Levels of Selection: Burying the Units-of-Selection Debate and Unearthing the Crucial New Issues
Kern Reeve and Laurent Keller
The purpose of this volume is to sample current theoretical and empirical research on (1) how natural selection among lower-level biological units (e.g., organisms) creates higher-level units (e.g., societies), and (2) given that multiple levels exist, how natural selection at one biological level affects selection at lower or higher levels. These two problems together constitute what Leigh (chap. 2) calls the "fundamental problem of ethology." Indeed, as Leigh further suggests, they could be viewed jointly as the "fundamental problem of biology," when genes and organisms are also included as adjacent levels in the biological hierarchy. This generalization has the desirable property of immediately removing the long-standing conceptual chasm between organismal and molecular biologists.
These two problems are just beginning to be addressed, but their study promises in the decades ahead to generate crucial insights (perhaps the crucial insight) into biological evolution both on our planet and on imagined planets. To appreciate just how intriguingly intricate these problems are, we use an analogy from particle physics to generate a heuristically useful picture of the myriad interlocking and concatenated selective forces acting simultaneously at different levels of biological organization (fig. 1.1). This picture can be thought of as a visual guide to the kinds of multilevel selection issues addressed in the chapters in this volume.
First, however, we wish to make yet one more attempt to bury the issue that usually usurps discussions of the levels of selection at the expense of the truly interesting issues raised by these two problems; that is, the question of what unit is the "true" fundamental unit of selection. This issue emerges in cyclic debates about (a) whether genes or individuals are best seen as the true units of selection, and (b) whether groups of individuals can be units of selection. In our opinion, these questions have been satisfactorily answered repeatedly, only to reappear subsequently with naive ferocity in new biological subdisciplines (e.g., the group-selection controversy is currently generating copious amounts of smoke within the human sciences; see, e.g., Wilson and Sober 1994 and responses; Sober and Wilson 1998). The particularly frustrating aspect of these constantly renewed debates is that, even though they seemed to be sparked by rival theories about how evolution works, in fact, they often involve only rival metaphors for the very same evolutionary logic and are thus empirically empty.
Thus, we first pause to heap one more shovelful of dirt on the units-of-selection debates (a) and (b) above by very briefly reviewing what we believe to be their well-established, correct (if not universally known) resolutions.
Burying the Debate over Whether Genes or Individuals Are the Units of Selection
Organisms themselves are not replicated in the process of reproduction. They die, and only their genes are passed on. This led Dawkins (1976, p. 12) to propose that "the fundamental unit of selection, and therefore of self-interest, is not the species, nor the group, nor even strictly, the individual. It is the gene, the unit of heredity." Dawkins referred to this unit of self-interest as the replicator, or enduring unit of replication. Dawkins's view, which builds on previous ideas by Hamilton (1964a) and Williams (1966a), has been criticized as too reductionist by those who argue that genes are not directly visible to natural selection (e.g., Gould 1984; Sober and Lewontin 1984). That is, selection simply cannot pick among genes directly, but must select among packages created by and containing these and other genes (e.g., organisms). Dawkins (1982), however, recognized organisms and perhaps higher-level or laterally extended units as being vehicles, that is, the units directly confronting selection. We expect that, as the result of natural selection, vehicles will possess properties that maximize the replication success of the set of genes that cocreated them. This picture is slightly modified by the possibility of intragenomic selection (e.g., meiotic drive) favoring certain genes. In this case, selection may seem to choose among genes directly. However, a useful distinction still can be made between the replicator as a piece of genetic information and the "vehicle" as the physical stretch of DNA containing this genetic information. Thus, even in this case, it is only a vehicle (albeit a replicator-level vehicle) that directly confronts selection.
Thus, one internally consistent logical picture is that the unit of replication is the gene (or, more precisely, the information contained in a gene), and the organism is one kind of vehicle for such genes, a vehicle being the entity on which selection acts directly. The debate is resolved: Dawkins (1976) emphasized that genes (i.e., bits of genetic information) are the enduring units of replication, whereas Sober (1984) and Sober and Lewontin (1984) emphasized that individuals and possibly higher-level units, and not genes (as bits of genetic information), are vehicles. Case closed.
