The Selfish Gene (Dawkins 1976) was published the year before I entered Macquarie University as an undergraduate and remains in print more than forty years later with millions of copies sold. It is a book that arouses strong emotions. For many of its admirers, its lucid presentation of difficult ideas made new sense of life. For many of its critics, it was a bad (even evil) book. Some saw its message as “We are fundamentally selfish and our better natures are an illusion”; others objected to the denigration of organisms and glorification of genes; yet others objected to the superficiality of viewing cultural change as the selfish propagation of alternative memes. A common thread runs through many of these negative (and contestable) reactions, a skein of unacknowledged mind–body dualism. The Selfish Gene unabashedly used the language of agency and purpose in describing the products of natural selection. Many biologists, especially those seeking a union of the life and physical sciences, would expunge all talk of meaning and purpose from biology because this sullies the purity of their science. Others objected to Richard Dawkins using the language of agency for genes because agency was the proper preserve of human beings. Sometimes these critics were the same reader.
The Selfish Gene emphatically rejected popular notions that evolution acts for the “good of the group” or “good of the species.” Natural selection, Dawkins argued instead, promotes the good of the individual or gene. The Selfish Gene, for the most part, equated individual and genetic fitness but favored the gene, rather than individual, as the fundamental beneficiary of adaptation. Genes were potentially immortal, passing through long series of mortal bodies, each of which was summarily discarded. These disposable bodies were viewed as elaborate survival machines evolved for the support and propagation of their genetic inhabitants.
“Good-of-the-group” arguments did not disappear, however. Group selection had its own passionate advocates who engaged in heated polemics with supporters of the selfish gene. David Sloan Wilson (1980), in particular, developed models in which individuals benefited from working together in groups. His mathematics was error-free, but the assumptions and interpretations of his models were vehemently attacked. His “groups,” it was argued, were not really groups; or his models described individual-level selection, not group-level selection, because individuals benefited from being members of groups rather than sacrificed themselves for the group; and so on. Wilson responded in kind and hostilities escalated.
The defenders of groups argued that selection and adaptation can occur at more than one level: among groups of individuals, among individuals within groups, among cells within individuals, and among genes within cells. Selection of genes occurred at the lowest level of this hierarchy, but genes did not constitute a privileged level of fitness or adaptation. This perspective was generally labeled “hierarchical selection” in the 1980s but is now more commonly known as multilevel selection. The Google Ngram Viewer suggests the crossover year, from numerical excess of “hierarchical selection” to a preponderance of “multilevel selection,” was 1996. I suspect that political considerations contributed to this intriguing memetic shift, with the pluralistic and inclusive connotations of “multilevel” found more appealing than the authoritarian and aristocratic associations of “hierarchy.”
It is impossible to read the arguments for and against multilevel selectionism without detecting political subtexts. The proponents of multiple levels of selection tended to present their models as showing that individuals (by implication ourselves) were kinder, gentler, and less selfish than how individuals were portrayed by gene-selectionists. This was resented by gene-selectionists who felt unfairly portrayed as defenders of selfishness. Beneath the acrimony, both camps were positing a less selfish individual, with selfishness either displaced onto genes or distributed across multiple levels. It seemed to me that the two sides were using a common vocabulary to mean different things. Their disagreements were largely semantic. Both sides were right in their own terms.
I did not read The Selfish Gene until sometime in the 1980s, and encountered the ideas of David Lack, George Williams, Bill Hamilton, Robert Trivers, and John Maynard Smith in their own words, rather than through Richard Dawkins’s elegant prose. Therefore, The Selfish Gene did not change my life as many have told me it changed their lives, but I did read Dawkins’s second book, The Extended Phenotype, soon after its publication in 1982, and that book was a revelation at the start of my doctoral studies. The Extended Phenotype was for me a radical book that undermined the unity of the individual. Dawkins wrote: “I see the concept of inclusive fitness as the instrument of a last-ditch rescue attempt, an attempt to save the individual organism as the level at which we think about natural selection.” In Dawkins’s view, a gene was subject to selection on how its phenotype—its effects in the world—promoted the gene’s own propagation. These selectable effects could be extended beyond the boundaries of an individual’s body into the inanimate environment and into the bodies of other individuals, even individuals of other species. Dawkins expressed his “central theorem” of the extended phenotype as follows: “An animal’s behaviour tends to maximize the survival of the genes ‘for’ that behaviour, whether or not those genes happen to be in the body of the particular animal performing it” (1982, 233).
A second strand of The Extended Phenotype was dissension among genes within organisms. Different genes pursued divergent ends. Genes on the sex chromosomes favored different actions than genes on other chromosomes. Genes in the nucleus and mitochondria disagreed about the value of sons. Segregation distorters were outlaws that flourished at the expense of law-abiding Mendelian conformists. Transposable elements favored their own proliferation with scant regard for benefits to their bodily hosts. I became a convinced gene-selectionist and an advocate of intragenomic conflict. The remainder of this chapter, beginning with the next sentence, is the oldest stratum of the book you are currently reading.
The Extended Phenotype viewed the complex behaviors and structures that have evolved by processes of natural selection as adaptations for the good of the relevant genes (replicators) rather than for the good of individual organisms (vehicles). A common criticism of this view has been that organisms are integrated wholes, in which no gene can replicate without the assistance of many others. The implicit metaphor is of the organism as a machine, with genes as instructions for the assembly of the machine’s component parts. But an alternative metaphor is possible: genes as members of social groups. Societies, like machines, can display intricate mutual dependence and elaborate divisions of labor; but, unlike machines, societies are not designed. Cooperation and coordination cannot be assumed; when present, they must be explained. Social theories vary in the causal relations they posit between individuals and the societies of which they are members: some theories emphasize the power of individual actions to shape society, whereas others emphasize the social constraints on individual freedom. This chapter views properties of organisms as social phenomena that arise from the actions of individual genes, and explores the internal conflicts that can disrupt genetic societies and the social contracts that have evolved to mitigate these conflicts.
