Cooperation is central to transitions in individuality 1234. Full individuality is achieved when components cooperate and relinquish their autonomy to the larger whole. Depending on the type of transition, this may necessitate the division of labor in growth, reproduction, development, feeding, movement, and protection against external aggression and internal conflict 56. In the evolution of multicellularity, the chain of events from autonomous individuals at one level to the incorporation of these individuals into a more complex entity remains unclear 5. However, some of the putative forces are likely to be general, since multicellularity has arisen many different times in evolutionary history 78. Moreover, that many groupings do not show sophisticated specialization and are characterized by substantial levels of internal conflict 910, suggests that incomplete multicellularity may be a frequent outcome. What mechanisms are essential to generate individuality? We believe that a general theory needs to explain both full and incomplete transitions towards multicellular individuals.

Previous work highlights group and kin selection 51011, organism size 1213, and the reorganization of fitness and specialization tradeoffs 14 as playing roles in the evolution of multicellularity. A feature common to these mechanisms is the establishment and maintenance of cooperative behaviors amongst subunits through, for example, conflict mediation (e.g. 1516). Based on a recent literature review, Grosberg and Strathmann 8 argued that for cooperation to emerge and favor the specialization of subunits, groups of cells need to reduce genetic conflicts arising in cell lineages 10. They conclude that several mechanisms can limit such conflicts, perhaps the most important being development from a single cell (e.g., 516).

A key type of subunit specialization in multicellular organisms is the separation of germ and soma 15101718. Separating germ and somatic functions amongst individual cells or cell lineages requires that each sacrifice autonomy. Theory predicts that such specialization is promoted by non-mutually exclusive mechanisms such as cooperation and relatedness amongst cell lineages 10, cheater control 11920 and adaptive responses to tradeoffs between survival and reproductive functions, i.e. a covariance effect augmenting the fitness of the group over the average fitness of its members 14. It is not known whether the alignment of fitness interests in emerging soma and germ lines tends to occur before, during or after other types of specialization characteristic of multicellular organisms 12.

A pervasive feature in a diverse array of social systems is that individuals not contributing to the common good either act as dispersers, or are either rewarded for, or coerced into, cooperating. Examples range from bacteria (e.g. Pseudomonas fluorescens) through protozoa (e.g. Volvox carteri) to metazoans, like eusocial insects and mammals (see Additional file 1). For example, in naturally occurring Dictyostelium slime molds prespores secrete a chlorinated hexaphenone (DIF-1) inhibiting redifferentiation of prestalk cells into prespores, which would transpose them from "cooperative" stalk building to "cheating" spore production (i.e. a transition into the dispersing and perennial germ line; 2122). Cheating is further curtailed by pleiotropic effects of a gene required to permit receipt of this signal, which affects also the probability of spore formation 23. In tunicates such as Botryllus schlosseri, natural chimeras consisting of genetically nonhomogenous organisms often show reproducible germ cell parasitism that is sexually inherited, with "parasitic forms" being expressed only in the germ line, i.e. in the dispersing entities 24. In the cooperatively breeding cichlid fish Neolamprologus pulcher, brood care helpers of both sexes are forced to pay rent for being tolerated in a safe territory 2526. To avoid being punished they preemptively appease dominants by cooperative and submissive behavior 27. Typically, in these cichlids and in cooparatively breeding meerkats Suricata suricatta, subordinates preparing for dispersal reduce helping 2829, which might be explained by reduced costs of potential punishment by eviction 3031. In eusocial mole rats (Heterocephalus glaber and Cryptomys damarensis) non-reproductive helpers and hardly helping dispersers coexist 323334. Policing of subordinates by dominant breeders may simultaneously maintain social order and stimulate cooperative behaviors 3536. This distinction of roles between individuals is particularly obvious in the separation between soma and germ that has apparently evolved many times independently 7. Nevertheless, there are examples where cooperative behaviors are associated with enhanced group dispersal (see Additional file 1). For example, in the soil-dwelling social bacterium Myxococcus xanthus, individualistic cell movement ('A-motility') promotes swarming on hard surfaces, whereas swarming on soft surfaces is a group function driven primarily by individually costly S-motility 37.