Burying the Old Group-Selection Debate
It is still embarrassingly common to read inaccurate statements in newspapers and even in professional biological literature that frogs have to produce many eggs to ensure the survival of the species because tadpoles suffer extremely high rates of predation, or that wolves have evolved ritualized displays to establish dominance hierarchies because physical combats would be too disadvantageous for the species. These naive statements betray a widespread and persistent misunderstanding of the level at which natural selection most commonly operates.
Wynne-Edwards (1962, 1993) has been the leading modern proponent of the idea that animals behave for the good of the group. He suggested that a population would become extinct if it overexploited its food resources; such between-population selection has fixed population-level adaptations to prevent extinction (such as animal displays to signal population density and thereby limit the risk of resource overexploitation). The most important criticism of this idea was formulated by Williams (1966a). Although selection at the population level is theoretically possible, in practice, such selection will be weak because of the high speed of within-population (between-individual) selection relative to that of between-population selection. Moreover, virtually all examples of group selection given by Wynne-Edwards (1962) have been shown to be better understood with the individual-selection paradigm (e.g., Alcock 1998; Kitchen and Packer, chap. 9). In Dawkins's terms, overwhelmingly strong theoretical arguments and empirical evidence tell us that individuals, far more commonly than populations, are the vehicles.
More recently, formal models of within-population group selection--that is, selection that occurs when a single breeding population is temporarily broken up into subgroups within which both cooperative and competitive actions can occur--have been developed under the rubric of "new group selection," "intrademic group selection," or "trait-group selection." These models simply partition ordinary individual fitness into within- and between-group components--often using the clever covariance approach of Price (1972)--and allow detailed predictions of the circumstances favoring the evolution of traits affecting both within- and between-group fitness in various ways (e.g., Wilson 1975; Wilson and Sober 1989). These models are mathematically equivalent to individual-selection (i.e., inclusive fitness) models, however, and therefore do not point to a fundamentally different kind of evolution (e.g., Dugatkin and Reeve 1994; Bourke and Franks 1995). Thus, acknowledging the utility of these models should not be taken as a tip-toed retreat to Wynne-Edwardsian interpopulation selection, as is often mistakenly feared because of the shared label of "group selection." Acceptance of these models also does not commit one to a particular view about the relative balance of cooperation and conflict in nature, because either can have any degree of strength in these models (Dugatkin and Reeve 1994). Furthermore, these models fit comfortably into Dawkins's (1982) conceptual scheme because the "groups" in these models (e.g., animal societies) can be viewed simply as vehicles above the level of the individual (Seeley 1997). A distinct virtue of intrademic group-selection models is that they provide a simple, standardized means of unveiling the structure of selection working simultaneously at different hierarchical levels (e.g., Dugatkin and Reeve 1994; Reeve and Keller 1997; Keller and Reeve, chap. 8).
Unearthing the New Issues
Most class lectures on levels of selection begin and end with a discussion of the two (now stale) debates above. However, the current theoretical excitement in theoretical and empirical research in multilevel selection centers on the two problems set forth at the beginning of this introduction, namely, (1) how natural selection among lower-level biological vehicles creates higher-level vehicles, and (2) given that multiple levels of vehicles exist, how natural selection at one level affects selection at lower or higher levels. The richness of these two questions can be conveyed with the help of figure 1.1, which pictures interactions within and between lower- and higher-level vehicles (e.g., for vehicles ranging from single-celled organisms, to multicellular individuals, to social groups of individuals).
