4 Differences That Make a Differance

My contemplation of the many meanings of “gene” identified a subtle ambiguity in usage. The classical gene was a hypothetical entity that was invoked to explain an observed difference. Genes were identified by the heritable differences they caused. Thus, the gene was implicitly defined as a difference because only a difference can cause a difference. On the other hand, genes were also thought of as physical things. The rival concepts of genes as differences and genes as things each have their uses but, when we address questions of adaptation and natural selection, it is the gene as heritable difference-maker that is the relevant concept. Natural selection “chooses” among alternatives by the criterion of the differences they cause. These alternatives are differences that make a differance. There can be no differential replication without a difference, just as there can be no choice without an alternative, and there can be no cumulative change without heritability of the causes of difference.

Gene-selectionism is the conceptual framework that views genes as the ultimate beneficiaries of adaptations and organisms or groups as means for genes’ ends. Rival conceptual frameworks exist. Multilevel selection theory views genes as the lowest level of a nested hierarchy in which each level is subject to selection and each level can be a beneficiary of adaptations (Sober and Wilson 1994; Wilson and Sober 1994). Developmental systems theory similarly denies a privileged role for genes in development and evolution. In this framework, many things other than genes are inherited and many things other than genes have a causal role in development. It is the entire developmental system, including developmental resources of the environment, that reconstructs itself from generation to generation (Gray 1992; Oyama 2000; Sterelny and Griffiths 1999). Arguments over the relative merits of the different frameworks can be heated. Some of this argument is substantive but much is semantic. Different frameworks define fundamental terms in different ways. Without close attention to these semantic differences, substantive issues can become obscured by mutual incomprehension.

This chapter explicates the nonstandard definitions of gene, phenotype, and environment used by one particular gene-selectionist (myself) in some particular contexts. Although I strive for precision in definition, no definition will be unassailable. Natural selection constantly undermines rigid definitions because it is a process by which things of one kind become things of a different kind. Definitions themselves evolve and words acquire different meanings in different contexts. No language is unobjectionable if a hostile reader can choose how to interpret terms. Mutual understanding should be facilitated by a clarification of how terms are used, but uniformity of definition is neither achievable nor desirable.

My intent is to clarify how central concepts should be defined to achieve a consistent gene-selectionism, not to argue that gene-selectionism is superior to other frameworks. Although I am a gene-selectionist by predilection, I have respect for and sometimes use multilevel selection theory. It is consistent and coherent. I also respect developmental systems theory, especially in its account of development. I see these frameworks as heuristic devices for thinking about evolutionary questions. These alternative frameworks may be better suited to particular temperaments and particular questions.

Phenotypes

Phenotypes have traditionally been defined as properties of organisms and thus must be redefined if phenotypes are to be considered properties of genes. A gene’s effects are its phenotype (Dawkins 1982, 4). In this definition, an effect is simply a difference from what would be observed in the absence of the gene or in the presence of a variant gene, other things being equal. A gene considered in isolation does not have a phenotype. All assignments of phenotypes are based on comparisons (Bouchard and Rosenberg 2004). These comparisons are implicit in the measurement of genic fitness as a change in relative allele frequency. Thus, a gene’s phenotype depends on the implicit or explicit alternative with which it is compared. For some evolutionary questions, comparisons are made between existing variants of a known sequence. For other questions, an existing gene might be compared to a hypothetical alternative, or a hypothetical gene and hypothetical alternative might be posited as making a specified difference in the world without tying the difference to a particular genetic locus.

Consider a sequence variant that has become fixed in a population by a selective sweep. The variant allele was initially present in a single copy in a single cell. Early in the sweep, the allele’s increase in frequency was determined by its effects relative to the previously established allele. Once the sweep is complete, the allele maintains its high frequency, or is itself replaced, depending on its effects relative to new mutations. Some mutations will result in complete loss of function. In this case, the phenotype that is subject to selection is the difference between a functional and nonfunctional allele. Other mutations will cause expression in a new cell type, or a change in alternative splicing, or a change in promoter activity, and so on. For each kind of mutation, there is a different difference under selection.

The definition of phenotype as a gene’s effects changes how environmental and genetic factors are conceptualized. Phenotype is no longer seen as a sum of genetic and environmental influences (plus interaction terms). All phenotypes are effects of genes, but a gene’s phenotype may encompass different effects in different environments (the gene’s repertoire of effects) and includes the gene’s effects on the environment. If there are differences among organisms that are uninfluenced by genes, then these differences are part of no gene’s phenotype. If a frog has a leg amputated by some purely random event, then the absence of the limb is not a phenotype, although how the frog copes, or fails to cope, with the amputation may be part of the phenotype of many genes. We are all interactionists. A definition of phenotype in which all effects are ascribed to genes, but genes’ effects may vary among environments, has much to recommend it if it helps us move beyond sterile debates about nature versus nurture.

