euan, and B. Bembe. 2011. Characterization of andultraconservedelementstoilluminate the phylogeny of Apidae. the orchid bee Euglossa viridissima (Apidae: Euglossini) and a Molecular Phylogenetics and Evolution 130:121–131. https://doi novel cryptic sibling species, by morphological, chemical, and genetic .org/10.1016/j.ympev.2018.10.012. characters. Zoological Journal of the Linnean Society 163:1064– Bourke, A. F. G. 1999. Colony size, social complexity and reproduc- 1076. https://doi.org/10.1111/j.1096-3642.2011.00740.x. tive conflict in social insects. Journal of Evolutionary Biology Evans, H. E., and M. J. West-Eberhard. 1973. The wasps. David & 12:245–257. Charles, Newton Abbot. ———. 2011. Principles of social evolution. Oxford University Ferguson-Gow, H., S. Sumner, A. F. G. Bourke, and K. E. Jones. 2014. Press, Oxford. Colony size predicts division of labour in attine ants. Proceedings Brown, M. J. F., R. Schmid-Hempel, and P. Schmid-Hempel. 2003. of the Royal Society B 281:20141411. https://doi.org/10.1098/rspb Queen-controlled sex ratios and worker reproduction in the bum- .2014.1411. ble bee Bombus hypnorum, as revealed by microsatellites. Molec- Ferreira, P. G., S. Patalano, R. Chauhan, R. Ffrench-Constant, T. ular Ecology 12:1599–1605. https://doi.org/10.1046/j.1365-294X.2003 Gabaldón, R. Guigó, and S. Sumner. 2013. Transcriptome analyses .01840.x. of primitively eusocial wasps reveal novel insights into the evolu- Cardinal, S., and B. N. Danforth. 2011. The antiquity and evolution of sociality and the origin of alternative phenotypes. Genome tionary history of social behavior in Bees. PLoS ONE 6:e21086. Biology 14:1–15. https://doi.org/10.1186/gb-2013-14-2-r20. https://doi.org/e21086 10.1371/journal.pone.0021086. Fischman, B. J., S. H. Woodard, and G. E. Robinson. 2011. Molec- Chole, H., S. H. Woodard, and G. Bloch. 2019. Body size variation ular evolutionary analyses of insect societies. Proceedings of the in bees: regulation, mechanisms, and relationship to social orga- National Academy of Sciences of the USA 108:10847–10854. nization. Current Opinion in Insect Science 35:77–87. https:// https://doi.org/10.1073/pnas.1100301108. doi.org/0.1016/j.cois.2019.07.006. Gardner, K. E., R. L. Foster, and S. O’Donnell. 2007. Experimental Cnaani, J., G. E. Robinson, G. Bloch, D. Borst, and A. Hefetz. 2000. analysis of worker division of labor in bumblebee nest thermo- The effect of queen-worker conflict on caste determination in regulation (Bombus huntii, Hymenoptera: Apidae). Behavioral the bumblebee Bombus terrestris. Behavioral Ecology and Socio- Ecology and Sociobiology 61:783–792. https://doi.org/10.1007 biology 47:346–352. https://doi.org/10.1007/s002650050675. /s00265-006-0309-7. Cocom Pech, M. E., W. D. May-Itza, L. A. M. Medina, and J. J. G. Glastad, K. M., S. V. Arsenault, K. L. Vertacnik, S. M. Geib, S. Kay, Quezada-Euan. 2008. Sociality in Euglossa (Euglossa) viridissima B. N. Danforth, S. M. Rehan, C. R. Linnen, S. D. Kocher, and Quantitative Social Complexity 000

B. G. Hunt. 2017. Variation in DNA methylation is not consis- Kapheim, K. M., H. Pan, C. Li, S. L. Salzberg, D. Puiu, T. Magoc, tently reflected by sociality in Hymenoptera. Genome Biology H. M. Robertson, M. E. Hudson, et al. 2015. Genomic signatures and Evolution 9:1687–1698. https://doi.org/10.1093/gbe/evx128. of evolutionary transitions from solitary to group living. Science Gorelick, R., and S. M. Bertram. 2007. Quantifying division of la- 348:1139–1143. https://doi.org/10.1126/science.aaa4788. bor: borrowing tools from sociology, sociobiology, information Keller, L. 1998. Lifespan and colony characteristics in ants and theory, landscape ecology, and biogeography. Insectes Sociaux termites. Insectes Sociaux 45:235–246. https://doi.org/10.1007 54:105–112. https://doi.org/10.1007/s00040-007-0923-z. /s000400050084. Goulson, D. 2010. Bumblebees: behaviour, ecology and evolution. ———. 