This section discusses outlaws and the genetic conflicts to which they lead. The focus is thus on the process of genic selection and its evolutionary consequences, rather than the gene's-eye perspective.
An outlaw, or SGE, is a gene that enjoys a transmission advantage over other genes in the same organism but does not increase the organism's fitness (Alexander and Borgia 1978; Dawkins 1982; Hurst et al. 1996; Hurst and Werren 2001). Most outlaws in fact reduce organismic fitness; this generates selection pressure at all unlinked loci for genes that suppress the outlaw's effects, leading to genetic conflict. Such conflicts are usually called 'intra-genomic', for they involve conflict between the different parts of a single genome. However, Maynard Smith and Szathmary (1995) argue that some SGEs are better thought of as parasites or endo-symbionts, rather than as part of host genome, hence do not generate intra-genomic conflict, but rather conflict between the multiple genomes inhabiting a single organism.
Genic selection, as defined above, occurs when there are fitness differences between the genes within an organism. Given this definition, it follows that all outlaws spread by genic selection. But the converse is not necessarily true. A gene might boost the fitness of its host organism and also secure a transmission advantage over its allele. Such a gene would not be an outlaw, and would not induce selection for suppressors at unlinked loci; but its spread would still be due, in part, to genic selection. The effect of such a gene would be to benefit the whole genome, but to take a disproportionate share of the benefit for itself.
Defining genic selection this way requires that the notion of 'genic fitness' be appropriately understood. Consider a simple one-locus model with two alleles A and B, and thus three diploid genotypes AA, AB, and BB. Suppose that segregation is Mendelian—neither A nor B is a segregation-distorter. This means that no genic selection occurs—both alleles within an AB heterozygote have identical fitness, for they share the organism's gametic output equally. However, the overall fitness of the A and B alleles, averaged over all the genotypes, may well differ.6 If so, one might be tempted to argue that there are fitness differences between the genes within an AB heterozygote, and thus that genic selection is occurring. But this would be a confusion. The crucial distinction is between the fitness of a token particle within a collective, and the fitness of that particle-type averaged across collectives. When we say that genic selection occurs if the genes within an organism differ in fitness, it is the former, not the latter, sense of 'genic fitness' that is pertinent.
This conception of genic selection sits well with the Price approach to multi-level selection. In Chapter 2, we saw how the Price approach can be applied to a one-locus model, by treating each diploid organism as a collective containing two particles. It then follows that organismic (or genotypic) selection occurs if there are fitness differences between organisms; while genic selection occurs if the alleles within the AB heterozygote differ in fitness, that is, if there is segregation distortion. The overall change in gene frequency depends on both levels of selection, as we saw.
This illustrates an important general point. By regarding an organism's genome as a 'group' of genes, and thus recognizing two potential levels of selection—within-group and between-group—one can apply the lessons of multi-level selection theory to the evolution of genomic organization. This approach has yielded substantial insights. For example, one lesson of multi-level theory is that the evolution of cooperative wholes requires suppression of competition among the parts (Frank 1995b; Michod 1999). This has been used to help explain why meiosis
6 These overall fitness values, usually denoted wa and wb, are sometimes called 'marginal' allelic fitnesses; they determine which allele, if either, will increase in frequency. By contrast, the within-genotype allelic fitnesses determine whether or not there is a component of genic selection, i.e. whether or not segregation is Mendelian. See Kerr and Godfrey-Smith (2002), Okasha (2004a), and Section 5.5 for further discussion of the difference between the two types of genic or allelic fitness.
is usually fair, and why mitochondria are inherited uniparentally in the vast majority of eukaryotes (Haig and Grafen 1991; Hurst and Hamilton 1992). Both of these genomic features can be interpreted as adaptations for minimizing the damaging effects of lower-level selection on the integrity of the whole.
Many types of SGEs have been discovered, in both plants and animals. Ultimately, they all owe their existence to one of two related factors (Pomiankowski 1999; Maynard Smith and Szathmary 1995). The first is that the different genetic entities within an organism, for example, autosomes, sex chromosomes, and organelles, do not always share the same mode of transmission. For example, mitochondrial genes are transmitted maternally while autosomal genes are transmitted biparentally. This creates a conflict of interest: mitochondrian genes favour a female-biased sex-ratio among the progeny of their host, while autosomal genes do not. The second is sexual reproduction. During meiosis, which precedes sexual fusion, only half the nuclear genes are passed to each gamete. So a gene which can increase its probability of surviving the meiotic cut will be selected, even if it reduces organismic fitness. This logic explains the behaviour of both segregation-distorters and transposons, which insert themselves at many chromosomal locations within the genome, thus boosting their chances of passing to a gamete.