These empirical patterns merit explanation, and we take a first step by employing optimization techniques to evaluate the conditions leading to associations between dispersal and social strategy. Sociality in our models takes the form of cooperation in the production of a public good. Previous study of public goods has shown how cheating, if left unchecked, potentially leads to a "tragedy of the commons" 3839, whereby individual selection tends to favor exploitation of the public good at some concurrent or future detriment of the group. Several non-mutually exclusive mechanisms may promote cooperation and group persistence, including kin selection (e.g., 404142), rewards and sanctions (e.g., 4344), spatial and network structure (e.g., 454647), and signals involving kin or non-kin (e.g., 484950). Recent reviews and perspectives can be found in Crespi 51, Sachs and colleagues 52, Lehmann and Keller 53, and West and coworkers 54.

We develop a model based on kin selection that incorporates dispersal specialization, as suggested by the case studies in Table S1 (see Additional file 1). We employ the terms "soma" and "germ" to represent the functions of within-group growth and dispersal leading to the founding of new groups, respectively. Our use of the terms "cooperators" and "cheaters" refers to social behaviors within the commons (e.g., soma), and this should be distinguished from the frequent usage of "cheaters" as cooperative somatic lineages trying to gain access to germ line (e.g., 1581022). Specifically, cooperators contribute to the public good within a distinct group at an individual cost, and cheaters exploit the public good. Cooperators and/or cheaters may be selected to either remain in a group, or to disperse (potentially founding new groups). Our theory proposes a mechanism leading to high overall cooperation, based on dispersal specialization. In addition to increasing our understanding of cooperative and dispersal behaviors, it could apply to the evolution of multicellularity in a range of contexts, including physiologically integrated organisms 5556, organisms with both solitary and integrated life-styles (e.g., 57), and complex societies 58.

We formalize our verbal arguments given above by developing and analyzing a model of coevolution between exploitation of the commons and dispersal. From the outset, we stress that our model is a highly simplified representation of this process, and not aimed to make quantitative predictions for any given system. Rather, our goal is to identify the qualitatively important drivers in the coevolution of individual strategies and the evolution of multicellularity.

In our model the focal units of selection are individuals themselves, rather than the higher-level unit. A transition to multicellularity is favored when the interests of the individual and the higher-level (the group) are aligned 5815. Previous models investigating the transition to multicellularity invoke a framework where the group is the focal unit of selection (see, for example 15). However, focusing on the higher-level as the focal unit does not easily allow the investigation of optimization at the lower level 59, and the individually-selected conditions leading to a major transition 60. Grosberg and Strathmann 8 have argued that many of the requirements for transitions to multicellularity exist in unicellular organisms (for social groups, see 61). Once a transition is in progress, and the "group" begins to behave as an individual entity, one can begin to treat this unit as an evolving individual in itself.

We analyze an optimization model that takes into account the effect of both the phenotype of the focal individual and the average phenotype of the group in which it lives, on the fitness of the focal individual (see Table 1 for descriptions of parameters and variables). The approach is based on the direct fitness method 4262 in that, by considering the effects of both individual and average group phenotypes on the fitness of a focal individual, we can apply the Price Equation to partition these effects as weighted by the relatedness of the focal individual to other members of the group 42. We can then assess the relative impacts of (1) costs and benefits of individual behaviors and (2) kin structure, on associations between exploitative strategy within a group, and dispersal to found new groups. Nevertheless, our model oversimplifies the complexity of social behavior and dispersal decisions (for review, see 63), and should thus be viewed as a preliminary attempt to identify patterns.

Our model makes several assumptions. First, we do not explicitly consider dynamics, such as group founding, group numbers, individual emigration and immigration, and competition for limiting resources within or between groups. Rather, we assume negligible variation in inter-group competition. Second, our model does not explicitly incorporate genetic polymorphisms, meaning that the heritable traits are probabilities to adopt alternatives of each strategy (disperse or stay; cooperate or cheat) depending on environmental and/or social conditions 11032646566. Third, there is a simple direct tradeoff between an individual's viability (growth, survival and reproduction) within the group and its ability to disperse and found new groups. This is based on the well established life-history trade-off between reproduction and dispersal (see 67), probably best studied in insects (on the physiological scale e.g. 686970; on the ecological scale e.g. 7172). Whereas growth and reproduction within the group impacts the production and consumption of the public good, the tendency to disperse reduces these impacts because of the limited presence of dispersers in the source group.