Gene-centered theories are often reviled because of their perceived implications for human societies. But even though genes may cajole, deceive, cheat, swindle, or steal, all in pursuit of their own replication, this does not mean that persons must be similarly self-interested. Organisms are collective entities (like firms, communes, unions, charities, teams), and the behaviors and decisions of collective bodies need not mirror those of their individual members. As I write this paragraph, my replicators—my genes and my memes—are in constant debate, even dissension, yet somehow I muddle through. I am glad I am not a unit of selection.
Genes are catalysts. They facilitate chemical reactions but are not themselves consumed. A gene influences its own probability of replication by the reactions it catalyzes, usually indirectly via transcripts and translated products. These effects can be likened to the gene’s strategy in an evolutionary game. When, where, and in what quantity the gene is expressed is part of that gene’s strategy to the extent that changes in the gene’s sequence (mutations) could produce a different pattern of expression. The evolutionary theory of games (Maynard Smith 1982) has usually been phrased in terms of payoffs to individuals rather than their genes. This sleight of hand is possible because outcomes that enhance an individual’s reproductive success also enhance the transmission of most (if not all) of the individual’s genes.
Individual and genetic payoffs are no longer in close harmony when an individual’s actions affect the reproductive success of relatives or when conflicts occur within an individual’s genome. Interactions among relatives can be reconciled with an individualistic perspective by recourse to the concept of inclusive fitness (Hamilton 1964); but intragenomic conflicts pose a more intractable problem, because an individual’s fitness, inclusive or otherwise, is ill-defined when different genes have different fitnesses. Such conceptual difficulties do not arise if genes, rather than individuals, are treated as the strategists.
Why use strategic thinking, which anthropomorphizes genes, instead of the well-developed infrastructure of population genetics? My reasons are pragmatic. Molecular biology reveals that genes are much more sophisticated than the stolid dominant or recessive caricatures of classical genetics. A gene may be expressed in some tissues, and some environments, but not in others; may have multiple alternative transcripts; may respond to signals from other genes; may have a history (be expressed when maternally derived but silent when paternally derived); and so on. Such complexities are difficult to model by traditional genetic methods. Game theory, however, allows evolutionarily stable strategies to be selected from among a large range of alternative patterns of gene expression. The realism of a strategic analysis depends on the realism of the set of alternatives from which candidate strategies are chosen. Some conceivable strategies may be unavailable in the real world, but too restricted a set of alternatives can also mislead.
There are at least two defensible answers to the question “How many words in this book?” The first is the number tallied by the word processor of my computer. In this answer, a word is a string of characters terminated by a space or punctuation mark. Each time “vehicle” appears it is an extra word added to the tally. The second is the size of my vocabulary. In this answer, “replicator” counts as a single word no matter how many times it appears. Gene has a similar ambiguity. It can refer to the group of atoms that is organized into a particular DNA sequence—each time the double helix replicates, the gene is replaced by two new genes—or it can refer to the abstract sequence that remains the same gene no matter how many times the sequence is copied. The material gene (first sense) can be considered to be a vehicle of the informational gene (second sense). In philosophical parlance, the informational gene is a type of which material genes are the tokens.
Debates about the “units of selection” are interminable, partly because different meanings of “gene” are conflated. When hierarchical-selectionists describe the gene as the lowest level in a nested hierarchy of units (species, populations, individuals, cells, genes: Wilson and Sober 1994), their sense is closer to the material gene, whereas when gene-selectionists refer to the gene as the unit of selection (Dawkins 1982), their sense is closer to the informational gene. The informational gene may be represented materially at multiple levels of the vehicular hierarchy, but it is not itself a level of the hierarchy. On this view, the material gene is an ephemeral vehicle of the informational gene. However, the informational gene is not precisely the intended meaning of gene-selectionists. I will call their gene the strategic gene because their sense corresponds to the gene that is a strategist in an evolutionary game.
Every genetic novelty (new informational gene) originates as a modification of an existing gene and is initially restricted to a few vehicles at lower levels of the material hierarchy, solely because it is rare. Therefore, the gene’s material copies will interact with each other only when they are present in different cells of the same body or in the bodies of closely related individuals. If such a gene is ever to become established, it must be able to increase in frequency under these circumstances. As the gene’s frequency increases, its fate may be influenced by selection at higher levels of the hierarchy, but it will still retain the features that ensured its success when rare. Thus, the gene can be said to commit itself to a strategy when rare that it must maintain at all frequencies. The phenotypic effects of successful genes will consequently appear to be adaptations for the good of groups of material genes that interact because of recent common descent. A strategic gene corresponds to such a coterie of material genes and can be considered the unit of adaptive innovation.
The meanings of words (like genes) evolve, and it would be futile to legislate a single meaning of “gene,” just as it would be futile to legislate a single meaning of “word.” Semantic flexibility can even be useful when precise distinctions are unimportant, because it allows subtle shifts of sense without becoming embroiled in long terminological explanations. Occasional inconsistency is sometimes the price of brevity.
A strategic gene is defined by the nature of the interactions among the material copies of an informational gene that influence transmission of its sequence when its copies are rare. If all copies of the informational gene acted in isolation, the only phenotypic effects that would promote its transmission would be effects that directly promote the replication of its individual copies. A strategic gene would then be coextensive with a material gene. Material genes need not act in isolation. For example, material genes that are expressed in the soma of multicellular organisms do not leave direct descendants but promote the transmission of their replicas in the organism’s germline. The strategic gene now corresponds to an organism-sized cluster of material genes. Similarly, a gene in the soma of one individual may promote the transmission of its copies in the germlines of relatives. In this case, the strategic gene becomes a cluster of material genes distributed among some, but not all, members of a family. Such a gene’s strategy could be “Treat all offspring equally,” not because all carry its copies, but because the gene has no way of directing benefits preferentially to the offspring who do.