Transposons are believed to be responsible for much of the non-coding DNA in the eukaryotic genome, a hypothesis first advanced by Doolittle and Sapienza (1980) and Orgel and Crick (1980). These authors argued that since transposition within the genome is sufficient to explain the prevalence of non-coding DNA, it is a mistake to think that such DNA must benefit the organism; it could simply be parasitic. This methodological point is clearly correct. In general, if selection at a given level can explain some feature, it is wrong to assume that that feature must benefit entities at another level; though cross-level synergism can occur. It is also possible that transposable elements were originally parasitic but later evolved more symbiotic relationships with their hosts (Pomiankowski 1999; Hoekstra 2003). In any case, the fact that transposons compose so large a fraction of the eukaryotic genome shows the evolutionary importance of selection at the genic level.
Further evidence of genic selection's importance comes from the discovery of segregation-distorters in numerous species. The best-known cases involve genes at two tightly linked autosomal loci, one producing a toxin, the other an antidote (Lyttle 1991; Maynard Smith and Szathmary 1995). For example, in Drosophila melanogaster, an allele at the 'toxin' locus, denoted Sd, produces a product that inactivates any gametes that do not produce antidote. Whether a gamete produces antidote depends on which allele it has at a second locus; the Rsp+ allele does not produce antidote, while the Rsp allele does. So if the two loci are tightly linked, then the Sd—Rsp pair constitutes an effective system for meiotic drive. Both alleles will achieve greater than fifty per cent representation in the successful gametes of their host organism, though they reduce organismic fitness.
This meiotic drive system, and similar systems in other species, depends crucially on the Sd and Rsp genes being in linkage disequilibrium; so recombination will tend to break it up. This is the basis of Haig and Grafen's (1991) suggestion that the function of recombination may be to prevent meiotic drive, and thus reduce the deleterious effects on organismic fitness. In effect, Haig and Grafen are suggesting that recombination evolved as a way of resolving the conflict between two levels of selection. Selection at the genic level led meiotic drive systems, consisting of a linked toxin—antidote pair, to evolve; because of the resulting negative effects on organismic fitness, selection at the organismic level led recombination to evolve, thus restoring fair meiosis.
A different type of conflict can occur between nuclear and cytoplasmic genes, arising because the latter are usually only transmitted maternally. Cytoplasmic outlaws exploit this fact to gain a transmission advantage for themselves at the expense of the nuclear genes. For example, in many angiosperm species, mitochondrial genes suppress male function—a phenomenon known as 'cytoplasmic male sterility'. The advantage to the mitochondrial gene is clear—it is transmitted in ovules but not pollen, so gains if the plant devotes more resources to making the former (Maynard Smith and Szathmary 1995). Nuclear genes do not benefit from male sterility, since they are transmitted in pollen, so are selected to suppress the effects of the mitochondrial gene, that is, to restore male sexual function. Other forms of cytoplasmic outlawry include converting males into females, and male killing (Hurst et al. 1996).
Though uniparental inheritance ofcytoplasmic genes leads to conflict over the sex ratio, many theorists believe that uniparental inheritance is itself an adaptation for eliminating another source of conflict (Hoekstra 1990; Hurst and Hamilton 1992). If mitochondria were biparentally inherited, then a mutant mitochondrian that abandoned normal cellular function for faster replication could gain access to the gametes and spread. Uniparental inheritance greatly reduces this problem, by ensuring that all the mitochondria in a cell are genetically similar, and making it easier for selection to weed out mutants. Thus the integrity of the whole is preserved by reducing the variance, hence opportunity for selection, among the parts. Ironically, uniparental inheritance eliminates one source of conflict—between variants in the cytoplasm—but creates another—between nuclear and cytoplasmic genes.
This brief survey of SGEs highlights a number of important points. The first is the importance of transmission asymmetry, which itself derives from sexual reproduction, in facilitating selection at the genic level. The second is that the 'interests' of different genetic elements converge to the extent that they are transmitted similarly, and diverge to the extent that they are transmitted dissimilarly. The third is the pervasive within-organism conflict to which outlaws give rise. Leigh (1977) observed that outlaws will typically be 'outnumbered' by the rest of the genome, so their effects will tend to be suppressed by the majority. This explains why organisms usually function as cohesive wholes despite the potential for conflict among their constituent genetic units. That is, genic selection may be relatively infrequent today precisely because organisms have evolved means to suppress it.