A gene achieves its effects by interacting with proteins, RNAs, DNA, and other molecules. The other molecules with which it interacts can be conceptualized as part of the gene’s environment. Of particular importance are a gene’s interactions with RNA polymerases in the process of transcription. But a gene’s phenotype need not be mediated solely by the coding and noncoding RNAs transcribed from its sequence. DNA sequences themselves can adopt multiple conformations depending on conditions in the nuclear environment and these conformations can influence whether the gene is transcribed. For example, the 5′-flanking region of human γ-globin genes responds to low pH by forming an intramolecular triplex in which a purine-rich strand inserts into the major groove of the contiguous double-helix, leaving an unpaired pyrimidine-rich strand. Point mutations that destabilize this structure are associated with hereditary persistence of fetal hemoglobin (Bacolla et al. 1995). Thus, the pH-dependent ability to fold back upon itself and thereby inhibit its own transcription can be considered part of the phenotype of the γ-globin gene. As another example, the imprinted H19 gene interacts directly with imprinted regions on other chromosomes and influences when in the cell cycle these other genes are replicated (Sandhu et al. 2009).

Genes have been viewed as catalysts that facilitate chemical reactions but are not changed by those reactions. Much recent attention, however, has focused on chemical reactions that alter genes. For example, a DNA sequence may be “epigenetically” modified by its interaction with methyltransferases. If alternative DNA sequences exist that are not subject to methylation, then methylation is part of a gene’s phenotype compared to those alternatives. If a methyl group, once attached, is faithfully inherited and more or less permanent, then a methylated and unmethylated sequence can be considered alternative genes compared to each other. But, if a genetic lineage switches back and forth between methylated and unmethylated states, then the ability to switch between states (the gene’s norm of reaction) can be interpreted as part of the gene’s phenotype compared to alternative alleles that do not undergo switching (Haig 2007).

Functions and Side Effects

A gene may have effects that influence the probability that it will be replicated. Genes that promote their own replication will be perpetuated, whereas alternative genes that are less effective replicators will be eliminated. The effects of a DNA sequence may thus be included among the causal factors that account for the presence of the sequence in a gene pool. It is this causal feedback between genotype and phenotype—when combined with a source of genetic novelty (mutation)—that explains how a purposeless process (natural selection) can produce purposeful structures and functions (adaptation). The environment selects among phenotypes and thereby selects among genes. By this means, gene sequences come to embody and represent “information” about what has worked in the environment (Frank 2009; Shea 2007).

The effects of a gene can be classified as either functions (effects that are beneficial for the gene) or side effects (effects that are neutral or harmful for the gene). Effects are judged as functions or side effects by their average contribution to replication over many occurrences not from a single occurrence. A gene’s functions consist of those of its effects that have contributed, however indirectly, to its own transmission from past generations. Insofar as the future repeats the past, such functions will contribute to the gene’s transmission to future generations. All effects of a gene comprise its phenotype and are subject to selection, but only those effects that promote a gene’s replication comprise its functions. There is selection for a gene’s functions but selection of its side effects. If selection is to choose a harmful effect then it must be associated with an even greater beneficial effect (a selective trade-off).

Functions are the adaptations of genes. For an effect to qualify as a function, variant genes must have been eliminated in the past because they lacked the effect, and, if the effect is to remain a function, such variants must continue to be eliminated when they arise. A gene’s effects can change when the environment changes, and a given effect can shift its status from function to side effect or the reverse. There is no inconsistency in saying that an effect was once a function but is so no longer. Teleological language is appropriate when referring to the functions of genes because functions are final causes. They are both causes of a gene’s persistence and effects of the gene.

Environments

A gene’s environment encompasses all factors that are shared with the alternative against which the gene’s effects are measured. It contains not only factors external to the cells and bodies of organisms, but also (and more immediately) these cells and bodies themselves. A body can be viewed as the collectively constructed niche of the genes of which it was the extended phenotype. Among the most important parts of a gene’s environment are the other molecules with which it interacts. Other genes, even other alleles at the same locus, are parts of a gene’s social environment (Fisher 1941; Okasha 2008; Sterelny and Kitcher 1988). Any factor that is experienced by a gene, but not by its alternative, belongs to the gene’s phenotype, not its environment.