If a gene copy confers a benefit B on another vehicle at cost C to its own vehicle, its costly action is strategically beneficial if pB > C, where p is the probability that a copy of the gene is present in the vehicle that benefits. Actions with substantial costs therefore require significant values of p. Two kinds of factors ensure high values of p: recognition (green beards) and relatedness (kinship). A “green beard” is present when genes are recognized directly or by their phenotypic effects. By contrast, genes recognize kinship by historical continuity: a mammalian mother learns to identify her own offspring in the act of giving birth; a male preferentially directs resources to the offspring of mothers with whom he has copulated; the other chicks in a nest are siblings; and so on. Under kin selection, a gene’s strategy is blind to the outcome of each toss of the meiotic coin. Thus, the treatment of the members of a class of relatives does not depend on which genes they actually inherit, and p corresponds to a conventional coefficient of relatedness. By contrast, green beard effects will discriminate between brothers with and without the relevant gene.
Green beards gained their name from a thought experiment by Dawkins (1976), who considered the possibility of a gene that caused its possessors to develop a green beard and to be nice to other green-bearded individuals. Since then, a “green beard effect” has come to refer to forms of genetic self-recognition in which a gene in one individual directs benefits to other individuals that possess the gene. The recognition of self and the recognition of non-self are two sides of the same coin. Thus, the rejection of individuals that do not possess a label can also be considered a green beard effect, if the absence of the label is correlated with the absence of the genes responsible for rejection. Green beard effects have often been dismissed as implausible because a single gene has been considered unlikely to specify a label, the ability to recognize the label, and the response to the label. However, these functions could also be performed by two or more closely linked genes that are consistently inherited together.
The distinction between cooperation because of self-recognition (green beards) and historical continuity (kinship) is applicable to many genetic interactions. When homologous centromeres segregate at anaphase I of meiosis, their orderly behavior is made possible by the prior recognition of some degree of sequence identity between homologous chromosomes (a green beard), whereas when sister centromeres segregate at anaphase II, recognition is not necessary because the centromeres have been physically associated since their joint origin from an ancestral sequence (kinship). Similarly, the physical cohesion of the body is made possible because sister cells have remained in intimate contact since their origin from a common zygote (kinship), but this rich source of nutrients is defended against interlopers by an immune system that distinguishes self from non-self (a green beard effect).
Genetic replication makes use of energy and substrates that are supplied by the metabolic economy in much greater quantities than would be possible without a genetic division of labor. These resources are common goods, available to every gene in the cytoplasm. Thus, genetic communities are potentially vulnerable to free riders, genes that take more than they contribute, and the gains of trade from biochemical specialization would not have been possible without the evolution of institutions and of procedures that limit the opportunities for social exploitation. In particular, strict controls are expected on access to the machinery of replication.
DNA-based replicators are believed to have evolved from RNA-based replicators, possibly because DNA is copied with greater fidelity than RNA (Lazcano et al. 1988). The change also had implications for cellular security. Communities in which RNA polymerases were responsible for both replication and transcription would have been less easily policed than communities in which replication (“self-aggrandizement”) was performed by DNA polymerases and transcription (“communal labor”) by RNA polymerases. As a bonus, a community that modified its own genes to DNA, and periodically cleansed its cytoplasm with ribonucleases, would in the process eliminate most RNA-based parasites.
An effective way to manage the cytoplasmic commons is to link genes to a single origin of replication, and to exclude nonmembers from the cytoplasm. The chromosome becomes a team whose members’ interests coincide. The solution is egalitarian, at least within the group. Each gene that joins the chromosome has equity and replicates once per cycle, no matter what its contribution during that cycle. Efficiency might conceivably be improved if genes that contributed more to productivity in the local environment were rewarded with increased copy number, but this argument ignores the costs of negotiating fair shares and of policing complex rules. The suppression of internal conflict by replication from a single origin has a price because genes can be copied more quickly from multiple origins than from one (Maynard Smith and Szathmáry 1993).
Leda Cosmides and John Tooby (1981) called a set of genes that replicated together, and whose fitness was maximized in the same way, a coreplicon. They argued that intragenomic conflicts are likely if an organism contains more than one coreplicon, because the members of a coreplicon will sometimes be selected to maximize their own propagation in ways that interfere with the propagation of other coreplicons. Selection for short-term replication and selection for long-term replication may be opposed. A coreplicon that replicated faster than other coreplicons within its cell lineage would increase in frequency. However, if differential replication were costly for cell survival, the cell lineage would eventually be eliminated in competition with other lineages. Thus, the long-term interests of coreplicons that share a cell lineage will coincide if they never have opportunities to form new combinations with other genes in other lineages. Recombination decouples genes’ fates and is therefore essential for the indefinite persistence of intragenomic conflict (Hickey 1982).
Many bacteria contain multiple circular genomes. By convention, one of these circles is designated the bacterial chromosome, and the extra circles are called plasmids. Plasmid replication consumes energy and substrates. Whether a plasmid pays for its keep—from the perspective of chromosomal genes—depends on the metabolic skills that its genes bring to the cell, on whether these skills are required in the current environment, and on the degree of coadaptation between plasmid and chromosome. Many of the genes for antibiotic resistance that are the scourge of modern hospitals are carried on plasmids. Such a plasmid may be essential in the presence of antibiotics but a burden in their absence (Eberhard 1980).
Most plasmids promote conjugation between their host and other bacteria, although some smaller plasmids rely on larger plasmids for these functions. In the process, a copy of the plasmid is retained by the donor cell and an uninfected bacterium acquires a plasmid. Thus, conjugation allows plasmids to colonize new cytoplasms. Chromosomal genes, by contrast, are usually not transferred. Therefore, the chromosome bears the costs of replicating the donated plasmid, and the costs of increased exposure to viruses during conjugation, but the donor chromosome seemingly gains little in return. If the plasmid encodes beneficial functions, these are transferred to a potential competitor. Perhaps, if a plasmid is a burden to chromosomal genes, the chromosome benefits from sharing its cold with rivals.
Plasmids cannot be categorized simply as parasites or mutualists. For example, a plasmid that initially reduced its host’s competitiveness enhanced fitness after plasmid and chromosome were propagated together for five hundred generations (Lenski, Simpson, and Nguyen 1994). A plasmid has two modes of transmission, vertical and horizontal, and selection can favor its propagation by either path. Selection for greater horizontal transmission, at the expense of vertical transmission, will generally increase the costs of the plasmid to chromosomal genes, whereas selection for increased vertical transmission will generally benefit the chromosome. In the limit, when there is no horizontal transmission of the plasmid, the long-term fates of chromosome and plasmid are inexorably linked and they effectively become a single coreplicon. Similar arguments apply to the viruses, transposons, and other coreplicons that populate bacterial cytoplasms.