There is a minor irony lurking here. Outlaws are often regarded as the ultimate vindication of the gene's-eye approach, proof that the traditional organism-first paradigm cannot be sustained. There is something to this view. Although the origins of gene's-eye thinking lie in kin selection theory, which does not involve outlaws or genic selection, the gene's-eye perspective provides a framework into which outlaws fit easily. Indeed without that framework, it is hard to see how evolutionary theory could begin to make sense of outlaws. On the other hand, multi-level selection theory, which is sometimes regarded as the antithesis of gene's-eye thinking, is also crucial to understanding outlaws and genetic conflicts. The idea that a genome is a collective of genes with partially overlapping interests, that internal competition must be suppressed if the collective is to function as a whole, that selection at the collective level will favour such suppression, and that reducing the variation among the parts is one way to achieve it, are all themes from multi-level selection theory. Therefore, the gene's-eye and multi-level approaches are both needed to understand genomic organization; the two approaches are complementary, not antithetical.
5.4 PRICE'S EQUATION VERSUS CONTEXTUAL ANALYSIS REVISITED
Above we discussed how a one-locus population genetics model can be regarded as a multi-level system of the MLS1 type. The Price approach to MLS1 then applies neatly: genic selection operates on fitness differences between genes within organisms, while organismic selection operates on fitness differences between organisms. This implies that genic selection occurs if segregation in the AB heterozygote is distorted, that is, if there is meiotic drive. Since this corresponds to one standard conception of what 'genic selection' means, it is a point in favour of the Price approach.
However, recall the theoretical flaw with the Price approach discussed in Chapter 3: it fails to deal adequately with cross-level by-products. As we saw, contextual analysis addresses this flaw; it constitutes a rival to the Price approach to MLS1, for it partitions the overall change into two components in a different way. What happens when we apply the contextual approach to diploid population genetics, considered as a multi-level system?
Recall the contextual partition:
Collective-level Particle-level selection selection wAz = p2 Var (Z) + ft Var (z)
where z is particle character, Z is collective character (defined as mean particle character), w is particle fitness, and fii and fa are the partial regressions of w on z and Z respectively. It is perfectly possible to apply this partition, rather than the Price partition, to the diploid population genetics model. But interestingly, doing so produces extremely counterintuitive results.
To see why, consider a situation analogous to the case of pure 'soft selection' discussed in Chapter 3. The three diploid genotypes, AA, AB, and BB, have identical fitnesses, that is, waa = wab = wbb. But there are 'organismic effects' on genic fitness—an A allele in an AA homozygote has lower fitness than an A allele in an AB heterozygote. This means that segregation in the heterozygote is distorted in favour of A. Intuitively, all the selection is at the genic level in this example—for the organisms themselves do not differ in fitness. However, the contextual approach will detect a component of organismic selection, for differences in organismic character will help predict differences in genic fitness, controlling for genic character.7 So in the equation above will be non-zero, indicating selection at the organismic level.
This is not the only unpalatable consequence of the contextual approach as applied to diploid population genetics. Consider the following example. Genotypic fitnesses are waa = 16, wab = 12, and wbb = 8. Segregation is distorted in favour of the A allele in the ratio 8:4, that is, of the 12 gametes that an AB organism contributes to the next generation, 8 are A, and 4 are B. Given this fitness scheme, contextual analysis implies that all the selection is at the genic level. For the fitness of a gene is independent of its organismic context—an A gene has a fitness of 8, irrespective of which organism it is in, and a B gene has a fitness of 4, irrespective of which organism it is in.8 So ^2 equals zero, implying that genic selection is the only force in operation. This is deeply implausible, and not something that any evolutionist would want to say.
In short, if we wish to treat diploid population genetics as a multi-level system, the Price approach seems clearly preferable to the contextual approach. This is interesting for two reasons. First, it shows that despite the theoretical argument in favour of the contextual approach, in some cases it gets the answer 'wrong' while the Price approach gets it 'right'. Secondly, it shows that multi-level systems that are formally isomorphic, but have different biological interpretations, can elicit from us very different intuitions about the level(s) at which selection is acting. This point is worth expanding on.
Recall the case that motivated contextual analysis originally: the particles are organisms, the collectives are groups, and there are no 'group effects' on organismic fitness, that is, the fitness of an individual organism depends only on its own phenotype. Biologists are unanimous that there is no group selection in this scenario —all the selection is at the lower level. But the case described in the paragraph before last, where the particles are genes, the collectives are diploid organisms, and there are no 'organismic effects' on genic fitness, is formally identical.