Genes may have effects that vary in different environments, in adaptive or nonadaptive ways. If alternative genes experience a similar range of environments but exhibit different responses to these environments, then their norms of reaction are phenotypes subject to selection. Genes may also have effects that modify the environment or phenotype of other genes.

Genetic inheritance and phenotypic development can be conceptualized as orthogonal axes (Bergstrom and Rosvall 2011). The vertical axis represents transmission of old information from progenitors to progeny whereas the horizontal axis represents development of each new generation. Relations between genes and environment differ on the two axes. The environment has primacy on the vertical axis. The information of genes comes from the environment via selection among genetic differences. But, on the horizontal axis of public performance, genes and environment interact to create form. Neither has explanatory primacy. On the vertical axis, genes refer to past environments, whereas on the horizontal axis, genes interact with the current environment.

Genealogical (vertical) and ontogenetic (horizontal) axes are not causally isolated. If a gene makes a selective difference in the current environment, this will be reflected in a changed composition of the gene pool. In any particular generation, entire genomes are selected and one cannot ascribe a selective difference to a particular gene. However, genomes are disassembled and reassembled in each generation by processes of recombination. Therefore, over a series of generations, shorter sequences of DNA are tested against multiple genetic backgrounds. On the ontogenetic axis, the effects of genes are highly nonadditive, because of complex interactions with the environment (which includes other genes), but on the genealogical axis, sustained changes in gene frequency are explained by the average additive effects of smaller parts of the genome (Ewens 2011; Fisher 1941).

Are Genes Dispensable?

To recapitulate, a gene’s world can be divided into phenotype and environment. Phenotype is the difference between the alternatives under selection and environment is the sameness. Phenotype consists of those parts of a gene’s world that differ from the world of the alternative gene to which it is compared whereas environment contains those parts that are shared with the alternative’s world. In this formulation, phenotype and environment are properties of comparisons between things not properties of the things themselves. The environment chooses between the alternative phenotypes and thereby chooses between the genes. Could a simpler story be told in which the environment selects among heritable phenotypic differences without the need to invoke heritable determinants of difference?

The principal reason for invoking determinants is that causality matters. Elsewhere I have considered a model in which heritability of birth weight was high and different birth weights were associated with different probabilities of survival, but there were no fitness differences associated with genetic causes of variation (Haig 2003). Rather, fitness differences were associated solely with environmental contributions to the variance. Perhaps all maternal genotypes produce heavier and healthier babies in better environments. If so, there would be the appearance of directional selection for heavier babies but no response to selection. Perhaps greater environmental perturbations from genotype-specific optimal birth weights result in reduced survival of babies. If so, there would be the appearance of stabilizing selection but no reduction in genetic variance. In both scenarios, birth weight and fitness are correlated, and birth weight is heritable but fitness is not. Heritability of a trait that is correlated with fitness is not enough. Natural selection requires heritability of the causes of variation in fitness. The second reason for invoking genes as heritable determinants of difference is that genes can be considered agents that benefit from the phenotypes they cause.

Are Genes Countable?

George Williams and Richard Dawkins defined a gene as a rarely recombining stretch of DNA that is transmitted intact over multiple generations (Williams 1966, 24; Dawkins 1976, 30). This definition includes stretches of DNA that are the “context-sensitive difference makers” of Sterelny and Griffiths’s (1999, 87) definition of a gene but need not be restricted to that category. The existence of DNA sequences without effects has been presented as a problem for the evolutionary gene concept but the difficulty evaporates if it is conceded that an evolutionary gene need not be subject to selection. Nor is there a problem if different DNA sequences have identical effects. One can identify two genes with the same effects as different genes on the basis of their sequence. A gene without effects is a gene without a phenotype. Such genes make no selective difference, although their relative frequency may change by genetic drift or draft. Similarly, a gene that has erratic effects on its own replication, such that there is no average effect, is not subject to selection, but remains a gene. These are differences that do not make a difference.

The linear extent of an evolutionary gene can be considered to be the distance along a chromosome over which genetic differences are correlated. Consider two neighboring DNA segments X and Y. If P(X) and P(Y) are the relative frequencies of X and Y, and P(XY) is the frequency with which X occurs together with Y, then the distributions of X and Y are statistically nonindependent if P(XY) ≠ P(X)P(Y). In this case, knowing whether X is present provides information about Y, and knowing whether Y is present provides information about X. Such nonindependence of DNA segments is known as linkage disequilibrium. In some cases, blocks of high linkage disequilibrium may be separated from each other by breaks at “hot spots” of recombination. But in many cases, linkage disequilibrium declines gradually with distance so that a chromosome cannot be divided into discrete evolutionary genes. Rather, there is a region of high linkage disequilibrium associated with each polymorphic site that can be considered the evolutionary gene with respect to that site, but evolutionary genes that are defined with respect to different polymorphic sites may overlap.