Plasmids, once acquired, are difficult to discard. Plasmid genes encode multiple functions that ensure their stable transmission within an infected lineage (Nordström and Austin 1989). Many plasmids encode a persistent “poison” and its short-lived “antidote.” Thus, if a cell segregates without the plasmid, it is cut off from its supply of antidote and succumbs to the poison. The gene for the poison can be said to recognize the presence or absence of the gene for the antidote. Because poison and antidote are inherited as a unit, the plasmid can be said to recognize itself (a green beard effect). Such protection rackets take many forms (Lehnherr et al. 1993; Salmon et al. 1994; Thisted et al. 1994). Some plasmids, for example, encode a methylase and its matching restriction enzyme. The methylase modifies bacterial DNA so that it is protected from the restriction enzyme, which cleaves unmodified DNA. Bacteria that lose the plasmid die, because methylation must be restored each time the chromosome replicates. The restriction enzyme simultaneously defends its cytoplasm against viruses and rival plasmids that lack the appropriate methylase, just as a gangster defends his turf (Kusano et al. 1995).
Mitochondria of trypanosomes contain a large DNA maxicircle and many small minicircles. The maxicircle encodes essential genes in garbled form, whereas the minicircles encode guide RNAs that edit the otherwise unreadable transcripts to yield translatable mRNAs (Benne 1994). Could RNA editing have evolved as a minicircle-maintenance system? If so, one would predict that minicircles could also edit DNA and encrypt maxicircle genes in ways that only they could decipher. Minicircles spread from one mitochondrial lineage to another (Gibson and Garside 1990), strengthening their analogy with bacterial plasmids.
A nonrecombining bacterial chromosome is a team that does not change its members (except by mutation). Its social contract is “All for one, and one for all,” not “Every gene for itself.” Chromosomal recombination occurs on rare occasions as a coincidental side effect of the horizontal transfer of plasmids (conjugation) or viruses (transduction). Some bacteria, however, have evolved mechanisms by which DNA is taken from the environment and used to replace homologous sequences of the chromosome, so-called natural transformation. Transformation, unlike conjugation and transduction, is controlled by chromosomal genes (Stewart and Carlson 1986). Uptake of DNA is induced under conditions of nutritional stress and may have evolved primarily as a means of gaining nutrients (Redfield 1993). Nevertheless, the expression of DNA-binding proteins that prevent the degradation of the donor sequence and the induction of the enzymatic machinery of recombination suggest that recombination is not a mere side effect but has been positively selected (Lorenz and Wackerknagel 1994; Stewart and Carlson 1986).
Why should a team replace one of its members? The repair hypothesis views transformation as a means of replacing injured team members (damaged DNA). However, repair is unlikely to be the principal function of transformation because uptake of DNA is not induced by damage to the chromosome (Redfield 1993). The recombinant-progeny hypothesis views transformation as a means of trying out new players. Replacement of one gene by another occurs in a single cell of a clone and does not commit other team members to the new combination, because the old combination survives in other cells of the clone. A team’s chances of remaining successful in a changing environment will presumably be improved by some degree of experimentation with new combinations. The problem is that, for each member of the team, the advantages arise only from changes at positions other than its own. A gene benefits from replacing an established gene on another chromosome, but suffers from being so replaced. The important social question becomes whether some positions are privileged and not subject to replacement—in particular, whether the genes responsible for transformation are themselves transformed.
The development of resistant spores by Bacillus subtilis illustrates the differentiation of soma and germline in simple form. A bacterium undergoes an unequal cell division to produce a mother cell (soma) and a prespore (germline). The mother cell engulfs the prespore, assists in formation of the spore coat, and is then discarded. The process is coordinated by an exchange of signals between mother cell and prespore (Errington 1996). The genes of the mother cell sacrifice themselves for their replicas in the spore. Some bacterial somas are more complex. Myxococcus xanthum is a motile, predatory bacterium that forms a multicellular fruiting body. Individual myxobacteria forage and divide in the soil, but, when nutrients become scarce, they aggregate to form a structure in which sacrificial stalk cells (soma) raise myxospores (germline) above the substrate (Shimkets 1990).
Organisms develop somas to gain the benefits of a cellular division of labor. A soma, however, is a rich resource that can be exploited by genes of other germlines. Therefore, the advantages of somatic specialization can be realized only if the genes of the soma have some degree of confidence that their copies are represented among the beneficiaries of their labor. The simplest means by which genes in somatic cells ensure that their efforts are well directed is physical cohesion between soma and germline. The genes of the Bacillus mother cell can be assured that their copies are present in the prespore because cell division and sporulation take place within an enclosed sporangium that excludes outsiders. However, as somas become larger and more complex, interactions between somatic cells and germ cells become less direct. This creates additional opportunities for parasites to misappropriate somatic effort, and necessitates more elaborate security systems to protect the soma from exploitation. The genes of my liver are almost certain to have copies in my testes (because of my body’s physical cohesion) but my lump of lard and mass of meat would not last long without a sophisticated system of immune surveillance. This example shows how relatedness (cohesion) and green beard effects (immune surveillance) interact to maintain somatic cooperation.
Uninterrupted physical cohesion cannot protect the genes of somatic cells from exploitation if sister cells become detached to forage (as in Myxococcus) or to form complex organs (as in multicellular animals). Some form of cellular recognition is required. When two cells meet, their responses can be influenced by what they learn about each other. Molecules on their surface can provide clues about which genes are present in a cell and whether the cell is friend, foe, or indifferent. Two categories of molecular interactions can be distinguished. Homotypic interactions occur between identical molecules on the two cells and are a particularly direct means for a gene to recognize itself in other cells. Heterotypic interactions occur between molecules encoded by different genes and can also provide a gene with information about its presence or absence in another cell if there is linkage disequilibrium between the interacting genes (Haig 1996a). Thus, green beard effects may play an important role in the somatic security systems of multicellular organisms (particularly making use of the complete linkage disequilibrium between genes of different species).