1999. Levels of selection in evolution. Princeton Univer- Oxford University Press, Oxford. sity Press, Princeton, NJ. Goulson, D., J. Peat, J. C. Stout, J. Tucker, B. Darvill, L. C. Kocher, S. D., L. Pellissier, C. Veller, J. Purcell, M. A. Nowak, M. Derwent, and W. O. H. Hughes. 2002. Can alloethism in Chapuisat, and N. E. Pierce. 2014. Transitions in social complex- workers of the bumblebee, Bombus terrestris,beexplainedin ity along elevational gradients reveal a combined impact of season terms of foraging efficiency? Animal Behaviour 64:123–130. length and development time on social evolution. Proceedings https://doi.org/10.1006/anbe.2002.3041. of the Royal Society B 281:20140627. https://doi.org/10.1098 Grüter, C., F. H. I. D. Segers, C. Menezes, A. Vollet-Neto, T. Fal- /rspb.2014.0627. cón, L. von Zuben, M. M. G. Bitondi, F. S. Nascimento, and Korb, J., M. Buschmann, S. Schafberg, J. Liebig, and A. G. Bag- E. A. B. Almeida. 2017. Repeated evolution of soldier sub-castes neres. 2012. Brood care and social evolution in termites. Pro- suggests parasitism drives social complexity in stingless bees. ceedings of the Royal Society B 279:2662–2671. https://doi.org Nature Communications 8:4. https://doi.org/10.1038/s41467-016 /10.1098/rspb.2011.2639. -0012-y. Kramer, B. H., and R. Schaible. 2013. Colony size explains the Hamilton, W. D. 1964. Genetical evolution of social behaviour. lifespan differences between queens and workers in eusocial Hy- I, II. Journal of Theoretical Biology 7:1–16, 17–52. https://doi.org menoptera. Biological Journal of the Linnean Society 109:710– /10.1016/0022-5193(64)90039-6. 724. https://doi.org/10.1111/bij.12072. Hammond, R. L., and L. Keller. 2004. Conflict over male parentage Linksvayer, T. A., and B. R. Johnson. 2019. Re-thinking the social in social insects. PLoS Biology 2:e248. https://doi.org/10.1371 ladder approach for elucidating the evolution and molecular ba- /journal.pbio.0020248. sis of insect societies. Current Opinion in Insect Science 34:123– Harpur, B. A., A. Dey, J. R. Albert, N. Patel, H. M. Hines, M. 129. https://doi.org/10.1016/j.cois.2019.07.003. Hasselmann, L. Packer, and A. Zayed. 2017. Queens and workers Mares, S., L. Ash, and W. Gronenberg. 2005. Brain allometry in contribute differently to adaptive evolution in bumble bees and bumblebee and honey bee workers. Brain, Behavior and Evolu- honey bees. Genome Biology and Evolution 9:2395–2402. https:// tion 66:50–61. https://doi.org/10.1159/000085047. doi.org/10.1093/gbe/evx182. Maynard Smith, J., and E. Szathmáry. 1995. The major transitions Helantera, H., and L. Sundstrom. 2007. Worker policing and nest in evolution. Oxford University Press, Oxford. mate recognition in the ant Formica fusca. Behavioral Ecology McShea, D. W. 2000. Functional complexity in organisms: parts and Sociobiology 61:1143–1149. https://doi.org/10.1007/s00265-006 as proxies. Biology and Philosophy 15:641–668. https://doi.org -0327-5. /10.1023/A:1006695908715. Holbrook, C. T., P. M. Barden, and J. H. Fewell. 2011. Division of Michener, C. 1974. The social behavior of the bees. Belknap, labor increases with colony size in the harvester ant Pogono- Cambridge, MA. myrmex californicus. Behavioral Ecology 22:960–966. https:// ———. 2007. The bees of the world. 2nd ed. Johns Hopkins Uni- doi.org/10.1093/beheco/arr075. versity Press, Baltimore. Holland, J. G., and G. Bloch. 2020. Data from: The complexity of Owen, R. E., and R. C. Plowright. 