We consider two scenarios. In the first (Model 1) only dispersal in cooperators (y) and cheaters (z) evolves, but not cooperation (n). This situation would be obtained if mechanisms not explicitly included in the model (e.g., policing, 44) controlled the level of cooperation, or if the frequencies of cooperative behaviors were either not subject to evolution, or labile to it over much longer time scales than dispersal. More generally however, empirical study suggests that cooperative behaviors are subject to selection 787980 and we consider the case (Model 2) in which dispersal and the frequency of cooperators (n) and cheaters (1-n) co-evolve.

In addition to optimal levels of dispersal (Model 1), and of cooperation and dispersal (Model 2), we examine the effects of model parameters on dispersal specialization σ = y*/(y*+z*), and for Model 2 only, overall cooperation Φ = n*(1-y*)+(1-n*)z* (i.e., the sum of cooperators not dispersing and of cheaters dispersing). Note that when σ = 0 (or σ = 1), although all cooperators (cheats for σ = 1) are sedentary it is not necessarily true that all cheats (cooperators for σ = 1) disperse.

Our results are in broad agreement with the tenets of kin selection theory for explaining dispersal 458182 and the maintenance of cooperative behaviors 838485. Specifically, we found that dispersal specialization leading to high levels of overall cooperation (Φ) is promoted by sufficient benefit to cost ratios (P : s) of cooperation and by kin selection (low k). The one apparent discrepancy to previous theory is, whereas higher benefit (P) to cost (s) ratios promote cooperation, higher kin selection (low k) was sometimes observed to reduce the relative frequency of cooperators (n*) (cf. Figs. 4c, d). This can be explained if we consider cheaters dispersing from the group as a type of cooperative behavior. Dispersing cheaters are effectively 'cooperative' because of the incurred individual cost of dispersal (e), and the benefits to the group in having less negative impact on the commons (Q) (cf. Figs 4c, d with Figs. 6c, d). Moreover, we found that partial or total specialization of otherwise somatic cheats as dispersing germ line occurred without the need for costly modifiers 86 or the repression of cheaters [8788, but see 22], suggesting that the mechanism identified here is applicable to a wider range of organisms where these mechanisms do not sufficiently reduce somatic cheating, or cannot evolve. Conversely, control mechanisms such as rewarding and punishment, which might be operating in many systems (see examples in Additional file 1), do not preclude the functionality of the mechanism demonstrated in this study (cf. Model 1).

The examples presented in Table S1 (see Additional file 1) and our theoretical findings suggest a common conceptual and mechanistic foundation for the evolution of cooperation and individual functional specialization within groups (e.g., multicellularity). Most of the empirical examples share the feature that cooperators are less dispersive than more competitive individuals. For instance, low dispersal coincides with physical binding in bacteria that generate biofilms as a public good by polymer production 7989 (but see ref. 90 for an alternative interpretation), with alloparental care of offspring in cooperative breeding 91, or with complete genetic altruism in certain eusocial insects 92. It is worth noting that a consistent differentiation of roles regarding sedentariness and dispersal in relation to cooperation and cheating may be much more common in nature than currently believed (e.g. 93). Because there is no prior formal theory predicting such a relationship, empirical research on this issue is rare and suitable data are therefore scant. We stress that our theory does not elucidate the precise evolutionary pathway leading to complete multicellularity 1677, but rather assesses the forces promoting or forestalling different levels of specialization of cooperators and cheaters as functional germ line and soma. As such, the observations of biased dispersal in Table S1 (see Additional file 1) have alternative explanations, including forced eviction 94 and individual-based habitat selection 95. Experimental (e.g., 7980969798), phylogenetic (e.g., 6), and theoretical (e.g., 1322 and see discussion below) approaches are fruitful avenues to explore alternative explanations and pathways.

Transitions in individuality and social complexity are generally thought to require some form of reduction in genetic variance during the reproductive process 2077. Genetic heterogeneity can emerge from many sources 99, and the recursive equation 6 in our framework greatly simplifies these, only explicitly including the effects of dispersal. Our results confirm the importance of relatedness in achieving multicellularity, but also show that the highest levels of relatedness did not necessarily yield full specialization of cooperators or cheaters as dispersers, and that complete specialization could occur at relatedness levels as low as 0.7 (Fig. 4d). As such, our findings could extend to some systems in which groups are formed by the initial aggregation of non-kin 10748799. Further study is needed to explore this prediction in detail, since our model did not explicitly account for different lineages, and as such we do not know how spatial heterogeneities in relatedness might influence our results 100.