The origin of molecules that were able to discriminate between themselves and closely similar molecules greatly expanded the strategies available to genes and made possible the evolution of large multicellular bodies. The ancestor of the immunoglobulin superfamily probably interacted with itself in homotypic adhesion or signaling, but the family now includes many heterophilic adhesion molecules as well as the T-cell receptors, MHC antigens, and immunoglobulins of the vertebrate immune system (A. F. Williams and Barclay 1988). The cadherins, to take another example, are a family of cell-surface proteins that bind to copies of themselves on other cells. Nose, Nagafuchi, and Takeichi (1988) introduced the genes for P-cadherin and E-cadherin into a cell line that lacked cadherin activity, creating two sublines that were identical except for this single gene difference. When the cells were mixed, they spontaneously segregated into discrete populations, like oil mixed with water. Cadherins play a pivotal role in organogenesis, but similar mechanisms could clearly be used to distinguish self from non-self.
Slime molds are eukaryotes with a life cycle remarkably similar to Myxococcus (Kaiser 1986). They feed as unicellular amoebae, but aggregate when starved to form a fruiting body with a simple division of labor between spores and somatic stalk. Slime molds are thus particularly vulnerable to somatic exploitation because there is no guarantee that the amoebae who respond to an aggregation signal are members of the same clone, or that a predator will not use the signal to lure amoebae to their doom. The dangers are real, although somewhat mitigated by the mechanisms of cell-surface recognition discussed in the previous section. Dictyostelium caveatum is a predator that responds to the aggregation signals of other species and devours their amoebae before forming its own fruiting body (Waddell 1982). Zygotes of Dictyostelium discoideum produce the aggregation signal and devour haploid amoebae of their own species as they respond to the signal (O’Day 1979). Some strains of Dictyostelium discoideum form chimeric fruiting bodies with amoebae of other strains without contributing to the stalk (Buss 1982).
Chimerism between members of a single species has also been described in animals. Vascular fusion frequently occurs between neighboring genotypes of the colonial urochordate Botryllus schlosseri. The progenitors of germ cells circulate in the blood and will colonize, and in some cases totally replace, the gonads of the neighboring soma (Pancer, Gershon, and Rinkevich 1995). As another example, “hermaphroditic” females of the haplodiploid scale insect Icerya purchasi are host to spermatogenic cells derived from sperm that entered the cytoplasm of an egg, but which failed to fertilize the egg nucleus because they were preempted by another sperm (Royer 1975). Thus, a sperm that fails to fertilize the eggs of the mother can try again with those of the daughter or granddaughter, or persist as a permanent haploid inhabitant of female somas. Occasional winged males are produced from unfertilized eggs (Hughes-Schrader 1948) but must compete for fertilization with the “reduced males” resident in female gonads.
Marmosets and tamarins regularly produce dizygotic twins in a uterus ancestrally designed for singletons (that is, in a simplex uterus, which lacks long uterine horns to keep squabbling offspring apart). The placental circulations of the twins fuse with the result that each adult marmoset carries blood cells derived from its twin (Benirschke, Anderson, and Brownhill 1962). If germ cells were also transferred, and equally mixed between twin brothers, the genes of their respective somas would be indifferent about which brother copulated, although competition within the brothers’ testes and ejaculates could be intense. Chimerism between dizygotic twins is the rule for marmosets but is the exception for human twins (van Dijk, Boomsma, and de Man 1996). It is common, however, between human mothers and their offspring. Fetal cells circulate in a mother’s blood from the early weeks of pregnancy and descendants of these cells may persist in a mother’s body for decades after the child’s birth (Bianchi et al. 1996). Are these cells simply lost, or do they manipulate the maternal soma for the offspring’s benefit?
Exploitation of host somas by pathogens and parasites remains a major problem for multicellular organisms. This section’s collection of intraspecific chimeras emphasizes that the risk of somatic exploitation is not restricted to members of different species. Of course, somatic exploitation within species usually involves the everyday strategies of coercion and deceit.
The speed of replication limits the amount of DNA that can be efficiently copied from a single origin of replication. The chromosome of Escherichia coli takes about forty minutes to replicate (Zyskind and Smith 1992). If the thousand-fold larger genome of Homo sapiens were similarly organized as a circular chromosome with a single bidirectional origin of replication, it would take almost a month to replicate (Fonstein and Haselkorn 1995; Morton 1991). Humans and other eukaryotes avoid this problem by using multiple origins of replication. The attendant risk that some parts of the genome will replicate faster than others is exacerbated because the alternation of gametic fusion and meiotic segregation creates ample opportunities for rogue elements to colonize new genomes (Hickey 1982). For these reasons, eukaryotes are expected to have evolved sophisticated systems for controlling unauthorized replication.
Two characteristic features of eukaryotic cells probably contribute to replicative security. The first is the separation of the machinery of protein synthesis (in the cytoplasm) from the genetic material (in the nucleus). Passage of large molecules to and from the nucleus is controlled at the nuclear pore complex. Before a protein can dock with this complex, it must possess nuclear localization signals that are recognized by docking molecules in the cytoplasm (Davis 1995; Hicks and Raikhel 1995). The second is the eukaryotic cell cycle. Replication is confined to a specific S phase. Before DNA can replicate, it must acquire a “replication licensing factor” that authorizes it to replicate once, but once only, per cycle (Rowley, Dowell, and Diffley 1994; Su, Follette, and O’Farrell 1995). The origin recognition complex (ORC) that marks a site for future initiation of replication causes transcription to be silenced in its vicinity (Rivier and Pillus 1994). Thus, genes near an ORC are prevented from producing locally acting RNAs that could tamper with the genes’ own replication.
The bread mold Neurospora crassa has evolved a highly effective defense against genetic elements that replicate more than once in a cell cycle. If a sequence is repeated within a haploid nucleus, both copies are inactivated by methylation and subject to a process of repeat-induced point mutation (RIP) until their sequences have diverged sufficiently to be no longer recognized as similar (Selker 1990). Thus, if a DNA sequence replicates faster than other members of its collective, both the additional copies and the master sequence are corrupted by a process of programmed mutation.