1982. Worker-queen conflict social complexity: a quantitative multidimensional approach for and male parentage in bumble bees. Behavioral Ecology and So- studies of social organization. American Naturalist, Dryad Dig- ciobiology 11:91–99. https://doi.org/10.1007/BF00300097. ital Repository, https://doi.org/10.5061/dryad.x3ffbg7g6. Pouvreau, A. 1989. Contribution à l’étude du polyéthisme chez les Holland, J. G., S. Nakayama, M. Porfiri, O. Nov, and G. Bloch. bourdons, Bombus Latr. (Hymenoptera, Apidae). Apidologie 20:229– 2020. Is diversity in worker body size important for the perfor- 244. mance of bumble bee colonies? bioRxiv, https://doi.org/10.1101 Prŷs-Jones, O. E., and S. A. Corbet. 2011. Bumblebees. 3rd ed. Pe- /2020.05.06.079525. lagic, Exeter. Holldobler, B., and E. O. Wilson. 1990. The ants. Harvard Univer- Rasmussen, C., and S. Cameron. 2009. Global stingless bee phy- sity Press, Cambridge, MA. logeny supports ancient divergence, vicariance, and long dis- Hughes, W. O. H., B. P. Oldroyd, M. Beekman, and F. L. W. tance dispersal. Biological Journal of the Linnean Society 99:206– Ratnieks. 2008. Ancestral monogamy shows kin selection is key 232. to the evolution of eusociality. Science 230:1213–1216. https:// Rehan, S. M., R. Leys, and M. P. Schwarz. 2012. A mid-Cretaceous doi.org/10.1126/science.1156108. origin of sociality in xylocopine bees with only two origins of Jandt, J. M., and A. Toth. 2015. Chapter three—physiological and true worker castes indicates severe barriers to eusociality. PLoS genomic mechanisms of social organization in wasps (family: ONE 7:e34690. https://doi.org/10.1371/journal.pone.0034690. Vespidae). Advances in Insect Physiology 48:95–130. https:// Rehan, S. M., and A. L. Toth. 2015. Climbing the social ladder: the doi.org/10.1016/bs.aiip.2015.01.003. molecular evolution of sociality. Trends in Ecology and Evolu- Johnson, B., M. L. Boroweic, J. C. Chiu, E. K. Lee, J. Atallah, and P. S. tion 30:426–433. https://doi.org/10.1016/j.tree.2015.05.004. Ward. 2013. Phylogenomics resolves evolutionary relationships Richards, M. H. 2019. Social trait definitions influence evolution- among ants, bees, and wasps. Current Biology 23:2058–2062. ary inferences: a phylogenetic approach to improving social 000 The American Naturalist

terminology for bees. Current Opinion in Insect Science 34:97– Toth, E., D. C. Queller, A. Dollin, and J. E. Strassmann, 2004. Con- 104. https://doi.org/10.1016/j.cois.2019.04.006. ﬂict over male parentage in stingless bees. Insectes Sociaux Röseler, P. F., and I. Röseler. 1986. Caste speciﬁc differences in fat- 51:1–11. https://doi.org/10.1007/s00040-003-0707-z. body glycogen-metabolism of the bumblebee, Bombus-terrestris. Trible, W., and D. J. Kronauer. 2017. Caste development and evo- Insect Biochemistry 16:501–508. https://doi.org/10.1016/0020 lution in ants: it’s all about size. Journal of Experimental Biology -1790(86)90027-2. 220:53–62. https://doi.org/10.1242/jeb.145292. Roubik, D. W. 1983. Nest and colony characteristics of stingless Valentine, J. W., A. G. Collins, and C. P. Meyer. 1994. Morpholog- bees from Panamá (Hymenoptera: Apidae). Journal of the Kan- ical complexity increase in metazoans. Paleobiology 20:131–142. sas Entomological Society 56:327–355. https://doi.org/10.1017/S0094837300012641. Roulston, T. A. H., and J. H. Cane. 2000. The effect of diet breadth Waddington, K. D., L. H. Herbst, and D. W. Roubik. 1986. Re- and nesting ecology on body size variation in bees (Apiformes). lationship between recruitment systems of stingless bees and Journal of the Kansas Entomological Society 73:129–142. within-nest worker size variation. Journal of the Kansas Ento- Rubenstein, D. R., C. A. Botero, and E. A. Lacey. 2016. Discrete mological Society 59:95–102. but variable structure of animal societies leads to the false Warner, M. R., A. S. Mikheyev, and T. A. Linksvayer. 