Our findings have precedent, both in the study of symbiotic associations, and investigations of cooperation within species. With regard to host-parasite and symbiotic interactions, previous research has considered how parasite virulence (which is analogous to cheaters exploiting cooperative groups) may evolve spatially (e.g., 101; for reviews see 102103). In the case of horizontal transmission in parasites, which is analogous to the level of dispersal in our model (see also 73), theory generally predicts that increased horizontal transmission (z in our model) associates with higher parasite virulence (Q (1-n) in our model) 103. Despite allowing for relatedness between potential cooperators and cheats we have a comparable finding, whereby an increasing tragedy of the commons pushes cheating individuals to disperse; this is both because of increased individual fitness opportunities through dispersal (z) and increased inclusive fitness through lowered group effects for those related individuals that do not disperse (Q(1-n) (1-z)).

In a model investigating cooperation in spatially viscous environments, van Baalen and Rand 45 suggested that non-altruists should disperse more readily than altruists and hypothesized that this could be viewed as a transition towards multicellularity. Koella 104 studied the independent dispersal of altruists and of cheaters in a spatially explicit setting and found that a polymorphism could arise in which altruists dispersed and interacted locally, whereas cheaters evolved longer dispersal distances and exploited altruistic clusters. Hamilton and Taborsky 95 showed that when the propensities to cooperate by generalized reciprocity and to disperse evolve independently, under a wide range of conditions either cooperation or defection is associated with dispersal, depending on the probability of finding new groups and on the costs of being alone. Over most of the range of mobility costs examined, cooperation was negatively correlated with mobility, while defection was not. Ultimately, this leads to assortment between altruists and defectors in the population (see also 105), which secondarily can generate group selection effects 106107. Hamilton and Taborsky 95 did not check for linkage effects, however. In another study of the joint evolution of altruism and mobility, Le Galliard and coworkers 108 found that more altruism enhancing local aggregation can select for increased mobility. The synergistic selective interaction between altruism and mobility may cause dispersal to be considerably higher than that predicted in a purely selfish population, if altruism costs accelerate slowly and mobility costs are moderate. However, their model did not reveal a polymorphism to occur between selfish-mobile and altruistic-sessile phenotypes as found so often in nature, from microbes and unicellular algae to mammals (e.g. 1832109; Additional file 1). Queller 10 argued that the resolution of within-organism conflicts could occur if an altruism allele is expressed conditional on the environment, the altruistic act being an individual removing itself from the germ line in order to perform an enhanced somatic activity. Rainey 110 verbally proposed an idea similar in some respects to these studies, in which group selection acts to promote the functional separation of germ and soma in bacterial biofilms through the dispersal of cheats (see also 79). Finally, Michod 14 showed how the specialization of lower level units into germ and soma could be associated with the transfer of fitness from lower units to the new higher individual. A critical feature of his model is the tradeoff between the viability and fecundity of lower level units, which, for convex relationships, creates disruptive selection for cooperative germ and soma. Our study, whilst generating congruent results, is to our knowledge the first to demonstrate that the evolution of lower level units based on their effects on the commons can yield dispersal specialization, one of the precursors for selection at the group level and the evolution of full multicellularity.

Our results suggest that the establishment of trait linkage between dispersal and the propensity of within-group cheating may be a general phenomenon promoting complex social organization and multicellularity. Importantly, we cautiously suggest this should be operative regardless of whether groups ever achieve higher levels of individuality, because selection on individual components will always tend to increase exploitation, and stronger group structure will tend to increase overall cooperation through kin selected benefits 4284. Partial or full reduction in the negative effects of cheaters on the commons through their specialization as dispersers offers partial solutions to two problems: the evolution of cooperation in social groups and the origin of the specialization of germ and soma in multicellular organisms. Our model is, nevertheless, a highly simplified caricature of real systems and future theoretical and empirical study is needed to explore its robustness.

MEH conceived the study, developed and analyzed the model and wrote the manuscript. DJR participated in the design of the study, developed the model and participated in writing the manuscript. MT participated in the design of the study, constructed Table S1, and participated in writing the manuscript. All authors read and approved the final manuscript.