Vertebrates compartmentalize their DNA into active regions and methylated regions that are maintained in a compact transcriptionally inactive state (Bestor 1990; Bird 1993). The inactive portion of the genome often contains large amounts of simple repetitive sequences that do not encode proteins and which are subject to high rates of sequence turnover because of replication slippage and unequal crossing over (Dover 1993). This arrangement may function, in part, as a system of defenses against intragenomic parasites. First, a higher proportion of insertions will occur in noncritical sequences. Second, foreign DNA (once inserted) is transcriptionally inactivated by a methylation process that may specifically recognize structural features of parasitic DNA (Bestor and Tycko 1996). Third, inserted DNA is subject to sequence degradation by replication slippage.
It would be misleading to argue that the sole function of the organizational changes of the eukaryotic nucleus has been internal security. Even when agents have identical interests there is still a problem of coordination. The genome of E. coli contains about 4,000 genes whereas the genomes of humans, mice, and pufferfish contain perhaps 20,000 different protein-coding genes. Bird (1995) has argued that the nuclear envelope and histone proteins of eukaryotic cells, and the extensive methylation of vertebrate chromosomes, are adaptations for the reduction of the transcriptional noise associated with larger genomes. New mechanisms of control and security measures would have evolved hand in hand.
Bacterial recombination involves the formation and dissolution of partnerships between coreplicons or the substitution of one gene for another in a process that has clear winners and losers. By contrast, meiotic recombination involves a symmetric relationship in which two temporary teams come together, swap members, and form new temporary teams. The members of successful teams get to play more often in the next generation than members of unsuccessful teams. Thus, a successful player is one who performs well as a member of many different teams, and the system favors teams of champions rather than champion teams. Team members pursue the same goals, not because their long-term destinies are indissolubly linked, but because the rules of meiosis ensure that all receive the same opportunities if only their team can make it through to the next lottery.
If every gene assorted independently at meiosis, players could not form long-term partnerships because any two players present in a haploid team before gametic fusion would have an even chance of parting at meiosis. This 50 percent probability of recombination per generation applies to almost all randomly chosen pairs of genes in organisms with multiple chromosomes. Genes that are linked on the same chromosome can expect to remain associated for longer periods. If some combinations of linked genes work more effectively together than others, these combinations will tend to occur in successful teams and leave more descendants than less favored combinations. By this process, selection generates nonrandom associations of players (i.e., linkage disequilibrium), but these associations are constantly being disrupted by recombination.
One of the major preoccupations of evolutionary genetics has been the question why so many genetic collectives regularly break up successful teams to take a chance on untried combinations. Zhivotovsky, Feldman, and Christiansen (1994) summarized numerous models that reached the conclusion “that, in a random mating population, if a pair of loci is under constant viability selection (the same in both sexes), with recombination between them controlled by a modifying gene, and if this system attains an equilibrium at which the major genes are in linkage disequilibrium, then new alleles at the modifying locus can invade only if they reduce the rate of recombination between the major loci.” A similar principle applies for an arbitrary number of loci (Zhivotovsky, Feldman, and Christiansen 1994). The intuitive explanation of this “reduction principle” is that new teams generated by recombination will, on average, be less successful than existing teams that have survived a generation of selection. Therefore, individual players are more likely to be successful in the next generation if there is less recombination of their current team.
Despite the reduction principle, recombination is widespread in nature. One or more of the assumptions of models that predict selection for reduced recombination must be violated. Genes that increase recombination can be favored if a population has not reached selective equilibrium, because recombination increases the efficiency with which currently favored players are brought together in the same team. Technically, the advantage a team gains from having both gene A and gene B must be less than the sum of their individual contributions to team success (Barton 1995). A similar process favors increased recombination if the cost of injury (mutation) to A and B is greater than the sum of the costs if A and B were damaged individually (Charlesworth 1990). In both these examples, increased recombination improves the efficiency of selection because it reduces the risk that inferior players will “hitch-hike” on the success of their teammates, or, what amounts to the same thing, that superior players will be dragged down by lesser players. Theories that ascribe the adaptive advantage of recombination to increased resistance to parasites are explanations of this sort because recombination rates evolve in a constantly deteriorating environment in which the most-favored allelic combination is always in flux (Hamilton, Axelrod, and Tenese 1990).
The reduction principle also breaks down in the presence of multilocus green beard effects. Green beard effects allow genes to direct benefits to teams in which they have a high probability of being present. As we have seen, a gene (or coalition of genes) can profit from conferring a benefit B on another team at cost C to its own team, if pB > C where p is the probability that the gene (or coalition) is present in the team that benefits. If this probability were the same for all genes in the donor team, all members would gain equally from the transaction. But, if the probabilities differ—as they do when benefits are directed to green-bearded relatives at the expense of other relatives (Ridley and Grafen 1981)—some team members will lose while others gain. Linkage disequilibrium can enable small coalitions of genes to conspire against the common good, but high levels of recombination will disrupt the persistent nonrandom associations on which multilocus green beard effects depend. Other team members would suspect the motivation of a group of players who were simultaneously members of a rival team, and can benefit from disrupting cliques before they form.
The best-studied conspiracies are systems of meiotic drive. A haplotype in a heterozygous diploid causes the failure of gametes that do not carry its copies, usually by means of a two-locus poison–antidote mechanism. If the haplotype does not go to fixation, it must be associated with countervailing fitness costs that will be experienced in full by team members that are unlinked to the haplotype. Therefore, selection at unlinked loci favors increased recombination to disrupt the conspiracy and separate the poison from its antidote (Haig and Grafen 1991). Leigh (1971; see also Eshel 1985) has compared the genome to “a parliament of genes: each acts in its own self-interest, but if its acts hurt the others, they will combine together to suppress it.” Segregation distortion and related phenomena are departures from fairness. “The transmission rules of meiosis,” Leigh suggested, “evolve as increasingly inviolable rules of fair play, a constitution designed to protect the parliament against the harmful acts of one or a few. . . . Just as too small a parliament may be perverted by the cabals of a few, a species with only one, tightly linked chromosome is an easy prey to distorters.”