If X and Y are perfectly associated, then natural selection is indifferent to whether an effect (relative to non-X and non-Y) is due to X or Y alone, to the sum of their effects, or to their interaction. (The causal question can be addressed by experimentally breaking the association and producing X without Y or Y without X.) By contrast, if X and Y are randomly associated, then X can be treated as a variable part of the environment of Y and Y as a variable part of the environment of X. Clearly, the extremes of perfect and random association are the ends of a continuum. As linkage disequilibrium increases, it becomes more convenient to treat X and Y as parts of the same evolutionary gene. As linkage disequilibrium decreases, it becomes more convenient to treat X and Y as part of each other’s environment.

Peter Godfrey-Smith (2009, 135–139) sees the inability to assign nonarbitrary boundaries to evolutionary genes as a major flaw of gene-selectionism. In his view, “A Darwinian population is made up of a collection of definite countable things,” but evolutionary genes are, at best, marginal Darwinian individuals because they fail to satisfy this “definite countable” criterion. He concedes that the lack of countability does not matter much “if one’s point of view is sufficiently pragmatic” and that similar problems sometimes arise in counting organisms, cells, and groups, but the difficulties are, in his view, particularly pronounced for evolutionary genes. They are not the sort of “real entities that undergo the kind of change that Darwin described.” Rather than talk of genes, “in an evolutionary context it is more accurate to talk of genetic material, which comes in smaller and larger chunks, all of which may be passed on and which have various causal roles.”

Godfrey-Smith confounds two kinds of count in his calculation of the number of genes in a bacterial population. First, he counts the number of genes in a bacterium (a few thousand). Second, he counts the number of bacteria in the population (a million). Then, he multiplies these numbers to obtain the number of genes in the population (a few billion). The first number is a count of different kinds of genes. This number is poorly defined because boundaries between genes are indeterminate. Moreover, the items so counted do not constitute a Darwinian population. The second number is a count of how many genes of each kind. This is the size of a Darwinian population, but the count is not affected by where one places the boundaries between genes.

Similar issues arise when counting evolutionary genes in sexual eukaryotes. The number of genes on a chromosome is poorly defined because boundaries between genes are fuzzy, but the number of copies of the X chromosome in a group of organisms is not affected by where one places boundaries between genes. Only the latter number measures the size of a Darwinian population. If the population at a particular site consists of different variants (“alleles”), then selection can be measured by changes in the nonarbitrary numbers of these variants. Linkage disequilibrium is a measure of how representative the count at one polymorphic site is of counts at nearby polymorphic sites.

Many significant things do not have precise boundaries. No line on the ground demarcates where the Rocky Mountains begin and there is no way to count the number of peaks in Colorado without making pragmatic, somewhat arbitrary, choices about what counts as a peak. One could, if one chose, think of North America as made up of smaller and larger chunks of landscape material without naming any topographical features with fuzzy boundaries. Locations and areas could be identified by latitude and longitude, but this would be cumbersome. Evolutionary genes were conceptually central for Williams and Dawkins because such stretches of DNA persist over many generations whereas organisms, cells, and groups are ephemeral. Within their conceptual framework, discrete boundaries are unimportant but persistence is central. Within Godfrey-Smith’s conceptual framework, Darwinian individuals should be clearly identifiable things, but persistence is not one of their essential properties.

DNA blocks defined by strong linkage disequilibrium need not respect the boundaries of protein-coding units. They may be smaller or bigger than such units. As George Williams noted, “Various kinds of suppression of recombination may cause a major chromosomal segment or even a whole chromosome to be transmitted entire for many generations in certain lines of descent. In such cases the segment or chromosome behaves in a way that approximates the population genetics of a single gene” (1966, 24). From this perspective, a mitochondrial genome or the nonrecombining portion of the Y chromosome can be considered a single evolutionary gene, as can the entire genome of an asexual organism.

The definition of linkage disequilibrium, P(XY) ≠ P(X)P(Y), can be generalized to all cases of nonindependence between X and Y. From this perspective, species boundaries are a major cause of linkage disequilibrium. For example, there is complete linkage disequilibrium between the genes of the native red squirrel and the introduced grey squirrel in the English countryside. Grey squirrel DNA has rapidly displaced red squirrel DNA from most British forests with the frequency of all parts of the red squirrel genome changing in concert relative to all parts of the grey squirrel genome. Some DNA segments from red squirrels might be selectively favored in grey squirrel bodies, but they never get the chance because red and grey squirrels cannot interbreed. Natural selection acts on phenotypic differences between the gene pools but does not “see” independent effects of smaller DNA segments. Ecological displacement can be considered a selective process in which the nonrecombining units are the gene pools of the competing species (G. C. Williams 1986). Proponents of the evolutionary gene concept could accommodate interspecific competition by identifying the gene pools as “evolutionary genes” or could avoid the issue by restricting the concept’s application to natural selection within sexually recombining populations.