An analogy from particle physics is useful here. Higher-level vehicles can be seen as composites of lower-level vehicles, each of which experiences both evolutionary repulsive and attractive bipolar forces with other units at the same level (fig. 1.1). The separated unipolar forces can be viewed as having magnitudes equal to the absolute inclusive fitnesses for peaceful cooperation with a same-level partner unit or for competitive suppression (e.g., killing) of the same partner unit. The outgoing arrows refer to the absolute inclusive fitness of a vehicle that leaves the group; thus, this represents a second evolutionary force, which we call "centrifugal force," that tends to break apart the group. In this scheme, a cooperative group of lower-level units will be stable only if, for every unit, the attractive force exceeds the maximum of the repulsive and centrifugal forces also acting on that unit. (See box 1.1 for elaboration of the exact nature of these forces.)
Figure 1.1 makes explicit several key features of the evolution of higher-level vehicles from lower-level ones. First, a higher-level vehicle is created from a lower-level vehicle whenever an attractive force arises that exceeds both the maximal repulsive and centrifugal forces. Interestingly, repulsive forces among unbound lower-level units can create binding forces between other such units, for example, as when ancestral multicellularity increased the fused cells' ability to outcompete single-celled organisms for resources, or when social grouping increased the ability of individuals to defend resources from intruding robbers.
Second, because the magnitude of each of the forces depends on inclusive fitness, which in turn depends on both genetic relatedness and multiple ecologically determined costs and benefits of cooperation and noncooperation, it follows that understanding higher-level vehicle formation requires knowing both genetic and ecological factors that generate attractive, repulsive, and centrifugal forces. Ecology will be crucially important in determining the magnitude of the centrifugal force, by strongly affecting the expected reproductive output of a dispersing, solitary vehicle. The most complete theories of vehicle formation will thus be those that specify both the ecological and the genetic contexts for vehicle creation.
Third, even if the creation of higher-level vehicles requires that attractive forces exceed repulsive and centrifugal forces, this does not imply that the latter two forces will disappear once the higher-level vehicles are formed. They may continue to operate and shape the features of the higher-level vehicle (just as the conformation of a stable molecule will depend on the internal electrical repulsive forces). Indeed, repulsive forces may sometimes strengthen sufficiently to cause subsequent vehicle breakdown. For example, the attractive forces will often be sufficiently weak and variable that a composite vehicle lasts only a short time, just as an unstable, heavy particle created in an particle accelerator may leave only a short track on a photographic plate before disintegrating into component particles. Analogously, in many if not most animal species, the only cooperative groups are fleeting associations of mates during courtship, copulation, and mate defense; that is, the inclusive fitness for cooperation (attractive force) exceeds that for noncooperation (repulsive and centrifugal forces) only until mating is completed. A complete theory of social evolution will tell us not only the contexts in which higher-level vehicles form, but also the contexts in which they break down.
Finally, this model, represented in figure 1.1, predicts that larger cooperative groups are inherently less likely to be stable. Suppose there are n lower-level vehicles within the cooperative group (i.e., higher-level vehicle). If the group is to be completely stable, the attractive forces must exceed the repulsive forces for all n(n - 1) = n2- n polar interactions, and, in addition, the attractive forces must exceed the centrifugal forces for all n cases, for a total of n2- n + n = n2 requirements. Thus, the number of Hamilton's rule requirements for group stability increases as the square of the number of group members! This immediately suggests that larger groups will be progressively less stable, unless high genetic relatedness, positive correlation among subunits in the values of their inclusive fitness parameters, or some kind of between-subunit interaction somehow forces the multiple Hamilton's rule requirements to be satisfied en masse. Furthermore, as lower-level vehicles are nested to form higher-level vehicles, say from lower-level vehicles consisting of nl subunits each to a higher-level vehicle of nh lower-level vehicles, the total number of Hamilton's rule requirements rapidly becomes compounded to (nlnh)2. This immediately suggests that vehicles created by the nesting of successively higher-level vehicles will become progressively less stable, again unless some condition or process causes these requirements to be satisfied at once.
Now we can represent the questions that together form the "fundamental problem of biology," and thus the conceptual structure of this book, in terms of the picture in figure 1.1. Questions A-C below refer to processes A-C in figure 1.2.