Most internal conflicts within the nucleus are defused by the procedures of fair segregation and allelic recombination. However, eukaryotes also contain genes, in mitochondria and plastids, that are not part of the meiotic compact. The eukaryotic cell originated as an alliance between nuclear genes and the genes of symbiotic bacteria. Many of the latter eventually joined the nuclear firm, but some retained a limited independence as the mitochondrial and plastid genes of today. We do not fully understand why some genes have accepted (or been granted) nuclear equity whereas others have maintained a separate contractual arrangement, nor why these shifts of allegiance have been predominantly one-way, from organelle to nucleus. Nuclear and organellar genes are mutually dependent, yet their different rules of transmission can be a source of conflict in their partnership.
If different organellar lineages occupied the same cytoplasm after gametic fusion, the lineages would be expected to compete for occupation of the cytoplasm, with concomitant costs to nuclear genes. Cosmides and Tooby (1981) suggested that nuclear genes have been selected to minimize conflicts among organellar genes by causing the destruction of the organelles of one gamete, either before or after the fusion of gametes. For this reason, they proposed, the nuclear genes of one kind of gamete (sperm) discard their organellar partners before fertilizing a different kind of gamete (eggs) that retains its organelles (for related arguments, see Hastings 1992; Hurst and Hamilton 1992; Law and Hutson 1992). Nuclear-enforced suppression of cytoplasmic conflict may thus have been the key factor in the evolution of eggs and sperm, with all other differences between the sexes arising from this initial dichotomy. In support of this conjecture, Hurst and Hamilton (1992) have noted that morphologically distinct sexes are absent in taxa that exchange nuclear genes without cytoplasmic fusion.
Uniparental inheritance of mitochondria and plastids resolves one conflict but creates another. Nuclear genes are transmitted by sperm and eggs, whereas organellar genes are transmitted by eggs alone. Organellar genes would therefore benefit from preventing reproduction by male function, if this increased the resources available for female function. Cytoplasmic male sterility has evolved many times in flowering plants. In all well-studied cases, male sterility is caused by mitochondrial genes, but their effects are often countered by nuclear genes that restore male fertility. Chloroplasts also have predominantly maternal inheritance but chloroplast genes are not known to cause male sterility (Saumitou-Laprade, Cuguen, and Vernet 1994). The plastid genome may lack mechanisms to abort male function, or, if such mechanisms exist, they may be easily circumvented by nuclear genes.
Despite its internal conflicts, the eukaryotic alliance has been an outstanding success. Daniel Dennett (1995, 340–341) considers human beings to be a radically new kind of entity, comparable in importance to the eukaryotic cell. In his view, we are a symbiosis between genetic replicators and cultural replicators (memes). Just as eukaryote cells cannot survive without both nucleus and organelles, we cannot survive without both genes and memes; neither genes nor memes are dispensable; and neither genes nor memes can claim priority as representing our true selves. Genes and memes have very different rules of transmission, and a meme cannot simply be incorporated into a chromosome where it follows the rules of meiosis. Conflicts are therefore expected. Some people will die for an idea. Others will abandon their faith for a sexual fling.
An average gene from a species with two sexes spends equal time in male and female bodies because every individual has a mother and a father. However, at a selective equilibrium some genes (or combinations of genes) may be more successful than average in one sex and less successful than average in the other. Such sexually antagonistic genes will benefit from being associated with other genes that bias sex determination toward the sex in which they have a relative advantage (Rice 1987), whether this is a conventional fitness advantage (viability) or a segregational advantage (meiotic drive). The process is self-reinforcing because genes that influence sex determination spend more time in one sex than the other and can thus persist in linkage disequilibrium with genes whose disadvantage in the less frequent sex is greater than their advantage in the more frequent sex. Genomes thus have a tendency to split into factions that spend equal time in the two sexes (autosomes) and factions that specialize in one sex or the other (sex chromosomes).
Meiotic drive in spermatogenesis or oogenesis (but not both) can favor the evolution of sex chromosomes, because a distorter has an advantage in one sex that is absent in the other. Segregation distorters will also be favored if they arise on existing sex chromosomes. Associations between agents of meiotic drive and the genetic determiners of sex result in biased sex ratios, but these biases will be opposed by the parliament of genes (or at least by its autosomal majority). Half of the genes in the next generation will come from males and half from females. This means that members of a minority sex will leave more descendants on average than members of the majority sex, and autosomal genes will benefit from being present in the minority sex (assuming that the sexes are equally costly). Autosomal genes are expected to enforce fair segregation of sex chromosomes in the heterogametic sex because neither sex will then be in a majority.
Hamilton (1967) recognized that autosomal genes will sometimes favor biased sex ratios. He considered a model in which small numbers of unrelated females founded local populations, their offspring mated among themselves, and the newly mated females dispersed to found new local populations. If males were heterogametic and segregation were strictly Mendelian, the sex ratio in the global population would be very close to unity, with some variation among local populations because of random fertilization by X- and Y-bearing sperm. The expected fitness of a female offspring would be the same as the average female fitness for the global population, regardless of the local sex ratio, but the expected fitness of a male offspring would increase with the local proportion of females. Therefore, an autosomal gene that caused itself to be present in female-biased local populations would have higher than average fitness, and a balanced sex ratio would no longer be the unbeatable strategy. In this example, the parliament of genes contains a number of parties with different policies concerning the sex ratio. The X party, the autosomal party, and the mitochondrial party would enter into coalition against the Y party to force a female-biased sex ratio among offspring, but the coalition partners would lack unanimity about precisely which ratio (Hamilton 1979). Sexual politics can profoundly destabilize the “parliamentary rules” of meiosis (Haig 1993a).
Relatives are genetic collectives that share some, but not all, of their members. A gene can benefit from employing a contingent strategy that treats collectives differently depending on information about the probability r that a collective includes one of its copies. This section will assume that a gene’s only information about r comes from the family tree (pedigree of collectives) and the Mendelian probabilities associated with the pedigree, in some cases supplemented with knowledge of parental origin and the uncertainty of paternity. Green beard effects will not be considered.