2019. perception of a social continuum. Royal Society Open Science Transcriptomic basis and evolution of the ant nurse-larval so- 3:160147. https://doi.org/10.1098/rsos.160147. cial interactome. PLoS Genetics 15:20. https://doi.org/10.1371 Seeley, T. D. 1978. Life history strategy of the honey bee, Apis mellifera. /journal.pgen.1008156. Oecologia 32:109–118. https://doi.org/10.1007/bf00344695. Weber, N. A. 1966. Fungus-growing ants. Science 153:587–604. Shell, W. A., and S. M. Rehan. 2018. Behavioral and genetic mech- https://doi.org/10.1126/science.153.3736.587. anisms of social evolution: insights from incipiently and facul- Wenseleers, T., and F. L. W. Ratnieks. 2006. Comparative analysis tatively social bees. Apidologie 49:13–30. https://doi.org/10.1007 of worker reproduction and policing in eusocial hymenoptera /s13592-017-0527-1. supports relatedness theory. American Naturalist 168:E163– Sherman, P. W., E. A. Lacey, H. K. Reeve, and L. Keller. 1995. Fo- E179. https://doi.org/10.1086/508619. rum: the eusociality continuum. Behavioral Ecology 6:102–108. Wenzel, J. W. 1992. Extreme queen-worker dimorphism in Ropalidia https://doi.org/10.1093/beheco/6.1.102. ignobilis, a small-colony wasp (Hymenoptera: Vespidae). Insectes Skov, C., and J. Wiley. 2005. Establishment of the neotropical orchid Sociaux 39:31–43. https://doi.org/10.1007/bf01240529. bee Euglossa viridissima (Hymenoptera: Apidae) in Florida. Flor- West, S. A., R. M. Fisher, A. Gardner, and E. T. Kiers. 2015. Major ida Entomologist 88:225–227. https://doi.org/10.1653/0015-4040 evolutionary transitions in individuality. Proceedings of the (2005)088[0225:eotnob]2.0.co;2. National Academy of Sciences of the USA 112:10112–10119. Spaethe, J., and A. Weidenmuller, 2002. Size variation and forag- https://doi.org/10.1073/pnas.1421402112. ing rate in bumblebees (Bombus terrestris). Insectes Sociaux Wheeler, D. E. 1986. Developmental and physiological deter- 49:142–146. https://doi.org/10.1007/s00040-002-8293-z. minants of caste in social hymenoptera: evolutionary implica- Sumner, S. 2014. The importance of genomic novelty in social evo- tions. American Naturalist 128:13–34. https://doi.org/10.1086 lution. Molecular Ecology 23:26–28. https://doi.org/10.1111/mec /284536. .12580. Wille, A. 1983. Biology of the stingless bees. Annual Review of En- Sumner, S., E. Bell, and D. Taylor. 2018. A molecular concept of caste tomology 28:41–64. in insect societies. Current Opinion in Insect Science 25:42–50. Wilson, E. O. 1971. The insect societies. Belknap, Cambridge, MA. https://doi.org/10.1016/j.cois.2017.11.010. Wittwer, B., A. Hefetz, T. Simon, L. E. K. Murphy, M. A. Elgar, Szathmary, E. 2015. Toward major evolutionary transitions theory 2.0. N. E. Pierce, and S. D. Kocher. 2017. Solitary bees reduce invest- Proceedings of the National Academy of Sciences of the USA ment in communication compared with their social relatives. 112:10104–10111. https://doi.org/10.1073/pnas.1421398112. Proceedings of the National Academy of Sciences of the USA Takahashi, J., M. Itoh, I. Shimizu, and M. Ono. 2008. Male parentage 114:6569–6574. https://doi.org/10.1073/pnas.1620780114. and queen mating frequency in the bumblebee Bombus ignitus (Hy- Woodard, S. H., B. J. Fischman, A. Venkat, M. E. Hudson, K. menoptera: bombinae). Ecological Research 23:937–942. https:// Varala, S. A. Cameron, A. G. Clark, and G. E. Robinson. 2011. doi.org/10.1007/s11284-007-0456-y. Genes involved in convergent evolution of eusociality in bees. Thomas, M. L., and M. A. Elgar. 2003. Colony size affects division of Proceedings of the National Academy of Sciences of the USA labour in the ponerine ant Rhytidoponera metallica.Naturwis- 108:7472–7477. https://doi.org/10.1073/pnas.1103457108. senschaften 90:88–92. https://doi.org/10.1007/s00114-002-0396-x. Toth, A. L., S. Sumner, and R. L. Jeanne. 2016. Patterns of longevity across a sociality gradient in vespid wasps. Current Opinion in Associate Editor: Timothy A. Linksvayer Insect Science 16:28–35. https://doi.org/10.1016/j.cois.2016.05.006. Editor: Daniel I. Bolnick