The Strategic Gene

What is the selfish gene? It is not just one single physical bit of DNA . . . it is all replicas of a particular bit of DNA distributed throughout the world.

—Richard Dawkins (1976)

A proper understanding of the units of selection problem must take account of an important symmetry: Just as organisms are parts of groups, so genes are parts of organisms.

—Elliott Sober and David Sloan Wilson (1994)

A gene that is distributed throughout the world cannot be part of an organism that is localized in space. Whether genes are the “unit of selection” has remained contentious, in part, because different meanings of “gene” are conflated. A first step to untangle this knot is to recognize that “gene” can refer both to a type and to tokens of the type (as well as to collections of tokens of a type). Gene tokens are physical objects but gene types are abstract kinds. It is tempting to simplify matters by suggesting that Dawkins refers to the type when he describes a gene as all replicas of a particular bit of DNA whereas Sober and Wilson refer to tokens when they identify genes as parts of organisms. However, such an attempt to cut the knot fails for the “selfish gene” because universal benevolence is not predicted when all members of a species possess tokens of the same type.

Evolution is often characterized as changes in gene frequency and the phenotypic effects of these changes. Changes in frequency imply counting, but gene tokens are rarely counted; rather, population geneticists usually lump together large numbers of tokens defined by the boundaries of individual organisms and count these collections as a single gene (Queller 2011). Thus, all tokens of a type in a haploid individual are counted as one gene (haploids have one allele), whereas all egg-derived tokens in a diploid individual are counted as one gene and all sperm-derived tokens as another gene (diploids have two alleles). This sleight of hand facilitates the development of simple mathematical models of evolutionary change.

Multilevel selection theory implicitly defines a gene as a single token within a cell in some contexts but as the collection of all tokens of a type within an organism in other contexts. Gene-selectionists also implicitly define the gene as a collection of tokens, but a collection that may be distributed across multiple organisms, for example across the members of a hive. I call a coterie of tokens that act together a strategic gene because it is this collection of gene tokens that can be considered a strategist in an evolutionary game played with other strategic genes.

A gene token is transcribed when an RNA polymerase copies its sequence into a functional RNA (such as a messenger RNA that is translated into a protein) and is replicated when a DNA polymerase takes apart the two strands of its double helix and uses each as a template to produce two new tokens. The strategic gene groups together tokens that cause an effect (actors) with tokens of the same type whose probability of replication is thereby affected (recipients). Actors may be located in somatic cells of a multicellular organism with recipients located in germ cells of the same organism, but actors and recipients may also be tokens of the same type located in different organisms. The strategic gene is not a fixed entity but can evolve to encompass more, or fewer, tokens of its type.

Consider a particular token in a germ cell (the focal token) and trace its ancestry back to the urtoken, the very first token of its type to arise by mutation. From the urtoken, a dichotomously branching tree can be envisaged that represents the history of all tokens of the type, with the focal token at one of the tips of the tree. The path through this tree from focal token to urtoken summarizes the selective history of the focal token (figure 4.1a). In organisms with a strong germ–soma distinction, the tokens on this path are located in germ cells, whereas most of the tokens on lateral branches are located in somatic cells. Tokens on the germ-path may be recipients of effects from tokens on lateral tips of the token-tree (figure 4.1b, c). Selection acts when these effects make a replicative difference (cause a change in relative frequency) relative to tokens of some other type. In this schema, phenotypic effects flow “inward” from somatic actors to germline recipients. These “causal arrows” influence which tokens are replicated but do not change the type of the token.

The extent of the strategic gene is determined by the number of replication cycles that separate the tokens responsible for a phenotypic effect from the tokens that thereby gain a selective advantage. Tokens on remote tips of a token-tree may be unable to exert selective effects on each other’s replication because token-trees are broken up into selectively isolated fragments by spatial dispersal of tokens and mixing with tokens of other types. Selectively isolated tips of a token-tree belong to different strategic genes. Which tokens belong to a strategic gene is determined by the answer to the question, what is it about the effects of tokens of this type that accounts for a focal token being present in the population rather than a token of another type.

