For natural selection to proceed there must be heritable variation in phenotypes and the variation in phenotype must be associated with differential survival and/or reproduction, i.e., there must be differential fitness. By inference then, any entities exhibiting heritable variation in rates of reproduction can evolve. We need not restrict our thinking to "individuals" in "populations" in the traditional senses of these words.

Nature is organized in a hierarchical fashion. In terms of entities that can be heritable we can consider genes, chromosomes, genomes, individuals, groups, demes, populations, species, etc. Each of these entities meets the requirements of units that can be acted upon by selection. At which level(s) does selection act? Answer: all of them. What then is the important unit of selection? Answer: it depends.

First, some historical context. Serious consideration of a unit of selection other than the individual was advanced by V. C. Wynne-Edwards (1962, Animal Dispersion in Relation to Social Behavior). Populations have their own rates of origination and extinction and selection could thus operate at the level of the group. Idea based on observation that many species tend to curb their reproductive rate/output when population densities are high. This behavior would favor groups that exhibited the behavior and select against those that did not; i.e., there would be group selection.

G. C. Williams responded to this idea with Adaptation and Natural Selection (1966) arguing that this behavior would be less fit than a cheating behavior where individuals did not reduce their reproductive output at times of high density/low food availability. In general selection at the level of the individual would be much stronger than selection at the level of groups. In keeping with Williams' claim that one should always seek the simplest explanation for selective/adaptive explanations, individual selection is usually sufficient to account for patterns.

Group selectionist thinking leads to statements such as "good for the species" when it is entirely likely that it may be good for the individual as well: reduced reproductive effort in times of low food may increase an individual's reproductive output at a later date.

Examples of selection at different hierarchical levels: Genic selection is selection at the gene level; best example is meiotic drive (segregation distortion) where one gametic type (often one chromosomal type) is transmitted into the gamete pool (or next generation) in excess (or deficiency). The T locus in mice: affects tail length but also viability. TT homozygotes have normal long tails; Tt heterozygotes have short tails and transmit ~ 90% of the t allele to their sperm; tt homozygotes are sterile. Meiotic drive will increase frequency of t allele to point where that become frequent enough to occur as tt homozygotes with appreciable frequency, whereupon selection works against t alleles. Opposite Selection at two levels selection for at the level of the gene; against at the level of the genotype (organism).

In this system the balance of opposing selection coefficients at different levels should give an equilibrium allele frequency of 0.7 for f(t) allele (using data not provided here). In nature f(t) = 0.36. Discrepancy due to small local groups and drift. Some local breeding groups (2 - 4 individuals) fixed for the t allele and since tt is sterile, these demes go extinct reducing the f(t). Thus we have selection at three levels: genic, individual and intergroup all contributing to the maintenance of the t/T polymorphism.

Would we expect to detect meiotic drive systems in natural populations? If a new mutation arose that introduced a bias in the transmission of the chromosome on which it was located, then it would sweep to fixation and the locus would be homozygous for the "drive" allele. Meiotic drive can only be detected in heterozygous state, so the drive system would disappear when the drive allele went to fixation. There will be a window of time where the allele is increasing in frequency, but this could be short-lived. If the drive allele reduced viability in the homozygous state (as the T locus example), then variation can be maintained and the drive system would persist longer, making it more likely to be detected.

Another case where genic selection may act: sex ratios. Why should the sex ratio be 1:1 in most diploid species? Assume a sex ratio of 40% males and 60% females. Males in this case are in limited supply. Any gene leading to the production of more males (a allele results in more males the A allele at a sex determination locus) will be favored until the frequency of males is >50%. Sex ratios tend to stabilize at 50:50 (R. A. Fisher, 1930).

Kin selection and altruistic behavior many species of animals that live in groups give warning calls which alert other individuals about predators, etc. How could this behavior evolve when making the call alerts the predator to the callers location and increases the possibility of the caller becoming prey. Mammals that nurse their young: major energy investment for the mother may be thought to reduce her fitness, but make obvious sense sine the individuals that benefit are close relatives (offspring). Put in fitness terms: how could a trait evolve that lowers individual fitness?

Key point is the term individual. If the ones that benefit from the behavior are related, the loss in individual fitness may be regained in inclusive fitness, i.e., individual fitness plus fitnesses of relatives. W. D. Hamilton argued that an altruistic trait could evolve if cost to altruist/benefit to recipient < genetic relatedness (C/B < r) where r is an estimate of the probability of the donor and recipient having allele identical by descent. For example parent - offspring have r=0.5; siblings have r=0.5; grandparent-grandchild have r=0.25.