A1. What attractive evolutionary forces bind low-level vehicles (i.e., vehicles nearly at the same level as the replicators themselves), like physical stretches of DNA (replicators being the genetic information encoded in such stretches), chromosomes, and cells, into intermediate-level vehicles, like multicellular organisms? Under what conditions do these attractive forces exceed the repulsive and centrifugal forces and under what conditions do they not?
This topic is addressed in chapters 3 and 4. A central question in the study of the origin of life is how cooperating groups of small replicator-level vehicles could have arisen and how they could have protected themselves against invasion by molecular parasites. Szathm ry (chap. 3) argues that synergism (i.e., division of labor and complementation of functions) provided the most important attractive force leading the first replicator-level vehicles to associate. Cooperation was also facilitated by genetic compartmentalization that resulted from limited dispersal and bonding of different replicator-level vehicles (which were therefore obliged to "sit in the same boat"; Szathm ry, chap. 3). Compartmentalization represented an important step in the overriding of repulsive and centrifugal forces and also probably led to their subsequent weakening. Finally, the benefits of division of labor, together with the many advantages of larger size, were probably the two important attractive forces that favored the transition from unicellular to multicellular life(Michod, chap. 4).
A2. Similarly (as we move up the hierarchy of nested vehicles), what attractive evolutionary forces bind intermediate-level vehicles, such as organisms, into higher-level vehicles, such as social groups of individuals? Under what conditions do these attractive forces exceed the repulsive and centrifugal forces?
This topic is addressed in chapters 5, 6, and 8-11. In sexual species, the necessity of finding a mate provides an inescapable attractive force. In most species, however, this attractive force is transient because males and females typically have low genetic interest in each other's future (Lessells, chap. 5). Another attractive force may keep parents together: their common genetic interest in rearing their offspring. The magnitude of this attractive force directly depends on the degree to which greater parental investment increases offspring reproductive success (Godfray, chap. 6). This positive force is opposed by the centrifugal force created by mating opportunities outside of the pair bond. Thus, the dynamics of the attractive and centrifugal forces set the stage for a variety of conflicts (new, subtle repulsive forces) between mates over their relative investment in parental care (Lessells, chap. 5; Godfray, chap. 6).
Attractive forces may also lead individuals other than mates to cooperate when this increases either their survival or number of offspring produced or the survival and fecundity of relatives. Higher relatedness increases the magnitude of the attractive forces and decreases the magnitude of the repulsive forces (because increased relatedness between interacting individuals enhances the inclusive fitness payoffs for cooperation and reduces the inclusive fitness payoffs for group-destructive selfishness). Increased relatedness thus increases the scope both for reproductive altruism (whereby individuals forgo direct reproduction to help others) and possibly for group stability, although models of optimal reproductive skew (Keller and Reeve, chap. 8) predict that dominant members of animal societies may actually increase the attractive force for potential subordinate helpers when the latter are less related, erasing any net effect of relatedness on group stability (Reeve and Ratnieks 1993; Reeve 1998a). Not surprisingly, unreciprocated altruism occurs nearly exclusively in groups formed by closely related individuals (Keller and Reeve, chap. 8; Kitchen and Packer, chap. 9; Maynard Smith, chap. 10). Groups of unrelated individuals are generally stable only when group living provides direct reproductive benefits to all group members, when it requires no reproductive altruism, and opportunities for cheating are limited (i.e., repulsive forces are weakened) (Kitchen and Packer, chap. 9). The other important factor shaping social life is the ability for individuals to disperse successfully and reproduce outside the group. Groups will be inherently more stable when such opportunities are limited (weak centrifugal forces).
Interspecific mutualism provides another interesting case of attractive forces being stronger than repulsive and centrifugal forces, the two forces that generally predominate in interspecific interactions. Interestingly, the same positive force (the benefits of division of labor) that facilitated the evolution of early life is also probably important in shaping the nature of interspecific cooperation (Herre, chap. 11). Moreover, Herre provides examples showing that stable interspecific cooperation (or reduced virulence) is facilitated by the long-term association of interspecific individuals and parallel vertical transmission of the symbionts (from parents to offspring). The consequence of symbionts being only vertically transmitted is similar to the effect of compartmentalization of replicator-level vehicles during the early evolution of life because, in both cases, the interests of replicators are aligned, increasing the magnitude of the net attractive force.