The strategic gene navigates a perilous path between the Scylla of the gene token (material gene) and the Charybdis of the gene type (informational gene). It is a collection of tokens but not the collection of all tokens of a type. Its tokens may be distributed across multiple levels of the hierarchy of interactors of multilevel selection theory, but it is not a level of this hierarchy. When viewed in this light, gene-selectionist and multilevel selection frameworks are fundamentally similar ways of describing the same phenomena. The strategic gene combines tokens responsible for a phenotypic effect with tokens of the same type that benefit, directly or indirectly, from the effect. It is the beneficiary of the effects it causes. As such, it is a unit of adaptive innovation and a unit of self-interest.

Historical Kinds

When I first used the type–token distinction to distinguish the informational gene (as type) from the material gene (as token), I thought of the informational gene as corresponding more or less to the classical notion of an eternal form. I now think it is more fruitful to think of the informational gene as a historical kind (Millikan 1999). It comes into existence, is copied, and ceases to exist when copying ceases or an imperfect copy becomes a new historical kind. Conceptualized in this way, the informational gene can be considered from either a diachronic (historical) or synchronic (contemporary) perspective. From the diachronic perspective, the informational gene is an individual with a distinct origin and history (Hull 1978), but from the synchronic perspective it is a historical kind, a collection of things that are similar because of common ancestry.

Strategic genes are also diachronic individuals, with a distinct origin and history. A strategic gene comes into existence when a token becomes causally “detached” from other tokens of its type, and consists of the descendants of this token until they lose their causal cohesion by “death” or “detachment.” Although a strategic gene is an individual, it is not a synchronic historical kind because its tokens are indistinguishable from tokens of other strategic genes that together are the present parts of an informational gene.

The definition of the informational gene as a historical kind creates the seeming difficulty that if the same sequence originated independently in two lineages, there would be two informational genes that were indistinguishable. Would it not be simpler to consider the informational gene an eternal kind defined by an essence rather than its ancestry? My pragmatic answer is that the same sequence, if it has appreciable length, is highly unlikely to have independent evolutionary origins within the age of this universe. Of course, it is possible that two sequences that are already highly similar because of shared ancestry could converge on the same sequence, but there are pragmatic ways of dealing with such cases within the concept of a historical kind. A definition of the informational gene as an eternal chemical kind would create a peculiar ontology. Consider a DNA molecule of length 1,000 nucleotides composed of the four nucleotides A, T, C, and G. There are 4^1,000 distinct eternal kinds of this length (for simplicity of exposition, I am ignoring the many molecular types that differ in ways other than nucleotide sequence). This number (2^2,000) vastly exceeds the number of elementary particles in our universe. Eternal genes are not parts of a practical ontology that carves nature at the joints because this would be an ontology with many more joints than things in the world.

Developmental Systems Framework

Developmental systems theory has been presented as a radical challenge to gene-centered accounts of development and, by extension, to gene-centered accounts of evolution (Gray 1992; Griffiths 1998; Oyama 2000). In the developmental systems framework, genes are just one among many components of a developmental matrix and have no privileged causal role in development. Gene-selectionism is seen as misguided because it implicitly endorses a dualistic account of development in which genes are the carriers of preformed instructions of how to construct phenotypes, with the environment cast in a subsidiary and passive role.

Adaptation by natural selection takes a back seat to ontogenetic questions for proponents of developmental systems theory, whereas adaptation occupies the front seat for proponents of gene-selectionism. A simple resolution would be to propose that the two frameworks address different questions. Such a proposal would probably be perceived as partisan because the premise that ontogenetic and evolutionary questions require different kinds of answer is generally accepted by one side (Dawkins 1982, 98; G. C. Williams 1986) but rejected by the other (Gray 1992, 187; Oyama 2000, 45). Gene-selectionists believe that a conceptual separation of developmental from adaptive explanations aids clarity of thought, whereas developmental systemists believe such separation obscures more than it illuminates.

My intent has been a defense of gene-selectionism, not an attack on the fastness of developmental systems theory. Developmental systemists are on firm ground when they argue that genes do not have a privileged role in a causal account of development, and that phenotypes (in the traditional sense) are constructed by complex, highly nonadditive interactions of genetic and environmental factors (Gray 1992, 172–174). For these reasons, effects of individual genes on the course of development cannot be isolated in developmental time, but this is exactly what natural selection does over the course of many generations as genes are tested in different genetic backgrounds and in a series of environments. Natural selection extracts the average additive effects of genes as the environment chooses among phenotypes.