The simplest relationship between a diploid mother and her sexually produced diploid offspring is one in which the mother produces a series of eggs that are provisioned, fertilized, and scattered, without subsequent maternal care. Each gene in the mother has an equal probability of being present in each offspring—determined by a flip of the meiotic coin—and the quantity of yolk received cannot be influenced by genes expressed in offspring because provisioning is completed before meiosis. Genes of the mother can do no better than produce the size and number of eggs that maximize the mother’s expected lifetime reproductive success. In a simple model in which eggs are produced sequentially on a production line from a fixed quantity of resources, maternal fitness is maximized when each egg is supplied with an amount of resources such that the marginal benefit (δB) that an offspring would gain from a little bit extra committed to its egg would equal the marginal cost of these resources (δC) to another offspring that develops from an egg at the end of the line. Marginal costs and marginal benefits are given equal weight because a gene in the mother has the same chance of being present in either offspring.
The relationship becomes more complex if offspring receive postzygotic maternal care, because the amount of care received can be influenced by genes expressed in offspring. A gene expressed in the current offspring gains the full marginal benefit of extra resources received from the mother but has only a probability r of being present in the offspring that experiences the marginal cost. Therefore, genes expressed in offspring will favor receiving extra resources as long as δB > rδC, whereas genes expressed in the mother will favor terminating investment once δC > δB. Thus, parent–offspring conflict exists whenever δC > δB > rδC (Trivers 1974; Haig 1992a). This conflict arises from the difficulty of making binding agreements. Even though genes in a parent would agree among themselves to terminate investment in each offspring when δB = δC, the agreement is generally unenforceable once genes find themselves in offspring. All genes would do better if offspring demanded less, but unilateral restraint will be exploited.
The probability r is a half for maternal genes expressed in offspring, whereas r will generally be less than half for paternal genes. This is because the offspring that gains a marginal benefit from extra maternal resources may have a different father from the offspring who suffers the marginal cost. Therefore, paternal genes in offspring are predicted to make greater demands on mothers than are maternal genes in the same offspring. Such conditional strategies are made possible by genomic imprinting, which causes genes to have different patterns of expression depending on whether they spent the previous generation in a male or female germline (Moore and Haig 1991). For example, during murine development, insulin-like growth factor 2 (Igf2) is expressed from the paternal allele and the maternal allele is silent, whereas the insulin-like growth factor 2 receptor (Igf2r) has the opposite pattern of expression. Mice with an inactivated paternal copy of Igf2 are 60 percent normal size at birth, whereas mice with an inactivated maternal copy of Igf2r are 20 percent larger than normal at birth. Birthweight is normal in mice that inherit the inactivated genes from the opposite parent (DeChiara, Robertson, and Efstratiadis 1991; Lau et al. 1994). Igf2r has been proposed to function by degrading the products of Igf2 (Haig and Graham 1991). Thus, these genes employ a conditional strategy of “Make greater demands when paternally derived than when maternally derived.”
The relatedness asymmetry between maternal and paternal genes is maximal for half-siblings, but most kinds of relatives will have different degrees of maternal and paternal relatedness. These patterns can be quite complex. Consider a hypothetical social system in which males disperse and females remain in their natal group. If all offspring within a group are fathered by a single male who maintains a monopoly on sexual access to the females until he is displaced by a new unrelated male, female group members of different ages will be closer relatives on the maternal side than on the paternal side because of female philopatry, whereas offspring of the same age will be either full-siblings or paternal half-siblings. It is not known whether genes have evolved conditional strategies that take account of such patterns of relatedness.
All the well-studied examples of genomic imprinting so far appear to be simple conditional strategies of the form “Do one thing when maternally derived and something else when paternally derived.” More complex conditional strategies are logically possible—for example, “Do one thing when derived from an egg; something else when derived from a sperm of a resident male; and something else when derived from a sperm of a cuckolding male”—but whether such logical possibilities are actually realized depends on costs, benefits, and the existence of appropriate mechanisms. In the social system of the previous paragraph, a gene’s probability of being present in other female group members increases as it passes through successive female germlines. One could imagine a gene subject to a cumulative imprinting effect that was reset to zero every time the gene passed through a male germline.
Our intuitive concept of the genetic boundaries of an organism approximates the membership of a coreplicon (that is, of a set of genes that are transmitted by the same rules). A coreplicon evolves as a unit with common goals because its members benefit from the same outcomes, whereas the genes of different coreplicons can have conflicting interests. Thus, viral genes inserted into a bacterial chromosome are distinguished from “true” bacterial genes because of their alternative mode of transmission. They can be mobilized to outreplicate their companions, package themselves into resistant viral particles, and then burst out of their dying host. The relationship between coreplicons need not, however, be strictly adversarial. One coreplicon can obtain a good from another by trade, as well as by theft, but with room for haggling over the price. A coreplicon functions as a commonwealth, without an internal market. Its members thus avoid the transaction costs of finding buyers and of learning the prices and quality of the goods on offer, and are protected from hucksters and frauds in the marketplace (Coase 1993).
Bacterial cells usually contain a small number of coreplicons (sometimes only one). Recombination between bacterial chromosomes is rare and, when it occurs, is a substitutional process in which one gene (a winner) is substituted for another (a loser). By contrast, eukaryotic recombination is much more frequent, and is a segregational process in which genes are swapped between chromosomes, without winners and losers. The coreplicon is no longer a group of nonrecombining genes who cooperate because their long-term fates are intimately bound; rather, it becomes a group of temporary associates who obey the same rules and who gain an equitable division of resources when their ephemeral partnership is dissolved. High-frequency recombination creates a market (of a sort) for currently favored team players.
The reasons for the eukaryotic sexual cycle of gametic fusion, recombination, and meiotic segregation remain somewhat unclear, but the process probably enhances individual genes’ chances of long-term survival in a changing selective environment. Recombination creates relatives—genetic collectives that share some, but not all, of their genes. Interactions with relatives are a potential source of internal dissension within the collective, because some members of the collective can gain at the expense of others, for example, by sabotaging other members’ gametes or by favoring some offspring over others. High levels of recombination can be a partial solution to the conflicts created by lower levels of recombination, because randomizing devices disrupt the “cabals of the few.”