Two major themes have characterized multilevel selection theory from Darwin to the present (Wilson Reference Wilson1997). One theme is centered on the concept of a social group as like an organism, whose members work together to benefit the common good. Another theme is centered on the concept of altruism, which is selectively disadvantageous within groups and therefore requires between-group selection to evolve. These two themes sound similar and compatible, but in fact there are important differences. For example, in models of altruism, the most adaptive group is one that contains 100% altruists. In contrast, the parts of an organism are differentiated into organs and cell types. As another example, highly altruistic traits require strong group selection to evolve, to counteract strong selection within groups. But dividing labor, organ-like, need not be self-sacrificial at all.
I applaud Smaldino for advancing the “group as organism” theme in his target article. In this commentary, I will argue that his points apply to all multilevel evolutionary processes, not just human cultural evolution.
It is helpful to begin with the familiar case of individuals as the unit of selection in genetic evolution. Momentarily forget about group selection or genomic conflict and imagine a species in which natural selection acts entirely at the individual level. Organisms are selected on the basis of their phenotypic properties, which are mechanistically caused by genes and their interactions. Occasionally a single phenotypic trait can be attributed to a single gene, but most often the phenotype-genotype relationship is more complex, with single phenotypic traits caused by many genes and their interactions. Moreover, when the same phenotypic trait is selected in different populations, the response to selection often involves different genes. As an example, for adults, the ability to digest lactose has evolved at least twice in human populations, but different mutations were selected in each case (Holden & Mace Reference Holden and Mace2009). Given the chance nature of mutations, it is unlikely that the same one would arise in different populations subjected to the same selection pressure.
In short, when individuals are the units of selection, they also become the unit of functional analysis. The elements that comprise the unit are studied as proximate mechanisms. This is often simpler in principle than in practice, because the proximate mechanisms can be so complex, involving gene–gene and gene–environment interactions operating throughout development. In fact, some authors would fault my description as too gene-centric, when genes are merely parts of self-replicating developmental systems (e.g., Oyama et al. Reference Oyama, Griffiths and Gray2003).
The complexity and distributed nature of the proximate mechanisms is a severe barrier to understanding. We cling to single-locus models and the idea that phenotypic traits are caused directly by genes to avoid the complexity, but this is at our peril because real-world complex systems have properties that our imaginary simple systems do not.
Frame-shifting upward, when groups become the unit of selection, they also need to become the unit of functional analysis. Individuals and their interactions must be studied as proximate mechanisms, similar to genes within organisms. Sometimes the relationship between a group-level phenotype and the actions of individuals will be simple, but often it will be complex and distributed. Sometimes individuals will be aware of the role that they are playing, but often they will not. For anyone accustomed to thinking of individual organisms as units of functional analysis, it can be unsettling to think of them as proximate mechanisms; however, that is precisely what needs to be done and what Smaldino is arguing for.
Smaldino orients his discussion toward human cultural evolution, but most of his major points apply with equal force to any multilevel evolutionary process, regardless of the inheritance mechanism. It is fascinating to compare the theoretical literature on multilevel selection with artificial selection experiments at the group level. The theoretical models make simplifying assumptions, such as phenotypic traits coded by genes at single loci. Those assumptions result in certain conclusions, such as phenotypic variation among groups declining with the number of individuals colonizing the groups. When real groups are created in the laboratory, however, phenotypic variation among groups remains high even with large numbers of initial colonists. And when groups are selected on the basis of their phenotypic properties, there is typically a response to selection (Goodnight & Stevens Reference Goodnight and Stevens1997). The theoretical models with their simplifying assumptions got it wrong. Theoretical models that assume more complex interactions do a better job (e.g., Bijma & Wade Reference Bijma and Wade2008; Gilpin Reference Gilpin1975; Goodnight Reference Goodnight2011; Wilson Reference Wilson1992).
These conclusions apply to social insect colonies, as well as to other groups. One of my few disagreements with Smaldino is that he endorses the formulaic statement that eusociality in insects can be explained by high genetic relatedness or as “the extended phenotype” of the queen. High genetic relatedness contributes to heritable phenotypic variation among groups, but the colony is the unit of selection, and the proximate mechanisms that evolve are highly distributed among individuals. It is hard to imagine the cavity selection process of honeybee swarms as the extended phenotype of the queen when the queen plays no role whatsoever (Seeley Reference Seeley2010).
I do agree with Smaldino's points about equivalence. It is important to establish the equivalence of multilevel selection theory and inclusive fitness theory for models of individual-level traits, so that they are not pointlessly argued against each other. The more individuals become part of a complex distributed system with a group-level adaptive function, however, the more difficult it becomes to imagine them as optimizing units.
To summarize, whenever selection operates at a given level of a multitier hierarchy, units at that level should become the object of functional analysis, and units at lower levels should be studied as proximate mechanisms. This intuition already exists for the study of genes in individuals, when individuals are the unit of selection. It is only beginning to be applied for the study of individuals in groups, when groups are the unit of selection. Smaldino's target article is an important step in this direction with an emphasis on human cultural evolution, but the same algorithm applies to all multilevel evolutionary processes.