B1. How do attractive, repulsive, and centrifugal forces among lower-level vehicles interact to shape the properties of intermediate-level vehicles like individuals? Can different repulsive forces sometimes nullify each other within intermediate-level vehicles (as when there is some mechanism of policing against intragenomic selfishness) and thus leave an imprint on the characteristics of the intermediate-level vehicle (such as increased reproductive efficiency resulting from greater internal cooperation)? Are there attractive forces (perhaps arising only after the creation of the intermediate-level vehicle) that would overcome all or some of the original repulsive forces and thus leave an imprint on the characteristics of the intermediate-level vehicle?
These topics are addressed in chapters 4, 7, and 12. The two main forces shaping the integrity of the organism are the attractive and repulsive forces because lower-level vehicles (genes and cells) have little or no opportunity to leave the organism and embark on independent and solitary lives (except in some primitive multicellular organisms). The repulsive forces stem from the benefits that lower-level vehicles (gene-level vehicles and cells) may gain by increasing their reproductive rates at the expense of the other vehicles forming the organism. Thus, genes may increase their reproduction by subverting meiosis in diploid organisms. Similarly, cells may reap a short-term reproductive benefit at the long-term expense of the organism through uncontrolled cell proliferation (cancer). Michod (chap. 4) and Pomiankowksi (chap. 7) provide examples of how repulsive forces can nullify each other to decrease conflicts between genes and enforce fair meiosis. For example, it is the mutual interest of genes in multicellular organisms in decreasing repulsive forces that probably led to the sequestration of a cell lineage set early in development for the production of gametes (Michod, chap. 4). Mutual competition (repulsion) between cell lineages might result in no net advantage for either; moreover, such competition might greatly limit the efficiency of the vehicle formed by their cooperation. The separation of the germ line reduced the opportunity for conflict (greatly reducing repulsive forces) and thus was a first step toward the evolution of individuality (i.e., a higher-level vehicle with stronger attractive than repulsive forces). Similarly, because most genes in the genome suffer from the detrimental effects of meiotic-drive genes (unless linked with them), they are selectively favored to suppress the selfish actions of such genes (Pomiankowski, chap. 7). Nunney (chap. 12) suggests that between-lineage species selection may cause the long-run predominance of genetic architectures that decrease the risk of cancer (detrimental to the organism) and also that decrease the probability of a shift from sexual to asexual reproduction (the latter being detrimental to the species). (Note that this is not Wynne-Edwardsian group selection, because Nunney is only speaking of differential extinction among lineages that have different biological characteristics, the latter characteristics all having been fixed by within-population selection.) Under this intriguing view, lineages with relatively high repulsive and low attractive forces (i.e., those in which lower-level vehicles are less likely to form higher-level vehicles) are more likely to become extinct, leading to a long-term lineage selection for clades that exhibit well-elaborated, high-level vehicles.
B2. Similarly, how do attractive, repulsive, and centrifugal forces interact to shape the properties of high-level vehicles like animal societies? Can repulsive forces sometimes nullify each other within high-level vehicles (as in policing against selfishness) and thus leave an imprint on the characteristics of the high-level vehicle (such as increased efficiency resulting from greater internal cooperation)? Are there attractive forces (perhaps originating only after the initial creation of the high-level vehicle) that would nullify all or some of the original repulsive forces and thus leave an imprint on the characteristics of the high-level vehicle?