In the foreword to the second edition of Oyama’s Ontogeny of Information, Lewontin remarked:

Throughout the history of modern biology there has been a confusion between two basic questions about organisms: the problem of the origin of differences and the problem of the origin of state. At first sight these seem to be the same question, and taken in the right direction, they are. After all, if we could explain why each particular organism has its particular form, then we would have explained, pari passu, the differences between them. But the reverse is not true. A sufficient explanation of why two things are different may leave out everything needed to explain their nature. (2000, viii)1

By implication, understanding causes of difference is subsidiary to understanding causes of state. Lewontin (1974, 2000) and Oyama (2000, 52, 155) have perceptively identified an issue on which there is conceptual disagreement between the rival camps.

In the developmental systems framework, genes are nonprivileged components of a developmental matrix and it is a conceptual error to assign phenotypic features to genes rather than to the matrix as a whole. The entire matrix, or life cycle, constructs itself epigenetically in each generation. By the definitions of this chapter, a gene’s environment is co-extensive with the developmental matrix except that the environment is defined to exclude the gene. There are effects of the gene, and all else is environment.

Effects are differences. Natural selection chooses this versus that based on a phenotypic difference. Why we observe these developmental systems and not others is explained, in part, by a long history of selection among differences and thereby selection of particular heritable difference-makers. Selection of differences can result in profound changes of state. One might say that there is selection for the causes of difference and selection of the causes of state.

Gene-selectionists use the language of statistics, of variances, correlations and average effects, whereas developmental systemists prefer “causal” accounts. The contrast, within physics, between mechanics and thermodynamics provides a useful analogy. Thermodynamics is a statistical theory, not an exact causal theory. It makes predictions that are right on average. In principle, a thermodynamic account of any system could always be superseded by a complete mechanical account, but in many circumstances an exact causal account is not practical, nor even possible, nor would it add much to the thermodynamic explanation.

There is, in principle although not in practice, a complete account of all evolutionary change expressed in terms of proximate physical causes that makes no appeal to concepts of selection, information, average effects, and the like. But I will settle for what is practical and predictive. Sober writes, “The strategy of averaging over contexts is the magic wand of genic selectionism. It is a universal tool, allowing all selection processes, regardless of their causal structure, to be represented at the level of the single gene” (1984, 311). I agree, but see this as a strength rather than a weakness of gene-selectionism.

A developmental system exists in which thornbill chicks in a nest are fed by a family of thornbills. This system is reconstructed in each generation as part of the life cycle of thornbills. The nest is a key developmental resource that is constructed afresh in each generation. There is another developmental system in which a cuckoo chick, in a similar nest, is fed by a family of thornbills until the cuckoo is larger than a thornbill. This system, including the nest, is reconstructed in each generation of the thornbill–cuckoo symbiosis. Developmental resources are similar in the two systems. The key difference-maker is the placement, in the nest, of a cuckoo egg with its cargo of cuckoo genes. Developmental systemists see these systems as fundamentally similar, whereas gene-selectionists see them as fundamentally different. Both are right. The origins of adaptation, by selection for the causes of average additive differences, and processes of development, involving highly nonadditive interactions among causes of state, are both fundamental questions.

Are Genes Special?

Many things besides genes are replicated, including membranes, song traditions, burrows, and nests (Sterelny, Smith, and Dickison 1996; Sterelny and Griffiths 1999, 70). Genes, however, possess a peculiar property that distinguishes them from most other heritable difference-makers. Hermann Muller observed that genes catalyze their own replication:

But the most remarkable feature of the situation is not this oft-noted autocatalytic action in itself—it is the fact that, when the structure of the gene becomes changed, through some “chance variation,” the catalytic property of the gene may become correspondingly changed, in such a way as to leave it still autocatalytic. In other words, the change in gene structure—accidental though it was—has somehow resulted in a change of exactly appropriate nature in the catalytic reactions, so that the new reactions are now accurately adapted to produce more material just like that in the changed gene itself. (1922, 34)

This prescient passage was written before elucidation of the structure of DNA. We now understand, in considerable molecular detail, how this extraordinary property is achieved by each strand of the double helix acting as a template for the replication of the other strand. Not all chemical changes to DNA molecules are preserved through DNA replication, however (changes to the backbone are not maintained, but changes in the sequence of bases are). Moreover, mechanisms of proofreading and repair have evolved to correct “errors” of replication (Sterelny, Smith, and Dickison 1996). But some changes remain uncorrected, as are subsequent changes to these changes, allowing the exploration of a vast space of possible sequences. Muller recognized that this open-ended property of genetic change had far-reaching consequences:

Thus it is not inheritance and variation which bring about evolution, but the inheritance of variation, and this in turn is due to the general principle of gene construction which causes the persistence of autocatalysis despite the alteration in the structure of the gene itself. Given, now, any material or collection of materials having this one unusual characteristic, and evolution would automatically follow, for this material would, after a time, through the accumulation, competition and selective spreading of the self-propagated variations, come to differ from ordinary inorganic matter in innumerable respects, in addition to the original difference in its mode of catalysis. There would thus result a wide gap between this matter and other matter, which would keep growing wider, with the increased complexity, diversity and so-called “adaptation” of the selected mutable material. (1922, 35)

Not all things that are replicated have the property that changes in their structure, arising by chance and selected by the environment, are transmitted to future generations without compromising autocatalysis. And even if some nongenetic replicators transmit minor changes in this manner, few, if any, have potential for the open-ended adaptive change that is characteristic of DNA sequences. Human cultural evolution clearly has this open-ended quality (Boyd, Richerson, and Henrich 2011), although the nature of the heritable difference-makers of cultural change, if such exist, is disputed. One needs to have evolved very sophisticated organisms by other processes before meaningful cultural evolution can take place. Genes are special (and so, in its own way, is culture).

The strategic gene is a refinement of the metaphor of genes as self-interested agents. The phenotypes that are chosen by natural selection resemble those that would be chosen by a rational agent attempting to ensure its own transmission to future generations. Mindless genes can therefore be viewed as if they make strategic decisions. Some find this metaphor appealing (Dennett 2011; Queller 2011). Others consider it insidious and paranoid (Godfrey-Smith 2009, 144). Agential metaphors are less attractive (perhaps less seductive) for most other hereditary replicators, such as DNA methylation, membranes, nests, or money invested in the stock market. Are gene-selectionists inconsistent, or is there some principled difference between genes and nonagential replicators? I believe the difference resides in the sophistication of genes’ strategies made possible by the open-ended nature of genetic inheritance identified by Muller. Genes are “indefinite hereditary replicators” (Maynard Smith and Szathmáry 1995, 58) that accumulate functional information about what works in the environment to a much greater extent than other candidates for the replicator role.

From Whence Have We Come to Where Are We Going

The formalism of the strategic gene redefines phenotype and environment. All choices involve selection among things that are different against a backdrop of things that are the same. In a “choice” of nature, that which is different is phenotype and that which is the same is environment. Evolution by natural selection is a process that converts phenotype, that which is selected, into environment, that which selects. It is a process by which large events, at the level of ecology and organismal behavior, determine fine structure, at the level of molecular gene sequences. It is the means by which the macrocosm shapes the microcosm.

The heritable difference-maker when two alleles differ by a single nucleotide is the nucleotide difference. This heritable difference is subject to selection on its phenotypic effects in the environmental context of all nucleotides that are the same. If one of the nucleotides becomes fixed in the population, then that nucleotide becomes part of the environment that selects among remaining differences. The “extended phenotype” and “niche construction” are sometimes presented as alternative labels for a single concept, but I would prefer a division of labor in which the former refers to differences under selection and the latter to evolved samenesses of the environment (Haig 2017). Bodies and genomes are constructed niches that select extended phenotypes of genetic differences.

Many readers will think that a “difference that makes a difference” is not what most people mean when they speak of a gene. That is true. In some contexts, it might be convenient to have another term for the concept of a heritable difference-maker. I have already suggested a term that fits the bill—“usit” as shorthand for unit-of-selection—but I am loath to introduce a new term because there is a long history of implicitly defining a gene as a difference, because “unit-of-selection” already has a tangled history, and because I have only created confusion with a previously attempted neologism that most readers loathe (see the Supplement to the Supplement to the Supplement to the Appendix to this book). But usit is now a historical kind. If people “use it,” it will thrive.

Janus was the god of transitions. Natural selection involves a transition from a past difference to a present thing. Mutation is a transition from a past sameness to a present difference of possible futures. This chapter began by contrasting genes as differences and genes as things, but it is perhaps more fruitful to think of these as two faces of change. The selection of a gene, like the selection of a word, acquires meaning from what it is not. Each “choice” of nature requires a phenotypic difference but the outcome is a genotypic thing that embodies information about why it was chosen and that acts in the world. The next chapter considers genes as actors in present organisms and therefore focuses on genes as material things.

1 Isadore Nabi offered the rejoinder: “After all, if we could explain how each organism has evolved its particular form, by the selection of differences, then we would have explained, per stirpes, why it has its particular state. But the reverse is not true. A sufficient explanation of how an organism develops may tell us nothing about why it has its particular form.”