Two major themes have characterized multilevel selection theory from Darwin to the present (Wilson Reference Wilson1997). One theme is centered on the concept of a social group as like an organism, whose members work together to benefit the common good. Another theme is centered on the concept of altruism, which is selectively disadvantageous within groups and therefore requires between-group selection to evolve. These two themes sound similar and compatible, but in fact there are important differences. For example, in models of altruism, the most adaptive group is one that contains 100% altruists. In contrast, the parts of an organism are differentiated into organs and cell types. As another example, highly altruistic traits require strong group selection to evolve, to counteract strong selection within groups. But dividing labor, organ-like, need not be self-sacrificial at all.
I applaud Smaldino for advancing the “group as organism” theme in his target article. In this commentary, I will argue that his points apply to all multilevel evolutionary processes, not just human cultural evolution.
It is helpful to begin with the familiar case of individuals as the unit of selection in genetic evolution. Momentarily forget about group selection or genomic conflict and imagine a species in which natural selection acts entirely at the individual level. Organisms are selected on the basis of their phenotypic properties, which are mechanistically caused by genes and their interactions. Occasionally a single phenotypic trait can be attributed to a single gene, but most often the phenotype-genotype relationship is more complex, with single phenotypic traits caused by many genes and their interactions. Moreover, when the same phenotypic trait is selected in different populations, the response to selection often involves different genes. As an example, for adults, the ability to digest lactose has evolved at least twice in human populations, but different mutations were selected in each case (Holden & Mace Reference Holden and Mace2009). Given the chance nature of mutations, it is unlikely that the same one would arise in different populations subjected to the same selection pressure.
In short, when individuals are the units of selection, they also become the unit of functional analysis. The elements that comprise the unit are studied as proximate mechanisms. This is often simpler in principle than in practice, because the proximate mechanisms can be so complex, involving gene–gene and gene–environment interactions operating throughout development. In fact, some authors would fault my description as too gene-centric, when genes are merely parts of self-replicating developmental systems (e.g., Oyama et al. Reference Oyama, Griffiths and Gray2003).
The complexity and distributed nature of the proximate mechanisms is a severe barrier to understanding. We cling to single-locus models and the idea that phenotypic traits are caused directly by genes to avoid the complexity, but this is at our peril because real-world complex systems have properties that our imaginary simple systems do not.
Frame-shifting upward, when groups become the unit of selection, they also need to become the unit of functional analysis. Individuals and their interactions must be studied as proximate mechanisms, similar to genes within organisms. Sometimes the relationship between a group-level phenotype and the actions of individuals will be simple, but often it will be complex and distributed. Sometimes individuals will be aware of the role that they are playing, but often they will not. For anyone accustomed to thinking of individual organisms as units of functional analysis, it can be unsettling to think of them as proximate mechanisms; however, that is precisely what needs to be done and what Smaldino is arguing for.
Smaldino orients his discussion toward human cultural evolution, but most of his major points apply with equal force to any multilevel evolutionary process, regardless of the inheritance mechanism. It is fascinating to compare the theoretical literature on multilevel selection with artificial selection experiments at the group level. The theoretical models make simplifying assumptions, such as phenotypic traits coded by genes at single loci. Those assumptions result in certain conclusions, such as phenotypic variation among groups declining with the number of individuals colonizing the groups. When real groups are created in the laboratory, however, phenotypic variation among groups remains high even with large numbers of initial colonists. And when groups are selected on the basis of their phenotypic properties, there is typically a response to selection (Goodnight & Stevens Reference Goodnight and Stevens1997). The theoretical models with their simplifying assumptions got it wrong. Theoretical models that assume more complex interactions do a better job (e.g., Bijma & Wade Reference Bijma and Wade2008; Gilpin Reference Gilpin1975; Goodnight Reference Goodnight2011; Wilson Reference Wilson1992).
These conclusions apply to social insect colonies, as well as to other groups. One of my few disagreements with Smaldino is that he endorses the formulaic statement that eusociality in insects can be explained by high genetic relatedness or as “the extended phenotype” of the queen. High genetic relatedness contributes to heritable phenotypic variation among groups, but the colony is the unit of selection, and the proximate mechanisms that evolve are highly distributed among individuals. It is hard to imagine the cavity selection process of honeybee swarms as the extended phenotype of the queen when the queen plays no role whatsoever (Seeley Reference Seeley2010).
I do agree with Smaldino's points about equivalence. It is important to establish the equivalence of multilevel selection theory and inclusive fitness theory for models of individual-level traits, so that they are not pointlessly argued against each other. The more individuals become part of a complex distributed system with a group-level adaptive function, however, the more difficult it becomes to imagine them as optimizing units.
To summarize, whenever selection operates at a given level of a multitier hierarchy, units at that level should become the object of functional analysis, and units at lower levels should be studied as proximate mechanisms. This intuition already exists for the study of genes in individuals, when individuals are the unit of selection. It is only beginning to be applied for the study of individuals in groups, when groups are the unit of selection. Smaldino's target article is an important step in this direction with an emphasis on human cultural evolution, but the same algorithm applies to all multilevel evolutionary processes.