These topics are addressed in chapters 8-11. For example, Keller and Reeve (chap. 8) discuss how policing and bribing can promote intragroup cooperation within animal societies by in effect weakening repulsive forces or strengthening attractive forces. Similarly, Maynard Smith (chap. 10) investigates the conditions favoring the emergence and enforcement of social contract strategies to punish selfish behaviors in human societies. Finally, the evolution of co-adapted traits in obligately mutualistic species (e.g., figs and their associates; Herre, chap. 11) provide yet another example of attractive forces that arose or strengthened after the initial creation of a higher-level vehicle from the mutualistic pair of organisms, that is, following the evolution of complete reproductive interdependence.
C1. Perhaps the most unexplored question concerns how interactions between lower-level vehicles might affect the interactions between intermediate-level vehicles and thus affect the properties of the highest-level vehicle. For example, Keller and Reeve (chap. 8) describe one of Reeve's (1998b) hypotheses for the absence of nepotism within insect societies. Intragenomic selection on parentally imprinted alleles involved in kin recognition (lowest-level vehicles) might favor sabotaging of the potential nepotism-dispensing machinery of individuals (intermediate-level vehicles), leading to the lack of nepotism and thus increased cooperation within hymenopteran societies (highest-level vehicles).
BOX 1.1. THE ABSOLUTE INCLUSIVE FITNESS "FORCE"
By absolute inclusive fitness, we mean a focal vehicle's direct reproductive output plus the sum, over all related vehicles, of the product of its relatedness to the vehicle times the reproductive output of that vehicle. Note that we use absolute outputs, rather than changes in outputs caused by the focal vehicle (the latter is used in most verbal formulations of inclusive fitness). The outputs of vehicles unaffected by the focal vehicle's actions will appropriately vanish when the absolute inclusive fitnesses associated with the two actions by the focal vehicle are compared by subtraction, because such outputs will have exactly the same value in the compared inclusive fitnesses. For example, suppose a phenotype A causes the focal animal to have x offspring and a relative (of relatedness r) to have y offspring. The corresponding offspring numbers for phenotype B are z and w. The magnitudes of the corresponding absolute inclusive fitnesses (i.e., of the "forces") are
x + ry
for A and
z + rw
for B. The "net force" is then obtained by subtraction and is readily seen to equal
(x - z) + r(y - w),
which is the same as Hamilton's rule when set greater than zero (Grafen 1982, 1984, 1985). If action B did not change the reproductive output of the focal individual's relative, then y = w and the term r(y - w) would simply vanish. If the net force is greater than zero, phenotype A is favored.
The use of absolute offspring number in the above inclusive fitness calculations may sound wrong to some because of well-known theoretical admonitions against (1) including personal offspring added because of help received from others, and (2) giving inclusive fitness credit for the reproductive outputs of relatives that are unaffected by the phenotype, both of which can cause gross overestimation of the kin-selective value of a cooperative strategy (Grafen 1982, 1984). However, this is an error only when a strategy's absolute inclusive fitness is compared with zero, not when the absolute inclusive fitness for one strategy is compared (by subtraction) with the absolute inclusive fitness of another strategy. The latter procedure automatically yields the appropriate description of net selective force by generating Hamilton's rule. It should be mentioned, however, that the "absolute inclusive fitness force" approach is precisely true only if there are additive costs and benefits and weak selection (Grafen 1984, 1985). Conditionality of phenotypic expression can make these assumptions more likely to hold (Parker 1989).
Our characterization of the magnitudes of the attractive, repulsive, and centrifugal forces properly ties vehicle behavior to the interests of the ultimate replicators, the genes that create the vehicles. Why? This scheme correctly specifies when kin selection favors cooperation as summarized in Hamilton's rule. That is, the sign of net inclusive fitness force determines whether kin selection favors cooperation over killing or ejecting the partner and also over leaving the group to reproduce independently. (The physical analogy breaks down a bit here, because in the physical case, the net attractive force would simply be the vector sum of all three forces, not the difference between the attractive force and the maximum of the repulsive and centrifugal forces. Despite this, the physical picture is useful.)
We thank E. Leigh for useful comments on the manuscript. We were supported by grants from the Swiss and U.S. NSF.
Return to Book Description
File created: 8/7/2007
Questions and comments to: email@example.com
Princeton University Press