7 To see this, note that if you are trying to predict the fitness of a randomly picked gene from the population and you already know whether it is A or B, additional information about the genotype of its host organism does help you make your prediction.
8 The crucial feature of this example is that segregation in the AB heterozygote is distorted in favour of the A allele in the ratio waa/wbb. Wherever this condition is satisfied, then a gene's fitness will be independent of its organismic context.
However, in this case it seems crazy to assert that all the selection is at the genic level and none at the organismic level. The two cases are formally isomorphic, but they elicit very different intuitions about the levels of selection.
Why is this? The answer, I think, is that the formal isomorphism belies a biologically important difference. In the group selection case, the question we are critically interested in is whether there are 'group effects' on organismic fitness. In the diploid population genetics case, we are not especially interested in whether there are 'organismic effects' on genic fitness. The situation described three paragraphs back, where waa = 16, wab = 12, wbb = 8, and segregation is distorted in the ratio 8:4 in favour of A, is of no theoretical significance at all. This is because the explanation of why the fitness of an A gene is the same, whatever its organismic context, involves two quite disparate circumstances: the fact that segregation is distorted in a certain very specific way, and the fact that genotypic fitnesses are as they are. By contrast, where the fitness of an organism depends on its own phenotype alone, irrespective of group context, this is theoretically significant—it signals the absence of group effects on fitness. So although the fitness structures in the two cases are isomorphic, biologically they are quite unlike.
The disanalogy can be seen in another way. In the group selection case, fitnesses are possessed in the first instance by individual organisms. Group fitness is derivative—a group only has a fitness in virtue of the fitnesses of its constituent organisms. (Recall that we dealing with MLS1). In the diploid population genetics case it is the other way round. Fitnesses are again possessed in the first instance by individual organisms—but they are now the collectives, not the particles. The fitness of a gene within an organism, defined as the number of copies of the gene in the organism's successful gametes, is derivative—a gene only has a fitness in virtue of its host organism's fitness. So the biological explanation of why the fitnesses values are as they are is quite different in the two cases.
What moral should we draw? One might conclude that diploid population genetics should not be treated as a multi-level system at all. After all, diploid organisms are not really groups of two genes; to regard them as such is an idealization. But this conclusion is over hasty. The basic idea of treating an organism's genome as a group of genes and then applying multi-level selection theory has considerable explanatory power, as discussed. And the idea that mei-otic drive constitutes selection at the sub-organismic level is widely accepted. So although the gene—organism relation is in some respects unlike the organism—group relation, there seems nothing in principle wrong with treating diploid population genetics as a multi-level system.
I think the correct moral is twofold. First, since the Price approach sometimes works better than the contextual approach, theoretical arguments for the latter notwithstanding, there cannot be a fully general solution to the problem of causally decomposing the total evolutionary change in an MLS1 scenario. Secondly, the fact that two fitness structures can be formally isomorphic, yet generate different intuitions about the levels of selection, shows that the biological interpretation of the fitness structure is also crucial (cf. van der Steen and van den Berg 1999). This indicates a limit on the extent to which the levels-of-selection question can be addressed in purely abstract terms.
Finally, recall the discussion of the Wimsatt/Lloyd 'additivity criterion' in Chapter 4. There we showed that the relevance of additivity is different, depending on whether we favour the Price or the contextual approach to MLS1. On the Price approach, additivity of variance is wholly irrelevant to determining the level(s) of selection, but on the contextual approach it is partly relevant. We treated this as a partial vindication of Wimsatt and Lloyd, given the theoretical superiority of the contextual approach. But the foregoing results complicate matters, given that the contextual approach yields the 'wrong' answer when applied to diploid population genetics.
In a one-locus model, perfectly additive variance in collective fitness means the absence of any dominance or heterosis, that is, genotype fitness is a linear function of allelic 'dosage', so (waa — wab) = (wab — wbb). Certain theorists, notably Wright (1980), have argued that the difference between 'genic' and 'organismic' selection does indeed depend on whether the variance in genotype fitness is additive. But if we agree that the Price approach is the correct way of applying multi-level theory to diploid population genetics, Wright's conclusion must be rejected. Whether selection is at the genic or organismic level depends on whether the fitness differences are within or between organisms, or both; this has nothing to do with the additivity of variance.9
9 This explains why our partial defence of the additivity criterion in Chapter 3 is compatible with the criticisms of Wimsatt and Lloyd made by Sarkar (1994) and Godfrey-Smith (1992). For these authors' criticisms are directed against the additivity criterion as applied to diploid population genetics; and in this context the criterion does not work, for the reasons given in the text.
Was this article helpful?