Idea of inclusive fitness implies that one's fitness is determined by one's own life time reproductive output and the reproductive output of relatives, scaled by their degree of relatedness (r). In the warning call example, if calling out to warn about the arrival of a hawk killed you but saved three reproductively active siblings, it would have been worth it. If it only saved two siblings, it probability wasn't worth it. Obviously one cannot tabulate the payoff of event x, y, or z. The point is that the notion of inclusive fitness provides a fitness context where altruistic behavior could evolve even when it appears to decrease individual fitness. Two modes of selection: individual selection opposes altruism; selection among kin groups favors altruism.

Classic examples: helpers at the nest in birds. Young offspring remain at the nest to help their parents produce more siblings in subsequent years/seasons. Helpers may contribute more to their own fitness by aiding in the production of siblings than by trying to reproduce themselves and failing due to lack of experience or availability of nest sites. Sterile workers in hymenoptera (ants, bees, wasps): males are haploid (develop from unfertilized egg) so sisters have r=0.75 because male contributes the same allele (relatedness between sibs for paternal genes = 1.0; relatedness between sibs for maternal genes = 0.5; among diploid female workers this averages out to r=0.75). A female worker does more to propagate her own genes by staying in the nest and aiding in the production of sisters (r=0.75) than by going off and producing her own daughters (r=0.5). Used as an explanation for the evolution of sociality (e.g., colonies) in hymenoptera.

Group selection = variation in the rate of increase or extinction among groups as a function of their genetic composition. Again consider how an altruistic trait could increase in frequency. Differential rates of extinction: allele A confers altruistic behavior; at a selective disadvantage to allele a within the group. Should lead to the reduction of allele A. But groups with high frequency of A may be less likely to go extinct (due to better exploitation of resources). Over all groups with high frequency of A persist and f(A) increases.

Differential productivity: similar to model above but altruistic trait affects reproductive output of group. Selection against A allele within groups (selfish types have higher short term fitness) but groups with high frequency of A exploit resources more prudently and actually produce more offspring over the long term: f(A) increases. Model this as follows: assume a haploid trait with A=altruist, a=selfish; p=f(A). In each population p decreases within a population through time due to selfish individuals out competing altruistic individuals. But in all populations as a whole the altruist gene increases over time. Below, the average f(A) across all populations is 0.5 at the start:

Clearly, if this system were to continue for many generations, the frequency of the altruist gene would decrease within each population. But under conditions where the selection favoring selfish genes was weak and the group selection increasing the probability of staying extant (or the growth rate of the population) was strong, and altruist allele might be preserved. Because the conditions are so restrictive, group selection is presumed to be a rare phenomenon.

Group selection often involves plausible models but require that interdeme (group) selection be strong. Would have to be very strong to overcome selection among individuals within populations. Other complicating factors: turnover rate of individuals is faster than of populations/groups; fixation of less fit allele is unstable to invasion by new mutant allele or "selfish" allele introduced by gene flow. New research on multilevel selection suggests that there should not be the necessary association between altruism and "sacrifice" or genetic "suicide". Cooperation among individuals can actually result in higher group fitness without the assumed loss of individual fitness (see a meeting review in Science (9 August 1996) vol 273:739-740). D. S. Wilson makes the analogy between the optimal clutch size argument of D. Lack and the optimal group of Wynne-Edwards. With too many eggs in a clutch an individual may die trying to support them all, so some intermediate clutch size is "optimal" (see Life History Lecture). Optimal groups may evolve intermediate density by the same trade-off mechanism.

A further problem for group selection: with localize population structure, there can be considerable inbreeding which increases relatedness (r). Thus inter "group" selection that gives the appearance of evolution of altruistic traits may be mediated by kin selection due to the high relatedness among individuals.

Later we will consider species selection. Some lineages have more species than others, but are these lineages more fit? Is this simply a pattern (more species) or is it really a different process? Is it simple like bacteria in chemostats: a higher birth/death ratio; some lineages seem to speciate faster than their members go extinct? Is this mediated at the level of the species, or can we explain it (as G. C. Williams might like) at the level of individuals within populations?

Richard Dawkins likes to couch this discussion in terms of replicators and vehicles. Replicators are any entities of which copies are made; selection will favor replicators with the highest replication rate. Vehicles are survival machines: organisms are vehicles for replicators and selection will favor vehicles that are better at propagating the replicators that reside within them. There is a hierarchy of both replicators and vehicles. The key issues are that 1) the "unit" of selection is one that is potentially immortal: organisms die, but their genes could be passed on indefinitely. The heritability of a gene is greater than that of a chromosome is > that of a cell > organism > and so on. But , because of linkage we should not think of individual genes as the units; it is the stretch of chromosome upon which selection can select, given certain rates of recombination. Issue 2) is that selection acts on phenotypes that are the product of the replicators, not on the replicators themselves, but the vehicles have lower heritability and immortality than replicators. What then is the unit of selection?? All of them, just of different strengths and effects at different levels.